Differences between how paging is done on Solaris and Linux can cause
deadlocks if KM_SLEEP is used in any the following contexts.
* The txg_sync thread
* The zvol write/discard threads
* The zpl_putpage() VFS callback
This is because KM_SLEEP will allow for direct reclaim which may result
in the VM calling back in to the filesystem or block layer to write out
pages. If a lock is held over this operation the potential exists to
deadlock the system. To ensure forward progress all memory allocations
in these contexts must us KM_PUSHPAGE which disables performing any I/O
to accomplish the memory allocation.
Previously, this behavior was acheived by setting PF_MEMALLOC on the
thread. However, that resulted in unexpected side effects such as the
exhaustion of pages in ZONE_DMA. This approach touchs more of the zfs
code, but it is more consistent with the right way to handle these cases
under Linux.
This is patch lays the ground work for being able to safely revert the
following commits which used PF_MEMALLOC:
21ade34 Disable direct reclaim for z_wr_* threads
cfc9a5c Fix zpl_writepage() deadlock
eec8164 Fix ASSERTION(!dsl_pool_sync_context(tx->tx_pool))
Signed-off-by: Richard Yao <ryao@cs.stonybrook.edu>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #726
The original range lock implementation had to be modified by commit
8926ab7 because it was unsafe on Linux. In particular, calling
cv_destroy() immediately after cv_broadcast() is dangerous because
the waiters may still be asleep. Thus the following cv_destroy()
will free memory which may still be in use.
This was fixed by updating cv_destroy() to block on waiters but
this in turn introduced a deadlock. The deadlock was resolved
with the use of a taskq to move the offending free outside the
range lock. This worked well but using the taskq for the free
resulted in a serious performace hit. This is somewhat ironic
because at the time I felt using the taskq might improve things
by making the free asynchronous.
This patch refines the original fix and moves the free from the
taskq to a private free list. Then items which must be free'd
are simply inserted in to the list. When the range lock is dropped
it's safe to free the items. The list is walked and all rl_t
entries are freed.
This change improves small cached read performance by 26x. This
was expected because for small reads the number of locking calls
goes up significantly. More surprisingly this change significantly
improves large cache read performance. This probably attributable
to better cpu/memory locality. Very likely the same processor
which allocated the memory is now freeing it.
bs ext3 zfs zfs+fix faster
----------------------------------------------
512 435 3 79 26x
1k 820 7 160 22x
2k 1536 14 305 21x
4k 2764 28 572 20x
8k 3788 50 1024 20x
16k 4300 86 1843 21x
32k 4505 138 2560 18x
64k 5324 252 3891 15x
128k 5427 276 4710 17x
256k 5427 413 5017 12x
512k 5427 497 5324 10x
1m 5427 521 5632 10x
Closes#142
With the recent SPL change (d599e4fa) that forces cv_destroy()
to block until all waiters have been woken. It is now unsafe
to call cv_destroy() under the zp->z_range_lock() because it
is used as the condition variable mutex. If there are waiters
cv_destroy() will block until they wake up and aquire the mutex.
However, they will never aquire the mutex because cv_destroy()
will not return allowing it's caller to drop the lock. Deadlock.
To avoid this cv_destroy() is now run asynchronously in a taskq.
This solves two problems:
1) It is no longer run under the zp->z_range_lock so no deadlock.
2) Since cv_destroy() may now block we don't want this slowing
down zfs_range_unlock() and throttling the system.
This was not as much of an issue under OpenSolaris because their
cv_destroy() implementation does not do anything. They do however
risk a bad paging request if cv_destroy() returns, the memory holding
the condition variable is free'd, and then the waiters wake up and
try to reference it. It's a very small unlikely race, but it is
possible.
A new flag is required for the zfs_rlock code to determine if
it is operation of the zvol of zpl dataset. This used to be
keyed off the zp->z_vnode, which was a hack to begin with, but
with the removal of vnodes we needed a dedicated flag.
I appologize in advance why to many things ended up in this commit.
When it could be seperated in to a whole series of commits teasing
that all apart now would take considerable time and I'm not sure
there's much merrit in it. As such I'll just summerize the intent
of the changes which are all (or partly) in this commit. Broadly
the intent is to remove as much Solaris specific code as possible
and replace it with native Linux equivilants. More specifically:
1) Replace all instances of zfsvfs_t with zfs_sb_t. While the
type is largely the same calling it private super block data
rather than a zfsvfs is more consistent with how Linux names
this. While non critical it makes the code easier to read when
your thinking in Linux friendly VFS terms.
2) Replace vnode_t with struct inode. The Linux VFS doesn't have
the notion of a vnode and there's absolutely no good reason to
create one. There are in fact several good reasons to remove it.
It just adds overhead on Linux if we were to manage one, it
conplicates the code, and it likely will lead to bugs so there's
a good change it will be out of date. The code has been updated
to remove all need for this type.
3) Replace all vtype_t's with umode types. Along with this shift
all uses of types to mode bits. The Solaris code would pass a
vtype which is redundant with the Linux mode. Just update all the
code to use the Linux mode macros and remove this redundancy.
4) Remove using of vn_* helpers and replace where needed with
inode helpers. The big example here is creating iput_aync to
replace vn_rele_async. Other vn helpers will be addressed as
needed but they should be be emulated. They are a Solaris VFS'ism
and should simply be replaced with Linux equivilants.
5) Update znode alloc/free code. Under Linux it's common to
embed the inode specific data with the inode itself. This removes
the need for an extra memory allocation. In zfs this information
is called a znode and it now embeds the inode with it. Allocators
have been updated accordingly.
6) Minimal integration with the vfs flags for setting up the
super block and handling mount options has been added this
code will need to be refined but functionally it's all there.
This will be the first and last of these to large to review commits.
