1 #include <linux/ceph/ceph_debug.h>
3 #include <linux/module.h>
5 #include <linux/highmem.h>
7 #include <linux/pagemap.h>
8 #include <linux/slab.h>
9 #include <linux/uaccess.h>
11 #include <linux/bio.h>
14 #include <linux/ceph/libceph.h>
15 #include <linux/ceph/osd_client.h>
16 #include <linux/ceph/messenger.h>
17 #include <linux/ceph/decode.h>
18 #include <linux/ceph/auth.h>
19 #include <linux/ceph/pagelist.h>
21 #define OSD_OP_FRONT_LEN 4096
22 #define OSD_OPREPLY_FRONT_LEN 512
24 static const struct ceph_connection_operations osd_con_ops;
26 static void send_queued(struct ceph_osd_client *osdc);
27 static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd);
28 static void __register_request(struct ceph_osd_client *osdc,
29 struct ceph_osd_request *req);
30 static void __unregister_linger_request(struct ceph_osd_client *osdc,
31 struct ceph_osd_request *req);
32 static void __send_request(struct ceph_osd_client *osdc,
33 struct ceph_osd_request *req);
35 static int op_needs_trail(int op)
38 case CEPH_OSD_OP_GETXATTR:
39 case CEPH_OSD_OP_SETXATTR:
40 case CEPH_OSD_OP_CMPXATTR:
41 case CEPH_OSD_OP_CALL:
42 case CEPH_OSD_OP_NOTIFY:
49 static int op_has_extent(int op)
51 return (op == CEPH_OSD_OP_READ ||
52 op == CEPH_OSD_OP_WRITE);
55 int ceph_calc_raw_layout(struct ceph_osd_client *osdc,
56 struct ceph_file_layout *layout,
58 u64 off, u64 *plen, u64 *bno,
59 struct ceph_osd_request *req,
60 struct ceph_osd_req_op *op)
62 struct ceph_osd_request_head *reqhead = req->r_request->front.iov_base;
64 u64 objoff, objlen; /* extent in object */
67 reqhead->snapid = cpu_to_le64(snapid);
70 r = ceph_calc_file_object_mapping(layout, off, plen, bno,
75 dout(" skipping last %llu, final file extent %llu~%llu\n",
76 orig_len - *plen, off, *plen);
78 if (op_has_extent(op->op)) {
79 op->extent.offset = objoff;
80 op->extent.length = objlen;
82 req->r_num_pages = calc_pages_for(off, *plen);
83 req->r_page_alignment = off & ~PAGE_MASK;
84 if (op->op == CEPH_OSD_OP_WRITE)
85 op->payload_len = *plen;
87 dout("calc_layout bno=%llx %llu~%llu (%d pages)\n",
88 *bno, objoff, objlen, req->r_num_pages);
91 EXPORT_SYMBOL(ceph_calc_raw_layout);
94 * Implement client access to distributed object storage cluster.
96 * All data objects are stored within a cluster/cloud of OSDs, or
97 * "object storage devices." (Note that Ceph OSDs have _nothing_ to
98 * do with the T10 OSD extensions to SCSI.) Ceph OSDs are simply
99 * remote daemons serving up and coordinating consistent and safe
102 * Cluster membership and the mapping of data objects onto storage devices
103 * are described by the osd map.
105 * We keep track of pending OSD requests (read, write), resubmit
106 * requests to different OSDs when the cluster topology/data layout
107 * change, or retry the affected requests when the communications
108 * channel with an OSD is reset.
112 * calculate the mapping of a file extent onto an object, and fill out the
113 * request accordingly. shorten extent as necessary if it crosses an
116 * fill osd op in request message.
118 static int calc_layout(struct ceph_osd_client *osdc,
119 struct ceph_vino vino,
120 struct ceph_file_layout *layout,
122 struct ceph_osd_request *req,
123 struct ceph_osd_req_op *op)
128 r = ceph_calc_raw_layout(osdc, layout, vino.snap, off,
129 plen, &bno, req, op);
133 snprintf(req->r_oid, sizeof(req->r_oid), "%llx.%08llx", vino.ino, bno);
134 req->r_oid_len = strlen(req->r_oid);
142 void ceph_osdc_release_request(struct kref *kref)
144 struct ceph_osd_request *req = container_of(kref,
145 struct ceph_osd_request,
149 ceph_msg_put(req->r_request);
150 if (req->r_con_filling_msg) {
151 dout("%s revoking pages %p from con %p\n", __func__,
152 req->r_pages, req->r_con_filling_msg);
153 ceph_msg_revoke_incoming(req->r_reply);
154 req->r_con_filling_msg->ops->put(req->r_con_filling_msg);
157 ceph_msg_put(req->r_reply);
158 if (req->r_own_pages)
159 ceph_release_page_vector(req->r_pages,
165 ceph_put_snap_context(req->r_snapc);
167 ceph_pagelist_release(req->r_trail);
171 mempool_free(req, req->r_osdc->req_mempool);
175 EXPORT_SYMBOL(ceph_osdc_release_request);
177 static int get_num_ops(struct ceph_osd_req_op *ops, int *needs_trail)
184 if (needs_trail && op_needs_trail(ops[i].op))
192 struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
194 struct ceph_snap_context *snapc,
195 struct ceph_osd_req_op *ops,
201 struct ceph_osd_request *req;
202 struct ceph_msg *msg;
204 int num_op = get_num_ops(ops, &needs_trail);
205 size_t msg_size = sizeof(struct ceph_osd_request_head);
207 msg_size += num_op*sizeof(struct ceph_osd_op);
210 req = mempool_alloc(osdc->req_mempool, gfp_flags);
211 memset(req, 0, sizeof(*req));
213 req = kzalloc(sizeof(*req), gfp_flags);
219 req->r_mempool = use_mempool;
221 kref_init(&req->r_kref);
222 init_completion(&req->r_completion);
223 init_completion(&req->r_safe_completion);
224 INIT_LIST_HEAD(&req->r_unsafe_item);
225 INIT_LIST_HEAD(&req->r_linger_item);
226 INIT_LIST_HEAD(&req->r_linger_osd);
227 INIT_LIST_HEAD(&req->r_req_lru_item);
228 INIT_LIST_HEAD(&req->r_osd_item);
230 req->r_flags = flags;
232 WARN_ON((flags & (CEPH_OSD_FLAG_READ|CEPH_OSD_FLAG_WRITE)) == 0);
234 /* create reply message */
236 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0);
238 msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY,
239 OSD_OPREPLY_FRONT_LEN, gfp_flags, true);
241 ceph_osdc_put_request(req);
246 /* allocate space for the trailing data */
248 req->r_trail = kmalloc(sizeof(struct ceph_pagelist), gfp_flags);
250 ceph_osdc_put_request(req);
253 ceph_pagelist_init(req->r_trail);
256 /* create request message; allow space for oid */
257 msg_size += MAX_OBJ_NAME_SIZE;
259 msg_size += sizeof(u64) * snapc->num_snaps;
261 msg = ceph_msgpool_get(&osdc->msgpool_op, 0);
263 msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp_flags, true);
265 ceph_osdc_put_request(req);
269 memset(msg->front.iov_base, 0, msg->front.iov_len);
271 req->r_request = msg;
272 req->r_pages = pages;
282 EXPORT_SYMBOL(ceph_osdc_alloc_request);
284 static void osd_req_encode_op(struct ceph_osd_request *req,
285 struct ceph_osd_op *dst,
286 struct ceph_osd_req_op *src)
288 dst->op = cpu_to_le16(src->op);
291 case CEPH_OSD_OP_READ:
292 case CEPH_OSD_OP_WRITE:
294 cpu_to_le64(src->extent.offset);
296 cpu_to_le64(src->extent.length);
297 dst->extent.truncate_size =
298 cpu_to_le64(src->extent.truncate_size);
299 dst->extent.truncate_seq =
300 cpu_to_le32(src->extent.truncate_seq);
303 case CEPH_OSD_OP_GETXATTR:
304 case CEPH_OSD_OP_SETXATTR:
305 case CEPH_OSD_OP_CMPXATTR:
306 BUG_ON(!req->r_trail);
308 dst->xattr.name_len = cpu_to_le32(src->xattr.name_len);
309 dst->xattr.value_len = cpu_to_le32(src->xattr.value_len);
310 dst->xattr.cmp_op = src->xattr.cmp_op;
311 dst->xattr.cmp_mode = src->xattr.cmp_mode;
312 ceph_pagelist_append(req->r_trail, src->xattr.name,
313 src->xattr.name_len);
314 ceph_pagelist_append(req->r_trail, src->xattr.val,
315 src->xattr.value_len);
317 case CEPH_OSD_OP_CALL:
318 BUG_ON(!req->r_trail);
320 dst->cls.class_len = src->cls.class_len;
321 dst->cls.method_len = src->cls.method_len;
322 dst->cls.indata_len = cpu_to_le32(src->cls.indata_len);
324 ceph_pagelist_append(req->r_trail, src->cls.class_name,
326 ceph_pagelist_append(req->r_trail, src->cls.method_name,
327 src->cls.method_len);
328 ceph_pagelist_append(req->r_trail, src->cls.indata,
329 src->cls.indata_len);
331 case CEPH_OSD_OP_ROLLBACK:
332 dst->snap.snapid = cpu_to_le64(src->snap.snapid);
334 case CEPH_OSD_OP_STARTSYNC:
336 case CEPH_OSD_OP_NOTIFY:
338 __le32 prot_ver = cpu_to_le32(src->watch.prot_ver);
339 __le32 timeout = cpu_to_le32(src->watch.timeout);
341 BUG_ON(!req->r_trail);
343 ceph_pagelist_append(req->r_trail,
344 &prot_ver, sizeof(prot_ver));
345 ceph_pagelist_append(req->r_trail,
346 &timeout, sizeof(timeout));
348 case CEPH_OSD_OP_NOTIFY_ACK:
349 case CEPH_OSD_OP_WATCH:
350 dst->watch.cookie = cpu_to_le64(src->watch.cookie);
351 dst->watch.ver = cpu_to_le64(src->watch.ver);
352 dst->watch.flag = src->watch.flag;
355 pr_err("unrecognized osd opcode %d\n", dst->op);
359 dst->payload_len = cpu_to_le32(src->payload_len);
363 * build new request AND message
366 void ceph_osdc_build_request(struct ceph_osd_request *req,
368 struct ceph_osd_req_op *src_ops,
369 struct ceph_snap_context *snapc,
370 struct timespec *mtime,
374 struct ceph_msg *msg = req->r_request;
375 struct ceph_osd_request_head *head;
376 struct ceph_osd_req_op *src_op;
377 struct ceph_osd_op *op;
379 int num_op = get_num_ops(src_ops, NULL);
380 size_t msg_size = sizeof(*head) + num_op*sizeof(*op);
381 int flags = req->r_flags;
385 head = msg->front.iov_base;
386 op = (void *)(head + 1);
387 p = (void *)(op + num_op);
389 req->r_snapc = ceph_get_snap_context(snapc);
391 head->client_inc = cpu_to_le32(1); /* always, for now. */
392 head->flags = cpu_to_le32(flags);
393 if (flags & CEPH_OSD_FLAG_WRITE)
394 ceph_encode_timespec(&head->mtime, mtime);
395 head->num_ops = cpu_to_le16(num_op);
399 head->object_len = cpu_to_le32(oid_len);
400 memcpy(p, oid, oid_len);
405 osd_req_encode_op(req, op, src_op);
411 data_len += req->r_trail->length;
414 head->snap_seq = cpu_to_le64(snapc->seq);
415 head->num_snaps = cpu_to_le32(snapc->num_snaps);
416 for (i = 0; i < snapc->num_snaps; i++) {
417 put_unaligned_le64(snapc->snaps[i], p);
422 if (flags & CEPH_OSD_FLAG_WRITE) {
423 req->r_request->hdr.data_off = cpu_to_le16(off);
424 req->r_request->hdr.data_len = cpu_to_le32(*plen + data_len);
425 } else if (data_len) {
426 req->r_request->hdr.data_off = 0;
427 req->r_request->hdr.data_len = cpu_to_le32(data_len);
430 req->r_request->page_alignment = req->r_page_alignment;
432 BUG_ON(p > msg->front.iov_base + msg->front.iov_len);
433 msg_size = p - msg->front.iov_base;
434 msg->front.iov_len = msg_size;
435 msg->hdr.front_len = cpu_to_le32(msg_size);
438 EXPORT_SYMBOL(ceph_osdc_build_request);
441 * build new request AND message, calculate layout, and adjust file
444 * if the file was recently truncated, we include information about its
445 * old and new size so that the object can be updated appropriately. (we
446 * avoid synchronously deleting truncated objects because it's slow.)
448 * if @do_sync, include a 'startsync' command so that the osd will flush
451 struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
452 struct ceph_file_layout *layout,
453 struct ceph_vino vino,
455 int opcode, int flags,
456 struct ceph_snap_context *snapc,
460 struct timespec *mtime,
461 bool use_mempool, int num_reply,
464 struct ceph_osd_req_op ops[3];
465 struct ceph_osd_request *req;
469 ops[0].extent.truncate_seq = truncate_seq;
470 ops[0].extent.truncate_size = truncate_size;
471 ops[0].payload_len = 0;
474 ops[1].op = CEPH_OSD_OP_STARTSYNC;
475 ops[1].payload_len = 0;
480 req = ceph_osdc_alloc_request(osdc, flags,
483 GFP_NOFS, NULL, NULL);
485 return ERR_PTR(-ENOMEM);
487 /* calculate max write size */
488 r = calc_layout(osdc, vino, layout, off, plen, req, ops);
491 req->r_file_layout = *layout; /* keep a copy */
493 /* in case it differs from natural (file) alignment that
494 calc_layout filled in for us */
495 req->r_num_pages = calc_pages_for(page_align, *plen);
496 req->r_page_alignment = page_align;
498 ceph_osdc_build_request(req, off, plen, ops,
501 req->r_oid, req->r_oid_len);
505 EXPORT_SYMBOL(ceph_osdc_new_request);
508 * We keep osd requests in an rbtree, sorted by ->r_tid.
510 static void __insert_request(struct ceph_osd_client *osdc,
511 struct ceph_osd_request *new)
513 struct rb_node **p = &osdc->requests.rb_node;
514 struct rb_node *parent = NULL;
515 struct ceph_osd_request *req = NULL;
519 req = rb_entry(parent, struct ceph_osd_request, r_node);
520 if (new->r_tid < req->r_tid)
522 else if (new->r_tid > req->r_tid)
528 rb_link_node(&new->r_node, parent, p);
529 rb_insert_color(&new->r_node, &osdc->requests);
532 static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc,
535 struct ceph_osd_request *req;
536 struct rb_node *n = osdc->requests.rb_node;
539 req = rb_entry(n, struct ceph_osd_request, r_node);
540 if (tid < req->r_tid)
542 else if (tid > req->r_tid)
550 static struct ceph_osd_request *
551 __lookup_request_ge(struct ceph_osd_client *osdc,
554 struct ceph_osd_request *req;
555 struct rb_node *n = osdc->requests.rb_node;
558 req = rb_entry(n, struct ceph_osd_request, r_node);
559 if (tid < req->r_tid) {
563 } else if (tid > req->r_tid) {
573 * Resubmit requests pending on the given osd.
575 static void __kick_osd_requests(struct ceph_osd_client *osdc,
576 struct ceph_osd *osd)
578 struct ceph_osd_request *req, *nreq;
581 dout("__kick_osd_requests osd%d\n", osd->o_osd);
582 err = __reset_osd(osdc, osd);
586 list_for_each_entry(req, &osd->o_requests, r_osd_item) {
587 list_move(&req->r_req_lru_item, &osdc->req_unsent);
588 dout("requeued %p tid %llu osd%d\n", req, req->r_tid,
591 req->r_flags |= CEPH_OSD_FLAG_RETRY;
594 list_for_each_entry_safe(req, nreq, &osd->o_linger_requests,
597 * reregister request prior to unregistering linger so
598 * that r_osd is preserved.
600 BUG_ON(!list_empty(&req->r_req_lru_item));
601 __register_request(osdc, req);
602 list_add(&req->r_req_lru_item, &osdc->req_unsent);
603 list_add(&req->r_osd_item, &req->r_osd->o_requests);
604 __unregister_linger_request(osdc, req);
605 dout("requeued lingering %p tid %llu osd%d\n", req, req->r_tid,
610 static void kick_osd_requests(struct ceph_osd_client *osdc,
611 struct ceph_osd *kickosd)
613 mutex_lock(&osdc->request_mutex);
614 __kick_osd_requests(osdc, kickosd);
615 mutex_unlock(&osdc->request_mutex);
619 * If the osd connection drops, we need to resubmit all requests.
621 static void osd_reset(struct ceph_connection *con)
623 struct ceph_osd *osd = con->private;
624 struct ceph_osd_client *osdc;
628 dout("osd_reset osd%d\n", osd->o_osd);
630 down_read(&osdc->map_sem);
631 kick_osd_requests(osdc, osd);
633 up_read(&osdc->map_sem);
637 * Track open sessions with osds.
639 static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
641 struct ceph_osd *osd;
643 osd = kzalloc(sizeof(*osd), GFP_NOFS);
647 atomic_set(&osd->o_ref, 1);
650 INIT_LIST_HEAD(&osd->o_requests);
651 INIT_LIST_HEAD(&osd->o_linger_requests);
652 INIT_LIST_HEAD(&osd->o_osd_lru);
653 osd->o_incarnation = 1;
655 ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
657 INIT_LIST_HEAD(&osd->o_keepalive_item);
661 static struct ceph_osd *get_osd(struct ceph_osd *osd)
663 if (atomic_inc_not_zero(&osd->o_ref)) {
664 dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
665 atomic_read(&osd->o_ref));
668 dout("get_osd %p FAIL\n", osd);
673 static void put_osd(struct ceph_osd *osd)
675 dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
676 atomic_read(&osd->o_ref) - 1);
677 if (atomic_dec_and_test(&osd->o_ref) && osd->o_auth.authorizer) {
678 struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth;
680 if (ac->ops && ac->ops->destroy_authorizer)
681 ac->ops->destroy_authorizer(ac, osd->o_auth.authorizer);
687 * remove an osd from our map
689 static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
691 dout("__remove_osd %p\n", osd);
692 BUG_ON(!list_empty(&osd->o_requests));
693 rb_erase(&osd->o_node, &osdc->osds);
694 list_del_init(&osd->o_osd_lru);
695 ceph_con_close(&osd->o_con);
699 static void remove_all_osds(struct ceph_osd_client *osdc)
701 dout("%s %p\n", __func__, osdc);
702 mutex_lock(&osdc->request_mutex);
703 while (!RB_EMPTY_ROOT(&osdc->osds)) {
704 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
705 struct ceph_osd, o_node);
706 __remove_osd(osdc, osd);
708 mutex_unlock(&osdc->request_mutex);
711 static void __move_osd_to_lru(struct ceph_osd_client *osdc,
712 struct ceph_osd *osd)
714 dout("__move_osd_to_lru %p\n", osd);
715 BUG_ON(!list_empty(&osd->o_osd_lru));
716 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
717 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl * HZ;
720 static void __remove_osd_from_lru(struct ceph_osd *osd)
722 dout("__remove_osd_from_lru %p\n", osd);
723 if (!list_empty(&osd->o_osd_lru))
724 list_del_init(&osd->o_osd_lru);
727 static void remove_old_osds(struct ceph_osd_client *osdc)
729 struct ceph_osd *osd, *nosd;
731 dout("__remove_old_osds %p\n", osdc);
732 mutex_lock(&osdc->request_mutex);
733 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
734 if (time_before(jiffies, osd->lru_ttl))
736 __remove_osd(osdc, osd);
738 mutex_unlock(&osdc->request_mutex);
744 static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
746 struct ceph_osd_request *req;
749 dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
750 if (list_empty(&osd->o_requests) &&
751 list_empty(&osd->o_linger_requests)) {
752 __remove_osd(osdc, osd);
753 } else if (memcmp(&osdc->osdmap->osd_addr[osd->o_osd],
754 &osd->o_con.peer_addr,
755 sizeof(osd->o_con.peer_addr)) == 0 &&
756 !ceph_con_opened(&osd->o_con)) {
757 dout(" osd addr hasn't changed and connection never opened,"
758 " letting msgr retry");
759 /* touch each r_stamp for handle_timeout()'s benfit */
760 list_for_each_entry(req, &osd->o_requests, r_osd_item)
761 req->r_stamp = jiffies;
764 ceph_con_close(&osd->o_con);
765 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd,
766 &osdc->osdmap->osd_addr[osd->o_osd]);
767 osd->o_incarnation++;
772 static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new)
774 struct rb_node **p = &osdc->osds.rb_node;
775 struct rb_node *parent = NULL;
776 struct ceph_osd *osd = NULL;
778 dout("__insert_osd %p osd%d\n", new, new->o_osd);
781 osd = rb_entry(parent, struct ceph_osd, o_node);
782 if (new->o_osd < osd->o_osd)
784 else if (new->o_osd > osd->o_osd)
790 rb_link_node(&new->o_node, parent, p);
791 rb_insert_color(&new->o_node, &osdc->osds);
794 static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o)
796 struct ceph_osd *osd;
797 struct rb_node *n = osdc->osds.rb_node;
800 osd = rb_entry(n, struct ceph_osd, o_node);
803 else if (o > osd->o_osd)
811 static void __schedule_osd_timeout(struct ceph_osd_client *osdc)
813 schedule_delayed_work(&osdc->timeout_work,
814 osdc->client->options->osd_keepalive_timeout * HZ);
817 static void __cancel_osd_timeout(struct ceph_osd_client *osdc)
819 cancel_delayed_work(&osdc->timeout_work);
823 * Register request, assign tid. If this is the first request, set up
826 static void __register_request(struct ceph_osd_client *osdc,
827 struct ceph_osd_request *req)
829 req->r_tid = ++osdc->last_tid;
830 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
831 dout("__register_request %p tid %lld\n", req, req->r_tid);
832 __insert_request(osdc, req);
833 ceph_osdc_get_request(req);
834 osdc->num_requests++;
835 if (osdc->num_requests == 1) {
836 dout(" first request, scheduling timeout\n");
837 __schedule_osd_timeout(osdc);
841 static void register_request(struct ceph_osd_client *osdc,
842 struct ceph_osd_request *req)
844 mutex_lock(&osdc->request_mutex);
845 __register_request(osdc, req);
846 mutex_unlock(&osdc->request_mutex);
850 * called under osdc->request_mutex
852 static void __unregister_request(struct ceph_osd_client *osdc,
853 struct ceph_osd_request *req)
855 if (RB_EMPTY_NODE(&req->r_node)) {
856 dout("__unregister_request %p tid %lld not registered\n",
861 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
862 rb_erase(&req->r_node, &osdc->requests);
863 osdc->num_requests--;
866 /* make sure the original request isn't in flight. */
867 ceph_msg_revoke(req->r_request);
869 list_del_init(&req->r_osd_item);
870 if (list_empty(&req->r_osd->o_requests) &&
871 list_empty(&req->r_osd->o_linger_requests)) {
872 dout("moving osd to %p lru\n", req->r_osd);
873 __move_osd_to_lru(osdc, req->r_osd);
875 if (list_empty(&req->r_linger_item))
879 ceph_osdc_put_request(req);
881 list_del_init(&req->r_req_lru_item);
882 if (osdc->num_requests == 0) {
883 dout(" no requests, canceling timeout\n");
884 __cancel_osd_timeout(osdc);
889 * Cancel a previously queued request message
891 static void __cancel_request(struct ceph_osd_request *req)
893 if (req->r_sent && req->r_osd) {
894 ceph_msg_revoke(req->r_request);
899 static void __register_linger_request(struct ceph_osd_client *osdc,
900 struct ceph_osd_request *req)
902 dout("__register_linger_request %p\n", req);
903 list_add_tail(&req->r_linger_item, &osdc->req_linger);
905 list_add_tail(&req->r_linger_osd,
906 &req->r_osd->o_linger_requests);
909 static void __unregister_linger_request(struct ceph_osd_client *osdc,
910 struct ceph_osd_request *req)
912 dout("__unregister_linger_request %p\n", req);
914 list_del_init(&req->r_linger_item);
915 list_del_init(&req->r_linger_osd);
917 if (list_empty(&req->r_osd->o_requests) &&
918 list_empty(&req->r_osd->o_linger_requests)) {
919 dout("moving osd to %p lru\n", req->r_osd);
920 __move_osd_to_lru(osdc, req->r_osd);
922 if (list_empty(&req->r_osd_item))
927 void ceph_osdc_unregister_linger_request(struct ceph_osd_client *osdc,
928 struct ceph_osd_request *req)
930 mutex_lock(&osdc->request_mutex);
932 __unregister_linger_request(osdc, req);
933 ceph_osdc_put_request(req);
935 mutex_unlock(&osdc->request_mutex);
937 EXPORT_SYMBOL(ceph_osdc_unregister_linger_request);
939 void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
940 struct ceph_osd_request *req)
942 if (!req->r_linger) {
943 dout("set_request_linger %p\n", req);
946 * caller is now responsible for calling
947 * unregister_linger_request
949 ceph_osdc_get_request(req);
952 EXPORT_SYMBOL(ceph_osdc_set_request_linger);
955 * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
956 * (as needed), and set the request r_osd appropriately. If there is
957 * no up osd, set r_osd to NULL. Move the request to the appropriate list
958 * (unsent, homeless) or leave on in-flight lru.
960 * Return 0 if unchanged, 1 if changed, or negative on error.
962 * Caller should hold map_sem for read and request_mutex.
964 static int __map_request(struct ceph_osd_client *osdc,
965 struct ceph_osd_request *req, int force_resend)
967 struct ceph_osd_request_head *reqhead = req->r_request->front.iov_base;
969 int acting[CEPH_PG_MAX_SIZE];
973 dout("map_request %p tid %lld\n", req, req->r_tid);
974 err = ceph_calc_object_layout(&reqhead->layout, req->r_oid,
975 &req->r_file_layout, osdc->osdmap);
977 list_move(&req->r_req_lru_item, &osdc->req_notarget);
980 pgid = reqhead->layout.ol_pgid;
983 err = ceph_calc_pg_acting(osdc->osdmap, pgid, acting);
989 if ((!force_resend &&
990 req->r_osd && req->r_osd->o_osd == o &&
991 req->r_sent >= req->r_osd->o_incarnation &&
992 req->r_num_pg_osds == num &&
993 memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
994 (req->r_osd == NULL && o == -1))
995 return 0; /* no change */
997 dout("map_request tid %llu pgid %d.%x osd%d (was osd%d)\n",
998 req->r_tid, le32_to_cpu(pgid.pool), le16_to_cpu(pgid.ps), o,
999 req->r_osd ? req->r_osd->o_osd : -1);
1001 /* record full pg acting set */
1002 memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
1003 req->r_num_pg_osds = num;
1006 __cancel_request(req);
1007 list_del_init(&req->r_osd_item);
1011 req->r_osd = __lookup_osd(osdc, o);
1012 if (!req->r_osd && o >= 0) {
1014 req->r_osd = create_osd(osdc, o);
1016 list_move(&req->r_req_lru_item, &osdc->req_notarget);
1020 dout("map_request osd %p is osd%d\n", req->r_osd, o);
1021 __insert_osd(osdc, req->r_osd);
1023 ceph_con_open(&req->r_osd->o_con,
1024 CEPH_ENTITY_TYPE_OSD, o,
1025 &osdc->osdmap->osd_addr[o]);
1029 __remove_osd_from_lru(req->r_osd);
1030 list_add(&req->r_osd_item, &req->r_osd->o_requests);
1031 list_move(&req->r_req_lru_item, &osdc->req_unsent);
1033 list_move(&req->r_req_lru_item, &osdc->req_notarget);
1035 err = 1; /* osd or pg changed */
1042 * caller should hold map_sem (for read) and request_mutex
1044 static void __send_request(struct ceph_osd_client *osdc,
1045 struct ceph_osd_request *req)
1047 struct ceph_osd_request_head *reqhead;
1049 dout("send_request %p tid %llu to osd%d flags %d\n",
1050 req, req->r_tid, req->r_osd->o_osd, req->r_flags);
1052 reqhead = req->r_request->front.iov_base;
1053 reqhead->osdmap_epoch = cpu_to_le32(osdc->osdmap->epoch);
1054 reqhead->flags |= cpu_to_le32(req->r_flags); /* e.g., RETRY */
1055 reqhead->reassert_version = req->r_reassert_version;
1057 req->r_stamp = jiffies;
1058 list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
1060 ceph_msg_get(req->r_request); /* send consumes a ref */
1061 ceph_con_send(&req->r_osd->o_con, req->r_request);
1062 req->r_sent = req->r_osd->o_incarnation;
1066 * Send any requests in the queue (req_unsent).
1068 static void send_queued(struct ceph_osd_client *osdc)
1070 struct ceph_osd_request *req, *tmp;
1072 dout("send_queued\n");
1073 mutex_lock(&osdc->request_mutex);
1074 list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item) {
1075 __send_request(osdc, req);
1077 mutex_unlock(&osdc->request_mutex);
1081 * Timeout callback, called every N seconds when 1 or more osd
1082 * requests has been active for more than N seconds. When this
1083 * happens, we ping all OSDs with requests who have timed out to
1084 * ensure any communications channel reset is detected. Reset the
1085 * request timeouts another N seconds in the future as we go.
1086 * Reschedule the timeout event another N seconds in future (unless
1087 * there are no open requests).
1089 static void handle_timeout(struct work_struct *work)
1091 struct ceph_osd_client *osdc =
1092 container_of(work, struct ceph_osd_client, timeout_work.work);
1093 struct ceph_osd_request *req, *last_req = NULL;
1094 struct ceph_osd *osd;
1095 unsigned long timeout = osdc->client->options->osd_timeout * HZ;
1096 unsigned long keepalive =
1097 osdc->client->options->osd_keepalive_timeout * HZ;
1098 unsigned long last_stamp = 0;
1099 struct list_head slow_osds;
1101 down_read(&osdc->map_sem);
1103 ceph_monc_request_next_osdmap(&osdc->client->monc);
1105 mutex_lock(&osdc->request_mutex);
1108 * reset osds that appear to be _really_ unresponsive. this
1109 * is a failsafe measure.. we really shouldn't be getting to
1110 * this point if the system is working properly. the monitors
1111 * should mark the osd as failed and we should find out about
1112 * it from an updated osd map.
1114 while (timeout && !list_empty(&osdc->req_lru)) {
1115 req = list_entry(osdc->req_lru.next, struct ceph_osd_request,
1118 /* hasn't been long enough since we sent it? */
1119 if (time_before(jiffies, req->r_stamp + timeout))
1122 /* hasn't been long enough since it was acked? */
1123 if (req->r_request->ack_stamp == 0 ||
1124 time_before(jiffies, req->r_request->ack_stamp + timeout))
1127 BUG_ON(req == last_req && req->r_stamp == last_stamp);
1129 last_stamp = req->r_stamp;
1133 pr_warning(" tid %llu timed out on osd%d, will reset osd\n",
1134 req->r_tid, osd->o_osd);
1135 __kick_osd_requests(osdc, osd);
1139 * ping osds that are a bit slow. this ensures that if there
1140 * is a break in the TCP connection we will notice, and reopen
1141 * a connection with that osd (from the fault callback).
1143 INIT_LIST_HEAD(&slow_osds);
1144 list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
1145 if (time_before(jiffies, req->r_stamp + keepalive))
1150 dout(" tid %llu is slow, will send keepalive on osd%d\n",
1151 req->r_tid, osd->o_osd);
1152 list_move_tail(&osd->o_keepalive_item, &slow_osds);
1154 while (!list_empty(&slow_osds)) {
1155 osd = list_entry(slow_osds.next, struct ceph_osd,
1157 list_del_init(&osd->o_keepalive_item);
1158 ceph_con_keepalive(&osd->o_con);
1161 __schedule_osd_timeout(osdc);
1162 mutex_unlock(&osdc->request_mutex);
1164 up_read(&osdc->map_sem);
1167 static void handle_osds_timeout(struct work_struct *work)
1169 struct ceph_osd_client *osdc =
1170 container_of(work, struct ceph_osd_client,
1171 osds_timeout_work.work);
1172 unsigned long delay =
1173 osdc->client->options->osd_idle_ttl * HZ >> 2;
1175 dout("osds timeout\n");
1176 down_read(&osdc->map_sem);
1177 remove_old_osds(osdc);
1178 up_read(&osdc->map_sem);
1180 schedule_delayed_work(&osdc->osds_timeout_work,
1181 round_jiffies_relative(delay));
1184 static void complete_request(struct ceph_osd_request *req)
1186 if (req->r_safe_callback)
1187 req->r_safe_callback(req, NULL);
1188 complete_all(&req->r_safe_completion); /* fsync waiter */
1192 * handle osd op reply. either call the callback if it is specified,
1193 * or do the completion to wake up the waiting thread.
1195 static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg,
1196 struct ceph_connection *con)
1198 struct ceph_osd_reply_head *rhead = msg->front.iov_base;
1199 struct ceph_osd_request *req;
1201 int numops, object_len, flags;
1204 tid = le64_to_cpu(msg->hdr.tid);
1205 if (msg->front.iov_len < sizeof(*rhead))
1207 numops = le32_to_cpu(rhead->num_ops);
1208 object_len = le32_to_cpu(rhead->object_len);
1209 result = le32_to_cpu(rhead->result);
1210 if (msg->front.iov_len != sizeof(*rhead) + object_len +
1211 numops * sizeof(struct ceph_osd_op))
1213 dout("handle_reply %p tid %llu result %d\n", msg, tid, (int)result);
1215 mutex_lock(&osdc->request_mutex);
1216 req = __lookup_request(osdc, tid);
1218 dout("handle_reply tid %llu dne\n", tid);
1219 mutex_unlock(&osdc->request_mutex);
1222 ceph_osdc_get_request(req);
1223 flags = le32_to_cpu(rhead->flags);
1226 * if this connection filled our message, drop our reference now, to
1227 * avoid a (safe but slower) revoke later.
1229 if (req->r_con_filling_msg == con && req->r_reply == msg) {
1230 dout(" dropping con_filling_msg ref %p\n", con);
1231 req->r_con_filling_msg = NULL;
1235 if (!req->r_got_reply) {
1238 req->r_result = le32_to_cpu(rhead->result);
1239 bytes = le32_to_cpu(msg->hdr.data_len);
1240 dout("handle_reply result %d bytes %d\n", req->r_result,
1242 if (req->r_result == 0)
1243 req->r_result = bytes;
1245 /* in case this is a write and we need to replay, */
1246 req->r_reassert_version = rhead->reassert_version;
1248 req->r_got_reply = 1;
1249 } else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) {
1250 dout("handle_reply tid %llu dup ack\n", tid);
1251 mutex_unlock(&osdc->request_mutex);
1255 dout("handle_reply tid %llu flags %d\n", tid, flags);
1257 if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK))
1258 __register_linger_request(osdc, req);
1260 /* either this is a read, or we got the safe response */
1262 (flags & CEPH_OSD_FLAG_ONDISK) ||
1263 ((flags & CEPH_OSD_FLAG_WRITE) == 0))
1264 __unregister_request(osdc, req);
1266 mutex_unlock(&osdc->request_mutex);
1268 if (req->r_callback)
1269 req->r_callback(req, msg);
1271 complete_all(&req->r_completion);
1273 if (flags & CEPH_OSD_FLAG_ONDISK)
1274 complete_request(req);
1277 dout("req=%p req->r_linger=%d\n", req, req->r_linger);
1278 ceph_osdc_put_request(req);
1282 pr_err("corrupt osd_op_reply got %d %d expected %d\n",
1283 (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len),
1284 (int)sizeof(*rhead));
1288 static void reset_changed_osds(struct ceph_osd_client *osdc)
1290 struct rb_node *p, *n;
1292 for (p = rb_first(&osdc->osds); p; p = n) {
1293 struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
1296 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
1297 memcmp(&osd->o_con.peer_addr,
1298 ceph_osd_addr(osdc->osdmap,
1300 sizeof(struct ceph_entity_addr)) != 0)
1301 __reset_osd(osdc, osd);
1306 * Requeue requests whose mapping to an OSD has changed. If requests map to
1307 * no osd, request a new map.
1309 * Caller should hold map_sem for read and request_mutex.
1311 static void kick_requests(struct ceph_osd_client *osdc, int force_resend)
1313 struct ceph_osd_request *req, *nreq;
1318 dout("kick_requests %s\n", force_resend ? " (force resend)" : "");
1319 mutex_lock(&osdc->request_mutex);
1320 for (p = rb_first(&osdc->requests); p; ) {
1321 req = rb_entry(p, struct ceph_osd_request, r_node);
1323 err = __map_request(osdc, req, force_resend);
1325 continue; /* error */
1326 if (req->r_osd == NULL) {
1327 dout("%p tid %llu maps to no osd\n", req, req->r_tid);
1328 needmap++; /* request a newer map */
1329 } else if (err > 0) {
1330 if (!req->r_linger) {
1331 dout("%p tid %llu requeued on osd%d\n", req,
1333 req->r_osd ? req->r_osd->o_osd : -1);
1334 req->r_flags |= CEPH_OSD_FLAG_RETRY;
1337 if (req->r_linger && list_empty(&req->r_linger_item)) {
1339 * register as a linger so that we will
1340 * re-submit below and get a new tid
1342 dout("%p tid %llu restart on osd%d\n",
1344 req->r_osd ? req->r_osd->o_osd : -1);
1345 __register_linger_request(osdc, req);
1346 __unregister_request(osdc, req);
1350 list_for_each_entry_safe(req, nreq, &osdc->req_linger,
1352 dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
1354 err = __map_request(osdc, req, force_resend);
1356 continue; /* no change and no osd was specified */
1358 continue; /* hrm! */
1359 if (req->r_osd == NULL) {
1360 dout("tid %llu maps to no valid osd\n", req->r_tid);
1361 needmap++; /* request a newer map */
1365 dout("kicking lingering %p tid %llu osd%d\n", req, req->r_tid,
1366 req->r_osd ? req->r_osd->o_osd : -1);
1367 __unregister_linger_request(osdc, req);
1368 __register_request(osdc, req);
1370 mutex_unlock(&osdc->request_mutex);
1373 dout("%d requests for down osds, need new map\n", needmap);
1374 ceph_monc_request_next_osdmap(&osdc->client->monc);
1380 * Process updated osd map.
1382 * The message contains any number of incremental and full maps, normally
1383 * indicating some sort of topology change in the cluster. Kick requests
1384 * off to different OSDs as needed.
1386 void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1388 void *p, *end, *next;
1389 u32 nr_maps, maplen;
1391 struct ceph_osdmap *newmap = NULL, *oldmap;
1393 struct ceph_fsid fsid;
1395 dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
1396 p = msg->front.iov_base;
1397 end = p + msg->front.iov_len;
1400 ceph_decode_need(&p, end, sizeof(fsid), bad);
1401 ceph_decode_copy(&p, &fsid, sizeof(fsid));
1402 if (ceph_check_fsid(osdc->client, &fsid) < 0)
1405 down_write(&osdc->map_sem);
1407 /* incremental maps */
1408 ceph_decode_32_safe(&p, end, nr_maps, bad);
1409 dout(" %d inc maps\n", nr_maps);
1410 while (nr_maps > 0) {
1411 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
1412 epoch = ceph_decode_32(&p);
1413 maplen = ceph_decode_32(&p);
1414 ceph_decode_need(&p, end, maplen, bad);
1416 if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
1417 dout("applying incremental map %u len %d\n",
1419 newmap = osdmap_apply_incremental(&p, next,
1421 &osdc->client->msgr);
1422 if (IS_ERR(newmap)) {
1423 err = PTR_ERR(newmap);
1427 if (newmap != osdc->osdmap) {
1428 ceph_osdmap_destroy(osdc->osdmap);
1429 osdc->osdmap = newmap;
1431 kick_requests(osdc, 0);
1432 reset_changed_osds(osdc);
1434 dout("ignoring incremental map %u len %d\n",
1444 ceph_decode_32_safe(&p, end, nr_maps, bad);
1445 dout(" %d full maps\n", nr_maps);
1447 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
1448 epoch = ceph_decode_32(&p);
1449 maplen = ceph_decode_32(&p);
1450 ceph_decode_need(&p, end, maplen, bad);
1452 dout("skipping non-latest full map %u len %d\n",
1454 } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
1455 dout("skipping full map %u len %d, "
1456 "older than our %u\n", epoch, maplen,
1457 osdc->osdmap->epoch);
1459 int skipped_map = 0;
1461 dout("taking full map %u len %d\n", epoch, maplen);
1462 newmap = osdmap_decode(&p, p+maplen);
1463 if (IS_ERR(newmap)) {
1464 err = PTR_ERR(newmap);
1468 oldmap = osdc->osdmap;
1469 osdc->osdmap = newmap;
1471 if (oldmap->epoch + 1 < newmap->epoch)
1473 ceph_osdmap_destroy(oldmap);
1475 kick_requests(osdc, skipped_map);
1482 downgrade_write(&osdc->map_sem);
1483 ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch);
1486 * subscribe to subsequent osdmap updates if full to ensure
1487 * we find out when we are no longer full and stop returning
1490 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL))
1491 ceph_monc_request_next_osdmap(&osdc->client->monc);
1494 up_read(&osdc->map_sem);
1495 wake_up_all(&osdc->client->auth_wq);
1499 pr_err("osdc handle_map corrupt msg\n");
1501 up_write(&osdc->map_sem);
1506 * watch/notify callback event infrastructure
1508 * These callbacks are used both for watch and notify operations.
1510 static void __release_event(struct kref *kref)
1512 struct ceph_osd_event *event =
1513 container_of(kref, struct ceph_osd_event, kref);
1515 dout("__release_event %p\n", event);
1519 static void get_event(struct ceph_osd_event *event)
1521 kref_get(&event->kref);
1524 void ceph_osdc_put_event(struct ceph_osd_event *event)
1526 kref_put(&event->kref, __release_event);
1528 EXPORT_SYMBOL(ceph_osdc_put_event);
1530 static void __insert_event(struct ceph_osd_client *osdc,
1531 struct ceph_osd_event *new)
1533 struct rb_node **p = &osdc->event_tree.rb_node;
1534 struct rb_node *parent = NULL;
1535 struct ceph_osd_event *event = NULL;
1539 event = rb_entry(parent, struct ceph_osd_event, node);
1540 if (new->cookie < event->cookie)
1542 else if (new->cookie > event->cookie)
1543 p = &(*p)->rb_right;
1548 rb_link_node(&new->node, parent, p);
1549 rb_insert_color(&new->node, &osdc->event_tree);
1552 static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
1555 struct rb_node **p = &osdc->event_tree.rb_node;
1556 struct rb_node *parent = NULL;
1557 struct ceph_osd_event *event = NULL;
1561 event = rb_entry(parent, struct ceph_osd_event, node);
1562 if (cookie < event->cookie)
1564 else if (cookie > event->cookie)
1565 p = &(*p)->rb_right;
1572 static void __remove_event(struct ceph_osd_event *event)
1574 struct ceph_osd_client *osdc = event->osdc;
1576 if (!RB_EMPTY_NODE(&event->node)) {
1577 dout("__remove_event removed %p\n", event);
1578 rb_erase(&event->node, &osdc->event_tree);
1579 ceph_osdc_put_event(event);
1581 dout("__remove_event didn't remove %p\n", event);
1585 int ceph_osdc_create_event(struct ceph_osd_client *osdc,
1586 void (*event_cb)(u64, u64, u8, void *),
1587 int one_shot, void *data,
1588 struct ceph_osd_event **pevent)
1590 struct ceph_osd_event *event;
1592 event = kmalloc(sizeof(*event), GFP_NOIO);
1596 dout("create_event %p\n", event);
1597 event->cb = event_cb;
1598 event->one_shot = one_shot;
1601 INIT_LIST_HEAD(&event->osd_node);
1602 kref_init(&event->kref); /* one ref for us */
1603 kref_get(&event->kref); /* one ref for the caller */
1604 init_completion(&event->completion);
1606 spin_lock(&osdc->event_lock);
1607 event->cookie = ++osdc->event_count;
1608 __insert_event(osdc, event);
1609 spin_unlock(&osdc->event_lock);
1614 EXPORT_SYMBOL(ceph_osdc_create_event);
1616 void ceph_osdc_cancel_event(struct ceph_osd_event *event)
1618 struct ceph_osd_client *osdc = event->osdc;
1620 dout("cancel_event %p\n", event);
1621 spin_lock(&osdc->event_lock);
1622 __remove_event(event);
1623 spin_unlock(&osdc->event_lock);
1624 ceph_osdc_put_event(event); /* caller's */
1626 EXPORT_SYMBOL(ceph_osdc_cancel_event);
1629 static void do_event_work(struct work_struct *work)
1631 struct ceph_osd_event_work *event_work =
1632 container_of(work, struct ceph_osd_event_work, work);
1633 struct ceph_osd_event *event = event_work->event;
1634 u64 ver = event_work->ver;
1635 u64 notify_id = event_work->notify_id;
1636 u8 opcode = event_work->opcode;
1638 dout("do_event_work completing %p\n", event);
1639 event->cb(ver, notify_id, opcode, event->data);
1640 complete(&event->completion);
1641 dout("do_event_work completed %p\n", event);
1642 ceph_osdc_put_event(event);
1648 * Process osd watch notifications
1650 void handle_watch_notify(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1654 u64 cookie, ver, notify_id;
1656 struct ceph_osd_event *event;
1657 struct ceph_osd_event_work *event_work;
1659 p = msg->front.iov_base;
1660 end = p + msg->front.iov_len;
1662 ceph_decode_8_safe(&p, end, proto_ver, bad);
1663 ceph_decode_8_safe(&p, end, opcode, bad);
1664 ceph_decode_64_safe(&p, end, cookie, bad);
1665 ceph_decode_64_safe(&p, end, ver, bad);
1666 ceph_decode_64_safe(&p, end, notify_id, bad);
1668 spin_lock(&osdc->event_lock);
1669 event = __find_event(osdc, cookie);
1672 if (event->one_shot)
1673 __remove_event(event);
1675 spin_unlock(&osdc->event_lock);
1676 dout("handle_watch_notify cookie %lld ver %lld event %p\n",
1677 cookie, ver, event);
1679 event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
1681 dout("ERROR: could not allocate event_work\n");
1684 INIT_WORK(&event_work->work, do_event_work);
1685 event_work->event = event;
1686 event_work->ver = ver;
1687 event_work->notify_id = notify_id;
1688 event_work->opcode = opcode;
1689 if (!queue_work(osdc->notify_wq, &event_work->work)) {
1690 dout("WARNING: failed to queue notify event work\n");
1698 complete(&event->completion);
1699 ceph_osdc_put_event(event);
1703 pr_err("osdc handle_watch_notify corrupt msg\n");
1707 int ceph_osdc_wait_event(struct ceph_osd_event *event, unsigned long timeout)
1711 dout("wait_event %p\n", event);
1712 err = wait_for_completion_interruptible_timeout(&event->completion,
1714 ceph_osdc_put_event(event);
1717 dout("wait_event %p returns %d\n", event, err);
1720 EXPORT_SYMBOL(ceph_osdc_wait_event);
1723 * Register request, send initial attempt.
1725 int ceph_osdc_start_request(struct ceph_osd_client *osdc,
1726 struct ceph_osd_request *req,
1731 req->r_request->pages = req->r_pages;
1732 req->r_request->nr_pages = req->r_num_pages;
1734 req->r_request->bio = req->r_bio;
1736 req->r_request->trail = req->r_trail;
1738 register_request(osdc, req);
1740 down_read(&osdc->map_sem);
1741 mutex_lock(&osdc->request_mutex);
1743 * a racing kick_requests() may have sent the message for us
1744 * while we dropped request_mutex above, so only send now if
1745 * the request still han't been touched yet.
1747 if (req->r_sent == 0) {
1748 rc = __map_request(osdc, req, 0);
1751 dout("osdc_start_request failed map, "
1752 " will retry %lld\n", req->r_tid);
1757 if (req->r_osd == NULL) {
1758 dout("send_request %p no up osds in pg\n", req);
1759 ceph_monc_request_next_osdmap(&osdc->client->monc);
1761 __send_request(osdc, req);
1767 mutex_unlock(&osdc->request_mutex);
1768 up_read(&osdc->map_sem);
1771 EXPORT_SYMBOL(ceph_osdc_start_request);
1774 * wait for a request to complete
1776 int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
1777 struct ceph_osd_request *req)
1781 rc = wait_for_completion_interruptible(&req->r_completion);
1783 mutex_lock(&osdc->request_mutex);
1784 __cancel_request(req);
1785 __unregister_request(osdc, req);
1786 mutex_unlock(&osdc->request_mutex);
1787 complete_request(req);
1788 dout("wait_request tid %llu canceled/timed out\n", req->r_tid);
1792 dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result);
1793 return req->r_result;
1795 EXPORT_SYMBOL(ceph_osdc_wait_request);
1798 * sync - wait for all in-flight requests to flush. avoid starvation.
1800 void ceph_osdc_sync(struct ceph_osd_client *osdc)
1802 struct ceph_osd_request *req;
1803 u64 last_tid, next_tid = 0;
1805 mutex_lock(&osdc->request_mutex);
1806 last_tid = osdc->last_tid;
1808 req = __lookup_request_ge(osdc, next_tid);
1811 if (req->r_tid > last_tid)
1814 next_tid = req->r_tid + 1;
1815 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
1818 ceph_osdc_get_request(req);
1819 mutex_unlock(&osdc->request_mutex);
1820 dout("sync waiting on tid %llu (last is %llu)\n",
1821 req->r_tid, last_tid);
1822 wait_for_completion(&req->r_safe_completion);
1823 mutex_lock(&osdc->request_mutex);
1824 ceph_osdc_put_request(req);
1826 mutex_unlock(&osdc->request_mutex);
1827 dout("sync done (thru tid %llu)\n", last_tid);
1829 EXPORT_SYMBOL(ceph_osdc_sync);
1834 int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
1839 osdc->client = client;
1840 osdc->osdmap = NULL;
1841 init_rwsem(&osdc->map_sem);
1842 init_completion(&osdc->map_waiters);
1843 osdc->last_requested_map = 0;
1844 mutex_init(&osdc->request_mutex);
1846 osdc->osds = RB_ROOT;
1847 INIT_LIST_HEAD(&osdc->osd_lru);
1848 osdc->requests = RB_ROOT;
1849 INIT_LIST_HEAD(&osdc->req_lru);
1850 INIT_LIST_HEAD(&osdc->req_unsent);
1851 INIT_LIST_HEAD(&osdc->req_notarget);
1852 INIT_LIST_HEAD(&osdc->req_linger);
1853 osdc->num_requests = 0;
1854 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
1855 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
1856 spin_lock_init(&osdc->event_lock);
1857 osdc->event_tree = RB_ROOT;
1858 osdc->event_count = 0;
1860 schedule_delayed_work(&osdc->osds_timeout_work,
1861 round_jiffies_relative(osdc->client->options->osd_idle_ttl * HZ));
1864 osdc->req_mempool = mempool_create_kmalloc_pool(10,
1865 sizeof(struct ceph_osd_request));
1866 if (!osdc->req_mempool)
1869 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
1870 OSD_OP_FRONT_LEN, 10, true,
1874 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
1875 OSD_OPREPLY_FRONT_LEN, 10, true,
1880 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
1881 if (IS_ERR(osdc->notify_wq)) {
1882 err = PTR_ERR(osdc->notify_wq);
1883 osdc->notify_wq = NULL;
1889 ceph_msgpool_destroy(&osdc->msgpool_op);
1891 mempool_destroy(osdc->req_mempool);
1895 EXPORT_SYMBOL(ceph_osdc_init);
1897 void ceph_osdc_stop(struct ceph_osd_client *osdc)
1899 flush_workqueue(osdc->notify_wq);
1900 destroy_workqueue(osdc->notify_wq);
1901 cancel_delayed_work_sync(&osdc->timeout_work);
1902 cancel_delayed_work_sync(&osdc->osds_timeout_work);
1904 ceph_osdmap_destroy(osdc->osdmap);
1905 osdc->osdmap = NULL;
1907 remove_all_osds(osdc);
1908 mempool_destroy(osdc->req_mempool);
1909 ceph_msgpool_destroy(&osdc->msgpool_op);
1910 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
1912 EXPORT_SYMBOL(ceph_osdc_stop);
1915 * Read some contiguous pages. If we cross a stripe boundary, shorten
1916 * *plen. Return number of bytes read, or error.
1918 int ceph_osdc_readpages(struct ceph_osd_client *osdc,
1919 struct ceph_vino vino, struct ceph_file_layout *layout,
1921 u32 truncate_seq, u64 truncate_size,
1922 struct page **pages, int num_pages, int page_align)
1924 struct ceph_osd_request *req;
1927 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
1928 vino.snap, off, *plen);
1929 req = ceph_osdc_new_request(osdc, layout, vino, off, plen,
1930 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
1931 NULL, 0, truncate_seq, truncate_size, NULL,
1932 false, 1, page_align);
1934 return PTR_ERR(req);
1936 /* it may be a short read due to an object boundary */
1937 req->r_pages = pages;
1939 dout("readpages final extent is %llu~%llu (%d pages align %d)\n",
1940 off, *plen, req->r_num_pages, page_align);
1942 rc = ceph_osdc_start_request(osdc, req, false);
1944 rc = ceph_osdc_wait_request(osdc, req);
1946 ceph_osdc_put_request(req);
1947 dout("readpages result %d\n", rc);
1950 EXPORT_SYMBOL(ceph_osdc_readpages);
1953 * do a synchronous write on N pages
1955 int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
1956 struct ceph_file_layout *layout,
1957 struct ceph_snap_context *snapc,
1959 u32 truncate_seq, u64 truncate_size,
1960 struct timespec *mtime,
1961 struct page **pages, int num_pages,
1962 int flags, int do_sync, bool nofail)
1964 struct ceph_osd_request *req;
1966 int page_align = off & ~PAGE_MASK;
1968 BUG_ON(vino.snap != CEPH_NOSNAP);
1969 req = ceph_osdc_new_request(osdc, layout, vino, off, &len,
1971 flags | CEPH_OSD_FLAG_ONDISK |
1972 CEPH_OSD_FLAG_WRITE,
1974 truncate_seq, truncate_size, mtime,
1975 nofail, 1, page_align);
1977 return PTR_ERR(req);
1979 /* it may be a short write due to an object boundary */
1980 req->r_pages = pages;
1981 dout("writepages %llu~%llu (%d pages)\n", off, len,
1984 rc = ceph_osdc_start_request(osdc, req, nofail);
1986 rc = ceph_osdc_wait_request(osdc, req);
1988 ceph_osdc_put_request(req);
1991 dout("writepages result %d\n", rc);
1994 EXPORT_SYMBOL(ceph_osdc_writepages);
1997 * handle incoming message
1999 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
2001 struct ceph_osd *osd = con->private;
2002 struct ceph_osd_client *osdc;
2003 int type = le16_to_cpu(msg->hdr.type);
2010 case CEPH_MSG_OSD_MAP:
2011 ceph_osdc_handle_map(osdc, msg);
2013 case CEPH_MSG_OSD_OPREPLY:
2014 handle_reply(osdc, msg, con);
2016 case CEPH_MSG_WATCH_NOTIFY:
2017 handle_watch_notify(osdc, msg);
2021 pr_err("received unknown message type %d %s\n", type,
2022 ceph_msg_type_name(type));
2029 * lookup and return message for incoming reply. set up reply message
2032 static struct ceph_msg *get_reply(struct ceph_connection *con,
2033 struct ceph_msg_header *hdr,
2036 struct ceph_osd *osd = con->private;
2037 struct ceph_osd_client *osdc = osd->o_osdc;
2039 struct ceph_osd_request *req;
2040 int front = le32_to_cpu(hdr->front_len);
2041 int data_len = le32_to_cpu(hdr->data_len);
2044 tid = le64_to_cpu(hdr->tid);
2045 mutex_lock(&osdc->request_mutex);
2046 req = __lookup_request(osdc, tid);
2050 dout("get_reply unknown tid %llu from osd%d\n", tid,
2055 if (req->r_con_filling_msg) {
2056 dout("%s revoking msg %p from old con %p\n", __func__,
2057 req->r_reply, req->r_con_filling_msg);
2058 ceph_msg_revoke_incoming(req->r_reply);
2059 req->r_con_filling_msg->ops->put(req->r_con_filling_msg);
2060 req->r_con_filling_msg = NULL;
2063 if (front > req->r_reply->front.iov_len) {
2064 pr_warning("get_reply front %d > preallocated %d\n",
2065 front, (int)req->r_reply->front.iov_len);
2066 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front, GFP_NOFS, false);
2069 ceph_msg_put(req->r_reply);
2072 m = ceph_msg_get(req->r_reply);
2075 int want = calc_pages_for(req->r_page_alignment, data_len);
2077 if (unlikely(req->r_num_pages < want)) {
2078 pr_warning("tid %lld reply has %d bytes %d pages, we"
2079 " had only %d pages ready\n", tid, data_len,
2080 want, req->r_num_pages);
2086 m->pages = req->r_pages;
2087 m->nr_pages = req->r_num_pages;
2088 m->page_alignment = req->r_page_alignment;
2090 m->bio = req->r_bio;
2094 req->r_con_filling_msg = con->ops->get(con);
2095 dout("get_reply tid %lld %p\n", tid, m);
2098 mutex_unlock(&osdc->request_mutex);
2103 static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2104 struct ceph_msg_header *hdr,
2107 struct ceph_osd *osd = con->private;
2108 int type = le16_to_cpu(hdr->type);
2109 int front = le32_to_cpu(hdr->front_len);
2113 case CEPH_MSG_OSD_MAP:
2114 case CEPH_MSG_WATCH_NOTIFY:
2115 return ceph_msg_new(type, front, GFP_NOFS, false);
2116 case CEPH_MSG_OSD_OPREPLY:
2117 return get_reply(con, hdr, skip);
2119 pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
2127 * Wrappers to refcount containing ceph_osd struct
2129 static struct ceph_connection *get_osd_con(struct ceph_connection *con)
2131 struct ceph_osd *osd = con->private;
2137 static void put_osd_con(struct ceph_connection *con)
2139 struct ceph_osd *osd = con->private;
2147 * Note: returned pointer is the address of a structure that's
2148 * managed separately. Caller must *not* attempt to free it.
2150 static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
2151 int *proto, int force_new)
2153 struct ceph_osd *o = con->private;
2154 struct ceph_osd_client *osdc = o->o_osdc;
2155 struct ceph_auth_client *ac = osdc->client->monc.auth;
2156 struct ceph_auth_handshake *auth = &o->o_auth;
2158 if (force_new && auth->authorizer) {
2159 if (ac->ops && ac->ops->destroy_authorizer)
2160 ac->ops->destroy_authorizer(ac, auth->authorizer);
2161 auth->authorizer = NULL;
2163 if (!auth->authorizer && ac->ops && ac->ops->create_authorizer) {
2164 int ret = ac->ops->create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2167 return ERR_PTR(ret);
2169 *proto = ac->protocol;
2175 static int verify_authorizer_reply(struct ceph_connection *con, int len)
2177 struct ceph_osd *o = con->private;
2178 struct ceph_osd_client *osdc = o->o_osdc;
2179 struct ceph_auth_client *ac = osdc->client->monc.auth;
2182 * XXX If ac->ops or ac->ops->verify_authorizer_reply is null,
2183 * XXX which do we do: succeed or fail?
2185 return ac->ops->verify_authorizer_reply(ac, o->o_auth.authorizer, len);
2188 static int invalidate_authorizer(struct ceph_connection *con)
2190 struct ceph_osd *o = con->private;
2191 struct ceph_osd_client *osdc = o->o_osdc;
2192 struct ceph_auth_client *ac = osdc->client->monc.auth;
2194 if (ac->ops && ac->ops->invalidate_authorizer)
2195 ac->ops->invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
2197 return ceph_monc_validate_auth(&osdc->client->monc);
2200 static const struct ceph_connection_operations osd_con_ops = {
2203 .dispatch = dispatch,
2204 .get_authorizer = get_authorizer,
2205 .verify_authorizer_reply = verify_authorizer_reply,
2206 .invalidate_authorizer = invalidate_authorizer,
2207 .alloc_msg = alloc_msg,