It is invalid to call del_gendisk() when disk->queue is NULL. Fix error
handling in axon_ram_probe() to avoid doing that.
Also del_gendisk() does not drop a reference to gendisk allocated by
alloc_disk(). That has to be done by put_disk(). Add that call where
needed.
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@fb.com>
This reverts commit 0dba1314d4. It causes
leaking of device numbers for SCSI when SCSI registers multiple gendisks
for one request_queue in succession. It can be easily reproduced using
Omar's script [1] on kernel with CONFIG_DEBUG_TEST_DRIVER_REMOVE.
Furthermore the protection provided by this commit is not needed anymore
as the problem it was fixing got also fixed by commit 165a5e22fa
"block: Move bdi_unregister() to del_gendisk()".
[1]: http://marc.info/?l=linux-block&m=148554717109098&w=2
Signed-off-by: Jan Kara <jack@suse.cz>
Acked-by: Dan Williams <dan.j.williams@intel.com>
Tested-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
Commit 165a5e22fa "block: Move bdi_unregister() to del_gendisk()"
added disk->queue dereference to del_gendisk(). Although del_gendisk()
is not supposed to be called without disk->queue valid and
blk_unregister_queue() warns in that case, this change will make it oops
instead. Return to the old more robust behavior of just warning when
del_gendisk() gets called for gendisk with disk->queue being NULL.
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Tested-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
bdi_writeback_congested structures get created for each blkcg and bdi
regardless whether bdi is registered or not. When they are created in
unregistered bdi and the request queue (and thus bdi) is then destroyed
while blkg still holds reference to bdi_writeback_congested structure,
this structure will be referencing freed bdi and last wb_congested_put()
will try to remove the structure from already freed bdi.
With commit 165a5e22fa "block: Move bdi_unregister() to
del_gendisk()", SCSI started to destroy bdis without calling
bdi_unregister() first (previously it was calling bdi_unregister() even
for unregistered bdis) and thus the code detaching
bdi_writeback_congested in cgwb_bdi_destroy() was not triggered and we
started hitting this use-after-free bug. It is enough to boot a KVM
instance with virtio-scsi device to trigger this behavior.
Fix the problem by detaching bdi_writeback_congested structures in
bdi_exit() instead of bdi_unregister(). This is also more logical as
they can get attached to bdi regardless whether it ever got registered
or not.
Fixes: 165a5e22fa
Signed-off-by: Jan Kara <jack@suse.cz>
Tested-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
SCSI can call device_add_disk() several times for one request queue when
a device in unbound and bound, creating new gendisk each time. This will
lead to bdi being repeatedly registered and unregistered. This was not a
big problem until commit 165a5e22fa "block: Move bdi_unregister() to
del_gendisk()" since bdi was only registered repeatedly (bdi_register()
handles repeated calls fine, only we ended up leaking reference to
gendisk due to overwriting bdi->owner) but unregistered only in
blk_cleanup_queue() which didn't get called repeatedly. After
165a5e22fa we were doing correct bdi_register() - bdi_unregister()
cycles however bdi_unregister() is not prepared for it. So make sure
bdi_unregister() cleans up bdi in such a way that it is prepared for
a possible following bdi_register() call.
An easy way to provoke this behavior is to enable
CONFIG_DEBUG_TEST_DRIVER_REMOVE and use scsi_debug driver to create a
scsi disk which immediately hangs without this fix.
Fixes: 165a5e22fa
Signed-off-by: Jan Kara <jack@suse.cz>
Tested-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
Commit 6cd18e711d "block: destroy bdi before blockdev is
unregistered." moved bdi unregistration (at that time through
bdi_destroy()) from blk_release_queue() to blk_cleanup_queue() because
it needs to happen before blk_unregister_region() call in del_gendisk()
for MD. SCSI though will free up the device number from sd_remove()
called through a maze of callbacks from device_del() in
__scsi_remove_device() before blk_cleanup_queue() and thus similar races
as described in 6cd18e711d can happen for SCSI as well as reported by
Omar [1].
Moving bdi_unregister() to del_gendisk() works for MD and fixes the
problem for SCSI since del_gendisk() gets called from sd_remove() before
freeing the device number.
This also makes device_add_disk() (calling bdi_register_owner()) more
symmetric with del_gendisk().
[1] http://marc.info/?l=linux-block&m=148554717109098&w=2
Tested-by: Lekshmi Pillai <lekshmicpillai@in.ibm.com>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jan Kara <jack@suse.cz>
Tested-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
So far we initialized bd_bdi only in bdget(). That is fine for normal
bdev inodes however for the special case of the root inode of
blockdev_superblock that function is never called and thus bd_bdi is
left uninitialized. As a result bdev_evict_inode() may oops doing
bdi_put(root->bd_bdi) on that inode as can be seen when doing:
mount -t bdev none /mnt
Fix the problem by initializing bd_bdi when first allocating the inode
and then reinitializing bd_bdi in bdev_evict_inode().
Thanks to syzkaller team for finding the problem.
Reported-by: Dmitry Vyukov <dvyukov@google.com>
Fixes: b1d2dc5659 ("block: Make blk_get_backing_dev_info() safe without open bdev")
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@fb.com>
When a device gets removed, block device inode unhashed so that it is not
used anymore (bdget() will not find it anymore). Later when a new device
gets created with the same device number, we create new block device
inode. However there may be file system device inodes whose i_bdev still
points to the original block device inode and thus we get two active
block device inodes for the same device. They will share the same
gendisk so the only visible differences will be that page caches will
not be coherent and BDIs will be different (the old block device inode
still points to unregistered BDI).
Fix the problem by checking in bd_acquire() whether i_bdev still points
to active block device inode and re-lookup the block device if not. That
way any open of a block device happening after the old device has been
removed will get correct block device inode.
Tested-by: Lekshmi Pillai <lekshmicpillai@in.ibm.com>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@fb.com>
Iteration over partitions in del_gendisk() omits part0. Add
bdev_unhash_inode() call for the whole device. Otherwise if the device
number gets reused, bdev inode will be still associated with the old
(stale) bdi.
Tested-by: Lekshmi Pillai <lekshmicpillai@in.ibm.com>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@fb.com>
Move bdev_unhash_inode() after invalidate_partition() as
invalidate_partition() looks up bdev and it cannot find the right bdev
inode after bdev_unhash_inode() is called. Thus invalidate_partition()
would not invalidate page cache of the previously used bdev. Also use
part_devt() when calling bdev_unhash_inode() instead of manually
creating the device number.
Tested-by: Lekshmi Pillai <lekshmicpillai@in.ibm.com>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@fb.com>
Avoid using stripe_width for sbi->s_stripe value if it is not actually
set. It prevents using the stride for sbi->s_stripe.
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
When a filesystem is created using:
mkfs.ext4 -b 4096 -E stride=512 <dev>
and we try to allocate 64MB extent, we will end up directly in
ext4_mb_complex_scan_group(). This is because the request is detected
as power-of-two allocation (so we start in ext4_mb_regular_allocator()
with ac_criteria == 0) however the check before
ext4_mb_simple_scan_group() refuses the direct buddy scan because the
allocation request is too large. Since cr == 0, the check whether we
should use ext4_mb_scan_aligned() fails as well and we fall back to
ext4_mb_complex_scan_group().
Fix the problem by checking for upper limit on power-of-two requests
directly when detecting them.
Reported-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Some ->page_mkwrite handlers may return VM_FAULT_RETRY as its return
code (GFS2 or Lustre can definitely do this). However VM_FAULT_RETRY
from ->page_mkwrite is completely unhandled by the mm code and results
in locking and writeably mapping the page which definitely is not what
the caller wanted.
Fix Lustre and block_page_mkwrite_ret() used by other filesystems
(notably GFS2) to return VM_FAULT_NOPAGE instead which results in
bailing out from the fault code, the CPU then retries the access, and we
fault again effectively doing what the handler wanted.
Link: http://lkml.kernel.org/r/20170203150729.15863-1-jack@suse.cz
Signed-off-by: Jan Kara <jack@suse.cz>
Reported-by: Al Viro <viro@ZenIV.linux.org.uk>
Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
blk_get_backing_dev_info() is now a simple dereference. Remove that
function and simplify some code around that.
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@fb.com>
Currenly blk_get_backing_dev_info() is not safe to be called when the
block device is not open as bdev->bd_disk is NULL in that case. However
inode_to_bdi() uses this function and may be call called from flusher
worker or other writeback related functions without bdev being open
which leads to crashes such as:
[113031.075540] Unable to handle kernel paging request for data at address 0x00000000
[113031.075614] Faulting instruction address: 0xc0000000003692e0
0:mon> t
[c0000000fb65f900] c00000000036cb6c writeback_sb_inodes+0x30c/0x590
[c0000000fb65fa10] c00000000036ced4 __writeback_inodes_wb+0xe4/0x150
[c0000000fb65fa70] c00000000036d33c wb_writeback+0x30c/0x450
[c0000000fb65fb40] c00000000036e198 wb_workfn+0x268/0x580
[c0000000fb65fc50] c0000000000f3470 process_one_work+0x1e0/0x590
[c0000000fb65fce0] c0000000000f38c8 worker_thread+0xa8/0x660
[c0000000fb65fd80] c0000000000fc4b0 kthread+0x110/0x130
[c0000000fb65fe30] c0000000000098f0 ret_from_kernel_thread+0x5c/0x6c
Signed-off-by: Jens Axboe <axboe@fb.com>
Instead of storing backing_dev_info inside struct request_queue,
allocate it dynamically, reference count it, and free it when the last
reference is dropped. Currently only request_queue holds the reference
but in the following patch we add other users referencing
backing_dev_info.
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@fb.com>
We will want to have struct backing_dev_info allocated separately from
struct request_queue. As the first step add pointer to backing_dev_info
to request_queue and convert all users touching it. No functional
changes in this patch.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@fb.com>
Currently, block device inodes stay around after corresponding gendisk
hash died until memory reclaim finds them and frees them. Since we will
make block device inode pin the bdi, we want to free the block device
inode as soon as the device goes away so that bdi does not stay around
unnecessarily. Furthermore we need to avoid issues when new device with
the same major,minor pair gets created since reusing the bdi structure
would be rather difficult in this case.
Unhashing block device inode on gendisk destruction nicely deals with
these problems. Once last block device inode reference is dropped (which
may be directly in del_gendisk()), the inode gets evicted. Furthermore if
the major,minor pair gets reallocated, we are guaranteed to get new
block device inode even if old block device inode is not yet evicted and
thus we avoid issues with possible reuse of bdi.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@fb.com>
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: Al Viro <viro@ZenIV.linux.org.uk>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore 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: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@vger.kernel.org
Commit 99579ccec4 "xfs: skip dirty pages in ->releasepage()" started
to skip dirty pages in xfs_vm_releasepage() which also has the effect
that if a dirty page is truncated, it does not get freed by
block_invalidatepage() and is lingering in LRU list waiting for reclaim.
So a simple loop like:
while true; do
dd if=/dev/zero of=file bs=1M count=100
rm file
done
will keep using more and more memory until we hit low watermarks and
start pagecache reclaim which will eventually reclaim also the truncate
pages. Keeping these truncated (and thus never usable) pages in memory
is just a waste of memory, is unnecessarily stressing page cache
reclaim, and reportedly also leads to anonymous mmap(2) returning ENOMEM
prematurely.
So instead of just skipping dirty pages in xfs_vm_releasepage(), return
to old behavior of skipping them only if they have delalloc or unwritten
buffers and fix the spurious warnings by warning only if the page is
clean.
CC: stable@vger.kernel.org
CC: Brian Foster <bfoster@redhat.com>
CC: Vlastimil Babka <vbabka@suse.cz>
Reported-by: Petr Tůma <petr.tuma@d3s.mff.cuni.cz>
Fixes: 99579ccec4
Signed-off-by: Jan Kara <jack@suse.cz>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
UDF encodes symlinks in a more complex fashion and thus i_size of a
symlink does not match the lenght of a string returned by readlink(2).
This confuses some applications (see bug 191241) and may be considered a
violation of POSIX. Fix the problem by reading the link into page cache
in response to stat(2) call and report the length of the decoded path.
Signed-off-by: Jan Kara <jack@suse.cz>
Audit tree code was happily adding new notification marks while holding
spinlocks. Since fsnotify_add_mark() acquires group->mark_mutex this can
lead to sleeping while holding a spinlock, deadlocks due to lock
inversion, and probably other fun. Fix the problem by acquiring
group->mark_mutex earlier.
CC: Paul Moore <paul@paul-moore.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Paul Moore <paul@paul-moore.com>
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: Jan Kara <jack@suse.cz>
Signed-off-by: Dan 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: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan 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: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan 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: Christoph Hellwig <hch@lst.de>
Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan 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: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan 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: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
There are only two calls sites of fsnotify_duplicate_mark(). Those are
in kernel/audit_tree.c and both are bogus. Vfsmount pointer is unused
for audit tree, inode pointer and group gets set in
fsnotify_add_mark_locked() later anyway, mask and free_mark are already
set in alloc_chunk(). In fact, calling fsnotify_duplicate_mark() is
actively harmful because following fsnotify_add_mark_locked() will leak
group reference by overwriting the group pointer. So just remove the two
calls to fsnotify_duplicate_mark() and the function.
Signed-off-by: Jan Kara <jack@suse.cz>
[PM: line wrapping to fit in 80 chars]
Signed-off-by: Paul Moore <paul@paul-moore.com>
dquot_disable() was warning when sb_has_quota_loaded() was true when
invalidating page cache for quota files. The thinking behind this
warning was that we must have raced with somebody else turning quotas on
and this should not happen because all places modifying quota state must
hold s_umount exclusively now. However sb_has_quota_loaded() can be also
true at this point when we are just suspending quotas on remount
read-only. Just restore the behavior to situation before commit
c3b004460d ("quota: Remove dqonoff_mutex") which introduced the
warning.
The code in dquot_disable() can be further simplified with the new
locking of quota state changes however let's leave that to a separate
commit that can get more testing exposure.
Fixes: c3b004460d
Signed-off-by: Jan Kara <jack@suse.cz>
Currently finish_mkwrite_fault() returns 0 when PTE got changed before
we acquired PTE lock and VM_FAULT_WRITE when we succeeded in modifying
the PTE. This is somewhat confusing since 0 generally means success, it
is also inconsistent with finish_fault() which returns 0 on success.
Change finish_mkwrite_fault() to return 0 on success and VM_FAULT_NOPAGE
when PTE changed. Practically, there should be no behavioral difference
since we bail out from the fault the same way regardless whether we
return 0, VM_FAULT_NOPAGE, or VM_FAULT_WRITE. Also note that
VM_FAULT_WRITE has no effect for shared mappings since the only two
places that check it - KSM and GUP - care about private mappings only.
Generally the meaning of VM_FAULT_WRITE for shared mappings is not well
defined and we should probably clean that up.
Link: http://lkml.kernel.org/r/1479460644-25076-17-git-send-email-jack@suse.cz
Signed-off-by: Jan Kara <jack@suse.cz>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Patch series "dax: Clear dirty bits after flushing caches", v5.
Patchset to clear dirty bits from radix tree of DAX inodes when caches
for corresponding pfns have been flushed. In principle, these patches
enable handlers to easily update PTEs and do other work necessary to
finish the fault without duplicating the functionality present in the
generic code. I'd like to thank Kirill and Ross for reviews of the
series!
This patch (of 20):
To allow full handling of COW faults add memcg field to struct vm_fault
and a return value of ->fault() handler meaning that COW fault is fully
handled and memcg charge must not be canceled. This will allow us to
remove knowledge about special DAX locking from the generic fault code.
Link: http://lkml.kernel.org/r/1479460644-25076-9-git-send-email-jack@suse.cz
Signed-off-by: Jan Kara <jack@suse.cz>
Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Currently we have two different structures for passing fault information
around - struct vm_fault and struct fault_env. DAX will need more
information in struct vm_fault to handle its faults so the content of
that structure would become event closer to fault_env. Furthermore it
would need to generate struct fault_env to be able to call some of the
generic functions. So at this point I don't think there's much use in
keeping these two structures separate. Just embed into struct vm_fault
all that is needed to use it for both purposes.
Link: http://lkml.kernel.org/r/1479460644-25076-2-git-send-email-jack@suse.cz
Signed-off-by: Jan Kara <jack@suse.cz>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fsnotify_unmount_inodes() plays complex tricks to pin next inode in the
sb->s_inodes list when iterating over all inodes. Furthermore the code has a
bug that if the current inode is the last on i_sb_list that does not have e.g.
I_FREEING set, then we leave next_i pointing to inode which may get removed
from the i_sb_list once we drop s_inode_list_lock thus resulting in
use-after-free issues (usually manifesting as infinite looping in
fsnotify_unmount_inodes()).
Fix the problem by keeping current inode pinned somewhat longer. Then we can
make the code much simpler and standard.
CC: stable@vger.kernel.org
Signed-off-by: Jan Kara <jack@suse.cz>
Commit 642261ac99: "dax: add struct iomap based DAX PMD support" has
introduced unmapping of page tables if huge page needs to be split in
grab_mapping_entry(). However the unmapping happens after
radix_tree_preload() call which disables preemption and thus
unmap_mapping_range() tries to acquire i_mmap_lock in atomic context
which is a bug. Fix the problem by moving unmapping before
radix_tree_preload() call.
Fixes: 642261ac99
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Combination of data=ordered mode and journal_async_commit mount option
is invalid. However the check in parse_options() fails to detect the
case where we simply end up defaulting to data=ordered mode and we
detect the problem only on remount which triggers hard to understand
failure to remount the filesystem.
Fix the checking of mount options to take into account also the default
mode by moving the check somewhat later in the mount sequence.
Reported-by: Wolfgang Walter <linux@stwm.de>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Warn when a page is dirtied without buffers (as that will likely lead to
a crash in ext4_writepages()) or when it gets newly dirtied without the
page being locked (as there is nothing that prevents buffers to get
stripped just before calling set_page_dirty() under memory pressure).
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The only places that were grabbing dqonoff_mutex are functions turning
quotas on and off and these are properly serialized using s_umount
semaphore. Remove dqonoff_mutex.
Signed-off-by: Jan Kara <jack@suse.cz>
Currently we use dqonoff_mutex to serialize quota recovery protection
and turning of quotas on / off. Use s_umount semaphore instead.
Tested-by: Eric Ren <zren@suse.com>
Signed-off-by: Jan Kara <jack@suse.cz>
All callers of dquot_scan_active() now hold s_umount so we can rely on
that lock to protect us against quota state changes.
Signed-off-by: Jan Kara <jack@suse.cz>
New quota locking rules will require s_umount semaphore for all quota
scanning functions. Add is for periodic quota syncing.
Tested-by: Eric Ren <zren@suse.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Currently we just silently ignore flags that we don't understand (or
that cannot be manipulated) through EXT4_IOC_SETFLAGS and
EXT4_IOC_FSSETXATTR ioctls. This makes it problematic for the unused
flags to be used in future (some app may be inadvertedly setting them
and we won't notice until the flag gets used). Also this is inconsistent
with other filesystems like XFS or BTRFS which return EOPNOTSUPP when
they see a flag they cannot set.
ext4 has the additional problem that there are flags which are returned
by EXT4_IOC_GETFLAGS ioctl but which cannot be modified via
EXT4_IOC_SETFLAGS. So we have to be careful to ignore value of these
flags and not fail the ioctl when they are set (as e.g. chattr(1) passes
flags returned from EXT4_IOC_GETFLAGS to EXT4_IOC_SETFLAGS without any
masking and thus we'd break this utility).
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Add EXT4_JOURNAL_DATA_FL and EXT4_EXTENTS_FL to EXT4_FL_USER_MODIFIABLE
to recognize that they are modifiable by userspace. So far we got away
without having them there because ext4_ioctl_setflags() treats them in a
special way. But it was really confusing like that.
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Writeback quota is protected by s_umount semaphore held for reading
because every writeback must be protected by that lock (grabbed either
by the generic writeback code or by quotactl handler). Getting next
available ID in quota file, querying quota state, setting quota
information, getting quota format are all quotactl operations protected
by s_umount semaphore held for reading grabbed in quotactl handler.
This also fixes lockdep splat about possible deadlock during filesystem
freezing where sync_filesystem() is called with page-faults already
blocked but sync_filesystem() calls into dquot_writeback_dquots() which
grabs dqonoff_mutex which ranks above i_mutex (vfs_load_quota_inode()
grabs i_mutex under dqonoff_mutex) which clearly ranks below page fault
freeze protection (e.g. via mmap_sem dependencies). The reported problem
is not a real deadlock possibility since during quota on we check
whether filesystem freezing is not in progress but still it is good to
have this fixed.
Reported-by: Ted Tso <tytso@mit.edu>
Reported-by: Eric Whitney <enwlinux@gmail.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Currently we hold s_umount semaphore only in shared mode when enabling
or disabling quotas and use dqonoff_mutex for serializing quota state
changes on a filesystem and also quota state changes with other places
depending on current quota state. Using dedicated mutex for this causes
possible deadlocks during filesystem freezing (see following commit for
details) so we transition to using s_umount semaphore for the necessary
synchronization whose lock ordering is properly handled by the
filesystem freezing code. As a start grab s_umount in exclusive mode
when enabling / disabling quotas.
Signed-off-by: Jan Kara <jack@suse.cz>
Quota code will need a variant of get_super_thawed() that returns
superblock with s_umount held in exclusive mode to serialize quota on
and quota off operations. Provide this functionality.
Signed-off-by: Jan Kara <jack@suse.cz>
When ext4 is compiled with DAX support, it now needs the iomap code. Add
appropriate select to Kconfig.
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
No one uses functions using the get_block callback anymore. Rip them
out and update documentation.
Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Currently the last user of ext2_get_blocks() for DAX inodes was
dax_truncate_page(). Convert that to iomap_zero_range() so that all DAX
IO uses the iomap path.
Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Reads and writes for DAX inodes should no longer end up in direct IO
code. Rip out the support and add a warning.
Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Convert DAX faults to use iomap infrastructure. We would not have to start
transaction in ext4_dax_fault() anymore since ext4_iomap_begin takes
care of that but so far we do that to avoid lock inversion of
transaction start with DAX entry lock which gets acquired in
dax_iomap_fault() before calling ->iomap_begin handler.
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Currently mapping of blocks for DAX writes happen with
EXT4_GET_BLOCKS_PRE_IO flag set. That has a result that each
ext4_map_blocks() call creates a separate written extent, although it
could be merged to the neighboring extents in the extent tree. The
reason for using this flag is that in case the extent is unwritten, we
need to convert it to written one and zero it out. However this "convert
mapped range to written" operation is already implemented by
ext4_map_blocks() for the case of data writes into unwritten extent. So
just use flags for that mode of operation, simplify the code, and avoid
unnecessary split extents.
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Implement DAX writes using the new iomap infrastructure instead of
overloading the direct IO path.
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Use iomap infrastructure for zeroing blocks when in DAX mode.
ext4_iomap_begin() handles read requests just fine and that's all that
is needed for iomap_zero_range().
Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Implement basic iomap_begin function that handles reading and use it for
DAX reads.
Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Currently we have S_DAX set inode->i_flags for a regular file whenever
ext4 is mounted with dax mount option. However in some cases we cannot
really do DAX - e.g. when inode is marked to use data journalling, when
inode data is being encrypted, or when inode is stored inline. Make sure
S_DAX flag is appropriately set/cleared in these cases.
Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Factor out checks of 'from' and whether we are overwriting out of
ext4_file_write_iter() so that the function is easier to follow.
Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Introduce a flag telling iomap operations whether they are handling a
fault or other IO. That may influence behavior wrt inode size and
similar things.
Signed-off-by: Jan Kara <jack@suse.cz>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Add a helper function that clears buffer heads from a block device
aliasing passed bh. Use this helper function from filesystems instead of
the original unmap_underlying_metadata() to save some boiler plate code
and also have a better name for the functionalily since it is not
unmapping anything for a *long* time.
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@fb.com>
Use clean_bdev_aliases() instead of iterating through blocks one by one.
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@fb.com>
Use clean_bdev_aliases() instead of iterating through blocks one by one.
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@fb.com>
Use new provided function instead of an iteration through all allocated
blocks.
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@fb.com>
Provide function equivalent to unmap_underlying_metadata() for a range
of blocks. We somewhat optimize the function to use pagevec lookups
instead of looking up buffer heads one by one and use page lock to pin
buffer heads instead of mapping's private_lock to improve scalability.
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@fb.com>
Currently we dropped freeze protection of aio writes just after IO was
submitted. Thus aio write could be in flight while the filesystem was
frozen and that could result in unexpected situation like aio completion
wanting to convert extent type on frozen filesystem. Testcase from
Dmitry triggering this is like:
for ((i=0;i<60;i++));do fsfreeze -f /mnt ;sleep 1;fsfreeze -u /mnt;done &
fio --bs=4k --ioengine=libaio --iodepth=128 --size=1g --direct=1 \
--runtime=60 --filename=/mnt/file --name=rand-write --rw=randwrite
Fix the problem by dropping freeze protection only once IO is completed
in aio_complete().
Reported-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: Jan Kara <jack@suse.cz>
[hch: forward ported on top of various VFS and aio changes]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Currently rd_size was int which lead to overflow and bogus device size
once the requested ramdisk size was 1 TB or more. Although these days
ramdisks with 1 TB size are mostly a mistake, the days when they are
useful are not far.
Reported-by: Bart Van Assche <bart.vanassche@sandisk.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@fb.com>
iomap_page_mkwrite_actor() calls __block_write_begin_int() with position
masked as pos & ~PAGE_MASK which is equivalent to pos & (PAGE_SIZE-1).
Thus it masks off high bits of file position. However
__block_write_begin_int() expects full file position on input. This does
not cause any visible issues because all __block_write_begin_int()
really cares about are low file position bits but still it is a bug
waiting to happen.
Signed-off-by: Jan Kara <jack@suse.cz>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
When isofs_mount() is called to mount a device read-write, it returns
EACCES even before it checks that the device actually contains an isofs
filesystem. This may confuse mount(8) which then tries to mount all
subsequent filesystem types in read-only mode.
Fix the problem by returning EACCES only once we verify that the device
indeed contains an iso9660 filesystem.
CC: stable@vger.kernel.org
Fixes: 17b7f7cf58
Reported-by: Kent Overstreet <kent.overstreet@gmail.com>
Reported-by: Karel Zak <kzak@redhat.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Fanotify code has its own lock (access_lock) to protect a list of events
waiting for a response from userspace.
However this is somewhat awkward as the same list_head in the event is
protected by notification_lock if it is part of the notification queue
and by access_lock if it is part of the fanotify private queue which
makes it difficult for any reliable checks in the generic code. So make
fanotify use the same lock - notification_lock - for protecting its
private event list.
Link: http://lkml.kernel.org/r/1473797711-14111-6-git-send-email-jack@suse.cz
Signed-off-by: Jan Kara <jack@suse.cz>
Reviewed-by: Lino Sanfilippo <LinoSanfilippo@gmx.de>
Cc: Miklos Szeredi <mszeredi@redhat.com>
Cc: Eric Paris <eparis@redhat.com>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fsnotify_flush_notify() and fanotify_release() destroy notification
event while holding notification_mutex.
The destruction of fanotify event includes a path_put() call which may
end up calling into a filesystem to delete an inode if we happen to be
the last holders of dentry reference which happens to be the last holder
of inode reference.
That in turn may violate lock ordering for some filesystems since
notification_mutex is also acquired e. g. during write when generating
fanotify event.
Also this is the only thing that forces notification_mutex to be a
sleeping lock. So drop notification_mutex before destroying a
notification event.
Link: http://lkml.kernel.org/r/1473797711-14111-4-git-send-email-jack@suse.cz
Signed-off-by: Jan Kara <jack@suse.cz>
Cc: Miklos Szeredi <mszeredi@redhat.com>
Cc: Lino Sanfilippo <LinoSanfilippo@gmx.de>
Cc: Eric Paris <eparis@redhat.com>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When zeroing blocks for DAX allocations, we also have to unmap aliases
in the block device mappings. Otherwise writeback can overwrite zeros
with stale data from block device page cache.
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@kernel.org
do_blockdev_direct_IO() takes care of properly plugging direct IO so
there's no need to plug again inside ext4_file_write_iter().
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Currently we do not allow unlocked (meaning without inode_lock) direct
IO when the file has any pages cached. This check is not needed anymore
as we keep inode lock until ext4_direct_IO_write() and thus can happily
writeback and evict any pages conflicting with current direct IO write.
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
We can easily support parallel direct IO reads. We only have to make
sure we cannot expose uninitialized data by reading allocated block to
which data was not written yet, or which was already truncated. That is
easily achieved by holding inode_lock in shared mode - that excludes all
writes, truncates, hole punches. We also have to guard against page
writeback allocating blocks for delay-allocated pages - that race is
handled by the fact that we writeback all the pages in the affected
range and the lock protects us from new pages being created there.
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
When zeroing blocks for DAX allocations, we also have to unmap aliases
in the block device mappings. Otherwise writeback can overwrite zeros
with stale data from block device page cache.
Signed-off-by: Jan Kara <jack@suse.cz>
Thomas has reported a lockdep splat hitting in
add_transaction_credits(). The problem is that that function calls
jbd2_might_wait_for_commit() while holding j_state_lock which is wrong
(we do not really wait for transaction commit while holding that lock).
Fix the problem by moving jbd2_might_wait_for_commit() into places where
we are ready to wait for transaction commit and thus j_state_lock is
unlocked.
Cc: stable@vger.kernel.org
Fixes: 1eaa566d36
Reported-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Currently, notify_change() clears capabilities or IMA attributes by
calling security_inode_killpriv() before calling into ->setattr. Thus it
happens before any other permission checks in inode_change_ok() and user
is thus allowed to trigger clearing of capabilities or IMA attributes
for any file he can look up e.g. by calling chown for that file. This is
unexpected and can lead to user DoSing a system.
Fix the problem by calling security_inode_killpriv() at the end of
inode_change_ok() instead of from notify_change(). At that moment we are
sure user has permissions to do the requested change.
References: CVE-2015-1350
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jan Kara <jack@suse.cz>
inode_change_ok() will be resposible for clearing capabilities and IMA
extended attributes and as such will need dentry. Give it as an argument
to inode_change_ok() instead of an inode. Also rename inode_change_ok()
to setattr_prepare() to better relect that it does also some
modifications in addition to checks.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jan Kara <jack@suse.cz>
Jan Kara