r=adilger

Obtain necessary amount of peer credits for exery export per oss and for
every possible AST per MDS.

 r=adilger, deen

Description: When running with O_DIRECT I/O, neither the client rpc_stats
nor read_ahead_stats were updated. Copied the stats section from
osc_send_oap_rpc() into async_internal().

b=12916
i=green
i=shadow

handling of zero connect flags in self export for ELC code

b=13148
i=green
i=adilger

land 10800(adding llog ctxt refcount) to b1_6.
b=10800
i=green
i=shadow

Description: Async OSC create to avoid the blocking unnecessarily.
Details    : If a OST has no remain object, system will block on the creating
            when need to create a new object on this OST. Now, ways use
            pre-created objects when available, instead of blocking on an
            empty osc while others are not empty.  If we must block, we block
            for the shortest possible period of time.

Patch by shadow.

b=4900
i=wangdi
i=nathan

CONFIG_PAGE_ALLOC enabled

b=12975
i=adilger
i=green

To improve performance, disable most logging (for debug purposes)
by default.  VFSTRACE, RPCTRACE, and DLMTRACE are now off by
default, and HA includes contains fewer messages.

Original patch by adilger.

b=12417
i=scjody
i=green

i=adilger
i=green
fix mds_lov_synchronize race

b=11815
i=adilger
i=wangdi

Description: replace obdo_alloc() with OBDO_ALLOC macro
Details    : nothing special is done in obdo_alloc() function, and for
             debugging purpose, it needs to be replaced with macros.

b=11710
i=green
i=wangdi

land Early Lock Cancel code

rather than requested size.

b=11737
i=adilger
i=shadow

LCONSOLE_ERROR(fmt, arg) to LCONSOLE_ERROR_MSG(0x000, fmt, ## arg).

b=10595
i=adilger

It also includes code fixes for first five improved error messages.

b=10595
i=adilger
i=braam

Fix up error message for the case where a client hits a bogus checksum
error.  This only happens during mmap IO, and can be seen when the
client-side checksum matches the new checksum found on the server,
or has changed again since the message was originally sent.

In preparation for turning checksumming on by default.
b=11742
i=scjody
i=adilger (behlendorf original patch)

Always send IO to OST_IO_PORTAL regardless of what the default portal is.
Was previously broken for echo_client on catamount, but would also break
on 1.8 once the default request portal is REQPORTAL.

Detect if OST_WRITE or OST_READ requests go to non-IO portals, as that
would LBUG the server due to lack of IO buffers.
b=11813
i=green
i=shadow

The 11737 patch causes various kernel crashes; see bug 12549.
Verified by nathan and adilger.
Reverting.

Clear the grant when disconnect the import to prevent the grant being erased
wrongly when the import is evicted.
b=12371
i=green
i=adilger

rather than requested size.

b=11737
i=green
i=shadow

Add LNET router traceability for debug purposes

If a checksum failure occurs with a router as part of the IO path, the NID of
the last router that forwarded the bulk data is printed so it can be
identified.

Original patch by eeb.

b=11548
i=eeb
i=adilger

Patch from nic@cray.com: add spin locks around import/export bit flag changes.

b=11315
i=adilger
i=alex

According to shadow's advice, move BRW_WRITE check for grant to the right place
to avoid cl_loi_list_lock for read.
b:11662
r:adilger,green

b:11662
Fix grant leak that caused OST crashes.
r:adilger
r:green

b=11471

r=alex

 - limit number of in-flight async destroy rpcs MDS issues
   to destroy OST objects

change debug level.

missing update lvb_mtime

 - always close transaction before calling lov/osc

 - additional debug check

minor fix:
* for partial write only search capability once.

a minor reorganization of capability code.
some cleanup.

- all stats gatherd in OSC debugging time (cache loading, etc.) moved to proc. Added new proc file ./cache_stats showing them.
- some RPCs related new stats added to ./rpc_stats

fixes:
* previous implementation doesn't support multi-open/truncate by different
  users on the same inode.
* update DLD.
* cleanup.