linux-stable.git/include/linux, branch v4.9.46

Clarify (and fix) MAX_LFS_FILESIZE macros

2017-08-30T08:21:54+00:00

commit 0cc3b0ec23ce4c69e1e890ed2b8d2fa932b14aad upstream.

We have a MAX_LFS_FILESIZE macro that is meant to be filled in by
filesystems (and other IO targets) that know they are 64-bit clean and
don't have any 32-bit limits in their IO path.

It turns out that our 32-bit value for that limit was bogus.  On 32-bit,
the VM layer is limited by the page cache to only 32-bit index values,
but our logic for that was confusing and actually wrong.  We used to
define that value to

	(((loff_t)PAGE_SIZE << (BITS_PER_LONG-1))-1)

which is actually odd in several ways: it limits the index to 31 bits,
and then it limits files so that they can't have data in that last byte
of a page that has the highest 31-bit index (ie page index 0x7fffffff).

Neither of those limitations make sense.  The index is actually the full
32 bit unsigned value, and we can use that whole full page.  So the
maximum size of the file would logically be "PAGE_SIZE << BITS_PER_LONG".

However, we do wan tto avoid the maximum index, because we have code
that iterates over the page indexes, and we don't want that code to
overflow.  So the maximum size of a file on a 32-bit host should
actually be one page less than the full 32-bit index.

So the actual limit is ULONG_MAX << PAGE_SHIFT.  That means that we will
not actually be using the page of that last index (ULONG_MAX), but we
can grow a file up to that limit.

The wrong value of MAX_LFS_FILESIZE actually caused problems for Doug
Nazar, who was still using a 32-bit host, but with a 9.7TB 2 x RAID5
volume.  It turns out that our old MAX_LFS_FILESIZE was 8TiB (well, one
byte less), but the actual true VM limit is one page less than 16TiB.

This was invisible until commit c2a9737f45e2 ("vfs,mm: fix a dead loop
in truncate_inode_pages_range()"), which started applying that
MAX_LFS_FILESIZE limit to block devices too.

NOTE! On 64-bit, the page index isn't a limiter at all, and the limit is
actually just the offset type itself (loff_t), which is signed.  But for
clarity, on 64-bit, just use the maximum signed value, and don't make
people have to count the number of 'f' characters in the hex constant.

So just use LLONG_MAX for the 64-bit case.  That was what the value had
been before too, just written out as a hex constant.

Fixes: c2a9737f45e2 ("vfs,mm: fix a dead loop in truncate_inode_pages_range()")
Reported-and-tested-by: Doug Nazar 
Cc: Andreas Dilger 
Cc: Mark Fasheh 
Cc: Joel Becker 
Cc: Dave Kleikamp 
Signed-off-by: Linus Torvalds 
Signed-off-by: Greg Kroah-Hartman

perf/x86/intel/rapl: Make package handling more robust

2017-08-30T08:21:50+00:00

commit dd86e373e09fb16b83e8adf5c48c421a4ca76468 upstream.

The package management code in RAPL relies on package mapping being
available before a CPU is started. This changed with:

  9d85eb9119f4 ("x86/smpboot: Make logical package management more robust")

because the ACPI/BIOS information turned out to be unreliable, but that
left RAPL in broken state. This was not noticed because on a regular boot
all CPUs are online before RAPL is initialized.

A possible fix would be to reintroduce the mess which allocates a package
data structure in CPU prepare and when it turns out to already exist in
starting throw it away later in the CPU online callback. But that's a
horrible hack and not required at all because RAPL becomes functional for
perf only in the CPU online callback. That's correct because user space is
not yet informed about the CPU being onlined, so nothing caan rely on RAPL
being available on that particular CPU.

Move the allocation to the CPU online callback and simplify the hotplug
handling. At this point the package mapping is established and correct.

This also adds a missing check for available package data in the
event_init() function.

Reported-by: Yasuaki Ishimatsu 
Signed-off-by: Thomas Gleixner 
Cc: Alexander Shishkin 
Cc: Arnaldo Carvalho de Melo 
Cc: Jiri Olsa 
Cc: Linus Torvalds 
Cc: Peter Zijlstra 
Cc: Sebastian Siewior 
Cc: Stephane Eranian 
Cc: Vince Weaver 
Fixes: 9d85eb9119f4 ("x86/smpboot: Make logical package management more robust")
Link: http://lkml.kernel.org/r/20170131230141.212593966@linutronix.de
Signed-off-by: Ingo Molnar 
[ jwang: backport to 4.9 fix Null pointer deref during hotplug cpu.]
Signed-off-by: Jack Wang 
Signed-off-by: Greg Kroah-Hartman

bpf: fix mixed signed/unsigned derived min/max value bounds

2017-08-30T08:21:43+00:00

[ Upstream commit 4cabc5b186b5427b9ee5a7495172542af105f02b ]

Edward reported that there's an issue in min/max value bounds
tracking when signed and unsigned compares both provide hints
on limits when having unknown variables. E.g. a program such
as the following should have been rejected:

   0: (7a) *(u64 *)(r10 -8) = 0
   1: (bf) r2 = r10
   2: (07) r2 += -8
   3: (18) r1 = 0xffff8a94cda93400
   5: (85) call bpf_map_lookup_elem#1
   6: (15) if r0 == 0x0 goto pc+7
  R0=map_value(ks=8,vs=8,id=0),min_value=0,max_value=0 R10=fp
   7: (7a) *(u64 *)(r10 -16) = -8
   8: (79) r1 = *(u64 *)(r10 -16)
   9: (b7) r2 = -1
  10: (2d) if r1 > r2 goto pc+3
  R0=map_value(ks=8,vs=8,id=0),min_value=0,max_value=0 R1=inv,min_value=0
  R2=imm-1,max_value=18446744073709551615,min_align=1 R10=fp
  11: (65) if r1 s> 0x1 goto pc+2
  R0=map_value(ks=8,vs=8,id=0),min_value=0,max_value=0 R1=inv,min_value=0,max_value=1
  R2=imm-1,max_value=18446744073709551615,min_align=1 R10=fp
  12: (0f) r0 += r1
  13: (72) *(u8 *)(r0 +0) = 0
  R0=map_value_adj(ks=8,vs=8,id=0),min_value=0,max_value=1 R1=inv,min_value=0,max_value=1
  R2=imm-1,max_value=18446744073709551615,min_align=1 R10=fp
  14: (b7) r0 = 0
  15: (95) exit

What happens is that in the first part ...

   8: (79) r1 = *(u64 *)(r10 -16)
   9: (b7) r2 = -1
  10: (2d) if r1 > r2 goto pc+3

... r1 carries an unsigned value, and is compared as unsigned
against a register carrying an immediate. Verifier deduces in
reg_set_min_max() that since the compare is unsigned and operation
is greater than (>), that in the fall-through/false case, r1's
minimum bound must be 0 and maximum bound must be r2. Latter is
larger than the bound and thus max value is reset back to being
'invalid' aka BPF_REGISTER_MAX_RANGE. Thus, r1 state is now
'R1=inv,min_value=0'. The subsequent test ...

  11: (65) if r1 s> 0x1 goto pc+2

... is a signed compare of r1 with immediate value 1. Here,
verifier deduces in reg_set_min_max() that since the compare
is signed this time and operation is greater than (>), that
in the fall-through/false case, we can deduce that r1's maximum
bound must be 1, meaning with prior test, we result in r1 having
the following state: R1=inv,min_value=0,max_value=1. Given that
the actual value this holds is -8, the bounds are wrongly deduced.
When this is being added to r0 which holds the map_value(_adj)
type, then subsequent store access in above case will go through
check_mem_access() which invokes check_map_access_adj(), that
will then probe whether the map memory is in bounds based
on the min_value and max_value as well as access size since
the actual unknown value is min_value <= x <= max_value; commit
fce366a9dd0d ("bpf, verifier: fix alu ops against map_value{,
_adj} register types") provides some more explanation on the
semantics.

It's worth to note in this context that in the current code,
min_value and max_value tracking are used for two things, i)
dynamic map value access via check_map_access_adj() and since
commit 06c1c049721a ("bpf: allow helpers access to variable memory")
ii) also enforced at check_helper_mem_access() when passing a
memory address (pointer to packet, map value, stack) and length
pair to a helper and the length in this case is an unknown value
defining an access range through min_value/max_value in that
case. The min_value/max_value tracking is /not/ used in the
direct packet access case to track ranges. However, the issue
also affects case ii), for example, the following crafted program
based on the same principle must be rejected as well:

   0: (b7) r2 = 0
   1: (bf) r3 = r10
   2: (07) r3 += -512
   3: (7a) *(u64 *)(r10 -16) = -8
   4: (79) r4 = *(u64 *)(r10 -16)
   5: (b7) r6 = -1
   6: (2d) if r4 > r6 goto pc+5
  R1=ctx R2=imm0,min_value=0,max_value=0,min_align=2147483648 R3=fp-512
  R4=inv,min_value=0 R6=imm-1,max_value=18446744073709551615,min_align=1 R10=fp
   7: (65) if r4 s> 0x1 goto pc+4
  R1=ctx R2=imm0,min_value=0,max_value=0,min_align=2147483648 R3=fp-512
  R4=inv,min_value=0,max_value=1 R6=imm-1,max_value=18446744073709551615,min_align=1
  R10=fp
   8: (07) r4 += 1
   9: (b7) r5 = 0
  10: (6a) *(u16 *)(r10 -512) = 0
  11: (85) call bpf_skb_load_bytes#26
  12: (b7) r0 = 0
  13: (95) exit

Meaning, while we initialize the max_value stack slot that the
verifier thinks we access in the [1,2] range, in reality we
pass -7 as length which is interpreted as u32 in the helper.
Thus, this issue is relevant also for the case of helper ranges.
Resetting both bounds in check_reg_overflow() in case only one
of them exceeds limits is also not enough as similar test can be
created that uses values which are within range, thus also here
learned min value in r1 is incorrect when mixed with later signed
test to create a range:

   0: (7a) *(u64 *)(r10 -8) = 0
   1: (bf) r2 = r10
   2: (07) r2 += -8
   3: (18) r1 = 0xffff880ad081fa00
   5: (85) call bpf_map_lookup_elem#1
   6: (15) if r0 == 0x0 goto pc+7
  R0=map_value(ks=8,vs=8,id=0),min_value=0,max_value=0 R10=fp
   7: (7a) *(u64 *)(r10 -16) = -8
   8: (79) r1 = *(u64 *)(r10 -16)
   9: (b7) r2 = 2
  10: (3d) if r2 >= r1 goto pc+3
  R0=map_value(ks=8,vs=8,id=0),min_value=0,max_value=0 R1=inv,min_value=3
  R2=imm2,min_value=2,max_value=2,min_align=2 R10=fp
  11: (65) if r1 s> 0x4 goto pc+2
  R0=map_value(ks=8,vs=8,id=0),min_value=0,max_value=0
  R1=inv,min_value=3,max_value=4 R2=imm2,min_value=2,max_value=2,min_align=2 R10=fp
  12: (0f) r0 += r1
  13: (72) *(u8 *)(r0 +0) = 0
  R0=map_value_adj(ks=8,vs=8,id=0),min_value=3,max_value=4
  R1=inv,min_value=3,max_value=4 R2=imm2,min_value=2,max_value=2,min_align=2 R10=fp
  14: (b7) r0 = 0
  15: (95) exit

This leaves us with two options for fixing this: i) to invalidate
all prior learned information once we switch signed context, ii)
to track min/max signed and unsigned boundaries separately as
done in [0]. (Given latter introduces major changes throughout
the whole verifier, it's rather net-next material, thus this
patch follows option i), meaning we can derive bounds either
from only signed tests or only unsigned tests.) There is still the
case of adjust_reg_min_max_vals(), where we adjust bounds on ALU
operations, meaning programs like the following where boundaries
on the reg get mixed in context later on when bounds are merged
on the dst reg must get rejected, too:

   0: (7a) *(u64 *)(r10 -8) = 0
   1: (bf) r2 = r10
   2: (07) r2 += -8
   3: (18) r1 = 0xffff89b2bf87ce00
   5: (85) call bpf_map_lookup_elem#1
   6: (15) if r0 == 0x0 goto pc+6
  R0=map_value(ks=8,vs=8,id=0),min_value=0,max_value=0 R10=fp
   7: (7a) *(u64 *)(r10 -16) = -8
   8: (79) r1 = *(u64 *)(r10 -16)
   9: (b7) r2 = 2
  10: (3d) if r2 >= r1 goto pc+2
  R0=map_value(ks=8,vs=8,id=0),min_value=0,max_value=0 R1=inv,min_value=3
  R2=imm2,min_value=2,max_value=2,min_align=2 R10=fp
  11: (b7) r7 = 1
  12: (65) if r7 s> 0x0 goto pc+2
  R0=map_value(ks=8,vs=8,id=0),min_value=0,max_value=0 R1=inv,min_value=3
  R2=imm2,min_value=2,max_value=2,min_align=2 R7=imm1,max_value=0 R10=fp
  13: (b7) r0 = 0
  14: (95) exit

  from 12 to 15: R0=map_value(ks=8,vs=8,id=0),min_value=0,max_value=0
  R1=inv,min_value=3 R2=imm2,min_value=2,max_value=2,min_align=2 R7=imm1,min_value=1 R10=fp
  15: (0f) r7 += r1
  16: (65) if r7 s> 0x4 goto pc+2
  R0=map_value(ks=8,vs=8,id=0),min_value=0,max_value=0 R1=inv,min_value=3
  R2=imm2,min_value=2,max_value=2,min_align=2 R7=inv,min_value=4,max_value=4 R10=fp
  17: (0f) r0 += r7
  18: (72) *(u8 *)(r0 +0) = 0
  R0=map_value_adj(ks=8,vs=8,id=0),min_value=4,max_value=4 R1=inv,min_value=3
  R2=imm2,min_value=2,max_value=2,min_align=2 R7=inv,min_value=4,max_value=4 R10=fp
  19: (b7) r0 = 0
  20: (95) exit

Meaning, in adjust_reg_min_max_vals() we must also reset range
values on the dst when src/dst registers have mixed signed/
unsigned derived min/max value bounds with one unbounded value
as otherwise they can be added together deducing false boundaries.
Once both boundaries are established from either ALU ops or
compare operations w/o mixing signed/unsigned insns, then they
can safely be added to other regs also having both boundaries
established. Adding regs with one unbounded side to a map value
where the bounded side has been learned w/o mixing ops is
possible, but the resulting map value won't recover from that,
meaning such op is considered invalid on the time of actual
access. Invalid bounds are set on the dst reg in case i) src reg,
or ii) in case dst reg already had them. The only way to recover
would be to perform i) ALU ops but only 'add' is allowed on map
value types or ii) comparisons, but these are disallowed on
pointers in case they span a range. This is fine as only BPF_JEQ
and BPF_JNE may be performed on PTR_TO_MAP_VALUE_OR_NULL registers
which potentially turn them into PTR_TO_MAP_VALUE type depending
on the branch, so only here min/max value cannot be invalidated
for them.

In terms of state pruning, value_from_signed is considered
as well in states_equal() when dealing with adjusted map values.
With regards to breaking existing programs, there is a small
risk, but use-cases are rather quite narrow where this could
occur and mixing compares probably unlikely.

Joint work with Josef and Edward.

  [0] https://lists.iovisor.org/pipermail/iovisor-dev/2017-June/000822.html

Fixes: 484611357c19 ("bpf: allow access into map value arrays")
Reported-by: Edward Cree 
Signed-off-by: Daniel Borkmann 
Signed-off-by: Edward Cree 
Signed-off-by: Josef Bacik 
Signed-off-by: David S. Miller 
Signed-off-by: Greg Kroah-Hartman

ptr_ring: use kmalloc_array()

2017-08-30T08:21:40+00:00

[ Upstream commit 81fbfe8adaf38d4f5a98c19bebfd41c5d6acaee8 ]

As found by syzkaller, malicious users can set whatever tx_queue_len
on a tun device and eventually crash the kernel.

Lets remove the ALIGN(XXX, SMP_CACHE_BYTES) thing since a small
ring buffer is not fast anyway.

Fixes: 2e0ab8ca83c1 ("ptr_ring: array based FIFO for pointers")
Signed-off-by: Eric Dumazet 
Reported-by: Dmitry Vyukov 
Cc: Michael S. Tsirkin 
Cc: Jason Wang 
Signed-off-by: David S. Miller 
Signed-off-by: Greg Kroah-Hartman

pids: make task_tgid_nr_ns() safe

2017-08-25T00:12:21+00:00

commit dd1c1f2f2028a7b851f701fc6a8ebe39dcb95e7c upstream.

This was reported many times, and this was even mentioned in commit
52ee2dfdd4f5 ("pids: refactor vnr/nr_ns helpers to make them safe") but
somehow nobody bothered to fix the obvious problem: task_tgid_nr_ns() is
not safe because task->group_leader points to nowhere after the exiting
task passes exit_notify(), rcu_read_lock() can not help.

We really need to change __unhash_process() to nullify group_leader,
parent, and real_parent, but this needs some cleanups.  Until then we
can turn task_tgid_nr_ns() into another user of __task_pid_nr_ns() and
fix the problem.

Reported-by: Troy Kensinger 
Signed-off-by: Oleg Nesterov 
Signed-off-by: Linus Torvalds 
Signed-off-by: Greg Kroah-Hartman

mm: discard memblock data later

2017-08-25T00:12:19+00:00

commit 3010f876500f9ba921afaeccec30c45ca6584dc8 upstream.

There is existing use after free bug when deferred struct pages are
enabled:

The memblock_add() allocates memory for the memory array if more than
128 entries are needed.  See comment in e820__memblock_setup():

  * The bootstrap memblock region count maximum is 128 entries
  * (INIT_MEMBLOCK_REGIONS), but EFI might pass us more E820 entries
  * than that - so allow memblock resizing.

This memblock memory is freed here:
        free_low_memory_core_early()

We access the freed memblock.memory later in boot when deferred pages
are initialized in this path:

        deferred_init_memmap()
                for_each_mem_pfn_range()
                  __next_mem_pfn_range()
                    type = &memblock.memory;

One possible explanation for why this use-after-free hasn't been hit
before is that the limit of INIT_MEMBLOCK_REGIONS has never been
exceeded at least on systems where deferred struct pages were enabled.

Tested by reducing INIT_MEMBLOCK_REGIONS down to 4 from the current 128,
and verifying in qemu that this code is getting excuted and that the
freed pages are sane.

Link: http://lkml.kernel.org/r/1502485554-318703-2-git-send-email-pasha.tatashin@oracle.com
Fixes: 7e18adb4f80b ("mm: meminit: initialise remaining struct pages in parallel with kswapd")
Signed-off-by: Pavel Tatashin 
Reviewed-by: Steven Sistare 
Reviewed-by: Daniel Jordan 
Reviewed-by: Bob Picco 
Acked-by: Michal Hocko 
Cc: Mel Gorman 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds 
Signed-off-by: Greg Kroah-Hartman

workqueue: implicit ordered attribute should be overridable

2017-08-11T15:49:37+00:00

commit 0a94efb5acbb6980d7c9ab604372d93cd507e4d8 upstream.

5c0338c68706 ("workqueue: restore WQ_UNBOUND/max_active==1 to be
ordered") automatically enabled ordered attribute for unbound
workqueues w/ max_active == 1.  Because ordered workqueues reject
max_active and some attribute changes, this implicit ordered mode
broke cases where the user creates an unbound workqueue w/ max_active
== 1 and later explicitly changes the related attributes.

This patch distinguishes explicit and implicit ordered setting and
overrides from attribute changes if implict.

Signed-off-by: Tejun Heo 
Fixes: 5c0338c68706 ("workqueue: restore WQ_UNBOUND/max_active==1 to be ordered")
Cc: Holger Hoffstätte 
Signed-off-by: Greg Kroah-Hartman

signal: protect SIGNAL_UNKILLABLE from unintentional clearing.

2017-08-11T15:49:36+00:00

[ Upstream commit 2d39b3cd34e6d323720d4c61bd714f5ae202c022 ]

Since commit 00cd5c37afd5 ("ptrace: permit ptracing of /sbin/init") we
can now trace init processes.  init is initially protected with
SIGNAL_UNKILLABLE which will prevent fatal signals such as SIGSTOP, but
there are a number of paths during tracing where SIGNAL_UNKILLABLE can
be implicitly cleared.

This can result in init becoming stoppable/killable after tracing.  For
example, running:

  while true; do kill -STOP 1; done &
  strace -p 1

and then stopping strace and the kill loop will result in init being
left in state TASK_STOPPED.  Sending SIGCONT to init will resume it, but
init will now respond to future SIGSTOP signals rather than ignoring
them.

Make sure that when setting SIGNAL_STOP_CONTINUED/SIGNAL_STOP_STOPPED
that we don't clear SIGNAL_UNKILLABLE.

Link: http://lkml.kernel.org/r/20170104122017.25047-1-jamie.iles@oracle.com
Signed-off-by: Jamie Iles 
Acked-by: Oleg Nesterov 
Cc: Alexander Viro 
Cc: Ingo Molnar 
Cc: Peter Zijlstra 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds 
Signed-off-by: Sasha Levin 
Signed-off-by: Greg Kroah-Hartman

mm, slab: make sure that KMALLOC_MAX_SIZE will fit into MAX_ORDER

2017-08-11T15:49:36+00:00

[ Upstream commit bb1107f7c6052c863692a41f78c000db792334bf ]

Andrey Konovalov has reported the following warning triggered by the
syzkaller fuzzer.

  WARNING: CPU: 1 PID: 9935 at mm/page_alloc.c:3511 __alloc_pages_nodemask+0x159c/0x1e20
  Kernel panic - not syncing: panic_on_warn set ...
  CPU: 1 PID: 9935 Comm: syz-executor0 Not tainted 4.9.0-rc7+ #34
  Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Bochs 01/01/2011
  Call Trace:
    __alloc_pages_slowpath mm/page_alloc.c:3511
    __alloc_pages_nodemask+0x159c/0x1e20 mm/page_alloc.c:3781
    alloc_pages_current+0x1c7/0x6b0 mm/mempolicy.c:2072
    alloc_pages include/linux/gfp.h:469
    kmalloc_order+0x1f/0x70 mm/slab_common.c:1015
    kmalloc_order_trace+0x1f/0x160 mm/slab_common.c:1026
    kmalloc_large include/linux/slab.h:422
    __kmalloc+0x210/0x2d0 mm/slub.c:3723
    kmalloc include/linux/slab.h:495
    ep_write_iter+0x167/0xb50 drivers/usb/gadget/legacy/inode.c:664
    new_sync_write fs/read_write.c:499
    __vfs_write+0x483/0x760 fs/read_write.c:512
    vfs_write+0x170/0x4e0 fs/read_write.c:560
    SYSC_write fs/read_write.c:607
    SyS_write+0xfb/0x230 fs/read_write.c:599
    entry_SYSCALL_64_fastpath+0x1f/0xc2

The issue is caused by a lack of size check for the request size in
ep_write_iter which should be fixed.  It, however, points to another
problem, that SLUB defines KMALLOC_MAX_SIZE too large because the its
KMALLOC_SHIFT_MAX is (MAX_ORDER + PAGE_SHIFT) which means that the
resulting page allocator request might be MAX_ORDER which is too large
(see __alloc_pages_slowpath).

The same applies to the SLOB allocator which allows even larger sizes.
Make sure that they are capped properly and never request more than
MAX_ORDER order.

Link: http://lkml.kernel.org/r/20161220130659.16461-2-mhocko@kernel.org
Signed-off-by: Michal Hocko 
Reported-by: Andrey Konovalov 
Acked-by: Christoph Lameter 
Cc: Alexei Starovoitov 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds 
Signed-off-by: Sasha Levin 
Signed-off-by: Greg Kroah-Hartman

cpuset: fix a deadlock due to incomplete patching of cpusets_enabled()

2017-08-11T15:49:29+00:00

commit 89affbf5d9ebb15c6460596822e8857ea2f9e735 upstream.

In codepaths that use the begin/retry interface for reading
mems_allowed_seq with irqs disabled, there exists a race condition that
stalls the patch process after only modifying a subset of the
static_branch call sites.

This problem manifested itself as a deadlock in the slub allocator,
inside get_any_partial.  The loop reads mems_allowed_seq value (via
read_mems_allowed_begin), performs the defrag operation, and then
verifies the consistency of mem_allowed via the read_mems_allowed_retry
and the cookie returned by xxx_begin.

The issue here is that both begin and retry first check if cpusets are
enabled via cpusets_enabled() static branch.  This branch can be
rewritted dynamically (via cpuset_inc) if a new cpuset is created.  The
x86 jump label code fully synchronizes across all CPUs for every entry
it rewrites.  If it rewrites only one of the callsites (specifically the
one in read_mems_allowed_retry) and then waits for the
smp_call_function(do_sync_core) to complete while a CPU is inside the
begin/retry section with IRQs off and the mems_allowed value is changed,
we can hang.

This is because begin() will always return 0 (since it wasn't patched
yet) while retry() will test the 0 against the actual value of the seq
counter.

The fix is to use two different static keys: one for begin
(pre_enable_key) and one for retry (enable_key).  In cpuset_inc(), we
first bump the pre_enable key to ensure that cpuset_mems_allowed_begin()
always return a valid seqcount if are enabling cpusets.  Similarly, when
disabling cpusets via cpuset_dec(), we first ensure that callers of
cpuset_mems_allowed_retry() will start ignoring the seqcount value
before we let cpuset_mems_allowed_begin() return 0.

The relevant stack traces of the two stuck threads:

  CPU: 1 PID: 1415 Comm: mkdir Tainted: G L  4.9.36-00104-g540c51286237 #4
  Hardware name: Default string Default string/Hardware, BIOS 4.29.1-20170526215256 05/26/2017
  task: ffff8817f9c28000 task.stack: ffffc9000ffa4000
  RIP: smp_call_function_many+0x1f9/0x260
  Call Trace:
    smp_call_function+0x3b/0x70
    on_each_cpu+0x2f/0x90
    text_poke_bp+0x87/0xd0
    arch_jump_label_transform+0x93/0x100
    __jump_label_update+0x77/0x90
    jump_label_update+0xaa/0xc0
    static_key_slow_inc+0x9e/0xb0
    cpuset_css_online+0x70/0x2e0
    online_css+0x2c/0xa0
    cgroup_apply_control_enable+0x27f/0x3d0
    cgroup_mkdir+0x2b7/0x420
    kernfs_iop_mkdir+0x5a/0x80
    vfs_mkdir+0xf6/0x1a0
    SyS_mkdir+0xb7/0xe0
    entry_SYSCALL_64_fastpath+0x18/0xad

  ...

  CPU: 2 PID: 1 Comm: init Tainted: G L  4.9.36-00104-g540c51286237 #4
  Hardware name: Default string Default string/Hardware, BIOS 4.29.1-20170526215256 05/26/2017
  task: ffff8818087c0000 task.stack: ffffc90000030000
  RIP: int3+0x39/0x70
  Call Trace:
    <#DB> ? ___slab_alloc+0x28b/0x5a0
     ? copy_process.part.40+0xf7/0x1de0
    __slab_alloc.isra.80+0x54/0x90
    copy_process.part.40+0xf7/0x1de0
    copy_process.part.40+0xf7/0x1de0
    kmem_cache_alloc_node+0x8a/0x280
    copy_process.part.40+0xf7/0x1de0
    _do_fork+0xe7/0x6c0
    _raw_spin_unlock_irq+0x2d/0x60
    trace_hardirqs_on_caller+0x136/0x1d0
    entry_SYSCALL_64_fastpath+0x5/0xad
    do_syscall_64+0x27/0x350
    SyS_clone+0x19/0x20
    do_syscall_64+0x60/0x350
    entry_SYSCALL64_slow_path+0x25/0x25

Link: http://lkml.kernel.org/r/20170731040113.14197-1-dmitriyz@waymo.com
Fixes: 46e700abc44c ("mm, page_alloc: remove unnecessary taking of a seqlock when cpusets are disabled")
Signed-off-by: Dima Zavin 
Reported-by: Cliff Spradlin 
Acked-by: Vlastimil Babka 
Cc: Peter Zijlstra 
Cc: Christopher Lameter 
Cc: Li Zefan 
Cc: Pekka Enberg 
Cc: David Rientjes 
Cc: Joonsoo Kim 
Cc: Mel Gorman 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds 
Signed-off-by: Greg Kroah-Hartman