the ffs_copyonwrite routine to avoid a deadlock between the syncer
daemon trying to sync out a snapshot vnode and the bufdaemon
trying to write out a buffer containing the snapshot inode.
With any luck this will be the last snapshot race condition.
Sponsored by: DARPA & NAI Labs.
a full filesystem. Previously, if the allocation failed, we had to
fsync the file before rolling back any partial allocation of indirect
blocks. Most block allocation requests only need to allocate a single
data block and if that allocation fails, there is nothing to unroll.
So, before doing the fsync, we check to see if any rollback will
really be necessary. If none is necessary, then we simply return.
This update eliminates the flurry of disk activity that got triggered
whenever a filesystem would run out of space.
Sponsored by: DARPA & NAI Labs.
locks the mount point directory while waiting for vfs_busy to clear.
Meanwhile the unmount which holds the vfs_busy lock tried to lock
the mount point vnode. The fix is to observe that it is safe for the
unmount to remove the vnode from the mount point without locking it.
The lookup will wait for the unmount to complete, then recheck the
mount point when the vfs_busy lock clears.
Sponsored by: DARPA & NAI Labs.
that works in the new threaded kernel. It was commented out of
the disksort routine earlier this year for the reasons given in
kern/subr_disklabel.c (which is where this code used to reside
before it moved to kern/subr_disk.c):
----------------------------
revision 1.65
date: 2002/04/22 06:53:20; author: phk; state: Exp; lines: +5 -0
Comment out Kirks io-request priority hack until we can do this in a
civilized way which doesn't cause grief.
The problem is that it is not generally safe to cast a "struct bio
*" to a "struct buf *". Things like ccd, vinum, ata-raid and GEOM
constructs bio's which are not entrails of a struct buf.
Also, curthread may or may not have anything to do with the I/O request
at hand.
The correct solution can either be to tag struct bio's with a
priority derived from the requesting threads nice and have disksort
act on this field, this wouldn't address the "silly-seek syndrome"
where two equal processes bang the diskheads from one edge to the
other of the disk repeatedly.
Alternatively, and probably better: a sleep should be introduced
either at the time the I/O is requested or at the time it is completed
where we can be sure to sleep in the right thread.
The sleep also needs to be in constant timeunits, 1/hz can be practicaly
any sub-second size, at high HZ the current code practically doesn't
do anything.
----------------------------
As suggested in this comment, it is no longer located in the disk sort
routine, but rather now resides in spec_strategy where the disk operations
are being queued by the thread that is associated with the process that
is really requesting the I/O. At that point, the disk queues are not
visible, so the I/O for positively niced processes is always slowed
down whether or not there is other activity on the disk.
On the issue of scaling HZ, I believe that the current scheme is
better than using a fixed quantum of time. As machines and I/O
subsystems get faster, the resolution on the clock also rises.
So, ten years from now we will be slowing things down for shorter
periods of time, but the proportional effect on the system will
be about the same as it is today. So, I view this as a feature
rather than a drawback. Hence this patch sticks with using HZ.
Sponsored by: DARPA & NAI Labs.
Reviewed by: Poul-Henning Kamp <phk@critter.freebsd.dk>
Various cleanups, no functional changes:
- Fix a type in an entry point stub, socket checks accept
sockets, not vnodes.
- Trailing whitespace
- Entry point sort order
Obtained from: TrustedBSD Project
Sponsored by: DARPA, Network Associates Laboratories
It is never used. I left it there from pre-KSE days as I didn't know
if I'd need it or not but now I know I don't.. It's functionality
is in TDI_IWAIT in the thread.
copy elements of one Biba or MLS label to another based on the flags
on the source label element. Use this instead of
mac_{biba,mls}_{single,range}() to simplify the existing code, as
well as support partial label updates (we don't update if none is
requested).
Obtained from: TrustedBSD Project
Sponsored by: DARPA, Network Associates Laboratories
a server process bound to a wildcard UDP socket to select the IP
address from which outgoing packets are sent on a per-datagram
basis. When combined with IP_RECVDSTADDR, such a server process can
guarantee to reply to an incoming request using the same source IP
address as the destination IP address of the request, without having
to open one socket per server IP address.
Discussed on: -net
Approved by: re
to send datagrams from an unconnected socket, we used to first block
input, then connect the socket to the sendmsg/sendto destination,
send the datagram, and finally disconnect the socket and unblock
input.
We now use in_pcbconnect_setup() to check if a connect() would have
succeeded, but we never record the connection in the PCB (local
anonymous port allocation is still recorded, though). The result
from in_pcbconnect_setup() authorises the sending of the datagram
and selects the local address and port to use, so we just construct
the header and call ip_output().
Discussed on: -net
Approved by: re
revert to checking the name to determine if our root device is a ramdisk,
md(4) specifically to determine if we should attempt the root-mount RW
Sponsored by: DARPA & NAI Labs.
policies remains the same: subjects and objects are labeled for
integrity or sensitivity, and a dominance operator determines whether
or not subject/object accesses are permitted to limit inappropriate
information flow. Compartments are a non-hierarchal component to
the label, so add a bitfield to the label element for each, and a
set check as part of the dominance operator. This permits the
implementation of "need to know" elements of MLS.
Obtained from: TrustedBSD Project
Sponsored by: DARPA, Network Associates Laboratories
range on them, leaving process credentials as the only kernel
objects with label ranges in the Biba and MLS policies. We
weren't using the range in any access control decisions, so this
lets us garbage collect effectively unused code.
Obtained from: TrustedBSD Project
Sponsored by: DARPA, Network Associates Laboratories
collapse the two cases more cleanly: rather than wrapping an access
check around open, simply provide the open implementation for the
access vector entry. No functional change.
Obtained from: TrustedBSD Project
Sponsored by: DARPA, Network Associates Laboratories
we just break out some of the tests better. Minor change in that
we now better support incremental update of labels.
Obtained from: TrustedBSD Project
Sponsored by: DARPA, Network Associates Laboratories
in_pcbconnect() called in_pcbconnect_setup(). This version performs
all of the functions of in_pcbconnect() except for the final
committing of changes to the PCB. In the case of an EADDRINUSE error
it can also provide to the caller the PCB of the duplicate connection,
avoiding an extra in_pcblookup_hash() lookup in tcp_connect().
This change will allow the "temporary connect" hack in udp_output()
to be removed and is part of the preparation for adding the
IP_SENDSRCADDR control message.
Discussed on: -net
Approved by: re
and save/restore during a context switch.
The USER_SR could be overwritten when the current thread was switched
out with a faulting copyin/copyout.
Approved by: Benno
instead of the default biba/high, mls/low, making it easier to use
ptys with these policies. This isn't the final solution, but does
help.
Obtained from: TrustedBSD Project
Sponsored by: DARPA, Network Associates Laboratories
pointer to a dumperinfo instead of di. A brainfart, surely. This
bug went unnoticed for all this time because the pointer is only
used by buf_write() when it can write a completely filled buffer
to the dump device. This depends on the number of memory chunks
that needs to be dumped. This has apparently been low enough that
it has never happened up until this point.
contents. The code was subtracting two unsigned ints, stored the
result in a log and expected it to be the same as of a signed
subtraction; this does only work on platforms where int and long
have the same size (due to overflows).
Instead, cast to long before the subtraction; the numbers are
guaranteed to be small enough so that there will be no overflows
because of that.
