Age | Commit message (Collapse) | Author |
|
Renauld of Network Storage Solutions, Inc. Many fixes, wider device
support. In particular, the notorious 'Target 0' problem seems to be
fixed.
Does *not* include any updates to isa or eisa code beyond what was
necessary to compile.
Known issues:
1) Tagged Queuing is probably not optimal.
2) PPR negotiation may not be fully functional.
3) No support yet for freezing devices or channels.
4) The mechanism for preventing 'A' and 'B' channel confusion during probe
can fail if scsibus > 254 found.
5) Requeuing I/O's not working. A workaround will be committed almost
immediately. At the moment timeouts, SCSI message rejects, aborting
SCB's and trying to freeze a device may cause incomplete i/o's to be
reported as complete.
6) Verbosity and probe messages need work.
7) Last disk on bus seems to go through an extra re-negotiation.
8) >16 devices on an adapter will trigger the usual problems of total
openings exceeding available SCB's under heavy load.
Tested by deraadt@, beck@, miod@, naddy@, drahn@, marc@ amoung
others.
ok deraadt@.
|
|
and process the bgp messages after each other out of that big buffer.
gives a nice speedup, easier code and earlier connection close detection.
|
|
Renauld of Network Storage Solutions, Inc. Many fixes, wider device
support. In particular, the notorious 'Target 0' problem seems to be
fixed.
Does *not* include any updates to isa or eisa code beyond what was
necessary to compile.
Known issues:
1) Tagged Queuing is probably not optimal.
2) PPR negotiation may not be fully functional.
3) No support yet for freezing devices or channels.
4) The mechanism for preventing 'A' and 'B' channel confusion during probe
can fail if scsibus > 254 found.
5) Requeuing I/O's not working. A workaround will be committed almost
immediately. At the moment timeouts, SCSI message rejects, aborting
SCB's and trying to freeze a device may cause incomplete i/o's to be
reported as complete.
6) Verbosity and probe messages need work.
7) Last disk on bus seems to go through an extra re-negotiation.
8) >16 devices on an adapter will trigger the usual problems of total
openings exceeding available SCB's under heavy load.
Tested by deraadt@, beck@, miod@, naddy@, drahn@, marc@ amoung
others.
ok deraadt@.
|
|
Renauld of Network Storage Solutions, Inc. Many fixes, wider device
support. In particular, the notorious 'Target 0' problem seems to be
fixed.
Does *not* include any updates to isa or eisa code beyond what was
necessary to compile.
Known issues:
1) Tagged Queuing is probably not optimal.
2) PPR negotiation may not be fully functional.
3) No support yet for freezing devices or channels.
4) The mechanism for preventing 'A' and 'B' channel confusion during probe
can fail if scsibus > 254 found.
5) Requeuing I/O's not working. A workaround will be committed almost
immediately. At the moment timeouts, SCSI message rejects, aborting
SCB's and trying to freeze a device may cause incomplete i/o's to be
reported as complete.
6) Verbosity and probe messages need work.
7) Last disk on bus seems to go through an extra re-negotiation.
8) >16 devices on an adapter will trigger the usual problems of total
openings exceeding available SCB's under heavy load.
Tested by deraadt@, beck@, miod@, naddy@, drahn@, marc@ amoung
others.
ok deraadt@.
|
|
Every non-KNF file in the source tree contains at least one severe bug.
KNF and a timid start at cleaning, but not very far, so there is no
functional change at the moment.
This code is still pathetic for now, but it sorta works and the price
was right.
|
|
Renauld of Network Storage Solutions, Inc. Many fixes, wider device
support. In particular, the notorious 'Target 0' problem seems to be
fixed.
Does *not* include any updates to isa or eisa code beyond what was
necessary to compile.
Known issues:
1) Tagged Queuing is probably not optimal.
2) PPR negotiation may not be fully functional.
3) No support yet for freezing devices or channels.
4) The mechanism for preventing 'A' and 'B' channel confusion during probe
can fail if scsibus > 254 found.
5) Requeuing I/O's not working. A workaround will be committed almost
immediately. At the moment timeouts, SCSI message rejects, aborting
SCB's and trying to freeze a device may cause incomplete i/o's to be
reported as complete.
6) Verbosity and probe messages need work.
7) Last disk on bus seems to go through an extra re-negotiation.
8) >16 devices on an adapter will trigger the usual problems of total
openings exceeding available SCB's under heavy load.
Tested by deraadt@, beck@, miod@, naddy@, drahn@, marc@ amoung
others.
ok deraadt@.
|
|
|
|
|
|
|
|
at this point in development, bgpd establishes connections to neighbors fine
and does all the session handling as desired, processes the announcements
it gets from its neighbors, runs the decision process and enters the routes to
the kernel routing table. they are also updated as needed and cleaned up on
exit.
bgpd does not yet have the ability to announce anything, or send outgoing
UPDATE messages in general, nor does it have a filter language yet.
|
|
multiple '&'s. Resolves PR 3616.
ok deraadt@
|
|
somewhat fragile IMSG_SHUTDOWN_* stuff any more. speeds shutdown up
enourmously.
ok claudio@
|
|
them easily on shutdown without the RDE's help
|
|
Pointed out by Jorge Severino (jorge at netsecure dot cl)
|
|
do only ever delete those, and no others
|
|
|
|
|
|
sigsetjmp(foo, !0) to work. Sigh.
|
|
|
|
uvm_fault() returning EACCES as a segmentation fault rather than a bus
error, whatever address the fault is at.
As a result, this correctly delivers SIGSEGV, rather than SIGBUS, when
attempting to write to a page with only PROT_READ permissions.
|
|
|
|
|
|
requested by theo
|
|
it's better when documentation and code agree on the keyword...
|
|
|
|
and make all callers cope.
|
|
the error to be emitted.
ok miod@.
|
|
|
|
apparently these were already fixed in the diff jmc@ sent me and I somehow
screwed up when getting it in - sorry.
|
|
|
|
|
|
Found and tested by our ubiquitous afs user, Bob Beck.
|
|
|
|
|
|
|
|
treated as usermode traps, because they reference a user space address.
However, the tests for pcb_onfault being set were only present in the
kernelmode traps handling.
Since pcb_onfault is only set in those functions, move the associated
recovery code from the kernelmode part to the usermode part, and only
attempt to jump to pcb_onfault() if the access could not be resolved
by uvm_fault() earlier.
This lets things like setlogin(NULL) correctly return EFAULT, rather
than killing the process with SIGSEGV, and incidentally lets sendmail
in a non-default configuration run.
|
|
|
|
|
|
From: FreeBSD' libc_r
ok marc@
|
|
|
|
|
|
|
|
|
|
|
|
LIBC_R_DEBUG -> LIBPTHREAD_DEBUG
ok marc@
|
|
|
|
If an application closes one of its stdio descriptors (0..2),
an excessive close() on one of these descriptors would cause
a memory for this descriptor to be allocated in the internal
descriptor table. When this descriptor gets used again, e.g.
through the call to open() or socket(), the descriptor would
be erroneously left in the blocking mode, and the whole
application would get stuck on a blocking operation, e.g.,
in accept(2).
but changed to not eat fds when a file that the thread kernel doesn't
know about is closed.
|
|
|
|
- kill some .Pp's before and after a list
|
|
and sendto() cancellation points, as required by POSIX.1-2001.
From: FreeBSD' libc_r
ok marc@
|