]>
Commit | Line | Data |
---|---|---|
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 | */ | |
62 | void 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 |
80 | static 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 | */ | |
107 | static 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 |
135 | static 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 |
155 | struct 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 |
165 | static void __put_snap_realm(struct ceph_mds_client *mdsc, |
166 | struct ceph_snap_realm *realm); | |
167 | ||
168 | /* | |
169 | * called with snap_rwsem (write) | |
170 | */ | |
171 | static 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 | */ | |
192 | static 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 | */ | |
204 | void 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 | */ | |
229 | static 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 | ||
245 | void 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 | */ | |
260 | static 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 | ||
289 | static 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 | */ | |
3ae0bebc YZ |
302 | static int build_snap_context(struct ceph_snap_realm *realm, |
303 | struct list_head* dirty_realms) | |
963b61eb SW |
304 | { |
305 | struct ceph_snap_realm *parent = realm->parent; | |
306 | struct ceph_snap_context *snapc; | |
307 | int err = 0; | |
aa711ee3 | 308 | u32 num = realm->num_prior_parent_snaps + realm->num_snaps; |
963b61eb SW |
309 | |
310 | /* | |
311 | * build parent context, if it hasn't been built. | |
312 | * conservatively estimate that all parent snaps might be | |
313 | * included by us. | |
314 | */ | |
315 | if (parent) { | |
316 | if (!parent->cached_context) { | |
3ae0bebc | 317 | err = build_snap_context(parent, dirty_realms); |
963b61eb SW |
318 | if (err) |
319 | goto fail; | |
320 | } | |
321 | num += parent->cached_context->num_snaps; | |
322 | } | |
323 | ||
324 | /* do i actually need to update? not if my context seq | |
325 | matches realm seq, and my parents' does to. (this works | |
326 | because we rebuild_snap_realms() works _downward_ in | |
327 | hierarchy after each update.) */ | |
328 | if (realm->cached_context && | |
ec4318bc | 329 | realm->cached_context->seq == realm->seq && |
963b61eb | 330 | (!parent || |
ec4318bc | 331 | realm->cached_context->seq >= parent->cached_context->seq)) { |
aa711ee3 | 332 | dout("build_snap_context %llx %p: %p seq %lld (%u snaps)" |
963b61eb SW |
333 | " (unchanged)\n", |
334 | realm->ino, realm, realm->cached_context, | |
335 | realm->cached_context->seq, | |
3ae0bebc | 336 | (unsigned int)realm->cached_context->num_snaps); |
963b61eb SW |
337 | return 0; |
338 | } | |
339 | ||
340 | /* alloc new snap context */ | |
341 | err = -ENOMEM; | |
a3860c1c | 342 | if (num > (SIZE_MAX - sizeof(*snapc)) / sizeof(u64)) |
963b61eb | 343 | goto fail; |
812164f8 | 344 | snapc = ceph_create_snap_context(num, GFP_NOFS); |
963b61eb SW |
345 | if (!snapc) |
346 | goto fail; | |
963b61eb SW |
347 | |
348 | /* build (reverse sorted) snap vector */ | |
349 | num = 0; | |
350 | snapc->seq = realm->seq; | |
351 | if (parent) { | |
aa711ee3 AE |
352 | u32 i; |
353 | ||
25985edc | 354 | /* include any of parent's snaps occurring _after_ my |
963b61eb SW |
355 | parent became my parent */ |
356 | for (i = 0; i < parent->cached_context->num_snaps; i++) | |
357 | if (parent->cached_context->snaps[i] >= | |
358 | realm->parent_since) | |
359 | snapc->snaps[num++] = | |
360 | parent->cached_context->snaps[i]; | |
361 | if (parent->cached_context->seq > snapc->seq) | |
362 | snapc->seq = parent->cached_context->seq; | |
363 | } | |
364 | memcpy(snapc->snaps + num, realm->snaps, | |
365 | sizeof(u64)*realm->num_snaps); | |
366 | num += realm->num_snaps; | |
367 | memcpy(snapc->snaps + num, realm->prior_parent_snaps, | |
368 | sizeof(u64)*realm->num_prior_parent_snaps); | |
369 | num += realm->num_prior_parent_snaps; | |
370 | ||
371 | sort(snapc->snaps, num, sizeof(u64), cmpu64_rev, NULL); | |
372 | snapc->num_snaps = num; | |
aa711ee3 AE |
373 | dout("build_snap_context %llx %p: %p seq %lld (%u snaps)\n", |
374 | realm->ino, realm, snapc, snapc->seq, | |
375 | (unsigned int) snapc->num_snaps); | |
963b61eb | 376 | |
3ae0bebc YZ |
377 | if (realm->cached_context) { |
378 | ceph_put_snap_context(realm->cached_context); | |
379 | /* queue realm for cap_snap creation */ | |
380 | list_add_tail(&realm->dirty_item, dirty_realms); | |
381 | } | |
963b61eb SW |
382 | realm->cached_context = snapc; |
383 | return 0; | |
384 | ||
385 | fail: | |
386 | /* | |
387 | * if we fail, clear old (incorrect) cached_context... hopefully | |
388 | * we'll have better luck building it later | |
389 | */ | |
390 | if (realm->cached_context) { | |
391 | ceph_put_snap_context(realm->cached_context); | |
392 | realm->cached_context = NULL; | |
393 | } | |
394 | pr_err("build_snap_context %llx %p fail %d\n", realm->ino, | |
395 | realm, err); | |
396 | return err; | |
397 | } | |
398 | ||
399 | /* | |
400 | * rebuild snap context for the given realm and all of its children. | |
401 | */ | |
3ae0bebc YZ |
402 | static void rebuild_snap_realms(struct ceph_snap_realm *realm, |
403 | struct list_head *dirty_realms) | |
963b61eb SW |
404 | { |
405 | struct ceph_snap_realm *child; | |
406 | ||
407 | dout("rebuild_snap_realms %llx %p\n", realm->ino, realm); | |
3ae0bebc | 408 | build_snap_context(realm, dirty_realms); |
963b61eb SW |
409 | |
410 | list_for_each_entry(child, &realm->children, child_item) | |
3ae0bebc | 411 | rebuild_snap_realms(child, dirty_realms); |
963b61eb SW |
412 | } |
413 | ||
414 | ||
415 | /* | |
416 | * helper to allocate and decode an array of snapids. free prior | |
417 | * instance, if any. | |
418 | */ | |
aa711ee3 | 419 | static int dup_array(u64 **dst, __le64 *src, u32 num) |
963b61eb | 420 | { |
aa711ee3 | 421 | u32 i; |
963b61eb SW |
422 | |
423 | kfree(*dst); | |
424 | if (num) { | |
425 | *dst = kcalloc(num, sizeof(u64), GFP_NOFS); | |
426 | if (!*dst) | |
427 | return -ENOMEM; | |
428 | for (i = 0; i < num; i++) | |
429 | (*dst)[i] = get_unaligned_le64(src + i); | |
430 | } else { | |
431 | *dst = NULL; | |
432 | } | |
433 | return 0; | |
434 | } | |
435 | ||
86056090 YZ |
436 | static bool has_new_snaps(struct ceph_snap_context *o, |
437 | struct ceph_snap_context *n) | |
438 | { | |
439 | if (n->num_snaps == 0) | |
440 | return false; | |
441 | /* snaps are in descending order */ | |
442 | return n->snaps[0] > o->seq; | |
443 | } | |
963b61eb SW |
444 | |
445 | /* | |
446 | * When a snapshot is applied, the size/mtime inode metadata is queued | |
447 | * in a ceph_cap_snap (one for each snapshot) until writeback | |
448 | * completes and the metadata can be flushed back to the MDS. | |
449 | * | |
450 | * However, if a (sync) write is currently in-progress when we apply | |
451 | * the snapshot, we have to wait until the write succeeds or fails | |
452 | * (and a final size/mtime is known). In this case the | |
453 | * cap_snap->writing = 1, and is said to be "pending." When the write | |
454 | * finishes, we __ceph_finish_cap_snap(). | |
455 | * | |
456 | * Caller must hold snap_rwsem for read (i.e., the realm topology won't | |
457 | * change). | |
458 | */ | |
fc837c8f | 459 | void ceph_queue_cap_snap(struct ceph_inode_info *ci) |
963b61eb SW |
460 | { |
461 | struct inode *inode = &ci->vfs_inode; | |
462 | struct ceph_cap_snap *capsnap; | |
86056090 | 463 | struct ceph_snap_context *old_snapc, *new_snapc; |
4a625be4 | 464 | int used, dirty; |
963b61eb SW |
465 | |
466 | capsnap = kzalloc(sizeof(*capsnap), GFP_NOFS); | |
467 | if (!capsnap) { | |
468 | pr_err("ENOMEM allocating ceph_cap_snap on %p\n", inode); | |
469 | return; | |
470 | } | |
471 | ||
be655596 | 472 | spin_lock(&ci->i_ceph_lock); |
963b61eb | 473 | used = __ceph_caps_used(ci); |
4a625be4 | 474 | dirty = __ceph_caps_dirty(ci); |
af0ed569 | 475 | |
5dda377c | 476 | old_snapc = ci->i_head_snapc; |
86056090 | 477 | new_snapc = ci->i_snap_realm->cached_context; |
5dda377c | 478 | |
af0ed569 SW |
479 | /* |
480 | * If there is a write in progress, treat that as a dirty Fw, | |
481 | * even though it hasn't completed yet; by the time we finish | |
482 | * up this capsnap it will be. | |
483 | */ | |
484 | if (used & CEPH_CAP_FILE_WR) | |
485 | dirty |= CEPH_CAP_FILE_WR; | |
486 | ||
963b61eb SW |
487 | if (__ceph_have_pending_cap_snap(ci)) { |
488 | /* there is no point in queuing multiple "pending" cap_snaps, | |
489 | as no new writes are allowed to start when pending, so any | |
490 | writes in progress now were started before the previous | |
491 | cap_snap. lucky us. */ | |
fc837c8f | 492 | dout("queue_cap_snap %p already pending\n", inode); |
5dda377c YZ |
493 | goto update_snapc; |
494 | } | |
86056090 YZ |
495 | if (ci->i_wrbuffer_ref_head == 0 && |
496 | !(dirty & (CEPH_CAP_ANY_EXCL|CEPH_CAP_FILE_WR))) { | |
5dda377c YZ |
497 | dout("queue_cap_snap %p nothing dirty|writing\n", inode); |
498 | goto update_snapc; | |
499 | } | |
963b61eb | 500 | |
5dda377c | 501 | BUG_ON(!old_snapc); |
e20d258d | 502 | |
86056090 YZ |
503 | /* |
504 | * There is no need to send FLUSHSNAP message to MDS if there is | |
505 | * no new snapshot. But when there is dirty pages or on-going | |
506 | * writes, we still need to create cap_snap. cap_snap is needed | |
507 | * by the write path and page writeback path. | |
508 | * | |
509 | * also see ceph_try_drop_cap_snap() | |
510 | */ | |
511 | if (has_new_snaps(old_snapc, new_snapc)) { | |
512 | if (dirty & (CEPH_CAP_ANY_EXCL|CEPH_CAP_FILE_WR)) | |
513 | capsnap->need_flush = true; | |
514 | } else { | |
515 | if (!(used & CEPH_CAP_FILE_WR) && | |
516 | ci->i_wrbuffer_ref_head == 0) { | |
517 | dout("queue_cap_snap %p " | |
518 | "no new_snap|dirty_page|writing\n", inode); | |
519 | goto update_snapc; | |
520 | } | |
521 | } | |
522 | ||
523 | dout("queue_cap_snap %p cap_snap %p queuing under %p %s %s\n", | |
524 | inode, capsnap, old_snapc, ceph_cap_string(dirty), | |
525 | capsnap->need_flush ? "" : "no_flush"); | |
5dda377c YZ |
526 | ihold(inode); |
527 | ||
805692d0 | 528 | refcount_set(&capsnap->nref, 1); |
5dda377c | 529 | INIT_LIST_HEAD(&capsnap->ci_item); |
5dda377c YZ |
530 | |
531 | capsnap->follows = old_snapc->seq; | |
532 | capsnap->issued = __ceph_caps_issued(ci, NULL); | |
533 | capsnap->dirty = dirty; | |
534 | ||
535 | capsnap->mode = inode->i_mode; | |
536 | capsnap->uid = inode->i_uid; | |
537 | capsnap->gid = inode->i_gid; | |
538 | ||
539 | if (dirty & CEPH_CAP_XATTR_EXCL) { | |
540 | __ceph_build_xattrs_blob(ci); | |
541 | capsnap->xattr_blob = | |
542 | ceph_buffer_get(ci->i_xattrs.blob); | |
543 | capsnap->xattr_version = ci->i_xattrs.version; | |
963b61eb | 544 | } else { |
5dda377c YZ |
545 | capsnap->xattr_blob = NULL; |
546 | capsnap->xattr_version = 0; | |
963b61eb SW |
547 | } |
548 | ||
5dda377c YZ |
549 | capsnap->inline_data = ci->i_inline_version != CEPH_INLINE_NONE; |
550 | ||
551 | /* dirty page count moved from _head to this cap_snap; | |
552 | all subsequent writes page dirties occur _after_ this | |
553 | snapshot. */ | |
554 | capsnap->dirty_pages = ci->i_wrbuffer_ref_head; | |
555 | ci->i_wrbuffer_ref_head = 0; | |
556 | capsnap->context = old_snapc; | |
557 | list_add_tail(&capsnap->ci_item, &ci->i_cap_snaps); | |
5dda377c YZ |
558 | |
559 | if (used & CEPH_CAP_FILE_WR) { | |
560 | dout("queue_cap_snap %p cap_snap %p snapc %p" | |
561 | " seq %llu used WR, now pending\n", inode, | |
562 | capsnap, old_snapc, old_snapc->seq); | |
563 | capsnap->writing = 1; | |
564 | } else { | |
565 | /* note mtime, size NOW. */ | |
566 | __ceph_finish_cap_snap(ci, capsnap); | |
567 | } | |
568 | capsnap = NULL; | |
fce85157 | 569 | old_snapc = NULL; |
5dda377c YZ |
570 | |
571 | update_snapc: | |
572 | if (ci->i_head_snapc) { | |
86056090 YZ |
573 | ci->i_head_snapc = ceph_get_snap_context(new_snapc); |
574 | dout(" new snapc is %p\n", new_snapc); | |
5dda377c | 575 | } |
be655596 | 576 | spin_unlock(&ci->i_ceph_lock); |
5dda377c YZ |
577 | |
578 | kfree(capsnap); | |
579 | ceph_put_snap_context(old_snapc); | |
963b61eb SW |
580 | } |
581 | ||
582 | /* | |
583 | * Finalize the size, mtime for a cap_snap.. that is, settle on final values | |
584 | * to be used for the snapshot, to be flushed back to the mds. | |
585 | * | |
586 | * If capsnap can now be flushed, add to snap_flush list, and return 1. | |
587 | * | |
be655596 | 588 | * Caller must hold i_ceph_lock. |
963b61eb SW |
589 | */ |
590 | int __ceph_finish_cap_snap(struct ceph_inode_info *ci, | |
591 | struct ceph_cap_snap *capsnap) | |
592 | { | |
593 | struct inode *inode = &ci->vfs_inode; | |
3d14c5d2 | 594 | struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; |
963b61eb SW |
595 | |
596 | BUG_ON(capsnap->writing); | |
597 | capsnap->size = inode->i_size; | |
598 | capsnap->mtime = inode->i_mtime; | |
599 | capsnap->atime = inode->i_atime; | |
600 | capsnap->ctime = inode->i_ctime; | |
601 | capsnap->time_warp_seq = ci->i_time_warp_seq; | |
5f743e45 YZ |
602 | capsnap->truncate_size = ci->i_truncate_size; |
603 | capsnap->truncate_seq = ci->i_truncate_seq; | |
963b61eb | 604 | if (capsnap->dirty_pages) { |
819ccbfa | 605 | dout("finish_cap_snap %p cap_snap %p snapc %p %llu %s s=%llu " |
963b61eb SW |
606 | "still has %d dirty pages\n", inode, capsnap, |
607 | capsnap->context, capsnap->context->seq, | |
819ccbfa SW |
608 | ceph_cap_string(capsnap->dirty), capsnap->size, |
609 | capsnap->dirty_pages); | |
963b61eb SW |
610 | return 0; |
611 | } | |
70220ac8 YZ |
612 | |
613 | ci->i_ceph_flags |= CEPH_I_FLUSH_SNAPS; | |
819ccbfa | 614 | dout("finish_cap_snap %p cap_snap %p snapc %p %llu %s s=%llu\n", |
963b61eb | 615 | inode, capsnap, capsnap->context, |
819ccbfa SW |
616 | capsnap->context->seq, ceph_cap_string(capsnap->dirty), |
617 | capsnap->size); | |
963b61eb SW |
618 | |
619 | spin_lock(&mdsc->snap_flush_lock); | |
620 | list_add_tail(&ci->i_snap_flush_item, &mdsc->snap_flush_list); | |
621 | spin_unlock(&mdsc->snap_flush_lock); | |
622 | return 1; /* caller may want to ceph_flush_snaps */ | |
623 | } | |
624 | ||
ed326044 SW |
625 | /* |
626 | * Queue cap_snaps for snap writeback for this realm and its children. | |
627 | * Called under snap_rwsem, so realm topology won't change. | |
628 | */ | |
629 | static void queue_realm_cap_snaps(struct ceph_snap_realm *realm) | |
630 | { | |
631 | struct ceph_inode_info *ci; | |
632 | struct inode *lastinode = NULL; | |
ed326044 SW |
633 | |
634 | dout("queue_realm_cap_snaps %p %llx inodes\n", realm, realm->ino); | |
635 | ||
636 | spin_lock(&realm->inodes_with_caps_lock); | |
3ae0bebc | 637 | list_for_each_entry(ci, &realm->inodes_with_caps, i_snap_realm_item) { |
ed326044 SW |
638 | struct inode *inode = igrab(&ci->vfs_inode); |
639 | if (!inode) | |
640 | continue; | |
641 | spin_unlock(&realm->inodes_with_caps_lock); | |
e96a650a | 642 | iput(lastinode); |
ed326044 SW |
643 | lastinode = inode; |
644 | ceph_queue_cap_snap(ci); | |
645 | spin_lock(&realm->inodes_with_caps_lock); | |
646 | } | |
647 | spin_unlock(&realm->inodes_with_caps_lock); | |
e96a650a | 648 | iput(lastinode); |
ed326044 | 649 | |
ed326044 SW |
650 | dout("queue_realm_cap_snaps %p %llx done\n", realm, realm->ino); |
651 | } | |
963b61eb SW |
652 | |
653 | /* | |
654 | * Parse and apply a snapblob "snap trace" from the MDS. This specifies | |
655 | * the snap realm parameters from a given realm and all of its ancestors, | |
656 | * up to the root. | |
657 | * | |
658 | * Caller must hold snap_rwsem for write. | |
659 | */ | |
660 | int ceph_update_snap_trace(struct ceph_mds_client *mdsc, | |
982d6011 YZ |
661 | void *p, void *e, bool deletion, |
662 | struct ceph_snap_realm **realm_ret) | |
963b61eb SW |
663 | { |
664 | struct ceph_mds_snap_realm *ri; /* encoded */ | |
665 | __le64 *snaps; /* encoded */ | |
666 | __le64 *prior_parent_snaps; /* encoded */ | |
982d6011 YZ |
667 | struct ceph_snap_realm *realm = NULL; |
668 | struct ceph_snap_realm *first_realm = NULL; | |
963b61eb SW |
669 | int invalidate = 0; |
670 | int err = -ENOMEM; | |
ae00d4f3 | 671 | LIST_HEAD(dirty_realms); |
963b61eb SW |
672 | |
673 | dout("update_snap_trace deletion=%d\n", deletion); | |
674 | more: | |
675 | ceph_decode_need(&p, e, sizeof(*ri), bad); | |
676 | ri = p; | |
677 | p += sizeof(*ri); | |
678 | ceph_decode_need(&p, e, sizeof(u64)*(le32_to_cpu(ri->num_snaps) + | |
679 | le32_to_cpu(ri->num_prior_parent_snaps)), bad); | |
680 | snaps = p; | |
681 | p += sizeof(u64) * le32_to_cpu(ri->num_snaps); | |
682 | prior_parent_snaps = p; | |
683 | p += sizeof(u64) * le32_to_cpu(ri->num_prior_parent_snaps); | |
684 | ||
685 | realm = ceph_lookup_snap_realm(mdsc, le64_to_cpu(ri->ino)); | |
963b61eb SW |
686 | if (!realm) { |
687 | realm = ceph_create_snap_realm(mdsc, le64_to_cpu(ri->ino)); | |
688 | if (IS_ERR(realm)) { | |
689 | err = PTR_ERR(realm); | |
690 | goto fail; | |
691 | } | |
692 | } | |
693 | ||
963b61eb SW |
694 | /* ensure the parent is correct */ |
695 | err = adjust_snap_realm_parent(mdsc, realm, le64_to_cpu(ri->parent)); | |
696 | if (err < 0) | |
697 | goto fail; | |
698 | invalidate += err; | |
699 | ||
700 | if (le64_to_cpu(ri->seq) > realm->seq) { | |
ae00d4f3 SW |
701 | dout("update_snap_trace updating %llx %p %lld -> %lld\n", |
702 | realm->ino, realm, realm->seq, le64_to_cpu(ri->seq)); | |
963b61eb SW |
703 | /* update realm parameters, snap lists */ |
704 | realm->seq = le64_to_cpu(ri->seq); | |
705 | realm->created = le64_to_cpu(ri->created); | |
706 | realm->parent_since = le64_to_cpu(ri->parent_since); | |
707 | ||
708 | realm->num_snaps = le32_to_cpu(ri->num_snaps); | |
709 | err = dup_array(&realm->snaps, snaps, realm->num_snaps); | |
710 | if (err < 0) | |
711 | goto fail; | |
712 | ||
713 | realm->num_prior_parent_snaps = | |
714 | le32_to_cpu(ri->num_prior_parent_snaps); | |
715 | err = dup_array(&realm->prior_parent_snaps, prior_parent_snaps, | |
716 | realm->num_prior_parent_snaps); | |
717 | if (err < 0) | |
718 | goto fail; | |
719 | ||
affbc19a YZ |
720 | if (realm->seq > mdsc->last_snap_seq) |
721 | mdsc->last_snap_seq = realm->seq; | |
ae00d4f3 | 722 | |
963b61eb SW |
723 | invalidate = 1; |
724 | } else if (!realm->cached_context) { | |
ae00d4f3 SW |
725 | dout("update_snap_trace %llx %p seq %lld new\n", |
726 | realm->ino, realm, realm->seq); | |
963b61eb | 727 | invalidate = 1; |
ae00d4f3 SW |
728 | } else { |
729 | dout("update_snap_trace %llx %p seq %lld unchanged\n", | |
730 | realm->ino, realm, realm->seq); | |
963b61eb SW |
731 | } |
732 | ||
733 | dout("done with %llx %p, invalidated=%d, %p %p\n", realm->ino, | |
734 | realm, invalidate, p, e); | |
735 | ||
963b61eb | 736 | /* invalidate when we reach the _end_ (root) of the trace */ |
982d6011 | 737 | if (invalidate && p >= e) |
3ae0bebc | 738 | rebuild_snap_realms(realm, &dirty_realms); |
963b61eb | 739 | |
982d6011 YZ |
740 | if (!first_realm) |
741 | first_realm = realm; | |
742 | else | |
743 | ceph_put_snap_realm(mdsc, realm); | |
744 | ||
745 | if (p < e) | |
746 | goto more; | |
747 | ||
ae00d4f3 SW |
748 | /* |
749 | * queue cap snaps _after_ we've built the new snap contexts, | |
750 | * so that i_head_snapc can be set appropriately. | |
751 | */ | |
e8e1ba96 SW |
752 | while (!list_empty(&dirty_realms)) { |
753 | realm = list_first_entry(&dirty_realms, struct ceph_snap_realm, | |
754 | dirty_item); | |
3ae0bebc | 755 | list_del_init(&realm->dirty_item); |
ae00d4f3 SW |
756 | queue_realm_cap_snaps(realm); |
757 | } | |
758 | ||
982d6011 YZ |
759 | if (realm_ret) |
760 | *realm_ret = first_realm; | |
761 | else | |
762 | ceph_put_snap_realm(mdsc, first_realm); | |
763 | ||
963b61eb SW |
764 | __cleanup_empty_realms(mdsc); |
765 | return 0; | |
766 | ||
767 | bad: | |
768 | err = -EINVAL; | |
769 | fail: | |
982d6011 YZ |
770 | if (realm && !IS_ERR(realm)) |
771 | ceph_put_snap_realm(mdsc, realm); | |
772 | if (first_realm) | |
773 | ceph_put_snap_realm(mdsc, first_realm); | |
963b61eb SW |
774 | pr_err("update_snap_trace error %d\n", err); |
775 | return err; | |
776 | } | |
777 | ||
778 | ||
779 | /* | |
780 | * Send any cap_snaps that are queued for flush. Try to carry | |
781 | * s_mutex across multiple snap flushes to avoid locking overhead. | |
782 | * | |
783 | * Caller holds no locks. | |
784 | */ | |
785 | static void flush_snaps(struct ceph_mds_client *mdsc) | |
786 | { | |
787 | struct ceph_inode_info *ci; | |
788 | struct inode *inode; | |
789 | struct ceph_mds_session *session = NULL; | |
790 | ||
791 | dout("flush_snaps\n"); | |
792 | spin_lock(&mdsc->snap_flush_lock); | |
793 | while (!list_empty(&mdsc->snap_flush_list)) { | |
794 | ci = list_first_entry(&mdsc->snap_flush_list, | |
795 | struct ceph_inode_info, i_snap_flush_item); | |
796 | inode = &ci->vfs_inode; | |
70b666c3 | 797 | ihold(inode); |
963b61eb | 798 | spin_unlock(&mdsc->snap_flush_lock); |
ed9b430c | 799 | ceph_flush_snaps(ci, &session); |
963b61eb SW |
800 | iput(inode); |
801 | spin_lock(&mdsc->snap_flush_lock); | |
802 | } | |
803 | spin_unlock(&mdsc->snap_flush_lock); | |
804 | ||
805 | if (session) { | |
806 | mutex_unlock(&session->s_mutex); | |
807 | ceph_put_mds_session(session); | |
808 | } | |
809 | dout("flush_snaps done\n"); | |
810 | } | |
811 | ||
812 | ||
813 | /* | |
814 | * Handle a snap notification from the MDS. | |
815 | * | |
816 | * This can take two basic forms: the simplest is just a snap creation | |
817 | * or deletion notification on an existing realm. This should update the | |
818 | * realm and its children. | |
819 | * | |
820 | * The more difficult case is realm creation, due to snap creation at a | |
821 | * new point in the file hierarchy, or due to a rename that moves a file or | |
822 | * directory into another realm. | |
823 | */ | |
824 | void ceph_handle_snap(struct ceph_mds_client *mdsc, | |
2600d2dd | 825 | struct ceph_mds_session *session, |
963b61eb SW |
826 | struct ceph_msg *msg) |
827 | { | |
3d14c5d2 | 828 | struct super_block *sb = mdsc->fsc->sb; |
2600d2dd | 829 | int mds = session->s_mds; |
963b61eb SW |
830 | u64 split; |
831 | int op; | |
832 | int trace_len; | |
833 | struct ceph_snap_realm *realm = NULL; | |
834 | void *p = msg->front.iov_base; | |
835 | void *e = p + msg->front.iov_len; | |
836 | struct ceph_mds_snap_head *h; | |
837 | int num_split_inos, num_split_realms; | |
838 | __le64 *split_inos = NULL, *split_realms = NULL; | |
839 | int i; | |
840 | int locked_rwsem = 0; | |
841 | ||
963b61eb SW |
842 | /* decode */ |
843 | if (msg->front.iov_len < sizeof(*h)) | |
844 | goto bad; | |
845 | h = p; | |
846 | op = le32_to_cpu(h->op); | |
847 | split = le64_to_cpu(h->split); /* non-zero if we are splitting an | |
848 | * existing realm */ | |
849 | num_split_inos = le32_to_cpu(h->num_split_inos); | |
850 | num_split_realms = le32_to_cpu(h->num_split_realms); | |
851 | trace_len = le32_to_cpu(h->trace_len); | |
852 | p += sizeof(*h); | |
853 | ||
854 | dout("handle_snap from mds%d op %s split %llx tracelen %d\n", mds, | |
855 | ceph_snap_op_name(op), split, trace_len); | |
856 | ||
963b61eb SW |
857 | mutex_lock(&session->s_mutex); |
858 | session->s_seq++; | |
859 | mutex_unlock(&session->s_mutex); | |
860 | ||
861 | down_write(&mdsc->snap_rwsem); | |
862 | locked_rwsem = 1; | |
863 | ||
864 | if (op == CEPH_SNAP_OP_SPLIT) { | |
865 | struct ceph_mds_snap_realm *ri; | |
866 | ||
867 | /* | |
868 | * A "split" breaks part of an existing realm off into | |
869 | * a new realm. The MDS provides a list of inodes | |
870 | * (with caps) and child realms that belong to the new | |
871 | * child. | |
872 | */ | |
873 | split_inos = p; | |
874 | p += sizeof(u64) * num_split_inos; | |
875 | split_realms = p; | |
876 | p += sizeof(u64) * num_split_realms; | |
877 | ceph_decode_need(&p, e, sizeof(*ri), bad); | |
878 | /* we will peek at realm info here, but will _not_ | |
879 | * advance p, as the realm update will occur below in | |
880 | * ceph_update_snap_trace. */ | |
881 | ri = p; | |
882 | ||
883 | realm = ceph_lookup_snap_realm(mdsc, split); | |
963b61eb SW |
884 | if (!realm) { |
885 | realm = ceph_create_snap_realm(mdsc, split); | |
886 | if (IS_ERR(realm)) | |
887 | goto out; | |
888 | } | |
963b61eb SW |
889 | |
890 | dout("splitting snap_realm %llx %p\n", realm->ino, realm); | |
891 | for (i = 0; i < num_split_inos; i++) { | |
892 | struct ceph_vino vino = { | |
893 | .ino = le64_to_cpu(split_inos[i]), | |
894 | .snap = CEPH_NOSNAP, | |
895 | }; | |
896 | struct inode *inode = ceph_find_inode(sb, vino); | |
897 | struct ceph_inode_info *ci; | |
ae00d4f3 | 898 | struct ceph_snap_realm *oldrealm; |
963b61eb SW |
899 | |
900 | if (!inode) | |
901 | continue; | |
902 | ci = ceph_inode(inode); | |
903 | ||
be655596 | 904 | spin_lock(&ci->i_ceph_lock); |
963b61eb SW |
905 | if (!ci->i_snap_realm) |
906 | goto skip_inode; | |
907 | /* | |
908 | * If this inode belongs to a realm that was | |
909 | * created after our new realm, we experienced | |
910 | * a race (due to another split notifications | |
911 | * arriving from a different MDS). So skip | |
912 | * this inode. | |
913 | */ | |
914 | if (ci->i_snap_realm->created > | |
915 | le64_to_cpu(ri->created)) { | |
916 | dout(" leaving %p in newer realm %llx %p\n", | |
917 | inode, ci->i_snap_realm->ino, | |
918 | ci->i_snap_realm); | |
919 | goto skip_inode; | |
920 | } | |
921 | dout(" will move %p to split realm %llx %p\n", | |
922 | inode, realm->ino, realm); | |
923 | /* | |
ae00d4f3 | 924 | * Move the inode to the new realm |
963b61eb | 925 | */ |
052bb34a | 926 | spin_lock(&realm->inodes_with_caps_lock); |
963b61eb | 927 | list_del_init(&ci->i_snap_realm_item); |
ae00d4f3 SW |
928 | list_add(&ci->i_snap_realm_item, |
929 | &realm->inodes_with_caps); | |
930 | oldrealm = ci->i_snap_realm; | |
931 | ci->i_snap_realm = realm; | |
052bb34a | 932 | spin_unlock(&realm->inodes_with_caps_lock); |
be655596 | 933 | spin_unlock(&ci->i_ceph_lock); |
963b61eb | 934 | |
ae00d4f3 SW |
935 | ceph_get_snap_realm(mdsc, realm); |
936 | ceph_put_snap_realm(mdsc, oldrealm); | |
963b61eb SW |
937 | |
938 | iput(inode); | |
939 | continue; | |
940 | ||
941 | skip_inode: | |
be655596 | 942 | spin_unlock(&ci->i_ceph_lock); |
963b61eb SW |
943 | iput(inode); |
944 | } | |
945 | ||
946 | /* we may have taken some of the old realm's children. */ | |
947 | for (i = 0; i < num_split_realms; i++) { | |
948 | struct ceph_snap_realm *child = | |
982d6011 | 949 | __lookup_snap_realm(mdsc, |
963b61eb | 950 | le64_to_cpu(split_realms[i])); |
963b61eb SW |
951 | if (!child) |
952 | continue; | |
953 | adjust_snap_realm_parent(mdsc, child, realm->ino); | |
954 | } | |
955 | } | |
956 | ||
957 | /* | |
958 | * update using the provided snap trace. if we are deleting a | |
959 | * snap, we can avoid queueing cap_snaps. | |
960 | */ | |
961 | ceph_update_snap_trace(mdsc, p, e, | |
982d6011 | 962 | op == CEPH_SNAP_OP_DESTROY, NULL); |
963b61eb | 963 | |
ae00d4f3 | 964 | if (op == CEPH_SNAP_OP_SPLIT) |
963b61eb SW |
965 | /* we took a reference when we created the realm, above */ |
966 | ceph_put_snap_realm(mdsc, realm); | |
963b61eb SW |
967 | |
968 | __cleanup_empty_realms(mdsc); | |
969 | ||
970 | up_write(&mdsc->snap_rwsem); | |
971 | ||
972 | flush_snaps(mdsc); | |
973 | return; | |
974 | ||
975 | bad: | |
976 | pr_err("corrupt snap message from mds%d\n", mds); | |
9ec7cab1 | 977 | ceph_msg_dump(msg); |
963b61eb SW |
978 | out: |
979 | if (locked_rwsem) | |
980 | up_write(&mdsc->snap_rwsem); | |
981 | return; | |
982 | } |