dragonfly.git
15 years agoRemove outdated hammer.txt.
Sascha Wildner [Fri, 18 Jul 2008 23:29:24 +0000 (23:29 +0000)]
Remove outdated hammer.txt.

15 years agoFix message.
Sascha Wildner [Fri, 18 Jul 2008 23:23:21 +0000 (23:23 +0000)]
Fix message.

15 years ago* Add some lines about HAMMER mirroring which are based on a nice explanation
Sascha Wildner [Fri, 18 Jul 2008 22:54:52 +0000 (22:54 +0000)]
* Add some lines about HAMMER mirroring which are based on a nice explanation
  with examples sent to me by Michael Neumann <mneumann@ntecs.de>.

* Miscellaneous additions and fixes here and there.

15 years agoTry to make fwohci work more reliably. Stop printing 'phy int' to the
Matthew Dillon [Fri, 18 Jul 2008 03:51:28 +0000 (03:51 +0000)]
Try to make fwohci work more reliably.  Stop printing 'phy int' to the
console when it occurs.  Mask the interrupts which are not supported so
they do not reoccur.

There is still a problem with PHY resets... the bus does not always get
scanned after a reset.

15 years agoDetach the SIM when a firewire disk device is disconnected. Leave the
Matthew Dillon [Fri, 18 Jul 2008 03:49:13 +0000 (03:49 +0000)]
Detach the SIM when a firewire disk device is disconnected.  Leave the
code that supported reconnection commented out.

Fix an API violation in some of the code using SLISTs.

15 years agoSet count to a negative value for an initial burst.
Sascha Wildner [Fri, 18 Jul 2008 00:28:04 +0000 (00:28 +0000)]
Set count to a negative value for an initial burst.

15 years agoHAMMER 63/Many: IO Error handling features
Matthew Dillon [Fri, 18 Jul 2008 00:19:53 +0000 (00:19 +0000)]
HAMMER 63/Many: IO Error handling features

This commit removes all the remaining Debugger() calls and KKASSERTs in the
I/O error path.  Errors are now propagated up the call tree and properly
reported.

* Report I/O errors instead of asserting.

* Read or Write errors in the flush path disable flushing and force the mount
  into read-only mode.  Modified buffers are left locked in memory until
  umount to provide a consistent snapshot of the state of the filesystem.

  You must umount and remount to recover the filesystem.  The filesystem
  will automatically rollback to the last valid flush upon remounting.

* umount and umount -f are now able to unmount a HAMMER filesystem that has
  catastrophic write errors (e.g. pulling the USB cable on an external
  drive).

15 years agoWhen dealing with a failed read properly set B_INVAL.
Matthew Dillon [Fri, 18 Jul 2008 00:09:39 +0000 (00:09 +0000)]
When dealing with a failed read properly set B_INVAL.

15 years agoFix multiple bugs in CAM related devices which go away unexpectedly. This
Matthew Dillon [Fri, 18 Jul 2008 00:07:23 +0000 (00:07 +0000)]
Fix multiple bugs in CAM related devices which go away unexpectedly.  This
fixes numerous panics when pulling a USB mass media device in the midst of
heavy I/O.

* The SIM lock was being unlocked via the periph->sim path after periph
  was unheld.  periph can become free and blow up the unlock, so get
  the sim into a local variable first, then release periph.

* The code which waits for CCB completion needs to be a while loop, not
  an if.  It worked anyway, but wasn't very robust.

* Add CAM_SIM_DEREGISTERED to flag when a sim is undergoing deregistration.

* Beef up cam_dead_sim so it works more like a real sim.

* Properly install &cam_dead_sim in the device and periph structures
  related to a SCSI bus, when deregistering the bus.

* Disallow the addition of new devices when deregistering a bus.

* NULL out periph->softc when freeing it.

15 years agoMake some adjustments to the buffer cache:
Matthew Dillon [Fri, 18 Jul 2008 00:01:11 +0000 (00:01 +0000)]
Make some adjustments to the buffer cache:

* Retain B_ERROR instead of clearing it.

* Change B_ERROR's behavior.  It no longer causes the buffer to be
  invalidated on write.

* Change B_NOCACHE's behavior.  It no longer causes the buffer to be
  invalidated while the buffer is marked dirty.

* Code that was supposed to re-dirty a failed write buffer in brelse()
  was not running because biodone() cleared the fields brelse() was
  testing.  Move the code to biodone().

* When attempting to reflush B_DELWRI|B_ERROR'd buffers, sleep a tick
  to try to avoid a live-lock.

15 years agoGive krateprintf() an initial burst capability if count is set to
Matthew Dillon [Thu, 17 Jul 2008 23:56:23 +0000 (23:56 +0000)]
Give krateprintf() an initial burst capability if count is set to
a negative number like e.g. -5.

15 years agoCode documentation only: Describe B_NOCACHE
Matthew Dillon [Thu, 17 Jul 2008 23:55:24 +0000 (23:55 +0000)]
Code documentation only: Describe B_NOCACHE

15 years agoHAMMER can now be demand-loaded.
Thomas Nikolajsen [Thu, 17 Jul 2008 21:38:07 +0000 (21:38 +0000)]
HAMMER can now be demand-loaded.

15 years agoAdd missing space and newline to error message.
Thomas Nikolajsen [Thu, 17 Jul 2008 21:34:47 +0000 (21:34 +0000)]
Add missing space and newline to error message.

15 years agoStart describing snapshots.
Sascha Wildner [Thu, 17 Jul 2008 11:23:23 +0000 (11:23 +0000)]
Start describing snapshots.

15 years agoFix MODULES_WITH_WORLD build.
Sascha Wildner [Thu, 17 Jul 2008 10:11:58 +0000 (10:11 +0000)]
Fix MODULES_WITH_WORLD build.

15 years agoMinor corrections.
Sascha Wildner [Thu, 17 Jul 2008 09:41:19 +0000 (09:41 +0000)]
Minor corrections.

15 years agoHEAD is now 2.1, so reflect that change for the man pages as well.
Matthias Schmidt [Thu, 17 Jul 2008 09:24:00 +0000 (09:24 +0000)]
HEAD is now 2.1, so reflect that change for the man pages as well.

15 years agoIn DragonFly, gpt partitions look like slices in /dev, and we match the
Matthew Dillon [Thu, 17 Jul 2008 01:15:59 +0000 (01:15 +0000)]
In DragonFly, gpt partitions look like slices in /dev, and we match the
slice number to the GPT index to avoid confusion.

Fix the reporting of the slice in 'gpt add' (it was off by 1).

15 years agoMake HAMMER build and work as a module and extend hammer(5)'s SYNOPSIS
Sascha Wildner [Thu, 17 Jul 2008 00:16:14 +0000 (00:16 +0000)]
Make HAMMER build and work as a module and extend hammer(5)'s SYNOPSIS
accordingly.

15 years agoAdd a calendar.dragonfly file to remind us that on July 16, 2003, DragonFly
Sascha Wildner [Wed, 16 Jul 2008 23:09:04 +0000 (23:09 +0000)]
Add a calendar.dragonfly file to remind us that on July 16, 2003, DragonFly
was first announced, so the project just got 5 years old.

Noticed-by: Justin Sherrill <justin@shiningsilence.com> in his Digest
15 years agoAdd logic to warn of possible renames, and clearly state when failures may
Matthew Dillon [Wed, 16 Jul 2008 21:38:10 +0000 (21:38 +0000)]
Add logic to warn of possible renames, and clearly state when failures may
be due to prior renames.

Change -h to -a (-a for 'all'), which is a more obvious option name.

15 years agoAdd some examples and move some stuff around.
Sascha Wildner [Wed, 16 Jul 2008 20:39:28 +0000 (20:39 +0000)]
Add some examples and move some stuff around.

15 years agoAdd a pointer to hammer(5) and fix some minor issues.
Sascha Wildner [Wed, 16 Jul 2008 20:00:41 +0000 (20:00 +0000)]
Add a pointer to hammer(5) and fix some minor issues.

15 years agoHAMMER 62/Many: Stabilization, performance, and cleanup
Matthew Dillon [Wed, 16 Jul 2008 18:30:59 +0000 (18:30 +0000)]
HAMMER 62/Many: Stabilization, performance, and cleanup

* Fix 'hammer synctid'.  It was blocking until the next filesystem sync in
  some cases.  Also redo the flush sequencing for synctid so the inodes
  are flushed in large groups instead of individually, improving performance.

* Fix a case where reclaimed inodes were sometimes not being reclaimed on
  the backend in a timely fashion.  There are still some performance issues
  here.

* Fix a race in the buffer invalidation code that could cause an assertion.

* Remove a short-cut in hammer_checkspace() that was not taking into
  account reserved big-blocks and allowing operations to exhaust all
  free space on smaller medias and then assert, instead of returning ENOSPC.

* Clean up the flush_group append code by tracking the append point.

* Clean up documentation in the inode flush path.

* Allow the reblocker to dig deeper into available free space when run
  with a low fill level (<= 20%).

15 years agoFix a lock leak in nfs_create(), tracked down from a crash dump and
Matthew Dillon [Wed, 16 Jul 2008 18:20:40 +0000 (18:20 +0000)]
Fix a lock leak in nfs_create(), tracked down from a crash dump and
possibly responsible for a pkgbox crash while rsync()ing over a
NFS mount.

15 years agoAdd a incomplete hammer(5) manual page.
Sascha Wildner [Wed, 16 Jul 2008 12:45:13 +0000 (12:45 +0000)]
Add a incomplete hammer(5) manual page.

It's intention is to give an introduction to the hammer file system
features along with examples, a la vkernel(7).

15 years agoCapitalize TAP and add some markup.
Thomas Nikolajsen [Wed, 16 Jul 2008 01:30:47 +0000 (01:30 +0000)]
Capitalize TAP and add some markup.

15 years agoUpdate `-t' flag; now it is always transaction-id.
Thomas Nikolajsen [Wed, 16 Jul 2008 01:27:09 +0000 (01:27 +0000)]
Update `-t' flag; now it is always transaction-id.

15 years agoAdd VKERNEL to FILES and make some paths a bit shorter.
Thomas Nikolajsen [Wed, 16 Jul 2008 01:25:23 +0000 (01:25 +0000)]
Add VKERNEL to FILES and make some paths a bit shorter.

15 years agoAdd VKERNEL.
Thomas Nikolajsen [Wed, 16 Jul 2008 01:02:07 +0000 (01:02 +0000)]
Add VKERNEL.

15 years agoUpdate newfs_hammer doc:
Thomas Nikolajsen [Wed, 16 Jul 2008 00:58:22 +0000 (00:58 +0000)]
Update newfs_hammer doc:
 - sync usage() & newfs_hammer.8
 - add a cross reference

15 years agoUpdate hammer doc:
Thomas Nikolajsen [Wed, 16 Jul 2008 00:53:48 +0000 (00:53 +0000)]
Update hammer doc:
 - sync usage() & hammer.8
 - reblock* percentage is 100%, correct a few places still saying 90%
 - drop `-s' as it isn't used anymore
 - drop doc for blockmap, es code is #if 0'ed
 - add more description of pruning
 - add more markup
 - capitalize UUID
 - spell out TID
 - add a few more cross references

15 years agoUpdate mount_hammer doc:
Thomas Nikolajsen [Tue, 15 Jul 2008 23:42:21 +0000 (23:42 +0000)]
Update mount_hammer doc:
 - sync usage() & mount_hammer.8
 - add missing \n to usage()
 - rename transactionid to transaction-id to ease reading
 - describe that special files can be `:' separated
 - add some more cross references

15 years agoHAMMER 61H/Many: Stabilization
Matthew Dillon [Tue, 15 Jul 2008 22:13:16 +0000 (22:13 +0000)]
HAMMER 61H/Many: Stabilization

* Fix a nasty uninitialized variable in hammer_btree_iterate_reverse().

Submitted-by: Johannes Hofmann <hofmann@blob.baaderstrasse.com>
15 years agoHAMMER 61G/Many: Stabilization
Matthew Dillon [Tue, 15 Jul 2008 18:01:58 +0000 (18:01 +0000)]
HAMMER 61G/Many: Stabilization

* Fix a nasty uninitialized variable in hammer_btree_iterate().
  gcc-4 missed the complex case and didn't complain about it.

Submitted-by: Johannes Hofmann <hofmann@blob.baaderstrasse.com>
15 years agoundo(1) doc fixes:
Thomas Nikolajsen [Tue, 15 Jul 2008 17:04:35 +0000 (17:04 +0000)]
undo(1) doc fixes:
 - sync usage() to undo.1
 - improve wording a bit in a few places

15 years agoFix an issue where libthread_xu was not accepting the full priority
Matthew Dillon [Tue, 15 Jul 2008 01:18:53 +0000 (01:18 +0000)]
Fix an issue where libthread_xu was not accepting the full priority
range returned by sched_get_priority_{min,max}(), causing firefox
to assert.

Reported-by: Jordan Gordeev <jgordeev@dir.bg>
Tested-by: walt <wa1ter@myrealbox.com>
15 years agoRemove useless types(5) manpage.
Sascha Wildner [Mon, 14 Jul 2008 23:59:34 +0000 (23:59 +0000)]
Remove useless types(5) manpage.

15 years agoExclude acd* and fd* from the list of disks.
Sascha Wildner [Mon, 14 Jul 2008 23:57:10 +0000 (23:57 +0000)]
Exclude acd* and fd* from the list of disks.

15 years agoFix a bug where mount_nfs would properly parse an IP address, but would
Matthew Dillon [Mon, 14 Jul 2008 22:22:40 +0000 (22:22 +0000)]
Fix a bug where mount_nfs would properly parse an IP address, but would
still fail if name resolution was down because it also tried to do a
reverse lookup on it.

Reported-by: Vincent Stemen <vince.dragonfly@hightek.org>
15 years agoFix a bug in vmntvnodescan() revealed by the recent NFS sync fix. The
Matthew Dillon [Mon, 14 Jul 2008 22:16:35 +0000 (22:16 +0000)]
Fix a bug in vmntvnodescan() revealed by the recent NFS sync fix.  The
function can improperly return -1 if the last element scanned is skipped
by the fastfunc.

15 years agoHAMMER Utilities: Sync with 61E
Matthew Dillon [Mon, 14 Jul 2008 20:28:07 +0000 (20:28 +0000)]
HAMMER Utilities: Sync with 61E

* Implement hammer iostats

15 years agoHAMMER 61E/Many: Features
Matthew Dillon [Mon, 14 Jul 2008 20:27:54 +0000 (20:27 +0000)]
HAMMER 61E/Many: Features

* Implement hammer iostats.

15 years agoNFS performance fixes.
Matthew Dillon [Mon, 14 Jul 2008 17:45:49 +0000 (17:45 +0000)]
NFS performance fixes.

* sync on an NFS mount was a big NOP due to a silly bug.

* utimes (setattr w/ mtime-changed) was unconditionally flushing the file,
  causing programs such as cpdup, rsync, rdist, and tar xp to sync on each
  file.

  change it so it does not unconditionally flush the file.

15 years ago2.0 Release Engineering:
Matthew Dillon [Mon, 14 Jul 2008 04:01:45 +0000 (04:01 +0000)]
2.0 Release Engineering:

* Post-branch adjustments to HEAD (2.1)

15 years ago2.0 Release Engineering:
Matthew Dillon [Mon, 14 Jul 2008 03:50:00 +0000 (03:50 +0000)]
2.0 Release Engineering:

* cvsup file, slip tag target

15 years ago2.0 Release Engineering:
Matthew Dillon [Mon, 14 Jul 2008 03:46:52 +0000 (03:46 +0000)]
2.0 Release Engineering:

* Bump DragonFly_version
* Bump HEAD sub-vers

15 years agoHAMMER Utilities: Features
Matthew Dillon [Mon, 14 Jul 2008 03:21:34 +0000 (03:21 +0000)]
HAMMER Utilities: Features

* hammer show: report the CRC in verbose mode.

15 years agoHAMMER 61F/Many: Stabilization w/ simultanious pruning and reblocking
Matthew Dillon [Mon, 14 Jul 2008 03:20:49 +0000 (03:20 +0000)]
HAMMER 61F/Many: Stabilization w/ simultanious pruning and reblocking

* BUG FIX: When doing direct-read check to see if any device buffers
  are aliasing the disk block and flush any we find which are dirty.
  This ensures that reblocked data gets to disk before a direct-read
  tries to read them FROM the disk.

* BUG FIX: Fix a bug introduced in a recent commit where the flusher
  wlll not always completely flush the UNDO FIFO or completely flush
  all meta-data, resulting in a rollback after a normal umount/mount.

* BUG FIX: Direct-writes queue I/O independant of the in-memory record.
  When the backend flusher flushes the record, making it available in the
  B-Tree, make sure that the indepent I/O has completed.  Otherwise
  a later reblocking operation might read the media before the direct-write
  has actually completed.

* BUG FIX: In-memory records are not subject direct-IO, since their data
  is not yet on the media.

* BUG FIX: Do not allow mount to succeed unless all volumes have been found.
  (Reported-by: Sascha Wildner <saw@online.de>)

* BUG FIX: The bd_heatup() call in the reblocker was in the wrong place,
  potentially causing the cursor to shift unexpectedly.

* Reorient some of the buffer invalidation code by enhancing
  the reservation code.

* Add read CRC verification logic for some direct-reads, but comment it out
  because the VM system's bogus-page replacement breaks it.

15 years agoKernel support for HAMMER:
Matthew Dillon [Mon, 14 Jul 2008 03:09:00 +0000 (03:09 +0000)]
Kernel support for HAMMER:

* Add another type to the bio->bio_caller_info1 union

* Add two new flags to getblk(), used by the cluster code.

  GETBLK_SZMATCH - Tell getblk() to fail and return NULL if a
  pre-existing buffer's size does not match
  the requested size (this prevents getblk()
  from doing a potentially undesired bwrite()
  sequence).

  GETBLK_NOWAIT - Tell getblk() to use a non-blocking lock.

* pop_bio() now returns the previous BIO (or NULL if there is no previous
  BIO).  This allows HAMMER to chain bio_done()'s

* Fix a bug in cluster_read().  The cluster code's read-ahead at the
  end could go past the caller-specified limit and force a block to
  the wrong block size.

15 years agoFix an asf core dump.
Matthew Dillon [Mon, 14 Jul 2008 00:09:42 +0000 (00:09 +0000)]
Fix an asf core dump.

Submitted-by: ejc <eric.j.christeson@gmail.com>
15 years agoSync the 'hammer show' description with reality.
Sascha Wildner [Sun, 13 Jul 2008 22:18:18 +0000 (22:18 +0000)]
Sync the 'hammer show' description with reality.

15 years agoUniformly refer to 'B-tree'.
Sascha Wildner [Sun, 13 Jul 2008 22:14:16 +0000 (22:14 +0000)]
Uniformly refer to 'B-tree'.

15 years agoUnload the bus dmamap, which was successfully loaded
Sepherosa Ziehau [Sun, 13 Jul 2008 11:02:50 +0000 (11:02 +0000)]
Unload the bus dmamap, which was successfully loaded

15 years agoWe're building live CDs using the installer in base for some time now,
Sascha Wildner [Sun, 13 Jul 2008 10:55:05 +0000 (10:55 +0000)]
We're building live CDs using the installer in base for some time now,
so they don't need to identify anymore as being built with it.

15 years agoFix deadlock occuring when booting vkernel from gdb. One lwp in the vkernel
Nicolas Thery [Sun, 13 Jul 2008 10:28:51 +0000 (10:28 +0000)]
Fix deadlock occuring when booting vkernel from gdb.  One lwp in the vkernel
sends a signal to another one and is preempted by the signal receiving one in
userexit().  As the vkernel is traced, the vkernel process is stopped and the
receiving lwp stops too.  The emitting lwp then resumes and blocks later on a
condition variable.  The gdb process sleeps on wait(2) which never completes
because the emitting lwp is not stopped.

To prevent this, stop lwps in userexit(), after lwkt_reschedule, if their
owning process is stopped.

Reported-by: Simon Schubert <corecode@fs.ei.tum.de>
15 years agoHAMMER 61E/Many: Stabilization, Performance
Matthew Dillon [Sun, 13 Jul 2008 09:32:48 +0000 (09:32 +0000)]
HAMMER 61E/Many: Stabilization, Performance

* PERFORMANCE: hammer_sync_inode() was generating a new transaction id
  for each inode, causing hammer_btree_do_propagation() to have to
  modify B-Tree nodes all the way to root on a per-file basis when
  syncing a rm -rf.

  Change the code to use the flusher's transaction id so all inodes
  bundled into the same flush group use the same transaction id.

* BUG FIX:  The reblocker was able to blow out the buffer cache with
  dirty data buffers.  Even though HAMMER allows these buffers to be
  flushed to the disk at any time by the kernel, calls to bwillwrite()
  are still needed to prevent a buffer cache deadlock.

15 years agoIncreate receive max coalesced bds to 80 to prevent system from being
Sepherosa Ziehau [Sun, 13 Jul 2008 03:57:41 +0000 (03:57 +0000)]
Increate receive max coalesced bds to 80 to prevent system from being
livelocked when 64bytes packets are injected @1.48Mpps.

It's RX side interrupt moderation logic is quite similar to et(4)'s:
(# recved packets > recv max coalesced bds) || (ticks > recv coalescing ticks)

# The 5750A1 I have tested could only sink packets @1.26Mpps

15 years agoUse a 64 bit quantity to collect file size data instead of
Matthew Dillon [Sun, 13 Jul 2008 03:37:43 +0000 (03:37 +0000)]
Use a 64 bit quantity to collect file size data instead of
32 bits, fixing overflows in the printed output.

Requested-by: "Trevor Kendall" <trevorjkendall@gmail.com>
15 years agoperror() will print its own colon.
Sascha Wildner [Sun, 13 Jul 2008 02:02:40 +0000 (02:02 +0000)]
perror() will print its own colon.

15 years agoHAMMER 61G/Many: Stabilization of new flush_group code
Matthew Dillon [Sun, 13 Jul 2008 01:12:41 +0000 (01:12 +0000)]
HAMMER 61G/Many: Stabilization of new flush_group code

* The mirroring, reblocking, and pruning code are now more robust in
  checking that too many meta-data buffers have not built up.

* Fix a bug in the flusher (today's flusher work) where pure meta-data
  changes to the media were not being flushed unless there was also a
  modified inode or two to flush.

15 years agoTypo fixing & minor cleanup.
Sascha Wildner [Sun, 13 Jul 2008 00:56:06 +0000 (00:56 +0000)]
Typo fixing & minor cleanup.

15 years agoHAMMER 61F2/Many: Fix bug in last commit
Matthew Dillon [Sat, 12 Jul 2008 23:55:22 +0000 (23:55 +0000)]
HAMMER 61F2/Many: Fix bug in last commit

* ip -> dip (pun intended)

15 years agoHAMMER 61F/Many: Stabilization
Matthew Dillon [Sat, 12 Jul 2008 23:47:13 +0000 (23:47 +0000)]
HAMMER 61F/Many: Stabilization

* Fix a bug triggered by rm -rf where HAMMER was failling to wait for
  reclaimed inodes to drain, allowing a build-up that would result in
  a kmalloc() pool-full panic.

  Add a required call to hammer_inode_waitreclaims() in the dounlink
  code.  Because this code acquires and releases inodes without having
  to mess around with related vnodes, it bypassed the reclaim checks in
  the vnode path.

Reported-by: Michael Neumann <mneumann@ntecs.de>
15 years agoHAMMER Utilities: Cleanup
Matthew Dillon [Sat, 12 Jul 2008 23:05:30 +0000 (23:05 +0000)]
HAMMER Utilities: Cleanup

* Cleanup the printfs for mirror-read, mirror-write, and mirror-copy.

15 years agoHAMMER 61E/Many: Stabilization, Performance
Matthew Dillon [Sat, 12 Jul 2008 23:04:50 +0000 (23:04 +0000)]
HAMMER 61E/Many: Stabilization, Performance

* The UNDO FIFO could get over-full, causing an assertion.  This was because
  certain load tests could cuase directories to end up with tens of
  thousands of directory entries and all of them were being flushed in
  a single flush group.

  Rewrite the flush group infrastructure to fix the issue.  Instead of
  having a two-stage flush we now have flush compartmentalization with
  the new hammer_flush_group structure.  Flushes can thus be broken down
  into transactions that don't blow out the UNDO FIFO.

* Improve flush performance by quite a bit.  The new infrastructure
  generates fewer B-Tree conflicts and appears to flush the work more
  smoothly, probably owing to the fact that the upper size of each
  flush group is now bounded.

15 years agoAdd quirk for the Samsung YP-U3.
Peter Avalos [Sat, 12 Jul 2008 20:36:06 +0000 (20:36 +0000)]
Add quirk for the Samsung YP-U3.

Obtained-from: FreeBSD

15 years agoFix an venerable bug: if we're reducing a rule that has an empty
Michael Neumann [Sat, 12 Jul 2008 16:52:48 +0000 (16:52 +0000)]
Fix an venerable bug: if we're reducing a rule that has an empty
right hand side and the yacc stackpointer is pointing at the very
end of the allocated stack, we end up accessing the stack out of
bounds by the implicit $$ = $1 action.

Obtained-from: OpenBSD (skeleton.c rev 1.29)

15 years ago* Add support for the nVidia MCP61, MCP65, MCP67, MCP73 and MCP77
Michael Neumann [Sat, 12 Jul 2008 16:38:10 +0000 (16:38 +0000)]
* Add support for the nVidia MCP61, MCP65, MCP67, MCP73 and MCP77
  series chipsets.

  Obtained-from: FreeBSD
  (ata-chipset.c rev 1.198 and parts of 1.204)
  (ata-pci.h rev 1.77 and 1.82)

* Fix clearing of nVidia interrupts.

  Obtained-from: FreeBSD
  (ata-chipset.c rev 1.214)

* Add support for MCP67 (revision 0xa2) not found in FreeBSD.

15 years agoUpdate build structure for libarchive-2.5.5.
Peter Avalos [Sat, 12 Jul 2008 14:57:33 +0000 (14:57 +0000)]
Update build structure for libarchive-2.5.5.

Start building bsdcpio.  It is not the default, but it can be installed by
setting WITH_BSDCPIO.  bsdcpio will become the default in 2.1, and GNU cpio
will be removed in 2.3.

15 years agoMerge from vendor branch LIBARCHIVE:
Peter Avalos [Sat, 12 Jul 2008 14:42:23 +0000 (14:42 +0000)]
Merge from vendor branch LIBARCHIVE:
Import libarchive-2.5.5.

15 years agoImport libarchive-2.5.5.
Peter Avalos [Sat, 12 Jul 2008 14:42:23 +0000 (14:42 +0000)]
Import libarchive-2.5.5.

15 years ago- Check number of free TX descs before trying to xmit packets
Sepherosa Ziehau [Sat, 12 Jul 2008 11:44:17 +0000 (11:44 +0000)]
- Check number of free TX descs before trying to xmit packets
- Recollect TX descs in nfe_start() if nfe(4) is going to enter OACTIVE
- Use dynamic segment limit for bus_dmamap_load() in nfe_encap()

This change gives me following result on AMD64X2 3600+ with MCP51:

10 seconds, TX 64bytes packets, before/after

256 TX desc
 imtimer | packets     | if_snd full    | if_oerrors
  -125     1.08M/1.08M   84532/82786      90/0
  -250     990K/1.08M    3735886/157125   38872/0

4096 TX desc
 imtimer | packets     | if_snd full    | if_oerrors
  -125     1.08M/1.09M   0/0              45/0
  -250     1.08M/1.10M   0/0              39/0

It gives great improvement when number of TX desc is 256 (default),
and it at least will not loose any packets if number TX desc is 4096.

15 years agoAdd brief description about tx_ring_count
Sepherosa Ziehau [Sat, 12 Jul 2008 09:31:08 +0000 (09:31 +0000)]
Add brief description about tx_ring_count

15 years agoAdd tunable for number of TX desc
Sepherosa Ziehau [Sat, 12 Jul 2008 09:27:49 +0000 (09:27 +0000)]
Add tunable for number of TX desc

15 years agoPrepare to rework nfe_start():
Sepherosa Ziehau [Sat, 12 Jul 2008 06:50:04 +0000 (06:50 +0000)]
Prepare to rework nfe_start():
Allow number of segments to exceeds the limit set by the caller; caller
detects and handles this error accordingly.

15 years ago- Decrease NFE_MAX_SCATTER to 32, should be enough
Sepherosa Ziehau [Sat, 12 Jul 2008 06:16:22 +0000 (06:16 +0000)]
- Decrease NFE_MAX_SCATTER to 32, should be enough
- Increase ifnet.if_snd length to (NFE_TX_RING_COUNT - 1)

15 years agoUse softc sc_rx_ring_count to allocate jumbo frame instead of nfe_rx_ring_count
Sepherosa Ziehau [Sat, 12 Jul 2008 06:01:37 +0000 (06:01 +0000)]
Use softc sc_rx_ring_count to allocate jumbo frame instead of nfe_rx_ring_count

15 years ago- Correct max segment size passed to various bus_dma_tag_create()
Sepherosa Ziehau [Sat, 12 Jul 2008 05:48:32 +0000 (05:48 +0000)]
- Correct max segment size passed to various bus_dma_tag_create()
- Pass softc's sysctl context to SYSCTL_ADD_INT
- Allow jumbo frame allocation to fail during attach.  If jumbo frame
  allocation failed, then MTU was not allowed to set above ETHERMTU
- Don't reinitialize device, if MTU is changed but the device is not
  running yet

15 years agoHAMMER Utillities: Sync with 61D
Matthew Dillon [Sat, 12 Jul 2008 02:48:46 +0000 (02:48 +0000)]
HAMMER Utillities: Sync with 61D

* Add the new pfs-upgrade, pfs-downgrade, and pfs-destroy directives.

* Add more safeties to the PFS directives.

15 years agoHAMMER 61D/Many: Mirroring features
Matthew Dillon [Sat, 12 Jul 2008 02:47:39 +0000 (02:47 +0000)]
HAMMER 61D/Many: Mirroring features

* Split PFS ioctls into their own source file.

* Add additional PFS/mirroring directives:  pfs-upgrade, pfs-downgrade, and
  finish implementing pfs-destroy.  (Yes, that means you can change the
  master/slave mode for a PFS now).

* Consolidate some of the B-Tree deletion code.

* Fix another sync_lock deadlock.

15 years agoCorrect a bug in the last commit.
Matthew Dillon [Sat, 12 Jul 2008 02:44:59 +0000 (02:44 +0000)]
Correct a bug in the last commit.

15 years agoAdd a vclean_unlocked() call that allows HAMMER to try to get rid of a
Matthew Dillon [Sat, 12 Jul 2008 01:09:46 +0000 (01:09 +0000)]
Add a vclean_unlocked() call that allows HAMMER to try to get rid of a
vnode.

15 years agoTunables are called loader tunables; fix this in a few places.
Thomas Nikolajsen [Fri, 11 Jul 2008 23:42:46 +0000 (23:42 +0000)]
Tunables are called loader tunables; fix this in a few places.

15 years agoUpdate netstat help:
Thomas Nikolajsen [Fri, 11 Jul 2008 23:09:18 +0000 (23:09 +0000)]
Update netstat help:
 - sync usage() with netstat.1
 - describe that -W shows MPLS label operations
 - add description of protocol families and carp(4)
 - add cross references to inet6(4) and carp(4)

15 years agoAdd patch from 9.3.5 to 9.3.5-P1. This is for adding randomization
Jeremy C. Reed [Fri, 11 Jul 2008 22:42:30 +0000 (22:42 +0000)]
Add patch from 9.3.5 to 9.3.5-P1. This is for adding randomization
for every query.

The patch for the security issues from 9.3.5 to 9.3.5-P1 (minus
win32 and doc) applies fine to HEAD contrib/bind-9.3 (which is
9.3.4-P1) other than a few RCS ident conflicts.

I made up my own version name for this for DragonFly: 9.3.4-P2.

This is for
http://www.kb.cert.org/vuls/id/800113
http://cve.mitre.org/cgi-bin/cvename.cgi?name=CVE-2008-1447

Note that this -P1 patch is the unoptimized version.
Also on busy recursive caching systems you may need to set limits
to allow many sockets.

This also removes comment from DragonFly's named.conf as
that is now a "warning" and is not recommended. Remove any
query-source restriction for one port or this patch will not be used.

15 years agoUpdate LINT:
Thomas Nikolajsen [Fri, 11 Jul 2008 18:42:14 +0000 (18:42 +0000)]
Update LINT:
 - add reference to kernconf(5)
 - copy description of SMP from GENERIC,
espcially that APIC_IO can be laft out
 - truncate description of polling(4) & dummynet(4),
ensure reference to polling(4) & dummynet(4).
This also removed outdated description of dependency on HZ.
Corret information is in manual pages; it is best to have it in one place.
 - point out that SOFTUPDATES applies to UFS only.
 - add HAMMER to list of filesystems which cant be build as module.
 - point outthat ATA_STATIC_ID is for NATA also, not just ATA.

15 years agoAdd prototype for crc32_ext().
Sascha Wildner [Fri, 11 Jul 2008 11:59:44 +0000 (11:59 +0000)]
Add prototype for crc32_ext().

15 years agoMake support for polling(4) more visible,
Thomas Nikolajsen [Fri, 11 Jul 2008 10:38:50 +0000 (10:38 +0000)]
Make support for polling(4) more visible,
by including a sentence; not only a cross reference to polling(4).
bce.4 and bge.4 was missing any refence to polling(4).
Also add sentence with general reference to ifconfig(8) in manual pages missing it.

15 years agoChange how `options INCLUDE_CONFIG_FILE' is fixed:
Thomas Nikolajsen [Fri, 11 Jul 2008 09:28:20 +0000 (09:28 +0000)]
Change how `options INCLUDE_CONFIG_FILE' is fixed:
use gcc'ism, __used attribute,
instead of relying on compiler not doing global optimization.
This also doesn't introduce name space pollution.
Also add comment.

Suggested-by: corecode
15 years agoHAMMER 61C/Many: Stabilization
Matthew Dillon [Fri, 11 Jul 2008 05:44:23 +0000 (05:44 +0000)]
HAMMER 61C/Many: Stabilization

* Clean up flusher sequencing for the mirroring, pruning, and reblocking
  code.  Check free space and don't let the buffer cache implode.

15 years agoHAMMER 61B/Many: Stabilization
Matthew Dillon [Fri, 11 Jul 2008 01:22:29 +0000 (01:22 +0000)]
HAMMER 61B/Many: Stabilization

* Giving the sync_lock's exclusive locks priority over shared locks caused
  a 3-way deadlock, revert to using normal shared locks.

* Move the syncer_lock deeper, closer to the code that actually needs to
  hold but still paying attention to atomicy requirements.  This reduces
  lock contention and improves performance with parallel loads.

Reported-by: Michael Neumann (using his file-creator program)
15 years agoAdd remains from the libreadline removal.
Sascha Wildner [Thu, 10 Jul 2008 22:52:18 +0000 (22:52 +0000)]
Add remains from the libreadline removal.

15 years agoNuke libreadline, it's no longer being built.
Sascha Wildner [Thu, 10 Jul 2008 22:45:27 +0000 (22:45 +0000)]
Nuke libreadline, it's no longer being built.

15 years agoAdd some files that seem to have slipped during the conversion from
Sascha Wildner [Thu, 10 Jul 2008 22:24:18 +0000 (22:24 +0000)]
Add some files that seem to have slipped during the conversion from
etc/Makefile to Makefile_upgrade.inc.

15 years agoHAMMER 61A/Many: Stabilization
Matthew Dillon [Thu, 10 Jul 2008 21:23:58 +0000 (21:23 +0000)]
HAMMER 61A/Many: Stabilization

* Fix a buffer exhaustion issue.  When creating large numbers of empty files
  a single inode sync of the directory can exhaust the buffer cache because
  not enough other things get dirty enough to force a flush.

  Put a check in the record syncing code and do a partial finalization if
  necessary to avoid deadlocking the buffer cache.

* Fix a panic caused by a missing call to hammer_cursor_deleted_element().

Reported-by: Michael Neumann <mneumann@ntecs.de>,
     Gergo Szakal <bastyaelvtars@gmail.com>

15 years agoThe symlink generated during "hammer snapshot" should be "/filesystem/@@0x.."
Michael Neumann [Thu, 10 Jul 2008 18:47:22 +0000 (18:47 +0000)]
The symlink generated during "hammer snapshot" should be "/filesystem/@@0x.."
and not "/filesystem@@0x...".

15 years agostyle: compare return value of getopt() against -1 rather than EOF
Sascha Wildner [Thu, 10 Jul 2008 18:29:52 +0000 (18:29 +0000)]
style: compare return value of getopt() against -1 rather than EOF

15 years agoAdd crc32_ext() - allows continuation of a 32 bit crc.
Matthew Dillon [Thu, 10 Jul 2008 17:42:18 +0000 (17:42 +0000)]
Add crc32_ext() - allows continuation of a 32 bit crc.