- *ip is not initialized in the case of inet6 connection, but ip->ip_len is
being changed anyway
Now the question is, why does it think an ipv4 connection is an ipv6 connection?
xemacs still doesn't work over X11 forwarding, but the kernel no longer panics.
- SWAPLR quirk for (unknown, luckily it is mine) broken uaudio stick.
Fixing by rewiring is impossible without damaging it. Luckily,
we can fix it using "other" methods :) .
- Add uaudio_get_vendor(), _product() and _release() in uaudio.c
(currently used by uaudio_pcm quirk).
- Implement CHANNEL_SETFRAGMENTS().
- Drop channel locking in few places where it is about to sleep
somewhere. This should help eliminating illegal locking acquisition
where the current thread is about to sleep, and also few deadlock
cases. Dropping it right here is quite safe since it is already
protected by CHN_F_BUSY flag and other threads won't bother to touch it.
Solving other illegal locking issues are quite tricky without converting
most usbd_do_request() calls to its equivalent _async() calls,
which I intend to do it later after getting full test report from
other people with different uaudio hardwares.
- Fix memory leak issues during detach. This seems common to any drivers
(notably emu10kx, csapcm?) with bridge functions.
Implement CHANNEL_SETFRAGMENTS() for snd_atiixp, snd_es137x, snd_hda
and snd_via8233. CHANNEL_SETBLOCKSIZE() will basically call
CHANNEL_SETFRAGMENTS() internally using conservative blocksize /
blockcount hints. Other drivers will be converted later.
- Disable stray buffer management, since sample size aligned buffering
are pretty much guaranteed through out the entire feeder_* chain
processes.
- Few style(9) cleanups.
channel.c/channel_if.m:
- Macros cleanups, prefer inlined min() over MIN().
- Rework chn_read()/chn_write() for better dead interrupt detection
policy. Reduce scheduling overhead by doing pure 5 seconds sleep
before giving up, instead of several cycle of brute micro sleeping.
- Avoid calling wakeup_one() for non-sleeping channel (for example,
vchan parent channel).
- EWOULDBLOCK -> EAGAIN.
- Fix possible divide-by-zero panic on chn_sync().
- Re-enforce ^2 blocksize policy, since there are too many broken
userland apps that blindly assume it without even trying to do
serious calculations.
- New channel method - CHANNEL_SETFRAGMENTS(), a refined version of
CHANNEL_SETBLOCKSIZE(). It accept _both_ blocksize and blockcount
arguments, so the driver internals will have better hints for
buffering and timing calculations.
- Hook FEEDER_SWAPLR into feederchain building process.
feeder_fmt.c:
- Unified version of various filters, avoiding duplications.
- malloc()less feeder_fmt. Informations can be retrieved dynamically
by doing table lookup on static data. For cases such as converting
from stereo to mono or reducing bit depth where input data is larger
than output, cycle remaining available free space until it has been
exhausted and start kicking 8 bytes reservoir space from there to
complete the remaining requested count.
- Introduce FEEDER_SWAPLR. Few super broken hardwares (found on several
extremely cheap uaudio stick, possibly others) mistakenly wired left
and right channels wrongly, screwing output or input.
- Rearrange FEEDER_* constants starting from 0 to 31, so the future
additions will be much easier and consistent.
- Introduce FEEDER_SWAPLR. Few super broken hardwares (found on several
extremely cheap uaudio stick, possibly others) mistakenly wired left
and right channels wrongly, screwing output or input.
malloc()less feeder_vchan. Informations can be retrieved dynamically
by doing table lookup on static data. Reduce mixing overhead by
doing direct copy on first channel. Mixing process will begin starting
from second channel onwards.
malloc()less feeder_volume. Informations can be retrieved dynamically
by doing table lookup on static data. Increase resolution from 6bit
to PCM_FXSHIFT (8bit) for better resolution and finer volume changes.
- Convert sx lock to plain mutex. Since the access of /dev/sndstat
is pretty much exclusive and protected by toggling sndstat_isopen,
plain mutex is more than enough.
- Enable SBUF_AUTOEXTEND to avoid buffer truncation.
# ls -ld /mnt/{foo,bar}
drwxr-xr-x 3 root wheel 512 Mar 16 06:56 /mnt/bar
lrwxr-xr-x 1 root wheel 3 Mar 16 12:10 /mnt/foo -> bar
# grep /mnt/foo /etc/fstab
/dev/da1 /mnt/foo ufs rw 0 0
Which means, we give symbolic link as a mount point to mount(8), but mount(8)
use realpath(3) before mounting the file systems, so we get:
# mount | grep /dev/da1
/dev/da1 on /mnt/bar (ufs, local)
Before the commit:
# snapinfo /mnt/foo
usage: snapinfo [-v] -a
snapinfo [-v] mountpoint
# snapinfo /mnt/bar
/mnt/bar/snap
This commit makes snapinfo(8) to first realpath(3) the given mount point and
now we have:
# snapinfo /mnt/foo
/mnt/bar/snap
# snapinfo /mnt/bar
/mnt/bar/snap
point path. This way we properly handle the case when file system listed
in /etc/fstab was unmounted and another file system was mounted on the
same mount point.
prefix) as an argument and mount point path. At the end it has to find
device name file system is stored on, which means when mount point path is
given, it tries to look into /etc/fstab and find special device
corresponding to the given mount point. This is not perfect, because it
doesn't handle the case when file system is mounted by hand and mount point
is given as an argument.
I found this problem while trying to use snapinfo(8), which passes mount
points to the ufs_disk_fillout(3) function, but I had file system mounted
manually, so snapinfo(8) was exiting with the error below:
ufs_disk_fillout: No such file or directory
I modified libufs(3) to handle those arguments (the order is important):
1. special device with /dev/ prefix
2. special device without /dev/ prefix
3. mount point listed in /etc/fstab, directory exists
4. mount point listed in /etc/fstab, directory doesn't exist
5. mount point of a file system mounted by hand
- We need at least two OCBs with indirect pointers allocated in a 4KB page.
- SBP_MAXPHYS can increase to 1MB once we separate management OCB/ORB
which usually does not need indirect pointers.
- We have to increase SBP_DMA_SIZE for MAXPHYS larger than 1MB.
MFC after: 3 days
EVP_CIPHER_CTX_key_length() should return the set key length
in the EVP_CIPHER_CTX structure which may not be the same as
the underlying cipher key length for variable length ciphers.
This fixes problems in OpenSSH using some ciphers, and possibly other
applications.
See also: http://bugzilla.mindrot.org/show_bug.cgi?id=1291
cache coherency, besides of causing train wreck in other places
(especially on amd64, possibly on i386).
Discussed with: kib@, rafan@
Tested by: rafan@
confusions and panic provided that the following conditions are met:
1) WITNESS is enabled (watch/trace).
2) Using modules, instead of statically linked (Not a strict
requirement, but easier to reproduce this way).
3) 2 or more modules share the same mtx type ("sound softc").
- They might share the same name (strcmp() == 0), but it always
point to different address.
4) Repetitive kldunload/load on any module that shares the same mtx
type (Not a strict requirement, but easier to reproduce this way).
Consider module A and module B:
- From enroll() - subr_witness.c:
* Load module A. Everything seems fine right now.
wA-w_refcount == 1 ; wA-w_name = "sound softc"
* Load module B.
* w->w_name == description will always fail.
("sound softc" from A and B point to different address).
* wA->w_refcount > 0 && strcmp(description, wA->w_name) == 0
* enroll() will return wA instead of returning (possibly unique)
wB.
wA->w_refcount++ , == 2.
* Unload module A, mtx_destroy(), wA->w_name become invalid,
but wA->w_refcount-- become 1 instead of 0. wA will not be
removed from witness list.
* Some other places call mtx_init(), iterating witness list,
found wA, failed on wA->w_name == description
* wA->w_refcount > 0 && strcmp(description, wA->w_name)
* Panic on strcmp() since wA->w_name no longer point to valid
address.
Note that this could happened in other places as well, not just sound
(eg. consider lots of drivers that share simmilar MTX_NETWORK_LOCK).
Solutions (for sound case):
1) Provide unique mtx type string for each mutex creation (chosen)
or
2) Put "sound softc" global variable somewhere and use it.
and syncache_respond() into its own generic function tcp_addoptions().
tcp_addoptions() is alignment agnostic and does optimal packing in all cases.
In struct tcpopt rename to_requested_s_scale to just to_wscale.
Add a comment with quote from RFC1323: "The Window field in a SYN (i.e.,
a <SYN> or <SYN,ACK>) segment itself is never scaled."
Reviewed by: silby, mohans, julian
Sponsored by: TCP/IP Optimization Fundraise 2005
