- Move sonewconn(), which creates new sockets for incoming connections on
listen sockets, so that all socket allocate code is together in
uipc_socket.c.
- Move 'maxsockets' and associated sysctls to uipc_socket.c with the
socket allocation code.
- Move kern.ipc sysctl node to uipc_socket.c, add a SYSCTL_DECL() for it
to sysctl.h and remove lots of scattered implementations in various
IPC modules.
- Sort sodealloc() after soalloc() in uipc_socket.c for dependency order
reasons. Statisticize soalloc() and sodealloc() as they are now
required only in uipc_socket.c, and are internal to the socket
implementation.
After this change, socket allocation and deallocation is entirely
centralized in one file, and uipc_socket2.c consists entirely of socket
buffer manipulation and default protocol switch functions.
MFC after: 1 month
- Introducing the possibility of using locks different than mutexes
for the knlist locking. In order to do this, we add three arguments to
knlist_init() to specify the functions to use to lock, unlock and
check if the lock is owned. If these arguments are NULL, we assume
mtx_lock, mtx_unlock and mtx_owned, respectively.
- Using the vnode lock for the knlist locking, when doing kqueue operations
on a vnode. This way, we don't have to lock the vnode while holding a
mutex, in filt_vfsread.
Reviewed by: jmg
Approved by: re (scottl), scottl (mentor override)
Pointyhat to: ssouhlal
Will be happy: everyone
they both happen before pipe backing allocation occurs. Previously,
a pipe memory shortage would cause a panic due to a KNOTE call
on an uninitialized si_note.
Reported by: Peter Holm
MFC after: 1 week
release the pipe mutex before calling fsetown(), as fsetown()
may block. The sigio code protects the pipe sigio data using
its own mutex, and the pipe reference count held by the caller
will prevent the pipe from being prematurely garbage-collected.
Discovered by: imp
- pipespace is now able to resize non-empty pipes; this allows
for many more resizing opportunities
- Backing is no longer pre-allocated for the reverse direction
of pipes. This direction is rarely (if ever) used, so this cuts the
amount of map space allocated to a pipe in half.
- Pipe growth is now much more dynamic; a pipe will now grow when
the total amount of data it contains and the size of the write are
larger than the size of pipe. Previously, only individual writes greater
than the size of the pipe would cause growth.
- In low memory situations, pipes will now shrink during both read
and write operations, where possible. Once the memory shortage
ends, the growth code will cause these pipes to grow back to an appropriate
size.
- If the full PIPE_SIZE allocation fails when a new pipe is created, the
allocation will be retried with SMALL_PIPE_SIZE. This helps to deal
with the situation of a fragmented map after a low memory period has
ended.
- Minor documentation + code changes to support the above.
In total, these changes increase the total number of pipes that
can be allocated simultaneously, drastically reducing the chances that
pipe allocation will fail.
Performance appears unchanged due to dynamic resizing.
a more complete subsystem, and removes the knowlege of how things are
implemented from the drivers. Include locking around filter ops, so a
module like aio will know when not to be unloaded if there are outstanding
knotes using it's filter ops.
Currently, it uses the MTX_DUPOK even though it is not always safe to
aquire duplicate locks. Witness currently doesn't support the ability
to discover if a dup lock is ok (in some cases).
Reviewed by: green, rwatson (both earlier versions)
pipelock(), not via a mixture of mutexes and pipelock(). Additionally,
add a few KASSERTS, and change some statements that should have been
KASSERTS into KASSERTS.
As a result of these cleanups, some segments of code have become
significantly shorter and/or easier to read.
so that they know whether the allocation is supposed to be able to sleep
or not.
* Allow uma_zone constructors and initialation functions to return either
success or error. Almost all of the ones in the tree currently return
success unconditionally, but mbuf is a notable exception: the packet
zone constructor wants to be able to fail if it cannot suballocate an
mbuf cluster, and the mbuf allocators want to be able to fail in general
in a MAC kernel if the MAC mbuf initializer fails. This fixes the
panics people are seeing when they run out of memory for mbuf clusters.
* Allow debug.nosleepwithlocks on WITNESS to be disabled, without changing
the default.
Both bmilekic and jeff have reviewed the changes made to make failable
zone allocations work.
can't yet be referenced by other threads.
In microbenchmarks, this appears to reduce the cost of
pipe();close();close() on UP by 10%, and SMP by 7%. The vast majority
of the cost of allocating a pipe remains VM magic.
Suggested by: silby
instead of ephemeral mappings using pmap_qenter() by the writer. The
writer is still, however, responsible for wiring the pages, just not
mapping them. Consequently, the allocation of KVA for the direct case is
unnecessary. Remove it and the sysctls limiting it, i.e.,
kern.ipc.maxpipekvawired and kern.ipc.amountpipekvawired. The number
of temporarily wired pages is still, however, limited by
kern.ipc.maxpipekva.
Note: On platforms lacking a direct virtual-to-physical mapping,
uiomove_fromphys() uses sf_bufs to cache ephemeral mappings. Thus,
the number of available sf_bufs can influence the performance of pipes
on platforms such i386. Surprisingly, I saw the greatest gain from this
change on such a machine: lmbench's pipe bandwidth result increased from
~1050MB/s to ~1850MB/s on my 2.4GHz, 400MHz FSB P4 Xeon.
of all, PIPE_EOF is not checked pervasively after everything that can drop
the pipe mutex and msleep(), so fix. Additionally, though it might not
harm anything, pipelock() and pipeunlock() are not used consistently.
Third, the kqueue support functions do not use the pipe mutex correctly.
Last, but absolutely not least, is a race: if pipe_busy is not set on
the closing side of the pipe, the other side that is trying to write to
that will crash BECAUSE PIPE_EOF IS NOT SET! Unconditionally set
PIPE_EOF, and get rid of all the lockups/crashes I have seen trying
to build ports.
instead, just dec/inc in the ctor/dtor. For now, increment/decrement
in two's, since we're now performing the operation once per pair,
not once per pipe. Not really any measurable performance change
in my micro-benchmarks, but doing less work is good, especially when
it comes to atomic operations.
Suggested by: alc
changes to jointly allocated pipe pairs. Replace these checks
with pipe_present checks. This avoids a NULL pointer dereference
when a pipe is half-closed.
Submitted by: Peter Edwards <peter.edwards@openet-telecom.com>
would allocate two 'struct pipe's from the pipe zone, and malloc a
mutex.
- Create a new "struct pipepair" object holding the two 'struct
pipe' instances, struct mutex, and struct label reference. Pipe
structures now have a back-pointer to the pipe pair, and a
'pipe_present' flag to indicate whether the half has been
closed.
- Perform mutex init/destroy in zone init/destroy, avoiding
reallocating the mutex for each pipe. Perform most pipe structure
setup in zone constructor.
- VM memory mappings for pageable buffers are still done outside of
the UMA zone.
- Change MAC API to speak 'struct pipepair' instead of 'struct pipe',
update many policies. MAC labels are also handled outside of the
UMA zone for now. Label-only policy modules don't have to be
recompiled, but if a module is recompiled, its pipe entry points
will need to be updated. If a module actually reached into the
pipe structures (unlikely), that would also need to be modified.
These changes substantially simplify failure handling in the pipe
code as there are many fewer possible failure modes.
On half-close, pipes no longer free the 'struct pipe' for the closed
half until a full-close takes place. However, VM mapped buffers
are still released on half-close.
Some code refactoring is now possible to clean up some of the back
references, etc; this patch attempts not to change the structure
of most of the pipe implementation, only allocation/free code
paths, so as to avoid introducing bugs (hopefully).
This cuts about 8%-9% off the cost of sequential pipe allocation
and free in system call tests on UP and SMP in my micro-benchmarks.
May or may not make a difference in macro-benchmarks, but doing
less work is good.
Reviewed by: juli, tjr
Testing help: dwhite, fenestro, scottl, et al
in OpenBSD by Niels Provos. The patch introduces a bitmap of allocated
file descriptors which is used to locate available descriptors when a new
one is needed. It also moves the task of growing the file descriptor table
out of fdalloc(), reducing complexity in both fdalloc() and do_dup().
Debts of gratitude are owed to tjr@ (who provided the original patch on
which this work is based), grog@ (for the gdb(4) man page) and rwatson@
(for assistance with pxeboot(8)).
thread being waken up. The thread waken up can run at a priority as
high as after tsleep().
- Replace selwakeup()s with selwakeuppri()s and pass appropriate
priorities.
- Add cv_broadcastpri() which raises the priority of the broadcast
threads. Used by selwakeuppri() if collision occurs.
Not objected in: -arch, -current
Instead, let the vm objects be lazily instantiated at fault time. This
results in the allocation of fewer vm objects and vm map entries due to
aggregation in the vm system.
table, acquiring the necessary locks as it works. It usually returns
two references to the new descriptor: one in the descriptor table
and one via a pointer argument.
As falloc releases the FILEDESC lock before returning, there is a
potential for a process to close the reference in the file descriptor
table before falloc's caller gets to use the file. I don't think this
can happen in practice at the moment, because Giant indirectly protects
closes.
To stop the file being completly closed in this situation, this change
makes falloc set the refcount to two when both references are returned.
This makes life easier for several of falloc's callers, because the
first thing they previously did was grab an extra reference on the
file.
Reviewed by: iedowse
Idea run past: jhb
kqueue write events on a socket and you regularly create tons of pipes
which overwrites the structure causing a panic when removing the knote
from the list. If the peer has gone away (and it's a write knote), then
don't bother trying to remove the knote from the list.
Submitted by: Brian Buchanan and myself
Obtained from: nCircle
From alc:
Move pageable pipe memory to a seperate kernel submap to avoid awkward
vm map interlocking issues. (Bad explanation provided by me.)
From me:
Rework pipespace accounting code to handle this new layout, and adjust
our default values to account for the fact that we now have a solid
limit on allocations.
Also, remove the "maxpipes" limit, as it no longer has a purpose.
(The limit on kva usage solves the problem of having two many pipes.)
application could cause a wired page to be freed. In general,
vm_page_hold() should be preferred for ephemeral kernel mappings of pages
borrowed from a user-level address space. (vm_page_wire() should really be
reserved for indefinite duration pinning by the "owner" of the page.)
Discussed with: silby
Submitted by: tegge
on a non-blocking pipe in cases where select(2) returns the file
descriptor as ready for write. This in turns causes libc_r, for
one, to busy wait in such cases.
Note: it is a quick performance fix, a more complex fix might be
required in case this turns out to have unexpected side effects.
Reviewed by: silby
MFC after: 3 days
a long-standing mistake in the way a portion of a pipe's KVA is
allocated. Specifically, kmem_alloc_pageable() is inappropriate
for use in the "direct" case because it allows a preceding vm map entry
and vm object to be extended to support the new KVA allocation.
However, the direct case KVA allocation should not have a backing
vm object. This is corrected by using kmem_alloc_nofault().
Submitted by: tegge (with the above explanation by me)
- Use atomic ops to update the bigpipe count
- Make the bigpipe count sysctl readable
- Remove a duplicate comparison in an if statement
- Comment two SYSCTLs.
- Limit the total number of pipes so that we do not
exhaust all vm objects in the kernel map. When
this limit is reached, a ratelimited message will
be printed to the console.
- Put a soft limit on the amount of memory consumable
by pipes. Once the limit has been reached, all new
pipes will be limited to 4K in size, rather than the
default of 16K.
- Put a limit on the number of pages that may be used
for high speed page flipping in order to reduce the
amount of wired memory. Pipe writes that occur
while this limit is exceeded will fall back to
non-page flipping mode.
The above values are auto-tuned in subr_param.c and
are scaled to take into account both the size of
physical memory and the size of the kernel map.
These limits help to reduce the "kernel resources exhausted"
panics that could be caused by opening a large
number of pipes. (Pipes alone are no longer able
to exhaust all resources, but other kernel memory hogs
in league with pipes may still be able to do so.)
PR: 53627
Ideas / comments from: hsu, tjr, dillon@apollo.backplane.com
MFC after: 1 week
