2 #include <linux/ceph/ceph_debug.h>
4 #include <linux/module.h>
6 #include <linux/highmem.h>
8 #include <linux/pagemap.h>
9 #include <linux/slab.h>
10 #include <linux/uaccess.h>
12 #include <linux/bio.h>
15 #include <linux/ceph/libceph.h>
16 #include <linux/ceph/osd_client.h>
17 #include <linux/ceph/messenger.h>
18 #include <linux/ceph/decode.h>
19 #include <linux/ceph/auth.h>
20 #include <linux/ceph/pagelist.h>
22 #define OSD_OP_FRONT_LEN 4096
23 #define OSD_OPREPLY_FRONT_LEN 512
25 static struct kmem_cache *ceph_osd_request_cache;
27 static const struct ceph_connection_operations osd_con_ops;
29 static void __send_queued(struct ceph_osd_client *osdc);
30 static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd);
31 static void __register_request(struct ceph_osd_client *osdc,
32 struct ceph_osd_request *req);
33 static void __unregister_linger_request(struct ceph_osd_client *osdc,
34 struct ceph_osd_request *req);
35 static void __send_request(struct ceph_osd_client *osdc,
36 struct ceph_osd_request *req);
39 * Implement client access to distributed object storage cluster.
41 * All data objects are stored within a cluster/cloud of OSDs, or
42 * "object storage devices." (Note that Ceph OSDs have _nothing_ to
43 * do with the T10 OSD extensions to SCSI.) Ceph OSDs are simply
44 * remote daemons serving up and coordinating consistent and safe
47 * Cluster membership and the mapping of data objects onto storage devices
48 * are described by the osd map.
50 * We keep track of pending OSD requests (read, write), resubmit
51 * requests to different OSDs when the cluster topology/data layout
52 * change, or retry the affected requests when the communications
53 * channel with an OSD is reset.
57 * calculate the mapping of a file extent onto an object, and fill out the
58 * request accordingly. shorten extent as necessary if it crosses an
61 * fill osd op in request message.
63 static int calc_layout(struct ceph_file_layout *layout, u64 off, u64 *plen,
64 u64 *objnum, u64 *objoff, u64 *objlen)
70 r = ceph_calc_file_object_mapping(layout, off, orig_len, objnum,
74 if (*objlen < orig_len) {
76 dout(" skipping last %llu, final file extent %llu~%llu\n",
77 orig_len - *plen, off, *plen);
80 dout("calc_layout objnum=%llx %llu~%llu\n", *objnum, *objoff, *objlen);
85 static void ceph_osd_data_init(struct ceph_osd_data *osd_data)
87 memset(osd_data, 0, sizeof (*osd_data));
88 osd_data->type = CEPH_OSD_DATA_TYPE_NONE;
91 static void ceph_osd_data_pages_init(struct ceph_osd_data *osd_data,
92 struct page **pages, u64 length, u32 alignment,
93 bool pages_from_pool, bool own_pages)
95 osd_data->type = CEPH_OSD_DATA_TYPE_PAGES;
96 osd_data->pages = pages;
97 osd_data->length = length;
98 osd_data->alignment = alignment;
99 osd_data->pages_from_pool = pages_from_pool;
100 osd_data->own_pages = own_pages;
103 static void ceph_osd_data_pagelist_init(struct ceph_osd_data *osd_data,
104 struct ceph_pagelist *pagelist)
106 osd_data->type = CEPH_OSD_DATA_TYPE_PAGELIST;
107 osd_data->pagelist = pagelist;
111 static void ceph_osd_data_bio_init(struct ceph_osd_data *osd_data,
112 struct bio *bio, size_t bio_length)
114 osd_data->type = CEPH_OSD_DATA_TYPE_BIO;
116 osd_data->bio_length = bio_length;
118 #endif /* CONFIG_BLOCK */
120 #define osd_req_op_data(oreq, whch, typ, fld) \
122 BUG_ON(whch >= (oreq)->r_num_ops); \
123 &(oreq)->r_ops[whch].typ.fld; \
126 static struct ceph_osd_data *
127 osd_req_op_raw_data_in(struct ceph_osd_request *osd_req, unsigned int which)
129 BUG_ON(which >= osd_req->r_num_ops);
131 return &osd_req->r_ops[which].raw_data_in;
134 struct ceph_osd_data *
135 osd_req_op_extent_osd_data(struct ceph_osd_request *osd_req,
138 return osd_req_op_data(osd_req, which, extent, osd_data);
140 EXPORT_SYMBOL(osd_req_op_extent_osd_data);
142 struct ceph_osd_data *
143 osd_req_op_cls_response_data(struct ceph_osd_request *osd_req,
146 return osd_req_op_data(osd_req, which, cls, response_data);
148 EXPORT_SYMBOL(osd_req_op_cls_response_data); /* ??? */
150 void osd_req_op_raw_data_in_pages(struct ceph_osd_request *osd_req,
151 unsigned int which, struct page **pages,
152 u64 length, u32 alignment,
153 bool pages_from_pool, bool own_pages)
155 struct ceph_osd_data *osd_data;
157 osd_data = osd_req_op_raw_data_in(osd_req, which);
158 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
159 pages_from_pool, own_pages);
161 EXPORT_SYMBOL(osd_req_op_raw_data_in_pages);
163 void osd_req_op_extent_osd_data_pages(struct ceph_osd_request *osd_req,
164 unsigned int which, struct page **pages,
165 u64 length, u32 alignment,
166 bool pages_from_pool, bool own_pages)
168 struct ceph_osd_data *osd_data;
170 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
171 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
172 pages_from_pool, own_pages);
174 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pages);
176 void osd_req_op_extent_osd_data_pagelist(struct ceph_osd_request *osd_req,
177 unsigned int which, struct ceph_pagelist *pagelist)
179 struct ceph_osd_data *osd_data;
181 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
182 ceph_osd_data_pagelist_init(osd_data, pagelist);
184 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pagelist);
187 void osd_req_op_extent_osd_data_bio(struct ceph_osd_request *osd_req,
188 unsigned int which, struct bio *bio, size_t bio_length)
190 struct ceph_osd_data *osd_data;
192 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
193 ceph_osd_data_bio_init(osd_data, bio, bio_length);
195 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bio);
196 #endif /* CONFIG_BLOCK */
198 static void osd_req_op_cls_request_info_pagelist(
199 struct ceph_osd_request *osd_req,
200 unsigned int which, struct ceph_pagelist *pagelist)
202 struct ceph_osd_data *osd_data;
204 osd_data = osd_req_op_data(osd_req, which, cls, request_info);
205 ceph_osd_data_pagelist_init(osd_data, pagelist);
208 void osd_req_op_cls_request_data_pagelist(
209 struct ceph_osd_request *osd_req,
210 unsigned int which, struct ceph_pagelist *pagelist)
212 struct ceph_osd_data *osd_data;
214 osd_data = osd_req_op_data(osd_req, which, cls, request_data);
215 ceph_osd_data_pagelist_init(osd_data, pagelist);
217 EXPORT_SYMBOL(osd_req_op_cls_request_data_pagelist);
219 void osd_req_op_cls_request_data_pages(struct ceph_osd_request *osd_req,
220 unsigned int which, struct page **pages, u64 length,
221 u32 alignment, bool pages_from_pool, bool own_pages)
223 struct ceph_osd_data *osd_data;
225 osd_data = osd_req_op_data(osd_req, which, cls, request_data);
226 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
227 pages_from_pool, own_pages);
229 EXPORT_SYMBOL(osd_req_op_cls_request_data_pages);
231 void osd_req_op_cls_response_data_pages(struct ceph_osd_request *osd_req,
232 unsigned int which, struct page **pages, u64 length,
233 u32 alignment, bool pages_from_pool, bool own_pages)
235 struct ceph_osd_data *osd_data;
237 osd_data = osd_req_op_data(osd_req, which, cls, response_data);
238 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
239 pages_from_pool, own_pages);
241 EXPORT_SYMBOL(osd_req_op_cls_response_data_pages);
243 static u64 ceph_osd_data_length(struct ceph_osd_data *osd_data)
245 switch (osd_data->type) {
246 case CEPH_OSD_DATA_TYPE_NONE:
248 case CEPH_OSD_DATA_TYPE_PAGES:
249 return osd_data->length;
250 case CEPH_OSD_DATA_TYPE_PAGELIST:
251 return (u64)osd_data->pagelist->length;
253 case CEPH_OSD_DATA_TYPE_BIO:
254 return (u64)osd_data->bio_length;
255 #endif /* CONFIG_BLOCK */
257 WARN(true, "unrecognized data type %d\n", (int)osd_data->type);
262 static void ceph_osd_data_release(struct ceph_osd_data *osd_data)
264 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES && osd_data->own_pages) {
267 num_pages = calc_pages_for((u64)osd_data->alignment,
268 (u64)osd_data->length);
269 ceph_release_page_vector(osd_data->pages, num_pages);
271 ceph_osd_data_init(osd_data);
274 static void osd_req_op_data_release(struct ceph_osd_request *osd_req,
277 struct ceph_osd_req_op *op;
279 BUG_ON(which >= osd_req->r_num_ops);
280 op = &osd_req->r_ops[which];
283 case CEPH_OSD_OP_READ:
284 case CEPH_OSD_OP_WRITE:
285 ceph_osd_data_release(&op->extent.osd_data);
287 case CEPH_OSD_OP_CALL:
288 ceph_osd_data_release(&op->cls.request_info);
289 ceph_osd_data_release(&op->cls.request_data);
290 ceph_osd_data_release(&op->cls.response_data);
300 void ceph_osdc_release_request(struct kref *kref)
302 struct ceph_osd_request *req;
305 req = container_of(kref, struct ceph_osd_request, r_kref);
307 ceph_msg_put(req->r_request);
309 ceph_msg_revoke_incoming(req->r_reply);
310 ceph_msg_put(req->r_reply);
313 for (which = 0; which < req->r_num_ops; which++)
314 osd_req_op_data_release(req, which);
316 ceph_put_snap_context(req->r_snapc);
318 mempool_free(req, req->r_osdc->req_mempool);
320 kmem_cache_free(ceph_osd_request_cache, req);
323 EXPORT_SYMBOL(ceph_osdc_release_request);
325 struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
326 struct ceph_snap_context *snapc,
327 unsigned int num_ops,
331 struct ceph_osd_request *req;
332 struct ceph_msg *msg;
335 BUILD_BUG_ON(CEPH_OSD_MAX_OP > U16_MAX);
336 BUG_ON(num_ops > CEPH_OSD_MAX_OP);
338 msg_size = 4 + 4 + 8 + 8 + 4+8;
339 msg_size += 2 + 4 + 8 + 4 + 4; /* oloc */
340 msg_size += 1 + 8 + 4 + 4; /* pg_t */
341 msg_size += 4 + CEPH_MAX_OID_NAME_LEN; /* oid */
342 msg_size += 2 + num_ops*sizeof(struct ceph_osd_op);
343 msg_size += 8; /* snapid */
344 msg_size += 8; /* snap_seq */
345 msg_size += 8 * (snapc ? snapc->num_snaps : 0); /* snaps */
349 req = mempool_alloc(osdc->req_mempool, gfp_flags);
350 memset(req, 0, sizeof(*req));
352 req = kmem_cache_zalloc(ceph_osd_request_cache, gfp_flags);
358 req->r_mempool = use_mempool;
359 req->r_num_ops = num_ops;
361 kref_init(&req->r_kref);
362 init_completion(&req->r_completion);
363 init_completion(&req->r_safe_completion);
364 RB_CLEAR_NODE(&req->r_node);
365 INIT_LIST_HEAD(&req->r_unsafe_item);
366 INIT_LIST_HEAD(&req->r_linger_item);
367 INIT_LIST_HEAD(&req->r_linger_osd);
368 INIT_LIST_HEAD(&req->r_req_lru_item);
369 INIT_LIST_HEAD(&req->r_osd_item);
371 req->r_oloc.pool = -1;
373 /* create reply message */
375 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0);
377 msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY,
378 OSD_OPREPLY_FRONT_LEN, gfp_flags, true);
380 ceph_osdc_put_request(req);
385 /* create request message; allow space for oid */
387 msg = ceph_msgpool_get(&osdc->msgpool_op, 0);
389 msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp_flags, true);
391 ceph_osdc_put_request(req);
395 memset(msg->front.iov_base, 0, msg->front.iov_len);
397 req->r_request = msg;
401 EXPORT_SYMBOL(ceph_osdc_alloc_request);
403 static bool osd_req_opcode_valid(u16 opcode)
406 case CEPH_OSD_OP_READ:
407 case CEPH_OSD_OP_STAT:
408 case CEPH_OSD_OP_MAPEXT:
409 case CEPH_OSD_OP_MASKTRUNC:
410 case CEPH_OSD_OP_SPARSE_READ:
411 case CEPH_OSD_OP_NOTIFY:
412 case CEPH_OSD_OP_NOTIFY_ACK:
413 case CEPH_OSD_OP_ASSERT_VER:
414 case CEPH_OSD_OP_WRITE:
415 case CEPH_OSD_OP_WRITEFULL:
416 case CEPH_OSD_OP_TRUNCATE:
417 case CEPH_OSD_OP_ZERO:
418 case CEPH_OSD_OP_DELETE:
419 case CEPH_OSD_OP_APPEND:
420 case CEPH_OSD_OP_STARTSYNC:
421 case CEPH_OSD_OP_SETTRUNC:
422 case CEPH_OSD_OP_TRIMTRUNC:
423 case CEPH_OSD_OP_TMAPUP:
424 case CEPH_OSD_OP_TMAPPUT:
425 case CEPH_OSD_OP_TMAPGET:
426 case CEPH_OSD_OP_CREATE:
427 case CEPH_OSD_OP_ROLLBACK:
428 case CEPH_OSD_OP_WATCH:
429 case CEPH_OSD_OP_OMAPGETKEYS:
430 case CEPH_OSD_OP_OMAPGETVALS:
431 case CEPH_OSD_OP_OMAPGETHEADER:
432 case CEPH_OSD_OP_OMAPGETVALSBYKEYS:
433 case CEPH_OSD_OP_OMAPSETVALS:
434 case CEPH_OSD_OP_OMAPSETHEADER:
435 case CEPH_OSD_OP_OMAPCLEAR:
436 case CEPH_OSD_OP_OMAPRMKEYS:
437 case CEPH_OSD_OP_OMAP_CMP:
438 case CEPH_OSD_OP_CLONERANGE:
439 case CEPH_OSD_OP_ASSERT_SRC_VERSION:
440 case CEPH_OSD_OP_SRC_CMPXATTR:
441 case CEPH_OSD_OP_GETXATTR:
442 case CEPH_OSD_OP_GETXATTRS:
443 case CEPH_OSD_OP_CMPXATTR:
444 case CEPH_OSD_OP_SETXATTR:
445 case CEPH_OSD_OP_SETXATTRS:
446 case CEPH_OSD_OP_RESETXATTRS:
447 case CEPH_OSD_OP_RMXATTR:
448 case CEPH_OSD_OP_PULL:
449 case CEPH_OSD_OP_PUSH:
450 case CEPH_OSD_OP_BALANCEREADS:
451 case CEPH_OSD_OP_UNBALANCEREADS:
452 case CEPH_OSD_OP_SCRUB:
453 case CEPH_OSD_OP_SCRUB_RESERVE:
454 case CEPH_OSD_OP_SCRUB_UNRESERVE:
455 case CEPH_OSD_OP_SCRUB_STOP:
456 case CEPH_OSD_OP_SCRUB_MAP:
457 case CEPH_OSD_OP_WRLOCK:
458 case CEPH_OSD_OP_WRUNLOCK:
459 case CEPH_OSD_OP_RDLOCK:
460 case CEPH_OSD_OP_RDUNLOCK:
461 case CEPH_OSD_OP_UPLOCK:
462 case CEPH_OSD_OP_DNLOCK:
463 case CEPH_OSD_OP_CALL:
464 case CEPH_OSD_OP_PGLS:
465 case CEPH_OSD_OP_PGLS_FILTER:
473 * This is an osd op init function for opcodes that have no data or
474 * other information associated with them. It also serves as a
475 * common init routine for all the other init functions, below.
477 static struct ceph_osd_req_op *
478 _osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which,
481 struct ceph_osd_req_op *op;
483 BUG_ON(which >= osd_req->r_num_ops);
484 BUG_ON(!osd_req_opcode_valid(opcode));
486 op = &osd_req->r_ops[which];
487 memset(op, 0, sizeof (*op));
493 void osd_req_op_init(struct ceph_osd_request *osd_req,
494 unsigned int which, u16 opcode)
496 (void)_osd_req_op_init(osd_req, which, opcode);
498 EXPORT_SYMBOL(osd_req_op_init);
500 void osd_req_op_extent_init(struct ceph_osd_request *osd_req,
501 unsigned int which, u16 opcode,
502 u64 offset, u64 length,
503 u64 truncate_size, u32 truncate_seq)
505 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
506 size_t payload_len = 0;
508 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
509 opcode != CEPH_OSD_OP_DELETE && opcode != CEPH_OSD_OP_ZERO &&
510 opcode != CEPH_OSD_OP_TRUNCATE);
512 op->extent.offset = offset;
513 op->extent.length = length;
514 op->extent.truncate_size = truncate_size;
515 op->extent.truncate_seq = truncate_seq;
516 if (opcode == CEPH_OSD_OP_WRITE)
517 payload_len += length;
519 op->payload_len = payload_len;
521 EXPORT_SYMBOL(osd_req_op_extent_init);
523 void osd_req_op_extent_update(struct ceph_osd_request *osd_req,
524 unsigned int which, u64 length)
526 struct ceph_osd_req_op *op;
529 BUG_ON(which >= osd_req->r_num_ops);
530 op = &osd_req->r_ops[which];
531 previous = op->extent.length;
533 if (length == previous)
534 return; /* Nothing to do */
535 BUG_ON(length > previous);
537 op->extent.length = length;
538 op->payload_len -= previous - length;
540 EXPORT_SYMBOL(osd_req_op_extent_update);
542 void osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
543 u16 opcode, const char *class, const char *method)
545 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
546 struct ceph_pagelist *pagelist;
547 size_t payload_len = 0;
550 BUG_ON(opcode != CEPH_OSD_OP_CALL);
552 pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
554 ceph_pagelist_init(pagelist);
556 op->cls.class_name = class;
557 size = strlen(class);
558 BUG_ON(size > (size_t) U8_MAX);
559 op->cls.class_len = size;
560 ceph_pagelist_append(pagelist, class, size);
563 op->cls.method_name = method;
564 size = strlen(method);
565 BUG_ON(size > (size_t) U8_MAX);
566 op->cls.method_len = size;
567 ceph_pagelist_append(pagelist, method, size);
570 osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist);
572 op->cls.argc = 0; /* currently unused */
574 op->payload_len = payload_len;
576 EXPORT_SYMBOL(osd_req_op_cls_init);
578 void osd_req_op_watch_init(struct ceph_osd_request *osd_req,
579 unsigned int which, u16 opcode,
580 u64 cookie, u64 version, int flag)
582 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
584 BUG_ON(opcode != CEPH_OSD_OP_NOTIFY_ACK && opcode != CEPH_OSD_OP_WATCH);
586 op->watch.cookie = cookie;
587 op->watch.ver = version;
588 if (opcode == CEPH_OSD_OP_WATCH && flag)
589 op->watch.flag = (u8)1;
591 EXPORT_SYMBOL(osd_req_op_watch_init);
593 static void ceph_osdc_msg_data_add(struct ceph_msg *msg,
594 struct ceph_osd_data *osd_data)
596 u64 length = ceph_osd_data_length(osd_data);
598 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
599 BUG_ON(length > (u64) SIZE_MAX);
601 ceph_msg_data_add_pages(msg, osd_data->pages,
602 length, osd_data->alignment);
603 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
605 ceph_msg_data_add_pagelist(msg, osd_data->pagelist);
607 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
608 ceph_msg_data_add_bio(msg, osd_data->bio, length);
611 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
615 static u64 osd_req_encode_op(struct ceph_osd_request *req,
616 struct ceph_osd_op *dst, unsigned int which)
618 struct ceph_osd_req_op *src;
619 struct ceph_osd_data *osd_data;
620 u64 request_data_len = 0;
623 BUG_ON(which >= req->r_num_ops);
624 src = &req->r_ops[which];
625 if (WARN_ON(!osd_req_opcode_valid(src->op))) {
626 pr_err("unrecognized osd opcode %d\n", src->op);
632 case CEPH_OSD_OP_STAT:
633 osd_data = &src->raw_data_in;
634 ceph_osdc_msg_data_add(req->r_reply, osd_data);
636 case CEPH_OSD_OP_READ:
637 case CEPH_OSD_OP_WRITE:
638 case CEPH_OSD_OP_ZERO:
639 case CEPH_OSD_OP_DELETE:
640 case CEPH_OSD_OP_TRUNCATE:
641 if (src->op == CEPH_OSD_OP_WRITE)
642 request_data_len = src->extent.length;
643 dst->extent.offset = cpu_to_le64(src->extent.offset);
644 dst->extent.length = cpu_to_le64(src->extent.length);
645 dst->extent.truncate_size =
646 cpu_to_le64(src->extent.truncate_size);
647 dst->extent.truncate_seq =
648 cpu_to_le32(src->extent.truncate_seq);
649 osd_data = &src->extent.osd_data;
650 if (src->op == CEPH_OSD_OP_WRITE)
651 ceph_osdc_msg_data_add(req->r_request, osd_data);
653 ceph_osdc_msg_data_add(req->r_reply, osd_data);
655 case CEPH_OSD_OP_CALL:
656 dst->cls.class_len = src->cls.class_len;
657 dst->cls.method_len = src->cls.method_len;
658 osd_data = &src->cls.request_info;
659 ceph_osdc_msg_data_add(req->r_request, osd_data);
660 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGELIST);
661 request_data_len = osd_data->pagelist->length;
663 osd_data = &src->cls.request_data;
664 data_length = ceph_osd_data_length(osd_data);
666 BUG_ON(osd_data->type == CEPH_OSD_DATA_TYPE_NONE);
667 dst->cls.indata_len = cpu_to_le32(data_length);
668 ceph_osdc_msg_data_add(req->r_request, osd_data);
669 src->payload_len += data_length;
670 request_data_len += data_length;
672 osd_data = &src->cls.response_data;
673 ceph_osdc_msg_data_add(req->r_reply, osd_data);
675 case CEPH_OSD_OP_STARTSYNC:
677 case CEPH_OSD_OP_NOTIFY_ACK:
678 case CEPH_OSD_OP_WATCH:
679 dst->watch.cookie = cpu_to_le64(src->watch.cookie);
680 dst->watch.ver = cpu_to_le64(src->watch.ver);
681 dst->watch.flag = src->watch.flag;
684 pr_err("unsupported osd opcode %s\n",
685 ceph_osd_op_name(src->op));
690 dst->op = cpu_to_le16(src->op);
691 dst->payload_len = cpu_to_le32(src->payload_len);
693 return request_data_len;
697 * build new request AND message, calculate layout, and adjust file
700 * if the file was recently truncated, we include information about its
701 * old and new size so that the object can be updated appropriately. (we
702 * avoid synchronously deleting truncated objects because it's slow.)
704 * if @do_sync, include a 'startsync' command so that the osd will flush
707 struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
708 struct ceph_file_layout *layout,
709 struct ceph_vino vino,
710 u64 off, u64 *plen, int num_ops,
711 int opcode, int flags,
712 struct ceph_snap_context *snapc,
717 struct ceph_osd_request *req;
725 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
726 opcode != CEPH_OSD_OP_DELETE && opcode != CEPH_OSD_OP_ZERO &&
727 opcode != CEPH_OSD_OP_TRUNCATE);
729 req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool,
732 return ERR_PTR(-ENOMEM);
734 req->r_flags = flags;
736 /* calculate max write size */
737 r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
739 ceph_osdc_put_request(req);
743 object_size = le32_to_cpu(layout->fl_object_size);
744 object_base = off - objoff;
745 if (!(truncate_seq == 1 && truncate_size == -1ULL)) {
746 if (truncate_size <= object_base) {
749 truncate_size -= object_base;
750 if (truncate_size > object_size)
751 truncate_size = object_size;
755 osd_req_op_extent_init(req, 0, opcode, objoff, objlen,
756 truncate_size, truncate_seq);
759 * A second op in the ops array means the caller wants to
760 * also issue a include a 'startsync' command so that the
761 * osd will flush data quickly.
764 osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC);
766 req->r_oloc.pool = ceph_file_layout_pg_pool(*layout);
768 snprintf(req->r_oid.name, sizeof(req->r_oid.name),
769 "%llx.%08llx", vino.ino, objnum);
770 req->r_oid.name_len = strlen(req->r_oid.name);
774 EXPORT_SYMBOL(ceph_osdc_new_request);
777 * We keep osd requests in an rbtree, sorted by ->r_tid.
779 static void __insert_request(struct ceph_osd_client *osdc,
780 struct ceph_osd_request *new)
782 struct rb_node **p = &osdc->requests.rb_node;
783 struct rb_node *parent = NULL;
784 struct ceph_osd_request *req = NULL;
788 req = rb_entry(parent, struct ceph_osd_request, r_node);
789 if (new->r_tid < req->r_tid)
791 else if (new->r_tid > req->r_tid)
797 rb_link_node(&new->r_node, parent, p);
798 rb_insert_color(&new->r_node, &osdc->requests);
801 static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc,
804 struct ceph_osd_request *req;
805 struct rb_node *n = osdc->requests.rb_node;
808 req = rb_entry(n, struct ceph_osd_request, r_node);
809 if (tid < req->r_tid)
811 else if (tid > req->r_tid)
819 static struct ceph_osd_request *
820 __lookup_request_ge(struct ceph_osd_client *osdc,
823 struct ceph_osd_request *req;
824 struct rb_node *n = osdc->requests.rb_node;
827 req = rb_entry(n, struct ceph_osd_request, r_node);
828 if (tid < req->r_tid) {
832 } else if (tid > req->r_tid) {
842 * Resubmit requests pending on the given osd.
844 static void __kick_osd_requests(struct ceph_osd_client *osdc,
845 struct ceph_osd *osd)
847 struct ceph_osd_request *req, *nreq;
851 dout("__kick_osd_requests osd%d\n", osd->o_osd);
852 err = __reset_osd(osdc, osd);
856 * Build up a list of requests to resend by traversing the
857 * osd's list of requests. Requests for a given object are
858 * sent in tid order, and that is also the order they're
859 * kept on this list. Therefore all requests that are in
860 * flight will be found first, followed by all requests that
861 * have not yet been sent. And to resend requests while
862 * preserving this order we will want to put any sent
863 * requests back on the front of the osd client's unsent
866 * So we build a separate ordered list of already-sent
867 * requests for the affected osd and splice it onto the
868 * front of the osd client's unsent list. Once we've seen a
869 * request that has not yet been sent we're done. Those
870 * requests are already sitting right where they belong.
872 list_for_each_entry(req, &osd->o_requests, r_osd_item) {
875 list_move_tail(&req->r_req_lru_item, &resend);
876 dout("requeueing %p tid %llu osd%d\n", req, req->r_tid,
879 req->r_flags |= CEPH_OSD_FLAG_RETRY;
881 list_splice(&resend, &osdc->req_unsent);
884 * Linger requests are re-registered before sending, which
885 * sets up a new tid for each. We add them to the unsent
886 * list at the end to keep things in tid order.
888 list_for_each_entry_safe(req, nreq, &osd->o_linger_requests,
891 * reregister request prior to unregistering linger so
892 * that r_osd is preserved.
894 BUG_ON(!list_empty(&req->r_req_lru_item));
895 __register_request(osdc, req);
896 list_add_tail(&req->r_req_lru_item, &osdc->req_unsent);
897 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
898 __unregister_linger_request(osdc, req);
899 dout("requeued lingering %p tid %llu osd%d\n", req, req->r_tid,
905 * If the osd connection drops, we need to resubmit all requests.
907 static void osd_reset(struct ceph_connection *con)
909 struct ceph_osd *osd = con->private;
910 struct ceph_osd_client *osdc;
914 dout("osd_reset osd%d\n", osd->o_osd);
916 down_read(&osdc->map_sem);
917 mutex_lock(&osdc->request_mutex);
918 __kick_osd_requests(osdc, osd);
920 mutex_unlock(&osdc->request_mutex);
921 up_read(&osdc->map_sem);
925 * Track open sessions with osds.
927 static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
929 struct ceph_osd *osd;
931 osd = kzalloc(sizeof(*osd), GFP_NOFS);
935 atomic_set(&osd->o_ref, 1);
938 RB_CLEAR_NODE(&osd->o_node);
939 INIT_LIST_HEAD(&osd->o_requests);
940 INIT_LIST_HEAD(&osd->o_linger_requests);
941 INIT_LIST_HEAD(&osd->o_osd_lru);
942 osd->o_incarnation = 1;
944 ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
946 INIT_LIST_HEAD(&osd->o_keepalive_item);
950 static struct ceph_osd *get_osd(struct ceph_osd *osd)
952 if (atomic_inc_not_zero(&osd->o_ref)) {
953 dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
954 atomic_read(&osd->o_ref));
957 dout("get_osd %p FAIL\n", osd);
962 static void put_osd(struct ceph_osd *osd)
964 dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
965 atomic_read(&osd->o_ref) - 1);
966 if (atomic_dec_and_test(&osd->o_ref) && osd->o_auth.authorizer) {
967 struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth;
969 ceph_auth_destroy_authorizer(ac, osd->o_auth.authorizer);
975 * remove an osd from our map
977 static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
979 dout("__remove_osd %p\n", osd);
980 BUG_ON(!list_empty(&osd->o_requests));
981 rb_erase(&osd->o_node, &osdc->osds);
982 list_del_init(&osd->o_osd_lru);
983 ceph_con_close(&osd->o_con);
987 static void remove_all_osds(struct ceph_osd_client *osdc)
989 dout("%s %p\n", __func__, osdc);
990 mutex_lock(&osdc->request_mutex);
991 while (!RB_EMPTY_ROOT(&osdc->osds)) {
992 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
993 struct ceph_osd, o_node);
994 __remove_osd(osdc, osd);
996 mutex_unlock(&osdc->request_mutex);
999 static void __move_osd_to_lru(struct ceph_osd_client *osdc,
1000 struct ceph_osd *osd)
1002 dout("__move_osd_to_lru %p\n", osd);
1003 BUG_ON(!list_empty(&osd->o_osd_lru));
1004 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
1005 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl * HZ;
1008 static void __remove_osd_from_lru(struct ceph_osd *osd)
1010 dout("__remove_osd_from_lru %p\n", osd);
1011 if (!list_empty(&osd->o_osd_lru))
1012 list_del_init(&osd->o_osd_lru);
1015 static void remove_old_osds(struct ceph_osd_client *osdc)
1017 struct ceph_osd *osd, *nosd;
1019 dout("__remove_old_osds %p\n", osdc);
1020 mutex_lock(&osdc->request_mutex);
1021 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
1022 if (time_before(jiffies, osd->lru_ttl))
1024 __remove_osd(osdc, osd);
1026 mutex_unlock(&osdc->request_mutex);
1032 static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
1034 struct ceph_entity_addr *peer_addr;
1036 dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
1037 if (list_empty(&osd->o_requests) &&
1038 list_empty(&osd->o_linger_requests)) {
1039 __remove_osd(osdc, osd);
1044 peer_addr = &osdc->osdmap->osd_addr[osd->o_osd];
1045 if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1046 !ceph_con_opened(&osd->o_con)) {
1047 struct ceph_osd_request *req;
1049 dout("osd addr hasn't changed and connection never opened, "
1050 "letting msgr retry\n");
1051 /* touch each r_stamp for handle_timeout()'s benfit */
1052 list_for_each_entry(req, &osd->o_requests, r_osd_item)
1053 req->r_stamp = jiffies;
1058 ceph_con_close(&osd->o_con);
1059 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1060 osd->o_incarnation++;
1065 static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new)
1067 struct rb_node **p = &osdc->osds.rb_node;
1068 struct rb_node *parent = NULL;
1069 struct ceph_osd *osd = NULL;
1071 dout("__insert_osd %p osd%d\n", new, new->o_osd);
1074 osd = rb_entry(parent, struct ceph_osd, o_node);
1075 if (new->o_osd < osd->o_osd)
1077 else if (new->o_osd > osd->o_osd)
1078 p = &(*p)->rb_right;
1083 rb_link_node(&new->o_node, parent, p);
1084 rb_insert_color(&new->o_node, &osdc->osds);
1087 static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o)
1089 struct ceph_osd *osd;
1090 struct rb_node *n = osdc->osds.rb_node;
1093 osd = rb_entry(n, struct ceph_osd, o_node);
1096 else if (o > osd->o_osd)
1104 static void __schedule_osd_timeout(struct ceph_osd_client *osdc)
1106 schedule_delayed_work(&osdc->timeout_work,
1107 osdc->client->options->osd_keepalive_timeout * HZ);
1110 static void __cancel_osd_timeout(struct ceph_osd_client *osdc)
1112 cancel_delayed_work(&osdc->timeout_work);
1116 * Register request, assign tid. If this is the first request, set up
1117 * the timeout event.
1119 static void __register_request(struct ceph_osd_client *osdc,
1120 struct ceph_osd_request *req)
1122 req->r_tid = ++osdc->last_tid;
1123 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
1124 dout("__register_request %p tid %lld\n", req, req->r_tid);
1125 __insert_request(osdc, req);
1126 ceph_osdc_get_request(req);
1127 osdc->num_requests++;
1128 if (osdc->num_requests == 1) {
1129 dout(" first request, scheduling timeout\n");
1130 __schedule_osd_timeout(osdc);
1135 * called under osdc->request_mutex
1137 static void __unregister_request(struct ceph_osd_client *osdc,
1138 struct ceph_osd_request *req)
1140 if (RB_EMPTY_NODE(&req->r_node)) {
1141 dout("__unregister_request %p tid %lld not registered\n",
1146 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
1147 rb_erase(&req->r_node, &osdc->requests);
1148 osdc->num_requests--;
1151 /* make sure the original request isn't in flight. */
1152 ceph_msg_revoke(req->r_request);
1154 list_del_init(&req->r_osd_item);
1155 if (list_empty(&req->r_osd->o_requests) &&
1156 list_empty(&req->r_osd->o_linger_requests)) {
1157 dout("moving osd to %p lru\n", req->r_osd);
1158 __move_osd_to_lru(osdc, req->r_osd);
1160 if (list_empty(&req->r_linger_item))
1164 list_del_init(&req->r_req_lru_item);
1165 ceph_osdc_put_request(req);
1167 if (osdc->num_requests == 0) {
1168 dout(" no requests, canceling timeout\n");
1169 __cancel_osd_timeout(osdc);
1174 * Cancel a previously queued request message
1176 static void __cancel_request(struct ceph_osd_request *req)
1178 if (req->r_sent && req->r_osd) {
1179 ceph_msg_revoke(req->r_request);
1184 static void __register_linger_request(struct ceph_osd_client *osdc,
1185 struct ceph_osd_request *req)
1187 dout("__register_linger_request %p\n", req);
1188 ceph_osdc_get_request(req);
1189 list_add_tail(&req->r_linger_item, &osdc->req_linger);
1191 list_add_tail(&req->r_linger_osd,
1192 &req->r_osd->o_linger_requests);
1195 static void __unregister_linger_request(struct ceph_osd_client *osdc,
1196 struct ceph_osd_request *req)
1198 dout("__unregister_linger_request %p\n", req);
1199 list_del_init(&req->r_linger_item);
1201 list_del_init(&req->r_linger_osd);
1203 if (list_empty(&req->r_osd->o_requests) &&
1204 list_empty(&req->r_osd->o_linger_requests)) {
1205 dout("moving osd to %p lru\n", req->r_osd);
1206 __move_osd_to_lru(osdc, req->r_osd);
1208 if (list_empty(&req->r_osd_item))
1211 ceph_osdc_put_request(req);
1214 void ceph_osdc_unregister_linger_request(struct ceph_osd_client *osdc,
1215 struct ceph_osd_request *req)
1217 mutex_lock(&osdc->request_mutex);
1218 if (req->r_linger) {
1220 __unregister_linger_request(osdc, req);
1222 mutex_unlock(&osdc->request_mutex);
1224 EXPORT_SYMBOL(ceph_osdc_unregister_linger_request);
1226 void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
1227 struct ceph_osd_request *req)
1229 if (!req->r_linger) {
1230 dout("set_request_linger %p\n", req);
1234 EXPORT_SYMBOL(ceph_osdc_set_request_linger);
1237 * Returns whether a request should be blocked from being sent
1238 * based on the current osdmap and osd_client settings.
1240 * Caller should hold map_sem for read.
1242 static bool __req_should_be_paused(struct ceph_osd_client *osdc,
1243 struct ceph_osd_request *req)
1245 bool pauserd = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD);
1246 bool pausewr = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR) ||
1247 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
1248 return (req->r_flags & CEPH_OSD_FLAG_READ && pauserd) ||
1249 (req->r_flags & CEPH_OSD_FLAG_WRITE && pausewr);
1253 * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
1254 * (as needed), and set the request r_osd appropriately. If there is
1255 * no up osd, set r_osd to NULL. Move the request to the appropriate list
1256 * (unsent, homeless) or leave on in-flight lru.
1258 * Return 0 if unchanged, 1 if changed, or negative on error.
1260 * Caller should hold map_sem for read and request_mutex.
1262 static int __map_request(struct ceph_osd_client *osdc,
1263 struct ceph_osd_request *req, int force_resend)
1265 struct ceph_pg pgid;
1266 int acting[CEPH_PG_MAX_SIZE];
1267 int o = -1, num = 0;
1271 dout("map_request %p tid %lld\n", req, req->r_tid);
1272 err = ceph_calc_ceph_pg(&pgid, req->r_oid.name, osdc->osdmap,
1275 list_move(&req->r_req_lru_item, &osdc->req_notarget);
1280 err = ceph_calc_pg_acting(osdc->osdmap, pgid, acting);
1286 was_paused = req->r_paused;
1287 req->r_paused = __req_should_be_paused(osdc, req);
1288 if (was_paused && !req->r_paused)
1291 if ((!force_resend &&
1292 req->r_osd && req->r_osd->o_osd == o &&
1293 req->r_sent >= req->r_osd->o_incarnation &&
1294 req->r_num_pg_osds == num &&
1295 memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
1296 (req->r_osd == NULL && o == -1) ||
1298 return 0; /* no change */
1300 dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n",
1301 req->r_tid, pgid.pool, pgid.seed, o,
1302 req->r_osd ? req->r_osd->o_osd : -1);
1304 /* record full pg acting set */
1305 memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
1306 req->r_num_pg_osds = num;
1309 __cancel_request(req);
1310 list_del_init(&req->r_osd_item);
1314 req->r_osd = __lookup_osd(osdc, o);
1315 if (!req->r_osd && o >= 0) {
1317 req->r_osd = create_osd(osdc, o);
1319 list_move(&req->r_req_lru_item, &osdc->req_notarget);
1323 dout("map_request osd %p is osd%d\n", req->r_osd, o);
1324 __insert_osd(osdc, req->r_osd);
1326 ceph_con_open(&req->r_osd->o_con,
1327 CEPH_ENTITY_TYPE_OSD, o,
1328 &osdc->osdmap->osd_addr[o]);
1332 __remove_osd_from_lru(req->r_osd);
1333 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
1334 list_move_tail(&req->r_req_lru_item, &osdc->req_unsent);
1336 list_move_tail(&req->r_req_lru_item, &osdc->req_notarget);
1338 err = 1; /* osd or pg changed */
1345 * caller should hold map_sem (for read) and request_mutex
1347 static void __send_request(struct ceph_osd_client *osdc,
1348 struct ceph_osd_request *req)
1352 dout("send_request %p tid %llu to osd%d flags %d pg %lld.%x\n",
1353 req, req->r_tid, req->r_osd->o_osd, req->r_flags,
1354 (unsigned long long)req->r_pgid.pool, req->r_pgid.seed);
1356 /* fill in message content that changes each time we send it */
1357 put_unaligned_le32(osdc->osdmap->epoch, req->r_request_osdmap_epoch);
1358 put_unaligned_le32(req->r_flags, req->r_request_flags);
1359 put_unaligned_le64(req->r_oloc.pool, req->r_request_pool);
1360 p = req->r_request_pgid;
1361 ceph_encode_64(&p, req->r_pgid.pool);
1362 ceph_encode_32(&p, req->r_pgid.seed);
1363 put_unaligned_le64(1, req->r_request_attempts); /* FIXME */
1364 memcpy(req->r_request_reassert_version, &req->r_reassert_version,
1365 sizeof(req->r_reassert_version));
1367 req->r_stamp = jiffies;
1368 list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
1370 ceph_msg_get(req->r_request); /* send consumes a ref */
1372 req->r_sent = req->r_osd->o_incarnation;
1374 ceph_con_send(&req->r_osd->o_con, req->r_request);
1378 * Send any requests in the queue (req_unsent).
1380 static void __send_queued(struct ceph_osd_client *osdc)
1382 struct ceph_osd_request *req, *tmp;
1384 dout("__send_queued\n");
1385 list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item)
1386 __send_request(osdc, req);
1390 * Timeout callback, called every N seconds when 1 or more osd
1391 * requests has been active for more than N seconds. When this
1392 * happens, we ping all OSDs with requests who have timed out to
1393 * ensure any communications channel reset is detected. Reset the
1394 * request timeouts another N seconds in the future as we go.
1395 * Reschedule the timeout event another N seconds in future (unless
1396 * there are no open requests).
1398 static void handle_timeout(struct work_struct *work)
1400 struct ceph_osd_client *osdc =
1401 container_of(work, struct ceph_osd_client, timeout_work.work);
1402 struct ceph_osd_request *req;
1403 struct ceph_osd *osd;
1404 unsigned long keepalive =
1405 osdc->client->options->osd_keepalive_timeout * HZ;
1406 struct list_head slow_osds;
1408 down_read(&osdc->map_sem);
1410 ceph_monc_request_next_osdmap(&osdc->client->monc);
1412 mutex_lock(&osdc->request_mutex);
1415 * ping osds that are a bit slow. this ensures that if there
1416 * is a break in the TCP connection we will notice, and reopen
1417 * a connection with that osd (from the fault callback).
1419 INIT_LIST_HEAD(&slow_osds);
1420 list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
1421 if (time_before(jiffies, req->r_stamp + keepalive))
1426 dout(" tid %llu is slow, will send keepalive on osd%d\n",
1427 req->r_tid, osd->o_osd);
1428 list_move_tail(&osd->o_keepalive_item, &slow_osds);
1430 while (!list_empty(&slow_osds)) {
1431 osd = list_entry(slow_osds.next, struct ceph_osd,
1433 list_del_init(&osd->o_keepalive_item);
1434 ceph_con_keepalive(&osd->o_con);
1437 __schedule_osd_timeout(osdc);
1438 __send_queued(osdc);
1439 mutex_unlock(&osdc->request_mutex);
1440 up_read(&osdc->map_sem);
1443 static void handle_osds_timeout(struct work_struct *work)
1445 struct ceph_osd_client *osdc =
1446 container_of(work, struct ceph_osd_client,
1447 osds_timeout_work.work);
1448 unsigned long delay =
1449 osdc->client->options->osd_idle_ttl * HZ >> 2;
1451 dout("osds timeout\n");
1452 down_read(&osdc->map_sem);
1453 remove_old_osds(osdc);
1454 up_read(&osdc->map_sem);
1456 schedule_delayed_work(&osdc->osds_timeout_work,
1457 round_jiffies_relative(delay));
1460 static void complete_request(struct ceph_osd_request *req)
1462 complete_all(&req->r_safe_completion); /* fsync waiter */
1466 * handle osd op reply. either call the callback if it is specified,
1467 * or do the completion to wake up the waiting thread.
1469 static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg,
1470 struct ceph_connection *con)
1473 struct ceph_osd_request *req;
1476 unsigned int numops;
1477 int payload_len, flags;
1483 u64 reassert_version;
1485 int already_completed;
1489 tid = le64_to_cpu(msg->hdr.tid);
1490 dout("handle_reply %p tid %llu\n", msg, tid);
1492 p = msg->front.iov_base;
1493 end = p + msg->front.iov_len;
1495 ceph_decode_need(&p, end, 4, bad);
1496 object_len = ceph_decode_32(&p);
1497 ceph_decode_need(&p, end, object_len, bad);
1500 err = ceph_decode_pgid(&p, end, &pg);
1504 ceph_decode_need(&p, end, 8 + 4 + 4 + 8 + 4, bad);
1505 flags = ceph_decode_64(&p);
1506 result = ceph_decode_32(&p);
1507 reassert_epoch = ceph_decode_32(&p);
1508 reassert_version = ceph_decode_64(&p);
1509 osdmap_epoch = ceph_decode_32(&p);
1512 mutex_lock(&osdc->request_mutex);
1513 req = __lookup_request(osdc, tid);
1515 dout("handle_reply tid %llu dne\n", tid);
1518 ceph_osdc_get_request(req);
1520 dout("handle_reply %p tid %llu req %p result %d\n", msg, tid,
1523 ceph_decode_need(&p, end, 4, bad_put);
1524 numops = ceph_decode_32(&p);
1525 if (numops > CEPH_OSD_MAX_OP)
1527 if (numops != req->r_num_ops)
1530 ceph_decode_need(&p, end, numops * sizeof(struct ceph_osd_op), bad_put);
1531 for (i = 0; i < numops; i++) {
1532 struct ceph_osd_op *op = p;
1535 len = le32_to_cpu(op->payload_len);
1536 req->r_reply_op_len[i] = len;
1537 dout(" op %d has %d bytes\n", i, len);
1541 bytes = le32_to_cpu(msg->hdr.data_len);
1542 if (payload_len != bytes) {
1543 pr_warning("sum of op payload lens %d != data_len %d",
1544 payload_len, bytes);
1548 ceph_decode_need(&p, end, 4 + numops * 4, bad_put);
1549 retry_attempt = ceph_decode_32(&p);
1550 for (i = 0; i < numops; i++)
1551 req->r_reply_op_result[i] = ceph_decode_32(&p);
1553 already_completed = req->r_got_reply;
1555 if (!req->r_got_reply) {
1557 req->r_result = result;
1558 dout("handle_reply result %d bytes %d\n", req->r_result,
1560 if (req->r_result == 0)
1561 req->r_result = bytes;
1563 /* in case this is a write and we need to replay, */
1564 req->r_reassert_version.epoch = cpu_to_le32(reassert_epoch);
1565 req->r_reassert_version.version = cpu_to_le64(reassert_version);
1567 req->r_got_reply = 1;
1568 } else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) {
1569 dout("handle_reply tid %llu dup ack\n", tid);
1570 mutex_unlock(&osdc->request_mutex);
1574 dout("handle_reply tid %llu flags %d\n", tid, flags);
1576 if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK))
1577 __register_linger_request(osdc, req);
1579 /* either this is a read, or we got the safe response */
1581 (flags & CEPH_OSD_FLAG_ONDISK) ||
1582 ((flags & CEPH_OSD_FLAG_WRITE) == 0))
1583 __unregister_request(osdc, req);
1585 mutex_unlock(&osdc->request_mutex);
1587 if (!already_completed) {
1588 if (req->r_unsafe_callback &&
1589 result >= 0 && !(flags & CEPH_OSD_FLAG_ONDISK))
1590 req->r_unsafe_callback(req, true);
1591 if (req->r_callback)
1592 req->r_callback(req, msg);
1594 complete_all(&req->r_completion);
1597 if (flags & CEPH_OSD_FLAG_ONDISK) {
1598 if (req->r_unsafe_callback && already_completed)
1599 req->r_unsafe_callback(req, false);
1600 complete_request(req);
1604 dout("req=%p req->r_linger=%d\n", req, req->r_linger);
1605 ceph_osdc_put_request(req);
1609 req->r_result = -EIO;
1610 __unregister_request(osdc, req);
1611 if (req->r_callback)
1612 req->r_callback(req, msg);
1614 complete_all(&req->r_completion);
1615 complete_request(req);
1616 ceph_osdc_put_request(req);
1618 mutex_unlock(&osdc->request_mutex);
1620 pr_err("corrupt osd_op_reply got %d %d\n",
1621 (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len));
1625 static void reset_changed_osds(struct ceph_osd_client *osdc)
1627 struct rb_node *p, *n;
1629 for (p = rb_first(&osdc->osds); p; p = n) {
1630 struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
1633 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
1634 memcmp(&osd->o_con.peer_addr,
1635 ceph_osd_addr(osdc->osdmap,
1637 sizeof(struct ceph_entity_addr)) != 0)
1638 __reset_osd(osdc, osd);
1643 * Requeue requests whose mapping to an OSD has changed. If requests map to
1644 * no osd, request a new map.
1646 * Caller should hold map_sem for read.
1648 static void kick_requests(struct ceph_osd_client *osdc, bool force_resend,
1649 bool force_resend_writes)
1651 struct ceph_osd_request *req, *nreq;
1655 bool force_resend_req;
1657 dout("kick_requests %s %s\n", force_resend ? " (force resend)" : "",
1658 force_resend_writes ? " (force resend writes)" : "");
1659 mutex_lock(&osdc->request_mutex);
1660 for (p = rb_first(&osdc->requests); p; ) {
1661 req = rb_entry(p, struct ceph_osd_request, r_node);
1665 * For linger requests that have not yet been
1666 * registered, move them to the linger list; they'll
1667 * be sent to the osd in the loop below. Unregister
1668 * the request before re-registering it as a linger
1669 * request to ensure the __map_request() below
1670 * will decide it needs to be sent.
1672 if (req->r_linger && list_empty(&req->r_linger_item)) {
1673 dout("%p tid %llu restart on osd%d\n",
1675 req->r_osd ? req->r_osd->o_osd : -1);
1676 ceph_osdc_get_request(req);
1677 __unregister_request(osdc, req);
1678 __register_linger_request(osdc, req);
1679 ceph_osdc_put_request(req);
1683 force_resend_req = force_resend ||
1684 (force_resend_writes &&
1685 req->r_flags & CEPH_OSD_FLAG_WRITE);
1686 err = __map_request(osdc, req, force_resend_req);
1688 continue; /* error */
1689 if (req->r_osd == NULL) {
1690 dout("%p tid %llu maps to no osd\n", req, req->r_tid);
1691 needmap++; /* request a newer map */
1692 } else if (err > 0) {
1693 if (!req->r_linger) {
1694 dout("%p tid %llu requeued on osd%d\n", req,
1696 req->r_osd ? req->r_osd->o_osd : -1);
1697 req->r_flags |= CEPH_OSD_FLAG_RETRY;
1702 list_for_each_entry_safe(req, nreq, &osdc->req_linger,
1704 dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
1706 err = __map_request(osdc, req,
1707 force_resend || force_resend_writes);
1708 dout("__map_request returned %d\n", err);
1710 continue; /* no change and no osd was specified */
1712 continue; /* hrm! */
1713 if (req->r_osd == NULL) {
1714 dout("tid %llu maps to no valid osd\n", req->r_tid);
1715 needmap++; /* request a newer map */
1719 dout("kicking lingering %p tid %llu osd%d\n", req, req->r_tid,
1720 req->r_osd ? req->r_osd->o_osd : -1);
1721 __register_request(osdc, req);
1722 __unregister_linger_request(osdc, req);
1724 reset_changed_osds(osdc);
1725 mutex_unlock(&osdc->request_mutex);
1728 dout("%d requests for down osds, need new map\n", needmap);
1729 ceph_monc_request_next_osdmap(&osdc->client->monc);
1735 * Process updated osd map.
1737 * The message contains any number of incremental and full maps, normally
1738 * indicating some sort of topology change in the cluster. Kick requests
1739 * off to different OSDs as needed.
1741 void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1743 void *p, *end, *next;
1744 u32 nr_maps, maplen;
1746 struct ceph_osdmap *newmap = NULL, *oldmap;
1748 struct ceph_fsid fsid;
1751 dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
1752 p = msg->front.iov_base;
1753 end = p + msg->front.iov_len;
1756 ceph_decode_need(&p, end, sizeof(fsid), bad);
1757 ceph_decode_copy(&p, &fsid, sizeof(fsid));
1758 if (ceph_check_fsid(osdc->client, &fsid) < 0)
1761 down_write(&osdc->map_sem);
1763 was_full = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
1765 /* incremental maps */
1766 ceph_decode_32_safe(&p, end, nr_maps, bad);
1767 dout(" %d inc maps\n", nr_maps);
1768 while (nr_maps > 0) {
1769 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
1770 epoch = ceph_decode_32(&p);
1771 maplen = ceph_decode_32(&p);
1772 ceph_decode_need(&p, end, maplen, bad);
1774 if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
1775 dout("applying incremental map %u len %d\n",
1777 newmap = osdmap_apply_incremental(&p, next,
1779 &osdc->client->msgr);
1780 if (IS_ERR(newmap)) {
1781 err = PTR_ERR(newmap);
1785 if (newmap != osdc->osdmap) {
1786 ceph_osdmap_destroy(osdc->osdmap);
1787 osdc->osdmap = newmap;
1789 was_full = was_full ||
1790 ceph_osdmap_flag(osdc->osdmap,
1792 kick_requests(osdc, 0, was_full);
1794 dout("ignoring incremental map %u len %d\n",
1804 ceph_decode_32_safe(&p, end, nr_maps, bad);
1805 dout(" %d full maps\n", nr_maps);
1807 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
1808 epoch = ceph_decode_32(&p);
1809 maplen = ceph_decode_32(&p);
1810 ceph_decode_need(&p, end, maplen, bad);
1812 dout("skipping non-latest full map %u len %d\n",
1814 } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
1815 dout("skipping full map %u len %d, "
1816 "older than our %u\n", epoch, maplen,
1817 osdc->osdmap->epoch);
1819 int skipped_map = 0;
1821 dout("taking full map %u len %d\n", epoch, maplen);
1822 newmap = osdmap_decode(&p, p+maplen);
1823 if (IS_ERR(newmap)) {
1824 err = PTR_ERR(newmap);
1828 oldmap = osdc->osdmap;
1829 osdc->osdmap = newmap;
1831 if (oldmap->epoch + 1 < newmap->epoch)
1833 ceph_osdmap_destroy(oldmap);
1835 was_full = was_full ||
1836 ceph_osdmap_flag(osdc->osdmap,
1838 kick_requests(osdc, skipped_map, was_full);
1847 downgrade_write(&osdc->map_sem);
1848 ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch);
1851 * subscribe to subsequent osdmap updates if full to ensure
1852 * we find out when we are no longer full and stop returning
1855 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) ||
1856 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD) ||
1857 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR))
1858 ceph_monc_request_next_osdmap(&osdc->client->monc);
1860 mutex_lock(&osdc->request_mutex);
1861 __send_queued(osdc);
1862 mutex_unlock(&osdc->request_mutex);
1863 up_read(&osdc->map_sem);
1864 wake_up_all(&osdc->client->auth_wq);
1868 pr_err("osdc handle_map corrupt msg\n");
1870 up_write(&osdc->map_sem);
1875 * watch/notify callback event infrastructure
1877 * These callbacks are used both for watch and notify operations.
1879 static void __release_event(struct kref *kref)
1881 struct ceph_osd_event *event =
1882 container_of(kref, struct ceph_osd_event, kref);
1884 dout("__release_event %p\n", event);
1888 static void get_event(struct ceph_osd_event *event)
1890 kref_get(&event->kref);
1893 void ceph_osdc_put_event(struct ceph_osd_event *event)
1895 kref_put(&event->kref, __release_event);
1897 EXPORT_SYMBOL(ceph_osdc_put_event);
1899 static void __insert_event(struct ceph_osd_client *osdc,
1900 struct ceph_osd_event *new)
1902 struct rb_node **p = &osdc->event_tree.rb_node;
1903 struct rb_node *parent = NULL;
1904 struct ceph_osd_event *event = NULL;
1908 event = rb_entry(parent, struct ceph_osd_event, node);
1909 if (new->cookie < event->cookie)
1911 else if (new->cookie > event->cookie)
1912 p = &(*p)->rb_right;
1917 rb_link_node(&new->node, parent, p);
1918 rb_insert_color(&new->node, &osdc->event_tree);
1921 static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
1924 struct rb_node **p = &osdc->event_tree.rb_node;
1925 struct rb_node *parent = NULL;
1926 struct ceph_osd_event *event = NULL;
1930 event = rb_entry(parent, struct ceph_osd_event, node);
1931 if (cookie < event->cookie)
1933 else if (cookie > event->cookie)
1934 p = &(*p)->rb_right;
1941 static void __remove_event(struct ceph_osd_event *event)
1943 struct ceph_osd_client *osdc = event->osdc;
1945 if (!RB_EMPTY_NODE(&event->node)) {
1946 dout("__remove_event removed %p\n", event);
1947 rb_erase(&event->node, &osdc->event_tree);
1948 ceph_osdc_put_event(event);
1950 dout("__remove_event didn't remove %p\n", event);
1954 int ceph_osdc_create_event(struct ceph_osd_client *osdc,
1955 void (*event_cb)(u64, u64, u8, void *),
1956 void *data, struct ceph_osd_event **pevent)
1958 struct ceph_osd_event *event;
1960 event = kmalloc(sizeof(*event), GFP_NOIO);
1964 dout("create_event %p\n", event);
1965 event->cb = event_cb;
1966 event->one_shot = 0;
1969 INIT_LIST_HEAD(&event->osd_node);
1970 RB_CLEAR_NODE(&event->node);
1971 kref_init(&event->kref); /* one ref for us */
1972 kref_get(&event->kref); /* one ref for the caller */
1974 spin_lock(&osdc->event_lock);
1975 event->cookie = ++osdc->event_count;
1976 __insert_event(osdc, event);
1977 spin_unlock(&osdc->event_lock);
1982 EXPORT_SYMBOL(ceph_osdc_create_event);
1984 void ceph_osdc_cancel_event(struct ceph_osd_event *event)
1986 struct ceph_osd_client *osdc = event->osdc;
1988 dout("cancel_event %p\n", event);
1989 spin_lock(&osdc->event_lock);
1990 __remove_event(event);
1991 spin_unlock(&osdc->event_lock);
1992 ceph_osdc_put_event(event); /* caller's */
1994 EXPORT_SYMBOL(ceph_osdc_cancel_event);
1997 static void do_event_work(struct work_struct *work)
1999 struct ceph_osd_event_work *event_work =
2000 container_of(work, struct ceph_osd_event_work, work);
2001 struct ceph_osd_event *event = event_work->event;
2002 u64 ver = event_work->ver;
2003 u64 notify_id = event_work->notify_id;
2004 u8 opcode = event_work->opcode;
2006 dout("do_event_work completing %p\n", event);
2007 event->cb(ver, notify_id, opcode, event->data);
2008 dout("do_event_work completed %p\n", event);
2009 ceph_osdc_put_event(event);
2015 * Process osd watch notifications
2017 static void handle_watch_notify(struct ceph_osd_client *osdc,
2018 struct ceph_msg *msg)
2022 u64 cookie, ver, notify_id;
2024 struct ceph_osd_event *event;
2025 struct ceph_osd_event_work *event_work;
2027 p = msg->front.iov_base;
2028 end = p + msg->front.iov_len;
2030 ceph_decode_8_safe(&p, end, proto_ver, bad);
2031 ceph_decode_8_safe(&p, end, opcode, bad);
2032 ceph_decode_64_safe(&p, end, cookie, bad);
2033 ceph_decode_64_safe(&p, end, ver, bad);
2034 ceph_decode_64_safe(&p, end, notify_id, bad);
2036 spin_lock(&osdc->event_lock);
2037 event = __find_event(osdc, cookie);
2039 BUG_ON(event->one_shot);
2042 spin_unlock(&osdc->event_lock);
2043 dout("handle_watch_notify cookie %lld ver %lld event %p\n",
2044 cookie, ver, event);
2046 event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
2048 dout("ERROR: could not allocate event_work\n");
2051 INIT_WORK(&event_work->work, do_event_work);
2052 event_work->event = event;
2053 event_work->ver = ver;
2054 event_work->notify_id = notify_id;
2055 event_work->opcode = opcode;
2056 if (!queue_work(osdc->notify_wq, &event_work->work)) {
2057 dout("WARNING: failed to queue notify event work\n");
2065 ceph_osdc_put_event(event);
2069 pr_err("osdc handle_watch_notify corrupt msg\n");
2074 * build new request AND message
2077 void ceph_osdc_build_request(struct ceph_osd_request *req, u64 off,
2078 struct ceph_snap_context *snapc, u64 snap_id,
2079 struct timespec *mtime)
2081 struct ceph_msg *msg = req->r_request;
2084 int flags = req->r_flags;
2088 req->r_snapid = snap_id;
2089 req->r_snapc = ceph_get_snap_context(snapc);
2091 /* encode request */
2092 msg->hdr.version = cpu_to_le16(4);
2094 p = msg->front.iov_base;
2095 ceph_encode_32(&p, 1); /* client_inc is always 1 */
2096 req->r_request_osdmap_epoch = p;
2098 req->r_request_flags = p;
2100 if (req->r_flags & CEPH_OSD_FLAG_WRITE)
2101 ceph_encode_timespec(p, mtime);
2102 p += sizeof(struct ceph_timespec);
2103 req->r_request_reassert_version = p;
2104 p += sizeof(struct ceph_eversion); /* will get filled in */
2107 ceph_encode_8(&p, 4);
2108 ceph_encode_8(&p, 4);
2109 ceph_encode_32(&p, 8 + 4 + 4);
2110 req->r_request_pool = p;
2112 ceph_encode_32(&p, -1); /* preferred */
2113 ceph_encode_32(&p, 0); /* key len */
2115 ceph_encode_8(&p, 1);
2116 req->r_request_pgid = p;
2118 ceph_encode_32(&p, -1); /* preferred */
2121 ceph_encode_32(&p, req->r_oid.name_len);
2122 memcpy(p, req->r_oid.name, req->r_oid.name_len);
2123 dout("oid '%.*s' len %d\n", req->r_oid.name_len,
2124 req->r_oid.name, req->r_oid.name_len);
2125 p += req->r_oid.name_len;
2127 /* ops--can imply data */
2128 ceph_encode_16(&p, (u16)req->r_num_ops);
2130 for (i = 0; i < req->r_num_ops; i++) {
2131 data_len += osd_req_encode_op(req, p, i);
2132 p += sizeof(struct ceph_osd_op);
2136 ceph_encode_64(&p, req->r_snapid);
2137 ceph_encode_64(&p, req->r_snapc ? req->r_snapc->seq : 0);
2138 ceph_encode_32(&p, req->r_snapc ? req->r_snapc->num_snaps : 0);
2140 for (i = 0; i < snapc->num_snaps; i++) {
2141 ceph_encode_64(&p, req->r_snapc->snaps[i]);
2145 req->r_request_attempts = p;
2149 if (flags & CEPH_OSD_FLAG_WRITE) {
2153 * The header "data_off" is a hint to the receiver
2154 * allowing it to align received data into its
2155 * buffers such that there's no need to re-copy
2156 * it before writing it to disk (direct I/O).
2158 data_off = (u16) (off & 0xffff);
2159 req->r_request->hdr.data_off = cpu_to_le16(data_off);
2161 req->r_request->hdr.data_len = cpu_to_le32(data_len);
2163 BUG_ON(p > msg->front.iov_base + msg->front.iov_len);
2164 msg_size = p - msg->front.iov_base;
2165 msg->front.iov_len = msg_size;
2166 msg->hdr.front_len = cpu_to_le32(msg_size);
2168 dout("build_request msg_size was %d\n", (int)msg_size);
2170 EXPORT_SYMBOL(ceph_osdc_build_request);
2173 * Register request, send initial attempt.
2175 int ceph_osdc_start_request(struct ceph_osd_client *osdc,
2176 struct ceph_osd_request *req,
2181 down_read(&osdc->map_sem);
2182 mutex_lock(&osdc->request_mutex);
2183 __register_request(osdc, req);
2185 req->r_got_reply = 0;
2186 rc = __map_request(osdc, req, 0);
2189 dout("osdc_start_request failed map, "
2190 " will retry %lld\n", req->r_tid);
2193 __unregister_request(osdc, req);
2197 if (req->r_osd == NULL) {
2198 dout("send_request %p no up osds in pg\n", req);
2199 ceph_monc_request_next_osdmap(&osdc->client->monc);
2201 __send_queued(osdc);
2205 mutex_unlock(&osdc->request_mutex);
2206 up_read(&osdc->map_sem);
2209 EXPORT_SYMBOL(ceph_osdc_start_request);
2212 * wait for a request to complete
2214 int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
2215 struct ceph_osd_request *req)
2219 rc = wait_for_completion_interruptible(&req->r_completion);
2221 mutex_lock(&osdc->request_mutex);
2222 __cancel_request(req);
2223 __unregister_request(osdc, req);
2224 mutex_unlock(&osdc->request_mutex);
2225 complete_request(req);
2226 dout("wait_request tid %llu canceled/timed out\n", req->r_tid);
2230 dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result);
2231 return req->r_result;
2233 EXPORT_SYMBOL(ceph_osdc_wait_request);
2236 * sync - wait for all in-flight requests to flush. avoid starvation.
2238 void ceph_osdc_sync(struct ceph_osd_client *osdc)
2240 struct ceph_osd_request *req;
2241 u64 last_tid, next_tid = 0;
2243 mutex_lock(&osdc->request_mutex);
2244 last_tid = osdc->last_tid;
2246 req = __lookup_request_ge(osdc, next_tid);
2249 if (req->r_tid > last_tid)
2252 next_tid = req->r_tid + 1;
2253 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
2256 ceph_osdc_get_request(req);
2257 mutex_unlock(&osdc->request_mutex);
2258 dout("sync waiting on tid %llu (last is %llu)\n",
2259 req->r_tid, last_tid);
2260 wait_for_completion(&req->r_safe_completion);
2261 mutex_lock(&osdc->request_mutex);
2262 ceph_osdc_put_request(req);
2264 mutex_unlock(&osdc->request_mutex);
2265 dout("sync done (thru tid %llu)\n", last_tid);
2267 EXPORT_SYMBOL(ceph_osdc_sync);
2270 * Call all pending notify callbacks - for use after a watch is
2271 * unregistered, to make sure no more callbacks for it will be invoked
2273 extern void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc)
2275 flush_workqueue(osdc->notify_wq);
2277 EXPORT_SYMBOL(ceph_osdc_flush_notifies);
2283 int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
2288 osdc->client = client;
2289 osdc->osdmap = NULL;
2290 init_rwsem(&osdc->map_sem);
2291 init_completion(&osdc->map_waiters);
2292 osdc->last_requested_map = 0;
2293 mutex_init(&osdc->request_mutex);
2295 osdc->osds = RB_ROOT;
2296 INIT_LIST_HEAD(&osdc->osd_lru);
2297 osdc->requests = RB_ROOT;
2298 INIT_LIST_HEAD(&osdc->req_lru);
2299 INIT_LIST_HEAD(&osdc->req_unsent);
2300 INIT_LIST_HEAD(&osdc->req_notarget);
2301 INIT_LIST_HEAD(&osdc->req_linger);
2302 osdc->num_requests = 0;
2303 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
2304 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
2305 spin_lock_init(&osdc->event_lock);
2306 osdc->event_tree = RB_ROOT;
2307 osdc->event_count = 0;
2309 schedule_delayed_work(&osdc->osds_timeout_work,
2310 round_jiffies_relative(osdc->client->options->osd_idle_ttl * HZ));
2313 osdc->req_mempool = mempool_create_kmalloc_pool(10,
2314 sizeof(struct ceph_osd_request));
2315 if (!osdc->req_mempool)
2318 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
2319 OSD_OP_FRONT_LEN, 10, true,
2323 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
2324 OSD_OPREPLY_FRONT_LEN, 10, true,
2330 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
2331 if (!osdc->notify_wq)
2336 ceph_msgpool_destroy(&osdc->msgpool_op);
2338 mempool_destroy(osdc->req_mempool);
2343 void ceph_osdc_stop(struct ceph_osd_client *osdc)
2345 flush_workqueue(osdc->notify_wq);
2346 destroy_workqueue(osdc->notify_wq);
2347 cancel_delayed_work_sync(&osdc->timeout_work);
2348 cancel_delayed_work_sync(&osdc->osds_timeout_work);
2350 ceph_osdmap_destroy(osdc->osdmap);
2351 osdc->osdmap = NULL;
2353 remove_all_osds(osdc);
2354 mempool_destroy(osdc->req_mempool);
2355 ceph_msgpool_destroy(&osdc->msgpool_op);
2356 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
2360 * Read some contiguous pages. If we cross a stripe boundary, shorten
2361 * *plen. Return number of bytes read, or error.
2363 int ceph_osdc_readpages(struct ceph_osd_client *osdc,
2364 struct ceph_vino vino, struct ceph_file_layout *layout,
2366 u32 truncate_seq, u64 truncate_size,
2367 struct page **pages, int num_pages, int page_align)
2369 struct ceph_osd_request *req;
2372 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
2373 vino.snap, off, *plen);
2374 req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 1,
2375 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
2376 NULL, truncate_seq, truncate_size,
2379 return PTR_ERR(req);
2381 /* it may be a short read due to an object boundary */
2383 osd_req_op_extent_osd_data_pages(req, 0,
2384 pages, *plen, page_align, false, false);
2386 dout("readpages final extent is %llu~%llu (%llu bytes align %d)\n",
2387 off, *plen, *plen, page_align);
2389 ceph_osdc_build_request(req, off, NULL, vino.snap, NULL);
2391 rc = ceph_osdc_start_request(osdc, req, false);
2393 rc = ceph_osdc_wait_request(osdc, req);
2395 ceph_osdc_put_request(req);
2396 dout("readpages result %d\n", rc);
2399 EXPORT_SYMBOL(ceph_osdc_readpages);
2402 * do a synchronous write on N pages
2404 int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
2405 struct ceph_file_layout *layout,
2406 struct ceph_snap_context *snapc,
2408 u32 truncate_seq, u64 truncate_size,
2409 struct timespec *mtime,
2410 struct page **pages, int num_pages)
2412 struct ceph_osd_request *req;
2414 int page_align = off & ~PAGE_MASK;
2416 BUG_ON(vino.snap != CEPH_NOSNAP); /* snapshots aren't writeable */
2417 req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 1,
2419 CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
2420 snapc, truncate_seq, truncate_size,
2423 return PTR_ERR(req);
2425 /* it may be a short write due to an object boundary */
2426 osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align,
2428 dout("writepages %llu~%llu (%llu bytes)\n", off, len, len);
2430 ceph_osdc_build_request(req, off, snapc, CEPH_NOSNAP, mtime);
2432 rc = ceph_osdc_start_request(osdc, req, true);
2434 rc = ceph_osdc_wait_request(osdc, req);
2436 ceph_osdc_put_request(req);
2439 dout("writepages result %d\n", rc);
2442 EXPORT_SYMBOL(ceph_osdc_writepages);
2444 int ceph_osdc_setup(void)
2446 BUG_ON(ceph_osd_request_cache);
2447 ceph_osd_request_cache = kmem_cache_create("ceph_osd_request",
2448 sizeof (struct ceph_osd_request),
2449 __alignof__(struct ceph_osd_request),
2452 return ceph_osd_request_cache ? 0 : -ENOMEM;
2454 EXPORT_SYMBOL(ceph_osdc_setup);
2456 void ceph_osdc_cleanup(void)
2458 BUG_ON(!ceph_osd_request_cache);
2459 kmem_cache_destroy(ceph_osd_request_cache);
2460 ceph_osd_request_cache = NULL;
2462 EXPORT_SYMBOL(ceph_osdc_cleanup);
2465 * handle incoming message
2467 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
2469 struct ceph_osd *osd = con->private;
2470 struct ceph_osd_client *osdc;
2471 int type = le16_to_cpu(msg->hdr.type);
2478 case CEPH_MSG_OSD_MAP:
2479 ceph_osdc_handle_map(osdc, msg);
2481 case CEPH_MSG_OSD_OPREPLY:
2482 handle_reply(osdc, msg, con);
2484 case CEPH_MSG_WATCH_NOTIFY:
2485 handle_watch_notify(osdc, msg);
2489 pr_err("received unknown message type %d %s\n", type,
2490 ceph_msg_type_name(type));
2497 * lookup and return message for incoming reply. set up reply message
2500 static struct ceph_msg *get_reply(struct ceph_connection *con,
2501 struct ceph_msg_header *hdr,
2504 struct ceph_osd *osd = con->private;
2505 struct ceph_osd_client *osdc = osd->o_osdc;
2507 struct ceph_osd_request *req;
2508 int front_len = le32_to_cpu(hdr->front_len);
2509 int data_len = le32_to_cpu(hdr->data_len);
2512 tid = le64_to_cpu(hdr->tid);
2513 mutex_lock(&osdc->request_mutex);
2514 req = __lookup_request(osdc, tid);
2518 dout("get_reply unknown tid %llu from osd%d\n", tid,
2523 if (req->r_reply->con)
2524 dout("%s revoking msg %p from old con %p\n", __func__,
2525 req->r_reply, req->r_reply->con);
2526 ceph_msg_revoke_incoming(req->r_reply);
2528 if (front_len > req->r_reply->front_alloc_len) {
2529 pr_warning("get_reply front %d > preallocated %d (%u#%llu)\n",
2530 front_len, req->r_reply->front_alloc_len,
2531 (unsigned int)con->peer_name.type,
2532 le64_to_cpu(con->peer_name.num));
2533 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS,
2537 ceph_msg_put(req->r_reply);
2540 m = ceph_msg_get(req->r_reply);
2543 struct ceph_osd_data *osd_data;
2546 * XXX This is assuming there is only one op containing
2547 * XXX page data. Probably OK for reads, but this
2548 * XXX ought to be done more generally.
2550 osd_data = osd_req_op_extent_osd_data(req, 0);
2551 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
2552 if (osd_data->pages &&
2553 unlikely(osd_data->length < data_len)) {
2555 pr_warning("tid %lld reply has %d bytes "
2556 "we had only %llu bytes ready\n",
2557 tid, data_len, osd_data->length);
2566 dout("get_reply tid %lld %p\n", tid, m);
2569 mutex_unlock(&osdc->request_mutex);
2574 static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2575 struct ceph_msg_header *hdr,
2578 struct ceph_osd *osd = con->private;
2579 int type = le16_to_cpu(hdr->type);
2580 int front = le32_to_cpu(hdr->front_len);
2584 case CEPH_MSG_OSD_MAP:
2585 case CEPH_MSG_WATCH_NOTIFY:
2586 return ceph_msg_new(type, front, GFP_NOFS, false);
2587 case CEPH_MSG_OSD_OPREPLY:
2588 return get_reply(con, hdr, skip);
2590 pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
2598 * Wrappers to refcount containing ceph_osd struct
2600 static struct ceph_connection *get_osd_con(struct ceph_connection *con)
2602 struct ceph_osd *osd = con->private;
2608 static void put_osd_con(struct ceph_connection *con)
2610 struct ceph_osd *osd = con->private;
2618 * Note: returned pointer is the address of a structure that's
2619 * managed separately. Caller must *not* attempt to free it.
2621 static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
2622 int *proto, int force_new)
2624 struct ceph_osd *o = con->private;
2625 struct ceph_osd_client *osdc = o->o_osdc;
2626 struct ceph_auth_client *ac = osdc->client->monc.auth;
2627 struct ceph_auth_handshake *auth = &o->o_auth;
2629 if (force_new && auth->authorizer) {
2630 ceph_auth_destroy_authorizer(ac, auth->authorizer);
2631 auth->authorizer = NULL;
2633 if (!auth->authorizer) {
2634 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2637 return ERR_PTR(ret);
2639 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2642 return ERR_PTR(ret);
2644 *proto = ac->protocol;
2650 static int verify_authorizer_reply(struct ceph_connection *con, int len)
2652 struct ceph_osd *o = con->private;
2653 struct ceph_osd_client *osdc = o->o_osdc;
2654 struct ceph_auth_client *ac = osdc->client->monc.auth;
2656 return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer, len);
2659 static int invalidate_authorizer(struct ceph_connection *con)
2661 struct ceph_osd *o = con->private;
2662 struct ceph_osd_client *osdc = o->o_osdc;
2663 struct ceph_auth_client *ac = osdc->client->monc.auth;
2665 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
2666 return ceph_monc_validate_auth(&osdc->client->monc);
2669 static const struct ceph_connection_operations osd_con_ops = {
2672 .dispatch = dispatch,
2673 .get_authorizer = get_authorizer,
2674 .verify_authorizer_reply = verify_authorizer_reply,
2675 .invalidate_authorizer = invalidate_authorizer,
2676 .alloc_msg = alloc_msg,