It seems that the real cause of all the issues where that
we did not noticed in drbd_try_connect() when the other
guy closes one socket if the round trip time gets higher
than 100ms. There were that 100ms hard coded!
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
It is no longer sufficient to trigger on local WRITE,
we need to check on (rq_state & RQ_IN_ACT_LOG)
before calling drbd_al_complete_io also in the error path.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
If there is no replication traffic within the idle timeout
(ping-int seconds), DRBD will send a P_PING,
and adjust the timeout to ping-timeout.
If there is no P_PING_ACK received within this ping-timeout,
DRBD finally drops the connection, and tries to re-establish it.
To decide which timeout was active, we compared the current timeout
with the ping-timeout, and dropped the connection, if that was the case.
By default, ping-int is 10 seconds, ping-timeout is 500 ms.
Unfortunately, if you configure ping-timeout to be the same as ping-int,
expiry of the idle-timeout had been mistaken for a missing ping ack,
and caused an immediate reconnection attempt.
Fix:
Allow both timeouts to be equal, use a local variable
to store which timeout is active.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
We limit ourselves to a configurable maximum number of pages used as
temporary bio pages.
If the configured "max_buffers" is not big enough to match the bandwidth
of the respective deployment, a distributed deadlock could be triggered
by e.g. fast online verify and heavy application IO.
TCP connections would block on congestion, because both receivers
would wait on pages to become available.
Fortunately the respective senders in this case would be able to give
back some pages already. So do that.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
In case a write failes on the local disk, go into D_INCONSISTENT
disk state. That causes future reads of that block to be shipped
to the peer.
Read retry remote was already in place.
Actually the documentation needs to get fixed now. Since the
application is still shielded from the error. (as long as we have
only a single disk failing) The difference to detach is that
we keep the disk. And therefore might keep all the other, still
working sectors up to date.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
The sector number on empty barrier requests may (will?) be -1, which,
given that it's being treated as unsigned 64-bit quantity, will almost
always exceed the actual (virtual) disk's size.
Inspired by Konrad's "When writting barriers set the sector number to
zero...".
While at it also add overflow checking to the math in vbd_translate().
Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
vbd_resize() up_read()'s xs_state.suspend_mutex twice in a row via double
xenbus_transaction_end() calls. The next down_read() in
xenbus_transaction_start() (at eg. the next resize attempt) hangs.
Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=618317
Acked-by: Jan Beulich <jbeulich@novell.com>
Acked-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
If the backend supports the 'feature-flush-cache' mode, use that
instead of the 'feature-barrier' support.
Currently there are three backends that support the 'feature-flush-cache'
mode: NetBSD, Solaris and Linux kernel. The 'flush' option is much
light-weight version than the 'barrier' support so lets try to use as
there are no filesystems in the kernel that use full barriers anymore.
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
barrier variable is int, not long. This overflow caused another variable
override: "err" (in PV code) and "binfo" (in xenlinux code -
drivers/xen/blkfront/blkfront.c). The later caused incorrect device
flags (RO/removable etc).
Signed-off-by: Marek Marczykowski <marmarek@mimuw.edu.pl>
Acked-by: Ian Campbell <Ian.Campbell@citrix.com>
[v1: Changed title]
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
drivers/block/cciss.c: In function ‘cciss_send_reset’:
drivers/block/cciss.c:2515:2: error: implicit declaration of function ‘fill_cmd’
drivers/block/cciss.c: At top level:
drivers/block/cciss.c:2531:12: error: conflicting types for ‘fill_cmd’
drivers/block/cciss.c:2534:1: note: an argument type that has a default promotion can’t match an empty parameter name list declaration
drivers/block/cciss.c:2515:18: note: previous implicit declaration of ‘fill_cmd’ was here
make[1]: *** [drivers/block/cciss.o] Error 1
make: *** [drivers/block/cciss.o] Error 2
Move fill_cmd() to above where it is first used.
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
This is to allow number of commands reserved for use by SCSI tape drives
and medium changers to be adjusted at driver load time via the kernel
parameter cciss_tape_cmds, with a default value of 6, and a range
of 2 - 16 inclusive. Previously, the driver limited the number of
commands which could be queued to the SCSI half of the the driver
to only 2. This is to fix the problem that if you had more than
two tape drives, you couldn't, for example, erase or rewind them all
at the same time.
Signed-off-by: Stephen M. Cameron <scameron@beardog.cce.hp.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
It causes NMIs which are undesirable at best, unsurvivable at worst.
Prefer the soft reset instead.
Signed-off-by: Stephen M. Cameron <scameron@beardog.cce.hp.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Just go straight to the soft-reset method instead.
Signed-off-by: Stephen M. Cameron <scameron@beardog.cce.hp.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
on driver load, if reset_devices is set, and the hard reset
attempts fail, try to bring up the controller to the point that
a command can be sent, and send it a soft reset command, then
after the reset undo whatever driver initialization was done to get
it to the point to take a command, and re-do it after the reset.
This is to get kdump to work on all the "non-resettable" controllers
(except 64xx controllers which can't be reset due to the potentially
shared cache module.)
Signed-off-by: Stephen M. Cameron <scameron@beardog.cce.hp.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
The bit-2-doorbell reset method seemed to cause (survivable) NMIs
on some systems and (unsurvivable) IOCK NMIs on some G7 servers.
Firmware guys implemented a new doorbell method to alleviate these
problems triggered by bit 5 of the doorbell register. We want to
use it if it's available.
Signed-off-by: Stephen M. Cameron <scameron@beardog.cce.hp.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Just to reduce the messages about timeouts that appear.
Signed-off-by: Stephen M. Cameron <scameron@beardog.cce.hp.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
When waiting for the board to become "not ready"
don't print a message saying "waiting for board to
become ready" (possibly followed by a message saying
"failed waiting for board to become not ready". Instead,
it should be "waiting for board to reset" and "failed
waiting for board to reset."
Signed-off-by: Stephen M. Cameron <scameron@beardog.cce.hp.com>
"
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Detect failure of controller reset by noticing if the 32 bytes of
"driver version" we store on the hardware in the config table
fail to get zeroed out. Previously we noticed if the controller
did not transition to "simple mode", but this did not detect reset
failure if the controller was already in simple mode prior to
the reset attempt (e.g. due to module parameter hpsa_simple_mode=1).
Signed-off-by: Stephen M. Cameron <scameron@beardog.cce.hp.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
This is to ensure the board interrupts are really off when
these functions return.
Signed-off-by: Stephen M. Cameron <scameron@beardog.cce.hp.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
We do a check for the operations right before calling dispatch_rw_block_io.
And then we do the same check in dispatch_rw_block_io. This patch
squashes those checks into the 'dispatch_rw_block_io' function.
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
We drop the support for 'feature-barrier' and add in the support
for the 'feature-flush-cache' if the real backend storage supports
flushing.
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
