Blah, fsck_ifs uses `8' for all errors, including that it cannot check
in the background. fsck_ffs uses `5' for when SOFTUPDATES are not set,
and thus background cleaning cannot take place. That seems to [semi-]apply
here. So I don't know what non-zero value to use.
If anyone has an opinion, let me know.
Recognize the "-F" option which requests whether the filesystem needs to
be cleaned immediately in foreground, or if its cleaning can be deferred
to background.
More README file cleanup, mostly for style. Also punt the readers
to the Handbook for more documentation sources. Mention INSTALL.TXT
as one of the other release documentation files.
Unconditinally turn on VLAN support when building fxp as a module, since
it may be plugged into a kernel that supports VLANs. If the kernel is
not VLAN aware, things will still work as before.
Modules don't really have option support, so this is somewhat of a hack.
Only turn on MWI if the PCI configuration word indicates that it
is supported, in addition to checking for a valid cacheline size.
Add a missing splx() in fxp_tick that got dropped.
Instead of giving an example of rotating logs at the end of the month,
give an example of how to rotate logs at the beginning of the month.
Although they sound the same, since both of them rotate logs at the
beginning of the day, the former ended up taking place on, e.g., July
31 00:00 instead of the expected July 31 23:59. This is contraty to POLA.
Fix an underflow resulting from the assumption !O_ISSET(sp, O_LEFTRIGHT)
(ironically, the assumption is in a code block which is conditional on its
converse). This isn't strictly the correct fix; it's more of a workaround
to prevent an infinite loop. The correct fix (see
ports/editors/nvi-devel/files/patch-vi-relative r1.1) would take a file off
the vendor branch, but since the result for this version of nvi is
identical, this route was elected.
swtch.s: During context save, use the correct bit mask for clearing
the non-reserved bits of dr7.
During context restore, load dr7 in such a way as to not
disturb reserved bits.
machdep.c: Don't explicitly disallow the setting of the reserved bits
in dr7 since we now keep from setting them when we load dr7
from the PCB.
This allows one to write back the dr7 value obtained from
the system without triggering an EINVAL (one of the
reserved bits always seems to be set after taking a trace
trap).
ahc_pci.c:
If bus_dma will give us addresses > 32 bits, setup our dma tag
to accept up to 39bit addresses.
aic7770.c:
Update the softc directly rather than use an intermediate
"probe_config" structure.
aic7xxx.c:
Complete core work to support 39bit addresses for bulk data
dma operations. Controller data structures still must reside
under the 4GB boundary to reduce code/data size in the sequencer
and related data structures. This has been tested under Linux
IA64 and will be tested on IA64 for FreeBSD as soon as our port
can run there.
Add bus dmamap synchronization calls around manipulation of
all controller/kernel shared host data structures.
Implement data pointer reinitialation for a second data phase
in a single connection in the kernel rather than bloat the
sequencer. This is an extremely rare operation (does it ever
happen?) and the sequencer implementation was flawed for some
of the newest chips.
Don't ever allow our target role to initiate a PPR. This
is forbidden by the SCSI spec.
Add a few missing endian conversions in the ignore wide pointers
code. The core has been tested on the PPC under Linux and should
work for FreeBSD PPC. As soon as I can test the OSM layer for
FreeBSD PPC, I will.
Move some of ahc_softc_init() into ahc_alloc() now that the
probe_config structure is gone.
Add a 4GB boundary condition on all of our dma tags. 32bit
DAC under PCI only works on a single 4GB "page". Although
we can cross 4GB on a true 64bit bus, the card won't always
be installed in one and we can save code space and cost in
implementing high address support by assuming the high DWORD
address will never change.
Add diagnostics to ahc_search_qinfifo().
Correct a target mode issue with bus resets. To avoid an
interrupt storm from a malicious third party holding the
reset line, the sequencer would defer re-enabling the reset
interrupt until either a select-out or select-in. Unfortunately,
the select-in enable bit is cleared by a bus reset, so a second
reset will render the card deaf to an initiator's attempts to
contact it. We now re-enable bus reset interrupts immediately
if the target role is enabled.
aic7xxx.h:
Remove struct ahc_probe_config.
SCB's now contain a pointer to the sg_map_node so we can perfrom
bus dma sync operations on the SG list prior to queuing a command.
aic7xxx.reg:
Register the Perforce ID for this file with the VERSION keyword
so it is printed in generated files.
Add the DSCOMMAND1 register which is used to access the high
DWORD of address bits.
Add the data pointer reinitialize sequencer interrupt code.
aic7xxx.seq:
Register the Perforce ID for this file with the VERSION keyword
so it is printed in generated files.
Remove code to re-enable the bus reset interrupt after a select-in.
In target mode we cannot defer this operation as ENSELI is cleared
by a bus reset.
Complete 39bit support.
Generate a sequencer inteerrupt rather than handle the data
pointers re-initialitation in the sequencer.
Inline the "seen identify" assertion to save a few cycles.
Short circuit the update of our residual data if we have
fully completed a transfer. The residual is correct from
our last S/G load operation.
Short circuit full SDPTR processing if the residual is 0.
Just mark the transfer as complete.
Clear our residuals prior to starting a command. The
update residual code in the core only sets the residual
if there is one.
aic7xxx_freebsd.h:
Modeify ahc_dmamap_sync() macros to take an offset and a length.
This is how sync operations are performed in NetBSD, and we should
update our bus dma implementation to match.
aic7xxx_inline.h:
Add data structure synchronization helper functions.
Fix a bug in ahc_intr() where we would not clear our unsolicited
interrupt counter after running our PCI interrupt handler. This
may have been the cause of the spurious PCI interrupt messages.
aic7xxx_pci.c:
Adjust for loss of probe_config structure.
Guard against bogus 9005 subdevice information as seen on some
IBM MB configurations.
free_entry(): Don't free e->envp if it's already NULL; likewise for
e->cmd. free_entry() now does the right thing with
partially-initialized structures.
load_entry(): Don't call env_free() on e->envp throughout the routine
before jumping to eof; the free_entry() call at that label will take
care of it. The previous behavior resulted in e->envp being free'd
twice (well, the second time would usually result in a crash, but
that's besides the point); once in load_entry(), and once in
free_entry() after the former called the latter. Also note that the
check added to free_entry() (above) doesn't help, since e->envp wasn't
reset to NULL after env_free().
Set wollman as the MAINTAINER and mention the vendor contact. Ideally
this entire subtree would be in src/contrib, but if that isn't going
to happen at least this has a chance of warning off unsuspecting
committers.
Removed the comment that required all FreeBSD manpages
to have the $FreeBSD$ keyword, as this is now enforced
by the CVSROOT/commit_prep.pl script.
Fold multi-word macro arguments into a single argument
by putting the surrounding double quotes - this speeds
up the -mdoc processing drastically (of course if used
systematically).
Use the new features of -mdoc: exact -width specifiers,
.In macro as an ``.Fd #include'' replacement.
Make sure that installing a module complains if the target modules
directory does not exist, instead of creating/overwriting a file
with the name of the (expected) directory. Yes, this deviates a bit
from nearly all other install targets in the tree, but let's face it,
removing a modules directory is not all that uncommon a mistake,
and finding a file with the contents of the last module installed
is a baaad surprise at boot time..
PR: 26317
Submitted by: "T. William Wells" <bill@twwells.com> (the PR)
Gregory Bond <gnb@itga.com.au> (the actual patch)
Reviewed by: silence on -arch and -audit for the last 10 days
MFC after: 2 weeks
: 2001-07-18 Ruslan Ermilov <ru@FreeBSD.org>
:
: * tmac/groff_mdoc.man: Document new -width and -column syntax.
: Some other minor fixes.
: * tmac/an-old.tmac: Add `AT' and `UC' macros.
eric [Wed, 18 Jul 2001 05:21:37 +0000 (05:21 +0000)]
Improve the interface provided by libdialog. Move a cursor around over
the components and trigger actions based on its position. This reduces
the need to remember the functions of various keys, and makes the
interface more consistant across library.
~
(forced commit for more info on the previous change)
According to RFC 1112, which deals with multicasting, an application must
specify the interface on which it wished to send a multicast packet. In the
absence of such an interface, a default is selected.
Previous behavior in ip_output.c erroneously checked for a route to the
destination multicast address, rejecting the packet if none existed.
Applications got around this with a FAQ recommending a 224.0.0.0/29 route
to loopback. This worked because the multicast code in ip_output.c discards
route information if an interface was selected.
The previous commit skips the route check in the case where a multicast packet
is being send to an application-defined interface.
The only change in behavior, if no bugs were introduced, is the lack of
rejection of a multicast packet for which no route exists sent to an
application-defined interface.
mike [Tue, 17 Jul 2001 20:40:41 +0000 (20:40 +0000)]
Re-write the logic that finds the whois server to query.
[This fixes a bug where one would type 'whois foo.bar.'
and get an error because of the trailing period.]
The per-cpu temporary buffers are not needed since the pcb_save areas have
the proper alignment. Change dummy variable in npxinit from stack to bss
to ensure proper alignment.
Backout damage to the INADDR_TO_IFP() macro in revision 1.7.
This macro was supposed to only match local IP addresses of
interfaces, and all consumers of this macro assume this as
well. (See IP_MULTICAST_IF and IP_ADD_MEMBERSHIP socket
options in the ip(4) manpage.)
This fixes a major security breach in IPFW-based firewalls
where the `me' keyword would match the other end of a P2P
link.
Set the NOADDITIONAL flag for bind. This removes the 5% ttl reduction
for glue records and forces the glue record to be reloaded from the real NS.
The 5% ttl reduction can cause the glue IN A to timeout before the NS
record in certain situations, such as when the domain owner does not match
up NS records with the NIC. This behavior by domain owners is becoming
more common as primary zone serving iterates through another glue level
(i.e. exodus hosts the master NS's but the customer then redirects the
NS's to the real DNS servers). The result is that named would appear to
work properly for about 40 minutes, and then unexpectedly fail for that
zone. This causes named to behave very inconsistently and a google search
shows that it has obviously frustrated many, many people. So until the bind
guys make named behave consistently (either fail instantly or accomodate the
case), we need to set this option to accomodate the case. The result
will be much more consistent behavior and fewer head-scratching failures.