Add a shutdown bit that will cause ext4 processing to fail immediately
with EIO.
Signed-off-by: NTheodore Ts'o <tytso@mit.edu>

We are currently using one bit in s_resize_flags; rename it in order
to allow more of the bits in that unsigned long for other purposes.
Signed-off-by: NTheodore Ts'o <tytso@mit.edu>

If the file system requires journal recovery, and the device is
read-ony, return EROFS to the mount system call.  This allows xfstests
generic/050 to pass.
Signed-off-by: NTheodore Ts'o <tytso@mit.edu>
Cc: stable@vger.kernel.org

If the journal is aborted, the needs_recovery feature flag should not
be removed.  Otherwise, it's the journal might not get replayed and
this could lead to more data getting lost.
Signed-off-by: NTheodore Ts'o <tytso@mit.edu>
Cc: stable@vger.kernel.org

If the journal has been aborted, we shouldn't mark the underlying
buffer head as dirty, since that will cause the metadata block to get
modified.  And if the journal has been aborted, we shouldn't allow
this since it will almost certainly lead to a corrupted file system.
Signed-off-by: NTheodore Ts'o <tytso@mit.edu>
Cc: stable@vger.kernel.org

The write_end() function must always unlock the page and drop its ref
count, even on an error.
Signed-off-by: NTheodore Ts'o <tytso@mit.edu>
Cc: stable@vger.kernel.org

The "half md4" transform should not be used by any new code. And
fortunately, it's only used now by ext4. Since ext4 supports several
hashing methods, at some point it might be desirable to move to
something like SipHash. As an intermediate step, remove half md4 from
cryptohash.h and lib, and make it just a local function in ext4's
hash.c. There's precedent for doing this; the other function ext can use
for its hashes -- TEA -- is also implemented in the same place. Also, by
being a local function, this might allow gcc to perform some additional
optimizations.
Signed-off-by: NJason A. Donenfeld <Jason@zx2c4.com>
Reviewed-by: NAndreas Dilger <adilger@dilger.ca>
Cc: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: NTheodore Ts'o <tytso@mit.edu>

In the case where the child's encryption context was inconsistent with
its parent directory, we were using inode->i_sb and inode->i_ino after
the inode had already been iput().  Fix this by doing the iput() in the
correct places.

Note: only ext4 had this bug, not f2fs and ubifs.

Fixes: d9cdc903 ("ext4 crypto: enforce context consistency")
Cc: stable@vger.kernel.org
Signed-off-by: NEric Biggers <ebiggers@google.com>
Signed-off-by: NTheodore Ts'o <tytso@mit.edu>

Below is the synchronization issue between unmount and kjournald2
contexts, which results into use after free issue in kjournald2().
Fix this issue by using journal->j_state_lock to synchronize the
wait_event() done in journal_kill_thread() and the wake_up() done
in kjournald2().

TASK 1:
umount cmd:
   |--jbd2_journal_destroy() {
       |--journal_kill_thread() {
            write_lock(&journal->j_state_lock);
	    journal->j_flags |= JBD2_UNMOUNT;
	    ...
	    write_unlock(&journal->j_state_lock);
	    wake_up(&journal->j_wait_commit);	   TASK 2 wakes up here:
	    					   kjournald2() {
						     ...
						     checks JBD2_UNMOUNT flag and calls goto end-loop;
						     ...
						     end_loop:
						       write_unlock(&journal->j_state_lock);
						       journal->j_task = NULL; --> If this thread gets
						       pre-empted here, then TASK 1 wait_event will
						       exit even before this thread is completely
						       done.
	    wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
	    ...
	    write_lock(&journal->j_state_lock);
	    write_unlock(&journal->j_state_lock);
	  }
       |--kfree(journal);
     }
}
						       wake_up(&journal->j_wait_done_commit); --> this step
						       now results into use after free issue.
						   }
Signed-off-by: NSahitya Tummala <stummala@codeaurora.org>
Signed-off-by: NTheodore Ts'o <tytso@mit.edu>

ext4_journalled_write_end() did not propely handle all the cases when
generic_perform_write() did not copy all the data into the target page
and could mark buffers with uninitialized contents as uptodate and dirty
leading to possible data corruption (which would be quickly fixed by
generic_perform_write() retrying the write but still). Fix the problem
by carefully handling the case when the page that is written to is not
uptodate.

CC: stable@vger.kernel.org
Reported-by: NAl Viro <viro@ZenIV.linux.org.uk>
Signed-off-by: NJan Kara <jack@suse.cz>
Signed-off-by: NTheodore Ts'o <tytso@mit.edu>

If filesystem groups are artifically small (using parameter -g to
mkfs.ext4), ext4_mb_normalize_request() can result in a request that is
larger than a block group. Trim the request size to not confuse
allocation code.
Reported-by: N"Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Signed-off-by: NJan Kara <jack@suse.cz>
Signed-off-by: NTheodore Ts'o <tytso@mit.edu>
Cc: stable@vger.kernel.org

The last BUG_ON in mb_find_extent() is apparently triggering in some
rare cases.  Most of the time it indicates a bug in the buddy bitmap
algorithms, but there are some weird cases where it can trigger when
buddy bitmap is still in memory, but the block bitmap has to be read
from disk, and there is disk or memory corruption such that the block
bitmap and the buddy bitmap are out of sync.

Google-Bug-Id: #33702157
Signed-off-by: NTheodore Ts'o <tytso@mit.edu>

Signed-off-by: NTheodore Ts'o <tytso@mit.edu>

There is no need to call ext4_mark_inode_dirty while holding xattr_sem
or i_data_sem, so where it's easy to avoid it, move it out from the
critical region.
Signed-off-by: NTheodore Ts'o <tytso@mit.edu>

The xattr_sem deadlock problems fixed in commit 2e81a4ee: "ext4:
avoid deadlock when expanding inode size" didn't include the use of
xattr_sem in fs/ext4/inline.c.  With the addition of project quota
which added a new extra inode field, this exposed deadlocks in the
inline_data code similar to the ones fixed by 2e81a4ee.

The deadlock can be reproduced via:

   dmesg -n 7
   mke2fs -t ext4 -O inline_data -Fq -I 256 /dev/vdc 32768
   mount -t ext4 -o debug_want_extra_isize=24 /dev/vdc /vdc
   mkdir /vdc/a
   umount /vdc
   mount -t ext4 /dev/vdc /vdc
   echo foo > /vdc/a/foo

and looks like this:

[   11.158815] 
[   11.160276] =============================================
[   11.161960] [ INFO: possible recursive locking detected ]
[   11.161960] 4.10.0-rc3-00015-g011b30a8a3cf #160 Tainted: G        W      
[   11.161960] ---------------------------------------------
[   11.161960] bash/2519 is trying to acquire lock:
[   11.161960]  (&ei->xattr_sem){++++..}, at: [<c1225a4b>] ext4_expand_extra_isize_ea+0x3d/0x4cd
[   11.161960] 
[   11.161960] but task is already holding lock:
[   11.161960]  (&ei->xattr_sem){++++..}, at: [<c1227941>] ext4_try_add_inline_entry+0x3a/0x152
[   11.161960] 
[   11.161960] other info that might help us debug this:
[   11.161960]  Possible unsafe locking scenario:
[   11.161960] 
[   11.161960]        CPU0
[   11.161960]        ----
[   11.161960]   lock(&ei->xattr_sem);
[   11.161960]   lock(&ei->xattr_sem);
[   11.161960] 
[   11.161960]  *** DEADLOCK ***
[   11.161960] 
[   11.161960]  May be due to missing lock nesting notation
[   11.161960] 
[   11.161960] 4 locks held by bash/2519:
[   11.161960]  #0:  (sb_writers#3){.+.+.+}, at: [<c11a2414>] mnt_want_write+0x1e/0x3e
[   11.161960]  #1:  (&type->i_mutex_dir_key){++++++}, at: [<c119508b>] path_openat+0x338/0x67a
[   11.161960]  #2:  (jbd2_handle){++++..}, at: [<c123314a>] start_this_handle+0x582/0x622
[   11.161960]  #3:  (&ei->xattr_sem){++++..}, at: [<c1227941>] ext4_try_add_inline_entry+0x3a/0x152
[   11.161960] 
[   11.161960] stack backtrace:
[   11.161960] CPU: 0 PID: 2519 Comm: bash Tainted: G        W       4.10.0-rc3-00015-g011b30a8a3cf #160
[   11.161960] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.10.1-1 04/01/2014
[   11.161960] Call Trace:
[   11.161960]  dump_stack+0x72/0xa3
[   11.161960]  __lock_acquire+0xb7c/0xcb9
[   11.161960]  ? kvm_clock_read+0x1f/0x29
[   11.161960]  ? __lock_is_held+0x36/0x66
[   11.161960]  ? __lock_is_held+0x36/0x66
[   11.161960]  lock_acquire+0x106/0x18a
[   11.161960]  ? ext4_expand_extra_isize_ea+0x3d/0x4cd
[   11.161960]  down_write+0x39/0x72
[   11.161960]  ? ext4_expand_extra_isize_ea+0x3d/0x4cd
[   11.161960]  ext4_expand_extra_isize_ea+0x3d/0x4cd
[   11.161960]  ? _raw_read_unlock+0x22/0x2c
[   11.161960]  ? jbd2_journal_extend+0x1e2/0x262
[   11.161960]  ? __ext4_journal_get_write_access+0x3d/0x60
[   11.161960]  ext4_mark_inode_dirty+0x17d/0x26d
[   11.161960]  ? ext4_add_dirent_to_inline.isra.12+0xa5/0xb2
[   11.161960]  ext4_add_dirent_to_inline.isra.12+0xa5/0xb2
[   11.161960]  ext4_try_add_inline_entry+0x69/0x152
[   11.161960]  ext4_add_entry+0xa3/0x848
[   11.161960]  ? __brelse+0x14/0x2f
[   11.161960]  ? _raw_spin_unlock_irqrestore+0x44/0x4f
[   11.161960]  ext4_add_nondir+0x17/0x5b
[   11.161960]  ext4_create+0xcf/0x133
[   11.161960]  ? ext4_mknod+0x12f/0x12f
[   11.161960]  lookup_open+0x39e/0x3fb
[   11.161960]  ? __wake_up+0x1a/0x40
[   11.161960]  ? lock_acquire+0x11e/0x18a
[   11.161960]  path_openat+0x35c/0x67a
[   11.161960]  ? sched_clock_cpu+0xd7/0xf2
[   11.161960]  do_filp_open+0x36/0x7c
[   11.161960]  ? _raw_spin_unlock+0x22/0x2c
[   11.161960]  ? __alloc_fd+0x169/0x173
[   11.161960]  do_sys_open+0x59/0xcc
[   11.161960]  SyS_open+0x1d/0x1f
[   11.161960]  do_int80_syscall_32+0x4f/0x61
[   11.161960]  entry_INT80_32+0x2f/0x2f
[   11.161960] EIP: 0xb76ad469
[   11.161960] EFLAGS: 00000286 CPU: 0
[   11.161960] EAX: ffffffda EBX: 08168ac8 ECX: 00008241 EDX: 000001b6
[   11.161960] ESI: b75e46bc EDI: b7755000 EBP: bfbdb108 ESP: bfbdafc0
[   11.161960]  DS: 007b ES: 007b FS: 0000 GS: 0033 SS: 007b

Cc: stable@vger.kernel.org # 3.10 (requires 2e81a4ee as a prereq)
Reported-by: NGeorge Spelvin <linux@sciencehorizons.net>
Signed-off-by: NTheodore Ts'o <tytso@mit.edu>

In order to test the inode extra isize expansion code, it is useful to
be able to easily create file systems that have inodes with extra
isize values smaller than the current desired value.
Signed-off-by: NTheodore Ts'o <tytso@mit.edu>

Inside ext4_ext_shift_extents() function ext4_find_extent() is called
without EXT4_EX_NOCACHE flag, which should prevent cache population.

This leads to oudated offsets in the extents tree and wrong blocks
afterwards.

Patch fixes the problem providing EXT4_EX_NOCACHE flag for each
ext4_find_extents() call inside ext4_ext_shift_extents function.

Fixes: 331573feSigned-off-by: NRoman Pen <roman.penyaev@profitbricks.com>
Signed-off-by: NTheodore Ts'o <tytso@mit.edu>
Cc: Namjae Jeon <namjae.jeon@samsung.com>
Cc: Andreas Dilger <adilger.kernel@dilger.ca>
Cc: stable@vger.kernel.org

While doing 'insert range' start block should be also shifted right.
The bug can be easily reproduced by the following test:

    ptr = malloc(4096);
    assert(ptr);

    fd = open("./ext4.file", O_CREAT | O_TRUNC | O_RDWR, 0600);
    assert(fd >= 0);

    rc = fallocate(fd, 0, 0, 8192);
    assert(rc == 0);
    for (i = 0; i < 2048; i++)
            *((unsigned short *)ptr + i) = 0xbeef;
    rc = pwrite(fd, ptr, 4096, 0);
    assert(rc == 4096);
    rc = pwrite(fd, ptr, 4096, 4096);
    assert(rc == 4096);

    for (block = 2; block < 1000; block++) {
            rc = fallocate(fd, FALLOC_FL_INSERT_RANGE, 4096, 4096);
            assert(rc == 0);

            for (i = 0; i < 2048; i++)
                    *((unsigned short *)ptr + i) = block;

            rc = pwrite(fd, ptr, 4096, 4096);
            assert(rc == 4096);
    }

Because start block is not included in the range the hole appears at
the wrong offset (just after the desired offset) and the following
pwrite() overwrites already existent block, keeping hole untouched.

Simple way to verify wrong behaviour is to check zeroed blocks after
the test:

   $ hexdump ./ext4.file | grep '0000 0000'

The root cause of the bug is a wrong range (start, stop], where start
should be inclusive, i.e. [start, stop].

This patch fixes the problem by including start into the range.  But
not to break left shift (range collapse) stop points to the beginning
of the a block, not to the end.

The other not obvious change is an iterator check on validness in a
main loop.  Because iterator is unsigned the following corner case
should be considered with care: insert a block at 0 offset, when stop
variables overflows and never becomes less than start, which is 0.
To handle this special case iterator is set to NULL to indicate that
end of the loop is reached.

Fixes: 331573feSigned-off-by: NRoman Pen <roman.penyaev@profitbricks.com>
Signed-off-by: NTheodore Ts'o <tytso@mit.edu>
Cc: Namjae Jeon <namjae.jeon@samsung.com>
Cc: Andreas Dilger <adilger.kernel@dilger.ca>
Cc: stable@vger.kernel.org

There was an unnecessary amount of complexity around requesting the
filesystem-specific key prefix.  It was unclear why; perhaps it was
envisioned that different instances of the same filesystem type could
use different key prefixes, or that key prefixes could be binary.
However, neither of those things were implemented or really make sense
at all.  So simplify the code by making key_prefix a const char *.
Signed-off-by: NEric Biggers <ebiggers@google.com>
Reviewed-by: NRichard Weinberger <richard@nod.at>
Signed-off-by: NTheodore Ts'o <tytso@mit.edu>

While we allow deletes without the key, the following should not be
permitted:

# cd /vdc/encrypted-dir-without-key
# ls -l
total 4
-rw-r--r-- 1 root root   0 Dec 27 22:35 6,LKNRJsp209FbXoSvJWzB
-rw-r--r-- 1 root root 286 Dec 27 22:35 uRJ5vJh9gE7vcomYMqTAyD
# mv uRJ5vJh9gE7vcomYMqTAyD  6,LKNRJsp209FbXoSvJWzB
Signed-off-by: NTheodore Ts'o <tytso@mit.edu>

max_retries _show and _store functions should test against cfg->max_retries,
not cfg->retry_timeout
Signed-off-by: NCarlos Maiolino <cmaiolino@redhat.com>
Reviewed-by: NEric Sandeen <sandeen@redhat.com>
Signed-off-by: NDarrick J. Wong <darrick.wong@oracle.com>

There is a race window between write_cache_pages calling
clear_page_dirty_for_io and XFS calling set_page_writeback, in which
the mapping for an inode is tagged neither as dirty, nor as writeback.

If the COW shrinker hits in exactly that window we'll remove the delayed
COW extents and writepages trying to write it back, which in release
kernels will manifest as corruption of the bmap btree, and in debug
kernels will trip the ASSERT about now calling xfs_bmapi_write with the
COWFORK flag for holes.  A complex customer load manages to hit this
window fairly reliably, probably by always having COW writeback in flight
while the cow shrinker runs.

This patch adds another check for having the I_DIRTY_PAGES flag set,
which is still set during this race window.  While this fixes the problem
I'm still not overly happy about the way the COW shrinker works as it
still seems a bit fragile.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NDarrick J. Wong <darrick.wong@oracle.com>

We need to use the actual AG length when making per-AG reservations,
since we could otherwise end up reserving more blocks out of the last
AG than there are actual blocks.
Complained-about-by: NBrian Foster <bfoster@redhat.com>
Signed-off-by: NDarrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>

Dan Carpenter reported a double-free of rcur if _defer_finish fails
while we're recovering CUI items.  Fix the error recovery to prevent
this.
Reported-by: NDan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: NDarrick J. Wong <darrick.wong@oracle.com>

Currently, the test_dummy_encryption ext4 mount option, which exists
only to test encrypted I/O paths with xfstests, overrides all
per-inode encryption keys with a fixed key.

This change minimizes test_dummy_encryption-specific code path changes
by supplying a fake context for directories which are not encrypted
for use when creating new directories, files, or symlinks.  This
allows us to properly exercise the keyring lookup, derivation, and
context inheritance code paths.

Before mounting a file system using test_dummy_encryption, userspace
must execute the following shell commands:

    mode='\x00\x00\x00\x00'
    raw="$(printf ""\\\\x%02x"" $(seq 0 63))"
    if lscpu | grep "Byte Order" | grep -q Little ; then
        size='\x40\x00\x00\x00'
    else
        size='\x00\x00\x00\x40'
    fi
    key="${mode}${raw}${size}"
    keyctl new_session
    echo -n -e "${key}" | keyctl padd logon fscrypt:4242424242424242 @s
Signed-off-by: NTheodore Ts'o <tytso@mit.edu>

The first block to be cleaned may start at a non-zero page offset. In
such a scenario clean_bdev_aliases() will end up cleaning blocks that
do not fall in the range of blocks to be cleaned. This commit fixes the
issue by skipping blocks that do not fall in valid block range.
Signed-off-by: NChandan Rajendra <chandan@linux.vnet.ibm.com>
Reviewed-by: NJan Kara <jack@suse.cz>
Reviewed-by: NEryu Guan <eguan@redhat.com>
Signed-off-by: NJens Axboe <axboe@fb.com>

That way we can get rid of the direct dependency on CONFIG_BLOCK.

Fixes: d475a507 ("ubifs: Add skeleton for fscrypto")
Reported-by: NArnd Bergmann <arnd@arndb.de>
Reported-by: NRandy Dunlap <rdunlap@infradead.org>
Reviewed-by: NEric Biggers <ebiggers@google.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDavid Gstir <david@sigma-star.at>
Signed-off-by: NRichard Weinberger <richard@nod.at>
Signed-off-by: NTheodore Ts'o <tytso@mit.edu>

It was possible for the ->get_context() operation to fail with a
specific error code, which was then not returned to the caller of
FS_IOC_SET_ENCRYPTION_POLICY or FS_IOC_GET_ENCRYPTION_POLICY.  Make sure
to pass through these error codes.  Also reorganize the code so that
->get_context() only needs to be called one time when setting an
encryption policy, and handle contexts of unrecognized sizes more
appropriately.
Signed-off-by: NEric Biggers <ebiggers@google.com>
Signed-off-by: NTheodore Ts'o <tytso@mit.edu>

Several warning messages were not rate limited and were user-triggerable
from FS_IOC_SET_ENCRYPTION_POLICY.  These shouldn't really have been
there in the first place, but either way they aren't as useful now that
the error codes have been improved.  So just remove them.
Signed-off-by: NEric Biggers <ebiggers@google.com>
Signed-off-by: NTheodore Ts'o <tytso@mit.edu>

As part of an effort to clean up fscrypt-related error codes, make
FS_IOC_SET_ENCRYPTION_POLICY fail with EEXIST when the file already uses
a different encryption policy.  This is more descriptive than EINVAL,
which was ambiguous with some of the other error cases.

I am not aware of any users who might be relying on the previous error
code of EINVAL, which was never documented anywhere.

This failure case will be exercised by an xfstest.
Signed-off-by: NEric Biggers <ebiggers@google.com>
Signed-off-by: NTheodore Ts'o <tytso@mit.edu>

As part of an effort to clean up fscrypt-related error codes, make
FS_IOC_SET_ENCRYPTION_POLICY fail with ENOTDIR when the file descriptor
does not refer to a directory.  This is more descriptive than EINVAL,
which was ambiguous with some of the other error cases.

I am not aware of any users who might be relying on the previous error
code of EINVAL, which was never documented anywhere, and in some buggy
kernels did not exist at all as the S_ISDIR() check was missing.

This failure case will be exercised by an xfstest.
Signed-off-by: NEric Biggers <ebiggers@google.com>
Signed-off-by: NTheodore Ts'o <tytso@mit.edu>

As part of an effort to clean up fscrypt-related error codes, make
attempting to create a file in an encrypted directory that hasn't been
"unlocked" fail with ENOKEY.  Previously, several error codes were used
for this case, including ENOENT, EACCES, and EPERM, and they were not
consistent between and within filesystems.  ENOKEY is a better choice
because it expresses that the failure is due to lacking the encryption
key.  It also matches the error code returned when trying to open an
encrypted regular file without the key.

I am not aware of any users who might be relying on the previous
inconsistent error codes, which were never documented anywhere.

This failure case will be exercised by an xfstest.
Signed-off-by: NEric Biggers <ebiggers@google.com>
Signed-off-by: NTheodore Ts'o <tytso@mit.edu>

Attempting to link a device node, named pipe, or socket file into an
encrypted directory through rename(2) or link(2) always failed with
EPERM.  This happened because fscrypt_has_permitted_context() saw that
the file was unencrypted and forbid creating the link.  This behavior
was unexpected because such files are never encrypted; only regular
files, directories, and symlinks can be encrypted.

To fix this, make fscrypt_has_permitted_context() always return true on
special files.

This will be covered by a test in my encryption xfstests patchset.

Fixes: 9bd8212f ("ext4 crypto: add encryption policy and password salt support")
Signed-off-by: NEric Biggers <ebiggers@google.com>
Reviewed-by: NRichard Weinberger <richard@nod.at>
Cc: stable@vger.kernel.org
Signed-off-by: NTheodore Ts'o <tytso@mit.edu>

Commit f1c131b4: "crypto: xts - Convert to skcipher" now fails
the setkey operation if the AES key is the same as the tweak key.
Previously this check was only done if FIPS mode is enabled.  Now this
check is also done if weak key checking was requested.  This is
reasonable, but since we were using the dummy key which was a constant
series of 0x42 bytes, it now caused dummy encrpyption test mode to
fail.

Fix this by using 0x42... and 0x24... for the two keys, so they are
different.

Fixes: f1c131b4
Cc: stable@vger.kernel.org
Signed-off-by: NTheodore Ts'o <tytso@mit.edu>

Now that dax_iomap_fault() calls ->iomap_begin() without entry lock, we
can use transaction starting in ext4_iomap_begin() and thus simplify
ext4_dax_fault(). It also provides us proper retries in case of ENOSPC.
Signed-off-by: NJan Kara <jack@suse.cz>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Currently ->iomap_begin() handler is called with entry lock held. If the
filesystem held any locks between ->iomap_begin() and ->iomap_end()
(such as ext4 which will want to hold transaction open), this would cause
lock inversion with the iomap_apply() from standard IO path which first
calls ->iomap_begin() and only then calls ->actor() callback which grabs
entry locks for DAX (if it faults when copying from/to user provided
buffers).

Fix the problem by nesting grabbing of entry lock inside ->iomap_begin()
- ->iomap_end() pair.
Reviewed-by: NRoss Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: NJan Kara <jack@suse.cz>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

The only case when we do not finish the page fault completely is when we
are loading hole pages into a radix tree. Avoid this special case and
finish the fault in that case as well inside the DAX fault handler. It
will allow us for easier iomap handling.
Reviewed-by: NRoss Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: NJan Kara <jack@suse.cz>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Currently dax_iomap_rw() takes care of invalidating page tables and
evicting hole pages from the radix tree when write(2) to the file
happens. This invalidation is only necessary when there is some block
allocation resulting from write(2). Furthermore in current place the
invalidation is racy wrt page fault instantiating a hole page just after
we have invalidated it.

So perform the page invalidation inside dax_iomap_actor() where we can
do it only when really necessary and after blocks have been allocated so
nobody will be instantiating new hole pages anymore.
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NRoss Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: NJan Kara <jack@suse.cz>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Currently invalidate_inode_pages2_range() and invalidate_mapping_pages()
just delete all exceptional radix tree entries they find. For DAX this
is not desirable as we track cache dirtiness in these entries and when
they are evicted, we may not flush caches although it is necessary. This
can for example manifest when we write to the same block both via mmap
and via write(2) (to different offsets) and fsync(2) then does not
properly flush CPU caches when modification via write(2) was the last
one.

Create appropriate DAX functions to handle invalidation of DAX entries
for invalidate_inode_pages2_range() and invalidate_mapping_pages() and
wire them up into the corresponding mm functions.
Acked-by: NJohannes Weiner <hannes@cmpxchg.org>
Reviewed-by: NRoss Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: NJan Kara <jack@suse.cz>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

So far we did not return BH_New buffers from ext2_get_blocks() when we
allocated and zeroed-out a block for DAX inode to avoid racy zeroing in
DAX code. This zeroing is gone these days so we can remove the
workaround.
Reviewed-by: NRoss Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJan Kara <jack@suse.cz>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>