Coly Li [Thu, 9 Aug 2018 07:48:49 +0000 (15:48 +0800)]
bcache: set max writeback rate when I/O request is idle
Commit b1092c9af9ed ("bcache: allow quick writeback when backing idle")
allows the writeback rate to be faster if there is no I/O request on a
bcache device. It works well if there is only one bcache device attached
to the cache set. If there are many bcache devices attached to a cache
set, it may introduce performance regression because multiple faster
writeback threads of the idle bcache devices will compete the btree level
locks with the bcache device who have I/O requests coming.
This patch fixes the above issue by only permitting fast writebac when
all bcache devices attached on the cache set are idle. And if one of the
bcache devices has new I/O request coming, minimized all writeback
throughput immediately and let PI controller __update_writeback_rate()
to decide the upcoming writeback rate for each bcache device.
Also when all bcache devices are idle, limited wrieback rate to a small
number is wast of thoughput, especially when backing devices are slower
non-rotation devices (e.g. SATA SSD). This patch sets a max writeback
rate for each backing device if the whole cache set is idle. A faster
writeback rate in idle time means new I/Os may have more available space
for dirty data, and people may observe a better write performance then.
Please note bcache may change its cache mode in run time, and this patch
still works if the cache mode is switched from writeback mode and there
is still dirty data on cache.
Fixes: Commit b1092c9af9ed ("bcache: allow quick writeback when backing idle") Cc: stable@vger.kernel.org #4.16+ Signed-off-by: Coly Li <colyli@suse.de> Tested-by: Kai Krakow <kai@kaishome.de> Tested-by: Stefan Priebe <s.priebe@profihost.ag> Cc: Michael Lyle <mlyle@lyle.org> Signed-off-by: Jens Axboe <axboe@kernel.dk>
Coly Li [Thu, 9 Aug 2018 07:48:48 +0000 (15:48 +0800)]
bcache: add code comments for bset.c
This patch tries to add code comments in bset.c, to make some
tricky code and designment to be more comprehensible. Most information
of this patch comes from the discussion between Kent and I, he
offers very informative details. If there is any mistake
of the idea behind the code, no doubt that's from me misrepresentation.
Signed-off-by: Coly Li <colyli@suse.de> Signed-off-by: Jens Axboe <axboe@kernel.dk>
In bch_btree_node_get() the read-in btree node will be partially
prefetched into L1 cache for following bset iteration (if there is).
But if the btree node read is failed, the perfetch operations will
waste L1 cache space. This patch checkes whether read operation and
only does cache prefetch when read I/O succeeded.
Signed-off-by: Coly Li <colyli@suse.de> Signed-off-by: Jens Axboe <axboe@kernel.dk>
Coly Li [Thu, 9 Aug 2018 07:48:43 +0000 (15:48 +0800)]
bcache: display rate debug parameters to 0 when writeback is not running
When writeback is not running, writeback rate should be 0, other value is
misleading. And the following dyanmic writeback rate debug parameters
should be 0 too,
rate, proportional, integral, change
otherwise they are misleading when writeback is not running.
Signed-off-by: Coly Li <colyli@suse.de> Signed-off-by: Jens Axboe <axboe@kernel.dk>
Coly Li [Thu, 9 Aug 2018 07:48:42 +0000 (15:48 +0800)]
bcache: do not check return value of debugfs_create_dir()
Greg KH suggests that normal code should not care about debugfs. Therefore
no matter successful or failed of debugfs_create_dir() execution, it is
unncessary to check its return value.
There are two functions called debugfs_create_dir() and check the return
value, which are bch_debug_init() and closure_debug_init(). This patch
changes these two functions from int to void type, and ignore return values
of debugfs_create_dir().
This patch does not fix exact bug, just makes things work as they should.
Signed-off-by: Coly Li <colyli@suse.de> Suggested-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: stable@vger.kernel.org Cc: Kai Krakow <kai@kaishome.de> Cc: Kent Overstreet <kent.overstreet@gmail.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
Bart Van Assche [Tue, 7 Aug 2018 23:17:29 +0000 (16:17 -0700)]
cfq: Suppress compiler warnings about comparisons
This patch does not change any functionality but avoids that gcc
reports the following warnings when building with W=1:
block/cfq-iosched.c: In function ?cfq_back_seek_max_store?:
block/cfq-iosched.c:4741:13: warning: comparison of unsigned expression < 0 is always false [-Wtype-limits]
if (__data < (MIN)) \
^
block/cfq-iosched.c:4756:1: note: in expansion of macro ?STORE_FUNCTION?
STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0);
^~~~~~~~~~~~~~
block/cfq-iosched.c: In function ?cfq_slice_idle_store?:
block/cfq-iosched.c:4741:13: warning: comparison of unsigned expression < 0 is always false [-Wtype-limits]
if (__data < (MIN)) \
^
block/cfq-iosched.c:4759:1: note: in expansion of macro ?STORE_FUNCTION?
STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1);
^~~~~~~~~~~~~~
block/cfq-iosched.c: In function ?cfq_group_idle_store?:
block/cfq-iosched.c:4741:13: warning: comparison of unsigned expression < 0 is always false [-Wtype-limits]
if (__data < (MIN)) \
^
block/cfq-iosched.c:4760:1: note: in expansion of macro ?STORE_FUNCTION?
STORE_FUNCTION(cfq_group_idle_store, &cfqd->cfq_group_idle, 0, UINT_MAX, 1);
^~~~~~~~~~~~~~
block/cfq-iosched.c: In function ?cfq_low_latency_store?:
block/cfq-iosched.c:4741:13: warning: comparison of unsigned expression < 0 is always false [-Wtype-limits]
if (__data < (MIN)) \
^
block/cfq-iosched.c:4765:1: note: in expansion of macro ?STORE_FUNCTION?
STORE_FUNCTION(cfq_low_latency_store, &cfqd->cfq_latency, 0, 1, 0);
^~~~~~~~~~~~~~
block/cfq-iosched.c: In function ?cfq_slice_idle_us_store?:
block/cfq-iosched.c:4775:13: warning: comparison of unsigned expression < 0 is always false [-Wtype-limits]
if (__data < (MIN)) \
^
block/cfq-iosched.c:4782:1: note: in expansion of macro ?USEC_STORE_FUNCTION?
USEC_STORE_FUNCTION(cfq_slice_idle_us_store, &cfqd->cfq_slice_idle, 0, UINT_MAX);
^~~~~~~~~~~~~~~~~~~
block/cfq-iosched.c: In function ?cfq_group_idle_us_store?:
block/cfq-iosched.c:4775:13: warning: comparison of unsigned expression < 0 is always false [-Wtype-limits]
if (__data < (MIN)) \
^
block/cfq-iosched.c:4783:1: note: in expansion of macro ?USEC_STORE_FUNCTION?
USEC_STORE_FUNCTION(cfq_group_idle_us_store, &cfqd->cfq_group_idle, 0, UINT_MAX);
^~~~~~~~~~~~~~~~~~~
Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
Anchal Agarwal [Tue, 7 Aug 2018 20:40:49 +0000 (14:40 -0600)]
blk-wbt: Avoid lock contention and thundering herd issue in wbt_wait
I am currently running a large bare metal instance (i3.metal)
on EC2 with 72 cores, 512GB of RAM and NVME drives, with a
4.18 kernel. I have a workload that simulates a database
workload and I am running into lockup issues when writeback
throttling is enabled,with the hung task detector also
kicking in.
Crash dumps show that most CPUs (up to 50 of them) are
all trying to get the wbt wait queue lock while trying to add
themselves to it in __wbt_wait (see stack traces below).
In the original patch, wbt_done is waking up all the exclusive
processes in the wait queue, which can cause a thundering herd
if there is a large number of writer threads in the queue. The
original intention of the code seems to be to wake up one thread
only however, it uses wake_up_all() in __wbt_done(), and then
uses the following check in __wbt_wait to have only one thread
actually get out of the wait loop:
if (waitqueue_active(&rqw->wait) &&
rqw->wait.head.next != &wait->entry)
return false;
The problem with this is that the wait entry in wbt_wait is
define with DEFINE_WAIT, which uses the autoremove wakeup function.
That means that the above check is invalid - the wait entry will
have been removed from the queue already by the time we hit the
check in the loop.
Secondly, auto-removing the wait entries also means that the wait
queue essentially gets reordered "randomly" (e.g. threads re-add
themselves in the order they got to run after being woken up).
Additionally, new requests entering wbt_wait might overtake requests
that were queued earlier, because the wait queue will be
(temporarily) empty after the wake_up_all, so the waitqueue_active
check will not stop them. This can cause certain threads to starve
under high load.
The fix is to leave the woken up requests in the queue and remove
them in finish_wait() once the current thread breaks out of the
wait loop in __wbt_wait. This will ensure new requests always
end up at the back of the queue, and they won't overtake requests
that are already in the wait queue. With that change, the loop
in wbt_wait is also in line with many other wait loops in the kernel.
Waking up just one thread drastically reduces lock contention, as
does moving the wait queue add/remove out of the loop.
A significant drop in lockdep's lock contention numbers is seen when
running the test application on the patched kernel.
Signed-off-by: Anchal Agarwal <anchalag@amazon.com> Signed-off-by: Frank van der Linden <fllinden@amazon.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
xen-blkfront: use true and false for boolean values
Return statements in functions returning bool should use true or false
instead of an integer value.
This code was detected with the help of Coccinelle.
Acked-by: Roger Pau Monné <roger.pau@citrix.com> Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Mon, 6 Aug 2018 01:34:09 +0000 (19:34 -0600)]
Merge branch 'nvme-4.19' of git://git.infradead.org/nvme into for-4.19/block2
Pull NVMe changes from Christoph:
"This contains the support for TP4004, Asymmetric Namespace Access,
which makes NVMe multipathing usable in practice."
* 'nvme-4.19' of git://git.infradead.org/nvme:
nvmet: use Retain Async Event bit to clear AEN
nvmet: support configuring ANA groups
nvmet: add minimal ANA support
nvmet: track and limit the number of namespaces per subsystem
nvmet: keep a port pointer in nvmet_ctrl
nvme: add ANA support
nvme: remove nvme_req_needs_failover
nvme: simplify the API for getting log pages
nvme.h: add ANA definitions
nvme.h: add support for the log specific field
To avoid introducing problems like those fixed in commit f7068114d45e
("sr: pass down correctly sized SCSI sense buffer"), this creates a macro
wrapper for scsi_execute() that verifies the size of the sense buffer
similar to what was done for command string sizes in commit 3756f6401c30
("exec: avoid gcc-8 warning for get_task_comm").
Another solution could be to add a length argument to scsi_execute(),
but this function already takes a lot of arguments and Jens was not fond
of that approach.
Additionally, this moves the SCSI_SENSE_BUFFERSIZE definition into
scsi_device.h, and removes a redundant include for scsi_device.h from
scsi_cmnd.h.
To support future compile-time sizeof() checks that will be able to
validate the length of sense buffers, this removes the only dynamically
allocated sense buffers in the tree by putting the 96 byte sense buffers
on the stack.
This removes more casts of struct request_sense and uses the standard
struct scsi_sense_hdr instead. This also fixes any possible stale values
since the prior code did not check the sense length.
This is already able to process the sense buffer, so remove the redundant
parsing during the failure path. This also fixes any possible stale values
since the prior code did not check the sense length.
Acked-by: David S. Miller <davem@davemloft.net> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Kees Cook <keescook@chromium.org> Signed-off-by: Jens Axboe <axboe@kernel.dk>
Kees Cook [Thu, 2 Aug 2018 21:22:13 +0000 (15:22 -0600)]
block: Switch struct packet_command to use struct scsi_sense_hdr
There is a lot of needless struct request_sense usage in the CDROM
code. These can all be struct scsi_sense_hdr instead, to avoid any
confusion over their respective structure sizes. This patch is a lot
of noise changing "sense" to "sshdr", but the final code is more
readable to distinguish between "sense" meaning "struct request_sense"
and "sshdr" meaning "struct scsi_sense_hdr".
Acked-by: David S. Miller <davem@davemloft.net> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Kees Cook <keescook@chromium.org> Signed-off-by: Jens Axboe <axboe@kernel.dk>
Ming Lei [Thu, 2 Aug 2018 10:23:26 +0000 (18:23 +0800)]
blk-mq: fix updating tags depth
The passed 'nr' from userspace represents the total depth, meantime
inside 'struct blk_mq_tags', 'nr_tags' stores the total tag depth,
and 'nr_reserved_tags' stores the reserved part.
There are two issues in blk_mq_tag_update_depth() now:
1) for growing tags, we should have used the passed 'nr', and keep the
number of reserved tags not changed.
2) the passed 'nr' should have been used for checking against
'tags->nr_tags', instead of number of the normal part.
This patch fixes the above two cases, and avoids kernel crash caused
by wrong resizing sbitmap queue.
Cc: "Ewan D. Milne" <emilne@redhat.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Bart Van Assche <bart.vanassche@sandisk.com> Cc: Omar Sandoval <osandov@fb.com>
Tested by: Marco Patalano <mpatalan@redhat.com> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
Ming Lei [Mon, 30 Jul 2018 12:02:19 +0000 (20:02 +0800)]
block: really disable runtime-pm for blk-mq
Runtime PM isn't ready for blk-mq yet, and commit 765e40b675a9 ("block:
disable runtime-pm for blk-mq") tried to disable it. Unfortunately,
it can't take effect in that way since user space still can switch
it on via 'echo auto > /sys/block/sdN/device/power/control'.
This patch disables runtime-pm for blk-mq really by pm_runtime_disable()
and fixes all kinds of PM related kernel crash.
Cc: Tomas Janousek <tomi@nomi.cz> Cc: Przemek Socha <soprwa@gmail.com> Cc: Alan Stern <stern@rowland.harvard.edu> Cc: <stable@vger.kernel.org> Reviewed-by: Bart Van Assche <bart.vanassche@wdc.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Patrick Steinhardt <ps@pks.im> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
Currently, avg_lat is calculated by accumulating the mean of every
window in a long running cumulative average. As time goes on, the metric
becomes less and less useful due to the accumulated history.
This patch reuses the same calculation done in load averages to make the
avg_lat metric more lively. Unlike load averages, the avg only advances
when a window elapses (due to an io). Idle periods extend the most
recent window. Bucketing is used to limit the history of avg_lat by
binding it to the window size. So, the window range for 1/exp (decay
rate) is [1 min, 2.5 min) when windows elapse immediately.
The current sample window size is exposed in the debug info to enable
calculation of the window range.
Signed-off-by: Dennis Zhou <dennisszhou@gmail.com> Acked-by: Tejun Heo <tj@kernel.org> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Acked-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
Josef Bacik [Tue, 31 Jul 2018 16:39:04 +0000 (12:39 -0400)]
blk-cgroup: clear the throttle queue on fork
We were hitting a panic in production where we put too many times on the
request queue. This is because we'd get the throttle_queue of the
parent if we fork()'ed while we needed to be throttled, but we didn't
have a reference on it. Instead just clear these flags on fork so the
child doesn't pay for the sins of its father.
Signed-off-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
xiao jin [Mon, 30 Jul 2018 06:11:12 +0000 (14:11 +0800)]
block: blk_init_allocated_queue() set q->fq as NULL in the fail case
We find the memory use-after-free issue in __blk_drain_queue()
on the kernel 4.14. After read the latest kernel 4.18-rc6 we
think it has the same problem.
Memory is allocated for q->fq in the blk_init_allocated_queue().
If the elevator init function called with error return, it will
run into the fail case to free the q->fq.
Then the __blk_drain_queue() uses the same memory after the free
of the q->fq, it will lead to the unpredictable event.
The patch is to set q->fq as NULL in the fail case of
blk_init_allocated_queue().
Fixes: commit 7c94e1c157a2 ("block: introduce blk_flush_queue to drive flush machinery") Cc: <stable@vger.kernel.org> Reviewed-by: Ming Lei <ming.lei@redhat.com> Reviewed-by: Bart Van Assche <bart.vanassche@wdc.com> Signed-off-by: xiao jin <jin.xiao@intel.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
Max Gurtovoy [Sun, 29 Jul 2018 21:15:33 +0000 (00:15 +0300)]
nvme: use blk API to remap ref tags for IOs with metadata
Also moved the logic of the remapping to the nvme core driver instead
of implementing it in the nvme pci driver. This way all the other nvme
transport drivers will benefit from it (in case they'll implement metadata
support).
Suggested-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Acked-by: Keith Busch <keith.busch@intel.com> Signed-off-by: Max Gurtovoy <maxg@mellanox.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
Max Gurtovoy [Sun, 29 Jul 2018 21:15:32 +0000 (00:15 +0300)]
block: move dif_prepare/dif_complete functions to block layer
Currently these functions are implemented in the scsi layer, but their
actual place should be the block layer since T10-PI is a general data
integrity feature that is used in the nvme protocol as well. Also, use
the tuple size from the integrity profile since it may vary between
integrity types.
Suggested-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Signed-off-by: Max Gurtovoy <maxg@mellanox.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
Max Gurtovoy [Sun, 29 Jul 2018 21:15:31 +0000 (00:15 +0300)]
block: move ref_tag calculation func to the block layer
Currently this function is implemented in the scsi layer, but it's
actual place should be the block layer since T10-PI is a general
data integrity feature that is used in the nvme protocol as well.
Suggested-by: Christoph Hellwig <hch@lst.de> Cc: Martin K. Petersen <martin.petersen@oracle.com> Signed-off-by: Max Gurtovoy <maxg@mellanox.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
Josef Bacik [Mon, 30 Jul 2018 14:10:01 +0000 (10:10 -0400)]
block: don't account for split bio's size in cgroup stats
We need to check in blkcg_bio_issue_check if the bio is flagged as
QUEUE_ENTERED, because if it is then we've already accounted for the
size of the IO in the cgroup stats. We can still however account for
the extra IO since it'll be another request.
In the current implementation, we clear the AEN bit when we get the
"get log page" command if given log page is associated with AEN.
This patch allows optionally retaining the AEN for the ctrl
under consideration when Retain Asynchronous Event (RAE) bit is set
as a part of "get log page" command.
This allows the host to read the Log page and optionally retaining the
AEN associated with this log page when using userspace tools like
nvme-cli.
Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
[hch: also use the new helper in the just merged ANA code] Signed-off-by: Christoph Hellwig <hch@lst.de>
Allow creating non-default ANA groups (group ID > 1). Groups are created
either by assigning the group ID to a namespace, or by creating a configfs
group object under a specific port. All namespaces assigned to a group
that doesn't have a configfs object for a given port are marked as
inaccessible.
Allow changing the ANA state on a per-port basis by creating an
ana_groups directory under each port, and another directory with an
ana_state file in it. The default ANA group 1 directory is created
automatically for each port.
For all changes in ANA configuration the ANA change AEN is sent. We only
keep a global changecount instead of additional per-group changecounts to
keep the implementation as simple as possible.
Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Keith Busch <keith.busch@intel.com> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Add support for Asynchronous Namespace Access as specified in NVMe 1.3
TP 4004.
Just add a default ANA group 1 that is optimized on all ports. This is
(and will remain) the default assignment for any namespace not epxlicitly
assigned to another ANA group. The ANA state can be manually changed
through the configfs interface, including the change state.
Includes fixes and improvements from Hannes Reinecke.
Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Keith Busch <keith.busch@intel.com> Reviewed-by: Sagi Grimberg <sagi@grimberg.me> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
nvmet: track and limit the number of namespaces per subsystem
TP 4004 introduces a new 'Maximum Number of Allocated Namespaces' field
in the Identify controller data to help the host size resources. Put
an upper limit on the supported namespaces to be able to support this
value as supporting 32-bits worth of namespaces would lead to very
large buffers. The limit is completely arbitrary at this point.
Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Keith Busch <keith.busch@intel.com> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Keith Busch <keith.busch@intel.com> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Add support for Asynchronous Namespace Access as specified in NVMe 1.3
TP 4004. With ANA each namespace attached to a controller belongs to an
ANA group that describes the characteristics of accessing the namespaces
through this controller. In the optimized and non-optimized states
namespaces can be accessed regularly, although in a multi-pathing
environment we should always prefer to access a namespace through a
controller where an optimized relationship exists. Namespaces in
Inaccessible, Permanent-Loss or Change state for a given controller
should not be accessed.
The states are updated through reading the ANA log page, which is read
once during controller initialization, whenever the ANA change notice
AEN is received, or when one of the ANA specific status codes that
signal a state change is received on a command.
The ANA state is kept in the nvme_ns structure, which makes the checks in
the fast path very simple. Updating the ANA state when reading the log
page is also very simple, the only downside is that finding the initial
ANA state when scanning for namespaces is a bit cumbersome.
The gendisk for a ns_head is only registered once a live path for it
exists. Without that the kernel would hang during partition scanning.
Includes fixes and improvements from Hannes Reinecke.
Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Keith Busch <keith.busch@intel.com> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Now that we just call out to blk_path_error there isn't really any good
reason to not merge it into the only caller.
Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Keith Busch <keith.busch@intel.com> Reviewed-by: Sagi Grimberg <sagi@grimberg.me> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Merge nvme_get_log and nvme_get_log_ext into a single helper, which takes
a plain nsid instead of the nvme_ns pointer. Also add support for the
log specific field while we're at it.
Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Keith Busch <keith.busch@intel.com> Reviewed-by: Sagi Grimberg <sagi@grimberg.me> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Keith Busch <keith.busch@intel.com> Reviewed-by: Sagi Grimberg <sagi@grimberg.me> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
NVMe 1.3 added a new log specific field to the get log page CQ
defintion, add it to our get_log_page SQ structure.
Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Keith Busch <keith.busch@intel.com> Reviewed-by: Sagi Grimberg <sagi@grimberg.me> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
partitions/aix: append null character to print data from disk
Even if properly initialized, the lvname array (i.e., strings)
is read from disk, and might contain corrupt data (e.g., lack
the null terminating character for strings).
So, make sure the partition name string used in pr_warn() has
the null terminating character.
Fixes: 6ceea22bbbc8 ("partitions: add aix lvm partition support files") Suggested-by: Daniel J. Axtens <daniel.axtens@canonical.com> Signed-off-by: Mauricio Faria de Oliveira <mfo@canonical.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
partitions/aix: fix usage of uninitialized lv_info and lvname structures
The if-block that sets a successful return value in aix_partition()
uses 'lvip[].pps_per_lv' and 'n[].name' potentially uninitialized.
For example, if 'numlvs' is zero or alloc_lvn() fails, neither is
initialized, but are used anyway if alloc_pvd() succeeds after it.
So, make the alloc_pvd() call conditional on their initialization.
This has been hit when attaching an apparently corrupted/stressed
AIX LUN, misleading the kernel to pr_warn() invalid data and hang.
[...] partition (null) (11 pp's found) is not contiguous
[...] partition (null) (2 pp's found) is not contiguous
[...] partition (null) (3 pp's found) is not contiguous
[...] partition (null) (64 pp's found) is not contiguous
Fixes: 6ceea22bbbc8 ("partitions: add aix lvm partition support files") Signed-off-by: Mauricio Faria de Oliveira <mfo@canonical.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
The get_seconds function is deprecated now since it returns a 32-bit
value that will eventually overflow, and we are replacing it throughout
the kernel with ktime_get_seconds() or ktime_get_real_seconds() that
return a time64_t.
bcache uses get_seconds() to read the current system time and store it in
the superblock as well as in uuid_entry structures that are user visible.
Unfortunately, the two structures in are still limited to 32 bits, so this
won't fix any real problems but will still overflow in year 2106. Let's
at least document that properly, in case we get an updated format in the
future it can be fixed. We still have a long time before the overflow
and checking the tools at https://github.com/koverstreet/bcache-tools
reveals no access to any of them.
bcache: fix I/O significant decline while backend devices registering
I attached several backend devices in the same cache set, and produced lots
of dirty data by running small rand I/O writes in a long time, then I
continue run I/O in the others cached devices, and stopped a cached device,
after a mean while, I register the stopped device again, I see the running
I/O in the others cached devices dropped significantly, sometimes even
jumps to zero.
In currently code, bcache would traverse each keys and btree node to count
the dirty data under read locker, and the writes threads can not get the
btree write locker, and when there is a lot of keys and btree node in the
registering device, it would last several seconds, so the write I/Os in
others cached device are blocked and declined significantly.
In this patch, when a device registering to a ache set, which exist others
cached devices with running I/Os, we get the amount of dirty data of the
device in an incremental way, and do not block other cached devices all the
time.
Patch v2: Rename some variables and macros name as Coly suggested.
bcache: calculate the number of incremental GC nodes according to the total of btree nodes
This patch base on "[PATCH] bcache: finish incremental GC".
Since incremental GC would stop 100ms when front side I/O comes, so when
there are many btree nodes, if GC only processes constant (100) nodes each
time, GC would last a long time, and the front I/Os would run out of the
buckets (since no new bucket can be allocated during GC), and I/Os be
blocked again.
So GC should not process constant nodes, but varied nodes according to the
number of btree nodes. In this patch, GC is divided into constant (100)
times, so when there are many btree nodes, GC can process more nodes each
time, otherwise GC will process less nodes each time (but no less than
MIN_GC_NODES).
In GC thread, we record the latest GC key in gc_done, which is expected
to be used for incremental GC, but in currently code, we didn't realize
it. When GC runs, front side IO would be blocked until the GC over, it
would be a long time if there is a lot of btree nodes.
This patch realizes incremental GC, the main ideal is that, when there
are front side I/Os, after GC some nodes (100), we stop GC, release locker
of the btree node, and go to process the front side I/Os for some times
(100 ms), then go back to GC again.
By this patch, when we doing GC, I/Os are not blocked all the time, and
there is no obvious I/Os zero jump problem any more.
Patch v2: Rename some variables and macros name as Coly suggested.
bcache: simplify the calculation of the total amount of flash dirty data
Currently we calculate the total amount of flash only devices dirty data
by adding the dirty data of each flash only device under registering
locker. It is very inefficient.
In this patch, we add a member flash_dev_dirty_sectors in struct cache_set
to record the total amount of flash only devices dirty data in real time,
so we didn't need to calculate the total amount of dirty data any more.
ondemand_readahead() checks bdi->io_pages to cap the maximum pages
that need to be processed. This works until the readit section. If
we would do an async only readahead (async size = sync size) and
target is at beginning of window we expand the pages by another
get_next_ra_size() pages. Btrace for large reads shows that kernel
always issues a doubled size read at the beginning of processing.
Add an additional check for io_pages in the lower part of the func.
The fix helps devices that hard limit bio pages and rely on proper
handling of max_hw_read_sectors (e.g. older FusionIO cards). For
that reason it could qualify for stable.
Fixes: 9491ae4a ("mm: don't cap request size based on read-ahead setting") Cc: stable@vger.kernel.org Signed-off-by: Markus Stockhausen stockhausen@collogia.de Signed-off-by: Jens Axboe <axboe@kernel.dk>
scsi: virtio_scsi: fix pi_bytes{out,in} on 4 KiB block size devices
When the underlying device is a 4 KiB logical block size device with a
protection interval exponent of 0, i.e. 4096 bytes data + 8 bytes PI, the
driver miscalculates the pi_bytes{out,in} by a factor of 8x (64 bytes).
This leads to errors on all reads and writes on 4 KiB logical block size
devices when CONFIG_BLK_DEV_INTEGRITY is enabled and the
VIRTIO_SCSI_F_T10_PI feature bit has been negotiated.
Fixes: e6dc783a38ec0 ("virtio-scsi: Enable DIF/DIX modes in SCSI host LLD") Acked-by: Martin K. Petersen <martin.petersen@oracle.com> Signed-off-by: Greg Edwards <gedwards@ddn.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
Merge branch 'nvme-4.19' of git://git.infradead.org/nvme into for-4.19/block
Pull NVMe updates from Christoph:
"Highlights:
- massively improved tracepoints (Keith Busch)
- support for larger inline data in the RDMA host and target
(Steve Wise)
- RDMA setup/teardown path fixes and refactor (Sagi Grimberg)
- Command Supported and Effects log support for the NVMe target
(Chaitanya Kulkarni)
- buffered I/O support for the NVMe target (Chaitanya Kulkarni)
plus the usual set of cleanups and small enhancements."
* 'nvme-4.19' of git://git.infradead.org/nvme:
nvmet: don't use uuid_le type
nvmet: check fileio lba range access boundaries
nvmet: fix file discard return status
nvme-rdma: centralize admin/io queue teardown sequence
nvme-rdma: centralize controller setup sequence
nvme-rdma: unquiesce queues when deleting the controller
nvme-rdma: mark expected switch fall-through
nvme: add disk name to trace events
nvme: add controller name to trace events
nvme: use hw qid in trace events
nvme: cache struct nvme_ctrl reference to struct nvme_request
nvmet-rdma: add an error flow for post_recv failures
nvmet-rdma: add unlikely check in the fast path
nvmet-rdma: support max(16KB, PAGE_SIZE) inline data
nvme-rdma: support up to 4 segments of inline data
nvmet: add buffered I/O support for file backed ns
nvmet: add commands supported and effects log page
nvme: move init of keep_alive work item to controller initialization
nvme.h: resync with nvme-cli
Fixes: 1e739730c5b9e ("block: optionally merge discontiguous discard bios into a single request") Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
We immediately overwrite the biovec array, so instead just allocate
a new bio and copy over the disk, setor and size.
Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Ming Lei <ming.lei@redhat.com> Acked-by: Coly Li <colyli@suse.de> Signed-off-by: Jens Axboe <axboe@kernel.dk>
bio_check_pages_dirty currently inviolates the invariant that bv_page of
a bio_vec inside bi_vcnt shouldn't be zero, and that is going to become
really annoying with multpath biovecs. Fortunately there isn't any
all that good reason for it - once we decide to defer freeing the bio
to a workqueue holding onto a few additional pages isn't really an
issue anymore. So just check if there is a clean page that needs
dirtying in the first path, and do a second pass to free them if there
was none, while the cache is still hot.
Also use the chance to micro-optimize bio_dirty_fn a bit by not saving
irq state - we know we are called from a workqueue.
Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
block: Rename the null_blk_mod kernel module back into null_blk
Commit ca4b2a011948 ("null_blk: add zone support") breaks several
blktests scripts because it renamed the null_blk kernel module into
null_blk_mod. Hence rename null_blk_mod back into null_blk.
Fixes: ca4b2a011948 ("null_blk: add zone support") Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com> Cc: Matias Bjorling <matias.bjorling@wdc.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Ming Lei <ming.lei@redhat.com> Cc: Damien Le Moal <damien.lemoal@wdc.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
Andy Shevchenko [Tue, 17 Jul 2018 14:17:36 +0000 (17:17 +0300)]
nvmet: don't use uuid_le type
Don't use sizeof(uuid_le) where none of the parameters is type of uuid_le.
Since both arguments are u8 [16], use size of destination there.
Moreover, uuid_le is a deprecated type, and nvmet is using uuid_t
already.
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de> Signed-off-by: Christoph Hellwig <hch@lst.de>
Centralize controller sequence to a single routine that correctly cleans
up after failures instead of having multiple apperances in several flows
(create, reset, reconnect).
One thing that we also gain here are the sanity/boundary checks also
when connecting back to a dynamic controller.
Signed-off-by: Sagi Grimberg <sagi@grimberg.me> Signed-off-by: Christoph Hellwig <hch@lst.de>
nvme-rdma: unquiesce queues when deleting the controller
If the controller is going away, we need to unquiesce the IO queues so
that all pending request can fail gracefully before moving forward with
controller deletion. Do that before we destroy the IO queues so
blk_cleanup_queue won't block in freeze.
Signed-off-by: Sagi Grimberg <sagi@grimberg.me> Signed-off-by: Christoph Hellwig <hch@lst.de>
Keith Busch [Fri, 29 Jun 2018 22:50:03 +0000 (16:50 -0600)]
nvme: add disk name to trace events
This will print the disk name to the nvme event trace for io requests so
a user can better distinguish traffic to different disks. This can be used
to create disk based filters. For example, to see only nvme0n2 traffic:
Signed-off-by: Keith Busch <keith.busch@intel.com> Reviewed-by: Sagi Grimberg <sagi@grimberg.me> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
[hch: turned __assign_disk_name into an inline function] Signed-off-by: Christoph Hellwig <hch@lst.de>
Keith Busch [Fri, 29 Jun 2018 22:50:01 +0000 (16:50 -0600)]
nvme: use hw qid in trace events
We can not match a command to its completion based on the command
id alone. We need the submitting queue identifier to pair with the
completion, so this patch adds that to the trace buffer.
This patch is also collapsing the admin and IO submission traces into a
single one so we don't need to duplicate this and creating unnecessary
code branches: we know if the command is an admin vs IO based on the qid.
And since we're here, the patch fixes code formatting in the area.
Signed-off-by: Keith Busch <keith.busch@intel.com> Reviewed-by: Sagi Grimberg <sagi@grimberg.me> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
[hch: move the qid helper to nvme.h and made it an inline function] Signed-off-by: Christoph Hellwig <hch@lst.de>
Max Gurtovoy [Sun, 1 Jul 2018 09:20:24 +0000 (12:20 +0300)]
nvmet-rdma: add an error flow for post_recv failures
Posting receive buffer operation can fail, thus we should make
sure to have an error flow during initialization phase. While
we're here, add a debug print in case of a failure.
Signed-off-by: Max Gurtovoy <maxg@mellanox.com> Signed-off-by: Christoph Hellwig <hch@lst.de>
Steve Wise [Wed, 20 Jun 2018 14:15:10 +0000 (07:15 -0700)]
nvmet-rdma: support max(16KB, PAGE_SIZE) inline data
The patch enables inline data sizes using up to 4 recv sges, and capping
the size at 16KB or at least 1 page size. So on a 4K page system, up to
16KB is supported, and for a 64K page system 1 page of 64KB is supported.
We avoid > 0 order page allocations for the inline buffers by using
multiple recv sges, one for each page. If the device cannot support
the configured inline data size due to lack of enough recv sges, then
log a warning and reduce the inline size.
Add a new configfs port attribute, called param_inline_data_size,
to allow configuring the size of inline data for a given nvmf port.
The maximum size allowed is still enforced by nvmet-rdma with
NVMET_RDMA_MAX_INLINE_DATA_SIZE, which is now max(16KB, PAGE_SIZE).
And the default size, if not specified via configfs, is still PAGE_SIZE.
This preserves the existing behavior, but allows larger inline sizes
for small page systems. If the configured inline data size exceeds
NVMET_RDMA_MAX_INLINE_DATA_SIZE, a warning is logged and the size is
reduced. If param_inline_data_size is set to 0, then inline data is
disabled for that nvmf port.
Reviewed-by: Sagi Grimberg <sagi@grimberg.me> Reviewed-by: Max Gurtovoy <maxg@mellanox.com> Signed-off-by: Steve Wise <swise@opengridcomputing.com> Signed-off-by: Christoph Hellwig <hch@lst.de>
Steve Wise [Wed, 20 Jun 2018 14:15:05 +0000 (07:15 -0700)]
nvme-rdma: support up to 4 segments of inline data
Allow up to 4 segments of inline data for NVMF WRITE operations. This
reduces latency for small WRITEs by removing the need for the target to
issue a READ WR for IB, or a REG_MR + READ WR chain for iWarp.
Also cap the inline segments used based on the limitations of the
device.
Reviewed-by: Sagi Grimberg <sagi@grimberg.me> Reviewed-by: Max Gurtovoy <maxg@mellanox.com> Signed-off-by: Steve Wise <swise@opengridcomputing.com> Signed-off-by: Christoph Hellwig <hch@lst.de>
nvmet: add buffered I/O support for file backed ns
Add a new "buffered_io" attribute, which disabled direct I/O and thus
enables page cache based caching when enabled. The attribute can only
be changed when the namespace is disabled as the file has to be reopend
for the change to take effect.
The possibly blocking read/write are deferred to a newly introduced
global workqueue.
Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com> Signed-off-by: Christoph Hellwig <hch@lst.de>
nvmet: add commands supported and effects log page
This patch adds support for Commands Supported and Effects log page
(Log Identifier 05h) for NVMeOF. This also makes it easier to find
which commands are supported, e.g. :-
James Smart [Tue, 12 Jun 2018 23:28:24 +0000 (16:28 -0700)]
nvme: move init of keep_alive work item to controller initialization
Currently, the code initializes the keep alive work item whenever
nvme_start_keep_alive() is called. However, this routine is called
several times while reconnecting, etc. Although it's hoped that keep
alive is always disabled and not scheduled when start is called,
re-initing if it were scheduled or completing can have very bad
side effects. There's no need for re-initialization.
Move the keep_alive work item and cmd struct initialization to
controller init.
Signed-off-by: James Smart <james.smart@broadcom.com> Signed-off-by: Christoph Hellwig <hch@lst.de>
Ming Lei [Sun, 22 Jul 2018 06:10:15 +0000 (14:10 +0800)]
blk-mq: fail the request in case issue failure
Inside blk_mq_try_issue_list_directly(), if the request is issued as
failed, we shouldn't try to do it again, otherwise the warning in
blk_mq_start_request() will be triggered. This change is aligned to
behaviour of other ways of request issue & dispatch.
Fixes: 6ce3dd6eec1 ("blk-mq: issue directly if hw queue isn't busy in case of 'none'") Cc: Kashyap Desai <kashyap.desai@broadcom.com> Cc: Laurence Oberman <loberman@redhat.com> Cc: Omar Sandoval <osandov@fb.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Bart Van Assche <bart.vanassche@wdc.com> Cc: Hannes Reinecke <hare@suse.de> Cc: Kashyap Desai <kashyap.desai@broadcom.com> Cc: kernel test robot <rong.a.chen@intel.com> Cc: LKP <lkp@01.org> Reported-by: kernel test robot <rong.a.chen@intel.com> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
Merge tag 'nvme-for-4.18' of git://git.infradead.org/nvme
Pull NVMe fixes from Christoph Hellwig:
- fix a regression in 4.18 that causes a memory leak on probe failure
(Keith Bush)
- fix a deadlock in the passthrough ioctl code (Scott Bauer)
- don't enable AENs if not supported (Weiping Zhang)
- fix an old regression in metadata handling in the passthrough ioctl
code (Roland Dreier)
* tag 'nvme-for-4.18' of git://git.infradead.org/nvme:
nvme: fix handling of metadata_len for NVME_IOCTL_IO_CMD
nvme: don't enable AEN if not supported
nvme: ensure forward progress during Admin passthru
nvme-pci: fix memory leak on probe failure
projects / kernel.git / log