blob: 9ff5219d849e942c8f3a6a480d57daa52efeb8e0 [file] [log] [blame]
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001#include <linux/ceph/ceph_debug.h>
Sage Weil963b61e2009-10-06 11:31:12 -07002
Sage Weil963b61e2009-10-06 11:31:12 -07003#include <linux/sort.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +09004#include <linux/slab.h>
Sage Weil963b61e2009-10-06 11:31:12 -07005
6#include "super.h"
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07007#include "mds_client.h"
8
9#include <linux/ceph/decode.h>
Sage Weil963b61e2009-10-06 11:31:12 -070010
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 */
Yan, Zheng982d6012014-12-23 15:30:54 +080073 if (atomic_inc_return(&realm->nref) == 1) {
Sage Weil963b61e2009-10-06 11:31:12 -070074 spin_lock(&mdsc->snap_empty_lock);
75 list_del_init(&realm->empty_item);
76 spin_unlock(&mdsc->snap_empty_lock);
77 }
Sage Weil963b61e2009-10-06 11:31:12 -070078}
79
Sage Weila105f002010-02-15 14:37:55 -080080static 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
Sage Weil963b61e2009-10-06 11:31:12 -0700102/*
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
Yan, Zheng982d6012014-12-23 15:30:54 +0800117 atomic_set(&realm->nref, 1); /* for caller */
Sage Weil963b61e2009-10-06 11:31:12 -0700118 realm->ino = ino;
119 INIT_LIST_HEAD(&realm->children);
120 INIT_LIST_HEAD(&realm->child_item);
121 INIT_LIST_HEAD(&realm->empty_item);
Sage Weilae00d4f2010-09-16 16:26:51 -0700122 INIT_LIST_HEAD(&realm->dirty_item);
Sage Weil963b61e2009-10-06 11:31:12 -0700123 INIT_LIST_HEAD(&realm->inodes_with_caps);
124 spin_lock_init(&realm->inodes_with_caps_lock);
Sage Weila105f002010-02-15 14:37:55 -0800125 __insert_snap_realm(&mdsc->snap_realms, realm);
Sage Weil963b61e2009-10-06 11:31:12 -0700126 dout("create_snap_realm %llx %p\n", realm->ino, realm);
127 return realm;
128}
129
130/*
Sage Weila105f002010-02-15 14:37:55 -0800131 * lookup the realm rooted at @ino.
Sage Weil963b61e2009-10-06 11:31:12 -0700132 *
133 * caller must hold snap_rwsem for write.
134 */
Yan, Zheng982d6012014-12-23 15:30:54 +0800135static struct ceph_snap_realm *__lookup_snap_realm(struct ceph_mds_client *mdsc,
136 u64 ino)
Sage Weil963b61e2009-10-06 11:31:12 -0700137{
Sage Weila105f002010-02-15 14:37:55 -0800138 struct rb_node *n = mdsc->snap_realms.rb_node;
139 struct ceph_snap_realm *r;
Sage Weil963b61e2009-10-06 11:31:12 -0700140
Sage Weila105f002010-02-15 14:37:55 -0800141 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;
Sage Weil963b61e2009-10-06 11:31:12 -0700153}
154
Yan, Zheng982d6012014-12-23 15:30:54 +0800155struct 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
Sage Weil963b61e2009-10-06 11:31:12 -0700165static 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
Sage Weila105f002010-02-15 14:37:55 -0800176 rb_erase(&realm->node, &mdsc->snap_realms);
Sage Weil963b61e2009-10-06 11:31:12 -0700177
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);
Henry C Changa26a1852011-05-11 10:29:53 +0000217 list_add(&realm->empty_item, &mdsc->snap_empty);
Sage Weil963b61e2009-10-06 11:31:12 -0700218 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);
Sage Weil963b61e2009-10-06 11:31:12 -0700270 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;
Sage Weil963b61e2009-10-06 11:31:12 -0700284 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
Yan, Zheng97c85a82014-11-06 15:09:41 +0800298
Sage Weil963b61e2009-10-06 11:31:12 -0700299/*
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;
Alex Elderaa711ee2012-07-13 20:35:11 -0500307 u32 num = realm->num_prior_parent_snaps + realm->num_snaps;
Sage Weil963b61e2009-10-06 11:31:12 -0700308
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 &&
Sage Weilec4318bc2010-03-19 13:24:39 -0700328 realm->cached_context->seq == realm->seq &&
Sage Weil963b61e2009-10-06 11:31:12 -0700329 (!parent ||
Sage Weilec4318bc2010-03-19 13:24:39 -0700330 realm->cached_context->seq >= parent->cached_context->seq)) {
Alex Elderaa711ee2012-07-13 20:35:11 -0500331 dout("build_snap_context %llx %p: %p seq %lld (%u snaps)"
Sage Weil963b61e2009-10-06 11:31:12 -0700332 " (unchanged)\n",
333 realm->ino, realm, realm->cached_context,
334 realm->cached_context->seq,
Alex Elderaa711ee2012-07-13 20:35:11 -0500335 (unsigned int) realm->cached_context->num_snaps);
Sage Weil963b61e2009-10-06 11:31:12 -0700336 return 0;
337 }
338
339 /* alloc new snap context */
340 err = -ENOMEM;
Xi Wanga3860c12012-05-31 16:26:04 -0700341 if (num > (SIZE_MAX - sizeof(*snapc)) / sizeof(u64))
Sage Weil963b61e2009-10-06 11:31:12 -0700342 goto fail;
Alex Elder812164f82013-04-30 00:44:32 -0500343 snapc = ceph_create_snap_context(num, GFP_NOFS);
Sage Weil963b61e2009-10-06 11:31:12 -0700344 if (!snapc)
345 goto fail;
Sage Weil963b61e2009-10-06 11:31:12 -0700346
347 /* build (reverse sorted) snap vector */
348 num = 0;
349 snapc->seq = realm->seq;
350 if (parent) {
Alex Elderaa711ee2012-07-13 20:35:11 -0500351 u32 i;
352
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300353 /* include any of parent's snaps occurring _after_ my
Sage Weil963b61e2009-10-06 11:31:12 -0700354 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;
Alex Elderaa711ee2012-07-13 20:35:11 -0500372 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);
Sage Weil963b61e2009-10-06 11:31:12 -0700375
SF Markus Elfringe96a6502014-11-02 15:20:59 +0100376 ceph_put_snap_context(realm->cached_context);
Sage Weil963b61e2009-10-06 11:31:12 -0700377 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 */
Alex Elderaa711ee2012-07-13 20:35:11 -0500413static int dup_array(u64 **dst, __le64 *src, u32 num)
Sage Weil963b61e2009-10-06 11:31:12 -0700414{
Alex Elderaa711ee2012-07-13 20:35:11 -0500415 u32 i;
Sage Weil963b61e2009-10-06 11:31:12 -0700416
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
Yan, Zheng86056092015-05-01 16:57:16 +0800430static 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}
Sage Weil963b61e2009-10-06 11:31:12 -0700438
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 */
Sage Weilfc837c8f2010-04-13 11:41:22 -0700453void ceph_queue_cap_snap(struct ceph_inode_info *ci)
Sage Weil963b61e2009-10-06 11:31:12 -0700454{
455 struct inode *inode = &ci->vfs_inode;
456 struct ceph_cap_snap *capsnap;
Yan, Zheng86056092015-05-01 16:57:16 +0800457 struct ceph_snap_context *old_snapc, *new_snapc;
Sage Weil4a625be2010-08-22 15:03:56 -0700458 int used, dirty;
Sage Weil963b61e2009-10-06 11:31:12 -0700459
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
Sage Weilbe655592011-11-30 09:47:09 -0800466 spin_lock(&ci->i_ceph_lock);
Sage Weil963b61e2009-10-06 11:31:12 -0700467 used = __ceph_caps_used(ci);
Sage Weil4a625be2010-08-22 15:03:56 -0700468 dirty = __ceph_caps_dirty(ci);
Sage Weilaf0ed562011-07-26 11:26:31 -0700469
Yan, Zheng5dda377c2015-04-30 14:40:54 +0800470 old_snapc = ci->i_head_snapc;
Yan, Zheng86056092015-05-01 16:57:16 +0800471 new_snapc = ci->i_snap_realm->cached_context;
Yan, Zheng5dda377c2015-04-30 14:40:54 +0800472
Sage Weilaf0ed562011-07-26 11:26:31 -0700473 /*
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
Sage Weil963b61e2009-10-06 11:31:12 -0700481 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. */
Sage Weilfc837c8f2010-04-13 11:41:22 -0700486 dout("queue_cap_snap %p already pending\n", inode);
Yan, Zheng5dda377c2015-04-30 14:40:54 +0800487 goto update_snapc;
488 }
Yan, Zheng86056092015-05-01 16:57:16 +0800489 if (ci->i_wrbuffer_ref_head == 0 &&
490 !(dirty & (CEPH_CAP_ANY_EXCL|CEPH_CAP_FILE_WR))) {
Sage Weil963b61e2009-10-06 11:31:12 -0700491 dout("queue_cap_snap %p nothing dirty|writing\n", inode);
Yan, Zheng5dda377c2015-04-30 14:40:54 +0800492 goto update_snapc;
Sage Weil963b61e2009-10-06 11:31:12 -0700493 }
494
Yan, Zheng5dda377c2015-04-30 14:40:54 +0800495 BUG_ON(!old_snapc);
496
Yan, Zheng86056092015-05-01 16:57:16 +0800497 /*
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");
Yan, Zheng5dda377c2015-04-30 14:40:54 +0800520 ihold(inode);
521
522 atomic_set(&capsnap->nref, 1);
Yan, Zheng5dda377c2015-04-30 14:40:54 +0800523 INIT_LIST_HEAD(&capsnap->ci_item);
Yan, Zheng5dda377c2015-04-30 14:40:54 +0800524
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;
538 } else {
539 capsnap->xattr_blob = NULL;
540 capsnap->xattr_version = 0;
541 }
542
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);
Yan, Zheng5dda377c2015-04-30 14:40:54 +0800552
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;
Yan, Zhengfce85152016-06-15 20:51:22 +0800563 old_snapc = NULL;
Yan, Zheng5dda377c2015-04-30 14:40:54 +0800564
565update_snapc:
566 if (ci->i_head_snapc) {
Yan, Zheng86056092015-05-01 16:57:16 +0800567 ci->i_head_snapc = ceph_get_snap_context(new_snapc);
568 dout(" new snapc is %p\n", new_snapc);
Yan, Zheng5dda377c2015-04-30 14:40:54 +0800569 }
Sage Weilbe655592011-11-30 09:47:09 -0800570 spin_unlock(&ci->i_ceph_lock);
Yan, Zheng5dda377c2015-04-30 14:40:54 +0800571
572 kfree(capsnap);
573 ceph_put_snap_context(old_snapc);
Sage Weil963b61e2009-10-06 11:31:12 -0700574}
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 *
Sage Weilbe655592011-11-30 09:47:09 -0800582 * Caller must hold i_ceph_lock.
Sage Weil963b61e2009-10-06 11:31:12 -0700583 */
584int __ceph_finish_cap_snap(struct ceph_inode_info *ci,
585 struct ceph_cap_snap *capsnap)
586{
587 struct inode *inode = &ci->vfs_inode;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700588 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weil963b61e2009-10-06 11:31:12 -0700589
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;
596 if (capsnap->dirty_pages) {
Sage Weil819ccbf2010-04-01 09:33:46 -0700597 dout("finish_cap_snap %p cap_snap %p snapc %p %llu %s s=%llu "
Sage Weil963b61e2009-10-06 11:31:12 -0700598 "still has %d dirty pages\n", inode, capsnap,
599 capsnap->context, capsnap->context->seq,
Sage Weil819ccbf2010-04-01 09:33:46 -0700600 ceph_cap_string(capsnap->dirty), capsnap->size,
601 capsnap->dirty_pages);
Sage Weil963b61e2009-10-06 11:31:12 -0700602 return 0;
603 }
Yan, Zheng70220ac2016-07-06 16:21:30 +0800604
605 ci->i_ceph_flags |= CEPH_I_FLUSH_SNAPS;
Sage Weil819ccbf2010-04-01 09:33:46 -0700606 dout("finish_cap_snap %p cap_snap %p snapc %p %llu %s s=%llu\n",
Sage Weil963b61e2009-10-06 11:31:12 -0700607 inode, capsnap, capsnap->context,
Sage Weil819ccbf2010-04-01 09:33:46 -0700608 capsnap->context->seq, ceph_cap_string(capsnap->dirty),
609 capsnap->size);
Sage Weil963b61e2009-10-06 11:31:12 -0700610
611 spin_lock(&mdsc->snap_flush_lock);
612 list_add_tail(&ci->i_snap_flush_item, &mdsc->snap_flush_list);
613 spin_unlock(&mdsc->snap_flush_lock);
614 return 1; /* caller may want to ceph_flush_snaps */
615}
616
Sage Weiled326042010-08-16 13:37:31 -0700617/*
618 * Queue cap_snaps for snap writeback for this realm and its children.
619 * Called under snap_rwsem, so realm topology won't change.
620 */
621static void queue_realm_cap_snaps(struct ceph_snap_realm *realm)
622{
623 struct ceph_inode_info *ci;
624 struct inode *lastinode = NULL;
625 struct ceph_snap_realm *child;
626
627 dout("queue_realm_cap_snaps %p %llx inodes\n", realm, realm->ino);
628
629 spin_lock(&realm->inodes_with_caps_lock);
630 list_for_each_entry(ci, &realm->inodes_with_caps,
631 i_snap_realm_item) {
632 struct inode *inode = igrab(&ci->vfs_inode);
633 if (!inode)
634 continue;
635 spin_unlock(&realm->inodes_with_caps_lock);
SF Markus Elfringe96a6502014-11-02 15:20:59 +0100636 iput(lastinode);
Sage Weiled326042010-08-16 13:37:31 -0700637 lastinode = inode;
638 ceph_queue_cap_snap(ci);
639 spin_lock(&realm->inodes_with_caps_lock);
640 }
641 spin_unlock(&realm->inodes_with_caps_lock);
SF Markus Elfringe96a6502014-11-02 15:20:59 +0100642 iput(lastinode);
Sage Weiled326042010-08-16 13:37:31 -0700643
Sage Weile8e1ba962011-02-04 20:45:58 -0800644 list_for_each_entry(child, &realm->children, child_item) {
645 dout("queue_realm_cap_snaps %p %llx queue child %p %llx\n",
646 realm, realm->ino, child, child->ino);
647 list_del_init(&child->dirty_item);
648 list_add(&child->dirty_item, &realm->dirty_item);
649 }
Sage Weiled326042010-08-16 13:37:31 -0700650
Sage Weile8e1ba962011-02-04 20:45:58 -0800651 list_del_init(&realm->dirty_item);
Sage Weiled326042010-08-16 13:37:31 -0700652 dout("queue_realm_cap_snaps %p %llx done\n", realm, realm->ino);
653}
Sage Weil963b61e2009-10-06 11:31:12 -0700654
655/*
656 * Parse and apply a snapblob "snap trace" from the MDS. This specifies
657 * the snap realm parameters from a given realm and all of its ancestors,
658 * up to the root.
659 *
660 * Caller must hold snap_rwsem for write.
661 */
662int ceph_update_snap_trace(struct ceph_mds_client *mdsc,
Yan, Zheng982d6012014-12-23 15:30:54 +0800663 void *p, void *e, bool deletion,
664 struct ceph_snap_realm **realm_ret)
Sage Weil963b61e2009-10-06 11:31:12 -0700665{
666 struct ceph_mds_snap_realm *ri; /* encoded */
667 __le64 *snaps; /* encoded */
668 __le64 *prior_parent_snaps; /* encoded */
Yan, Zheng982d6012014-12-23 15:30:54 +0800669 struct ceph_snap_realm *realm = NULL;
670 struct ceph_snap_realm *first_realm = NULL;
Sage Weil963b61e2009-10-06 11:31:12 -0700671 int invalidate = 0;
672 int err = -ENOMEM;
Sage Weilae00d4f2010-09-16 16:26:51 -0700673 LIST_HEAD(dirty_realms);
Sage Weil963b61e2009-10-06 11:31:12 -0700674
675 dout("update_snap_trace deletion=%d\n", deletion);
676more:
677 ceph_decode_need(&p, e, sizeof(*ri), bad);
678 ri = p;
679 p += sizeof(*ri);
680 ceph_decode_need(&p, e, sizeof(u64)*(le32_to_cpu(ri->num_snaps) +
681 le32_to_cpu(ri->num_prior_parent_snaps)), bad);
682 snaps = p;
683 p += sizeof(u64) * le32_to_cpu(ri->num_snaps);
684 prior_parent_snaps = p;
685 p += sizeof(u64) * le32_to_cpu(ri->num_prior_parent_snaps);
686
687 realm = ceph_lookup_snap_realm(mdsc, le64_to_cpu(ri->ino));
Sage Weil963b61e2009-10-06 11:31:12 -0700688 if (!realm) {
689 realm = ceph_create_snap_realm(mdsc, le64_to_cpu(ri->ino));
690 if (IS_ERR(realm)) {
691 err = PTR_ERR(realm);
692 goto fail;
693 }
694 }
695
Sage Weil963b61e2009-10-06 11:31:12 -0700696 /* ensure the parent is correct */
697 err = adjust_snap_realm_parent(mdsc, realm, le64_to_cpu(ri->parent));
698 if (err < 0)
699 goto fail;
700 invalidate += err;
701
702 if (le64_to_cpu(ri->seq) > realm->seq) {
Sage Weilae00d4f2010-09-16 16:26:51 -0700703 dout("update_snap_trace updating %llx %p %lld -> %lld\n",
704 realm->ino, realm, realm->seq, le64_to_cpu(ri->seq));
Sage Weil963b61e2009-10-06 11:31:12 -0700705 /* update realm parameters, snap lists */
706 realm->seq = le64_to_cpu(ri->seq);
707 realm->created = le64_to_cpu(ri->created);
708 realm->parent_since = le64_to_cpu(ri->parent_since);
709
710 realm->num_snaps = le32_to_cpu(ri->num_snaps);
711 err = dup_array(&realm->snaps, snaps, realm->num_snaps);
712 if (err < 0)
713 goto fail;
714
715 realm->num_prior_parent_snaps =
716 le32_to_cpu(ri->num_prior_parent_snaps);
717 err = dup_array(&realm->prior_parent_snaps, prior_parent_snaps,
718 realm->num_prior_parent_snaps);
719 if (err < 0)
720 goto fail;
721
Sage Weilae00d4f2010-09-16 16:26:51 -0700722 /* queue realm for cap_snap creation */
723 list_add(&realm->dirty_item, &dirty_realms);
Yan, Zhengaffbc192015-05-05 21:22:13 +0800724 if (realm->seq > mdsc->last_snap_seq)
725 mdsc->last_snap_seq = realm->seq;
Sage Weilae00d4f2010-09-16 16:26:51 -0700726
Sage Weil963b61e2009-10-06 11:31:12 -0700727 invalidate = 1;
728 } else if (!realm->cached_context) {
Sage Weilae00d4f2010-09-16 16:26:51 -0700729 dout("update_snap_trace %llx %p seq %lld new\n",
730 realm->ino, realm, realm->seq);
Sage Weil963b61e2009-10-06 11:31:12 -0700731 invalidate = 1;
Sage Weilae00d4f2010-09-16 16:26:51 -0700732 } else {
733 dout("update_snap_trace %llx %p seq %lld unchanged\n",
734 realm->ino, realm, realm->seq);
Sage Weil963b61e2009-10-06 11:31:12 -0700735 }
736
737 dout("done with %llx %p, invalidated=%d, %p %p\n", realm->ino,
738 realm, invalidate, p, e);
739
Yan, Zheng982d6012014-12-23 15:30:54 +0800740 /* invalidate when we reach the _end_ (root) of the trace */
741 if (invalidate && p >= e)
742 rebuild_snap_realms(realm);
743
744 if (!first_realm)
745 first_realm = realm;
746 else
747 ceph_put_snap_realm(mdsc, realm);
748
Sage Weil963b61e2009-10-06 11:31:12 -0700749 if (p < e)
750 goto more;
751
Sage Weilae00d4f2010-09-16 16:26:51 -0700752 /*
753 * queue cap snaps _after_ we've built the new snap contexts,
754 * so that i_head_snapc can be set appropriately.
755 */
Sage Weile8e1ba962011-02-04 20:45:58 -0800756 while (!list_empty(&dirty_realms)) {
757 realm = list_first_entry(&dirty_realms, struct ceph_snap_realm,
758 dirty_item);
Sage Weilae00d4f2010-09-16 16:26:51 -0700759 queue_realm_cap_snaps(realm);
760 }
761
Yan, Zheng982d6012014-12-23 15:30:54 +0800762 if (realm_ret)
763 *realm_ret = first_realm;
764 else
765 ceph_put_snap_realm(mdsc, first_realm);
766
Sage Weil963b61e2009-10-06 11:31:12 -0700767 __cleanup_empty_realms(mdsc);
768 return 0;
769
770bad:
771 err = -EINVAL;
772fail:
Yan, Zheng982d6012014-12-23 15:30:54 +0800773 if (realm && !IS_ERR(realm))
774 ceph_put_snap_realm(mdsc, realm);
775 if (first_realm)
776 ceph_put_snap_realm(mdsc, first_realm);
Sage Weil963b61e2009-10-06 11:31:12 -0700777 pr_err("update_snap_trace error %d\n", err);
778 return err;
779}
780
781
782/*
783 * Send any cap_snaps that are queued for flush. Try to carry
784 * s_mutex across multiple snap flushes to avoid locking overhead.
785 *
786 * Caller holds no locks.
787 */
788static void flush_snaps(struct ceph_mds_client *mdsc)
789{
790 struct ceph_inode_info *ci;
791 struct inode *inode;
792 struct ceph_mds_session *session = NULL;
793
794 dout("flush_snaps\n");
795 spin_lock(&mdsc->snap_flush_lock);
796 while (!list_empty(&mdsc->snap_flush_list)) {
797 ci = list_first_entry(&mdsc->snap_flush_list,
798 struct ceph_inode_info, i_snap_flush_item);
799 inode = &ci->vfs_inode;
Sage Weil70b666c2011-05-27 09:24:26 -0700800 ihold(inode);
Sage Weil963b61e2009-10-06 11:31:12 -0700801 spin_unlock(&mdsc->snap_flush_lock);
Yan, Zhenged9b4302016-07-05 21:08:07 +0800802 ceph_flush_snaps(ci, &session);
Sage Weil963b61e2009-10-06 11:31:12 -0700803 iput(inode);
804 spin_lock(&mdsc->snap_flush_lock);
805 }
806 spin_unlock(&mdsc->snap_flush_lock);
807
808 if (session) {
809 mutex_unlock(&session->s_mutex);
810 ceph_put_mds_session(session);
811 }
812 dout("flush_snaps done\n");
813}
814
815
816/*
817 * Handle a snap notification from the MDS.
818 *
819 * This can take two basic forms: the simplest is just a snap creation
820 * or deletion notification on an existing realm. This should update the
821 * realm and its children.
822 *
823 * The more difficult case is realm creation, due to snap creation at a
824 * new point in the file hierarchy, or due to a rename that moves a file or
825 * directory into another realm.
826 */
827void ceph_handle_snap(struct ceph_mds_client *mdsc,
Sage Weil2600d2d2010-02-22 15:12:16 -0800828 struct ceph_mds_session *session,
Sage Weil963b61e2009-10-06 11:31:12 -0700829 struct ceph_msg *msg)
830{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700831 struct super_block *sb = mdsc->fsc->sb;
Sage Weil2600d2d2010-02-22 15:12:16 -0800832 int mds = session->s_mds;
Sage Weil963b61e2009-10-06 11:31:12 -0700833 u64 split;
834 int op;
835 int trace_len;
836 struct ceph_snap_realm *realm = NULL;
837 void *p = msg->front.iov_base;
838 void *e = p + msg->front.iov_len;
839 struct ceph_mds_snap_head *h;
840 int num_split_inos, num_split_realms;
841 __le64 *split_inos = NULL, *split_realms = NULL;
842 int i;
843 int locked_rwsem = 0;
844
Sage Weil963b61e2009-10-06 11:31:12 -0700845 /* decode */
846 if (msg->front.iov_len < sizeof(*h))
847 goto bad;
848 h = p;
849 op = le32_to_cpu(h->op);
850 split = le64_to_cpu(h->split); /* non-zero if we are splitting an
851 * existing realm */
852 num_split_inos = le32_to_cpu(h->num_split_inos);
853 num_split_realms = le32_to_cpu(h->num_split_realms);
854 trace_len = le32_to_cpu(h->trace_len);
855 p += sizeof(*h);
856
857 dout("handle_snap from mds%d op %s split %llx tracelen %d\n", mds,
858 ceph_snap_op_name(op), split, trace_len);
859
Sage Weil963b61e2009-10-06 11:31:12 -0700860 mutex_lock(&session->s_mutex);
861 session->s_seq++;
862 mutex_unlock(&session->s_mutex);
863
864 down_write(&mdsc->snap_rwsem);
865 locked_rwsem = 1;
866
867 if (op == CEPH_SNAP_OP_SPLIT) {
868 struct ceph_mds_snap_realm *ri;
869
870 /*
871 * A "split" breaks part of an existing realm off into
872 * a new realm. The MDS provides a list of inodes
873 * (with caps) and child realms that belong to the new
874 * child.
875 */
876 split_inos = p;
877 p += sizeof(u64) * num_split_inos;
878 split_realms = p;
879 p += sizeof(u64) * num_split_realms;
880 ceph_decode_need(&p, e, sizeof(*ri), bad);
881 /* we will peek at realm info here, but will _not_
882 * advance p, as the realm update will occur below in
883 * ceph_update_snap_trace. */
884 ri = p;
885
886 realm = ceph_lookup_snap_realm(mdsc, split);
Sage Weil963b61e2009-10-06 11:31:12 -0700887 if (!realm) {
888 realm = ceph_create_snap_realm(mdsc, split);
889 if (IS_ERR(realm))
890 goto out;
891 }
Sage Weil963b61e2009-10-06 11:31:12 -0700892
893 dout("splitting snap_realm %llx %p\n", realm->ino, realm);
894 for (i = 0; i < num_split_inos; i++) {
895 struct ceph_vino vino = {
896 .ino = le64_to_cpu(split_inos[i]),
897 .snap = CEPH_NOSNAP,
898 };
899 struct inode *inode = ceph_find_inode(sb, vino);
900 struct ceph_inode_info *ci;
Sage Weilae00d4f2010-09-16 16:26:51 -0700901 struct ceph_snap_realm *oldrealm;
Sage Weil963b61e2009-10-06 11:31:12 -0700902
903 if (!inode)
904 continue;
905 ci = ceph_inode(inode);
906
Sage Weilbe655592011-11-30 09:47:09 -0800907 spin_lock(&ci->i_ceph_lock);
Sage Weil963b61e2009-10-06 11:31:12 -0700908 if (!ci->i_snap_realm)
909 goto skip_inode;
910 /*
911 * If this inode belongs to a realm that was
912 * created after our new realm, we experienced
913 * a race (due to another split notifications
914 * arriving from a different MDS). So skip
915 * this inode.
916 */
917 if (ci->i_snap_realm->created >
918 le64_to_cpu(ri->created)) {
919 dout(" leaving %p in newer realm %llx %p\n",
920 inode, ci->i_snap_realm->ino,
921 ci->i_snap_realm);
922 goto skip_inode;
923 }
924 dout(" will move %p to split realm %llx %p\n",
925 inode, realm->ino, realm);
926 /*
Sage Weilae00d4f2010-09-16 16:26:51 -0700927 * Move the inode to the new realm
Sage Weil963b61e2009-10-06 11:31:12 -0700928 */
Sage Weil052bb342010-03-09 12:52:26 -0800929 spin_lock(&realm->inodes_with_caps_lock);
Sage Weil963b61e2009-10-06 11:31:12 -0700930 list_del_init(&ci->i_snap_realm_item);
Sage Weilae00d4f2010-09-16 16:26:51 -0700931 list_add(&ci->i_snap_realm_item,
932 &realm->inodes_with_caps);
933 oldrealm = ci->i_snap_realm;
934 ci->i_snap_realm = realm;
Sage Weil052bb342010-03-09 12:52:26 -0800935 spin_unlock(&realm->inodes_with_caps_lock);
Sage Weilbe655592011-11-30 09:47:09 -0800936 spin_unlock(&ci->i_ceph_lock);
Sage Weil963b61e2009-10-06 11:31:12 -0700937
Sage Weilae00d4f2010-09-16 16:26:51 -0700938 ceph_get_snap_realm(mdsc, realm);
939 ceph_put_snap_realm(mdsc, oldrealm);
Sage Weil963b61e2009-10-06 11:31:12 -0700940
941 iput(inode);
942 continue;
943
944skip_inode:
Sage Weilbe655592011-11-30 09:47:09 -0800945 spin_unlock(&ci->i_ceph_lock);
Sage Weil963b61e2009-10-06 11:31:12 -0700946 iput(inode);
947 }
948
949 /* we may have taken some of the old realm's children. */
950 for (i = 0; i < num_split_realms; i++) {
951 struct ceph_snap_realm *child =
Yan, Zheng982d6012014-12-23 15:30:54 +0800952 __lookup_snap_realm(mdsc,
Sage Weil963b61e2009-10-06 11:31:12 -0700953 le64_to_cpu(split_realms[i]));
Sage Weil963b61e2009-10-06 11:31:12 -0700954 if (!child)
955 continue;
956 adjust_snap_realm_parent(mdsc, child, realm->ino);
957 }
958 }
959
960 /*
961 * update using the provided snap trace. if we are deleting a
962 * snap, we can avoid queueing cap_snaps.
963 */
964 ceph_update_snap_trace(mdsc, p, e,
Yan, Zheng982d6012014-12-23 15:30:54 +0800965 op == CEPH_SNAP_OP_DESTROY, NULL);
Sage Weil963b61e2009-10-06 11:31:12 -0700966
Sage Weilae00d4f2010-09-16 16:26:51 -0700967 if (op == CEPH_SNAP_OP_SPLIT)
Sage Weil963b61e2009-10-06 11:31:12 -0700968 /* we took a reference when we created the realm, above */
969 ceph_put_snap_realm(mdsc, realm);
Sage Weil963b61e2009-10-06 11:31:12 -0700970
971 __cleanup_empty_realms(mdsc);
972
973 up_write(&mdsc->snap_rwsem);
974
975 flush_snaps(mdsc);
976 return;
977
978bad:
979 pr_err("corrupt snap message from mds%d\n", mds);
Sage Weil9ec7cab2009-12-14 15:13:47 -0800980 ceph_msg_dump(msg);
Sage Weil963b61e2009-10-06 11:31:12 -0700981out:
982 if (locked_rwsem)
983 up_write(&mdsc->snap_rwsem);
984 return;
985}