OSDN Git Service

uclinux-h8/linux.git
4 years agoio_uring: clean file_data access in files_register
Pavel Begunkov [Sat, 10 Oct 2020 17:34:14 +0000 (18:34 +0100)]
io_uring: clean file_data access in files_register

Keep file_data in a local var and replace with it complex references
such as ctx->file_data.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: don't delay io_init_req() error check
Pavel Begunkov [Sat, 10 Oct 2020 17:34:13 +0000 (18:34 +0100)]
io_uring: don't delay io_init_req() error check

Don't postpone io_init_req() error checks and do that right after
calling it. There is no control-flow statements or dependencies with
sqe/submitted accounting, so do those earlier, that makes the code flow
a bit more natural.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: clean leftovers after splitting issue
Pavel Begunkov [Sat, 10 Oct 2020 17:34:12 +0000 (18:34 +0100)]
io_uring: clean leftovers after splitting issue

Kill extra if in io_issue_sqe() and place send/recv[msg] calls
appropriately under switch's cases.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: remove timeout.list after hrtimer cancel
Pavel Begunkov [Sat, 10 Oct 2020 17:34:11 +0000 (18:34 +0100)]
io_uring: remove timeout.list after hrtimer cancel

Remove timeouts from ctx->timeout_list after hrtimer_try_to_cancel()
successfully cancels it. With this we don't need to care whether there
was a race and it was removed in io_timeout_fn(), and that will be handy
for following patches.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: use a separate struct for timeout_remove
Pavel Begunkov [Sat, 10 Oct 2020 17:34:10 +0000 (18:34 +0100)]
io_uring: use a separate struct for timeout_remove

Don't use struct io_timeout for both IORING_OP_TIMEOUT and
IORING_OP_TIMEOUT_REMOVE, they're quite different. Split them in two,
that allows to remove an unused field in struct io_timeout, and btw kill
->flags not used by either. This also easier to follow, especially for
timeout remove.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: improve submit_state.ios_left accounting
Pavel Begunkov [Sat, 10 Oct 2020 17:34:09 +0000 (18:34 +0100)]
io_uring: improve submit_state.ios_left accounting

state->ios_left isn't decremented for requests that don't need a file,
so it might be larger than number of SQEs left. That in some
circumstances makes us to grab more files that is needed so imposing
extra put.
Deaccount one ios_left for each request.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: simplify io_file_get()
Pavel Begunkov [Sat, 10 Oct 2020 17:34:08 +0000 (18:34 +0100)]
io_uring: simplify io_file_get()

Keep ->needs_file_no_error check out of io_file_get(), and let callers
handle it. It makes it more straightforward. Also, as the only error it
can hand back -EBADF, make it return a file or NULL.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: kill extra check in fixed io_file_get()
Pavel Begunkov [Sat, 10 Oct 2020 17:34:07 +0000 (18:34 +0100)]
io_uring: kill extra check in fixed io_file_get()

ctx->nr_user_files == 0 IFF ctx->file_data == NULL and there fixed files
are not used. Hence, verifying fds only against ctx->nr_user_files is
enough. Remove the other check from hot path.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: clean up ->files grabbing
Pavel Begunkov [Sat, 10 Oct 2020 17:34:06 +0000 (18:34 +0100)]
io_uring: clean up ->files grabbing

Move work.files grabbing into io_prep_async_work() to all other work
resources initialisation. We don't need to keep it separately now, as
->ring_fd/file are gone. It also allows to not grab it when a request
is not going to io-wq.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: don't io_prep_async_work() linked reqs
Pavel Begunkov [Sat, 10 Oct 2020 17:34:05 +0000 (18:34 +0100)]
io_uring: don't io_prep_async_work() linked reqs

There is no real reason left for preparing io-wq work context for linked
requests in advance, remove it as this might become a bottleneck in some
cases.

Reported-by: Roman Gershman <romger@amazon.com>
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: Convert advanced XArray uses to the normal API
Matthew Wilcox (Oracle) [Fri, 9 Oct 2020 12:49:53 +0000 (13:49 +0100)]
io_uring: Convert advanced XArray uses to the normal API

There are no bugs here that I've spotted, it's just easier to use the
normal API and there are no performance advantages to using the more
verbose advanced API.

Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: Fix XArray usage in io_uring_add_task_file
Matthew Wilcox (Oracle) [Fri, 9 Oct 2020 12:49:52 +0000 (13:49 +0100)]
io_uring: Fix XArray usage in io_uring_add_task_file

The xas_store() wasn't paired with an xas_nomem() loop, so if it couldn't
allocate memory using GFP_NOWAIT, it would leak the reference to the file
descriptor.  Also the node pointed to by the xas could be freed between
the call to xas_load() under the rcu_read_lock() and the acquisition of
the xa_lock.

It's easier to just use the normal xa_load/xa_store interface here.

Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
[axboe: fix missing assign after alloc, cur_uring -> tctx rename]
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: Fix use of XArray in __io_uring_files_cancel
Matthew Wilcox (Oracle) [Fri, 9 Oct 2020 12:49:51 +0000 (13:49 +0100)]
io_uring: Fix use of XArray in __io_uring_files_cancel

We have to drop the lock during each iteration, so there's no advantage
to using the advanced API.  Convert this to a standard xa_for_each() loop.

Reported-by: syzbot+27c12725d8ff0bfe1a13@syzkaller.appspotmail.com
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: fix break condition for __io_uring_register() waiting
Jens Axboe [Fri, 9 Oct 2020 01:09:46 +0000 (19:09 -0600)]
io_uring: fix break condition for __io_uring_register() waiting

Colin reports that there's unreachable code, since we only ever break
if ret == 0. This is correct, and is due to a reversed logic condition
in when to break or not.

Break out of the loop if we don't process any task work, in that case
we do want to return -EINTR.

Fixes: af9c1a44f8de ("io_uring: process task work in io_uring_register()")
Reported-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: no need to call xa_destroy() on empty xarray
Jens Axboe [Thu, 8 Oct 2020 13:46:52 +0000 (07:46 -0600)]
io_uring: no need to call xa_destroy() on empty xarray

The kernel test robot reports this lockdep issue:

[child1:659] mbind (274) returned ENOSYS, marking as inactive.
[child1:659] mq_timedsend (279) returned ENOSYS, marking as inactive.
[main] 10175 iterations. [F:7781 S:2344 HI:2397]
[   24.610601]
[   24.610743] ================================
[   24.611083] WARNING: inconsistent lock state
[   24.611437] 5.9.0-rc7-00017-g0f2122045b9462 #5 Not tainted
[   24.611861] --------------------------------
[   24.612193] inconsistent {SOFTIRQ-ON-W} -> {IN-SOFTIRQ-W} usage.
[   24.612660] ksoftirqd/0/7 [HC0[0]:SC1[3]:HE0:SE0] takes:
[   24.613086] f00ed998 (&xa->xa_lock#4){+.?.}-{2:2}, at: xa_destroy+0x43/0xc1
[   24.613642] {SOFTIRQ-ON-W} state was registered at:
[   24.614024]   lock_acquire+0x20c/0x29b
[   24.614341]   _raw_spin_lock+0x21/0x30
[   24.614636]   io_uring_add_task_file+0xe8/0x13a
[   24.614987]   io_uring_create+0x535/0x6bd
[   24.615297]   io_uring_setup+0x11d/0x136
[   24.615606]   __ia32_sys_io_uring_setup+0xd/0xf
[   24.615977]   do_int80_syscall_32+0x53/0x6c
[   24.616306]   restore_all_switch_stack+0x0/0xb1
[   24.616677] irq event stamp: 939881
[   24.616968] hardirqs last  enabled at (939880): [<8105592d>] __local_bh_enable_ip+0x13c/0x145
[   24.617642] hardirqs last disabled at (939881): [<81b6ace3>] _raw_spin_lock_irqsave+0x1b/0x4e
[   24.618321] softirqs last  enabled at (939738): [<81b6c7c8>] __do_softirq+0x3f0/0x45a
[   24.618924] softirqs last disabled at (939743): [<81055741>] run_ksoftirqd+0x35/0x61
[   24.619521]
[   24.619521] other info that might help us debug this:
[   24.620028]  Possible unsafe locking scenario:
[   24.620028]
[   24.620492]        CPU0
[   24.620685]        ----
[   24.620894]   lock(&xa->xa_lock#4);
[   24.621168]   <Interrupt>
[   24.621381]     lock(&xa->xa_lock#4);
[   24.621695]
[   24.621695]  *** DEADLOCK ***
[   24.621695]
[   24.622154] 1 lock held by ksoftirqd/0/7:
[   24.622468]  #0: 823bfb94 (rcu_callback){....}-{0:0}, at: rcu_process_callbacks+0xc0/0x155
[   24.623106]
[   24.623106] stack backtrace:
[   24.623454] CPU: 0 PID: 7 Comm: ksoftirqd/0 Not tainted 5.9.0-rc7-00017-g0f2122045b9462 #5
[   24.624090] Call Trace:
[   24.624284]  ? show_stack+0x40/0x46
[   24.624551]  dump_stack+0x1b/0x1d
[   24.624809]  print_usage_bug+0x17a/0x185
[   24.625142]  mark_lock+0x11d/0x1db
[   24.625474]  ? print_shortest_lock_dependencies+0x121/0x121
[   24.625905]  __lock_acquire+0x41e/0x7bf
[   24.626206]  lock_acquire+0x20c/0x29b
[   24.626517]  ? xa_destroy+0x43/0xc1
[   24.626810]  ? lock_acquire+0x20c/0x29b
[   24.627110]  _raw_spin_lock_irqsave+0x3e/0x4e
[   24.627450]  ? xa_destroy+0x43/0xc1
[   24.627725]  xa_destroy+0x43/0xc1
[   24.627989]  __io_uring_free+0x57/0x71
[   24.628286]  ? get_pid+0x22/0x22
[   24.628544]  __put_task_struct+0xf2/0x163
[   24.628865]  put_task_struct+0x1f/0x2a
[   24.629161]  delayed_put_task_struct+0xe2/0xe9
[   24.629509]  rcu_process_callbacks+0x128/0x155
[   24.629860]  __do_softirq+0x1a3/0x45a
[   24.630151]  run_ksoftirqd+0x35/0x61
[   24.630443]  smpboot_thread_fn+0x304/0x31a
[   24.630763]  kthread+0x124/0x139
[   24.631016]  ? sort_range+0x18/0x18
[   24.631290]  ? kthread_create_worker_on_cpu+0x17/0x17
[   24.631682]  ret_from_fork+0x1c/0x28

which is complaining about xa_destroy() grabbing the xa lock in an
IRQ disabling fashion, whereas the io_uring uses cases aren't interrupt
safe. This is really an xarray issue, since it should not assume the
lock type. But for our use case, since we know the xarray is empty at
this point, there's no need to actually call xa_destroy(). So just get
rid of it.

Fixes: 0f2122045b94 ("io_uring: don't rely on weak ->files references")
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: batch account ->req_issue and task struct references
Jens Axboe [Wed, 7 Oct 2020 18:48:53 +0000 (12:48 -0600)]
io_uring: batch account ->req_issue and task struct references

Identical to how we handle the ctx reference counts, increase by the
batch we're expecting to submit, and handle any slow path residual,
if any. The request alloc-and-issue path is very hot, and this makes
a noticeable difference by avoiding an two atomic incs for each
individual request.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: kill callback_head argument for io_req_task_work_add()
Jens Axboe [Thu, 1 Oct 2020 03:00:14 +0000 (21:00 -0600)]
io_uring: kill callback_head argument for io_req_task_work_add()

We always use &req->task_work anyway, no point in passing it in.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: move req preps out of io_issue_sqe()
Pavel Begunkov [Wed, 30 Sep 2020 19:57:56 +0000 (22:57 +0300)]
io_uring: move req preps out of io_issue_sqe()

All request preparations are done only during submission, reflect it in
the code by moving io_req_prep() much earlier into io_queue_sqe().

That's much cleaner, because it doen't expose bits to async code which
it won't ever use. Also it makes the interface harder to misuse, and
there are potential places for bugs.

For instance, __io_queue() doesn't clear @sqe before proceeding to a
next linked request, that could have been disastrous, but hopefully
there are linked requests IFF sqe==NULL, so not actually a bug.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: decouple issuing and req preparation
Pavel Begunkov [Wed, 30 Sep 2020 19:57:55 +0000 (22:57 +0300)]
io_uring: decouple issuing and req preparation

io_issue_sqe() does two things at once, trying to prepare request and
issuing them. Split it in two and deduplicate with io_defer_prep().

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: remove nonblock arg from io_{rw}_prep()
Pavel Begunkov [Wed, 30 Sep 2020 19:57:54 +0000 (22:57 +0300)]
io_uring: remove nonblock arg from io_{rw}_prep()

All io_*_prep() functions including io_{read,write}_prep() are called
only during submission where @force_nonblock is always true. Don't keep
propagating it and instead remove the @force_nonblock argument
from prep() altogether.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: set/clear IOCB_NOWAIT into io_read/write
Pavel Begunkov [Wed, 30 Sep 2020 19:57:53 +0000 (22:57 +0300)]
io_uring: set/clear IOCB_NOWAIT into io_read/write

Move setting IOCB_NOWAIT from io_prep_rw() into io_read()/io_write(), so
it's set/cleared in a single place. Also remove @force_nonblock
parameter from io_prep_rw().

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: remove F_NEED_CLEANUP check in *prep()
Pavel Begunkov [Wed, 30 Sep 2020 19:57:35 +0000 (22:57 +0300)]
io_uring: remove F_NEED_CLEANUP check in *prep()

REQ_F_NEED_CLEANUP is set only by io_*_prep() and they're guaranteed to
be called only once, so there is no one who may have set the flag
before. Kill REQ_F_NEED_CLEANUP check in these *prep() handlers.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: io_kiocb_ppos() style change
Pavel Begunkov [Wed, 30 Sep 2020 19:57:15 +0000 (22:57 +0300)]
io_uring: io_kiocb_ppos() style change

Put brackets around bitwise ops in a complex expression

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: simplify io_alloc_req()
Pavel Begunkov [Wed, 30 Sep 2020 19:57:01 +0000 (22:57 +0300)]
io_uring: simplify io_alloc_req()

Extract common code from if/else branches. That is cleaner and optimised
even better.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio-wq: kill unused IO_WORKER_F_EXITING
Jens Axboe [Sat, 26 Sep 2020 18:37:46 +0000 (12:37 -0600)]
io-wq: kill unused IO_WORKER_F_EXITING

This flag is no longer used, remove it.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio-wq: fix use-after-free in io_wq_worker_running
Hillf Danton [Sat, 26 Sep 2020 13:26:55 +0000 (21:26 +0800)]
io-wq: fix use-after-free in io_wq_worker_running

The smart syzbot has found a reproducer for the following issue:

 ==================================================================
 BUG: KASAN: use-after-free in instrument_atomic_write include/linux/instrumented.h:71 [inline]
 BUG: KASAN: use-after-free in atomic_inc include/asm-generic/atomic-instrumented.h:240 [inline]
 BUG: KASAN: use-after-free in io_wqe_inc_running fs/io-wq.c:301 [inline]
 BUG: KASAN: use-after-free in io_wq_worker_running+0xde/0x110 fs/io-wq.c:613
 Write of size 4 at addr ffff8882183db08c by task io_wqe_worker-0/7771

 CPU: 0 PID: 7771 Comm: io_wqe_worker-0 Not tainted 5.9.0-rc4-syzkaller #0
 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
 Call Trace:
  __dump_stack lib/dump_stack.c:77 [inline]
  dump_stack+0x198/0x1fd lib/dump_stack.c:118
  print_address_description.constprop.0.cold+0xae/0x497 mm/kasan/report.c:383
  __kasan_report mm/kasan/report.c:513 [inline]
  kasan_report.cold+0x1f/0x37 mm/kasan/report.c:530
  check_memory_region_inline mm/kasan/generic.c:186 [inline]
  check_memory_region+0x13d/0x180 mm/kasan/generic.c:192
  instrument_atomic_write include/linux/instrumented.h:71 [inline]
  atomic_inc include/asm-generic/atomic-instrumented.h:240 [inline]
  io_wqe_inc_running fs/io-wq.c:301 [inline]
  io_wq_worker_running+0xde/0x110 fs/io-wq.c:613
  schedule_timeout+0x148/0x250 kernel/time/timer.c:1879
  io_wqe_worker+0x517/0x10e0 fs/io-wq.c:580
  kthread+0x3b5/0x4a0 kernel/kthread.c:292
  ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:294

 Allocated by task 7768:
  kasan_save_stack+0x1b/0x40 mm/kasan/common.c:48
  kasan_set_track mm/kasan/common.c:56 [inline]
  __kasan_kmalloc.constprop.0+0xbf/0xd0 mm/kasan/common.c:461
  kmem_cache_alloc_node_trace+0x17b/0x3f0 mm/slab.c:3594
  kmalloc_node include/linux/slab.h:572 [inline]
  kzalloc_node include/linux/slab.h:677 [inline]
  io_wq_create+0x57b/0xa10 fs/io-wq.c:1064
  io_init_wq_offload fs/io_uring.c:7432 [inline]
  io_sq_offload_start fs/io_uring.c:7504 [inline]
  io_uring_create fs/io_uring.c:8625 [inline]
  io_uring_setup+0x1836/0x28e0 fs/io_uring.c:8694
  do_syscall_64+0x2d/0x70 arch/x86/entry/common.c:46
  entry_SYSCALL_64_after_hwframe+0x44/0xa9

 Freed by task 21:
  kasan_save_stack+0x1b/0x40 mm/kasan/common.c:48
  kasan_set_track+0x1c/0x30 mm/kasan/common.c:56
  kasan_set_free_info+0x1b/0x30 mm/kasan/generic.c:355
  __kasan_slab_free+0xd8/0x120 mm/kasan/common.c:422
  __cache_free mm/slab.c:3418 [inline]
  kfree+0x10e/0x2b0 mm/slab.c:3756
  __io_wq_destroy fs/io-wq.c:1138 [inline]
  io_wq_destroy+0x2af/0x460 fs/io-wq.c:1146
  io_finish_async fs/io_uring.c:6836 [inline]
  io_ring_ctx_free fs/io_uring.c:7870 [inline]
  io_ring_exit_work+0x1e4/0x6d0 fs/io_uring.c:7954
  process_one_work+0x94c/0x1670 kernel/workqueue.c:2269
  worker_thread+0x64c/0x1120 kernel/workqueue.c:2415
  kthread+0x3b5/0x4a0 kernel/kthread.c:292
  ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:294

 The buggy address belongs to the object at ffff8882183db000
  which belongs to the cache kmalloc-1k of size 1024
 The buggy address is located 140 bytes inside of
  1024-byte region [ffff8882183db000ffff8882183db400)
 The buggy address belongs to the page:
 page:000000009bada22b refcount:1 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x2183db
 flags: 0x57ffe0000000200(slab)
 raw: 057ffe0000000200 ffffea0008604c48 ffffea00086a8648 ffff8880aa040700
 raw: 0000000000000000 ffff8882183db000 0000000100000002 0000000000000000
 page dumped because: kasan: bad access detected

 Memory state around the buggy address:
  ffff8882183daf80: ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff
  ffff8882183db000: fa fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
 >ffff8882183db080: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
                       ^
  ffff8882183db100: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
  ffff8882183db180: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
 ==================================================================

which is down to the comment below,

/* all workers gone, wq exit can proceed */
if (!nr_workers && refcount_dec_and_test(&wqe->wq->refs))
complete(&wqe->wq->done);

because there might be multiple cases of wqe in a wq and we would wait
for every worker in every wqe to go home before releasing wq's resources
on destroying.

To that end, rework wq's refcount by making it independent of the tracking
of workers because after all they are two different things, and keeping
it balanced when workers come and go. Note the manager kthread, like
other workers, now holds a grab to wq during its lifetime.

Finally to help destroy wq, check IO_WQ_BIT_EXIT upon creating worker
and do nothing for exiting wq.

Cc: stable@vger.kernel.org # v5.5+
Reported-by: syzbot+45fa0a195b941764e0f0@syzkaller.appspotmail.com
Reported-by: syzbot+9af99580130003da82b1@syzkaller.appspotmail.com
Cc: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Hillf Danton <hdanton@sina.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: show sqthread pid and cpu in fdinfo
Joseph Qi [Tue, 29 Sep 2020 15:01:22 +0000 (09:01 -0600)]
io_uring: show sqthread pid and cpu in fdinfo

In most cases we'll specify IORING_SETUP_SQPOLL and run multiple
io_uring instances in a host. Since all sqthreads are named
"io_uring-sq", it's hard to distinguish the relations between
application process and its io_uring sqthread.
With this patch, application can get its corresponding sqthread pid
and cpu through show_fdinfo.
Steps:
1. Get io_uring fd first.
$ ls -l /proc/<pid>/fd | grep -w io_uring
2. Then get io_uring instance related info, including corresponding
sqthread pid and cpu.
$ cat /proc/<pid>/fdinfo/<io_uring_fd>

pos: 0
flags: 02000002
mnt_id: 13
SqThread: 6929
SqThreadCpu: 2
UserFiles: 1
    0: testfile
UserBufs: 0
PollList:

Signed-off-by: Joseph Qi <joseph.qi@linux.alibaba.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
[axboe: fixed for new shared SQPOLL]
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: process task work in io_uring_register()
Jens Axboe [Thu, 24 Sep 2020 19:32:18 +0000 (13:32 -0600)]
io_uring: process task work in io_uring_register()

We do this for CQ ring wait, in case task_work completions come in. We
should do the same in io_uring_register(), to avoid spurious -EINTR
if the ring quiescing ends up having to process task_work to complete
the operation

Reported-by: Dan Melnic <dmm@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: add blkcg accounting to offloaded operations
Dennis Zhou [Wed, 16 Sep 2020 20:41:05 +0000 (13:41 -0700)]
io_uring: add blkcg accounting to offloaded operations

There are a few operations that are offloaded to the worker threads. In
this case, we lose process context and end up in kthread context. This
results in ios to be not accounted to the issuing cgroup and
consequently end up as issued by root. Just like others, adopt the
personality of the blkcg too when issuing via the workqueues.

For the SQPOLL thread, it will live and attach in the inited cgroup's
context.

Signed-off-by: Dennis Zhou <dennis@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: improve registered buffer accounting for huge pages
Jens Axboe [Thu, 17 Sep 2020 22:19:16 +0000 (16:19 -0600)]
io_uring: improve registered buffer accounting for huge pages

io_uring does account any registered buffer as pinned/locked memory, and
checks limit and fails if the given user doesn't have a big enough limit
to register the ranges specified. However, if huge pages are used, we
are potentially under-accounting the memory in terms of what gets pinned
on the vm side.

This patch rectifies that, by ensuring that we account the full size of
a compound page, regardless of how much of it is being registered. Huge
pages are not accounted mulitple times - if multiple sections of a huge
page is registered, then the page is only accounted once.

Reported-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: remove unneeded semicolon
Zheng Bin [Wed, 9 Sep 2020 12:12:37 +0000 (20:12 +0800)]
io_uring: remove unneeded semicolon

Fixes coccicheck warning:

fs/io_uring.c:4242:13-14: Unneeded semicolon

Signed-off-by: Zheng Bin <zhengbin13@huawei.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: cap SQ submit size for SQPOLL with multiple rings
Jens Axboe [Tue, 8 Sep 2020 15:11:32 +0000 (09:11 -0600)]
io_uring: cap SQ submit size for SQPOLL with multiple rings

In the spirit of fairness, cap the max number of SQ entries we'll submit
for SQPOLL if we have multiple rings. If we don't do that, we could be
submitting tons of entries for one ring, while others are waiting to get
service.

The value of 8 is somewhat arbitrarily chosen as something that allows
a fair bit of batching, without using an excessive time per ring.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: get rid of req->io/io_async_ctx union
Jens Axboe [Sun, 16 Aug 2020 01:44:09 +0000 (18:44 -0700)]
io_uring: get rid of req->io/io_async_ctx union

There's really no point in having this union, it just means that we're
always allocating enough room to cater to any command. But that's
pointless, as the ->io field is request type private anyway.

This gets rid of the io_async_ctx structure, and fills in the required
size in the io_op_defs[] instead.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: kill extra user_bufs check
Pavel Begunkov [Sat, 5 Sep 2020 21:45:48 +0000 (00:45 +0300)]
io_uring: kill extra user_bufs check

Testing ctx->user_bufs for NULL in io_import_fixed() is not neccessary,
because in that case ctx->nr_user_bufs would be zero, and the following
check would fail.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: fix overlapped memcpy in io_req_map_rw()
Pavel Begunkov [Sat, 5 Sep 2020 21:45:47 +0000 (00:45 +0300)]
io_uring: fix overlapped memcpy in io_req_map_rw()

When io_req_map_rw() is called from io_rw_prep_async(), it memcpy()
iorw->iter into itself. Even though it doesn't lead to an error, such a
memcpy()'s aliasing rules violation is considered to be a bad practise.

Inline io_req_map_rw() into io_rw_prep_async(). We don't really need any
remapping there, so it's much simpler than the generic implementation.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: refactor io_req_map_rw()
Pavel Begunkov [Sat, 5 Sep 2020 21:45:46 +0000 (00:45 +0300)]
io_uring: refactor io_req_map_rw()

Set rw->free_iovec to @iovec, that gives an identical result and stresses
that @iovec param rw->free_iovec play the same role.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: simplify io_rw_prep_async()
Pavel Begunkov [Sat, 5 Sep 2020 21:45:45 +0000 (00:45 +0300)]
io_uring: simplify io_rw_prep_async()

Don't touch iter->iov and iov in between __io_import_iovec() and
io_req_map_rw(), the former function aleady sets it correctly, because it
creates one more case with NULL'ed iov to consider in io_req_map_rw().

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: provide IORING_ENTER_SQ_WAIT for SQPOLL SQ ring waits
Jens Axboe [Thu, 3 Sep 2020 18:12:41 +0000 (12:12 -0600)]
io_uring: provide IORING_ENTER_SQ_WAIT for SQPOLL SQ ring waits

When using SQPOLL, applications can run into the issue of running out of
SQ ring entries because the thread hasn't consumed them yet. The only
option for dealing with that is checking later, or busy checking for the
condition.

Provide IORING_ENTER_SQ_WAIT if applications want to wait on this
condition.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: mark io_uring_fops/io_op_defs as __read_mostly
Jens Axboe [Thu, 3 Sep 2020 11:54:56 +0000 (05:54 -0600)]
io_uring: mark io_uring_fops/io_op_defs as __read_mostly

These structures are never written, move them appropriately.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: enable IORING_SETUP_ATTACH_WQ to attach to SQPOLL thread too
Jens Axboe [Wed, 2 Sep 2020 20:50:27 +0000 (14:50 -0600)]
io_uring: enable IORING_SETUP_ATTACH_WQ to attach to SQPOLL thread too

We support using IORING_SETUP_ATTACH_WQ to share async backends between
rings created by the same process, this now also allows the same to
happen with SQPOLL. The setup procedure remains the same, the caller
sets io_uring_params->wq_fd to the 'parent' context, and then the newly
created ring will attach to that async backend.

This means that multiple rings can share the same SQPOLL thread, saving
resources.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: base SQPOLL handling off io_sq_data
Jens Axboe [Mon, 14 Sep 2020 17:16:23 +0000 (11:16 -0600)]
io_uring: base SQPOLL handling off io_sq_data

Remove the SQPOLL thread from the ctx, and use the io_sq_data as the
data structure we pass in. io_sq_data has a list of ctx's that we can
then iterate over and handle.

As of now we're ready to handle multiple ctx's, though we're still just
handling a single one after this patch.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: split SQPOLL data into separate structure
Jens Axboe [Wed, 2 Sep 2020 19:52:19 +0000 (13:52 -0600)]
io_uring: split SQPOLL data into separate structure

Move all the necessary state out of io_ring_ctx, and into a new
structure, io_sq_data. The latter now deals with any state or
variables associated with the SQPOLL thread itself.

In preparation for supporting more than one io_ring_ctx per SQPOLL
thread.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: split work handling part of SQPOLL into helper
Jens Axboe [Mon, 14 Sep 2020 17:07:26 +0000 (11:07 -0600)]
io_uring: split work handling part of SQPOLL into helper

This is done in preparation for handling more than one ctx, but it also
cleans up the code a bit since io_sq_thread() was a bit too unwieldy to
get a get overview on.

__io_sq_thread() is now the main handler, and it returns an enum sq_ret
that tells io_sq_thread() what it ended up doing. The parent then makes
a decision on idle, spinning, or work handling based on that.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: move SQPOLL post-wakeup ring need wakeup flag into wake handler
Jens Axboe [Wed, 2 Sep 2020 18:42:47 +0000 (12:42 -0600)]
io_uring: move SQPOLL post-wakeup ring need wakeup flag into wake handler

We need to decouple the clearing on wakeup from the the inline schedule,
as that is going to be required for handling multiple rings in one
thread.

Wrap our wakeup handler so we can clear it when we get the wakeup, by
definition that is when we no longer need the flag set.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: use private ctx wait queue entries for SQPOLL
Jens Axboe [Wed, 2 Sep 2020 18:21:41 +0000 (12:21 -0600)]
io_uring: use private ctx wait queue entries for SQPOLL

This is in preparation to sharing the poller thread between rings. For
that we need per-ring wait_queue_entry storage, and we can't easily put
that on the stack if one thread is managing multiple rings.

We'll also be sharing the wait_queue_head across rings for the purposes
of wakeups, provide the usual private ring wait_queue_head for now but
make it a pointer so we can easily override it when sharing.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agofs: align IOCB_* flags with RWF_* flags
Jens Axboe [Mon, 31 Aug 2020 18:08:10 +0000 (12:08 -0600)]
fs: align IOCB_* flags with RWF_* flags

We have a set of flags that are shared between the two and inherired
in kiocb_set_rw_flags(), but we check and set these individually.
Reorder the IOCB flags so that the bottom part of the space is synced
with the RWF flag space, and then we can do them all in one mask and
set operation.

The only exception is RWF_SYNC, which needs to mark IOCB_SYNC and
IOCB_DSYNC. Do that one separately.

This shaves 15 bytes of text from kiocb_set_rw_flags() for me.

Suggested-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: io_sq_thread() doesn't need to flush signals
Jens Axboe [Wed, 2 Sep 2020 18:44:20 +0000 (12:44 -0600)]
io_uring: io_sq_thread() doesn't need to flush signals

We're not handling signals by default in kernel threads, and we never
use TWA_SIGNAL for the SQPOLL thread internally. Hence we can never
have a signal pending, and we don't need to check for it (nor flush it).

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_wq: Make io_wqe::lock a raw_spinlock_t
Sebastian Andrzej Siewior [Tue, 1 Sep 2020 08:41:46 +0000 (10:41 +0200)]
io_wq: Make io_wqe::lock a raw_spinlock_t

During a context switch the scheduler invokes wq_worker_sleeping() with
disabled preemption. Disabling preemption is needed because it protects
access to `worker->sleeping'. As an optimisation it avoids invoking
schedule() within the schedule path as part of possible wake up (thus
preempt_enable_no_resched() afterwards).

The io-wq has been added to the mix in the same section with disabled
preemption. This breaks on PREEMPT_RT because io_wq_worker_sleeping()
acquires a spinlock_t. Also within the schedule() the spinlock_t must be
acquired after tsk_is_pi_blocked() otherwise it will block on the
sleeping lock again while scheduling out.

While playing with `io_uring-bench' I didn't notice a significant
latency spike after converting io_wqe::lock to a raw_spinlock_t. The
latency was more or less the same.

In order to keep the spinlock_t it would have to be moved after the
tsk_is_pi_blocked() check which would introduce a branch instruction
into the hot path.

The lock is used to maintain the `work_list' and wakes one task up at
most.
Should io_wqe_cancel_pending_work() cause latency spikes, while
searching for a specific item, then it would need to drop the lock
during iterations.
revert_creds() is also invoked under the lock. According to debug
cred::non_rcu is 0. Otherwise it should be moved outside of the locked
section because put_cred_rcu()->free_uid() acquires a sleeping lock.

Convert io_wqe::lock to a raw_spinlock_t.c

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: allow disabling rings during the creation
Stefano Garzarella [Thu, 27 Aug 2020 14:58:31 +0000 (16:58 +0200)]
io_uring: allow disabling rings during the creation

This patch adds a new IORING_SETUP_R_DISABLED flag to start the
rings disabled, allowing the user to register restrictions,
buffers, files, before to start processing SQEs.

When IORING_SETUP_R_DISABLED is set, SQE are not processed and
SQPOLL kthread is not started.

The restrictions registration are allowed only when the rings
are disable to prevent concurrency issue while processing SQEs.

The rings can be enabled using IORING_REGISTER_ENABLE_RINGS
opcode with io_uring_register(2).

Suggested-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Reviewed-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: add IOURING_REGISTER_RESTRICTIONS opcode
Stefano Garzarella [Thu, 27 Aug 2020 14:58:30 +0000 (16:58 +0200)]
io_uring: add IOURING_REGISTER_RESTRICTIONS opcode

The new io_uring_register(2) IOURING_REGISTER_RESTRICTIONS opcode
permanently installs a feature allowlist on an io_ring_ctx.
The io_ring_ctx can then be passed to untrusted code with the
knowledge that only operations present in the allowlist can be
executed.

The allowlist approach ensures that new features added to io_uring
do not accidentally become available when an existing application
is launched on a newer kernel version.

Currently is it possible to restrict sqe opcodes, sqe flags, and
register opcodes.

IOURING_REGISTER_RESTRICTIONS can only be made once. Afterwards
it is not possible to change restrictions anymore.
This prevents untrusted code from removing restrictions.

Suggested-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Reviewed-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: use an enumeration for io_uring_register(2) opcodes
Stefano Garzarella [Thu, 27 Aug 2020 14:58:29 +0000 (16:58 +0200)]
io_uring: use an enumeration for io_uring_register(2) opcodes

The enumeration allows us to keep track of the last
io_uring_register(2) opcode available.

Behaviour and opcodes names don't change.

Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Reviewed-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: move io_uring_get_socket() into io_uring.h
Jens Axboe [Sat, 19 Sep 2020 02:41:00 +0000 (20:41 -0600)]
io_uring: move io_uring_get_socket() into io_uring.h

Now we have a io_uring kernel header, move this definition out of fs.h
and into io_uring.h where it belongs.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: reference ->nsproxy for file table commands
Jens Axboe [Sat, 19 Sep 2020 02:13:06 +0000 (20:13 -0600)]
io_uring: reference ->nsproxy for file table commands

If we don't get and assign the namespace for the async work, then certain
paths just don't work properly (like /dev/stdin, /proc/mounts, etc).
Anything that references the current namespace of the given task should
be assigned for async work on behalf of that task.

Cc: stable@vger.kernel.org # v5.5+
Reported-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: don't rely on weak ->files references
Jens Axboe [Sun, 13 Sep 2020 19:09:39 +0000 (13:09 -0600)]
io_uring: don't rely on weak ->files references

Grab actual references to the files_struct. To avoid circular references
issues due to this, we add a per-task note that keeps track of what
io_uring contexts a task has used. When the tasks execs or exits its
assigned files, we cancel requests based on this tracking.

With that, we can grab proper references to the files table, and no
longer need to rely on stashing away ring_fd and ring_file to check
if the ring_fd may have been closed.

Cc: stable@vger.kernel.org # v5.5+
Reviewed-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: enable task/files specific overflow flushing
Jens Axboe [Mon, 28 Sep 2020 19:10:13 +0000 (13:10 -0600)]
io_uring: enable task/files specific overflow flushing

This allows us to selectively flush out pending overflows, depending on
the task and/or files_struct being passed in.

No intended functional changes in this patch.

Reviewed-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: return cancelation status from poll/timeout/files handlers
Jens Axboe [Sat, 26 Sep 2020 21:05:03 +0000 (15:05 -0600)]
io_uring: return cancelation status from poll/timeout/files handlers

Return whether we found and canceled requests or not. This is in
preparation for using this information, no functional changes in this
patch.

Reviewed-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: unconditionally grab req->task
Jens Axboe [Thu, 24 Sep 2020 14:45:57 +0000 (08:45 -0600)]
io_uring: unconditionally grab req->task

Sometimes we assign a weak reference to it, sometimes we grab a
reference to it. Clean this up and make it unconditional, and drop the
flag related to tracking this state.

Reviewed-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: stash ctx task reference for SQPOLL
Jens Axboe [Mon, 14 Sep 2020 16:45:53 +0000 (10:45 -0600)]
io_uring: stash ctx task reference for SQPOLL

We can grab a reference to the task instead of stashing away the task
files_struct. This is doable without creating a circular reference
between the ring fd and the task itself.

Reviewed-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: move dropping of files into separate helper
Jens Axboe [Tue, 22 Sep 2020 16:19:24 +0000 (10:19 -0600)]
io_uring: move dropping of files into separate helper

No functional changes in this patch, prep patch for grabbing references
to the files_struct.

Reviewed-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: allow timeout/poll/files killing to take task into account
Jens Axboe [Tue, 22 Sep 2020 14:18:24 +0000 (08:18 -0600)]
io_uring: allow timeout/poll/files killing to take task into account

We currently cancel these when the ring exits, and we cancel all of
them. This is in preparation for killing only the ones associated
with a given task.

Reviewed-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoMerge branch 'io_uring-5.9' into for-5.10/io_uring
Jens Axboe [Thu, 1 Oct 2020 02:32:25 +0000 (20:32 -0600)]
Merge branch 'io_uring-5.9' into for-5.10/io_uring

* io_uring-5.9:
  io_uring: fix async buffered reads when readahead is disabled
  io_uring: fix potential ABBA deadlock in ->show_fdinfo()
  io_uring: always delete double poll wait entry on match

4 years agoio_uring: fix async buffered reads when readahead is disabled
Hao Xu [Tue, 29 Sep 2020 12:00:45 +0000 (20:00 +0800)]
io_uring: fix async buffered reads when readahead is disabled

The async buffered reads feature is not working when readahead is
turned off. There are two things to concern:

- when doing retry in io_read, not only the IOCB_WAITQ flag but also
  the IOCB_NOWAIT flag is still set, which makes it goes to would_block
  phase in generic_file_buffered_read() and then return -EAGAIN. After
  that, the io-wq thread work is queued, and later doing the async
  reads in the old way.

- even if we remove IOCB_NOWAIT when doing retry, the feature is still
  not running properly, since in generic_file_buffered_read() it goes to
  lock_page_killable() after calling mapping->a_ops->readpage() to do
  IO, and thus causing process to sleep.

Fixes: 1a0a7853b901 ("mm: support async buffered reads in generic_file_buffered_read()")
Fixes: 3b2a4439e0ae ("io_uring: get rid of kiocb_wait_page_queue_init()")
Signed-off-by: Hao Xu <haoxu@linux.alibaba.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoMerge tag 'nfs-for-5.9-3' of git://git.linux-nfs.org/projects/trondmy/linux-nfs
Linus Torvalds [Mon, 28 Sep 2020 18:05:56 +0000 (11:05 -0700)]
Merge tag 'nfs-for-5.9-3' of git://git.linux-nfs.org/projects/trondmy/linux-nfs

Pull NFS client bugfixes from Trond Myklebust:
 "Highlights include:

   - NFSv4.2: copy_file_range needs to invalidate caches on success

   - NFSv4.2: Fix security label length not being reset

   - pNFS/flexfiles: Ensure we initialise the mirror bsizes correctly
     on read

   - pNFS/flexfiles: Fix signed/unsigned type issues with mirror
     indices"

* tag 'nfs-for-5.9-3' of git://git.linux-nfs.org/projects/trondmy/linux-nfs:
  pNFS/flexfiles: Be consistent about mirror index types
  pNFS/flexfiles: Ensure we initialise the mirror bsizes correctly on read
  NFSv4.2: fix client's attribute cache management for copy_file_range
  nfs: Fix security label length not being reset

4 years agomm: do not rely on mm == current->mm in __get_user_pages_locked
Jason A. Donenfeld [Mon, 28 Sep 2020 10:35:07 +0000 (12:35 +0200)]
mm: do not rely on mm == current->mm in __get_user_pages_locked

It seems likely this block was pasted from internal_get_user_pages_fast,
which is not passed an mm struct and therefore uses current's.  But
__get_user_pages_locked is passed an explicit mm, and current->mm is not
always valid. This was hit when being called from i915, which uses:

  pin_user_pages_remote->
    __get_user_pages_remote->
      __gup_longterm_locked->
        __get_user_pages_locked

Before, this would lead to an OOPS:

  BUG: kernel NULL pointer dereference, address: 0000000000000064
  #PF: supervisor write access in kernel mode
  #PF: error_code(0x0002) - not-present page
  CPU: 10 PID: 1431 Comm: kworker/u33:1 Tainted: P S   U     O      5.9.0-rc7+ #140
  Hardware name: LENOVO 20QTCTO1WW/20QTCTO1WW, BIOS N2OET47W (1.34 ) 08/06/2020
  Workqueue: i915-userptr-acquire __i915_gem_userptr_get_pages_worker [i915]
  RIP: 0010:__get_user_pages_remote+0xd7/0x310
  Call Trace:
   __i915_gem_userptr_get_pages_worker+0xc8/0x260 [i915]
   process_one_work+0x1ca/0x390
   worker_thread+0x48/0x3c0
   kthread+0x114/0x130
   ret_from_fork+0x1f/0x30
  CR2: 0000000000000064

This commit fixes the problem by using the mm pointer passed to the
function rather than the bogus one in current.

Fixes: 008cfe4418b3 ("mm: Introduce mm_struct.has_pinned")
Tested-by: Chris Wilson <chris@chris-wilson.co.uk>
Reported-by: Harald Arnesen <harald@skogtun.org>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agoio_uring: fix potential ABBA deadlock in ->show_fdinfo()
Jens Axboe [Mon, 28 Sep 2020 14:57:48 +0000 (08:57 -0600)]
io_uring: fix potential ABBA deadlock in ->show_fdinfo()

syzbot reports a potential lock deadlock between the normal IO path and
->show_fdinfo():

======================================================
WARNING: possible circular locking dependency detected
5.9.0-rc6-syzkaller #0 Not tainted
------------------------------------------------------
syz-executor.2/19710 is trying to acquire lock:
ffff888098ddc450 (sb_writers#4){.+.+}-{0:0}, at: io_write+0x6b5/0xb30 fs/io_uring.c:3296

but task is already holding lock:
ffff8880a11b8428 (&ctx->uring_lock){+.+.}-{3:3}, at: __do_sys_io_uring_enter+0xe9a/0x1bd0 fs/io_uring.c:8348

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:

-> #2 (&ctx->uring_lock){+.+.}-{3:3}:
       __mutex_lock_common kernel/locking/mutex.c:956 [inline]
       __mutex_lock+0x134/0x10e0 kernel/locking/mutex.c:1103
       __io_uring_show_fdinfo fs/io_uring.c:8417 [inline]
       io_uring_show_fdinfo+0x194/0xc70 fs/io_uring.c:8460
       seq_show+0x4a8/0x700 fs/proc/fd.c:65
       seq_read+0x432/0x1070 fs/seq_file.c:208
       do_loop_readv_writev fs/read_write.c:734 [inline]
       do_loop_readv_writev fs/read_write.c:721 [inline]
       do_iter_read+0x48e/0x6e0 fs/read_write.c:955
       vfs_readv+0xe5/0x150 fs/read_write.c:1073
       kernel_readv fs/splice.c:355 [inline]
       default_file_splice_read.constprop.0+0x4e6/0x9e0 fs/splice.c:412
       do_splice_to+0x137/0x170 fs/splice.c:871
       splice_direct_to_actor+0x307/0x980 fs/splice.c:950
       do_splice_direct+0x1b3/0x280 fs/splice.c:1059
       do_sendfile+0x55f/0xd40 fs/read_write.c:1540
       __do_sys_sendfile64 fs/read_write.c:1601 [inline]
       __se_sys_sendfile64 fs/read_write.c:1587 [inline]
       __x64_sys_sendfile64+0x1cc/0x210 fs/read_write.c:1587
       do_syscall_64+0x2d/0x70 arch/x86/entry/common.c:46
       entry_SYSCALL_64_after_hwframe+0x44/0xa9

-> #1 (&p->lock){+.+.}-{3:3}:
       __mutex_lock_common kernel/locking/mutex.c:956 [inline]
       __mutex_lock+0x134/0x10e0 kernel/locking/mutex.c:1103
       seq_read+0x61/0x1070 fs/seq_file.c:155
       pde_read fs/proc/inode.c:306 [inline]
       proc_reg_read+0x221/0x300 fs/proc/inode.c:318
       do_loop_readv_writev fs/read_write.c:734 [inline]
       do_loop_readv_writev fs/read_write.c:721 [inline]
       do_iter_read+0x48e/0x6e0 fs/read_write.c:955
       vfs_readv+0xe5/0x150 fs/read_write.c:1073
       kernel_readv fs/splice.c:355 [inline]
       default_file_splice_read.constprop.0+0x4e6/0x9e0 fs/splice.c:412
       do_splice_to+0x137/0x170 fs/splice.c:871
       splice_direct_to_actor+0x307/0x980 fs/splice.c:950
       do_splice_direct+0x1b3/0x280 fs/splice.c:1059
       do_sendfile+0x55f/0xd40 fs/read_write.c:1540
       __do_sys_sendfile64 fs/read_write.c:1601 [inline]
       __se_sys_sendfile64 fs/read_write.c:1587 [inline]
       __x64_sys_sendfile64+0x1cc/0x210 fs/read_write.c:1587
       do_syscall_64+0x2d/0x70 arch/x86/entry/common.c:46
       entry_SYSCALL_64_after_hwframe+0x44/0xa9

-> #0 (sb_writers#4){.+.+}-{0:0}:
       check_prev_add kernel/locking/lockdep.c:2496 [inline]
       check_prevs_add kernel/locking/lockdep.c:2601 [inline]
       validate_chain kernel/locking/lockdep.c:3218 [inline]
       __lock_acquire+0x2a96/0x5780 kernel/locking/lockdep.c:4441
       lock_acquire+0x1f3/0xaf0 kernel/locking/lockdep.c:5029
       percpu_down_read include/linux/percpu-rwsem.h:51 [inline]
       __sb_start_write+0x228/0x450 fs/super.c:1672
       io_write+0x6b5/0xb30 fs/io_uring.c:3296
       io_issue_sqe+0x18f/0x5c50 fs/io_uring.c:5719
       __io_queue_sqe+0x280/0x1160 fs/io_uring.c:6175
       io_queue_sqe+0x692/0xfa0 fs/io_uring.c:6254
       io_submit_sqe fs/io_uring.c:6324 [inline]
       io_submit_sqes+0x1761/0x2400 fs/io_uring.c:6521
       __do_sys_io_uring_enter+0xeac/0x1bd0 fs/io_uring.c:8349
       do_syscall_64+0x2d/0x70 arch/x86/entry/common.c:46
       entry_SYSCALL_64_after_hwframe+0x44/0xa9

other info that might help us debug this:

Chain exists of:
  sb_writers#4 --> &p->lock --> &ctx->uring_lock

 Possible unsafe locking scenario:

       CPU0                    CPU1
       ----                    ----
  lock(&ctx->uring_lock);
                               lock(&p->lock);
                               lock(&ctx->uring_lock);
  lock(sb_writers#4);

 *** DEADLOCK ***

1 lock held by syz-executor.2/19710:
 #0: ffff8880a11b8428 (&ctx->uring_lock){+.+.}-{3:3}, at: __do_sys_io_uring_enter+0xe9a/0x1bd0 fs/io_uring.c:8348

stack backtrace:
CPU: 0 PID: 19710 Comm: syz-executor.2 Not tainted 5.9.0-rc6-syzkaller #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
Call Trace:
 __dump_stack lib/dump_stack.c:77 [inline]
 dump_stack+0x198/0x1fd lib/dump_stack.c:118
 check_noncircular+0x324/0x3e0 kernel/locking/lockdep.c:1827
 check_prev_add kernel/locking/lockdep.c:2496 [inline]
 check_prevs_add kernel/locking/lockdep.c:2601 [inline]
 validate_chain kernel/locking/lockdep.c:3218 [inline]
 __lock_acquire+0x2a96/0x5780 kernel/locking/lockdep.c:4441
 lock_acquire+0x1f3/0xaf0 kernel/locking/lockdep.c:5029
 percpu_down_read include/linux/percpu-rwsem.h:51 [inline]
 __sb_start_write+0x228/0x450 fs/super.c:1672
 io_write+0x6b5/0xb30 fs/io_uring.c:3296
 io_issue_sqe+0x18f/0x5c50 fs/io_uring.c:5719
 __io_queue_sqe+0x280/0x1160 fs/io_uring.c:6175
 io_queue_sqe+0x692/0xfa0 fs/io_uring.c:6254
 io_submit_sqe fs/io_uring.c:6324 [inline]
 io_submit_sqes+0x1761/0x2400 fs/io_uring.c:6521
 __do_sys_io_uring_enter+0xeac/0x1bd0 fs/io_uring.c:8349
 do_syscall_64+0x2d/0x70 arch/x86/entry/common.c:46
 entry_SYSCALL_64_after_hwframe+0x44/0xa9
RIP: 0033:0x45e179
Code: 3d b2 fb ff c3 66 2e 0f 1f 84 00 00 00 00 00 66 90 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 0f 83 0b b2 fb ff c3 66 2e 0f 1f 84 00 00 00 00
RSP: 002b:00007f1194e74c78 EFLAGS: 00000246 ORIG_RAX: 00000000000001aa
RAX: ffffffffffffffda RBX: 00000000000082c0 RCX: 000000000045e179
RDX: 0000000000000000 RSI: 0000000000000001 RDI: 0000000000000004
RBP: 000000000118cf98 R08: 0000000000000000 R09: 0000000000000000
R10: 0000000000000000 R11: 0000000000000246 R12: 000000000118cf4c
R13: 00007ffd1aa5756f R14: 00007f1194e759c0 R15: 000000000118cf4c

Fix this by just not diving into details if we fail to trylock the
io_uring mutex. We know the ctx isn't going away during this operation,
but we cannot safely iterate buffers/files/personalities if we don't
hold the io_uring mutex.

Reported-by: syzbot+2f8fa4e860edc3066aba@syzkaller.appspotmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: always delete double poll wait entry on match
Jens Axboe [Mon, 28 Sep 2020 14:38:54 +0000 (08:38 -0600)]
io_uring: always delete double poll wait entry on match

syzbot reports a crash with tty polling, which is using the double poll
handling:

general protection fault, probably for non-canonical address 0xdffffc0000000009: 0000 [#1] PREEMPT SMP KASAN
KASAN: null-ptr-deref in range [0x0000000000000048-0x000000000000004f]
CPU: 0 PID: 6874 Comm: syz-executor749 Not tainted 5.9.0-rc6-next-20200924-syzkaller #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
RIP: 0010:io_poll_get_single fs/io_uring.c:4778 [inline]
RIP: 0010:io_poll_double_wake+0x51/0x510 fs/io_uring.c:4845
Code: fc ff df 48 c1 ea 03 80 3c 02 00 0f 85 9e 03 00 00 48 b8 00 00 00 00 00 fc ff df 49 8b 5d 08 48 8d 7b 48 48 89 fa 48 c1 ea 03 <0f> b6 04 02 84 c0 74 06 0f 8e 63 03 00 00 0f b6 6b 48 bf 06 00 00
RSP: 0018:ffffc90001c1fb70 EFLAGS: 00010006
RAX: dffffc0000000000 RBX: 0000000000000000 RCX: 0000000000000004
RDX: 0000000000000009 RSI: ffffffff81d9b3ad RDI: 0000000000000048
RBP: dffffc0000000000 R08: ffff8880a3cac798 R09: ffffc90001c1fc60
R10: fffff52000383f73 R11: 0000000000000000 R12: 0000000000000004
R13: ffff8880a3cac798 R14: ffff8880a3cac7a0 R15: 0000000000000004
FS:  0000000001f98880(0000) GS:ffff8880ae400000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00007f18886916c0 CR3: 0000000094c5a000 CR4: 00000000001506f0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:
 __wake_up_common+0x147/0x650 kernel/sched/wait.c:93
 __wake_up_common_lock+0xd0/0x130 kernel/sched/wait.c:123
 tty_ldisc_hangup+0x1cf/0x680 drivers/tty/tty_ldisc.c:735
 __tty_hangup.part.0+0x403/0x870 drivers/tty/tty_io.c:625
 __tty_hangup drivers/tty/tty_io.c:575 [inline]
 tty_vhangup+0x1d/0x30 drivers/tty/tty_io.c:698
 pty_close+0x3f5/0x550 drivers/tty/pty.c:79
 tty_release+0x455/0xf60 drivers/tty/tty_io.c:1679
 __fput+0x285/0x920 fs/file_table.c:281
 task_work_run+0xdd/0x190 kernel/task_work.c:141
 tracehook_notify_resume include/linux/tracehook.h:188 [inline]
 exit_to_user_mode_loop kernel/entry/common.c:165 [inline]
 exit_to_user_mode_prepare+0x1e2/0x1f0 kernel/entry/common.c:192
 syscall_exit_to_user_mode+0x7a/0x2c0 kernel/entry/common.c:267
 entry_SYSCALL_64_after_hwframe+0x44/0xa9
RIP: 0033:0x401210

which is due to a failure in removing the double poll wait entry if we
hit a wakeup match. This can cause multiple invocations of the wakeup,
which isn't safe.

Cc: stable@vger.kernel.org # v5.8
Reported-by: syzbot+81b3883093f772addf6d@syzkaller.appspotmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoLinux 5.9-rc7
Linus Torvalds [Sun, 27 Sep 2020 21:38:10 +0000 (14:38 -0700)]
Linux 5.9-rc7

4 years agoMerge tag 'kbuild-fixes-v5.9-4' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 27 Sep 2020 19:18:57 +0000 (12:18 -0700)]
Merge tag 'kbuild-fixes-v5.9-4' of git://git./linux/kernel/git/masahiroy/linux-kbuild

Pull Kbuild fixes from Masahiro Yamada:

 - ignore compiler stubs for PPC to fix builds

 - fix the usage of --target mentioned in the LLVM document

* tag 'kbuild-fixes-v5.9-4' of git://git.kernel.org/pub/scm/linux/kernel/git/masahiroy/linux-kbuild:
  Documentation/llvm: Fix clang target examples
  scripts/kallsyms: skip ppc compiler stub *.long_branch.* / *.plt_branch.*

4 years agoMerge tag 'x86-urgent-2020-09-27' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 27 Sep 2020 19:15:21 +0000 (12:15 -0700)]
Merge tag 'x86-urgent-2020-09-27' of git://git./linux/kernel/git/tip/tip

Pull x86 fixes from Thomas Gleixner:
 "Two fixes for the x86 interrupt code:

   - Unbreak the magic 'search the timer interrupt' logic in IO/APIC
     code which got wreckaged when the core interrupt code made the
     state tracking logic stricter.

     That caused the interrupt line to stay masked after switching from
     IO/APIC to PIC delivery mode, which obviously prevents interrupts
     from being delivered.

   - Make run_on_irqstack_code() typesafe. The function argument is a
     void pointer which is then cast to 'void (*fun)(void *).

     This breaks Control Flow Integrity checking in clang. Use proper
     helper functions for the three variants reuqired"

* tag 'x86-urgent-2020-09-27' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/ioapic: Unbreak check_timer()
  x86/irq: Make run_on_irqstack_cond() typesafe

4 years agoMerge tag 'timers-urgent-2020-09-27' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 27 Sep 2020 19:11:35 +0000 (12:11 -0700)]
Merge tag 'timers-urgent-2020-09-27' of git://git./linux/kernel/git/tip/tip

Pull timer updates from Thomas Gleixner:
 "A set of clocksource/clockevents updates:

   - Reset the TI/DM timer before enabling it instead of doing it the
     other way round.

   - Initialize the reload value for the GX6605s timer correctly so the
     hardware counter starts at 0 again after overrun.

   - Make error return value negative in the h8300 timer init function"

* tag 'timers-urgent-2020-09-27' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  clocksource/drivers/timer-gx6605s: Fixup counter reload
  clocksource/drivers/timer-ti-dm: Do reset before enable
  clocksource/drivers/h8300_timer8: Fix wrong return value in h8300_8timer_init()

4 years agomm/thp: Split huge pmds/puds if they're pinned when fork()
Peter Xu [Fri, 25 Sep 2020 22:26:00 +0000 (18:26 -0400)]
mm/thp: Split huge pmds/puds if they're pinned when fork()

Pinned pages shouldn't be write-protected when fork() happens, because
follow up copy-on-write on these pages could cause the pinned pages to
be replaced by random newly allocated pages.

For huge PMDs, we split the huge pmd if pinning is detected.  So that
future handling will be done by the PTE level (with our latest changes,
each of the small pages will be copied).  We can achieve this by let
copy_huge_pmd() return -EAGAIN for pinned pages, so that we'll
fallthrough in copy_pmd_range() and finally land the next
copy_pte_range() call.

Huge PUDs will be even more special - so far it does not support
anonymous pages.  But it can actually be done the same as the huge PMDs
even if the split huge PUDs means to erase the PUD entries.  It'll
guarantee the follow up fault ins will remap the same pages in either
parent/child later.

This might not be the most efficient way, but it should be easy and
clean enough.  It should be fine, since we're tackling with a very rare
case just to make sure userspaces that pinned some thps will still work
even without MADV_DONTFORK and after they fork()ed.

Signed-off-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agomm: Do early cow for pinned pages during fork() for ptes
Peter Xu [Fri, 25 Sep 2020 22:25:59 +0000 (18:25 -0400)]
mm: Do early cow for pinned pages during fork() for ptes

This allows copy_pte_range() to do early cow if the pages were pinned on
the source mm.

Currently we don't have an accurate way to know whether a page is pinned
or not.  The only thing we have is page_maybe_dma_pinned().  However
that's good enough for now.  Especially, with the newly added
mm->has_pinned flag to make sure we won't affect processes that never
pinned any pages.

It would be easier if we can do GFP_KERNEL allocation within
copy_one_pte().  Unluckily, we can't because we're with the page table
locks held for both the parent and child processes.  So the page
allocation needs to be done outside copy_one_pte().

Some trick is there in copy_present_pte(), majorly the wrprotect trick
to block concurrent fast-gup.  Comments in the function should explain
better in place.

Oleg Nesterov reported a (probably harmless) bug during review that we
didn't reset entry.val properly in copy_pte_range() so that potentially
there's chance to call add_swap_count_continuation() multiple times on
the same swp entry.  However that should be harmless since even if it
happens, the same function (add_swap_count_continuation()) will return
directly noticing that there're enough space for the swp counter.  So
instead of a standalone stable patch, it is touched up in this patch
directly.

Link: https://lore.kernel.org/lkml/20200914143829.GA1424636@nvidia.com/
Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agomm/fork: Pass new vma pointer into copy_page_range()
Peter Xu [Fri, 25 Sep 2020 22:25:58 +0000 (18:25 -0400)]
mm/fork: Pass new vma pointer into copy_page_range()

This prepares for the future work to trigger early cow on pinned pages
during fork().

No functional change intended.

Signed-off-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agomm: Introduce mm_struct.has_pinned
Peter Xu [Fri, 25 Sep 2020 22:25:57 +0000 (18:25 -0400)]
mm: Introduce mm_struct.has_pinned

(Commit message majorly collected from Jason Gunthorpe)

Reduce the chance of false positive from page_maybe_dma_pinned() by
keeping track if the mm_struct has ever been used with pin_user_pages().
This allows cases that might drive up the page ref_count to avoid any
penalty from handling dma_pinned pages.

Future work is planned, to provide a more sophisticated solution, likely
to turn it into a real counter.  For now, make it atomic_t but use it as
a boolean for simplicity.

Suggested-by: Jason Gunthorpe <jgg@ziepe.ca>
Signed-off-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agoMerge tag 'timers-v5.9-rc4' of https://git.linaro.org/people/daniel.lezcano/linux...
Thomas Gleixner [Sun, 27 Sep 2020 09:24:34 +0000 (11:24 +0200)]
Merge tag 'timers-v5.9-rc4' of https://git.linaro.org/people/daniel.lezcano/linux into timers/urgent

Pull clocksource/clockevent fixes from Daniel Lezcano:

 - Fix wrong signed return value when checking of_iomap in the probe
   function for the h8300 timer (Tianjia Zhang)

 - Fix reset sequence when setting up the timer on the dm_timer (Tony
   Lindgren)

 - Fix counter reload when the interrupt fires on gx6605s (Guo Ren)

4 years agoMerge tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi
Linus Torvalds [Sat, 26 Sep 2020 18:18:37 +0000 (11:18 -0700)]
Merge tag 'scsi-fixes' of git://git./linux/kernel/git/jejb/scsi

Pull SCSI fixes from James Bottomley:
 "Three fixes: one in drivers (lpfc) and two for zoned block devices.

  The latter also impinges on the block layer but only to introduce a
  new block API for setting the zone model rather than fiddling with the
  queue directly in the zoned block driver"

* tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi:
  scsi: sd: sd_zbc: Fix ZBC disk initialization
  scsi: sd: sd_zbc: Fix handling of host-aware ZBC disks
  scsi: lpfc: Fix initial FLOGI failure due to BBSCN not supported

4 years agoMerge tag 'io_uring-5.9-2020-09-25' of git://git.kernel.dk/linux-block
Linus Torvalds [Sat, 26 Sep 2020 18:13:51 +0000 (11:13 -0700)]
Merge tag 'io_uring-5.9-2020-09-25' of git://git.kernel.dk/linux-block

Pull io_uring fixes from Jens Axboe:
 "Two fixes for regressions in this cycle, and one that goes to 5.8
  stable:

   - fix leak of getname() retrieved filename

   - remove plug->nowait assignment, fixing a regression with btrfs

   - fix for async buffered retry"

* tag 'io_uring-5.9-2020-09-25' of git://git.kernel.dk/linux-block:
  io_uring: ensure async buffered read-retry is setup properly
  io_uring: don't unconditionally set plug->nowait = true
  io_uring: ensure open/openat2 name is cleaned on cancelation

4 years agoMerge tag 'block-5.9-2020-09-25' of git://git.kernel.dk/linux-block
Linus Torvalds [Sat, 26 Sep 2020 18:07:36 +0000 (11:07 -0700)]
Merge tag 'block-5.9-2020-09-25' of git://git.kernel.dk/linux-block

Pull block fixes from Jens Axboe:
 "NVMe pull request from Christoph, and removal of a dead define.

   - fix error during controller probe that cause double free irqs
     (Keith Busch)

   - FC connection establishment fix (James Smart)

   - properly handle completions for invalid tags (Xianting Tian)

   - pass the correct nsid to the command effects and supported log
     (Chaitanya Kulkarni)"

* tag 'block-5.9-2020-09-25' of git://git.kernel.dk/linux-block:
  block: remove unused BLK_QC_T_EAGAIN flag
  nvme-core: don't use NVME_NSID_ALL for command effects and supported log
  nvme-fc: fail new connections to a deleted host or remote port
  nvme-pci: fix NULL req in completion handler
  nvme: return errors for hwmon init

4 years agoMerge tag 's390-5.9-7' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux
Linus Torvalds [Sat, 26 Sep 2020 18:01:18 +0000 (11:01 -0700)]
Merge tag 's390-5.9-7' of git://git./linux/kernel/git/s390/linux

Pull s390 fix from Vasily Gorbik:
 "Fix truncated ZCRYPT_PERDEV_REQCNT ioctl result. Copy entire reqcnt
  list"

* tag 's390-5.9-7' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux:
  s390/zcrypt: Fix ZCRYPT_PERDEV_REQCNT ioctl

4 years agoMerge branch 'akpm' (patches from Andrew)
Linus Torvalds [Sat, 26 Sep 2020 17:53:35 +0000 (10:53 -0700)]
Merge branch 'akpm' (patches from Andrew)

Merge misc fixes from Andrew Morton:
 "9 patches.

  Subsystems affected by this patch series: mm (thp, memcg, gup,
  migration, memory-hotplug), lib, and x86"

* emailed patches from Andrew Morton <akpm@linux-foundation.org>:
  mm: don't rely on system state to detect hot-plug operations
  mm: replace memmap_context by meminit_context
  arch/x86/lib/usercopy_64.c: fix __copy_user_flushcache() cache writeback
  lib/memregion.c: include memregion.h
  lib/string.c: implement stpcpy
  mm/migrate: correct thp migration stats
  mm/gup: fix gup_fast with dynamic page table folding
  mm: memcontrol: fix missing suffix of workingset_restore
  mm, THP, swap: fix allocating cluster for swapfile by mistake

4 years agomm: validate pmd after splitting
Minchan Kim [Tue, 15 Sep 2020 06:32:15 +0000 (23:32 -0700)]
mm: validate pmd after splitting

syzbot reported the following KASAN splat:

  general protection fault, probably for non-canonical address 0xdffffc0000000003: 0000 [#1] PREEMPT SMP KASAN
  KASAN: null-ptr-deref in range [0x0000000000000018-0x000000000000001f]
  CPU: 1 PID: 6826 Comm: syz-executor142 Not tainted 5.9.0-rc4-syzkaller #0
  Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
  RIP: 0010:__lock_acquire+0x84/0x2ae0 kernel/locking/lockdep.c:4296
  Code: ff df 8a 04 30 84 c0 0f 85 e3 16 00 00 83 3d 56 58 35 08 00 0f 84 0e 17 00 00 83 3d 25 c7 f5 07 00 74 2c 4c 89 e8 48 c1 e8 03 <80> 3c 30 00 74 12 4c 89 ef e8 3e d1 5a 00 48 be 00 00 00 00 00 fc
  RSP: 0018:ffffc90004b9f850 EFLAGS: 00010006
  Call Trace:
    lock_acquire+0x140/0x6f0 kernel/locking/lockdep.c:5006
    __raw_spin_lock include/linux/spinlock_api_smp.h:142 [inline]
    _raw_spin_lock+0x2a/0x40 kernel/locking/spinlock.c:151
    spin_lock include/linux/spinlock.h:354 [inline]
    madvise_cold_or_pageout_pte_range+0x52f/0x25c0 mm/madvise.c:389
    walk_pmd_range mm/pagewalk.c:89 [inline]
    walk_pud_range mm/pagewalk.c:160 [inline]
    walk_p4d_range mm/pagewalk.c:193 [inline]
    walk_pgd_range mm/pagewalk.c:229 [inline]
    __walk_page_range+0xe7b/0x1da0 mm/pagewalk.c:331
    walk_page_range+0x2c3/0x5c0 mm/pagewalk.c:427
    madvise_pageout_page_range mm/madvise.c:521 [inline]
    madvise_pageout mm/madvise.c:557 [inline]
    madvise_vma mm/madvise.c:946 [inline]
    do_madvise+0x12d0/0x2090 mm/madvise.c:1145
    __do_sys_madvise mm/madvise.c:1171 [inline]
    __se_sys_madvise mm/madvise.c:1169 [inline]
    __x64_sys_madvise+0x76/0x80 mm/madvise.c:1169
    do_syscall_64+0x31/0x70 arch/x86/entry/common.c:46
    entry_SYSCALL_64_after_hwframe+0x44/0xa9

The backing vma was shmem.

In case of split page of file-backed THP, madvise zaps the pmd instead
of remapping of sub-pages.  So we need to check pmd validity after
split.

Reported-by: syzbot+ecf80462cb7d5d552bc7@syzkaller.appspotmail.com
Fixes: 1a4e58cce84e ("mm: introduce MADV_PAGEOUT")
Signed-off-by: Minchan Kim <minchan@kernel.org>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agomm: don't rely on system state to detect hot-plug operations
Laurent Dufour [Sat, 26 Sep 2020 04:19:31 +0000 (21:19 -0700)]
mm: don't rely on system state to detect hot-plug operations

In register_mem_sect_under_node() the system_state's value is checked to
detect whether the call is made during boot time or during an hot-plug
operation.  Unfortunately, that check against SYSTEM_BOOTING is wrong
because regular memory is registered at SYSTEM_SCHEDULING state.  In
addition, memory hot-plug operation can be triggered at this system
state by the ACPI [1].  So checking against the system state is not
enough.

The consequence is that on system with interleaved node's ranges like this:

 Early memory node ranges
   node   1: [mem 0x0000000000000000-0x000000011fffffff]
   node   2: [mem 0x0000000120000000-0x000000014fffffff]
   node   1: [mem 0x0000000150000000-0x00000001ffffffff]
   node   0: [mem 0x0000000200000000-0x000000048fffffff]
   node   2: [mem 0x0000000490000000-0x00000007ffffffff]

This can be seen on PowerPC LPAR after multiple memory hot-plug and
hot-unplug operations are done.  At the next reboot the node's memory
ranges can be interleaved and since the call to link_mem_sections() is
made in topology_init() while the system is in the SYSTEM_SCHEDULING
state, the node's id is not checked, and the sections registered to
multiple nodes:

  $ ls -l /sys/devices/system/memory/memory21/node*
  total 0
  lrwxrwxrwx 1 root root     0 Aug 24 05:27 node1 -> ../../node/node1
  lrwxrwxrwx 1 root root     0 Aug 24 05:27 node2 -> ../../node/node2

In that case, the system is able to boot but if later one of theses
memory blocks is hot-unplugged and then hot-plugged, the sysfs
inconsistency is detected and this is triggering a BUG_ON():

  kernel BUG at /Users/laurent/src/linux-ppc/mm/memory_hotplug.c:1084!
  Oops: Exception in kernel mode, sig: 5 [#1]
  LE PAGE_SIZE=64K MMU=Hash SMP NR_CPUS=2048 NUMA pSeries
  Modules linked in: rpadlpar_io rpaphp pseries_rng rng_core vmx_crypto gf128mul binfmt_misc ip_tables x_tables xfs libcrc32c crc32c_vpmsum autofs4
  CPU: 8 PID: 10256 Comm: drmgr Not tainted 5.9.0-rc1+ #25
  Call Trace:
    add_memory_resource+0x23c/0x340 (unreliable)
    __add_memory+0x5c/0xf0
    dlpar_add_lmb+0x1b4/0x500
    dlpar_memory+0x1f8/0xb80
    handle_dlpar_errorlog+0xc0/0x190
    dlpar_store+0x198/0x4a0
    kobj_attr_store+0x30/0x50
    sysfs_kf_write+0x64/0x90
    kernfs_fop_write+0x1b0/0x290
    vfs_write+0xe8/0x290
    ksys_write+0xdc/0x130
    system_call_exception+0x160/0x270
    system_call_common+0xf0/0x27c

This patch addresses the root cause by not relying on the system_state
value to detect whether the call is due to a hot-plug operation.  An
extra parameter is added to link_mem_sections() detailing whether the
operation is due to a hot-plug operation.

[1] According to Oscar Salvador, using this qemu command line, ACPI
memory hotplug operations are raised at SYSTEM_SCHEDULING state:

  $QEMU -enable-kvm -machine pc -smp 4,sockets=4,cores=1,threads=1 -cpu host -monitor pty \
        -m size=$MEM,slots=255,maxmem=4294967296k  \
        -numa node,nodeid=0,cpus=0-3,mem=512 -numa node,nodeid=1,mem=512 \
        -object memory-backend-ram,id=memdimm0,size=134217728 -device pc-dimm,node=0,memdev=memdimm0,id=dimm0,slot=0 \
        -object memory-backend-ram,id=memdimm1,size=134217728 -device pc-dimm,node=0,memdev=memdimm1,id=dimm1,slot=1 \
        -object memory-backend-ram,id=memdimm2,size=134217728 -device pc-dimm,node=0,memdev=memdimm2,id=dimm2,slot=2 \
        -object memory-backend-ram,id=memdimm3,size=134217728 -device pc-dimm,node=0,memdev=memdimm3,id=dimm3,slot=3 \
        -object memory-backend-ram,id=memdimm4,size=134217728 -device pc-dimm,node=1,memdev=memdimm4,id=dimm4,slot=4 \
        -object memory-backend-ram,id=memdimm5,size=134217728 -device pc-dimm,node=1,memdev=memdimm5,id=dimm5,slot=5 \
        -object memory-backend-ram,id=memdimm6,size=134217728 -device pc-dimm,node=1,memdev=memdimm6,id=dimm6,slot=6 \

Fixes: 4fbce633910e ("mm/memory_hotplug.c: make register_mem_sect_under_node() a callback of walk_memory_range()")
Signed-off-by: Laurent Dufour <ldufour@linux.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: "Rafael J. Wysocki" <rafael@kernel.org>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: Nathan Lynch <nathanl@linux.ibm.com>
Cc: Scott Cheloha <cheloha@linux.ibm.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: <stable@vger.kernel.org>
Link: https://lkml.kernel.org/r/20200915094143.79181-3-ldufour@linux.ibm.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agomm: replace memmap_context by meminit_context
Laurent Dufour [Sat, 26 Sep 2020 04:19:28 +0000 (21:19 -0700)]
mm: replace memmap_context by meminit_context

Patch series "mm: fix memory to node bad links in sysfs", v3.

Sometimes, firmware may expose interleaved memory layout like this:

 Early memory node ranges
   node   1: [mem 0x0000000000000000-0x000000011fffffff]
   node   2: [mem 0x0000000120000000-0x000000014fffffff]
   node   1: [mem 0x0000000150000000-0x00000001ffffffff]
   node   0: [mem 0x0000000200000000-0x000000048fffffff]
   node   2: [mem 0x0000000490000000-0x00000007ffffffff]

In that case, we can see memory blocks assigned to multiple nodes in
sysfs:

  $ ls -l /sys/devices/system/memory/memory21
  total 0
  lrwxrwxrwx 1 root root     0 Aug 24 05:27 node1 -> ../../node/node1
  lrwxrwxrwx 1 root root     0 Aug 24 05:27 node2 -> ../../node/node2
  -rw-r--r-- 1 root root 65536 Aug 24 05:27 online
  -r--r--r-- 1 root root 65536 Aug 24 05:27 phys_device
  -r--r--r-- 1 root root 65536 Aug 24 05:27 phys_index
  drwxr-xr-x 2 root root     0 Aug 24 05:27 power
  -r--r--r-- 1 root root 65536 Aug 24 05:27 removable
  -rw-r--r-- 1 root root 65536 Aug 24 05:27 state
  lrwxrwxrwx 1 root root     0 Aug 24 05:25 subsystem -> ../../../../bus/memory
  -rw-r--r-- 1 root root 65536 Aug 24 05:25 uevent
  -r--r--r-- 1 root root 65536 Aug 24 05:27 valid_zones

The same applies in the node's directory with a memory21 link in both
the node1 and node2's directory.

This is wrong but doesn't prevent the system to run.  However when
later, one of these memory blocks is hot-unplugged and then hot-plugged,
the system is detecting an inconsistency in the sysfs layout and a
BUG_ON() is raised:

  kernel BUG at /Users/laurent/src/linux-ppc/mm/memory_hotplug.c:1084!
  LE PAGE_SIZE=64K MMU=Hash SMP NR_CPUS=2048 NUMA pSeries
  Modules linked in: rpadlpar_io rpaphp pseries_rng rng_core vmx_crypto gf128mul binfmt_misc ip_tables x_tables xfs libcrc32c crc32c_vpmsum autofs4
  CPU: 8 PID: 10256 Comm: drmgr Not tainted 5.9.0-rc1+ #25
  Call Trace:
    add_memory_resource+0x23c/0x340 (unreliable)
    __add_memory+0x5c/0xf0
    dlpar_add_lmb+0x1b4/0x500
    dlpar_memory+0x1f8/0xb80
    handle_dlpar_errorlog+0xc0/0x190
    dlpar_store+0x198/0x4a0
    kobj_attr_store+0x30/0x50
    sysfs_kf_write+0x64/0x90
    kernfs_fop_write+0x1b0/0x290
    vfs_write+0xe8/0x290
    ksys_write+0xdc/0x130
    system_call_exception+0x160/0x270
    system_call_common+0xf0/0x27c

This has been seen on PowerPC LPAR.

The root cause of this issue is that when node's memory is registered,
the range used can overlap another node's range, thus the memory block
is registered to multiple nodes in sysfs.

There are two issues here:

 (a) The sysfs memory and node's layouts are broken due to these
     multiple links

 (b) The link errors in link_mem_sections() should not lead to a system
     panic.

To address (a) register_mem_sect_under_node should not rely on the
system state to detect whether the link operation is triggered by a hot
plug operation or not.  This is addressed by the patches 1 and 2 of this
series.

Issue (b) will be addressed separately.

This patch (of 2):

The memmap_context enum is used to detect whether a memory operation is
due to a hot-add operation or happening at boot time.

Make it general to the hotplug operation and rename it as
meminit_context.

There is no functional change introduced by this patch

Suggested-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Laurent Dufour <ldufour@linux.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: "Rafael J . Wysocki" <rafael@kernel.org>
Cc: Nathan Lynch <nathanl@linux.ibm.com>
Cc: Scott Cheloha <cheloha@linux.ibm.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: <stable@vger.kernel.org>
Link: https://lkml.kernel.org/r/20200915094143.79181-1-ldufour@linux.ibm.com
Link: https://lkml.kernel.org/r/20200915132624.9723-1-ldufour@linux.ibm.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agoarch/x86/lib/usercopy_64.c: fix __copy_user_flushcache() cache writeback
Mikulas Patocka [Sat, 26 Sep 2020 04:19:24 +0000 (21:19 -0700)]
arch/x86/lib/usercopy_64.c: fix __copy_user_flushcache() cache writeback

If we copy less than 8 bytes and if the destination crosses a cache
line, __copy_user_flushcache would invalidate only the first cache line.

This patch makes it invalidate the second cache line as well.

Fixes: 0aed55af88345b ("x86, uaccess: introduce copy_from_iter_flushcache for pmem / cache-bypass operations")
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Dan Williams <dan.j.wiilliams@intel.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Toshi Kani <toshi.kani@hpe.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: <stable@vger.kernel.org>
Link: https://lkml.kernel.org/r/alpine.LRH.2.02.2009161451140.21915@file01.intranet.prod.int.rdu2.redhat.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agolib/memregion.c: include memregion.h
Jason Yan [Sat, 26 Sep 2020 04:19:21 +0000 (21:19 -0700)]
lib/memregion.c: include memregion.h

This addresses the following sparse warning:

  lib/memregion.c:8:5: warning: symbol 'memregion_alloc' was not declared. Should it be static?
  lib/memregion.c:14:6: warning: symbol 'memregion_free' was not declared. Should it be static?

Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: Jason Yan <yanaijie@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Link: https://lkml.kernel.org/r/20200921142852.875312-1-yanaijie@huawei.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agolib/string.c: implement stpcpy
Nick Desaulniers [Sat, 26 Sep 2020 04:19:18 +0000 (21:19 -0700)]
lib/string.c: implement stpcpy

LLVM implemented a recent "libcall optimization" that lowers calls to
`sprintf(dest, "%s", str)` where the return value is used to
`stpcpy(dest, str) - dest`.

This generally avoids the machinery involved in parsing format strings.
`stpcpy` is just like `strcpy` except it returns the pointer to the new
tail of `dest`.  This optimization was introduced into clang-12.

Implement this so that we don't observe linkage failures due to missing
symbol definitions for `stpcpy`.

Similar to last year's fire drill with: commit 5f074f3e192f
("lib/string.c: implement a basic bcmp")

The kernel is somewhere between a "freestanding" environment (no full
libc) and "hosted" environment (many symbols from libc exist with the
same type, function signature, and semantics).

As Peter Anvin notes, there's not really a great way to inform the
compiler that you're targeting a freestanding environment but would like
to opt-in to some libcall optimizations (see pr/47280 below), rather
than opt-out.

Arvind notes, -fno-builtin-* behaves slightly differently between GCC
and Clang, and Clang is missing many __builtin_* definitions, which I
consider a bug in Clang and am working on fixing.

Masahiro summarizes the subtle distinction between compilers justly:
  To prevent transformation from foo() into bar(), there are two ways in
  Clang to do that; -fno-builtin-foo, and -fno-builtin-bar.  There is
  only one in GCC; -fno-buitin-foo.

(Any difference in that behavior in Clang is likely a bug from a missing
__builtin_* definition.)

Masahiro also notes:
  We want to disable optimization from foo() to bar(),
  but we may still benefit from the optimization from
  foo() into something else. If GCC implements the same transform, we
  would run into a problem because it is not -fno-builtin-bar, but
  -fno-builtin-foo that disables that optimization.

  In this regard, -fno-builtin-foo would be more future-proof than
  -fno-built-bar, but -fno-builtin-foo is still potentially overkill. We
  may want to prevent calls from foo() being optimized into calls to
  bar(), but we still may want other optimization on calls to foo().

It seems that compilers today don't quite provide the fine grain control
over which libcall optimizations pseudo-freestanding environments would
prefer.

Finally, Kees notes that this interface is unsafe, so we should not
encourage its use.  As such, I've removed the declaration from any
header, but it still needs to be exported to avoid linkage errors in
modules.

Reported-by: Sami Tolvanen <samitolvanen@google.com>
Suggested-by: Andy Lavr <andy.lavr@gmail.com>
Suggested-by: Arvind Sankar <nivedita@alum.mit.edu>
Suggested-by: Joe Perches <joe@perches.com>
Suggested-by: Kees Cook <keescook@chromium.org>
Suggested-by: Masahiro Yamada <masahiroy@kernel.org>
Suggested-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Signed-off-by: Nick Desaulniers <ndesaulniers@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Nathan Chancellor <natechancellor@gmail.com>
Cc: <stable@vger.kernel.org>
Link: https://lkml.kernel.org/r/20200914161643.938408-1-ndesaulniers@google.com
Link: https://bugs.llvm.org/show_bug.cgi?id=47162
Link: https://bugs.llvm.org/show_bug.cgi?id=47280
Link: https://github.com/ClangBuiltLinux/linux/issues/1126
Link: https://man7.org/linux/man-pages/man3/stpcpy.3.html
Link: https://pubs.opengroup.org/onlinepubs/9699919799/functions/stpcpy.html
Link: https://reviews.llvm.org/D85963
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agomm/migrate: correct thp migration stats
Zi Yan [Sat, 26 Sep 2020 04:19:14 +0000 (21:19 -0700)]
mm/migrate: correct thp migration stats

PageTransHuge returns true for both thp and hugetlb, so thp stats was
counting both thp and hugetlb migrations.  Exclude hugetlb migration by
setting is_thp variable right.

Clean up thp handling code too when we are there.

Fixes: 1a5bae25e3cf ("mm/vmstat: add events for THP migration without split")
Signed-off-by: Zi Yan <ziy@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Daniel Jordan <daniel.m.jordan@oracle.com>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Link: https://lkml.kernel.org/r/20200917210413.1462975-1-zi.yan@sent.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agomm/gup: fix gup_fast with dynamic page table folding
Vasily Gorbik [Sat, 26 Sep 2020 04:19:10 +0000 (21:19 -0700)]
mm/gup: fix gup_fast with dynamic page table folding

Currently to make sure that every page table entry is read just once
gup_fast walks perform READ_ONCE and pass pXd value down to the next
gup_pXd_range function by value e.g.:

  static int gup_pud_range(p4d_t p4d, unsigned long addr, unsigned long end,
                           unsigned int flags, struct page **pages, int *nr)
  ...
          pudp = pud_offset(&p4d, addr);

This function passes a reference on that local value copy to pXd_offset,
and might get the very same pointer in return.  This happens when the
level is folded (on most arches), and that pointer should not be
iterated.

On s390 due to the fact that each task might have different 5,4 or
3-level address translation and hence different levels folded the logic
is more complex and non-iteratable pointer to a local copy leads to
severe problems.

Here is an example of what happens with gup_fast on s390, for a task
with 3-level paging, crossing a 2 GB pud boundary:

  // addr = 0x1007ffff000, end = 0x10080001000
  static int gup_pud_range(p4d_t p4d, unsigned long addr, unsigned long end,
                           unsigned int flags, struct page **pages, int *nr)
  {
        unsigned long next;
        pud_t *pudp;

        // pud_offset returns &p4d itself (a pointer to a value on stack)
        pudp = pud_offset(&p4d, addr);
        do {
                // on second iteratation reading "random" stack value
                pud_t pud = READ_ONCE(*pudp);

                // next = 0x10080000000, due to PUD_SIZE/MASK != PGDIR_SIZE/MASK on s390
                next = pud_addr_end(addr, end);
                ...
        } while (pudp++, addr = next, addr != end); // pudp++ iterating over stack

        return 1;
  }

This happens since s390 moved to common gup code with commit
d1874a0c2805 ("s390/mm: make the pxd_offset functions more robust") and
commit 1a42010cdc26 ("s390/mm: convert to the generic
get_user_pages_fast code").

s390 tried to mimic static level folding by changing pXd_offset
primitives to always calculate top level page table offset in pgd_offset
and just return the value passed when pXd_offset has to act as folded.

What is crucial for gup_fast and what has been overlooked is that
PxD_SIZE/MASK and thus pXd_addr_end should also change correspondingly.
And the latter is not possible with dynamic folding.

To fix the issue in addition to pXd values pass original pXdp pointers
down to gup_pXd_range functions.  And introduce pXd_offset_lockless
helpers, which take an additional pXd entry value parameter.  This has
already been discussed in

  https://lkml.kernel.org/r/20190418100218.0a4afd51@mschwideX1

Fixes: 1a42010cdc26 ("s390/mm: convert to the generic get_user_pages_fast code")
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Reviewed-by: Alexander Gordeev <agordeev@linux.ibm.com>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Reviewed-by: Mike Rapoport <rppt@linux.ibm.com>
Reviewed-by: John Hubbard <jhubbard@nvidia.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Russell King <linux@armlinux.org.uk>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will@kernel.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Jeff Dike <jdike@addtoit.com>
Cc: Richard Weinberger <richard@nod.at>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Cc: Claudio Imbrenda <imbrenda@linux.ibm.com>
Cc: <stable@vger.kernel.org> [5.2+]
Link: https://lkml.kernel.org/r/patch.git-943f1e5dcff2.your-ad-here.call-01599856292-ext-8676@work.hours
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agomm: memcontrol: fix missing suffix of workingset_restore
Muchun Song [Sat, 26 Sep 2020 04:19:05 +0000 (21:19 -0700)]
mm: memcontrol: fix missing suffix of workingset_restore

We forget to add the suffix to the workingset_restore string, so fix it.

And also update the documentation of cgroup-v2.rst.

Fixes: 170b04b7ae49 ("mm/workingset: prepare the workingset detection infrastructure for anon LRU")
Signed-off-by: Muchun Song <songmuchun@bytedance.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Tejun Heo <tj@kernel.org>
Cc: Zefan Li <lizefan@huawei.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Randy Dunlap <rdunlap@infradead.org>
Link: https://lkml.kernel.org/r/20200916100030.71698-1-songmuchun@bytedance.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agomm, THP, swap: fix allocating cluster for swapfile by mistake
Gao Xiang [Sat, 26 Sep 2020 04:19:01 +0000 (21:19 -0700)]
mm, THP, swap: fix allocating cluster for swapfile by mistake

SWP_FS is used to make swap_{read,write}page() go through the
filesystem, and it's only used for swap files over NFS.  So, !SWP_FS
means non NFS for now, it could be either file backed or device backed.
Something similar goes with legacy SWP_FILE.

So in order to achieve the goal of the original patch, SWP_BLKDEV should
be used instead.

FS corruption can be observed with SSD device + XFS + fragmented
swapfile due to CONFIG_THP_SWAP=y.

I reproduced the issue with the following details:

Environment:

  QEMU + upstream kernel + buildroot + NVMe (2 GB)

Kernel config:

  CONFIG_BLK_DEV_NVME=y
  CONFIG_THP_SWAP=y

Some reproducible steps:

  mkfs.xfs -f /dev/nvme0n1
  mkdir /tmp/mnt
  mount /dev/nvme0n1 /tmp/mnt
  bs="32k"
  sz="1024m"    # doesn't matter too much, I also tried 16m
  xfs_io -f -c "pwrite -R -b $bs 0 $sz" -c "fdatasync" /tmp/mnt/sw
  xfs_io -f -c "pwrite -R -b $bs 0 $sz" -c "fdatasync" /tmp/mnt/sw
  xfs_io -f -c "pwrite -R -b $bs 0 $sz" -c "fdatasync" /tmp/mnt/sw
  xfs_io -f -c "pwrite -F -S 0 -b $bs 0 $sz" -c "fdatasync" /tmp/mnt/sw
  xfs_io -f -c "pwrite -R -b $bs 0 $sz" -c "fsync" /tmp/mnt/sw

  mkswap /tmp/mnt/sw
  swapon /tmp/mnt/sw

  stress --vm 2 --vm-bytes 600M   # doesn't matter too much as well

Symptoms:
 - FS corruption (e.g. checksum failure)
 - memory corruption at: 0xd2808010
 - segfault

Fixes: f0eea189e8e9 ("mm, THP, swap: Don't allocate huge cluster for file backed swap device")
Fixes: 38d8b4e6bdc8 ("mm, THP, swap: delay splitting THP during swap out")
Signed-off-by: Gao Xiang <hsiangkao@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: "Huang, Ying" <ying.huang@intel.com>
Reviewed-by: Yang Shi <shy828301@gmail.com>
Acked-by: Rafael Aquini <aquini@redhat.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Carlos Maiolino <cmaiolino@redhat.com>
Cc: Eric Sandeen <esandeen@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: <stable@vger.kernel.org>
Link: https://lkml.kernel.org/r/20200820045323.7809-1-hsiangkao@redhat.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agomm: slab: fix potential double free in ___cache_free
Shakeel Butt [Sat, 26 Sep 2020 14:13:41 +0000 (07:13 -0700)]
mm: slab: fix potential double free in ___cache_free

With the commit 10befea91b61 ("mm: memcg/slab: use a single set of
kmem_caches for all allocations"), it becomes possible to call kfree()
from the slabs_destroy().

The functions cache_flusharray() and do_drain() calls slabs_destroy() on
array_cache of the local CPU without updating the size of the
array_cache.  This enables the kfree() call from the slabs_destroy() to
recursively call cache_flusharray() which can potentially call
free_block() on the same elements of the array_cache of the local CPU
and causing double free and memory corruption.

To fix the issue, simply update the local CPU array_cache cache before
calling slabs_destroy().

Fixes: 10befea91b61 ("mm: memcg/slab: use a single set of kmem_caches for all allocations")
Signed-off-by: Shakeel Butt <shakeelb@google.com>
Reviewed-by: Roman Gushchin <guro@fb.com>
Tested-by: Ming Lei <ming.lei@redhat.com>
Reported-by: kernel test robot <rong.a.chen@intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Ted Ts'o <tytso@mit.edu>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agoDocumentation/llvm: Fix clang target examples
Florian Fainelli [Fri, 25 Sep 2020 15:21:14 +0000 (08:21 -0700)]
Documentation/llvm: Fix clang target examples

clang --target=<triple> is how we can specify a particular toolchain
triple to be use, fix the two occurences in the documentation.

Fixes: fcf1b6a35c16 ("Documentation/llvm: add documentation on building w/ Clang/LLVM")
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
Reviewed-by: Nathan Chancellor <natechancellor@gmail.com>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
4 years agoMerge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
Linus Torvalds [Sat, 26 Sep 2020 00:15:19 +0000 (17:15 -0700)]
Merge tag 'for-linus' of git://git./virt/kvm/kvm

Pull more kvm fixes from Paolo Bonzini:
 "Five small fixes.

  The nested migration bug will be fixed with a better API in 5.10 or
  5.11, for now this is a fix that works with existing userspace but
  keeps the current ugly API"

* tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm:
  KVM: SVM: Add a dedicated INVD intercept routine
  KVM: x86: Reset MMU context if guest toggles CR4.SMAP or CR4.PKE
  KVM: x86: fix MSR_IA32_TSC read for nested migration
  selftests: kvm: Fix assert failure in single-step test
  KVM: x86: VMX: Make smaller physical guest address space support user-configurable

4 years agoMerge tag 'mips_fixes_5.9_3' of git://git.kernel.org/pub/scm/linux/kernel/git/mips...
Linus Torvalds [Fri, 25 Sep 2020 22:24:04 +0000 (15:24 -0700)]
Merge tag 'mips_fixes_5.9_3' of git://git./linux/kernel/git/mips/linux

Pull MIPS fixes from Thomas Bogendoerfer:

 - fixed FP register access on Loongsoon-3

 - added missing 1074 cpu handling

 - fixed Loongson2ef build error

* tag 'mips_fixes_5.9_3' of git://git.kernel.org/pub/scm/linux/kernel/git/mips/linux:
  MIPS: BCM47XX: Remove the needless check with the 1074K
  MIPS: Add the missing 'CPU_1074K' into __get_cpu_type()
  MIPS: Loongson2ef: Disable Loongson MMI instructions
  MIPS: Loongson-3: Fix fp register access if MSA enabled

4 years agoMerge tag 'spi-fix-v5.9-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie/spi
Linus Torvalds [Fri, 25 Sep 2020 22:21:54 +0000 (15:21 -0700)]
Merge tag 'spi-fix-v5.9-rc6' of git://git./linux/kernel/git/broonie/spi

Pull spi fixes from Mark Brown:
 "A small collection of driver specific fixes, the fsl-espi and bcm-qspi
  changes in particular have been causing breakage for users"

* tag 'spi-fix-v5.9-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie/spi:
  spi: bcm-qspi: Fix probe regression on iProc platforms
  spi: fsl-dspi: fix use-after-free in remove path
  spi: fsl-espi: Only process interrupts for expected events
  spi: bcm2835: Make polling_limit_us static
  spi: spi-fsl-dspi: use XSPI mode instead of DMA for DPAA2 SoCs

4 years agoMerge tag 'regulator-fix-v5.9-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Fri, 25 Sep 2020 22:16:01 +0000 (15:16 -0700)]
Merge tag 'regulator-fix-v5.9-rc6' of git://git./linux/kernel/git/broonie/regulator

Pull regulator fix from Mark Brown:
 "A single fix for incorrect specification of some of the register
  fields on axp20x devices which would break voltage setting on affected
  systems"

* tag 'regulator-fix-v5.9-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie/regulator:
  regulator: axp20x: fix LDO2/4 description

4 years agoMerge tag 'regmap-fix-v5.9-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Fri, 25 Sep 2020 22:11:24 +0000 (15:11 -0700)]
Merge tag 'regmap-fix-v5.9-rc6' of git://git./linux/kernel/git/broonie/regmap

Pull regmap fixes from Mark Brown:
 "Two issues here - one is a fix for use after free issues in the case
  where a regmap overrides its name using something dynamically
  generated, the other is that we weren't handling access checks
  non-incrementing I/O on registers within paged register regions
  correctly resulting in spurious errors.

  Both of these are quite rare but serious if they occur"

* tag 'regmap-fix-v5.9-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie/regmap:
  regmap: fix page selection for noinc writes
  regmap: fix page selection for noinc reads
  regmap: debugfs: Add back in erroneously removed initialisation of ret
  regmap: debugfs: Fix handling of name string for debugfs init delays

4 years agoio_uring: ensure async buffered read-retry is setup properly
Jens Axboe [Fri, 25 Sep 2020 21:23:43 +0000 (15:23 -0600)]
io_uring: ensure async buffered read-retry is setup properly

A previous commit for fixing up short reads botched the async retry
path, so we ended up going to worker threads more often than we should.
Fix this up, so retries work the way they originally were intended to.

Fixes: 227c0c9673d8 ("io_uring: internally retry short reads")
Reported-by: Hao_Xu <haoxu@linux.alibaba.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoMerge tag 'nfsd-5.9-2' of git://git.linux-nfs.org/projects/cel/cel-2.6
Linus Torvalds [Fri, 25 Sep 2020 17:46:11 +0000 (10:46 -0700)]
Merge tag 'nfsd-5.9-2' of git://git.linux-nfs.org/projects/cel/cel-2.6

Pull NFS server fix from Chuck Lever:
 "Fix incorrect calculation on platforms that implement
  flush_dcache_page()"

* tag 'nfsd-5.9-2' of git://git.linux-nfs.org/projects/cel/cel-2.6:
  SUNRPC: Fix svc_flush_dcache()

4 years agoMerge tag 'pm-5.9-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm
Linus Torvalds [Fri, 25 Sep 2020 17:39:22 +0000 (10:39 -0700)]
Merge tag 'pm-5.9-rc7' of git://git./linux/kernel/git/rafael/linux-pm

Pull power management fixes from Rafael Wysocki:
 "These fix more fallout of recent RCU-lockdep changes in CPU idle code
  and two devfreq issues.

  Specifics:

   - Export rcu_idle_{enter,exit} to modules to fix build issues
     introduced by recent RCU-lockdep fixes (Borislav Petkov)

   - Add missing return statement to a stub function in the ACPI
     processor driver to fix a build issue introduced by recent
     RCU-lockdep fixes (Rafael Wysocki)

   - Fix recently introduced suspicious RCU usage warnings in the PSCI
     cpuidle driver and drop stale comments regarding RCU_NONIDLE()
     usage from enter_s2idle_proper() (Ulf Hansson)

   - Fix error code path in the tegra30 devfreq driver (Dan Carpenter)

   - Add missing information to devfreq_summary debugfs (Chanwoo Choi)"

* tag 'pm-5.9-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  ACPI: processor: Fix build for ARCH_APICTIMER_STOPS_ON_C3 unset
  PM / devfreq: tegra30: Disable clock on error in probe
  PM / devfreq: Add timer type to devfreq_summary debugfs
  cpuidle: Drop misleading comments about RCU usage
  cpuidle: psci: Fix suspicious RCU usage
  rcu/tree: Export rcu_idle_{enter,exit} to modules