1 // SPDX-License-Identifier: GPL-2.0 2 #include <linux/ceph/ceph_debug.h> 3 4 #include <linux/sort.h> 5 #include <linux/slab.h> 6 #include <linux/iversion.h> 7 #include "super.h" 8 #include "mds_client.h" 9 #include <linux/ceph/decode.h> 10 11 /* unused map expires after 5 minutes */ 12 #define CEPH_SNAPID_MAP_TIMEOUT (5 * 60 * HZ) 13 14 /* 15 * Snapshots in ceph are driven in large part by cooperation from the 16 * client. In contrast to local file systems or file servers that 17 * implement snapshots at a single point in the system, ceph's 18 * distributed access to storage requires clients to help decide 19 * whether a write logically occurs before or after a recently created 20 * snapshot. 21 * 22 * This provides a perfect instantanous client-wide snapshot. Between 23 * clients, however, snapshots may appear to be applied at slightly 24 * different points in time, depending on delays in delivering the 25 * snapshot notification. 26 * 27 * Snapshots are _not_ file system-wide. Instead, each snapshot 28 * applies to the subdirectory nested beneath some directory. This 29 * effectively divides the hierarchy into multiple "realms," where all 30 * of the files contained by each realm share the same set of 31 * snapshots. An individual realm's snap set contains snapshots 32 * explicitly created on that realm, as well as any snaps in its 33 * parent's snap set _after_ the point at which the parent became it's 34 * parent (due to, say, a rename). Similarly, snaps from prior parents 35 * during the time intervals during which they were the parent are included. 36 * 37 * The client is spared most of this detail, fortunately... it must only 38 * maintains a hierarchy of realms reflecting the current parent/child 39 * realm relationship, and for each realm has an explicit list of snaps 40 * inherited from prior parents. 41 * 42 * A snap_realm struct is maintained for realms containing every inode 43 * with an open cap in the system. (The needed snap realm information is 44 * provided by the MDS whenever a cap is issued, i.e., on open.) A 'seq' 45 * version number is used to ensure that as realm parameters change (new 46 * snapshot, new parent, etc.) the client's realm hierarchy is updated. 47 * 48 * The realm hierarchy drives the generation of a 'snap context' for each 49 * realm, which simply lists the resulting set of snaps for the realm. This 50 * is attached to any writes sent to OSDs. 51 */ 52 /* 53 * Unfortunately error handling is a bit mixed here. If we get a snap 54 * update, but don't have enough memory to update our realm hierarchy, 55 * it's not clear what we can do about it (besides complaining to the 56 * console). 57 */ 58 59 60 /* 61 * increase ref count for the realm 62 * 63 * caller must hold snap_rwsem for write. 64 */ 65 void ceph_get_snap_realm(struct ceph_mds_client *mdsc, 66 struct ceph_snap_realm *realm) 67 { 68 dout("get_realm %p %d -> %d\n", realm, 69 atomic_read(&realm->nref), atomic_read(&realm->nref)+1); 70 /* 71 * since we _only_ increment realm refs or empty the empty 72 * list with snap_rwsem held, adjusting the empty list here is 73 * safe. we do need to protect against concurrent empty list 74 * additions, however. 75 */ 76 if (atomic_inc_return(&realm->nref) == 1) { 77 spin_lock(&mdsc->snap_empty_lock); 78 list_del_init(&realm->empty_item); 79 spin_unlock(&mdsc->snap_empty_lock); 80 } 81 } 82 83 static void __insert_snap_realm(struct rb_root *root, 84 struct ceph_snap_realm *new) 85 { 86 struct rb_node **p = &root->rb_node; 87 struct rb_node *parent = NULL; 88 struct ceph_snap_realm *r = NULL; 89 90 while (*p) { 91 parent = *p; 92 r = rb_entry(parent, struct ceph_snap_realm, node); 93 if (new->ino < r->ino) 94 p = &(*p)->rb_left; 95 else if (new->ino > r->ino) 96 p = &(*p)->rb_right; 97 else 98 BUG(); 99 } 100 101 rb_link_node(&new->node, parent, p); 102 rb_insert_color(&new->node, root); 103 } 104 105 /* 106 * create and get the realm rooted at @ino and bump its ref count. 107 * 108 * caller must hold snap_rwsem for write. 109 */ 110 static struct ceph_snap_realm *ceph_create_snap_realm( 111 struct ceph_mds_client *mdsc, 112 u64 ino) 113 { 114 struct ceph_snap_realm *realm; 115 116 realm = kzalloc(sizeof(*realm), GFP_NOFS); 117 if (!realm) 118 return ERR_PTR(-ENOMEM); 119 120 atomic_set(&realm->nref, 1); /* for caller */ 121 realm->ino = ino; 122 INIT_LIST_HEAD(&realm->children); 123 INIT_LIST_HEAD(&realm->child_item); 124 INIT_LIST_HEAD(&realm->empty_item); 125 INIT_LIST_HEAD(&realm->dirty_item); 126 INIT_LIST_HEAD(&realm->inodes_with_caps); 127 spin_lock_init(&realm->inodes_with_caps_lock); 128 __insert_snap_realm(&mdsc->snap_realms, realm); 129 mdsc->num_snap_realms++; 130 131 dout("create_snap_realm %llx %p\n", realm->ino, realm); 132 return realm; 133 } 134 135 /* 136 * lookup the realm rooted at @ino. 137 * 138 * caller must hold snap_rwsem for write. 139 */ 140 static struct ceph_snap_realm *__lookup_snap_realm(struct ceph_mds_client *mdsc, 141 u64 ino) 142 { 143 struct rb_node *n = mdsc->snap_realms.rb_node; 144 struct ceph_snap_realm *r; 145 146 while (n) { 147 r = rb_entry(n, struct ceph_snap_realm, node); 148 if (ino < r->ino) 149 n = n->rb_left; 150 else if (ino > r->ino) 151 n = n->rb_right; 152 else { 153 dout("lookup_snap_realm %llx %p\n", r->ino, r); 154 return r; 155 } 156 } 157 return NULL; 158 } 159 160 struct ceph_snap_realm *ceph_lookup_snap_realm(struct ceph_mds_client *mdsc, 161 u64 ino) 162 { 163 struct ceph_snap_realm *r; 164 r = __lookup_snap_realm(mdsc, ino); 165 if (r) 166 ceph_get_snap_realm(mdsc, r); 167 return r; 168 } 169 170 static void __put_snap_realm(struct ceph_mds_client *mdsc, 171 struct ceph_snap_realm *realm); 172 173 /* 174 * called with snap_rwsem (write) 175 */ 176 static void __destroy_snap_realm(struct ceph_mds_client *mdsc, 177 struct ceph_snap_realm *realm) 178 { 179 dout("__destroy_snap_realm %p %llx\n", realm, realm->ino); 180 181 rb_erase(&realm->node, &mdsc->snap_realms); 182 mdsc->num_snap_realms--; 183 184 if (realm->parent) { 185 list_del_init(&realm->child_item); 186 __put_snap_realm(mdsc, realm->parent); 187 } 188 189 kfree(realm->prior_parent_snaps); 190 kfree(realm->snaps); 191 ceph_put_snap_context(realm->cached_context); 192 kfree(realm); 193 } 194 195 /* 196 * caller holds snap_rwsem (write) 197 */ 198 static void __put_snap_realm(struct ceph_mds_client *mdsc, 199 struct ceph_snap_realm *realm) 200 { 201 dout("__put_snap_realm %llx %p %d -> %d\n", realm->ino, realm, 202 atomic_read(&realm->nref), atomic_read(&realm->nref)-1); 203 if (atomic_dec_and_test(&realm->nref)) 204 __destroy_snap_realm(mdsc, realm); 205 } 206 207 /* 208 * caller needn't hold any locks 209 */ 210 void ceph_put_snap_realm(struct ceph_mds_client *mdsc, 211 struct ceph_snap_realm *realm) 212 { 213 dout("put_snap_realm %llx %p %d -> %d\n", realm->ino, realm, 214 atomic_read(&realm->nref), atomic_read(&realm->nref)-1); 215 if (!atomic_dec_and_test(&realm->nref)) 216 return; 217 218 if (down_write_trylock(&mdsc->snap_rwsem)) { 219 __destroy_snap_realm(mdsc, realm); 220 up_write(&mdsc->snap_rwsem); 221 } else { 222 spin_lock(&mdsc->snap_empty_lock); 223 list_add(&realm->empty_item, &mdsc->snap_empty); 224 spin_unlock(&mdsc->snap_empty_lock); 225 } 226 } 227 228 /* 229 * Clean up any realms whose ref counts have dropped to zero. Note 230 * that this does not include realms who were created but not yet 231 * used. 232 * 233 * Called under snap_rwsem (write) 234 */ 235 static void __cleanup_empty_realms(struct ceph_mds_client *mdsc) 236 { 237 struct ceph_snap_realm *realm; 238 239 spin_lock(&mdsc->snap_empty_lock); 240 while (!list_empty(&mdsc->snap_empty)) { 241 realm = list_first_entry(&mdsc->snap_empty, 242 struct ceph_snap_realm, empty_item); 243 list_del(&realm->empty_item); 244 spin_unlock(&mdsc->snap_empty_lock); 245 __destroy_snap_realm(mdsc, realm); 246 spin_lock(&mdsc->snap_empty_lock); 247 } 248 spin_unlock(&mdsc->snap_empty_lock); 249 } 250 251 void ceph_cleanup_empty_realms(struct ceph_mds_client *mdsc) 252 { 253 down_write(&mdsc->snap_rwsem); 254 __cleanup_empty_realms(mdsc); 255 up_write(&mdsc->snap_rwsem); 256 } 257 258 /* 259 * adjust the parent realm of a given @realm. adjust child list, and parent 260 * pointers, and ref counts appropriately. 261 * 262 * return true if parent was changed, 0 if unchanged, <0 on error. 263 * 264 * caller must hold snap_rwsem for write. 265 */ 266 static int adjust_snap_realm_parent(struct ceph_mds_client *mdsc, 267 struct ceph_snap_realm *realm, 268 u64 parentino) 269 { 270 struct ceph_snap_realm *parent; 271 272 if (realm->parent_ino == parentino) 273 return 0; 274 275 parent = ceph_lookup_snap_realm(mdsc, parentino); 276 if (!parent) { 277 parent = ceph_create_snap_realm(mdsc, parentino); 278 if (IS_ERR(parent)) 279 return PTR_ERR(parent); 280 } 281 dout("adjust_snap_realm_parent %llx %p: %llx %p -> %llx %p\n", 282 realm->ino, realm, realm->parent_ino, realm->parent, 283 parentino, parent); 284 if (realm->parent) { 285 list_del_init(&realm->child_item); 286 ceph_put_snap_realm(mdsc, realm->parent); 287 } 288 realm->parent_ino = parentino; 289 realm->parent = parent; 290 list_add(&realm->child_item, &parent->children); 291 return 1; 292 } 293 294 295 static int cmpu64_rev(const void *a, const void *b) 296 { 297 if (*(u64 *)a < *(u64 *)b) 298 return 1; 299 if (*(u64 *)a > *(u64 *)b) 300 return -1; 301 return 0; 302 } 303 304 305 /* 306 * build the snap context for a given realm. 307 */ 308 static int build_snap_context(struct ceph_snap_realm *realm, 309 struct list_head* dirty_realms) 310 { 311 struct ceph_snap_realm *parent = realm->parent; 312 struct ceph_snap_context *snapc; 313 int err = 0; 314 u32 num = realm->num_prior_parent_snaps + realm->num_snaps; 315 316 /* 317 * build parent context, if it hasn't been built. 318 * conservatively estimate that all parent snaps might be 319 * included by us. 320 */ 321 if (parent) { 322 if (!parent->cached_context) { 323 err = build_snap_context(parent, dirty_realms); 324 if (err) 325 goto fail; 326 } 327 num += parent->cached_context->num_snaps; 328 } 329 330 /* do i actually need to update? not if my context seq 331 matches realm seq, and my parents' does to. (this works 332 because we rebuild_snap_realms() works _downward_ in 333 hierarchy after each update.) */ 334 if (realm->cached_context && 335 realm->cached_context->seq == realm->seq && 336 (!parent || 337 realm->cached_context->seq >= parent->cached_context->seq)) { 338 dout("build_snap_context %llx %p: %p seq %lld (%u snaps)" 339 " (unchanged)\n", 340 realm->ino, realm, realm->cached_context, 341 realm->cached_context->seq, 342 (unsigned int)realm->cached_context->num_snaps); 343 return 0; 344 } 345 346 /* alloc new snap context */ 347 err = -ENOMEM; 348 if (num > (SIZE_MAX - sizeof(*snapc)) / sizeof(u64)) 349 goto fail; 350 snapc = ceph_create_snap_context(num, GFP_NOFS); 351 if (!snapc) 352 goto fail; 353 354 /* build (reverse sorted) snap vector */ 355 num = 0; 356 snapc->seq = realm->seq; 357 if (parent) { 358 u32 i; 359 360 /* include any of parent's snaps occurring _after_ my 361 parent became my parent */ 362 for (i = 0; i < parent->cached_context->num_snaps; i++) 363 if (parent->cached_context->snaps[i] >= 364 realm->parent_since) 365 snapc->snaps[num++] = 366 parent->cached_context->snaps[i]; 367 if (parent->cached_context->seq > snapc->seq) 368 snapc->seq = parent->cached_context->seq; 369 } 370 memcpy(snapc->snaps + num, realm->snaps, 371 sizeof(u64)*realm->num_snaps); 372 num += realm->num_snaps; 373 memcpy(snapc->snaps + num, realm->prior_parent_snaps, 374 sizeof(u64)*realm->num_prior_parent_snaps); 375 num += realm->num_prior_parent_snaps; 376 377 sort(snapc->snaps, num, sizeof(u64), cmpu64_rev, NULL); 378 snapc->num_snaps = num; 379 dout("build_snap_context %llx %p: %p seq %lld (%u snaps)\n", 380 realm->ino, realm, snapc, snapc->seq, 381 (unsigned int) snapc->num_snaps); 382 383 ceph_put_snap_context(realm->cached_context); 384 realm->cached_context = snapc; 385 /* queue realm for cap_snap creation */ 386 list_add_tail(&realm->dirty_item, dirty_realms); 387 return 0; 388 389 fail: 390 /* 391 * if we fail, clear old (incorrect) cached_context... hopefully 392 * we'll have better luck building it later 393 */ 394 if (realm->cached_context) { 395 ceph_put_snap_context(realm->cached_context); 396 realm->cached_context = NULL; 397 } 398 pr_err("build_snap_context %llx %p fail %d\n", realm->ino, 399 realm, err); 400 return err; 401 } 402 403 /* 404 * rebuild snap context for the given realm and all of its children. 405 */ 406 static void rebuild_snap_realms(struct ceph_snap_realm *realm, 407 struct list_head *dirty_realms) 408 { 409 struct ceph_snap_realm *child; 410 411 dout("rebuild_snap_realms %llx %p\n", realm->ino, realm); 412 build_snap_context(realm, dirty_realms); 413 414 list_for_each_entry(child, &realm->children, child_item) 415 rebuild_snap_realms(child, dirty_realms); 416 } 417 418 419 /* 420 * helper to allocate and decode an array of snapids. free prior 421 * instance, if any. 422 */ 423 static int dup_array(u64 **dst, __le64 *src, u32 num) 424 { 425 u32 i; 426 427 kfree(*dst); 428 if (num) { 429 *dst = kcalloc(num, sizeof(u64), GFP_NOFS); 430 if (!*dst) 431 return -ENOMEM; 432 for (i = 0; i < num; i++) 433 (*dst)[i] = get_unaligned_le64(src + i); 434 } else { 435 *dst = NULL; 436 } 437 return 0; 438 } 439 440 static bool has_new_snaps(struct ceph_snap_context *o, 441 struct ceph_snap_context *n) 442 { 443 if (n->num_snaps == 0) 444 return false; 445 /* snaps are in descending order */ 446 return n->snaps[0] > o->seq; 447 } 448 449 /* 450 * When a snapshot is applied, the size/mtime inode metadata is queued 451 * in a ceph_cap_snap (one for each snapshot) until writeback 452 * completes and the metadata can be flushed back to the MDS. 453 * 454 * However, if a (sync) write is currently in-progress when we apply 455 * the snapshot, we have to wait until the write succeeds or fails 456 * (and a final size/mtime is known). In this case the 457 * cap_snap->writing = 1, and is said to be "pending." When the write 458 * finishes, we __ceph_finish_cap_snap(). 459 * 460 * Caller must hold snap_rwsem for read (i.e., the realm topology won't 461 * change). 462 */ 463 void ceph_queue_cap_snap(struct ceph_inode_info *ci) 464 { 465 struct inode *inode = &ci->vfs_inode; 466 struct ceph_cap_snap *capsnap; 467 struct ceph_snap_context *old_snapc, *new_snapc; 468 struct ceph_buffer *old_blob = NULL; 469 int used, dirty; 470 471 capsnap = kzalloc(sizeof(*capsnap), GFP_NOFS); 472 if (!capsnap) { 473 pr_err("ENOMEM allocating ceph_cap_snap on %p\n", inode); 474 return; 475 } 476 477 spin_lock(&ci->i_ceph_lock); 478 used = __ceph_caps_used(ci); 479 dirty = __ceph_caps_dirty(ci); 480 481 old_snapc = ci->i_head_snapc; 482 new_snapc = ci->i_snap_realm->cached_context; 483 484 /* 485 * If there is a write in progress, treat that as a dirty Fw, 486 * even though it hasn't completed yet; by the time we finish 487 * up this capsnap it will be. 488 */ 489 if (used & CEPH_CAP_FILE_WR) 490 dirty |= CEPH_CAP_FILE_WR; 491 492 if (__ceph_have_pending_cap_snap(ci)) { 493 /* there is no point in queuing multiple "pending" cap_snaps, 494 as no new writes are allowed to start when pending, so any 495 writes in progress now were started before the previous 496 cap_snap. lucky us. */ 497 dout("queue_cap_snap %p already pending\n", inode); 498 goto update_snapc; 499 } 500 if (ci->i_wrbuffer_ref_head == 0 && 501 !(dirty & (CEPH_CAP_ANY_EXCL|CEPH_CAP_FILE_WR))) { 502 dout("queue_cap_snap %p nothing dirty|writing\n", inode); 503 goto update_snapc; 504 } 505 506 BUG_ON(!old_snapc); 507 508 /* 509 * There is no need to send FLUSHSNAP message to MDS if there is 510 * no new snapshot. But when there is dirty pages or on-going 511 * writes, we still need to create cap_snap. cap_snap is needed 512 * by the write path and page writeback path. 513 * 514 * also see ceph_try_drop_cap_snap() 515 */ 516 if (has_new_snaps(old_snapc, new_snapc)) { 517 if (dirty & (CEPH_CAP_ANY_EXCL|CEPH_CAP_FILE_WR)) 518 capsnap->need_flush = true; 519 } else { 520 if (!(used & CEPH_CAP_FILE_WR) && 521 ci->i_wrbuffer_ref_head == 0) { 522 dout("queue_cap_snap %p " 523 "no new_snap|dirty_page|writing\n", inode); 524 goto update_snapc; 525 } 526 } 527 528 dout("queue_cap_snap %p cap_snap %p queuing under %p %s %s\n", 529 inode, capsnap, old_snapc, ceph_cap_string(dirty), 530 capsnap->need_flush ? "" : "no_flush"); 531 ihold(inode); 532 533 refcount_set(&capsnap->nref, 1); 534 INIT_LIST_HEAD(&capsnap->ci_item); 535 536 capsnap->follows = old_snapc->seq; 537 capsnap->issued = __ceph_caps_issued(ci, NULL); 538 capsnap->dirty = dirty; 539 540 capsnap->mode = inode->i_mode; 541 capsnap->uid = inode->i_uid; 542 capsnap->gid = inode->i_gid; 543 544 if (dirty & CEPH_CAP_XATTR_EXCL) { 545 old_blob = __ceph_build_xattrs_blob(ci); 546 capsnap->xattr_blob = 547 ceph_buffer_get(ci->i_xattrs.blob); 548 capsnap->xattr_version = ci->i_xattrs.version; 549 } else { 550 capsnap->xattr_blob = NULL; 551 capsnap->xattr_version = 0; 552 } 553 554 capsnap->inline_data = ci->i_inline_version != CEPH_INLINE_NONE; 555 556 /* dirty page count moved from _head to this cap_snap; 557 all subsequent writes page dirties occur _after_ this 558 snapshot. */ 559 capsnap->dirty_pages = ci->i_wrbuffer_ref_head; 560 ci->i_wrbuffer_ref_head = 0; 561 capsnap->context = old_snapc; 562 list_add_tail(&capsnap->ci_item, &ci->i_cap_snaps); 563 564 if (used & CEPH_CAP_FILE_WR) { 565 dout("queue_cap_snap %p cap_snap %p snapc %p" 566 " seq %llu used WR, now pending\n", inode, 567 capsnap, old_snapc, old_snapc->seq); 568 capsnap->writing = 1; 569 } else { 570 /* note mtime, size NOW. */ 571 __ceph_finish_cap_snap(ci, capsnap); 572 } 573 capsnap = NULL; 574 old_snapc = NULL; 575 576 update_snapc: 577 if (ci->i_wrbuffer_ref_head == 0 && 578 ci->i_wr_ref == 0 && 579 ci->i_dirty_caps == 0 && 580 ci->i_flushing_caps == 0) { 581 ci->i_head_snapc = NULL; 582 } else { 583 ci->i_head_snapc = ceph_get_snap_context(new_snapc); 584 dout(" new snapc is %p\n", new_snapc); 585 } 586 spin_unlock(&ci->i_ceph_lock); 587 588 ceph_buffer_put(old_blob); 589 kfree(capsnap); 590 ceph_put_snap_context(old_snapc); 591 } 592 593 /* 594 * Finalize the size, mtime for a cap_snap.. that is, settle on final values 595 * to be used for the snapshot, to be flushed back to the mds. 596 * 597 * If capsnap can now be flushed, add to snap_flush list, and return 1. 598 * 599 * Caller must hold i_ceph_lock. 600 */ 601 int __ceph_finish_cap_snap(struct ceph_inode_info *ci, 602 struct ceph_cap_snap *capsnap) 603 { 604 struct inode *inode = &ci->vfs_inode; 605 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; 606 607 BUG_ON(capsnap->writing); 608 capsnap->size = inode->i_size; 609 capsnap->mtime = inode->i_mtime; 610 capsnap->atime = inode->i_atime; 611 capsnap->ctime = inode->i_ctime; 612 capsnap->btime = ci->i_btime; 613 capsnap->change_attr = inode_peek_iversion_raw(inode); 614 capsnap->time_warp_seq = ci->i_time_warp_seq; 615 capsnap->truncate_size = ci->i_truncate_size; 616 capsnap->truncate_seq = ci->i_truncate_seq; 617 if (capsnap->dirty_pages) { 618 dout("finish_cap_snap %p cap_snap %p snapc %p %llu %s s=%llu " 619 "still has %d dirty pages\n", inode, capsnap, 620 capsnap->context, capsnap->context->seq, 621 ceph_cap_string(capsnap->dirty), capsnap->size, 622 capsnap->dirty_pages); 623 return 0; 624 } 625 626 ci->i_ceph_flags |= CEPH_I_FLUSH_SNAPS; 627 dout("finish_cap_snap %p cap_snap %p snapc %p %llu %s s=%llu\n", 628 inode, capsnap, capsnap->context, 629 capsnap->context->seq, ceph_cap_string(capsnap->dirty), 630 capsnap->size); 631 632 spin_lock(&mdsc->snap_flush_lock); 633 if (list_empty(&ci->i_snap_flush_item)) 634 list_add_tail(&ci->i_snap_flush_item, &mdsc->snap_flush_list); 635 spin_unlock(&mdsc->snap_flush_lock); 636 return 1; /* caller may want to ceph_flush_snaps */ 637 } 638 639 /* 640 * Queue cap_snaps for snap writeback for this realm and its children. 641 * Called under snap_rwsem, so realm topology won't change. 642 */ 643 static void queue_realm_cap_snaps(struct ceph_snap_realm *realm) 644 { 645 struct ceph_inode_info *ci; 646 struct inode *lastinode = NULL; 647 648 dout("queue_realm_cap_snaps %p %llx inodes\n", realm, realm->ino); 649 650 spin_lock(&realm->inodes_with_caps_lock); 651 list_for_each_entry(ci, &realm->inodes_with_caps, i_snap_realm_item) { 652 struct inode *inode = igrab(&ci->vfs_inode); 653 if (!inode) 654 continue; 655 spin_unlock(&realm->inodes_with_caps_lock); 656 /* avoid calling iput_final() while holding 657 * mdsc->snap_rwsem or in mds dispatch threads */ 658 ceph_async_iput(lastinode); 659 lastinode = inode; 660 ceph_queue_cap_snap(ci); 661 spin_lock(&realm->inodes_with_caps_lock); 662 } 663 spin_unlock(&realm->inodes_with_caps_lock); 664 ceph_async_iput(lastinode); 665 666 dout("queue_realm_cap_snaps %p %llx done\n", realm, realm->ino); 667 } 668 669 /* 670 * Parse and apply a snapblob "snap trace" from the MDS. This specifies 671 * the snap realm parameters from a given realm and all of its ancestors, 672 * up to the root. 673 * 674 * Caller must hold snap_rwsem for write. 675 */ 676 int ceph_update_snap_trace(struct ceph_mds_client *mdsc, 677 void *p, void *e, bool deletion, 678 struct ceph_snap_realm **realm_ret) 679 { 680 struct ceph_mds_snap_realm *ri; /* encoded */ 681 __le64 *snaps; /* encoded */ 682 __le64 *prior_parent_snaps; /* encoded */ 683 struct ceph_snap_realm *realm = NULL; 684 struct ceph_snap_realm *first_realm = NULL; 685 int invalidate = 0; 686 int err = -ENOMEM; 687 LIST_HEAD(dirty_realms); 688 689 dout("update_snap_trace deletion=%d\n", deletion); 690 more: 691 ceph_decode_need(&p, e, sizeof(*ri), bad); 692 ri = p; 693 p += sizeof(*ri); 694 ceph_decode_need(&p, e, sizeof(u64)*(le32_to_cpu(ri->num_snaps) + 695 le32_to_cpu(ri->num_prior_parent_snaps)), bad); 696 snaps = p; 697 p += sizeof(u64) * le32_to_cpu(ri->num_snaps); 698 prior_parent_snaps = p; 699 p += sizeof(u64) * le32_to_cpu(ri->num_prior_parent_snaps); 700 701 realm = ceph_lookup_snap_realm(mdsc, le64_to_cpu(ri->ino)); 702 if (!realm) { 703 realm = ceph_create_snap_realm(mdsc, le64_to_cpu(ri->ino)); 704 if (IS_ERR(realm)) { 705 err = PTR_ERR(realm); 706 goto fail; 707 } 708 } 709 710 /* ensure the parent is correct */ 711 err = adjust_snap_realm_parent(mdsc, realm, le64_to_cpu(ri->parent)); 712 if (err < 0) 713 goto fail; 714 invalidate += err; 715 716 if (le64_to_cpu(ri->seq) > realm->seq) { 717 dout("update_snap_trace updating %llx %p %lld -> %lld\n", 718 realm->ino, realm, realm->seq, le64_to_cpu(ri->seq)); 719 /* update realm parameters, snap lists */ 720 realm->seq = le64_to_cpu(ri->seq); 721 realm->created = le64_to_cpu(ri->created); 722 realm->parent_since = le64_to_cpu(ri->parent_since); 723 724 realm->num_snaps = le32_to_cpu(ri->num_snaps); 725 err = dup_array(&realm->snaps, snaps, realm->num_snaps); 726 if (err < 0) 727 goto fail; 728 729 realm->num_prior_parent_snaps = 730 le32_to_cpu(ri->num_prior_parent_snaps); 731 err = dup_array(&realm->prior_parent_snaps, prior_parent_snaps, 732 realm->num_prior_parent_snaps); 733 if (err < 0) 734 goto fail; 735 736 if (realm->seq > mdsc->last_snap_seq) 737 mdsc->last_snap_seq = realm->seq; 738 739 invalidate = 1; 740 } else if (!realm->cached_context) { 741 dout("update_snap_trace %llx %p seq %lld new\n", 742 realm->ino, realm, realm->seq); 743 invalidate = 1; 744 } else { 745 dout("update_snap_trace %llx %p seq %lld unchanged\n", 746 realm->ino, realm, realm->seq); 747 } 748 749 dout("done with %llx %p, invalidated=%d, %p %p\n", realm->ino, 750 realm, invalidate, p, e); 751 752 /* invalidate when we reach the _end_ (root) of the trace */ 753 if (invalidate && p >= e) 754 rebuild_snap_realms(realm, &dirty_realms); 755 756 if (!first_realm) 757 first_realm = realm; 758 else 759 ceph_put_snap_realm(mdsc, realm); 760 761 if (p < e) 762 goto more; 763 764 /* 765 * queue cap snaps _after_ we've built the new snap contexts, 766 * so that i_head_snapc can be set appropriately. 767 */ 768 while (!list_empty(&dirty_realms)) { 769 realm = list_first_entry(&dirty_realms, struct ceph_snap_realm, 770 dirty_item); 771 list_del_init(&realm->dirty_item); 772 queue_realm_cap_snaps(realm); 773 } 774 775 if (realm_ret) 776 *realm_ret = first_realm; 777 else 778 ceph_put_snap_realm(mdsc, first_realm); 779 780 __cleanup_empty_realms(mdsc); 781 return 0; 782 783 bad: 784 err = -EINVAL; 785 fail: 786 if (realm && !IS_ERR(realm)) 787 ceph_put_snap_realm(mdsc, realm); 788 if (first_realm) 789 ceph_put_snap_realm(mdsc, first_realm); 790 pr_err("update_snap_trace error %d\n", err); 791 return err; 792 } 793 794 795 /* 796 * Send any cap_snaps that are queued for flush. Try to carry 797 * s_mutex across multiple snap flushes to avoid locking overhead. 798 * 799 * Caller holds no locks. 800 */ 801 static void flush_snaps(struct ceph_mds_client *mdsc) 802 { 803 struct ceph_inode_info *ci; 804 struct inode *inode; 805 struct ceph_mds_session *session = NULL; 806 807 dout("flush_snaps\n"); 808 spin_lock(&mdsc->snap_flush_lock); 809 while (!list_empty(&mdsc->snap_flush_list)) { 810 ci = list_first_entry(&mdsc->snap_flush_list, 811 struct ceph_inode_info, i_snap_flush_item); 812 inode = &ci->vfs_inode; 813 ihold(inode); 814 spin_unlock(&mdsc->snap_flush_lock); 815 ceph_flush_snaps(ci, &session); 816 /* avoid calling iput_final() while holding 817 * session->s_mutex or in mds dispatch threads */ 818 ceph_async_iput(inode); 819 spin_lock(&mdsc->snap_flush_lock); 820 } 821 spin_unlock(&mdsc->snap_flush_lock); 822 823 if (session) { 824 mutex_unlock(&session->s_mutex); 825 ceph_put_mds_session(session); 826 } 827 dout("flush_snaps done\n"); 828 } 829 830 831 /* 832 * Handle a snap notification from the MDS. 833 * 834 * This can take two basic forms: the simplest is just a snap creation 835 * or deletion notification on an existing realm. This should update the 836 * realm and its children. 837 * 838 * The more difficult case is realm creation, due to snap creation at a 839 * new point in the file hierarchy, or due to a rename that moves a file or 840 * directory into another realm. 841 */ 842 void ceph_handle_snap(struct ceph_mds_client *mdsc, 843 struct ceph_mds_session *session, 844 struct ceph_msg *msg) 845 { 846 struct super_block *sb = mdsc->fsc->sb; 847 int mds = session->s_mds; 848 u64 split; 849 int op; 850 int trace_len; 851 struct ceph_snap_realm *realm = NULL; 852 void *p = msg->front.iov_base; 853 void *e = p + msg->front.iov_len; 854 struct ceph_mds_snap_head *h; 855 int num_split_inos, num_split_realms; 856 __le64 *split_inos = NULL, *split_realms = NULL; 857 int i; 858 int locked_rwsem = 0; 859 860 /* decode */ 861 if (msg->front.iov_len < sizeof(*h)) 862 goto bad; 863 h = p; 864 op = le32_to_cpu(h->op); 865 split = le64_to_cpu(h->split); /* non-zero if we are splitting an 866 * existing realm */ 867 num_split_inos = le32_to_cpu(h->num_split_inos); 868 num_split_realms = le32_to_cpu(h->num_split_realms); 869 trace_len = le32_to_cpu(h->trace_len); 870 p += sizeof(*h); 871 872 dout("handle_snap from mds%d op %s split %llx tracelen %d\n", mds, 873 ceph_snap_op_name(op), split, trace_len); 874 875 mutex_lock(&session->s_mutex); 876 session->s_seq++; 877 mutex_unlock(&session->s_mutex); 878 879 down_write(&mdsc->snap_rwsem); 880 locked_rwsem = 1; 881 882 if (op == CEPH_SNAP_OP_SPLIT) { 883 struct ceph_mds_snap_realm *ri; 884 885 /* 886 * A "split" breaks part of an existing realm off into 887 * a new realm. The MDS provides a list of inodes 888 * (with caps) and child realms that belong to the new 889 * child. 890 */ 891 split_inos = p; 892 p += sizeof(u64) * num_split_inos; 893 split_realms = p; 894 p += sizeof(u64) * num_split_realms; 895 ceph_decode_need(&p, e, sizeof(*ri), bad); 896 /* we will peek at realm info here, but will _not_ 897 * advance p, as the realm update will occur below in 898 * ceph_update_snap_trace. */ 899 ri = p; 900 901 realm = ceph_lookup_snap_realm(mdsc, split); 902 if (!realm) { 903 realm = ceph_create_snap_realm(mdsc, split); 904 if (IS_ERR(realm)) 905 goto out; 906 } 907 908 dout("splitting snap_realm %llx %p\n", realm->ino, realm); 909 for (i = 0; i < num_split_inos; i++) { 910 struct ceph_vino vino = { 911 .ino = le64_to_cpu(split_inos[i]), 912 .snap = CEPH_NOSNAP, 913 }; 914 struct inode *inode = ceph_find_inode(sb, vino); 915 struct ceph_inode_info *ci; 916 struct ceph_snap_realm *oldrealm; 917 918 if (!inode) 919 continue; 920 ci = ceph_inode(inode); 921 922 spin_lock(&ci->i_ceph_lock); 923 if (!ci->i_snap_realm) 924 goto skip_inode; 925 /* 926 * If this inode belongs to a realm that was 927 * created after our new realm, we experienced 928 * a race (due to another split notifications 929 * arriving from a different MDS). So skip 930 * this inode. 931 */ 932 if (ci->i_snap_realm->created > 933 le64_to_cpu(ri->created)) { 934 dout(" leaving %p in newer realm %llx %p\n", 935 inode, ci->i_snap_realm->ino, 936 ci->i_snap_realm); 937 goto skip_inode; 938 } 939 dout(" will move %p to split realm %llx %p\n", 940 inode, realm->ino, realm); 941 /* 942 * Move the inode to the new realm 943 */ 944 oldrealm = ci->i_snap_realm; 945 spin_lock(&oldrealm->inodes_with_caps_lock); 946 list_del_init(&ci->i_snap_realm_item); 947 spin_unlock(&oldrealm->inodes_with_caps_lock); 948 949 spin_lock(&realm->inodes_with_caps_lock); 950 list_add(&ci->i_snap_realm_item, 951 &realm->inodes_with_caps); 952 ci->i_snap_realm = realm; 953 if (realm->ino == ci->i_vino.ino) 954 realm->inode = inode; 955 spin_unlock(&realm->inodes_with_caps_lock); 956 957 spin_unlock(&ci->i_ceph_lock); 958 959 ceph_get_snap_realm(mdsc, realm); 960 ceph_put_snap_realm(mdsc, oldrealm); 961 962 /* avoid calling iput_final() while holding 963 * mdsc->snap_rwsem or mds in dispatch threads */ 964 ceph_async_iput(inode); 965 continue; 966 967 skip_inode: 968 spin_unlock(&ci->i_ceph_lock); 969 ceph_async_iput(inode); 970 } 971 972 /* we may have taken some of the old realm's children. */ 973 for (i = 0; i < num_split_realms; i++) { 974 struct ceph_snap_realm *child = 975 __lookup_snap_realm(mdsc, 976 le64_to_cpu(split_realms[i])); 977 if (!child) 978 continue; 979 adjust_snap_realm_parent(mdsc, child, realm->ino); 980 } 981 } 982 983 /* 984 * update using the provided snap trace. if we are deleting a 985 * snap, we can avoid queueing cap_snaps. 986 */ 987 ceph_update_snap_trace(mdsc, p, e, 988 op == CEPH_SNAP_OP_DESTROY, NULL); 989 990 if (op == CEPH_SNAP_OP_SPLIT) 991 /* we took a reference when we created the realm, above */ 992 ceph_put_snap_realm(mdsc, realm); 993 994 __cleanup_empty_realms(mdsc); 995 996 up_write(&mdsc->snap_rwsem); 997 998 flush_snaps(mdsc); 999 return; 1000 1001 bad: 1002 pr_err("corrupt snap message from mds%d\n", mds); 1003 ceph_msg_dump(msg); 1004 out: 1005 if (locked_rwsem) 1006 up_write(&mdsc->snap_rwsem); 1007 return; 1008 } 1009 1010 struct ceph_snapid_map* ceph_get_snapid_map(struct ceph_mds_client *mdsc, 1011 u64 snap) 1012 { 1013 struct ceph_snapid_map *sm, *exist; 1014 struct rb_node **p, *parent; 1015 int ret; 1016 1017 exist = NULL; 1018 spin_lock(&mdsc->snapid_map_lock); 1019 p = &mdsc->snapid_map_tree.rb_node; 1020 while (*p) { 1021 exist = rb_entry(*p, struct ceph_snapid_map, node); 1022 if (snap > exist->snap) { 1023 p = &(*p)->rb_left; 1024 } else if (snap < exist->snap) { 1025 p = &(*p)->rb_right; 1026 } else { 1027 if (atomic_inc_return(&exist->ref) == 1) 1028 list_del_init(&exist->lru); 1029 break; 1030 } 1031 exist = NULL; 1032 } 1033 spin_unlock(&mdsc->snapid_map_lock); 1034 if (exist) { 1035 dout("found snapid map %llx -> %x\n", exist->snap, exist->dev); 1036 return exist; 1037 } 1038 1039 sm = kmalloc(sizeof(*sm), GFP_NOFS); 1040 if (!sm) 1041 return NULL; 1042 1043 ret = get_anon_bdev(&sm->dev); 1044 if (ret < 0) { 1045 kfree(sm); 1046 return NULL; 1047 } 1048 1049 INIT_LIST_HEAD(&sm->lru); 1050 atomic_set(&sm->ref, 1); 1051 sm->snap = snap; 1052 1053 exist = NULL; 1054 parent = NULL; 1055 p = &mdsc->snapid_map_tree.rb_node; 1056 spin_lock(&mdsc->snapid_map_lock); 1057 while (*p) { 1058 parent = *p; 1059 exist = rb_entry(*p, struct ceph_snapid_map, node); 1060 if (snap > exist->snap) 1061 p = &(*p)->rb_left; 1062 else if (snap < exist->snap) 1063 p = &(*p)->rb_right; 1064 else 1065 break; 1066 exist = NULL; 1067 } 1068 if (exist) { 1069 if (atomic_inc_return(&exist->ref) == 1) 1070 list_del_init(&exist->lru); 1071 } else { 1072 rb_link_node(&sm->node, parent, p); 1073 rb_insert_color(&sm->node, &mdsc->snapid_map_tree); 1074 } 1075 spin_unlock(&mdsc->snapid_map_lock); 1076 if (exist) { 1077 free_anon_bdev(sm->dev); 1078 kfree(sm); 1079 dout("found snapid map %llx -> %x\n", exist->snap, exist->dev); 1080 return exist; 1081 } 1082 1083 dout("create snapid map %llx -> %x\n", sm->snap, sm->dev); 1084 return sm; 1085 } 1086 1087 void ceph_put_snapid_map(struct ceph_mds_client* mdsc, 1088 struct ceph_snapid_map *sm) 1089 { 1090 if (!sm) 1091 return; 1092 if (atomic_dec_and_lock(&sm->ref, &mdsc->snapid_map_lock)) { 1093 if (!RB_EMPTY_NODE(&sm->node)) { 1094 sm->last_used = jiffies; 1095 list_add_tail(&sm->lru, &mdsc->snapid_map_lru); 1096 spin_unlock(&mdsc->snapid_map_lock); 1097 } else { 1098 /* already cleaned up by 1099 * ceph_cleanup_snapid_map() */ 1100 spin_unlock(&mdsc->snapid_map_lock); 1101 kfree(sm); 1102 } 1103 } 1104 } 1105 1106 void ceph_trim_snapid_map(struct ceph_mds_client *mdsc) 1107 { 1108 struct ceph_snapid_map *sm; 1109 unsigned long now; 1110 LIST_HEAD(to_free); 1111 1112 spin_lock(&mdsc->snapid_map_lock); 1113 now = jiffies; 1114 1115 while (!list_empty(&mdsc->snapid_map_lru)) { 1116 sm = list_first_entry(&mdsc->snapid_map_lru, 1117 struct ceph_snapid_map, lru); 1118 if (time_after(sm->last_used + CEPH_SNAPID_MAP_TIMEOUT, now)) 1119 break; 1120 1121 rb_erase(&sm->node, &mdsc->snapid_map_tree); 1122 list_move(&sm->lru, &to_free); 1123 } 1124 spin_unlock(&mdsc->snapid_map_lock); 1125 1126 while (!list_empty(&to_free)) { 1127 sm = list_first_entry(&to_free, struct ceph_snapid_map, lru); 1128 list_del(&sm->lru); 1129 dout("trim snapid map %llx -> %x\n", sm->snap, sm->dev); 1130 free_anon_bdev(sm->dev); 1131 kfree(sm); 1132 } 1133 } 1134 1135 void ceph_cleanup_snapid_map(struct ceph_mds_client *mdsc) 1136 { 1137 struct ceph_snapid_map *sm; 1138 struct rb_node *p; 1139 LIST_HEAD(to_free); 1140 1141 spin_lock(&mdsc->snapid_map_lock); 1142 while ((p = rb_first(&mdsc->snapid_map_tree))) { 1143 sm = rb_entry(p, struct ceph_snapid_map, node); 1144 rb_erase(p, &mdsc->snapid_map_tree); 1145 RB_CLEAR_NODE(p); 1146 list_move(&sm->lru, &to_free); 1147 } 1148 spin_unlock(&mdsc->snapid_map_lock); 1149 1150 while (!list_empty(&to_free)) { 1151 sm = list_first_entry(&to_free, struct ceph_snapid_map, lru); 1152 list_del(&sm->lru); 1153 free_anon_bdev(sm->dev); 1154 if (WARN_ON_ONCE(atomic_read(&sm->ref))) { 1155 pr_err("snapid map %llx -> %x still in use\n", 1156 sm->snap, sm->dev); 1157 } 1158 } 1159 } 1160