dm raid: enhance reshape check and factor out reshape setup

Enhance rs_reshape_requested() check function to be more transparent and
fix its raid10 check.

Streamline the constructor by factoring out reshaping preparation into
fucntion rs_prepare_reshape().

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: allow resize during recovery

Resizing a RAID set during recovery can be allowed, because the MD
resynchronization thread will either stop any ongoing recovery in case
of shrinking below the current recovery position or carry on recovery
to the new size if the set is growing.

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: fix rs_is_recovering() to allow for lvextend

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: fix rebuild and catch bogus sync/resync flags

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: fix ctr memory leaks on error paths

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: fix typo in write_mostly flag

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: also reject size change during recovery

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: fix new superblock/bitmap creation on disk addition

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: add comments and fix typos

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: fix raid10 device size error on out-of-place reshape

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: prohibit 'nosync' on new raid6 and reject resize during reshape

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: clarify and fix recovery

Add function rs_setup_recovery() to allow for defined setup of RAID set
recovery in the constructor.

Will be called with dev_sectors={0, rdev->sectors, MaxSectors} to
recover a new or enforced sync, grown or not to be synhronized RAID set
respectively.

Prevents recovery on raid0, which doesn't support it.

Enforces recovery on raid6 to ensure properly defined Syndromes
mandatory for that MD personality are being created.

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: fix rs_set_capacity on growing reshape

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: make rs_set_capacity to work on shrinking reshape

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: enhance comments in takeover checks

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: remove bogus comment and fix comment typos

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: more restricting data_offset value checks

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: reject too many write_mostly devices

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: the sync_page_io() metadata_op argument is bool

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: prohibit to pass in both sync and nosync ctr flags

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: avoid superfluous memory barriers on static metadata

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm rq: check kthread_run return for .request_fn request-based DM

Check return value of kthread_run() in dm_old_init_request_queue().

Reported-by: Minfei Huang <mnghuan@gmail.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm verity fec: fix block calculation

do_div was replaced with div64_u64 at some point, causing a bug with
block calculation due to incompatible semantics of the two functions.

Signed-off-by: Sami Tolvanen <samitolvanen@google.com>
Fixes: a739ff3f543a ("dm verity: add support for forward error correction")
Cc: stable@vger.kernel.org # v4.5+
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm ioctl: Simplify parameter buffer management code

Merge the two DM_PARAMS_[KV]MALLOC flags into a single flag.

Doing so avoids the crashes seen with previous attempts to consolidate
buffer management to use kvfree() without first flagging that memory had
actually been allocated.

Signed-off-by: Bart Van Assche <bart.vanassche@sandisk.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm crypt: Fix sparse complaints

Avoid that sparse complains about assigning a __le64 value to a u64
variable.  Remove the (u64) casts since these are superfluous.  This
patch does not change the behavior of the source code.

Signed-off-by: Bart Van Assche <bart.vanassche@sandisk.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: don't use 'const' in function return

A newly introduced function has 'const int' as the return type,
but as "make W=1" reports, that has no meaning:

drivers/md/dm-raid.c:510:18: error: type qualifiers ignored on function return type [-Werror=ignored-qualifiers]

This changes the return type to plain 'int'.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Fixes: 33e53f06850f ("dm raid: introduce extended superblock and new raid types to support takeover/reshaping")
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: fix failed takeover/reshapes by keeping raid set frozen

Superblock updates where bogus causing some takovers/reshapes to fail.

Introduce new runtime flag (RT_FLAG_KEEP_RS_FROZEN) to keep a raid set
frozen when a layout change was requested.  Userpace will immediately
reload the table w/o the flags requesting such change once they made it
to the superblocks and any change of recovery/reshape offsets has to be
avoided until after read.

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: support to change bitmap region size

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: update Documentation about reshaping/takeover/additonal RAID types

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: add reshaping support to the target

Add bool functions rs_is_recovering and rs_is_reshaping()
to test for ongoing recovery/reshaping respectively in order
to reject respective requests on ongoing ones.

Remove ctr array size check, because ti->len and array
sectors will differ during disk addition/removal reshape.

Use __is_raid10_near() rather than type string compare.

Introduce rs_check_reshape() and rs_start_reshape(),
use the former in the ctr to reject bogus rehsape requests
and the latter in preresume to actually start a reshape.

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: add prerequisite functions and definitions for reshaping

Add rs_is_reshapable(), rs_data_stripes(), rs_reshape_requested(),
rs_set_dev_and_array_sectors() and rs_adjust_data_offsets()

Remove superfluous check for reshape message

Correct runtime bit definitions to be incremental

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: inverse check for flags from invalid to valid flags

It is more intuitive to manage each raid level's features in terms of
what is supported rather than what isn't supported.

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: various code cleanups

Renamed functions and variables with leading single underscore to have a
double underscore.  Renamed some functions to have better names.  Folded
functions that were split out without reason.

Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: rename functions that alloc and free struct raid_set

Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: remove all the bitops wrappers

Removes obfuscation that is of little value.

Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: rename _in_range to __within_range

Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: add missing "dm-raid0" module alias

Also update module description to "raid0/1/10/4/5/6 target"

Reported by Alasdair G Kergon <agk@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: rename _argname_by_flag to dm_raid_arg_name_by_flag

Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: bump to v1.9.0 and make the extended SB feature flag reflect it

No idea what Heinz was doing with the versioning but upstream commit
4c9971ca6a ("dm raid: make sure no feature flags are set in metadata")
bumped to 1.8.0 already.

Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: remove ti_error_* wrappers

There ti_error_* wrappers added very little.  No other DM target has
ever gone to such lengths to wrap setting ti->error.

Also fixes some NULL derefences via rs->ti->error.

Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: tabify appropriate whitespace

Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: enhance status interface and fixup takeover/raid0

The target's status interface has to provide the new 'data_offset' value
to allow userspace to retrieve the kernels offset to the data on each
raid device of a raid set.  This is the base for out-of-place reshaping
required to not write over any data during reshaping (e.g. change
raid6_zr -> raid6_nc):

 - add rs_set_cur() to be able to start up existing array in case of no
   takeover; use in ctr on takeover check

 - enhance raid_status()

 - add supporting functions to get resync/reshape progress and raid
   device status chars

 - fixup rebuild table line output race, which does miss to emit
   'rebuild N' on fully synced/rebuild devices, because it is relying on
   the transient 'In_sync' raid device flag

 - add new status line output for 'data_offset', which'll later be used
   for out-of-place reshaping

 - fixup takeover not working for all levels

 - fixup raid0 message interface oops caused by missing checks
   for the md threads, which don't exist in case of raid0

 - remove ALL_FREEZE_FLAGS not needed for takeover

 - adjust comments

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: add raid level takeover support

Add raid level takeover support allowing arbitrary takeovers between
raid levels supported by md personalities (i.e. raid0, raid1/10 and
raid4/5/6):

 - add rs_config_{backup|restore} function to allow for temporary
   storing ctr requested layout changes and restore them for takeover
   conersion decision after the superblocks got loaded and analyzed

 - add members to store layout to 'struct raid_set' (not mandatory
   for takeover but needed for reshape in later patch)

 - add rebuild_disks bitfield to 'struct raid_set' and set bits in ctr
   to use in setting up takeover (base to address a 'rebuild' related
   raid_status() table line bug and needed as well for reshape in future
   patch)

 - add runtime flags and respective manipulation functions to be able to
   control e.g. wrting of superlocks to the preresume function on
   takeover and (later) reshape

 - add functions to detect takeover, check it's valid (mandatory here to
   avoid failing on md_run()), setup for it and use in the ctr; those
   will be likely moved out once reshaping gets added to simplify the
   ctr

 - start raid set readonly in ctr and switch to readwrite, optionally
   updating superblocks, in preresume in order to allow suspend to
   quiesce any active table before (which involves superblock updates);
   this ensures the proper sequence of writing the current and any new
   takeover(/reshape) metadata

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: enhance super_sync() to support new superblock members

Add transferring the new takeover/reshape related superblock
members introduced to the super_sync() function:

 - add/move supporting functions

 - add failed devices bitfield transfer functions to retrieve the
   bitfield from superblock format or update it in the superblock

 - add code to transfer all new members

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: add new reshaping/raid10 format table line options to parameter parser

Support the follwoing arguments in the ctr parameter parser:

 - add 'delta_disks', 'data_offset' taking int and sector respectively

 - 'raid10_use_near_sets' bool argument to optionally select
   near sets with supporting raid10 mappings

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: introduce extended superblock and new raid types to support takeover/reshaping

Add new members to the dm-raid superblock and new raid types to support
takeover/reshape.

Add all necessary members needed to support takeover and reshape in one
go -- aiming to limit the amount of changes to the superblock layout.

This is a larger patch due to the new superblock members, their related
flags, validation of both and involved API additions/changes:

 - add additional members to keep track of:
   - state about forward/backward reshaping
   - reshape position
   - new level, layout, stripe size and delta disks
   - data offset to current and new data for out-of-place reshapes
   - failed devices bitfield extensions to keep track of max raid devices

 - adjust super_validate() to cope with new superblock members

 - adjust super_init_validation() to cope with new superblock members

 - add definitions for ctr flags supporting delta disks etc.

 - add new raid types (raid6_n_6 etc.)

 - add new raid10 supporting function API (_is_raid10_*())

 - adjust to changed raid10 supporting function API

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: use rt_is_raid*() in all appropriate checks

Make use if raid type rt_is_*() bool functions for simplification and
consistency reasons.

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: more use of flag testing wrappers

 - add _test_flags() function

 - use it to simplify rs_check_for_invalid_flags()

 - use _test_flag() throughout

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: check constructor arguments for invalid raid level/argument combinations

Reject invalid flag combinations to avoid potential data corruption or
failing raid set construction:

 - add definitions for constructor flag combinations and invalid flags
   per level

 - add bool test functions for the various raid types
   (also will be used by future reshaping enhancements)

 - introduce rs_check_for_invalid_flags() and _invalid_flags()
   to perform the validity checks

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: cleanup / provide infrastructure

Provide necessary infrastructure to handle ctr flags and their names
and cleanup setting ti->error:

 - comment constructor flags

 - introduce constructor flag manipulation

 - introduce ti_error_*() functions to simplify
   setting the error message (use in other targets?)

 - introduce array to hold ctr flag <-> flag name mapping

 - introduce argument name by flag functions for that array

 - use those functions throughout the ctr call path

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: use dm_arg_set API in constructor

- use dm_arg_set API in ctr and its callees parse_raid_params() and dev_parms()

- introduce _in_range() function to check a value is in a [ min, max ] range;
  this is to support more callers in parsing parameters etc. in the future

- correct comment on MAX_RAID_DEVICES

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm raid: rename variable 'ret' to 'r' to conform to other dm code

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm mpath: add optional "queue_mode" feature

Allow a user to specify an optional feature 'queue_mode <mode>' where
<mode> may be "bio", "rq" or "mq" -- which corresponds to bio-based,
request_fn rq-based, and blk-mq rq-based respectively.

If the queue_mode feature isn't specified the default for the
"multipath" target is still "rq" but if dm_mod.use_blk_mq is set to Y
it'll default to mode "mq".

This new queue_mode feature introduces the ability for each multipath
device to have its own queue_mode (whereas before this feature all
multipath devices effectively had to have the same queue_mode).

This commit also goes a long way to eliminate the awkward (ab)use of
DM_TYPE_*, the associated filter_md_type() and other relatively fragile
and difficult to maintain code.

Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm mpath: remove bio-based bloat from struct dm_mpath_io

Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm mpath: reinstate bio-based support

Add "multipath-bio" target that offers a bio-based multipath target as
an alternative to the request-based "multipath" target -- but in a
following commit "multipath-bio" will immediately be replaced by a new
"queue_mode" feature for the "multipath" target which will allow
bio-based mode to be selected.

When DM multipath was originally converted from bio-based to
request-based the motivation for the change was better dynamic load
balancing (by leveraging block core's request-based IO schedulers, for
merging and sorting, _before_ DM multipath would make the decision on
where to steer the IO -- based on path load and/or availability).

More background is available in this "Request-based Device-mapper
multipath and Dynamic load balancing" paper:
https://www.kernel.org/doc/ols/2007/ols2007v2-pages-235-244.pdf

But we've now come full circle where significantly faster storage
devices no longer need IOs to be made larger to drive optimal IO
performance.  And even if they do there have been changes to the block
and filesystem layers that help ensure upper layers are constructing
larger IOs.  In addition, SCSI's differentiated IO errors will propagate
through to bio-based IO completion hooks -- so that eliminates another
historic justiciation for request-based DM multipath.  Lastly, the block
layer's immutable biovec changes have made bio cloning cheaper than it
has ever been; whereas request cloning is still relatively expensive
(both on a CPU usage and memory footprint level).

As such, bio-based DM multipath offers the promise of a more efficient
IO path for high IOPs devices that are, or will be, emerging.

Signed-off-by: Mike Snitzer <snitzer@redhat.com>

dm: move request-based code out to dm-rq.[hc]

Add some seperation between bio-based and request-based DM core code.

'struct mapped_device' and other DM core only structures and functions
have been moved to dm-core.h and all relevant DM core .c files have been
updated to include dm-core.h rather than dm.h

DM targets should _never_ include dm-core.h!

[block core merge conflict resolution from Stephen Rothwell]
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>

block: bio: kill BIO_MAX_SIZE

No one need this macro now, so remove it. Basically
only how many bvecs in one bio matters instead
of how many bytes in this bio.

The motivation is for supporting multipage bvecs, in
which we only know what the max count of bvecs is supported
in the bio.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Jens Axboe <axboe@fb.com>

cfq-iosched: temporarily boost queue priority for idle classes

If we're queuing REQ_PRIO IO and the task is running at an idle IO
class, then temporarily boost the priority. This prevents livelocks
due to priority inversion, when a low priority task is holding file
system resources while attempting to do IO.

An example of that is shown below. An ioniced idle task is holding
the directory mutex, while a normal priority task is trying to do
a directory lookup.

[478381.198925] ------------[ cut here ]------------
[478381.200315] INFO: task ionice:1168369 blocked for more than 120 seconds.
[478381.201324]       Not tainted 4.0.9-38_fbk5_hotfix1_2936_g85409c6 #1
[478381.202278] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
[478381.203462] ionice          D ffff8803692736a8     0 1168369      1 0x00000080
[478381.203466]  ffff8803692736a8 ffff880399c21300 ffff880276adcc00 ffff880369273698
[478381.204589]  ffff880369273fd8 0000000000000000 7fffffffffffffff 0000000000000002
[478381.205752]  ffffffff8177d5e0 ffff8803692736c8 ffffffff8177cea7 0000000000000000
[478381.206874] Call Trace:
[478381.207253]  [<ffffffff8177d5e0>] ? bit_wait_io_timeout+0x80/0x80
[478381.208175]  [<ffffffff8177cea7>] schedule+0x37/0x90
[478381.208932]  [<ffffffff8177f5fc>] schedule_timeout+0x1dc/0x250
[478381.209805]  [<ffffffff81421c17>] ? __blk_run_queue+0x37/0x50
[478381.210706]  [<ffffffff810ca1c5>] ? ktime_get+0x45/0xb0
[478381.211489]  [<ffffffff8177c407>] io_schedule_timeout+0xa7/0x110
[478381.212402]  [<ffffffff810a8c2b>] ? prepare_to_wait+0x5b/0x90
[478381.213280]  [<ffffffff8177d616>] bit_wait_io+0x36/0x50
[478381.214063]  [<ffffffff8177d325>] __wait_on_bit+0x65/0x90
[478381.214961]  [<ffffffff8177d5e0>] ? bit_wait_io_timeout+0x80/0x80
[478381.215872]  [<ffffffff8177d47c>] out_of_line_wait_on_bit+0x7c/0x90
[478381.216806]  [<ffffffff810a89f0>] ? wake_atomic_t_function+0x40/0x40
[478381.217773]  [<ffffffff811f03aa>] __wait_on_buffer+0x2a/0x30
[478381.218641]  [<ffffffff8123c557>] ext4_bread+0x57/0x70
[478381.219425]  [<ffffffff8124498c>] __ext4_read_dirblock+0x3c/0x380
[478381.220467]  [<ffffffff8124665d>] ext4_dx_find_entry+0x7d/0x170
[478381.221357]  [<ffffffff8114c49e>] ? find_get_entry+0x1e/0xa0
[478381.222208]  [<ffffffff81246bd4>] ext4_find_entry+0x484/0x510
[478381.223090]  [<ffffffff812471a2>] ext4_lookup+0x52/0x160
[478381.223882]  [<ffffffff811c401d>] lookup_real+0x1d/0x60
[478381.224675]  [<ffffffff811c4698>] __lookup_hash+0x38/0x50
[478381.225697]  [<ffffffff817745bd>] lookup_slow+0x45/0xab
[478381.226941]  [<ffffffff811c690e>] link_path_walk+0x7ae/0x820
[478381.227880]  [<ffffffff811c6a42>] path_init+0xc2/0x430
[478381.228677]  [<ffffffff813e6e26>] ? security_file_alloc+0x16/0x20
[478381.229776]  [<ffffffff811c8c57>] path_openat+0x77/0x620
[478381.230767]  [<ffffffff81185c6e>] ? page_add_file_rmap+0x2e/0x70
[478381.232019]  [<ffffffff811cb253>] do_filp_open+0x43/0xa0
[478381.233016]  [<ffffffff8108c4a9>] ? creds_are_invalid+0x29/0x70
[478381.234072]  [<ffffffff811c0cb0>] do_open_execat+0x70/0x170
[478381.235039]  [<ffffffff811c1bf8>] do_execveat_common.isra.36+0x1b8/0x6e0
[478381.236051]  [<ffffffff811c214c>] do_execve+0x2c/0x30
[478381.236809]  [<ffffffff811ca392>] ? getname+0x12/0x20
[478381.237564]  [<ffffffff811c23be>] SyS_execve+0x2e/0x40
[478381.238338]  [<ffffffff81780a1d>] stub_execve+0x6d/0xa0
[478381.239126] ------------[ cut here ]------------
[478381.239915] ------------[ cut here ]------------
[478381.240606] INFO: task python2.7:1168375 blocked for more than 120 seconds.
[478381.242673]       Not tainted 4.0.9-38_fbk5_hotfix1_2936_g85409c6 #1
[478381.243653] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
[478381.244902] python2.7       D ffff88005cf8fb98     0 1168375 1168248 0x00000080
[478381.244904]  ffff88005cf8fb98 ffff88016c1f0980 ffffffff81c134c0 ffff88016c1f11a0
[478381.246023]  ffff88005cf8ffd8 ffff880466cd0cbc ffff88016c1f0980 00000000ffffffff
[478381.247138]  ffff880466cd0cc0 ffff88005cf8fbb8 ffffffff8177cea7 ffff88005cf8fcc8
[478381.248252] Call Trace:
[478381.248630]  [<ffffffff8177cea7>] schedule+0x37/0x90
[478381.249382]  [<ffffffff8177d08e>] schedule_preempt_disabled+0xe/0x10
[478381.250465]  [<ffffffff8177e892>] __mutex_lock_slowpath+0x92/0x100
[478381.251409]  [<ffffffff8177e91b>] mutex_lock+0x1b/0x2f
[478381.252199]  [<ffffffff817745ae>] lookup_slow+0x36/0xab
[478381.253023]  [<ffffffff811c690e>] link_path_walk+0x7ae/0x820
[478381.253877]  [<ffffffff811aeb41>] ? try_charge+0xc1/0x700
[478381.254690]  [<ffffffff811c6a42>] path_init+0xc2/0x430
[478381.255525]  [<ffffffff813e6e26>] ? security_file_alloc+0x16/0x20
[478381.256450]  [<ffffffff811c8c57>] path_openat+0x77/0x620
[478381.257256]  [<ffffffff8115b2fb>] ? lru_cache_add_active_or_unevictable+0x2b/0xa0
[478381.258390]  [<ffffffff8117b623>] ? handle_mm_fault+0x13f3/0x1720
[478381.259309]  [<ffffffff811cb253>] do_filp_open+0x43/0xa0
[478381.260139]  [<ffffffff811d7ae2>] ? __alloc_fd+0x42/0x120
[478381.260962]  [<ffffffff811b95ac>] do_sys_open+0x13c/0x230
[478381.261779]  [<ffffffff81011393>] ? syscall_trace_enter_phase1+0x113/0x170
[478381.262851]  [<ffffffff811b96c2>] SyS_open+0x22/0x30
[478381.263598]  [<ffffffff81780532>] system_call_fastpath+0x12/0x17
[478381.264551] ------------[ cut here ]------------
[478381.265377] ------------[ cut here ]------------

Signed-off-by: Jens Axboe <axboe@fb.com>
Reviewed-by: Jeff Moyer <jmoyer@redhat.com>

block: drbd: avoid to use BIO_MAX_SIZE

Use BIO_MAX_PAGES instead and we will remove BIO_MAX_SIZE.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Tested-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>

block: bio: remove BIO_MAX_SECTORS

No one need this macro, so remove it. The motivation is for supporting
multipage bvecs, in which we only know what the max count of bvecs is
supported in the bio, instead of max size or max sectors.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Tested-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>

fs: xfs: replace BIO_MAX_SECTORS with BIO_MAX_PAGES

BIO_MAX_PAGES is used as maximum count of bvecs, so
replace BIO_MAX_SECTORS with BIO_MAX_PAGES since
BIO_MAX_SECTORS is to be removed.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Tested-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>

iov_iter: use bvec iterator to implement iterate_bvec()

bvec has one native/mature iterator for long time, so not
necessary to use the reinvented wheel for iterating bvecs
in lib/iov_iter.c.

Two ITER_BVEC test cases are run:
- xfstest(-g auto) on loop dio/aio, no regression found
- swap file works well under extreme stress(stress-ng --all 64 -t
  800 -v), and lots of OOMs are triggerd, and the whole
system still survives

Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Tested-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>

block: mark 1st parameter of bvec_iter_advance as const

bvec_iter_advance() only writes the parameter of iterator,
so the base address of bvec can be marked as const safely.

Without the change, we can see compiling warning in the
following patch for implementing iterate_bvec(): lib/iov_iter.c
with bvec iterator.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Tested-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>

block: move two bvec structure into bvec.h

This patch moves 'struct bio_vec' and 'struct bvec_iter'
into 'include/linux/bvec.h', then always include this header
into 'include/linux/blk_types.h'.

With this change, both 'struct bvec_iter' and bvec iterator
helpers don't depend on CONFIG_BLOCK any more, then we can
use bvec iterator to implement iterate_bvec(): lib/iov_iter.c.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Suggested-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Tested-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>

block: move bvec iterator into include/linux/bvec.h

bvec iterator helpers should be used to implement by
iterate_bvec():lib/iov_iter.c too, and move them into
one header, so that we can keep bvec iterator header
out of CONFIG_BLOCK. Then we can remove the reinventing
of wheel in iterate_bvec().

Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Tested-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>

blk-mq: actually hook up defer list when running requests

If ->queue_rq() returns BLK_MQ_RQ_QUEUE_OK, we use continue and skip
over the rest of the loop body. However, dptr is assigned later in the
loop body, and the BLK_MQ_RQ_QUEUE_OK case is exactly the case that we'd
want it for.

NVMe isn't actually using BLK_MQ_F_DEFER_ISSUE yet, nor is any other
in-tree driver, but if the code's going to be there, it might as well
work.

Fixes: 74c450521dd8 ("blk-mq: add a 'list' parameter to ->queue_rq()")
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>

block: better packing for struct request

Keep the 32-bit CPU and cmd_type flags together to avoid holes on 64-bit
architectures.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>

ext4: use bio op helprs in ext4 crypto code

This was missed from my last patchset.

This patch has ext4 crypto code use the bio op helper
to set the operation. The operation (discard, write, writesame,
etc) is now defined seperately from the other REQ bits. They
still share the bi_rw field to save space, so we use these
helpers so modules do not have to worry about setting/overwriting
info.

Jens, I am not sure how you handle patches on top of patches
in the next branches. If you merge patches that fix issues
in previous patches in next, then this patch could be part
of

commit 95fe6c1a209ef89d9f94dd04a0ad72be1487d5d5
Author: Mike Christie <mchristi@redhat.com>
Date:   Sun Jun 5 14:31:48 2016 -0500

    block, fs, mm, drivers: use bio set/get op accessors

Signed-off-by: Mike Christie <mchristi@redhat.com>
Signed-off-by: Jens Axboe <axboe@fb.com>

cfq-iosched: Convert to use highres timers

Reviewed-by: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Jan Kara <jack@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>

cfq-iosched: Expose microsecond interfaces

Expose interfaces to tune time slices of CFQ IO scheduler in
microseconds.

Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@fb.com>

cfq-iosched: Convert from jiffies to nanoseconds

Convert all time-keeping in CFQ IO scheduler from jiffies to nanoseconds
so that we can later make the intervals more fine-grained than jiffies.
One jiffie is several miliseconds and even for today's rotating disks
that is a noticeable amount of time and thus we leave disk unnecessarily
idle.

Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@fb.com>

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>

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>

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>

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>

block, drivers, fs: shrink bi_rw from long to int

We don't need bi_rw to be so large on 64 bit archs, so
reduce it to unsigned int.

Signed-off-by: Mike Christie <mchristi@redhat.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>

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>

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>

blktrace: use op accessors

Have blktrace use the req/bio op accessor to get the REQ_OP.

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>

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>

block: convert is_sync helpers to use REQ_OPs.

This patch converts the is_sync helpers to use separate variables
for the operation and flags.

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>

block: convert merge/insert code to check for REQ_OPs.

This patch converts the block layer merging code to use separate variables
for the operation and flags, and to check req_op for the REQ_OP.

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>

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>

block: prepare elevator to use REQ_OPs.

This patch converts the elevator code to use separate variables
for the operation and flags, and to check req_op for the REQ_OP.

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>

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>

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>

block: copy bio op to request op

The bio users should now always be setting up the bio op. This patch
has the block layer copy that to the request.

Signed-off-by: Mike Christie <mchristi@redhat.com>
Signed-off-by: Jens Axboe <axboe@fb.com>

xen: use bio op accessors

Separate the op from the rq_flag_bits and have xen
set/get the bio using bio_set_op_attrs/bio_op.

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>

target: use bio op accessors

Separate the op from the rq_flag_bits and have the target layer
set/get the bio using bio_set_op_attrs/bio_op.

Signed-off-by: Mike Christie <mchristi@redhat.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>

md: use bio op accessors

Separate the op from the rq_flag_bits and have md
set/get the bio using bio_set_op_attrs/bio_op.

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>

drbd: use bio op accessors

Separate the op from the rq_flag_bits and have drbd
set/get the bio using bio_set_op_attrs/bio_op.

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>

bcache: use bio op accessors

Separate the op from the rq_flag_bits and have bcache
set/get the bio using bio_set_op_attrs/bio_op.

Signed-off-by: Mike Christie <mchristi@redhat.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>

dm: use bio op accessors

Separate the op from the rq_flag_bits and have dm
set/get the bio using bio_set_op_attrs/bio_op.

Signed-off-by: Mike Christie <mchristi@redhat.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>

dm: pass dm stats data dir instead of bi_rw

It looks like dm stats cares about the data direction
(READ vs WRITE) and does not need the bio/request flags.
Commands like REQ_FLUSH, REQ_DISCARD and REQ_WRITE_SAME
are currently always set with REQ_WRITE, so the extra check for
REQ_DISCARD in dm_stats_account_io is not needed.

This patch has it use the bio and request data_dir helpers
instead of accessing the bi_rw/cmd_flags directly. This makes
the next patches that remove the operation from the cmd_flags
and bi_rw easier, because we will no longer have the REQ_WRITE
bit set for operations like discards.

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>

pm: use bio op accessors

Separate the op from the rq_flag_bits and have the pm code
set/get the bio using bio_set_op_attrs/bio_op.

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>

ocfs2: use bio op accessors

Separate the op from the rq_flag_bits and have ocfs2
set/get the bio using bio_set_op_attrs/bio_op.

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>

nilfs: use bio op accessors

Separate the op from the rq_flag_bits and have nilfs
set/get the bio using bio_set_op_attrs/bio_op.

Signed-off-by: Mike Christie <mchristi@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Acked-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Signed-off-by: Jens Axboe <axboe@fb.com>

mpage: use bio op accessors

Separate the op from the rq_flag_bits and have the mpage code
set/get the bio using bio_set_op_attrs/bio_op.

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>

hfsplus: use bio op accessors

Separate the op from the rq_flag_bits and have gfs2
set/get the bio using bio_set_op_attrs/bio_op.

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>

xfs: use bio op accessors

Separate the op from the rq_flag_bits and have xfs
set/get the bio using bio_set_op_attrs/bio_op.

Signed-off-by: Mike Christie <mchristi@redhat.com>
Signed-off-by: Jens Axboe <axboe@fb.com>