Hasso Tepper [Tue, 15 Apr 2008 23:41:28 +0000 (23:41 +0000)]
Add libc support for gcc stack protector. Compatibility with gcc34 propolice
is preserved.
Sascha Wildner [Tue, 15 Apr 2008 23:00:52 +0000 (23:00 +0000)]
Fix reference. While here, remove trailing whitespace.
Sascha Wildner [Tue, 15 Apr 2008 22:22:24 +0000 (22:22 +0000)]
VOP_ABORTOP(9) was removed before we forked.
Sascha Wildner [Tue, 15 Apr 2008 22:05:38 +0000 (22:05 +0000)]
Add section.
Sascha Wildner [Tue, 15 Apr 2008 22:02:59 +0000 (22:02 +0000)]
Add a device_set_driver(9) manual page.
Taken-from: FreeBSD
Sascha Wildner [Tue, 15 Apr 2008 21:00:56 +0000 (21:00 +0000)]
Add a vflush(9) manual page.
Taken-from: FreeBSD
Sascha Wildner [Tue, 15 Apr 2008 19:38:26 +0000 (19:38 +0000)]
Improve markup.
Sascha Wildner [Tue, 15 Apr 2008 19:19:49 +0000 (19:19 +0000)]
a -> an
Sascha Wildner [Tue, 15 Apr 2008 18:11:32 +0000 (18:11 +0000)]
Sync with FreeBSD (documents -P).
Sascha Wildner [Tue, 15 Apr 2008 18:11:00 +0000 (18:11 +0000)]
Sync with FreeBSD (adds reload).
Sascha Wildner [Tue, 15 Apr 2008 16:26:02 +0000 (16:26 +0000)]
vfs_export_lookup() doesn't have a manual page.
Sascha Wildner [Tue, 15 Apr 2008 10:11:35 +0000 (10:11 +0000)]
Remove hardcoded paths from etc/rc.d/ldconfig and move them to
ldconfig_paths.
Submitted-by: Andreas Hauser <andy@splashground.de>
Dragonfly-bug: <https://bugs.dragonflybsd.org/issue209>
Sascha Wildner [Tue, 15 Apr 2008 08:11:50 +0000 (08:11 +0000)]
Add <sys/types.h> for size_t.
Matthew Dillon [Tue, 15 Apr 2008 01:45:22 +0000 (01:45 +0000)]
Remove debugging assertion.
Matthew Dillon [Tue, 15 Apr 2008 01:16:14 +0000 (01:16 +0000)]
Add __sreadahead() to help with pkgsrc's devel/m4.
Sascha Wildner [Mon, 14 Apr 2008 22:18:14 +0000 (22:18 +0000)]
s/.Nm/.Xr/
Matthew Dillon [Mon, 14 Apr 2008 20:17:41 +0000 (20:17 +0000)]
Update the documentation for umtx_sleep() and umtx_wakeup().
Matthew Dillon [Mon, 14 Apr 2008 20:12:41 +0000 (20:12 +0000)]
Fix some issues in libthread_xu's condvar implementation.
* Non-broadcast mode is not guaranteed to signal just one waiter, loosen
it up a bit to close race conditions and signal more if necessary.
* Clean up the condition structure. Do not try to track non-broadcast
wakeups. Do not try to block waiting for individual wakeups... the
spec does not require it and, in fact, doing so can create more
problems then it solves.
* Load oldseq from cv->c_seqno *BEFORE* releasing the passed mutex to
close a race. The mutex is there precisely so that userland can
guarantee that no race will occur between waiter and signaler.
Reported-by: Jordan Gordeev <jgordeev@dir.bg>,
"Simon 'corecode' Schubert" <corecode@xxxxxxxxxxxx>
Matthew Dillon [Mon, 14 Apr 2008 20:00:29 +0000 (20:00 +0000)]
Fix a bug in umtx_sleep(). This function sleeps on the mutex's physical
address and will get lost if the physical page underlying the VM address is
copied on write. This case can occur when a threaded program fork()'s.
Introduce a VM page event notification mechanism and use it to wake-up
the umtx_sleep() if the underlying page takes a COW fault.
Reported-by: Jordan Gordeev <jgordeev@dir.bg>,
"Simon 'corecode' Schubert" <corecode@xxxxxxxxxxxx>
Matthew Dillon [Mon, 14 Apr 2008 19:44:18 +0000 (19:44 +0000)]
Minor optimization to LIST_FOREACH_MUTABLE taken from FreeBSD.
Matthew Dillon [Mon, 14 Apr 2008 19:43:02 +0000 (19:43 +0000)]
Don't free held clean pages when asked to clean.
Matthew Dillon [Mon, 14 Apr 2008 12:01:53 +0000 (12:01 +0000)]
Fix ktrace for threaded processes. Move the KTRFAC_ACTIVE flag to the LWP
so a ktrace occuring on one LWP does not cause another LWP to fail to log
a trace entry.
Sascha Wildner [Mon, 14 Apr 2008 08:17:09 +0000 (08:17 +0000)]
Fix various typos in our manual pages.
Matthew Dillon [Mon, 14 Apr 2008 05:40:51 +0000 (05:40 +0000)]
Fix multiple issues with -p<parallel>, including several data corruption
issues.
* Use malloc'd buffers instead of static buffers
* Fix memory leaks
* Fix races with hardlink tracking structures
* Fix bug in remote mknod (it wasn't creating the node on the remote host)
* Fix stack blowouts from deep recursions by starting a new thread.
* Postpend the pid for tmp file creation.
Matthew Dillon [Sat, 12 Apr 2008 17:39:41 +0000 (17:39 +0000)]
pfsync_state doesn't have or need a hash field, the state will be hashed
when it is moved into a real pf_state structure.
Reported-by: Sascha Wildner <saw@online.de>
Sascha Wildner [Sat, 12 Apr 2008 08:38:40 +0000 (08:38 +0000)]
Add MLINKS for MD[245]Pad.3.
Sascha Wildner [Fri, 11 Apr 2008 20:22:18 +0000 (20:22 +0000)]
Start sentence on a new line (and remove trailing space).
Sascha Wildner [Fri, 11 Apr 2008 18:27:44 +0000 (18:27 +0000)]
* Bump date for recent new features.
* Add -l to the synopsis.
* Some little mdoc fixes.
Matthew Dillon [Fri, 11 Apr 2008 18:21:49 +0000 (18:21 +0000)]
Implement a number of major new features to PF.
* Implement several new options to keep/modulate/synproxy state.
'pickups', 'no-pickups', and 'hash-only'. Example:
pass on $ext_if inet from any to any keep state (pickups)
* Implement a SET directive to set the default keep policy. Example:
set keep-policy keep state (pickups)
pass on $ext_if inet from any to any
* Implement the 'no state' phrase for rules, disabling any previously
specified default state policy for that rule (ala OpenBSD).
'pickups' and 'hash-only' imply no flag restrictions, no-pickups implies
flags S/SA. When pickups are enabled the state code will re-establish state
for existing TCP connections. Because the tcp options passed in the SYNs
are not known when this occurs, sequence space checking will be disabled
for those connections. The 'hash-only' mode allows pickups but disables
sequence space checking regardless of whether the SYN packets were observed
or not.
The state code no longer attempts to do sequence space checking if it has
no information on TCP options passed in SYN packets. The SYN from both
ends must be observed for sequence space checking to occur (which is what
happens by default due to flag restrictions).
It should be noted that OpenBSD defaults to 'keep state' with the equivalent
of no-pickups. The problem with this is that rebooting you router or
otherwise losing state will cause existing TCP connections to be reset,
and I do not think this is an appropriate default for PF. DragonFly does
not default to using keep state but introduces the global SET directive
to allow it to be enabled by default.
Finally, note that when using pickups or hash-only, if you desire PF to
not drop TCP connections on loss of state you need to have a few more
rules then usual to pickup on packets going the opposite direction.
For example, if you allow outgoing connections then you ALSO have to keep
state on INCOMING packets with 'flags /S' (i.e. 'established') in order
to ensure that active connections can be re-established on loss of state
by packets in either direction.
This isn't too bad since anyone using queueing has to use multiple rule
sets anyway to ensure that state is established on the proper queue.
In anycase, DragonFly now allows you to pick your poison. One global SET
directive and you can use OpenBSD's policy, or you can make your own.
In-consultation-with: Max Laier <max@love2party.net>
Matthew Dillon [Fri, 11 Apr 2008 17:33:11 +0000 (17:33 +0000)]
Update PORTING instructions for linux to support pthreads.
Matthew Dillon [Fri, 11 Apr 2008 17:18:21 +0000 (17:18 +0000)]
Properly mark a transaction has being completed so the slave side of
the connection does not loop forever on its data.
Sascha Wildner [Fri, 11 Apr 2008 11:18:16 +0000 (11:18 +0000)]
Clean up remains after the Citrus update.
Matthew Dillon [Fri, 11 Apr 2008 08:44:07 +0000 (08:44 +0000)]
Properly detach children so we dont have to pthread_join() them. Fixes
a resource exhaustion issue which could stall cpdup when used with -pN.
Matthew Dillon [Fri, 11 Apr 2008 07:31:05 +0000 (07:31 +0000)]
More cpdup work.
* Add -C which is passed down to ssh to turn on compression.
* Fix a race condition in the hardlink handling code that could occur w/ -pN
* Fix a deadlock in the socket read code. The connection's read mutex
must be released before the master mutex can be re-acquired.
* Reduce the I/O block size when running parallel threads to try to avoid
blocking on write().
Matthew Dillon [Thu, 10 Apr 2008 22:09:08 +0000 (22:09 +0000)]
Add parallel transaction support for remote source or target specifications.
The implementation is a bit crude because I don't want to take too many
chances on a codebase that wasn't originally designed to be multi-threaded,
so the master mutex is only released when a thread is waiting for input
on a socket.
* Add the -p<threads> option and compile -pthread by default. This is only
useful when the source and/or destination is a remote host. Note that
parallel transaction mode will not work with older cpdup binaries on the
remote end.
This greatly improves cpdup's performance when operating on a remote
source and/or target.
* Add -l to force stdout and stderr to be line-buffered.
Hasso Tepper [Thu, 10 Apr 2008 10:21:13 +0000 (10:21 +0000)]
Sync Citrus iconv support with NetBSD.
Obtained-from: NetBSD
Sascha Wildner [Wed, 9 Apr 2008 21:41:59 +0000 (21:41 +0000)]
Bring the list of variables and function docs in time(9) closer to
reality.
Derived-from: FreeBSD
Hasso Tepper [Wed, 9 Apr 2008 07:05:54 +0000 (07:05 +0000)]
Fix ifdefs to make it possible to use time.h in standards compilant code.
Obtained-from: FreeBSD
Hasso Tepper [Wed, 9 Apr 2008 06:44:37 +0000 (06:44 +0000)]
Add double_t and float_t typedefs for both i386 and amd64 as required by C99.
Obtained-from: FreeBSD with modifications
Sascha Wildner [Tue, 8 Apr 2008 13:23:38 +0000 (13:23 +0000)]
Sync with FreeBSD:
* Change semantics of -i (in-place editing) so that it treats each file
independently from other files (desired in most practical cases). Keep
the previous semantics (use a single continuous address space covering
all files to edit in-place) under -I.
* Add -l option (make output line buffered).
* Make the 'y' command (translate) aware of multibyte characters.
* Add case-insensitive matching, using the 'I' flag, a la GNU sed.
* Remove 3rd clause.
* Raise WARNS to 6.
* Miscellaneous fixes and style(9) issues.
Sascha Wildner [Tue, 8 Apr 2008 12:12:53 +0000 (12:12 +0000)]
Remove references to patches that no longer exist.
Peter Avalos [Mon, 7 Apr 2008 01:20:18 +0000 (01:20 +0000)]
Update build for OpenSSH 5.0p1.
Matthew Dillon [Mon, 7 Apr 2008 00:43:44 +0000 (00:43 +0000)]
MFC 1.33/pf.c from NetBSD. Don't apply a window scale to the window
size in a SYN packet.
Partial MFC 1.25/pf.c from NetBSD (specifically 1.487 from OpenBSD).
Fix a bug in the setting of sequence windows in the synproxy case.
Matthew Dillon [Sun, 6 Apr 2008 21:12:42 +0000 (21:12 +0000)]
Bring the 'probability' keyword into PF from NetBSD. This feature allows
a rule to have a probability associated with it which governs whether the
rule is run or not for any given packet.
Suggested-by: =?ISO-8859-1?Q?C=E9dric_Berger?= <cedric@berger.to>
Obtained-from: NetBSD-current
Sascha Wildner [Sun, 6 Apr 2008 19:29:47 +0000 (19:29 +0000)]
Mention ALTQ_FAIRQ.
Sascha Wildner [Sun, 6 Apr 2008 19:29:23 +0000 (19:29 +0000)]
Start sentences on a new line and bump date for fair queueing.
Peter Avalos [Sun, 6 Apr 2008 19:03:18 +0000 (19:03 +0000)]
LK_EXCLUSIVE has no business being in lockinit.
Matthew Dillon [Sun, 6 Apr 2008 18:58:16 +0000 (18:58 +0000)]
Add fairq to altq. Fairq is a fair queueing algorithm with bandwidth
prioritization and a bandwidth delimiter (hogs) to allow low bandwidth
buckets to jump the round robin. This fairq algorithm is currently unweighted
but traffic can still be classified with the global priority model. For
each queue traffic is normally round robined by taking a packet from each
bucket in turn.
This feature is primarily intended for edge routers and egress points with
bandwidth constrictions.
* Hogs feature allows low bandwidth buckets to burst. Low bandwidth can
mean, e.g. an interactive shell or even simply ack traffic, without
the need to explicitly classify it. Bandwidth is managed on a per-bucket
basis.
* Prioritization feature allows minimum guaranteed bandwidths based on
service classifications. e.g. VOIP, web, mail, PtP, etc.
* Weighted fairq not implemented (beyond using classification into priority
queues), but the circular bucket design makes it a fairly easy task if
someone wants to do it.
* Add ALTQ_MBUF_STATE_HASHED and generate a hash of the connection state
in the mbuf header for any packet that you have set 'keep state' for
in pf. This is done in PF and is needed by fairq to bucketize
'connections'.
* Add the fairq implementation and a new ALTQ_FAIRQ kernel build option.
* Simple example included below.
ports="{ 25, 80 }"
altq on vke0 fairq bandwidth 500Kb queue { normal, bulk }
queue bulk priority 1 bandwidth 100Kb \
fairq(buckets 64, hogs 25Kb) qlimit 50
queue normal priority 2 bandwidth 400Kb \
fairq(buckets 64, hogs 25Kb, default) qlimit 50
pass out on vke0 inet proto tcp from any to any \
keep state queue normal
pass out on vke0 inet proto tcp from any to any port $ports \
keep state queue bulk
Peter Avalos [Sun, 6 Apr 2008 17:29:06 +0000 (17:29 +0000)]
Merge from vendor branch OPENSSH:
Import OpenSSH 5.0p1.
Peter Avalos [Sun, 6 Apr 2008 17:29:06 +0000 (17:29 +0000)]
Import OpenSSH 5.0p1.
Sepherosa Ziehau [Sun, 6 Apr 2008 03:23:49 +0000 (03:23 +0000)]
Add a line about netrate/
Reminded-by: swildner@
Sascha Wildner [Sat, 5 Apr 2008 22:10:14 +0000 (22:10 +0000)]
Ansify a bit.
I've left out raising WARNS since there are still warnings left.
Submitted-by: Kevin L. Kane <kevin.kane@gmail.com>
Dragonfly-bug: <https://bugs.dragonflybsd.org/issue77>
Matthew Dillon [Sat, 5 Apr 2008 20:13:04 +0000 (20:13 +0000)]
Fix a snafu with the last commit. Not all of the new AHCI detection support
was properly brought in, causing intel AHCI detection to panic with a
NULL pointer indirection.
Reported-by: Cristi Magherusan <cristi.magherusan@net.utcluj.ro>
Hasso Tepper [Sat, 5 Apr 2008 18:12:30 +0000 (18:12 +0000)]
DRM update to git snapshot from 2008-01-04.
Sascha Wildner [Sat, 5 Apr 2008 09:13:24 +0000 (09:13 +0000)]
Remove trailing space.
Sascha Wildner [Sat, 5 Apr 2008 09:12:45 +0000 (09:12 +0000)]
Fix typo in comment.
Sascha Wildner [Sat, 5 Apr 2008 08:50:41 +0000 (08:50 +0000)]
Sync with FreeBSD (adds -n, -s and -x).
Matthias Schmidt [Sat, 5 Apr 2008 08:07:11 +0000 (08:07 +0000)]
Fix typo.
Noticed-by: swildner@
Sepherosa Ziehau [Sat, 5 Apr 2008 07:57:22 +0000 (07:57 +0000)]
When aarpwhohas() is called, interface's serializer has already been held:
ether_output() -> aarpresolve() -> aarpwhohas()
Sepherosa Ziehau [Sat, 5 Apr 2008 07:40:28 +0000 (07:40 +0000)]
Hold serializer of the correct interface when calling if_output
Sepherosa Ziehau [Sat, 5 Apr 2008 06:28:35 +0000 (06:28 +0000)]
tunoutput() is declared as static function and is used in this way too.
Sepherosa Ziehau [Sat, 5 Apr 2008 05:57:07 +0000 (05:57 +0000)]
sppp_output() is installed in ifnet.if_output and is only called through that
interface, so it has already been serialized.
Sepherosa Ziehau [Sat, 5 Apr 2008 05:30:29 +0000 (05:30 +0000)]
atm_output() must be serialized.
Sascha Wildner [Fri, 4 Apr 2008 19:25:46 +0000 (19:25 +0000)]
Treat the argument as a file name only if it contains a '/'.
This catches the case where a user would do 'man foo' from a directory
which contains a file named foo.
Sascha Wildner [Fri, 4 Apr 2008 19:09:50 +0000 (19:09 +0000)]
Start sentences on a new line and add some markup.
Sepherosa Ziehau [Thu, 3 Apr 2008 16:02:32 +0000 (16:02 +0000)]
Unbreak UP building
Sepherosa Ziehau [Thu, 3 Apr 2008 13:43:29 +0000 (13:43 +0000)]
Parallelize ip_flow:
- Create per-CPU ipflow hash table. During initial ipflow installation, route
entry to the destination is allocated on each CPU and cached in the ipflow,
so each CPU will only access the route entry owned by itself.
- Timeout processing on per-CPU ipflow hash table is initialized by broadcasting
IPI to each CPU. In IPI, a per-CPU netmsg is used to inform CPU local netisr
to do the real timout processing.
- Add sysctl node to show how many ipflow entries are used on the each CPU.
Sepherosa Ziehau [Thu, 3 Apr 2008 12:55:15 +0000 (12:55 +0000)]
Add counters for serializer enter/try
Suggested-by: aggelos@
Sascha Wildner [Wed, 2 Apr 2008 21:25:30 +0000 (21:25 +0000)]
Add support for the Core 2 Duo T7500.
Submitted-by: Cristi Magherusan <cristi.magherusan@net.utcluj.ro>
Joerg Sonnenberger [Wed, 2 Apr 2008 14:46:37 +0000 (14:46 +0000)]
Fix buffer overflow in config parser.
Sepherosa Ziehau [Wed, 2 Apr 2008 14:18:55 +0000 (14:18 +0000)]
- Save statistics
- Factor out common routine to end threads
Sepherosa Ziehau [Wed, 2 Apr 2008 14:16:16 +0000 (14:16 +0000)]
Constify second parameter of timeval{add,sub}()
Sepherosa Ziehau [Wed, 2 Apr 2008 13:27:24 +0000 (13:27 +0000)]
Correct print format
Sepherosa Ziehau [Wed, 2 Apr 2008 13:11:48 +0000 (13:11 +0000)]
- Add two fields in lwkt_serialize to profile serializer contention.
- Expose serializer contention data through em's private sysctl tree.
Sepherosa Ziehau [Wed, 2 Apr 2008 11:48:30 +0000 (11:48 +0000)]
Use sysctl_ctx in softc
Sascha Wildner [Tue, 1 Apr 2008 21:10:26 +0000 (21:10 +0000)]
* MLINK nrelease(7) -> release(7).
* Merge relevant information from nrelease/README.
* Prerequisite for realquickrel is buildworld/-kernel and not necessarily
make release (reported by Max Herrgaard <herrgard@gmail.com>).
Sascha Wildner [Tue, 1 Apr 2008 20:55:06 +0000 (20:55 +0000)]
Make ISOROOT overridable.
Nicolas Thery [Tue, 1 Apr 2008 20:20:29 +0000 (20:20 +0000)]
Use M_ZERO instead of bzero().
Nicolas Thery [Tue, 1 Apr 2008 18:06:34 +0000 (18:06 +0000)]
Migrate allocation of proc structures from zones to kmalloc(). This gives us
MP-safety and does not seem to decrease performance (using make buildworld as
benchmark). Using an objcache seems unnecessary because there isn't much
pre-initialization code we could move to a constructor and there is no
performance gain.
Debatted-With: dillon@, hsu@, corecode@
Sascha Wildner [Tue, 1 Apr 2008 15:22:44 +0000 (15:22 +0000)]
Make the value of PTT_RELAY_SID match the RFC.
This fixes problems with relayed PPPoE.
Reported-by: Jordan Gordeev <jgordeev@dir.bg>
Taken-from: FreeBSD
Matthew Dillon [Sun, 30 Mar 2008 21:33:42 +0000 (21:33 +0000)]
HAMMER 36B/many: Misc debugging.
* Add vfs.hammer.contention_count to report B-Tree node contention
* Conditionalize debugging kprintf's for B-Tree node contention
around vfs.hammer.debug_locks, default disabled.
Matthew Dillon [Sun, 30 Mar 2008 20:39:01 +0000 (20:39 +0000)]
It's frankly long past time that we turn net.inet.tcp.always_keepalive
on by default.
The default test internal is 2 hours of idle (left unchanged). I am not
going to make it shorter but I will note that while you might not have wanted
to have the feature enabled 15 years ago (because many people were using
demand-dial), these days you actually almost universally want the feature
enabled for a multitude of reasons:
(1) Because you will often be running over a NAT and can lose the NAT
entry if the connection stays idle with no traffic at all for too
long a period of time.
(2) Because you will often be running over stateful filters which can also
get confused and lose the table entry if they see no traffic for too
long a period of time.
(3) Because having this turned on guarantees that stuck tcp connections,
especially those generated through batch operations, will eventually
get unstuck. A TCP connection can become stuck if it is waiting on
data and the remote end has closed the connection, but the RST got lost.
Without this feature such connections can remain stuck literally
forever.
In some cases you might even need a far shorter interval but that is another
issue not addressed by this commit.
Sascha Wildner [Sun, 30 Mar 2008 18:11:58 +0000 (18:11 +0000)]
Fix 'testmain' building.
Simon Schubert [Sun, 30 Mar 2008 16:45:04 +0000 (16:45 +0000)]
Sync audio data with other output.
Hasso Tepper [Sun, 30 Mar 2008 14:40:13 +0000 (14:40 +0000)]
Decrease the number of reported stray interrupts from 100 to 10. Problems
with stray interrupt 7 seem to be common nowadays when many computers don't
have parallel ports at all.
Sepherosa Ziehau [Sun, 30 Mar 2008 10:15:46 +0000 (10:15 +0000)]
Clean it up a little bit
Sascha Wildner [Sun, 30 Mar 2008 00:32:33 +0000 (00:32 +0000)]
Remove old time zone files.
Sascha Wildner [Sat, 29 Mar 2008 23:31:07 +0000 (23:31 +0000)]
* Update to Ficl 3.03:
- Newly implemented CORE EXT words: CASE, OF, ENDOF, and ENDCASE. Also
added FALLTHROUGH, which works like ENDOF but jumps to the instruction
just after the next OF.
- Bugfix: John-Hopkins locals syntax now accepts | and -- in the comment
(between the first -- and the }.)
- Bugfix: Changed vmGetWord0() to make Purify happier. The resulting
code is no slower, no larger, and slightly more robust.
* Instead of just not defining a bunch of words when TESTMAIN is set,
provide stubs that at least handle the stack correctly. This makes it
much easier to experiment with loader scripts from userland.
Taken-from: FreeBSD
Matthew Dillon [Sat, 29 Mar 2008 20:12:54 +0000 (20:12 +0000)]
HAMMER 36/many: Stabilization pass.
* Assert that no cursor linkages remain when releasing the last ref on
an inode.
* Rewrite the node lock upgrade and downgrade code and fix a bug where
a no-error (0) from locking the cursors parent was overriding a
possible error from locking the cursors current node, which could
result in B-Tree node corruption or an assertion later on.
Sepherosa Ziehau [Sat, 29 Mar 2008 11:45:46 +0000 (11:45 +0000)]
Add configuration to generate packets with different [sd]port and/or [sd]addr
in round-robin fashion.
Sepherosa Ziehau [Sat, 29 Mar 2008 05:27:07 +0000 (05:27 +0000)]
Add tunable for polling burst max, which is a significant tuning parameter
in addition to polling hz for polling(4) operation
Sepherosa Ziehau [Sat, 29 Mar 2008 04:45:47 +0000 (04:45 +0000)]
Add two tunables to run netisr and udp_thread without mplock, so experiment
could be conducted under controlled environment.
Default values of these two tunables are to run netisr/udp_thread with mplock.
Sepherosa Ziehau [Sat, 29 Mar 2008 03:38:53 +0000 (03:38 +0000)]
ifaddr threads does not need MP lock
Sepherosa Ziehau [Sat, 29 Mar 2008 03:37:15 +0000 (03:37 +0000)]
Add MPSAFE version of netmsg_service_loop()
Sepherosa Ziehau [Fri, 28 Mar 2008 11:03:44 +0000 (11:03 +0000)]
Pull back part of rev1.7
Simon Schubert [Thu, 27 Mar 2008 19:37:15 +0000 (19:37 +0000)]
Fix layout.
Submitted-by: Andreas Burghardt <a-burghardt@gmx.de>
Sepherosa Ziehau [Thu, 27 Mar 2008 11:42:11 +0000 (11:42 +0000)]
Install into /usr/sbin
Matthew Dillon [Thu, 27 Mar 2008 04:28:09 +0000 (04:28 +0000)]
Make sure there is no possibility of a cothread trying to access the
per-cpu globaldata space. cothreads operate like coprocessors and run
outside the cpu abstraction, and thus have no per-cpu abstraction.
Sepherosa Ziehau [Wed, 26 Mar 2008 15:05:33 +0000 (15:05 +0000)]
Remove debug prints
Sepherosa Ziehau [Wed, 26 Mar 2008 14:44:59 +0000 (14:44 +0000)]
Fix bugs concerning cached route entry in UDP inpcb.
For an unconnected and unbound UDP socket, first sending calls in_pcbladdr()
to fix the local port, which may change the target CPU of the next sending.
in_pcbladdr() has a side effect to allocate the route entry cached in inpcb.
If the target CPU after in_pcbladdr() is no longer the current CPU, then
the route entry will be accessed/freed on non-owner CPU during later sending.
Similarly, connect/disconnect a UDP socket may change the target CPU too; the
target CPU may no longer the owner of the cached route entry.
So, for the first sending happens on an unconnected and unbound UDP socket,
the target CPU of next sending is compared with the current CPU. If they
are different, then cached route entry will be freed, so next time a packet
sent on this socket, a new route entry owned by the correct CPU will be
cached. Same target CPU check is applied to UDP socket connect/disconnect.
Originally UDP PRU_CONNECT always happens on CPU0, which will cause problem if
following conditions are met:
- Dst of the cached route entry is different from the dst to be connected
- Cached route entry is not allocated on CPU0
This could happen if two datagram are sent on an unbounded and unconnected UDP
socket, then later connectting this UDP socket will cause cached route entry
being freed on different CPU. To solve this problem, PRU_CONNECT is dispatched
according to existing [lf]{addr,port} pairs.
If in_pcbladdr() fails after altering the cached route entry, the cached route
entry is freed to make sure that freeing this cached route entry happens on
its owner CPU.
Reported-by: y0netan1@
Tested-by: y0netan1@