direct-io: only inc/dec inode->i_dio_count for file systems
do_blockdev_direct_IO() increments and decrements the inode
->i_dio_count for each IO operation. It does this to protect against
truncate of a file. Block devices don't need this sort of protection.
For a capable multiqueue setup, this atomic int is the only shared
state between applications accessing the device for O_DIRECT, and it
presents a scaling wall for that. In my testing, as much as 30% of
system time is spent incrementing and decrementing this value. A mixed
read/write workload improved from ~2.5M IOPS to ~9.6M IOPS, with
better latencies too. Before:
In other setups, Robert Elliott reported seeing good performance
improvements:
https://lkml.org/lkml/2015/4/3/557
The more applications accessing the device, the worse it gets.
Add a new direct-io flags, DIO_SKIP_DIO_COUNT, which tells
do_blockdev_direct_IO() that it need not worry about incrementing
or decrementing the inode i_dio_count for this caller.
Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Christoph Hellwig <hch@lst.de> Cc: Theodore Ts'o <tytso@mit.edu> Cc: Elliott, Robert (Server Storage) <elliott@hp.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Jens Axboe <axboe@fb.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Johannes Berg [Wed, 22 Apr 2015 09:55:14 +0000 (11:55 +0200)]
fs/9p: fix readdir()
Al Viro's IOV changes broke 9p readdir() because the new code
didn't abort the read when it returned nothing. The original
code checked if the combined error/length was <= 0 but in the
new code that accidentally got changed to just an error check.
Add back the return from the function when nothing is read.
Cc: Al Viro <viro@zeniv.linux.org.uk> Fixes: e1200fe68f20 ("9p: switch p9_client_read() to passing struct iov_iter *") Signed-off-by: Johannes Berg <johannes.berg@intel.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
David Howells [Mon, 2 Mar 2015 16:40:32 +0000 (16:40 +0000)]
configfs: Fix inconsistent use of file_inode() vs file->f_path.dentry->d_inode
Fix inconsistent use of file_inode() vs file->f_path.dentry->d_inode.
Reported-by: Dan Carpenter <dan.carpenter@oracle.com> Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
David Howells [Tue, 17 Mar 2015 22:16:40 +0000 (22:16 +0000)]
VFS: Make pathwalk use d_is_reg() rather than S_ISREG()
Make pathwalk use d_is_reg() rather than S_ISREG() to determine whether to
honour O_TRUNC. Since this occurs after complete_walk(), the dentry type
field cannot change and the inode pointer cannot change as we hold a ref on
the dentry, so this should be safe.
Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
David Howells [Tue, 17 Mar 2015 17:33:52 +0000 (17:33 +0000)]
VFS: Combine inode checks with d_is_negative() and d_is_positive() in pathwalk
Where we have:
if (!dentry->d_inode || d_is_negative(dentry)) {
type constructions in pathwalk we should be able to eliminate the check of
d_inode and rely solely on the result of d_is_negative() or d_is_positive().
What we do have to take care to do is to read d_inode after calling a
d_is_xxx() typecheck function to get the barriering right.
Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
David Howells [Thu, 5 Mar 2015 14:09:22 +0000 (14:09 +0000)]
VFS: Impose ordering on accesses of d_inode and d_flags
Impose ordering on accesses of d_inode and d_flags to avoid the need to do
this:
if (!dentry->d_inode || d_is_negative(dentry)) {
when this:
if (d_is_negative(dentry)) {
should suffice.
This check is especially problematic if a dentry can have its type field set
to something other than DENTRY_MISS_TYPE when d_inode is NULL (as in
unionmount).
What we really need to do is stick a write barrier between setting d_inode and
setting d_flags and a read barrier between reading d_flags and reading
d_inode.
Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Supply two functions to test whether a filesystem's own dentries are positive
or negative (d_really_is_positive() and d_really_is_negative()).
The problem is that the DCACHE_ENTRY_TYPE field of dentry->d_flags may be
overridden by the union part of a layered filesystem and isn't thus
necessarily indicative of the type of dentry.
Normally, this would involve a negative dentry (ie. ->d_inode == NULL) having
->d_layer.lower pointed to a lower layer dentry, DCACHE_PINNING_LOWER set and
the DCACHE_ENTRY_TYPE field set to something other than DCACHE_MISS_TYPE - but
it could also involve, say, a DCACHE_SPECIAL_TYPE being overridden to
DCACHE_WHITEOUT_TYPE if a 0,0 chardev is detected in the top layer.
However, inside a filesystem, when that fs is looking at its own dentries, it
probably wants to know if they are really negative or not - and doesn't care
about the fallthrough bits used by the union.
To this end, a filesystem should normally use d_really_is_positive/negative()
when looking at its own dentries rather than d_is_positive/negative() and
should use d_inode() to get at the inode.
Anyone looking at someone else's dentries (this includes pathwalk) should use
d_is_xxx() and d_backing_inode().
Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Al Viro [Thu, 9 Apr 2015 16:55:47 +0000 (12:55 -0400)]
switch generic_write_checks() to iocb and iter
... returning -E... upon error and amount of data left in iter after
(possible) truncation upon success. Note, that normal case gives
a non-zero (positive) return value, so any tests for != 0 _must_ be
updated.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Conflicts:
fs/ext4/file.c
Al Viro [Thu, 9 Apr 2015 15:14:45 +0000 (11:14 -0400)]
ocfs2: move generic_write_checks() before the alignment checks
Alignment checks for dio depend upon the range truncation done by
generic_write_checks(). They can be done as soon as we got ocfs2_rw_lock()
and that actually makes ocfs2_prepare_inode_for_write() simpler.
The only thing to watch out for is restoring the original count
in "unlock and redo without dio" case. Position doesn't need to be
restored, since we change it only in O_APPEND case and in that case it
will be reassigned anyway.
Al Viro [Tue, 7 Apr 2015 19:06:19 +0000 (15:06 -0400)]
fuse: ->direct_IO() doesn't need generic_write_checks()
already done by caller. We used to call __fuse_direct_write(), which
called generic_write_checks(); now the former got expanded, bringing
the latter to the surface. It used to be called all along and calling
it from there had been wrong all along...
Al Viro [Tue, 7 Apr 2015 14:22:53 +0000 (10:22 -0400)]
__generic_file_write_iter: keep ->ki_pos and return value consistent
A side effect worth noting: in O_APPEND case we set ->ki_pos early,
so if it turns out to be an error or a zero-length write, we'll
end up with ->ki_pos modified. Safe, since all callers never
look at the ->ki_pos after the call of __generic_file_write_iter()
returning non-positive, all the way to caller of ->write_iter() and
those discard ->ki_pos when getting that.
Al Viro [Tue, 7 Apr 2015 00:50:38 +0000 (20:50 -0400)]
new_sync_write(): discard ->ki_pos unless the return value is positive
That allows ->write_iter() instances much more convenient life wrt
iocb->ki_pos (and fixes several filesystems with borderline POSIX
violations when zero-length write succeeds and changes the current
position).
Omar Sandoval [Mon, 16 Mar 2015 11:33:52 +0000 (04:33 -0700)]
direct_IO: use iov_iter_rw() instead of rw everywhere
The rw parameter to direct_IO is redundant with iov_iter->type, and
treated slightly differently just about everywhere it's used: some users
do rw & WRITE, and others do rw == WRITE where they should be doing a
bitwise check. Simplify this with the new iov_iter_rw() helper, which
always returns either READ or WRITE.
Signed-off-by: Omar Sandoval <osandov@osandov.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Al Viro [Sat, 4 Apr 2015 04:19:32 +0000 (00:19 -0400)]
pcm: another weird API abuse
readv() and writev() should _not_ ignore all but the first ->iov_len,
among other things. Really weird abuse of those syscalls - it
expects a vector element per channel, with identical lengths (it
actually assumes them to be identical - no checking is done).
readv() and writev() are really bad match for that. Unfortunately,
userland API is userland API and we can't do anything about them.
Converted to ->read_iter/->write_iter. Please, _please_ don't do
anything of that kind when designing new interfaces.
Al Viro [Sat, 4 Apr 2015 04:11:32 +0000 (00:11 -0400)]
infinibad: weird APIs switched to ->write_iter()
Things Not To Do When Writing A Driver, part 1001st:
have writev() and write() on the same file doing completely
different things. As in, "interpret very different sets of
commands".
We _can_ handle that, but it's a bloody bad idea.
Don't do that in new drivers. Ever.
Al Viro [Sat, 4 Apr 2015 02:10:20 +0000 (22:10 -0400)]
kill do_sync_read/do_sync_write
all remaining instances of aio_{read,write} (all 4 of them) have explicit
->read and ->write resp.; do_sync_read/do_sync_write is never called by
__vfs_read/__vfs_write anymore and no other users had been left.
Al Viro [Fri, 3 Apr 2015 19:57:04 +0000 (15:57 -0400)]
switch drivers/char/mem.c to ->read_iter/->write_iter
Note that _these_ guys have ->read() and ->write() left in place - they are
eqiuvalent to what we'd get if we replaced those with NULL, but we are
talking about hot paths here.
Al Viro [Fri, 3 Apr 2015 19:41:18 +0000 (15:41 -0400)]
make new_sync_{read,write}() static
All places outside of core VFS that checked ->read and ->write for being NULL or
called the methods directly are gone now, so NULL {read,write} with non-NULL
{read,write}_iter will do the right thing in all cases.
Al Viro [Fri, 3 Apr 2015 01:47:49 +0000 (21:47 -0400)]
p9_client_attach(): set fid->uid correctly
it's almost always equal to current_fsuid(), but there's an exception -
if the first writeback fid is opened by non-root *and* that happens before
root has done any lookups in /, we end up doing attach for root. The
current code leaves the resulting FID owned by root from the server POV
and by non-root from the client one. Unfortunately, it means that e.g.
massive dcache eviction will leave that user buggered - they'll end
up redoing walks from / *and* picking that FID every time. As soon as
they try to create something, the things will get nasty.
Al Viro [Tue, 31 Mar 2015 15:54:59 +0000 (11:54 -0400)]
aio_run_iocb(): kill dead check
We check if ->ki_pos is positive. However, by that point we have
already done rw_verify_area(), which would have rejected such
unless the file had been one of /dev/mem, /dev/kmem and /proc/kcore.
All of which do not have vectored rw methods, so we would've bailed
out even earlier.
This check had been introduced before rw_verify_area() had been added there
- in fact, it was a subset of checks done on sync paths by rw_verify_area()
(back then the /dev/mem exception didn't exist at all). The rest of checks
(mandatory locking, etc.) hadn't been added until later. Unfortunately,
by the time the call of rw_verify_area() got added, the /dev/mem exception
had already appeared, so it wasn't obvious that the older explicit check
downstream had become dead code. It *is* a dead code, though, since the few
files for which the exception applies do not have ->aio_{read,write}() or
->{read,write}_iter() and for them we won't reach that check anyway.
What's more, even if we ever introduce vectored methods for /dev/mem
and friends, they'll have to cope with negative positions anyway, since
readv(2) and writev(2) are using the same checks as read(2) and write(2) -
i.e. rw_verify_area().
Al Viro [Tue, 31 Mar 2015 15:43:52 +0000 (11:43 -0400)]
ioctx_alloc(): remove pointless check
Way, way back kiocb used to be picked from arrays, so ioctx_alloc()
checked for multiplication overflow when calculating the size of
such array. By the time fs/aio.c went into the tree (in 2002) they
were already allocated one-by-one by kmem_cache_alloc(), so that
check had already become pointless. Let's bury it...
Al Viro [Sun, 22 Mar 2015 00:08:18 +0000 (20:08 -0400)]
sg_start_req(): make sure that there's not too many elements in iovec
unfortunately, allowing an arbitrary 16bit value means a possibility of
overflow in the calculation of total number of pages in bio_map_user_iov() -
we rely on there being no more than PAGE_SIZE members of sum in the
first loop there. If that sum wraps around, we end up allocating
too small array of pointers to pages and it's easy to overflow it in
the second loop.
X-Coverup: TINC (and there's no lumber cartel either) Cc: stable@vger.kernel.org # way, way back Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
projects / kernel.git / log