ARM: 7409/1: Do not call flush_cache_user_range with mmap_sem held
[linux/fpc-iii.git] / fs / ceph / snap.c
blob54b14de2e729114d50d91fde0f9c5ab5bf7d29d7
1 #include <linux/ceph/ceph_debug.h>
3 #include <linux/sort.h>
4 #include <linux/slab.h>
6 #include "super.h"
7 #include "mds_client.h"
9 #include <linux/ceph/decode.h>
12 * Snapshots in ceph are driven in large part by cooperation from the
13 * client. In contrast to local file systems or file servers that
14 * implement snapshots at a single point in the system, ceph's
15 * distributed access to storage requires clients to help decide
16 * whether a write logically occurs before or after a recently created
17 * snapshot.
19 * This provides a perfect instantanous client-wide snapshot. Between
20 * clients, however, snapshots may appear to be applied at slightly
21 * different points in time, depending on delays in delivering the
22 * snapshot notification.
24 * Snapshots are _not_ file system-wide. Instead, each snapshot
25 * applies to the subdirectory nested beneath some directory. This
26 * effectively divides the hierarchy into multiple "realms," where all
27 * of the files contained by each realm share the same set of
28 * snapshots. An individual realm's snap set contains snapshots
29 * explicitly created on that realm, as well as any snaps in its
30 * parent's snap set _after_ the point at which the parent became it's
31 * parent (due to, say, a rename). Similarly, snaps from prior parents
32 * during the time intervals during which they were the parent are included.
34 * The client is spared most of this detail, fortunately... it must only
35 * maintains a hierarchy of realms reflecting the current parent/child
36 * realm relationship, and for each realm has an explicit list of snaps
37 * inherited from prior parents.
39 * A snap_realm struct is maintained for realms containing every inode
40 * with an open cap in the system. (The needed snap realm information is
41 * provided by the MDS whenever a cap is issued, i.e., on open.) A 'seq'
42 * version number is used to ensure that as realm parameters change (new
43 * snapshot, new parent, etc.) the client's realm hierarchy is updated.
45 * The realm hierarchy drives the generation of a 'snap context' for each
46 * realm, which simply lists the resulting set of snaps for the realm. This
47 * is attached to any writes sent to OSDs.
50 * Unfortunately error handling is a bit mixed here. If we get a snap
51 * update, but don't have enough memory to update our realm hierarchy,
52 * it's not clear what we can do about it (besides complaining to the
53 * console).
58 * increase ref count for the realm
60 * caller must hold snap_rwsem for write.
62 void ceph_get_snap_realm(struct ceph_mds_client *mdsc,
63 struct ceph_snap_realm *realm)
65 dout("get_realm %p %d -> %d\n", realm,
66 atomic_read(&realm->nref), atomic_read(&realm->nref)+1);
68 * since we _only_ increment realm refs or empty the empty
69 * list with snap_rwsem held, adjusting the empty list here is
70 * safe. we do need to protect against concurrent empty list
71 * additions, however.
73 if (atomic_read(&realm->nref) == 0) {
74 spin_lock(&mdsc->snap_empty_lock);
75 list_del_init(&realm->empty_item);
76 spin_unlock(&mdsc->snap_empty_lock);
79 atomic_inc(&realm->nref);
82 static void __insert_snap_realm(struct rb_root *root,
83 struct ceph_snap_realm *new)
85 struct rb_node **p = &root->rb_node;
86 struct rb_node *parent = NULL;
87 struct ceph_snap_realm *r = NULL;
89 while (*p) {
90 parent = *p;
91 r = rb_entry(parent, struct ceph_snap_realm, node);
92 if (new->ino < r->ino)
93 p = &(*p)->rb_left;
94 else if (new->ino > r->ino)
95 p = &(*p)->rb_right;
96 else
97 BUG();
100 rb_link_node(&new->node, parent, p);
101 rb_insert_color(&new->node, root);
105 * create and get the realm rooted at @ino and bump its ref count.
107 * caller must hold snap_rwsem for write.
109 static struct ceph_snap_realm *ceph_create_snap_realm(
110 struct ceph_mds_client *mdsc,
111 u64 ino)
113 struct ceph_snap_realm *realm;
115 realm = kzalloc(sizeof(*realm), GFP_NOFS);
116 if (!realm)
117 return ERR_PTR(-ENOMEM);
119 atomic_set(&realm->nref, 0); /* tree does not take a ref */
120 realm->ino = ino;
121 INIT_LIST_HEAD(&realm->children);
122 INIT_LIST_HEAD(&realm->child_item);
123 INIT_LIST_HEAD(&realm->empty_item);
124 INIT_LIST_HEAD(&realm->dirty_item);
125 INIT_LIST_HEAD(&realm->inodes_with_caps);
126 spin_lock_init(&realm->inodes_with_caps_lock);
127 __insert_snap_realm(&mdsc->snap_realms, realm);
128 dout("create_snap_realm %llx %p\n", realm->ino, realm);
129 return realm;
133 * lookup the realm rooted at @ino.
135 * caller must hold snap_rwsem for write.
137 struct ceph_snap_realm *ceph_lookup_snap_realm(struct ceph_mds_client *mdsc,
138 u64 ino)
140 struct rb_node *n = mdsc->snap_realms.rb_node;
141 struct ceph_snap_realm *r;
143 while (n) {
144 r = rb_entry(n, struct ceph_snap_realm, node);
145 if (ino < r->ino)
146 n = n->rb_left;
147 else if (ino > r->ino)
148 n = n->rb_right;
149 else {
150 dout("lookup_snap_realm %llx %p\n", r->ino, r);
151 return r;
154 return NULL;
157 static void __put_snap_realm(struct ceph_mds_client *mdsc,
158 struct ceph_snap_realm *realm);
161 * called with snap_rwsem (write)
163 static void __destroy_snap_realm(struct ceph_mds_client *mdsc,
164 struct ceph_snap_realm *realm)
166 dout("__destroy_snap_realm %p %llx\n", realm, realm->ino);
168 rb_erase(&realm->node, &mdsc->snap_realms);
170 if (realm->parent) {
171 list_del_init(&realm->child_item);
172 __put_snap_realm(mdsc, realm->parent);
175 kfree(realm->prior_parent_snaps);
176 kfree(realm->snaps);
177 ceph_put_snap_context(realm->cached_context);
178 kfree(realm);
182 * caller holds snap_rwsem (write)
184 static void __put_snap_realm(struct ceph_mds_client *mdsc,
185 struct ceph_snap_realm *realm)
187 dout("__put_snap_realm %llx %p %d -> %d\n", realm->ino, realm,
188 atomic_read(&realm->nref), atomic_read(&realm->nref)-1);
189 if (atomic_dec_and_test(&realm->nref))
190 __destroy_snap_realm(mdsc, realm);
194 * caller needn't hold any locks
196 void ceph_put_snap_realm(struct ceph_mds_client *mdsc,
197 struct ceph_snap_realm *realm)
199 dout("put_snap_realm %llx %p %d -> %d\n", realm->ino, realm,
200 atomic_read(&realm->nref), atomic_read(&realm->nref)-1);
201 if (!atomic_dec_and_test(&realm->nref))
202 return;
204 if (down_write_trylock(&mdsc->snap_rwsem)) {
205 __destroy_snap_realm(mdsc, realm);
206 up_write(&mdsc->snap_rwsem);
207 } else {
208 spin_lock(&mdsc->snap_empty_lock);
209 list_add(&realm->empty_item, &mdsc->snap_empty);
210 spin_unlock(&mdsc->snap_empty_lock);
215 * Clean up any realms whose ref counts have dropped to zero. Note
216 * that this does not include realms who were created but not yet
217 * used.
219 * Called under snap_rwsem (write)
221 static void __cleanup_empty_realms(struct ceph_mds_client *mdsc)
223 struct ceph_snap_realm *realm;
225 spin_lock(&mdsc->snap_empty_lock);
226 while (!list_empty(&mdsc->snap_empty)) {
227 realm = list_first_entry(&mdsc->snap_empty,
228 struct ceph_snap_realm, empty_item);
229 list_del(&realm->empty_item);
230 spin_unlock(&mdsc->snap_empty_lock);
231 __destroy_snap_realm(mdsc, realm);
232 spin_lock(&mdsc->snap_empty_lock);
234 spin_unlock(&mdsc->snap_empty_lock);
237 void ceph_cleanup_empty_realms(struct ceph_mds_client *mdsc)
239 down_write(&mdsc->snap_rwsem);
240 __cleanup_empty_realms(mdsc);
241 up_write(&mdsc->snap_rwsem);
245 * adjust the parent realm of a given @realm. adjust child list, and parent
246 * pointers, and ref counts appropriately.
248 * return true if parent was changed, 0 if unchanged, <0 on error.
250 * caller must hold snap_rwsem for write.
252 static int adjust_snap_realm_parent(struct ceph_mds_client *mdsc,
253 struct ceph_snap_realm *realm,
254 u64 parentino)
256 struct ceph_snap_realm *parent;
258 if (realm->parent_ino == parentino)
259 return 0;
261 parent = ceph_lookup_snap_realm(mdsc, parentino);
262 if (!parent) {
263 parent = ceph_create_snap_realm(mdsc, parentino);
264 if (IS_ERR(parent))
265 return PTR_ERR(parent);
267 dout("adjust_snap_realm_parent %llx %p: %llx %p -> %llx %p\n",
268 realm->ino, realm, realm->parent_ino, realm->parent,
269 parentino, parent);
270 if (realm->parent) {
271 list_del_init(&realm->child_item);
272 ceph_put_snap_realm(mdsc, realm->parent);
274 realm->parent_ino = parentino;
275 realm->parent = parent;
276 ceph_get_snap_realm(mdsc, parent);
277 list_add(&realm->child_item, &parent->children);
278 return 1;
282 static int cmpu64_rev(const void *a, const void *b)
284 if (*(u64 *)a < *(u64 *)b)
285 return 1;
286 if (*(u64 *)a > *(u64 *)b)
287 return -1;
288 return 0;
292 * build the snap context for a given realm.
294 static int build_snap_context(struct ceph_snap_realm *realm)
296 struct ceph_snap_realm *parent = realm->parent;
297 struct ceph_snap_context *snapc;
298 int err = 0;
299 int i;
300 int num = realm->num_prior_parent_snaps + realm->num_snaps;
303 * build parent context, if it hasn't been built.
304 * conservatively estimate that all parent snaps might be
305 * included by us.
307 if (parent) {
308 if (!parent->cached_context) {
309 err = build_snap_context(parent);
310 if (err)
311 goto fail;
313 num += parent->cached_context->num_snaps;
316 /* do i actually need to update? not if my context seq
317 matches realm seq, and my parents' does to. (this works
318 because we rebuild_snap_realms() works _downward_ in
319 hierarchy after each update.) */
320 if (realm->cached_context &&
321 realm->cached_context->seq == realm->seq &&
322 (!parent ||
323 realm->cached_context->seq >= parent->cached_context->seq)) {
324 dout("build_snap_context %llx %p: %p seq %lld (%d snaps)"
325 " (unchanged)\n",
326 realm->ino, realm, realm->cached_context,
327 realm->cached_context->seq,
328 realm->cached_context->num_snaps);
329 return 0;
332 /* alloc new snap context */
333 err = -ENOMEM;
334 if (num > ULONG_MAX / sizeof(u64) - sizeof(*snapc))
335 goto fail;
336 snapc = kzalloc(sizeof(*snapc) + num*sizeof(u64), GFP_NOFS);
337 if (!snapc)
338 goto fail;
339 atomic_set(&snapc->nref, 1);
341 /* build (reverse sorted) snap vector */
342 num = 0;
343 snapc->seq = realm->seq;
344 if (parent) {
345 /* include any of parent's snaps occurring _after_ my
346 parent became my parent */
347 for (i = 0; i < parent->cached_context->num_snaps; i++)
348 if (parent->cached_context->snaps[i] >=
349 realm->parent_since)
350 snapc->snaps[num++] =
351 parent->cached_context->snaps[i];
352 if (parent->cached_context->seq > snapc->seq)
353 snapc->seq = parent->cached_context->seq;
355 memcpy(snapc->snaps + num, realm->snaps,
356 sizeof(u64)*realm->num_snaps);
357 num += realm->num_snaps;
358 memcpy(snapc->snaps + num, realm->prior_parent_snaps,
359 sizeof(u64)*realm->num_prior_parent_snaps);
360 num += realm->num_prior_parent_snaps;
362 sort(snapc->snaps, num, sizeof(u64), cmpu64_rev, NULL);
363 snapc->num_snaps = num;
364 dout("build_snap_context %llx %p: %p seq %lld (%d snaps)\n",
365 realm->ino, realm, snapc, snapc->seq, snapc->num_snaps);
367 if (realm->cached_context)
368 ceph_put_snap_context(realm->cached_context);
369 realm->cached_context = snapc;
370 return 0;
372 fail:
374 * if we fail, clear old (incorrect) cached_context... hopefully
375 * we'll have better luck building it later
377 if (realm->cached_context) {
378 ceph_put_snap_context(realm->cached_context);
379 realm->cached_context = NULL;
381 pr_err("build_snap_context %llx %p fail %d\n", realm->ino,
382 realm, err);
383 return err;
387 * rebuild snap context for the given realm and all of its children.
389 static void rebuild_snap_realms(struct ceph_snap_realm *realm)
391 struct ceph_snap_realm *child;
393 dout("rebuild_snap_realms %llx %p\n", realm->ino, realm);
394 build_snap_context(realm);
396 list_for_each_entry(child, &realm->children, child_item)
397 rebuild_snap_realms(child);
402 * helper to allocate and decode an array of snapids. free prior
403 * instance, if any.
405 static int dup_array(u64 **dst, __le64 *src, int num)
407 int i;
409 kfree(*dst);
410 if (num) {
411 *dst = kcalloc(num, sizeof(u64), GFP_NOFS);
412 if (!*dst)
413 return -ENOMEM;
414 for (i = 0; i < num; i++)
415 (*dst)[i] = get_unaligned_le64(src + i);
416 } else {
417 *dst = NULL;
419 return 0;
424 * When a snapshot is applied, the size/mtime inode metadata is queued
425 * in a ceph_cap_snap (one for each snapshot) until writeback
426 * completes and the metadata can be flushed back to the MDS.
428 * However, if a (sync) write is currently in-progress when we apply
429 * the snapshot, we have to wait until the write succeeds or fails
430 * (and a final size/mtime is known). In this case the
431 * cap_snap->writing = 1, and is said to be "pending." When the write
432 * finishes, we __ceph_finish_cap_snap().
434 * Caller must hold snap_rwsem for read (i.e., the realm topology won't
435 * change).
437 void ceph_queue_cap_snap(struct ceph_inode_info *ci)
439 struct inode *inode = &ci->vfs_inode;
440 struct ceph_cap_snap *capsnap;
441 int used, dirty;
443 capsnap = kzalloc(sizeof(*capsnap), GFP_NOFS);
444 if (!capsnap) {
445 pr_err("ENOMEM allocating ceph_cap_snap on %p\n", inode);
446 return;
449 spin_lock(&inode->i_lock);
450 used = __ceph_caps_used(ci);
451 dirty = __ceph_caps_dirty(ci);
452 if (__ceph_have_pending_cap_snap(ci)) {
453 /* there is no point in queuing multiple "pending" cap_snaps,
454 as no new writes are allowed to start when pending, so any
455 writes in progress now were started before the previous
456 cap_snap. lucky us. */
457 dout("queue_cap_snap %p already pending\n", inode);
458 kfree(capsnap);
459 } else if (ci->i_wrbuffer_ref_head || (used & CEPH_CAP_FILE_WR) ||
460 (dirty & (CEPH_CAP_AUTH_EXCL|CEPH_CAP_XATTR_EXCL|
461 CEPH_CAP_FILE_EXCL|CEPH_CAP_FILE_WR))) {
462 struct ceph_snap_context *snapc = ci->i_head_snapc;
464 dout("queue_cap_snap %p cap_snap %p queuing under %p\n", inode,
465 capsnap, snapc);
466 ihold(inode);
468 atomic_set(&capsnap->nref, 1);
469 capsnap->ci = ci;
470 INIT_LIST_HEAD(&capsnap->ci_item);
471 INIT_LIST_HEAD(&capsnap->flushing_item);
473 capsnap->follows = snapc->seq;
474 capsnap->issued = __ceph_caps_issued(ci, NULL);
475 capsnap->dirty = dirty;
477 capsnap->mode = inode->i_mode;
478 capsnap->uid = inode->i_uid;
479 capsnap->gid = inode->i_gid;
481 if (dirty & CEPH_CAP_XATTR_EXCL) {
482 __ceph_build_xattrs_blob(ci);
483 capsnap->xattr_blob =
484 ceph_buffer_get(ci->i_xattrs.blob);
485 capsnap->xattr_version = ci->i_xattrs.version;
486 } else {
487 capsnap->xattr_blob = NULL;
488 capsnap->xattr_version = 0;
491 /* dirty page count moved from _head to this cap_snap;
492 all subsequent writes page dirties occur _after_ this
493 snapshot. */
494 capsnap->dirty_pages = ci->i_wrbuffer_ref_head;
495 ci->i_wrbuffer_ref_head = 0;
496 capsnap->context = snapc;
497 ci->i_head_snapc =
498 ceph_get_snap_context(ci->i_snap_realm->cached_context);
499 dout(" new snapc is %p\n", ci->i_head_snapc);
500 list_add_tail(&capsnap->ci_item, &ci->i_cap_snaps);
502 if (used & CEPH_CAP_FILE_WR) {
503 dout("queue_cap_snap %p cap_snap %p snapc %p"
504 " seq %llu used WR, now pending\n", inode,
505 capsnap, snapc, snapc->seq);
506 capsnap->writing = 1;
507 } else {
508 /* note mtime, size NOW. */
509 __ceph_finish_cap_snap(ci, capsnap);
511 } else {
512 dout("queue_cap_snap %p nothing dirty|writing\n", inode);
513 kfree(capsnap);
516 spin_unlock(&inode->i_lock);
520 * Finalize the size, mtime for a cap_snap.. that is, settle on final values
521 * to be used for the snapshot, to be flushed back to the mds.
523 * If capsnap can now be flushed, add to snap_flush list, and return 1.
525 * Caller must hold i_lock.
527 int __ceph_finish_cap_snap(struct ceph_inode_info *ci,
528 struct ceph_cap_snap *capsnap)
530 struct inode *inode = &ci->vfs_inode;
531 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
533 BUG_ON(capsnap->writing);
534 capsnap->size = inode->i_size;
535 capsnap->mtime = inode->i_mtime;
536 capsnap->atime = inode->i_atime;
537 capsnap->ctime = inode->i_ctime;
538 capsnap->time_warp_seq = ci->i_time_warp_seq;
539 if (capsnap->dirty_pages) {
540 dout("finish_cap_snap %p cap_snap %p snapc %p %llu %s s=%llu "
541 "still has %d dirty pages\n", inode, capsnap,
542 capsnap->context, capsnap->context->seq,
543 ceph_cap_string(capsnap->dirty), capsnap->size,
544 capsnap->dirty_pages);
545 return 0;
547 dout("finish_cap_snap %p cap_snap %p snapc %p %llu %s s=%llu\n",
548 inode, capsnap, capsnap->context,
549 capsnap->context->seq, ceph_cap_string(capsnap->dirty),
550 capsnap->size);
552 spin_lock(&mdsc->snap_flush_lock);
553 list_add_tail(&ci->i_snap_flush_item, &mdsc->snap_flush_list);
554 spin_unlock(&mdsc->snap_flush_lock);
555 return 1; /* caller may want to ceph_flush_snaps */
559 * Queue cap_snaps for snap writeback for this realm and its children.
560 * Called under snap_rwsem, so realm topology won't change.
562 static void queue_realm_cap_snaps(struct ceph_snap_realm *realm)
564 struct ceph_inode_info *ci;
565 struct inode *lastinode = NULL;
566 struct ceph_snap_realm *child;
568 dout("queue_realm_cap_snaps %p %llx inodes\n", realm, realm->ino);
570 spin_lock(&realm->inodes_with_caps_lock);
571 list_for_each_entry(ci, &realm->inodes_with_caps,
572 i_snap_realm_item) {
573 struct inode *inode = igrab(&ci->vfs_inode);
574 if (!inode)
575 continue;
576 spin_unlock(&realm->inodes_with_caps_lock);
577 if (lastinode)
578 iput(lastinode);
579 lastinode = inode;
580 ceph_queue_cap_snap(ci);
581 spin_lock(&realm->inodes_with_caps_lock);
583 spin_unlock(&realm->inodes_with_caps_lock);
584 if (lastinode)
585 iput(lastinode);
587 list_for_each_entry(child, &realm->children, child_item) {
588 dout("queue_realm_cap_snaps %p %llx queue child %p %llx\n",
589 realm, realm->ino, child, child->ino);
590 list_del_init(&child->dirty_item);
591 list_add(&child->dirty_item, &realm->dirty_item);
594 list_del_init(&realm->dirty_item);
595 dout("queue_realm_cap_snaps %p %llx done\n", realm, realm->ino);
599 * Parse and apply a snapblob "snap trace" from the MDS. This specifies
600 * the snap realm parameters from a given realm and all of its ancestors,
601 * up to the root.
603 * Caller must hold snap_rwsem for write.
605 int ceph_update_snap_trace(struct ceph_mds_client *mdsc,
606 void *p, void *e, bool deletion)
608 struct ceph_mds_snap_realm *ri; /* encoded */
609 __le64 *snaps; /* encoded */
610 __le64 *prior_parent_snaps; /* encoded */
611 struct ceph_snap_realm *realm;
612 int invalidate = 0;
613 int err = -ENOMEM;
614 LIST_HEAD(dirty_realms);
616 dout("update_snap_trace deletion=%d\n", deletion);
617 more:
618 ceph_decode_need(&p, e, sizeof(*ri), bad);
619 ri = p;
620 p += sizeof(*ri);
621 ceph_decode_need(&p, e, sizeof(u64)*(le32_to_cpu(ri->num_snaps) +
622 le32_to_cpu(ri->num_prior_parent_snaps)), bad);
623 snaps = p;
624 p += sizeof(u64) * le32_to_cpu(ri->num_snaps);
625 prior_parent_snaps = p;
626 p += sizeof(u64) * le32_to_cpu(ri->num_prior_parent_snaps);
628 realm = ceph_lookup_snap_realm(mdsc, le64_to_cpu(ri->ino));
629 if (!realm) {
630 realm = ceph_create_snap_realm(mdsc, le64_to_cpu(ri->ino));
631 if (IS_ERR(realm)) {
632 err = PTR_ERR(realm);
633 goto fail;
637 /* ensure the parent is correct */
638 err = adjust_snap_realm_parent(mdsc, realm, le64_to_cpu(ri->parent));
639 if (err < 0)
640 goto fail;
641 invalidate += err;
643 if (le64_to_cpu(ri->seq) > realm->seq) {
644 dout("update_snap_trace updating %llx %p %lld -> %lld\n",
645 realm->ino, realm, realm->seq, le64_to_cpu(ri->seq));
646 /* update realm parameters, snap lists */
647 realm->seq = le64_to_cpu(ri->seq);
648 realm->created = le64_to_cpu(ri->created);
649 realm->parent_since = le64_to_cpu(ri->parent_since);
651 realm->num_snaps = le32_to_cpu(ri->num_snaps);
652 err = dup_array(&realm->snaps, snaps, realm->num_snaps);
653 if (err < 0)
654 goto fail;
656 realm->num_prior_parent_snaps =
657 le32_to_cpu(ri->num_prior_parent_snaps);
658 err = dup_array(&realm->prior_parent_snaps, prior_parent_snaps,
659 realm->num_prior_parent_snaps);
660 if (err < 0)
661 goto fail;
663 /* queue realm for cap_snap creation */
664 list_add(&realm->dirty_item, &dirty_realms);
666 invalidate = 1;
667 } else if (!realm->cached_context) {
668 dout("update_snap_trace %llx %p seq %lld new\n",
669 realm->ino, realm, realm->seq);
670 invalidate = 1;
671 } else {
672 dout("update_snap_trace %llx %p seq %lld unchanged\n",
673 realm->ino, realm, realm->seq);
676 dout("done with %llx %p, invalidated=%d, %p %p\n", realm->ino,
677 realm, invalidate, p, e);
679 if (p < e)
680 goto more;
682 /* invalidate when we reach the _end_ (root) of the trace */
683 if (invalidate)
684 rebuild_snap_realms(realm);
687 * queue cap snaps _after_ we've built the new snap contexts,
688 * so that i_head_snapc can be set appropriately.
690 while (!list_empty(&dirty_realms)) {
691 realm = list_first_entry(&dirty_realms, struct ceph_snap_realm,
692 dirty_item);
693 queue_realm_cap_snaps(realm);
696 __cleanup_empty_realms(mdsc);
697 return 0;
699 bad:
700 err = -EINVAL;
701 fail:
702 pr_err("update_snap_trace error %d\n", err);
703 return err;
708 * Send any cap_snaps that are queued for flush. Try to carry
709 * s_mutex across multiple snap flushes to avoid locking overhead.
711 * Caller holds no locks.
713 static void flush_snaps(struct ceph_mds_client *mdsc)
715 struct ceph_inode_info *ci;
716 struct inode *inode;
717 struct ceph_mds_session *session = NULL;
719 dout("flush_snaps\n");
720 spin_lock(&mdsc->snap_flush_lock);
721 while (!list_empty(&mdsc->snap_flush_list)) {
722 ci = list_first_entry(&mdsc->snap_flush_list,
723 struct ceph_inode_info, i_snap_flush_item);
724 inode = &ci->vfs_inode;
725 ihold(inode);
726 spin_unlock(&mdsc->snap_flush_lock);
727 spin_lock(&inode->i_lock);
728 __ceph_flush_snaps(ci, &session, 0);
729 spin_unlock(&inode->i_lock);
730 iput(inode);
731 spin_lock(&mdsc->snap_flush_lock);
733 spin_unlock(&mdsc->snap_flush_lock);
735 if (session) {
736 mutex_unlock(&session->s_mutex);
737 ceph_put_mds_session(session);
739 dout("flush_snaps done\n");
744 * Handle a snap notification from the MDS.
746 * This can take two basic forms: the simplest is just a snap creation
747 * or deletion notification on an existing realm. This should update the
748 * realm and its children.
750 * The more difficult case is realm creation, due to snap creation at a
751 * new point in the file hierarchy, or due to a rename that moves a file or
752 * directory into another realm.
754 void ceph_handle_snap(struct ceph_mds_client *mdsc,
755 struct ceph_mds_session *session,
756 struct ceph_msg *msg)
758 struct super_block *sb = mdsc->fsc->sb;
759 int mds = session->s_mds;
760 u64 split;
761 int op;
762 int trace_len;
763 struct ceph_snap_realm *realm = NULL;
764 void *p = msg->front.iov_base;
765 void *e = p + msg->front.iov_len;
766 struct ceph_mds_snap_head *h;
767 int num_split_inos, num_split_realms;
768 __le64 *split_inos = NULL, *split_realms = NULL;
769 int i;
770 int locked_rwsem = 0;
772 /* decode */
773 if (msg->front.iov_len < sizeof(*h))
774 goto bad;
775 h = p;
776 op = le32_to_cpu(h->op);
777 split = le64_to_cpu(h->split); /* non-zero if we are splitting an
778 * existing realm */
779 num_split_inos = le32_to_cpu(h->num_split_inos);
780 num_split_realms = le32_to_cpu(h->num_split_realms);
781 trace_len = le32_to_cpu(h->trace_len);
782 p += sizeof(*h);
784 dout("handle_snap from mds%d op %s split %llx tracelen %d\n", mds,
785 ceph_snap_op_name(op), split, trace_len);
787 mutex_lock(&session->s_mutex);
788 session->s_seq++;
789 mutex_unlock(&session->s_mutex);
791 down_write(&mdsc->snap_rwsem);
792 locked_rwsem = 1;
794 if (op == CEPH_SNAP_OP_SPLIT) {
795 struct ceph_mds_snap_realm *ri;
798 * A "split" breaks part of an existing realm off into
799 * a new realm. The MDS provides a list of inodes
800 * (with caps) and child realms that belong to the new
801 * child.
803 split_inos = p;
804 p += sizeof(u64) * num_split_inos;
805 split_realms = p;
806 p += sizeof(u64) * num_split_realms;
807 ceph_decode_need(&p, e, sizeof(*ri), bad);
808 /* we will peek at realm info here, but will _not_
809 * advance p, as the realm update will occur below in
810 * ceph_update_snap_trace. */
811 ri = p;
813 realm = ceph_lookup_snap_realm(mdsc, split);
814 if (!realm) {
815 realm = ceph_create_snap_realm(mdsc, split);
816 if (IS_ERR(realm))
817 goto out;
819 ceph_get_snap_realm(mdsc, realm);
821 dout("splitting snap_realm %llx %p\n", realm->ino, realm);
822 for (i = 0; i < num_split_inos; i++) {
823 struct ceph_vino vino = {
824 .ino = le64_to_cpu(split_inos[i]),
825 .snap = CEPH_NOSNAP,
827 struct inode *inode = ceph_find_inode(sb, vino);
828 struct ceph_inode_info *ci;
829 struct ceph_snap_realm *oldrealm;
831 if (!inode)
832 continue;
833 ci = ceph_inode(inode);
835 spin_lock(&inode->i_lock);
836 if (!ci->i_snap_realm)
837 goto skip_inode;
839 * If this inode belongs to a realm that was
840 * created after our new realm, we experienced
841 * a race (due to another split notifications
842 * arriving from a different MDS). So skip
843 * this inode.
845 if (ci->i_snap_realm->created >
846 le64_to_cpu(ri->created)) {
847 dout(" leaving %p in newer realm %llx %p\n",
848 inode, ci->i_snap_realm->ino,
849 ci->i_snap_realm);
850 goto skip_inode;
852 dout(" will move %p to split realm %llx %p\n",
853 inode, realm->ino, realm);
855 * Move the inode to the new realm
857 spin_lock(&realm->inodes_with_caps_lock);
858 list_del_init(&ci->i_snap_realm_item);
859 list_add(&ci->i_snap_realm_item,
860 &realm->inodes_with_caps);
861 oldrealm = ci->i_snap_realm;
862 ci->i_snap_realm = realm;
863 spin_unlock(&realm->inodes_with_caps_lock);
864 spin_unlock(&inode->i_lock);
866 ceph_get_snap_realm(mdsc, realm);
867 ceph_put_snap_realm(mdsc, oldrealm);
869 iput(inode);
870 continue;
872 skip_inode:
873 spin_unlock(&inode->i_lock);
874 iput(inode);
877 /* we may have taken some of the old realm's children. */
878 for (i = 0; i < num_split_realms; i++) {
879 struct ceph_snap_realm *child =
880 ceph_lookup_snap_realm(mdsc,
881 le64_to_cpu(split_realms[i]));
882 if (!child)
883 continue;
884 adjust_snap_realm_parent(mdsc, child, realm->ino);
889 * update using the provided snap trace. if we are deleting a
890 * snap, we can avoid queueing cap_snaps.
892 ceph_update_snap_trace(mdsc, p, e,
893 op == CEPH_SNAP_OP_DESTROY);
895 if (op == CEPH_SNAP_OP_SPLIT)
896 /* we took a reference when we created the realm, above */
897 ceph_put_snap_realm(mdsc, realm);
899 __cleanup_empty_realms(mdsc);
901 up_write(&mdsc->snap_rwsem);
903 flush_snaps(mdsc);
904 return;
906 bad:
907 pr_err("corrupt snap message from mds%d\n", mds);
908 ceph_msg_dump(msg);
909 out:
910 if (locked_rwsem)
911 up_write(&mdsc->snap_rwsem);
912 return;