blob: 4aa7122a8d38c18dd4fe7443fb64f46765b83280 [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
Yan, Zheng7b06a822015-05-01 10:03:40 +0800299struct ceph_snap_context *ceph_empty_snapc;
Yan, Zheng97c85a82014-11-06 15:09:41 +0800300
Sage Weil963b61e2009-10-06 11:31:12 -0700301/*
302 * build the snap context for a given realm.
303 */
304static int build_snap_context(struct ceph_snap_realm *realm)
305{
306 struct ceph_snap_realm *parent = realm->parent;
307 struct ceph_snap_context *snapc;
308 int err = 0;
Alex Elderaa711ee2012-07-13 20:35:11 -0500309 u32 num = realm->num_prior_parent_snaps + realm->num_snaps;
Sage Weil963b61e2009-10-06 11:31:12 -0700310
311 /*
312 * build parent context, if it hasn't been built.
313 * conservatively estimate that all parent snaps might be
314 * included by us.
315 */
316 if (parent) {
317 if (!parent->cached_context) {
318 err = build_snap_context(parent);
319 if (err)
320 goto fail;
321 }
322 num += parent->cached_context->num_snaps;
323 }
324
325 /* do i actually need to update? not if my context seq
326 matches realm seq, and my parents' does to. (this works
327 because we rebuild_snap_realms() works _downward_ in
328 hierarchy after each update.) */
329 if (realm->cached_context &&
Sage Weilec4318bc2010-03-19 13:24:39 -0700330 realm->cached_context->seq == realm->seq &&
Sage Weil963b61e2009-10-06 11:31:12 -0700331 (!parent ||
Sage Weilec4318bc2010-03-19 13:24:39 -0700332 realm->cached_context->seq >= parent->cached_context->seq)) {
Alex Elderaa711ee2012-07-13 20:35:11 -0500333 dout("build_snap_context %llx %p: %p seq %lld (%u snaps)"
Sage Weil963b61e2009-10-06 11:31:12 -0700334 " (unchanged)\n",
335 realm->ino, realm, realm->cached_context,
336 realm->cached_context->seq,
Alex Elderaa711ee2012-07-13 20:35:11 -0500337 (unsigned int) realm->cached_context->num_snaps);
Sage Weil963b61e2009-10-06 11:31:12 -0700338 return 0;
339 }
340
341 /* alloc new snap context */
342 err = -ENOMEM;
Xi Wanga3860c12012-05-31 16:26:04 -0700343 if (num > (SIZE_MAX - sizeof(*snapc)) / sizeof(u64))
Sage Weil963b61e2009-10-06 11:31:12 -0700344 goto fail;
Alex Elder812164f82013-04-30 00:44:32 -0500345 snapc = ceph_create_snap_context(num, GFP_NOFS);
Sage Weil963b61e2009-10-06 11:31:12 -0700346 if (!snapc)
347 goto fail;
Sage Weil963b61e2009-10-06 11:31:12 -0700348
349 /* build (reverse sorted) snap vector */
350 num = 0;
351 snapc->seq = realm->seq;
352 if (parent) {
Alex Elderaa711ee2012-07-13 20:35:11 -0500353 u32 i;
354
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300355 /* include any of parent's snaps occurring _after_ my
Sage Weil963b61e2009-10-06 11:31:12 -0700356 parent became my parent */
357 for (i = 0; i < parent->cached_context->num_snaps; i++)
358 if (parent->cached_context->snaps[i] >=
359 realm->parent_since)
360 snapc->snaps[num++] =
361 parent->cached_context->snaps[i];
362 if (parent->cached_context->seq > snapc->seq)
363 snapc->seq = parent->cached_context->seq;
364 }
365 memcpy(snapc->snaps + num, realm->snaps,
366 sizeof(u64)*realm->num_snaps);
367 num += realm->num_snaps;
368 memcpy(snapc->snaps + num, realm->prior_parent_snaps,
369 sizeof(u64)*realm->num_prior_parent_snaps);
370 num += realm->num_prior_parent_snaps;
371
372 sort(snapc->snaps, num, sizeof(u64), cmpu64_rev, NULL);
373 snapc->num_snaps = num;
Alex Elderaa711ee2012-07-13 20:35:11 -0500374 dout("build_snap_context %llx %p: %p seq %lld (%u snaps)\n",
375 realm->ino, realm, snapc, snapc->seq,
376 (unsigned int) snapc->num_snaps);
Sage Weil963b61e2009-10-06 11:31:12 -0700377
SF Markus Elfringe96a6502014-11-02 15:20:59 +0100378 ceph_put_snap_context(realm->cached_context);
Sage Weil963b61e2009-10-06 11:31:12 -0700379 realm->cached_context = snapc;
380 return 0;
381
382fail:
383 /*
384 * if we fail, clear old (incorrect) cached_context... hopefully
385 * we'll have better luck building it later
386 */
387 if (realm->cached_context) {
388 ceph_put_snap_context(realm->cached_context);
389 realm->cached_context = NULL;
390 }
391 pr_err("build_snap_context %llx %p fail %d\n", realm->ino,
392 realm, err);
393 return err;
394}
395
396/*
397 * rebuild snap context for the given realm and all of its children.
398 */
399static void rebuild_snap_realms(struct ceph_snap_realm *realm)
400{
401 struct ceph_snap_realm *child;
402
403 dout("rebuild_snap_realms %llx %p\n", realm->ino, realm);
404 build_snap_context(realm);
405
406 list_for_each_entry(child, &realm->children, child_item)
407 rebuild_snap_realms(child);
408}
409
410
411/*
412 * helper to allocate and decode an array of snapids. free prior
413 * instance, if any.
414 */
Alex Elderaa711ee2012-07-13 20:35:11 -0500415static int dup_array(u64 **dst, __le64 *src, u32 num)
Sage Weil963b61e2009-10-06 11:31:12 -0700416{
Alex Elderaa711ee2012-07-13 20:35:11 -0500417 u32 i;
Sage Weil963b61e2009-10-06 11:31:12 -0700418
419 kfree(*dst);
420 if (num) {
421 *dst = kcalloc(num, sizeof(u64), GFP_NOFS);
422 if (!*dst)
423 return -ENOMEM;
424 for (i = 0; i < num; i++)
425 (*dst)[i] = get_unaligned_le64(src + i);
426 } else {
427 *dst = NULL;
428 }
429 return 0;
430}
431
Yan, Zheng86056092015-05-01 16:57:16 +0800432static bool has_new_snaps(struct ceph_snap_context *o,
433 struct ceph_snap_context *n)
434{
435 if (n->num_snaps == 0)
436 return false;
437 /* snaps are in descending order */
438 return n->snaps[0] > o->seq;
439}
Sage Weil963b61e2009-10-06 11:31:12 -0700440
441/*
442 * When a snapshot is applied, the size/mtime inode metadata is queued
443 * in a ceph_cap_snap (one for each snapshot) until writeback
444 * completes and the metadata can be flushed back to the MDS.
445 *
446 * However, if a (sync) write is currently in-progress when we apply
447 * the snapshot, we have to wait until the write succeeds or fails
448 * (and a final size/mtime is known). In this case the
449 * cap_snap->writing = 1, and is said to be "pending." When the write
450 * finishes, we __ceph_finish_cap_snap().
451 *
452 * Caller must hold snap_rwsem for read (i.e., the realm topology won't
453 * change).
454 */
Sage Weilfc837c8f2010-04-13 11:41:22 -0700455void ceph_queue_cap_snap(struct ceph_inode_info *ci)
Sage Weil963b61e2009-10-06 11:31:12 -0700456{
457 struct inode *inode = &ci->vfs_inode;
458 struct ceph_cap_snap *capsnap;
Yan, Zheng86056092015-05-01 16:57:16 +0800459 struct ceph_snap_context *old_snapc, *new_snapc;
Sage Weil4a625be2010-08-22 15:03:56 -0700460 int used, dirty;
Sage Weil963b61e2009-10-06 11:31:12 -0700461
462 capsnap = kzalloc(sizeof(*capsnap), GFP_NOFS);
463 if (!capsnap) {
464 pr_err("ENOMEM allocating ceph_cap_snap on %p\n", inode);
465 return;
466 }
467
Sage Weilbe655592011-11-30 09:47:09 -0800468 spin_lock(&ci->i_ceph_lock);
Sage Weil963b61e2009-10-06 11:31:12 -0700469 used = __ceph_caps_used(ci);
Sage Weil4a625be2010-08-22 15:03:56 -0700470 dirty = __ceph_caps_dirty(ci);
Sage Weilaf0ed562011-07-26 11:26:31 -0700471
Yan, Zheng5dda377c2015-04-30 14:40:54 +0800472 old_snapc = ci->i_head_snapc;
Yan, Zheng86056092015-05-01 16:57:16 +0800473 new_snapc = ci->i_snap_realm->cached_context;
Yan, Zheng5dda377c2015-04-30 14:40:54 +0800474
Sage Weilaf0ed562011-07-26 11:26:31 -0700475 /*
476 * If there is a write in progress, treat that as a dirty Fw,
477 * even though it hasn't completed yet; by the time we finish
478 * up this capsnap it will be.
479 */
480 if (used & CEPH_CAP_FILE_WR)
481 dirty |= CEPH_CAP_FILE_WR;
482
Sage Weil963b61e2009-10-06 11:31:12 -0700483 if (__ceph_have_pending_cap_snap(ci)) {
484 /* there is no point in queuing multiple "pending" cap_snaps,
485 as no new writes are allowed to start when pending, so any
486 writes in progress now were started before the previous
487 cap_snap. lucky us. */
Sage Weilfc837c8f2010-04-13 11:41:22 -0700488 dout("queue_cap_snap %p already pending\n", inode);
Yan, Zheng5dda377c2015-04-30 14:40:54 +0800489 goto update_snapc;
490 }
Yan, Zheng86056092015-05-01 16:57:16 +0800491 if (ci->i_wrbuffer_ref_head == 0 &&
492 !(dirty & (CEPH_CAP_ANY_EXCL|CEPH_CAP_FILE_WR))) {
Sage Weil963b61e2009-10-06 11:31:12 -0700493 dout("queue_cap_snap %p nothing dirty|writing\n", inode);
Yan, Zheng5dda377c2015-04-30 14:40:54 +0800494 goto update_snapc;
Sage Weil963b61e2009-10-06 11:31:12 -0700495 }
496
Yan, Zheng5dda377c2015-04-30 14:40:54 +0800497 BUG_ON(!old_snapc);
498
Yan, Zheng86056092015-05-01 16:57:16 +0800499 /*
500 * There is no need to send FLUSHSNAP message to MDS if there is
501 * no new snapshot. But when there is dirty pages or on-going
502 * writes, we still need to create cap_snap. cap_snap is needed
503 * by the write path and page writeback path.
504 *
505 * also see ceph_try_drop_cap_snap()
506 */
507 if (has_new_snaps(old_snapc, new_snapc)) {
508 if (dirty & (CEPH_CAP_ANY_EXCL|CEPH_CAP_FILE_WR))
509 capsnap->need_flush = true;
510 } else {
511 if (!(used & CEPH_CAP_FILE_WR) &&
512 ci->i_wrbuffer_ref_head == 0) {
513 dout("queue_cap_snap %p "
514 "no new_snap|dirty_page|writing\n", inode);
515 goto update_snapc;
516 }
517 }
518
519 dout("queue_cap_snap %p cap_snap %p queuing under %p %s %s\n",
520 inode, capsnap, old_snapc, ceph_cap_string(dirty),
521 capsnap->need_flush ? "" : "no_flush");
Yan, Zheng5dda377c2015-04-30 14:40:54 +0800522 ihold(inode);
523
524 atomic_set(&capsnap->nref, 1);
525 capsnap->ci = ci;
526 INIT_LIST_HEAD(&capsnap->ci_item);
527 INIT_LIST_HEAD(&capsnap->flushing_item);
528
529 capsnap->follows = old_snapc->seq;
530 capsnap->issued = __ceph_caps_issued(ci, NULL);
531 capsnap->dirty = dirty;
532
533 capsnap->mode = inode->i_mode;
534 capsnap->uid = inode->i_uid;
535 capsnap->gid = inode->i_gid;
536
537 if (dirty & CEPH_CAP_XATTR_EXCL) {
538 __ceph_build_xattrs_blob(ci);
539 capsnap->xattr_blob =
540 ceph_buffer_get(ci->i_xattrs.blob);
541 capsnap->xattr_version = ci->i_xattrs.version;
542 } else {
543 capsnap->xattr_blob = NULL;
544 capsnap->xattr_version = 0;
545 }
546
547 capsnap->inline_data = ci->i_inline_version != CEPH_INLINE_NONE;
548
549 /* dirty page count moved from _head to this cap_snap;
550 all subsequent writes page dirties occur _after_ this
551 snapshot. */
552 capsnap->dirty_pages = ci->i_wrbuffer_ref_head;
553 ci->i_wrbuffer_ref_head = 0;
554 capsnap->context = old_snapc;
555 list_add_tail(&capsnap->ci_item, &ci->i_cap_snaps);
556 old_snapc = NULL;
557
558 if (used & CEPH_CAP_FILE_WR) {
559 dout("queue_cap_snap %p cap_snap %p snapc %p"
560 " seq %llu used WR, now pending\n", inode,
561 capsnap, old_snapc, old_snapc->seq);
562 capsnap->writing = 1;
563 } else {
564 /* note mtime, size NOW. */
565 __ceph_finish_cap_snap(ci, capsnap);
566 }
567 capsnap = NULL;
568
569update_snapc:
570 if (ci->i_head_snapc) {
Yan, Zheng86056092015-05-01 16:57:16 +0800571 ci->i_head_snapc = ceph_get_snap_context(new_snapc);
572 dout(" new snapc is %p\n", new_snapc);
Yan, Zheng5dda377c2015-04-30 14:40:54 +0800573 }
Sage Weilbe655592011-11-30 09:47:09 -0800574 spin_unlock(&ci->i_ceph_lock);
Yan, Zheng5dda377c2015-04-30 14:40:54 +0800575
576 kfree(capsnap);
577 ceph_put_snap_context(old_snapc);
Sage Weil963b61e2009-10-06 11:31:12 -0700578}
579
580/*
581 * Finalize the size, mtime for a cap_snap.. that is, settle on final values
582 * to be used for the snapshot, to be flushed back to the mds.
583 *
584 * If capsnap can now be flushed, add to snap_flush list, and return 1.
585 *
Sage Weilbe655592011-11-30 09:47:09 -0800586 * Caller must hold i_ceph_lock.
Sage Weil963b61e2009-10-06 11:31:12 -0700587 */
588int __ceph_finish_cap_snap(struct ceph_inode_info *ci,
589 struct ceph_cap_snap *capsnap)
590{
591 struct inode *inode = &ci->vfs_inode;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700592 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weil963b61e2009-10-06 11:31:12 -0700593
594 BUG_ON(capsnap->writing);
595 capsnap->size = inode->i_size;
596 capsnap->mtime = inode->i_mtime;
597 capsnap->atime = inode->i_atime;
598 capsnap->ctime = inode->i_ctime;
599 capsnap->time_warp_seq = ci->i_time_warp_seq;
600 if (capsnap->dirty_pages) {
Sage Weil819ccbf2010-04-01 09:33:46 -0700601 dout("finish_cap_snap %p cap_snap %p snapc %p %llu %s s=%llu "
Sage Weil963b61e2009-10-06 11:31:12 -0700602 "still has %d dirty pages\n", inode, capsnap,
603 capsnap->context, capsnap->context->seq,
Sage Weil819ccbf2010-04-01 09:33:46 -0700604 ceph_cap_string(capsnap->dirty), capsnap->size,
605 capsnap->dirty_pages);
Sage Weil963b61e2009-10-06 11:31:12 -0700606 return 0;
607 }
Sage Weil819ccbf2010-04-01 09:33:46 -0700608 dout("finish_cap_snap %p cap_snap %p snapc %p %llu %s s=%llu\n",
Sage Weil963b61e2009-10-06 11:31:12 -0700609 inode, capsnap, capsnap->context,
Sage Weil819ccbf2010-04-01 09:33:46 -0700610 capsnap->context->seq, ceph_cap_string(capsnap->dirty),
611 capsnap->size);
Sage Weil963b61e2009-10-06 11:31:12 -0700612
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
Sage Weiled326042010-08-16 13:37:31 -0700619/*
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);
SF Markus Elfringe96a6502014-11-02 15:20:59 +0100638 iput(lastinode);
Sage Weiled326042010-08-16 13:37:31 -0700639 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);
SF Markus Elfringe96a6502014-11-02 15:20:59 +0100644 iput(lastinode);
Sage Weiled326042010-08-16 13:37:31 -0700645
Sage Weile8e1ba962011-02-04 20:45:58 -0800646 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 }
Sage Weiled326042010-08-16 13:37:31 -0700652
Sage Weile8e1ba962011-02-04 20:45:58 -0800653 list_del_init(&realm->dirty_item);
Sage Weiled326042010-08-16 13:37:31 -0700654 dout("queue_realm_cap_snaps %p %llx done\n", realm, realm->ino);
655}
Sage Weil963b61e2009-10-06 11:31:12 -0700656
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,
Yan, Zheng982d6012014-12-23 15:30:54 +0800665 void *p, void *e, bool deletion,
666 struct ceph_snap_realm **realm_ret)
Sage Weil963b61e2009-10-06 11:31:12 -0700667{
668 struct ceph_mds_snap_realm *ri; /* encoded */
669 __le64 *snaps; /* encoded */
670 __le64 *prior_parent_snaps; /* encoded */
Yan, Zheng982d6012014-12-23 15:30:54 +0800671 struct ceph_snap_realm *realm = NULL;
672 struct ceph_snap_realm *first_realm = NULL;
Sage Weil963b61e2009-10-06 11:31:12 -0700673 int invalidate = 0;
674 int err = -ENOMEM;
Sage Weilae00d4f2010-09-16 16:26:51 -0700675 LIST_HEAD(dirty_realms);
Sage Weil963b61e2009-10-06 11:31:12 -0700676
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));
Sage Weil963b61e2009-10-06 11:31:12 -0700690 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
Sage Weil963b61e2009-10-06 11:31:12 -0700698 /* 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) {
Sage Weilae00d4f2010-09-16 16:26:51 -0700705 dout("update_snap_trace updating %llx %p %lld -> %lld\n",
706 realm->ino, realm, realm->seq, le64_to_cpu(ri->seq));
Sage Weil963b61e2009-10-06 11:31:12 -0700707 /* 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
Sage Weilae00d4f2010-09-16 16:26:51 -0700724 /* queue realm for cap_snap creation */
725 list_add(&realm->dirty_item, &dirty_realms);
Yan, Zhengaffbc192015-05-05 21:22:13 +0800726 if (realm->seq > mdsc->last_snap_seq)
727 mdsc->last_snap_seq = realm->seq;
Sage Weilae00d4f2010-09-16 16:26:51 -0700728
Sage Weil963b61e2009-10-06 11:31:12 -0700729 invalidate = 1;
730 } else if (!realm->cached_context) {
Sage Weilae00d4f2010-09-16 16:26:51 -0700731 dout("update_snap_trace %llx %p seq %lld new\n",
732 realm->ino, realm, realm->seq);
Sage Weil963b61e2009-10-06 11:31:12 -0700733 invalidate = 1;
Sage Weilae00d4f2010-09-16 16:26:51 -0700734 } else {
735 dout("update_snap_trace %llx %p seq %lld unchanged\n",
736 realm->ino, realm, realm->seq);
Sage Weil963b61e2009-10-06 11:31:12 -0700737 }
738
739 dout("done with %llx %p, invalidated=%d, %p %p\n", realm->ino,
740 realm, invalidate, p, e);
741
Yan, Zheng982d6012014-12-23 15:30:54 +0800742 /* invalidate when we reach the _end_ (root) of the trace */
743 if (invalidate && p >= e)
744 rebuild_snap_realms(realm);
745
746 if (!first_realm)
747 first_realm = realm;
748 else
749 ceph_put_snap_realm(mdsc, realm);
750
Sage Weil963b61e2009-10-06 11:31:12 -0700751 if (p < e)
752 goto more;
753
Sage Weilae00d4f2010-09-16 16:26:51 -0700754 /*
755 * queue cap snaps _after_ we've built the new snap contexts,
756 * so that i_head_snapc can be set appropriately.
757 */
Sage Weile8e1ba962011-02-04 20:45:58 -0800758 while (!list_empty(&dirty_realms)) {
759 realm = list_first_entry(&dirty_realms, struct ceph_snap_realm,
760 dirty_item);
Sage Weilae00d4f2010-09-16 16:26:51 -0700761 queue_realm_cap_snaps(realm);
762 }
763
Yan, Zheng982d6012014-12-23 15:30:54 +0800764 if (realm_ret)
765 *realm_ret = first_realm;
766 else
767 ceph_put_snap_realm(mdsc, first_realm);
768
Sage Weil963b61e2009-10-06 11:31:12 -0700769 __cleanup_empty_realms(mdsc);
770 return 0;
771
772bad:
773 err = -EINVAL;
774fail:
Yan, Zheng982d6012014-12-23 15:30:54 +0800775 if (realm && !IS_ERR(realm))
776 ceph_put_snap_realm(mdsc, realm);
777 if (first_realm)
778 ceph_put_snap_realm(mdsc, first_realm);
Sage Weil963b61e2009-10-06 11:31:12 -0700779 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;
Sage Weil70b666c2011-05-27 09:24:26 -0700802 ihold(inode);
Sage Weil963b61e2009-10-06 11:31:12 -0700803 spin_unlock(&mdsc->snap_flush_lock);
Sage Weilbe655592011-11-30 09:47:09 -0800804 spin_lock(&ci->i_ceph_lock);
Sage Weile8351242010-09-17 08:03:08 -0700805 __ceph_flush_snaps(ci, &session, 0);
Sage Weilbe655592011-11-30 09:47:09 -0800806 spin_unlock(&ci->i_ceph_lock);
Sage Weil963b61e2009-10-06 11:31:12 -0700807 iput(inode);
808 spin_lock(&mdsc->snap_flush_lock);
809 }
810 spin_unlock(&mdsc->snap_flush_lock);
811
812 if (session) {
813 mutex_unlock(&session->s_mutex);
814 ceph_put_mds_session(session);
815 }
816 dout("flush_snaps done\n");
817}
818
819
820/*
821 * Handle a snap notification from the MDS.
822 *
823 * This can take two basic forms: the simplest is just a snap creation
824 * or deletion notification on an existing realm. This should update the
825 * realm and its children.
826 *
827 * The more difficult case is realm creation, due to snap creation at a
828 * new point in the file hierarchy, or due to a rename that moves a file or
829 * directory into another realm.
830 */
831void ceph_handle_snap(struct ceph_mds_client *mdsc,
Sage Weil2600d2d2010-02-22 15:12:16 -0800832 struct ceph_mds_session *session,
Sage Weil963b61e2009-10-06 11:31:12 -0700833 struct ceph_msg *msg)
834{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700835 struct super_block *sb = mdsc->fsc->sb;
Sage Weil2600d2d2010-02-22 15:12:16 -0800836 int mds = session->s_mds;
Sage Weil963b61e2009-10-06 11:31:12 -0700837 u64 split;
838 int op;
839 int trace_len;
840 struct ceph_snap_realm *realm = NULL;
841 void *p = msg->front.iov_base;
842 void *e = p + msg->front.iov_len;
843 struct ceph_mds_snap_head *h;
844 int num_split_inos, num_split_realms;
845 __le64 *split_inos = NULL, *split_realms = NULL;
846 int i;
847 int locked_rwsem = 0;
848
Sage Weil963b61e2009-10-06 11:31:12 -0700849 /* decode */
850 if (msg->front.iov_len < sizeof(*h))
851 goto bad;
852 h = p;
853 op = le32_to_cpu(h->op);
854 split = le64_to_cpu(h->split); /* non-zero if we are splitting an
855 * existing realm */
856 num_split_inos = le32_to_cpu(h->num_split_inos);
857 num_split_realms = le32_to_cpu(h->num_split_realms);
858 trace_len = le32_to_cpu(h->trace_len);
859 p += sizeof(*h);
860
861 dout("handle_snap from mds%d op %s split %llx tracelen %d\n", mds,
862 ceph_snap_op_name(op), split, trace_len);
863
Sage Weil963b61e2009-10-06 11:31:12 -0700864 mutex_lock(&session->s_mutex);
865 session->s_seq++;
866 mutex_unlock(&session->s_mutex);
867
868 down_write(&mdsc->snap_rwsem);
869 locked_rwsem = 1;
870
871 if (op == CEPH_SNAP_OP_SPLIT) {
872 struct ceph_mds_snap_realm *ri;
873
874 /*
875 * A "split" breaks part of an existing realm off into
876 * a new realm. The MDS provides a list of inodes
877 * (with caps) and child realms that belong to the new
878 * child.
879 */
880 split_inos = p;
881 p += sizeof(u64) * num_split_inos;
882 split_realms = p;
883 p += sizeof(u64) * num_split_realms;
884 ceph_decode_need(&p, e, sizeof(*ri), bad);
885 /* we will peek at realm info here, but will _not_
886 * advance p, as the realm update will occur below in
887 * ceph_update_snap_trace. */
888 ri = p;
889
890 realm = ceph_lookup_snap_realm(mdsc, split);
Sage Weil963b61e2009-10-06 11:31:12 -0700891 if (!realm) {
892 realm = ceph_create_snap_realm(mdsc, split);
893 if (IS_ERR(realm))
894 goto out;
895 }
Sage Weil963b61e2009-10-06 11:31:12 -0700896
897 dout("splitting snap_realm %llx %p\n", realm->ino, realm);
898 for (i = 0; i < num_split_inos; i++) {
899 struct ceph_vino vino = {
900 .ino = le64_to_cpu(split_inos[i]),
901 .snap = CEPH_NOSNAP,
902 };
903 struct inode *inode = ceph_find_inode(sb, vino);
904 struct ceph_inode_info *ci;
Sage Weilae00d4f2010-09-16 16:26:51 -0700905 struct ceph_snap_realm *oldrealm;
Sage Weil963b61e2009-10-06 11:31:12 -0700906
907 if (!inode)
908 continue;
909 ci = ceph_inode(inode);
910
Sage Weilbe655592011-11-30 09:47:09 -0800911 spin_lock(&ci->i_ceph_lock);
Sage Weil963b61e2009-10-06 11:31:12 -0700912 if (!ci->i_snap_realm)
913 goto skip_inode;
914 /*
915 * If this inode belongs to a realm that was
916 * created after our new realm, we experienced
917 * a race (due to another split notifications
918 * arriving from a different MDS). So skip
919 * this inode.
920 */
921 if (ci->i_snap_realm->created >
922 le64_to_cpu(ri->created)) {
923 dout(" leaving %p in newer realm %llx %p\n",
924 inode, ci->i_snap_realm->ino,
925 ci->i_snap_realm);
926 goto skip_inode;
927 }
928 dout(" will move %p to split realm %llx %p\n",
929 inode, realm->ino, realm);
930 /*
Sage Weilae00d4f2010-09-16 16:26:51 -0700931 * Move the inode to the new realm
Sage Weil963b61e2009-10-06 11:31:12 -0700932 */
Sage Weil052bb342010-03-09 12:52:26 -0800933 spin_lock(&realm->inodes_with_caps_lock);
Sage Weil963b61e2009-10-06 11:31:12 -0700934 list_del_init(&ci->i_snap_realm_item);
Sage Weilae00d4f2010-09-16 16:26:51 -0700935 list_add(&ci->i_snap_realm_item,
936 &realm->inodes_with_caps);
937 oldrealm = ci->i_snap_realm;
938 ci->i_snap_realm = realm;
Sage Weil052bb342010-03-09 12:52:26 -0800939 spin_unlock(&realm->inodes_with_caps_lock);
Sage Weilbe655592011-11-30 09:47:09 -0800940 spin_unlock(&ci->i_ceph_lock);
Sage Weil963b61e2009-10-06 11:31:12 -0700941
Sage Weilae00d4f2010-09-16 16:26:51 -0700942 ceph_get_snap_realm(mdsc, realm);
943 ceph_put_snap_realm(mdsc, oldrealm);
Sage Weil963b61e2009-10-06 11:31:12 -0700944
945 iput(inode);
946 continue;
947
948skip_inode:
Sage Weilbe655592011-11-30 09:47:09 -0800949 spin_unlock(&ci->i_ceph_lock);
Sage Weil963b61e2009-10-06 11:31:12 -0700950 iput(inode);
951 }
952
953 /* we may have taken some of the old realm's children. */
954 for (i = 0; i < num_split_realms; i++) {
955 struct ceph_snap_realm *child =
Yan, Zheng982d6012014-12-23 15:30:54 +0800956 __lookup_snap_realm(mdsc,
Sage Weil963b61e2009-10-06 11:31:12 -0700957 le64_to_cpu(split_realms[i]));
Sage Weil963b61e2009-10-06 11:31:12 -0700958 if (!child)
959 continue;
960 adjust_snap_realm_parent(mdsc, child, realm->ino);
961 }
962 }
963
964 /*
965 * update using the provided snap trace. if we are deleting a
966 * snap, we can avoid queueing cap_snaps.
967 */
968 ceph_update_snap_trace(mdsc, p, e,
Yan, Zheng982d6012014-12-23 15:30:54 +0800969 op == CEPH_SNAP_OP_DESTROY, NULL);
Sage Weil963b61e2009-10-06 11:31:12 -0700970
Sage Weilae00d4f2010-09-16 16:26:51 -0700971 if (op == CEPH_SNAP_OP_SPLIT)
Sage Weil963b61e2009-10-06 11:31:12 -0700972 /* we took a reference when we created the realm, above */
973 ceph_put_snap_realm(mdsc, realm);
Sage Weil963b61e2009-10-06 11:31:12 -0700974
975 __cleanup_empty_realms(mdsc);
976
977 up_write(&mdsc->snap_rwsem);
978
979 flush_snaps(mdsc);
980 return;
981
982bad:
983 pr_err("corrupt snap message from mds%d\n", mds);
Sage Weil9ec7cab2009-12-14 15:13:47 -0800984 ceph_msg_dump(msg);
Sage Weil963b61e2009-10-06 11:31:12 -0700985out:
986 if (locked_rwsem)
987 up_write(&mdsc->snap_rwsem);
988 return;
989}
990
Yan, Zheng97c85a82014-11-06 15:09:41 +0800991int __init ceph_snap_init(void)
992{
Yan, Zheng7b06a822015-05-01 10:03:40 +0800993 ceph_empty_snapc = ceph_create_snap_context(0, GFP_NOFS);
994 if (!ceph_empty_snapc)
Yan, Zheng97c85a82014-11-06 15:09:41 +0800995 return -ENOMEM;
Yan, Zheng7b06a822015-05-01 10:03:40 +0800996 ceph_empty_snapc->seq = 1;
Yan, Zheng97c85a82014-11-06 15:09:41 +0800997 return 0;
998}
Sage Weil963b61e2009-10-06 11:31:12 -0700999
Yan, Zheng97c85a82014-11-06 15:09:41 +08001000void ceph_snap_exit(void)
1001{
Yan, Zheng7b06a822015-05-01 10:03:40 +08001002 ceph_put_snap_context(ceph_empty_snapc);
Yan, Zheng97c85a82014-11-06 15:09:41 +08001003}