For ERESTARTSYS/EIO/EROFS/ENOSPC/E2BIG in layoutget, we
should just bail out instead of hiding the error and
retrying inband IO.

Change all the call sites to pop the error all the way up.
Signed-off-by: NPeng Tao <tao.peng@primarydata.com>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

When lseg's commit_through_mds is set, pnfs client always WARN once
in nfs_direct_select_verf after checking ds_cinfo.nbuckets.

nfs should use the DS verf except commit_through_mds is set for
layout segment where nbuckets is zero.

[17844.666094] ------------[ cut here ]------------
[17844.667071] WARNING: CPU: 0 PID: 21758 at /root/source/linux-pnfs/fs/nfs/direct.c:174 nfs_direct_select_verf+0x5a/0x70 [nfs]()
[17844.668650] Modules linked in: nfs_layout_nfsv41_files(OE) nfsv4(OE) nfs(OE) fscache(E) nfsd(OE) xfs libcrc32c btrfs ppdev coretemp crct10dif_pclmul auth_rpcgss crc32_pclmul crc32c_intel nfs_acl ghash_clmulni_intel lockd vmw_balloon xor vmw_vmci grace raid6_pq shpchp sunrpc parport_pc i2c_piix4 parport vmwgfx drm_kms_helper ttm drm serio_raw mptspi e1000 scsi_transport_spi mptscsih mptbase ata_generic pata_acpi [last unloaded: fscache]
[17844.686676] CPU: 0 PID: 21758 Comm: kworker/0:1 Tainted: G        W  OE   4.3.0-rc1-pnfs+ #245
[17844.687352] Hardware name: VMware, Inc. VMware Virtual Platform/440BX Desktop Reference Platform, BIOS 6.00 05/20/2014
[17844.698502] Workqueue: nfsiod rpc_async_release [sunrpc]
[17844.699212]  0000000000000009 0000000043e58010 ffff8800454fbc10 ffffffff813680c4
[17844.699990]  ffff8800454fbc48 ffffffff8108b49d ffff88004eb20000 ffff88004eb20000
[17844.700844]  ffff880062e26000 0000000000000000 0000000000000001 ffff8800454fbc58
[17844.701637] Call Trace:
[17844.725252]  [<ffffffff813680c4>] dump_stack+0x19/0x25
[17844.732693]  [<ffffffff8108b49d>] warn_slowpath_common+0x7d/0xb0
[17844.733855]  [<ffffffff8108b5da>] warn_slowpath_null+0x1a/0x20
[17844.735015]  [<ffffffffa04a27ca>] nfs_direct_select_verf+0x5a/0x70 [nfs]
[17844.735999]  [<ffffffffa04a2b83>] nfs_direct_set_hdr_verf+0x23/0x90 [nfs]
[17844.736846]  [<ffffffffa04a2e17>] nfs_direct_write_completion+0x227/0x260 [nfs]
[17844.737782]  [<ffffffffa04a433c>] nfs_pgio_release+0x1c/0x20 [nfs]
[17844.738597]  [<ffffffffa0502df3>] pnfs_generic_rw_release+0x23/0x30 [nfsv4]
[17844.739486]  [<ffffffffa01cbbea>] rpc_free_task+0x2a/0x70 [sunrpc]
[17844.740326]  [<ffffffffa01cbcd5>] rpc_async_release+0x15/0x20 [sunrpc]
[17844.741173]  [<ffffffff810a387c>] process_one_work+0x21c/0x4c0
[17844.741984]  [<ffffffff810a37cd>] ? process_one_work+0x16d/0x4c0
[17844.742837]  [<ffffffff810a3b6a>] worker_thread+0x4a/0x440
[17844.743639]  [<ffffffff810a3b20>] ? process_one_work+0x4c0/0x4c0
[17844.744399]  [<ffffffff810a3b20>] ? process_one_work+0x4c0/0x4c0
[17844.745176]  [<ffffffff810a8d75>] kthread+0xf5/0x110
[17844.745927]  [<ffffffff810a8c80>] ? kthread_create_on_node+0x240/0x240
[17844.747105]  [<ffffffff8172ce1f>] ret_from_fork+0x3f/0x70
[17844.747856]  [<ffffffff810a8c80>] ? kthread_create_on_node+0x240/0x240
[17844.748642] ---[ end trace 336a2845d42b83f0 ]---
Signed-off-by: NKinglong Mee <kinglongmee@gmail.com>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

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:

clat percentiles (usec):
 |  1.00th=[   33],  5.00th=[   34], 10.00th=[   34], 20.00th=[   34],
 | 30.00th=[   34], 40.00th=[   34], 50.00th=[   35], 60.00th=[   35],
 | 70.00th=[   35], 80.00th=[   35], 90.00th=[   37], 95.00th=[   80],
 | 99.00th=[   98], 99.50th=[  151], 99.90th=[  155], 99.95th=[  155],
 | 99.99th=[  165]

After:

clat percentiles (usec):
 |  1.00th=[   95],  5.00th=[  108], 10.00th=[  129], 20.00th=[  149],
 | 30.00th=[  155], 40.00th=[  161], 50.00th=[  167], 60.00th=[  171],
 | 70.00th=[  177], 80.00th=[  185], 90.00th=[  201], 95.00th=[  270],
 | 99.00th=[  390], 99.50th=[  398], 99.90th=[  418], 99.95th=[  422],
 | 99.99th=[  438]

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: NJens Axboe <axboe@fb.com>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

For flexfiles driver, we might choose to read from mirror index other
than 0 while mirror_count is always 1 for read.
Reported-by: NJean Spector <jean@primarydata.com>
Cc: <stable@vger.kernel.org> # v3.19+
Cc: Weston Andros Adamson <dros@primarydata.com>
Signed-off-by: NPeng Tao <tao.peng@primarydata.com>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

For direct read that has IO size larger than rsize, we'll split
it into several READ requests and nfs_direct_good_bytes() would
count completed bytes incorrectly by eating last zero count reply.

Fix it by handling mirror and non-mirror cases differently such that
we only count mirrored writes differently.

This fixes 5fadeb47("nfs: count DIO good bytes correctly with mirroring").
Reported-by: NJean Spector <jean@primarydata.com>
Cc: <stable@vger.kernel.org> # v3.19+
Signed-off-by: NPeng Tao <tao.peng@primarydata.com>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

that's the bulk of filesystem drivers dealing with inodes of their own
Signed-off-by: NDavid Howells <dhowells@redhat.com>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

... 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: NAl Viro <viro@zeniv.linux.org.uk>

Conflicts:
	fs/ext4/file.c

all remaining callers are passing 0; some just obscure that fact.
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Now that no one is using rw, remove it completely.
Signed-off-by: NOmar Sandoval <osandov@osandov.com>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

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: NOmar Sandoval <osandov@osandov.com>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

If the caller does not specify the O_SYNC flag, then it is legitimate
to return from O_DIRECT without doing a pNFS layoutcommit operation.
However if the file is opened O_DIRECT|O_SYNC then we'd better get it
right.
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

Most callers in the kernel want to perform synchronous file I/O, but
still have to bloat the stack with a full struct kiocb.  Split out
the parts needed in filesystem code from those in the aio code, and
only allocate those needed to pass down argument on the stack.  The
aio code embedds the generic iocb in the one it allocates and can
easily get back to it by using container_of.

Also add a ->ki_complete method to struct kiocb, this is used to call
into the aio code and thus removes the dependency on aio for filesystems
impementing asynchronous operations.  It will also allow other callers
to substitute their own completion callback.

We also add a new ->ki_flags field to work around the nasty layering
violation recently introduced in commit 5e33f6 ("usb: gadget: ffs: add
eventfd notification about ffs events").
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

This follows up "nfs: fix dio deadlock when O_DIRECT flag is flipped"
and removes the unnecessary CONFIG_NFS_SWAP switch.
Signed-off-by: NPeng Tao <tao.peng@primarydata.com>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

There is no need to pass the total request length in the kiocb, as
we already get passed in through the iov_iter argument.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Commit 411a99ad (nfs: clear_request_commit while holding i_lock)
assumes that the nfs_commit_info always points to the inode->i_lock.
For historical reasons, that is not the case for O_DIRECT writes.

Cc: Weston Andros Adamson <dros@primarydata.com>
Fixes: 411a99ad ("nfs: clear_request_commit while holding i_lock")
Cc: stable@vger.kernel.org # 3.17.x
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

When resending to MDS, we might resend multiple mirroring
requests to MDS. As a result, nfs_direct_good_bytes() ends
up counting bytes multiple times, causing application to
get wrong return results in read/write syscalls.

Fix it by tracking start of a dreq and checking the range of
pgio header.

Cc: Weston Andros Adamson <dros@primarydata.com>
Signed-off-by: NPeng Tao <tao.peng@primarydata.com>

To allow pnfs LD to ask direct writes to be resend.
Signed-off-by: NPeng Tao <tao.peng@primarydata.com>

This skips the WARN_ON_ONCE, but doesnt change behavior (the memcmp would
fail).
Signed-off-by: NWeston Andros Adamson <dros@primarydata.com>
Signed-off-by: NTom Haynes <Thomas.Haynes@primarydata.com>

The current mirroring code only notices short writes to the first
mirror. This patch keeps per-mirror byte counts and only considers
a byte to be written once all mirrors report so.
Signed-off-by: NWeston Andros Adamson <dros@primarydata.com>

This patch adds mirrored write support to the pgio layer. The default
is to use one mirror, but pgio callers may define callbacks to change
this to any value up to the (arbitrarily selected) limit of 16.

The basic idea is to break out members of nfs_pageio_descriptor that cannot
be shared between mirrored DSes and put them in a new structure.
Signed-off-by: NWeston Andros Adamson <dros@primarydata.com>

Pass ds_commit_idx through the nfs commit path. It's used to select
the commit bucket when using pnfs and is ignored when not using pnfs.
Several functions had to be changed: nfs_retry_commit,
nfs_mark_request_commit, pnfs_mark_request_commit and the pnfs layout
driver .mark_request_commit functions.
Signed-off-by: NTom Haynes <loghyr@primarydata.com>

'ds_commit_idx' is a better name - it is used to select the right
commit bucket for pnfs.
Signed-off-by: NWeston Andros Adamson <dros@primarydata.com>

Acked-by: NJeff Layton <jlayton@primarydata.com>
Signed-off-by: NTom Haynes <loghyr@primarydata.com>

We only support swap file calling nfs_direct_IO. However, application
might be able to get to nfs_direct_IO if it toggles O_DIRECT flag
during IO and it can deadlock because we grab inode->i_mutex in
nfs_file_direct_write(). So return 0 for such case. Then the generic
layer will fall back to buffer IO.
Signed-off-by: NPeng Tao <tao.peng@primarydata.com>
Cc: stable@vger.kernel.org
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

For pNFS direct writes, layout driver may dynamically allocate ds_cinfo.buckets.
So we need to take care to free them when freeing dreq.

Ideally this needs to be done inside layout driver where ds_cinfo.buckets
are allocated. But buckets are attached to dreq and reused across LD IO iterations.
So I feel it's OK to free them in the generic layer.

Cc: stable@vger.kernel.org [v3.4+]
Signed-off-by: NPeng Tao <tao.peng@primarydata.com>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

REQ_KERNEL is no longer used. Remove it and drop the redundant uio
argument to nfs_file_direct_{read,write}.
Signed-off-by: NMartin K. Petersen <martin.petersen@oracle.com>
Cc: Christoph Hellwig <hch@infradead.org>
Reported-by: NDan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: NJens Axboe <axboe@fb.com>

The goal is to create a generic NFS module with code that does not
depend on what versions of NFS are enabled.
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

Cc: Weston Andros Adamson <dros@primarydata.com>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

Remove duplicate writeverf structure from merge of nfs_pgio_header and
nfs_pgio_data and remove writeverf related flags and logic to handle
more than one RPC per nfs_pgio_header.
Signed-off-by: NWeston Andros Adamson <dros@primarydata.com>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

struct nfs_pgio_data only exists as a member of nfs_pgio_header, but is
passed around everywhere, because there used to be multiple _data structs
per _header. Many of these functions then use the _data to find a pointer
to the _header.  This patch cleans this up by merging the nfs_pgio_data
structure into nfs_pgio_header and passing nfs_pgio_header around instead.
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NWeston Andros Adamson <dros@primarydata.com>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

nfs_rw_header was used to allocate an nfs_pgio_header along with an
nfs_pgio_data, because a _header would need at least one _data.

Now there is only ever one nfs_pgio_data for each nfs_pgio_header -- move
it to nfs_pgio_header and get rid of nfs_rw_header.
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NWeston Andros Adamson <dros@primarydata.com>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

Support direct requests that span multiple pnfs data servers by
comparing nfs_pgio_header->verf to a cached verf in pnfs_commit_bucket.
Continue to use dreq->verf if the MDS is used / non-pNFS.
Signed-off-by: NWeston Andros Adamson <dros@primarydata.com>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

Add "page groups" - a circular list of nfs requests (struct nfs_page)
that all reference the same page. This gives nfs read and write paths
the ability to account for sub-page regions independently.  This
somewhat follows the design of struct buffer_head's sub-page
accounting.

Only "head" requests are ever added/removed from the inode list in
the buffered write path. "head" and "sub" requests are treated the
same through the read path and the rest of the write/commit path.
Requests are given an extra reference across the life of the list.

Page groups are never rejoined after being split. If the read/write
request fails and the client falls back to another path (ie revert
to MDS in PNFS case), the already split requests are pushed through
the recoalescing code again, which may split them further and then
coalesce them into properly sized requests on the wire. Fragmentation
shouldn't be a problem with the current design, because we flush all
requests in page group when a non-contiguous request is added, so
the only time resplitting should occur is on a resend of a read or
write.

This patch lays the groundwork for sub-page splitting, but does not
actually do any splitting. For now all page groups have one request
as pg_test functions don't yet split pages. There are several related
patches that are needed support multiple requests per page group.
Signed-off-by: NWeston Andros Adamson <dros@primarydata.com>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

@inode is passed but not used.
Signed-off-by: NWeston Andros Adamson <dros@primarydata.com>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

The header had a pointer to the verifier that was set from the old write
data struct.  We don't need to keep the pointer around now that we have
shared structures.
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

The read_pageio_init method is just a very convoluted way to grab the
right nfs_pageio_ops vector.  The vector to chose is not a choice of
protocol version, but just a pNFS vs MDS I/O choice that can simply be
done inside nfs_pageio_init_read based on the presence of a layout
driver, and a new force_mds flag to the special case of falling back
to MDS I/O on a pNFS-capable volume.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Tested-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

The write_pageio_init method is just a very convoluted way to grab the
right nfs_pageio_ops vector.  The vector to chose is not a choice of
protocol version, but just a pNFS vs MDS I/O choice that can simply be
done inside nfs_pageio_init_write based on the presence of a layout
driver, and a new force_mds flag to the special case of falling back
to MDS I/O on a pNFS-capable volume.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Tested-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>

same as iov_iter_get_pages(), except that pages array is allocated
(kmalloc if possible, vmalloc if that fails) and left for caller to
free.  Lustre and NFS ->direct_IO() switched to it.
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

all callers have iov_length(iter->iov, iter->nr_segs) == iov_iter_count(iter)
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>