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);
36 * Implement client access to distributed object storage cluster.
38 * All data objects are stored within a cluster/cloud of OSDs, or
39 * "object storage devices." (Note that Ceph OSDs have _nothing_ to
40 * do with the T10 OSD extensions to SCSI.) Ceph OSDs are simply
41 * remote daemons serving up and coordinating consistent and safe
44 * Cluster membership and the mapping of data objects onto storage devices
45 * are described by the osd map.
47 * We keep track of pending OSD requests (read, write), resubmit
48 * requests to different OSDs when the cluster topology/data layout
49 * change, or retry the affected requests when the communications
50 * channel with an OSD is reset.
54 * calculate the mapping of a file extent onto an object, and fill out the
55 * request accordingly. shorten extent as necessary if it crosses an
58 * fill osd op in request message.
60 static int calc_layout(struct ceph_file_layout *layout, u64 off, u64 *plen,
61 u64 *objnum, u64 *objoff, u64 *objlen)
67 r = ceph_calc_file_object_mapping(layout, off, orig_len, objnum,
71 if (*objlen < orig_len) {
73 dout(" skipping last %llu, final file extent %llu~%llu\n",
74 orig_len - *plen, off, *plen);
77 dout("calc_layout objnum=%llx %llu~%llu\n", *objnum, *objoff, *objlen);
82 static void ceph_osd_data_init(struct ceph_osd_data *osd_data)
84 memset(osd_data, 0, sizeof (*osd_data));
85 osd_data->type = CEPH_OSD_DATA_TYPE_NONE;
88 void ceph_osd_data_pages_init(struct ceph_osd_data *osd_data,
89 struct page **pages, u64 length, u32 alignment,
90 bool pages_from_pool, bool own_pages)
92 osd_data->type = CEPH_OSD_DATA_TYPE_PAGES;
93 osd_data->pages = pages;
94 osd_data->length = length;
95 osd_data->alignment = alignment;
96 osd_data->pages_from_pool = pages_from_pool;
97 osd_data->own_pages = own_pages;
99 EXPORT_SYMBOL(ceph_osd_data_pages_init);
101 void ceph_osd_data_pagelist_init(struct ceph_osd_data *osd_data,
102 struct ceph_pagelist *pagelist)
104 osd_data->type = CEPH_OSD_DATA_TYPE_PAGELIST;
105 osd_data->pagelist = pagelist;
107 EXPORT_SYMBOL(ceph_osd_data_pagelist_init);
110 void ceph_osd_data_bio_init(struct ceph_osd_data *osd_data,
111 struct bio *bio, size_t bio_length)
113 osd_data->type = CEPH_OSD_DATA_TYPE_BIO;
115 osd_data->bio_length = bio_length;
117 EXPORT_SYMBOL(ceph_osd_data_bio_init);
118 #endif /* CONFIG_BLOCK */
120 static u64 ceph_osd_data_length(struct ceph_osd_data *osd_data)
122 switch (osd_data->type) {
123 case CEPH_OSD_DATA_TYPE_NONE:
125 case CEPH_OSD_DATA_TYPE_PAGES:
126 return osd_data->length;
127 case CEPH_OSD_DATA_TYPE_PAGELIST:
128 return (u64)osd_data->pagelist->length;
130 case CEPH_OSD_DATA_TYPE_BIO:
131 return (u64)osd_data->bio_length;
132 #endif /* CONFIG_BLOCK */
134 WARN(true, "unrecognized data type %d\n", (int)osd_data->type);
139 static void ceph_osd_data_release(struct ceph_osd_data *osd_data)
141 if (osd_data->type != CEPH_OSD_DATA_TYPE_PAGES)
144 if (osd_data->own_pages) {
147 num_pages = calc_pages_for((u64)osd_data->alignment,
148 (u64)osd_data->length);
149 ceph_release_page_vector(osd_data->pages, num_pages);
156 void ceph_osdc_release_request(struct kref *kref)
158 struct ceph_osd_request *req;
160 req = container_of(kref, struct ceph_osd_request, r_kref);
162 ceph_msg_put(req->r_request);
164 ceph_msg_revoke_incoming(req->r_reply);
165 ceph_msg_put(req->r_reply);
168 ceph_osd_data_release(&req->r_data_in);
169 ceph_osd_data_release(&req->r_data_out);
171 ceph_put_snap_context(req->r_snapc);
173 mempool_free(req, req->r_osdc->req_mempool);
177 EXPORT_SYMBOL(ceph_osdc_release_request);
179 struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
180 struct ceph_snap_context *snapc,
181 unsigned int num_ops,
185 struct ceph_osd_request *req;
186 struct ceph_msg *msg;
189 BUILD_BUG_ON(CEPH_OSD_MAX_OP > U16_MAX);
190 BUG_ON(num_ops > CEPH_OSD_MAX_OP);
192 msg_size = 4 + 4 + 8 + 8 + 4+8;
193 msg_size += 2 + 4 + 8 + 4 + 4; /* oloc */
194 msg_size += 1 + 8 + 4 + 4; /* pg_t */
195 msg_size += 4 + MAX_OBJ_NAME_SIZE;
196 msg_size += 2 + num_ops*sizeof(struct ceph_osd_op);
197 msg_size += 8; /* snapid */
198 msg_size += 8; /* snap_seq */
199 msg_size += 8 * (snapc ? snapc->num_snaps : 0); /* snaps */
203 req = mempool_alloc(osdc->req_mempool, gfp_flags);
204 memset(req, 0, sizeof(*req));
206 req = kzalloc(sizeof(*req), gfp_flags);
212 req->r_mempool = use_mempool;
213 req->r_num_ops = num_ops;
215 kref_init(&req->r_kref);
216 init_completion(&req->r_completion);
217 init_completion(&req->r_safe_completion);
218 RB_CLEAR_NODE(&req->r_node);
219 INIT_LIST_HEAD(&req->r_unsafe_item);
220 INIT_LIST_HEAD(&req->r_linger_item);
221 INIT_LIST_HEAD(&req->r_linger_osd);
222 INIT_LIST_HEAD(&req->r_req_lru_item);
223 INIT_LIST_HEAD(&req->r_osd_item);
225 /* create reply message */
227 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0);
229 msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY,
230 OSD_OPREPLY_FRONT_LEN, gfp_flags, true);
232 ceph_osdc_put_request(req);
237 ceph_osd_data_init(&req->r_data_in);
238 ceph_osd_data_init(&req->r_data_out);
240 /* create request message; allow space for oid */
242 msg = ceph_msgpool_get(&osdc->msgpool_op, 0);
244 msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp_flags, true);
246 ceph_osdc_put_request(req);
250 memset(msg->front.iov_base, 0, msg->front.iov_len);
252 req->r_request = msg;
256 EXPORT_SYMBOL(ceph_osdc_alloc_request);
258 static bool osd_req_opcode_valid(u16 opcode)
261 case CEPH_OSD_OP_READ:
262 case CEPH_OSD_OP_STAT:
263 case CEPH_OSD_OP_MAPEXT:
264 case CEPH_OSD_OP_MASKTRUNC:
265 case CEPH_OSD_OP_SPARSE_READ:
266 case CEPH_OSD_OP_NOTIFY:
267 case CEPH_OSD_OP_NOTIFY_ACK:
268 case CEPH_OSD_OP_ASSERT_VER:
269 case CEPH_OSD_OP_WRITE:
270 case CEPH_OSD_OP_WRITEFULL:
271 case CEPH_OSD_OP_TRUNCATE:
272 case CEPH_OSD_OP_ZERO:
273 case CEPH_OSD_OP_DELETE:
274 case CEPH_OSD_OP_APPEND:
275 case CEPH_OSD_OP_STARTSYNC:
276 case CEPH_OSD_OP_SETTRUNC:
277 case CEPH_OSD_OP_TRIMTRUNC:
278 case CEPH_OSD_OP_TMAPUP:
279 case CEPH_OSD_OP_TMAPPUT:
280 case CEPH_OSD_OP_TMAPGET:
281 case CEPH_OSD_OP_CREATE:
282 case CEPH_OSD_OP_ROLLBACK:
283 case CEPH_OSD_OP_WATCH:
284 case CEPH_OSD_OP_OMAPGETKEYS:
285 case CEPH_OSD_OP_OMAPGETVALS:
286 case CEPH_OSD_OP_OMAPGETHEADER:
287 case CEPH_OSD_OP_OMAPGETVALSBYKEYS:
288 case CEPH_OSD_OP_OMAPSETVALS:
289 case CEPH_OSD_OP_OMAPSETHEADER:
290 case CEPH_OSD_OP_OMAPCLEAR:
291 case CEPH_OSD_OP_OMAPRMKEYS:
292 case CEPH_OSD_OP_OMAP_CMP:
293 case CEPH_OSD_OP_CLONERANGE:
294 case CEPH_OSD_OP_ASSERT_SRC_VERSION:
295 case CEPH_OSD_OP_SRC_CMPXATTR:
296 case CEPH_OSD_OP_GETXATTR:
297 case CEPH_OSD_OP_GETXATTRS:
298 case CEPH_OSD_OP_CMPXATTR:
299 case CEPH_OSD_OP_SETXATTR:
300 case CEPH_OSD_OP_SETXATTRS:
301 case CEPH_OSD_OP_RESETXATTRS:
302 case CEPH_OSD_OP_RMXATTR:
303 case CEPH_OSD_OP_PULL:
304 case CEPH_OSD_OP_PUSH:
305 case CEPH_OSD_OP_BALANCEREADS:
306 case CEPH_OSD_OP_UNBALANCEREADS:
307 case CEPH_OSD_OP_SCRUB:
308 case CEPH_OSD_OP_SCRUB_RESERVE:
309 case CEPH_OSD_OP_SCRUB_UNRESERVE:
310 case CEPH_OSD_OP_SCRUB_STOP:
311 case CEPH_OSD_OP_SCRUB_MAP:
312 case CEPH_OSD_OP_WRLOCK:
313 case CEPH_OSD_OP_WRUNLOCK:
314 case CEPH_OSD_OP_RDLOCK:
315 case CEPH_OSD_OP_RDUNLOCK:
316 case CEPH_OSD_OP_UPLOCK:
317 case CEPH_OSD_OP_DNLOCK:
318 case CEPH_OSD_OP_CALL:
319 case CEPH_OSD_OP_PGLS:
320 case CEPH_OSD_OP_PGLS_FILTER:
328 * This is an osd op init function for opcodes that have no data or
329 * other information associated with them. It also serves as a
330 * common init routine for all the other init functions, below.
332 static struct ceph_osd_req_op *
333 osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which,
336 struct ceph_osd_req_op *op;
338 BUG_ON(which >= osd_req->r_num_ops);
339 BUG_ON(!osd_req_opcode_valid(opcode));
341 op = &osd_req->r_ops[which];
342 memset(op, 0, sizeof (*op));
348 void osd_req_op_extent_init(struct ceph_osd_request *osd_req,
349 unsigned int which, u16 opcode,
350 u64 offset, u64 length,
351 u64 truncate_size, u32 truncate_seq)
353 struct ceph_osd_req_op *op = osd_req_op_init(osd_req, which, opcode);
354 size_t payload_len = 0;
356 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE);
358 op->extent.offset = offset;
359 op->extent.length = length;
360 op->extent.truncate_size = truncate_size;
361 op->extent.truncate_seq = truncate_seq;
362 if (opcode == CEPH_OSD_OP_WRITE)
363 payload_len += length;
365 op->payload_len = payload_len;
367 EXPORT_SYMBOL(osd_req_op_extent_init);
369 void osd_req_op_extent_update(struct ceph_osd_request *osd_req,
370 unsigned int which, u64 length)
372 struct ceph_osd_req_op *op;
375 BUG_ON(which >= osd_req->r_num_ops);
376 op = &osd_req->r_ops[which];
377 previous = op->extent.length;
379 if (length == previous)
380 return; /* Nothing to do */
381 BUG_ON(length > previous);
383 op->extent.length = length;
384 op->payload_len -= previous - length;
386 EXPORT_SYMBOL(osd_req_op_extent_update);
388 void osd_req_op_extent_osd_data(struct ceph_osd_request *osd_req,
390 struct ceph_osd_data *osd_data)
392 BUG_ON(which >= osd_req->r_num_ops);
393 osd_req->r_ops[which].extent.osd_data = osd_data;
395 EXPORT_SYMBOL(osd_req_op_extent_osd_data);
397 void osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
398 u16 opcode, const char *class, const char *method,
399 const void *request_data, size_t request_data_size)
401 struct ceph_osd_req_op *op = osd_req_op_init(osd_req, which, opcode);
402 struct ceph_pagelist *pagelist;
403 size_t payload_len = 0;
406 BUG_ON(opcode != CEPH_OSD_OP_CALL);
408 pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
410 ceph_pagelist_init(pagelist);
412 op->cls.class_name = class;
413 size = strlen(class);
414 BUG_ON(size > (size_t) U8_MAX);
415 op->cls.class_len = size;
416 ceph_pagelist_append(pagelist, class, size);
419 op->cls.method_name = method;
420 size = strlen(method);
421 BUG_ON(size > (size_t) U8_MAX);
422 op->cls.method_len = size;
423 ceph_pagelist_append(pagelist, method, size);
426 op->cls.request_data = request_data;
427 BUG_ON(request_data_size > (size_t) U32_MAX);
428 op->cls.request_data_len = (u32) request_data_size;
429 ceph_pagelist_append(pagelist, request_data, request_data_size);
430 payload_len += request_data_size;
432 op->cls.request_info = &osd_req->r_data_out;
433 ceph_osd_data_pagelist_init(op->cls.request_info, pagelist);
435 op->cls.argc = 0; /* currently unused */
437 op->payload_len = payload_len;
439 EXPORT_SYMBOL(osd_req_op_cls_init);
440 void osd_req_op_cls_response_data(struct ceph_osd_request *osd_req,
442 struct ceph_osd_data *response_data)
444 BUG_ON(which >= osd_req->r_num_ops);
445 osd_req->r_ops[which].cls.response_data = response_data;
447 EXPORT_SYMBOL(osd_req_op_cls_response_data);
449 void osd_req_op_watch_init(struct ceph_osd_request *osd_req,
450 unsigned int which, u16 opcode,
451 u64 cookie, u64 version, int flag)
453 struct ceph_osd_req_op *op = osd_req_op_init(osd_req, which, opcode);
455 BUG_ON(opcode != CEPH_OSD_OP_NOTIFY_ACK && opcode != CEPH_OSD_OP_WATCH);
457 op->watch.cookie = cookie;
458 /* op->watch.ver = version; */ /* XXX 3847 */
459 op->watch.ver = cpu_to_le64(version);
460 if (opcode == CEPH_OSD_OP_WATCH && flag)
461 op->watch.flag = (u8)1;
463 EXPORT_SYMBOL(osd_req_op_watch_init);
465 static u64 osd_req_encode_op(struct ceph_osd_request *req,
466 struct ceph_osd_op *dst, unsigned int which)
468 struct ceph_osd_req_op *src;
469 u64 request_data_len = 0;
471 BUG_ON(which >= req->r_num_ops);
472 src = &req->r_ops[which];
473 if (WARN_ON(!osd_req_opcode_valid(src->op))) {
474 pr_err("unrecognized osd opcode %d\n", src->op);
480 case CEPH_OSD_OP_STAT:
482 case CEPH_OSD_OP_READ:
483 case CEPH_OSD_OP_WRITE:
484 if (src->op == CEPH_OSD_OP_WRITE)
485 request_data_len = src->extent.length;
486 dst->extent.offset = cpu_to_le64(src->extent.offset);
487 dst->extent.length = cpu_to_le64(src->extent.length);
488 dst->extent.truncate_size =
489 cpu_to_le64(src->extent.truncate_size);
490 dst->extent.truncate_seq =
491 cpu_to_le32(src->extent.truncate_seq);
492 if (src->op == CEPH_OSD_OP_WRITE)
493 WARN_ON(src->extent.osd_data != &req->r_data_out);
495 WARN_ON(src->extent.osd_data != &req->r_data_in);
497 case CEPH_OSD_OP_CALL:
498 dst->cls.class_len = src->cls.class_len;
499 dst->cls.method_len = src->cls.method_len;
500 dst->cls.indata_len = cpu_to_le32(src->cls.request_data_len);
501 WARN_ON(src->cls.response_data != &req->r_data_in);
502 WARN_ON(src->cls.request_info != &req->r_data_out);
503 BUG_ON(src->cls.request_info->type !=
504 CEPH_OSD_DATA_TYPE_PAGELIST);
505 request_data_len = src->cls.request_info->pagelist->length;
507 case CEPH_OSD_OP_STARTSYNC:
509 case CEPH_OSD_OP_NOTIFY_ACK:
510 case CEPH_OSD_OP_WATCH:
511 dst->watch.cookie = cpu_to_le64(src->watch.cookie);
512 dst->watch.ver = cpu_to_le64(src->watch.ver);
513 dst->watch.flag = src->watch.flag;
516 pr_err("unsupported osd opcode %s\n",
517 ceph_osd_op_name(src->op));
522 dst->op = cpu_to_le16(src->op);
523 dst->payload_len = cpu_to_le32(src->payload_len);
525 return request_data_len;
529 * build new request AND message, calculate layout, and adjust file
532 * if the file was recently truncated, we include information about its
533 * old and new size so that the object can be updated appropriately. (we
534 * avoid synchronously deleting truncated objects because it's slow.)
536 * if @do_sync, include a 'startsync' command so that the osd will flush
539 struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
540 struct ceph_file_layout *layout,
541 struct ceph_vino vino,
542 u64 off, u64 *plen, int num_ops,
543 int opcode, int flags,
544 struct ceph_snap_context *snapc,
549 struct ceph_osd_request *req;
550 struct ceph_osd_data *osd_data;
558 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE);
560 req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool,
563 return ERR_PTR(-ENOMEM);
564 osd_data = opcode == CEPH_OSD_OP_WRITE ? &req->r_data_out
567 req->r_flags = flags;
569 /* calculate max write size */
570 r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
572 ceph_osdc_put_request(req);
576 object_size = le32_to_cpu(layout->fl_object_size);
577 object_base = off - objoff;
578 if (truncate_size <= object_base) {
581 truncate_size -= object_base;
582 if (truncate_size > object_size)
583 truncate_size = object_size;
586 osd_req_op_extent_init(req, 0, opcode, objoff, objlen,
587 truncate_size, truncate_seq);
588 osd_req_op_extent_osd_data(req, 0, osd_data);
591 * A second op in the ops array means the caller wants to
592 * also issue a include a 'startsync' command so that the
593 * osd will flush data quickly.
596 osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC);
598 req->r_file_layout = *layout; /* keep a copy */
600 snprintf(req->r_oid, sizeof(req->r_oid), "%llx.%08llx",
602 req->r_oid_len = strlen(req->r_oid);
606 EXPORT_SYMBOL(ceph_osdc_new_request);
609 * We keep osd requests in an rbtree, sorted by ->r_tid.
611 static void __insert_request(struct ceph_osd_client *osdc,
612 struct ceph_osd_request *new)
614 struct rb_node **p = &osdc->requests.rb_node;
615 struct rb_node *parent = NULL;
616 struct ceph_osd_request *req = NULL;
620 req = rb_entry(parent, struct ceph_osd_request, r_node);
621 if (new->r_tid < req->r_tid)
623 else if (new->r_tid > req->r_tid)
629 rb_link_node(&new->r_node, parent, p);
630 rb_insert_color(&new->r_node, &osdc->requests);
633 static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc,
636 struct ceph_osd_request *req;
637 struct rb_node *n = osdc->requests.rb_node;
640 req = rb_entry(n, struct ceph_osd_request, r_node);
641 if (tid < req->r_tid)
643 else if (tid > req->r_tid)
651 static struct ceph_osd_request *
652 __lookup_request_ge(struct ceph_osd_client *osdc,
655 struct ceph_osd_request *req;
656 struct rb_node *n = osdc->requests.rb_node;
659 req = rb_entry(n, struct ceph_osd_request, r_node);
660 if (tid < req->r_tid) {
664 } else if (tid > req->r_tid) {
674 * Resubmit requests pending on the given osd.
676 static void __kick_osd_requests(struct ceph_osd_client *osdc,
677 struct ceph_osd *osd)
679 struct ceph_osd_request *req, *nreq;
683 dout("__kick_osd_requests osd%d\n", osd->o_osd);
684 err = __reset_osd(osdc, osd);
688 * Build up a list of requests to resend by traversing the
689 * osd's list of requests. Requests for a given object are
690 * sent in tid order, and that is also the order they're
691 * kept on this list. Therefore all requests that are in
692 * flight will be found first, followed by all requests that
693 * have not yet been sent. And to resend requests while
694 * preserving this order we will want to put any sent
695 * requests back on the front of the osd client's unsent
698 * So we build a separate ordered list of already-sent
699 * requests for the affected osd and splice it onto the
700 * front of the osd client's unsent list. Once we've seen a
701 * request that has not yet been sent we're done. Those
702 * requests are already sitting right where they belong.
704 list_for_each_entry(req, &osd->o_requests, r_osd_item) {
707 list_move_tail(&req->r_req_lru_item, &resend);
708 dout("requeueing %p tid %llu osd%d\n", req, req->r_tid,
711 req->r_flags |= CEPH_OSD_FLAG_RETRY;
713 list_splice(&resend, &osdc->req_unsent);
716 * Linger requests are re-registered before sending, which
717 * sets up a new tid for each. We add them to the unsent
718 * list at the end to keep things in tid order.
720 list_for_each_entry_safe(req, nreq, &osd->o_linger_requests,
723 * reregister request prior to unregistering linger so
724 * that r_osd is preserved.
726 BUG_ON(!list_empty(&req->r_req_lru_item));
727 __register_request(osdc, req);
728 list_add_tail(&req->r_req_lru_item, &osdc->req_unsent);
729 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
730 __unregister_linger_request(osdc, req);
731 dout("requeued lingering %p tid %llu osd%d\n", req, req->r_tid,
737 * If the osd connection drops, we need to resubmit all requests.
739 static void osd_reset(struct ceph_connection *con)
741 struct ceph_osd *osd = con->private;
742 struct ceph_osd_client *osdc;
746 dout("osd_reset osd%d\n", osd->o_osd);
748 down_read(&osdc->map_sem);
749 mutex_lock(&osdc->request_mutex);
750 __kick_osd_requests(osdc, osd);
752 mutex_unlock(&osdc->request_mutex);
753 up_read(&osdc->map_sem);
757 * Track open sessions with osds.
759 static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
761 struct ceph_osd *osd;
763 osd = kzalloc(sizeof(*osd), GFP_NOFS);
767 atomic_set(&osd->o_ref, 1);
770 RB_CLEAR_NODE(&osd->o_node);
771 INIT_LIST_HEAD(&osd->o_requests);
772 INIT_LIST_HEAD(&osd->o_linger_requests);
773 INIT_LIST_HEAD(&osd->o_osd_lru);
774 osd->o_incarnation = 1;
776 ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
778 INIT_LIST_HEAD(&osd->o_keepalive_item);
782 static struct ceph_osd *get_osd(struct ceph_osd *osd)
784 if (atomic_inc_not_zero(&osd->o_ref)) {
785 dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
786 atomic_read(&osd->o_ref));
789 dout("get_osd %p FAIL\n", osd);
794 static void put_osd(struct ceph_osd *osd)
796 dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
797 atomic_read(&osd->o_ref) - 1);
798 if (atomic_dec_and_test(&osd->o_ref) && osd->o_auth.authorizer) {
799 struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth;
801 ceph_auth_destroy_authorizer(ac, osd->o_auth.authorizer);
807 * remove an osd from our map
809 static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
811 dout("__remove_osd %p\n", osd);
812 BUG_ON(!list_empty(&osd->o_requests));
813 rb_erase(&osd->o_node, &osdc->osds);
814 list_del_init(&osd->o_osd_lru);
815 ceph_con_close(&osd->o_con);
819 static void remove_all_osds(struct ceph_osd_client *osdc)
821 dout("%s %p\n", __func__, osdc);
822 mutex_lock(&osdc->request_mutex);
823 while (!RB_EMPTY_ROOT(&osdc->osds)) {
824 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
825 struct ceph_osd, o_node);
826 __remove_osd(osdc, osd);
828 mutex_unlock(&osdc->request_mutex);
831 static void __move_osd_to_lru(struct ceph_osd_client *osdc,
832 struct ceph_osd *osd)
834 dout("__move_osd_to_lru %p\n", osd);
835 BUG_ON(!list_empty(&osd->o_osd_lru));
836 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
837 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl * HZ;
840 static void __remove_osd_from_lru(struct ceph_osd *osd)
842 dout("__remove_osd_from_lru %p\n", osd);
843 if (!list_empty(&osd->o_osd_lru))
844 list_del_init(&osd->o_osd_lru);
847 static void remove_old_osds(struct ceph_osd_client *osdc)
849 struct ceph_osd *osd, *nosd;
851 dout("__remove_old_osds %p\n", osdc);
852 mutex_lock(&osdc->request_mutex);
853 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
854 if (time_before(jiffies, osd->lru_ttl))
856 __remove_osd(osdc, osd);
858 mutex_unlock(&osdc->request_mutex);
864 static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
866 struct ceph_entity_addr *peer_addr;
868 dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
869 if (list_empty(&osd->o_requests) &&
870 list_empty(&osd->o_linger_requests)) {
871 __remove_osd(osdc, osd);
876 peer_addr = &osdc->osdmap->osd_addr[osd->o_osd];
877 if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
878 !ceph_con_opened(&osd->o_con)) {
879 struct ceph_osd_request *req;
881 dout(" osd addr hasn't changed and connection never opened,"
882 " letting msgr retry");
883 /* touch each r_stamp for handle_timeout()'s benfit */
884 list_for_each_entry(req, &osd->o_requests, r_osd_item)
885 req->r_stamp = jiffies;
890 ceph_con_close(&osd->o_con);
891 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
892 osd->o_incarnation++;
897 static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new)
899 struct rb_node **p = &osdc->osds.rb_node;
900 struct rb_node *parent = NULL;
901 struct ceph_osd *osd = NULL;
903 dout("__insert_osd %p osd%d\n", new, new->o_osd);
906 osd = rb_entry(parent, struct ceph_osd, o_node);
907 if (new->o_osd < osd->o_osd)
909 else if (new->o_osd > osd->o_osd)
915 rb_link_node(&new->o_node, parent, p);
916 rb_insert_color(&new->o_node, &osdc->osds);
919 static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o)
921 struct ceph_osd *osd;
922 struct rb_node *n = osdc->osds.rb_node;
925 osd = rb_entry(n, struct ceph_osd, o_node);
928 else if (o > osd->o_osd)
936 static void __schedule_osd_timeout(struct ceph_osd_client *osdc)
938 schedule_delayed_work(&osdc->timeout_work,
939 osdc->client->options->osd_keepalive_timeout * HZ);
942 static void __cancel_osd_timeout(struct ceph_osd_client *osdc)
944 cancel_delayed_work(&osdc->timeout_work);
948 * Register request, assign tid. If this is the first request, set up
951 static void __register_request(struct ceph_osd_client *osdc,
952 struct ceph_osd_request *req)
954 req->r_tid = ++osdc->last_tid;
955 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
956 dout("__register_request %p tid %lld\n", req, req->r_tid);
957 __insert_request(osdc, req);
958 ceph_osdc_get_request(req);
959 osdc->num_requests++;
960 if (osdc->num_requests == 1) {
961 dout(" first request, scheduling timeout\n");
962 __schedule_osd_timeout(osdc);
967 * called under osdc->request_mutex
969 static void __unregister_request(struct ceph_osd_client *osdc,
970 struct ceph_osd_request *req)
972 if (RB_EMPTY_NODE(&req->r_node)) {
973 dout("__unregister_request %p tid %lld not registered\n",
978 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
979 rb_erase(&req->r_node, &osdc->requests);
980 osdc->num_requests--;
983 /* make sure the original request isn't in flight. */
984 ceph_msg_revoke(req->r_request);
986 list_del_init(&req->r_osd_item);
987 if (list_empty(&req->r_osd->o_requests) &&
988 list_empty(&req->r_osd->o_linger_requests)) {
989 dout("moving osd to %p lru\n", req->r_osd);
990 __move_osd_to_lru(osdc, req->r_osd);
992 if (list_empty(&req->r_linger_item))
996 list_del_init(&req->r_req_lru_item);
997 ceph_osdc_put_request(req);
999 if (osdc->num_requests == 0) {
1000 dout(" no requests, canceling timeout\n");
1001 __cancel_osd_timeout(osdc);
1006 * Cancel a previously queued request message
1008 static void __cancel_request(struct ceph_osd_request *req)
1010 if (req->r_sent && req->r_osd) {
1011 ceph_msg_revoke(req->r_request);
1016 static void __register_linger_request(struct ceph_osd_client *osdc,
1017 struct ceph_osd_request *req)
1019 dout("__register_linger_request %p\n", req);
1020 list_add_tail(&req->r_linger_item, &osdc->req_linger);
1022 list_add_tail(&req->r_linger_osd,
1023 &req->r_osd->o_linger_requests);
1026 static void __unregister_linger_request(struct ceph_osd_client *osdc,
1027 struct ceph_osd_request *req)
1029 dout("__unregister_linger_request %p\n", req);
1030 list_del_init(&req->r_linger_item);
1032 list_del_init(&req->r_linger_osd);
1034 if (list_empty(&req->r_osd->o_requests) &&
1035 list_empty(&req->r_osd->o_linger_requests)) {
1036 dout("moving osd to %p lru\n", req->r_osd);
1037 __move_osd_to_lru(osdc, req->r_osd);
1039 if (list_empty(&req->r_osd_item))
1044 void ceph_osdc_unregister_linger_request(struct ceph_osd_client *osdc,
1045 struct ceph_osd_request *req)
1047 mutex_lock(&osdc->request_mutex);
1048 if (req->r_linger) {
1049 __unregister_linger_request(osdc, req);
1050 ceph_osdc_put_request(req);
1052 mutex_unlock(&osdc->request_mutex);
1054 EXPORT_SYMBOL(ceph_osdc_unregister_linger_request);
1056 void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
1057 struct ceph_osd_request *req)
1059 if (!req->r_linger) {
1060 dout("set_request_linger %p\n", req);
1063 * caller is now responsible for calling
1064 * unregister_linger_request
1066 ceph_osdc_get_request(req);
1069 EXPORT_SYMBOL(ceph_osdc_set_request_linger);
1072 * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
1073 * (as needed), and set the request r_osd appropriately. If there is
1074 * no up osd, set r_osd to NULL. Move the request to the appropriate list
1075 * (unsent, homeless) or leave on in-flight lru.
1077 * Return 0 if unchanged, 1 if changed, or negative on error.
1079 * Caller should hold map_sem for read and request_mutex.
1081 static int __map_request(struct ceph_osd_client *osdc,
1082 struct ceph_osd_request *req, int force_resend)
1084 struct ceph_pg pgid;
1085 int acting[CEPH_PG_MAX_SIZE];
1086 int o = -1, num = 0;
1089 dout("map_request %p tid %lld\n", req, req->r_tid);
1090 err = ceph_calc_ceph_pg(&pgid, req->r_oid, osdc->osdmap,
1091 ceph_file_layout_pg_pool(req->r_file_layout));
1093 list_move(&req->r_req_lru_item, &osdc->req_notarget);
1098 err = ceph_calc_pg_acting(osdc->osdmap, pgid, acting);
1104 if ((!force_resend &&
1105 req->r_osd && req->r_osd->o_osd == o &&
1106 req->r_sent >= req->r_osd->o_incarnation &&
1107 req->r_num_pg_osds == num &&
1108 memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
1109 (req->r_osd == NULL && o == -1))
1110 return 0; /* no change */
1112 dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n",
1113 req->r_tid, pgid.pool, pgid.seed, o,
1114 req->r_osd ? req->r_osd->o_osd : -1);
1116 /* record full pg acting set */
1117 memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
1118 req->r_num_pg_osds = num;
1121 __cancel_request(req);
1122 list_del_init(&req->r_osd_item);
1126 req->r_osd = __lookup_osd(osdc, o);
1127 if (!req->r_osd && o >= 0) {
1129 req->r_osd = create_osd(osdc, o);
1131 list_move(&req->r_req_lru_item, &osdc->req_notarget);
1135 dout("map_request osd %p is osd%d\n", req->r_osd, o);
1136 __insert_osd(osdc, req->r_osd);
1138 ceph_con_open(&req->r_osd->o_con,
1139 CEPH_ENTITY_TYPE_OSD, o,
1140 &osdc->osdmap->osd_addr[o]);
1144 __remove_osd_from_lru(req->r_osd);
1145 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
1146 list_move_tail(&req->r_req_lru_item, &osdc->req_unsent);
1148 list_move_tail(&req->r_req_lru_item, &osdc->req_notarget);
1150 err = 1; /* osd or pg changed */
1157 * caller should hold map_sem (for read) and request_mutex
1159 static void __send_request(struct ceph_osd_client *osdc,
1160 struct ceph_osd_request *req)
1164 dout("send_request %p tid %llu to osd%d flags %d pg %lld.%x\n",
1165 req, req->r_tid, req->r_osd->o_osd, req->r_flags,
1166 (unsigned long long)req->r_pgid.pool, req->r_pgid.seed);
1168 /* fill in message content that changes each time we send it */
1169 put_unaligned_le32(osdc->osdmap->epoch, req->r_request_osdmap_epoch);
1170 put_unaligned_le32(req->r_flags, req->r_request_flags);
1171 put_unaligned_le64(req->r_pgid.pool, req->r_request_pool);
1172 p = req->r_request_pgid;
1173 ceph_encode_64(&p, req->r_pgid.pool);
1174 ceph_encode_32(&p, req->r_pgid.seed);
1175 put_unaligned_le64(1, req->r_request_attempts); /* FIXME */
1176 memcpy(req->r_request_reassert_version, &req->r_reassert_version,
1177 sizeof(req->r_reassert_version));
1179 req->r_stamp = jiffies;
1180 list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
1182 ceph_msg_get(req->r_request); /* send consumes a ref */
1183 ceph_con_send(&req->r_osd->o_con, req->r_request);
1184 req->r_sent = req->r_osd->o_incarnation;
1188 * Send any requests in the queue (req_unsent).
1190 static void __send_queued(struct ceph_osd_client *osdc)
1192 struct ceph_osd_request *req, *tmp;
1194 dout("__send_queued\n");
1195 list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item)
1196 __send_request(osdc, req);
1200 * Timeout callback, called every N seconds when 1 or more osd
1201 * requests has been active for more than N seconds. When this
1202 * happens, we ping all OSDs with requests who have timed out to
1203 * ensure any communications channel reset is detected. Reset the
1204 * request timeouts another N seconds in the future as we go.
1205 * Reschedule the timeout event another N seconds in future (unless
1206 * there are no open requests).
1208 static void handle_timeout(struct work_struct *work)
1210 struct ceph_osd_client *osdc =
1211 container_of(work, struct ceph_osd_client, timeout_work.work);
1212 struct ceph_osd_request *req;
1213 struct ceph_osd *osd;
1214 unsigned long keepalive =
1215 osdc->client->options->osd_keepalive_timeout * HZ;
1216 struct list_head slow_osds;
1218 down_read(&osdc->map_sem);
1220 ceph_monc_request_next_osdmap(&osdc->client->monc);
1222 mutex_lock(&osdc->request_mutex);
1225 * ping osds that are a bit slow. this ensures that if there
1226 * is a break in the TCP connection we will notice, and reopen
1227 * a connection with that osd (from the fault callback).
1229 INIT_LIST_HEAD(&slow_osds);
1230 list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
1231 if (time_before(jiffies, req->r_stamp + keepalive))
1236 dout(" tid %llu is slow, will send keepalive on osd%d\n",
1237 req->r_tid, osd->o_osd);
1238 list_move_tail(&osd->o_keepalive_item, &slow_osds);
1240 while (!list_empty(&slow_osds)) {
1241 osd = list_entry(slow_osds.next, struct ceph_osd,
1243 list_del_init(&osd->o_keepalive_item);
1244 ceph_con_keepalive(&osd->o_con);
1247 __schedule_osd_timeout(osdc);
1248 __send_queued(osdc);
1249 mutex_unlock(&osdc->request_mutex);
1250 up_read(&osdc->map_sem);
1253 static void handle_osds_timeout(struct work_struct *work)
1255 struct ceph_osd_client *osdc =
1256 container_of(work, struct ceph_osd_client,
1257 osds_timeout_work.work);
1258 unsigned long delay =
1259 osdc->client->options->osd_idle_ttl * HZ >> 2;
1261 dout("osds timeout\n");
1262 down_read(&osdc->map_sem);
1263 remove_old_osds(osdc);
1264 up_read(&osdc->map_sem);
1266 schedule_delayed_work(&osdc->osds_timeout_work,
1267 round_jiffies_relative(delay));
1270 static void complete_request(struct ceph_osd_request *req)
1272 if (req->r_safe_callback)
1273 req->r_safe_callback(req, NULL);
1274 complete_all(&req->r_safe_completion); /* fsync waiter */
1278 * handle osd op reply. either call the callback if it is specified,
1279 * or do the completion to wake up the waiting thread.
1281 static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg,
1282 struct ceph_connection *con)
1285 struct ceph_osd_request *req;
1288 unsigned int numops;
1289 int payload_len, flags;
1295 u64 reassert_version;
1297 int already_completed;
1301 tid = le64_to_cpu(msg->hdr.tid);
1302 dout("handle_reply %p tid %llu\n", msg, tid);
1304 p = msg->front.iov_base;
1305 end = p + msg->front.iov_len;
1307 ceph_decode_need(&p, end, 4, bad);
1308 object_len = ceph_decode_32(&p);
1309 ceph_decode_need(&p, end, object_len, bad);
1312 err = ceph_decode_pgid(&p, end, &pg);
1316 ceph_decode_need(&p, end, 8 + 4 + 4 + 8 + 4, bad);
1317 flags = ceph_decode_64(&p);
1318 result = ceph_decode_32(&p);
1319 reassert_epoch = ceph_decode_32(&p);
1320 reassert_version = ceph_decode_64(&p);
1321 osdmap_epoch = ceph_decode_32(&p);
1324 mutex_lock(&osdc->request_mutex);
1325 req = __lookup_request(osdc, tid);
1327 dout("handle_reply tid %llu dne\n", tid);
1330 ceph_osdc_get_request(req);
1332 dout("handle_reply %p tid %llu req %p result %d\n", msg, tid,
1335 ceph_decode_need(&p, end, 4, bad);
1336 numops = ceph_decode_32(&p);
1337 if (numops > CEPH_OSD_MAX_OP)
1339 if (numops != req->r_num_ops)
1342 ceph_decode_need(&p, end, numops * sizeof(struct ceph_osd_op), bad);
1343 for (i = 0; i < numops; i++) {
1344 struct ceph_osd_op *op = p;
1347 len = le32_to_cpu(op->payload_len);
1348 req->r_reply_op_len[i] = len;
1349 dout(" op %d has %d bytes\n", i, len);
1353 bytes = le32_to_cpu(msg->hdr.data_len);
1354 if (payload_len != bytes) {
1355 pr_warning("sum of op payload lens %d != data_len %d",
1356 payload_len, bytes);
1360 ceph_decode_need(&p, end, 4 + numops * 4, bad);
1361 retry_attempt = ceph_decode_32(&p);
1362 for (i = 0; i < numops; i++)
1363 req->r_reply_op_result[i] = ceph_decode_32(&p);
1365 if (!req->r_got_reply) {
1367 req->r_result = result;
1368 dout("handle_reply result %d bytes %d\n", req->r_result,
1370 if (req->r_result == 0)
1371 req->r_result = bytes;
1373 /* in case this is a write and we need to replay, */
1374 req->r_reassert_version.epoch = cpu_to_le32(reassert_epoch);
1375 req->r_reassert_version.version = cpu_to_le64(reassert_version);
1377 req->r_got_reply = 1;
1378 } else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) {
1379 dout("handle_reply tid %llu dup ack\n", tid);
1380 mutex_unlock(&osdc->request_mutex);
1384 dout("handle_reply tid %llu flags %d\n", tid, flags);
1386 if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK))
1387 __register_linger_request(osdc, req);
1389 /* either this is a read, or we got the safe response */
1391 (flags & CEPH_OSD_FLAG_ONDISK) ||
1392 ((flags & CEPH_OSD_FLAG_WRITE) == 0))
1393 __unregister_request(osdc, req);
1395 already_completed = req->r_completed;
1396 req->r_completed = 1;
1397 mutex_unlock(&osdc->request_mutex);
1398 if (already_completed)
1401 if (req->r_callback)
1402 req->r_callback(req, msg);
1404 complete_all(&req->r_completion);
1406 if (flags & CEPH_OSD_FLAG_ONDISK)
1407 complete_request(req);
1410 dout("req=%p req->r_linger=%d\n", req, req->r_linger);
1411 ceph_osdc_put_request(req);
1415 ceph_osdc_put_request(req);
1417 mutex_unlock(&osdc->request_mutex);
1419 pr_err("corrupt osd_op_reply got %d %d\n",
1420 (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len));
1424 static void reset_changed_osds(struct ceph_osd_client *osdc)
1426 struct rb_node *p, *n;
1428 for (p = rb_first(&osdc->osds); p; p = n) {
1429 struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
1432 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
1433 memcmp(&osd->o_con.peer_addr,
1434 ceph_osd_addr(osdc->osdmap,
1436 sizeof(struct ceph_entity_addr)) != 0)
1437 __reset_osd(osdc, osd);
1442 * Requeue requests whose mapping to an OSD has changed. If requests map to
1443 * no osd, request a new map.
1445 * Caller should hold map_sem for read.
1447 static void kick_requests(struct ceph_osd_client *osdc, int force_resend)
1449 struct ceph_osd_request *req, *nreq;
1454 dout("kick_requests %s\n", force_resend ? " (force resend)" : "");
1455 mutex_lock(&osdc->request_mutex);
1456 for (p = rb_first(&osdc->requests); p; ) {
1457 req = rb_entry(p, struct ceph_osd_request, r_node);
1461 * For linger requests that have not yet been
1462 * registered, move them to the linger list; they'll
1463 * be sent to the osd in the loop below. Unregister
1464 * the request before re-registering it as a linger
1465 * request to ensure the __map_request() below
1466 * will decide it needs to be sent.
1468 if (req->r_linger && list_empty(&req->r_linger_item)) {
1469 dout("%p tid %llu restart on osd%d\n",
1471 req->r_osd ? req->r_osd->o_osd : -1);
1472 __unregister_request(osdc, req);
1473 __register_linger_request(osdc, req);
1477 err = __map_request(osdc, req, force_resend);
1479 continue; /* error */
1480 if (req->r_osd == NULL) {
1481 dout("%p tid %llu maps to no osd\n", req, req->r_tid);
1482 needmap++; /* request a newer map */
1483 } else if (err > 0) {
1484 if (!req->r_linger) {
1485 dout("%p tid %llu requeued on osd%d\n", req,
1487 req->r_osd ? req->r_osd->o_osd : -1);
1488 req->r_flags |= CEPH_OSD_FLAG_RETRY;
1493 list_for_each_entry_safe(req, nreq, &osdc->req_linger,
1495 dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
1497 err = __map_request(osdc, req, force_resend);
1498 dout("__map_request returned %d\n", err);
1500 continue; /* no change and no osd was specified */
1502 continue; /* hrm! */
1503 if (req->r_osd == NULL) {
1504 dout("tid %llu maps to no valid osd\n", req->r_tid);
1505 needmap++; /* request a newer map */
1509 dout("kicking lingering %p tid %llu osd%d\n", req, req->r_tid,
1510 req->r_osd ? req->r_osd->o_osd : -1);
1511 __register_request(osdc, req);
1512 __unregister_linger_request(osdc, req);
1514 mutex_unlock(&osdc->request_mutex);
1517 dout("%d requests for down osds, need new map\n", needmap);
1518 ceph_monc_request_next_osdmap(&osdc->client->monc);
1520 reset_changed_osds(osdc);
1525 * Process updated osd map.
1527 * The message contains any number of incremental and full maps, normally
1528 * indicating some sort of topology change in the cluster. Kick requests
1529 * off to different OSDs as needed.
1531 void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1533 void *p, *end, *next;
1534 u32 nr_maps, maplen;
1536 struct ceph_osdmap *newmap = NULL, *oldmap;
1538 struct ceph_fsid fsid;
1540 dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
1541 p = msg->front.iov_base;
1542 end = p + msg->front.iov_len;
1545 ceph_decode_need(&p, end, sizeof(fsid), bad);
1546 ceph_decode_copy(&p, &fsid, sizeof(fsid));
1547 if (ceph_check_fsid(osdc->client, &fsid) < 0)
1550 down_write(&osdc->map_sem);
1552 /* incremental maps */
1553 ceph_decode_32_safe(&p, end, nr_maps, bad);
1554 dout(" %d inc maps\n", nr_maps);
1555 while (nr_maps > 0) {
1556 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
1557 epoch = ceph_decode_32(&p);
1558 maplen = ceph_decode_32(&p);
1559 ceph_decode_need(&p, end, maplen, bad);
1561 if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
1562 dout("applying incremental map %u len %d\n",
1564 newmap = osdmap_apply_incremental(&p, next,
1566 &osdc->client->msgr);
1567 if (IS_ERR(newmap)) {
1568 err = PTR_ERR(newmap);
1572 if (newmap != osdc->osdmap) {
1573 ceph_osdmap_destroy(osdc->osdmap);
1574 osdc->osdmap = newmap;
1576 kick_requests(osdc, 0);
1578 dout("ignoring incremental map %u len %d\n",
1588 ceph_decode_32_safe(&p, end, nr_maps, bad);
1589 dout(" %d full maps\n", nr_maps);
1591 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
1592 epoch = ceph_decode_32(&p);
1593 maplen = ceph_decode_32(&p);
1594 ceph_decode_need(&p, end, maplen, bad);
1596 dout("skipping non-latest full map %u len %d\n",
1598 } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
1599 dout("skipping full map %u len %d, "
1600 "older than our %u\n", epoch, maplen,
1601 osdc->osdmap->epoch);
1603 int skipped_map = 0;
1605 dout("taking full map %u len %d\n", epoch, maplen);
1606 newmap = osdmap_decode(&p, p+maplen);
1607 if (IS_ERR(newmap)) {
1608 err = PTR_ERR(newmap);
1612 oldmap = osdc->osdmap;
1613 osdc->osdmap = newmap;
1615 if (oldmap->epoch + 1 < newmap->epoch)
1617 ceph_osdmap_destroy(oldmap);
1619 kick_requests(osdc, skipped_map);
1626 downgrade_write(&osdc->map_sem);
1627 ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch);
1630 * subscribe to subsequent osdmap updates if full to ensure
1631 * we find out when we are no longer full and stop returning
1634 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL))
1635 ceph_monc_request_next_osdmap(&osdc->client->monc);
1637 mutex_lock(&osdc->request_mutex);
1638 __send_queued(osdc);
1639 mutex_unlock(&osdc->request_mutex);
1640 up_read(&osdc->map_sem);
1641 wake_up_all(&osdc->client->auth_wq);
1645 pr_err("osdc handle_map corrupt msg\n");
1647 up_write(&osdc->map_sem);
1652 * watch/notify callback event infrastructure
1654 * These callbacks are used both for watch and notify operations.
1656 static void __release_event(struct kref *kref)
1658 struct ceph_osd_event *event =
1659 container_of(kref, struct ceph_osd_event, kref);
1661 dout("__release_event %p\n", event);
1665 static void get_event(struct ceph_osd_event *event)
1667 kref_get(&event->kref);
1670 void ceph_osdc_put_event(struct ceph_osd_event *event)
1672 kref_put(&event->kref, __release_event);
1674 EXPORT_SYMBOL(ceph_osdc_put_event);
1676 static void __insert_event(struct ceph_osd_client *osdc,
1677 struct ceph_osd_event *new)
1679 struct rb_node **p = &osdc->event_tree.rb_node;
1680 struct rb_node *parent = NULL;
1681 struct ceph_osd_event *event = NULL;
1685 event = rb_entry(parent, struct ceph_osd_event, node);
1686 if (new->cookie < event->cookie)
1688 else if (new->cookie > event->cookie)
1689 p = &(*p)->rb_right;
1694 rb_link_node(&new->node, parent, p);
1695 rb_insert_color(&new->node, &osdc->event_tree);
1698 static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
1701 struct rb_node **p = &osdc->event_tree.rb_node;
1702 struct rb_node *parent = NULL;
1703 struct ceph_osd_event *event = NULL;
1707 event = rb_entry(parent, struct ceph_osd_event, node);
1708 if (cookie < event->cookie)
1710 else if (cookie > event->cookie)
1711 p = &(*p)->rb_right;
1718 static void __remove_event(struct ceph_osd_event *event)
1720 struct ceph_osd_client *osdc = event->osdc;
1722 if (!RB_EMPTY_NODE(&event->node)) {
1723 dout("__remove_event removed %p\n", event);
1724 rb_erase(&event->node, &osdc->event_tree);
1725 ceph_osdc_put_event(event);
1727 dout("__remove_event didn't remove %p\n", event);
1731 int ceph_osdc_create_event(struct ceph_osd_client *osdc,
1732 void (*event_cb)(u64, u64, u8, void *),
1733 void *data, struct ceph_osd_event **pevent)
1735 struct ceph_osd_event *event;
1737 event = kmalloc(sizeof(*event), GFP_NOIO);
1741 dout("create_event %p\n", event);
1742 event->cb = event_cb;
1743 event->one_shot = 0;
1746 INIT_LIST_HEAD(&event->osd_node);
1747 RB_CLEAR_NODE(&event->node);
1748 kref_init(&event->kref); /* one ref for us */
1749 kref_get(&event->kref); /* one ref for the caller */
1751 spin_lock(&osdc->event_lock);
1752 event->cookie = ++osdc->event_count;
1753 __insert_event(osdc, event);
1754 spin_unlock(&osdc->event_lock);
1759 EXPORT_SYMBOL(ceph_osdc_create_event);
1761 void ceph_osdc_cancel_event(struct ceph_osd_event *event)
1763 struct ceph_osd_client *osdc = event->osdc;
1765 dout("cancel_event %p\n", event);
1766 spin_lock(&osdc->event_lock);
1767 __remove_event(event);
1768 spin_unlock(&osdc->event_lock);
1769 ceph_osdc_put_event(event); /* caller's */
1771 EXPORT_SYMBOL(ceph_osdc_cancel_event);
1774 static void do_event_work(struct work_struct *work)
1776 struct ceph_osd_event_work *event_work =
1777 container_of(work, struct ceph_osd_event_work, work);
1778 struct ceph_osd_event *event = event_work->event;
1779 u64 ver = event_work->ver;
1780 u64 notify_id = event_work->notify_id;
1781 u8 opcode = event_work->opcode;
1783 dout("do_event_work completing %p\n", event);
1784 event->cb(ver, notify_id, opcode, event->data);
1785 dout("do_event_work completed %p\n", event);
1786 ceph_osdc_put_event(event);
1792 * Process osd watch notifications
1794 static void handle_watch_notify(struct ceph_osd_client *osdc,
1795 struct ceph_msg *msg)
1799 u64 cookie, ver, notify_id;
1801 struct ceph_osd_event *event;
1802 struct ceph_osd_event_work *event_work;
1804 p = msg->front.iov_base;
1805 end = p + msg->front.iov_len;
1807 ceph_decode_8_safe(&p, end, proto_ver, bad);
1808 ceph_decode_8_safe(&p, end, opcode, bad);
1809 ceph_decode_64_safe(&p, end, cookie, bad);
1810 ceph_decode_64_safe(&p, end, ver, bad);
1811 ceph_decode_64_safe(&p, end, notify_id, bad);
1813 spin_lock(&osdc->event_lock);
1814 event = __find_event(osdc, cookie);
1816 BUG_ON(event->one_shot);
1819 spin_unlock(&osdc->event_lock);
1820 dout("handle_watch_notify cookie %lld ver %lld event %p\n",
1821 cookie, ver, event);
1823 event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
1825 dout("ERROR: could not allocate event_work\n");
1828 INIT_WORK(&event_work->work, do_event_work);
1829 event_work->event = event;
1830 event_work->ver = ver;
1831 event_work->notify_id = notify_id;
1832 event_work->opcode = opcode;
1833 if (!queue_work(osdc->notify_wq, &event_work->work)) {
1834 dout("WARNING: failed to queue notify event work\n");
1842 ceph_osdc_put_event(event);
1846 pr_err("osdc handle_watch_notify corrupt msg\n");
1850 static void ceph_osdc_msg_data_set(struct ceph_msg *msg,
1851 struct ceph_osd_data *osd_data)
1853 u64 length = ceph_osd_data_length(osd_data);
1855 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
1856 BUG_ON(length > (u64) SIZE_MAX);
1858 ceph_msg_data_set_pages(msg, osd_data->pages,
1859 length, osd_data->alignment);
1860 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
1862 ceph_msg_data_set_pagelist(msg, osd_data->pagelist);
1864 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
1865 ceph_msg_data_set_bio(msg, osd_data->bio, length);
1868 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
1873 * build new request AND message
1876 void ceph_osdc_build_request(struct ceph_osd_request *req, u64 off,
1877 struct ceph_snap_context *snapc, u64 snap_id,
1878 struct timespec *mtime)
1880 struct ceph_msg *msg = req->r_request;
1883 int flags = req->r_flags;
1887 /* Set up response incoming data and request outgoing data fields */
1889 ceph_osdc_msg_data_set(req->r_reply, &req->r_data_in);
1890 ceph_osdc_msg_data_set(req->r_request, &req->r_data_out);
1892 req->r_snapid = snap_id;
1893 req->r_snapc = ceph_get_snap_context(snapc);
1895 /* encode request */
1896 msg->hdr.version = cpu_to_le16(4);
1898 p = msg->front.iov_base;
1899 ceph_encode_32(&p, 1); /* client_inc is always 1 */
1900 req->r_request_osdmap_epoch = p;
1902 req->r_request_flags = p;
1904 if (req->r_flags & CEPH_OSD_FLAG_WRITE)
1905 ceph_encode_timespec(p, mtime);
1906 p += sizeof(struct ceph_timespec);
1907 req->r_request_reassert_version = p;
1908 p += sizeof(struct ceph_eversion); /* will get filled in */
1911 ceph_encode_8(&p, 4);
1912 ceph_encode_8(&p, 4);
1913 ceph_encode_32(&p, 8 + 4 + 4);
1914 req->r_request_pool = p;
1916 ceph_encode_32(&p, -1); /* preferred */
1917 ceph_encode_32(&p, 0); /* key len */
1919 ceph_encode_8(&p, 1);
1920 req->r_request_pgid = p;
1922 ceph_encode_32(&p, -1); /* preferred */
1925 ceph_encode_32(&p, req->r_oid_len);
1926 memcpy(p, req->r_oid, req->r_oid_len);
1927 dout("oid '%.*s' len %d\n", req->r_oid_len, req->r_oid, req->r_oid_len);
1928 p += req->r_oid_len;
1930 /* ops--can imply data */
1931 ceph_encode_16(&p, (u16)req->r_num_ops);
1933 for (i = 0; i < req->r_num_ops; i++) {
1934 data_len += osd_req_encode_op(req, p, i);
1935 p += sizeof(struct ceph_osd_op);
1939 ceph_encode_64(&p, req->r_snapid);
1940 ceph_encode_64(&p, req->r_snapc ? req->r_snapc->seq : 0);
1941 ceph_encode_32(&p, req->r_snapc ? req->r_snapc->num_snaps : 0);
1943 for (i = 0; i < snapc->num_snaps; i++) {
1944 ceph_encode_64(&p, req->r_snapc->snaps[i]);
1948 req->r_request_attempts = p;
1952 if (flags & CEPH_OSD_FLAG_WRITE) {
1956 * The header "data_off" is a hint to the receiver
1957 * allowing it to align received data into its
1958 * buffers such that there's no need to re-copy
1959 * it before writing it to disk (direct I/O).
1961 data_off = (u16) (off & 0xffff);
1962 req->r_request->hdr.data_off = cpu_to_le16(data_off);
1964 req->r_request->hdr.data_len = cpu_to_le32(data_len);
1966 BUG_ON(p > msg->front.iov_base + msg->front.iov_len);
1967 msg_size = p - msg->front.iov_base;
1968 msg->front.iov_len = msg_size;
1969 msg->hdr.front_len = cpu_to_le32(msg_size);
1971 dout("build_request msg_size was %d\n", (int)msg_size);
1973 EXPORT_SYMBOL(ceph_osdc_build_request);
1976 * Register request, send initial attempt.
1978 int ceph_osdc_start_request(struct ceph_osd_client *osdc,
1979 struct ceph_osd_request *req,
1984 down_read(&osdc->map_sem);
1985 mutex_lock(&osdc->request_mutex);
1986 __register_request(osdc, req);
1987 WARN_ON(req->r_sent);
1988 rc = __map_request(osdc, req, 0);
1991 dout("osdc_start_request failed map, "
1992 " will retry %lld\n", req->r_tid);
1997 if (req->r_osd == NULL) {
1998 dout("send_request %p no up osds in pg\n", req);
1999 ceph_monc_request_next_osdmap(&osdc->client->monc);
2001 __send_queued(osdc);
2005 mutex_unlock(&osdc->request_mutex);
2006 up_read(&osdc->map_sem);
2009 EXPORT_SYMBOL(ceph_osdc_start_request);
2012 * wait for a request to complete
2014 int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
2015 struct ceph_osd_request *req)
2019 rc = wait_for_completion_interruptible(&req->r_completion);
2021 mutex_lock(&osdc->request_mutex);
2022 __cancel_request(req);
2023 __unregister_request(osdc, req);
2024 mutex_unlock(&osdc->request_mutex);
2025 complete_request(req);
2026 dout("wait_request tid %llu canceled/timed out\n", req->r_tid);
2030 dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result);
2031 return req->r_result;
2033 EXPORT_SYMBOL(ceph_osdc_wait_request);
2036 * sync - wait for all in-flight requests to flush. avoid starvation.
2038 void ceph_osdc_sync(struct ceph_osd_client *osdc)
2040 struct ceph_osd_request *req;
2041 u64 last_tid, next_tid = 0;
2043 mutex_lock(&osdc->request_mutex);
2044 last_tid = osdc->last_tid;
2046 req = __lookup_request_ge(osdc, next_tid);
2049 if (req->r_tid > last_tid)
2052 next_tid = req->r_tid + 1;
2053 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
2056 ceph_osdc_get_request(req);
2057 mutex_unlock(&osdc->request_mutex);
2058 dout("sync waiting on tid %llu (last is %llu)\n",
2059 req->r_tid, last_tid);
2060 wait_for_completion(&req->r_safe_completion);
2061 mutex_lock(&osdc->request_mutex);
2062 ceph_osdc_put_request(req);
2064 mutex_unlock(&osdc->request_mutex);
2065 dout("sync done (thru tid %llu)\n", last_tid);
2067 EXPORT_SYMBOL(ceph_osdc_sync);
2072 int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
2077 osdc->client = client;
2078 osdc->osdmap = NULL;
2079 init_rwsem(&osdc->map_sem);
2080 init_completion(&osdc->map_waiters);
2081 osdc->last_requested_map = 0;
2082 mutex_init(&osdc->request_mutex);
2084 osdc->osds = RB_ROOT;
2085 INIT_LIST_HEAD(&osdc->osd_lru);
2086 osdc->requests = RB_ROOT;
2087 INIT_LIST_HEAD(&osdc->req_lru);
2088 INIT_LIST_HEAD(&osdc->req_unsent);
2089 INIT_LIST_HEAD(&osdc->req_notarget);
2090 INIT_LIST_HEAD(&osdc->req_linger);
2091 osdc->num_requests = 0;
2092 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
2093 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
2094 spin_lock_init(&osdc->event_lock);
2095 osdc->event_tree = RB_ROOT;
2096 osdc->event_count = 0;
2098 schedule_delayed_work(&osdc->osds_timeout_work,
2099 round_jiffies_relative(osdc->client->options->osd_idle_ttl * HZ));
2102 osdc->req_mempool = mempool_create_kmalloc_pool(10,
2103 sizeof(struct ceph_osd_request));
2104 if (!osdc->req_mempool)
2107 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
2108 OSD_OP_FRONT_LEN, 10, true,
2112 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
2113 OSD_OPREPLY_FRONT_LEN, 10, true,
2118 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
2119 if (IS_ERR(osdc->notify_wq)) {
2120 err = PTR_ERR(osdc->notify_wq);
2121 osdc->notify_wq = NULL;
2127 ceph_msgpool_destroy(&osdc->msgpool_op);
2129 mempool_destroy(osdc->req_mempool);
2134 void ceph_osdc_stop(struct ceph_osd_client *osdc)
2136 flush_workqueue(osdc->notify_wq);
2137 destroy_workqueue(osdc->notify_wq);
2138 cancel_delayed_work_sync(&osdc->timeout_work);
2139 cancel_delayed_work_sync(&osdc->osds_timeout_work);
2141 ceph_osdmap_destroy(osdc->osdmap);
2142 osdc->osdmap = NULL;
2144 remove_all_osds(osdc);
2145 mempool_destroy(osdc->req_mempool);
2146 ceph_msgpool_destroy(&osdc->msgpool_op);
2147 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
2151 * Read some contiguous pages. If we cross a stripe boundary, shorten
2152 * *plen. Return number of bytes read, or error.
2154 int ceph_osdc_readpages(struct ceph_osd_client *osdc,
2155 struct ceph_vino vino, struct ceph_file_layout *layout,
2157 u32 truncate_seq, u64 truncate_size,
2158 struct page **pages, int num_pages, int page_align)
2160 struct ceph_osd_request *req;
2163 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
2164 vino.snap, off, *plen);
2165 req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 1,
2166 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
2167 NULL, truncate_seq, truncate_size,
2170 return PTR_ERR(req);
2172 /* it may be a short read due to an object boundary */
2174 ceph_osd_data_pages_init(&req->r_data_in, pages, *plen, page_align,
2177 dout("readpages final extent is %llu~%llu (%llu bytes align %d)\n",
2178 off, *plen, *plen, page_align);
2180 ceph_osdc_build_request(req, off, NULL, vino.snap, NULL);
2182 rc = ceph_osdc_start_request(osdc, req, false);
2184 rc = ceph_osdc_wait_request(osdc, req);
2186 ceph_osdc_put_request(req);
2187 dout("readpages result %d\n", rc);
2190 EXPORT_SYMBOL(ceph_osdc_readpages);
2193 * do a synchronous write on N pages
2195 int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
2196 struct ceph_file_layout *layout,
2197 struct ceph_snap_context *snapc,
2199 u32 truncate_seq, u64 truncate_size,
2200 struct timespec *mtime,
2201 struct page **pages, int num_pages)
2203 struct ceph_osd_request *req;
2205 int page_align = off & ~PAGE_MASK;
2207 BUG_ON(vino.snap != CEPH_NOSNAP); /* snapshots aren't writeable */
2208 req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 1,
2210 CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
2211 snapc, truncate_seq, truncate_size,
2214 return PTR_ERR(req);
2216 /* it may be a short write due to an object boundary */
2217 ceph_osd_data_pages_init(&req->r_data_out, pages, len, page_align,
2219 dout("writepages %llu~%llu (%llu bytes)\n", off, len, len);
2221 ceph_osdc_build_request(req, off, snapc, CEPH_NOSNAP, mtime);
2223 rc = ceph_osdc_start_request(osdc, req, true);
2225 rc = ceph_osdc_wait_request(osdc, req);
2227 ceph_osdc_put_request(req);
2230 dout("writepages result %d\n", rc);
2233 EXPORT_SYMBOL(ceph_osdc_writepages);
2236 * handle incoming message
2238 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
2240 struct ceph_osd *osd = con->private;
2241 struct ceph_osd_client *osdc;
2242 int type = le16_to_cpu(msg->hdr.type);
2249 case CEPH_MSG_OSD_MAP:
2250 ceph_osdc_handle_map(osdc, msg);
2252 case CEPH_MSG_OSD_OPREPLY:
2253 handle_reply(osdc, msg, con);
2255 case CEPH_MSG_WATCH_NOTIFY:
2256 handle_watch_notify(osdc, msg);
2260 pr_err("received unknown message type %d %s\n", type,
2261 ceph_msg_type_name(type));
2268 * lookup and return message for incoming reply. set up reply message
2271 static struct ceph_msg *get_reply(struct ceph_connection *con,
2272 struct ceph_msg_header *hdr,
2275 struct ceph_osd *osd = con->private;
2276 struct ceph_osd_client *osdc = osd->o_osdc;
2278 struct ceph_osd_request *req;
2279 int front = le32_to_cpu(hdr->front_len);
2280 int data_len = le32_to_cpu(hdr->data_len);
2283 tid = le64_to_cpu(hdr->tid);
2284 mutex_lock(&osdc->request_mutex);
2285 req = __lookup_request(osdc, tid);
2289 dout("get_reply unknown tid %llu from osd%d\n", tid,
2294 if (req->r_reply->con)
2295 dout("%s revoking msg %p from old con %p\n", __func__,
2296 req->r_reply, req->r_reply->con);
2297 ceph_msg_revoke_incoming(req->r_reply);
2299 if (front > req->r_reply->front.iov_len) {
2300 pr_warning("get_reply front %d > preallocated %d\n",
2301 front, (int)req->r_reply->front.iov_len);
2302 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front, GFP_NOFS, false);
2305 ceph_msg_put(req->r_reply);
2308 m = ceph_msg_get(req->r_reply);
2311 struct ceph_osd_data *osd_data = &req->r_data_in;
2313 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
2314 if (osd_data->pages &&
2315 unlikely(osd_data->length < data_len)) {
2317 pr_warning("tid %lld reply has %d bytes "
2318 "we had only %llu bytes ready\n",
2319 tid, data_len, osd_data->length);
2328 dout("get_reply tid %lld %p\n", tid, m);
2331 mutex_unlock(&osdc->request_mutex);
2336 static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2337 struct ceph_msg_header *hdr,
2340 struct ceph_osd *osd = con->private;
2341 int type = le16_to_cpu(hdr->type);
2342 int front = le32_to_cpu(hdr->front_len);
2346 case CEPH_MSG_OSD_MAP:
2347 case CEPH_MSG_WATCH_NOTIFY:
2348 return ceph_msg_new(type, front, GFP_NOFS, false);
2349 case CEPH_MSG_OSD_OPREPLY:
2350 return get_reply(con, hdr, skip);
2352 pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
2360 * Wrappers to refcount containing ceph_osd struct
2362 static struct ceph_connection *get_osd_con(struct ceph_connection *con)
2364 struct ceph_osd *osd = con->private;
2370 static void put_osd_con(struct ceph_connection *con)
2372 struct ceph_osd *osd = con->private;
2380 * Note: returned pointer is the address of a structure that's
2381 * managed separately. Caller must *not* attempt to free it.
2383 static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
2384 int *proto, int force_new)
2386 struct ceph_osd *o = con->private;
2387 struct ceph_osd_client *osdc = o->o_osdc;
2388 struct ceph_auth_client *ac = osdc->client->monc.auth;
2389 struct ceph_auth_handshake *auth = &o->o_auth;
2391 if (force_new && auth->authorizer) {
2392 ceph_auth_destroy_authorizer(ac, auth->authorizer);
2393 auth->authorizer = NULL;
2395 if (!auth->authorizer) {
2396 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2399 return ERR_PTR(ret);
2401 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2404 return ERR_PTR(ret);
2406 *proto = ac->protocol;
2412 static int verify_authorizer_reply(struct ceph_connection *con, int len)
2414 struct ceph_osd *o = con->private;
2415 struct ceph_osd_client *osdc = o->o_osdc;
2416 struct ceph_auth_client *ac = osdc->client->monc.auth;
2418 return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer, len);
2421 static int invalidate_authorizer(struct ceph_connection *con)
2423 struct ceph_osd *o = con->private;
2424 struct ceph_osd_client *osdc = o->o_osdc;
2425 struct ceph_auth_client *ac = osdc->client->monc.auth;
2427 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
2428 return ceph_monc_validate_auth(&osdc->client->monc);
2431 static const struct ceph_connection_operations osd_con_ops = {
2434 .dispatch = dispatch,
2435 .get_authorizer = get_authorizer,
2436 .verify_authorizer_reply = verify_authorizer_reply,
2437 .invalidate_authorizer = invalidate_authorizer,
2438 .alloc_msg = alloc_msg,