linux.git
5 years agoceph: fix snapshot directory timestamps
Luis Henriques [Wed, 11 Mar 2020 11:05:38 +0000 (11:05 +0000)]
ceph: fix snapshot directory timestamps

The .snap directory timestamps are kept at 0 (1970-01-01 00:00), which
isn't consistent with what the fuse client does.  This patch makes the
behaviour consistent, by setting these timestamps (atime, btime, ctime,
mtime) to those of the parent directory.

Cc: Marc Roos <M.Roos@f1-outsourcing.eu>
Signed-off-by: Luis Henriques <lhenriques@suse.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: wait for async creating inode before requesting new max size
Yan, Zheng [Tue, 10 Mar 2020 11:34:21 +0000 (19:34 +0800)]
ceph: wait for async creating inode before requesting new max size

ceph_check_caps() can't request new max size for async creating inode.
This may make ceph_get_caps() loop busily until getting reply of the
async create. Also, wait for async creating reply before calling
ceph_renew_caps().

Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: don't skip updating wanted caps when cap is stale
Yan, Zheng [Tue, 10 Mar 2020 11:34:20 +0000 (19:34 +0800)]
ceph: don't skip updating wanted caps when cap is stale

1. try_get_cap_refs() fails to get caps and finds that mds_wanted
   does not include what it wants. It returns -ESTALE.
2. ceph_get_caps() calls ceph_renew_caps(). ceph_renew_caps() finds
   that inode has cap, so it calls ceph_check_caps().
3. ceph_check_caps() finds that issued caps (without checking if it's
   stale) already includes caps wanted by open file, so it skips
   updating wanted caps.

Above events can cause an infinite loop inside ceph_get_caps().

Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: request new max size only when there is auth cap
Yan, Zheng [Tue, 10 Mar 2020 11:34:19 +0000 (19:34 +0800)]
ceph: request new max size only when there is auth cap

When there is no auth cap, check_max_size() can't do anything and may
cause an infinite loop inside ceph_get_caps().

Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: cleanup return error of try_get_cap_refs()
Yan, Zheng [Tue, 10 Mar 2020 11:34:18 +0000 (19:34 +0800)]
ceph: cleanup return error of try_get_cap_refs()

Returns 0 if caps were not able to be acquired (yet), 1 if cap
acquisition succeeded, or a negative error code. There are 3 special
error codes:

-EAGAIN: need to sleep but non-blocking is specified
-EFBIG:  ask caller to call check_max_size() and try again.
-ESTALE: ask caller to call ceph_renew_caps() and try again.

[ jlayton: add WARN_ON_ONCE check for -EAGAIN ]

Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: return ceph_mdsc_do_request() errors from __get_parent()
Qiujun Huang [Fri, 6 Mar 2020 01:34:20 +0000 (09:34 +0800)]
ceph: return ceph_mdsc_do_request() errors from __get_parent()

Return the error returned by ceph_mdsc_do_request(). Otherwise,
r_target_inode ends up being NULL this ends up returning ENOENT
regardless of the error.

Signed-off-by: Qiujun Huang <hqjagain@gmail.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: check all mds' caps after page writeback
Yan, Zheng [Thu, 5 Mar 2020 12:21:04 +0000 (20:21 +0800)]
ceph: check all mds' caps after page writeback

If an inode has caps from multiple mds's, the following can happen:

- non-auth mds revokes Fsc. Fcb is used, so page writeback is queued.
- when writeback finishes, ceph_check_caps() is called with auth only
  flag. ceph_check_caps() invalidates pagecache, but skips checking any
  non-auth caps.

Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: update i_requested_max_size only when sending cap msg to auth mds
Yan, Zheng [Thu, 5 Mar 2020 12:21:03 +0000 (20:21 +0800)]
ceph: update i_requested_max_size only when sending cap msg to auth mds

Non-auth mds can't do anything to 'update max' cap message.

Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: simplify calling of ceph_get_fmode()
Yan, Zheng [Thu, 5 Mar 2020 12:21:02 +0000 (20:21 +0800)]
ceph: simplify calling of ceph_get_fmode()

Originally, calling ceph_get_fmode() for open files is by thread that
handles request reply. There is a small window between updating caps and
and waking the request initiator. We need to prevent ceph_check_caps()
from releasing wanted caps in the window.

Previous patches made fill_inode() call __ceph_touch_fmode() for open file
requests. This prevented ceph_check_caps() from releasing wanted caps for
'caps_wanted_delay_min' seconds, enough for request initiator to get
woken up and call ceph_get_fmode().

This allows us to now call ceph_get_fmode() in ceph_open() instead.

Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: remove delay check logic from ceph_check_caps()
Yan, Zheng [Thu, 5 Mar 2020 12:21:01 +0000 (20:21 +0800)]
ceph: remove delay check logic from ceph_check_caps()

__ceph_caps_file_wanted() already checks 'caps_wanted_delay_min' and
'caps_wanted_delay_max'. There is no need to duplicate the logic in
ceph_check_caps() and __send_cap()

Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: consider inode's last read/write when calculating wanted caps
Yan, Zheng [Thu, 5 Mar 2020 12:21:00 +0000 (20:21 +0800)]
ceph: consider inode's last read/write when calculating wanted caps

Add i_last_rd and i_last_wr to ceph_inode_info. These fields are
used to track the last time the client acquired read/write caps for
the inode.

If there is no read/write on an inode for 'caps_wanted_delay_max'
seconds, __ceph_caps_file_wanted() does not request caps for read/write
even there are open files.

Call __ceph_touch_fmode() for dir operations. __ceph_caps_file_wanted()
calculates dir's wanted caps according to last dir read/modification. If
there is recent dir read, dir inode wants CEPH_CAP_ANY_SHARED caps. If
there is recent dir modification, also wants CEPH_CAP_FILE_EXCL.

Readdir is a special case. Dir inode wants CEPH_CAP_FILE_EXCL after
readdir, as with that, modifications do not need to release
CEPH_CAP_FILE_SHARED or invalidate all dentry leases issued by readdir.

Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: always renew caps if mds_wanted is insufficient
Yan, Zheng [Thu, 5 Mar 2020 12:20:59 +0000 (20:20 +0800)]
ceph: always renew caps if mds_wanted is insufficient

Original code only renews caps for inodes with CEPH_I_CAP_DROPPED flag,
which indicates that mds has closed the session and caps were dropped.
Remove this flag in preparation for not requesting caps for idle open
files.

Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: update dentry lease for async create
Yan, Zheng [Wed, 4 Mar 2020 13:22:20 +0000 (21:22 +0800)]
ceph: update dentry lease for async create

Otherwise ceph_d_delete() may return 1 for the dentry, which makes
dput() prune the dentry and clear parent dir's complete flag.

Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: attempt to do async create when possible
Jeff Layton [Wed, 27 Nov 2019 17:06:14 +0000 (12:06 -0500)]
ceph: attempt to do async create when possible

With the Octopus release, the MDS will hand out directory create caps.

If we have Fxc caps on the directory, and complete directory information
or a known negative dentry, then we can return without waiting on the
reply, allowing the open() call to return very quickly to userland.

We use the normal ceph_fill_inode() routine to fill in the inode, so we
have to gin up some reply inode information with what we'd expect the
newly-created inode to have. The client assumes that it has a full set
of caps on the new inode, and that the MDS will revoke them when there
is conflicting access.

This functionality is gated on the wsync/nowsync mount options.

Signed-off-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: "Yan, Zheng" <zyan@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: cache layout in parent dir on first sync create
Jeff Layton [Thu, 2 Jan 2020 12:11:38 +0000 (07:11 -0500)]
ceph: cache layout in parent dir on first sync create

If a create is done, then typically we'll end up writing to the file
soon afterward. We don't want to wait for the reply before doing that
when doing an async create, so that means we need the layout for the
new file before we've gotten the response from the MDS.

All files created in a directory will initially inherit the same layout,
so copy off the requisite info from the first synchronous create in the
directory, and save it in a new i_cached_layout field. Zero out the
layout when we lose Dc caps in the dir.

Signed-off-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: "Yan, Zheng" <zyan@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: add new MDS req field to hold delegated inode number
Jeff Layton [Mon, 13 Jan 2020 18:04:08 +0000 (13:04 -0500)]
ceph: add new MDS req field to hold delegated inode number

Add new request field to hold the delegated inode number. Encode that
into the message when it's set.

Signed-off-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: "Yan, Zheng" <zyan@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: decode interval_sets for delegated inos
Jeff Layton [Fri, 15 Nov 2019 16:51:55 +0000 (11:51 -0500)]
ceph: decode interval_sets for delegated inos

Starting in Octopus, the MDS will hand out caps that allow the client
to do asynchronous file creates under certain conditions. As part of
that, the MDS will delegate ranges of inode numbers to the client.

Add the infrastructure to decode these ranges, and stuff them into an
xarray for later consumption by the async creation code.

Because the xarray code currently only handles unsigned long indexes,
and those are 32-bits on 32-bit arches, we only enable the decoding when
running on a 64-bit arch.

Signed-off-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: "Yan, Zheng" <zyan@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: make ceph_fill_inode non-static
Jeff Layton [Thu, 5 Dec 2019 14:09:25 +0000 (09:09 -0500)]
ceph: make ceph_fill_inode non-static

Signed-off-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: "Yan, Zheng" <zyan@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: perform asynchronous unlink if we have sufficient caps
Jeff Layton [Tue, 2 Apr 2019 19:35:56 +0000 (15:35 -0400)]
ceph: perform asynchronous unlink if we have sufficient caps

The MDS is getting a new lock-caching facility that will allow it
to cache the necessary locks to allow asynchronous directory operations.
Since the CEPH_CAP_FILE_* caps are currently unused on directories,
we can repurpose those bits for this purpose.

When performing an unlink, if we have Fx on the parent directory,
and CEPH_CAP_DIR_UNLINK (aka Fr), and we know that the dentry being
removed is the primary link, then then we can fire off an unlink
request immediately and don't need to wait on reply before returning.

In that situation, just fix up the dcache and link count and return
immediately after issuing the call to the MDS. This does mean that we
need to hold an extra reference to the inode being unlinked, and extra
references to the caps to avoid races. Those references are put and
error handling is done in the r_callback routine.

If the operation ends up failing, then set a writeback error on the
directory inode, and the inode itself that can be fetched later by
an fsync on the dir.

The behavior of dir caps is slightly different from caps on normal
files. Because these are just considered an optimization, if the
session is reconnected, we will not automatically reclaim them. They
are instead considered lost until we do another synchronous op in the
parent directory.

Async dirops are enabled via the "nowsync" mount option, which is
patterned after the xfs "wsync" mount option. For now, the default
is "wsync", but eventually we may flip that.

Signed-off-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: "Yan, Zheng" <zyan@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: don't take refs to want mask unless we have all bits
Yan, Zheng [Tue, 18 Feb 2020 13:17:08 +0000 (08:17 -0500)]
ceph: don't take refs to want mask unless we have all bits

If we don't have all of the cap bits for the want mask in
try_get_cap_refs, then just take refs on the need bits.

Signed-off-by: "Yan, Zheng" <ukernel@gmail.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: cap tracking for async directory operations
Jeff Layton [Tue, 18 Feb 2020 19:12:45 +0000 (14:12 -0500)]
ceph: cap tracking for async directory operations

Track and correctly handle directory caps for asynchronous operations.
Add aliases for Frc caps that we now designate at Dcu caps (when dealing
with directories).

Unlike file caps, we don't reclaim these when the session goes away, and
instead preemptively release them. In-flight async dirops are instead
handled during reconnect phase. The client needs to re-do a synchronous
operation in order to re-get directory caps.

Signed-off-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: "Yan, Zheng" <zyan@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: make __take_cap_refs non-static
Jeff Layton [Tue, 14 Jan 2020 14:23:49 +0000 (09:23 -0500)]
ceph: make __take_cap_refs non-static

Rename it to ceph_take_cap_refs and make it available to other files.
Also replace a comment with a lockdep assertion.

Signed-off-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: "Yan, Zheng" <zyan@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: add infrastructure for waiting for async create to complete
Jeff Layton [Tue, 14 Jan 2020 20:06:40 +0000 (15:06 -0500)]
ceph: add infrastructure for waiting for async create to complete

When we issue an async create, we must ensure that any later on-the-wire
requests involving it wait for the create reply.

Expand i_ceph_flags to be an unsigned long, and add a new bit that
MDS requests can wait on. If the bit is set in the inode when sending
caps, then don't send it and just return that it has been delayed.

Signed-off-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: "Yan, Zheng" <zyan@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: track primary dentry link
Jeff Layton [Tue, 18 Feb 2020 19:12:32 +0000 (14:12 -0500)]
ceph: track primary dentry link

Newer versions of the MDS will flag a dentry as "primary". In later
patches, we'll need to consult this info, so track it in di->flags.

Signed-off-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: "Yan, Zheng" <zyan@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: add flag to designate that a request is asynchronous
Jeff Layton [Mon, 2 Dec 2019 18:47:57 +0000 (13:47 -0500)]
ceph: add flag to designate that a request is asynchronous

...and ensure that such requests are never queued. The MDS has need to
know that a request is asynchronous so add flags and proper
infrastructure for that.

Also, delegated inode numbers and directory caps are associated with the
session, so ensure that async requests are always transmitted on the
first attempt and are never queued to wait for session reestablishment.

If it does end up looking like we'll need to queue the request, then
have it return -EJUKEBOX so the caller can reattempt with a synchronous
request.

Signed-off-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: "Yan, Zheng" <zyan@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: more caps.c lockdep assertions
Jeff Layton [Tue, 25 Feb 2020 19:08:33 +0000 (11:08 -0800)]
ceph: more caps.c lockdep assertions

Signed-off-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: "Yan, Zheng" <zyan@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: clean up kick_flushing_inode_caps()
Jeff Layton [Tue, 25 Feb 2020 19:49:53 +0000 (11:49 -0800)]
ceph: clean up kick_flushing_inode_caps()

The last thing that this function does is release i_ceph_lock, so
have the caller do that instead. Add a lockdep assertion to
ensure that the function is always called with i_ceph_lock held.
Change the prototype to take a ceph_inode_info pointer and drop
the separate mdsc argument as we can get that from the session.

While at it, make it non-static.  We'll need this to kick any
flushing caps once the create reply comes in.

Signed-off-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: Ilya Dryomov <idryomov@gmail.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agolibceph: directly skip to the end of redirect reply
Ilya Dryomov [Fri, 30 Aug 2019 15:38:31 +0000 (17:38 +0200)]
libceph: directly skip to the end of redirect reply

Coverity complains about a double write to *p.  Don't bother with
osd_instructions and directly skip to the end of redirect reply.

Reported-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agolibceph: simplify ceph_monc_handle_map()
Ilya Dryomov [Wed, 26 Feb 2020 15:37:55 +0000 (16:37 +0100)]
libceph: simplify ceph_monc_handle_map()

ceph_monc_handle_map() confuses static checkers which report a
false use-after-free on monc->monmap, missing that monc->monmap and
client->monc.monmap is the same pointer.

Use monc->monmap consistently and get rid of "old", which is redundant.

Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: return ETIMEDOUT errno to userland when request timed out
Xiubo Li [Mon, 24 Feb 2020 03:23:11 +0000 (22:23 -0500)]
ceph: return ETIMEDOUT errno to userland when request timed out

req->r_timeout is only used during mounting, so this error will
be more accurate.

URL: https://tracker.ceph.com/issues/44215
Signed-off-by: Xiubo Li <xiubli@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: re-org copy_file_range and fix some error paths
Luis Henriques [Mon, 24 Feb 2020 13:44:32 +0000 (13:44 +0000)]
ceph: re-org copy_file_range and fix some error paths

This patch re-organizes copy_file_range, trying to fix a few issues in the
error handling.  Here's the summary:

- Abort copy if initial do_splice_direct() returns fewer bytes than
  requested.

- Move the 'size' initialization (with i_size_read()) further down in the
  code, after the initial call to do_splice_direct().  This avoids issues
  with a possibly stale value if a manual copy is done.

- Move the object copy loop into a separate function.  This makes it
  easier to handle errors (e.g, dirtying caps and updating the MDS
  metadata if only some objects have been copied before an error has
  occurred).

- Added calls to ceph_oloc_destroy() to avoid leaking memory with src_oloc
  and dst_oloc

- After the object copy loop, the new file size to be reported to the MDS
  (if there's file size change) is now the actual file size, and not the
  size after an eventual extra manual copy.

- Added a few dout() to show the number of bytes copied in the two manual
  copies and in the object copy loop.

Signed-off-by: Luis Henriques <lhenriques@suse.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: move to a dedicated slabcache for mds requests
Jeff Layton [Mon, 17 Feb 2020 23:38:37 +0000 (18:38 -0500)]
ceph: move to a dedicated slabcache for mds requests

On my machine (x86_64) this struct is 952 bytes, which gets rounded up
to 1024 by kmalloc. Move this to a dedicated slabcache, so we can
allocate them without the extra 72 bytes of overhead per.

Signed-off-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: Ilya Dryomov <idryomov@gmail.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: reorganize fields in ceph_mds_request
Jeff Layton [Mon, 17 Feb 2020 15:19:14 +0000 (10:19 -0500)]
ceph: reorganize fields in ceph_mds_request

This shrinks the struct size by 16 bytes.

Signed-off-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: switch to page_mkwrite_check_truncate in ceph_page_mkwrite
Andreas Gruenbacher [Thu, 13 Feb 2020 20:24:22 +0000 (21:24 +0100)]
ceph: switch to page_mkwrite_check_truncate in ceph_page_mkwrite

Use the "page has been truncated" logic in page_mkwrite_check_truncate
instead of reimplementing it here.  Other than with the existing code,
fail with -EFAULT / VM_FAULT_NOPAGE when page_offset(page) == size here
as well, as should be expected.

Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: replace zero-length array with flexible-array member
Gustavo A. R. Silva [Thu, 13 Feb 2020 16:00:04 +0000 (10:00 -0600)]
ceph: replace zero-length array with flexible-array member

The current codebase makes use of the zero-length array language
extension to the C90 standard, but the preferred mechanism to declare
variable-length types such as these ones is a flexible array member[1][2],
introduced in C99:

struct foo {
        int stuff;
        struct boo array[];
};

By making use of the mechanism above, we will get a compiler warning
in case the flexible array does not occur last in the structure, which
will help us prevent some kind of undefined behavior bugs from being
inadvertently introduced[3] to the codebase from now on.

Also, notice that, dynamic memory allocations won't be affected by
this change:

"Flexible array members have incomplete type, and so the sizeof operator
may not be applied. As a quirk of the original implementation of
zero-length arrays, sizeof evaluates to zero."[1]

This issue was found with the help of Coccinelle.

[1] https://gcc.gnu.org/onlinedocs/gcc/Zero-Length.html
[2] https://github.com/KSPP/linux/issues/21
[3] commit 76497732932f ("cxgb3/l2t: Fix undefined behaviour")

Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agorbd: enable multiple blk-mq queues
Hannes Reinecke [Fri, 31 Jan 2020 10:37:39 +0000 (11:37 +0100)]
rbd: enable multiple blk-mq queues

Allocate one queue per CPU and get a performance boost from
higher parallelism.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Ilya Dryomov <idryomov@gmail.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agorbd: embed image request in blk-mq pdu
Ilya Dryomov [Wed, 12 Feb 2020 14:23:58 +0000 (15:23 +0100)]
rbd: embed image request in blk-mq pdu

Avoid making allocations for !IMG_REQ_CHILD image requests.  Only
IMG_REQ_CHILD image requests need to be freed now.

Move the initial request checks to rbd_queue_rq().  Unfortunately we
can't fill the image request and kick the state machine directly from
rbd_queue_rq() because ->queue_rq() isn't allowed to block.

Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agorbd: acquire header_rwsem just once in rbd_queue_workfn()
Ilya Dryomov [Wed, 12 Feb 2020 14:08:39 +0000 (15:08 +0100)]
rbd: acquire header_rwsem just once in rbd_queue_workfn()

Currently header_rwsem is acquired twice: once in rbd_dev_parent_get()
when the image request is being created and then in rbd_queue_workfn()
to capture mapping_size and snapc.  Introduce rbd_img_capture_header()
and move image request allocation so that header_rwsem can be acquired
just once.

Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agorbd: get rid of img_request_layered_clear()
Ilya Dryomov [Wed, 12 Feb 2020 13:34:03 +0000 (14:34 +0100)]
rbd: get rid of img_request_layered_clear()

No need to clear IMG_REQ_LAYERED before destroying the request.

Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agorbd: kill img_request kref
Hannes Reinecke [Fri, 31 Jan 2020 10:37:36 +0000 (11:37 +0100)]
rbd: kill img_request kref

The reference counter is never increased, so we can as well call
rbd_img_request_destroy() directly and drop the kref.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Ilya Dryomov <idryomov@gmail.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: check if file lock exists before sending unlock request
Yan, Zheng [Tue, 11 Feb 2020 14:54:43 +0000 (22:54 +0800)]
ceph: check if file lock exists before sending unlock request

When a process exits, kernel closes its files. locks_remove_file()
is called to remove file locks on these files. locks_remove_file()
tries unlocking files even there is no file lock.

Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: fix description of some mount options
Xiubo Li [Tue, 11 Feb 2020 15:31:20 +0000 (10:31 -0500)]
ceph: fix description of some mount options

Based on the latest code, the default value for wsize/rsize is
64MB and the default value for the mount_timeout is 60 seconds.

Signed-off-by: Xiubo Li <xiubli@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: move ceph_osdc_{read,write}pages to ceph.ko
Xiubo Li [Wed, 29 Jan 2020 08:27:07 +0000 (03:27 -0500)]
ceph: move ceph_osdc_{read,write}pages to ceph.ko

Since these helpers are only used by ceph.ko, move them there and
rename them with _sync_ qualifiers.

Signed-off-by: Xiubo Li <xiubli@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: don't ClearPageChecked in ceph_invalidatepage()
Jeff Layton [Tue, 4 Feb 2020 14:37:48 +0000 (09:37 -0500)]
ceph: don't ClearPageChecked in ceph_invalidatepage()

CephFS doesn't set this bit to begin with, so there should be no need
to clear it.

Reported-by: David Howells <dhowells@redhat.com>
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: Ilya Dryomov <idryomov@gmail.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agorbd: remove barriers from img_request_layered_{set,clear,test}()
Ilya Dryomov [Thu, 30 Jan 2020 12:54:59 +0000 (13:54 +0100)]
rbd: remove barriers from img_request_layered_{set,clear,test}()

IMG_REQ_LAYERED is set in rbd_img_request_create(), and tested and
cleared in rbd_img_request_destroy() when the image request is about to
be destroyed.  The barriers are unnecessary.

Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agolibceph: drop CEPH_DEFINE_SHOW_FUNC
Ilya Dryomov [Tue, 28 Jan 2020 19:12:22 +0000 (20:12 +0100)]
libceph: drop CEPH_DEFINE_SHOW_FUNC

Although CEPH_DEFINE_SHOW_FUNC is much older, it now duplicates
DEFINE_SHOW_ATTRIBUTE from linux/seq_file.h.

Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
5 years agoceph: check inode type for CEPH_CAP_FILE_{CACHE,RD,REXTEND,LAZYIO}
Yan, Zheng [Sat, 11 May 2019 09:27:59 +0000 (17:27 +0800)]
ceph: check inode type for CEPH_CAP_FILE_{CACHE,RD,REXTEND,LAZYIO}

These bits will have new meaning for directory inodes.

Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: add refcounting for Fx caps
Jeff Layton [Tue, 2 Apr 2019 12:04:30 +0000 (08:04 -0400)]
ceph: add refcounting for Fx caps

In future patches we'll be taking and relying on Fx caps. Add proper
refcounting for them.

Signed-off-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: "Yan, Zheng" <zyan@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoceph: register MDS request with dir inode from the start
Jeff Layton [Thu, 4 Apr 2019 12:05:38 +0000 (08:05 -0400)]
ceph: register MDS request with dir inode from the start

When the unsafe reply to a request comes in, the request is put on the
r_unsafe_dir inode's list. In future patches, we're going to need to
wait on requests that may not have gotten an unsafe reply yet.

Change __register_request to put the entry on the dir inode's list when
the pointer is set in the request, and don't check the
CEPH_MDS_R_GOT_UNSAFE flag when unregistering it.

The only place that uses this list today is fsync codepath, and with
the coming changes, we'll want to wait on all operations whether it has
gotten an unsafe reply or not.

Signed-off-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: "Yan, Zheng" <zyan@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoLinux 5.6
Linus Torvalds [Sun, 29 Mar 2020 22:25:41 +0000 (15:25 -0700)]
Linux 5.6

5 years agoMerge branch 'akpm' (patches from Andrew)
Linus Torvalds [Sun, 29 Mar 2020 17:40:31 +0000 (10:40 -0700)]
Merge branch 'akpm' (patches from Andrew)

Merge vm fixes from Andrew Morton:
 "5 fixes"

* emailed patches from Andrew Morton <akpm@linux-foundation.org>:
  mm/sparse: fix kernel crash with pfn_section_valid check
  mm: fork: fix kernel_stack memcg stats for various stack implementations
  hugetlb_cgroup: fix illegal access to memory
  drivers/base/memory.c: indicate all memory blocks as removable
  mm/swapfile.c: move inode_lock out of claim_swapfile

5 years agoMerge tag 'timers-urgent-2020-03-29' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 29 Mar 2020 17:36:29 +0000 (10:36 -0700)]
Merge tag 'timers-urgent-2020-03-29' of git://git./linux/kernel/git/tip/tip

Pull timer fix from Thomas Gleixner:
 "A single fix for the Hyper-V clocksource driver to make sched clock
  actually return nanoseconds and not the virtual clock value which
  increments at 10e7 HZ (100ns)"

* tag 'timers-urgent-2020-03-29' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  clocksource/drivers/hyper-v: Make sched clock return nanoseconds correctly

5 years agoMerge tag 'irq-urgent-2020-03-29' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 29 Mar 2020 17:07:00 +0000 (10:07 -0700)]
Merge tag 'irq-urgent-2020-03-29' of git://git./linux/kernel/git/tip/tip

Pull irq fix from Thomas Gleixner:
 "A single bugfix to prevent reference leaks in irq affinity notifiers"

* tag 'irq-urgent-2020-03-29' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  genirq: Fix reference leaks on irq affinity notifiers

5 years agomm/sparse: fix kernel crash with pfn_section_valid check
Aneesh Kumar K.V [Sun, 29 Mar 2020 02:17:29 +0000 (19:17 -0700)]
mm/sparse: fix kernel crash with pfn_section_valid check

Fix the crash like this:

    BUG: Kernel NULL pointer dereference on read at 0x00000000
    Faulting instruction address: 0xc000000000c3447c
    Oops: Kernel access of bad area, sig: 11 [#1]
    LE PAGE_SIZE=64K MMU=Hash SMP NR_CPUS=2048 NUMA pSeries
    CPU: 11 PID: 7519 Comm: lt-ndctl Not tainted 5.6.0-rc7-autotest #1
    ...
    NIP [c000000000c3447c] vmemmap_populated+0x98/0xc0
    LR [c000000000088354] vmemmap_free+0x144/0x320
    Call Trace:
       section_deactivate+0x220/0x240
       __remove_pages+0x118/0x170
       arch_remove_memory+0x3c/0x150
       memunmap_pages+0x1cc/0x2f0
       devm_action_release+0x30/0x50
       release_nodes+0x2f8/0x3e0
       device_release_driver_internal+0x168/0x270
       unbind_store+0x130/0x170
       drv_attr_store+0x44/0x60
       sysfs_kf_write+0x68/0x80
       kernfs_fop_write+0x100/0x290
       __vfs_write+0x3c/0x70
       vfs_write+0xcc/0x240
       ksys_write+0x7c/0x140
       system_call+0x5c/0x68

The crash is due to NULL dereference at

test_bit(idx, ms->usage->subsection_map);

due to ms->usage = NULL in pfn_section_valid()

With commit d41e2f3bd546 ("mm/hotplug: fix hot remove failure in
SPARSEMEM|!VMEMMAP case") section_mem_map is set to NULL after
depopulate_section_mem().  This was done so that pfn_page() can work
correctly with kernel config that disables SPARSEMEM_VMEMMAP.  With that
config pfn_to_page does

__section_mem_map_addr(__sec) + __pfn;

where

  static inline struct page *__section_mem_map_addr(struct mem_section *section)
  {
unsigned long map = section->section_mem_map;
map &= SECTION_MAP_MASK;
return (struct page *)map;
  }

Now with SPASEMEM_VMEMAP enabled, mem_section->usage->subsection_map is
used to check the pfn validity (pfn_valid()).  Since section_deactivate
release mem_section->usage if a section is fully deactivated,
pfn_valid() check after a subsection_deactivate cause a kernel crash.

  static inline int pfn_valid(unsigned long pfn)
  {
  ...
return early_section(ms) || pfn_section_valid(ms, pfn);
  }

where

  static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
  {
int idx = subsection_map_index(pfn);

return test_bit(idx, ms->usage->subsection_map);
  }

Avoid this by clearing SECTION_HAS_MEM_MAP when mem_section->usage is
freed.  For architectures like ppc64 where large pages are used for
vmmemap mapping (16MB), a specific vmemmap mapping can cover multiple
sections.  Hence before a vmemmap mapping page can be freed, the kernel
needs to make sure there are no valid sections within that mapping.
Clearing the section valid bit before depopulate_section_memap enables
this.

[aneesh.kumar@linux.ibm.com: add comment]
Link: http://lkml.kernel.org/r/20200326133235.343616-1-aneesh.kumar@linux.ibm.comLink:
Fixes: d41e2f3bd546 ("mm/hotplug: fix hot remove failure in SPARSEMEM|!VMEMMAP case")
Reported-by: Sachin Sant <sachinp@linux.vnet.ibm.com>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Sachin Sant <sachinp@linux.vnet.ibm.com>
Reviewed-by: Baoquan He <bhe@redhat.com>
Reviewed-by: Wei Yang <richard.weiyang@gmail.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Mike Rapoport <rppt@linux.ibm.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
5 years agomm: fork: fix kernel_stack memcg stats for various stack implementations
Roman Gushchin [Sun, 29 Mar 2020 02:17:25 +0000 (19:17 -0700)]
mm: fork: fix kernel_stack memcg stats for various stack implementations

Depending on CONFIG_VMAP_STACK and the THREAD_SIZE / PAGE_SIZE ratio the
space for task stacks can be allocated using __vmalloc_node_range(),
alloc_pages_node() and kmem_cache_alloc_node().

In the first and the second cases page->mem_cgroup pointer is set, but
in the third it's not: memcg membership of a slab page should be
determined using the memcg_from_slab_page() function, which looks at
page->slab_cache->memcg_params.memcg .  In this case, using
mod_memcg_page_state() (as in account_kernel_stack()) is incorrect:
page->mem_cgroup pointer is NULL even for pages charged to a non-root
memory cgroup.

It can lead to kernel_stack per-memcg counters permanently showing 0 on
some architectures (depending on the configuration).

In order to fix it, let's introduce a mod_memcg_obj_state() helper,
which takes a pointer to a kernel object as a first argument, uses
mem_cgroup_from_obj() to get a RCU-protected memcg pointer and calls
mod_memcg_state().  It allows to handle all possible configurations
(CONFIG_VMAP_STACK and various THREAD_SIZE/PAGE_SIZE values) without
spilling any memcg/kmem specifics into fork.c .

Note: This is a special version of the patch created for stable
backports.  It contains code from the following two patches:
  - mm: memcg/slab: introduce mem_cgroup_from_obj()
  - mm: fork: fix kernel_stack memcg stats for various stack implementations

[guro@fb.com: introduce mem_cgroup_from_obj()]
Link: http://lkml.kernel.org/r/20200324004221.GA36662@carbon.dhcp.thefacebook.com
Fixes: 4d96ba353075 ("mm: memcg/slab: stop setting page->mem_cgroup pointer for slab pages")
Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Bharata B Rao <bharata@linux.ibm.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: <stable@vger.kernel.org>
Link: http://lkml.kernel.org/r/20200303233550.251375-1-guro@fb.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
5 years agohugetlb_cgroup: fix illegal access to memory
Mina Almasry [Sun, 29 Mar 2020 02:17:22 +0000 (19:17 -0700)]
hugetlb_cgroup: fix illegal access to memory

This appears to be a mistake in commit faced7e0806cf ("mm: hugetlb
controller for cgroups v2").

Essentially that commit does a hugetlb_cgroup_from_counter assuming that
page_counter_try_charge has initialized counter.

But if that has failed then it seems will not initialize counter, so
hugetlb_cgroup_from_counter(counter) ends up pointing to random memory,
causing kasan to complain.

The solution is to simply use 'h_cg', instead of
hugetlb_cgroup_from_counter(counter), since that is a reference to the
hugetlb_cgroup anyway.  After this change kasan ceases to complain.

Fixes: faced7e0806cf ("mm: hugetlb controller for cgroups v2")
Reported-by: syzbot+cac0c4e204952cf449b1@syzkaller.appspotmail.com
Signed-off-by: Mina Almasry <almasrymina@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Giuseppe Scrivano <gscrivan@redhat.com>
Acked-by: Tejun Heo <tj@kernel.org>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: David Rientjes <rientjes@google.com>
Link: http://lkml.kernel.org/r/20200313223920.124230-1-almasrymina@google.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
5 years agodrivers/base/memory.c: indicate all memory blocks as removable
David Hildenbrand [Sun, 29 Mar 2020 02:17:19 +0000 (19:17 -0700)]
drivers/base/memory.c: indicate all memory blocks as removable

We see multiple issues with the implementation/interface to compute
whether a memory block can be offlined (exposed via
/sys/devices/system/memory/memoryX/removable) and would like to simplify
it (remove the implementation).

1. It runs basically lockless. While this might be good for performance,
   we see possible races with memory offlining that will require at
   least some sort of locking to fix.

2. Nowadays, more false positives are possible. No arch-specific checks
   are performed that validate if memory offlining will not be denied
   right away (and such check will require locking). For example, arm64
   won't allow to offline any memory block that was added during boot -
   which will imply a very high error rate. Other archs have other
   constraints.

3. The interface is inherently racy. E.g., if a memory block is detected
   to be removable (and was not a false positive at that time), there is
   still no guarantee that offlining will actually succeed. So any
   caller already has to deal with false positives.

4. It is unclear which performance benefit this interface actually
   provides. The introducing commit 5c755e9fd813 ("memory-hotplug: add
   sysfs removable attribute for hotplug memory remove") mentioned

"A user-level agent must be able to identify which sections
 of memory are likely to be removable before attempting the
 potentially expensive operation."

   However, no actual performance comparison was included.

Known users:

 - lsmem: Will group memory blocks based on the "removable" property. [1]

 - chmem: Indirect user. It has a RANGE mode where one can specify
          removable ranges identified via lsmem to be offlined. However,
          it also has a "SIZE" mode, which allows a sysadmin to skip the
          manual "identify removable blocks" step. [2]

 - powerpc-utils: Uses the "removable" attribute to skip some memory
          blocks right away when trying to find some to offline+remove.
          However, with ballooning enabled, it already skips this
          information completely (because it once resulted in many false
          negatives). Therefore, the implementation can deal with false
          positives properly already. [3]

According to Nathan Fontenot, DLPAR on powerpc is nowadays no longer
driven from userspace via the drmgr command (powerpc-utils).  Nowadays
it's managed in the kernel - including onlining/offlining of memory
blocks - triggered by drmgr writing to /sys/kernel/dlpar.  So the
affected legacy userspace handling is only active on old kernels.  Only
very old versions of drmgr on a new kernel (unlikely) might execute
slower - totally acceptable.

With CONFIG_MEMORY_HOTREMOVE, always indicating "removable" should not
break any user space tool.  We implement a very bad heuristic now.
Without CONFIG_MEMORY_HOTREMOVE we cannot offline anything, so report
"not removable" as before.

Original discussion can be found in [4] ("[PATCH RFC v1] mm:
is_mem_section_removable() overhaul").

Other users of is_mem_section_removable() will be removed next, so that
we can remove is_mem_section_removable() completely.

[1] http://man7.org/linux/man-pages/man1/lsmem.1.html
[2] http://man7.org/linux/man-pages/man8/chmem.8.html
[3] https://github.com/ibm-power-utilities/powerpc-utils
[4] https://lkml.kernel.org/r/20200117105759.27905-1-david@redhat.com

Also, this patch probably fixes a crash reported by Steve.
http://lkml.kernel.org/r/CAPcyv4jpdaNvJ67SkjyUJLBnBnXXQv686BiVW042g03FUmWLXw@mail.gmail.com

Reported-by: "Scargall, Steve" <steve.scargall@intel.com>
Suggested-by: Michal Hocko <mhocko@kernel.org>
Signed-off-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Nathan Fontenot <ndfont@gmail.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: "Rafael J. Wysocki" <rafael@kernel.org>
Cc: Badari Pulavarty <pbadari@us.ibm.com>
Cc: Robert Jennings <rcj@linux.vnet.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Karel Zak <kzak@redhat.com>
Cc: <stable@vger.kernel.org>
Link: http://lkml.kernel.org/r/20200128093542.6908-1-david@redhat.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
5 years agomm/swapfile.c: move inode_lock out of claim_swapfile
Naohiro Aota [Sun, 29 Mar 2020 02:17:15 +0000 (19:17 -0700)]
mm/swapfile.c: move inode_lock out of claim_swapfile

claim_swapfile() currently keeps the inode locked when it is successful,
or the file is already swapfile (with -EBUSY).  And, on the other error
cases, it does not lock the inode.

This inconsistency of the lock state and return value is quite confusing
and actually causing a bad unlock balance as below in the "bad_swap"
section of __do_sys_swapon().

This commit fixes this issue by moving the inode_lock() and IS_SWAPFILE
check out of claim_swapfile().  The inode is unlocked in
"bad_swap_unlock_inode" section, so that the inode is ensured to be
unlocked at "bad_swap".  Thus, error handling codes after the locking now
jumps to "bad_swap_unlock_inode" instead of "bad_swap".

    =====================================
    WARNING: bad unlock balance detected!
    5.5.0-rc7+ #176 Not tainted
    -------------------------------------
    swapon/4294 is trying to release lock (&sb->s_type->i_mutex_key) at: __do_sys_swapon+0x94b/0x3550
    but there are no more locks to release!

    other info that might help us debug this:
    no locks held by swapon/4294.

    stack backtrace:
    CPU: 5 PID: 4294 Comm: swapon Not tainted 5.5.0-rc7-BTRFS-ZNS+ #176
    Hardware name: ASUS All Series/H87-PRO, BIOS 2102 07/29/2014
    Call Trace:
     dump_stack+0xa1/0xea
     print_unlock_imbalance_bug.cold+0x114/0x123
     lock_release+0x562/0xed0
     up_write+0x2d/0x490
     __do_sys_swapon+0x94b/0x3550
     __x64_sys_swapon+0x54/0x80
     do_syscall_64+0xa4/0x4b0
     entry_SYSCALL_64_after_hwframe+0x49/0xbe
    RIP: 0033:0x7f15da0a0dc7

Fixes: 1638045c3677 ("mm: set S_SWAPFILE on blockdev swap devices")
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Qais Youef <qais.yousef@arm.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: <stable@vger.kernel.org>
Link: http://lkml.kernel.org/r/20200206090132.154869-1-naohiro.aota@wdc.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
5 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
Linus Torvalds [Sun, 29 Mar 2020 01:55:15 +0000 (18:55 -0700)]
Merge git://git./linux/kernel/git/netdev/net

Pull networking fixes from David Miller:

 1) Fix memory leak in vti6, from Torsten Hilbrich.

 2) Fix double free in xfrm_policy_timer, from YueHaibing.

 3) NL80211_ATTR_CHANNEL_WIDTH attribute is put with wrong type, from
    Johannes Berg.

 4) Wrong allocation failure check in qlcnic driver, from Xu Wang.

 5) Get ks8851-ml IO operations right, for real this time, from Marek
    Vasut.

* git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (22 commits)
  r8169: fix PHY driver check on platforms w/o module softdeps
  net: ks8851-ml: Fix IO operations, again
  mlxsw: spectrum_mr: Fix list iteration in error path
  qlcnic: Fix bad kzalloc null test
  mac80211: set IEEE80211_TX_CTRL_PORT_CTRL_PROTO for nl80211 TX
  mac80211: mark station unauthorized before key removal
  mac80211: Check port authorization in the ieee80211_tx_dequeue() case
  cfg80211: Do not warn on same channel at the end of CSA
  mac80211: drop data frames without key on encrypted links
  ieee80211: fix HE SPR size calculation
  nl80211: fix NL80211_ATTR_CHANNEL_WIDTH attribute type
  xfrm: policy: Fix doulbe free in xfrm_policy_timer
  bpf: Explicitly memset some bpf info structures declared on the stack
  bpf: Explicitly memset the bpf_attr structure
  bpf: Sanitize the bpf_struct_ops tcp-cc name
  vti6: Fix memory leak of skb if input policy check fails
  esp: remove the skb from the chain when it's enqueued in cryptd_wq
  ipv6: xfrm6_tunnel.c: Use built-in RCU list checking
  xfrm: add the missing verify_sec_ctx_len check in xfrm_add_acquire
  xfrm: fix uctx len check in verify_sec_ctx_len
  ...

5 years agoMerge branch 'i2c/for-current' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa...
Linus Torvalds [Sat, 28 Mar 2020 20:11:26 +0000 (13:11 -0700)]
Merge branch 'i2c/for-current' of git://git./linux/kernel/git/wsa/linux

Pull i2c fixes from Wolfram Sang:
 "Three more driver bugfixes, and two doc improvements fixing build
  warnings while we are here"

* 'i2c/for-current' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa/linux:
  i2c: pca-platform: Use platform_irq_get_optional
  i2c: st: fix missing struct parameter description
  i2c: nvidia-gpu: Handle timeout correctly in gpu_i2c_check_status()
  i2c: fix a doc warning
  i2c: hix5hd2: add missed clk_disable_unprepare in remove

5 years agoMerge tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi
Linus Torvalds [Sat, 28 Mar 2020 16:14:16 +0000 (09:14 -0700)]
Merge tag 'scsi-fixes' of git://git./linux/kernel/git/jejb/scsi

Pull SCSI fixes from James Bottomley:
 "Two small fixes: one in drivers (qla2xxx), and one in the core (sd) to
  try to cope with USB enclosures that silently change reported
  parameters"

* tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi:
  scsi: sd: Fix optimal I/O size for devices that change reported values
  scsi: qla2xxx: Fix I/Os being passed down when FC device is being deleted

5 years agoi2c: pca-platform: Use platform_irq_get_optional
Chris Packham [Thu, 26 Mar 2020 22:44:22 +0000 (11:44 +1300)]
i2c: pca-platform: Use platform_irq_get_optional

The interrupt is not required so use platform_irq_get_optional() to
avoid error messages like

  i2c-pca-platform 22080000.i2c: IRQ index 0 not found

Signed-off-by: Chris Packham <chris.packham@alliedtelesis.co.nz>
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
5 years agoi2c: st: fix missing struct parameter description
Alain Volmat [Thu, 26 Mar 2020 21:22:43 +0000 (22:22 +0100)]
i2c: st: fix missing struct parameter description

Fix a missing struct parameter description to allow
warning free W=1 compilation.

Signed-off-by: Alain Volmat <avolmat@me.com>
Reviewed-by: Patrice Chotard <patrice.chotard@st.com>
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
5 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf
David S. Miller [Fri, 27 Mar 2020 23:18:51 +0000 (16:18 -0700)]
Merge git://git./pub/scm/linux/kernel/git/bpf/bpf

Daniel Borkmann says:

====================
pull-request: bpf 2020-03-27

The following pull-request contains BPF updates for your *net* tree.

We've added 3 non-merge commits during the last 4 day(s) which contain
a total of 4 files changed, 25 insertions(+), 20 deletions(-).

The main changes are:

1) Explicitly memset the bpf_attr structure on bpf() syscall to avoid
   having to rely on compiler to do so. Issues have been noticed on
   some compilers with padding and other oddities where the request was
   then unexpectedly rejected, from Greg Kroah-Hartman.

2) Sanitize the bpf_struct_ops TCP congestion control name in order to
   avoid problematic characters such as whitespaces, from Martin KaFai Lau.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agor8169: fix PHY driver check on platforms w/o module softdeps
Heiner Kallweit [Fri, 27 Mar 2020 16:33:32 +0000 (17:33 +0100)]
r8169: fix PHY driver check on platforms w/o module softdeps

On Android/x86 the module loading infrastructure can't deal with
softdeps. Therefore the check for presence of the Realtek PHY driver
module fails. mdiobus_register() will try to load the PHY driver
module, therefore move the check to after this call and explicitly
check that a dedicated PHY driver is bound to the PHY device.

Fixes: f32593773549 ("r8169: check that Realtek PHY driver module is loaded")
Reported-by: Chih-Wei Huang <cwhuang@android-x86.org>
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/klassert/ipsec
David S. Miller [Fri, 27 Mar 2020 21:56:55 +0000 (14:56 -0700)]
Merge branch 'master' of git://git./linux/kernel/git/klassert/ipsec

Steffen Klassert says:

====================
pull request (net): ipsec 2020-03-27

1) Handle NETDEV_UNREGISTER for xfrm device to handle asynchronous
   unregister events cleanly. From Raed Salem.

2) Fix vti6 tunnel inter address family TX through bpf_redirect().
   From Nicolas Dichtel.

3) Fix lenght check in verify_sec_ctx_len() to avoid a
   slab-out-of-bounds. From Xin Long.

4) Add a missing verify_sec_ctx_len check in xfrm_add_acquire
   to avoid a possible out-of-bounds to access. From Xin Long.

5) Use built-in RCU list checking of hlist_for_each_entry_rcu
   to silence false lockdep warning in __xfrm6_tunnel_spi_lookup
   when CONFIG_PROVE_RCU_LIST is enabled. From Madhuparna Bhowmik.

6) Fix a panic on esp offload when crypto is done asynchronously.
   From Xin Long.

7) Fix a skb memory leak in an error path of vti6_rcv.
   From Torsten Hilbrich.

8) Fix a race that can lead to a doulbe free in xfrm_policy_timer.
   From Xin Long.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'parisc-5.6-2' of git://git.kernel.org/pub/scm/linux/kernel/git/deller...
Linus Torvalds [Fri, 27 Mar 2020 21:34:03 +0000 (14:34 -0700)]
Merge branch 'parisc-5.6-2' of git://git./linux/kernel/git/deller/parisc-linux

Pull parsic fix from Helge Deller:
 "Fix a recursive loop when running 'make ARCH=parisc defconfig'"

* 'parisc-5.6-2' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/parisc-linux:
  parisc: Fix defconfig selection

5 years agoMerge tag 'arm-soc-fixes-5.6' of git://git.kernel.org/pub/scm/linux/kernel/git/soc/soc
Linus Torvalds [Fri, 27 Mar 2020 20:52:32 +0000 (13:52 -0700)]
Merge tag 'arm-soc-fixes-5.6' of git://git./linux/kernel/git/soc/soc

Pull ARM DT and driver fixes from Arnd Bergmann:
 "For the devicetree files, there are a total of 20 patches, almost
  entirely for 32-bit machines:

   - The Allwinner/sun9i r40 SoC dtsi file contains a number of issues,
     both for correctness and for style that are addressed in separate
     patches. This causes most of the changed lines of the DT updates
     this time.

   - More Allwinner updates fixing the identification of the security
     system on sun8i/A33, a recent regression of the A83t ethernet, and
     a few board specific issues on the TBS-A711 macine.

   - Several bug fixes for OMAP dts files, most notably fixing the
     timings for the NAND flash on the Nokia N900 that regressed a while
     ago after the move to configuring them from DT. Some other OMAPs
     now set the correct dma limits on the L3 bus, and a regression fix
     addresses lost Ethernet on dm814x

   - One incorrect setting in the newly added Raspberry Pi Zero W that
     may cause issues with the SD card controller.

   - A missing property on the bcm2835 firmware node caused incorrect
     DMA settings.

   - An old bug on the oxnas platform causing spurious interrupts is
     finally addressed.

   - A regression on the Exynos Midas board broke the OLED panel power
     supply.

   - The i.MX6 phycore SoM specified the wrong voltage for the SoC, this
     is now set to the values from the datasheet.

   - Some 64-bit machines use a deprecated string to identify the PSCI
     firmware.

  There are also several small code fixes addressing mostly serious
  issues:

   - Fix the sunxi rsb bus access to no longer return incorrect data
     when mixing 8 and 16 bit I/O.

   - Fix a suspend/resume regression on the OMAP2+ lcdc from a missing
     quirk in the ti-sysc driver

   - Fix a NULL pointer access from a race in the fsl dpio driver

   - Fix a v5.5 regression in the exynos-chipid driver that caused an
     invalid error code probing the device on non-exynos platforms

   - Fix an out-of-bounds access in the AMD TEE driver"

* tag 'arm-soc-fixes-5.6' of git://git.kernel.org/pub/scm/linux/kernel/git/soc/soc: (24 commits)
  soc: samsung: chipid: Fix return value on non-Exynos platforms
  arm64: dts: Fix leftover entry-methods for PSCI
  ARM: dts: exynos: Fix regulator node aliasing on Midas-based boards
  ARM: dts: oxnas: Fix clear-mask property
  ARM: dts: bcm283x: Fix vc4's firmware bus DMA limitations
  ARM: dts: omap5: Add bus_dma_limit for L3 bus
  ARM: dts: omap4-droid4: Fix lost touchscreen interrupts
  ARM: dts: dra7: Add bus_dma_limit for L3 bus
  ARM: bcm2835-rpi-zero-w: Add missing pinctrl name
  ARM: dts: sun8i: a33: add the new SS compatible
  dt-bindings: crypto: add new compatible for A33 SS
  ARM: dts: sun8i: r40: Move SPI device nodes based on address order
  ARM: dts: sun8i: r40: Fix register base address for SPI2 and SPI3
  ARM: dts: sun8i: r40: Move AHCI device node based on address order
  ARM: dts: imx6: phycore-som: fix arm and soc minimum voltage
  soc: fsl: dpio: register dpio irq handlers after dpio create
  tee: amdtee: out of bounds read in find_session()
  ARM: dts: N900: fix onenand timings
  bus: ti-sysc: Fix quirk flags for lcdc on am335x
  ARM: dts: Fix dm814x Ethernet by changing to use rgmii-id mode
  ...

5 years agoMerge tag 'riscv-for-linus-5.6' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Fri, 27 Mar 2020 18:06:10 +0000 (11:06 -0700)]
Merge tag 'riscv-for-linus-5.6' of git://git./linux/kernel/git/riscv/linux

Pull RISC-V fixes from Palmer Dabbelt:
 "Sorry for the last minute patches, but a few things fell through the
  cracks recently. I was on the fence about sending a late pull request
  just for the M-mode fixes, as we don't really have any users, but the
  last patch fixes the build for Fedora which I consider pretty
  important.

  Given that the M-mode fixes should be very low risk, I figured it's
  worth sending them along as well.

  Thhis passes my standard 'boot in QEMU' test"

* tag 'riscv-for-linus-5.6' of git://git.kernel.org/pub/scm/linux/kernel/git/riscv/linux:
  RISC-V: Move all address space definition macros to one place
  RISC-V: Only select essential drivers for SOC_VIRT config
  riscv: fix the IPI missing issue in nommu mode
  riscv: uaccess should be used in nommu mode

5 years agoMerge tag 'devicetree-fixes-for-5.6-4' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Fri, 27 Mar 2020 18:02:52 +0000 (11:02 -0700)]
Merge tag 'devicetree-fixes-for-5.6-4' of git://git./linux/kernel/git/robh/linux

Pull Devicetree fix from Rob Herring:
 "A single fix for building dtc with GCC 10"

* tag 'devicetree-fixes-for-5.6-4' of git://git.kernel.org/pub/scm/linux/kernel/git/robh/linux:
  scripts/dtc: Remove redundant YYLOC global declaration

5 years agoMerge tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux
Linus Torvalds [Fri, 27 Mar 2020 17:50:31 +0000 (10:50 -0700)]
Merge tag 'arm64-fixes' of git://git./linux/kernel/git/arm64/linux

Pull arm64 fix from Will Deacon:
 "Fix defconfig build when using Clang's integrated assembler"

* tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux:
  arm64: alternative: fix build with clang integrated assembler

5 years agoMerge tag 'clk-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Fri, 27 Mar 2020 16:33:48 +0000 (09:33 -0700)]
Merge tag 'clk-fixes-for-linus' of git://git./linux/kernel/git/clk/linux

Pull clk fixes from Stephen Boyd:
 "A handful of clk driver fixes.

  Mostly they're around the i.MX drivers fixing the parents of a few
  clks and making KASAN happy with how the message passing code works.

  Besides that we have a TI driver fix for the RTC parent and a fix for
  the basic gate type registration functions introduced this release
  where they didn't actually pass the arguments in the right places to
  the multiplexer function down below"

* tag 'clk-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/clk/linux:
  clk: imx: Align imx sc clock parent msg structs to 4
  clk: imx: Align imx sc clock msg structs to 4
  clk: Pass correct arguments to __clk_hw_register_gate()
  clk: ti: am43xx: Fix clock parent for RTC clock
  clk: imx8mp: Correct the enet_qos parent clock
  clk: imx8mp: Correct IMX8MP_CLK_HDMI_AXI clock parent

5 years agoMerge tag 'drm-fixes-2020-03-27' of git://anongit.freedesktop.org/drm/drm
Linus Torvalds [Fri, 27 Mar 2020 16:21:52 +0000 (09:21 -0700)]
Merge tag 'drm-fixes-2020-03-27' of git://anongit.freedesktop.org/drm/drm

Pull drm fixes from Dave Airlie:
 "Pretty quiet: some minor sg mapping fixes for 3 drivers, and a single
  oops fix for the scheduler. I'm hoping nobody tries to send me a fixes
  pull today but I'll keep an eye out of the weekend.

  radeon/amdgpu/dma-buf:
   - sg list fixes

  scheduler:
   - oops fix"

* tag 'drm-fixes-2020-03-27' of git://anongit.freedesktop.org/drm/drm:
  drm/scheduler: fix rare NULL ptr race
  drm/radeon: fix scatter-gather mapping with user pages
  drm/amdgpu: fix scatter-gather mapping with user pages
  drm/prime: use dma length macro when mapping sg

5 years agoparisc: Fix defconfig selection
Helge Deller [Thu, 26 Mar 2020 22:31:43 +0000 (23:31 +0100)]
parisc: Fix defconfig selection

Fix the recursive loop when running "make ARCH=parisc defconfig".

Fixes: 84669923e1ed ("parisc: Regenerate parisc defconfigs")
Noticed-by: Guenter Roeck <linux@roeck-us.net>
Tested-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: Helge Deller <deller@gmx.de>
5 years agoscripts/dtc: Remove redundant YYLOC global declaration
Dirk Mueller [Tue, 14 Jan 2020 17:53:41 +0000 (18:53 +0100)]
scripts/dtc: Remove redundant YYLOC global declaration

gcc 10 will default to -fno-common, which causes this error at link
time:

  (.text+0x0): multiple definition of `yylloc'; dtc-lexer.lex.o (symbol from plugin):(.text+0x0): first defined here

This is because both dtc-lexer as well as dtc-parser define the same
global symbol yyloc. Before with -fcommon those were merged into one
defintion. The proper solution would be to to mark this as "extern",
however that leads to:

  dtc-lexer.l:26:16: error: redundant redeclaration of 'yylloc' [-Werror=redundant-decls]
   26 | extern YYLTYPE yylloc;
      |                ^~~~~~
In file included from dtc-lexer.l:24:
dtc-parser.tab.h:127:16: note: previous declaration of 'yylloc' was here
  127 | extern YYLTYPE yylloc;
      |                ^~~~~~
cc1: all warnings being treated as errors

which means the declaration is completely redundant and can just be
dropped.

Signed-off-by: Dirk Mueller <dmueller@suse.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
[robh: cherry-pick from upstream]
Cc: stable@vger.kernel.org
Signed-off-by: Rob Herring <robh@kernel.org>
5 years agoclocksource/drivers/hyper-v: Make sched clock return nanoseconds correctly
Yubo Xie [Fri, 27 Mar 2020 02:11:59 +0000 (19:11 -0700)]
clocksource/drivers/hyper-v: Make sched clock return nanoseconds correctly

The sched clock read functions return the HV clock (100ns granularity)
without converting it to nanoseconds.

Add the missing conversion.

Fixes: bd00cd52d5be ("clocksource/drivers/hyperv: Add Hyper-V specific sched clock function")
Signed-off-by: Yubo Xie <yuboxie@microsoft.com>
Signed-off-by: Tianyu Lan <Tianyu.Lan@microsoft.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Cc: stable@vger.kernel.org
Link: https://lkml.kernel.org/r/20200327021159.31429-1-Tianyu.Lan@microsoft.com
5 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input
Linus Torvalds [Fri, 27 Mar 2020 03:49:44 +0000 (20:49 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/dtor/input

Pull input fixes from Dmitry Torokhov:

 - a fix to generate proper timestamps on key autorepeat events that
   were broken recently

 - a fix for Synaptics driver to only activate reduced reporting mode
   when explicitly requested

 - a new keycode for "selective screenshot" function

 - other assorted fixes

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input:
  Input: fix stale timestamp on key autorepeat events
  Input: move the new KEY_SELECTIVE_SCREENSHOT keycode
  Input: avoid BIT() macro usage in the serio.h UAPI header
  Input: synaptics-rmi4 - set reduced reporting mode only when requested
  Input: synaptics - enable RMI on HP Envy 13-ad105ng
  Input: allocate keycode for "Selective Screenshot" key
  Input: tm2-touchkey - add support for Coreriver TC360 variant
  dt-bindings: input: add Coreriver TC360 binding
  dt-bindings: vendor-prefixes: Add Coreriver vendor prefix
  Input: raydium_i2c_ts - fix error codes in raydium_i2c_boot_trigger()

5 years agonet: ks8851-ml: Fix IO operations, again
Marek Vasut [Wed, 25 Mar 2020 14:25:47 +0000 (15:25 +0100)]
net: ks8851-ml: Fix IO operations, again

This patch reverts 58292104832f ("net: ks8851-ml: Fix 16-bit IO operation")
and edacb098ea9c ("net: ks8851-ml: Fix 16-bit data access"), because it
turns out these were only necessary due to buggy hardware. This patch adds
a check for such a buggy hardware to prevent any such mistakes again.

While working further on the KS8851 driver, it came to light that the
KS8851-16MLL is capable of switching bus endianness by a hardware strap,
EESK pin. If this strap is incorrect, the IO accesses require such endian
swapping as is being reverted by this patch. Such swapping also impacts
the performance significantly.

Hence, in addition to removing it, detect that the hardware is broken,
report to user, and fail to bind with such hardware.

Fixes: 58292104832f ("net: ks8851-ml: Fix 16-bit IO operation")
Fixes: edacb098ea9c ("net: ks8851-ml: Fix 16-bit data access")
Signed-off-by: Marek Vasut <marex@denx.de>
Cc: David S. Miller <davem@davemloft.net>
Cc: Lukas Wunner <lukas@wunner.de>
Cc: Petr Stetiar <ynezz@true.cz>
Cc: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge tag 'amd-drm-fixes-5.6-2020-03-26' of git://people.freedesktop.org/~agd5f/linux...
Dave Airlie [Fri, 27 Mar 2020 03:03:17 +0000 (13:03 +1000)]
Merge tag 'amd-drm-fixes-5.6-2020-03-26' of git://people.freedesktop.org/~agd5f/linux into drm-fixes

amd-drm-fixes-5.6-2020-03-26:

Scheduler:
- Fix a race condition that could result in a segfault

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Alex Deucher <alexdeucher@gmail.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200326144538.3937-1-alexander.deucher@amd.com
5 years agoMerge tag 'drm-misc-fixes-2020-03-26' of git://anongit.freedesktop.org/drm/drm-misc...
Dave Airlie [Fri, 27 Mar 2020 02:33:13 +0000 (12:33 +1000)]
Merge tag 'drm-misc-fixes-2020-03-26' of git://anongit.freedesktop.org/drm/drm-misc into drm-fixes

drm-misc-fixes for v5.6:
- SG fixes for prime, radeon and amdgpu.

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/ef10e822-76dd-125d-ec1f-9a78c5f76bc3@linux.intel.com
5 years agoRISC-V: Move all address space definition macros to one place
Atish Patra [Thu, 26 Mar 2020 22:55:46 +0000 (15:55 -0700)]
RISC-V: Move all address space definition macros to one place

We get the following compilation error if CONFIG_SPARSEMEM_VMEMMAP is set.

---------------------------------------------------------------
./arch/riscv/include/asm/pgtable-64.h: In function ‘pud_page’:
./include/asm-generic/memory_model.h:54:29: error: ‘vmemmap’ undeclared
(first use in this function); did you mean ‘mem_map’?
 #define __pfn_to_page(pfn) (vmemmap + (pfn))
                             ^~~~~~~
./include/asm-generic/memory_model.h:82:21: note: in expansion of
macro ‘__pfn_to_page’

 #define pfn_to_page __pfn_to_page
                     ^~~~~~~~~~~~~
./arch/riscv/include/asm/pgtable-64.h:70:9: note: in expansion of macro
‘pfn_to_page’
  return pfn_to_page(pud_val(pud) >> _PAGE_PFN_SHIFT);
---------------------------------------------------------------

Fix the compliation errors by moving all the address space definition
macros before including pgtable-64.h.

Fixes: 8ad8b72721d0 (riscv: Add KASAN support)
Signed-off-by: Atish Patra <atish.patra@wdc.com>
Reviewed-by: Anup Patel <anup@brainfault.org>
Signed-off-by: Palmer Dabbelt <palmerdabbelt@google.com>
5 years agoInput: fix stale timestamp on key autorepeat events
Dmitry Torokhov [Wed, 25 Mar 2020 17:57:54 +0000 (10:57 -0700)]
Input: fix stale timestamp on key autorepeat events

We need to refresh timestamp when emitting key autorepeat events, otherwise
they will carry timestamp of the original key press event.

Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=206929
Fixes: 3b51c44bd693 ("Input: allow drivers specify timestamp for input events")
Cc: stable@vger.kernel.org
Reported-by: teika kazura <teika@gmx.com>
Tested-by: teika kazura <teika@gmx.com>
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
5 years agoafs: Fix unpinned address list during probing
David Howells [Thu, 26 Mar 2020 15:24:07 +0000 (15:24 +0000)]
afs: Fix unpinned address list during probing

When it's probing all of a fileserver's interfaces to find which one is
best to use, afs_do_probe_fileserver() takes a lock on the server record
and notes the pointer to the address list.

It doesn't, however, pin the address list, so as soon as it drops the
lock, there's nothing to stop the address list from being freed under
us.

Fix this by taking a ref on the address list inside the locked section
and dropping it at the end of the function.

Fixes: 3bf0fb6f33dd ("afs: Probe multiple fileservers simultaneously")
Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: Marc Dionne <marc.dionne@auristor.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
5 years agoMerge tag 'ceph-for-5.6-rc8' of git://github.com/ceph/ceph-client
Linus Torvalds [Thu, 26 Mar 2020 22:44:41 +0000 (15:44 -0700)]
Merge tag 'ceph-for-5.6-rc8' of git://github.com/ceph/ceph-client

Pull ceph fixes from Ilya Dryomov:
 "A patch for a rather old regression in fullness handling and two
  memory leak fixes, marked for stable"

* tag 'ceph-for-5.6-rc8' of git://github.com/ceph/ceph-client:
  ceph: fix memory leak in ceph_cleanup_snapid_map()
  libceph: fix alloc_msg_with_page_vector() memory leaks
  ceph: check POOL_FLAG_FULL/NEARFULL in addition to OSDMAP_FULL/NEARFULL

5 years agoMerge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
Linus Torvalds [Thu, 26 Mar 2020 22:30:49 +0000 (15:30 -0700)]
Merge tag 'for-linus' of git://git./virt/kvm/kvm

Pull KVM fixes from Paolo Bonzini:
 "x86 bug fixes"

* tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm:
  KVM: X86: Narrow down the IPI fastpath to single target IPI
  KVM: LAPIC: Also cancel preemption timer when disarm LAPIC timer
  KVM: VMX: don't allow memory operands for inline asm that modifies SP
  KVM: LAPIC: Mark hrtimer for period or oneshot mode to expire in hard interrupt context
  KVM: SVM: Issue WBINVD after deactivating an SEV guest
  KVM: SVM: document KVM_MEM_ENCRYPT_OP, let userspace detect if SEV is available
  KVM: x86: remove bogus user-triggerable WARN_ON

5 years agoMAINTAINERS: fix bad file pattern
Linus Torvalds [Thu, 26 Mar 2020 22:12:19 +0000 (15:12 -0700)]
MAINTAINERS: fix bad file pattern

Testing 'parse-maintainers' due to the previous commit shows a bad file
pattern for the "TI VPE/CAL DRIVERS" entry in the MAINTAINERS file.

There's also a lot of mis-ordered entries, but I'm still a bit nervous
about the inevitable and annoying merge problems it would probably cause
to fix them up.

The MAINTAINERS file is one of my least favorite files due to being huge
and centralized, but fixing it is also horribly painful for that reason.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
5 years agoparse-maintainers: Do not sort section content by default
Joe Perches [Sun, 8 Mar 2020 02:59:05 +0000 (18:59 -0800)]
parse-maintainers: Do not sort section content by default

Add an --order switch to control section reordering.
Default for --order is off.

Change the default ordering to a slightly more sensible:

M:  Person acting as a maintainer
R:  Person acting as a patch reviewer
L:  Mailing list where patches should be sent
S:  Maintenance status
W:  URI for general information
Q:  URI for patchwork tracking
B:  URI for bug tracking/submission
C:  URI for chat
P:  URI or file for subsystem specific coding styles
T:  SCM tree type and location
F:  File and directory pattern
X:  File and directory exclusion pattern
N:  File glob
K:  Keyword - patch content regex

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
5 years agoInput: move the new KEY_SELECTIVE_SCREENSHOT keycode
Dmitry Torokhov [Thu, 26 Mar 2020 19:54:02 +0000 (12:54 -0700)]
Input: move the new KEY_SELECTIVE_SCREENSHOT keycode

We should try to keep keycodes sequential unless there is a reason to leave
a gap in numbering, so let's move it from 0x280 to 0x27a while we still
can.

Fixes: 3b059da9835c ("Input: allocate keycode for Selective Screenshot key")
Acked-by: Rajat Jain <rajatja@google.com>
Link: https://lore.kernel.org/r/20200326182711.GA259753@dtor-ws
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
5 years agoRISC-V: Only select essential drivers for SOC_VIRT config
Anup Patel [Tue, 10 Mar 2020 11:59:25 +0000 (17:29 +0530)]
RISC-V: Only select essential drivers for SOC_VIRT config

The kconfig select causes build failues for SOC_VIRT config becaus
we are selecting lot of VIRTIO drivers without selecting all required
dependencies.

Better approach is to only select essential drivers from SOC_VIRT
config option and enable required VIRTIO drivers using defconfigs.

Fixes: 759bdc168181 ("RISC-V: Add kconfig option for QEMU virt machine")
Signed-off-by: Anup Patel <anup.patel@wdc.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Palmer Dabbelt <palmerdabbelt@google.com>
5 years agoMerge tag 'mac80211-for-net-2020-03-26' of git://git.kernel.org/pub/scm/linux/kernel...
David S. Miller [Thu, 26 Mar 2020 19:03:02 +0000 (12:03 -0700)]
Merge tag 'mac80211-for-net-2020-03-26' of git://git./linux/kernel/git/jberg/mac80211

Johannes Berg says:

====================
We have the following fixes:
 * drop data packets if there's no key for them anymore, after
   there had been one, to avoid sending them in clear when
   hostapd removes the key before it removes the station and
   the packets are still queued
 * check port authorization again after dequeue, to avoid
   sending packets if the station is no longer authorized
 * actually remove the authorization flag before the key so
   packets are also dropped properly because of this
 * fix nl80211 control port packet tagging to handle them as
   packets allowed to go out without encryption
 * fix NL80211_ATTR_CHANNEL_WIDTH outgoing netlink attribute
   width (should be 32 bits, not 8)
 * don't WARN in a CSA scenario that happens on some APs
 * fix HE spatial reuse element size calculation
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: spectrum_mr: Fix list iteration in error path
Ido Schimmel [Thu, 26 Mar 2020 14:17:33 +0000 (16:17 +0200)]
mlxsw: spectrum_mr: Fix list iteration in error path

list_for_each_entry_from_reverse() iterates backwards over the list from
the current position, but in the error path we should start from the
previous position.

Fix this by using list_for_each_entry_continue_reverse() instead.

This suppresses the following error from coccinelle:

drivers/net/ethernet/mellanox/mlxsw//spectrum_mr.c:655:34-38: ERROR:
invalid reference to the index variable of the iterator on line 636

Fixes: c011ec1bbfd6 ("mlxsw: spectrum: Add the multicast routing offloading logic")
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoqlcnic: Fix bad kzalloc null test
Xu Wang [Thu, 26 Mar 2020 10:14:29 +0000 (18:14 +0800)]
qlcnic: Fix bad kzalloc null test

In qlcnic_83xx_get_reset_instruction_template, the variable
of null test is bad, so correct it.

Signed-off-by: Xu Wang <vulab@iscas.ac.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma
Linus Torvalds [Thu, 26 Mar 2020 17:39:36 +0000 (10:39 -0700)]
Merge tag 'for-linus' of git://git./linux/kernel/git/rdma/rdma

Pull rdma fixes from Jason Gunthorpe:
 "A small set of late-rc patches, mostly fixes for various crashers,
  some syzkaller fixes and a mlx5 HW limitation:

   - Several MAINTAINERS updates

   - Memory leak regression in ODP

   - Several fixes for syzkaller related crashes. Google recently taught
     syzkaller to create the software RDMA devices

   - Crash fixes for HFI1

   - Several fixes for mlx5 crashes

   - Prevent unprivileged access to an unsafe mlx5 HW resource"

* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma:
  RDMA/mlx5: Block delay drop to unprivileged users
  RDMA/mlx5: Fix access to wrong pointer while performing flush due to error
  RDMA/core: Ensure security pkey modify is not lost
  MAINTAINERS: Clean RXE section and add Zhu as RXE maintainer
  IB/hfi1: Ensure pq is not left on waitlist
  IB/rdmavt: Free kernel completion queue when done
  RDMA/mad: Do not crash if the rdma device does not have a umad interface
  RDMA/core: Fix missing error check on dev_set_name()
  RDMA/nl: Do not permit empty devices names during RDMA_NLDEV_CMD_NEWLINK/SET
  RDMA/mlx5: Fix the number of hwcounters of a dynamic counter
  MAINTAINERS: Update maintainers for HISILICON ROCE DRIVER
  RDMA/odp: Fix leaking the tgid for implicit ODP

5 years agomac80211: set IEEE80211_TX_CTRL_PORT_CTRL_PROTO for nl80211 TX
Johannes Berg [Thu, 26 Mar 2020 14:53:34 +0000 (15:53 +0100)]
mac80211: set IEEE80211_TX_CTRL_PORT_CTRL_PROTO for nl80211 TX

When a frame is transmitted via the nl80211 TX rather than as a
normal frame, IEEE80211_TX_CTRL_PORT_CTRL_PROTO wasn't set and
this will lead to wrong decisions (rate control etc.) being made
about the frame; fix this.

Fixes: 911806491425 ("mac80211: Add support for tx_control_port")
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Link: https://lore.kernel.org/r/20200326155333.f183f52b02f0.I4054e2a8c11c2ddcb795a0103c87be3538690243@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
5 years agomac80211: mark station unauthorized before key removal
Johannes Berg [Thu, 26 Mar 2020 14:51:35 +0000 (15:51 +0100)]
mac80211: mark station unauthorized before key removal

If a station is still marked as authorized, mark it as no longer
so before removing its keys. This allows frames transmitted to it
to be rejected, providing additional protection against leaking
plain text data during the disconnection flow.

Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/20200326155133.ccb4fb0bb356.If48f0f0504efdcf16b8921f48c6d3bb2cb763c99@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
5 years agomac80211: Check port authorization in the ieee80211_tx_dequeue() case
Jouni Malinen [Thu, 26 Mar 2020 14:51:34 +0000 (15:51 +0100)]
mac80211: Check port authorization in the ieee80211_tx_dequeue() case

mac80211 used to check port authorization in the Data frame enqueue case
when going through start_xmit(). However, that authorization status may
change while the frame is waiting in a queue. Add a similar check in the
dequeue case to avoid sending previously accepted frames after
authorization change. This provides additional protection against
potential leaking of frames after a station has been disconnected and
the keys for it are being removed.

Cc: stable@vger.kernel.org
Signed-off-by: Jouni Malinen <jouni@codeaurora.org>
Link: https://lore.kernel.org/r/20200326155133.ced84317ea29.I34d4c47cd8cc8a4042b38a76f16a601fbcbfd9b3@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
5 years agocfg80211: Do not warn on same channel at the end of CSA
Ilan Peer [Thu, 26 Mar 2020 13:09:43 +0000 (15:09 +0200)]
cfg80211: Do not warn on same channel at the end of CSA

When cfg80211_update_assoc_bss_entry() is called, there is a
verification that the BSS channel actually changed. As some APs use
CSA also for bandwidth changes, this would result with a kernel
warning.

Fix this by removing the WARN_ON().

Signed-off-by: Ilan Peer <ilan.peer@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Link: https://lore.kernel.org/r/iwlwifi.20200326150855.96316ada0e8d.I6710376b1b4257e5f4712fc7ab16e2b638d512aa@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
5 years agomac80211: drop data frames without key on encrypted links
Johannes Berg [Thu, 26 Mar 2020 13:09:42 +0000 (15:09 +0200)]
mac80211: drop data frames without key on encrypted links

If we know that we have an encrypted link (based on having had
a key configured for TX in the past) then drop all data frames
in the key selection handler if there's no key anymore.

This fixes an issue with mac80211 internal TXQs - there we can
buffer frames for an encrypted link, but then if the key is no
longer there when they're dequeued, the frames are sent without
encryption. This happens if a station is disconnected while the
frames are still on the TXQ.

Detecting that a link should be encrypted based on a first key
having been configured for TX is fine as there are no use cases
for a connection going from with encryption to no encryption.
With extended key IDs, however, there is a case of having a key
configured for only decryption, so we can't just trigger this
behaviour on a key being configured.

Cc: stable@vger.kernel.org
Reported-by: Jouni Malinen <j@w1.fi>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Link: https://lore.kernel.org/r/iwlwifi.20200326150855.6865c7f28a14.I9fb1d911b064262d33e33dfba730cdeef83926ca@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
5 years agodrm/scheduler: fix rare NULL ptr race
Yintian Tao [Mon, 23 Mar 2020 11:19:37 +0000 (19:19 +0800)]
drm/scheduler: fix rare NULL ptr race

There is one one corner case at dma_fence_signal_locked
which will raise the NULL pointer problem just like below.
->dma_fence_signal
    ->dma_fence_signal_locked
->test_and_set_bit
here trigger dma_fence_release happen due to the zero of fence refcount.

->dma_fence_put
    ->dma_fence_release
->drm_sched_fence_release_scheduled
    ->call_rcu
here make the union fled “cb_list” at finished fence
to NULL because struct rcu_head contains two pointer
which is same as struct list_head cb_list

Therefore, to hold the reference of finished fence at drm_sched_process_job
to prevent the null pointer during finished fence dma_fence_signal

[  732.912867] BUG: kernel NULL pointer dereference, address: 0000000000000008
[  732.914815] #PF: supervisor write access in kernel mode
[  732.915731] #PF: error_code(0x0002) - not-present page
[  732.916621] PGD 0 P4D 0
[  732.917072] Oops: 0002 [#1] SMP PTI
[  732.917682] CPU: 7 PID: 0 Comm: swapper/7 Tainted: G           OE     5.4.0-rc7 #1
[  732.918980] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.8.2-0-g33fbe13 by qemu-project.org 04/01/2014
[  732.920906] RIP: 0010:dma_fence_signal_locked+0x3e/0x100
[  732.938569] Call Trace:
[  732.939003]  <IRQ>
[  732.939364]  dma_fence_signal+0x29/0x50
[  732.940036]  drm_sched_fence_finished+0x12/0x20 [gpu_sched]
[  732.940996]  drm_sched_process_job+0x34/0xa0 [gpu_sched]
[  732.941910]  dma_fence_signal_locked+0x85/0x100
[  732.942692]  dma_fence_signal+0x29/0x50
[  732.943457]  amdgpu_fence_process+0x99/0x120 [amdgpu]
[  732.944393]  sdma_v4_0_process_trap_irq+0x81/0xa0 [amdgpu]

v2: hold the finished fence at drm_sched_process_job instead of
    amdgpu_fence_process
v3: resume the blank line

Signed-off-by: Yintian Tao <yttao@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
5 years agoKVM: X86: Narrow down the IPI fastpath to single target IPI
Wanpeng Li [Thu, 26 Mar 2020 02:20:01 +0000 (10:20 +0800)]
KVM: X86: Narrow down the IPI fastpath to single target IPI

The original single target IPI fastpath patch forgot to filter the
ICR destination shorthand field. Multicast IPI is not suitable for
this feature since wakeup the multiple sleeping vCPUs will extend
the interrupt disabled time, it especially worse in the over-subscribe
and VM has a little bit more vCPUs scenario. Let's narrow it down to
single target IPI.

Two VMs, each is 76 vCPUs, one running 'ebizzy -M', the other
running cyclictest on all vCPUs, w/ this patch, the avg score
of cyclictest can improve more than 5%. (pv tlb, pv ipi, pv
sched yield are disabled during testing to avoid the disturb).

Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
Message-Id: <1585189202-1708-3-git-send-email-wanpengli@tencent.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>