15 Oct, 2012
3 commits
-
Don't circumvent the array size checks.
Signed-off-by: Trond Myklebust
-
Move the call to pnfs_return_layout() to the read and write rpc_release()
callbacks, so that it gets called from nfsiod, which is a more appropriate
context.Signed-off-by: Trond Myklebust
-
There is nothing to prevent another thread from dereferencing ds->ds_clp
during or after the call to nfs4_ds_disconnect(), and Oopsing due to the
resulting NULL pointer.Instead, we should just rely on filelayout_mark_devid_invalid() to keep
us out of trouble by avoiding that deviceid.Signed-off-by: Trond Myklebust
29 Sep, 2012
5 commits
-
We already have a mechanism for blocking LAYOUTGET by means of the
plh_block_lgets counter. The only "service" that NFS_LAYOUT_DESTROYED
provides at this point is to block layoutget once the layout segment
list is empty, which basically means that you have to wait until
the pnfs_layout_hdr is destroyed before you can do pNFS on that file
again.This patch enables the reuse of the pnfs_layout_hdr if the layout
segment list is empty.Signed-off-by: Trond Myklebust
-
In all cases where we set NFS_LAYOUT_INVALID, we also set NFS_LAYOUT_DESTROYED.
Furthermore, in all cases where we test for NFS_LAYOUT_INVALID, we should
also be testing for NFS_LAYOUT_DESTROYED, since the latter means that
we hold no valid layout segments.
Ergo the two are redundant.Signed-off-by: Trond Myklebust
-
In cases where the pNFS data server is just temporarily out of service,
we want to mark it as such, and then try again later. Typically that will
be in cases of network connection errors etc.
This patch allows us to mark the devices as being "unavailable" for such
transient errors, and will make them available for retries after a
2 minute timeout period.Signed-off-by: Trond Myklebust
-
Fix the namespace pollution issue.
Signed-off-by: Trond Myklebust
-
Signed-off-by: Trond Myklebust
17 Jul, 2012
2 commits
-
Signed-off-by: Andy Adamson
Signed-off-by: Trond Myklebust -
First mark the deviceid invalid to prevent any future use. Then fence all
files involved in I/O to a DS with a connection error by sending a
LAYOUTRETURN.Signed-off-by: Andy Adamson
Signed-off-by: Trond Myklebust
29 Jun, 2012
1 commit
-
The 'committed' field is not needed once we have put the struct nfs_page
on the right list.Also correct the type of the verifier: it is not an array of __be32, but
simply an 8 byte long opaque array.Signed-off-by: Trond Myklebust
28 May, 2012
1 commit
-
Let nfs4_schedule_session_recovery() handle the details of choosing
between resetting the session, and other session related recovery.Signed-off-by: Trond Myklebust
23 May, 2012
5 commits
-
The "struct inode *inode" was only used in a dprintk, so compiling with
CONFIG_SUNRPC_DEBUG off triggers a warning. To get around this, I
remove the "struct inode *inode" variable and instead change the
dprintk()s to use hdr->inode instead.Signed-off-by: Bryan Schumaker
Signed-off-by: Trond Myklebust -
We reset all I/O on a disconnected data server through the pgio layer indicated
by the NFS_IOHDR_REDO flag.Differentiate between on-the-wire tasks returning with an error which must
call rpc_call_done and tasks woken from the data server slot_table_waitq
waiting for a session slot with a status of zero which call rpc_exit in
rpc_prepare and need to skip rpc_call_done.Signed-off-by: Andy Adamson
Signed-off-by: Trond Myklebust -
Signed-off-by: Andy Adamson
Signed-off-by: Trond Myklebust -
Signed-off-by: Andy Adamson
Signed-off-by: Trond Myklebust -
filelayout_scan_commit_lists needs to bump the reference count on
the struct nfs_page just like nfs_scan_commit_list().Reported-by: Andy Adamson
Signed-off-by: Trond Myklebust
20 May, 2012
10 commits
-
The "invalid layout" class of errors is handled by destroying the layout and
getting a new layout from the server. Currently, the layout must be
destroyed before a new layout can be obtained.This means that all references (e.g.lsegs) to the "to be destroyed" layout
header must be dropped before it can be destroyed. This in turn means waiting
for all in flight RPC's using the old layout as well as draining the data
server session slot table wait queue.Set the NFS_LAYOUT_INVALID flag to redirect I/O to the MDS while waiting for
the old layout to be destroyed.Signed-off-by: Andy Adamson
Signed-off-by: Trond Myklebust -
When the last DS io is processed, the data server client record will be
freed.Signed-off-by: Andy Adamson
Signed-off-by: Trond Myklebust -
Prepare to put a dis-connected DS client record.
Signed-off-by: Andy Adamson
Signed-off-by: Trond Myklebust -
Let the MDS know that you are redirecting I/O from pNFS to MDS.
Signed-off-by: Andy Adamson
Signed-off-by: Trond Myklebust -
The DS has a connection error (invalid deviceid). Drain the fore channel
slot table waitq.Signed-off-by: Andy Adamson
Signed-off-by: Trond Myklebust -
Tasks sleeping on the slot table waitq wake to the rpc_prepare_task state.
Reset the task for io through the MDS if the deviceid is invalid.The reset functions put the io pages through the pageio layer which has the
advantage of re-coalescing which allows for the MDS and DS having different
r/wsizes. Exit the awakened task without executing the rpc_call_done routine.Signed-off-by: Andy Adamson
Signed-off-by: Trond Myklebust -
This prevents the use of any layout for i/o that references the deviceid.
I/O is redirected through the MDS.Redirect the unhandled failed I/O to the MDS without marking either the
layout or the deviceid invalid.Signed-off-by: Andy Adamson
Signed-off-by: Trond Myklebust -
RPC_TASK_SOFTCONN returns connection errors to the caller which allows the pNFS
file layout to quickly try the MDS or perhaps another DS.Signed-off-by: Andy Adamson
Signed-off-by: Trond Myklebust -
The invalid layout bits are should only be used to block LAYOUTGETs.
Do not invalidate a layout on deviceid invalidation.
Do not invalidate a layout on un-handled READ, WRITE, COMMIT errors.Signed-off-by: Andy Adamson
Signed-off-by: Trond Myklebust -
Move the invalid deviceid test into nfs4_fl_prepare_ds, called by the
filelayout read, write, and commit routines. NFS4_DEVICE_ID_NEG_ENTRY
is no longer needed.
Remove redundant printk's - filelayout_mark_devid_invalid prints a KERN_WARNING.An invalid device prevents pNFS io.
Signed-off-by: Andy Adamson
Signed-off-by: Trond Myklebust
28 Apr, 2012
13 commits
-
This is a bugfix that applies on top of the previous directio patches,
that fixes a bug introduced in "NFS: create struct nfs_commit_info".Signed-off-by: Fred Isaman
Signed-off-by: Trond Myklebust -
This also has the advantage that it allows directio to use pnfs.
Signed-off-by: Fred Isaman
Signed-off-by: Trond Myklebust -
Factors out the code that needs to change when directio
starts using these code paths.Signed-off-by: Fred Isaman
Signed-off-by: Trond Myklebust -
It is COMMIT that is handled the most differently between
the paged and direct paths. Create a structure that encapsulates
everything either path needs to know about the commit state.We could use void to hide some of the layout driver stuff, but
Trond suggests pulling it out to ensure type checking, given the
huge changes being made, and the fact that it doesn't interfere
with other drivers.Signed-off-by: Fred Isaman
Signed-off-by: Trond Myklebust -
The coalesce code made assumptions that will no longer be true once
non-page aligned io occurs. This introduces no change in
current behavior, but allows for more general situations to come.Signed-off-by: Fred Isaman
Signed-off-by: Trond Myklebust -
Decouple nfs_pgio_header and nfs_write_data, and have (possibly
multiple) nfs_write_datas each take a refcount on nfs_pgio_header.For the moment keeps nfs_write_header as a way to preallocate a single
nfs_write_data with the nfs_pgio_header. The code doesn't need this,
and would be prettier without, but given the amount of churn I am
already introducing I didn't want to play with tuning new mempools.This also fixes bug in pnfs_ld_handle_write_error. In the case of
desc->pg_bsize < PAGE_CACHE_SIZE, the pages list was empty, causing
replay attempt to do nothing.Signed-off-by: Fred Isaman
Signed-off-by: Trond Myklebust -
Decouple nfs_pgio_header and nfs_read_data, and have (possibly
multiple) nfs_read_datas each take a refcount on nfs_pgio_header.For the moment keeps nfs_read_header as a way to preallocate a single
nfs_read_data with the nfs_pgio_header. The code doesn't need this,
and would be prettier without, but given the amount of churn I am
already introducing I didn't want to play with tuning new mempools.This also fixes bug in pnfs_ld_handle_read_error. In the case of
desc->pg_bsize < PAGE_CACHE_SIZE, the pages list was empty, causing
replay attempt to do nothing.Signed-off-by: Fred Isaman
Signed-off-by: Trond Myklebust -
In order to avoid duplicating all the data in nfs_read_data whenever we
split it up into multiple RPC calls (either due to a short read result
or due to rsize < PAGE_SIZE), we split out the bits that are the same
per RPC call into a separate "header" structure.The goal this patch moves towards is to have a single header
refcounted by several rpc_data structures. Thus, want to always refer
from rpc_data to the header, and not the other way. This patch comes
close to that ideal, but the directio code currently needs some
special casing, isolated in the nfs_direct_[read_write]hdr_release()
functions. This will be dealt with in a future patch.Signed-off-by: Fred Isaman
Signed-off-by: Trond Myklebust -
Signed-off-by: Fred Isaman
Signed-off-by: Trond Myklebust -
Signed-off-by: Fred Isaman
Signed-off-by: Trond Myklebust -
Make it consistent with nfs_initiate_commit.
Signed-off-by: Fred Isaman
Signed-off-by: Trond Myklebust -
Commits don't need the vectors of pages, etc. that writes do. Split out
a separate structure for the commit operation.Signed-off-by: Fred Isaman
Signed-off-by: Trond Myklebust -
Also create a commit_info structure to hold the bucket array and push
it up from the lseg to the layout where it really belongs.While we are at it, fix a refcounting bug due to an (incorrect)
implicit assumption that filelayout_scan_ds_commit_list always
completely emptied the src list.This clarifies refcounting, removes the ugly find_only_write_lseg
functions, and pushes the file layout commit code along on the path to
supporting multiple lsegs.Signed-off-by: Fred Isaman
Signed-off-by: Trond Myklebust
