sched/headers: Remove <linux/sched.h> from <linux/sched/mm.h>

The <linux/sched/mm.h> file is a self-contained header and users of
it either don't need <linux/sched.h> - or have already included it.

Include kernel.h and atomic.h.

This reduces the size of the header dependency graph.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Remove <linux/sched.h> from <linux/sched/autogroup.h>

The <linux/sched/autogroup.h> file is a largely self-contained header and users of
it either don't need <linux/sched.h> - or have already included it.

Add a 'task_struct' predeclaration to make it build standalone.

This reduces the size of the header dependency graph.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Remove <linux/sched.h> from <linux/sched/loadavg.h>

The <linux/sched/loadavg.h> file is a self-contained header and users of
it either don't need <linux/sched.h> - or have already included it.

This reduces the size of the header dependency graph.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Remove <linux/sched.h> and <linux/slab.h> from <linux/delayacct.h>

The <linux/delayacct.h> file is a self-contained header and users of
it either don't need <linux/sched.h> and <linux/slab.h> - or have
already included it.

This reduces the size of the header dependency graph.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Remove <linux/sched.h> from <linux/sched/clock.h>

The <linux/sched/clock.h> file is a largely self-contained header and users of
it either don't need <linux/sched.h> - or have already included it.

This reduces the size of the header dependency graph.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Remove tsk_is_polling()

It's not used by anything.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Remove <linux/sched.h> from <linux/sched/topology.h>

The <linux/sched/topology.h> file is a self-contained header and users of
it either don't need <linux/sched.h> - or have already included it.

This reduces the size of the header dependency graph.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Move the sched_exec() prototype to <linux/sched/task.h>

sched_exec() better fits into the task lifetime APIs than into the core scheduler
APIs.

This reduces the size of <linux/sched.h> a bit.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Move the get_task_struct()/put_task_struct() and related APIs from <linux/sched.h> to <linux/sched/task.h>

These belong into the task lifetime API header.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers, mm: Move 'struct tlbflush_unmap_batch' from <linux/sched.h> to <linux/mm_types_task.h>

Unclutter <linux/sched.h> some more.

Also move the CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH condition inside the
structure body definition, to remove a pair of #ifdefs from sched.h.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers, timekeeping: Move the timer tick function prototypes to <linux/timekeeping.h>

Move the update_process_times() and xtime_update() prototypes to <linux/timekeeping.h>.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Move _init() prototypes from <linux/sched.h> to <linux/sched/init.h>

trap_init() and cpu_init() belong into <linux/cpu.h>, sched_init*() into <linux/sched/init.h>.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Move exit_files() and exit_itimers() from <linux/sched.h> to <linux/sched/task.h>

These two functions are task management related, not core scheduler APIs.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Move kstack_end() from <linux/sched.h> to <linux/sched/task_stack.h>

This is a task-stack related API, not core scheduler functionality.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers, vfs/execve: Move the do_execve*() prototypes from <linux/sched.h> to <linux/binfmts.h>

These are not scheduler related.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Move sched_info_on() and force_schedstat_enabled() from <linux/sched.h> to <linux/sched/stat.h>

These APIs are not core scheduler but scheduler statistics related.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Move cputime functionality from <linux/sched.h> and <linux/cputime.h> into <linux/sched/cputime.h>

Move cputime related functionality out of <linux/sched.h>, as most code
that includes <linux/sched.h> does not use that functionality.

Move data types that are not included in task_struct directly to
the signal definitions, into <linux/sched/signal.h>.

Also merge the (small) existing <linux/cputime.h> header into <linux/sched/cputime.h>.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Move the task_lock()/unlock() APIs to <linux/sched/task.h>

The task_lock()/task_unlock() APIs are not realated to core scheduling,
they are task lifetime APIs, i.e. they belong into <linux/sched/task.h>.

Move them.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Move 'init_task' and 'init_thread_union' from <linux/sched.h> to <linux/sched/task.h>

'init_task' is really not part of core scheduler APIs but part of
the fork() interface between the scheduler and process management.

So move the declarations.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Remove the <linux/mm_types.h> dependency from <linux/sched.h>

Use the freshly introduced, reduced size <linux/mm_types_task.h> header instead.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

mm/headers, sched/headers: Move task related MM types from <linux/mm_types.> to <linux/mm_types_task.h>

Separate all the MM types that are embedded directly in 'struct task_struct'
into the <linux/mm_types_task.h> header.

The goal is to include this header in <linux/sched.h>, not the full <linux/mm_types.h>
header, to reduce the size, complexity and coupling of <linux/sched.h>.

(This patch does not change <linux/sched.h> yet.)

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

signals: Move signal data types from <linux/signal.h> to <linux/signal_types.h>

Separate out just the pure data types - sched.h will be able to use
this reduced size header.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Move the 'root_task_group' declaration to <linux/sched/autogroup.h>

Also remove the duplicate declaration from <linux/init_task.h>.

( That declaration was originally duplicated for dependency hell reasons,
  but there's no problem including the much smaller <linux/sched/autogroup.h>
  header now, to pick up the right prototype. )

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Move task-stack related APIs from <linux/sched.h> to <linux/sched/task_stack.h>

Split out the task->stack related functionality, which is not really
part of the core scheduler APIs.

Only keep task_thread_info() because it's used by sched.h.

Update the code that uses those facilities.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Move the PREEMPT_COUNT defines from <linux/sched.h> to <linux/preempt.h>

These defines are not really part of the scheduler's driver API, but are
related to the preempt count - so move them to <linux/preempt.h>.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/core: Move, sort and clean up <linux/sched.h> structure predeclarations

Most of the structure predeclarations were at the head of sched.h, but not
all of them - there were a number of lines spread around sched.h, in
random places.

Move them to the head, and also sort them alphabetically.

Remove unused entries.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers, RCU: Move rcu_copy_process() from <linux/sched/task.h> to kernel/fork.c

Move rcu_copy_process() into kernel/fork.c, which is the only
user of this inline function.

This simplifies <linux/sched/task.h> to the level that <linux/sched.h>
does not have to be included in it anymore - which change is done
in a subsequent patch.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Move task lifetime APIs from <linux/sched.h> to <linux/sched/task.h>

There's a fair amount of task lifetime management (a.k.a fork()/exit())
related APIs in <linux/sched.h>, but only a small fraction of
the users of the generic sched.h header make use of them.

Move these functions to the <linux/sched/task.h> header.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Split hotplug CPU interfaces out of <linux/sched.h> into <linux/sched/hotplug.h>

Split the CPU hotplug scheduler APIs out of the common header.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Move <asm/current.h> include from the middle of <linux/sched.h> to the header portion

Linux-0.01 already defined 'current' in the middle of sched.h, so this
is an ancient historical precedent - but still in a modern kernel it
looks a bit weird that we have:

#include <asm/current.h>

in the middle of the header.

Move it further up. If this was done for some obscure dependency
reasons then we'll trigger and document it.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers, time/timekeeping: Move the xtime_update() prototype from <linux/sched.h> to <linux/time.h>

This was in <linux/sched.h> only for hysterical raisins.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Move debugging functions from <linux/sched.h> to <linux/sched/debug.h>

Collect the various scheduler and task state debugging APIs scattered
around <linux/sched.h> into the new <linux/sched/debug.h> header.

In particular the show_regs() and show_stack() prototype affects many files,
update them.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Move the NOHZ APIs from <linux/sched.h> to <linux/sched/nohz.h>

There's a number of NOHZ/dyntics related functionality in <linux/sched.h>,
but only a handful of timer files are making use of them.

Move them into their own header. This better documents these APIs
and unclutters <linux/sched.h>.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Move task statistics APIs from <linux/sched.h> to <linux/sched/stat.h>

There are a number of task statistics related variables and methods exported
via sched.h - collect them into <linux/sched/stat.h> and include it from
their usage sites.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Move the memalloc_noio_*() APIs to <linux/sched/mm.h>

In preparation to remove the <linux/gfp.h> include from <linux/sched.h>.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Move signal wakeup & sigpending methods from <linux/sched.h> into <linux/sched/signal.h>

This reduces the size of <linux/sched.h>.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Move CONFIG_TASK_XACCT bits from <linux/sched.h> to <linux/sched/xacct.h>

The CONFIG_TASK_XACCT=y accounting inline functions are only used by
fs/read_write.c, so move them into their separate header.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Remove various unrelated headers from <linux/sched.h>

Remove the following header inclusions from <linux/sched.h>:

#include <asm/param.h>
#include <linux/threads.h>
#include <linux/kernel.h>
#include <linux/types.h>
#include <linux/timex.h>
#include <linux/jiffies.h>
#include <linux/rbtree.h>
#include <linux/thread_info.h>
#include <linux/cpumask.h>
#include <linux/errno.h>
#include <linux/nodemask.h>
#include <linux/preempt.h>
#include <asm/page.h>
#include <linux/smp.h>
#include <linux/compiler.h>
#include <linux/completion.h>
#include <linux/percpu.h>
#include <linux/topology.h>
#include <linux/rcupdate.h>
#include <linux/time.h>
#include <linux/timer.h>
#include <linux/llist.h>
#include <linux/uidgid.h>
#include <asm/processor.h>

because they are either not required, or are already included
naturally as part of the remaining headers.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Move the JOBCTL_ defines and methods from <linux/sched.h> to <linux/sched/jobctl.h>

Only a small fraction of sched.h users actually utilizes these defines,
and they are not scheduler functionality in any case, so move them
into their separate header.

(Also make <linux/sched/jobctl.h> a self-contained header.)

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Move the NUMA balancing interfaces from <linux/sched.h> to <linux/sched/numa_balancing.h>

Split out the interface between the scheduler and the MM which
deals with page fault driven NUMA balancing, into the new
<linux/sched/numa_balancing.h> header.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Move in_vfork() from <linux/sched.h> to <linux/sched/mm.h>

The in_vfork() function deals with task->mm, so it better belongs
into <linux/sched/mm.h>.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Move the 'init_mm' declaration from <linux/sched.h> to <linux/mm_types.h>

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Move more mm_struct related functionality from <linux/sched.h> to <linux/sched/mm.h>

Neither the mmap_layout nor the mm_update_next_owner() methods need to be
in <linux/sched.h> - move them to the more appropriate <linux/sched/mm.h> header.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Move the cpufreq interfaces to <linux/sched/cpufreq.h>

No need to have this in the generic <linux/sched.h> header.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Move 'struct pacct_struct' and 'struct cpu_itimer' form <linux/sched.h> to <linux/sched/signal.h>

These structures are actually part of 'struct signal', so move them to <linux/sched/signal.h>
where they belong.

This further decreases the size and complexity of <linux/sched.h>.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Move softlockup detector watchdog methods to <linux/nmi.h>

These methods don't belong into <linux/sched.h>, they are neither directly
related to task_struct or are scheduler functionality.

Put them next to the other watchdog methods in <linux/nmi.h>.

( Arguably that header's name is a misnomer, and this patch makes it
  more so - but it should be renamed in another patch. )

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Move 'struct user_struct' definition and APIs to the new <linux/sched/user.h> header

'struct user_struct' was added to sched.h historically, but it's actually
entirely independent of task_struct and of scheduler details, so move
it to its own header.

Fix up .c files using those facilities.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Move task_struct::signal and task_struct::sighand types and accessors into <linux/sched/signal.h>

task_struct::signal and task_struct::sighand are pointers, which would normally make it
straightforward to not define those types in sched.h.

That is not so, because the types are accompanied by a myriad of APIs (macros and inline
functions) that dereference them.

Split the types and the APIs out of sched.h and move them into a new header, <linux/sched/signal.h>.

With this change sched.h does not know about 'struct signal' and 'struct sighand' anymore,
trying to put accessors into sched.h as a test fails the following way:

  ./include/linux/sched.h: In function ‘test_signal_types’:
  ./include/linux/sched.h:2461:18: error: dereferencing pointer to incomplete type ‘struct signal_struct’
                    ^

This reduces the size and complexity of sched.h significantly.

Update all headers and .c code that relied on getting the signal handling
functionality from <linux/sched.h> to include <linux/sched/signal.h>.

The list of affected files in the preparatory patch was partly generated by
grepping for the APIs, and partly by doing coverage build testing, both
all[yes|mod|def|no]config builds on 64-bit and 32-bit x86, and an array of
cross-architecture builds.

Nevertheless some (trivial) build breakage is still expected related to rare
Kconfig combinations and in-flight patches to various kernel code, but most
of it should be handled by this patch.

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Move task->mm coredumping related defines and methods from <linux/sched.h> to <linux/sched/coredump.h>

This further reduces the size and complexity of <linux/sched.h>.

These are the definitions and APIs that are moved:

  # MMF_*:
  fs/binfmt_elf.c
  fs/binfmt_elf_fdpic.c
  fs/exec.c
  fs/proc/base.c
  include/linux/khugepaged.h
  include/linux/ksm.h
  include/linux/sched/coredump.h
  kernel/events/uprobes.c
  kernel/fork.c
  mm/huge_memory.c
  mm/khugepaged.c
  mm/ksm.c
  mm/memory.c
  mm/oom_kill.c

  # SUID_DUMP_*:
  arch/ia64/include/asm/processor.h
  fs/coredump.c
  fs/exec.c
  fs/proc/internal.h
  include/linux/sched/coredump.h
  kernel/ptrace.c
  kernel/sys.c
  kernel/sysctl.c

  # get_dumpable():
  arch/ia64/include/asm/processor.h
  fs/coredump.c
  fs/exec.c
  fs/proc/internal.h
  include/linux/sched/coredump.h
  kernel/ptrace.c
  kernel/sys.c

  # set_dumpable():
  fs/exec.c
  include/linux/sched/coredump.h
  kernel/cred.c
  kernel/sys.c

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>

sched/headers: Move task->mm handling methods to <linux/sched/mm.h>

Move the following task->mm helper APIs into a new header file,
<linux/sched/mm.h>, to further reduce the size and complexity
of <linux/sched.h>.

Here are how the APIs are used in various kernel files:

  # mm_alloc():
  arch/arm/mach-rpc/ecard.c
  fs/exec.c
  include/linux/sched/mm.h
  kernel/fork.c

  # __mmdrop():
  arch/arc/include/asm/mmu_context.h
  include/linux/sched/mm.h
  kernel/fork.c

  # mmdrop():
  arch/arm/mach-rpc/ecard.c
  arch/m68k/sun3/mmu_emu.c
  arch/x86/mm/tlb.c
  drivers/gpu/drm/amd/amdkfd/kfd_process.c
  drivers/gpu/drm/i915/i915_gem_userptr.c
  drivers/infiniband/hw/hfi1/file_ops.c
  drivers/vfio/vfio_iommu_spapr_tce.c
  fs/exec.c
  fs/proc/base.c
  fs/proc/task_mmu.c
  fs/proc/task_nommu.c
  fs/userfaultfd.c
  include/linux/mmu_notifier.h
  include/linux/sched/mm.h
  kernel/fork.c
  kernel/futex.c
  kernel/sched/core.c
  mm/khugepaged.c
  mm/ksm.c
  mm/mmu_context.c
  mm/mmu_notifier.c
  mm/oom_kill.c
  virt/kvm/kvm_main.c

  # mmdrop_async_fn():
  include/linux/sched/mm.h

  # mmdrop_async():
  include/linux/sched/mm.h
  kernel/fork.c

  # mmget_not_zero():
  fs/userfaultfd.c
  include/linux/sched/mm.h
  mm/oom_kill.c

  # mmput():
  arch/arc/include/asm/mmu_context.h
  arch/arc/kernel/troubleshoot.c
  arch/frv/mm/mmu-context.c
  arch/powerpc/platforms/cell/spufs/context.c
  arch/sparc/include/asm/mmu_context_32.h
  drivers/android/binder.c
  drivers/gpu/drm/etnaviv/etnaviv_gem.c
  drivers/gpu/drm/i915/i915_gem_userptr.c
  drivers/infiniband/core/umem.c
  drivers/infiniband/core/umem_odp.c
  drivers/infiniband/core/uverbs_main.c
  drivers/infiniband/hw/mlx4/main.c
  drivers/infiniband/hw/mlx5/main.c
  drivers/infiniband/hw/usnic/usnic_uiom.c
  drivers/iommu/amd_iommu_v2.c
  drivers/iommu/intel-svm.c
  drivers/lguest/lguest_user.c
  drivers/misc/cxl/fault.c
  drivers/misc/mic/scif/scif_rma.c
  drivers/oprofile/buffer_sync.c
  drivers/vfio/vfio_iommu_type1.c
  drivers/vhost/vhost.c
  drivers/xen/gntdev.c
  fs/exec.c
  fs/proc/array.c
  fs/proc/base.c
  fs/proc/task_mmu.c
  fs/proc/task_nommu.c
  fs/userfaultfd.c
  include/linux/sched/mm.h
  kernel/cpuset.c
  kernel/events/core.c
  kernel/events/uprobes.c
  kernel/exit.c
  kernel/fork.c
  kernel/ptrace.c
  kernel/sys.c
  kernel/trace/trace_output.c
  kernel/tsacct.c
  mm/memcontrol.c
  mm/memory.c
  mm/mempolicy.c
  mm/migrate.c
  mm/mmu_notifier.c
  mm/nommu.c
  mm/oom_kill.c
  mm/process_vm_access.c
  mm/rmap.c
  mm/swapfile.c
  mm/util.c
  virt/kvm/async_pf.c

  # mmput_async():
  include/linux/sched/mm.h
  kernel/fork.c
  mm/oom_kill.c

  # get_task_mm():
  arch/arc/kernel/troubleshoot.c
  arch/powerpc/platforms/cell/spufs/context.c
  drivers/android/binder.c
  drivers/gpu/drm/etnaviv/etnaviv_gem.c
  drivers/infiniband/core/umem.c
  drivers/infiniband/core/umem_odp.c
  drivers/infiniband/hw/mlx4/main.c
  drivers/infiniband/hw/mlx5/main.c
  drivers/infiniband/hw/usnic/usnic_uiom.c
  drivers/iommu/amd_iommu_v2.c
  drivers/iommu/intel-svm.c
  drivers/lguest/lguest_user.c
  drivers/misc/cxl/fault.c
  drivers/misc/mic/scif/scif_rma.c
  drivers/oprofile/buffer_sync.c
  drivers/vfio/vfio_iommu_type1.c
  drivers/vhost/vhost.c
  drivers/xen/gntdev.c
  fs/proc/array.c
  fs/proc/base.c
  fs/proc/task_mmu.c
  include/linux/sched/mm.h
  kernel/cpuset.c
  kernel/events/core.c
  kernel/exit.c
  kernel/fork.c
  kernel/ptrace.c
  kernel/sys.c
  kernel/trace/trace_output.c
  kernel/tsacct.c
  mm/memcontrol.c
  mm/memory.c
  mm/mempolicy.c
  mm/migrate.c
  mm/mmu_notifier.c
  mm/nommu.c
  mm/util.c

  # mm_access():
  fs/proc/base.c
  include/linux/sched/mm.h
  kernel/fork.c
  mm/process_vm_access.c

  # mm_release():
  arch/arc/include/asm/mmu_context.h
  fs/exec.c
  include/linux/sched/mm.h
  include/uapi/linux/sched.h
  kernel/exit.c
  kernel/fork.c

Acked-by: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

Merge branch 'for-linus-4.11' of git://git.kernel.org/pub/scm/linux/kernel/git/mason/linux-btrfs

Pull more btrfs updates from Chris Mason:
 "Btrfs round two.

  These are mostly a continuation of Dave Sterba's collection of
  cleanups, but Filipe also has some bug fixes and performance
  improvements"

* 'for-linus-4.11' of git://git.kernel.org/pub/scm/linux/kernel/git/mason/linux-btrfs: (69 commits)
  btrfs: add dummy callback for readpage_io_failed and drop checks
  btrfs: drop checks for mandatory extent_io_ops callbacks
  btrfs: document existence of extent_io ops callbacks
  btrfs: let writepage_end_io_hook return void
  btrfs: do proper error handling in btrfs_insert_xattr_item
  btrfs: handle allocation error in update_dev_stat_item
  btrfs: remove BUG_ON from __tree_mod_log_insert
  btrfs: derive maximum output size in the compression implementation
  btrfs: use predefined limits for calculating maximum number of pages for compression
  btrfs: export compression buffer limits in a header
  btrfs: merge nr_pages input and output parameter in compress_pages
  btrfs: merge length input and output parameter in compress_pages
  btrfs: constify name of subvolume in creation helpers
  btrfs: constify buffers used by compression helpers
  btrfs: constify input buffer of btrfs_csum_data
  btrfs: constify device path passed to relevant helpers
  btrfs: make btrfs_inode_resume_unlocked_dio take btrfs_inode
  btrfs: make btrfs_inode_block_unlocked_dio take btrfs_inode
  btrfs: Make btrfs_add_nondir take btrfs_inode
  btrfs: Make btrfs_add_link take btrfs_inode
  ...

Merge branch 'acpi-apei'

* acpi-apei:
  ACPI: APEI: Fix BERT resources conflict with ACPI NVS area

Merge branches 'pm-cpuidle', 'pm-cpufreq' and 'pm-sleep'

* pm-cpuidle:
  intel_idle: stop exposing platform acronyms in sysfs
  cpuidle: menu: Avoid taking spinlock for accessing QoS values

* pm-cpufreq:
  cpufreq: intel_pstate: Fix limits issue with operation mode switching
  cpufreq: qoriq: clean up unused code

* pm-sleep:
  PM / hibernate: Define pr_fmt() and use pr_*() instead of printk()
  PM / hibernate: Untangle power_down()

Merge branches 'pm-core', 'pm-qos', 'pm-domains' and 'pm-opp'

* pm-core:
  PM / runtime: Fix some typos

* pm-qos:
  PM / QoS: Remove global notifiers

* pm-domains:
  PM / Domains: Power off masters immediately in the power off sequence
  PM / Domains: Rename is_async to one_dev_on for genpd_power_off()
  PM / Domains: Move genpd_power_off() above genpd_power_on()

* pm-opp:
  PM / OPP: Documentation: Fix opp-microvolt in examples
  PM / OPP: fix off-by-one bug in dev_pm_opp_get_max_volt_latency loop

Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs

Pull vfs pile two from Al Viro:

 - orangefs fix

 - series of fs/namei.c cleanups from me

 - VFS stuff coming from overlayfs tree

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
  orangefs: Use RCU for destroy_inode
  vfs: use helper for calling f_op->fsync()
  mm: use helper for calling f_op->mmap()
  vfs: use helpers for calling f_op->{read,write}_iter()
  vfs: pass type instead of fn to do_{loop,iter}_readv_writev()
  vfs: extract common parts of {compat_,}do_readv_writev()
  vfs: wrap write f_ops with file_{start,end}_write()
  vfs: deny copy_file_range() for non regular files
  vfs: deny fallocate() on directory
  vfs: create vfs helper vfs_tmpfile()
  namei.c: split unlazy_walk()
  namei.c: fold the check for DCACHE_OP_REVALIDATE into d_revalidate()
  lookup_fast(): clean up the logics around the fallback to non-rcu mode
  namei: fold unlazy_link() into its sole caller

Merge branch 'work.sendmsg' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs

Pull vfs sendmsg updates from Al Viro:
 "More sendmsg work.

  This is a fairly separate isolated stuff (there's a continuation
  around lustre, but that one was too late to soak in -next), thus the
  separate pull request"

* 'work.sendmsg' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
  ncpfs: switch to sock_sendmsg()
  ncpfs: don't mess with manually advancing iovec on send
  ncpfs: sendmsg does *not* bugger iovec these days
  ceph_tcp_sendpage(): use ITER_BVEC sendmsg
  afs_send_pages(): use ITER_BVEC
  rds: remove dead code
  ceph: switch to sock_recvmsg()
  usbip_recv(): switch to sock_recvmsg()
  iscsi_target: deal with short writes on the tx side
  [nbd] pass iov_iter to nbd_xmit()
  [nbd] switch sock_xmit() to sock_{send,recv}msg()
  [drbd] use sock_sendmsg()

block: Move bdi_unregister() to del_gendisk()

Commit 6cd18e711dd8 "block: destroy bdi before blockdev is
unregistered." moved bdi unregistration (at that time through
bdi_destroy()) from blk_release_queue() to blk_cleanup_queue() because
it needs to happen before blk_unregister_region() call in del_gendisk()
for MD. SCSI though will free up the device number from sd_remove()
called through a maze of callbacks from device_del() in
__scsi_remove_device() before blk_cleanup_queue() and thus similar races
as described in 6cd18e711dd8 can happen for SCSI as well as reported by
Omar [1].

Moving bdi_unregister() to del_gendisk() works for MD and fixes the
problem for SCSI since del_gendisk() gets called from sd_remove() before
freeing the device number.

This also makes device_add_disk() (calling bdi_register_owner()) more
symmetric with del_gendisk().

[1] http://marc.info/?l=linux-block&m=148554717109098&w=2

Tested-by: Lekshmi Pillai <[email protected]>
Acked-by: Tejun Heo <[email protected]>
Signed-off-by: Jan Kara <[email protected]>
Tested-by: Omar Sandoval <[email protected]>
Signed-off-by: Jens Axboe <[email protected]>

CIFS: implement get_dfs_refer for SMB2+

in SMB2+ the get_dfs_refer operation uses a FSCTL. The request can be
made on any Tree Connection according to the specs. Since Samba only
accepted it on an IPC connection until recently, try that first.

https://lists.samba.org/archive/samba-technical/2017-February/118859.html

3.2.4.20.3 Application Requests DFS Referral Information:
> The client MUST search for an existing Session and TreeConnect to any
> share on the server identified by ServerName for the user identified by
> UserCredentials. If no Session and TreeConnect are found, the client
> MUST establish a new Session and TreeConnect to IPC$ on the target
> server as described in section 3.2.4.2 using the supplied ServerName and
> UserCredentials.

Signed-off-by: Aurelien Aptel <[email protected]>
Reviewed-by: Pavel Shilovsky <[email protected]>
Signed-off-by: Steve French <[email protected]>

CIFS: use DFS pathnames in SMB2+ Create requests

When connected to a DFS capable share, the client must set the
SMB2_FLAGS_DFS_OPERATIONS flag in the SMB2 header and use
DFS path names: "<server>\<share>\<path>" *without* leading \\.

Sources:

[MS-SMB2] 3.2.5.5 Receiving an SMB2 TREE_CONNECT Response
> TreeConnect.IsDfsShare MUST be set to TRUE, if the SMB2_SHARE_CAP_DFS
> bit is set in the Capabilities field of the response.

[MS-SMB2] 3.2.4.3 Application Requests Opening a File
> If TreeConnect.IsDfsShare is TRUE, the SMB2_FLAGS_DFS_OPERATIONS flag
> is set in the Flags field.

[MS-SMB2] 2.2.13 SMB2 CREATE Request, NameOffset:
> If SMB2_FLAGS_DFS_OPERATIONS is set in the Flags field of the SMB2
> header, the file name includes a prefix that will be processed during
> DFS name normalization as specified in section 3.3.5.9. Otherwise, the
> file name is relative to the share that is identified by the TreeId in
> the SMB2 header.

Signed-off-by: Aurelien Aptel <[email protected]>
Acked-by: Pavel Shilovsky <[email protected]>
Signed-off-by: Steve French <[email protected]>

Merge branch 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/nab/target-pending

Pull SCSI target updates from Nicholas Bellinger:
 "The highlights this round include:

   - enable dual mode (initiator + target) qla2xxx operation. (Quinn +
     Himanshu)

   - add a framework for qla2xxx async fabric discovery. (Quinn +
     Himanshu)

   - enable iscsi PDU DDP completion offload in cxgbit/T6 NICs. (Varun)

   - fix target-core handling of aborted failed commands. (Bart)

   - fix a long standing target-core issue NULL pointer dereference with
     active I/O LUN shutdown. (Rob Millner + Bryant + nab)"

* 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/nab/target-pending: (44 commits)
  target: Add counters for ABORT_TASK success + failure
  iscsi-target: Fix early login failure statistics misses
  target: Fix NULL dereference during LUN lookup + active I/O shutdown
  target: Delete tmr from list before processing
  target: Fix handling of aborted failed commands
  uapi: fix linux/target_core_user.h userspace compilation errors
  target: export protocol identifier
  qla2xxx: Fix a warning reported by the "smatch" static checker
  target/iscsi: Fix unsolicited data seq_end_offset calculation
  target/cxgbit: add T6 iSCSI DDP completion feature
  target/cxgbit: Enable DDP for T6 only if data sequence and pdu are in order
  target/cxgbit: Use T6 specific macros to get ETH/IP hdr len
  target/cxgbit: use cxgb4_tp_smt_idx() to get smt idx
  target/iscsi: split iscsit_check_dataout_hdr()
  target: Remove command flag CMD_T_DEV_ACTIVE
  target: Remove command flag CMD_T_BUSY
  target: Move session check from target_put_sess_cmd() into target_release_cmd_kref()
  target: Inline transport_cmd_check_stop()
  target: Remove an overly chatty debug message
  target: Stop execution if CMD_T_STOP has been set
  ...

xen-netback: Use GFP_ATOMIC to allocate hash

Allocation of new_hash, inside xenvif_new_hash(), always happen
in softirq context, so use GFP_ATOMIC instead of GFP_KERNEL for new
hash allocation.

Signed-off-by: Anoob Soman <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

bonding: use ETH_MAX_MTU as max mtu

This restores the ability of setting bond device's mtu to 9000.

Fixes: 91572088e3fd ("net: use core MTU range checking in core net infra")
Reported-by: [email protected]
Reported-by: Brad Campbell <[email protected]>
Cc: Jarod Wilson <[email protected]>
Signed-off-by: Cong Wang <[email protected]>
Signed-off-by: Jay Vosburgh <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

netvsc: fix use-after-free in netvsc_change_mtu()

'nvdev' is freed in rndis_filter_device_remove -> netvsc_device_remove ->
free_netvsc_device, so we mustn't access it, before it's re-created in
rndis_filter_device_add -> netvsc_device_add.

Signed-off-by: Dexuan Cui <[email protected]>
Cc: "K. Y. Srinivasan" <[email protected]>
Cc: Haiyang Zhang <[email protected]>
Cc: Stephen Hemminger <[email protected]>
Reviewed-by: Stephen Hemminger <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

Merge tag 'mac80211-for-davem-2017-03-02' of git://git.kernel.org/pub/scm/linux/kernel/git/jberg/mac80211

Johannes Berg says:

====================
This contains just the average.h change in order to get it
into the tree before adding new users through -next trees.
====================

Signed-off-by: David S. Miller <[email protected]>

Merge tag 'wireless-drivers-for-davem-2017-03-02' of git://git.kernel.org/pub/scm/linux/kernel/git/kvalo/wireless-drivers

Kalle Valo says:

====================
wireless-drivers fixes for 4.11

Add back a fixed version of an ath10k patch which was reverted during
the merge window.
====================

Signed-off-by: David S. Miller <[email protected]>

Merge tag 'dm-4.11-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm

Pull device mapper fixes from Mike Snitzer:

 - a dm-raid stable@ fix for possible corruption when triggering a raid
   reshape via lvm2; and an additional small patch ontop to bump version
   of the dm-raid target outside of the stable@ fix

 - a dm-raid fix for a 'dm-4.11-changes' regression introduced by a
   commit that was meant to only cleanup confusing branching.

* tag 'dm-4.11-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm:
  dm raid: bump the target version
  dm raid: fix data corruption on reshape request
  dm raid: fix raid "check" regression due to improper cleanup in raid_message()

ipv6: ignore null_entry in inet6_rtm_getroute() too

Like commit 1f17e2f2c8a8 ("net: ipv6: ignore null_entry on route dumps"),
we need to ignore null entry in inet6_rtm_getroute() too.

Return -ENETUNREACH here to sync with IPv4 behavior, as suggested by David.

Fixes: a1a22c1206 ("net: ipv6: Keep nexthop of multipath route on admin down")
Reported-by: Dmitry Vyukov <[email protected]>
Cc: David Ahern <[email protected]>
Signed-off-by: Cong Wang <[email protected]>
Acked-by: David Ahern <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

drivers: net: ethernet: remove incorrect __exit markups

Even if bus is not hot-pluggable, devices can be unbound from the
driver via sysfs, so we should not be using __exit annotations on
remove() methods. The only exception is drivers registered with
platform_driver_probe() which specifically disables sysfs bind/unbind
attributes.

Signed-off-by: Dmitry Torokhov <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

Merge tag 'juno-fixes-4.11' of git://git.kernel.org/pub/scm/linux/kernel/git/sudeep.holla/linux into next/late

Merge "ARMv8 Juno DT fix for v4.11" from Sudeep Holla:

Just single patch to fix replicator in order to prevent overflows at
the source and reduce the back pressure by splitting the trace output
to TPIU and ETR.

* tag 'juno-fixes-4.11' of git://git.kernel.org/pub/scm/linux/kernel/git/sudeep.holla/linux:
  arm64: dts: juno: update definition for programmable replicator

ARM: deconfig: fix the moxart defconfig

The moxart defconfig wasn't even building a kernel for Moxart,
it was building a kernel for V4T on the nothing platform. Switch
to MULTI_V4 and keep the right drivers, update a few selections.
Now it (presumably) builds a minimalist Moxart kernel again.

Cc: Jonas Jensen <[email protected]>
Signed-off-by: Linus Walleij <[email protected]>
Signed-off-by: Arnd Bergmann <[email protected]>

tcp: fix potential double free issue for fastopen_req

tp->fastopen_req could potentially be double freed if a malicious
user does the following:
1. Enable TCP_FASTOPEN_CONNECT sockopt and do a connect() on the socket.
2. Call connect() with AF_UNSPEC to disconnect the socket.
3. Make this socket a listening socket by calling listen().
4. Accept incoming connections and generate child sockets. All child
   sockets will get a copy of the pointer of fastopen_req.
5. Call close() on all sockets. fastopen_req will get freed multiple
   times.

Fixes: 19f6d3f3c842 ("net/tcp-fastopen: Add new API support")
Reported-by: Andrey Konovalov <[email protected]>
Signed-off-by: Wei Wang <[email protected]>
Signed-off-by: Eric Dumazet <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

Merge tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost

Pull vhost updates from Michael Tsirkin:
 "virtio, vhost: optimizations, fixes

  Looks like a quiet cycle for vhost/virtio, just a couple of minor
  tweaks. Most notable is automatic interrupt affinity for blk and scsi.
  Hopefully other devices are not far behind"

* tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost:
  virtio-console: avoid DMA from stack
  vhost: introduce O(1) vq metadata cache
  virtio_scsi: use virtio IRQ affinity
  virtio_blk: use virtio IRQ affinity
  blk-mq: provide a default queue mapping for virtio device
  virtio: provide a method to get the IRQ affinity mask for a virtqueue
  virtio: allow drivers to request IRQ affinity when creating VQs
  virtio_pci: simplify MSI-X setup
  virtio_pci: don't duplicate the msix_enable flag in struct pci_dev
  virtio_pci: use shared interrupts for virtqueues
  virtio_pci: remove struct virtio_pci_vq_info
  vhost: try avoiding avail index access when getting descriptor
  virtio_mmio: expose header to userspace

blk-mq: ensure that bd->last is always set correctly

When drivers are called with a request in blk-mq, blk-mq flags the
state such that the driver knows if this is the last request in
this call chain or not. The driver can then use that information
to defer kicking off IO until bd->last is true. However, with blk-mq
and scheduling, we need to allocate a driver tag for a request before
it can be issued. If we fail to allocate such a tag, we could end up
in the situation where the last request issued did not have
bd->last == true set. This can then cause a driver hang.

This fixes a hang with virtio-blk, which uses bd->last as a hint
on whether to kick the queue or not.

Reported-by: Chris Mason <[email protected]>
Tested-by: Chris Mason <[email protected]>
Reviewed-by: Omar Sandoval <[email protected]>
Signed-off-by: Jens Axboe <[email protected]>

Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jmorris/linux-security

Pull security subsystem fixes from James Morris:
 "Two fixes for the security subsystem:

   - keys: split both rcu_dereference_key() and user_key_payload() into
     versions which can be called with or without holding the key
     semaphore.

   - SELinux: fix Android init(8) breakage due to new cgroup security
     labeling support when using older policy"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jmorris/linux-security:
  selinux: wrap cgroup seclabel support with its own policy capability
  KEYS: Differentiate uses of rcu_dereference_key() and user_key_payload()

net: Introduce sk_clone_lock() error path routine

When handling problems in cloning a socket with the sk_clone_locked()
function we need to perform several steps that were open coded in it and
its callers, so introduce a routine to avoid this duplication:
sk_free_unlock_clone().

Cc: Cong Wang <[email protected]>
Cc: Dmitry Vyukov <[email protected]>
Cc: Eric Dumazet <[email protected]>
Cc: Gerrit Renker <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Link: http://lkml.kernel.org/n/[email protected]
Signed-off-by: Arnaldo Carvalho de Melo <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

dccp: Unlock sock before calling sk_free()

The code where sk_clone() came from created a new socket and locked it,
but then, on the error path didn't unlock it.

This problem stayed there for a long while, till b0691c8ee7c2 ("net:
Unlock sock before calling sk_free()") fixed it, but unfortunately the
callers of sk_clone() (now sk_clone_locked()) were not audited and the
one in dccp_create_openreq_child() remained.

Now in the age of the syskaller fuzzer, this was finally uncovered, as
reported by Dmitry:

 ---- 8< ----

I've got the following report while running syzkaller fuzzer on
86292b33d4b7 ("Merge branch 'akpm' (patches from Andrew)")

  [ BUG: held lock freed! ]
  4.10.0+ #234 Not tainted
  -------------------------
  syz-executor6/6898 is freeing memory
  ffff88006286cac0-ffff88006286d3b7, with a lock still held there!
   (slock-AF_INET6){+.-...}, at: [<ffffffff8362c2c9>] spin_lock
  include/linux/spinlock.h:299 [inline]
   (slock-AF_INET6){+.-...}, at: [<ffffffff8362c2c9>]
  sk_clone_lock+0x3d9/0x12c0 net/core/sock.c:1504
  5 locks held by syz-executor6/6898:
   #0:  (sk_lock-AF_INET6){+.+.+.}, at: [<ffffffff839a34b4>] lock_sock
  include/net/sock.h:1460 [inline]
   #0:  (sk_lock-AF_INET6){+.+.+.}, at: [<ffffffff839a34b4>]
  inet_stream_connect+0x44/0xa0 net/ipv4/af_inet.c:681
   #1:  (rcu_read_lock){......}, at: [<ffffffff83bc1c2a>]
  inet6_csk_xmit+0x12a/0x5d0 net/ipv6/inet6_connection_sock.c:126
   #2:  (rcu_read_lock){......}, at: [<ffffffff8369b424>] __skb_unlink
  include/linux/skbuff.h:1767 [inline]
   #2:  (rcu_read_lock){......}, at: [<ffffffff8369b424>] __skb_dequeue
  include/linux/skbuff.h:1783 [inline]
   #2:  (rcu_read_lock){......}, at: [<ffffffff8369b424>]
  process_backlog+0x264/0x730 net/core/dev.c:4835
   #3:  (rcu_read_lock){......}, at: [<ffffffff83aeb5c0>]
  ip6_input_finish+0x0/0x1700 net/ipv6/ip6_input.c:59
   #4:  (slock-AF_INET6){+.-...}, at: [<ffffffff8362c2c9>] spin_lock
  include/linux/spinlock.h:299 [inline]
   #4:  (slock-AF_INET6){+.-...}, at: [<ffffffff8362c2c9>]
  sk_clone_lock+0x3d9/0x12c0 net/core/sock.c:1504

Fix it just like was done by b0691c8ee7c2 ("net: Unlock sock before calling
sk_free()").

Reported-by: Dmitry Vyukov <[email protected]>
Cc: Cong Wang <[email protected]>
Cc: Eric Dumazet <[email protected]>
Cc: Gerrit Renker <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Arnaldo Carvalho de Melo <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

Merge tag 'batadv-net-for-davem-20170301' of git://git.open-mesh.org/linux-merge

Simon Wunderlich says:

====================
Here are two batman-adv bugfixes:

 - fix a potential double free when fragment merges fail,
   by Sven Eckelmann

 - fix failing tranmission of the 16th (last) fragment if that exists,
   by Linus Lüssing
====================

Signed-off-by: David S. Miller <[email protected]>

openvswitch: actions: fixed a brace coding style warning

Fixed a brace coding style warning reported by checkpatch.pl

Signed-off-by: Peter Downs <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

cxgb4: update latest firmware version supported

Change t4fw_version.h to update latest firmware version
number to 1.16.33.0.

Signed-off-by: Ganesh Goudar <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

block: don't call ioc_exit_icq() with the queue lock held for blk-mq

For legacy scheduling, we always call ioc_exit_icq() with both the
ioc and queue lock held. This poses a problem for blk-mq with
scheduling, since the queue lock isn't what we use in the scheduler.
And since we don't need the queue lock held for ioc exit there,
don't grab it and leave any extra locking up to the blk-mq scheduler.

Reported-by: Paolo Valente <[email protected]>
Tested-by: Paolo Valente <[email protected]>
Reviewed-by: Omar Sandoval <[email protected]>
Signed-off-by: Jens Axboe <[email protected]>

Merge branch 'amd-xgbe-fixes'

Tom Lendacky says:

====================
amd-xgbe: AMD XGBE driver fixes 2017-02-28

This patch series addresses some issues in the AMD XGBE driver.

The following fixes are included in this driver update series:

- Stop the PHY before disabling and releasing device interrupts so that
  MDIO requests issued by the device can be properly handled
- Set the MDIO communication mode on device startup, not just device
  probe
- Do not overwrite SFP settings when mod_absent is detected

This patch series is based on net.
====================

Signed-off-by: David S. Miller <[email protected]>

amd-xgbe: Don't overwrite SFP PHY mod_absent settings

If an SFP module is not present, xgbe_phy_sfp_phy_settings() should
return after applying the default settings. Currently there is no return
statement and the default settings are overwritten.

Signed-off-by: Tom Lendacky <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

amd-xgbe: Be sure to set MDIO modes on device (re)start

The MDIO register mode is set when the device is probed. But when the
device is brought down and then back up, the MDIO register mode has been
reset.  Be sure to reset the mode during device startup and only change
the mode of the address specified.

Signed-off-by: Tom Lendacky <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

amd-xgbe: Stop the PHY before releasing interrupts

Some configurations require the use of the hardware's MDIO support to
communicate with external PHYs. The MDIO commands indicate completion
through the device interrupt. When bringing down the device the interrupts
were released before stopping the external PHY, resulting in MDIO command
timeouts. Move the stopping of the PHY to before the releasing of the
interrupts.

Signed-off-by: Tom Lendacky <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

Merge branch 'kvm-ppc-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/paulus/powerpc into next

Two bug fixes for HV KVM on POWER9 machines.

drivers: net: xgene: Fix crash on DT systems

On DT systems the driver require a clock, but the probe just print a
warning and continue, leading to a crash when resetting the device.
To fix this crash and properly handle probe deferals only ignore the
missing clock if DT isn't used or if the clock doesn't exist.

Signed-off-by: Alban Bedel <[email protected]>
Acked-by: Iyappan Subramanian <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

Merge tag 'watchdog-for-linus-v4.11-2' of git://git.kernel.org/pub/scm/linux/kernel/git/groeck/linux-staging

Pull more watchdog updates from Guenter Roeck:

 - fix fallout from enabling COMPILE_TEST

 - fix gcc-4.3 build of kempld watchdog driver

 - use hrtimer in softdog

* tag 'watchdog-for-linus-v4.11-2' of git://git.kernel.org/pub/scm/linux/kernel/git/groeck/linux-staging:
  watchdog: retu: restore MFD dependency
  watchdog: db8500: add back prmcu dependency
  watchdog: kempld: fix gcc-4.3 build
  watchdog: softdog: fire watchdog even if softirqs do not get to run
  watchdog: kempld: revert to full dependency
  watchdog: bcm2835: add CONFIG_OF dependency
  watchdog: sp805: add back AMBA dependency
  watchdog: menf21bmc: add I2C dependency
  watchdog: geode: restore hard CS5535_MFGPT dependency
  watchdog: wm831x watchdog really needs mfd

ipv6: check for ip6_null_entry in __ip6_del_rt_siblings()

Andrey reported a NULL pointer deref bug in ipv6_route_ioctl()
-> ip6_route_del() -> __ip6_del_rt_siblings() code path. This is
because ip6_null_entry is returned in this path since ip6_null_entry
is kinda default for a ipv6 route table root node. Quote from
David Ahern:

 ip6_null_entry is the root of all ipv6 fib tables making it integrated
 into the table ...

We should ignore any attempt of trying to delete it, like we do in
__ip6_del_rt() path and several others.

Reported-by: Andrey Konovalov <[email protected]>
Fixes: 0ae8133586ad ("net: ipv6: Allow shorthand delete of all nexthops in multipath route")
Cc: David Ahern <[email protected]>
Cc: Eric Dumazet <[email protected]>
Signed-off-by: Cong Wang <[email protected]>
Acked-by: David Ahern <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

give up on gcc ilog2() constant optimizations

gcc-7 has an "optimization" pass that completely screws up, and
generates the code expansion for the (impossible) case of calling
ilog2() with a zero constant, even when the code gcc compiles does not
actually have a zero constant.

And we try to generate a compile-time error for anybody doing ilog2() on
a constant where that doesn't make sense (be it zero or negative).  So
now gcc7 will fail the build due to our sanity checking, because it
created that constant-zero case that didn't actually exist in the source
code.

There's a whole long discussion on the kernel mailing about how to work
around this gcc bug.  The gcc people themselevs have discussed their
"feature" in

   https://gcc.gnu.org/bugzilla/show_bug.cgi?id=72785

but it's all water under the bridge, because while it looked at one
point like it would be solved by the time gcc7 was released, that was
not to be.

So now we have to deal with this compiler braindamage.

And the only simple approach seems to be to just delete the code that
tries to warn about bad uses of ilog2().

So now "ilog2()" will just return 0 not just for the value 1, but for
any non-positive value too.

It's not like I can recall anybody having ever actually tried to use
this function on any invalid value, but maybe the sanity check just
meant that such code never made it out in public.

Reported-by: Laura Abbott <[email protected]>
Cc: John Stultz <[email protected]>,
Cc: Thomas Gleixner <[email protected]>
Cc: Ard Biesheuvel <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

ARM: ux500: resume the second core properly

The pen hold/release scheme was copied over to Ux500 from the ARM
reference designs like most of these at the time. It is not needed
at all, and was mostly removed in commit c00def71efd9
"ARM: ux500: simplify secondary CPU boot".

However on the suspend/resume path and hot plug/unplug of CPUs,
the .cpu_die() callback was still waiting for the pen to be
released which made it spin forever and the second core never come
back online after suspend/resume.

Fix this by simply replacing the strange custom .cpu_die() with
a oneline wfi() just like e.g. the qcom platform does. This fixes
the issue and makes the second core come up properly after
suspend/resume.

As a side effect, this rids us of the completely surplus local
setup.h and hotplug.c files, and we just compile this into platsmp.c
with everything else SMP.

Cc: [email protected]
Fixes: c00def71efd9 ("ARM: ux500: simplify secondary CPU boot")
Cc: Ulf Hansson <[email protected]>
Signed-off-by: Linus Walleij <[email protected]>
Signed-off-by: Arnd Bergmann <[email protected]>

Merge branch 'next/late' with mainline

* next/late: (25 commits)
  arm64: dts: exynos: Add regulators for Vbus and Vbus-Boost
  arm64: dts: exynos: Add USB 3.0 controller node for Exynos7
  arm64: dts: exynos: Use macros for pinctrl configuration on Exynos7
  pinctrl: dt-bindings: samsung: Add Exynos7 specific pinctrl macro definitions
  arm64: dts: exynos: Add initial configuration for DISP clocks for TM2/TM2e
  ARM64: dts: meson-gxbb-p200: add ADC laddered keys
  ARM64: dts: meson: meson-gx: add the SAR ADC
  ARM64: dts: meson-gxl: add the pwm_ao_b pin
  ARM64: dts: meson-gx: add the missing pwm_AO_ab node
  clk: gxbb: fix CLKID_ETH defined twice
  clk: samsung: exynos5433: Add data for 250MHz and 278MHz PLL rates
  clk: samsung: exynos5433: Add IDs for PHYCLK_MIPIDPHY0_* clocks
  ARM64: dts: meson-gxl: rename Nexbox A95x for consistency
  clk: gxbb: add the SAR ADC clocks and expose them
  dt-bindings: amlogic: Add WeTek boards
  ARM64: dts: meson-gxbb: Add support for WeTek Hub and Play
  dt-bindings: vendor-prefix: Add wetek vendor prefix
  ARM64: dts: meson-gxm: Rename q200 and q201 DT files for consistency
  ARM64: dts: meson-gx: Add HDMI HPD/DDC pinctrl nodes
  ARM64: dts: meson-gxbb-vega-s95: Add LED
  ...

Signed-off-by: Arnd Bergmann <[email protected]>

block: Initialize bd_bdi on inode initialization

So far we initialized bd_bdi only in bdget(). That is fine for normal
bdev inodes however for the special case of the root inode of
blockdev_superblock that function is never called and thus bd_bdi is
left uninitialized. As a result bdev_evict_inode() may oops doing
bdi_put(root->bd_bdi) on that inode as can be seen when doing:

mount -t bdev none /mnt

Fix the problem by initializing bd_bdi when first allocating the inode
and then reinitializing bd_bdi in bdev_evict_inode().

Thanks to syzkaller team for finding the problem.

Reported-by: Dmitry Vyukov <[email protected]>
Fixes: b1d2dc5659b4 ("block: Make blk_get_backing_dev_info() safe without open bdev")
Signed-off-by: Jan Kara <[email protected]>
Signed-off-by: Jens Axboe <[email protected]>

loop: fix LO_FLAGS_PARTSCAN hang

loop_reread_partitions() needs to do I/O, but we just froze the queue,
so we end up waiting forever. This can easily be reproduced with losetup
-P. Fix it by moving the reread to after we unfreeze the queue.

Fixes: ecdd09597a57 ("block/loop: fix race between I/O and set_status")
Reported-by: Tejun Heo <[email protected]>
Cc: [email protected]
Signed-off-by: Omar Sandoval <[email protected]>
Reviewed-by: Ming Lei <[email protected]>
Signed-off-by: Jens Axboe <[email protected]>

nvme: Complete all stuck requests

If the nvme driver is shutting down its controller, the drievr will not
start the queues up again, preventing blk-mq's hot CPU notifier from
making forward progress.

To fix that, this patch starts a request_queue freeze when the driver
resets a controller so no new requests may enter. The driver will wait
for frozen after IO queues are restarted to ensure the queue reference
can be reinitialized when nvme requests to unfreeze the queues.

If the driver is doing a safe shutdown, the driver will wait for the
controller to successfully complete all inflight requests so that we
don't unnecessarily fail them. Once the controller has been disabled,
the queues will be restarted to force remaining entered requests to end
in failure so that blk-mq's hot cpu notifier may progress.

Signed-off-by: Keith Busch <[email protected]>
Reviewed-by: Sagi Grimberg <[email protected]>
Signed-off-by: Jens Axboe <[email protected]>

blk-mq: Provide freeze queue timeout

A driver may wish to take corrective action if queued requests do not
complete within a set time.

Signed-off-by: Keith Busch <[email protected]>
Signed-off-by: Jens Axboe <[email protected]>

blk-mq: Export blk_mq_freeze_queue_wait

Drivers can start a freeze, so this provides a way to wait for frozen.

Signed-off-by: Keith Busch <[email protected]>
Reviewed-by: Christoph Hellwig <[email protected]>
Signed-off-by: Sagi Grimberg <[email protected]>
Signed-off-by: Jens Axboe <[email protected]>

nbd: stop leaking sockets

This was introduced in the multi-connection patch, we've been leaking
socket's ever since.

Fixes: 9561a7a ("nbd: add multi-connection support")
cc: [email protected]
Signed-off-by: Josef Bacik <[email protected]>
Signed-off-by: Jens Axboe <[email protected]>

blk-mq: move update of tags->rqs to __blk_mq_alloc_request()

No functional difference, it just makes a little more sense to update
the tag map where we actually allocate the tag.

Signed-off-by: Omar Sandoval <[email protected]>
Signed-off-by: Jens Axboe <[email protected]>
Tested-by: Sagi Grimberg <[email protected]>

blk-mq: kill blk_mq_set_alloc_data()

Nothing is using it anymore.

Signed-off-by: Omar Sandoval <[email protected]>
Signed-off-by: Jens Axboe <[email protected]>
Tested-by: Sagi Grimberg <[email protected]>

blk-mq: make blk_mq_alloc_request_hctx() allocate a scheduler request

blk_mq_alloc_request_hctx() allocates a driver request directly, unlike
its blk_mq_alloc_request() counterpart. It also crashes because it
doesn't update the tags->rqs map.

Fix it by making it allocate a scheduler request.

Reported-by: Sagi Grimberg <[email protected]>
Signed-off-by: Omar Sandoval <[email protected]>
Signed-off-by: Jens Axboe <[email protected]>
Tested-by: Sagi Grimberg <[email protected]>