late stages of unmount). On failure, the vnode is recycled.
Add insmntque1(), to allow for file system specific cleanup when
recycling vnode on failure.
Change getnewvnode() to no longer call insmntque(). Previously,
embryonic vnodes were put onto the list of vnode belonging to a file
system, which is unsafe for a file system marked MPSAFE.
Change vfs_hash_insert() to no longer lock the vnode. The caller now
has that responsibility.
Change most file systems to lock the vnode and call insmntque() or
insmntque1() after a new vnode has been sufficiently setup. Handle
failed insmntque*() calls by propagating errors to callers, possibly
after some file system specific cleanup.
Approved by: re (kensmith)
Reviewed by: kib
In collaboration with: kib
vnode into the vfs hash. Otherwise, another thread walking the hash can trip
on an nfsnode with an uninitialized or partially initialized fh.
Thanks to ups@ for spotting this race.
file. Leave the loser's lock(s) initialized, so the reclaim logic can
unconditionally destroy them when that race occurs (or if the vfs hash
insert happened to fail for some other reason). Thanks to ups@ for a
careful review of the code.
Reported by : Kris Kennaway
adds a FS type specific flag indicating that the FS supports shared
vnode lock lookups, adds some logic in vfs_lookup.c to test this flag
and set lock flags appropriately.
- amd on 6.x is a non-starter (without this change). Using amd under
heavy load results in a deadlock (with cascading vnode locks all the
way to the root) very quickly.
- This change should also fix the more general problem of cascading
vnode deadlocks when an NFS server goes down.
Ideally, we wouldn't need these changes, as enabling shared vnode lock
lookups globally would work. Unfortunately, UFS, for example isn't
ready for shared vnode lock lookups, crashing pretty quickly.
This change is the result of discussions with Stephan Uphoff (ups@).
Reviewed by: ups@
last few days. I tracked it down to the fact that nfs_reclaim()
is setting vp->v_data to NULL _before_ calling vnode_destroy_object().
After silence from the mailing list I checked further and discovered
that ufs_reclaim() is unique among FreeBSD filesystems for calling
vnode_destroy_object() early, long before tossing v_data or much
of anything else, for that matter. The rest, including NFS, appear
to be identical, as if they were just clones of one original routine.
The enclosed patch fixes all file systems in essentially the same
way, by moving the call to vnode_destroy_object() to early in the
routine (before the call to vfs_hash_remove(), if any). I have
only tested NFS, but I've now run for over eighteen hours with the
patch where I wouldn't get past four or five without it.
Submitted by: Frank Mayhar
Requested by: Mohan Srinivasan
MFC After: 1 week
writers that want to extend the file. It was also used to serialize
readers that might want to read the last block of the file (with a
writer extending the file). Now that we support vnode locking for
NFS, the rslock is unnecessary. Writers grab the exclusive vnode
lock before writing and readers grab the shared (or in some cases
the exclusive) lock.
Submitted by: Mohan Srinivasan
vnode EXCLUSIVE lock. This prevents threads from adding pages to
the vnode while an invalidation is in progress, closing potential
races. In the bioread() path, callers acquire the SHARED vnode lock
- so while an invalidate was in progress, it was possible to fault
in new pages onto the vnode causing the invalidation to take a while
or fail. We saw these races at Yahoo! with very large files+heavy
concurrent access. Forcing an upgrade to EXCLUSIVE lock before doing
the invalidation closes all these races.
Submitted by: Mohan Srinivasan mohans at yahoo-inc dot com
initializations but we did have lofty goals and big ideals.
Adjust to more contemporary circumstances and gain type checking.
Replace the entire vop_t frobbing thing with properly typed
structures. The only casualty is that we can not add a new
VOP_ method with a loadable module. History has not given
us reason to belive this would ever be feasible in the the
first place.
Eliminate in toto VOCALL(), vop_t, VNODEOP_SET() etc.
Give coda correct prototypes and function definitions for
all vop_()s.
Generate a bit more data from the vnode_if.src file: a
struct vop_vector and protype typedefs for all vop methods.
Add a new vop_bypass() and make vop_default be a pointer
to another struct vop_vector.
Remove a lot of vfs_init since vop_vector is ready to use
from the compiler.
Cast various vop_mumble() to void * with uppercase name,
for instance VOP_PANIC, VOP_NULL etc.
Implement VCALL() by making vdesc_offset the offsetof() the
relevant function pointer in vop_vector. This is disgusting
but since the code is generated by a script comparatively
safe. The alternative for nullfs etc. would be much worse.
Fix up all vnode method vectors to remove casts so they
become typesafe. (The bulk of this is generated by scripts)
Extend it with a strategy method.
Add bufstrategy() which do the usual VOP_SPECSTRATEGY/VOP_STRATEGY
song and dance.
Rename ibwrite to bufwrite().
Move the two NFS buf_ops to more sensible places, add bufstrategy
to them.
Add inlines for bwrite() and bstrategy() which calls through
buf->b_bufobj->b_ops->b_{write,strategy}().
Replace almost all VOP_STRATEGY()/VOP_SPECSTRATEGY() calls with bstrategy().
This includes a modified form of some code from Thomas Moestl (tmm@)
to properly clean up the UMA zone and the "nfsnodehashtbl" hash
table.
Reviewed By: iedowse
PR: 16299
VOP_INACTIVE routines need not worry about their vnode getting
recycled if they block. Remove the code from nfs_inactive() that
used vget() to get an extra vnode reference that was held during
the nfs_vinvalbuf() call.
that use it. Specifically, vop_stdlock uses the lock pointed to by
vp->v_vnlock. By default, getnewvnode sets up vp->v_vnlock to
reference vp->v_lock. Filesystems that wish to use the default
do not need to allocate a lock at the front of their node structure
(as some still did) or do a lockinit. They can simply start using
vn_lock/VOP_UNLOCK. Filesystems that wish to manage their own locks,
but still use the vop_stdlock functions (such as nullfs) can simply
replace vp->v_vnlock with a pointer to the lock that they wish to
have used for the vnode. Such filesystems are responsible for
setting the vp->v_vnlock back to the default in their vop_reclaim
routine (e.g., vp->v_vnlock = &vp->v_lock).
In theory, this set of changes cleans up the existing filesystem
lock interface and should have no function change to the existing
locking scheme.
Sponsored by: DARPA & NAI Labs.
v_tag is now const char * and should only be used for debugging.
Additionally:
1. All users of VT_NTS now check vfsconf->vf_type VFCF_NETWORK
2. The user of VT_PROCFS now checks for the new flag VV_PROCDEP, which
is propagated by pseudofs to all child vnodes if the fs sets PFS_PROCDEP.
Suggested by: phk
Reviewed by: bde, rwatson (earlier version)
Note ALL MODULES MUST BE RECOMPILED
make the kernel aware that there are smaller units of scheduling than the
process. (but only allow one thread per process at this time).
This is functionally equivalent to teh previousl -current except
that there is a thread associated with each process.
Sorry john! (your next MFC will be a doosie!)
Reviewed by: peter@freebsd.org, dillon@freebsd.org
X-MFC after: ha ha ha ha
other "system" header files.
Also help the deprecation of lockmgr.h by making it a sub-include of
sys/lock.h and removing sys/lockmgr.h form kernel .c files.
Sort sys/*.h includes where possible in affected files.
OK'ed by: bde (with reservations)
enable easy access to the hash chain stats. The raw prefixed versions
dump an integer array to userland with the chain lengths. This cheats
and calls it an array of 'struct int' rather than 'int' or sysctl -a
faithfully dumps out the 128K array on an average machine. The non-raw
versions return 4 integers: count, number of chains used, maximum chain
length, and percentage utilization (fixed point, multiplied by 100).
The raw forms are more useful for analyzing the hash distribution, while
the other form can be read easily by humans and stats loggers.
Allow the initial hash value to be passed in, as the examples do.
Incrementally hash in the dvp->v_id (using the official api) rather than
add it. This seems to help power-of-two predictable filename trees
where the filenames repeat on a power-of-two cycle and the directory trees
have power-of-two components in it. The simple add then mask was causing
things like 12000+ entry collision chains while most other entries have
between 0 and 3 entries each. This way seems to improve things.
Make the name cache hash as well as the nfsnode hash use it.
As a special tweak, create an unsigned version of register_t. This allows
us to use a special tweak for the 64 bit versions that significantly
speeds up the i386 version (ie: int64 XOR int64 is slower than int64
XOR int32).
The code layout is a little strange for the string function, but I was
able to get between 5 to 10% improvement over the original version I
started with. The layout affects gcc code generation choices and this way
was fastest on x86 and alpha.
Note that 'CPUTYPE=p3' etc makes a fair difference to this. It is
around 45% faster with -march=pentiumpro on a p6 cpu.
Fowler / Noll / Vo Hash (http://www.isthe.com/chongo/tech/comp/fnv/).
This improves hash coverage a *massive* amount. We were seeing one
set of machines that were using 0.84% of their 131072 entry nfsnode
hash buckets with maximum chain lengths of up to ~500 entries. The
machine was spending nearly 100% of its time in 'system'.
A test with this has pushed the coverage from a few perCent up to 91%
utilization with a max chain length of 11.
Submitted by: David Filo
Add lockdestroy() and appropriate invocations, which corresponds to
lockinit() and must be called to clean up after a lockmgr lock is no
longer needed.
