]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - fs/ceph/snap.c
Merge tag 'trace-v4.12-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt...
[mirror_ubuntu-bionic-kernel.git] / fs / ceph / snap.c
CommitLineData
3d14c5d2 1#include <linux/ceph/ceph_debug.h>
963b61eb 2
963b61eb 3#include <linux/sort.h>
5a0e3ad6 4#include <linux/slab.h>
963b61eb
SW
5
6#include "super.h"
3d14c5d2
YS
7#include "mds_client.h"
8
9#include <linux/ceph/decode.h>
963b61eb
SW
10
11/*
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.
18 *
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.
23 *
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.
33 *
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.
38 *
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.
44 *
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.
48 */
49/*
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).
54 */
55
56
57/*
58 * increase ref count for the realm
59 *
60 * caller must hold snap_rwsem for write.
61 */
62void ceph_get_snap_realm(struct ceph_mds_client *mdsc,
63 struct ceph_snap_realm *realm)
64{
65 dout("get_realm %p %d -> %d\n", realm,
66 atomic_read(&realm->nref), atomic_read(&realm->nref)+1);
67 /*
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.
72 */
982d6011 73 if (atomic_inc_return(&realm->nref) == 1) {
963b61eb
SW
74 spin_lock(&mdsc->snap_empty_lock);
75 list_del_init(&realm->empty_item);
76 spin_unlock(&mdsc->snap_empty_lock);
77 }
963b61eb
SW
78}
79
a105f00c
SW
80static void __insert_snap_realm(struct rb_root *root,
81 struct ceph_snap_realm *new)
82{
83 struct rb_node **p = &root->rb_node;
84 struct rb_node *parent = NULL;
85 struct ceph_snap_realm *r = NULL;
86
87 while (*p) {
88 parent = *p;
89 r = rb_entry(parent, struct ceph_snap_realm, node);
90 if (new->ino < r->ino)
91 p = &(*p)->rb_left;
92 else if (new->ino > r->ino)
93 p = &(*p)->rb_right;
94 else
95 BUG();
96 }
97
98 rb_link_node(&new->node, parent, p);
99 rb_insert_color(&new->node, root);
100}
101
963b61eb
SW
102/*
103 * create and get the realm rooted at @ino and bump its ref count.
104 *
105 * caller must hold snap_rwsem for write.
106 */
107static struct ceph_snap_realm *ceph_create_snap_realm(
108 struct ceph_mds_client *mdsc,
109 u64 ino)
110{
111 struct ceph_snap_realm *realm;
112
113 realm = kzalloc(sizeof(*realm), GFP_NOFS);
114 if (!realm)
115 return ERR_PTR(-ENOMEM);
116
982d6011 117 atomic_set(&realm->nref, 1); /* for caller */
963b61eb
SW
118 realm->ino = ino;
119 INIT_LIST_HEAD(&realm->children);
120 INIT_LIST_HEAD(&realm->child_item);
121 INIT_LIST_HEAD(&realm->empty_item);
ae00d4f3 122 INIT_LIST_HEAD(&realm->dirty_item);
963b61eb
SW
123 INIT_LIST_HEAD(&realm->inodes_with_caps);
124 spin_lock_init(&realm->inodes_with_caps_lock);
a105f00c 125 __insert_snap_realm(&mdsc->snap_realms, realm);
963b61eb
SW
126 dout("create_snap_realm %llx %p\n", realm->ino, realm);
127 return realm;
128}
129
130/*
a105f00c 131 * lookup the realm rooted at @ino.
963b61eb
SW
132 *
133 * caller must hold snap_rwsem for write.
134 */
982d6011
YZ
135static struct ceph_snap_realm *__lookup_snap_realm(struct ceph_mds_client *mdsc,
136 u64 ino)
963b61eb 137{
a105f00c
SW
138 struct rb_node *n = mdsc->snap_realms.rb_node;
139 struct ceph_snap_realm *r;
140
141 while (n) {
142 r = rb_entry(n, struct ceph_snap_realm, node);
143 if (ino < r->ino)
144 n = n->rb_left;
145 else if (ino > r->ino)
146 n = n->rb_right;
147 else {
148 dout("lookup_snap_realm %llx %p\n", r->ino, r);
149 return r;
150 }
151 }
152 return NULL;
963b61eb
SW
153}
154
982d6011
YZ
155struct ceph_snap_realm *ceph_lookup_snap_realm(struct ceph_mds_client *mdsc,
156 u64 ino)
157{
158 struct ceph_snap_realm *r;
159 r = __lookup_snap_realm(mdsc, ino);
160 if (r)
161 ceph_get_snap_realm(mdsc, r);
162 return r;
163}
164
963b61eb
SW
165static void __put_snap_realm(struct ceph_mds_client *mdsc,
166 struct ceph_snap_realm *realm);
167
168/*
169 * called with snap_rwsem (write)
170 */
171static void __destroy_snap_realm(struct ceph_mds_client *mdsc,
172 struct ceph_snap_realm *realm)
173{
174 dout("__destroy_snap_realm %p %llx\n", realm, realm->ino);
175
a105f00c 176 rb_erase(&realm->node, &mdsc->snap_realms);
963b61eb
SW
177
178 if (realm->parent) {
179 list_del_init(&realm->child_item);
180 __put_snap_realm(mdsc, realm->parent);
181 }
182
183 kfree(realm->prior_parent_snaps);
184 kfree(realm->snaps);
185 ceph_put_snap_context(realm->cached_context);
186 kfree(realm);
187}
188
189/*
190 * caller holds snap_rwsem (write)
191 */
192static void __put_snap_realm(struct ceph_mds_client *mdsc,
193 struct ceph_snap_realm *realm)
194{
195 dout("__put_snap_realm %llx %p %d -> %d\n", realm->ino, realm,
196 atomic_read(&realm->nref), atomic_read(&realm->nref)-1);
197 if (atomic_dec_and_test(&realm->nref))
198 __destroy_snap_realm(mdsc, realm);
199}
200
201/*
202 * caller needn't hold any locks
203 */
204void ceph_put_snap_realm(struct ceph_mds_client *mdsc,
205 struct ceph_snap_realm *realm)
206{
207 dout("put_snap_realm %llx %p %d -> %d\n", realm->ino, realm,
208 atomic_read(&realm->nref), atomic_read(&realm->nref)-1);
209 if (!atomic_dec_and_test(&realm->nref))
210 return;
211
212 if (down_write_trylock(&mdsc->snap_rwsem)) {
213 __destroy_snap_realm(mdsc, realm);
214 up_write(&mdsc->snap_rwsem);
215 } else {
216 spin_lock(&mdsc->snap_empty_lock);
a26a185d 217 list_add(&realm->empty_item, &mdsc->snap_empty);
963b61eb
SW
218 spin_unlock(&mdsc->snap_empty_lock);
219 }
220}
221
222/*
223 * Clean up any realms whose ref counts have dropped to zero. Note
224 * that this does not include realms who were created but not yet
225 * used.
226 *
227 * Called under snap_rwsem (write)
228 */
229static void __cleanup_empty_realms(struct ceph_mds_client *mdsc)
230{
231 struct ceph_snap_realm *realm;
232
233 spin_lock(&mdsc->snap_empty_lock);
234 while (!list_empty(&mdsc->snap_empty)) {
235 realm = list_first_entry(&mdsc->snap_empty,
236 struct ceph_snap_realm, empty_item);
237 list_del(&realm->empty_item);
238 spin_unlock(&mdsc->snap_empty_lock);
239 __destroy_snap_realm(mdsc, realm);
240 spin_lock(&mdsc->snap_empty_lock);
241 }
242 spin_unlock(&mdsc->snap_empty_lock);
243}
244
245void ceph_cleanup_empty_realms(struct ceph_mds_client *mdsc)
246{
247 down_write(&mdsc->snap_rwsem);
248 __cleanup_empty_realms(mdsc);
249 up_write(&mdsc->snap_rwsem);
250}
251
252/*
253 * adjust the parent realm of a given @realm. adjust child list, and parent
254 * pointers, and ref counts appropriately.
255 *
256 * return true if parent was changed, 0 if unchanged, <0 on error.
257 *
258 * caller must hold snap_rwsem for write.
259 */
260static int adjust_snap_realm_parent(struct ceph_mds_client *mdsc,
261 struct ceph_snap_realm *realm,
262 u64 parentino)
263{
264 struct ceph_snap_realm *parent;
265
266 if (realm->parent_ino == parentino)
267 return 0;
268
269 parent = ceph_lookup_snap_realm(mdsc, parentino);
963b61eb
SW
270 if (!parent) {
271 parent = ceph_create_snap_realm(mdsc, parentino);
272 if (IS_ERR(parent))
273 return PTR_ERR(parent);
274 }
275 dout("adjust_snap_realm_parent %llx %p: %llx %p -> %llx %p\n",
276 realm->ino, realm, realm->parent_ino, realm->parent,
277 parentino, parent);
278 if (realm->parent) {
279 list_del_init(&realm->child_item);
280 ceph_put_snap_realm(mdsc, realm->parent);
281 }
282 realm->parent_ino = parentino;
283 realm->parent = parent;
963b61eb
SW
284 list_add(&realm->child_item, &parent->children);
285 return 1;
286}
287
288
289static int cmpu64_rev(const void *a, const void *b)
290{
291 if (*(u64 *)a < *(u64 *)b)
292 return 1;
293 if (*(u64 *)a > *(u64 *)b)
294 return -1;
295 return 0;
296}
297
97c85a82 298
963b61eb
SW
299/*
300 * build the snap context for a given realm.
301 */
302static int build_snap_context(struct ceph_snap_realm *realm)
303{
304 struct ceph_snap_realm *parent = realm->parent;
305 struct ceph_snap_context *snapc;
306 int err = 0;
aa711ee3 307 u32 num = realm->num_prior_parent_snaps + realm->num_snaps;
963b61eb
SW
308
309 /*
310 * build parent context, if it hasn't been built.
311 * conservatively estimate that all parent snaps might be
312 * included by us.
313 */
314 if (parent) {
315 if (!parent->cached_context) {
316 err = build_snap_context(parent);
317 if (err)
318 goto fail;
319 }
320 num += parent->cached_context->num_snaps;
321 }
322
323 /* do i actually need to update? not if my context seq
324 matches realm seq, and my parents' does to. (this works
325 because we rebuild_snap_realms() works _downward_ in
326 hierarchy after each update.) */
327 if (realm->cached_context &&
ec4318bc 328 realm->cached_context->seq == realm->seq &&
963b61eb 329 (!parent ||
ec4318bc 330 realm->cached_context->seq >= parent->cached_context->seq)) {
aa711ee3 331 dout("build_snap_context %llx %p: %p seq %lld (%u snaps)"
963b61eb
SW
332 " (unchanged)\n",
333 realm->ino, realm, realm->cached_context,
334 realm->cached_context->seq,
aa711ee3 335 (unsigned int) realm->cached_context->num_snaps);
963b61eb
SW
336 return 0;
337 }
338
339 /* alloc new snap context */
340 err = -ENOMEM;
a3860c1c 341 if (num > (SIZE_MAX - sizeof(*snapc)) / sizeof(u64))
963b61eb 342 goto fail;
812164f8 343 snapc = ceph_create_snap_context(num, GFP_NOFS);
963b61eb
SW
344 if (!snapc)
345 goto fail;
963b61eb
SW
346
347 /* build (reverse sorted) snap vector */
348 num = 0;
349 snapc->seq = realm->seq;
350 if (parent) {
aa711ee3
AE
351 u32 i;
352
25985edc 353 /* include any of parent's snaps occurring _after_ my
963b61eb
SW
354 parent became my parent */
355 for (i = 0; i < parent->cached_context->num_snaps; i++)
356 if (parent->cached_context->snaps[i] >=
357 realm->parent_since)
358 snapc->snaps[num++] =
359 parent->cached_context->snaps[i];
360 if (parent->cached_context->seq > snapc->seq)
361 snapc->seq = parent->cached_context->seq;
362 }
363 memcpy(snapc->snaps + num, realm->snaps,
364 sizeof(u64)*realm->num_snaps);
365 num += realm->num_snaps;
366 memcpy(snapc->snaps + num, realm->prior_parent_snaps,
367 sizeof(u64)*realm->num_prior_parent_snaps);
368 num += realm->num_prior_parent_snaps;
369
370 sort(snapc->snaps, num, sizeof(u64), cmpu64_rev, NULL);
371 snapc->num_snaps = num;
aa711ee3
AE
372 dout("build_snap_context %llx %p: %p seq %lld (%u snaps)\n",
373 realm->ino, realm, snapc, snapc->seq,
374 (unsigned int) snapc->num_snaps);
963b61eb 375
e96a650a 376 ceph_put_snap_context(realm->cached_context);
963b61eb
SW
377 realm->cached_context = snapc;
378 return 0;
379
380fail:
381 /*
382 * if we fail, clear old (incorrect) cached_context... hopefully
383 * we'll have better luck building it later
384 */
385 if (realm->cached_context) {
386 ceph_put_snap_context(realm->cached_context);
387 realm->cached_context = NULL;
388 }
389 pr_err("build_snap_context %llx %p fail %d\n", realm->ino,
390 realm, err);
391 return err;
392}
393
394/*
395 * rebuild snap context for the given realm and all of its children.
396 */
397static void rebuild_snap_realms(struct ceph_snap_realm *realm)
398{
399 struct ceph_snap_realm *child;
400
401 dout("rebuild_snap_realms %llx %p\n", realm->ino, realm);
402 build_snap_context(realm);
403
404 list_for_each_entry(child, &realm->children, child_item)
405 rebuild_snap_realms(child);
406}
407
408
409/*
410 * helper to allocate and decode an array of snapids. free prior
411 * instance, if any.
412 */
aa711ee3 413static int dup_array(u64 **dst, __le64 *src, u32 num)
963b61eb 414{
aa711ee3 415 u32 i;
963b61eb
SW
416
417 kfree(*dst);
418 if (num) {
419 *dst = kcalloc(num, sizeof(u64), GFP_NOFS);
420 if (!*dst)
421 return -ENOMEM;
422 for (i = 0; i < num; i++)
423 (*dst)[i] = get_unaligned_le64(src + i);
424 } else {
425 *dst = NULL;
426 }
427 return 0;
428}
429
86056090
YZ
430static bool has_new_snaps(struct ceph_snap_context *o,
431 struct ceph_snap_context *n)
432{
433 if (n->num_snaps == 0)
434 return false;
435 /* snaps are in descending order */
436 return n->snaps[0] > o->seq;
437}
963b61eb
SW
438
439/*
440 * When a snapshot is applied, the size/mtime inode metadata is queued
441 * in a ceph_cap_snap (one for each snapshot) until writeback
442 * completes and the metadata can be flushed back to the MDS.
443 *
444 * However, if a (sync) write is currently in-progress when we apply
445 * the snapshot, we have to wait until the write succeeds or fails
446 * (and a final size/mtime is known). In this case the
447 * cap_snap->writing = 1, and is said to be "pending." When the write
448 * finishes, we __ceph_finish_cap_snap().
449 *
450 * Caller must hold snap_rwsem for read (i.e., the realm topology won't
451 * change).
452 */
fc837c8f 453void ceph_queue_cap_snap(struct ceph_inode_info *ci)
963b61eb
SW
454{
455 struct inode *inode = &ci->vfs_inode;
456 struct ceph_cap_snap *capsnap;
86056090 457 struct ceph_snap_context *old_snapc, *new_snapc;
4a625be4 458 int used, dirty;
963b61eb
SW
459
460 capsnap = kzalloc(sizeof(*capsnap), GFP_NOFS);
461 if (!capsnap) {
462 pr_err("ENOMEM allocating ceph_cap_snap on %p\n", inode);
463 return;
464 }
465
be655596 466 spin_lock(&ci->i_ceph_lock);
963b61eb 467 used = __ceph_caps_used(ci);
4a625be4 468 dirty = __ceph_caps_dirty(ci);
af0ed569 469
5dda377c 470 old_snapc = ci->i_head_snapc;
86056090 471 new_snapc = ci->i_snap_realm->cached_context;
5dda377c 472
af0ed569
SW
473 /*
474 * If there is a write in progress, treat that as a dirty Fw,
475 * even though it hasn't completed yet; by the time we finish
476 * up this capsnap it will be.
477 */
478 if (used & CEPH_CAP_FILE_WR)
479 dirty |= CEPH_CAP_FILE_WR;
480
963b61eb
SW
481 if (__ceph_have_pending_cap_snap(ci)) {
482 /* there is no point in queuing multiple "pending" cap_snaps,
483 as no new writes are allowed to start when pending, so any
484 writes in progress now were started before the previous
485 cap_snap. lucky us. */
fc837c8f 486 dout("queue_cap_snap %p already pending\n", inode);
5dda377c
YZ
487 goto update_snapc;
488 }
86056090
YZ
489 if (ci->i_wrbuffer_ref_head == 0 &&
490 !(dirty & (CEPH_CAP_ANY_EXCL|CEPH_CAP_FILE_WR))) {
5dda377c
YZ
491 dout("queue_cap_snap %p nothing dirty|writing\n", inode);
492 goto update_snapc;
493 }
963b61eb 494
5dda377c 495 BUG_ON(!old_snapc);
e20d258d 496
86056090
YZ
497 /*
498 * There is no need to send FLUSHSNAP message to MDS if there is
499 * no new snapshot. But when there is dirty pages or on-going
500 * writes, we still need to create cap_snap. cap_snap is needed
501 * by the write path and page writeback path.
502 *
503 * also see ceph_try_drop_cap_snap()
504 */
505 if (has_new_snaps(old_snapc, new_snapc)) {
506 if (dirty & (CEPH_CAP_ANY_EXCL|CEPH_CAP_FILE_WR))
507 capsnap->need_flush = true;
508 } else {
509 if (!(used & CEPH_CAP_FILE_WR) &&
510 ci->i_wrbuffer_ref_head == 0) {
511 dout("queue_cap_snap %p "
512 "no new_snap|dirty_page|writing\n", inode);
513 goto update_snapc;
514 }
515 }
516
517 dout("queue_cap_snap %p cap_snap %p queuing under %p %s %s\n",
518 inode, capsnap, old_snapc, ceph_cap_string(dirty),
519 capsnap->need_flush ? "" : "no_flush");
5dda377c
YZ
520 ihold(inode);
521
805692d0 522 refcount_set(&capsnap->nref, 1);
5dda377c 523 INIT_LIST_HEAD(&capsnap->ci_item);
5dda377c
YZ
524
525 capsnap->follows = old_snapc->seq;
526 capsnap->issued = __ceph_caps_issued(ci, NULL);
527 capsnap->dirty = dirty;
528
529 capsnap->mode = inode->i_mode;
530 capsnap->uid = inode->i_uid;
531 capsnap->gid = inode->i_gid;
532
533 if (dirty & CEPH_CAP_XATTR_EXCL) {
534 __ceph_build_xattrs_blob(ci);
535 capsnap->xattr_blob =
536 ceph_buffer_get(ci->i_xattrs.blob);
537 capsnap->xattr_version = ci->i_xattrs.version;
963b61eb 538 } else {
5dda377c
YZ
539 capsnap->xattr_blob = NULL;
540 capsnap->xattr_version = 0;
963b61eb
SW
541 }
542
5dda377c
YZ
543 capsnap->inline_data = ci->i_inline_version != CEPH_INLINE_NONE;
544
545 /* dirty page count moved from _head to this cap_snap;
546 all subsequent writes page dirties occur _after_ this
547 snapshot. */
548 capsnap->dirty_pages = ci->i_wrbuffer_ref_head;
549 ci->i_wrbuffer_ref_head = 0;
550 capsnap->context = old_snapc;
551 list_add_tail(&capsnap->ci_item, &ci->i_cap_snaps);
5dda377c
YZ
552
553 if (used & CEPH_CAP_FILE_WR) {
554 dout("queue_cap_snap %p cap_snap %p snapc %p"
555 " seq %llu used WR, now pending\n", inode,
556 capsnap, old_snapc, old_snapc->seq);
557 capsnap->writing = 1;
558 } else {
559 /* note mtime, size NOW. */
560 __ceph_finish_cap_snap(ci, capsnap);
561 }
562 capsnap = NULL;
fce85157 563 old_snapc = NULL;
5dda377c
YZ
564
565update_snapc:
566 if (ci->i_head_snapc) {
86056090
YZ
567 ci->i_head_snapc = ceph_get_snap_context(new_snapc);
568 dout(" new snapc is %p\n", new_snapc);
5dda377c 569 }
be655596 570 spin_unlock(&ci->i_ceph_lock);
5dda377c
YZ
571
572 kfree(capsnap);
573 ceph_put_snap_context(old_snapc);
963b61eb
SW
574}
575
576/*
577 * Finalize the size, mtime for a cap_snap.. that is, settle on final values
578 * to be used for the snapshot, to be flushed back to the mds.
579 *
580 * If capsnap can now be flushed, add to snap_flush list, and return 1.
581 *
be655596 582 * Caller must hold i_ceph_lock.
963b61eb
SW
583 */
584int __ceph_finish_cap_snap(struct ceph_inode_info *ci,
585 struct ceph_cap_snap *capsnap)
586{
587 struct inode *inode = &ci->vfs_inode;
3d14c5d2 588 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
963b61eb
SW
589
590 BUG_ON(capsnap->writing);
591 capsnap->size = inode->i_size;
592 capsnap->mtime = inode->i_mtime;
593 capsnap->atime = inode->i_atime;
594 capsnap->ctime = inode->i_ctime;
595 capsnap->time_warp_seq = ci->i_time_warp_seq;
5f743e45
YZ
596 capsnap->truncate_size = ci->i_truncate_size;
597 capsnap->truncate_seq = ci->i_truncate_seq;
963b61eb 598 if (capsnap->dirty_pages) {
819ccbfa 599 dout("finish_cap_snap %p cap_snap %p snapc %p %llu %s s=%llu "
963b61eb
SW
600 "still has %d dirty pages\n", inode, capsnap,
601 capsnap->context, capsnap->context->seq,
819ccbfa
SW
602 ceph_cap_string(capsnap->dirty), capsnap->size,
603 capsnap->dirty_pages);
963b61eb
SW
604 return 0;
605 }
70220ac8
YZ
606
607 ci->i_ceph_flags |= CEPH_I_FLUSH_SNAPS;
819ccbfa 608 dout("finish_cap_snap %p cap_snap %p snapc %p %llu %s s=%llu\n",
963b61eb 609 inode, capsnap, capsnap->context,
819ccbfa
SW
610 capsnap->context->seq, ceph_cap_string(capsnap->dirty),
611 capsnap->size);
963b61eb
SW
612
613 spin_lock(&mdsc->snap_flush_lock);
614 list_add_tail(&ci->i_snap_flush_item, &mdsc->snap_flush_list);
615 spin_unlock(&mdsc->snap_flush_lock);
616 return 1; /* caller may want to ceph_flush_snaps */
617}
618
ed326044
SW
619/*
620 * Queue cap_snaps for snap writeback for this realm and its children.
621 * Called under snap_rwsem, so realm topology won't change.
622 */
623static void queue_realm_cap_snaps(struct ceph_snap_realm *realm)
624{
625 struct ceph_inode_info *ci;
626 struct inode *lastinode = NULL;
627 struct ceph_snap_realm *child;
628
629 dout("queue_realm_cap_snaps %p %llx inodes\n", realm, realm->ino);
630
631 spin_lock(&realm->inodes_with_caps_lock);
632 list_for_each_entry(ci, &realm->inodes_with_caps,
633 i_snap_realm_item) {
634 struct inode *inode = igrab(&ci->vfs_inode);
635 if (!inode)
636 continue;
637 spin_unlock(&realm->inodes_with_caps_lock);
e96a650a 638 iput(lastinode);
ed326044
SW
639 lastinode = inode;
640 ceph_queue_cap_snap(ci);
641 spin_lock(&realm->inodes_with_caps_lock);
642 }
643 spin_unlock(&realm->inodes_with_caps_lock);
e96a650a 644 iput(lastinode);
ed326044 645
e8e1ba96
SW
646 list_for_each_entry(child, &realm->children, child_item) {
647 dout("queue_realm_cap_snaps %p %llx queue child %p %llx\n",
648 realm, realm->ino, child, child->ino);
649 list_del_init(&child->dirty_item);
650 list_add(&child->dirty_item, &realm->dirty_item);
651 }
ed326044 652
e8e1ba96 653 list_del_init(&realm->dirty_item);
ed326044
SW
654 dout("queue_realm_cap_snaps %p %llx done\n", realm, realm->ino);
655}
963b61eb
SW
656
657/*
658 * Parse and apply a snapblob "snap trace" from the MDS. This specifies
659 * the snap realm parameters from a given realm and all of its ancestors,
660 * up to the root.
661 *
662 * Caller must hold snap_rwsem for write.
663 */
664int ceph_update_snap_trace(struct ceph_mds_client *mdsc,
982d6011
YZ
665 void *p, void *e, bool deletion,
666 struct ceph_snap_realm **realm_ret)
963b61eb
SW
667{
668 struct ceph_mds_snap_realm *ri; /* encoded */
669 __le64 *snaps; /* encoded */
670 __le64 *prior_parent_snaps; /* encoded */
982d6011
YZ
671 struct ceph_snap_realm *realm = NULL;
672 struct ceph_snap_realm *first_realm = NULL;
963b61eb
SW
673 int invalidate = 0;
674 int err = -ENOMEM;
ae00d4f3 675 LIST_HEAD(dirty_realms);
963b61eb
SW
676
677 dout("update_snap_trace deletion=%d\n", deletion);
678more:
679 ceph_decode_need(&p, e, sizeof(*ri), bad);
680 ri = p;
681 p += sizeof(*ri);
682 ceph_decode_need(&p, e, sizeof(u64)*(le32_to_cpu(ri->num_snaps) +
683 le32_to_cpu(ri->num_prior_parent_snaps)), bad);
684 snaps = p;
685 p += sizeof(u64) * le32_to_cpu(ri->num_snaps);
686 prior_parent_snaps = p;
687 p += sizeof(u64) * le32_to_cpu(ri->num_prior_parent_snaps);
688
689 realm = ceph_lookup_snap_realm(mdsc, le64_to_cpu(ri->ino));
963b61eb
SW
690 if (!realm) {
691 realm = ceph_create_snap_realm(mdsc, le64_to_cpu(ri->ino));
692 if (IS_ERR(realm)) {
693 err = PTR_ERR(realm);
694 goto fail;
695 }
696 }
697
963b61eb
SW
698 /* ensure the parent is correct */
699 err = adjust_snap_realm_parent(mdsc, realm, le64_to_cpu(ri->parent));
700 if (err < 0)
701 goto fail;
702 invalidate += err;
703
704 if (le64_to_cpu(ri->seq) > realm->seq) {
ae00d4f3
SW
705 dout("update_snap_trace updating %llx %p %lld -> %lld\n",
706 realm->ino, realm, realm->seq, le64_to_cpu(ri->seq));
963b61eb
SW
707 /* update realm parameters, snap lists */
708 realm->seq = le64_to_cpu(ri->seq);
709 realm->created = le64_to_cpu(ri->created);
710 realm->parent_since = le64_to_cpu(ri->parent_since);
711
712 realm->num_snaps = le32_to_cpu(ri->num_snaps);
713 err = dup_array(&realm->snaps, snaps, realm->num_snaps);
714 if (err < 0)
715 goto fail;
716
717 realm->num_prior_parent_snaps =
718 le32_to_cpu(ri->num_prior_parent_snaps);
719 err = dup_array(&realm->prior_parent_snaps, prior_parent_snaps,
720 realm->num_prior_parent_snaps);
721 if (err < 0)
722 goto fail;
723
ae00d4f3
SW
724 /* queue realm for cap_snap creation */
725 list_add(&realm->dirty_item, &dirty_realms);
affbc19a
YZ
726 if (realm->seq > mdsc->last_snap_seq)
727 mdsc->last_snap_seq = realm->seq;
ae00d4f3 728
963b61eb
SW
729 invalidate = 1;
730 } else if (!realm->cached_context) {
ae00d4f3
SW
731 dout("update_snap_trace %llx %p seq %lld new\n",
732 realm->ino, realm, realm->seq);
963b61eb 733 invalidate = 1;
ae00d4f3
SW
734 } else {
735 dout("update_snap_trace %llx %p seq %lld unchanged\n",
736 realm->ino, realm, realm->seq);
963b61eb
SW
737 }
738
739 dout("done with %llx %p, invalidated=%d, %p %p\n", realm->ino,
740 realm, invalidate, p, e);
741
963b61eb 742 /* invalidate when we reach the _end_ (root) of the trace */
982d6011 743 if (invalidate && p >= e)
963b61eb
SW
744 rebuild_snap_realms(realm);
745
982d6011
YZ
746 if (!first_realm)
747 first_realm = realm;
748 else
749 ceph_put_snap_realm(mdsc, realm);
750
751 if (p < e)
752 goto more;
753
ae00d4f3
SW
754 /*
755 * queue cap snaps _after_ we've built the new snap contexts,
756 * so that i_head_snapc can be set appropriately.
757 */
e8e1ba96
SW
758 while (!list_empty(&dirty_realms)) {
759 realm = list_first_entry(&dirty_realms, struct ceph_snap_realm,
760 dirty_item);
ae00d4f3
SW
761 queue_realm_cap_snaps(realm);
762 }
763
982d6011
YZ
764 if (realm_ret)
765 *realm_ret = first_realm;
766 else
767 ceph_put_snap_realm(mdsc, first_realm);
768
963b61eb
SW
769 __cleanup_empty_realms(mdsc);
770 return 0;
771
772bad:
773 err = -EINVAL;
774fail:
982d6011
YZ
775 if (realm && !IS_ERR(realm))
776 ceph_put_snap_realm(mdsc, realm);
777 if (first_realm)
778 ceph_put_snap_realm(mdsc, first_realm);
963b61eb
SW
779 pr_err("update_snap_trace error %d\n", err);
780 return err;
781}
782
783
784/*
785 * Send any cap_snaps that are queued for flush. Try to carry
786 * s_mutex across multiple snap flushes to avoid locking overhead.
787 *
788 * Caller holds no locks.
789 */
790static void flush_snaps(struct ceph_mds_client *mdsc)
791{
792 struct ceph_inode_info *ci;
793 struct inode *inode;
794 struct ceph_mds_session *session = NULL;
795
796 dout("flush_snaps\n");
797 spin_lock(&mdsc->snap_flush_lock);
798 while (!list_empty(&mdsc->snap_flush_list)) {
799 ci = list_first_entry(&mdsc->snap_flush_list,
800 struct ceph_inode_info, i_snap_flush_item);
801 inode = &ci->vfs_inode;
70b666c3 802 ihold(inode);
963b61eb 803 spin_unlock(&mdsc->snap_flush_lock);
ed9b430c 804 ceph_flush_snaps(ci, &session);
963b61eb
SW
805 iput(inode);
806 spin_lock(&mdsc->snap_flush_lock);
807 }
808 spin_unlock(&mdsc->snap_flush_lock);
809
810 if (session) {
811 mutex_unlock(&session->s_mutex);
812 ceph_put_mds_session(session);
813 }
814 dout("flush_snaps done\n");
815}
816
817
818/*
819 * Handle a snap notification from the MDS.
820 *
821 * This can take two basic forms: the simplest is just a snap creation
822 * or deletion notification on an existing realm. This should update the
823 * realm and its children.
824 *
825 * The more difficult case is realm creation, due to snap creation at a
826 * new point in the file hierarchy, or due to a rename that moves a file or
827 * directory into another realm.
828 */
829void ceph_handle_snap(struct ceph_mds_client *mdsc,
2600d2dd 830 struct ceph_mds_session *session,
963b61eb
SW
831 struct ceph_msg *msg)
832{
3d14c5d2 833 struct super_block *sb = mdsc->fsc->sb;
2600d2dd 834 int mds = session->s_mds;
963b61eb
SW
835 u64 split;
836 int op;
837 int trace_len;
838 struct ceph_snap_realm *realm = NULL;
839 void *p = msg->front.iov_base;
840 void *e = p + msg->front.iov_len;
841 struct ceph_mds_snap_head *h;
842 int num_split_inos, num_split_realms;
843 __le64 *split_inos = NULL, *split_realms = NULL;
844 int i;
845 int locked_rwsem = 0;
846
963b61eb
SW
847 /* decode */
848 if (msg->front.iov_len < sizeof(*h))
849 goto bad;
850 h = p;
851 op = le32_to_cpu(h->op);
852 split = le64_to_cpu(h->split); /* non-zero if we are splitting an
853 * existing realm */
854 num_split_inos = le32_to_cpu(h->num_split_inos);
855 num_split_realms = le32_to_cpu(h->num_split_realms);
856 trace_len = le32_to_cpu(h->trace_len);
857 p += sizeof(*h);
858
859 dout("handle_snap from mds%d op %s split %llx tracelen %d\n", mds,
860 ceph_snap_op_name(op), split, trace_len);
861
963b61eb
SW
862 mutex_lock(&session->s_mutex);
863 session->s_seq++;
864 mutex_unlock(&session->s_mutex);
865
866 down_write(&mdsc->snap_rwsem);
867 locked_rwsem = 1;
868
869 if (op == CEPH_SNAP_OP_SPLIT) {
870 struct ceph_mds_snap_realm *ri;
871
872 /*
873 * A "split" breaks part of an existing realm off into
874 * a new realm. The MDS provides a list of inodes
875 * (with caps) and child realms that belong to the new
876 * child.
877 */
878 split_inos = p;
879 p += sizeof(u64) * num_split_inos;
880 split_realms = p;
881 p += sizeof(u64) * num_split_realms;
882 ceph_decode_need(&p, e, sizeof(*ri), bad);
883 /* we will peek at realm info here, but will _not_
884 * advance p, as the realm update will occur below in
885 * ceph_update_snap_trace. */
886 ri = p;
887
888 realm = ceph_lookup_snap_realm(mdsc, split);
963b61eb
SW
889 if (!realm) {
890 realm = ceph_create_snap_realm(mdsc, split);
891 if (IS_ERR(realm))
892 goto out;
893 }
963b61eb
SW
894
895 dout("splitting snap_realm %llx %p\n", realm->ino, realm);
896 for (i = 0; i < num_split_inos; i++) {
897 struct ceph_vino vino = {
898 .ino = le64_to_cpu(split_inos[i]),
899 .snap = CEPH_NOSNAP,
900 };
901 struct inode *inode = ceph_find_inode(sb, vino);
902 struct ceph_inode_info *ci;
ae00d4f3 903 struct ceph_snap_realm *oldrealm;
963b61eb
SW
904
905 if (!inode)
906 continue;
907 ci = ceph_inode(inode);
908
be655596 909 spin_lock(&ci->i_ceph_lock);
963b61eb
SW
910 if (!ci->i_snap_realm)
911 goto skip_inode;
912 /*
913 * If this inode belongs to a realm that was
914 * created after our new realm, we experienced
915 * a race (due to another split notifications
916 * arriving from a different MDS). So skip
917 * this inode.
918 */
919 if (ci->i_snap_realm->created >
920 le64_to_cpu(ri->created)) {
921 dout(" leaving %p in newer realm %llx %p\n",
922 inode, ci->i_snap_realm->ino,
923 ci->i_snap_realm);
924 goto skip_inode;
925 }
926 dout(" will move %p to split realm %llx %p\n",
927 inode, realm->ino, realm);
928 /*
ae00d4f3 929 * Move the inode to the new realm
963b61eb 930 */
052bb34a 931 spin_lock(&realm->inodes_with_caps_lock);
963b61eb 932 list_del_init(&ci->i_snap_realm_item);
ae00d4f3
SW
933 list_add(&ci->i_snap_realm_item,
934 &realm->inodes_with_caps);
935 oldrealm = ci->i_snap_realm;
936 ci->i_snap_realm = realm;
052bb34a 937 spin_unlock(&realm->inodes_with_caps_lock);
be655596 938 spin_unlock(&ci->i_ceph_lock);
963b61eb 939
ae00d4f3
SW
940 ceph_get_snap_realm(mdsc, realm);
941 ceph_put_snap_realm(mdsc, oldrealm);
963b61eb
SW
942
943 iput(inode);
944 continue;
945
946skip_inode:
be655596 947 spin_unlock(&ci->i_ceph_lock);
963b61eb
SW
948 iput(inode);
949 }
950
951 /* we may have taken some of the old realm's children. */
952 for (i = 0; i < num_split_realms; i++) {
953 struct ceph_snap_realm *child =
982d6011 954 __lookup_snap_realm(mdsc,
963b61eb 955 le64_to_cpu(split_realms[i]));
963b61eb
SW
956 if (!child)
957 continue;
958 adjust_snap_realm_parent(mdsc, child, realm->ino);
959 }
960 }
961
962 /*
963 * update using the provided snap trace. if we are deleting a
964 * snap, we can avoid queueing cap_snaps.
965 */
966 ceph_update_snap_trace(mdsc, p, e,
982d6011 967 op == CEPH_SNAP_OP_DESTROY, NULL);
963b61eb 968
ae00d4f3 969 if (op == CEPH_SNAP_OP_SPLIT)
963b61eb
SW
970 /* we took a reference when we created the realm, above */
971 ceph_put_snap_realm(mdsc, realm);
963b61eb
SW
972
973 __cleanup_empty_realms(mdsc);
974
975 up_write(&mdsc->snap_rwsem);
976
977 flush_snaps(mdsc);
978 return;
979
980bad:
981 pr_err("corrupt snap message from mds%d\n", mds);
9ec7cab1 982 ceph_msg_dump(msg);
963b61eb
SW
983out:
984 if (locked_rwsem)
985 up_write(&mdsc->snap_rwsem);
986 return;
987}