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
->dirty_item
);
123 INIT_LIST_HEAD(&realm
->inodes_with_caps
);
124 spin_lock_init(&realm
->inodes_with_caps_lock
);
125 __insert_snap_realm(&mdsc
->snap_realms
, realm
);
126 dout("create_snap_realm %llx %p\n", realm
->ino
, realm
);
131 * lookup the realm rooted at @ino.
133 * caller must hold snap_rwsem for write.
135 struct ceph_snap_realm
*ceph_lookup_snap_realm(struct ceph_mds_client
*mdsc
,
138 struct rb_node
*n
= mdsc
->snap_realms
.rb_node
;
139 struct ceph_snap_realm
*r
;
142 r
= rb_entry(n
, struct ceph_snap_realm
, node
);
145 else if (ino
> r
->ino
)
148 dout("lookup_snap_realm %llx %p\n", r
->ino
, r
);
155 static void __put_snap_realm(struct ceph_mds_client
*mdsc
,
156 struct ceph_snap_realm
*realm
);
159 * called with snap_rwsem (write)
161 static void __destroy_snap_realm(struct ceph_mds_client
*mdsc
,
162 struct ceph_snap_realm
*realm
)
164 dout("__destroy_snap_realm %p %llx\n", realm
, realm
->ino
);
166 rb_erase(&realm
->node
, &mdsc
->snap_realms
);
169 list_del_init(&realm
->child_item
);
170 __put_snap_realm(mdsc
, realm
->parent
);
173 kfree(realm
->prior_parent_snaps
);
175 ceph_put_snap_context(realm
->cached_context
);
180 * caller holds snap_rwsem (write)
182 static void __put_snap_realm(struct ceph_mds_client
*mdsc
,
183 struct ceph_snap_realm
*realm
)
185 dout("__put_snap_realm %llx %p %d -> %d\n", realm
->ino
, realm
,
186 atomic_read(&realm
->nref
), atomic_read(&realm
->nref
)-1);
187 if (atomic_dec_and_test(&realm
->nref
))
188 __destroy_snap_realm(mdsc
, realm
);
192 * caller needn't hold any locks
194 void ceph_put_snap_realm(struct ceph_mds_client
*mdsc
,
195 struct ceph_snap_realm
*realm
)
197 dout("put_snap_realm %llx %p %d -> %d\n", realm
->ino
, realm
,
198 atomic_read(&realm
->nref
), atomic_read(&realm
->nref
)-1);
199 if (!atomic_dec_and_test(&realm
->nref
))
202 if (down_write_trylock(&mdsc
->snap_rwsem
)) {
203 __destroy_snap_realm(mdsc
, realm
);
204 up_write(&mdsc
->snap_rwsem
);
206 spin_lock(&mdsc
->snap_empty_lock
);
207 list_add(&mdsc
->snap_empty
, &realm
->empty_item
);
208 spin_unlock(&mdsc
->snap_empty_lock
);
213 * Clean up any realms whose ref counts have dropped to zero. Note
214 * that this does not include realms who were created but not yet
217 * Called under snap_rwsem (write)
219 static void __cleanup_empty_realms(struct ceph_mds_client
*mdsc
)
221 struct ceph_snap_realm
*realm
;
223 spin_lock(&mdsc
->snap_empty_lock
);
224 while (!list_empty(&mdsc
->snap_empty
)) {
225 realm
= list_first_entry(&mdsc
->snap_empty
,
226 struct ceph_snap_realm
, empty_item
);
227 list_del(&realm
->empty_item
);
228 spin_unlock(&mdsc
->snap_empty_lock
);
229 __destroy_snap_realm(mdsc
, realm
);
230 spin_lock(&mdsc
->snap_empty_lock
);
232 spin_unlock(&mdsc
->snap_empty_lock
);
235 void ceph_cleanup_empty_realms(struct ceph_mds_client
*mdsc
)
237 down_write(&mdsc
->snap_rwsem
);
238 __cleanup_empty_realms(mdsc
);
239 up_write(&mdsc
->snap_rwsem
);
243 * adjust the parent realm of a given @realm. adjust child list, and parent
244 * pointers, and ref counts appropriately.
246 * return true if parent was changed, 0 if unchanged, <0 on error.
248 * caller must hold snap_rwsem for write.
250 static int adjust_snap_realm_parent(struct ceph_mds_client
*mdsc
,
251 struct ceph_snap_realm
*realm
,
254 struct ceph_snap_realm
*parent
;
256 if (realm
->parent_ino
== parentino
)
259 parent
= ceph_lookup_snap_realm(mdsc
, parentino
);
261 parent
= ceph_create_snap_realm(mdsc
, parentino
);
263 return PTR_ERR(parent
);
265 dout("adjust_snap_realm_parent %llx %p: %llx %p -> %llx %p\n",
266 realm
->ino
, realm
, realm
->parent_ino
, realm
->parent
,
269 list_del_init(&realm
->child_item
);
270 ceph_put_snap_realm(mdsc
, realm
->parent
);
272 realm
->parent_ino
= parentino
;
273 realm
->parent
= parent
;
274 ceph_get_snap_realm(mdsc
, parent
);
275 list_add(&realm
->child_item
, &parent
->children
);
280 static int cmpu64_rev(const void *a
, const void *b
)
282 if (*(u64
*)a
< *(u64
*)b
)
284 if (*(u64
*)a
> *(u64
*)b
)
290 * build the snap context for a given realm.
292 static int build_snap_context(struct ceph_snap_realm
*realm
)
294 struct ceph_snap_realm
*parent
= realm
->parent
;
295 struct ceph_snap_context
*snapc
;
298 int num
= realm
->num_prior_parent_snaps
+ realm
->num_snaps
;
301 * build parent context, if it hasn't been built.
302 * conservatively estimate that all parent snaps might be
306 if (!parent
->cached_context
) {
307 err
= build_snap_context(parent
);
311 num
+= parent
->cached_context
->num_snaps
;
314 /* do i actually need to update? not if my context seq
315 matches realm seq, and my parents' does to. (this works
316 because we rebuild_snap_realms() works _downward_ in
317 hierarchy after each update.) */
318 if (realm
->cached_context
&&
319 realm
->cached_context
->seq
== realm
->seq
&&
321 realm
->cached_context
->seq
>= parent
->cached_context
->seq
)) {
322 dout("build_snap_context %llx %p: %p seq %lld (%d snaps)"
324 realm
->ino
, realm
, realm
->cached_context
,
325 realm
->cached_context
->seq
,
326 realm
->cached_context
->num_snaps
);
330 /* alloc new snap context */
332 if (num
> ULONG_MAX
/ sizeof(u64
) - sizeof(*snapc
))
334 snapc
= kzalloc(sizeof(*snapc
) + num
*sizeof(u64
), GFP_NOFS
);
337 atomic_set(&snapc
->nref
, 1);
339 /* build (reverse sorted) snap vector */
341 snapc
->seq
= realm
->seq
;
343 /* include any of parent's snaps occuring _after_ my
344 parent became my parent */
345 for (i
= 0; i
< parent
->cached_context
->num_snaps
; i
++)
346 if (parent
->cached_context
->snaps
[i
] >=
348 snapc
->snaps
[num
++] =
349 parent
->cached_context
->snaps
[i
];
350 if (parent
->cached_context
->seq
> snapc
->seq
)
351 snapc
->seq
= parent
->cached_context
->seq
;
353 memcpy(snapc
->snaps
+ num
, realm
->snaps
,
354 sizeof(u64
)*realm
->num_snaps
);
355 num
+= realm
->num_snaps
;
356 memcpy(snapc
->snaps
+ num
, realm
->prior_parent_snaps
,
357 sizeof(u64
)*realm
->num_prior_parent_snaps
);
358 num
+= realm
->num_prior_parent_snaps
;
360 sort(snapc
->snaps
, num
, sizeof(u64
), cmpu64_rev
, NULL
);
361 snapc
->num_snaps
= num
;
362 dout("build_snap_context %llx %p: %p seq %lld (%d snaps)\n",
363 realm
->ino
, realm
, snapc
, snapc
->seq
, snapc
->num_snaps
);
365 if (realm
->cached_context
)
366 ceph_put_snap_context(realm
->cached_context
);
367 realm
->cached_context
= snapc
;
372 * if we fail, clear old (incorrect) cached_context... hopefully
373 * we'll have better luck building it later
375 if (realm
->cached_context
) {
376 ceph_put_snap_context(realm
->cached_context
);
377 realm
->cached_context
= NULL
;
379 pr_err("build_snap_context %llx %p fail %d\n", realm
->ino
,
385 * rebuild snap context for the given realm and all of its children.
387 static void rebuild_snap_realms(struct ceph_snap_realm
*realm
)
389 struct ceph_snap_realm
*child
;
391 dout("rebuild_snap_realms %llx %p\n", realm
->ino
, realm
);
392 build_snap_context(realm
);
394 list_for_each_entry(child
, &realm
->children
, child_item
)
395 rebuild_snap_realms(child
);
400 * helper to allocate and decode an array of snapids. free prior
403 static int dup_array(u64
**dst
, __le64
*src
, int num
)
409 *dst
= kcalloc(num
, sizeof(u64
), GFP_NOFS
);
412 for (i
= 0; i
< num
; i
++)
413 (*dst
)[i
] = get_unaligned_le64(src
+ i
);
422 * When a snapshot is applied, the size/mtime inode metadata is queued
423 * in a ceph_cap_snap (one for each snapshot) until writeback
424 * completes and the metadata can be flushed back to the MDS.
426 * However, if a (sync) write is currently in-progress when we apply
427 * the snapshot, we have to wait until the write succeeds or fails
428 * (and a final size/mtime is known). In this case the
429 * cap_snap->writing = 1, and is said to be "pending." When the write
430 * finishes, we __ceph_finish_cap_snap().
432 * Caller must hold snap_rwsem for read (i.e., the realm topology won't
435 void ceph_queue_cap_snap(struct ceph_inode_info
*ci
)
437 struct inode
*inode
= &ci
->vfs_inode
;
438 struct ceph_cap_snap
*capsnap
;
441 capsnap
= kzalloc(sizeof(*capsnap
), GFP_NOFS
);
443 pr_err("ENOMEM allocating ceph_cap_snap on %p\n", inode
);
447 spin_lock(&inode
->i_lock
);
448 used
= __ceph_caps_used(ci
);
449 dirty
= __ceph_caps_dirty(ci
);
450 if (__ceph_have_pending_cap_snap(ci
)) {
451 /* there is no point in queuing multiple "pending" cap_snaps,
452 as no new writes are allowed to start when pending, so any
453 writes in progress now were started before the previous
454 cap_snap. lucky us. */
455 dout("queue_cap_snap %p already pending\n", inode
);
457 } else if (ci
->i_wrbuffer_ref_head
|| (used
& CEPH_CAP_FILE_WR
) ||
458 (dirty
& (CEPH_CAP_AUTH_EXCL
|CEPH_CAP_XATTR_EXCL
|
459 CEPH_CAP_FILE_EXCL
|CEPH_CAP_FILE_WR
))) {
460 struct ceph_snap_context
*snapc
= ci
->i_head_snapc
;
462 dout("queue_cap_snap %p cap_snap %p queuing under %p\n", inode
,
466 atomic_set(&capsnap
->nref
, 1);
468 INIT_LIST_HEAD(&capsnap
->ci_item
);
469 INIT_LIST_HEAD(&capsnap
->flushing_item
);
471 capsnap
->follows
= snapc
->seq
;
472 capsnap
->issued
= __ceph_caps_issued(ci
, NULL
);
473 capsnap
->dirty
= dirty
;
475 capsnap
->mode
= inode
->i_mode
;
476 capsnap
->uid
= inode
->i_uid
;
477 capsnap
->gid
= inode
->i_gid
;
479 if (dirty
& CEPH_CAP_XATTR_EXCL
) {
480 __ceph_build_xattrs_blob(ci
);
481 capsnap
->xattr_blob
=
482 ceph_buffer_get(ci
->i_xattrs
.blob
);
483 capsnap
->xattr_version
= ci
->i_xattrs
.version
;
485 capsnap
->xattr_blob
= NULL
;
486 capsnap
->xattr_version
= 0;
489 /* dirty page count moved from _head to this cap_snap;
490 all subsequent writes page dirties occur _after_ this
492 capsnap
->dirty_pages
= ci
->i_wrbuffer_ref_head
;
493 ci
->i_wrbuffer_ref_head
= 0;
494 capsnap
->context
= snapc
;
496 ceph_get_snap_context(ci
->i_snap_realm
->cached_context
);
497 dout(" new snapc is %p\n", ci
->i_head_snapc
);
498 list_add_tail(&capsnap
->ci_item
, &ci
->i_cap_snaps
);
500 if (used
& CEPH_CAP_FILE_WR
) {
501 dout("queue_cap_snap %p cap_snap %p snapc %p"
502 " seq %llu used WR, now pending\n", inode
,
503 capsnap
, snapc
, snapc
->seq
);
504 capsnap
->writing
= 1;
506 /* note mtime, size NOW. */
507 __ceph_finish_cap_snap(ci
, capsnap
);
510 dout("queue_cap_snap %p nothing dirty|writing\n", inode
);
514 spin_unlock(&inode
->i_lock
);
518 * Finalize the size, mtime for a cap_snap.. that is, settle on final values
519 * to be used for the snapshot, to be flushed back to the mds.
521 * If capsnap can now be flushed, add to snap_flush list, and return 1.
523 * Caller must hold i_lock.
525 int __ceph_finish_cap_snap(struct ceph_inode_info
*ci
,
526 struct ceph_cap_snap
*capsnap
)
528 struct inode
*inode
= &ci
->vfs_inode
;
529 struct ceph_mds_client
*mdsc
= &ceph_sb_to_client(inode
->i_sb
)->mdsc
;
531 BUG_ON(capsnap
->writing
);
532 capsnap
->size
= inode
->i_size
;
533 capsnap
->mtime
= inode
->i_mtime
;
534 capsnap
->atime
= inode
->i_atime
;
535 capsnap
->ctime
= inode
->i_ctime
;
536 capsnap
->time_warp_seq
= ci
->i_time_warp_seq
;
537 if (capsnap
->dirty_pages
) {
538 dout("finish_cap_snap %p cap_snap %p snapc %p %llu %s s=%llu "
539 "still has %d dirty pages\n", inode
, capsnap
,
540 capsnap
->context
, capsnap
->context
->seq
,
541 ceph_cap_string(capsnap
->dirty
), capsnap
->size
,
542 capsnap
->dirty_pages
);
545 dout("finish_cap_snap %p cap_snap %p snapc %p %llu %s s=%llu\n",
546 inode
, capsnap
, capsnap
->context
,
547 capsnap
->context
->seq
, ceph_cap_string(capsnap
->dirty
),
550 spin_lock(&mdsc
->snap_flush_lock
);
551 list_add_tail(&ci
->i_snap_flush_item
, &mdsc
->snap_flush_list
);
552 spin_unlock(&mdsc
->snap_flush_lock
);
553 return 1; /* caller may want to ceph_flush_snaps */
557 * Queue cap_snaps for snap writeback for this realm and its children.
558 * Called under snap_rwsem, so realm topology won't change.
560 static void queue_realm_cap_snaps(struct ceph_snap_realm
*realm
)
562 struct ceph_inode_info
*ci
;
563 struct inode
*lastinode
= NULL
;
564 struct ceph_snap_realm
*child
;
566 dout("queue_realm_cap_snaps %p %llx inodes\n", realm
, realm
->ino
);
568 spin_lock(&realm
->inodes_with_caps_lock
);
569 list_for_each_entry(ci
, &realm
->inodes_with_caps
,
571 struct inode
*inode
= igrab(&ci
->vfs_inode
);
574 spin_unlock(&realm
->inodes_with_caps_lock
);
578 ceph_queue_cap_snap(ci
);
579 spin_lock(&realm
->inodes_with_caps_lock
);
581 spin_unlock(&realm
->inodes_with_caps_lock
);
585 dout("queue_realm_cap_snaps %p %llx children\n", realm
, realm
->ino
);
586 list_for_each_entry(child
, &realm
->children
, child_item
)
587 queue_realm_cap_snaps(child
);
589 dout("queue_realm_cap_snaps %p %llx done\n", realm
, realm
->ino
);
593 * Parse and apply a snapblob "snap trace" from the MDS. This specifies
594 * the snap realm parameters from a given realm and all of its ancestors,
597 * Caller must hold snap_rwsem for write.
599 int ceph_update_snap_trace(struct ceph_mds_client
*mdsc
,
600 void *p
, void *e
, bool deletion
)
602 struct ceph_mds_snap_realm
*ri
; /* encoded */
603 __le64
*snaps
; /* encoded */
604 __le64
*prior_parent_snaps
; /* encoded */
605 struct ceph_snap_realm
*realm
;
608 LIST_HEAD(dirty_realms
);
610 dout("update_snap_trace deletion=%d\n", deletion
);
612 ceph_decode_need(&p
, e
, sizeof(*ri
), bad
);
615 ceph_decode_need(&p
, e
, sizeof(u64
)*(le32_to_cpu(ri
->num_snaps
) +
616 le32_to_cpu(ri
->num_prior_parent_snaps
)), bad
);
618 p
+= sizeof(u64
) * le32_to_cpu(ri
->num_snaps
);
619 prior_parent_snaps
= p
;
620 p
+= sizeof(u64
) * le32_to_cpu(ri
->num_prior_parent_snaps
);
622 realm
= ceph_lookup_snap_realm(mdsc
, le64_to_cpu(ri
->ino
));
624 realm
= ceph_create_snap_realm(mdsc
, le64_to_cpu(ri
->ino
));
626 err
= PTR_ERR(realm
);
631 /* ensure the parent is correct */
632 err
= adjust_snap_realm_parent(mdsc
, realm
, le64_to_cpu(ri
->parent
));
637 if (le64_to_cpu(ri
->seq
) > realm
->seq
) {
638 dout("update_snap_trace updating %llx %p %lld -> %lld\n",
639 realm
->ino
, realm
, realm
->seq
, le64_to_cpu(ri
->seq
));
640 /* update realm parameters, snap lists */
641 realm
->seq
= le64_to_cpu(ri
->seq
);
642 realm
->created
= le64_to_cpu(ri
->created
);
643 realm
->parent_since
= le64_to_cpu(ri
->parent_since
);
645 realm
->num_snaps
= le32_to_cpu(ri
->num_snaps
);
646 err
= dup_array(&realm
->snaps
, snaps
, realm
->num_snaps
);
650 realm
->num_prior_parent_snaps
=
651 le32_to_cpu(ri
->num_prior_parent_snaps
);
652 err
= dup_array(&realm
->prior_parent_snaps
, prior_parent_snaps
,
653 realm
->num_prior_parent_snaps
);
657 /* queue realm for cap_snap creation */
658 list_add(&realm
->dirty_item
, &dirty_realms
);
661 } else if (!realm
->cached_context
) {
662 dout("update_snap_trace %llx %p seq %lld new\n",
663 realm
->ino
, realm
, realm
->seq
);
666 dout("update_snap_trace %llx %p seq %lld unchanged\n",
667 realm
->ino
, realm
, realm
->seq
);
670 dout("done with %llx %p, invalidated=%d, %p %p\n", realm
->ino
,
671 realm
, invalidate
, p
, e
);
676 /* invalidate when we reach the _end_ (root) of the trace */
678 rebuild_snap_realms(realm
);
681 * queue cap snaps _after_ we've built the new snap contexts,
682 * so that i_head_snapc can be set appropriately.
684 list_for_each_entry(realm
, &dirty_realms
, dirty_item
) {
685 queue_realm_cap_snaps(realm
);
688 __cleanup_empty_realms(mdsc
);
694 pr_err("update_snap_trace error %d\n", err
);
700 * Send any cap_snaps that are queued for flush. Try to carry
701 * s_mutex across multiple snap flushes to avoid locking overhead.
703 * Caller holds no locks.
705 static void flush_snaps(struct ceph_mds_client
*mdsc
)
707 struct ceph_inode_info
*ci
;
709 struct ceph_mds_session
*session
= NULL
;
711 dout("flush_snaps\n");
712 spin_lock(&mdsc
->snap_flush_lock
);
713 while (!list_empty(&mdsc
->snap_flush_list
)) {
714 ci
= list_first_entry(&mdsc
->snap_flush_list
,
715 struct ceph_inode_info
, i_snap_flush_item
);
716 inode
= &ci
->vfs_inode
;
718 spin_unlock(&mdsc
->snap_flush_lock
);
719 spin_lock(&inode
->i_lock
);
720 __ceph_flush_snaps(ci
, &session
, 0);
721 spin_unlock(&inode
->i_lock
);
723 spin_lock(&mdsc
->snap_flush_lock
);
725 spin_unlock(&mdsc
->snap_flush_lock
);
728 mutex_unlock(&session
->s_mutex
);
729 ceph_put_mds_session(session
);
731 dout("flush_snaps done\n");
736 * Handle a snap notification from the MDS.
738 * This can take two basic forms: the simplest is just a snap creation
739 * or deletion notification on an existing realm. This should update the
740 * realm and its children.
742 * The more difficult case is realm creation, due to snap creation at a
743 * new point in the file hierarchy, or due to a rename that moves a file or
744 * directory into another realm.
746 void ceph_handle_snap(struct ceph_mds_client
*mdsc
,
747 struct ceph_mds_session
*session
,
748 struct ceph_msg
*msg
)
750 struct super_block
*sb
= mdsc
->client
->sb
;
751 int mds
= session
->s_mds
;
755 struct ceph_snap_realm
*realm
= NULL
;
756 void *p
= msg
->front
.iov_base
;
757 void *e
= p
+ msg
->front
.iov_len
;
758 struct ceph_mds_snap_head
*h
;
759 int num_split_inos
, num_split_realms
;
760 __le64
*split_inos
= NULL
, *split_realms
= NULL
;
762 int locked_rwsem
= 0;
765 if (msg
->front
.iov_len
< sizeof(*h
))
768 op
= le32_to_cpu(h
->op
);
769 split
= le64_to_cpu(h
->split
); /* non-zero if we are splitting an
771 num_split_inos
= le32_to_cpu(h
->num_split_inos
);
772 num_split_realms
= le32_to_cpu(h
->num_split_realms
);
773 trace_len
= le32_to_cpu(h
->trace_len
);
776 dout("handle_snap from mds%d op %s split %llx tracelen %d\n", mds
,
777 ceph_snap_op_name(op
), split
, trace_len
);
779 mutex_lock(&session
->s_mutex
);
781 mutex_unlock(&session
->s_mutex
);
783 down_write(&mdsc
->snap_rwsem
);
786 if (op
== CEPH_SNAP_OP_SPLIT
) {
787 struct ceph_mds_snap_realm
*ri
;
790 * A "split" breaks part of an existing realm off into
791 * a new realm. The MDS provides a list of inodes
792 * (with caps) and child realms that belong to the new
796 p
+= sizeof(u64
) * num_split_inos
;
798 p
+= sizeof(u64
) * num_split_realms
;
799 ceph_decode_need(&p
, e
, sizeof(*ri
), bad
);
800 /* we will peek at realm info here, but will _not_
801 * advance p, as the realm update will occur below in
802 * ceph_update_snap_trace. */
805 realm
= ceph_lookup_snap_realm(mdsc
, split
);
807 realm
= ceph_create_snap_realm(mdsc
, split
);
811 ceph_get_snap_realm(mdsc
, realm
);
813 dout("splitting snap_realm %llx %p\n", realm
->ino
, realm
);
814 for (i
= 0; i
< num_split_inos
; i
++) {
815 struct ceph_vino vino
= {
816 .ino
= le64_to_cpu(split_inos
[i
]),
819 struct inode
*inode
= ceph_find_inode(sb
, vino
);
820 struct ceph_inode_info
*ci
;
821 struct ceph_snap_realm
*oldrealm
;
825 ci
= ceph_inode(inode
);
827 spin_lock(&inode
->i_lock
);
828 if (!ci
->i_snap_realm
)
831 * If this inode belongs to a realm that was
832 * created after our new realm, we experienced
833 * a race (due to another split notifications
834 * arriving from a different MDS). So skip
837 if (ci
->i_snap_realm
->created
>
838 le64_to_cpu(ri
->created
)) {
839 dout(" leaving %p in newer realm %llx %p\n",
840 inode
, ci
->i_snap_realm
->ino
,
844 dout(" will move %p to split realm %llx %p\n",
845 inode
, realm
->ino
, realm
);
847 * Move the inode to the new realm
849 spin_lock(&realm
->inodes_with_caps_lock
);
850 list_del_init(&ci
->i_snap_realm_item
);
851 list_add(&ci
->i_snap_realm_item
,
852 &realm
->inodes_with_caps
);
853 oldrealm
= ci
->i_snap_realm
;
854 ci
->i_snap_realm
= realm
;
855 spin_unlock(&realm
->inodes_with_caps_lock
);
856 spin_unlock(&inode
->i_lock
);
858 ceph_get_snap_realm(mdsc
, realm
);
859 ceph_put_snap_realm(mdsc
, oldrealm
);
865 spin_unlock(&inode
->i_lock
);
869 /* we may have taken some of the old realm's children. */
870 for (i
= 0; i
< num_split_realms
; i
++) {
871 struct ceph_snap_realm
*child
=
872 ceph_lookup_snap_realm(mdsc
,
873 le64_to_cpu(split_realms
[i
]));
876 adjust_snap_realm_parent(mdsc
, child
, realm
->ino
);
881 * update using the provided snap trace. if we are deleting a
882 * snap, we can avoid queueing cap_snaps.
884 ceph_update_snap_trace(mdsc
, p
, e
,
885 op
== CEPH_SNAP_OP_DESTROY
);
887 if (op
== CEPH_SNAP_OP_SPLIT
)
888 /* we took a reference when we created the realm, above */
889 ceph_put_snap_realm(mdsc
, realm
);
891 __cleanup_empty_realms(mdsc
);
893 up_write(&mdsc
->snap_rwsem
);
899 pr_err("corrupt snap message from mds%d\n", mds
);
903 up_write(&mdsc
->snap_rwsem
);
This page took 0.048069 seconds and 5 git commands to generate.