nvmet-rdma: add a NVMe over Fabrics RDMA target driver
This patch implements the RDMA transport for the NVMe over Fabrics target,
which allows exporting NVMe over Fabrics functionality over RDMA fabrics
(Infiniband, RoCE, iWARP).
All NVMe logic is in the generic target and this module just provides a
small glue between it and the generic code in the RDMA subsystem.
Signed-off-by: Armen Baloyan <armenx.baloyan@intel.com>, Signed-off-by: Jay Freyensee <james.p.freyensee@intel.com> Signed-off-by: Ming Lin <ming.l@ssi.samsung.com> Signed-off-by: Sagi Grimberg <sagi@grimberg.me> Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Steve Wise <swise@opengridcomputing.com> Tested-by: Steve Wise <swise@opengridcomputing.com> Signed-off-by: Jens Axboe <axboe@fb.com>
nvme-rdma.h: Add includes for nvme rdma_cm negotiation
NVMe over Fabrics RDMA transport defines a connection establishment
protocol over the RDMA connection manager. This header will be used by
both the host and target drivers to negotiate the connection
establishment parameters.
Signed-off-by: Jay Freyensee <james.p.freyensee@intel.com> Signed-off-by: Ming Lin <ming.l@ssi.samsung.com> Signed-off-by: Sagi Grimberg <sagi@grimberg.me> Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Steve Wise <swise@opengridcomputing.com> Tested-by: Steve Wise <swise@opengridcomputing.com> Signed-off-by: Jens Axboe <axboe@fb.com>
The new nvme-rdma driver will need to reinitialize all the tags as part of
the error recovery procedure (realloc the tag memory region). Add a helper
in blk-mq for it that can iterate over all requests in a tagset to make
this easier.
Signed-off-by: Sagi Grimberg <sagi@grimberg.me> Tested-by: Ming Lin <ming.l@ssi.samsung.com> Reviewed-by: Stephen Bates <Stephen.Bates@pmcs.com> Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Steve Wise <swise@opengridcomputing.com> Tested-by: Steve Wise <swise@opengridcomputing.com> Signed-off-by: Jens Axboe <axboe@fb.com>
The passed by reference ppa list in nvm_set_rqd_list() is updated when
multiple planes are available. In that case, each PPA plane is
incremented when the device side PPA list is created. This prevents the
caller to rely on the PPA list to be unmodified after a call.
The gen_mark_blk_bad function marks the wrong block when a block is on
a different channel. Fix the index calculation, so that it updates the
correct block.
lightnvm: make rrpc_map_page call nvm_get_blk outside locks
The nvm_get_blk() function is called with rlun->lock held. This is ok
when the media manager implementation doesn't go out of its atomic
context. However, if a media manager persists its metadata, and
guarantees that the block is given to the target, this is no longer
a viable approach. Therefore, clean up the flow of rrpc_map_page,
and make sure that nvm_get_blk() is called without any locks acquired.
lightnvm: remove _unlocked variant of [get/put]_blk
The [get/put]_blk API enables targets to get ownership of blocks at
runtime. This information is currently not recorded on disk, and the
information is therefore lost on power failure. To restore the
metadata, the [get/put]_blk must persist its metadata. In that case,
we need to control the outer lock, so that we can disable them while
updating the on-disk metadata. Fortunately, the _unlocked versions can
be removed, which allows us to move the lock into the [get/put]_blk
functions.
lightnvm: remove unused lists from struct rrpc_block
The ->list, ->open_list, and ->closed_list lists were previously used
for statistics. However, their usage have been removed, and thus these
can safely be removed.
If a media manager tries to initialize it targets upon media manager
initialization, the media manager will need to know which target types
are available in LightNVM. The lists of which managers and target types
are available shares the same lock.
Therefore, on initialization, the nvm_lock is taken by LightNVM core,
which later leads to a deadlock when target types are enumerated by the
media manager.
Add an exclusive lock for target types to resolve this conflict.
To enable persistent block management to easily control creation and
removal of targets, we move target management into the media
manager. The LightNVM core continues to maintain which target types are
registered, while the media manager now keeps track of its initialized
targets.
Two new callbacks for the media manager are introduced. create_tgt and
remove_tgt. Note that remove_tgt returns 0 on successfully removing a
target, and returns 1 if the target was not found.
The generic manager should be called the general media manager, and
instead of using the rather long name of "gennvm" in front of each data
structures, use "gen" instead to shorten it. Update the description of
the media manager as well to make the media manager purpose clearer.
The responsibility of the media manager is not to keep track of
open/closed blocks. This is better maintained within a target,
that already manages this information on writes.
Remove the statistics and merge the states NVM_BLK_ST_OPEN and
NVM_BLK_ST_CLOSED.
A couple of small checkpatch fixups to stop it from complaining.
./drivers/lightnvm/core.c:360: WARNING: line over 80 characters
./drivers/lightnvm/core.c:360: ERROR: trailing statements should be on
next line
./drivers/lightnvm/core.c:503: WARNING: Block comments use a trailing */
on a separate line
lightnvm: remove checkpatch warning for unsigned ints
Checkpatch found two incidents where the type was preferred to be
written out in full.
./drivers/lightnvm/rrpc.h:184: WARNING: Prefer 'unsigned int' to bare
use of 'unsigned'
./drivers/lightnvm/rrpc.h:209: WARNING: Prefer 'unsigned int' to bare
use of 'unsigned'
./drivers/lightnvm/rrpc.c:51: WARNING: Prefer 'unsigned int' to bare use
of 'unsigned'
According to the OpenChannel SSD interface specification the NAND flash
MLC page pairing information's number of page page pairings field is the
first two bytes in the MLC Page Pairing data structure. The hardware's
data structure itself is little endian so annotate it as such, like the
rest of lighnvm's data structures.
CONFIG_NVME_TARGET has a correct CONFIG_CONFIGFS_FS dependency, but the
newly added NVME_TARGET_LOOP is missing this, resulting in a link
failure:
drivers/nvme/built-in.o: In function `nvmet_init_configfs':
loop.c:(.init.text+0x2a0): undefined reference to `config_group_init'
loop.c:(.init.text+0x2c0): undefined reference to `config_group_init_type_name'
loop.c:(.init.text+0x318): undefined reference to `configfs_register_subsystem'
drivers/nvme/built-in.o: In function `nvmet_exit_configfs':
loop.c:(.exit.text+0x9c): undefined reference to `configfs_unregister_subsystem'
This patch implements adds nvme-loop which allows to access local devices
exported as NVMe over Fabrics namespaces. This module can be useful for
easy evaluation, testing and also feature experimentation.
To createa nvme-loop device you need to configure the NVMe target to
export a loop port (see the nvmetcli documentaton for that) and then
connect to it using
nvme connect-all -t loop
which requires the very latest nvme-cli version with Fabrics support.
Signed-off-by: Jay Freyensee <james.p.freyensee@intel.com> Signed-off-by: Ming Lin <ming.l@ssi.samsung.com> Signed-off-by: Sagi Grimberg <sagi@grimberg.me> Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Steve Wise <swise@opengridcomputing.com> Signed-off-by: Jens Axboe <axboe@fb.com>
This patch introduces a implementation of NVMe subsystems,
controllers and discovery service which allows to export
NVMe namespaces across fabrics such as Ethernet, FC etc.
The implementation conforms to the NVMe 1.2.1 specification
and interoperates with NVMe over fabrics host implementations.
Configuration works using configfs, and is best performed using
the nvmetcli tool from http://git.infradead.org/users/hch/nvmetcli.git,
which also has a detailed explanation of the required steps in the
README file.
Signed-off-by: Armen Baloyan <armenx.baloyan@intel.com> Signed-off-by: Anthony Knapp <anthony.j.knapp@intel.com> Signed-off-by: Jay Freyensee <james.p.freyensee@intel.com> Signed-off-by: Ming Lin <ming.l@ssi.samsung.com> Signed-off-by: Sagi Grimberg <sagi@grimberg.me> Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Steve Wise <swise@opengridcomputing.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Sagi Grimberg [Mon, 13 Jun 2016 14:45:28 +0000 (16:45 +0200)]
nvme: add keep-alive support
Periodic keep-alive is a mandatory feature in NVMe over Fabrics, and
optional in NVMe 1.2.1 for PCIe. This patch adds periodic keep-alive
sent from the host to verify that the controller is still responsive
and vice-versa. The keep-alive timeout is user-defined (with
keep_alive_tmo connection parameter) and defaults to 5 seconds.
In order to avoid a race condition where the host sends a keep-alive
competing with the target side keep-alive timeout expiration, the host
adds a grace period of 10 seconds when publishing the keep-alive timeout
to the target.
In case a keep-alive failed (or timed out), a transport specific error
recovery kicks in.
For now only NVMe over Fabrics is wired up to support keep alive, but
we can add PCIe support easily once controllers actually supporting it
become available.
Signed-off-by: Sagi Grimberg <sagi@grimberg.me> Reviewed-by: Steve Wise <swise@chelsio.com> Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Keith Busch <keith.busch@intel.com> Signed-off-by: Jens Axboe <axboe@fb.com>
nvme-fabrics: add a generic NVMe over Fabrics library
The NVMe over Fabrics library provides an interface for both transports
and the nvme core to handle fabrics specific commands and attributes
independent of the underlying transport.
In addition, the fabrics library adds a misc device interface that allow
actually creating a fabrics controller, as we can't just autodiscover
it like in the PCI case. The nvme-cli utility has been enhanced to use
this interface to support fabric connect and discovery.
Signed-off-by: Armen Baloyan <armenx.baloyan@intel.com>, Signed-off-by: Jay Freyensee <james.p.freyensee@intel.com>, Signed-off-by: Ming Lin <ming.l@ssi.samsung.com> Signed-off-by: Sagi Grimberg <sagi@grimberg.me> Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Keith Busch <keith.busch@intel.com> Signed-off-by: Jens Axboe <axboe@fb.com>
The NVMe over Fabrics specification defines a protocol interface and
related extensions to NVMe that enable operation over network protocols.
The NVMe over Fabrics specification has an NVMe Transport binding for
each NVMe Transport.
This patch adds the fabrics related definitions:
- fabric specific command set and error codes
- transport addressing and binding definitions
- fabrics sgl extensions
- controller identification fabrics enhancements
- discovery log page definition
Signed-off-by: Armen Baloyan <armenx.baloyan@intel.com> Signed-off-by: James Smart <james.smart@broadcom.com> Signed-off-by: Jay Freyensee <james.p.freyensee@intel.com> Signed-off-by: Ming Lin <ming.l@ssi.samsung.com> Signed-off-by: Sagi Grimberg <sagi@grimberg.me> Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Keith Busch <keith.busch@intel.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Ming Lin [Mon, 13 Jun 2016 14:45:24 +0000 (16:45 +0200)]
nvme: add fabrics sysfs attributes
- delete_controller: This attribute allows to delete a controller.
A driver is not obligated to support it (pci doesn't) so it is
created only if the driver supports it. The new fabrics drivers
will support it (essentialy a disconnect operation).
- subsysnqn: This attribute shows the subsystem nqn of the configured
device. If a driver does not implement the get_subsysnqn method, the
file will not appear in sysfs.
- transport: This attribute shows the transport name. Added a "name"
field to struct nvme_ctrl_ops.
For loop,
cat /sys/class/nvme/nvme0/transport
loop
For RDMA,
cat /sys/class/nvme/nvme0/transport
rdma
For PCIe,
cat /sys/class/nvme/nvme0/transport
pcie
- address: This attributes shows the controller address. The fabrics
drivers that will implement get_address can show the address of the
connected controller.
Signed-off-by: Ming Lin <ming.l@ssi.samsung.com> Reviewed-by: Jay Freyensee <james.p.freyensee@intel.com> Reviewed-by: Sagi Grimberg <sagi@grimberg.me> Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Keith Busch <keith.busch@intel.com> Signed-off-by: Jens Axboe <axboe@fb.com>
nvme: Modify and export sync command submission for fabrics
NVMe over fabrics will use __nvme_submit_sync_cmd in the the
transport and require a few tweaks to it. For that we export it
and add a few more paramters:
1. allow passing a queue ID to the block layer
For the NVMe over Fabrics connect command we need to able to specify a
queue ID that we want to send the command on. Add a qid parameter to
the relevant functions to enable this behavior.
2. allow submitting at_head commands
In cases where we want to (re)connect to a controller
where we have inflight queued commands we want to first
connect and only then allow the other queued commands to
be kicked. This will prevents failures in controller resets
and reconnects.
3. allow passing flags to blk_mq_allocate_request
Both for Fabrics connect the the keep-alive feature in NVMe 1.2.1 we
want to be able to use reserved requests.
Reviewed-by: Jay Freyensee <james.p.freyensee@intel.com> Reviewed-by: Sagi Grimberg <sagi@grimberg.me> Tested-by: Ming Lin <ming.l@ssi.samsung.com> Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Keith Busch <keith.busch@intel.com> Signed-off-by: Jens Axboe <axboe@fb.com>
MG_DISK_MAJ is defined as 0 so dynamic block major number
allocation is used by the driver and the assigned major
number is stored in host->major. This patch fixes error
path in mg_probe() to use host->major instead of using
MG_DISK_MAJ.
Lars Ellenberg [Mon, 13 Jun 2016 22:26:38 +0000 (00:26 +0200)]
drbd: al_write_transaction: skip re-scanning of bitmap page pointer array
For larger devices, the array of bitmap page pointers can grow very
large (8000 pointers per TB of storage).
For each activity log transaction, we need to flush the associated
bitmap pages to stable storage. Currently, we just "mark" the respective
pages while setting up the transaction, then tell the bitmap code to
write out all marked pages, but skip unchanged pages.
But one such transaction can affect only a small number of bitmap pages,
there is no need to scan the full array of several (ten-)thousand
page pointers to find the few marked ones.
Instead, remember the index numbers of the few affected pages,
and later only re-check those to skip duplicates and unchanged ones.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Roland Kammerer [Mon, 13 Jun 2016 22:26:36 +0000 (00:26 +0200)]
drbd: get rid of empty statement in is_valid_state
This should silence a warning about an empty statement. Thanks to Fabian
Frederick <fabf@skynet.be> who sent a patch I modified to be smaller and
avoids an additional indent level.
Signed-off-by: Roland Kammerer <roland.kammerer@linbit.com> Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Fabian Frederick [Mon, 13 Jun 2016 22:26:35 +0000 (00:26 +0200)]
drbd: code cleanups without semantic changes
This contains various cosmetic fixes ranging from simple typos to
const-ifying, and using booleans properly.
Original commit messages from Fabian's patch set:
drbd: debugfs: constify drbd_version_fops
drbd: use seq_put instead of seq_print where possible
drbd: include linux/uaccess.h instead of asm/uaccess.h
drbd: use const char * const for drbd strings
drbd: kerneldoc warning fix in w_e_end_data_req()
drbd: use unsigned for one bit fields
drbd: use bool for peer is_ states
drbd: fix typo
drbd: use | for bitmask combination
drbd: use true/false for bool
drbd: fix drbd_bm_init() comments
drbd: introduce peer state union
drbd: fix maybe_pull_ahead() locking comments
drbd: use bool for growing
drbd: remove redundant declarations
drbd: replace if/BUG by BUG_ON
Signed-off-by: Fabian Frederick <fabf@skynet.be> Signed-off-by: Roland Kammerer <roland.kammerer@linbit.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Lars Ellenberg [Mon, 13 Jun 2016 22:26:34 +0000 (00:26 +0200)]
drbd: bump current uuid when resuming IO with diskless peer
Scenario, starting with normal operation
Connected Primary/Secondary UpToDate/UpToDate
NetworkFailure Primary/Unknown UpToDate/DUnknown (frozen)
... more failures happen, secondary loses it's disk,
but eventually is able to re-establish the replication link ...
Connected Primary/Secondary UpToDate/Diskless (resumed; needs to bump uuid!)
We used to just resume/resent suspended requests,
without bumping the UUID.
Which will lead to problems later, when we want to re-attach the disk on
the peer, without first disconnecting, or if we experience additional
failures, because we now have diverging data without being able to
recognize it.
Make sure we also bump the current data generation UUID,
if we notice "peer disk unknown" -> "peer disk known bad".
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Lars Ellenberg [Mon, 13 Jun 2016 22:26:33 +0000 (00:26 +0200)]
drbd: disallow promotion during resync handshake, avoid deadlock and hard reset
We already serialize connection state changes,
and other, non-connection state changes (role changes)
while we are establishing a connection.
But if we have an established connection,
then trigger a resync handshake (by primary --force or similar),
until now we just had to be "lucky".
Consider this sequence (e.g. deployment scenario):
create-md; up;
-> Connected Secondary/Secondary Inconsistent/Inconsistent
then do a racy primary --force on both peers.
block drbd0: drbd_sync_handshake:
block drbd0: self 0000000000000004:0000000000000000:0000000000000000:0000000000000000 bits:25590 flags:0
block drbd0: peer 0000000000000004:0000000000000000:0000000000000000:0000000000000000 bits:25590 flags:0
block drbd0: peer( Unknown -> Secondary ) conn( WFReportParams -> Connected ) pdsk( DUnknown -> Inconsistent )
block drbd0: peer( Secondary -> Primary ) pdsk( Inconsistent -> UpToDate )
*** HERE things go wrong. ***
block drbd0: role( Secondary -> Primary )
block drbd0: drbd_sync_handshake:
block drbd0: self 0000000000000005:0000000000000000:0000000000000000:0000000000000000 bits:25590 flags:0
block drbd0: peer C90D2FC716D232AB:0000000000000004:0000000000000000:0000000000000000 bits:25590 flags:0
block drbd0: Becoming sync target due to disk states.
block drbd0: Writing the whole bitmap, full sync required after drbd_sync_handshake.
block drbd0: Remote failed to finish a request within 6007ms > ko-count (2) * timeout (30 * 0.1s)
drbd s0: peer( Primary -> Unknown ) conn( Connected -> Timeout ) pdsk( UpToDate -> DUnknown )
The problem here is that the local promotion happens before the sync handshake
triggered by the remote promotion was completed. Some assumptions elsewhere
become wrong, and when the expected resync handshake is then received and
processed, we get stuck in a deadlock, which can only be recovered by reboot :-(
Fix: if we know the peer has good data,
and our own disk is present, but NOT good,
and there is no resync going on yet,
we expect a sync handshake to happen "soon".
So reject a racy promotion with SS_IN_TRANSIENT_STATE.
Lars Ellenberg [Mon, 13 Jun 2016 22:26:32 +0000 (00:26 +0200)]
drbd: sync_handshake: handle identical uuids with current (frozen) Primary
If in a two-primary scenario, we lost our peer, freeze IO,
and are still frozen (no UUID rotation) when the peer comes back
as Secondary after a hard crash, we will see identical UUIDs.
The "rule_nr = 40" chose to use the "CRASHED_PRIMARY" bit as
arbitration, but that would cause the still running (but frozen) Primary
to become SyncTarget (which it typically refuses), and the handshake is
declined.
Fix: check current roles.
If we have *one* current primary, the Primary wins.
(rule_nr = 41)
Since that is a protocol change, use the newly introduced DRBD_FF_WSAME
to determine if rule_nr = 41 can be applied.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Lars Ellenberg [Mon, 13 Jun 2016 22:26:31 +0000 (00:26 +0200)]
drbd: introduce WRITE_SAME support
We will support WRITE_SAME, if
* all peers support WRITE_SAME (both in kernel and DRBD version),
* all peer devices support WRITE_SAME
* logical_block_size is identical on all peers.
We may at some point introduce a fallback on the receiving side
for devices/kernels that do not support WRITE_SAME,
by open-coding a submit loop. But not yet.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Lars Ellenberg [Mon, 13 Jun 2016 22:26:29 +0000 (00:26 +0200)]
drbd: discard_zeroes_if_aligned allows "thin" resync for discard_zeroes_data=0
Even if discard_zeroes_data != 0,
if discard_zeroes_if_aligned is set, we assume we can reliably
zero-out/discard using the drbd_issue_peer_discard() helper.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Lars Ellenberg [Mon, 13 Jun 2016 22:26:28 +0000 (00:26 +0200)]
drbd: only restart frozen disk io when D_UP_TO_DATE
When re-attaching the local backend device to a C_STANDALONE D_DISKLESS
R_PRIMARY with OND_SUSPEND_IO, we may only resume IO if we recognize the
backend that is being attached as D_UP_TO_DATE.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Lars Ellenberg [Mon, 13 Jun 2016 22:26:27 +0000 (00:26 +0200)]
drbd: if there is no good data accessible, writes should be IO errors
If DRBD lost all path to good data,
and the on-no-data-accessible policy is OND_SUSPEND_IO,
all pending and new IO requests are suspended (will block).
If that setting is OND_IO_ERROR, IO will still be completed.
READ to "clean" areas (e.g. on an D_INCONSISTENT device,
and bitmap indicates a block is already in sync) will succeed.
READ to "unclean" areas (bitmap indicates block is out-of-sync),
will return EIO.
If we are already D_DISKLESS (or D_FAILED), we also return EIO.
Unfortunately, on a former R_PRIMARY C_SYNC_TARGET D_INCONSISTENT,
after replication link loss, new WRITE requests still went through OK.
The would also set the "out-of-sync" bit on their way, so READ after
WRITE would still return EIO. Also, the data generation UUIDs had not
been bumped, we would cause data divergence, without being able to
detect it on the next sync handshake, given the right sequence of events
in a multiple error scenario and "improper" order of recovery actions.
The right thing to do is to return EIO for all new writes,
unless we have access to good, current, D_UP_TO_DATE data.
The "established best practices" way to avoid these situations in the
first place is to set OND_SUSPEND_IO, or even do a hard-reset from
the pri-on-incon-degr policy helper hook.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Lars Ellenberg [Mon, 13 Jun 2016 22:26:26 +0000 (00:26 +0200)]
drbd: don't forget error completion when "unsuspending" IO
Possibly sequence of events:
SyncTarget is made Primary, then loses replication link
(only path to good data on SyncSource).
Behavior is then controlled by the on-no-data-accessible policy,
which defaults to OND_IO_ERROR (may be set to OND_SUSPEND_IO).
If OND_IO_ERROR is in fact the current policy, we clear the susp_fen
(IO suspended due to fencing policy) flag, do NOT set the susp_nod
(IO suspended due to no data) flag.
But we forgot to call the IO error completion for all pending,
suspended, requests.
While at it, also add a race check for a theoretically possible
race with a new handshake (network hickup), we may be able to
re-send requests, and can avoid passing IO errors up the stack.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Lars Ellenberg [Mon, 13 Jun 2016 22:26:25 +0000 (00:26 +0200)]
drbd: introduce unfence-peer handler
When resync is finished, we already call the "after-resync-target"
handler (on the former sync target, obviously), once per volume.
Paired with the before-resync-target handler, you can create snapshots,
before the resync causes the volumes to become inconsistent,
and discard those snapshots again, once they are no longer needed.
It was also overloaded to be paired with the "fence-peer" handler,
to "unfence" once the volumes are up-to-date and known good.
This has some disadvantages, though: we call "fence-peer" for the whole
connection (once for the group of volumes), but would call unfence as
side-effect of after-resync-target once for each volume.
Also, we fence on a (current, or about to become) Primary,
which will later become the sync-source.
Calling unfence only as a side effect of the after-resync-target
handler opens a race window, between a new fence on the Primary
(SyncTarget) and the unfence on the SyncTarget, which is difficult to
close without some kind of "cluster wide lock" in those handlers.
We would not need those handlers if we could still communicate.
Which makes trying to aquire a cluster wide lock from those handlers
seem like a very bad idea.
This introduces the "unfence-peer" handler, which will be called
per connection (once for the group of volumes), just like the fence
handler, only once all volumes are back in sync, and on the SyncSource.
Which is expected to be the node that previously called "fence", the
node that is currently allowed to be Primary, and thus the only node
that could trigger a new "fence" that could race with this unfence.
Which makes us not need any cluster wide synchronization here,
serializing two scripts running on the same node is trivial.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Lars Ellenberg [Mon, 13 Jun 2016 22:26:24 +0000 (00:26 +0200)]
drbd: finish resync on sync source only by notification from sync target
If the replication link breaks exactly during "resync finished" detection,
finishing too early on the sync source could again lead to UUIDs rotated
too fast, and potentially a spurious full resync on next handshake.
Always wait for explicit resync finished state change notification from
the sync target.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Lars Ellenberg [Mon, 13 Jun 2016 22:26:21 +0000 (00:26 +0200)]
drbd: possibly disable discard support, if backend has discard_zeroes_data=0
Now that we have the discard_zeroes_if_aligned setting, we should also
check it when setting up our queue parameters on the primary,
not only on the receiving side.
We announce discard support,
UNLESS
* we are connected to a peer that does not support TRIM
on the DRBD protocol level. Otherwise, it would either discard, or
do a fallback to zero-out, depending on its backend and configuration.
* our local backend does not support discards,
or (discard_zeroes_data=0 AND discard_zeroes_if_aligned=no).
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Lars Ellenberg [Mon, 13 Jun 2016 22:26:20 +0000 (00:26 +0200)]
drbd: when receiving P_TRIM, zero-out partial unaligned chunks
We can avoid spurious data divergence caused by partially-ignored
discards on certain backends with discard_zeroes_data=0, if we
translate partial unaligned discard requests into explicit zero-out.
The relevant use case is LVM/DM thin.
If on different nodes, DRBD is backed by devices with differing
discard characteristics, discards may lead to data divergence
(old data or garbage left over on one backend, zeroes due to
unmapped areas on the other backend). Online verify would now
potentially report tons of spurious differences.
While probably harmless for most use cases (fstrim on a file system),
DRBD cannot have that, it would violate our promise to upper layers
that our data instances on the nodes are identical.
To be correct and play safe (make sure data is identical on both copies),
we would have to disable discard support, if our local backend (on a
Primary) does not support "discard_zeroes_data=true".
We'd also have to translate discards to explicit zero-out on the
receiving (typically: Secondary) side, unless the receiving side
supports "discard_zeroes_data=true".
Which both would allocate those blocks, instead of unmapping them,
in contrast with expectations.
LVM/DM thin does set discard_zeroes_data=0,
because it silently ignores discards to partial chunks.
We can work around this by checking the alignment first.
For unaligned (wrt. alignment and granularity) or too small discards,
we zero-out the initial (and/or) trailing unaligned partial chunks,
but discard all the aligned full chunks.
At least for LVM/DM thin, the result is effectively "discard_zeroes_data=1".
Arguably it should behave this way internally, by default,
and we'll try to make that happen.
But our workaround is still valid for already deployed setups,
and for other devices that may behave this way.
Setting discard-zeroes-if-aligned=yes will allow DRBD to use
discards, and to announce discard_zeroes_data=true, even on
backends that announce discard_zeroes_data=false.
Setting discard-zeroes-if-aligned=no will cause DRBD to always
fall-back to zero-out on the receiving side, and to not even
announce discard capabilities on the Primary, if the respective
backend announces discard_zeroes_data=false.
We used to ignore the discard_zeroes_data setting completely.
To not break established and expected behaviour, and suddenly
cause fstrim on thin-provisioned LVs to run out-of-space,
instead of freeing up space, the default value is "yes".
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Lars Ellenberg [Mon, 13 Jun 2016 22:26:19 +0000 (00:26 +0200)]
drbd: allow parallel flushes for multi-volume resources
To maintain write-order fidelity accros all volumes in a DRBD resource,
the receiver of a P_BARRIER needs to issue flushes to all volumes.
We used to do this by calling blkdev_issue_flush(), synchronously,
one volume at a time.
We now submit all flushes to all volumes in parallel, then wait for all
completions, to reduce worst-case latencies on multi-volume resources.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Lars Ellenberg [Mon, 13 Jun 2016 22:26:18 +0000 (00:26 +0200)]
drbd: fix for truncated minor number in callback command line
The command line parameter the kernel module uses to communicate the
device minor to userland helper is flawed in a way that the device
indentifier "minor-%d" is being truncated to minors with a maximum
of 5 digits.
But DRBD 8.4 allows 2^20 == 1048576 minors,
thus a minimum of 7 digits must be supported.
Reported by Veit Wahlich on drbd-dev.
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Lars Ellenberg [Mon, 13 Jun 2016 22:26:17 +0000 (00:26 +0200)]
drbd: fix regression: protocol A sometimes synchronous, C sometimes double-latency
Regression introduced with 8.4.5
drbd: application writes may set-in-sync in protocol != C
Overwriting the same block (LBA) while a former version is still
"in-flight" to the peer (to be exact: we did not receive the
P_BARRIER_ACK for its epoch yet) would wait for the full epoch of that
former version to be acknowledged by the peer.
In synchronous and quasi-synchronous protocols C and B,
this may double the latency on overwrites.
With protocol A, which is supposed to be asynchronous and only wait for
local completion, it is even worse: it would make overwrites
quasi-synchronous, they would be hit by the full RTT, which protocol A
was specifically meant to avoid, and possibly the additional time it
takes to drain the buffers first.
Particularly bad for databases, or anything else that
does frequent updates to the same blocks (various file system meta data).
No impact if >= rtt passes between updates to the same block.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Philipp Reisner [Mon, 13 Jun 2016 22:26:15 +0000 (00:26 +0200)]
drbd: Create the protocol feature THIN_RESYNC
If thinly provisioned volumes are used, during a resync the sync source
tries to find out if a block is deallocated. If it is deallocated, then
the resync target uses block_dev_issue_zeroout() on the range in
question.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Philipp Reisner [Mon, 13 Jun 2016 22:26:14 +0000 (00:26 +0200)]
drbd: Introduce new disk config option rs-discard-granularity
As long as the value is 0 the feature is disabled. With setting
it to a positive value, DRBD limits and aligns its resync requests
to the rs-discard-granularity setting. If the sync source detects
all zeros in such a block, the resync target discards the range
on disk.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Lars Ellenberg [Mon, 13 Jun 2016 22:26:11 +0000 (00:26 +0200)]
drbd: change bitmap write-out when leaving resync states
When leaving resync states because of disconnect,
do the bitmap write-out synchronously in the drbd_disconnected() path.
When leaving resync states because we go back to AHEAD/BEHIND, or
because resync actually finished, or some disk was lost during resync,
trigger the write-out from after_state_ch().
The bitmap write-out for resync -> ahead/behind was missing completely before.
Note that this is all only an optimization to avoid double-resyncs of
already completed blocks in case this node crashes.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Lars Ellenberg [Mon, 13 Jun 2016 22:26:10 +0000 (00:26 +0200)]
drbd: bitmap bulk IO: do not always suspend IO
The intention was to only suspend IO if some normal bitmap operation is
supposed to be locked out, not always. If the bulk operation is flaged
as BM_LOCKED_CHANGE_ALLOWED, we do not need to suspend IO.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com> Signed-off-by: Jens Axboe <axboe@fb.com>
nvme: move the workaround for I/O queue-less controllers from PCIe to core
We want to apply this to Fabrics drivers as well, so move it to common
code.
Reviewed-by: Jay Freyensee <james.p.freyensee@intel.com> Reviewed-by: Sagi Grimberg <sagi@grimberg.me> Tested-by: Ming Lin <ming.l@ssi.samsung.com> Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Keith Busch <keith.busch@intel.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Reviewed-by: Jay Freyensee <james.p.freyensee@intel.com> Reviewed-by: Sagi Grimberg <sagi@grimberg.me> Reviewed-by: Ming Lin <ming.l@ssi.samsung.com> Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Keith Busch <keith.busch@intel.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Dedicated workqueues have been used since bcache_wq and moving_gc_wq
are workqueues for writes and are being used on a memory reclaim path.
WQ_MEM_RECLAIM has been set to ensure forward progress under memory
pressure.
Since there are only a fixed number of work items, explicit concurrency
limit is unnecessary here.
block: add a separate operation type for secure erase
Instead of overloading the discard support with the REQ_SECURE flag.
Use the opportunity to rename the queue flag as well, and remove the
dead checks for this flag in the RAID 1 and RAID 10 drivers that don't
claim support for secure erase.
Signed-off-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Jens Axboe <axboe@fb.com>
Sunad Bhandary [Fri, 27 May 2016 10:29:43 +0000 (15:59 +0530)]
NVMe: Fix removal in case of active namespace list scanning method
In case of the active namespace list scanning method, a namespace that
is detached is not removed from the host if it was the last entry in
the list. Fix this by adding a scan to validate namespaces greater than
the value of prev.
This also handles the case of removing namespaces whose value exceed
the device's reported number of namespaces.
Signed-off-by: Sunad Bhandary S <sunad.s@samsung.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Jens Axboe <axboe@fb.com>
Ming Lin [Wed, 18 May 2016 21:05:02 +0000 (14:05 -0700)]
nvme: move nvme_cancel_request() to common code
So it can be used by fabrics driver also.
Signed-off-by: Ming Lin <ming.l@samsung.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de> Reviewed-by: Keith Busch <keith.bsuch@intel.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Ming Lin [Wed, 18 May 2016 21:05:01 +0000 (14:05 -0700)]
nvme: update and rename nvme_cancel_io to nvme_cancel_request
nvme_cancel_io is a bit confusing (given the distinction of io/admin),
so rename it to nvme_cancel_request.
And update it a bit to pass in struct nvme_ctrl, so it can be used
by Fabrics driver also.
Signed-off-by: Ming Lin <ming.l@samsung.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Suggested-by: Sagi Grimberg <sagi@grimberg.me> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de> Reviewed-by: Keith Busch <keith.bsuch@intel.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Mike Christie [Sun, 5 Jun 2016 19:32:25 +0000 (14:32 -0500)]
block, drivers, fs: rename REQ_FLUSH to REQ_PREFLUSH
To avoid confusion between REQ_OP_FLUSH, which is handled by
request_fn drivers, and upper layers requesting the block layer
perform a flush sequence along with possibly a WRITE, this patch
renames REQ_FLUSH to REQ_PREFLUSH.
Signed-off-by: Mike Christie <mchristi@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Mike Christie [Sun, 5 Jun 2016 19:32:24 +0000 (14:32 -0500)]
block: do not use REQ_FLUSH for tracking flush support
The last patch added a REQ_OP_FLUSH for request_fn drivers
and the next patch renames REQ_FLUSH to REQ_PREFLUSH which
will be used by file systems and make_request_fn drivers so
they can send a write/flush combo.
This patch drops xen's use of REQ_FLUSH to track if it supports
REQ_OP_FLUSH requests, so REQ_FLUSH can be deleted.
Signed-off-by: Mike Christie <mchristi@redhat.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Juergen Gross <kernel@pfupf.net> Signed-off-by: Jens Axboe <axboe@fb.com>
Mike Christie [Sun, 5 Jun 2016 19:32:23 +0000 (14:32 -0500)]
block, drivers: add REQ_OP_FLUSH operation
This adds a REQ_OP_FLUSH operation that is sent to request_fn
based drivers by the block layer's flush code, instead of
sending requests with the request->cmd_flags REQ_FLUSH bit set.
Signed-off-by: Mike Christie <mchristi@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Mike Christie [Sun, 5 Jun 2016 19:32:22 +0000 (14:32 -0500)]
block, fs, drivers: remove REQ_OP compat defs and related code
This patch drops the compat definition of req_op where it matches
the rq_flag_bits definitions, and drops the related old and compat
code that allowed users to set either the op or flags for the operation.
We also then store the operation in the bi_rw/cmd_flags field similar
to how we used to store the bio ioprio where it sat in the upper bits
of the field.
Signed-off-by: Mike Christie <mchristi@redhat.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Mike Christie [Sun, 5 Jun 2016 19:32:20 +0000 (14:32 -0500)]
block: move bio io prio to a new field
In the next patch, we move drop the compat code and make
the op a separate value that is hidden in bi_rw. To give
the op and rq bits flags room to grow this moves prio to
its own field.
Signed-off-by: Mike Christie <mchristi@redhat.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Mike Christie [Sun, 5 Jun 2016 19:32:19 +0000 (14:32 -0500)]
ide cd: do not set REQ_WRITE on requests.
The block layer will set the correct READ/WRITE operation flags/fields
when creating a request, so there is not need for drivers to set the
REQ_WRITE flag.
Signed-off-by: Mike Christie <mchristi@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Mike Christie [Sun, 5 Jun 2016 19:32:17 +0000 (14:32 -0500)]
drivers: use req op accessor
The req operation REQ_OP is separated from the rq_flag_bits
definition. This converts the block layer drivers to
use req_op to get the op from the request struct.
Signed-off-by: Mike Christie <mchristi@redhat.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Mike Christie [Sun, 5 Jun 2016 19:32:14 +0000 (14:32 -0500)]
blkg_rwstat: separate op from flags
The bio and request operation and flags are going to be separate
definitions, so we cannot pass them in as a bitmap. This patch
converts the blkg_rwstat code and its caller, cfq, to pass in the
values separately.
Signed-off-by: Mike Christie <mchristi@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Mike Christie [Sun, 5 Jun 2016 19:32:12 +0000 (14:32 -0500)]
block: prepare mq request creation to use REQ_OPs
This patch modifies the blk mq request creation code to use
separate variables for the operation and flags, because in the
the next patches the struct request users will be converted like
was done for bios.
Signed-off-by: Mike Christie <mchristi@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Mike Christie [Sun, 5 Jun 2016 19:32:11 +0000 (14:32 -0500)]
block: prepare request creation/destruction code to use REQ_OPs
This patch prepares *_get_request/*_put_request and freed_request,
to use separate variables for the operation and flags. In the
next patches the struct request users will be converted like
was done for bios where the op and flags are set separately.
Signed-off-by: Mike Christie <mchristi@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@fb.com>
projects / linux-beck.git / log