sfrench/cifs-2.6.git
4 years agodax: Fix condition for filling of PMD holes
Jan Kara [Thu, 12 May 2016 16:29:14 +0000 (18:29 +0200)]
dax: Fix condition for filling of PMD holes

Currently dax_pmd_fault() decides to fill a PMD-sized hole only if
returned buffer has BH_Uptodate set. However that doesn't get set for
any mapping buffer so that branch is actually a dead code. The
BH_Uptodate check doesn't make any sense so just remove it.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
4 years agodax: fix a comment in dax_zero_page_range and dax_truncate_page
Vishal Verma [Thu, 21 Apr 2016 20:45:56 +0000 (16:45 -0400)]
dax: fix a comment in dax_zero_page_range and dax_truncate_page

The distinction between PAGE_SIZE and PAGE_CACHE_SIZE was removed in

09cbfea mm, fs: get rid of PAGE_CACHE_* and page_cache_{get,release}
macros

The comments for the above functions described a distinction between
those, that is now redundant, so remove those paragraphs

Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Vishal Verma <vishal.l.verma@intel.com>
4 years agodax: for truncate/hole-punch, do zeroing through the driver if possible
Vishal Verma [Thu, 21 Apr 2016 19:13:46 +0000 (15:13 -0400)]
dax: for truncate/hole-punch, do zeroing through the driver if possible

In the truncate or hole-punch path in dax, we clear out sub-page ranges.
If these sub-page ranges are sector aligned and sized, we can do the
zeroing through the driver instead so that error-clearing is handled
automatically.

For sub-sector ranges, we still have to rely on clear_pmem and have the
possibility of tripping over errors.

Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Jan Kara <jack@suse.cz>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Vishal Verma <vishal.l.verma@intel.com>
4 years agodax: export a low-level __dax_zero_page_range helper
Christoph Hellwig [Mon, 9 May 2016 08:47:04 +0000 (10:47 +0200)]
dax: export a low-level __dax_zero_page_range helper

This allows XFS to perform zeroing using the iomap infrastructure and
avoid buffer heads.

Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Christoph Hellwig <hch@lst.de>
[vishal: fix conflicts with dax-error-handling]
Signed-off-by: Vishal Verma <vishal.l.verma@intel.com>
4 years agodax: use sb_issue_zerout instead of calling dax_clear_sectors
Matthew Wilcox [Tue, 15 Mar 2016 17:20:41 +0000 (11:20 -0600)]
dax: use sb_issue_zerout instead of calling dax_clear_sectors

dax_clear_sectors() cannot handle poisoned blocks.  These must be
zeroed using the BIO interface instead.  Convert ext2 and XFS to use
only sb_issue_zerout().

Reviewed-by: Jeff Moyer <jmoyer@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Matthew Wilcox <matthew.r.wilcox@intel.com>
[vishal: Also remove the dax_clear_sectors function entirely]
Signed-off-by: Vishal Verma <vishal.l.verma@intel.com>
4 years agodax: enable dax in the presence of known media errors (badblocks)
Dan Williams [Wed, 24 Feb 2016 22:02:11 +0000 (14:02 -0800)]
dax: enable dax in the presence of known media errors (badblocks)

1/ If a mapping overlaps a bad sector fail the request.

2/ Do not opportunistically report more dax-capable capacity than is
   requested when errors present.

Reviewed-by: Jeff Moyer <jmoyer@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
[vishal: fix a conflict with system RAM collision patches]
[vishal: add a 'size' parameter to ->direct_access]
[vishal: fix a conflict with DAX alignment check patches]
Signed-off-by: Vishal Verma <vishal.l.verma@intel.com>
4 years agodax: fallback from pmd to pte on error
Dan Williams [Wed, 24 Feb 2016 22:02:06 +0000 (14:02 -0800)]
dax: fallback from pmd to pte on error

In preparation for consulting a badblocks list in pmem_direct_access(),
teach dax_pmd_fault() to fallback rather than fail immediately upon
encountering an error.  The thought being that reducing the span of the
dax request may avoid the error region.

Reviewed-by: Jeff Moyer <jmoyer@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Vishal Verma <vishal.l.verma@intel.com>
4 years agoblock: Update blkdev_dax_capable() for consistency
Toshi Kani [Tue, 10 May 2016 16:23:57 +0000 (10:23 -0600)]
block: Update blkdev_dax_capable() for consistency

blkdev_dax_capable() is similar to bdev_dax_supported(), but needs
to remain as a separate interface for checking dax capability of
a raw block device.

Rename and relocate blkdev_dax_capable() to keep them maintained
consistently, and call bdev_direct_access() for the dax capability
check.

There is no change in the behavior.

Link: https://lkml.org/lkml/2016/5/9/950
Signed-off-by: Toshi Kani <toshi.kani@hpe.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Jens Axboe <axboe@fb.com>
Cc: Andreas Dilger <adilger.kernel@dilger.ca>
Cc: Jan Kara <jack@suse.cz>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Boaz Harrosh <boaz@plexistor.com>
Signed-off-by: Vishal Verma <vishal.l.verma@intel.com>
4 years agoxfs: Add alignment check for DAX mount
Toshi Kani [Tue, 10 May 2016 16:23:56 +0000 (10:23 -0600)]
xfs: Add alignment check for DAX mount

When a partition is not aligned by 4KB, mount -o dax succeeds,
but any read/write access to the filesystem fails, except for
metadata update.

Call bdev_dax_supported() to perform proper precondition checks
which includes this partition alignment check.

Signed-off-by: Toshi Kani <toshi.kani@hpe.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Boaz Harrosh <boaz@plexistor.com>
Signed-off-by: Vishal Verma <vishal.l.verma@intel.com>
4 years agoext2: Add alignment check for DAX mount
Toshi Kani [Tue, 10 May 2016 16:23:55 +0000 (10:23 -0600)]
ext2: Add alignment check for DAX mount

When a partition is not aligned by 4KB, mount -o dax succeeds,
but any read/write access to the filesystem fails, except for
metadata update.

Call bdev_dax_supported() to perform proper precondition checks
which includes this partition alignment check.

Signed-off-by: Toshi Kani <toshi.kani@hpe.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Jan Kara <jack@suse.cz>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Boaz Harrosh <boaz@plexistor.com>
Signed-off-by: Vishal Verma <vishal.l.verma@intel.com>
4 years agoext4: Add alignment check for DAX mount
Toshi Kani [Tue, 10 May 2016 16:23:54 +0000 (10:23 -0600)]
ext4: Add alignment check for DAX mount

When a partition is not aligned by 4KB, mount -o dax succeeds,
but any read/write access to the filesystem fails, except for
metadata update.

Call bdev_dax_supported() to perform proper precondition checks
which includes this partition alignment check.

Reported-by: Micah Parrish <micah.parrish@hpe.com>
Signed-off-by: Toshi Kani <toshi.kani@hpe.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Andreas Dilger <adilger.kernel@dilger.ca>
Cc: Jan Kara <jack@suse.cz>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Boaz Harrosh <boaz@plexistor.com>
Signed-off-by: Vishal Verma <vishal.l.verma@intel.com>
4 years agoblock: Add bdev_dax_supported() for dax mount checks
Toshi Kani [Tue, 10 May 2016 16:23:53 +0000 (10:23 -0600)]
block: Add bdev_dax_supported() for dax mount checks

DAX imposes additional requirements to a device.  Add
bdev_dax_supported() which performs all the precondition checks
necessary for filesystem to mount the device with dax option.

Also add a new check to verify if a partition is aligned by 4KB.
When a partition is unaligned, any dax read/write access fails,
except for metadata update.

Signed-off-by: Toshi Kani <toshi.kani@hpe.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Jens Axboe <axboe@fb.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Andreas Dilger <adilger.kernel@dilger.ca>
Cc: Jan Kara <jack@suse.cz>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Boaz Harrosh <boaz@plexistor.com>
Signed-off-by: Vishal Verma <vishal.l.verma@intel.com>
4 years agoblock: Add vfs_msg() interface
Toshi Kani [Tue, 10 May 2016 16:23:52 +0000 (10:23 -0600)]
block: Add vfs_msg() interface

In preparation of moving DAX capability checks to the block layer
from filesystem code, add a VFS message interface that aligns with
filesystem's message format.

For instance, a vfs_msg() message followed by XFS messages in case
of a dax mount error may look like:

  VFS (pmem0p1): error: unaligned partition for dax
  XFS (pmem0p1): DAX unsupported by block device. Turning off DAX.
  XFS (pmem0p1): Mounting V5 Filesystem
   :

vfs_msg() is largely based on ext4_msg().

Signed-off-by: Toshi Kani <toshi.kani@hpe.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Jens Axboe <axboe@fb.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Andreas Dilger <adilger.kernel@dilger.ca>
Cc: Jan Kara <jack@suse.cz>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Boaz Harrosh <boaz@plexistor.com>
Signed-off-by: Vishal Verma <vishal.l.verma@intel.com>
4 years agodax: Remove redundant inode size checks
Jan Kara [Wed, 11 May 2016 09:58:53 +0000 (11:58 +0200)]
dax: Remove redundant inode size checks

Callers of dax fault handlers must make sure these calls cannot race
with truncate. Thus it is enough to check inode size when entering the
function and we don't have to recheck it again later in the handler.
Note that inode size itself can be decreased while the fault handler
runs but filesystem locking prevents against any radix tree or block
mapping information changes resulting from the truncate and that is what
we really care about.

Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Vishal Verma <vishal.l.verma@intel.com>
4 years agodax: Remove pointless writeback from dax_do_io()
Jan Kara [Wed, 11 May 2016 09:58:52 +0000 (11:58 +0200)]
dax: Remove pointless writeback from dax_do_io()

dax_do_io() is calling filemap_write_and_wait() if DIO_LOCKING flags is
set. Presumably this was copied over from direct IO code. However DAX
inodes have no pagecache pages to write so the call is pointless. Remove
it.

Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Vishal Verma <vishal.l.verma@intel.com>
4 years agodax: Remove zeroing from dax_io()
Jan Kara [Wed, 11 May 2016 09:58:51 +0000 (11:58 +0200)]
dax: Remove zeroing from dax_io()

All the filesystems are now zeroing blocks themselves for DAX IO to avoid
races between dax_io() and dax_fault(). Remove the zeroing code from
dax_io() and add warning to catch the case when somebody unexpectedly
returns new or unwritten buffer.

Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Vishal Verma <vishal.l.verma@intel.com>
4 years agodax: Remove dead zeroing code from fault handlers
Jan Kara [Wed, 11 May 2016 09:58:50 +0000 (11:58 +0200)]
dax: Remove dead zeroing code from fault handlers

Now that all filesystems zero out blocks allocated for a fault handler,
we can just remove the zeroing from the handler itself. Also add checks
that no filesystem returns to us unwritten or new buffer.

Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Vishal Verma <vishal.l.verma@intel.com>
4 years agoext2: Avoid DAX zeroing to corrupt data
Jan Kara [Wed, 11 May 2016 09:58:49 +0000 (11:58 +0200)]
ext2: Avoid DAX zeroing to corrupt data

Currently ext2 zeroes any data blocks allocated for DAX inode however it
still returns them as BH_New. Thus DAX code zeroes them again in
dax_insert_mapping() which can possibly overwrite the data that has been
already stored to those blocks by a racing dax_io(). Avoid marking
pre-zeroed buffers as new.

Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Vishal Verma <vishal.l.verma@intel.com>
4 years agoext2: Fix block zeroing in ext2_get_blocks() for DAX
Jan Kara [Mon, 16 May 2016 15:17:04 +0000 (17:17 +0200)]
ext2: Fix block zeroing in ext2_get_blocks() for DAX

When zeroing allocated blocks for DAX, we accidentally zeroed only the
first allocated block instead of all of them. So far this problem is
hidden by the fact that page faults always need only a single block and
DAX write code zeroes blocks again. But the zeroing in DAX code is racy
and needs to be removed so fix the zeroing in ext2 to zero all allocated
blocks.

Reported-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Vishal Verma <vishal.l.verma@intel.com>
4 years agodax: Remove complete_unwritten argument
Jan Kara [Wed, 11 May 2016 09:58:48 +0000 (11:58 +0200)]
dax: Remove complete_unwritten argument

Fault handlers currently take complete_unwritten argument to convert
unwritten extents after PTEs are updated. However no filesystem uses
this anymore as the code is racy. Remove the unused argument.

Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Vishal Verma <vishal.l.verma@intel.com>
4 years agoDAX: move RADIX_DAX_ definitions to dax.c
NeilBrown [Wed, 11 May 2016 09:58:47 +0000 (11:58 +0200)]
DAX: move RADIX_DAX_ definitions to dax.c

These don't belong in radix-tree.c any more than PAGECACHE_TAG_* do.
Let's try to maintain the idea that radix-tree simply implements an
abstract data type.

Acked-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Matthew Wilcox <willy@linux.intel.com>
Signed-off-by: NeilBrown <neilb@suse.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Vishal Verma <vishal.l.verma@intel.com>
4 years agoext4: pre-zero allocated blocks for DAX IO
Jan Kara [Fri, 13 May 2016 04:51:15 +0000 (00:51 -0400)]
ext4: pre-zero allocated blocks for DAX IO

Currently ext4 treats DAX IO the same way as direct IO. I.e., it
allocates unwritten extents before IO is done and converts unwritten
extents afterwards. However this way DAX IO can race with page fault to
the same area:

ext4_ext_direct_IO() dax_fault()
  dax_io()
    get_block() - allocates unwritten extent
    copy_from_iter_pmem()
  get_block() - converts
    unwritten block to
    written and zeroes it
    out
  ext4_convert_unwritten_extents()

So data written with DAX IO gets lost. Similarly dax_new_buf() called
from dax_io() can overwrite data that has been already written to the
block via mmap.

Fix the problem by using pre-zeroed blocks for DAX IO the same way as we
use them for DAX mmap. The downside of this solution is that every
allocating write writes each block twice (once zeros, once data). Fixing
the race with locking is possible as well however we would need to
lock-out faults for the whole range written to by DAX IO. And that is
not easy to do without locking-out faults for the whole file which seems
too aggressive.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
4 years agoext4: refactor direct IO code
Jan Kara [Fri, 13 May 2016 04:44:16 +0000 (00:44 -0400)]
ext4: refactor direct IO code

Currently ext4 direct IO handling is split between ext4_ext_direct_IO()
and ext4_ind_direct_IO(). However the extent based function calls into
the indirect based one for some cases and for example it is not able to
handle file extending. Previously it was not also properly handling
retries in case of ENOSPC errors. With DAX things would get even more
contrieved so just refactor the direct IO code and instead of indirect /
extent split do the split to read vs writes.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
4 years agoext4: fix race in transient ENOSPC detection
Jan Kara [Fri, 13 May 2016 04:42:40 +0000 (00:42 -0400)]
ext4: fix race in transient ENOSPC detection

When there are blocks to free in the running transaction, block
allocator can return ENOSPC although the filesystem has some blocks to
free. We use ext4_should_retry_alloc() to force commit of the current
transaction and return whether anything was committed so that it makes
sense to retry the allocation. However the transaction may get committed
after block allocation fails but before we call
ext4_should_retry_alloc(). So ext4_should_retry_alloc() returns false
because there is nothing to commit and we wrongly return ENOSPC.

Fix the race by unconditionally returning 1 from ext4_should_retry_alloc()
when we tried to commit a transaction. This should not add any
unnecessary retries since we had a transaction running a while ago when
trying to allocate blocks and we want to retry the allocation once that
transaction has committed anyway.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
4 years agoext4: handle transient ENOSPC properly for DAX
Jan Kara [Fri, 13 May 2016 04:38:16 +0000 (00:38 -0400)]
ext4: handle transient ENOSPC properly for DAX

ext4_dax_get_blocks() was accidentally omitted fixing get blocks
handlers to properly handle transient ENOSPC errors. Fix it now to use
ext4_get_blocks_trans() helper which takes care of these errors.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
4 years agodax: call get_blocks() with create == 1 for write faults to unwritten extents
Jan Kara [Fri, 13 May 2016 04:38:15 +0000 (00:38 -0400)]
dax: call get_blocks() with create == 1 for write faults to unwritten extents

Currently, __dax_fault() does not call get_blocks() callback with create
argument set, when we got back unwritten extent from the initial
get_blocks() call during a write fault. This is because originally
filesystems were supposed to convert unwritten extents to written ones
using complete_unwritten() callback. Later this was abandoned in favor of
using pre-zeroed blocks however the condition whether get_blocks() needs
to be called with create == 1 remained.

Fix the condition so that filesystems are not forced to zero-out and
convert unwritten extents when get_blocks() is called with create == 0
(which introduces unnecessary overhead for read faults and can be
problematic as the filesystem may possibly be read-only).

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
4 years agoext4: remove unmeetable inconsisteny check from ext4_find_extent()
Nicolai Stange [Fri, 6 May 2016 02:43:04 +0000 (22:43 -0400)]
ext4: remove unmeetable inconsisteny check from ext4_find_extent()

ext4_find_extent(), stripped down to the parts relevant to this patch,
reads as

  ppos = 0;
  i = depth;
  while (i) {
    --i;
    ++ppos;
    if (unlikely(ppos > depth)) {
      ...
      ret = -EFSCORRUPTED;
      goto err;
    }
  }

Due to the loop's bounds, the condition ppos > depth can never be met.

Remove this dead code.

Signed-off-by: Nicolai Stange <nicstange@gmail.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
4 years agojbd2: remove excess descriptions for handle_s
Luis de Bethencourt [Fri, 6 May 2016 02:35:54 +0000 (22:35 -0400)]
jbd2: remove excess descriptions for handle_s

Commit bf6993276f74 ("jbd2: Use tracepoints for history file")
removed the members j_history, j_history_max and j_history_cur from struct
handle_s but the descriptions stayed lingering. Removing them.

Signed-off-by: Luis de Bethencourt <luisbg@osg.samsung.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Reviewed-by: Jan Kara <jack@suse.cz>
4 years agoext4: remove unnecessary bio get/put
Jens Axboe [Fri, 6 May 2016 02:09:49 +0000 (22:09 -0400)]
ext4: remove unnecessary bio get/put

ext4_io_submit() used to check for EOPNOTSUPP after bio submission,
which is why it had to get an extra reference to the bio before
submitting it. But since we no longer touch the bio after submission,
get rid of the redundant get/put of the bio. If we do get the extra
reference, we enter the slower path of having to flag this bio as now
having external references.

Signed-off-by: Jens Axboe <axboe@fb.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
4 years agoext4: silence UBSAN in ext4_mb_init()
Nicolai Stange [Thu, 5 May 2016 23:46:19 +0000 (19:46 -0400)]
ext4: silence UBSAN in ext4_mb_init()

Currently, in ext4_mb_init(), there's a loop like the following:

  do {
    ...
    offset += 1 << (sb->s_blocksize_bits - i);
    i++;
  } while (i <= sb->s_blocksize_bits + 1);

Note that the updated offset is used in the loop's next iteration only.

However, at the last iteration, that is at i == sb->s_blocksize_bits + 1,
the shift count becomes equal to (unsigned)-1 > 31 (c.f. C99 6.5.7(3))
and UBSAN reports

  UBSAN: Undefined behaviour in fs/ext4/mballoc.c:2621:15
  shift exponent 4294967295 is too large for 32-bit type 'int'
  [...]
  Call Trace:
   [<ffffffff818c4d25>] dump_stack+0xbc/0x117
   [<ffffffff818c4c69>] ? _atomic_dec_and_lock+0x169/0x169
   [<ffffffff819411ab>] ubsan_epilogue+0xd/0x4e
   [<ffffffff81941cac>] __ubsan_handle_shift_out_of_bounds+0x1fb/0x254
   [<ffffffff81941ab1>] ? __ubsan_handle_load_invalid_value+0x158/0x158
   [<ffffffff814b6dc1>] ? kmem_cache_alloc+0x101/0x390
   [<ffffffff816fc13b>] ? ext4_mb_init+0x13b/0xfd0
   [<ffffffff814293c7>] ? create_cache+0x57/0x1f0
   [<ffffffff8142948a>] ? create_cache+0x11a/0x1f0
   [<ffffffff821c2168>] ? mutex_lock+0x38/0x60
   [<ffffffff821c23ab>] ? mutex_unlock+0x1b/0x50
   [<ffffffff814c26ab>] ? put_online_mems+0x5b/0xc0
   [<ffffffff81429677>] ? kmem_cache_create+0x117/0x2c0
   [<ffffffff816fcc49>] ext4_mb_init+0xc49/0xfd0
   [...]

Observe that the mentioned shift exponent, 4294967295, equals (unsigned)-1.

Unless compilers start to do some fancy transformations (which at least
GCC 6.0.0 doesn't currently do), the issue is of cosmetic nature only: the
such calculated value of offset is never used again.

Silence UBSAN by introducing another variable, offset_incr, holding the
next increment to apply to offset and adjust that one by right shifting it
by one position per loop iteration.

Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=114701
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=112161

Cc: stable@vger.kernel.org
Signed-off-by: Nicolai Stange <nicstange@gmail.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
4 years agoext4: address UBSAN warning in mb_find_order_for_block()
Nicolai Stange [Thu, 5 May 2016 21:38:03 +0000 (17:38 -0400)]
ext4: address UBSAN warning in mb_find_order_for_block()

Currently, in mb_find_order_for_block(), there's a loop like the following:

  while (order <= e4b->bd_blkbits + 1) {
    ...
    bb += 1 << (e4b->bd_blkbits - order);
  }

Note that the updated bb is used in the loop's next iteration only.

However, at the last iteration, that is at order == e4b->bd_blkbits + 1,
the shift count becomes negative (c.f. C99 6.5.7(3)) and UBSAN reports

  UBSAN: Undefined behaviour in fs/ext4/mballoc.c:1281:11
  shift exponent -1 is negative
  [...]
  Call Trace:
   [<ffffffff818c4d35>] dump_stack+0xbc/0x117
   [<ffffffff818c4c79>] ? _atomic_dec_and_lock+0x169/0x169
   [<ffffffff819411bb>] ubsan_epilogue+0xd/0x4e
   [<ffffffff81941cbc>] __ubsan_handle_shift_out_of_bounds+0x1fb/0x254
   [<ffffffff81941ac1>] ? __ubsan_handle_load_invalid_value+0x158/0x158
   [<ffffffff816e93a0>] ? ext4_mb_generate_from_pa+0x590/0x590
   [<ffffffff816502c8>] ? ext4_read_block_bitmap_nowait+0x598/0xe80
   [<ffffffff816e7b7e>] mb_find_order_for_block+0x1ce/0x240
   [...]

Unless compilers start to do some fancy transformations (which at least
GCC 6.0.0 doesn't currently do), the issue is of cosmetic nature only: the
such calculated value of bb is never used again.

Silence UBSAN by introducing another variable, bb_incr, holding the next
increment to apply to bb and adjust that one by right shifting it by one
position per loop iteration.

Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=114701
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=112161

Cc: stable@vger.kernel.org
Signed-off-by: Nicolai Stange <nicstange@gmail.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
4 years agoext4: fix oops on corrupted filesystem
Jan Kara [Thu, 5 May 2016 15:10:15 +0000 (11:10 -0400)]
ext4: fix oops on corrupted filesystem

When filesystem is corrupted in the right way, it can happen
ext4_mark_iloc_dirty() in ext4_orphan_add() returns error and we
subsequently remove inode from the in-memory orphan list. However this
deletion is done with list_del(&EXT4_I(inode)->i_orphan) and thus we
leave i_orphan list_head with a stale content. Later we can look at this
content causing list corruption, oops, or other issues. The reported
trace looked like:

WARNING: CPU: 0 PID: 46 at lib/list_debug.c:53 __list_del_entry+0x6b/0x100()
list_del corruption, 0000000061c1d6e0->next is LIST_POISON1
0000000000100100)
CPU: 0 PID: 46 Comm: ext4.exe Not tainted 4.1.0-rc4+ #250
Stack:
 60462947 62219960 602ede24 62219960
 602ede24 603ca293 622198f0 602f02eb
 62219950 6002c12c 62219900 601b4d6b
Call Trace:
 [<6005769c>] ? vprintk_emit+0x2dc/0x5c0
 [<602ede24>] ? printk+0x0/0x94
 [<600190bc>] show_stack+0xdc/0x1a0
 [<602ede24>] ? printk+0x0/0x94
 [<602ede24>] ? printk+0x0/0x94
 [<602f02eb>] dump_stack+0x2a/0x2c
 [<6002c12c>] warn_slowpath_common+0x9c/0xf0
 [<601b4d6b>] ? __list_del_entry+0x6b/0x100
 [<6002c254>] warn_slowpath_fmt+0x94/0xa0
 [<602f4d09>] ? __mutex_lock_slowpath+0x239/0x3a0
 [<6002c1c0>] ? warn_slowpath_fmt+0x0/0xa0
 [<60023ebf>] ? set_signals+0x3f/0x50
 [<600a205a>] ? kmem_cache_free+0x10a/0x180
 [<602f4e88>] ? mutex_lock+0x18/0x30
 [<601b4d6b>] __list_del_entry+0x6b/0x100
 [<601177ec>] ext4_orphan_del+0x22c/0x2f0
 [<6012f27c>] ? __ext4_journal_start_sb+0x2c/0xa0
 [<6010b973>] ? ext4_truncate+0x383/0x390
 [<6010bc8b>] ext4_write_begin+0x30b/0x4b0
 [<6001bb50>] ? copy_from_user+0x0/0xb0
 [<601aa840>] ? iov_iter_fault_in_readable+0xa0/0xc0
 [<60072c4f>] generic_perform_write+0xaf/0x1e0
 [<600c4166>] ? file_update_time+0x46/0x110
 [<60072f0f>] __generic_file_write_iter+0x18f/0x1b0
 [<6010030f>] ext4_file_write_iter+0x15f/0x470
 [<60094e10>] ? unlink_file_vma+0x0/0x70
 [<6009b180>] ? unlink_anon_vmas+0x0/0x260
 [<6008f169>] ? free_pgtables+0xb9/0x100
 [<600a6030>] __vfs_write+0xb0/0x130
 [<600a61d5>] vfs_write+0xa5/0x170
 [<600a63d6>] SyS_write+0x56/0xe0
 [<6029fcb0>] ? __libc_waitpid+0x0/0xa0
 [<6001b698>] handle_syscall+0x68/0x90
 [<6002633d>] userspace+0x4fd/0x600
 [<6002274f>] ? save_registers+0x1f/0x40
 [<60028bd7>] ? arch_prctl+0x177/0x1b0
 [<60017bd5>] fork_handler+0x85/0x90

Fix the problem by using list_del_init() as we always should with
i_orphan list.

CC: stable@vger.kernel.org
Reported-by: Vegard Nossum <vegard.nossum@oracle.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
4 years agoext4: fix check of dqget() return value in ext4_ioctl_setproject()
Seth Forshee [Thu, 5 May 2016 14:52:38 +0000 (10:52 -0400)]
ext4: fix check of dqget() return value in ext4_ioctl_setproject()

A failed call to dqget() returns an ERR_PTR() and not null. Fix
the check in ext4_ioctl_setproject() to handle this correctly.

Fixes: 9b7365fc1c82 ("ext4: add FS_IOC_FSSETXATTR/FS_IOC_FSGETXATTR interface support")
Cc: stable@vger.kernel.org # v4.5
Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Reviewed-by: Jan Kara <jack@suse.cz>
4 years agoext4: clean up error handling when orphan list is corrupted
Theodore Ts'o [Sat, 30 Apr 2016 04:49:54 +0000 (00:49 -0400)]
ext4: clean up error handling when orphan list is corrupted

Instead of just printing warning messages, if the orphan list is
corrupted, declare the file system is corrupted.  If there are any
reserved inodes in the orphaned inode list, declare the file system
corrupted and stop right away to avoid doing more potential damage to
the file system.

Cc: stable@vger.kernel.org
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
4 years agoext4: fix hang when processing corrupted orphaned inode list
Theodore Ts'o [Sat, 30 Apr 2016 04:48:54 +0000 (00:48 -0400)]
ext4: fix hang when processing corrupted orphaned inode list

If the orphaned inode list contains inode #5, ext4_iget() returns a
bad inode (since the bootloader inode should never be referenced
directly).  Because of the bad inode, we end up processing the inode
repeatedly and this hangs the machine.

This can be reproduced via:

   mke2fs -t ext4 /tmp/foo.img 100
   debugfs -w -R "ssv last_orphan 5" /tmp/foo.img
   mount -o loop /tmp/foo.img /mnt

(But don't do this if you are using an unpatched kernel if you care
about the system staying functional.  :-)

This bug was found by the port of American Fuzzy Lop into the kernel
to find file system problems[1].  (Since it *only* happens if inode #5
shows up on the orphan list --- 3, 7, 8, etc. won't do it, it's not
surprising that AFL needed two hours before it found it.)

[1] http://events.linuxfoundation.org/sites/events/files/slides/AFL%20filesystem%20fuzzing%2C%20Vault%202016_0.pdf

Cc: stable@vger.kernel.org
Reported by: Vegard Nossum <vegard.nossum@oracle.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
4 years agoext4: remove trailing \n from ext4_warning/ext4_error calls
Jakub Wilk [Wed, 27 Apr 2016 05:11:21 +0000 (01:11 -0400)]
ext4: remove trailing \n from ext4_warning/ext4_error calls

Messages passed to ext4_warning() or ext4_error() don't need trailing
newlines, because these function add the newlines themselves.

Signed-off-by: Jakub Wilk <jwilk@jwilk.net>
4 years agoext4: fix races between changing inode journal mode and ext4_writepages
Daeho Jeong [Tue, 26 Apr 2016 03:22:35 +0000 (23:22 -0400)]
ext4: fix races between changing inode journal mode and ext4_writepages

In ext4, there is a race condition between changing inode journal mode
and ext4_writepages(). While ext4_writepages() is executed on a
non-journalled mode inode, the inode's journal mode could be enabled
by ioctl() and then, some pages dirtied after switching the journal
mode will be still exposed to ext4_writepages() in non-journaled mode.
To resolve this problem, we use fs-wide per-cpu rw semaphore by Jan
Kara's suggestion because we don't want to waste ext4_inode_info's
space for this extra rare case.

Signed-off-by: Daeho Jeong <daeho.jeong@samsung.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Reviewed-by: Jan Kara <jack@suse.cz>
4 years agoext4: handle unwritten or delalloc buffers before enabling data journaling
Daeho Jeong [Tue, 26 Apr 2016 03:21:00 +0000 (23:21 -0400)]
ext4: handle unwritten or delalloc buffers before enabling data journaling

We already allocate delalloc blocks before changing the inode mode into
"per-file data journal" mode to prevent delalloc blocks from remaining
not allocated, but another issue concerned with "BH_Unwritten" status
still exists. For example, by fallocate(), several buffers' status
change into "BH_Unwritten", but these buffers cannot be processed by
ext4_alloc_da_blocks(). So, they still remain in unwritten status after
per-file data journaling is enabled and they cannot be changed into
written status any more and, if they are journaled and eventually
checkpointed, these unwritten buffer will cause a kernel panic by the
below BUG_ON() function of submit_bh_wbc() when they are submitted
during checkpointing.

static int submit_bh_wbc(int rw, struct buffer_head *bh,...
{
        ...
        BUG_ON(buffer_unwritten(bh));

Moreover, when "dioread_nolock" option is enabled, the status of a
buffer is changed into "BH_Unwritten" after write_begin() completes and
the "BH_Unwritten" status will be cleared after I/O is done. Therefore,
if a buffer's status is changed into unwrutten but the buffer's I/O is
not submitted and completed, it can cause the same problem after
enabling per-file data journaling. You can easily generate this bug by
executing the following command.

./kvm-xfstests -C 10000 -m nodelalloc,dioread_nolock generic/269

To resolve these problems and define a boundary between the previous
mode and per-file data journaling mode, we need to flush and wait all
the I/O of buffers of a file before enabling per-file data journaling
of the file.

Signed-off-by: Daeho Jeong <daeho.jeong@samsung.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Reviewed-by: Jan Kara <jack@suse.cz>
4 years agoext4: fix jbd2 handle extension in ext4_ext_truncate_extend_restart()
Theodore Ts'o [Tue, 26 Apr 2016 03:13:17 +0000 (23:13 -0400)]
ext4: fix jbd2 handle extension in ext4_ext_truncate_extend_restart()

The function jbd2_journal_extend() takes as its argument the number of
new credits to be added to the handle.  We weren't taking into account
the currently unused handle credits; worse, we would try to extend the
handle by N credits when it had N credits available.

In the case where jbd2_journal_extend() fails because the transaction
is too large, when jbd2_journal_restart() gets called, the N credits
owned by the handle gets returned to the transaction, and the
transaction commit is asynchronously requested, and then
start_this_handle() will be able to successfully attach the handle to
the current transaction since the required credits are now available.

This is mostly harmless, but since ext4_ext_truncate_extend_restart()
returns EAGAIN, the truncate machinery will once again try to call
ext4_ext_truncate_extend_restart(), which will do the above sequence
over and over again until the transaction has committed.

This was found while I was debugging a lockup in caused by running
xfstests generic/074 in the data=journal case.  I'm still not sure why
we ended up looping forever, which suggests there may still be another
bug hiding in the transaction accounting machinery, but this commit
prevents us from looping in the first place.

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
4 years agoext4: do not ask jbd2 to write data for delalloc buffers
Jan Kara [Sun, 24 Apr 2016 04:56:08 +0000 (00:56 -0400)]
ext4: do not ask jbd2 to write data for delalloc buffers

Currently we ask jbd2 to write all dirty allocated buffers before
committing a transaction when doing writeback of delay allocated blocks.
However this is unnecessary since we move all pages to writeback state
before dropping a transaction handle and then submit all the necessary
IO. We still need the transaction commit to wait for all the outstanding
writeback before flushing disk caches during transaction commit to avoid
data exposure issues though. Use the new jbd2 capability and ask it to
only wait for outstanding writeback during transaction commit when
writing back data in ext4_writepages().

Tested-by: "HUANG Weller (CM/ESW12-CN)" <Weller.Huang@cn.bosch.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
4 years agojbd2: add support for avoiding data writes during transaction commits
Jan Kara [Sun, 24 Apr 2016 04:56:07 +0000 (00:56 -0400)]
jbd2: add support for avoiding data writes during transaction commits

Currently when filesystem needs to make sure data is on permanent
storage before committing a transaction it adds inode to transaction's
inode list. During transaction commit, jbd2 writes back all dirty
buffers that have allocated underlying blocks and waits for the IO to
finish. However when doing writeback for delayed allocated data, we
allocate blocks and immediately submit the data. Thus asking jbd2 to
write dirty pages just unnecessarily adds more work to jbd2 possibly
writing back other redirtied blocks.

Add support to jbd2 to allow filesystem to ask jbd2 to only wait for
outstanding data writes before committing a transaction and thus avoid
unnecessary writes.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
4 years agoext4: remove EXT4_STATE_ORDERED_MODE
Jan Kara [Sun, 24 Apr 2016 04:56:05 +0000 (00:56 -0400)]
ext4: remove EXT4_STATE_ORDERED_MODE

This flag is just duplicating what ext4_should_order_data() tells you
and is used in a single place. Furthermore it doesn't reflect changes to
inode data journalling flag so it may be possibly misleading. Just
remove it.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
4 years agoext4: fix data exposure after a crash
Jan Kara [Sun, 24 Apr 2016 04:56:03 +0000 (00:56 -0400)]
ext4: fix data exposure after a crash

Huang has reported that in his powerfail testing he is seeing stale
block contents in some of recently allocated blocks although he mounts
ext4 in data=ordered mode. After some investigation I have found out
that indeed when delayed allocation is used, we don't add inode to
transaction's list of inodes needing flushing before commit. Originally
we were doing that but commit f3b59291a69d removed the logic with a
flawed argument that it is not needed.

The problem is that although for delayed allocated blocks we write their
contents immediately after allocating them, there is no guarantee that
the IO scheduler or device doesn't reorder things and thus transaction
allocating blocks and attaching them to inode can reach stable storage
before actual block contents. Actually whenever we attach freshly
allocated blocks to inode using a written extent, we should add inode to
transaction's ordered inode list to make sure we properly wait for block
contents to be written before committing the transaction. So that is
what we do in this patch. This also handles other cases where stale data
exposure was possible - like filling hole via mmap in
data=ordered,nodelalloc mode.

The only exception to the above rule are extending direct IO writes where
blkdev_direct_IO() waits for IO to complete before increasing i_size and
thus stale data exposure is not possible. For now we don't complicate
the code with optimizing this special case since the overhead is pretty
low. In case this is observed to be a performance problem we can always
handle it using a special flag to ext4_map_blocks().

CC: stable@vger.kernel.org
Fixes: f3b59291a69d0b734be1fc8be489fef2dd846d3d
Reported-by: "HUANG Weller (CM/ESW12-CN)" <Weller.Huang@cn.bosch.com>
Tested-by: "HUANG Weller (CM/ESW12-CN)" <Weller.Huang@cn.bosch.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
4 years agoext4: allow readdir()'s of large empty directories to be interrupted
Theodore Ts'o [Sun, 24 Apr 2016 02:50:07 +0000 (22:50 -0400)]
ext4: allow readdir()'s of large empty directories to be interrupted

If a directory has a large number of empty blocks, iterating over all
of them can take a long time, leading to scheduler warnings and users
getting irritated when they can't kill a process in the middle of one
of these long-running readdir operations.  Fix this by adding checks to
ext4_readdir() and ext4_htree_fill_tree().

This was reverted earlier due to a typo in the original commit where I
experimented with using signal_pending() instead of
fatal_signal_pending().  The test was in the wrong place if we were
going to return signal_pending() since we would end up returning
duplicant entries.  See 9f2394c9be47 for a more detailed explanation.

Added fix as suggested by Linus to check for signal_pending() in
in the filldir() functions.

Reported-by: Benjamin LaHaise <bcrl@kvack.org>
Google-Bug-Id: 27880676
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
4 years agoLinux 4.6-rc4 v4.6-rc4
Linus Torvalds [Mon, 18 Apr 2016 02:13:32 +0000 (19:13 -0700)]
Linux 4.6-rc4

4 years agoMerge tag 'dm-4.6-fix-2' of git://git.kernel.org/pub/scm/linux/kernel/git/device...
Linus Torvalds [Sun, 17 Apr 2016 19:30:06 +0000 (12:30 -0700)]
Merge tag 'dm-4.6-fix-2' of git://git./linux/kernel/git/device-mapper/linux-dm

Pull device mapper fix from Mike Snitzer:
 "Fix for earlier 4.6-rc4 stable@ commit that introduced improper use of
  write lock in cmd_read_lock() -- due to cut-n-paste gone awry (and
  sparse didn't catch it)"

* tag 'dm-4.6-fix-2' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm:
  dm cache metadata: fix cmd_read_lock() acquiring write lock

4 years agodm cache metadata: fix cmd_read_lock() acquiring write lock
Ahmed Samy [Sun, 17 Apr 2016 05:37:09 +0000 (05:37 +0000)]
dm cache metadata: fix cmd_read_lock() acquiring write lock

Commit 9567366fefdd ("dm cache metadata: fix READ_LOCK macros and
cleanup WRITE_LOCK macros") uses down_write() instead of down_read() in
cmd_read_lock(), yet up_read() is used to release the lock in
READ_UNLOCK().  Fix it.

Fixes: 9567366fefdd ("dm cache metadata: fix READ_LOCK macros and cleanup WRITE_LOCK macros")
Cc: stable@vger.kernel.org
Signed-off-by: Ahmed Samy <f.fallen45@gmail.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
4 years agoMerge tag 'char-misc-4.6-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh...
Linus Torvalds [Sun, 17 Apr 2016 03:59:06 +0000 (20:59 -0700)]
Merge tag 'char-misc-4.6-rc4' of git://git./linux/kernel/git/gregkh/char-misc

Pull char/misc fixes from Greg KH:
 "Here are some small char/misc driver fixes for 4.6-rc4.  Full details
  are in the shortlog, nothing major here.

  These have all been in linux-next for a while with no reported issues"

* tag 'char-misc-4.6-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/char-misc:
  lkdtm: do not leak free page on kmalloc failure
  lkdtm: fix memory leak of base
  lkdtm: fix memory leak of val
  extcon: palmas: Drop stray IRQF_EARLY_RESUME flag

4 years agoMerge tag 'driver-core-4.6-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 17 Apr 2016 03:53:50 +0000 (20:53 -0700)]
Merge tag 'driver-core-4.6-rc4' of git://git./linux/kernel/git/gregkh/driver-core

Pull misc fixes from Greg KH:
 "Here are three small fixes for 4.6-rc4.

  Two fix up some lz4 issues with big endian systems, and the remaining
  one resolves a minor debugfs issue that was reported.

  All have been in linux-next with no reported issues"

* tag 'driver-core-4.6-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/driver-core:
  lib: lz4: cleanup unaligned access efficiency detection
  lib: lz4: fixed zram with lz4 on big endian machines
  debugfs: Make automount point inodes permanently empty

4 years agoMerge tag 'usb-4.6-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb
Linus Torvalds [Sun, 17 Apr 2016 03:48:14 +0000 (20:48 -0700)]
Merge tag 'usb-4.6-rc4' of git://git./linux/kernel/git/gregkh/usb

Pull USB driver fixes from Greg KH:
 "Here are some small USB fixes for 4.6-rc4.

  Mostly xhci fixes for reported issues, a UAS bug that has hit a number
  of people, including stable tree users, and a few other minor things.

  All have been in linux-next for a while with no reported issues"

* tag 'usb-4.6-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb:
  usb: hcd: out of bounds access in for_each_companion
  USB: uas: Add a new NO_REPORT_LUNS quirk
  USB: uas: Limit qdepth at the scsi-host level
  doc: usb: Fix typo in gadget_multi documentation
  usb: host: xhci-plat: Make enum xhci_plat_type start at a non zero value
  xhci: fix 10 second timeout on removal of PCI hotpluggable xhci controllers
  usb: xhci: fix wild pointers in xhci_mem_cleanup
  usb: host: xhci-plat: fix cannot work if R-Car Gen2/3 run on above 4GB phys
  usb: host: xhci: add a new quirk XHCI_NO_64BIT_SUPPORT
  xhci: resume USB 3 roothub first
  usb: xhci: applying XHCI_PME_STUCK_QUIRK to Intel BXT B0 host
  cdc-acm: fix crash if flushed with nothing buffered

4 years agoMerge tag 'dmaengine-fix-4.6-rc4' of git://git.infradead.org/users/vkoul/slave-dma
Linus Torvalds [Sat, 16 Apr 2016 22:52:38 +0000 (15:52 -0700)]
Merge tag 'dmaengine-fix-4.6-rc4' of git://git.infradead.org/users/vkoul/slave-dma

Pull dmaengine fixes from Vinod Koul:
 "This time we have some odd fixes in hsu, edma, omap and xilinx.

  Usual fixes and nothing special"

* tag 'dmaengine-fix-4.6-rc4' of git://git.infradead.org/users/vkoul/slave-dma:
  dmaengine: dw: fix master selection
  dmaengine: edma: special case slot limit workaround
  dmaengine: edma: Remove dynamic TPTC power management feature
  dmaengine: vdma: don't crash when bad channel is requested
  dmaengine: omap-dma: Do not suppress interrupts for memcpy
  dmaengine: omap-dma: Fix polled channel completion detection and handling
  dmaengine: hsu: correct use of channel status register
  dmaengine: hsu: correct residue calculation of active descriptor
  dmaengine: hsu: set HSU_CH_MTSR to memory width

4 years agoMerge branch 'locking-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sat, 16 Apr 2016 22:43:19 +0000 (15:43 -0700)]
Merge branch 'locking-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull locking fixlet from Ingo Molnar:
 "Fixes a build warning on certain Kconfig combinations"

* 'locking-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  locking/lockdep: Fix print_collision() unused warning

4 years agoMerge branch 'efi-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sat, 16 Apr 2016 22:37:05 +0000 (15:37 -0700)]
Merge branch 'efi-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull EFI fix from Ingo Molnar:
 "An arm64 boot crash fix"

* 'efi-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  efi/arm64: Don't apply MEMBLOCK_NOMAP to UEFI memory map mapping

4 years agoMerge branch 'fix/edma' into fixes
Vinod Koul [Sat, 16 Apr 2016 17:22:03 +0000 (22:52 +0530)]
Merge branch 'fix/edma' into fixes

4 years agoMerge branch 'fix/xilinx' into fixes
Vinod Koul [Sat, 16 Apr 2016 17:15:26 +0000 (22:45 +0530)]
Merge branch 'fix/xilinx' into fixes

4 years agoMerge branch 'fix/omap' into fixes
Vinod Koul [Sat, 16 Apr 2016 17:15:17 +0000 (22:45 +0530)]
Merge branch 'fix/omap' into fixes

4 years agoMerge branch 'fix/hsu' into fixes
Vinod Koul [Sat, 16 Apr 2016 17:14:32 +0000 (22:44 +0530)]
Merge branch 'fix/hsu' into fixes

4 years agoMerge branch 'for-linus' of git://git.kernel.dk/linux-block
Linus Torvalds [Fri, 15 Apr 2016 22:44:10 +0000 (15:44 -0700)]
Merge branch 'for-linus' of git://git.kernel.dk/linux-block

Pull block fixes from Jens Axboe:
 "A few fixes for the current series. This contains:

   - Two fixes for NVMe:

     One fixes a reset race that can be triggered by repeated
     insert/removal of the module.

     The other fixes an issue on some platforms, where we get probe
     timeouts since legacy interrupts isn't working.  This used not to
     be a problem since we had the worker thread poll for completions,
     but since that was killed off, it means those poor souls can't
     successfully probe their NVMe device.  Use a proper IRQ check and
     probe (msi-x -> msi ->legacy), like most other drivers to work
     around this.  Both from Keith.

   - A loop corruption issue with offset in iters, from Ming Lei.

   - A fix for not having the partition stat per cpu ref count
     initialized before sending out the KOBJ_ADD, which could cause user
     space to access the counter prior to initialization.  Also from
     Ming Lei.

   - A fix for using the wrong congestion state, from Kaixu Xia"

* 'for-linus' of git://git.kernel.dk/linux-block:
  block: loop: fix filesystem corruption in case of aio/dio
  NVMe: Always use MSI/MSI-x interrupts
  NVMe: Fix reset/remove race
  writeback: fix the wrong congested state variable definition
  block: partition: initialize percpuref before sending out KOBJ_ADD

4 years agoMerge branch 'libnvdimm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdim...
Linus Torvalds [Fri, 15 Apr 2016 22:34:27 +0000 (15:34 -0700)]
Merge branch 'libnvdimm-fixes' of git://git./linux/kernel/git/nvdimm/nvdimm

Pull libnvdimm fixes from Ross Zwisler:
 "Two fixes:

   - Fix memcpy_from_pmem() to fallback to memcpy() for architectures
     where CONFIG_ARCH_HAS_PMEM_API=n.

   - Add a comment explaining why we write data twice when clearing
     poison in pmem_do_bvec().

  This has passed a boot test on an X86_32 config, which was the
  architecture where issue #1 above was first noticed"

Dan Williams adds:
 "We're giving this multi-maintainer setup a shot, so expect libnvdimm
  pull requests from either Ross or I going forward"

* 'libnvdimm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdimm/nvdimm:
  libnvdimm, pmem: clarify the write+clear_poison+write flow
  pmem: fix BUG() error in pmem.h:48 on X86_32

4 years agoMerge tag 'for-linus-20160415' of git://git.infradead.org/linux-mtd
Linus Torvalds [Fri, 15 Apr 2016 22:25:09 +0000 (15:25 -0700)]
Merge tag 'for-linus-20160415' of git://git.infradead.org/linux-mtd

Pull MTD fix from Brian Norris:
 "One MTD fix for v4.6-rc4:

  In the v4.4 cycle, we relaxed the requirement for assigning
  mtd->owner, but we didn't remove this error case.  It's hit only
  by drivers that are both:

   (a) using nand_scan() directly
  and
   (b) built as modules

  We haven't seen explicit complaints about this (most use cases don't
  fit one or both of the above), but we should definitely not be
  BUG()'ing here"

* tag 'for-linus-20160415' of git://git.infradead.org/linux-mtd:
  mtd: nand: Drop mtd.owner requirement in nand_scan

4 years agoMerge tag 'mmc-v4.6-rc3' of git://git.linaro.org/people/ulf.hansson/mmc
Linus Torvalds [Fri, 15 Apr 2016 22:10:32 +0000 (15:10 -0700)]
Merge tag 'mmc-v4.6-rc3' of git://git.linaro.org/people/ulf.hansson/mmc

Pull MMC fixes from Ulf Hansson:
 "Here are a couple of mmc fixes intended for v4.6 rc4.

  Regarding the fix for the regression about mmcblk device indexes.  The
  approach taken to solve the problem seems to be good enough.  There
  were some discussions around the solution, but it seems like people
  were happy about it in the end.

  MMC core:
   - Restore similar old behaviour when assigning mmcblk device indexes

  MMC host:
   - tegra: Disable UHS-I modes for Tegra124 to fix regression"

* tag 'mmc-v4.6-rc3' of git://git.linaro.org/people/ulf.hansson/mmc:
  mmc: tegra: Disable UHS-I modes for Tegra124
  mmc: block: Use the mmc host device index as the mmcblk device index

4 years agoMerge branch 'drm-fixes' of git://people.freedesktop.org/~airlied/linux
Linus Torvalds [Fri, 15 Apr 2016 21:59:28 +0000 (14:59 -0700)]
Merge branch 'drm-fixes' of git://people.freedesktop.org/~airlied/linux

Pull drm fixes from Dave Airlie:
 "This contains fixes for exynos, amdgpu, radeon, i915 and qxl.

  It also contains some fixes to the core drm edid parser.

  qxl:
   - fix for a cursor hotspot issue

  radeon:
   - some MST fixes that I've been running locally and make my monitor a
     bit happier

  exynos:
   - fix some regressions and build fixes

  amdgpu:
   - a couple of small fixes

  i915:
   - two DP MST fixes and a couple of other regression fixes

  Nothing too out of the ordinary or surprising at this point"

* 'drm-fixes' of git://people.freedesktop.org/~airlied/linux:
  drm/exynos: Use VIDEO_SAMSUNG_S5P_G2D=n as G2D Kconfig dependency
  drm/exynos: fix a warning message
  drm/exynos: mic: fix an error code
  drm/exynos: fimd: fix broken dp_clock control
  drm/exynos: build fbdev code conditionally
  drm/exynos: fix adjusted_mode pointer in exynos_plane_mode_set
  drm/exynos: fix error handling in exynos_drm_subdrv_open
  drm/amd/amdgpu: fix irq domain remove for tonga ih
  drm/i915: fix deadlock on lid open
  drm/radeon: use helper for mst connector dpms.
  drm/radeon/mst: port some MST setup code from DAL.
  drm/amdgpu: add invisible pin size statistic
  drm/edid: Fix DMT 1024x768@43Hz (interlaced) timings
  drm/i915: Exit cherryview_irq_handler() after one pass
  drm/i915: Call intel_dp_mst_resume() before resuming displays
  drm/i915: Fix race condition in intel_dp_destroy_mst_connector()
  drm/edid: Fix parsing of EDID 1.4 Established Timings III descriptor
  drm/edid: Fix EDID Established Timings I and II
  drm/qxl: fix cursor position with non-zero hotspot

4 years agoMerge branch 'parisc-4.6-4' of git://git.kernel.org/pub/scm/linux/kernel/git/deller...
Linus Torvalds [Fri, 15 Apr 2016 21:51:45 +0000 (14:51 -0700)]
Merge branch 'parisc-4.6-4' of git://git./linux/kernel/git/deller/parisc-linux

Pull parisc ftrace fixes from Helge Deller:
 "This is (most likely) the last pull request for v4.6 for the parisc
  architecture.

  It fixes the FTRACE feature for parisc, which is horribly broken since
   quite some time and doesn't even compile.  This patch just fixes the
  bare minimum (it actually removes more lines than it adds), so that
  the function tracer works again on 32- and 64bit kernels.

  I've queued up additional patches on top of this patch which e.g. add
  the syscall tracer, but those have to wait for the merge window for
  v4.7."

* 'parisc-4.6-4' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/parisc-linux:
  parisc: Fix ftrace function tracer

4 years agolibnvdimm, pmem: clarify the write+clear_poison+write flow
Dan Williams [Fri, 15 Apr 2016 02:40:47 +0000 (19:40 -0700)]
libnvdimm, pmem: clarify the write+clear_poison+write flow

The ACPI specification does not specify the state of data after a clear
poison operation.  Potential future libnvdimm bus implementations for
other architectures also might not specify or disagree on the state of
data after clear poison.  Clarify why we write twice.

Reported-by: Jeff Moyer <jmoyer@redhat.com>
Reported-by: Vishal Verma <vishal.l.verma@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Jeff Moyer <jmoyer@redhat.com>
Reviewed-by: Vishal Verma <vishal.l.verma@intel.com>
4 years agoblock: loop: fix filesystem corruption in case of aio/dio
Ming Lei [Fri, 15 Apr 2016 10:51:28 +0000 (18:51 +0800)]
block: loop: fix filesystem corruption in case of aio/dio

Starting from commit e36f620428(block: split bios to max possible length),
block core starts to split bio in the middle of bvec.

Unfortunately loop dio/aio doesn't consider this situation, and
always treat 'iter.iov_offset' as zero. Then filesystem corruption
is observed.

This patch figures out the offset of the base bvevc via
'bio->bi_iter.bi_bvec_done' and fixes the issue by passing the offset
to iov iterator.

Fixes: e36f6204288088f (block: split bios to max possible length)
Cc: Keith Busch <keith.busch@intel.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: stable@vger.kernel.org (4.5)
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
4 years agoMerge branch 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Fri, 15 Apr 2016 02:53:46 +0000 (19:53 -0700)]
Merge branch 'x86-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull x86 fixes from Ingo Molnar:
 "Misc fixes: a binutils fix, an lguest fix, an mcelog fix and a missing
  documentation fix"

* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/mce: Avoid using object after free in genpool
  lguest, x86/entry/32: Fix handling of guest syscalls using interrupt gates
  x86/build: Build compressed x86 kernels as PIE
  x86/mm/pkeys: Add missing Documentation

4 years agoMerge branch 'mm-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Fri, 15 Apr 2016 02:31:34 +0000 (19:31 -0700)]
Merge branch 'mm-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull mm gup cleanup from Ingo Molnar:
 "This removes the ugly get-user-pages API hack, now that all upstream
  code has been migrated to it"

("ugly" is putting it mildly. But it worked.. - Linus)

* 'mm-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  mm/gup: Remove the macro overload API migration helpers from the get_user*() APIs

4 years agoMerge tag 'dm-4.6-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/device...
Linus Torvalds [Fri, 15 Apr 2016 02:07:45 +0000 (19:07 -0700)]
Merge tag 'dm-4.6-fixes' of git://git./linux/kernel/git/device-mapper/linux-dm

Pull device mapper fixes from Mike Snitzer:

 - fix a 4.6-rc1 bio-based DM 'struct dm_target_io' leak in an error
   path

 - stable@ fix for DM cache metadata's READ_LOCK macros that were
   incorrectly returning error if the block manager was in read-only
   mode; also cleanup multi-statement macros to use do {} while(0)

* tag 'dm-4.6-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm:
  dm cache metadata: fix READ_LOCK macros and cleanup WRITE_LOCK macros
  dm: fix dm_target_io leak if clone_bio() returns an error

4 years agoMerge tag 'pwm/for-4.6-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/thierry...
Linus Torvalds [Fri, 15 Apr 2016 02:02:06 +0000 (19:02 -0700)]
Merge tag 'pwm/for-4.6-rc4' of git://git./linux/kernel/git/thierry.reding/linux-pwm

Pull pwm fix from Thierry Reding:
 "A single one-line fix to turn the regmap cache from an RB-tree to a
  flat cache to avoid lockdep and abort issues"

* tag 'pwm/for-4.6-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/thierry.reding/linux-pwm:
  pwm: fsl-ftm: Use flat regmap cache

4 years agoMerge tag 'sound-4.6-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai...
Linus Torvalds [Fri, 15 Apr 2016 01:47:51 +0000 (18:47 -0700)]
Merge tag 'sound-4.6-rc4' of git://git./linux/kernel/git/tiwai/sound

Pull sound fixes from Takashi Iwai:
 "We've had a very calm development cycle, so far.  Here are the few
  fixes for HD-audio and USB-audio, all of which are small and easy"

* tag 'sound-4.6-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound:
  ALSA: hda - Fix inconsistent monitor_present state until repoll
  ALSA: hda - Fix regression of monitor_present flag in eld proc file
  ALSA: usb-audio: Skip volume controls triggers hangup on Dell USB Dock
  ALSA: hda/realtek - Enable the ALC292 dock fixup on the Thinkpad T460s
  ALSA: sscape: Use correct format identifier for size_t
  ALSA: usb-audio: Add a quirk for Plantronics BT300
  ALSA: usb-audio: Add a sample rate quirk for Phoenix Audio TMX320
  ALSA: hda - Bind with i915 only when Intel graphics is present

4 years agoMerge branch 'mailbox-devel' of git://git.linaro.org/landing-teams/working/fujitsu...
Linus Torvalds [Fri, 15 Apr 2016 01:40:47 +0000 (18:40 -0700)]
Merge branch 'mailbox-devel' of git://git.linaro.org/landing-teams/working/fujitsu/integration

Pull mailbox fixes from Jussi Brar:
 "Misc fixes:

  mailbox-test driver:
   - prevent memory leak and another cosmetic change

  mailbox:
   - change the returned error code

  Xgene driver:
   - return -ENOMEM instead of PTR_ERR for failed devm_kzalloc"

* 'mailbox-devel' of git://git.linaro.org/landing-teams/working/fujitsu/integration:
  mailbox: Stop using ENOSYS for anything other than unimplemented syscalls
  mailbox: mailbox-test: Prevent memory leak
  mailbox: mailbox-test: Use more consistent format for calling copy_from_user()
  mailbox: xgene-slimpro: Fix wrong test for devm_kzalloc

4 years agoMerge tag 'for-linus-4.6-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/jaegeu...
Linus Torvalds [Fri, 15 Apr 2016 01:22:42 +0000 (18:22 -0700)]
Merge tag 'for-linus-4.6-rc4' of git://git./linux/kernel/git/jaegeuk/f2fs

Pull f2fs/fscrypto fixes from Jaegeuk Kim:
 "In addition to f2fs/fscrypto fixes, I've added one patch which
  prevents RCU mode lookup in d_revalidate, as Al mentioned.

  These patches fix f2fs and fscrypto based on -rc3 bug fixes in ext4
  crypto, which have not yet been fully propagated as follows.

   - use of dget_parent and file_dentry to avoid crashes
   - disallow RCU-mode lookup in d_invalidate
   - disallow -ENOMEM in the core data encryption path"

* tag 'for-linus-4.6-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/jaegeuk/f2fs:
  ext4/fscrypto: avoid RCU lookup in d_revalidate
  fscrypto: don't let data integrity writebacks fail with ENOMEM
  f2fs: use dget_parent and file_dentry in f2fs_file_open
  fscrypto: use dget_parent() in fscrypt_d_revalidate()

4 years agoMerge branch 'linus' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6
Linus Torvalds [Fri, 15 Apr 2016 01:15:40 +0000 (18:15 -0700)]
Merge branch 'linus' of git://git./linux/kernel/git/herbert/crypto-2.6

Pull crypto fixes from Herbert Xu:
 "This fixes an NFS regression caused by the skcipher/hash conversion in
  sunrpc.  It also fixes a build problem in certain configurations with
  bcm63xx"

* 'linus' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6:
  hwrng: bcm63xx - fix device tree compilation
  sunrpc: Fix skcipher/shash conversion

4 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jmorris...
Linus Torvalds [Fri, 15 Apr 2016 01:03:29 +0000 (18:03 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/jmorris/linux-security

Pull keys bugfixes from James Morris:
 "Two bugfixes for Keys related code"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jmorris/linux-security:
  ASN.1: fix open failure check on headername
  assoc_array: don't call compare_object() on a node

4 years agodm cache metadata: fix READ_LOCK macros and cleanup WRITE_LOCK macros
Mike Snitzer [Tue, 12 Apr 2016 16:14:46 +0000 (12:14 -0400)]
dm cache metadata: fix READ_LOCK macros and cleanup WRITE_LOCK macros

The READ_LOCK macro was incorrectly returning -EINVAL if
dm_bm_is_read_only() was true -- it will always be true once the cache
metadata transitions to read-only by dm_cache_metadata_set_read_only().

Wrap READ_LOCK and WRITE_LOCK multi-statement macros in do {} while(0).
Also, all accesses of the 'cmd' argument passed to these related macros
are now encapsulated in parenthesis.

A follow-up patch can be developed to eliminate the use of macros in
favor of pure C code.  Avoiding that now given that this needs to apply
to stable@.

Reported-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Fixes: d14fcf3dd79 ("dm cache: make sure every metadata function checks fail_io")
Cc: stable@vger.kernel.org
4 years agoNVMe: Always use MSI/MSI-x interrupts
Keith Busch [Fri, 8 Apr 2016 22:09:10 +0000 (16:09 -0600)]
NVMe: Always use MSI/MSI-x interrupts

Multiple users have reported device initialization failure due the driver
not receiving legacy PCI interrupts. This is not unique to any particular
controller, but has been observed on multiple platforms.

There have been no issues reported or observed when with message signaled
interrupts, so this patch attempts to use MSI-x during initialization,
falling back to MSI. If that fails, legacy would become the default.

The setup_io_queues error handling had to change as a result: the admin
queue's msix_entry used to be initialized to the legacy IRQ. The case
where nr_io_queues is 0 would fail request_irq when setting up the admin
queue's interrupt since re-enabling MSI-x fails with 0 vectors, leaving
the admin queue's msix_entry invalid. Instead, return success immediately.

Reported-by: Tim Muhlemmer <muhlemmer@gmail.com>
Reported-by: Jon Derrick <jonathan.derrick@intel.com>
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
4 years ago/proc/iomem: only expose physical resource addresses to privileged users
Linus Torvalds [Thu, 14 Apr 2016 19:05:37 +0000 (12:05 -0700)]
/proc/iomem: only expose physical resource addresses to privileged users

In commit c4004b02f8e5b ("x86: remove the kernel code/data/bss resources
from /proc/iomem") I was hoping to remove the phyiscal kernel address
data from /proc/iomem entirely, but that had to be reverted because some
system programs actually use it.

This limits all the detailed resource information to properly
credentialed users instead.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agopci-sysfs: use proper file capability helper function
Linus Torvalds [Thu, 14 Apr 2016 19:00:21 +0000 (12:00 -0700)]
pci-sysfs: use proper file capability helper function

The PCI config access checked the file capabilities correctly, but used
the itnernal security capability check rather than the helper function
that is actually meant for that.

The security_capable() has unusual return values and is not meant to be
used elsewhere (the only other use is in the capability checking
functions that we actually intend people to use, and this odd PCI usage
really stood out when looking around the capability code.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agoMake file credentials available to the seqfile interfaces
Linus Torvalds [Thu, 14 Apr 2016 18:22:00 +0000 (11:22 -0700)]
Make file credentials available to the seqfile interfaces

A lot of seqfile users seem to be using things like %pK that uses the
credentials of the current process, but that is actually completely
wrong for filesystem interfaces.

The unix semantics for permission checking files is to check permissions
at _open_ time, not at read or write time, and that is not just a small
detail: passing off stdin/stdout/stderr to a suid application and making
the actual IO happen in privileged context is a classic exploit
technique.

So if we want to be able to look at permissions at read time, we need to
use the file open credentials, not the current ones.  Normal file
accesses can just use "f_cred" (or any of the helper functions that do
that, like file_ns_capable()), but the seqfile interfaces do not have
any such options.

It turns out that seq_file _does_ save away the user_ns information of
the file, though.  Since user_ns is just part of the full credential
information, replace that special case with saving off the cred pointer
instead, and suddenly seq_file has all the permission information it
needs.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agoRevert "x86: remove the kernel code/data/bss resources from /proc/iomem"
Linus Torvalds [Thu, 14 Apr 2016 18:18:57 +0000 (11:18 -0700)]
Revert "x86: remove the kernel code/data/bss resources from /proc/iomem"

This reverts commit c4004b02f8e5b9ce357a0bb1641756cc86962664.

Sadly, my hope that nobody would actually use the special kernel entries
in /proc/iomem were dashed by kexec.  Which reads /proc/iomem explicitly
to find the kernel base address.  Nasty.

Anyway, that means we can't do the sane and simple thing and just remove
the entries, and we'll instead have to mask them out based on permissions.

Reported-by: Zhengyu Zhang <zhezhang@redhat.com>
Reported-by: Dave Young <dyoung@redhat.com>
Reported-by: Freeman Zhang <freeman.zhang1992@gmail.com>
Reported-by: Emrah Demir <ed@abdsec.com>
Reported-by: Baoquan He <bhe@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agoparisc: Fix ftrace function tracer
Helge Deller [Wed, 13 Apr 2016 20:27:22 +0000 (22:27 +0200)]
parisc: Fix ftrace function tracer

Fix the FTRACE function tracer for 32- and 64-bit kernel.
The former code was horribly broken.

Reimplement most coding in assembly and utilize optimizations, e.g. put
mcount() and ftrace_stub() into one L1 cacheline.

Signed-off-by: Helge Deller <deller@gmx.de>
4 years agopmem: fix BUG() error in pmem.h:48 on X86_32
Toshi Kani [Wed, 13 Apr 2016 00:10:52 +0000 (18:10 -0600)]
pmem: fix BUG() error in pmem.h:48 on X86_32

After 'commit fc0c2028135c ("x86, pmem: use memcpy_mcsafe()
for memcpy_from_pmem()")', probing a PMEM device hits the BUG()
error below on X86_32 kernel.

 kernel BUG at include/linux/pmem.h:48!

memcpy_from_pmem() calls arch_memcpy_from_pmem(), which is
unimplemented since CONFIG_ARCH_HAS_PMEM_API is undefined on
X86_32.

Fix the BUG() error by adding default_memcpy_from_pmem().

Acked-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Toshi Kani <toshi.kani@hpe.com>
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
4 years agopwm: fsl-ftm: Use flat regmap cache
Stefan Agner [Thu, 21 Jan 2016 02:56:22 +0000 (18:56 -0800)]
pwm: fsl-ftm: Use flat regmap cache

Use flat regmap cache to avoid lockdep warning at probe:

[    0.697285] WARNING: CPU: 0 PID: 1 at kernel/locking/lockdep.c:2755 lockdep_trace_alloc+0x15c/0x160()
[    0.697449] DEBUG_LOCKS_WARN_ON(irqs_disabled_flags(flags))

The RB-tree regmap cache needs to allocate new space on first writes.
However, allocations in an atomic context (e.g. when a spinlock is held)
are not allowed. The function regmap_write calls map->lock, which
acquires a spinlock in the fast_io case. Since the pwm-fsl-ftm driver
uses MMIO, the regmap bus of type regmap_mmio is being used which has
fast_io set to true.

The MMIO space of the pwm-fsl-ftm driver is reasonable condense, hence
using the much faster flat regmap cache is anyway the better choice.

Signed-off-by: Stefan Agner <stefan@agner.ch>
Cc: Mark Brown <broonie@kernel.org>
Signed-off-by: Thierry Reding <thierry.reding@gmail.com>
4 years agommc: tegra: Disable UHS-I modes for Tegra124
Jon Hunter [Wed, 13 Apr 2016 14:35:56 +0000 (15:35 +0100)]
mmc: tegra: Disable UHS-I modes for Tegra124

Tegra124 has been randomly hanging during system suspend when entering
the Tegra LP1 low power state. The hang is caused by the Tegra SDHCI
driver and linked to the UHS-I tuning sequence. Disabling the UHS-I
modes for Tegra124 prevents any hangs from occurring when entering
system suspend.

Unfortunately, the tuning sequence described in the public Tegra
documentation is incomplete and on inspection of the current tuning
sequence that has been implemented is also incomplete and may cause
problems. In the short-term it is safer to disable UHS-I modes for now
and fix later because it would be too large of a change to simply patch
now. Therefore, disable UHS-I modes for Tegra124.

Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Acked-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
4 years agommc: block: Use the mmc host device index as the mmcblk device index
Ulf Hansson [Wed, 6 Apr 2016 14:12:08 +0000 (16:12 +0200)]
mmc: block: Use the mmc host device index as the mmcblk device index

Commit 520bd7a8b415 ("mmc: core: Optimize boot time by detecting cards
simultaneously") causes regressions for some platforms.

These platforms relies on fixed mmcblk device indexes, instead of
deploying the defacto standard with UUID/PARTUUID. In other words their
rootfs needs to be available at hardcoded paths, like /dev/mmcblk0p2.

Such guarantees have never been made by the kernel, but clearly the above
commit changes the behaviour. More precisely, because of that the order
changes of how cards becomes detected, so do their corresponding mmcblk
device indexes.

As the above commit significantly improves boot time for some platforms
(magnitude of seconds), let's avoid reverting this change but instead
restore the behaviour of how mmcblk device indexes becomes picked.

By using the same index for the mmcblk device as for the corresponding mmc
host device, the probe order of mmc host devices decides the index we get
for the mmcblk device.

For those platforms that suffers from a regression, one could expect that
this updated behaviour should be sufficient to meet their expectations of
"fixed" mmcblk device indexes.

Another side effect from this change, is that the same index is used for
the mmc host device, the mmcblk device and the mmc block queue. That
should clarify their relationship.

Reported-by: Peter Hurley <peter@hurleysoftware.com>
Reported-by: Laszlo Fiat <laszlo.fiat@gmail.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Fixes: 520bd7a8b415 ("mmc: core: Optimize boot time by detecting cards
simultaneously")
Cc: <stable@vger.kernel.org>
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
4 years agoMerge branch 'exynos-drm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git...
Dave Airlie [Thu, 14 Apr 2016 03:06:19 +0000 (13:06 +1000)]
Merge branch 'exynos-drm-fixes' of git://git./linux/kernel/git/daeinki/drm-exynos into drm-fixes

fix some exynos regressions.

* 'exynos-drm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/daeinki/drm-exynos:
  drm/exynos: Use VIDEO_SAMSUNG_S5P_G2D=n as G2D Kconfig dependency
  drm/exynos: fix a warning message
  drm/exynos: mic: fix an error code
  drm/exynos: fimd: fix broken dp_clock control
  drm/exynos: build fbdev code conditionally
  drm/exynos: fix adjusted_mode pointer in exynos_plane_mode_set
  drm/exynos: fix error handling in exynos_drm_subdrv_open

4 years agoMerge branch 'drm-fixes-4.6' of git://people.freedesktop.org/~agd5f/linux into drm...
Dave Airlie [Thu, 14 Apr 2016 03:05:56 +0000 (13:05 +1000)]
Merge branch 'drm-fixes-4.6' of git://people.freedesktop.org/~agd5f/linux into drm-fixes

some misc radeon fixes.

* 'drm-fixes-4.6' of git://people.freedesktop.org/~agd5f/linux:
  drm/amd/amdgpu: fix irq domain remove for tonga ih
  drm/radeon: use helper for mst connector dpms.
  drm/radeon/mst: port some MST setup code from DAL.
  drm/amdgpu: add invisible pin size statistic

4 years agoMerge tag 'sh-fixes-4.6-rc1' of git://git.libc.org/linux-sh
Linus Torvalds [Wed, 13 Apr 2016 20:02:06 +0000 (13:02 -0700)]
Merge tag 'sh-fixes-4.6-rc1' of git://git.libc.org/linux-sh

Pull arch/sh fixes from Rich Felker:
 "Fixes for two arch/sh build regressions that appeared in 4.6-rc1, one
  introduced by me, and one caused by changes elsewhere"

* tag 'sh-fixes-4.6-rc1' of git://git.libc.org/linux-sh:
  sh: fix function signature of cpu_coregroup_mask to match pointer type
  sh: fix smp-shx3 build regression from removal of arch localtimer

4 years agousb: hcd: out of bounds access in for_each_companion
Robert Dobrowolski [Thu, 24 Mar 2016 10:30:07 +0000 (03:30 -0700)]
usb: hcd: out of bounds access in for_each_companion

On BXT platform Host Controller and Device Controller figure as
same PCI device but with different device function. HCD should
not pass data to Device Controller but only to Host Controllers.
Checking if companion device is Host Controller, otherwise skip.

Cc: <stable@vger.kernel.org>
Signed-off-by: Robert Dobrowolski <robert.dobrowolski@linux.intel.com>
Acked-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agoUSB: uas: Add a new NO_REPORT_LUNS quirk
Hans de Goede [Tue, 12 Apr 2016 10:27:09 +0000 (12:27 +0200)]
USB: uas: Add a new NO_REPORT_LUNS quirk

Add a new NO_REPORT_LUNS quirk and set it for Seagate drives with
an usb-id of: 0bc2:331a, as these will fail to respond to a
REPORT_LUNS command.

Cc: stable@vger.kernel.org
Reported-and-tested-by: David Webb <djw@noc.ac.uk>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Acked-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agoUSB: uas: Limit qdepth at the scsi-host level
Hans de Goede [Tue, 12 Apr 2016 10:27:08 +0000 (12:27 +0200)]
USB: uas: Limit qdepth at the scsi-host level

Commit 64d513ac31bd ("scsi: use host wide tags by default") causes
the SCSI core to queue more commands then we can handle on devices with
multiple LUNs, limit the queue depth at the scsi-host level instead of
per slave to fix this.

BugLink: https://bugzilla.redhat.com/show_bug.cgi?id=1315013
Cc: stable@vger.kernel.org # 4.4.x and 4.5.x
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agodoc: usb: Fix typo in gadget_multi documentation
Diego Herranz [Tue, 12 Apr 2016 17:13:27 +0000 (18:13 +0100)]
doc: usb: Fix typo in gadget_multi documentation

It tries to "match" drivers for each interface (not "much").

Signed-off-by: Diego Herranz <diegoherranz@diegoherranz.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agousb: host: xhci-plat: Make enum xhci_plat_type start at a non zero value
Peter Griffin [Wed, 13 Apr 2016 16:58:44 +0000 (19:58 +0300)]
usb: host: xhci-plat: Make enum xhci_plat_type start at a non zero value

Otherwise generic-xhci and xhci-platform which have no data get wrongly
detected as XHCI_PLAT_TYPE_MARVELL_ARMADA by xhci_plat_type_is().

This fixes a regression in v4.5 for STiH407 family SoC's which use the
synopsis dwc3 IP, whereby the disable_clk error path gets taken due to
wrongly being detected as XHCI_PLAT_TYPE_MARVELL_ARMADA and the hcd never
gets added.

I suspect this will also fix other dwc3 DT platforms such as Exynos,
although I've only tested on STih410 SoC.

Fixes: 4efb2f694114 ("usb: host: xhci-plat: add struct xhci_plat_priv")
Cc: stable@vger.kernel.org
Cc: gregory.clement@free-electrons.com
Cc: yoshihiro.shimoda.uh@renesas.com
Signed-off-by: Peter Griffin <peter.griffin@linaro.org>
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agoxhci: fix 10 second timeout on removal of PCI hotpluggable xhci controllers
Mathias Nyman [Fri, 8 Apr 2016 13:25:10 +0000 (16:25 +0300)]
xhci: fix 10 second timeout on removal of PCI hotpluggable xhci controllers

PCI hotpluggable xhci controllers such as some Alpine Ridge solutions will
remove the xhci controller from the PCI bus when the last USB device is
disconnected.

Add a flag to indicate that the host is being removed to avoid queueing
configure_endpoint commands for the dropped endpoints.
For PCI hotplugged controllers this will prevent 5 second command timeouts
For static xhci controllers the configure_endpoint command is not needed
in the removal case as everything will be returned, freed, and the
controller is reset.

For now the flag is only set for PCI connected host controllers.

Cc: <stable@vger.kernel.org>
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agousb: xhci: fix wild pointers in xhci_mem_cleanup
Lu Baolu [Fri, 8 Apr 2016 13:25:09 +0000 (16:25 +0300)]
usb: xhci: fix wild pointers in xhci_mem_cleanup

This patch fixes some wild pointers produced by xhci_mem_cleanup.
These wild pointers will cause system crash if xhci_mem_cleanup()
is called twice.

Reported-and-tested-by: Pengcheng Li <lpc.li@hisilicon.com>
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
Cc: stable@vger.kernel.org
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agousb: host: xhci-plat: fix cannot work if R-Car Gen2/3 run on above 4GB phys
Yoshihiro Shimoda [Fri, 8 Apr 2016 13:25:08 +0000 (16:25 +0300)]
usb: host: xhci-plat: fix cannot work if R-Car Gen2/3 run on above 4GB phys

This patch fixes an issue that cannot work if R-Car Gen2/3 run on
above 4GB physical memory environment to use a quirk XHCI_NO_64BIT_SUPPORT.

Cc: <stable@vger.kernel.org>
Signed-off-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>
Reviewed-by: Felipe Balbi <felipe.balbi@linux.intel.com>
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agousb: host: xhci: add a new quirk XHCI_NO_64BIT_SUPPORT
Yoshihiro Shimoda [Fri, 8 Apr 2016 13:25:07 +0000 (16:25 +0300)]
usb: host: xhci: add a new quirk XHCI_NO_64BIT_SUPPORT

On some xHCI controllers (e.g. R-Car SoCs), the AC64 bit (bit 0) of
HCCPARAMS1 is set to 1. However, the xHCs don't support 64-bit
address memory pointers actually. So, in this case, this driver should
call dma_set_coherent_mask(dev, DMA_BIT_MASK(32)) in xhci_gen_setup().
Otherwise, the xHCI controller will be died after a usb device is
connected if it runs on above 4GB physical memory environment.

So, this patch adds a new quirk XHCI_NO_64BIT_SUPPORT to resolve
such an issue.

Cc: <stable@vger.kernel.org>
Signed-off-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>
Reviewed-by: Felipe Balbi <felipe.balbi@linux.intel.com>
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agoxhci: resume USB 3 roothub first
Mathias Nyman [Fri, 8 Apr 2016 13:25:06 +0000 (16:25 +0300)]
xhci: resume USB 3 roothub first

Give USB3 devices a better chance to enumerate at USB 3 speeds if
they are connected to a suspended host.
Solves an issue with NEC uPD720200 host hanging when partially
enumerating a USB3 device as USB2 after host controller runtime resume.

Cc: <stable@vger.kernel.org>
Tested-by: Mike Murdoch <main.haarp@gmail.com>
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agousb: xhci: applying XHCI_PME_STUCK_QUIRK to Intel BXT B0 host
Rafal Redzimski [Fri, 8 Apr 2016 13:25:05 +0000 (16:25 +0300)]
usb: xhci: applying XHCI_PME_STUCK_QUIRK to Intel BXT B0 host

Broxton B0 also requires XHCI_PME_STUCK_QUIRK.
Adding PCI device ID for Broxton B and adding to quirk.

Cc: <stable@vger.kernel.org>
Signed-off-by: Rafal Redzimski <rafal.f.redzimski@intel.com>
Signed-off-by: Robert Dobrowolski <robert.dobrowolski@linux.intel.com>
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agocdc-acm: fix crash if flushed with nothing buffered
Oliver Neukum [Mon, 4 Apr 2016 12:30:53 +0000 (14:30 +0200)]
cdc-acm: fix crash if flushed with nothing buffered

Under some circumstances acm_tty_flush_chars() is called
with no buffer to flush. We simply need to do nothing.

Signed-off-by: Oliver Neukum <ONeukum@suse.com>
Reported-by: Torsten Hilbrich <torsten.hilbrich@secunet.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>