~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/ceph/snap.c

Version: ~ [ linux-5.2 ] ~ [ linux-5.1.16 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.57 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.132 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.184 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.184 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.69 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.39.4 ] ~ [ linux-2.6.38.8 ] ~ [ linux-2.6.37.6 ] ~ [ linux-2.6.36.4 ] ~ [ linux-2.6.35.14 ] ~ [ linux-2.6.34.15 ] ~ [ linux-2.6.33.20 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

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

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp