If the refcount of a page is lowered between the time that it is returned
by dax_busy_page() and when the refcount is again checked in
ext4_break_layouts() => ___wait_var_event(), the waiting function
ext4_wait_dax_page() will never be called. This means that
ext4_break_layouts() will still have 'retry' set to false, so we'll stop
looping and never check the refcount of other pages in this inode.
Instead, always continue looping as long as dax_layout_busy_page() gives us
a page which it found with an elevated refcount.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@vger.kernel.org
Follow the lead of xfs_break_dax_layouts() and add synchronization between
operations in ext4 which remove blocks from an inode (hole punch, truncate
down, etc.) and pages which are pinned due to DAX DMA operations.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Lukas Czerner <lczerner@redhat.com>
Inodes using DAX should only ever have exceptional entries in their page
caches. Make this clear by warning if the iteration in
dax_layout_busy_page() ever sees a non-exceptional entry, and by adding a
comment for the pagevec_release() call which only deals with struct page
pointers.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Reviewed-by: Jan Kara <jack@suse.cz>
Jan has been developing and reviewing filesystem DAX related changes for
quite a while now, and has agreed to help maintain this code going forward.
Thanks, Jan!
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Currently device_supports_dax() just checks to see if the QUEUE_FLAG_DAX
flag is set on the device's request queue to decide whether or not the
device supports filesystem DAX. Really we should be using
bdev_dax_supported() like filesystems do at mount time. This performs
other tests like checking to make sure the dax_direct_access() path works.
We also explicitly clear QUEUE_FLAG_DAX on the DM device's request queue if
any of the underlying devices do not support DAX. This makes the handling
of QUEUE_FLAG_DAX consistent with the setting/clearing of most other flags
in dm_table_set_restrictions().
Now that bdev_dax_supported() explicitly checks for QUEUE_FLAG_DAX, this
will ensure that filesystems built upon DM devices will only be able to
mount with DAX if all underlying devices also support DAX.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Fixes: commit 545ed20e6d ("dm: add infrastructure for DAX support")
Cc: stable@vger.kernel.org
Acked-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Toshi Kani <toshi.kani@hpe.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Add an explicit check for QUEUE_FLAG_DAX to __bdev_dax_supported(). This
is needed for DM configurations where the first element in the dm-linear or
dm-stripe target supports DAX, but other elements do not. Without this
check __bdev_dax_supported() will pass for such devices, letting a
filesystem on that device mount with the DAX option.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Suggested-by: Mike Snitzer <snitzer@redhat.com>
Fixes: commit 545ed20e6d ("dm: add infrastructure for DAX support")
Cc: stable@vger.kernel.org
Acked-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Toshi Kani <toshi.kani@hpe.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
QUEUE_FLAG_DAX is an indication that a given block device supports
filesystem DAX and should not be set for PMEM namespaces which are in "raw"
mode. These namespaces lack struct page and are prevented from
participating in filesystem DAX as of commit 569d0365f5 ("dax: require
'struct page' by default for filesystem dax").
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Suggested-by: Mike Snitzer <snitzer@redhat.com>
Fixes: 569d0365f5 ("dax: require 'struct page' by default for filesystem dax")
Cc: stable@vger.kernel.org
Acked-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Toshi Kani <toshi.kani@hpe.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Use dax_write_cache() and dax_write_cache_enabled() instead of open coding
the bit operations.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
This commit:
5fdf8e5ba5 ("libnvdimm: re-enable deep flush for pmem devices via fsync()")
intended to make sure that deep flush was always available even on
platforms which support a power-fail protected CPU cache. An unintended
side effect of this change was that we also lost the ability to skip
flushing CPU caches on those power-fail protected CPU cache.
Fix this by skipping the low level cache flushing in dax_flush() if we have
CPU caches which are power-fail protected. The user can still override this
behavior by manually setting the write_cache state of a namespace. See
libndctl's ndctl_namespace_write_cache_is_enabled(),
ndctl_namespace_enable_write_cache() and
ndctl_namespace_disable_write_cache() functions.
Cc: <stable@vger.kernel.org>
Fixes: 5fdf8e5ba5 ("libnvdimm: re-enable deep flush for pmem devices via fsync()")
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Prior to this commit we would only do a "deep flush" (have nvdimm_flush()
write to each of the flush hints for a region) in response to an
msync/fsync/sync call if the nvdimm_has_cache() returned true at the time
we were setting up the request queue. This happens due to the write cache
value passed in to blk_queue_write_cache(), which then causes the block
layer to send down BIOs with REQ_FUA and REQ_PREFLUSH set. We do have a
"write_cache" sysfs entry for namespaces, i.e.:
/sys/bus/nd/devices/pfn0.1/block/pmem0/dax/write_cache
which can be used to control whether or not the kernel thinks a given
namespace has a write cache, but this didn't modify the deep flush behavior
that we set up when the driver was initialized. Instead, it only modified
whether or not DAX would flush CPU caches via dax_flush() in response to
*sync calls.
Simplify this by making the *sync deep flush always happen, regardless of
the write cache setting of a namespace. The DAX CPU cache flushing will
still be controlled the write_cache setting of the namespace.
Cc: <stable@vger.kernel.org>
Fixes: 5fdf8e5ba5 ("libnvdimm: re-enable deep flush for pmem devices via fsync()")
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Fix a race in the multi-order iteration code which causes the kernel to
hit a GP fault. This was first seen with a production v4.15 based
kernel (4.15.6-300.fc27.x86_64) utilizing a DAX workload which used
order 9 PMD DAX entries.
The race has to do with how we tear down multi-order sibling entries
when we are removing an item from the tree. Remember for example that
an order 2 entry looks like this:
struct radix_tree_node.slots[] = [entry][sibling][sibling][sibling]
where 'entry' is in some slot in the struct radix_tree_node, and the
three slots following 'entry' contain sibling pointers which point back
to 'entry.'
When we delete 'entry' from the tree, we call :
radix_tree_delete()
radix_tree_delete_item()
__radix_tree_delete()
replace_slot()
replace_slot() first removes the siblings in order from the first to the
last, then at then replaces 'entry' with NULL. This means that for a
brief period of time we end up with one or more of the siblings removed,
so:
struct radix_tree_node.slots[] = [entry][NULL][sibling][sibling]
This causes an issue if you have a reader iterating over the slots in
the tree via radix_tree_for_each_slot() while only under
rcu_read_lock()/rcu_read_unlock() protection. This is a common case in
mm/filemap.c.
The issue is that when __radix_tree_next_slot() => skip_siblings() tries
to skip over the sibling entries in the slots, it currently does so with
an exact match on the slot directly preceding our current slot.
Normally this works:
V preceding slot
struct radix_tree_node.slots[] = [entry][sibling][sibling][sibling]
^ current slot
This lets you find the first sibling, and you skip them all in order.
But in the case where one of the siblings is NULL, that slot is skipped
and then our sibling detection is interrupted:
V preceding slot
struct radix_tree_node.slots[] = [entry][NULL][sibling][sibling]
^ current slot
This means that the sibling pointers aren't recognized since they point
all the way back to 'entry', so we think that they are normal internal
radix tree pointers. This causes us to think we need to walk down to a
struct radix_tree_node starting at the address of 'entry'.
In a real running kernel this will crash the thread with a GP fault when
you try and dereference the slots in your broken node starting at
'entry'.
We fix this race by fixing the way that skip_siblings() detects sibling
nodes. Instead of testing against the preceding slot we instead look
for siblings via is_sibling_entry() which compares against the position
of the struct radix_tree_node.slots[] array. This ensures that sibling
entries are properly identified, even if they are no longer contiguous
with the 'entry' they point to.
Link: http://lkml.kernel.org/r/20180503192430.7582-6-ross.zwisler@linux.intel.com
Fixes: 148deab223 ("radix-tree: improve multiorder iterators")
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reported-by: CR, Sapthagirish <sapthagirish.cr@intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Add a test which shows a race in the multi-order iteration code. This
test reliably hits the race in under a second on my machine, and is the
result of a real bug report against kernel a production v4.15 based
kernel (4.15.6-300.fc27.x86_64). With a real kernel this issue is hit
when using order 9 PMD DAX radix tree entries.
The race has to do with how we tear down multi-order sibling entries
when we are removing an item from the tree. Remember that an order 2
entry looks like this:
struct radix_tree_node.slots[] = [entry][sibling][sibling][sibling]
where 'entry' is in some slot in the struct radix_tree_node, and the
three slots following 'entry' contain sibling pointers which point back
to 'entry.'
When we delete 'entry' from the tree, we call :
radix_tree_delete()
radix_tree_delete_item()
__radix_tree_delete()
replace_slot()
replace_slot() first removes the siblings in order from the first to the
last, then at then replaces 'entry' with NULL. This means that for a
brief period of time we end up with one or more of the siblings removed,
so:
struct radix_tree_node.slots[] = [entry][NULL][sibling][sibling]
This causes an issue if you have a reader iterating over the slots in
the tree via radix_tree_for_each_slot() while only under
rcu_read_lock()/rcu_read_unlock() protection. This is a common case in
mm/filemap.c.
The issue is that when __radix_tree_next_slot() => skip_siblings() tries
to skip over the sibling entries in the slots, it currently does so with
an exact match on the slot directly preceding our current slot.
Normally this works:
V preceding slot
struct radix_tree_node.slots[] = [entry][sibling][sibling][sibling]
^ current slot
This lets you find the first sibling, and you skip them all in order.
But in the case where one of the siblings is NULL, that slot is skipped
and then our sibling detection is interrupted:
V preceding slot
struct radix_tree_node.slots[] = [entry][NULL][sibling][sibling]
^ current slot
This means that the sibling pointers aren't recognized since they point
all the way back to 'entry', so we think that they are normal internal
radix tree pointers. This causes us to think we need to walk down to a
struct radix_tree_node starting at the address of 'entry'.
In a real running kernel this will crash the thread with a GP fault when
you try and dereference the slots in your broken node starting at
'entry'.
In the radix tree test suite this will be caught by the address
sanitizer:
==27063==ERROR: AddressSanitizer: heap-buffer-overflow on address
0x60c0008ae400 at pc 0x00000040ce4f bp 0x7fa89b8fcad0 sp 0x7fa89b8fcac0
READ of size 8 at 0x60c0008ae400 thread T3
#0 0x40ce4e in __radix_tree_next_slot /home/rzwisler/project/linux/tools/testing/radix-tree/radix-tree.c:1660
#1 0x4022cc in radix_tree_next_slot linux/../../../../include/linux/radix-tree.h:567
#2 0x4022cc in iterator_func /home/rzwisler/project/linux/tools/testing/radix-tree/multiorder.c:655
#3 0x7fa8a088d50a in start_thread (/lib64/libpthread.so.0+0x750a)
#4 0x7fa8a03bd16e in clone (/lib64/libc.so.6+0xf516e)
Link: http://lkml.kernel.org/r/20180503192430.7582-5-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: CR, Sapthagirish <sapthagirish.cr@intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Currently the lifetime of "struct item" entries in the radix tree are
not controlled by RCU, but are instead deleted inline as they are
removed from the tree.
In the following patches we add a test which has threads iterating over
items pulled from the tree and verifying them in an
rcu_read_lock()/rcu_read_unlock() section. This means that though an
item has been removed from the tree it could still be being worked on by
other threads until the RCU grace period expires. So, we need to
actually free the "struct item" structures at the end of the grace
period, just as we do with "struct radix_tree_node" items.
Link: http://lkml.kernel.org/r/20180503192430.7582-4-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: CR, Sapthagirish <sapthagirish.cr@intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Pulled from a patch from Matthew Wilcox entitled "xarray: Add definition
of struct xarray":
> From: Matthew Wilcox <mawilcox@microsoft.com>
> Signed-off-by: Matthew Wilcox <mawilcox@microsoft.com>
https://patchwork.kernel.org/patch/10341249/
These defines fix this compilation error:
In file included from ./linux/radix-tree.h:6:0,
from ./linux/../../../../include/linux/idr.h:15,
from ./linux/idr.h:1,
from idr.c:4:
./linux/../../../../include/linux/idr.h: In function `idr_init_base':
./linux/../../../../include/linux/radix-tree.h:129:2: warning: implicit declaration of function `spin_lock_init'; did you mean `spinlock_t'? [-Wimplicit-function-declaration]
spin_lock_init(&(root)->xa_lock); \
^
./linux/../../../../include/linux/idr.h:126:2: note: in expansion of macro `INIT_RADIX_TREE'
INIT_RADIX_TREE(&idr->idr_rt, IDR_RT_MARKER);
^~~~~~~~~~~~~~~
by providing a spin_lock_init() wrapper for the v4.17-rc* version of the
radix tree test suite.
Link: http://lkml.kernel.org/r/20180503192430.7582-3-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: CR, Sapthagirish <sapthagirish.cr@intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Commit c6ce3e2fe3 ("radix tree test suite: Add config option for map
shift") introduced a phony makefile target called 'mapshift' that ends
up generating the file generated/map-shift.h. This phony target was
then added as a dependency of the top level 'targets' build target,
which is what is run when you go to tools/testing/radix-tree and just
type 'make'.
Unfortunately, this phony target doesn't actually work as a dependency,
so you end up getting:
$ make
make: *** No rule to make target 'generated/map-shift.h', needed by 'main.o'. Stop.
make: *** Waiting for unfinished jobs....
Fix this by making the file generated/map-shift.h our real makefile
target, and add this a dependency of the top level build target.
Link: http://lkml.kernel.org/r/20180503192430.7582-2-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: CR, Sapthagirish <sapthagirish.cr@intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The following commit:
commit aa4d86163e ("block: loop: switch to VFS ITER_BVEC")
replaced __do_lo_send_write(), which used ITER_KVEC iterators, with
lo_write_bvec() which uses ITER_BVEC iterators. In this change, though,
the WRITE flag was lost:
- iov_iter_kvec(&from, ITER_KVEC | WRITE, &kvec, 1, len);
+ iov_iter_bvec(&i, ITER_BVEC, bvec, 1, bvec->bv_len);
This flag is necessary for the DAX case because we make decisions based on
whether or not the iterator is a READ or a WRITE in dax_iomap_actor() and
in dax_iomap_rw().
We end up going through this path in configurations where we combine a PMEM
device with 4k sectors, a loopback device and DAX. The consequence of this
missed flag is that what we intend as a write actually turns into a read in
the DAX code, so no data is ever written.
The very simplest test case is to create a loopback device and try and
write a small string to it, then hexdump a few bytes of the device to see
if the write took. Without this patch you read back all zeros, with this
you read back the string you wrote.
For XFS this causes us to fail or panic during the following xfstests:
xfs/074 xfs/078 xfs/216 xfs/217 xfs/250
For ext4 we have a similar issue where writes never happen, but we don't
currently have any xfstests that use loopback and show this issue.
Fix this by restoring the WRITE flag argument to iov_iter_bvec(). This
causes the xfstests to all pass.
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: stable@vger.kernel.org
Fixes: commit aa4d86163e ("block: loop: switch to VFS ITER_BVEC")
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
When you load nfit_test you currently see the following error in dmesg:
nfit_test nfit_test.0: found a zero length table '0' parsing nfit
This happens because when we parse the nfit_test.0 table via
acpi_nfit_init(), we specify a size of nfit_test->nfit_size. For the first
pass through nfit_test.0 where (t->setup_hotplug == 0) this is the size of
the entire buffer we allocated, including space for the hot plug
structures, not the size that we've actually filled in.
Fix this by only trying to parse the size of the structures that we've
filled in.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
It turns out that we were overrunning the 'nfit_buf' buffer in
nfit_test0_setup() in the (t->setup_hotplug == 1) case because we failed to
correctly account for all of the acpi_nfit_memory_map structures.
Fix the structure count which will increase the allocation size of
'nfit_buf' in nfit_test0_alloc(). Also add some WARN_ON()s to
nfit_test0_setup() and nfit_test1_setup() to catch future issues where the
size of the buffer doesn't match the amount of data we're writing.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
In nfit_test0_setup() and nfit_test1_setup() we keep an 'offset' value
which we use to calculate where in our 'nfit_buf' we will place our next
structure. The handling of 'offset' and the calculation of the placement
of the next structure is a bit inconsistent, though. We don't update
'offset' after we insert each structure, sometimes causing us to update it
for multiple structures' sizes at once. When calculating the position of
the next structure we aren't always able to just use 'offset', but
sometimes have to add in other structure sizes as well.
Fix this by updating 'offset' after each structure insertion in a
consistent way, allowing us to always calculate the position of the next
structure to be inserted by just using 'nfit_buf + offset'.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Now that we have the ability log filesystem writes using a flat buffer, add
support for DAX.
The motivation for this support is the need for an xfstest that can test
the new MAP_SYNC DAX flag. By logging the filesystem activity with
dm-log-writes we can show that the MAP_SYNC page faults are writing out
their metadata as they happen, instead of requiring an explicit
msync/fsync.
Unfortunately we can't easily track data that has been written via
mmap() now that the dax_flush() abstraction was removed by commit
c3ca015fab ("dax: remove the pmem_dax_ops->flush abstraction").
Otherwise we could just treat each flush as a big write, and store the
data that is being synced to media. It may be worthwhile to add the
dax_flush() entry point back, just as a notifier so we can do this
logging.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Currently dm-log-writes supports writing filesystem data via BIOs, and
writing internal metadata from a flat buffer via write_metadata().
For DAX writes, though, we won't have a BIO, but will instead have an
iterator that we'll want to use to fill a flat data buffer.
So, create write_inline_data() which allows us to write filesystem data
using a flat buffer as a source, and wire it up in log_one_block().
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
The following commit:
commit 9b7365fc1c ("ext4: add FS_IOC_FSSETXATTR/FS_IOC_FSGETXATTR
interface support")
added several defines related to extended attributes to ext4.h. They were
added within an #ifndef FS_IOC_FSGETXATTR block with the comment:
/* Until the uapi changes get merged for project quota... */
Those uapi changes were merged by this commit:
commit 334e580a6f ("fs: XFS_IOC_FS[SG]SETXATTR to FS_IOC_FS[SG]ETXATTR
promotion")
so all the definitions needed by ext4 are available in
include/uapi/linux/fs.h. Remove the duplicates from ext4.h.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Reviewed-by: Jan Kara <jack@suse.cz>
This helper, in the spirit of ext4_should_dioread_nolock() et al., replaces
the complex conditional in ext4_set_inode_flags().
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Reviewed-by: Jan Kara <jack@suse.cz>
We prevent DAX from being used on inodes which are using ext4's built in
encryption via a check in ext4_set_inode_flags(). We do have what appears
to be an unsafe transition of S_DAX in ext4_set_context(), though, where
S_DAX can get disabled without us doing a proper writeback + invalidate.
There are also issues with mm-level races when changing the value of S_DAX,
as well as issues with the VM_MIXEDMAP flag:
https://www.spinics.net/lists/linux-xfs/msg09859.html
I actually think we are safe in this case because of the following:
1) You can't encrypt an existing file. Encryption can only be set on an
empty directory, with new inodes in that directory being created with
encryption turned on, so I don't think it's possible to turn encryption on
for a file that has open DAX mmaps or outstanding I/Os.
2) There is no way to turn encryption off on a given file. Once an inode
is encrypted, it stays encrypted for the life of that inode, so we don't
have to worry about the case where we turn encryption off and S_DAX
suddenly turns on.
3) The only way we end up in ext4_set_context() to turn on encryption is
when we are creating a new file in the encrypted directory. This happens
as part of ext4_create() before the inode has been allowed to do any I/O.
Here's the call tree:
ext4_create()
__ext4_new_inode()
ext4_set_inode_flags() // sets S_DAX
fscrypt_inherit_context()
fscrypt_get_encryption_info();
ext4_set_context() // sets EXT4_INODE_ENCRYPT, clears S_DAX
So, I actually think it's safe to transition S_DAX in ext4_set_context()
without any locking, writebacks or invalidations. I've added a
WARN_ON_ONCE() sanity check to make sure that we are notified if we ever
encounter a case where we are encrypting an inode that already has data,
in which case we need to add code to safely transition S_DAX.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Reviewed-by: Jan Kara <jack@suse.cz>
The current code has the potential for data corruption when changing an
inode's journaling mode, as that can result in a subsequent unsafe change
in S_DAX.
I've captured an instance of this data corruption in the following fstest:
https://patchwork.kernel.org/patch/9948377/
Prevent this data corruption from happening by disallowing changes to the
journaling mode if the '-o dax' mount option was used. This means that for
a given filesystem we could have a mix of inodes using either DAX or
data journaling, but whatever state the inodes are in will be held for the
duration of the mount.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: stable@vger.kernel.org
If an inode has inline data it is currently prevented from using DAX by a
check in ext4_set_inode_flags(). When the inode grows inline data via
ext4_create_inline_data() or removes its inline data via
ext4_destroy_inline_data_nolock(), the value of S_DAX can change.
Currently these changes are unsafe because we don't hold off page faults
and I/O, write back dirty radix tree entries and invalidate all mappings.
There are also issues with mm-level races when changing the value of S_DAX,
as well as issues with the VM_MIXEDMAP flag:
https://www.spinics.net/lists/linux-xfs/msg09859.html
The unsafe transition of S_DAX can reliably cause data corruption, as shown
by the following fstest:
https://patchwork.kernel.org/patch/9948381/
Fix this issue by preventing the DAX mount option from being used on
filesystems that were created to support inline data. Inline data is an
option given to mkfs.ext4.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Reviewed-by: Jan Kara <jack@suse.cz>
CC: stable@vger.kernel.org
Currently only the blocksize is checked, but we should really be calling
bdev_dax_supported() which also tests to make sure we can get a
struct dax_device and that the dax_direct_access() path is working.
This is the same check that we do for the "-o dax" mount option in
xfs_fs_fill_super().
This does not fix the race issues that caused the XFS DAX inode option to
be disabled, so that option will still be disabled. If/when we re-enable
it, though, I think we will want this issue to have been fixed. I also do
think that we want to fix this in stable kernels.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
CC: stable@vger.kernel.org
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
When servicing mmap() reads from file holes the current DAX code
allocates a page cache page of all zeroes and places the struct page
pointer in the mapping->page_tree radix tree.
This has three major drawbacks:
1) It consumes memory unnecessarily. For every 4k page that is read via
a DAX mmap() over a hole, we allocate a new page cache page. This
means that if you read 1GiB worth of pages, you end up using 1GiB of
zeroed memory. This is easily visible by looking at the overall
memory consumption of the system or by looking at /proc/[pid]/smaps:
7f62e72b3000-7f63272b3000 rw-s 00000000 103:00 12 /root/dax/data
Size: 1048576 kB
Rss: 1048576 kB
Pss: 1048576 kB
Shared_Clean: 0 kB
Shared_Dirty: 0 kB
Private_Clean: 1048576 kB
Private_Dirty: 0 kB
Referenced: 1048576 kB
Anonymous: 0 kB
LazyFree: 0 kB
AnonHugePages: 0 kB
ShmemPmdMapped: 0 kB
Shared_Hugetlb: 0 kB
Private_Hugetlb: 0 kB
Swap: 0 kB
SwapPss: 0 kB
KernelPageSize: 4 kB
MMUPageSize: 4 kB
Locked: 0 kB
2) It is slower than using a common zero page because each page fault
has more work to do. Instead of just inserting a common zero page we
have to allocate a page cache page, zero it, and then insert it. Here
are the average latencies of dax_load_hole() as measured by ftrace on
a random test box:
Old method, using zeroed page cache pages: 3.4 us
New method, using the common 4k zero page: 0.8 us
This was the average latency over 1 GiB of sequential reads done by
this simple fio script:
[global]
size=1G
filename=/root/dax/data
fallocate=none
[io]
rw=read
ioengine=mmap
3) The fact that we had to check for both DAX exceptional entries and
for page cache pages in the radix tree made the DAX code more
complex.
Solve these issues by following the lead of the DAX PMD code and using a
common 4k zero page instead. As with the PMD code we will now insert a
DAX exceptional entry into the radix tree instead of a struct page
pointer which allows us to remove all the special casing in the DAX
code.
Note that we do still pretty aggressively check for regular pages in the
DAX radix tree, especially where we take action based on the bits set in
the page. If we ever find a regular page in our radix tree now that
most likely means that someone besides DAX is inserting pages (which has
happened lots of times in the past), and we want to find that out early
and fail loudly.
This solution also removes the extra memory consumption. Here is that
same /proc/[pid]/smaps after 1GiB of reading from a hole with the new
code:
7f2054a74000-7f2094a74000 rw-s 00000000 103:00 12 /root/dax/data
Size: 1048576 kB
Rss: 0 kB
Pss: 0 kB
Shared_Clean: 0 kB
Shared_Dirty: 0 kB
Private_Clean: 0 kB
Private_Dirty: 0 kB
Referenced: 0 kB
Anonymous: 0 kB
LazyFree: 0 kB
AnonHugePages: 0 kB
ShmemPmdMapped: 0 kB
Shared_Hugetlb: 0 kB
Private_Hugetlb: 0 kB
Swap: 0 kB
SwapPss: 0 kB
KernelPageSize: 4 kB
MMUPageSize: 4 kB
Locked: 0 kB
Overall system memory consumption is similarly improved.
Another major change is that we remove dax_pfn_mkwrite() from our fault
flow, and instead rely on the page fault itself to make the PTE dirty
and writeable. The following description from the patch adding the
vm_insert_mixed_mkwrite() call explains this a little more:
"To be able to use the common 4k zero page in DAX we need to have our
PTE fault path look more like our PMD fault path where a PTE entry
can be marked as dirty and writeable as it is first inserted rather
than waiting for a follow-up dax_pfn_mkwrite() =>
finish_mkwrite_fault() call.
Right now we can rely on having a dax_pfn_mkwrite() call because we
can distinguish between these two cases in do_wp_page():
case 1: 4k zero page => writable DAX storage
case 2: read-only DAX storage => writeable DAX storage
This distinction is made by via vm_normal_page(). vm_normal_page()
returns false for the common 4k zero page, though, just as it does
for DAX ptes. Instead of special casing the DAX + 4k zero page case
we will simplify our DAX PTE page fault sequence so that it matches
our DAX PMD sequence, and get rid of the dax_pfn_mkwrite() helper.
We will instead use dax_iomap_fault() to handle write-protection
faults.
This means that insert_pfn() needs to follow the lead of
insert_pfn_pmd() and allow us to pass in a 'mkwrite' flag. If
'mkwrite' is set insert_pfn() will do the work that was previously
done by wp_page_reuse() as part of the dax_pfn_mkwrite() call path"
Link: http://lkml.kernel.org/r/20170724170616.25810-4-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Andreas Dilger <adilger.kernel@dilger.ca>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When servicing mmap() reads from file holes the current DAX code
allocates a page cache page of all zeroes and places the struct page
pointer in the mapping->page_tree radix tree. This has three major
drawbacks:
1) It consumes memory unnecessarily. For every 4k page that is read via
a DAX mmap() over a hole, we allocate a new page cache page. This
means that if you read 1GiB worth of pages, you end up using 1GiB of
zeroed memory.
2) It is slower than using a common zero page because each page fault
has more work to do. Instead of just inserting a common zero page we
have to allocate a page cache page, zero it, and then insert it.
3) The fact that we had to check for both DAX exceptional entries and
for page cache pages in the radix tree made the DAX code more
complex.
This series solves these issues by following the lead of the DAX PMD
code and using a common 4k zero page instead. This reduces memory usage
and decreases latencies for some workloads, and it simplifies the DAX
code, removing over 100 lines in total.
This patch (of 5):
To be able to use the common 4k zero page in DAX we need to have our PTE
fault path look more like our PMD fault path where a PTE entry can be
marked as dirty and writeable as it is first inserted rather than
waiting for a follow-up dax_pfn_mkwrite() => finish_mkwrite_fault()
call.
Right now we can rely on having a dax_pfn_mkwrite() call because we can
distinguish between these two cases in do_wp_page():
case 1: 4k zero page => writable DAX storage
case 2: read-only DAX storage => writeable DAX storage
This distinction is made by via vm_normal_page(). vm_normal_page()
returns false for the common 4k zero page, though, just as it does for
DAX ptes. Instead of special casing the DAX + 4k zero page case we will
simplify our DAX PTE page fault sequence so that it matches our DAX PMD
sequence, and get rid of the dax_pfn_mkwrite() helper. We will instead
use dax_iomap_fault() to handle write-protection faults.
This means that insert_pfn() needs to follow the lead of
insert_pfn_pmd() and allow us to pass in a 'mkwrite' flag. If 'mkwrite'
is set insert_pfn() will do the work that was previously done by
wp_page_reuse() as part of the dax_pfn_mkwrite() call path.
Link: http://lkml.kernel.org/r/20170724170616.25810-2-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Andreas Dilger <adilger.kernel@dilger.ca>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
In DAX there are two separate places where the 2MiB range of a PMD is
defined.
The first is in the page tables, where a PMD mapping inserted for a
given address spans from (vmf->address & PMD_MASK) to ((vmf->address &
PMD_MASK) + PMD_SIZE - 1). That is, from the 2MiB boundary below the
address to the 2MiB boundary above the address.
So, for example, a fault at address 3MiB (0x30 0000) falls within the
PMD that ranges from 2MiB (0x20 0000) to 4MiB (0x40 0000).
The second PMD range is in the mapping->page_tree, where a given file
offset is covered by a radix tree entry that spans from one 2MiB aligned
file offset to another 2MiB aligned file offset.
So, for example, the file offset for 3MiB (pgoff 768) falls within the
PMD range for the order 9 radix tree entry that ranges from 2MiB (pgoff
512) to 4MiB (pgoff 1024).
This system works so long as the addresses and file offsets for a given
mapping both have the same offsets relative to the start of each PMD.
Consider the case where the starting address for a given file isn't 2MiB
aligned - say our faulting address is 3 MiB (0x30 0000), but that
corresponds to the beginning of our file (pgoff 0). Now all the PMDs in
the mapping are misaligned so that the 2MiB range defined in the page
tables never matches up with the 2MiB range defined in the radix tree.
The current code notices this case for DAX faults to storage with the
following test in dax_pmd_insert_mapping():
if (pfn_t_to_pfn(pfn) & PG_PMD_COLOUR)
goto unlock_fallback;
This test makes sure that the pfn we get from the driver is 2MiB
aligned, and relies on the assumption that the 2MiB alignment of the pfn
we get back from the driver matches the 2MiB alignment of the faulting
address.
However, faults to holes were not checked and we could hit the problem
described above.
This was reported in response to the NVML nvml/src/test/pmempool_sync
TEST5:
$ cd nvml/src/test/pmempool_sync
$ make TEST5
You can grab NVML here:
https://github.com/pmem/nvml/
The dmesg warning you see when you hit this error is:
WARNING: CPU: 13 PID: 2900 at fs/dax.c:641 dax_insert_mapping_entry+0x2df/0x310
Where we notice in dax_insert_mapping_entry() that the radix tree entry
we are about to replace doesn't match the locked entry that we had
previously inserted into the tree. This happens because the initial
insertion was done in grab_mapping_entry() using a pgoff calculated from
the faulting address (vmf->address), and the replacement in
dax_pmd_load_hole() => dax_insert_mapping_entry() is done using
vmf->pgoff.
In our failure case those two page offsets (one calculated from
vmf->address, one using vmf->pgoff) point to different order 9 radix
tree entries.
This failure case can result in a deadlock because the radix tree unlock
also happens on the pgoff calculated from vmf->address. This means that
the locked radix tree entry that we swapped in to the tree in
dax_insert_mapping_entry() using vmf->pgoff is never unlocked, so all
future faults to that 2MiB range will block forever.
Fix this by validating that the faulting address's PMD offset matches
the PMD offset from the start of the file. This check is done at the
very beginning of the fault and covers faults that would have mapped to
storage as well as faults to holes. I left the COLOUR check in
dax_pmd_insert_mapping() in place in case we ever hit the insanity
condition where the alignment of the pfn we get from the driver doesn't
match the alignment of the userspace address.
Link: http://lkml.kernel.org/r/20170822222436.18926-1-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reported-by: "Slusarz, Marcin" <marcin.slusarz@intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Right now if a file includes acpi_numa.h and they don't happen to include
linux/numa.h before it, they get the following warning:
./include/acpi/acpi_numa.h:9:5: warning: "MAX_NUMNODES" is not defined [-Wundef]
#if MAX_NUMNODES > 256
^~~~~~~~~~~~
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
We currently have two related PMD vs PTE races in the DAX code. These
can both be easily triggered by having two threads reading and writing
simultaneously to the same private mapping, with the key being that
private mapping reads can be handled with PMDs but private mapping
writes are always handled with PTEs so that we can COW.
Here is the first race:
CPU 0 CPU 1
(private mapping write)
__handle_mm_fault()
create_huge_pmd() - FALLBACK
handle_pte_fault()
passes check for pmd_devmap()
(private mapping read)
__handle_mm_fault()
create_huge_pmd()
dax_iomap_pmd_fault() inserts PMD
dax_iomap_pte_fault() does a PTE fault, but we already have a DAX PMD
installed in our page tables at this spot.
Here's the second race:
CPU 0 CPU 1
(private mapping read)
__handle_mm_fault()
passes check for pmd_none()
create_huge_pmd()
dax_iomap_pmd_fault() inserts PMD
(private mapping write)
__handle_mm_fault()
create_huge_pmd() - FALLBACK
(private mapping read)
__handle_mm_fault()
passes check for pmd_none()
create_huge_pmd()
handle_pte_fault()
dax_iomap_pte_fault() inserts PTE
dax_iomap_pmd_fault() inserts PMD,
but we already have a PTE at
this spot.
The core of the issue is that while there is isolation between faults to
the same range in the DAX fault handlers via our DAX entry locking,
there is no isolation between faults in the code in mm/memory.c. This
means for instance that this code in __handle_mm_fault() can run:
if (pmd_none(*vmf.pmd) && transparent_hugepage_enabled(vma)) {
ret = create_huge_pmd(&vmf);
But by the time we actually get to run the fault handler called by
create_huge_pmd(), the PMD is no longer pmd_none() because a racing PTE
fault has installed a normal PMD here as a parent. This is the cause of
the 2nd race. The first race is similar - there is the following check
in handle_pte_fault():
} else {
/* See comment in pte_alloc_one_map() */
if (pmd_devmap(*vmf->pmd) || pmd_trans_unstable(vmf->pmd))
return 0;
So if a pmd_devmap() PMD (a DAX PMD) has been installed at vmf->pmd, we
will bail and retry the fault. This is correct, but there is nothing
preventing the PMD from being installed after this check but before we
actually get to the DAX PTE fault handlers.
In my testing these races result in the following types of errors:
BUG: Bad rss-counter state mm:ffff8800a817d280 idx:1 val:1
BUG: non-zero nr_ptes on freeing mm: 15
Fix this issue by having the DAX fault handlers verify that it is safe
to continue their fault after they have taken an entry lock to block
other racing faults.
[ross.zwisler@linux.intel.com: improve fix for colliding PMD & PTE entries]
Link: http://lkml.kernel.org/r/20170526195932.32178-1-ross.zwisler@linux.intel.com
Link: http://lkml.kernel.org/r/20170522215749.23516-2-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reported-by: Pawel Lebioda <pawel.lebioda@intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: "Kirill A . Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Pawel Lebioda <pawel.lebioda@intel.com>
Cc: Dave Jiang <dave.jiang@intel.com>
Cc: Xiong Zhou <xzhou@redhat.com>
Cc: Eryu Guan <eguan@redhat.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When the pmd_devmap() checks were added by 5c7fb56e5e ("mm, dax:
dax-pmd vs thp-pmd vs hugetlbfs-pmd") to add better support for DAX huge
pages, they were all added to the end of if() statements after existing
pmd_trans_huge() checks. So, things like:
- if (pmd_trans_huge(*pmd))
+ if (pmd_trans_huge(*pmd) || pmd_devmap(*pmd))
When further checks were added after pmd_trans_unstable() checks by
commit 7267ec008b ("mm: postpone page table allocation until we have
page to map") they were also added at the end of the conditional:
+ if (pmd_trans_unstable(fe->pmd) || pmd_devmap(*fe->pmd))
This ordering is fine for pmd_trans_huge(), but doesn't work for
pmd_trans_unstable(). This is because DAX huge pages trip the bad_pmd()
check inside of pmd_none_or_trans_huge_or_clear_bad() (called by
pmd_trans_unstable()), which prints out a warning and returns 1. So, we
do end up doing the right thing, but only after spamming dmesg with
suspicious looking messages:
mm/pgtable-generic.c:39: bad pmd ffff8808daa49b88(84000001006000a5)
Reorder these checks in a helper so that pmd_devmap() is checked first,
avoiding the error messages, and add a comment explaining why the
ordering is important.
Fixes: commit 7267ec008b ("mm: postpone page table allocation until we have page to map")
Link: http://lkml.kernel.org/r/20170522215749.23516-1-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Pawel Lebioda <pawel.lebioda@intel.com>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: "Kirill A . Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Dave Jiang <dave.jiang@intel.com>
Cc: Xiong Zhou <xzhou@redhat.com>
Cc: Eryu Guan <eguan@redhat.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This is based on a patch from Jan Kara that fixed the equivalent race in
the DAX PTE fault path.
Currently DAX PMD read fault can race with write(2) in the following
way:
CPU1 - write(2) CPU2 - read fault
dax_iomap_pmd_fault()
->iomap_begin() - sees hole
dax_iomap_rw()
iomap_apply()
->iomap_begin - allocates blocks
dax_iomap_actor()
invalidate_inode_pages2_range()
- there's nothing to invalidate
grab_mapping_entry()
- we add huge zero page to the radix tree
and map it to page tables
The result is that hole page is mapped into page tables (and thus zeros
are seen in mmap) while file has data written in that place.
Fix the problem by locking exception entry before mapping blocks for the
fault. That way we are sure invalidate_inode_pages2_range() call for
racing write will either block on entry lock waiting for the fault to
finish (and unmap stale page tables after that) or read fault will see
already allocated blocks by write(2).
Fixes: 9f141d6ef6 ("dax: Call ->iomap_begin without entry lock during dax fault")
Link: http://lkml.kernel.org/r/20170510172700.18991-1-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Patch series "mm,dax: Fix data corruption due to mmap inconsistency",
v4.
This series fixes data corruption that can happen for DAX mounts when
page faults race with write(2) and as a result page tables get out of
sync with block mappings in the filesystem and thus data seen through
mmap is different from data seen through read(2).
The series passes testing with t_mmap_stale test program from Ross and
also other mmap related tests on DAX filesystem.
This patch (of 4):
dax_invalidate_mapping_entry() currently removes DAX exceptional entries
only if they are clean and unlocked. This is done via:
invalidate_mapping_pages()
invalidate_exceptional_entry()
dax_invalidate_mapping_entry()
However, for page cache pages removed in invalidate_mapping_pages()
there is an additional criteria which is that the page must not be
mapped. This is noted in the comments above invalidate_mapping_pages()
and is checked in invalidate_inode_page().
For DAX entries this means that we can can end up in a situation where a
DAX exceptional entry, either a huge zero page or a regular DAX entry,
could end up mapped but without an associated radix tree entry. This is
inconsistent with the rest of the DAX code and with what happens in the
page cache case.
We aren't able to unmap the DAX exceptional entry because according to
its comments invalidate_mapping_pages() isn't allowed to block, and
unmap_mapping_range() takes a write lock on the mapping->i_mmap_rwsem.
Since we essentially never have unmapped DAX entries to evict from the
radix tree, just remove dax_invalidate_mapping_entry().
Fixes: c6dcf52c23 ("mm: Invalidate DAX radix tree entries only if appropriate")
Link: http://lkml.kernel.org/r/20170510085419.27601-2-jack@suse.cz
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Reported-by: Jan Kara <jack@suse.cz>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: <stable@vger.kernel.org> [4.10+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Add a tracepoint to dax_insert_mapping(), following the same logging
conventions as the rest of DAX. This tracepoint, along with the one in
dax_load_hole(), lets us know how a DAX PTE fault was serviced.
Here is an example DAX fault that inserts a PTE mapping:
small-1126 [007] ....
145.451604: dax_pte_fault: dev 259:0 ino 0x1003 shared WRITE|ALLOW_RETRY|KILLABLE|USER address 0x10420000 pgoff 0x220
small-1126 [007] ....
145.452317: dax_insert_mapping: dev 259:0 ino 0x1003 shared write address 0x10420000 radix_entry 0x100006
small-1126 [007] ....
145.452399: dax_pte_fault_done: dev 259:0 ino 0x1003 shared WRITE|ALLOW_RETRY|KILLABLE|USER address 0x10420000 pgoff 0x220 MAJOR|NOPAGE
Link: http://lkml.kernel.org/r/20170221195116.13278-7-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Add a tracepoint to dax_writeback_one(), following the same logging
conventions as the rest of DAX.
Here is an example range writeback which ends up flushing one PMD and
one PTE:
test-1265 [003] ....
496.615250: dax_writeback_range: dev 259:0 ino 0x1003 pgoff 0x0-0x7ffffffffffff
test-1265 [003] ....
496.616263: dax_writeback_one: dev 259:0 ino 0x1003 pgoff 0x0 pglen 0x200
test-1265 [003] ....
496.616270: dax_writeback_one: dev 259:0 ino 0x1003 pgoff 0x305 pglen 0x1
test-1265 [003] ....
496.616272: dax_writeback_range_done: dev 259:0 ino 0x1003 pgoff 0x0-0x7ffffffffffff
[akpm@linux-foundation.org: struct blk_dax_ctl has disappeared]
Link: http://lkml.kernel.org/r/20170221195116.13278-6-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Add tracepoints to dax_pfn_mkwrite(), following the same logging
conventions as the rest of DAX.
Here is an example PTE fault followed by a pfn_mkwrite:
small_aligned-1094 [002] ....
374.084998: dax_pte_fault: dev 259:0 ino 0x1003 shared WRITE|ALLOW_RETRY|KILLABLE|USER address 0x10400000 pgoff 0x200
small_aligned-1094 [002] ....
374.085145: dax_pte_fault_done: dev 259:0 ino 0x1003 shared WRITE|ALLOW_RETRY|KILLABLE|USER address 0x10400000 pgoff 0x200 MAJOR|NOPAGE
small_aligned-1094 [002] ....
374.085165: dax_pfn_mkwrite: dev 259:0 ino 0x1003 shared WRITE|MKWRITE|ALLOW_RETRY|KILLABLE|USER address 0x10400000 pgoff 0x200 NOPAGE
Link: http://lkml.kernel.org/r/20170221195116.13278-3-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Patch series "second round of tracepoints for DAX".
This second round of DAX tracepoint patches adds tracing to the PTE
fault path (dax_iomap_pte_fault(), dax_pfn_mkwrite(), dax_load_hole(),
dax_insert_mapping()) and to the writeback path
(dax_writeback_mapping_range(), dax_writeback_one()).
The purpose of this tracing is to give us a high level view of what DAX
is doing, whether faults are being serviced by PMDs or PTEs, and by real
storage or by zero pages covering holes.
I do have some patches nearly ready which also add tracing to
grab_mapping_entry() and dax_insert_mapping_entry(). These are more
targeted at logging how we are interacting with the radix tree, how we
use empty entries for locking, whether we "downgrade" huge zero pages to
4k PTE sized allocations, etc. In the end it seemed to me that this
might be too detailed to have as constantly present tracepoints, but if
anyone sees value in having tracepoints like this in the DAX code
permanently (Jan?), please let me know and I'll add those last two
patches.
All these tracepoints were done to be consistent with the style of the
XFS tracepoints and with the existing DAX PMD tracepoints.
This patch (of 6):
Add tracepoints to dax_iomap_pte_fault(), following the same logging
conventions as the rest of DAX.
Here is an example fault that initially tries to be serviced by the PMD
fault handler but which falls back to PTEs because the VMA isn't large
enough to hold a PMD:
small-1086 [005] ....
71.140014: xfs_filemap_huge_fault: dev 259:0 ino 0x1003
small-1086 [005] ....
71.140027: dax_pmd_fault: dev 259:0 ino 0x1003 shared WRITE|ALLOW_RETRY|KILLABLE|USER address 0x10420000 vm_start 0x10200000 vm_end 0x10500000 pgoff 0x220 max_pgoff 0x1400
small-1086 [005] ....
71.140028: dax_pmd_fault_done: dev 259:0 ino 0x1003 shared WRITE|ALLOW_RETRY|KILLABLE|USER address 0x10420000 vm_start 0x10200000 vm_end 0x10500000 pgoff 0x220 max_pgoff 0x1400 FALLBACK
small-1086 [005] ....
71.140035: dax_pte_fault: dev 259:0 ino 0x1003 shared WRITE|ALLOW_RETRY|KILLABLE|USER address 0x10420000 pgoff 0x220
small-1086 [005] ....
71.140396: dax_pte_fault_done: dev 259:0 ino 0x1003 shared WRITE|ALLOW_RETRY|KILLABLE|USER address 0x10420000 pgoff 0x220 MAJOR|NOPAGE
Link: http://lkml.kernel.org/r/20170221195116.13278-2-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
While running generic/340 in my test setup I hit the following race. It
can happen with kernels that support FS DAX PMDs, so v4.10 thru
v4.11-rc5.
Thread 1 Thread 2
-------- --------
dax_iomap_pmd_fault()
grab_mapping_entry()
spin_lock_irq()
get_unlocked_mapping_entry()
'entry' is NULL, can't call lock_slot()
spin_unlock_irq()
radix_tree_preload()
dax_iomap_pmd_fault()
grab_mapping_entry()
spin_lock_irq()
get_unlocked_mapping_entry()
...
lock_slot()
spin_unlock_irq()
dax_pmd_insert_mapping()
<inserts a PMD mapping>
spin_lock_irq()
__radix_tree_insert() fails with -EEXIST
<fall back to 4k fault, and die horribly
when inserting a 4k entry where a PMD exists>
The issue is that we have to drop mapping->tree_lock while calling
radix_tree_preload(), but since we didn't have a radix tree entry to
lock (unlike in the pmd_downgrade case) we have no protection against
Thread 2 coming along and inserting a PMD at the same index. For 4k
entries we handled this with a special-case response to -EEXIST coming
from the __radix_tree_insert(), but this doesn't save us for PMDs
because the -EEXIST case can also mean that we collided with a 4k entry
in the radix tree at a different index, but one that is covered by our
PMD range.
So, correctly handle both the 4k and 2M collision cases by explicitly
re-checking the radix tree for an entry at our index once we reacquire
mapping->tree_lock.
This patch has made it through a clean xfstests run with the current
v4.11-rc5 based linux/master, and it also ran generic/340 500 times in a
loop. It used to fail within the first 10 iterations.
Link: http://lkml.kernel.org/r/20170406212944.2866-1-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: <stable@vger.kernel.org> [4.10+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Tracepoints are the standard way to capture debugging and tracing
information in many parts of the kernel, including the XFS and ext4
filesystems. Create a tracepoint header for FS DAX and add the first DAX
tracepoints to the PMD fault handler. This allows the tracing for DAX to
be done in the same way as the filesystem tracing so that developers can
look at them together and get a coherent idea of what the system is doing.
I added both an entry and exit tracepoint because future patches will add
tracepoints to child functions of dax_iomap_pmd_fault() like
dax_pmd_load_hole() and dax_pmd_insert_mapping(). We want those messages
to be wrapped by the parent function tracepoints so the code flow is more
easily understood. Having entry and exit tracepoints for faults also
allows us to easily see what filesystems functions were called during the
fault. These filesystem functions get executed via iomap_begin() and
iomap_end() calls, for example, and will have their own tracepoints.
For PMD faults we primarily want to understand the type of mapping, the
fault flags, the faulting address and whether it fell back to 4k faults.
If it fell back to 4k faults the tracepoints should let us understand why.
I named the new tracepoint header file "fs_dax.h" to allow for device DAX
to have its own separate tracing header in the same directory at some
point.
Here is an example output for these events from a successful PMD fault:
big-1441 [005] .... 32.582758: xfs_filemap_pmd_fault: dev 259:0 ino 0x1003
big-1441 [005] .... 32.582776: dax_pmd_fault: dev 259:0 ino 0x1003
shared WRITE|ALLOW_RETRY|KILLABLE|USER address 0x10505000 vm_start 0x10200000 vm_end 0x10700000 pgoff 0x200 max_pgoff 0x1400
big-1441 [005] .... 32.583292: dax_pmd_fault_done: dev 259:0 ino 0x1003
shared WRITE|ALLOW_RETRY|KILLABLE|USER address 0x10505000 vm_start 0x10200000 vm_end 0x10700000 pgoff 0x200 max_pgoff 0x1400 NOPAGE
Link: http://lkml.kernel.org/r/1484085142-2297-3-git-send-email-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Suggested-by: Dave Chinner <david@fromorbit.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Cc: Dave Jiang <dave.jiang@intel.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Patch series "DAX tracepoints, mm argument simplification", v4.
This contains both my DAX tracepoint code and Dave Jiang's MM argument
simplifications. Dave's code was written with my tracepoint code as a
baseline, so it seemed simplest to keep them together in a single series.
This patch (of 7):
Add __print_flags_u64() and the helper trace_print_flags_seq_u64() in the
same spirit as __print_symbolic_u64() and trace_print_symbols_seq_u64().
These functions allow us to print symbols associated with flags that are
64 bits wide even on 32 bit machines.
These will be used by the DAX code so that we can print the flags set in a
pfn_t such as PFN_SG_CHAIN, PFN_SG_LAST, PFN_DEV and PFN_MAP.
Without this new function I was getting errors like the following when
compiling for i386:
include/linux/pfn_t.h:13:22: warning: large integer implicitly truncated to unsigned type [-Woverflow]
#define PFN_SG_CHAIN (1ULL << (BITS_PER_LONG_LONG - 1))
^
Link: http://lkml.kernel.org/r/1484085142-2297-2-git-send-email-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Steven Rostedt <rostedt@goodmis.org>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Dave Jiang <dave.jiang@intel.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
As reported by Arnd:
https://lkml.org/lkml/2017/1/10/756
Compiling with the following configuration:
# CONFIG_EXT2_FS is not set
# CONFIG_EXT4_FS is not set
# CONFIG_XFS_FS is not set
# CONFIG_FS_IOMAP depends on the above filesystems, as is not set
CONFIG_FS_DAX=y
generates build warnings about unused functions in fs/dax.c:
fs/dax.c:878:12: warning: `dax_insert_mapping' defined but not used [-Wunused-function]
static int dax_insert_mapping(struct address_space *mapping,
^~~~~~~~~~~~~~~~~~
fs/dax.c:572:12: warning: `copy_user_dax' defined but not used [-Wunused-function]
static int copy_user_dax(struct block_device *bdev, sector_t sector, size_t size,
^~~~~~~~~~~~~
fs/dax.c:542:12: warning: `dax_load_hole' defined but not used [-Wunused-function]
static int dax_load_hole(struct address_space *mapping, void **entry,
^~~~~~~~~~~~~
fs/dax.c:312:14: warning: `grab_mapping_entry' defined but not used [-Wunused-function]
static void *grab_mapping_entry(struct address_space *mapping, pgoff_t index,
^~~~~~~~~~~~~~~~~~
Now that the struct buffer_head based DAX fault paths and I/O path have
been removed we really depend on iomap support being present for DAX.
Make this explicit by selecting FS_IOMAP if we compile in DAX support.
This allows us to remove conditional selections of FS_IOMAP when FS_DAX
was present for ext2 and ext4, and to remove an #ifdef in fs/dax.c.
Link: http://lkml.kernel.org/r/1484087383-29478-1-git-send-email-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reported-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Currently dax_mapping_entry_mkclean() fails to clean and write protect
the pmd_t of a DAX PMD entry during an *sync operation. This can result
in data loss in the following sequence:
1) mmap write to DAX PMD, dirtying PMD radix tree entry and making the
pmd_t dirty and writeable
2) fsync, flushing out PMD data and cleaning the radix tree entry. We
currently fail to mark the pmd_t as clean and write protected.
3) more mmap writes to the PMD. These don't cause any page faults since
the pmd_t is dirty and writeable. The radix tree entry remains clean.
4) fsync, which fails to flush the dirty PMD data because the radix tree
entry was clean.
5) crash - dirty data that should have been fsync'd as part of 4) could
still have been in the processor cache, and is lost.
Fix this by marking the pmd_t clean and write protected in
dax_mapping_entry_mkclean(), which is called as part of the fsync
operation 2). This will cause the writes in step 3) above to generate
page faults where we'll re-dirty the PMD radix tree entry, resulting in
flushes in the fsync that happens in step 4).
Fixes: 4b4bb46d00 ("dax: clear dirty entry tags on cache flush")
Link: http://lkml.kernel.org/r/1482272586-21177-3-git-send-email-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Currently in DAX if we have three read faults on the same hole address we
can end up with the following:
Thread 0 Thread 1 Thread 2
-------- -------- --------
dax_iomap_fault
grab_mapping_entry
lock_slot
<locks empty DAX entry>
dax_iomap_fault
grab_mapping_entry
get_unlocked_mapping_entry
<sleeps on empty DAX entry>
dax_iomap_fault
grab_mapping_entry
get_unlocked_mapping_entry
<sleeps on empty DAX entry>
dax_load_hole
find_or_create_page
...
page_cache_tree_insert
dax_wake_mapping_entry_waiter
<wakes one sleeper>
__radix_tree_replace
<swaps empty DAX entry with 4k zero page>
<wakes>
get_page
lock_page
...
put_locked_mapping_entry
unlock_page
put_page
<sleeps forever on the DAX
wait queue>
The crux of the problem is that once we insert a 4k zero page, all
locking from then on is done in terms of that 4k zero page and any
additional threads sleeping on the empty DAX entry will never be woken.
Fix this by waking all sleepers when we replace the DAX radix tree entry
with a 4k zero page. This will allow all sleeping threads to
successfully transition from locking based on the DAX empty entry to
locking on the 4k zero page.
With the test case reported by Xiong this happens very regularly in my
test setup, with some runs resulting in 9+ threads in this deadlocked
state. With this fix I've been able to run that same test dozens of
times in a loop without issue.
Fixes: ac401cc782 ("dax: New fault locking")
Link: http://lkml.kernel.org/r/1483479365-13607-1-git-send-email-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reported-by: Xiong Zhou <xzhou@redhat.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: <stable@vger.kernel.org> [4.7+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The last user of ext4_aligned_io() was the DAX path in
ext4_direct_IO_write(). This usage was removed by Jan Kara's patch
entitled "ext4: Rip out DAX handling from direct IO path".
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Now that DAX PMD faults are once again working and are now participating in
DAX's radix tree locking scheme, allow their config option to be enabled.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Switch xfs_filemap_pmd_fault() from using dax_pmd_fault() to the new and
improved dax_iomap_pmd_fault(). Also, now that it has no more users,
remove xfs_get_blocks_dax_fault().
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
DAX PMDs have been disabled since Jan Kara introduced DAX radix tree based
locking. This patch allows DAX PMDs to participate in the DAX radix tree
based locking scheme so that they can be re-enabled using the new struct
iomap based fault handlers.
There are currently three types of DAX 4k entries: 4k zero pages, 4k DAX
mappings that have an associated block allocation, and 4k DAX empty
entries. The empty entries exist to provide locking for the duration of a
given page fault.
This patch adds three equivalent 2MiB DAX entries: Huge Zero Page (HZP)
entries, PMD DAX entries that have associated block allocations, and 2 MiB
DAX empty entries.
Unlike the 4k case where we insert a struct page* into the radix tree for
4k zero pages, for HZP we insert a DAX exceptional entry with the new
RADIX_DAX_HZP flag set. This is because we use a single 2 MiB zero page in
every 2MiB hole mapping, and it doesn't make sense to have that same struct
page* with multiple entries in multiple trees. This would cause contention
on the single page lock for the one Huge Zero Page, and it would break the
page->index and page->mapping associations that are assumed to be valid in
many other places in the kernel.
One difficult use case is when one thread is trying to use 4k entries in
radix tree for a given offset, and another thread is using 2 MiB entries
for that same offset. The current code handles this by making the 2 MiB
user fall back to 4k entries for most cases. This was done because it is
the simplest solution, and because the use of 2MiB pages is already
opportunistic.
If we were to try to upgrade from 4k pages to 2MiB pages for a given range,
we run into the problem of how we lock out 4k page faults for the entire
2MiB range while we clean out the radix tree so we can insert the 2MiB
entry. We can solve this problem if we need to, but I think that the cases
where both 2MiB entries and 4K entries are being used for the same range
will be rare enough and the gain small enough that it probably won't be
worth the complexity.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
No functional change.
The static functions put_locked_mapping_entry() and
put_unlocked_mapping_entry() will soon be used in error cases in
grab_mapping_entry(), so move their definitions above this function.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
The RADIX_DAX_* defines currently mostly live in fs/dax.c, with just
RADIX_DAX_ENTRY_LOCK being in include/linux/dax.h so it can be used in
mm/filemap.c. When we add PMD support, though, mm/filemap.c will also need
access to the RADIX_DAX_PTE type so it can properly construct a 4k sized
empty entry.
Instead of shifting the defines between dax.c and dax.h as they are
individually used in other code, just move them wholesale to dax.h so
they'll be available when we need them.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Currently iomap_end() doesn't do anything for DAX page faults for both ext2
and XFS. ext2_iomap_end() just checks for a write underrun, and
xfs_file_iomap_end() checks to see if it needs to finish a delayed
allocation. However, in the future iomap_end() calls might be needed to
make sure we have balanced allocations, locks, etc. So, add calls to
iomap_end() with appropriate error handling to dax_iomap_fault().
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Suggested-by: Jan Kara <jack@suse.cz>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
To be able to correctly calculate the sector from a file position and a
struct iomap there is a complex little bit of logic that currently happens
in both dax_iomap_actor() and dax_iomap_fault(). This will need to be
repeated yet again in the DAX PMD fault handler when it is added, so break
it out into a helper function.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
The recently added DAX functions that use the new struct iomap data
structure were named iomap_dax_rw(), iomap_dax_fault() and
iomap_dax_actor(). These are actually defined in fs/dax.c, though, so
should be part of the "dax" namespace and not the "iomap" namespace.
Rename them to dax_iomap_rw(), dax_iomap_fault() and dax_iomap_actor()
respectively.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Suggested-by: Dave Chinner <david@fromorbit.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
dax_pmd_fault() is the old struct buffer_head + get_block_t based 2 MiB DAX
fault handler. This fault handler has been disabled for several kernel
releases, and support for PMDs will be reintroduced using the struct iomap
interface instead.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
DAX radix tree locking currently locks entries based on the unique
combination of the 'mapping' pointer and the pgoff_t 'index' for the entry.
This works for PTEs, but as we move to PMDs we will need to have all the
offsets within the range covered by the PMD to map to the same bit lock.
To accomplish this, for ranges covered by a PMD entry we will instead lock
based on the page offset of the beginning of the PMD entry. The 'mapping'
pointer is still used in the same way.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
No functional change.
Consistently use the variable name 'entry' instead of 'ret' for DAX radix
tree entries. This was already happening in most of the code, so update
get_unlocked_mapping_entry(), grab_mapping_entry() and
dax_unlock_mapping_entry().
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Don't take down the kernel if we get an invalid 'from' and 'length'
argument pair. Just warn once and return an error.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
The global 'wait_table' variable is only used within fs/dax.c, and
generates the following sparse warning:
fs/dax.c:39:19: warning: symbol 'wait_table' was not declared. Should it be static?
Make it static so it has scope local to fs/dax.c, and to make sparse happy.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
DAX PMD support was added via the following commit:
commit e7b1ea2ad6 ("ext2: huge page fault support")
I believe this path to be untested as ext2 doesn't reliably provide block
allocations that are aligned to 2MiB. In my testing I've been unable to
get ext2 to actually fault in a PMD. It always fails with a "pfn
unaligned" message because the sector returned by ext2_get_block() isn't
aligned.
I've tried various settings for the "stride" and "stripe_width" extended
options to mkfs.ext2, without any luck.
Since we can't reliably get PMDs, remove support so that we don't have an
untested code path that we may someday traverse when we happen to get an
aligned block allocation. This should also make 4k DAX faults in ext2 a
bit faster since they will no longer have to call the PMD fault handler
only to get a response of VM_FAULT_FALLBACK.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Now that ext4 properly sets bh.b_size when we call get_block() for a hole,
rely on that value and remove the buffer_size_valid() sanity check.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
When DAX calls _ext4_get_block() and the file offset points to a hole we
currently don't set bh->b_size. This is current worked around via
buffer_size_valid() in fs/dax.c.
_ext4_get_block() has the hole size information from ext4_map_blocks(), so
populate bh->b_size so we can remove buffer_size_valid() in a later patch.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
There are four cases I can see where we could end up with a NULL 'slot' in
radix_tree_next_slot(). This unit test exercises all four of them, making
sure that if in the future we have an unsafe path through
radix_tree_next_slot(), we'll catch it.
Here are details on the four cases:
1) radix_tree_iter_retry() via a non-tagged iteration like
radix_tree_for_each_slot(). In this case we currently aren't seeing a bug
because radix_tree_iter_retry() sets
iter->next_index = iter->index;
which means that in in the else case in radix_tree_next_slot(), 'count' is
zero, so we skip over the while() loop and effectively just return NULL
without ever dereferencing 'slot'.
2) radix_tree_iter_retry() via tagged iteration like
radix_tree_for_each_tagged(). This case was giving us NULL pointer
dereferences in testing, and was fixed with this commit:
commit 3cb9185c67 ("radix-tree: fix radix_tree_iter_retry() for tagged
iterators.")
This fix doesn't explicitly check for 'slot' being NULL, though, it works
around the NULL pointer dereference by instead zeroing iter->tags in
radix_tree_iter_retry(), which makes us bail out of the if() case in
radix_tree_next_slot() before we dereference 'slot'.
3) radix_tree_iter_next() via via a non-tagged iteration like
radix_tree_for_each_slot(). This currently happens in shmem_tag_pins()
and shmem_partial_swap_usage().
As with non-tagged iteration, 'count' in the else case of
radix_tree_next_slot() is zero, so we skip over the while() loop and
effectively just return NULL without ever dereferencing 'slot'.
4) radix_tree_iter_next() via tagged iteration like
radix_tree_for_each_tagged(). This happens in shmem_wait_for_pins().
radix_tree_iter_next() zeros out iter->tags, so we end up exiting
radix_tree_next_slot() here:
if (flags & RADIX_TREE_ITER_TAGGED) {
void *canon = slot;
iter->tags >>= 1;
if (unlikely(!iter->tags))
return NULL;
Link: http://lkml.kernel.org/r/20160815194237.25967-3-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Konstantin Khlebnikov <koct9i@gmail.com>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Shuah Khan <shuahkh@osg.samsung.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
There are four cases I can see where we could end up with a NULL 'slot' in
radix_tree_next_slot(). Yet radix_tree_next_slot() never actually checks
whether 'slot' is NULL. It just happens that for the cases where 'slot'
is NULL, some other combination of factors prevents us from dereferencing
it.
It would be very easy for someone to unwittingly change one of these
factors without realizing that we are implicitly depending on it to save
us from a NULL pointer dereference.
Add a comment documenting the things that allow 'slot' to be safely passed
as NULL to radix_tree_next_slot().
Here are details on the four cases:
1) radix_tree_iter_retry() via a non-tagged iteration like
radix_tree_for_each_slot(). In this case we currently aren't seeing a bug
because radix_tree_iter_retry() sets
iter->next_index = iter->index;
which means that in in the else case in radix_tree_next_slot(), 'count' is
zero, so we skip over the while() loop and effectively just return NULL
without ever dereferencing 'slot'.
2) radix_tree_iter_retry() via tagged iteration like
radix_tree_for_each_tagged(). This case was giving us NULL pointer
dereferences in testing, and was fixed with this commit:
commit 3cb9185c67 ("radix-tree: fix radix_tree_iter_retry() for tagged
iterators.")
This fix doesn't explicitly check for 'slot' being NULL, though, it works
around the NULL pointer dereference by instead zeroing iter->tags in
radix_tree_iter_retry(), which makes us bail out of the if() case in
radix_tree_next_slot() before we dereference 'slot'.
3) radix_tree_iter_next() via via a non-tagged iteration like
radix_tree_for_each_slot(). This currently happens in shmem_tag_pins()
and shmem_partial_swap_usage().
As with non-tagged iteration, 'count' in the else case of
radix_tree_next_slot() is zero, so we skip over the while() loop and
effectively just return NULL without ever dereferencing 'slot'.
4) radix_tree_iter_next() via tagged iteration like
radix_tree_for_each_tagged(). This happens in shmem_wait_for_pins().
radix_tree_iter_next() zeros out iter->tags, so we end up exiting
radix_tree_next_slot() here:
if (flags & RADIX_TREE_ITER_TAGGED) {
void *canon = slot;
iter->tags >>= 1;
if (unlikely(!iter->tags))
return NULL;
Link: http://lkml.kernel.org/r/20160815194237.25967-2-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Konstantin Khlebnikov <koct9i@gmail.com>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Shuah Khan <shuahkh@osg.samsung.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The function dax_pmem_probe() in drivers/dax/pmem.c is compiled under the
CONFIG_DEV_DAX_PMEM tri-state config option. This config option currently
only depends on CONFIG_NVDIMM_DAX, a bool, which means that the following
configuration is possible:
CONFIG_LIBNVDIMM=m
...
CONFIG_NVDIMM_DAX=y
CONFIG_DEV_DAX=y
CONFIG_DEV_DAX_PMEM=y
With this config LIBNVDIMM is compiled as a module with NVDIMM_DAX=y just
meaning that we will compile drivers/nvdimm/dax_devs.c into that module.
However, dax_pmem_probe() depends on several symbols defined in
drivers/nvdimm/dax_devs.c, which results in the following build errors:
drivers/built-in.o: In function `dax_pmem_probe':
linux/drivers/dax/pmem.c:70: undefined reference to `to_nd_dax'
linux/drivers/dax/pmem.c:74: undefined reference to
`nvdimm_namespace_common_probe'
linux/drivers/dax/pmem.c:80: undefined reference to `devm_nsio_enable'
linux/drivers/dax/pmem.c:81: undefined reference to `nvdimm_setup_pfn'
linux/drivers/dax/pmem.c:84: undefined reference to `devm_nsio_disable'
linux/drivers/dax/pmem.c:122: undefined reference to `to_nd_region'
drivers/built-in.o: In function `dax_pmem_init':
linux/drivers/dax/pmem.c:147: undefined reference to `__nd_driver_register'
Fix this by making NVDIMM_DAX a tristate. DEV_DAX_PMEM depends on
NVDIMM_DAX which depends on LIBNVDIMM. Since they are all now tristates,
if LIBNVDIMM is built as a kernel module DEV_DAX_PMEM will be as well.
This prevents dax_devs.c from being built as a built-in while its
dependencies are in the libnvdimm.ko module.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Currently when doing a DAX hole punch with ext4 we fail to do a writeback.
This is because the logic around filemap_write_and_wait_range() in
ext4_punch_hole() only looks for dirty page cache pages in the radix tree,
not for dirty DAX exceptional entries.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: <stable@vger.kernel.org>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The "NVDIMM Block Window Driver Writer's Guide":
http://pmem.io/documents/NVDIMM_DriverWritersGuide-July-2016.pdf
...defines the layout of the block window status register. For the July
2016 version of the spec linked to above, this happens in Figure 4 on
page 26.
The only bits defined in this spec are bits 31, 5, 4, 2, 1 and 0. The
rest of the bits in the status register are reserved, and there is a
warning following the diagram that says:
Note: The driver cannot assume the value of the RESERVED bits in the
status register are zero. These reserved bits need to be masked off, and
the driver must avoid checking the state of those bits.
This change ensures that for hardware implementations that set these
reserved bits in the status register, the driver won't incorrectly fail the
block I/Os.
Cc: <stable@vger.kernel.org> #v4.2+
Reviewed-by: Lee, Chun-Yi <jlee@suse.com>
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
The bottom two bits of radix tree entries are reserved for special use
by the radix tree code itself. A comment detailing their usage was
added by commit 3bcadd6fa6 ("radix-tree: free up the bottom bit of
exceptional entries for reuse")
This comment states that if the bottom two bits are '11', this means
that this is a locked exceptional entry.
It turns out that this bit combination was never actually used. Radix
tree locking for DAX was indeed implemented, but it actually used the
third LSB:
/* We use lowest available exceptional entry bit for locking */
#define RADIX_DAX_ENTRY_LOCK (1 << RADIX_TREE_EXCEPTIONAL_SHIFT)
This locking code was also made specific to the DAX code instead of
being generally implemented in radix-tree.h.
So, fix the comment.
Link: http://lkml.kernel.org/r/1468997731-2155-1-git-send-email-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Jan Kara <jack@suse.com>
Cc: Konstantin Khlebnikov <koct9i@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Remove the unused wrappers dax_fault() and dax_pmd_fault(). After this
removal, rename __dax_fault() and __dax_pmd_fault() to dax_fault() and
dax_pmd_fault() respectively, and update all callers.
The dax_fault() and dax_pmd_fault() wrappers were initially intended to
capture some filesystem independent functionality around page faults
(calling sb_start_pagefault() & sb_end_pagefault(), updating file mtime
and ctime).
However, the following commits:
5726b27b09 ("ext2: Add locking for DAX faults")
ea3d7209ca ("ext4: fix races between page faults and hole punching")
added locking to the ext2 and ext4 filesystems after these common
operations but before __dax_fault() and __dax_pmd_fault() were called.
This means that these wrappers are no longer used, and are unlikely to
be used in the future.
XFS has had locking analogous to what was recently added to ext2 and
ext4 since DAX support was initially introduced by:
6b698edeee ("xfs: add DAX file operations support")
Link: http://lkml.kernel.org/r/20160714214049.20075-2-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Andreas Dilger <adilger.kernel@dilger.ca>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Jonathan Corbet <corbet@lwn.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Add a generic test for multi-order tag verification, and call it using
several different configurations.
This test creates a multi-order radix tree using the given index and
order, and then sets, checks and clears tags using the indices covered
by the single multi-order radix tree entry.
With the various calls done by this test we verify root multi-order
entries without siblings, multi-order entries without siblings in a
radix tree node, as well as multi-order entries with siblings of various
sizes.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Matthew Wilcox <willy@linux.intel.com>
Cc: Konstantin Khlebnikov <koct9i@gmail.com>
Cc: Kirill Shutemov <kirill.shutemov@linux.intel.com>
Cc: Jan Kara <jack@suse.com>
Cc: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Ross Zwisler