discussed on the arch@ mailinglist (after repo-copy).
sys.mk will .error if it finds /etc/defaults/make.conf but include
it anyways (this is the same behaviour as with the make.conf.local
removal).
/usr/share/examples/etc/make.conf has BDEFLAGS commented out now,
since it's only an example file.
Adjust all textes that talk about make.conf or defaults/make.conf to
match the new situation.
When file offset tends to be negative due to internal and ungetc buffers
additions counted, try to discard some ungetc data first, then return EBADF.
Later one can happens if lseek(fileno(fd),...) called f.e. POSIX says that
ungetc beyond beginning of the file results are undefined, so we can just
discard some of ungetc data in that case.
Don't rely on gcc cast when checking for overflow, use OFF_MAX.
Cosmetique.
Instead introduce the [M] prefix to existing keywords. e.g.
MSTD is the MP SAFE version of STD. This is prepatory for a
massive Giant lock pushdown. The old MPSAFE keyword made
syscalls.master too messy.
Begin comments MP-Safe procedures with the comment:
/*
* MPSAFE
*/
This comments means that the procedure may be called without
Giant held (The procedure itself may still need to obtain
Giant temporarily to do its thing).
sv_prepsyscall() is now MP SAFE and assumed to be MP SAFE
sv_transtrap() is now MP SAFE and assumed to be MP SAFE
ktrsyscall() and ktrsysret() are now MP SAFE (Giant Pushdown)
trapsignal() is now MP SAFE (Giant Pushdown)
Places which used to do the if (mtx_owned(&Giant)) mtx_unlock(&Giant)
test in syscall[2]() in */*/trap.c now do not. Instead they
explicitly unlock Giant if they previously obtained it, and then
assert that it is no longer held to catch broken system calls.
Rebuild syscall tables.
- build_x.sh now does all the steps to build an X dist rather than being
a child script of package_x_dists.sh
- Update the usage information to list the ports you need to install
before running this script as well as needing to set CVSROOT.
- Make sure CVSROOT is set. If not, exit with an error.
- We now take two parameters: a work directory and an output directory.
The work directory is used as scratch space. All of the bindist will
end up in the output directory.
- Only apply XF86.patch to the checked out XFree86 port if it exists.
- Use XFree86's build-bindist tool to package up the dists using their
distfiles rather than using our own packing lists that have to be
manually updated each time the port changes.
Example usage:
env CVSROOT=/home/ncvs ./build_x.sh /usr/xtmp /usr/x11dists
Clear residual counts after a successful samount (the user doesn't
care that we got an N-kbyte residual on our test read).
Change a lot of error handling code.
1. If we end up in saerror, check more carefully about the kind of
error. If it is a CAM_SCSI_STATUS_ERROR and it is a read/write
command, we'll be handling this in saerror. If it isn't a read/write
command, check to see whether this is just an EOM/EOP check condition-
if it is, just set residual and return normally. A residual and
then a NO SENSE check condiftion with the ASC of 0 and ASCQ of
between 1 and 4 are normal 'signifying' events, not errors per se,
and we shouldn't give the command to cam_periph_error to do something
relatively unpredictable with.
2. If we get a Bus Reset, had a BDR sent, or get the cam status of
CAM_REQUEUE_REQ, check the retry count on the command. The default
error handler, cam_periph_error, doesn't honor retry count in these
cases. This may change in the future, but for now, make sure we
set EIO and return without calling cam_periph_error if the retry
count for the command with an error is zero.
3. Clean up the pending error case goop and handle cases more
sensibly.
The rules are:
If command was a Write:
If we got a SSD_KEY_VOLUME_OVERFLOW, the resid is
propagated and we set ENOSPC as the error.
Else if we got an EOM condition- just mark EOM pending.
And set a residual of zero. For the longest time I was just
propagating residual from the sense data- but my tape
comparison tests were always failing because all drives I
tested with actually *do* write the data anyway- the EOM
(early warning) condition occurred *prior* to all of the
data going out to media- that is, it was still buffered by
the drive. This case is described in SCSI-2, 10.2.14,
paragraph #d for the meaning of 'information field'. A
better fix for this would be to issue a WFM command of zero
to cause the drive to flush any buffered data, but this
would require a fairly extensive rewrite.
Else if the command was a READ:
If we got a SSD_KEY_BLANK_CHECK-
If we have a One Filemark EOT model- mark EOM as pending,
otherwise set EIO as the erorr.
Else if we found a Filemark-
If we're in Fixed Block mode- mark EOF pending.
If we had an ILI (Incorrect Length Indicator)-
If the residual is less than zero, whine about tape record
being too big for user's buffer, otherwise if we were in
Fixed Block mode, mark EIO as pending.
All 'pending' conditions mean that the command in question completes
without error indication. It had succeeded, but a signifying event
occurred during its execution which will apply to the *next* command
that would be exexcuted. Except for the one EOM case above, we always
propagate residual.
Now, way back in sastart- if we notice any of the PENDING bits set,
we don't run the command we've just pulled off the wait queue. Instead,
we then figure out it's disposition based upon a previous command's
association with a signifying event.
If SA_FLAG_EOM_PENDING is set, we don't set an error. We just complete
the command with residual set to the request count (not data moved,
but no error). We continue on.
If SA_FLAG_EOF_PENDING- if we have this, it's only because we're in
Fixed Block mode- in which case we traverse all waiting buffers (which
we can get in fixed block mode because physio has split things up) and
mark them all as no error, but no data moved and complete them.
If SA_FLAG_EIO_PENDING, just mark the buffer with an EIO error
and complete it.
Then we clear all of the pending state bits- we're done.
MFC after: 4 weeks
added but not its postrequisite -ltermcap.
Fixed breakage of DPADD in previous commit. ${LIBREADLINE} was misspelled
-lreadline. This should have been fatal since there is no file named
-lreadline, but it worked because of an undcumented bugfeature in make(1)
(or its configuration files): missing source files named -l* are silently
assumed to be up to date libraries. `make checkdpadd' also fails to detect
this error.
At the times, restore(8) and rrestore(8) were the different
utilities. rrestore(8) was installed setuid `root', while
restore(8) with usual ownership and privileges. Later on,
on August 28, 1991 (what a coincidence!), rrestore(8) code
was merged with restore(8). The setgid `tty' bit then was
accidentally put.
assignment of `l' in `gr_update' to the return value of snprintf. It
claimed to have fixed the case where snprintf returned -1--in fact, it
broke the entire routine. Not setting `l' here causes fileupdate() to
invariably fail with EINVAL because it does its own check to assert
that the parameter isn't -1.
- Decrease reseeding interval from every 64 to every 16384 runs
to reduce entropy usage.
- Add time based reseeding. (Every 5 minutes.)
- Throw away the first 256 words of output as suggested in
"Weaknesses in the Key Scheduling Algorithm of RC4."
Reviewed by: Mark Murray
MFC After: jkh says ok
asprintf(3) when creating the request string, as the length of
a path is defined as unlimited by the standard and limiting the
total request to 4K is awfully arbitrary.
PR: 30054
Submitted by: Joseph Mallett <jmallett@xMach.org>
MFC after: 8 days
