This reverts commit 3c9fe8cd.

As Miklos points out in commit c1b2cc1a, the "lookup_hash()" helper
is now unused, and in fact, with the hash salting changes, since the
hash of a dentry name now depends on the directory dentry it is in, the
helper function isn't even really likely to be useful.

So rather than keep it around in case somebody else might end up finding
a use for it, let's just remove the helper and not trick people into
thinking it might be a useful thing.

For example, I had obviously completely missed how the helper didn't
follow the normal dentry hashing patterns, and how the hash salting
patch broke overlayfs.  Things would quietly build and look sane, but
not work.
Suggested-by: Miklos Szeredi <mszeredi@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Pull overlayfs update from Miklos Szeredi:
 "First of all, this fixes a regression in overlayfs introduced by the
  dentry hash salting.  I've moved the patch fixing this to the front of
  the queue, so if (god forbid) something needs to be bisected in
  overlayfs this regression won't interfere with that.

  The biggest part is preparation for selinux support, done by Vivek
  Goyal.  Essentially this makes all operations on underlying
  filesystems be done with credentials of mounter.  This makes
  everything nicely consistent.

  There are also fixes for a number of known and recently discovered
  non-standard behavior (thanks to Eryu Guan for testing and improving
  the test suites)"

* 'overlayfs-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mszeredi/vfs: (23 commits)
  ovl: simplify empty checking
  qstr: constify instances in overlayfs
  ovl: clear nlink on rmdir
  ovl: disallow overlayfs as upperdir
  ovl: fix warning
  ovl: remove duplicated include from super.c
  ovl: append MAY_READ when diluting write checks
  ovl: dilute permission checks on lower only if not special file
  ovl: fix POSIX ACL setting
  ovl: share inode for hard link
  ovl: store real inode pointer in ->i_private
  ovl: permission: return ECHILD instead of ENOENT
  ovl: update atime on upper
  ovl: fix sgid on directory
  ovl: simplify permission checking
  ovl: do not require mounter to have MAY_WRITE on lower
  ovl: do operations on underlying file system in mounter's context
  ovl: modify ovl_permission() to do checks on two inodes
  ovl: define ->get_acl() for overlay inodes
  ovl: move some common code in a function
  ...

Pull freevxfs updates from Christoph Hellwig:
 "Support for foreign endianess and HP-UP superblocks from
  Krzysztof Błaszkowski"

* tag 'freevxfs-for-4.8' of git://git.infradead.org/users/hch/freevxfs:
  freevxfs: update Kconfig information
  freevxfs: refactor readdir and lookup code
  freevxfs: fix lack of inode initialization
  freevxfs: fix memory leak in vxfs_read_fshead()
  freevxfs: update documentation and cresdits for HP-UX support
  freevxfs: implement ->alloc_inode and ->destroy_inode
  freevxfs: avoid the need for forward declaring the super operations
  freevxfs: move VFS inode allocation into vxfs_blkiget and vxfs_stiget
  freevxfs: remove vxfs_put_fake_inode
  freevxfs: handle big endian HP-UX file systems

Pull configfs update from Christoph Hellwig:
 "A simple error handling fix from Tal Shorer"

* tag 'configfs-for-4.8' of git://git.infradead.org/users/hch/configfs:
  configfs: don't set buffer_needs_fill to zero if show() returns error

Pull CIFS/SMB3 fixes from Steve French:
 "Various CIFS/SMB3 fixes, most for stable"

* 'for-next' of git://git.samba.org/sfrench/cifs-2.6:
  CIFS: Fix a possible invalid memory access in smb2_query_symlink()
  fs/cifs: make share unaccessible at root level mountable
  cifs: fix crash due to race in hmac(md5) handling
  cifs: unbreak TCP session reuse
  cifs: Check for existing directory when opening file with O_CREAT
  Add MF-Symlinks support for SMB 2.0

The empty checking logic is duplicated in ovl_check_empty_and_clear() and
ovl_remove_and_whiteout(), except the condition for clearing whiteouts is
different:

ovl_check_empty_and_clear() checked for being upper

ovl_remove_and_whiteout() checked for merge OR lower

Move the intersection of those checks (upper AND merge) into
ovl_check_empty_and_clear() and simplify ovl_remove_and_whiteout().
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

To make delete notification work on fa/inotify.
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

This does not work and does not make sense.  So instead of fixing it
(probably not hard) just disallow.
Reported-by: Andrei Vagin <avagin@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Cc: <stable@vger.kernel.org>

There's a superfluous newline in the warning message in ovl_d_real().
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

Remove duplicated include.
Signed-off-by: Wei Yongjun <yongjun_wei@trendmicro.com.cn>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

Right now we remove MAY_WRITE/MAY_APPEND bits from mask if realfile is on
lower/. This is done as files on lower will never be written and will be
copied up. But to copy up a file, mounter should have MAY_READ permission
otherwise copy up will fail. So set MAY_READ in mask when MAY_WRITE is
reset.

Dan Walsh noticed this when he did access(lowerfile, W_OK) and it returned
True (context mounts) but when he tried to actually write to file, it
failed as mounter did not have permission on lower file.

[SzM] don't set MAY_READ if only MAY_APPEND is set without MAY_WRITE; this
won't trigger a copy-up.
Reported-by: Dan Walsh <dwalsh@redhat.com>
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

Right now if file is on lower/, we remove MAY_WRITE/MAY_APPEND bits from
mask as lower/ will never be written and file will be copied up. But this
is not true for special files. These files are not copied up and are opened
in place. So don't dilute the checks for these types of files.
Reported-by: Dan Walsh <dwalsh@redhat.com>
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

Setting POSIX ACL needs special handling:

1) Some permission checks are done by ->setxattr() which now uses mounter's
creds ("ovl: do operations on underlying file system in mounter's
context").  These permission checks need to be done with current cred as
well.

2) Setting ACL can fail for various reasons.  We do not need to copy up in
these cases.

In the mean time switch to using generic_setxattr.

[Arnd Bergmann] Fix link error without POSIX ACL. posix_acl_from_xattr()
doesn't have a 'static inline' implementation when CONFIG_FS_POSIX_ACL is
disabled, and I could not come up with an obvious way to do it.

This instead avoids the link error by defining two sets of ACL operations
and letting the compiler drop one of the two at compile time depending
on CONFIG_FS_POSIX_ACL. This avoids all references to the ACL code,
also leading to smaller code.
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

Inode attributes are copied up to overlay inode (uid, gid, mode, atime,
mtime, ctime) so generic code using these fields works correcty.  If a hard
link is created in overlayfs separate inodes are allocated for each link.
If chmod/chown/etc. is performed on one of the links then the inode
belonging to the other ones won't be updated.

This patch attempts to fix this by sharing inodes for hard links.

Use inode hash (with real inode pointer as a key) to make sure overlay
inodes are shared for hard links on upper.  Hard links on lower are still
split (which is not user observable until the copy-up happens, see
Documentation/filesystems/overlayfs.txt under "Non-standard behavior").

The inode is only inserted in the hash if it is non-directoy and upper.
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

To get from overlay inode to real inode we currently use 'struct
ovl_entry', which has lifetime connected to overlay dentry.  This is okay,
since each overlay dentry had a new overlay inode allocated.

Following patch will break that assumption, so need to leave out ovl_entry.
This patch stores the real inode directly in i_private, with the lowest bit
used to indicate whether the inode is upper or lower.

Lifetime rules remain, using ovl_inode_real() must only be done while
caller holds ref on overlay dentry (and hence on real dentry), or within
RCU protected regions.
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

The error is due to RCU and is temporary.
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

Fix atime update logic in overlayfs.

This patch adds an i_op->update_time() handler to overlayfs inodes.  This
forwards atime updates to the upper layer only.  No atime updates are done
on lower layers.

Remove implicit atime updates to underlying files and directories with
O_NOATIME.  Remove explicit atime update in ovl_readlink().

Clear atime related mnt flags from cloned upper mount.  This means atime
updates are controlled purely by overlayfs mount options.

Reported-by: Konstantin Khlebnikov <koct9i@gmail.com> 
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

When creating directory in workdir, the group/sgid inheritance from the
parent dir was omitted completely.  Fix this by calling inode_init_owner()
on overlay inode and using the resulting uid/gid/mode to create the file.

Unfortunately the sgid bit can be stripped off due to umask, so need to
reset the mode in this case in workdir before moving the directory in
place.
Reported-by: Eryu Guan <eguan@redhat.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

The fact that we always do permission checking on the overlay inode and
clear MAY_WRITE for checking access to the lower inode allows cruft to be
removed from ovl_permission().

1) "default_permissions" option effectively did generic_permission() on the
overlay inode with i_mode, i_uid and i_gid updated from underlying
filesystem.  This is what we do by default now.  It did the update using
vfs_getattr() but that's only needed if the underlying filesystem can
change (which is not allowed).  We may later introduce a "paranoia_mode"
that verifies that mode/uid/gid are not changed.

2) splitting out the IS_RDONLY() check from inode_permission() also becomes
unnecessary once we remove the MAY_WRITE from the lower inode check.
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

Now we have two levels of checks in ovl_permission(). overlay inode
is checked with the creds of task while underlying inode is checked
with the creds of mounter.

Looks like mounter does not have to have WRITE access to files on lower/.
So remove the MAY_WRITE from access mask for checks on underlying
lower inode.

This means task should still have the MAY_WRITE permission on lower
inode and mounter is not required to have MAY_WRITE.

It also solves the problem of read only NFS mounts being used as lower.
If __inode_permission(lower_inode, MAY_WRITE) is called on read only
NFS, it fails. By resetting MAY_WRITE, check succeeds and case of
read only NFS shold work with overlay without having to specify any
special mount options (default permission).
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

Given we are now doing checks both on overlay inode as well underlying
inode, we should be able to do checks and operations on underlying file
system using mounter's context.

So modify all operations to do checks/operations on underlying dentry/inode
in the context of mounter.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

Right now ovl_permission() calls __inode_permission(realinode), to do
permission checks on real inode and no checks are done on overlay inode.

Modify it to do checks both on overlay inode as well as underlying inode.
Checks on overlay inode will be done with the creds of calling task while
checks on underlying inode will be done with the creds of mounter.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

Now we are planning to do DAC permission checks on overlay inode
itself. And to make it work, we will need to make sure we can get acls from
underlying inode. So define ->get_acl() for overlay inodes and this in turn
calls into underlying filesystem to get acls, if any.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

ovl_create_upper() and ovl_create_over_whiteout() seem to be sharing some
common code which can be moved into a separate function.  No functionality
change.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

Previously this was only done for directory inodes.  Doing so for all
inodes makes for a nice cleanup in ovl_permission at zero cost.

Inodes are not shared for hard links on the overlay, so this works fine.
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

No point in keeping overlay inodes around since they will never be reused.
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

The hash salting changes meant that we can no longer reuse the hash in the
overlay dentry to look up the underlying dentry.

Instead of lookup_hash(), use lookup_one_len_unlocked() and swith to
mounter's creds (like we do for all other operations later in the series).

Now the lookup_hash() export introduced in 4.6 by 3c9fe8cd ("vfs: add
lookup_hash() helper") is unused and can possibly be removed; its
usefulness negated by the hash salting and the idea that mounter's creds
should be used on operations on underlying filesystems.
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Fixes: 8387ff25 ("vfs: make the string hashes salt the hash")

Pull tracing updates from Steven Rostedt:
 "This is mostly clean ups and small fixes.  Some of the more visible
  changes are:

   - The function pid code uses the event pid filtering logic
   - [ku]probe events have access to current->comm
   - trace_printk now has sample code
   - PCI devices now trace physical addresses
   - stack tracing has less unnessary functions traced"

* tag 'trace-v4.8' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace:
  printk, tracing: Avoiding unneeded blank lines
  tracing: Use __get_str() when manipulating strings
  tracing, RAS: Cleanup on __get_str() usage
  tracing: Use outer () on __get_str() definition
  ftrace: Reduce size of function graph entries
  tracing: Have HIST_TRIGGERS select TRACING
  tracing: Using for_each_set_bit() to simplify trace_pid_write()
  ftrace: Move toplevel init out of ftrace_init_tracefs()
  tracing/function_graph: Fix filters for function_graph threshold
  tracing: Skip more functions when doing stack tracing of events
  tracing: Expose CPU physical addresses (resource values) for PCI devices
  tracing: Show the preempt count of when the event was called
  tracing: Add trace_printk sample code
  tracing: Choose static tp_printk buffer by explicit nesting count
  tracing: expose current->comm to [ku]probe events
  ftrace: Have set_ftrace_pid use the bitmap like events do
  tracing: Move pid_list write processing into its own function
  tracing: Move the pid_list seq_file functions to be global
  tracing: Move filtered_pid helper functions into trace.c
  tracing: Make the pid filtering helper functions global

Pull VFIO updates from Alex Williamson:
 - Enable no-iommu mode for platform devices (Peng Fan)
 - Sub-page mmap for exclusive pages (Yongji Xie)
 - Use-after-free fix (Ilya Lesokhin)
 - Support for ACPI-based platform devices (Sinan Kaya)

* tag 'vfio-v4.8-rc1' of git://github.com/awilliam/linux-vfio:
  vfio: platform: check reset call return code during release
  vfio: platform: check reset call return code during open
  vfio, platform: make reset driver a requirement by default
  vfio: platform: call _RST method when using ACPI
  vfio: platform: add extra debug info argument to call reset
  vfio: platform: add support for ACPI probe
  vfio: platform: determine reset capability
  vfio: platform: move reset call to a common function
  vfio: platform: rename reset function
  vfio: fix possible use after free of vfio group
  vfio-pci: Allow to mmap sub-page MMIO BARs if the mmio page is exclusive
  vfio: platform: support No-IOMMU mode

Pull MD updates from Shaohua Li:
 - A bunch of patches from Neil Brown to fix RCU usage
 - Two performance improvement patches from Tomasz Majchrzak
 - Alexey Obitotskiy fixes module refcount issue
 - Arnd Bergmann fixes time granularity
 - Cong Wang fixes a list corruption issue
 - Guoqing Jiang fixes a deadlock in md-cluster
 - A null pointer deference fix from me
 - Song Liu fixes misuse of raid6 rmw
 - Other trival/cleanup fixes from Guoqing Jiang and Xiao Ni

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/shli/md: (28 commits)
  MD: fix null pointer deference
  raid10: improve random reads performance
  md: add missing sysfs_notify on array_state update
  Fix kernel module refcount handling
  md: use seconds granularity for error logging
  md: reduce the number of synchronize_rcu() calls when multiple devices fail.
  md: be extra careful not to take a reference to a Faulty device.
  md/multipath: add rcu protection to rdev access in multipath_status.
  md/raid5: add rcu protection to rdev accesses in raid5_status.
  md/raid5: add rcu protection to rdev accesses in want_replace
  md/raid5: add rcu protection to rdev accesses in handle_failed_sync.
  md/raid1: add rcu protection to rdev in fix_read_error
  md/raid1: small code cleanup in end_sync_write
  md/raid1: small cleanup in raid1_end_read/write_request
  md/raid10: simplify print_conf a little.
  md/raid10: minor code improvement in fix_read_error()
  md/raid10: add rcu protection to rdev access during reshape.
  md/raid10: add rcu protection to rdev access in raid10_sync_request.
  md/raid10: add rcu protection in raid10_status.
  md/raid10: fix refounct imbalance when resyncing an array with a replacement device.
  ...

Pull libnvdimm updates from Dan Williams:

 - Replace pcommit with ADR / directed-flushing.

   The pcommit instruction, which has not shipped on any product, is
   deprecated.  Instead, the requirement is that platforms implement
   either ADR, or provide one or more flush addresses per nvdimm.

   ADR (Asynchronous DRAM Refresh) flushes data in posted write buffers
   to the memory controller on a power-fail event.

   Flush addresses are defined in ACPI 6.x as an NVDIMM Firmware
   Interface Table (NFIT) sub-structure: "Flush Hint Address Structure".
   A flush hint is an mmio address that when written and fenced assures
   that all previous posted writes targeting a given dimm have been
   flushed to media.

 - On-demand ARS (address range scrub).

   Linux uses the results of the ACPI ARS commands to track bad blocks
   in pmem devices.  When latent errors are detected we re-scrub the
   media to refresh the bad block list, userspace can also request a
   re-scrub at any time.

 - Support for the Microsoft DSM (device specific method) command
   format.

 - Support for EDK2/OVMF virtual disk device memory ranges.

 - Various fixes and cleanups across the subsystem.

* tag 'libnvdimm-for-4.8' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdimm/nvdimm: (41 commits)
  libnvdimm-btt: Delete an unnecessary check before the function call "__nd_device_register"
  nfit: do an ARS scrub on hitting a latent media error
  nfit: move to nfit/ sub-directory
  nfit, libnvdimm: allow an ARS scrub to be triggered on demand
  libnvdimm: register nvdimm_bus devices with an nd_bus driver
  pmem: clarify a debug print in pmem_clear_poison
  x86/insn: remove pcommit
  Revert "KVM: x86: add pcommit support"
  nfit, tools/testing/nvdimm/: unify shutdown paths
  libnvdimm: move ->module to struct nvdimm_bus_descriptor
  nfit: cleanup acpi_nfit_init calling convention
  nfit: fix _FIT evaluation memory leak + use after free
  tools/testing/nvdimm: add manufacturing_{date|location} dimm properties
  tools/testing/nvdimm: add virtual ramdisk range
  acpi, nfit: treat virtual ramdisk SPA as pmem region
  pmem: kill __pmem address space
  pmem: kill wmb_pmem()
  libnvdimm, pmem: use nvdimm_flush() for namespace I/O writes
  fs/dax: remove wmb_pmem()
  libnvdimm, pmem: flush posted-write queues on shutdown
  ...

Pull pin control updates from Linus Walleij:
 "This is the bulk of pin control changes for the v4.8 kernel cycle.

  Nothing stands out as especially exiting: new drivers, new subdrivers,
  lots of cleanups and incremental features.

  Business as usual.

  New drivers:

   - New driver for Oxnas pin control and GPIO.  This ARM-based chipset
     is used in a few storage (NAS) type devices.

   - New driver for the MAX77620/MAX20024 pin controller portions.

   - New driver for the Intel Merrifield pin controller.

  New subdrivers:

   - New subdriver for the Qualcomm MDM9615

   - New subdriver for the STM32F746 MCU

   - New subdriver for the Broadcom NSP SoC.

  Cleanups:

   - Demodularization of bool compiled-in drivers.

  Apart from this there is just regular incremental improvements to a
  lot of drivers, especially Uniphier and PFC"

* tag 'pinctrl-v4.8-1' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw/linux-pinctrl: (131 commits)
  pinctrl: fix pincontrol definition for marvell
  pinctrl: xway: fix typo
  Revert "pinctrl: amd: make it explicitly non-modular"
  pinctrl: iproc: Add NSP and Stingray GPIO support
  pinctrl: Update iProc GPIO DT bindings
  pinctrl: bcm: add OF dependencies
  pinctrl: ns2: remove redundant dev_err call in ns2_pinmux_probe()
  pinctrl: Add STM32F746 MCU support
  pinctrl: intel: Protect set wake flow by spin lock
  pinctrl: nsp: remove redundant dev_err call in nsp_pinmux_probe()
  pinctrl: uniphier: add Ethernet pin-mux settings
  sh-pfc: Use PTR_ERR_OR_ZERO() to simplify the code
  pinctrl: ns2: fix return value check in ns2_pinmux_probe()
  pinctrl: qcom: update DT bindings with ebi2 groups
  pinctrl: qcom: establish proper EBI2 pin groups
  pinctrl: imx21: Remove the MODULE_DEVICE_TABLE() macro
  Documentation: dt: Add new compatible to STM32 pinctrl driver bindings
  includes: dt-bindings: Add STM32F746 pinctrl DT bindings
  pinctrl: sunxi: fix nand0 function name for sun8i
  pinctrl: uniphier: remove pointless pin-mux settings for PH1-LD11
  ...

Merge more updates from Andrew Morton:
 "The rest of MM"

* emailed patches from Andrew Morton <akpm@linux-foundation.org>: (101 commits)
  mm, compaction: simplify contended compaction handling
  mm, compaction: introduce direct compaction priority
  mm, thp: remove __GFP_NORETRY from khugepaged and madvised allocations
  mm, page_alloc: make THP-specific decisions more generic
  mm, page_alloc: restructure direct compaction handling in slowpath
  mm, page_alloc: don't retry initial attempt in slowpath
  mm, page_alloc: set alloc_flags only once in slowpath
  lib/stackdepot.c: use __GFP_NOWARN for stack allocations
  mm, kasan: switch SLUB to stackdepot, enable memory quarantine for SLUB
  mm, kasan: account for object redzone in SLUB's nearest_obj()
  mm: fix use-after-free if memory allocation failed in vma_adjust()
  zsmalloc: Delete an unnecessary check before the function call "iput"
  mm/memblock.c: fix index adjustment error in __next_mem_range_rev()
  mem-hotplug: alloc new page from a nearest neighbor node when mem-offline
  mm: optimize copy_page_to/from_iter_iovec
  mm: add cond_resched() to generic_swapfile_activate()
  Revert "mm, mempool: only set __GFP_NOMEMALLOC if there are free elements"
  mm, compaction: don't isolate PageWriteback pages in MIGRATE_SYNC_LIGHT mode
  mm: hwpoison: remove incorrect comments
  make __section_nr() more efficient
  ...

Async compaction detects contention either due to failing trylock on
zone->lock or lru_lock, or by need_resched().  Since 1f9efdef ("mm,
compaction: khugepaged should not give up due to need_resched()") the
code got quite complicated to distinguish these two up to the
__alloc_pages_slowpath() level, so different decisions could be taken
for khugepaged allocations.

After the recent changes, khugepaged allocations don't check for
contended compaction anymore, so we again don't need to distinguish lock
and sched contention, and simplify the current convoluted code a lot.

However, I believe it's also possible to simplify even more and
completely remove the check for contended compaction after the initial
async compaction for costly orders, which was originally aimed at THP
page fault allocations.  There are several reasons why this can be done
now:

- with the new defaults, THP page faults no longer do reclaim/compaction at
  all, unless the system admin has overridden the default, or application has
  indicated via madvise that it can benefit from THP's. In both cases, it
  means that the potential extra latency is expected and worth the benefits.
- even if reclaim/compaction proceeds after this patch where it previously
  wouldn't, the second compaction attempt is still async and will detect the
  contention and back off, if the contention persists
- there are still heuristics like deferred compaction and pageblock skip bits
  in place that prevent excessive THP page fault latencies

Link: http://lkml.kernel.org/r/20160721073614.24395-9-vbabka@suse.czSigned-off-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

In the context of direct compaction, for some types of allocations we
would like the compaction to either succeed or definitely fail while
trying as hard as possible.  Current async/sync_light migration mode is
insufficient, as there are heuristics such as caching scanner positions,
marking pageblocks as unsuitable or deferring compaction for a zone.  At
least the final compaction attempt should be able to override these
heuristics.

To communicate how hard compaction should try, we replace migration mode
with a new enum compact_priority and change the relevant function
signatures.  In compact_zone_order() where struct compact_control is
constructed, the priority is mapped to suitable control flags.  This
patch itself has no functional change, as the current priority levels
are mapped back to the same migration modes as before.  Expanding them
will be done next.

Note that !CONFIG_COMPACTION variant of try_to_compact_pages() is
removed, as the only caller exists under CONFIG_COMPACTION.

Link: http://lkml.kernel.org/r/20160721073614.24395-8-vbabka@suse.czSigned-off-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

After the previous patch, we can distinguish costly allocations that
should be really lightweight, such as THP page faults, with
__GFP_NORETRY.  This means we don't need to recognize khugepaged
allocations via PF_KTHREAD anymore.  We can also change THP page faults
in areas where madvise(MADV_HUGEPAGE) was used to try as hard as
khugepaged, as the process has indicated that it benefits from THP's and
is willing to pay some initial latency costs.

We can also make the flags handling less cryptic by distinguishing
GFP_TRANSHUGE_LIGHT (no reclaim at all, default mode in page fault) from
GFP_TRANSHUGE (only direct reclaim, khugepaged default).  Adding
__GFP_NORETRY or __GFP_KSWAPD_RECLAIM is done where needed.

The patch effectively changes the current GFP_TRANSHUGE users as
follows:

* get_huge_zero_page() - the zero page lifetime should be relatively
  long and it's shared by multiple users, so it's worth spending some
  effort on it.  We use GFP_TRANSHUGE, and __GFP_NORETRY is not added.
  This also restores direct reclaim to this allocation, which was
  unintentionally removed by commit e4a49efe4e7e ("mm: thp: set THP defrag
  by default to madvise and add a stall-free defrag option")

* alloc_hugepage_khugepaged_gfpmask() - this is khugepaged, so latency
  is not an issue.  So if khugepaged "defrag" is enabled (the default), do
  reclaim via GFP_TRANSHUGE without __GFP_NORETRY.  We can remove the
  PF_KTHREAD check from page alloc.

  As a side-effect, khugepaged will now no longer check if the initial
  compaction was deferred or contended.  This is OK, as khugepaged sleep
  times between collapsion attempts are long enough to prevent noticeable
  disruption, so we should allow it to spend some effort.

* migrate_misplaced_transhuge_page() - already was masking out
  __GFP_RECLAIM, so just convert to GFP_TRANSHUGE_LIGHT which is
  equivalent.

* alloc_hugepage_direct_gfpmask() - vma's with VM_HUGEPAGE (via madvise)
  are now allocating without __GFP_NORETRY.  Other vma's keep using
  __GFP_NORETRY if direct reclaim/compaction is at all allowed (by default
  it's allowed only for madvised vma's).  The rest is conversion to
  GFP_TRANSHUGE(_LIGHT).

[mhocko@suse.com: suggested GFP_TRANSHUGE_LIGHT]
Link: http://lkml.kernel.org/r/20160721073614.24395-7-vbabka@suse.czSigned-off-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Since THP allocations during page faults can be costly, extra decisions
are employed for them to avoid excessive reclaim and compaction, if the
initial compaction doesn't look promising.  The detection has never been
perfect as there is no gfp flag specific to THP allocations.  At this
moment it checks the whole combination of flags that makes up
GFP_TRANSHUGE, and hopes that no other users of such combination exist,
or would mind being treated the same way.  Extra care is also taken to
separate allocations from khugepaged, where latency doesn't matter that
much.

It is however possible to distinguish these allocations in a simpler and
more reliable way.  The key observation is that after the initial
compaction followed by the first iteration of "standard"
reclaim/compaction, both __GFP_NORETRY allocations and costly
allocations without __GFP_REPEAT are declared as failures:

        /* Do not loop if specifically requested */
        if (gfp_mask & __GFP_NORETRY)
                goto nopage;

        /*
         * Do not retry costly high order allocations unless they are
         * __GFP_REPEAT
         */
        if (order > PAGE_ALLOC_COSTLY_ORDER && !(gfp_mask & __GFP_REPEAT))
                goto nopage;

This means we can further distinguish allocations that are costly order
*and* additionally include the __GFP_NORETRY flag.  As it happens,
GFP_TRANSHUGE allocations do already fall into this category.  This will
also allow other costly allocations with similar high-order benefit vs
latency considerations to use this semantic.  Furthermore, we can
distinguish THP allocations that should try a bit harder (such as from
khugepageed) by removing __GFP_NORETRY, as will be done in the next
patch.

Link: http://lkml.kernel.org/r/20160721073614.24395-6-vbabka@suse.czSigned-off-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

The retry loop in __alloc_pages_slowpath is supposed to keep trying
reclaim and compaction (and OOM), until either the allocation succeeds,
or returns with failure.  Success here is more probable when reclaim
precedes compaction, as certain watermarks have to be met for compaction
to even try, and more free pages increase the probability of compaction
success.  On the other hand, starting with light async compaction (if
the watermarks allow it), can be more efficient, especially for smaller
orders, if there's enough free memory which is just fragmented.

Thus, the current code starts with compaction before reclaim, and to
make sure that the last reclaim is always followed by a final
compaction, there's another direct compaction call at the end of the
loop.  This makes the code hard to follow and adds some duplicated
handling of migration_mode decisions.  It's also somewhat inefficient
that even if reclaim or compaction decides not to retry, the final
compaction is still attempted.  Some gfp flags combination also shortcut
these retry decisions by "goto noretry;", making it even harder to
follow.

This patch attempts to restructure the code with only minimal functional
changes.  The call to the first compaction and THP-specific checks are
now placed above the retry loop, and the "noretry" direct compaction is
removed.

The initial compaction is additionally restricted only to costly orders,
as we can expect smaller orders to be held back by watermarks, and only
larger orders to suffer primarily from fragmentation.  This better
matches the checks in reclaim's shrink_zones().

There are two other smaller functional changes.  One is that the upgrade
from async migration to light sync migration will always occur after the
initial compaction.  This is how it has been until recent patch "mm,
oom: protect !costly allocations some more", which introduced upgrading
the mode based on COMPACT_COMPLETE result, but kept the final compaction
always upgraded, which made it even more special.  It's better to return
to the simpler handling for now, as migration modes will be further
modified later in the series.

The second change is that once both reclaim and compaction declare it's
not worth to retry the reclaim/compact loop, there is no final
compaction attempt.  As argued above, this is intentional.  If that
final compaction were to succeed, it would be due to a wrong retry
decision, or simply a race with somebody else freeing memory for us.

The main outcome of this patch should be simpler code.  Logically, the
initial compaction without reclaim is the exceptional case to the
reclaim/compaction scheme, but prior to the patch, it was the last loop
iteration that was exceptional.  Now the code matches the logic better.
The change also enable the following patches.

Link: http://lkml.kernel.org/r/20160721073614.24395-5-vbabka@suse.czSigned-off-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

After __alloc_pages_slowpath() sets up new alloc_flags and wakes up
kswapd, it first tries get_page_from_freelist() with the new
alloc_flags, as it may succeed e.g. due to using min watermark instead
of low watermark.  It makes sense to to do this attempt before adjusting
zonelist based on alloc_flags/gfp_mask, as it's still relatively a fast
path if we just wake up kswapd and successfully allocate.

This patch therefore moves the initial attempt above the retry label and
reorganizes a bit the part below the retry label.  We still have to
attempt get_page_from_freelist() on each retry, as some allocations
cannot do that as part of direct reclaim or compaction, and yet are not
allowed to fail (even though they do a WARN_ON_ONCE() and thus should
not exist).  We can reuse the call meant for ALLOC_NO_WATERMARKS attempt
and just set alloc_flags to ALLOC_NO_WATERMARKS if the context allows
it.  As a side-effect, the attempts from direct reclaim/compaction will
also no longer obey watermarks once this is set, but there's little harm
in that.

Kswapd wakeups are also done on each retry to be safe from potential
races resulting in kswapd going to sleep while a process (that may not
be able to reclaim by itself) is still looping.

Link: http://lkml.kernel.org/r/20160721073614.24395-4-vbabka@suse.czSigned-off-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>