GitLab

Commit 4cf9d544

 recorded when an outgoing ceph message was ACKed,
in order to avoid unnecessary connection resets when an OSD is busy.

However, ack_stamp is uninitialized, so there is a window between
when the message is sent and when it is ACKed in which handle_timeout()
interprets the unitialized value as an expired timeout, and resets
the connection unnecessarily.

Close the window by initializing ack_stamp.
Signed-off-by: Jim Schutt <jaschut@sandia.gov>
Signed-off-by: Sage Weil <sage@newdream.net>

We want to remove all OSDs, not just those on the idle LRU.
Signed-off-by: Sage Weil <sage@newdream.net>

There were several problems here:

 1- we weren't tagging allocations with the pool, so they were never
    returned to the pool.
 2- msgpool_put didn't add back to the mempool, even it were called.
 3- msgpool_release didn't clear the pool pointer, so it would have looped
    had #1 not been broken.

These may or may not have been responsible for #1136 or #1381 (BUG due to
non-empty mempool on umount).  I can't seem to trigger the crash now using
the method I was using before.
Signed-off-by: Sage Weil <sage@newdream.net>

Keep track of when an outgoing message is ACKed (i.e., the server fully
received it and, presumably, queued it for processing).  Time out OSD
requests only if it's been too long since they've been received.

This prevents timeouts and connection thrashing when the OSDs are simply
busy and are throttling the requests they read off the network.
Reviewed-by: Yehuda Sadeh <yehuda@hq.newdream.net>
Signed-off-by: Sage Weil <sage@newdream.net>

open(2) must always include one of O_RDONLY, O_WRONLY, or O_RDWR.  No need
for any O_APPEND special case.

Passing O_WRONLY|O_RDWR is undefined according to the man page, but the
Linux VFS interprets this as O_RDWR, so we'll do the same.

This fixes open(2) with flags O_RDWR|O_APPEND, which was incorrectly being
translated to readonly.
Reported-by: Fyodor Ustinov <ufm@ufm.su>
Signed-off-by: Sage Weil <sage@newdream.net>

Set the page count correctly for non-page-aligned IO.  We were already
doing this correctly for alignment, but not the page count.  Fixes
DIRECT_IO writes from unaligned pages.
Signed-off-by: Sage Weil <sage@newdream.net>

If we cancel a write, trigger the safe completions to prevent a sync from
blocking indefinitely in ceph_osdc_sync().
Signed-off-by: Sage Weil <sage@newdream.net>

When the cluster is marked full, subscribe to subsequent map updates to
ensure we find out promptly when it is no longer full.  This will prevent
us from spewing ENOSPC for (much) longer than necessary.
Signed-off-by: Sage Weil <sage@newdream.net>

Old incrementals encode a 0 value (nearly always) when an osd goes down.
Change that to allow any state bit(s) to be flipped.  Special case 0 to
mean flip the CEPH_OSD_UP bit to mimic the old behavior.
Signed-off-by: Sage Weil <sage@newdream.net>

Since we pass the nofail arg, we should never get an error; BUG if we do.
(And fix the function to not return an error if __map_request fails.)
Signed-off-by: Sage Weil <sage@newdream.net>

Signed-off-by: Sage Weil <sage@newdream.net>

If we get a WAIT as a client something went wrong; error out.  And don't
fall through to an unrelated case.
Signed-off-by: Sage Weil <sage@newdream.net>

Signed-off-by: Sage Weil <sage@newdream.net>

Signed-off-by: Sage Weil <sage@newdream.net>

Signed-off-by: Sage Weil <sage@newdream.net>

If there is no get_authorizer method we set the out_kvec to a bogus
pointer.  The length is also zero in that case, so it doesn't much matter,
but it's better not to add the empty item in the first place.
Signed-off-by: Sage Weil <sage@newdream.net>

If a connection is closed and/or reopened (ceph_con_close, ceph_con_open)
it can race with a callback.  con_work does various state checks for
closed or reopened sockets at the beginning, but drops con->mutex before
making callbacks.  We need to check for state bit changes after retaking
the lock to ensure we restart con_work and execute those CLOSED/OPENING
tests or else we may end up operating under stale assumptions.

In Jim's case, this was causing 'bad tag' errors.

There are four cases where we re-take the con->mutex inside con_work: catch
them all and return EAGAIN from try_{read,write} so that we can restart
con_work.
Reported-by: Jim Schutt <jaschut@sandia.gov>
Tested-by: Jim Schutt <jaschut@sandia.gov>
Signed-off-by: Sage Weil <sage@newdream.net>

ceph_osdc_alloc_request returns NULL on failure.
Signed-off-by: Sage Weil <sage@newdream.net>

If memory allocation failed, calling ceph_msg_put() will cause GPF
since some of ceph_msg variables are not initialized first.

Fix Bug #970.
Signed-off-by: Henry C Chang <henry_c_chang@tcloudcomputing.com>
Signed-off-by: Sage Weil <sage@newdream.net>

Fix the request transition from linger -> normal request.  The key is to
preserve r_osd and requeue on the same OSD.  Reregister as a normal request,
add the request to the proper queues, then unregister the linger.  Fix the
unregister helper to avoid clearing r_osd (and also simplify the parallel
check in __unregister_request()).
Reported-by: Henry Chang <henry.cy.chang@gmail.com>
Signed-off-by: Sage Weil <sage@newdream.net>

Fixes generated by 'codespell' and manually reviewed.
Signed-off-by: Lucas De Marchi <lucas.demarchi@profusion.mobi>

This allows us to use existence of the key type as a feature test,
from userspace.
Signed-off-by: Tommi Virtanen <tommi.virtanen@dreamhost.com>
Signed-off-by: Sage Weil <sage@newdream.net>

Signed-off-by: Tommi Virtanen <tommi.virtanen@dreamhost.com>
Signed-off-by: Sage Weil <sage@newdream.net>

This makes the base64 logic be contained in mount option parsing,
and prepares us for replacing the homebew key management with the
kernel key retention service.
Signed-off-by: Tommi Virtanen <tommi.virtanen@dreamhost.com>
Signed-off-by: Sage Weil <sage@newdream.net>

We should only clear r_osd if we are neither registered as a linger or a
regular request.  We may unregister as a linger while still registered as
a regular request (e.g., in reset_osd).  Incorrectly clearing r_osd there
leads to a null pointer dereference in __send_request.

Also simplify the parallel check in __unregister_request() where we just
removed r_osd_item and know it's empty.
Signed-off-by: Sage Weil <sage@newdream.net>

There was a missing unlock on the error path if __map_request() failed.
Signed-off-by: Dan Carpenter <error27@gmail.com>
Signed-off-by: Sage Weil <sage@newdream.net>

This patch fixes 'event_work' dereference before it is checked for NULL.
Signed-off-by: Mariusz Kozlowski <mk@lab.zgora.pl>
Signed-off-by: Sage Weil <sage@newdream.net>

The release method for mds connections uses a backpointer to the
mds_client, so we need to flush the workqueue of any pending work (and
ceph_connection references) prior to freeing the mds_client.  This fixes
an oops easily triggered under UML by

 while true ; do mount ... ; umount ... ; done

Also fix an outdated comment: the flush in ceph_destroy_client only flushes
OSD connections out.  This bug is basically an artifact of the ceph ->
ceph+libceph conversion.
Signed-off-by: Sage Weil <sage@newdream.net>

Lingering requests are requests that are sent to the OSD normally but
tracked also after we get a successful request.  This keeps the OSD
connection open and resends the original request if the object moves to
another OSD.  The OSD can then send notification messages back to us
if another client initiates a notify.

This framework will be used by RBD so that the client gets notification
when a snapshot is created by another node or tool.
Signed-off-by: Yehuda Sadeh <yehuda@hq.newdream.net>
Signed-off-by: Sage Weil <sage@newdream.net>

If we send a request to osd A, and the request's pg remaps to osd B and
then back to A in quick succession, we need to resend the request to A. The
old code was only calling kick_requests after processing all incremental
maps in a message, so it was very possible to not resend a request that
needed to be resent.  This would make the osd eventually time out (at least
with the current default of osd timeouts enabled).

The correct approach is to scan requests on every map incremental.  This
patch refactors the kick code in a few ways:
 - all requests are either on req_lru (in flight), req_unsent (ready to
   send), or req_notarget (currently map to no up osd)
 - mapping always done by map_request (previous map_osds)
 - if the mapping changes, we requeue.  requests are resent only after all
   map incrementals are processed.
 - some osd reset code is moved out of kick_requests into a separate
   function
 - the "kick this osd" functionality is moved to kick_osd_requests, as it
   is unrelated to scanning for request->pg->osd mapping changes
Signed-off-by: Sage Weil <sage@newdream.net>

It used to return -EINVAL because it thought the end was not aligned
to 4 bytes.

Clean up superfluous src < end test in if, the while itself guarantees
that.
Signed-off-by: Tommi Virtanen <tommi.virtanen@dreamhost.com>
Signed-off-by: Sage Weil <sage@newdream.net>

The standby logic used to be pretty dependent on the work requeueing
behavior that changed when we switched to WQ_NON_REENTRANT.  It was also
very fragile.

Restructure things so that:
 - We clear WRITE_PENDING when we set STANDBY.  This ensures we will
   requeue work when we wake up later.
 - con_work backs off if STANDBY is set.  There is nothing to do if we are
   in standby.
 - clear_standby() helper is called by both con_send() and con_keepalive(),
   the two actions that can wake us up again.  Move the connect_seq++
   logic here.
Signed-off-by: Sage Weil <sage@newdream.net>

There was some broken keepalive code using a dead variable.  Shift to using
the proper bit flag.
Signed-off-by: Sage Weil <sage@newdream.net>

With commit f363e45f

 we replaced a bunch of hacky workqueue mutual
exclusion logic with the WQ_NON_REENTRANT flag.  One pieces of fallout is
that the exponential backoff breaks in certain cases:

 * con_work attempts to connect.
 * we get an immediate failure, and the socket state change handler queues
   immediate work.
 * con_work calls con_fault, we decide to back off, but can't queue delayed
   work.

In this case, we add a BACKOFF bit to make con_work reschedule delayed work
next time it runs (which should be immediately).
Signed-off-by: Sage Weil <sage@newdream.net>

If we mark the connection CLOSED we will give up trying to reconnect to
this server instance.  That is appropriate for things like a protocol
version mismatch that won't change until the server is restarted, at which
point we'll get a new addr and reconnect.  An authorization failure like
this is probably due to the server not properly rotating it's secret keys,
however, and should be treated as transient so that the normal backoff and
retry behavior kicks in.
Signed-off-by: Sage Weil <sage@newdream.net>

get_user_pages() can return fewer pages than we ask for.  We were returning
a bogus pointer/error code in that case.  Instead, loop until we get all
the pages we want or get an error we can return to the caller.
Signed-off-by: Sage Weil <sage@newdream.net>

Pass errors from writing to the socket up the stack.  If we get -EAGAIN,
return 0 from the helper to simplify the callers' checks.
Signed-off-by: Sage Weil <sage@newdream.net>

If we get EAGAIN when trying to read from the socket, it is not an error.
Return 0 from the helper in this case to simplify the error handling cases
in the caller (indirectly, try_read).

Fix try_read to pass any error to it's caller (con_work) instead of almost
always returning 0.  This let's us respond to things like socket
disconnects.
Signed-off-by: Sage Weil <sage@newdream.net>

ceph messenger code does a rather complex dancing around multithread
workqueue to make sure the same work item isn't executed concurrently
on different CPUs.  This restriction can be provided by workqueue with
WQ_NON_REENTRANT.

Make ceph_msgr_wq non-reentrant workqueue with the default concurrency
level and remove the QUEUED/BUSY logic.

* This removes backoff handling in con_work() but it couldn't reliably
  block execution of con_work() to begin with - queue_con() can be
  called after the work started but before BUSY is set.  It seems that
  it was an optimization for a rather cold path and can be safely
  removed.

* The number of concurrent work items is bound by the number of
  connections and connetions are independent from each other.  With
  the default concurrency level, different connections will be
  executed independently.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Sage Weil <sage@newdream.net>
Cc: ceph-devel@vger.kernel.org
Signed-off-by: Sage Weil <sage@newdream.net>

Always free memory allocated to 'pi' in
net/ceph/osdmap.c::osdmap_decode().
Signed-off-by: Jesper Juhl <jj@chaosbits.net>
Signed-off-by: Sage Weil <sage@newdream.net>