]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
15 years agoMFV "xmalloc: zero size" fix.
des [Wed, 24 Sep 2008 21:20:44 +0000 (21:20 +0000)]
MFV "xmalloc: zero size" fix.

MFC after: 1 week

15 years agoUndocument dead option.
des [Wed, 24 Sep 2008 21:07:11 +0000 (21:07 +0000)]
Undocument dead option.

MFC after: 3 days

15 years agoAdd back ar(1) which was accidentally decoupled in r183242.
marius [Wed, 24 Sep 2008 20:41:35 +0000 (20:41 +0000)]
Add back ar(1) which was accidentally decoupled in r183242.

Approved by: sam

15 years agoAdd another TTY locking migration macro: tty_getlock().
ed [Wed, 24 Sep 2008 19:43:09 +0000 (19:43 +0000)]
Add another TTY locking migration macro: tty_getlock().

Some subsystems (HPS-USB) like to lock down the TTY through a more
generic approach, namely the regular mtx(9) macro's. Allow the TTY lock
to be obtained through the new macro.

Discussed with: hps, thompsa

15 years agoEnable shared lookups on UFS. There are some remaining issues with forced
jhb [Wed, 24 Sep 2008 18:53:04 +0000 (18:53 +0000)]
Enable shared lookups on UFS.  There are some remaining issues with forced
unmounts, but those are in the VFS lookup code are not UFS specific.

Tested by: pho, kris

15 years agoPart 1 of making shared lookups more resilient with respect to forced
jhb [Wed, 24 Sep 2008 18:51:33 +0000 (18:51 +0000)]
Part 1 of making shared lookups more resilient with respect to forced
unmounts.  When we upgrade a vnode lock from shared to exclusive during
a name cache lookup, fail the lookup with EBADF if the vnode is invalidated
while we are waiting for the exclusive lock.

Also, for correctness (though I'm not sure it can occur in practice),
downgrade an exclusively locked vnode if it should be share locked.

Tested by: pho

15 years agoUpdate description of witness_watch.
jhb [Wed, 24 Sep 2008 18:47:24 +0000 (18:47 +0000)]
Update description of witness_watch.

15 years agoAdd multiple include guards.
jhb [Wed, 24 Sep 2008 18:46:39 +0000 (18:46 +0000)]
Add multiple include guards.

15 years agoAllow a jail's IP alias to be created with an arbitrary netmask.
ru [Wed, 24 Sep 2008 15:18:27 +0000 (15:18 +0000)]
Allow a jail's IP alias to be created with an arbitrary netmask.

MFC after: 3 days

15 years agoFix a crash when calling tty_rel_free() while draining during closure.
ed [Wed, 24 Sep 2008 11:16:09 +0000 (11:16 +0000)]
Fix a crash when calling tty_rel_free() while draining during closure.

Yesterday I got two reports of potential crashes, related to TTY
deallocation during device closure. When a thread is in TF_OPENCLOSE,
draining its output upon closure, we should not allow calls to
tty_rel_free() to happen at the same time. This could cause the TTY to
be torn down twice.

PR: kern/127561
Reported by: KOIE Hidetaka <koie suri co jp>
Discussed with: thompsa

15 years agoWhen dropping a packet and issuing a reset during TCP segment handling,
rwatson [Wed, 24 Sep 2008 11:07:03 +0000 (11:07 +0000)]
When dropping a packet and issuing a reset during TCP segment handling,
unconditionally drop the tcbinfo lock (after all, we assert it lines
before), but call tcp_dropwithreset() under both inpcb and inpcbinfo
locks only if we pass in an tcpcb.  Otherwise, if the pointer is NULL,
firewall code may later recurse the global tcbinfo lock trying to look
up an inpcb.

This is an instance where a layering violation leads not only
potentially to code reentrace and recursion, but also to lock
recursion, and was revealed by the conversion to rwlocks because
acquiring a read lock on an rwlock already held with a write lock is
forbidden.  When these locks were mutexes, they simply recursed.

Reported by: Stefan Ehmann <shoesoft at gmx dot net>
MFC after: 3 days

15 years agoChange the static struct sysentvec and struct Elf_Brandinfo initializers
kib [Wed, 24 Sep 2008 10:14:37 +0000 (10:14 +0000)]
Change the static struct sysentvec and struct Elf_Brandinfo initializers
to the C99 style. At least, it is easier to read sysent definitions
that way, and search for the actual instances of sigcode etc.

Explicitely initialize sysentvec.sv_maxssiz that was missed in most
sysvecs.

No objection from: jhb
MFC after: 1 month

15 years ago- Remove default NIC dependency on ulp headers
kmacy [Wed, 24 Sep 2008 01:19:08 +0000 (01:19 +0000)]
- Remove default NIC dependency on ulp headers
- make toe module build dependent on kernel support

Submitted by: Chelsio Inc.
MFC after: 1 week

15 years agoFix some ifdef logic in the libc syncicache() bit that I somehow missed in the
nwhitehorn [Wed, 24 Sep 2008 01:06:11 +0000 (01:06 +0000)]
Fix some ifdef logic in the libc syncicache() bit that I somehow missed in the
last commit. Now it really can use a runtime-set cacheline size.

15 years agoAllow the cacheline size on PowerPC to be set at runtime. This is essential for
nwhitehorn [Wed, 24 Sep 2008 00:28:46 +0000 (00:28 +0000)]
Allow the cacheline size on PowerPC to be set at runtime. This is essential for
supporting 64-bit CPUs, which often have 128-byte cache lines instead of the
standard 32.

15 years agoRemove spurious duplicated defination of sock.
delphij [Wed, 24 Sep 2008 00:04:51 +0000 (00:04 +0000)]
Remove spurious duplicated defination of sock.

15 years agoImprove rev 183168, so that if /chosen/stdout is connected to the serial
sobomax [Tue, 23 Sep 2008 22:16:23 +0000 (22:16 +0000)]
Improve rev 183168, so that if /chosen/stdout is connected to the serial
port by OF the syscons won't take over console. Only attach syscons to "screen"
if /chosen/stdout is not connected, which could be the case when loader(8)
is booted directly from the OF. This fixes Marcel's Xserver.

Reported by: marcel

15 years agoIn cust_install_files() we ignore CVS directories. In a similar way now
simon [Tue, 23 Sep 2008 18:54:56 +0000 (18:54 +0000)]
In cust_install_files() we ignore CVS directories.  In a similar way now
also ignore .svn directories.

15 years ago- Change all "echo #..." into using a progress print function to make it
simon [Tue, 23 Sep 2008 18:42:35 +0000 (18:42 +0000)]
- Change all "echo #..." into using a progress print function to make it
  possible to make NanoBSD output more quite or verbose.  The default
  output should remain mostly unchanged. [1]
- Add missing shift for -i.
- Clean up usage() so it's now (mostly) sorted alphabetically.
- Make command line argument handling more consistent in the code and
  remove redundant semicolons.

Reviwed by: phk [1]

15 years agoTrack state to determine if the associated TTY device node has been used.
ed [Tue, 23 Sep 2008 17:12:25 +0000 (17:12 +0000)]
Track state to determine if the associated TTY device node has been used.

It turns out our old TTY layer (and other implementations) block when
you read() on a PTY master device of which the slave device node has not
been opened yet. Our new implementation just returned 0. This caused
applications like telnetd to die in a very subtle way (when child
processes would open the TTY later than the first call to select()).

Introduce a new flag called PTS_FINISHED, which indicates whether we
should block or bail out of a read() or write() occurs.

Reported by: Claude Buisson <clbuisson orange fr>

15 years agoregen
sam [Tue, 23 Sep 2008 16:18:15 +0000 (16:18 +0000)]
regen

15 years agoadd missing options
sam [Tue, 23 Sep 2008 16:15:42 +0000 (16:15 +0000)]
add missing options

15 years agoadd missing options
sam [Tue, 23 Sep 2008 16:11:15 +0000 (16:11 +0000)]
add missing options

15 years agoThe kernel implemented 'memcmp' is an alias for 'bcmp'. However, memcmp
obrien [Tue, 23 Sep 2008 14:45:10 +0000 (14:45 +0000)]
The kernel implemented 'memcmp' is an alias for 'bcmp'.  However, memcmp
and bcmp are not the same thing.  'man bcmp' states that the return is
"non-zero" if the two byte strings are not identical.  Where as,
'man memcmp' states that the return is the "difference between the
first two differing bytes (treated as unsigned char values" if the
two byte strings are not identical.

So provide a proper memcmp(9), but it is a C implementation not a tuned
assembly implementation.  Therefore bcmp(9) should be preferred over memcmp(9).

15 years agostyle(9)
obrien [Tue, 23 Sep 2008 14:25:56 +0000 (14:25 +0000)]
style(9)

15 years agoReverse if() logic to improve readability.
obrien [Tue, 23 Sep 2008 14:25:38 +0000 (14:25 +0000)]
Reverse if() logic to improve readability.

Reviewed by: ru

15 years agoAdd missing library dependencies.
ru [Tue, 23 Sep 2008 13:12:51 +0000 (13:12 +0000)]
Add missing library dependencies.

PR: bin/127573
Submitted by: Eygene Ryabinkin

15 years agoUpdate cxgb include paths to not require prefixing with dev/cxgb
kmacy [Tue, 23 Sep 2008 03:16:54 +0000 (03:16 +0000)]
Update cxgb include paths to not require prefixing with dev/cxgb

Submitted by: Chelsio Inc.

15 years agoUpdate to xen specific files for SMP
kmacy [Tue, 23 Sep 2008 03:13:33 +0000 (03:13 +0000)]
Update to xen specific files for SMP

MFC after: 1 month

15 years agoIn preparation for PowerPC G5 support, allow PVO objects to contain page
nwhitehorn [Tue, 23 Sep 2008 03:02:57 +0000 (03:02 +0000)]
In preparation for PowerPC G5 support, allow PVO objects to contain page
table entries for both the 32-bit and 64-bit AIM MMUs.

15 years agoAllow cxgb to be unified across versions by making newer features conditional
kmacy [Tue, 23 Sep 2008 02:22:24 +0000 (02:22 +0000)]
Allow cxgb to be unified across versions by making newer features conditional

Submitted by: Chelsio Inc
MFC after: 3 days

15 years agoChange the DBDMA API to allow DBDMA registers in a subregion of a resource. This...
nwhitehorn [Tue, 23 Sep 2008 02:12:47 +0000 (02:12 +0000)]
Change the DBDMA API to allow DBDMA registers in a subregion of a resource. This is necessary to allow future support of DMA for the various Apple on-board ATA controllers.

MFC after: 1 week

15 years ago- Fix flag check
kmacy [Tue, 23 Sep 2008 01:55:36 +0000 (01:55 +0000)]
- Fix flag check
- Fix adaptive thread sleep
- set oactive when queue is full

15 years ago- Track number of times that the transmit queue overflowed
kmacy [Tue, 23 Sep 2008 01:27:19 +0000 (01:27 +0000)]
- Track number of times that the transmit queue overflowed
- Trivial whitespace cleanup

MFC after: 3 days

15 years ago* Add USB boot support.
bms [Mon, 22 Sep 2008 23:56:36 +0000 (23:56 +0000)]
* Add USB boot support.
* Allow the image name to be renamed via NANO_IMGNAME.
* Propagate TARGET_ARCH into src top level make targets
  explicitly to support cross-building.
* Increase the default size of NanoBSD media from 488MB to
  584MB to accomodate a -CURRENT world.

Reviewed by: phk

15 years agoUse Giant for kbdmux(4) locking. This is to workaround the problem
emax [Mon, 22 Sep 2008 22:08:43 +0000 (22:08 +0000)]
Use Giant for kbdmux(4) locking. This is to workaround the problem
where interrupt handlers may race with kbdmux(4) in polling mode.

PR: kern/127446
Reported by: Eygene Ryabinkin rea-fbsd at codelabs dot ru
Tested by: Eygene Ryabinkin rea-fbsd at codelabs dot ru
MFC after: 1 week

15 years agoAdd my entry to committers-src.dot
rnoland [Mon, 22 Sep 2008 21:53:00 +0000 (21:53 +0000)]
Add my entry to committers-src.dot

Approved by: jhb(mentor)

15 years ago- Don't read the identifier string from the VPD if there's no chip
marius [Mon, 22 Sep 2008 20:55:19 +0000 (20:55 +0000)]
- Don't read the identifier string from the VPD if there's no chip
  containing an Ethernet address fitted as this is yet another thing
  that fails in that case in order to avoid the one second delay
  until pci_read_vpd_reg() times out.
- Const'ify the bge_devs array.

15 years agoClose a race between concurrent calls to ufsdirhash_recycle() and
jhb [Mon, 22 Sep 2008 20:53:22 +0000 (20:53 +0000)]
Close a race between concurrent calls to ufsdirhash_recycle() and
ufsdirhash_free() introduced in my last commit by removing the dirhash
about to be free'd in ufsdirhash_free() from the global dirhash list
before dropping the sx lock.

Tested by: kris

15 years agoMark the first slice of a NanoBSD image 'active' by default.
bms [Mon, 22 Sep 2008 20:21:39 +0000 (20:21 +0000)]
Mark the first slice of a NanoBSD image 'active' by default.
This fixes USB boot (not yet merged to HEAD) with 3 flavours
of BIOS I've seen.

Approved by: phk

15 years agoAdd ng_rmnode_flags() so the caller can pass NG_QUEUE and have the node
thompsa [Mon, 22 Sep 2008 19:28:18 +0000 (19:28 +0000)]
Add ng_rmnode_flags() so the caller can pass NG_QUEUE and have the node
destroyed asynchronously due to locking or other constraints.

Reviewed by: julian

15 years agoIntroduce a hooks layer for the MPSAFE TTY layer.
ed [Mon, 22 Sep 2008 19:25:14 +0000 (19:25 +0000)]
Introduce a hooks layer for the MPSAFE TTY layer.

One of the features that prevented us from fixing some of the TTY
consumers to work once again, was an interface that allowed consumers to
do the following:

- `Sniff' incoming data, which is used by the snp(4) driver.

- Take direct control of the input and output paths of a TTY, which is
  used by ng_tty(4), ppp(4), sl(4), etc.

There's no practical advantage in committing a hooks layer without
having any consumers. In P4 there is a preliminary port of snp(4) and
thompsa@ is busy porting ng_tty(4) to this interface. I already want to
have it in the tree, because this may stimulate others to work on the
remaining modules.

Discussed with: thompsa
Obtained from: //depot/projects/mpsafetty/...

15 years agoFix usage of mac_vnode_check_open() in linuxulator - last argument
trasz [Mon, 22 Sep 2008 18:59:24 +0000 (18:59 +0000)]
Fix usage of mac_vnode_check_open() in linuxulator - last argument
should be VREAD, not FREAD.

Approved by: rwatson (mentor)

15 years agoFix style(9) issue in TTY header files: document function argument names.
ed [Mon, 22 Sep 2008 18:44:09 +0000 (18:44 +0000)]
Fix style(9) issue in TTY header files: document function argument names.

According to style(9), function argument names should only be omitted
for prototypes that are exported to userspace. This means we should
document the function arguments in the TTY header files, because they
are only used in userspace.

While there, change the type of the buffer argument of
ttydisc_rint_bypass() to `const void *' instead of `char *'.

Requested by: attilio
Obtained from: //depot/projects/mpsafetty/...

15 years agoAdd freebsd32 compat shims for ioctl(2)
obrien [Mon, 22 Sep 2008 16:24:36 +0000 (16:24 +0000)]
Add freebsd32 compat shims for ioctl(2)
CDIOREADTOCHEADER and CDIOREADTOCENTRYS requests.

15 years agoMake the make_pseudo_driver.sh shellscript work again.
ed [Mon, 22 Sep 2008 16:10:12 +0000 (16:10 +0000)]
Make the make_pseudo_driver.sh shellscript work again.

It seems this script was broken because of the SYSINIT changes and a
rather awkward variable initialisation. For some reason the
make_device_driver.sh script is also broken, related to BUS_SETUP_INTR.
I have no experience with FreeBSD's interrupt handling, so I hope
someone else is willing to take a look at that shell script.

PR: misc/126435
MFC after: 1 month

15 years agoRegenerate for r183270.
obrien [Mon, 22 Sep 2008 16:09:43 +0000 (16:09 +0000)]
Regenerate for r183270.

15 years agoAdd freebsd32 compat shims for ioctl(2)
obrien [Mon, 22 Sep 2008 16:09:16 +0000 (16:09 +0000)]
Add freebsd32 compat shims for ioctl(2)
MDIOCATTACH, MDIOCDETACH, MDIOCQUERY, and MDIOCLIST requests.

15 years agoadd back regdomain.xml
sam [Mon, 22 Sep 2008 15:37:47 +0000 (15:37 +0000)]
add back regdomain.xml

Noticed by: jhay

15 years agoDocument changes in behaviour due to sparse CPU numbering support.
jkoshy [Mon, 22 Sep 2008 14:31:24 +0000 (14:31 +0000)]
Document changes in behaviour due to sparse CPU numbering support.

15 years agoSupport sparsely numbered CPUs.
jkoshy [Mon, 22 Sep 2008 10:37:02 +0000 (10:37 +0000)]
Support sparsely numbered CPUs.

Requested by: obrien, alfred (long ago)

15 years agoWhen invoking the udp_send() from udp6_send() due to use of a v6-mapped
rwatson [Mon, 22 Sep 2008 06:44:03 +0000 (06:44 +0000)]
When invoking the udp_send() from udp6_send() due to use of a v6-mapped
IPv4 address, first drop the udbinfo and inpcb locks, which will otherwise
be recursed.  This leads to a potential minor race, but is preferable to a
deadlock when acquiring a read lock after a write lock on the inpcb.

MFC after: 3 days
Reported by: Norbert Papke <fbsd-ml@scrapper.ca>, lioux

15 years agoAdd workaround for occasional packet loss issue of JMC250 A2
yongari [Mon, 22 Sep 2008 06:17:21 +0000 (06:17 +0000)]
Add workaround for occasional packet loss issue of JMC250 A2
when it runs on half-duplex media.
While I'm here add register definition for GPREG1. ATM the GPREG1
register is only valid for JMC250 A1/A2.

Submitted by: Ethan at JMicron

15 years agoUnbreak the build.
keramida [Mon, 22 Sep 2008 04:12:27 +0000 (04:12 +0000)]
Unbreak the build.

15 years agoUnbreak G3 support. G3 processors don't have an L3 cache, so we shouldn't try to...
nwhitehorn [Mon, 22 Sep 2008 03:21:02 +0000 (03:21 +0000)]
Unbreak G3 support. G3 processors don't have an L3 cache, so we shouldn't try to program it.

Approved by: marcel (mentor)

15 years agoMIMO power save and RIFS; while here also update per-node state/flags for ampdu
sam [Mon, 22 Sep 2008 00:30:26 +0000 (00:30 +0000)]
MIMO power save and RIFS; while here also update per-node state/flags for ampdu

15 years ago"ampdudensity -" is too subtle; use "NA" for any and accept it as a parameter
sam [Mon, 22 Sep 2008 00:24:47 +0000 (00:24 +0000)]
"ampdudensity -" is too subtle; use "NA" for any and accept it as a parameter

15 years agocleanup bss node setup; syncs w/ p4 state that's been in use for months
sam [Mon, 22 Sep 2008 00:22:09 +0000 (00:22 +0000)]
cleanup bss node setup; syncs w/ p4 state that's been in use for months

15 years agocleanup after merge
sam [Mon, 22 Sep 2008 00:21:24 +0000 (00:21 +0000)]
cleanup after merge

15 years agoFix handling of shortgi: use the local configuration (and implicitly
sam [Mon, 22 Sep 2008 00:14:50 +0000 (00:14 +0000)]
Fix handling of shortgi: use the local configuration (and implicitly
device capabilities) to decide whether to use short gi.  Drivers
inspect ni_flags to decide whether to send a frame w/ short sgi.

15 years agoRIFS support; needs driver callback for dynamic state change
sam [Mon, 22 Sep 2008 00:10:22 +0000 (00:10 +0000)]
RIFS support; needs driver callback for dynamic state change

15 years agoMIMO power save support; still needs callbacks for notifying drivers
sam [Sun, 21 Sep 2008 23:59:14 +0000 (23:59 +0000)]
MIMO power save support; still needs callbacks for notifying drivers
of dynamic state change in station mode.

15 years agoRevamp ht ie handling:
sam [Sun, 21 Sep 2008 23:44:15 +0000 (23:44 +0000)]
Revamp ht ie handling:
o change ieee80211_parse_htcap and ieee80211_parse_htinfo to save only
  internal state obtained from the ie's; no dynamic state such as
  ni_chw is altered
o add ieee80211_ht_updateparams to parse ht cap+info ie's and update
  dynamic node state
o change ieee80211_ht_node_init to not take an htcap ie that is parsed;
  instead have the caller make a separate call as one caller wants to
  parse the ie while another wants to parse both cap+info ie's and
  update state so can better do this with ieee80211_ht_updateparams

These changes fix sta mode state handling where the node's channel
width was shifted to ht20/ht40 prematurely.

15 years agodon't propagate ht bss occupancy changes to legacy vaps;
sam [Sun, 21 Sep 2008 23:32:48 +0000 (23:32 +0000)]
don't propagate ht bss occupancy changes to legacy vaps;
they won't have an htinfo ie to update

15 years agoo widen ni_flags as it's going to be full shortly
sam [Sun, 21 Sep 2008 23:31:17 +0000 (23:31 +0000)]
o widen ni_flags as it's going to be full shortly
o shuffle members to minimize holes

15 years agoAdd current transmit parameters for fixed rate handling so drivers
sam [Sun, 21 Sep 2008 23:25:33 +0000 (23:25 +0000)]
Add current transmit parameters for fixed rate handling so drivers
don't duplicate this.  These are setup according to the role of the
node--the bss node for ap and adhoc modes need to use parameters
that are the least common denomimator of all nodes in the bss;
otherwise we are setting up params for a station joining a bss and
we select those according to the capabilities of the station.

This stuff needs more work as we do extra work due to having setup
in common code paths shared by nodes using both roles.

15 years agochange ieee80211_ampdu_stop to take an explicit reason code
sam [Sun, 21 Sep 2008 23:20:04 +0000 (23:20 +0000)]
change ieee80211_ampdu_stop to take an explicit reason code

15 years agosimplify debug msg
sam [Sun, 21 Sep 2008 23:18:36 +0000 (23:18 +0000)]
simplify debug msg

15 years agoCrypto api changes:
sam [Sun, 21 Sep 2008 23:16:19 +0000 (23:16 +0000)]
Crypto api changes:
o don't use the key index to identify when the driver has been
  asked to allocate a key slot, use an explicit flag; allows
  drivers to force s/w fallback for entries in the global table
o change callback api to allocate driver resources for a crypto key:
  - de-const the key parameter so drivers can muck with the flags
  - on callback failure don't automatically try to setup s/w crypto;
    instead the driver must now mark the key entry for s/w crypto and
    the caller will re-attach the cipher module

NB: api change permits drivers more control over fallback to s/w
    crypto (e.g. based on a limited number of h/w key slots)

15 years agoCleanup AMPDU handling:
sam [Sun, 21 Sep 2008 23:00:19 +0000 (23:00 +0000)]
Cleanup AMPDU handling:

For receive:
o explicitly tag rx frames w/ M_AMPDU instead of passing frames through
  the reorder processing according to the node having HT and the frame
  being QoS data
o relax ieee80211_ampdu_reorder asserts to allow any frame to be passed
  in, unsuitable frames are returned to the caller for normal processing;
  this permits drivers that cannot inspect the PLCP to mark all data
  frames as potential ampdu candidates with only a small penalty
o add M_AMPDU_MPDU to identify frames resubmitted from the reorder q

For transmit:
o tag aggregation candidates with M_AMPDU_MPDU
o fix the QoS ack policy set in ampdu subframes; we only support immediate
  BA streams which should be marked for "normal ack" to get implicit block
  ack behaviour; interestingly certain vendor parts BA'd frames with the
  11e BA ack policy set
o do not assign a sequence # to aggregation candidates; this must be done
  when frames are submitted for transmit (NB: this can/will be handled
  better when aggregation is pulled up to net80211)

15 years agoAlways update ic_lastnonht so we track overlapping legacy bss
sam [Sun, 21 Sep 2008 22:24:07 +0000 (22:24 +0000)]
Always update ic_lastnonht so we track overlapping legacy bss
properly; was doing this only on a change in the protection mode
so the advertised protection mode in the beacon would oscillate.

Submitted by: Chris Zimmermann

15 years agoCleanup starting seq# for tx ampdu:
sam [Sun, 21 Sep 2008 22:22:28 +0000 (22:22 +0000)]
Cleanup starting seq# for tx ampdu:
o use txa_start to form the addba request and purge txa_seqstart
o fill txa_start before calling ic_addba_request to permit drivers
  to override when they handle seq# generation (e.g. mwl)

15 years agocheck for undefined identifiers
sam [Sun, 21 Sep 2008 22:16:03 +0000 (22:16 +0000)]
check for undefined identifiers

15 years agoo fix spelling of EOSP
sam [Sun, 21 Sep 2008 22:09:18 +0000 (22:09 +0000)]
o fix spelling of EOSP
o correct SMPS values

15 years agoadd new build knobs and jigger some existing controls to improve
sam [Sun, 21 Sep 2008 22:02:26 +0000 (22:02 +0000)]
add new build knobs and jigger some existing controls to improve
control over the result of buildworld and installworld; this especially
helps packaging systems such as nanobsd

Reviewed by: various (posted to arch)
MFC after: 1 month

15 years agoAdd the check of the table number.
rik [Sun, 21 Sep 2008 21:46:56 +0000 (21:46 +0000)]
Add the check of the table number.

15 years agoExport IPFW_TABLES_MAX value for compiled in defaults.
rik [Sun, 21 Sep 2008 20:42:42 +0000 (20:42 +0000)]
Export IPFW_TABLES_MAX value for compiled in defaults.

15 years agoMake fstat() on a pseudo-terminal master return sane timestamps.
ed [Sun, 21 Sep 2008 19:24:15 +0000 (19:24 +0000)]
Make fstat() on a pseudo-terminal master return sane timestamps.

Because pseudo-terminal master file descriptors no longer have a vnode
underneath, we have to fill in fstat() values ourselves. Make our
implementation somewhat sane by returning the timestamps of the TTY
device node that corresponds with our file descriptor.

Obtained from: //depot/projects/mpsafettty/...

15 years agoNow that the number of clist consumers have dropped massively, trim down
ed [Sun, 21 Sep 2008 18:12:18 +0000 (18:12 +0000)]
Now that the number of clist consumers have dropped massively, trim down
the code to prevent useless waste of space.

- Remove support for quote bits. There is not a single driver that needs
  these bits anymore. This means putc() now accepts a char instead of an
  int.

- Remove the unneeded catq() and nextc() routines. They were only used
  by the old TTY layer.

- Convert the clist code to use ANSI C prototypes.

15 years agoInclude $NANO_NAME in the completed message. This is nice if you are
simon [Sun, 21 Sep 2008 18:02:00 +0000 (18:02 +0000)]
Include $NANO_NAME in the completed message.  This is nice if you are
building multiple NanoBSD images at once to keep track of what is
running and what isn't.

15 years agoPrevent an integer overflow in vm_pageout_page_stats() on machines with a
alc [Sun, 21 Sep 2008 18:01:34 +0000 (18:01 +0000)]
Prevent an integer overflow in vm_pageout_page_stats() on machines with a
large number of physical pages.

PR: 126158
Submitted by: Dmitry Tejblum
MFC after: 3 days

15 years agoCorrect suser(9) entry.
antoine [Sun, 21 Sep 2008 16:11:11 +0000 (16:11 +0000)]
Correct suser(9) entry.

15 years agoAlready initialize the vfs timestamps inside the cdev upon allocation.
ed [Sun, 21 Sep 2008 14:02:43 +0000 (14:02 +0000)]
Already initialize the vfs timestamps inside the cdev upon allocation.

In the MPSAFE TTY branch I noticed the vfs timestamps inside devfs were
allocated with 0, where the getattr() routine bumps the timestamps to
boottime if the value is below 3600. The reason why it has been designed
like this, is because timestamps during boot are likely to be invalid.

This means that device nodes that are created on demand (posix_openpt())
have timestamps with a value of boottime, which is not what we want.
Solve this by calling vfs_timestamp() inside devfs_alloc().

Discussed with: kib

15 years agoThe vendor area is the proper home for these files now.
simon [Sun, 21 Sep 2008 13:18:25 +0000 (13:18 +0000)]
The vendor area is the proper home for these files now.

15 years agoMove table list to a separate function.
rik [Sun, 21 Sep 2008 12:54:09 +0000 (12:54 +0000)]
Move table list to a separate function.

15 years agoFix error message content.
zec [Sun, 21 Sep 2008 07:33:33 +0000 (07:33 +0000)]
Fix error message content.

Approved by: julian (mentor)
MFC after: 3 days

15 years agofix compilation on 64-bit platform w/ ATH_DEBUG
sam [Sun, 21 Sep 2008 00:30:33 +0000 (00:30 +0000)]
fix compilation on 64-bit platform w/ ATH_DEBUG

15 years agofix memory smash on lp64 platforms; mostly noticeable in user mode
sam [Sun, 21 Sep 2008 00:28:54 +0000 (00:28 +0000)]
fix memory smash on lp64 platforms; mostly noticeable in user mode
as being unable to associate

15 years agoFix build.
kaiw [Sun, 21 Sep 2008 00:27:16 +0000 (00:27 +0000)]
Fix build.

15 years agoeliminate hardwired lists; use the media type to autoconfig
sam [Sun, 21 Sep 2008 00:26:13 +0000 (00:26 +0000)]
eliminate hardwired lists; use the media type to autoconfig

15 years agoAdd support for option "-M", which is used to operate ar(1) in a
kaiw [Sat, 20 Sep 2008 22:10:10 +0000 (22:10 +0000)]
Add support for option "-M", which is used to operate ar(1) in a
script mode like the MRI(Microtec Research Inc.) "librarian" program.

Originally this option is provided by Binutils ar(1) to ease the
transition for developers who are used to writing "librarian" scripts.

We added this option to BSD ar(1) because:

1. Further improve the compatibility with Binutils ar(1).
2. There are still a few software using this -M option. (at least one
in our ports collection)

Suggested by: rink & erwin

15 years ago<osreldate.h> does not have a forward declration of getosreldate(), so
rodrigc [Sat, 20 Sep 2008 21:12:54 +0000 (21:12 +0000)]
<osreldate.h> does not have a forward declration of getosreldate(), so
remove it from this man page.

MFC after: 3 days

15 years agoAllow the d_mmap driver methods to use cdevpriv KPI during verification
kib [Sat, 20 Sep 2008 19:56:02 +0000 (19:56 +0000)]
Allow the d_mmap driver methods to use cdevpriv KPI during verification
phase of establishing mapping.

Discussed with: rwatson, jhb, rnoland
Tested by: rnoland
MFC after: 3 days

15 years agofdescfs, devfs, mqueuefs, nfs, portalfs, pseudofs, tmpfs and xfs
kib [Sat, 20 Sep 2008 19:50:52 +0000 (19:50 +0000)]
fdescfs, devfs, mqueuefs, nfs, portalfs, pseudofs, tmpfs and xfs
initialize the vattr structure in VOP_GETATTR() with VATTR_NULL(),
vattr_null() or by zeroing it. Remove these to allow preinitialization
of fields work in vn_stat(). This is needed to get birthtime initialized
correctly.

Submitted by:   Jaakko Heinonen <jh saunalahti fi>
Discussed on:   freebsd-fs
MFC after: 1 month

15 years agoInitialize va_rdev to NODEV instead of 0 or VNOVAL in VOP_GETATTR().
kib [Sat, 20 Sep 2008 19:49:15 +0000 (19:49 +0000)]
Initialize va_rdev to NODEV instead of 0 or VNOVAL in VOP_GETATTR().
NODEV is more appropriate when va_rdev doesn't have a meaningful value.

Submitted by:   Jaakko Heinonen <jh saunalahti fi>
Suggested by:   bde
Discussed on:   freebsd-fs
MFC after: 1 month

15 years agoInitialize va_rdev to NODEV and va_fsid to VNOVAL before the
kib [Sat, 20 Sep 2008 19:48:24 +0000 (19:48 +0000)]
Initialize va_rdev to NODEV and va_fsid to VNOVAL before the
VOP_GETATTR() call in vn_stat(). Thus if a file system doesn't
initialize those fields in VOP_GETATTR() they will have a sane default
value.

Submitted by:   Jaakko Heinonen <jh saunalahti fi>
Discussed on:   freebsd-fs
MFC after: 1 month

15 years agoInitialize va_flags and va_filerev properly in VOP_GETATTR(). Don't
kib [Sat, 20 Sep 2008 19:46:45 +0000 (19:46 +0000)]
Initialize va_flags and va_filerev properly in VOP_GETATTR(). Don't
initialize va_vaflags and va_spare because they are not part of the
VOP_GETATTR() API. Also don't initialize birthtime to ctime or zero.

Submitted by:   Jaakko Heinonen <jh saunalahti fi>
Reviewed by: bde
Discussed on:   freebsd-fs
MFC after: 1 month

15 years agoInitialize birthtime fields in vn_stat() to prevent stat(2) from
kib [Sat, 20 Sep 2008 19:43:22 +0000 (19:43 +0000)]
Initialize birthtime fields in vn_stat() to prevent stat(2) from
returning uninitialized birthtime. Most file systems don't initialize
birthtime properly in their VOP_GETTATTR().

Submitted by:   Jaakko Heinonen <jh saunalahti fi>
Reviewed by: bde
Discussed on:   freebsd-fs
MFC after: 1 month

15 years agoFix clone destruction, can't use the simple api because that does not remove
thompsa [Sat, 20 Sep 2008 19:38:37 +0000 (19:38 +0000)]
Fix clone destruction, can't use the simple api because that does not remove
the ifnet from cloner's list.  Expose if_clone_destroyif api to do this.

Submitted by: sam

15 years agoFree allocated memory.
rik [Sat, 20 Sep 2008 19:25:02 +0000 (19:25 +0000)]
Free allocated memory.