This is trickier than the previous conversions.  We have backref_node's
that need to hold onto their root for their lifetime.  Do the read of
the root and grab the ref.  If at any point we don't use the root we
discard it, however if we use it in our backref node we don't free it
until we free the backref node.  Any time we switch the root's for the
backref node we need to drop our ref on the old root and grab the ref on
the new root, and if we dupe a node we need to get a ref on the root
there as well.
Signed-off-by: NJosef Bacik <josef@toxicpanda.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

We look up an arbitrary fs root here, we need to hold a ref on the root
for the duration.
Signed-off-by: NJosef Bacik <josef@toxicpanda.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

We look up whatever root userspace has given us, we need to hold a ref
throughout this operation. Use 'root' only for the on fs root and not as
a temporary variable elsewhere.
Signed-off-by: NJosef Bacik <josef@toxicpanda.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

We can wander into a different root, so grab a ref on the root we look
up.  Later on we make root = fs_info->tree_root so we need this separate
out label to make sure we do the right cleanup only in the case we're
looking up a different root.
Signed-off-by: NJosef Bacik <josef@toxicpanda.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

We look up an arbitrary fs root, we need to hold a ref on it while we're
doing our search.
Signed-off-by: NJosef Bacik <josef@toxicpanda.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

We lookup a arbitrary fs root, we need to hold a ref on that root.  If
we're using our own inodes root then grab a ref on that as well to make
the cleanup easier.
Signed-off-by: NJosef Bacik <josef@toxicpanda.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

We're creating the new root here, but we should hold the ref until after
we've initialized the inode for it.
Signed-off-by: NJosef Bacik <josef@toxicpanda.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Looking up the inode from an arbitrary tree means we need to hold a ref
on that root.
Signed-off-by: NJosef Bacik <josef@toxicpanda.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

We are looking up an arbitrary inode, we need to hold a ref on the root
while we're doing this.
Signed-off-by: NJosef Bacik <josef@toxicpanda.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Looking up the inode we need to search the root, make sure we hold a
reference on that root while we're doing the lookup.
Signed-off-by: NJosef Bacik <josef@toxicpanda.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

We're looking up a random root, we need to hold a ref on it while we're
using it.
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NJosef Bacik <josef@toxicpanda.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

If the root is sitting in the radix tree, we should probably have a ref
for the radix tree.  Grab a ref on the root when we insert it, and drop
it when it gets deleted.
Reviewed-by: NNikolay Borisov <nborisov@suse.com>
Signed-off-by: NJosef Bacik <josef@toxicpanda.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Add another comment to cover how the space reservation system works
generally.  This covers the actual reservation flow, as well as how
flushing is handled.
Signed-off-by: NJosef Bacik <josef@toxicpanda.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

delalloc space reservation is tricky because it encompasses both data
and metadata.  Make it clear what each side does, the general flow of
how space is moved throughout the lifetime of a write, and what goes
into the calculations.
Signed-off-by: NJosef Bacik <josef@toxicpanda.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

This is a giant comment at the top of block-rsv.c describing generally
how block reserves work.  It is purely about the block reserves
themselves, and nothing to do with how the actual reservation system
works.
Signed-off-by: NJosef Bacik <josef@toxicpanda.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

We want to use this for dropping all roots, and in some error cases we
may not have a root, so handle this to make the cleanup code easier.
Make btrfs_grab_fs_root the same so we can use it in cases where the
root may not exist (like the quota root).
Reviewed-by: NNikolay Borisov <nborisov@suse.com>
Signed-off-by: NJosef Bacik <josef@toxicpanda.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Now that the orphan cleanup stuff doesn't use this directly we can just
make them static.
Reviewed-by: NNikolay Borisov <nborisov@suse.com>
Signed-off-by: NJosef Bacik <josef@toxicpanda.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

All this does is call btrfs_get_fs_root() with check_ref == true.  Just
use btrfs_get_fs_root() so we don't have a bunch of different helpers
that do the same thing.
Reviewed-by: NNikolay Borisov <nborisov@suse.com>
Signed-off-by: NJosef Bacik <josef@toxicpanda.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

All helpers should either be using btrfs_get_fs_root() or
btrfs_read_tree_root().
Reviewed-by: NNikolay Borisov <nborisov@suse.com>
Signed-off-by: NJosef Bacik <josef@toxicpanda.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Relocation has it's special roots, we don't want to save these in the
root cache either, so swap it to use btrfs_read_tree_root().  However
the reloc root does need REF_COWS set, so make sure we set it everywhere
we use this helper, as it no longer does the REF_COWS setting.
Reviewed-by: NNikolay Borisov <nborisov@suse.com>
Signed-off-by: NJosef Bacik <josef@toxicpanda.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Tree-log uses btrfs_read_fs_root to load its log, but this just calls
btrfs_read_tree_root.  We don't save the log roots in our root cache, so
just export this helper and use it in the logging code.
Reviewed-by: NNikolay Borisov <nborisov@suse.com>
Signed-off-by: NJosef Bacik <josef@toxicpanda.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

btrfs_find_orphan_roots has this weird thing where it looks up the root
in cache to see if it is there before just reading the root.  But the
read it uses just reads the root, it doesn't do any of the init work, we
do that by hand here.  But this is unnecessary, all we really want is to
see if the root still exists and add it to the dead roots list to be
cleaned up, otherwise we delete the orphan item.

Fix this by just using btrfs_get_fs_root directly with check_ref set to
false so we get the orphan root items.  Then we just handle in cache and
out of cache roots the same, add them to the dead roots list and carry
on.
Reviewed-by: NNikolay Borisov <nborisov@suse.com>
Signed-off-by: NJosef Bacik <josef@toxicpanda.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

We have a helper for reading fs roots that just reads the fs root off
the disk and then sets REF_COWS and init's the inheritable flags.  Move
this into btrfs_init_fs_root so we can later get rid of this helper and
consolidate all of the fs root reading into one helper.
Reviewed-by: NNikolay Borisov <nborisov@suse.com>
Signed-off-by: NJosef Bacik <josef@toxicpanda.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

There's no reason to not init the root at alloc time, and with later
patches it actually causes problems if we error out mounting the fs
before the tree_root is init'ed because we expect it to have a valid ref
count.  Fix this by pushing __setup_root into btrfs_alloc_root.
Reviewed-by: NNikolay Borisov <nborisov@suse.com>
Signed-off-by: NJosef Bacik <josef@toxicpanda.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Now that we have a safe way to update the isize, remove all of this code
as it's no longer needed.
Reviewed-by: NFilipe Manana <fdmanana@suse.com>
Signed-off-by: NJosef Bacik <josef@toxicpanda.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Now that we have a safe way to update the i_size, replace all uses of
btrfs_ordered_update_i_size with btrfs_inode_safe_disk_i_size_write.
Reviewed-by: NFilipe Manana <fdmanana@suse.com>
Signed-off-by: NJosef Bacik <josef@toxicpanda.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

We want to use this everywhere we modify the file extent items
permanently.  These include:

  1) Inserting new file extents for writes and prealloc extents.
  2) Truncating inode items.
  3) btrfs_cont_expand().
  4) Insert inline extents.
  5) Insert new extents from log replay.
  6) Insert a new extent for clone, as it could be past i_size.
  7) Hole punching

For hole punching in particular it might seem it's not necessary because
anybody extending would use btrfs_cont_expand, however there is a corner
that still can give us trouble.  Start with an empty file and

fallocate KEEP_SIZE 1M-2M

We now have a 0 length file, and a hole file extent from 0-1M, and a
prealloc extent from 1M-2M.  Now

punch 1M-1.5M

Because this is past i_size we have

[HOLE EXTENT][ NOTHING ][PREALLOC]
[0        1M][1M   1.5M][1.5M  2M]

with an i_size of 0.  Now if we pwrite 0-1.5M we'll increas our i_size
to 1.5M, but our disk_i_size is still 0 until the ordered extent
completes.

However if we now immediately truncate 2M on the file we'll just call
btrfs_cont_expand(inode, 1.5M, 2M), since our old i_size is 1.5M.  If we
commit the transaction here and crash we'll expose the gap.

To fix this we need to clear the file extent mapping for the range that
we punched but didn't insert a corresponding file extent for.  This will
mean the truncate will only get an disk_i_size set to 1M if we crash
before the finish ordered io happens.

I've written an xfstest to reproduce the problem and validate this fix.
Reviewed-by: NFilipe Manana <fdmanana@suse.com>
Signed-off-by: NJosef Bacik <josef@toxicpanda.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

In order to keep track of where we have file extents on disk, and thus
where it is safe to adjust the i_size to, we need to have a tree in
place to keep track of the contiguous areas we have file extents for.

Add helpers to use this tree, as it's not required for NO_HOLES file
systems.  We will use this by setting DIRTY for areas we know we have
file extent item's set, and clearing it when we remove file extent items
for truncation.
Reviewed-by: NFilipe Manana <fdmanana@suse.com>
Signed-off-by: NJosef Bacik <josef@toxicpanda.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

We were using btrfs_i_size_write(), which unconditionally jacks up
inode->disk_i_size.  However since clone can operate on ranges we could
have pending ordered extents for a range prior to the start of our clone
operation and thus increase disk_i_size too far and have a hole with no
file extent.

Fix this by using the btrfs_ordered_update_i_size helper which will do
the right thing in the face of pending ordered extents outside of our
clone range.
Reviewed-by: NFilipe Manana <fdmanana@suse.com>
Signed-off-by: NJosef Bacik <josef@toxicpanda.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Btrfsctl was removed in 2012, now the function btrfs_control_ioctl()
is only used for devices ioctls. So update the comment.
Reviewed-by: NNikolay Borisov <nborisov@suse.com>
Signed-off-by: NSu Yue <Damenly_Su@gmx.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Relocation is one of the most complex part of btrfs, while it's also the
foundation stone for online resizing, profile converting.

For such a complex facility, we should at least have some introduction
to it.

This patch will add an basic introduction at pretty a high level,
explaining:

- What relocation does
- How relocation is done
  Only mentioning how data reloc tree and reloc tree are involved in the
  operation.
  No details like the backref cache, or the data reloc tree contents.
- Which function to refer.

More detailed comments will be added for reloc tree creation, data reloc
tree creation and backref cache.
Signed-off-by: NQu Wenruo <wqu@suse.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Each new element added to the mod seq list is always appended to the list,
and each one gets a sequence number coming from a counter which gets
incremented everytime a new element is added to the list (or a new node
is added to the tree mod log rbtree). Therefore the element with the
lowest sequence number is always the first element in the list.

So just remove the list iteration at btrfs_put_tree_mod_seq() that
computes the minimum sequence number in the list and replace it with
a check for the first element's sequence number.
Reviewed-by: NJosef Bacik <josef@toxicpanda.com>
Signed-off-by: NFilipe Manana <fdmanana@suse.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

The overview of btrfs dev-replace.  It mentions some corner cases caused
by the write duplication and scrub based data copy.
Reviewed-by: NAnand Jain <anand.jain@oracle.com>
Signed-off-by: NQu Wenruo <wqu@suse.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
[ adjust wording ]
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Pull btrfs fixes from David Sterba:
 "Two fixes.

  The first is a regression: when dropping some incompat bits the
  conditions were reversed. The other is a fix for rename whiteout
  potentially leaving stack memory linked to a list"

* tag 'for-5.6-rc6-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux:
  btrfs: fix removal of raid[56|1c34} incompat flags after removing block group
  btrfs: fix log context list corruption after rename whiteout error

Merge misc fixes from Andrew Morton:
 "10 fixes"

* emailed patches from Andrew Morton <akpm@linux-foundation.org>:
  x86/mm: split vmalloc_sync_all()
  mm, slub: prevent kmalloc_node crashes and memory leaks
  mm/mmu_notifier: silence PROVE_RCU_LIST warnings
  epoll: fix possible lost wakeup on epoll_ctl() path
  mm: do not allow MADV_PAGEOUT for CoW pages
  mm, memcg: throttle allocators based on ancestral memory.high
  mm, memcg: fix corruption on 64-bit divisor in memory.high throttling
  page-flags: fix a crash at SetPageError(THP_SWAP)
  mm/hotplug: fix hot remove failure in SPARSEMEM|!VMEMMAP case
  memcg: fix NULL pointer dereference in __mem_cgroup_usage_unregister_event

Commit 3f8fd02b ("mm/vmalloc: Sync unmappings in
__purge_vmap_area_lazy()") introduced a call to vmalloc_sync_all() in
the vunmap() code-path.  While this change was necessary to maintain
correctness on x86-32-pae kernels, it also adds additional cycles for
architectures that don't need it.

Specifically on x86-64 with CONFIG_VMAP_STACK=y some people reported
severe performance regressions in micro-benchmarks because it now also
calls the x86-64 implementation of vmalloc_sync_all() on vunmap().  But
the vmalloc_sync_all() implementation on x86-64 is only needed for newly
created mappings.

To avoid the unnecessary work on x86-64 and to gain the performance
back, split up vmalloc_sync_all() into two functions:

	* vmalloc_sync_mappings(), and
	* vmalloc_sync_unmappings()

Most call-sites to vmalloc_sync_all() only care about new mappings being
synchronized.  The only exception is the new call-site added in the
above mentioned commit.

Shile Zhang directed us to a report of an 80% regression in reaim
throughput.

Fixes: 3f8fd02b ("mm/vmalloc: Sync unmappings in __purge_vmap_area_lazy()")
Reported-by: Nkernel test robot <oliver.sang@intel.com>
Reported-by: NShile Zhang <shile.zhang@linux.alibaba.com>
Signed-off-by: NJoerg Roedel <jroedel@suse.de>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Tested-by: NBorislav Petkov <bp@suse.de>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>	[GHES]
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: <stable@vger.kernel.org>
Link: http://lkml.kernel.org/r/20191009124418.8286-1-joro@8bytes.org
Link: https://lists.01.org/hyperkitty/list/lkp@lists.01.org/thread/4D3JPPHBNOSPFK2KEPC6KGKS6J25AIDB/
Link: http://lkml.kernel.org/r/20191113095530.228959-1-shile.zhang@linux.alibaba.comSigned-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Sachin reports [1] a crash in SLUB __slab_alloc():

  BUG: Kernel NULL pointer dereference on read at 0x000073b0
  Faulting instruction address: 0xc0000000003d55f4
  Oops: Kernel access of bad area, sig: 11 [#1]
  LE PAGE_SIZE=64K MMU=Hash SMP NR_CPUS=2048 NUMA pSeries
  Modules linked in:
  CPU: 19 PID: 1 Comm: systemd Not tainted 5.6.0-rc2-next-20200218-autotest #1
  NIP:  c0000000003d55f4 LR: c0000000003d5b94 CTR: 0000000000000000
  REGS: c0000008b37836d0 TRAP: 0300   Not tainted  (5.6.0-rc2-next-20200218-autotest)
  MSR:  8000000000009033 <SF,EE,ME,IR,DR,RI,LE>  CR: 24004844  XER: 00000000
  CFAR: c00000000000dec4 DAR: 00000000000073b0 DSISR: 40000000 IRQMASK: 1
  GPR00: c0000000003d5b94 c0000008b3783960 c00000000155d400 c0000008b301f500
  GPR04: 0000000000000dc0 0000000000000002 c0000000003443d8 c0000008bb398620
  GPR08: 00000008ba2f0000 0000000000000001 0000000000000000 0000000000000000
  GPR12: 0000000024004844 c00000001ec52a00 0000000000000000 0000000000000000
  GPR16: c0000008a1b20048 c000000001595898 c000000001750c18 0000000000000002
  GPR20: c000000001750c28 c000000001624470 0000000fffffffe0 5deadbeef0000122
  GPR24: 0000000000000001 0000000000000dc0 0000000000000002 c0000000003443d8
  GPR28: c0000008b301f500 c0000008bb398620 0000000000000000 c00c000002287180
  NIP ___slab_alloc+0x1f4/0x760
  LR __slab_alloc+0x34/0x60
  Call Trace:
    ___slab_alloc+0x334/0x760 (unreliable)
    __slab_alloc+0x34/0x60
    __kmalloc_node+0x110/0x490
    kvmalloc_node+0x58/0x110
    mem_cgroup_css_online+0x108/0x270
    online_css+0x48/0xd0
    cgroup_apply_control_enable+0x2ec/0x4d0
    cgroup_mkdir+0x228/0x5f0
    kernfs_iop_mkdir+0x90/0xf0
    vfs_mkdir+0x110/0x230
    do_mkdirat+0xb0/0x1a0
    system_call+0x5c/0x68

This is a PowerPC platform with following NUMA topology:

  available: 2 nodes (0-1)
  node 0 cpus:
  node 0 size: 0 MB
  node 0 free: 0 MB
  node 1 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
  node 1 size: 35247 MB
  node 1 free: 30907 MB
  node distances:
  node   0   1
    0:  10  40
    1:  40  10

  possible numa nodes: 0-31

This only happens with a mmotm patch "mm/memcontrol.c: allocate
shrinker_map on appropriate NUMA node" [2] which effectively calls
kmalloc_node for each possible node.  SLUB however only allocates
kmem_cache_node on online N_NORMAL_MEMORY nodes, and relies on
node_to_mem_node to return such valid node for other nodes since commit
a561ce00 ("slub: fall back to node_to_mem_node() node if allocating
on memoryless node").  This is however not true in this configuration
where the _node_numa_mem_ array is not initialized for nodes 0 and 2-31,
thus it contains zeroes and get_partial() ends up accessing
non-allocated kmem_cache_node.

A related issue was reported by Bharata (originally by Ramachandran) [3]
where a similar PowerPC configuration, but with mainline kernel without
patch [2] ends up allocating large amounts of pages by kmalloc-1k
kmalloc-512.  This seems to have the same underlying issue with
node_to_mem_node() not behaving as expected, and might probably also
lead to an infinite loop with CONFIG_SLUB_CPU_PARTIAL [4].

This patch should fix both issues by not relying on node_to_mem_node()
anymore and instead simply falling back to NUMA_NO_NODE, when
kmalloc_node(node) is attempted for a node that's not online, or has no
usable memory.  The "usable memory" condition is also changed from
node_present_pages() to N_NORMAL_MEMORY node state, as that is exactly
the condition that SLUB uses to allocate kmem_cache_node structures.
The check in get_partial() is removed completely, as the checks in
___slab_alloc() are now sufficient to prevent get_partial() being
reached with an invalid node.

[1] https://lore.kernel.org/linux-next/3381CD91-AB3D-4773-BA04-E7A072A63968@linux.vnet.ibm.com/
[2] https://lore.kernel.org/linux-mm/fff0e636-4c36-ed10-281c-8cdb0687c839@virtuozzo.com/
[3] https://lore.kernel.org/linux-mm/20200317092624.GB22538@in.ibm.com/
[4] https://lore.kernel.org/linux-mm/088b5996-faae-8a56-ef9c-5b567125ae54@suse.cz/

Fixes: a561ce00 ("slub: fall back to node_to_mem_node() node if allocating on memoryless node")
Reported-by: NSachin Sant <sachinp@linux.vnet.ibm.com>
Reported-by: NPUVICHAKRAVARTHY RAMACHANDRAN <puvichakravarthy@in.ibm.com>
Signed-off-by: NVlastimil Babka <vbabka@suse.cz>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Tested-by: NSachin Sant <sachinp@linux.vnet.ibm.com>
Tested-by: NBharata B Rao <bharata@linux.ibm.com>
Reviewed-by: NSrikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Christopher Lameter <cl@linux.com>
Cc: linuxppc-dev@lists.ozlabs.org
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Kirill Tkhai <ktkhai@virtuozzo.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Nathan Lynch <nathanl@linux.ibm.com>
Cc: <stable@vger.kernel.org>
Link: http://lkml.kernel.org/r/20200320115533.9604-1-vbabka@suse.czDebugged-by: NSrikar Dronamraju <srikar@linux.vnet.ibm.com>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

It is safe to traverse mm->notifier_subscriptions->list either under
SRCU read lock or mm->notifier_subscriptions->lock using
hlist_for_each_entry_rcu().  Silence the PROVE_RCU_LIST false positives,
for example,

  WARNING: suspicious RCU usage
  -----------------------------
  mm/mmu_notifier.c:484 RCU-list traversed in non-reader section!!

  other info that might help us debug this:

  rcu_scheduler_active = 2, debug_locks = 1
  3 locks held by libvirtd/802:
   #0: ffff9321e3f58148 (&mm->mmap_sem#2){++++}, at: do_mprotect_pkey+0xe1/0x3e0
   #1: ffffffff91ae6160 (mmu_notifier_invalidate_range_start){+.+.}, at: change_p4d_range+0x5fa/0x800
   #2: ffffffff91ae6e08 (srcu){....}, at: __mmu_notifier_invalidate_range_start+0x178/0x460

  stack backtrace:
  CPU: 7 PID: 802 Comm: libvirtd Tainted: G          I       5.6.0-rc6-next-20200317+ #2
  Hardware name: HP ProLiant BL460c Gen8, BIOS I31 11/02/2014
  Call Trace:
    dump_stack+0xa4/0xfe
    lockdep_rcu_suspicious+0xeb/0xf5
    __mmu_notifier_invalidate_range_start+0x3ff/0x460
    change_p4d_range+0x746/0x800
    change_protection+0x1df/0x300
    mprotect_fixup+0x245/0x3e0
    do_mprotect_pkey+0x23b/0x3e0
    __x64_sys_mprotect+0x51/0x70
    do_syscall_64+0x91/0xae8
    entry_SYSCALL_64_after_hwframe+0x49/0xb3
Signed-off-by: NQian Cai <cai@lca.pw>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Reviewed-by: NPaul E. McKenney <paulmck@kernel.org>
Reviewed-by: NJason Gunthorpe <jgg@mellanox.com>
Link: http://lkml.kernel.org/r/20200317175640.2047-1-cai@lca.pwSigned-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

This fixes possible lost wakeup introduced by commit a218cc49.
Originally modifications to ep->wq were serialized by ep->wq.lock, but
in commit a218cc49 ("epoll: use rwlock in order to reduce
ep_poll_callback() contention") a new rw lock was introduced in order to
relax fd event path, i.e. callers of ep_poll_callback() function.

After the change ep_modify and ep_insert (both are called on epoll_ctl()
path) were switched to ep->lock, but ep_poll (epoll_wait) was using
ep->wq.lock on wqueue list modification.

The bug doesn't lead to any wqueue list corruptions, because wake up
path and list modifications were serialized by ep->wq.lock internally,
but actual waitqueue_active() check prior wake_up() call can be
reordered with modifications of ep ready list, thus wake up can be lost.

And yes, can be healed by explicit smp_mb():

  list_add_tail(&epi->rdlink, &ep->rdllist);
  smp_mb();
  if (waitqueue_active(&ep->wq))
	wake_up(&ep->wp);

But let's make it simple, thus current patch replaces ep->wq.lock with
the ep->lock for wqueue modifications, thus wake up path always observes
activeness of the wqueue correcty.

Fixes: a218cc49 ("epoll: use rwlock in order to reduce ep_poll_callback() contention")
Reported-by: NMax Neunhoeffer <max@arangodb.com>
Signed-off-by: NRoman Penyaev <rpenyaev@suse.de>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Tested-by: NMax Neunhoeffer <max@arangodb.com>
Cc: Jakub Kicinski <kuba@kernel.org>
Cc: Christopher Kohlhoff <chris.kohlhoff@clearpool.io>
Cc: Davidlohr Bueso <dbueso@suse.de>
Cc: Jason Baron <jbaron@akamai.com>
Cc: Jes Sorensen <jes.sorensen@gmail.com>
Cc: <stable@vger.kernel.org>	[5.1+]
Link: http://lkml.kernel.org/r/20200214170211.561524-1-rpenyaev@suse.de
References: https://bugzilla.kernel.org/show_bug.cgi?id=205933Bisected-by: NMax Neunhoeffer <max@arangodb.com>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>