1 #include "ceph_debug.h"
3 #include <linux/sort.h>
4 #include <linux/slab.h>
10 * Snapshots in ceph are driven in large part by cooperation from the
11 * client. In contrast to local file systems or file servers that
12 * implement snapshots at a single point in the system, ceph's
13 * distributed access to storage requires clients to help decide
14 * whether a write logically occurs before or after a recently created
17 * This provides a perfect instantanous client-wide snapshot. Between
18 * clients, however, snapshots may appear to be applied at slightly
19 * different points in time, depending on delays in delivering the
20 * snapshot notification.
22 * Snapshots are _not_ file system-wide. Instead, each snapshot
23 * applies to the subdirectory nested beneath some directory. This
24 * effectively divides the hierarchy into multiple "realms," where all
25 * of the files contained by each realm share the same set of
26 * snapshots. An individual realm's snap set contains snapshots
27 * explicitly created on that realm, as well as any snaps in its
28 * parent's snap set _after_ the point at which the parent became it's
29 * parent (due to, say, a rename). Similarly, snaps from prior parents
30 * during the time intervals during which they were the parent are included.
32 * The client is spared most of this detail, fortunately... it must only
33 * maintains a hierarchy of realms reflecting the current parent/child
34 * realm relationship, and for each realm has an explicit list of snaps
35 * inherited from prior parents.
37 * A snap_realm struct is maintained for realms containing every inode
38 * with an open cap in the system. (The needed snap realm information is
39 * provided by the MDS whenever a cap is issued, i.e., on open.) A 'seq'
40 * version number is used to ensure that as realm parameters change (new
41 * snapshot, new parent, etc.) the client's realm hierarchy is updated.
43 * The realm hierarchy drives the generation of a 'snap context' for each
44 * realm, which simply lists the resulting set of snaps for the realm. This
45 * is attached to any writes sent to OSDs.
48 * Unfortunately error handling is a bit mixed here. If we get a snap
49 * update, but don't have enough memory to update our realm hierarchy,
50 * it's not clear what we can do about it (besides complaining to the
56 * increase ref count for the realm
58 * caller must hold snap_rwsem for write.
60 void ceph_get_snap_realm(struct ceph_mds_client
*mdsc
,
61 struct ceph_snap_realm
*realm
)
63 dout("get_realm %p %d -> %d\n", realm
,
64 atomic_read(&realm
->nref
), atomic_read(&realm
->nref
)+1);
66 * since we _only_ increment realm refs or empty the empty
67 * list with snap_rwsem held, adjusting the empty list here is
68 * safe. we do need to protect against concurrent empty list
71 if (atomic_read(&realm
->nref
) == 0) {
72 spin_lock(&mdsc
->snap_empty_lock
);
73 list_del_init(&realm
->empty_item
);
74 spin_unlock(&mdsc
->snap_empty_lock
);
77 atomic_inc(&realm
->nref
);
80 static void __insert_snap_realm(struct rb_root
*root
,
81 struct ceph_snap_realm
*new)
83 struct rb_node
**p
= &root
->rb_node
;
84 struct rb_node
*parent
= NULL
;
85 struct ceph_snap_realm
*r
= NULL
;
89 r
= rb_entry(parent
, struct ceph_snap_realm
, node
);
90 if (new->ino
< r
->ino
)
92 else if (new->ino
> r
->ino
)
98 rb_link_node(&new->node
, parent
, p
);
99 rb_insert_color(&new->node
, root
);
103 * create and get the realm rooted at @ino and bump its ref count.
105 * caller must hold snap_rwsem for write.
107 static struct ceph_snap_realm
*ceph_create_snap_realm(
108 struct ceph_mds_client
*mdsc
,
111 struct ceph_snap_realm
*realm
;
113 realm
= kzalloc(sizeof(*realm
), GFP_NOFS
);
115 return ERR_PTR(-ENOMEM
);
117 atomic_set(&realm
->nref
, 0); /* tree does not take a ref */
119 INIT_LIST_HEAD(&realm
->children
);
120 INIT_LIST_HEAD(&realm
->child_item
);
121 INIT_LIST_HEAD(&realm
->empty_item
);
122 INIT_LIST_HEAD(&realm
->inodes_with_caps
);
123 spin_lock_init(&realm
->inodes_with_caps_lock
);
124 __insert_snap_realm(&mdsc
->snap_realms
, realm
);
125 dout("create_snap_realm %llx %p\n", realm
->ino
, realm
);
130 * lookup the realm rooted at @ino.
132 * caller must hold snap_rwsem for write.
134 struct ceph_snap_realm
*ceph_lookup_snap_realm(struct ceph_mds_client
*mdsc
,
137 struct rb_node
*n
= mdsc
->snap_realms
.rb_node
;
138 struct ceph_snap_realm
*r
;
141 r
= rb_entry(n
, struct ceph_snap_realm
, node
);
144 else if (ino
> r
->ino
)
147 dout("lookup_snap_realm %llx %p\n", r
->ino
, r
);
154 static void __put_snap_realm(struct ceph_mds_client
*mdsc
,
155 struct ceph_snap_realm
*realm
);
158 * called with snap_rwsem (write)
160 static void __destroy_snap_realm(struct ceph_mds_client
*mdsc
,
161 struct ceph_snap_realm
*realm
)
163 dout("__destroy_snap_realm %p %llx\n", realm
, realm
->ino
);
165 rb_erase(&realm
->node
, &mdsc
->snap_realms
);
168 list_del_init(&realm
->child_item
);
169 __put_snap_realm(mdsc
, realm
->parent
);
172 kfree(realm
->prior_parent_snaps
);
174 ceph_put_snap_context(realm
->cached_context
);
179 * caller holds snap_rwsem (write)
181 static void __put_snap_realm(struct ceph_mds_client
*mdsc
,
182 struct ceph_snap_realm
*realm
)
184 dout("__put_snap_realm %llx %p %d -> %d\n", realm
->ino
, realm
,
185 atomic_read(&realm
->nref
), atomic_read(&realm
->nref
)-1);
186 if (atomic_dec_and_test(&realm
->nref
))
187 __destroy_snap_realm(mdsc
, realm
);
191 * caller needn't hold any locks
193 void ceph_put_snap_realm(struct ceph_mds_client
*mdsc
,
194 struct ceph_snap_realm
*realm
)
196 dout("put_snap_realm %llx %p %d -> %d\n", realm
->ino
, realm
,
197 atomic_read(&realm
->nref
), atomic_read(&realm
->nref
)-1);
198 if (!atomic_dec_and_test(&realm
->nref
))
201 if (down_write_trylock(&mdsc
->snap_rwsem
)) {
202 __destroy_snap_realm(mdsc
, realm
);
203 up_write(&mdsc
->snap_rwsem
);
205 spin_lock(&mdsc
->snap_empty_lock
);
206 list_add(&mdsc
->snap_empty
, &realm
->empty_item
);
207 spin_unlock(&mdsc
->snap_empty_lock
);
212 * Clean up any realms whose ref counts have dropped to zero. Note
213 * that this does not include realms who were created but not yet
216 * Called under snap_rwsem (write)
218 static void __cleanup_empty_realms(struct ceph_mds_client
*mdsc
)
220 struct ceph_snap_realm
*realm
;
222 spin_lock(&mdsc
->snap_empty_lock
);
223 while (!list_empty(&mdsc
->snap_empty
)) {
224 realm
= list_first_entry(&mdsc
->snap_empty
,
225 struct ceph_snap_realm
, empty_item
);
226 list_del(&realm
->empty_item
);
227 spin_unlock(&mdsc
->snap_empty_lock
);
228 __destroy_snap_realm(mdsc
, realm
);
229 spin_lock(&mdsc
->snap_empty_lock
);
231 spin_unlock(&mdsc
->snap_empty_lock
);
234 void ceph_cleanup_empty_realms(struct ceph_mds_client
*mdsc
)
236 down_write(&mdsc
->snap_rwsem
);
237 __cleanup_empty_realms(mdsc
);
238 up_write(&mdsc
->snap_rwsem
);
242 * adjust the parent realm of a given @realm. adjust child list, and parent
243 * pointers, and ref counts appropriately.
245 * return true if parent was changed, 0 if unchanged, <0 on error.
247 * caller must hold snap_rwsem for write.
249 static int adjust_snap_realm_parent(struct ceph_mds_client
*mdsc
,
250 struct ceph_snap_realm
*realm
,
253 struct ceph_snap_realm
*parent
;
255 if (realm
->parent_ino
== parentino
)
258 parent
= ceph_lookup_snap_realm(mdsc
, parentino
);
260 parent
= ceph_create_snap_realm(mdsc
, parentino
);
262 return PTR_ERR(parent
);
264 dout("adjust_snap_realm_parent %llx %p: %llx %p -> %llx %p\n",
265 realm
->ino
, realm
, realm
->parent_ino
, realm
->parent
,
268 list_del_init(&realm
->child_item
);
269 ceph_put_snap_realm(mdsc
, realm
->parent
);
271 realm
->parent_ino
= parentino
;
272 realm
->parent
= parent
;
273 ceph_get_snap_realm(mdsc
, parent
);
274 list_add(&realm
->child_item
, &parent
->children
);
279 static int cmpu64_rev(const void *a
, const void *b
)
281 if (*(u64
*)a
< *(u64
*)b
)
283 if (*(u64
*)a
> *(u64
*)b
)
289 * build the snap context for a given realm.
291 static int build_snap_context(struct ceph_snap_realm
*realm
)
293 struct ceph_snap_realm
*parent
= realm
->parent
;
294 struct ceph_snap_context
*snapc
;
297 int num
= realm
->num_prior_parent_snaps
+ realm
->num_snaps
;
300 * build parent context, if it hasn't been built.
301 * conservatively estimate that all parent snaps might be
305 if (!parent
->cached_context
) {
306 err
= build_snap_context(parent
);
310 num
+= parent
->cached_context
->num_snaps
;
313 /* do i actually need to update? not if my context seq
314 matches realm seq, and my parents' does to. (this works
315 because we rebuild_snap_realms() works _downward_ in
316 hierarchy after each update.) */
317 if (realm
->cached_context
&&
318 realm
->cached_context
->seq
== realm
->seq
&&
320 realm
->cached_context
->seq
>= parent
->cached_context
->seq
)) {
321 dout("build_snap_context %llx %p: %p seq %lld (%d snaps)"
323 realm
->ino
, realm
, realm
->cached_context
,
324 realm
->cached_context
->seq
,
325 realm
->cached_context
->num_snaps
);
329 /* alloc new snap context */
331 if (num
> ULONG_MAX
/ sizeof(u64
) - sizeof(*snapc
))
333 snapc
= kzalloc(sizeof(*snapc
) + num
*sizeof(u64
), GFP_NOFS
);
336 atomic_set(&snapc
->nref
, 1);
338 /* build (reverse sorted) snap vector */
340 snapc
->seq
= realm
->seq
;
342 /* include any of parent's snaps occuring _after_ my
343 parent became my parent */
344 for (i
= 0; i
< parent
->cached_context
->num_snaps
; i
++)
345 if (parent
->cached_context
->snaps
[i
] >=
347 snapc
->snaps
[num
++] =
348 parent
->cached_context
->snaps
[i
];
349 if (parent
->cached_context
->seq
> snapc
->seq
)
350 snapc
->seq
= parent
->cached_context
->seq
;
352 memcpy(snapc
->snaps
+ num
, realm
->snaps
,
353 sizeof(u64
)*realm
->num_snaps
);
354 num
+= realm
->num_snaps
;
355 memcpy(snapc
->snaps
+ num
, realm
->prior_parent_snaps
,
356 sizeof(u64
)*realm
->num_prior_parent_snaps
);
357 num
+= realm
->num_prior_parent_snaps
;
359 sort(snapc
->snaps
, num
, sizeof(u64
), cmpu64_rev
, NULL
);
360 snapc
->num_snaps
= num
;
361 dout("build_snap_context %llx %p: %p seq %lld (%d snaps)\n",
362 realm
->ino
, realm
, snapc
, snapc
->seq
, snapc
->num_snaps
);
364 if (realm
->cached_context
)
365 ceph_put_snap_context(realm
->cached_context
);
366 realm
->cached_context
= snapc
;
371 * if we fail, clear old (incorrect) cached_context... hopefully
372 * we'll have better luck building it later
374 if (realm
->cached_context
) {
375 ceph_put_snap_context(realm
->cached_context
);
376 realm
->cached_context
= NULL
;
378 pr_err("build_snap_context %llx %p fail %d\n", realm
->ino
,
384 * rebuild snap context for the given realm and all of its children.
386 static void rebuild_snap_realms(struct ceph_snap_realm
*realm
)
388 struct ceph_snap_realm
*child
;
390 dout("rebuild_snap_realms %llx %p\n", realm
->ino
, realm
);
391 build_snap_context(realm
);
393 list_for_each_entry(child
, &realm
->children
, child_item
)
394 rebuild_snap_realms(child
);
399 * helper to allocate and decode an array of snapids. free prior
402 static int dup_array(u64
**dst
, __le64
*src
, int num
)
408 *dst
= kcalloc(num
, sizeof(u64
), GFP_NOFS
);
411 for (i
= 0; i
< num
; i
++)
412 (*dst
)[i
] = get_unaligned_le64(src
+ i
);
421 * When a snapshot is applied, the size/mtime inode metadata is queued
422 * in a ceph_cap_snap (one for each snapshot) until writeback
423 * completes and the metadata can be flushed back to the MDS.
425 * However, if a (sync) write is currently in-progress when we apply
426 * the snapshot, we have to wait until the write succeeds or fails
427 * (and a final size/mtime is known). In this case the
428 * cap_snap->writing = 1, and is said to be "pending." When the write
429 * finishes, we __ceph_finish_cap_snap().
431 * Caller must hold snap_rwsem for read (i.e., the realm topology won't
434 void ceph_queue_cap_snap(struct ceph_inode_info
*ci
)
436 struct inode
*inode
= &ci
->vfs_inode
;
437 struct ceph_cap_snap
*capsnap
;
440 capsnap
= kzalloc(sizeof(*capsnap
), GFP_NOFS
);
442 pr_err("ENOMEM allocating ceph_cap_snap on %p\n", inode
);
446 spin_lock(&inode
->i_lock
);
447 used
= __ceph_caps_used(ci
);
448 dirty
= __ceph_caps_dirty(ci
);
449 if (__ceph_have_pending_cap_snap(ci
)) {
450 /* there is no point in queuing multiple "pending" cap_snaps,
451 as no new writes are allowed to start when pending, so any
452 writes in progress now were started before the previous
453 cap_snap. lucky us. */
454 dout("queue_cap_snap %p already pending\n", inode
);
456 } else if (ci
->i_wrbuffer_ref_head
|| (used
& CEPH_CAP_FILE_WR
) ||
457 (dirty
& (CEPH_CAP_AUTH_EXCL
|CEPH_CAP_XATTR_EXCL
|
458 CEPH_CAP_FILE_EXCL
|CEPH_CAP_FILE_WR
))) {
459 struct ceph_snap_context
*snapc
= ci
->i_head_snapc
;
461 dout("queue_cap_snap %p cap_snap %p queuing under %p\n", inode
,
465 atomic_set(&capsnap
->nref
, 1);
467 INIT_LIST_HEAD(&capsnap
->ci_item
);
468 INIT_LIST_HEAD(&capsnap
->flushing_item
);
470 capsnap
->follows
= snapc
->seq
- 1;
471 capsnap
->issued
= __ceph_caps_issued(ci
, NULL
);
472 capsnap
->dirty
= dirty
;
474 capsnap
->mode
= inode
->i_mode
;
475 capsnap
->uid
= inode
->i_uid
;
476 capsnap
->gid
= inode
->i_gid
;
478 if (dirty
& CEPH_CAP_XATTR_EXCL
) {
479 __ceph_build_xattrs_blob(ci
);
480 capsnap
->xattr_blob
=
481 ceph_buffer_get(ci
->i_xattrs
.blob
);
482 capsnap
->xattr_version
= ci
->i_xattrs
.version
;
484 capsnap
->xattr_blob
= NULL
;
485 capsnap
->xattr_version
= 0;
488 /* dirty page count moved from _head to this cap_snap;
489 all subsequent writes page dirties occur _after_ this
491 capsnap
->dirty_pages
= ci
->i_wrbuffer_ref_head
;
492 ci
->i_wrbuffer_ref_head
= 0;
493 capsnap
->context
= snapc
;
495 ceph_get_snap_context(ci
->i_snap_realm
->cached_context
);
496 dout(" new snapc is %p\n", ci
->i_head_snapc
);
497 list_add_tail(&capsnap
->ci_item
, &ci
->i_cap_snaps
);
499 if (used
& CEPH_CAP_FILE_WR
) {
500 dout("queue_cap_snap %p cap_snap %p snapc %p"
501 " seq %llu used WR, now pending\n", inode
,
502 capsnap
, snapc
, snapc
->seq
);
503 capsnap
->writing
= 1;
505 /* note mtime, size NOW. */
506 __ceph_finish_cap_snap(ci
, capsnap
);
509 dout("queue_cap_snap %p nothing dirty|writing\n", inode
);
513 spin_unlock(&inode
->i_lock
);
517 * Finalize the size, mtime for a cap_snap.. that is, settle on final values
518 * to be used for the snapshot, to be flushed back to the mds.
520 * If capsnap can now be flushed, add to snap_flush list, and return 1.
522 * Caller must hold i_lock.
524 int __ceph_finish_cap_snap(struct ceph_inode_info
*ci
,
525 struct ceph_cap_snap
*capsnap
)
527 struct inode
*inode
= &ci
->vfs_inode
;
528 struct ceph_mds_client
*mdsc
= &ceph_sb_to_client(inode
->i_sb
)->mdsc
;
530 BUG_ON(capsnap
->writing
);
531 capsnap
->size
= inode
->i_size
;
532 capsnap
->mtime
= inode
->i_mtime
;
533 capsnap
->atime
= inode
->i_atime
;
534 capsnap
->ctime
= inode
->i_ctime
;
535 capsnap
->time_warp_seq
= ci
->i_time_warp_seq
;
536 if (capsnap
->dirty_pages
) {
537 dout("finish_cap_snap %p cap_snap %p snapc %p %llu %s s=%llu "
538 "still has %d dirty pages\n", inode
, capsnap
,
539 capsnap
->context
, capsnap
->context
->seq
,
540 ceph_cap_string(capsnap
->dirty
), capsnap
->size
,
541 capsnap
->dirty_pages
);
544 dout("finish_cap_snap %p cap_snap %p snapc %p %llu %s s=%llu\n",
545 inode
, capsnap
, capsnap
->context
,
546 capsnap
->context
->seq
, ceph_cap_string(capsnap
->dirty
),
549 spin_lock(&mdsc
->snap_flush_lock
);
550 list_add_tail(&ci
->i_snap_flush_item
, &mdsc
->snap_flush_list
);
551 spin_unlock(&mdsc
->snap_flush_lock
);
552 return 1; /* caller may want to ceph_flush_snaps */
556 * Queue cap_snaps for snap writeback for this realm and its children.
557 * Called under snap_rwsem, so realm topology won't change.
559 static void queue_realm_cap_snaps(struct ceph_snap_realm
*realm
)
561 struct ceph_inode_info
*ci
;
562 struct inode
*lastinode
= NULL
;
563 struct ceph_snap_realm
*child
;
565 dout("queue_realm_cap_snaps %p %llx inodes\n", realm
, realm
->ino
);
567 spin_lock(&realm
->inodes_with_caps_lock
);
568 list_for_each_entry(ci
, &realm
->inodes_with_caps
,
570 struct inode
*inode
= igrab(&ci
->vfs_inode
);
573 spin_unlock(&realm
->inodes_with_caps_lock
);
577 ceph_queue_cap_snap(ci
);
578 spin_lock(&realm
->inodes_with_caps_lock
);
580 spin_unlock(&realm
->inodes_with_caps_lock
);
584 dout("queue_realm_cap_snaps %p %llx children\n", realm
, realm
->ino
);
585 list_for_each_entry(child
, &realm
->children
, child_item
)
586 queue_realm_cap_snaps(child
);
588 dout("queue_realm_cap_snaps %p %llx done\n", realm
, realm
->ino
);
592 * Parse and apply a snapblob "snap trace" from the MDS. This specifies
593 * the snap realm parameters from a given realm and all of its ancestors,
596 * Caller must hold snap_rwsem for write.
598 int ceph_update_snap_trace(struct ceph_mds_client
*mdsc
,
599 void *p
, void *e
, bool deletion
)
601 struct ceph_mds_snap_realm
*ri
; /* encoded */
602 __le64
*snaps
; /* encoded */
603 __le64
*prior_parent_snaps
; /* encoded */
604 struct ceph_snap_realm
*realm
;
608 dout("update_snap_trace deletion=%d\n", deletion
);
610 ceph_decode_need(&p
, e
, sizeof(*ri
), bad
);
613 ceph_decode_need(&p
, e
, sizeof(u64
)*(le32_to_cpu(ri
->num_snaps
) +
614 le32_to_cpu(ri
->num_prior_parent_snaps
)), bad
);
616 p
+= sizeof(u64
) * le32_to_cpu(ri
->num_snaps
);
617 prior_parent_snaps
= p
;
618 p
+= sizeof(u64
) * le32_to_cpu(ri
->num_prior_parent_snaps
);
620 realm
= ceph_lookup_snap_realm(mdsc
, le64_to_cpu(ri
->ino
));
622 realm
= ceph_create_snap_realm(mdsc
, le64_to_cpu(ri
->ino
));
624 err
= PTR_ERR(realm
);
629 if (le64_to_cpu(ri
->seq
) > realm
->seq
) {
630 dout("update_snap_trace updating %llx %p %lld -> %lld\n",
631 realm
->ino
, realm
, realm
->seq
, le64_to_cpu(ri
->seq
));
633 * if the realm seq has changed, queue a cap_snap for every
634 * inode with open caps. we do this _before_ we update
635 * the realm info so that we prepare for writeback under the
636 * _previous_ snap context.
638 * ...unless it's a snap deletion!
641 queue_realm_cap_snaps(realm
);
643 dout("update_snap_trace %llx %p seq %lld unchanged\n",
644 realm
->ino
, realm
, realm
->seq
);
647 /* ensure the parent is correct */
648 err
= adjust_snap_realm_parent(mdsc
, realm
, le64_to_cpu(ri
->parent
));
653 if (le64_to_cpu(ri
->seq
) > realm
->seq
) {
654 /* update realm parameters, snap lists */
655 realm
->seq
= le64_to_cpu(ri
->seq
);
656 realm
->created
= le64_to_cpu(ri
->created
);
657 realm
->parent_since
= le64_to_cpu(ri
->parent_since
);
659 realm
->num_snaps
= le32_to_cpu(ri
->num_snaps
);
660 err
= dup_array(&realm
->snaps
, snaps
, realm
->num_snaps
);
664 realm
->num_prior_parent_snaps
=
665 le32_to_cpu(ri
->num_prior_parent_snaps
);
666 err
= dup_array(&realm
->prior_parent_snaps
, prior_parent_snaps
,
667 realm
->num_prior_parent_snaps
);
672 } else if (!realm
->cached_context
) {
676 dout("done with %llx %p, invalidated=%d, %p %p\n", realm
->ino
,
677 realm
, invalidate
, p
, e
);
682 /* invalidate when we reach the _end_ (root) of the trace */
684 rebuild_snap_realms(realm
);
686 __cleanup_empty_realms(mdsc
);
692 pr_err("update_snap_trace error %d\n", err
);
698 * Send any cap_snaps that are queued for flush. Try to carry
699 * s_mutex across multiple snap flushes to avoid locking overhead.
701 * Caller holds no locks.
703 static void flush_snaps(struct ceph_mds_client
*mdsc
)
705 struct ceph_inode_info
*ci
;
707 struct ceph_mds_session
*session
= NULL
;
709 dout("flush_snaps\n");
710 spin_lock(&mdsc
->snap_flush_lock
);
711 while (!list_empty(&mdsc
->snap_flush_list
)) {
712 ci
= list_first_entry(&mdsc
->snap_flush_list
,
713 struct ceph_inode_info
, i_snap_flush_item
);
714 inode
= &ci
->vfs_inode
;
716 spin_unlock(&mdsc
->snap_flush_lock
);
717 spin_lock(&inode
->i_lock
);
718 __ceph_flush_snaps(ci
, &session
);
719 spin_unlock(&inode
->i_lock
);
721 spin_lock(&mdsc
->snap_flush_lock
);
723 spin_unlock(&mdsc
->snap_flush_lock
);
726 mutex_unlock(&session
->s_mutex
);
727 ceph_put_mds_session(session
);
729 dout("flush_snaps done\n");
734 * Handle a snap notification from the MDS.
736 * This can take two basic forms: the simplest is just a snap creation
737 * or deletion notification on an existing realm. This should update the
738 * realm and its children.
740 * The more difficult case is realm creation, due to snap creation at a
741 * new point in the file hierarchy, or due to a rename that moves a file or
742 * directory into another realm.
744 void ceph_handle_snap(struct ceph_mds_client
*mdsc
,
745 struct ceph_mds_session
*session
,
746 struct ceph_msg
*msg
)
748 struct super_block
*sb
= mdsc
->client
->sb
;
749 int mds
= session
->s_mds
;
753 struct ceph_snap_realm
*realm
= NULL
;
754 void *p
= msg
->front
.iov_base
;
755 void *e
= p
+ msg
->front
.iov_len
;
756 struct ceph_mds_snap_head
*h
;
757 int num_split_inos
, num_split_realms
;
758 __le64
*split_inos
= NULL
, *split_realms
= NULL
;
760 int locked_rwsem
= 0;
763 if (msg
->front
.iov_len
< sizeof(*h
))
766 op
= le32_to_cpu(h
->op
);
767 split
= le64_to_cpu(h
->split
); /* non-zero if we are splitting an
769 num_split_inos
= le32_to_cpu(h
->num_split_inos
);
770 num_split_realms
= le32_to_cpu(h
->num_split_realms
);
771 trace_len
= le32_to_cpu(h
->trace_len
);
774 dout("handle_snap from mds%d op %s split %llx tracelen %d\n", mds
,
775 ceph_snap_op_name(op
), split
, trace_len
);
777 mutex_lock(&session
->s_mutex
);
779 mutex_unlock(&session
->s_mutex
);
781 down_write(&mdsc
->snap_rwsem
);
784 if (op
== CEPH_SNAP_OP_SPLIT
) {
785 struct ceph_mds_snap_realm
*ri
;
788 * A "split" breaks part of an existing realm off into
789 * a new realm. The MDS provides a list of inodes
790 * (with caps) and child realms that belong to the new
794 p
+= sizeof(u64
) * num_split_inos
;
796 p
+= sizeof(u64
) * num_split_realms
;
797 ceph_decode_need(&p
, e
, sizeof(*ri
), bad
);
798 /* we will peek at realm info here, but will _not_
799 * advance p, as the realm update will occur below in
800 * ceph_update_snap_trace. */
803 realm
= ceph_lookup_snap_realm(mdsc
, split
);
805 realm
= ceph_create_snap_realm(mdsc
, split
);
809 ceph_get_snap_realm(mdsc
, realm
);
811 dout("splitting snap_realm %llx %p\n", realm
->ino
, realm
);
812 for (i
= 0; i
< num_split_inos
; i
++) {
813 struct ceph_vino vino
= {
814 .ino
= le64_to_cpu(split_inos
[i
]),
817 struct inode
*inode
= ceph_find_inode(sb
, vino
);
818 struct ceph_inode_info
*ci
;
822 ci
= ceph_inode(inode
);
824 spin_lock(&inode
->i_lock
);
825 if (!ci
->i_snap_realm
)
828 * If this inode belongs to a realm that was
829 * created after our new realm, we experienced
830 * a race (due to another split notifications
831 * arriving from a different MDS). So skip
834 if (ci
->i_snap_realm
->created
>
835 le64_to_cpu(ri
->created
)) {
836 dout(" leaving %p in newer realm %llx %p\n",
837 inode
, ci
->i_snap_realm
->ino
,
841 dout(" will move %p to split realm %llx %p\n",
842 inode
, realm
->ino
, realm
);
844 * Remove the inode from the realm's inode
845 * list, but don't add it to the new realm
846 * yet. We don't want the cap_snap to be
847 * queued (again) by ceph_update_snap_trace()
848 * below. Queue it _now_, under the old context.
850 spin_lock(&realm
->inodes_with_caps_lock
);
851 list_del_init(&ci
->i_snap_realm_item
);
852 spin_unlock(&realm
->inodes_with_caps_lock
);
853 spin_unlock(&inode
->i_lock
);
855 ceph_queue_cap_snap(ci
);
861 spin_unlock(&inode
->i_lock
);
865 /* we may have taken some of the old realm's children. */
866 for (i
= 0; i
< num_split_realms
; i
++) {
867 struct ceph_snap_realm
*child
=
868 ceph_lookup_snap_realm(mdsc
,
869 le64_to_cpu(split_realms
[i
]));
872 adjust_snap_realm_parent(mdsc
, child
, realm
->ino
);
877 * update using the provided snap trace. if we are deleting a
878 * snap, we can avoid queueing cap_snaps.
880 ceph_update_snap_trace(mdsc
, p
, e
,
881 op
== CEPH_SNAP_OP_DESTROY
);
883 if (op
== CEPH_SNAP_OP_SPLIT
) {
885 * ok, _now_ add the inodes into the new realm.
887 for (i
= 0; i
< num_split_inos
; i
++) {
888 struct ceph_vino vino
= {
889 .ino
= le64_to_cpu(split_inos
[i
]),
892 struct inode
*inode
= ceph_find_inode(sb
, vino
);
893 struct ceph_inode_info
*ci
;
897 ci
= ceph_inode(inode
);
898 spin_lock(&inode
->i_lock
);
899 if (list_empty(&ci
->i_snap_realm_item
)) {
900 struct ceph_snap_realm
*oldrealm
=
903 dout(" moving %p to split realm %llx %p\n",
904 inode
, realm
->ino
, realm
);
905 spin_lock(&realm
->inodes_with_caps_lock
);
906 list_add(&ci
->i_snap_realm_item
,
907 &realm
->inodes_with_caps
);
908 ci
->i_snap_realm
= realm
;
909 spin_unlock(&realm
->inodes_with_caps_lock
);
910 ceph_get_snap_realm(mdsc
, realm
);
911 ceph_put_snap_realm(mdsc
, oldrealm
);
913 spin_unlock(&inode
->i_lock
);
917 /* we took a reference when we created the realm, above */
918 ceph_put_snap_realm(mdsc
, realm
);
921 __cleanup_empty_realms(mdsc
);
923 up_write(&mdsc
->snap_rwsem
);
929 pr_err("corrupt snap message from mds%d\n", mds
);
933 up_write(&mdsc
->snap_rwsem
);