Jason Thorpe suggests removing the limit altogether; anyone who needs
more than 1024 nfsds can still do "nfsd -n 1024" multiple times, and
this provides _some_ saftey-net against nfsd (issued by root) from
eating all the avaiable process slots.
Increase NFS_MAXRAHEAD to 32. With 32k read or write requests, that
amounts to 1 Mbyte of read-ahead, enough to cover about 10 ms latency
at gigabit Ethernet speeds. Increase the table of nfsiod kthreads
(NFS_MAXASYNCDAEMON) from 20 to 128, to match the raised value of
NFS_MAXRAHEAD. (Making the limit dynamic requires replacing the
compile-time array with a dynamic structure.)
Add a comment explaining that each read-ahead requires an I/O thread.
Wrap both parameters with an #ifdef <parameter>/#endif, to allow
hand-tuned values or (later) a kernel config-file option override.
the tag before forwarding the packet, make sure the packet+tag is at least
68 bytes long.
This is necessary because our parent will only pad to 64 bytes (ETHER_MIN_LEN)
and some switches will not pad by themselves after deleting a tag.
* Make tok_init(), tok_end(), tok_reset(), tok_line() and tok_str()
publically available in <histedit.h>
* Documented the public functions in editline(3)
* Renamed tok_line() -> tok_str()
* Added new tok_line() which takes a "const LineInfo *" instead of
"const char *" (the former has "cursor" information), and optionally
return the argv index ("int *cursorc") and offset within that index
("int *cursorv"). This means that completion routines can use the
tokenization code to crack the line and easily find which word the
cursor is at. (mmm, context sensitive completion :)
* Fixed TEST/test.c when using "continuation" lines (unmatched quote
or \ at EOL), and added some more DEBUG messages including highlighting
where the cursor is (with a `_').
* don't let cac_ccb_alloc() sleep since it regularly runs in an
interrupt context
* return EAGAIN instead EBUSY (or -1), which is what the upper ld
layer expects to get on transient resource shortages.
* ignore error from cac_ccb_start(), since a `fifo full' condition is
handled internally in the lower layer, i.e. the transaction started
from ldstart() has been queued at the lower layer and must be taken
off ld's disk queue.
