large (1G) memory machine configurations. I was able to run 'dbench 32'
on a 32MB system without bring the machine to a grinding halt.
* buffer cache hash table now dynamically allocated. This will
have no effect on memory consumption for smaller systems and
will help scale the buffer cache for larger systems.
* minor enhancement to pmap_clearbit(). I noticed that
all the calls to it used constant arguments. Making
it an inline allows the constants to propogate to
deeper inlines and should produce better code.
* removal of inherent vfs_ioopt support through the emplacement
of appropriate #ifdef's, with John's permission. If we do not
find a use for it by the end of the year we will remove it entirely.
* removal of getnewbufloops* counters & sysctl's - no longer
necessary for debugging, getnewbuf() is now optimal.
* buffer hash table functions removed from sys/buf.h and localized
to vfs_bio.c
* VFS_BIO_NEED_DIRTYFLUSH flag and support code added
( bwillwrite() ), allowing processes to block when too many dirty
buffers are present in the system.
* removal of a softdep test in bdwrite() that is no longer necessary
now that bdwrite() no longer attempts to flush dirty buffers.
* slight optimization added to bqrelse() - there is no reason
to test for available buffer space on B_DELWRI buffers.
* addition of reverse-scanning code to vfs_bio_awrite().
vfs_bio_awrite() will attempt to locate clusterable areas
in both the forward and reverse direction relative to the
offset of the buffer passed to it. This will probably not
make much of a difference now, but I believe we will start
to rely on it heavily in the future if we decide to shift
some of the burden of the clustering closer to the actual
I/O initiation.
* Removal of the newbufcnt and lastnewbuf counters that Kirk
added. They do not fix any race conditions that haven't already
been fixed by the gbincore() test done after the only call
to getnewbuf(). getnewbuf() is a static, so there is no chance
of it being misused by other modules. ( Unless Kirk can think
of a specific thing that this code fixes. I went through it
very carefully and didn't see anything ).
* removal of VOP_ISLOCKED() check in flushbufqueues(). I do not
think this check is necessary, the buffer should flush properly
whether the vnode is locked or not. ( yes? ).
* removal of extra arguments passed to getnewbuf() that are not
necessary.
* missed cluster_wbuild() that had to be a cluster_wbuild_wb() in
vfs_cluster.c
* vn_write() now calls bwillwrite() *PRIOR* to locking the vnode,
which should greatly aid flushing operations in heavy load
situations - both the pageout and update daemons will be able
to operate more efficiently.
* removal of b_usecount. We may add it back in later but for now
it is useless. Prior implementations of the buffer cache never
had enough buffers for it to be useful, and current implementations
which make more buffers available might not benefit relative to
the amount of sophistication required to implement a b_usecount.
Straight LRU should work just as well, especially when most things
are VMIO backed. I expect that (even though John will not like
this assumption) directories will become VMIO backed some point soon.
Submitted by: Matthew Dillon <dillon@backplane.com>
Reviewed by: Kirk McKusick <mckusick@mckusick.com>
QUEUE_AGE, QUEUE_LRU, and QUEUE_EMPTY we instead have QUEUE_CLEAN,
QUEUE_DIRTY, QUEUE_EMPTY, and QUEUE_EMPTYKVA. With this patch clean
and dirty buffers have been separated. Empty buffers with KVM
assignments have been separated from truely empty buffers. getnewbuf()
has been rewritten and now operates in a 100% optimal fashion. That is,
it is able to find precisely the right kind of buffer it needs to
allocate a new buffer, defragment KVM, or to free-up an existing buffer
when the buffer cache is full (which is a steady-state situation for
the buffer cache).
Buffer flushing has been reorganized. Previously buffers were flushed
in the context of whatever process hit the conditions forcing buffer
flushing to occur. This resulted in processes blocking on conditions
unrelated to what they were doing. This also resulted in inappropriate
VFS stacking chains due to multiple processes getting stuck trying to
flush dirty buffers or due to a single process getting into a situation
where it might attempt to flush buffers recursively - a situation that
was only partially fixed in prior commits. We have added a new daemon
called the buf_daemon which is responsible for flushing dirty buffers
when the number of dirty buffers exceeds the vfs.hidirtybuffers limit.
This daemon attempts to dynamically adjust the rate at which dirty buffers
are flushed such that getnewbuf() calls (almost) never block.
The number of nbufs and amount of buffer space is now scaled past the
8MB limit that was previously imposed for systems with over 64MB of
memory, and the vfs.{lo,hi}dirtybuffers limits have been relaxed
somewhat. The number of physical buffers has been increased with the
intention that we will manage physical I/O differently in the future.
reassignbuf previously attempted to keep the dirtyblkhd list sorted which
could result in non-deterministic operation under certain conditions,
such as when a large number of dirty buffers are being managed. This
algorithm has been changed. reassignbuf now keeps buffers locally sorted
if it can do so cheaply, and otherwise gives up and adds buffers to
the head of the dirtyblkhd list. The new algorithm is deterministic but
not perfect. The new algorithm greatly reduces problems that previously
occured when write_behind was turned off in the system.
The P_FLSINPROG proc->p_flag bit has been replaced by the more descriptive
P_BUFEXHAUST bit. This bit allows processes working with filesystem
buffers to use available emergency reserves. Normal processes do not set
this bit and are not allowed to dig into emergency reserves. The purpose
of this bit is to avoid low-memory deadlocks.
A small race condition was fixed in getpbuf() in vm/vm_pager.c.
Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
Reviewed by: Kirk McKusick <mckusick@mckusick.com>
(really this time) fix pageout to swap and a couple of clustering cases.
This simplifies BUF_KERNPROC() so that it unconditionally reassigns the
lock owner rather than testing B_ASYNC and having the caller decide when
to do the reassign. At present this is required because some places use
B_CALL/b_iodone to free the buffers without B_ASYNC being set. Also,
vfs_cluster.c explicitly calls BUF_KERNPROC() when attaching the buffers
rather than the parent walking the cluster_head tailq.
Reviewed by: Kirk McKusick <mckusick@mckusick.com>
lockmgr locks. This commit should be functionally equivalent to the old
semantics. That is, all buffer locking is done with LK_EXCLUSIVE
requests. Changes to take advantage of LK_SHARED and LK_RECURSIVE will
be done in future commits.
at which we may sleep. So, after completing our buffer allocation
we must ensure that another process has not come along and allocated
a different buffer with the same identity. We do this by keeping a
global counter of the number of buffers that getnewbuf has allocated.
We save this count when we enter getnewbuf and check it when we are
about to return. If it has changed, then other buffers were allocated
while we were in getnewbuf, so we must return NULL to let our parent
know that it must recheck to see if it still needs the new buffer.
Hopefully this fix will eliminate the creation of duplicate buffers
with the same identity and the obscure corruptions that they cause.
piecemeal, middle-of-file writes for NFS. These hacks have caused no
end of trouble, especially when combined with mmap(). I've removed
them. Instead, NFS will issue a read-before-write to fully
instantiate the struct buf containing the write. NFS does, however,
optimize piecemeal appends to files. For most common file operations,
you will not notice the difference. The sole remaining fragment in
the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache
coherency issues with read-merge-write style operations. NFS also
optimizes the write-covers-entire-buffer case by avoiding the
read-before-write. There is quite a bit of room for further
optimization in these areas.
The VM system marks pages fully-valid (AKA vm_page_t->valid =
VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This
is not correct operation. The vm_pager_get_pages() code is now
responsible for marking VM pages all-valid. A number of VM helper
routines have been added to aid in zeroing-out the invalid portions of
a VM page prior to the page being marked all-valid. This operation is
necessary to properly support mmap(). The zeroing occurs most often
when dealing with file-EOF situations. Several bugs have been fixed
in the NFS subsystem, including bits handling file and directory EOF
situations and buf->b_flags consistancy issues relating to clearing
B_ERROR & B_INVAL, and handling B_DONE.
getblk() and allocbuf() have been rewritten. B_CACHE operation is now
formally defined in comments and more straightforward in
implementation. B_CACHE for VMIO buffers is based on the validity of
the backing store. B_CACHE for non-VMIO buffers is based simply on
whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear,
and vise-versa). biodone() is now responsible for setting B_CACHE
when a successful read completes. B_CACHE is also set when a bdwrite()
is initiated and when a bwrite() is initiated. VFS VOP_BWRITE
routines (there are only two - nfs_bwrite() and bwrite()) are now
expected to set B_CACHE. This means that bowrite() and bawrite() also
set B_CACHE indirectly.
There are a number of places in the code which were previously using
buf->b_bufsize (which is DEV_BSIZE aligned) when they should have
been using buf->b_bcount. These have been fixed. getblk() now clears
B_DONE on return because the rest of the system is so bad about
dealing with B_DONE.
Major fixes to NFS/TCP have been made. A server-side bug could cause
requests to be lost by the server due to nfs_realign() overwriting
other rpc's in the same TCP mbuf chain. The server's kernel must be
recompiled to get the benefit of the fixes.
Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
In heavy-writing situations, QUEUE_LRU can contain a large number
of DELWRI buffers at its head. These buffers must be moved
to the tail if they cannot be written async in order to reduce
the scanning time required to skip past these buffers in later
getnewbuf() calls.
Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
condition ( bufspace > hibufspace ), an inappropriate scan of the empty
queue was performed looking for buffer space to free up.
Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
unallocated parts of the last page when the file ended on a frag
but not a page boundary.
Delimitted by tags PRE_MATT_MMAP_EOF and POST_MATT_MMAP_EOF,
in files alpha/alpha/pmap.c i386/i386/pmap.c nfs/nfs_bio.c vm/pmap.h
vm/vm_page.c vm/vm_page.h vm/vnode_pager.c miscfs/specfs/spec_vnops.c
ufs/ufs/ufs_readwrite.c kern/vfs_bio.c
Submitted by: Matt Dillon <dillon@freebsd.org>
Reviewed by: Alan Cox <alc@freebsd.org>
including alan, john, me, luoqi, and kirk
Submitted by: Matt Dillon <dillon@frebsd.org>
This change implements a relatively sophisticated fix to getnewbuf().
There were two problems with getnewbuf(). First, the writerecursion
can lead to a system stack overflow when you have NFS and/or VN
devices in the system. Second, the free/dirty buffer accounting was
completely broken. Not only did the nfs routines blow it trying to
manually account for the buffer state, but the accounting that was
done did not work well with the purpose of their existance: figuring
out when getnewbuf() needs to sleep.
The meat of the change is to kern/vfs_bio.c. The remaining diffs are
all minor except for NFS, which includes both the fixes for bp
interaction AND fixes for a 'biodone(): buffer already done' lockup.
Sys/buf.h also contains a chaining structure which is not used by
this patchset but is used by other patches that are coming soon.
This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF.
(sorry for the missing T matt)
This produced races resulting in panics and filesystem corruptions
under some circumstances.
Reviewed by: luoqi chen <luoqi@freebsd.org>
Reviewed by: Kirk McKusick <mckusick@mckusick.com>
Submitted by: Matt Dillon <dillon@freebsd.org>
the buffer as still being dirty. This isn't a perfect solution, but
throwing away the buffer contents will often result in filesystem
corruption and this solution will at least correctly deal with transient
errors.
Submitted by: Kirk McKusick <mckusick@mckusick.com>
B_DELWRI and B_CACHE flags, fixing a bug that showed up with NFS.
Also, a number of cases where manually inserted code has been removed
and replaced with an inline function call giving us better functional
isolation in the source.
changes to the VM system to support the new swapper, VM bug
fixes, several VM optimizations, and some additional revamping of the
VM code. The specific bug fixes will be documented with additional
forced commits. This commit is somewhat rough in regards to code
cleanup issues.
Reviewed by: "John S. Dyson" <root@dyson.iquest.net>, "David Greenman" <dg@root.com>
Fix NFS file corruption problem introduced in 1.188. The valid range
was not being set properly, causing a later reference to the buffer
to clear the B_CACHE bit.
object are not page aligned). This should fix the mount_msdos panic after a
failed attemp to mount as ffs.
Reviewed By: Matthew Dillon <dillon@apollo.backplane.com>
Archie Cobbs <archie@whistle.com>
Dmitrij Tejblum <dima@tejblum.dnttm.rssi.ru>
bio interrupts, and a truncated file that along with the precise alignment
of the planets could result in a page being freed multiple times or a
just-freed page being put onto the inactive queue.
1) The vnode pager wasn't properly tracking the file size due to
"size" being page rounded in some cases and not in others.
This sometimes resulted in corrupted files. First noticed by
Terry Lambert.
Fixed by changing the "size" pager_alloc parameter to be a 64bit
byte value (as opposed to a 32bit page index) and changing the
pagers and their callers to deal with this properly.
2) Fixed a bogus type cast in round_page() and trunc_page() that
caused some 64bit offsets and sizes to be scrambled. Removing
the cast required adding casts at a few dozen callers.
There may be problems with other bogus casts in close-by
macros. A quick check seemed to indicate that those were okay,
however.
