Pull Ceph updates from Sage Weil:
"This changeset has a few main parts:
- Ilya has finished a huge refactoring effort to sync up the
client-side logic in libceph with the user-space client code, which
has evolved significantly over the last couple years, with lots of
additional behaviors (e.g., how requests are handled when cluster
is full and transitions from full to non-full).
This structure of the code is more closely aligned with userspace
now such that it will be much easier to maintain going forward when
behavior changes take place. There are some locking improvements
bundled in as well.
- Zheng adds multi-filesystem support (multiple namespaces within the
same Ceph cluster)
- Zheng has changed the readdir offsets and directory enumeration so
that dentry offsets are hash-based and therefore stable across
directory fragmentation events on the MDS.
- Zheng has a smorgasbord of bug fixes across fs/ceph"
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph-client: (71 commits)
ceph: fix wake_up_session_cb()
ceph: don't use truncate_pagecache() to invalidate read cache
ceph: SetPageError() for writeback pages if writepages fails
ceph: handle interrupted ceph_writepage()
ceph: make ceph_update_writeable_page() uninterruptible
libceph: make ceph_osdc_wait_request() uninterruptible
ceph: handle -EAGAIN returned by ceph_update_writeable_page()
ceph: make fault/page_mkwrite return VM_FAULT_OOM for -ENOMEM
ceph: block non-fatal signals for fault/page_mkwrite
ceph: make logical calculation functions return bool
ceph: tolerate bad i_size for symlink inode
ceph: improve fragtree change detection
ceph: keep leaf frag when updating fragtree
ceph: fix dir_auth check in ceph_fill_dirfrag()
ceph: don't assume frag tree splits in mds reply are sorted
ceph: fix inode reference leak
ceph: using hash value to compose dentry offset
ceph: don't forbid marking directory complete after forward seek
ceph: record 'offset' for each entry of readdir result
ceph: define 'end/complete' in readdir reply as bit flags
...
If MDS sorts dentries in dirfrag in hash order, we use hash value to
compose dentry offset. dentry offset is:
(0xff << 52) | ((24 bits hash) << 28) |
(the nth entry hash hash collision)
This offset is stable across directory fragmentation. This alos means
there is no need to reset readdir offset if directory get fragmented
in the middle of readdir.
Signed-off-by: Yan, Zheng <zyan@redhat.com>
When mds session gets killed, read/write operation may hang.
Client waits for Frw caps, but mds does not know what caps client
wants. To recover this, client sends an open request to mds. The
request will tell mds what caps client wants.
Signed-off-by: Yan, Zheng <zyan@redhat.com>
To access non-default filesystem, we just need to subscribe to
mdsmap.<MDS_NAMESPACE_ID> and add a new mount option for mds
namespace id.
Signed-off-by: Yan, Zheng <zyan@redhat.com>
[idryomov@gmail.com: switch to a new libceph API]
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Add a catch-all xattr handler at the end of ceph_xattr_handlers. Check
for valid attribute names there, and remove those checks from
__ceph_{get,set,remove}xattr instead. No "system.*" xattrs need to be
handled by the catch-all handler anymore.
The set xattr handler is called with a NULL value to indicate that the
attribute should be removed; __ceph_setxattr already handles that case
correctly (ceph_set_acl could already calling __ceph_setxattr with a NULL
value).
Move the check for snapshots from ceph_{set,remove}xattr into
__ceph_{set,remove}xattr. With that, ceph_{get,set,remove}xattr can be
replaced with the generic iops.
Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Create a variant of ceph_setattr that takes an inode instead of a
dentry. Change __ceph_setxattr (and also __ceph_removexattr) to take an
inode instead of a dentry. Use those in ceph_set_acl so that we no
longer need a dentry there.
Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
When security is enabled, security module can call filesystem's
getxattr/setxattr callbacks during d_instantiate(). For cephfs,
d_instantiate() is usually called by MDS' dispatch thread, while
handling MDS reply. If the MDS reply does not include xattrs and
corresponding caps, getxattr/setxattr need to send a new request
to MDS and waits for the reply. This makes MDS' dispatch sleep,
nobody handles later MDS replies.
The fix is make sure lookup/atomic_open reply include xattrs and
corresponding caps. So getxattr can be handled by cached xattrs.
This requires some modification to both MDS and request message.
(Client tells MDS what caps it wants; MDS encodes proper caps in
the reply)
Smack security module may call setxattr during d_instantiate().
Unlike getxattr, we can't force MDS to issue CEPH_CAP_XATTR_EXCL
to us. So just make setxattr return error when called by MDS'
dispatch thread.
Signed-off-by: Yan, Zheng <zyan@redhat.com>
ceph_empty_snapc->num_snaps == 0 at all times. Passing such a snapc to
ceph_osdc_alloc_request() (possibly through ceph_osdc_new_request()) is
equivalent to passing NULL, as ceph_osdc_alloc_request() uses it only
for sizing the request message.
Further, in all four cases the subsequent ceph_osdc_build_request() is
passed NULL for snapc, meaning that 0 is encoded for seq and num_snaps
and making ceph_empty_snapc entirely useless. The two cases where it
actually mattered were removed in commits 8605609049 ("ceph: avoid
sending unnessesary FLUSHSNAP message") and 23078637e0 ("ceph: fix
queuing inode to mdsdir's snaprealm").
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Yan, Zheng <zyan@redhat.com>
When rbytes mount option is enabled, directory size is recursive
size. Recursive size is not updated instantly. This can cause
directory size to change between successive stat(1)
Signed-off-by: Yan, Zheng <zyan@redhat.com>
Add support for the format change of MClientReply/MclientCaps.
Also add code that denies access to inodes with pool_ns layouts.
Signed-off-by: Yan, Zheng <zyan@redhat.com>
Reviewed-by: Sage Weil <sage@redhat.com>
If we get a unsafe reply for request that created/modified inode,
add the unsafe request to a list in the newly created/modified
inode. So we can make fsync() wait these unsafe requests.
Signed-off-by: Yan, Zheng <zyan@redhat.com>
commit e548e9b93d makes the kclient
only re-send cap flush once during MDS failover. If the kclient sends
a cap flush after MDS enters reconnect stage but before MDS recovers.
The kclient will skip re-sending the same cap flush when MDS recovers.
This causes problem for newly created inode. The MDS handles cap
flushes before replaying unsafe requests, so it's possible that MDS
find corresponding inode is missing when handling cap flush. The fix
is reverting to old behaviour: always re-send when MDS recovers
Signed-off-by: Yan, Zheng <zyan@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Previously our dcache readdir code relies on that child dentries in
directory dentry's d_subdir list are sorted by dentry's offset in
descending order. When adding dentries to the dcache, if a dentry
already exists, our readdir code moves it to head of directory
dentry's d_subdir list. This design relies on dcache internals.
Al Viro suggests using ncpfs's approach: keeping array of pointers
to dentries in page cache of directory inode. the validity of those
pointers are presented by directory inode's complete and ordered
flags. When a dentry gets pruned, we clear directory inode's complete
flag in the d_prune() callback. Before moving a dentry to other
directory, we clear the ordered flag for both old and new directory.
Signed-off-by: Yan, Zheng <zyan@redhat.com>
if flushing caps were revoked, we should re-send the cap flush in
client reconnect stage. This guarantees that MDS processes the cap
flush message before issuing the flushing caps to other client.
Signed-off-by: Yan, Zheng <zyan@redhat.com>
So we know TID of the oldest pending caps flushing. Later patch will
send this information to MDS, so that MDS can trim its completed caps
flush list.
Tracking pending caps flushing globally also simplifies syncfs code.
Signed-off-by: Yan, Zheng <zyan@redhat.com>
Previously we do not trace accurate TID for flushing caps. when
MDS failovers, we have no choice but to re-send all flushing caps
with a new TID. This can cause problem because MDS can has already
flushed some caps and has issued the same caps to other client.
The re-sent cap flush has a new TID, which makes MDS unable to
detect if it has already processed the cap flush.
This patch adds code to track pending caps flushing accurately.
When re-sending cap flush is needed, we use its original flush
TID.
Signed-off-by: Yan, Zheng <zyan@redhat.com>
Previously we pre-allocate cap release messages for each caps. This
wastes lots of memory when there are large amount of caps. This patch
make the code not pre-allocate the cap release messages. Instead,
we add the corresponding ceph_cap struct to a list when releasing a
cap. Later when flush cap releases is needed, we allocate the cap
release messages dynamically.
Signed-off-by: Yan, Zheng <zyan@redhat.com>
when a snap notification contains no new snapshot, we can avoid
sending FLUSHSNAP message to MDS. But we still need to create
cap_snap in some case because it's required by write path and
page writeback path
Signed-off-by: Yan, Zheng <zyan@redhat.com>
Cached_context in ceph_snap_realm is directly accessed by
uninline_data() and get_pool_perm(). This is racy in theory.
both uninline_data() and get_pool_perm() do not modify existing
object, they only create new object. So we can pass the empty
snap context to them. Unlike cached_context in ceph_snap_realm,
we do not need to protect the empty snap context.
Signed-off-by: Yan, Zheng <zyan@redhat.com>
When snaprealm is created, its initial reference count is zero.
But in some rare cases, the newly created snaprealm is not referenced
by anyone. This causes snaprealm with zero reference count not freed.
The fix is set reference count of newly snaprealm to 1. The reference
is return the function who requests to create the snaprealm. When the
function finishes its job, it releases the reference.
Signed-off-by: Yan, Zheng <zyan@redhat.com>
After converting inline data to normal data, client need to flush
the new i_inline_version (CEPH_INLINE_NONE) to MDS. This commit makes
cap messages (sent to MDS) contain inline_version and inline_data.
Client always converts inline data to normal data before data write,
so the inline data length part is always zero.
Signed-off-by: Yan, Zheng <zyan@redhat.com>
Before any data write, convert inline data to normal data and set
i_inline_version to CEPH_INLINE_NONE. The OSD request that saves
inline data to object contains 3 operations (CMPXATTR, WRITE and
SETXATTR). It compares a xattr named 'inline_version' to prevent
old data overwrites newer data.
Signed-off-by: Yan, Zheng <zyan@redhat.com>
Add a new parameter 'locked_page' to ceph_do_getattr(). If inline data
in getattr reply will be copied to the page.
Signed-off-by: Yan, Zheng <zyan@redhat.com>
Request reply and cap message can contain inline data. add inline data
to the page cache if there is Fc cap.
Signed-off-by: Yan, Zheng <zyan@redhat.com>
Current snaphost code does not properly handle moving inode from one
empty snap realm to another empty snap realm. After changing inode's
snap realm, some dirty pages' snap context can be not equal to inode's
i_head_snap. This can trigger BUG() in ceph_put_wrbuffer_cap_refs()
The fix is introduce a global empty snap context for all empty snap
realm. This avoids triggering the BUG() for filesystem with no snapshot.
Fixes: http://tracker.ceph.com/issues/9928
Signed-off-by: Yan, Zheng <zyan@redhat.com>
Reviewed-by: Ilya Dryomov <idryomov@redhat.com>
After creating/deleting/renaming file, offsets of sibling dentries may
change. So we can not use cached dentries to satisfy readdir. But we can
still use the cached dentries to conclude -ENOENT for lookup.
This patch introduces a new inode flag indicating if child dentries are
ordered. The flag is set at the same time marking a directory complete.
After creating/deleting/renaming file, we clear the flag on directory
inode. This prevents ceph_readdir() from using cached dentries to satisfy
readdir syscall.
Signed-off-by: Yan, Zheng <zyan@redhat.com>
Current code set new file/directory's initial ACL in a non-atomic
manner.
Client first sends request to MDS to create new file/directory, then set
the initial ACL after the new file/directory is successfully created.
The fix is include the initial ACL in create/mkdir/mknod MDS requests.
So MDS can handle creating file/directory and setting the initial ACL in
one request.
Signed-off-by: Yan, Zheng <zyan@redhat.com>
Reviewed-by: Sage Weil <sage@redhat.com>
Following sequence of events can happen.
- Client releases an inode, queues cap release message.
- A 'lookup' reply brings the same inode back, but the reply
doesn't contain xattrs because MDS didn't receive the cap release
message and thought client already has up-to-data xattrs.
The fix is force sending a getattr request to MDS if xattrs_version
is 0. The getattr mask is set to CEPH_STAT_CAP_XATTR, so MDS knows client
does not have xattr.
Signed-off-by: Yan, Zheng <zyan@redhat.com>
So that ceph_add_cap() can be used while i_ceph_lock is locked.
This simplifies the code that handle cap import/export.
Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
When creating a file, ceph_set_dentry_offset() puts the new dentry
at the end of directory's d_subdirs, then set the dentry's offset
based on directory's max offset. The offset does not reflect the
real postion of the dentry in directory. Later readdir reply from
MDS may change the dentry's position/offset. This inconsistency
can cause missing/duplicate entries in readdir result if readdir
is partly satisfied by dcache_readdir().
The fix is clear directory's completeness after creating/renaming
file. It prevents later readdir from using dcache_readdir().
Fixes: http://tracker.ceph.com/issues/8025
Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Reviewed-by: Sage Weil <sage@inktank.com>
flock and posix lock should use fl->fl_file instead of process ID
as owner identifier. (posix lock uses fl->fl_owner. fl->fl_owner
is usually equal to fl->fl_file, but it also can be a customized
value). The process ID of who holds the lock is just for F_GETLK
fcntl(2).
The fix is rename the 'pid' fields of struct ceph_mds_request_args
and struct ceph_filelock to 'owner', rename 'pid_namespace' fields
to 'pid'. Assign fl->fl_file to the 'owner' field of lock messages.
We also set the most significant bit of the 'owner' field. MDS can
use that bit to distinguish between old and new clients.
The MDS counterpart of this patch modifies the flock code to not
take the 'pid_namespace' into consideration when checking conflict
locks.
Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Comparing offset with inode->i_sb->s_maxbytes doesn't make sense for
directory. For a fragmented directory, offset (frag_t, off) can be
larger than inode->i_sb->s_maxbytes.
At the very beginning of ceph_dir_llseek(), local variable old_offset
is initialized to parameter offset. This doesn't make sense neither.
Old_offset should be ceph_make_fpos(fi->frag, fi->next_offset).
Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Reviewed-by: Alex Elder <elder@linaro.org>
The merge of commit 7221fe4c2e ("ceph: add acl for cephfs") raced with
upstream changes in the generic POSIX ACL code (eg commit 2aeccbe957
"fs: add generic xattr_acl handlers" and others).
Some of the fallout was fixed in commit 4db658ea0c ("ceph: Fix up after
semantic merge conflict"), but it was incomplete: the set_acl
inode_operation wasn't getting set, and the prototype needed to be
adjusted a bit (it doesn't take a dentry anymore).
Signed-off-by: Sage Weil <sage@inktank.com>
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The previous ceph-client merge resulted in ceph not even building,
because there was a merge conflict that wasn't visible as an actual data
conflict: commit 7221fe4c2e ("ceph: add acl for cephfs") added support
for POSIX ACL's into Ceph, but unluckily we also had the VFS tree change
a lot of the POSIX ACL helper functions to be much more helpful to
filesystems (see for example commits 2aeccbe957 "fs: add generic
xattr_acl handlers", 5bf3258fd2 "fs: make posix_acl_chmod more useful"
and 37bc15392a "fs: make posix_acl_create more useful")
The reason this conflict wasn't obvious was many-fold: because it was a
semantic conflict rather than a data conflict, it wasn't visible in the
git merge as a conflict. And because the VFS tree hadn't been in
linux-next, people hadn't become aware of it that way. And because I
was at jury duty this morning, I was using my laptop and as a result not
doing constant "allmodconfig" builds.
Anyway, this fixes the build and generally removes a fair chunk of the
Ceph POSIX ACL support code, since the improved helpers seem to match
really well for Ceph too. But I don't actually have any way to *test*
the end result, and I was really hoping for some ACK's for this. Oh,
well.
Not compiling certainly doesn't make things easier to test, so I'm
committing this without the acks after having waited for four hours...
Plus it's what I would have done for the merge had I noticed the
semantic conflict..
Reported-by: Dave Jones <davej@redhat.com>
Cc: Sage Weil <sage@inktank.com>
Cc: Guangliang Zhao <lucienchao@gmail.com>
Cc: Li Wang <li.wang@ubuntykylin.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Version 3 cap export message includes information about the imported
caps. It allows us to add the imported caps if the corresponding cap
import message still hasn't been received.
This allow us to handle situation that the importer MDS crashes and
the cap import message is missing.
Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Some inodes in readdir reply may have no caps. Getattr mds request
for these inodes can return -ESTALE. The fix is consider dentry that
links to inode with no caps as invalid. Invalid dentry causes a
lookup request to send to the mds, the MDS will send caps back.
Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
handle following sequence of events:
- non-auth MDS revokes Fc cap. queue invalidate work
- auth MDS issues Fc cap through request reply. i_rdcache_gen gets
increased.
- invalidate work runs. it finds i_rdcache_revoking != i_rdcache_gen,
so it does nothing.
Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Linus Torvalds
Yan, Zheng
Andreas Gruenbacher
Al Viro
Ilya Dryomov
Fabian Frederick
John Spray
Yan, Zheng
Guangliang Zhao
Peter Rosin
Sage Weil