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 + MAX_OBJ_NAME_SIZE;
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 /* create reply message */
373 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0);
375 msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY,
376 OSD_OPREPLY_FRONT_LEN, gfp_flags, true);
378 ceph_osdc_put_request(req);
383 /* create request message; allow space for oid */
385 msg = ceph_msgpool_get(&osdc->msgpool_op, 0);
387 msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp_flags, true);
389 ceph_osdc_put_request(req);
393 memset(msg->front.iov_base, 0, msg->front.iov_len);
395 req->r_request = msg;
399 EXPORT_SYMBOL(ceph_osdc_alloc_request);
401 static bool osd_req_opcode_valid(u16 opcode)
404 case CEPH_OSD_OP_READ:
405 case CEPH_OSD_OP_STAT:
406 case CEPH_OSD_OP_MAPEXT:
407 case CEPH_OSD_OP_MASKTRUNC:
408 case CEPH_OSD_OP_SPARSE_READ:
409 case CEPH_OSD_OP_NOTIFY:
410 case CEPH_OSD_OP_NOTIFY_ACK:
411 case CEPH_OSD_OP_ASSERT_VER:
412 case CEPH_OSD_OP_WRITE:
413 case CEPH_OSD_OP_WRITEFULL:
414 case CEPH_OSD_OP_TRUNCATE:
415 case CEPH_OSD_OP_ZERO:
416 case CEPH_OSD_OP_DELETE:
417 case CEPH_OSD_OP_APPEND:
418 case CEPH_OSD_OP_STARTSYNC:
419 case CEPH_OSD_OP_SETTRUNC:
420 case CEPH_OSD_OP_TRIMTRUNC:
421 case CEPH_OSD_OP_TMAPUP:
422 case CEPH_OSD_OP_TMAPPUT:
423 case CEPH_OSD_OP_TMAPGET:
424 case CEPH_OSD_OP_CREATE:
425 case CEPH_OSD_OP_ROLLBACK:
426 case CEPH_OSD_OP_WATCH:
427 case CEPH_OSD_OP_OMAPGETKEYS:
428 case CEPH_OSD_OP_OMAPGETVALS:
429 case CEPH_OSD_OP_OMAPGETHEADER:
430 case CEPH_OSD_OP_OMAPGETVALSBYKEYS:
431 case CEPH_OSD_OP_OMAPSETVALS:
432 case CEPH_OSD_OP_OMAPSETHEADER:
433 case CEPH_OSD_OP_OMAPCLEAR:
434 case CEPH_OSD_OP_OMAPRMKEYS:
435 case CEPH_OSD_OP_OMAP_CMP:
436 case CEPH_OSD_OP_CLONERANGE:
437 case CEPH_OSD_OP_ASSERT_SRC_VERSION:
438 case CEPH_OSD_OP_SRC_CMPXATTR:
439 case CEPH_OSD_OP_GETXATTR:
440 case CEPH_OSD_OP_GETXATTRS:
441 case CEPH_OSD_OP_CMPXATTR:
442 case CEPH_OSD_OP_SETXATTR:
443 case CEPH_OSD_OP_SETXATTRS:
444 case CEPH_OSD_OP_RESETXATTRS:
445 case CEPH_OSD_OP_RMXATTR:
446 case CEPH_OSD_OP_PULL:
447 case CEPH_OSD_OP_PUSH:
448 case CEPH_OSD_OP_BALANCEREADS:
449 case CEPH_OSD_OP_UNBALANCEREADS:
450 case CEPH_OSD_OP_SCRUB:
451 case CEPH_OSD_OP_SCRUB_RESERVE:
452 case CEPH_OSD_OP_SCRUB_UNRESERVE:
453 case CEPH_OSD_OP_SCRUB_STOP:
454 case CEPH_OSD_OP_SCRUB_MAP:
455 case CEPH_OSD_OP_WRLOCK:
456 case CEPH_OSD_OP_WRUNLOCK:
457 case CEPH_OSD_OP_RDLOCK:
458 case CEPH_OSD_OP_RDUNLOCK:
459 case CEPH_OSD_OP_UPLOCK:
460 case CEPH_OSD_OP_DNLOCK:
461 case CEPH_OSD_OP_CALL:
462 case CEPH_OSD_OP_PGLS:
463 case CEPH_OSD_OP_PGLS_FILTER:
471 * This is an osd op init function for opcodes that have no data or
472 * other information associated with them. It also serves as a
473 * common init routine for all the other init functions, below.
475 static struct ceph_osd_req_op *
476 _osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which,
479 struct ceph_osd_req_op *op;
481 BUG_ON(which >= osd_req->r_num_ops);
482 BUG_ON(!osd_req_opcode_valid(opcode));
484 op = &osd_req->r_ops[which];
485 memset(op, 0, sizeof (*op));
491 void osd_req_op_init(struct ceph_osd_request *osd_req,
492 unsigned int which, u16 opcode)
494 (void)_osd_req_op_init(osd_req, which, opcode);
496 EXPORT_SYMBOL(osd_req_op_init);
498 void osd_req_op_extent_init(struct ceph_osd_request *osd_req,
499 unsigned int which, u16 opcode,
500 u64 offset, u64 length,
501 u64 truncate_size, u32 truncate_seq)
503 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
504 size_t payload_len = 0;
506 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE);
508 op->extent.offset = offset;
509 op->extent.length = length;
510 op->extent.truncate_size = truncate_size;
511 op->extent.truncate_seq = truncate_seq;
512 if (opcode == CEPH_OSD_OP_WRITE)
513 payload_len += length;
515 op->payload_len = payload_len;
517 EXPORT_SYMBOL(osd_req_op_extent_init);
519 void osd_req_op_extent_update(struct ceph_osd_request *osd_req,
520 unsigned int which, u64 length)
522 struct ceph_osd_req_op *op;
525 BUG_ON(which >= osd_req->r_num_ops);
526 op = &osd_req->r_ops[which];
527 previous = op->extent.length;
529 if (length == previous)
530 return; /* Nothing to do */
531 BUG_ON(length > previous);
533 op->extent.length = length;
534 op->payload_len -= previous - length;
536 EXPORT_SYMBOL(osd_req_op_extent_update);
538 void osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
539 u16 opcode, const char *class, const char *method)
541 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
542 struct ceph_pagelist *pagelist;
543 size_t payload_len = 0;
546 BUG_ON(opcode != CEPH_OSD_OP_CALL);
548 pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
550 ceph_pagelist_init(pagelist);
552 op->cls.class_name = class;
553 size = strlen(class);
554 BUG_ON(size > (size_t) U8_MAX);
555 op->cls.class_len = size;
556 ceph_pagelist_append(pagelist, class, size);
559 op->cls.method_name = method;
560 size = strlen(method);
561 BUG_ON(size > (size_t) U8_MAX);
562 op->cls.method_len = size;
563 ceph_pagelist_append(pagelist, method, size);
566 osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist);
568 op->cls.argc = 0; /* currently unused */
570 op->payload_len = payload_len;
572 EXPORT_SYMBOL(osd_req_op_cls_init);
574 void osd_req_op_watch_init(struct ceph_osd_request *osd_req,
575 unsigned int which, u16 opcode,
576 u64 cookie, u64 version, int flag)
578 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
580 BUG_ON(opcode != CEPH_OSD_OP_NOTIFY_ACK && opcode != CEPH_OSD_OP_WATCH);
582 op->watch.cookie = cookie;
583 op->watch.ver = version;
584 if (opcode == CEPH_OSD_OP_WATCH && flag)
585 op->watch.flag = (u8)1;
587 EXPORT_SYMBOL(osd_req_op_watch_init);
589 static void ceph_osdc_msg_data_add(struct ceph_msg *msg,
590 struct ceph_osd_data *osd_data)
592 u64 length = ceph_osd_data_length(osd_data);
594 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
595 BUG_ON(length > (u64) SIZE_MAX);
597 ceph_msg_data_add_pages(msg, osd_data->pages,
598 length, osd_data->alignment);
599 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
601 ceph_msg_data_add_pagelist(msg, osd_data->pagelist);
603 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
604 ceph_msg_data_add_bio(msg, osd_data->bio, length);
607 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
611 static u64 osd_req_encode_op(struct ceph_osd_request *req,
612 struct ceph_osd_op *dst, unsigned int which)
614 struct ceph_osd_req_op *src;
615 struct ceph_osd_data *osd_data;
616 u64 request_data_len = 0;
619 BUG_ON(which >= req->r_num_ops);
620 src = &req->r_ops[which];
621 if (WARN_ON(!osd_req_opcode_valid(src->op))) {
622 pr_err("unrecognized osd opcode %d\n", src->op);
628 case CEPH_OSD_OP_STAT:
629 osd_data = &src->raw_data_in;
630 ceph_osdc_msg_data_add(req->r_reply, osd_data);
632 case CEPH_OSD_OP_READ:
633 case CEPH_OSD_OP_WRITE:
634 if (src->op == CEPH_OSD_OP_WRITE)
635 request_data_len = src->extent.length;
636 dst->extent.offset = cpu_to_le64(src->extent.offset);
637 dst->extent.length = cpu_to_le64(src->extent.length);
638 dst->extent.truncate_size =
639 cpu_to_le64(src->extent.truncate_size);
640 dst->extent.truncate_seq =
641 cpu_to_le32(src->extent.truncate_seq);
642 osd_data = &src->extent.osd_data;
643 if (src->op == CEPH_OSD_OP_WRITE)
644 ceph_osdc_msg_data_add(req->r_request, osd_data);
646 ceph_osdc_msg_data_add(req->r_reply, osd_data);
648 case CEPH_OSD_OP_CALL:
649 dst->cls.class_len = src->cls.class_len;
650 dst->cls.method_len = src->cls.method_len;
651 osd_data = &src->cls.request_info;
652 ceph_osdc_msg_data_add(req->r_request, osd_data);
653 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGELIST);
654 request_data_len = osd_data->pagelist->length;
656 osd_data = &src->cls.request_data;
657 data_length = ceph_osd_data_length(osd_data);
659 BUG_ON(osd_data->type == CEPH_OSD_DATA_TYPE_NONE);
660 dst->cls.indata_len = cpu_to_le32(data_length);
661 ceph_osdc_msg_data_add(req->r_request, osd_data);
662 src->payload_len += data_length;
663 request_data_len += data_length;
665 osd_data = &src->cls.response_data;
666 ceph_osdc_msg_data_add(req->r_reply, osd_data);
668 case CEPH_OSD_OP_STARTSYNC:
670 case CEPH_OSD_OP_NOTIFY_ACK:
671 case CEPH_OSD_OP_WATCH:
672 dst->watch.cookie = cpu_to_le64(src->watch.cookie);
673 dst->watch.ver = cpu_to_le64(src->watch.ver);
674 dst->watch.flag = src->watch.flag;
677 pr_err("unsupported osd opcode %s\n",
678 ceph_osd_op_name(src->op));
683 dst->op = cpu_to_le16(src->op);
684 dst->payload_len = cpu_to_le32(src->payload_len);
686 return request_data_len;
690 * build new request AND message, calculate layout, and adjust file
693 * if the file was recently truncated, we include information about its
694 * old and new size so that the object can be updated appropriately. (we
695 * avoid synchronously deleting truncated objects because it's slow.)
697 * if @do_sync, include a 'startsync' command so that the osd will flush
700 struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
701 struct ceph_file_layout *layout,
702 struct ceph_vino vino,
703 u64 off, u64 *plen, int num_ops,
704 int opcode, int flags,
705 struct ceph_snap_context *snapc,
710 struct ceph_osd_request *req;
718 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE);
720 req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool,
723 return ERR_PTR(-ENOMEM);
725 req->r_flags = flags;
727 /* calculate max write size */
728 r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
730 ceph_osdc_put_request(req);
734 object_size = le32_to_cpu(layout->fl_object_size);
735 object_base = off - objoff;
736 if (!(truncate_seq == 1 && truncate_size == -1ULL)) {
737 if (truncate_size <= object_base) {
740 truncate_size -= object_base;
741 if (truncate_size > object_size)
742 truncate_size = object_size;
746 osd_req_op_extent_init(req, 0, opcode, objoff, objlen,
747 truncate_size, truncate_seq);
750 * A second op in the ops array means the caller wants to
751 * also issue a include a 'startsync' command so that the
752 * osd will flush data quickly.
755 osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC);
757 req->r_file_layout = *layout; /* keep a copy */
759 snprintf(req->r_oid, sizeof(req->r_oid), "%llx.%08llx",
761 req->r_oid_len = strlen(req->r_oid);
765 EXPORT_SYMBOL(ceph_osdc_new_request);
768 * We keep osd requests in an rbtree, sorted by ->r_tid.
770 static void __insert_request(struct ceph_osd_client *osdc,
771 struct ceph_osd_request *new)
773 struct rb_node **p = &osdc->requests.rb_node;
774 struct rb_node *parent = NULL;
775 struct ceph_osd_request *req = NULL;
779 req = rb_entry(parent, struct ceph_osd_request, r_node);
780 if (new->r_tid < req->r_tid)
782 else if (new->r_tid > req->r_tid)
788 rb_link_node(&new->r_node, parent, p);
789 rb_insert_color(&new->r_node, &osdc->requests);
792 static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc,
795 struct ceph_osd_request *req;
796 struct rb_node *n = osdc->requests.rb_node;
799 req = rb_entry(n, struct ceph_osd_request, r_node);
800 if (tid < req->r_tid)
802 else if (tid > req->r_tid)
810 static struct ceph_osd_request *
811 __lookup_request_ge(struct ceph_osd_client *osdc,
814 struct ceph_osd_request *req;
815 struct rb_node *n = osdc->requests.rb_node;
818 req = rb_entry(n, struct ceph_osd_request, r_node);
819 if (tid < req->r_tid) {
823 } else if (tid > req->r_tid) {
833 * Resubmit requests pending on the given osd.
835 static void __kick_osd_requests(struct ceph_osd_client *osdc,
836 struct ceph_osd *osd)
838 struct ceph_osd_request *req, *nreq;
842 dout("__kick_osd_requests osd%d\n", osd->o_osd);
843 err = __reset_osd(osdc, osd);
847 * Build up a list of requests to resend by traversing the
848 * osd's list of requests. Requests for a given object are
849 * sent in tid order, and that is also the order they're
850 * kept on this list. Therefore all requests that are in
851 * flight will be found first, followed by all requests that
852 * have not yet been sent. And to resend requests while
853 * preserving this order we will want to put any sent
854 * requests back on the front of the osd client's unsent
857 * So we build a separate ordered list of already-sent
858 * requests for the affected osd and splice it onto the
859 * front of the osd client's unsent list. Once we've seen a
860 * request that has not yet been sent we're done. Those
861 * requests are already sitting right where they belong.
863 list_for_each_entry(req, &osd->o_requests, r_osd_item) {
866 list_move_tail(&req->r_req_lru_item, &resend);
867 dout("requeueing %p tid %llu osd%d\n", req, req->r_tid,
870 req->r_flags |= CEPH_OSD_FLAG_RETRY;
872 list_splice(&resend, &osdc->req_unsent);
875 * Linger requests are re-registered before sending, which
876 * sets up a new tid for each. We add them to the unsent
877 * list at the end to keep things in tid order.
879 list_for_each_entry_safe(req, nreq, &osd->o_linger_requests,
882 * reregister request prior to unregistering linger so
883 * that r_osd is preserved.
885 BUG_ON(!list_empty(&req->r_req_lru_item));
886 __register_request(osdc, req);
887 list_add_tail(&req->r_req_lru_item, &osdc->req_unsent);
888 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
889 __unregister_linger_request(osdc, req);
890 dout("requeued lingering %p tid %llu osd%d\n", req, req->r_tid,
896 * If the osd connection drops, we need to resubmit all requests.
898 static void osd_reset(struct ceph_connection *con)
900 struct ceph_osd *osd = con->private;
901 struct ceph_osd_client *osdc;
905 dout("osd_reset osd%d\n", osd->o_osd);
907 down_read(&osdc->map_sem);
908 mutex_lock(&osdc->request_mutex);
909 __kick_osd_requests(osdc, osd);
911 mutex_unlock(&osdc->request_mutex);
912 up_read(&osdc->map_sem);
916 * Track open sessions with osds.
918 static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
920 struct ceph_osd *osd;
922 osd = kzalloc(sizeof(*osd), GFP_NOFS);
926 atomic_set(&osd->o_ref, 1);
929 RB_CLEAR_NODE(&osd->o_node);
930 INIT_LIST_HEAD(&osd->o_requests);
931 INIT_LIST_HEAD(&osd->o_linger_requests);
932 INIT_LIST_HEAD(&osd->o_osd_lru);
933 osd->o_incarnation = 1;
935 ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
937 INIT_LIST_HEAD(&osd->o_keepalive_item);
941 static struct ceph_osd *get_osd(struct ceph_osd *osd)
943 if (atomic_inc_not_zero(&osd->o_ref)) {
944 dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
945 atomic_read(&osd->o_ref));
948 dout("get_osd %p FAIL\n", osd);
953 static void put_osd(struct ceph_osd *osd)
955 dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
956 atomic_read(&osd->o_ref) - 1);
957 if (atomic_dec_and_test(&osd->o_ref) && osd->o_auth.authorizer) {
958 struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth;
960 ceph_auth_destroy_authorizer(ac, osd->o_auth.authorizer);
966 * remove an osd from our map
968 static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
970 dout("%s %p osd%d\n", __func__, osd, osd->o_osd);
971 WARN_ON(!list_empty(&osd->o_requests));
972 WARN_ON(!list_empty(&osd->o_linger_requests));
974 list_del_init(&osd->o_osd_lru);
975 rb_erase(&osd->o_node, &osdc->osds);
976 RB_CLEAR_NODE(&osd->o_node);
979 static void remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
981 dout("%s %p osd%d\n", __func__, osd, osd->o_osd);
983 if (!RB_EMPTY_NODE(&osd->o_node)) {
984 ceph_con_close(&osd->o_con);
985 __remove_osd(osdc, osd);
990 static void remove_all_osds(struct ceph_osd_client *osdc)
992 dout("%s %p\n", __func__, osdc);
993 mutex_lock(&osdc->request_mutex);
994 while (!RB_EMPTY_ROOT(&osdc->osds)) {
995 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
996 struct ceph_osd, o_node);
997 remove_osd(osdc, osd);
999 mutex_unlock(&osdc->request_mutex);
1002 static void __move_osd_to_lru(struct ceph_osd_client *osdc,
1003 struct ceph_osd *osd)
1005 dout("__move_osd_to_lru %p\n", osd);
1006 BUG_ON(!list_empty(&osd->o_osd_lru));
1007 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
1008 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl * HZ;
1011 static void __remove_osd_from_lru(struct ceph_osd *osd)
1013 dout("__remove_osd_from_lru %p\n", osd);
1014 if (!list_empty(&osd->o_osd_lru))
1015 list_del_init(&osd->o_osd_lru);
1018 static void remove_old_osds(struct ceph_osd_client *osdc)
1020 struct ceph_osd *osd, *nosd;
1022 dout("__remove_old_osds %p\n", osdc);
1023 mutex_lock(&osdc->request_mutex);
1024 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
1025 if (time_before(jiffies, osd->lru_ttl))
1027 remove_osd(osdc, osd);
1029 mutex_unlock(&osdc->request_mutex);
1035 static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
1037 struct ceph_entity_addr *peer_addr;
1039 dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
1040 if (list_empty(&osd->o_requests) &&
1041 list_empty(&osd->o_linger_requests)) {
1042 remove_osd(osdc, osd);
1046 peer_addr = &osdc->osdmap->osd_addr[osd->o_osd];
1047 if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1048 !ceph_con_opened(&osd->o_con)) {
1049 struct ceph_osd_request *req;
1051 dout(" osd addr hasn't changed and connection never opened,"
1052 " letting msgr retry");
1053 /* touch each r_stamp for handle_timeout()'s benfit */
1054 list_for_each_entry(req, &osd->o_requests, r_osd_item)
1055 req->r_stamp = jiffies;
1060 ceph_con_close(&osd->o_con);
1061 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1062 osd->o_incarnation++;
1067 static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new)
1069 struct rb_node **p = &osdc->osds.rb_node;
1070 struct rb_node *parent = NULL;
1071 struct ceph_osd *osd = NULL;
1073 dout("__insert_osd %p osd%d\n", new, new->o_osd);
1076 osd = rb_entry(parent, struct ceph_osd, o_node);
1077 if (new->o_osd < osd->o_osd)
1079 else if (new->o_osd > osd->o_osd)
1080 p = &(*p)->rb_right;
1085 rb_link_node(&new->o_node, parent, p);
1086 rb_insert_color(&new->o_node, &osdc->osds);
1089 static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o)
1091 struct ceph_osd *osd;
1092 struct rb_node *n = osdc->osds.rb_node;
1095 osd = rb_entry(n, struct ceph_osd, o_node);
1098 else if (o > osd->o_osd)
1106 static void __schedule_osd_timeout(struct ceph_osd_client *osdc)
1108 schedule_delayed_work(&osdc->timeout_work,
1109 osdc->client->options->osd_keepalive_timeout * HZ);
1112 static void __cancel_osd_timeout(struct ceph_osd_client *osdc)
1114 cancel_delayed_work(&osdc->timeout_work);
1118 * Register request, assign tid. If this is the first request, set up
1119 * the timeout event.
1121 static void __register_request(struct ceph_osd_client *osdc,
1122 struct ceph_osd_request *req)
1124 req->r_tid = ++osdc->last_tid;
1125 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
1126 dout("__register_request %p tid %lld\n", req, req->r_tid);
1127 __insert_request(osdc, req);
1128 ceph_osdc_get_request(req);
1129 osdc->num_requests++;
1130 if (osdc->num_requests == 1) {
1131 dout(" first request, scheduling timeout\n");
1132 __schedule_osd_timeout(osdc);
1137 * called under osdc->request_mutex
1139 static void __unregister_request(struct ceph_osd_client *osdc,
1140 struct ceph_osd_request *req)
1142 if (RB_EMPTY_NODE(&req->r_node)) {
1143 dout("__unregister_request %p tid %lld not registered\n",
1148 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
1149 rb_erase(&req->r_node, &osdc->requests);
1150 osdc->num_requests--;
1153 /* make sure the original request isn't in flight. */
1154 ceph_msg_revoke(req->r_request);
1156 list_del_init(&req->r_osd_item);
1157 if (list_empty(&req->r_osd->o_requests) &&
1158 list_empty(&req->r_osd->o_linger_requests)) {
1159 dout("moving osd to %p lru\n", req->r_osd);
1160 __move_osd_to_lru(osdc, req->r_osd);
1162 if (list_empty(&req->r_linger_item))
1166 list_del_init(&req->r_req_lru_item);
1167 ceph_osdc_put_request(req);
1169 if (osdc->num_requests == 0) {
1170 dout(" no requests, canceling timeout\n");
1171 __cancel_osd_timeout(osdc);
1176 * Cancel a previously queued request message
1178 static void __cancel_request(struct ceph_osd_request *req)
1180 if (req->r_sent && req->r_osd) {
1181 ceph_msg_revoke(req->r_request);
1186 static void __register_linger_request(struct ceph_osd_client *osdc,
1187 struct ceph_osd_request *req)
1189 dout("__register_linger_request %p\n", req);
1190 ceph_osdc_get_request(req);
1191 list_add_tail(&req->r_linger_item, &osdc->req_linger);
1193 list_add_tail(&req->r_linger_osd,
1194 &req->r_osd->o_linger_requests);
1197 static void __unregister_linger_request(struct ceph_osd_client *osdc,
1198 struct ceph_osd_request *req)
1200 dout("__unregister_linger_request %p\n", req);
1201 list_del_init(&req->r_linger_item);
1203 list_del_init(&req->r_linger_osd);
1205 if (list_empty(&req->r_osd->o_requests) &&
1206 list_empty(&req->r_osd->o_linger_requests)) {
1207 dout("moving osd to %p lru\n", req->r_osd);
1208 __move_osd_to_lru(osdc, req->r_osd);
1210 if (list_empty(&req->r_osd_item))
1213 ceph_osdc_put_request(req);
1216 void ceph_osdc_unregister_linger_request(struct ceph_osd_client *osdc,
1217 struct ceph_osd_request *req)
1219 mutex_lock(&osdc->request_mutex);
1220 if (req->r_linger) {
1222 __unregister_linger_request(osdc, req);
1224 mutex_unlock(&osdc->request_mutex);
1226 EXPORT_SYMBOL(ceph_osdc_unregister_linger_request);
1228 void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
1229 struct ceph_osd_request *req)
1231 if (!req->r_linger) {
1232 dout("set_request_linger %p\n", req);
1236 EXPORT_SYMBOL(ceph_osdc_set_request_linger);
1239 * Returns whether a request should be blocked from being sent
1240 * based on the current osdmap and osd_client settings.
1242 * Caller should hold map_sem for read.
1244 static bool __req_should_be_paused(struct ceph_osd_client *osdc,
1245 struct ceph_osd_request *req)
1247 bool pauserd = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD);
1248 bool pausewr = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR) ||
1249 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
1250 return (req->r_flags & CEPH_OSD_FLAG_READ && pauserd) ||
1251 (req->r_flags & CEPH_OSD_FLAG_WRITE && pausewr);
1255 * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
1256 * (as needed), and set the request r_osd appropriately. If there is
1257 * no up osd, set r_osd to NULL. Move the request to the appropriate list
1258 * (unsent, homeless) or leave on in-flight lru.
1260 * Return 0 if unchanged, 1 if changed, or negative on error.
1262 * Caller should hold map_sem for read and request_mutex.
1264 static int __map_request(struct ceph_osd_client *osdc,
1265 struct ceph_osd_request *req, int force_resend)
1267 struct ceph_pg pgid;
1268 int acting[CEPH_PG_MAX_SIZE];
1269 int o = -1, num = 0;
1273 dout("map_request %p tid %lld\n", req, req->r_tid);
1274 err = ceph_calc_ceph_pg(&pgid, req->r_oid, osdc->osdmap,
1275 ceph_file_layout_pg_pool(req->r_file_layout));
1277 list_move(&req->r_req_lru_item, &osdc->req_notarget);
1282 err = ceph_calc_pg_acting(osdc->osdmap, pgid, acting);
1288 was_paused = req->r_paused;
1289 req->r_paused = __req_should_be_paused(osdc, req);
1290 if (was_paused && !req->r_paused)
1293 if ((!force_resend &&
1294 req->r_osd && req->r_osd->o_osd == o &&
1295 req->r_sent >= req->r_osd->o_incarnation &&
1296 req->r_num_pg_osds == num &&
1297 memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
1298 (req->r_osd == NULL && o == -1) ||
1300 return 0; /* no change */
1302 dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n",
1303 req->r_tid, pgid.pool, pgid.seed, o,
1304 req->r_osd ? req->r_osd->o_osd : -1);
1306 /* record full pg acting set */
1307 memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
1308 req->r_num_pg_osds = num;
1311 __cancel_request(req);
1312 list_del_init(&req->r_osd_item);
1316 req->r_osd = __lookup_osd(osdc, o);
1317 if (!req->r_osd && o >= 0) {
1319 req->r_osd = create_osd(osdc, o);
1321 list_move(&req->r_req_lru_item, &osdc->req_notarget);
1325 dout("map_request osd %p is osd%d\n", req->r_osd, o);
1326 __insert_osd(osdc, req->r_osd);
1328 ceph_con_open(&req->r_osd->o_con,
1329 CEPH_ENTITY_TYPE_OSD, o,
1330 &osdc->osdmap->osd_addr[o]);
1334 __remove_osd_from_lru(req->r_osd);
1335 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
1336 list_move_tail(&req->r_req_lru_item, &osdc->req_unsent);
1338 list_move_tail(&req->r_req_lru_item, &osdc->req_notarget);
1340 err = 1; /* osd or pg changed */
1347 * caller should hold map_sem (for read) and request_mutex
1349 static void __send_request(struct ceph_osd_client *osdc,
1350 struct ceph_osd_request *req)
1354 dout("send_request %p tid %llu to osd%d flags %d pg %lld.%x\n",
1355 req, req->r_tid, req->r_osd->o_osd, req->r_flags,
1356 (unsigned long long)req->r_pgid.pool, req->r_pgid.seed);
1358 /* fill in message content that changes each time we send it */
1359 put_unaligned_le32(osdc->osdmap->epoch, req->r_request_osdmap_epoch);
1360 put_unaligned_le32(req->r_flags, req->r_request_flags);
1361 put_unaligned_le64(req->r_pgid.pool, req->r_request_pool);
1362 p = req->r_request_pgid;
1363 ceph_encode_64(&p, req->r_pgid.pool);
1364 ceph_encode_32(&p, req->r_pgid.seed);
1365 put_unaligned_le64(1, req->r_request_attempts); /* FIXME */
1366 memcpy(req->r_request_reassert_version, &req->r_reassert_version,
1367 sizeof(req->r_reassert_version));
1369 req->r_stamp = jiffies;
1370 list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
1372 ceph_msg_get(req->r_request); /* send consumes a ref */
1374 req->r_sent = req->r_osd->o_incarnation;
1376 ceph_con_send(&req->r_osd->o_con, req->r_request);
1380 * Send any requests in the queue (req_unsent).
1382 static void __send_queued(struct ceph_osd_client *osdc)
1384 struct ceph_osd_request *req, *tmp;
1386 dout("__send_queued\n");
1387 list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item)
1388 __send_request(osdc, req);
1392 * Timeout callback, called every N seconds when 1 or more osd
1393 * requests has been active for more than N seconds. When this
1394 * happens, we ping all OSDs with requests who have timed out to
1395 * ensure any communications channel reset is detected. Reset the
1396 * request timeouts another N seconds in the future as we go.
1397 * Reschedule the timeout event another N seconds in future (unless
1398 * there are no open requests).
1400 static void handle_timeout(struct work_struct *work)
1402 struct ceph_osd_client *osdc =
1403 container_of(work, struct ceph_osd_client, timeout_work.work);
1404 struct ceph_osd_request *req;
1405 struct ceph_osd *osd;
1406 unsigned long keepalive =
1407 osdc->client->options->osd_keepalive_timeout * HZ;
1408 struct list_head slow_osds;
1410 down_read(&osdc->map_sem);
1412 ceph_monc_request_next_osdmap(&osdc->client->monc);
1414 mutex_lock(&osdc->request_mutex);
1417 * ping osds that are a bit slow. this ensures that if there
1418 * is a break in the TCP connection we will notice, and reopen
1419 * a connection with that osd (from the fault callback).
1421 INIT_LIST_HEAD(&slow_osds);
1422 list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
1423 if (time_before(jiffies, req->r_stamp + keepalive))
1428 dout(" tid %llu is slow, will send keepalive on osd%d\n",
1429 req->r_tid, osd->o_osd);
1430 list_move_tail(&osd->o_keepalive_item, &slow_osds);
1432 while (!list_empty(&slow_osds)) {
1433 osd = list_entry(slow_osds.next, struct ceph_osd,
1435 list_del_init(&osd->o_keepalive_item);
1436 ceph_con_keepalive(&osd->o_con);
1439 __schedule_osd_timeout(osdc);
1440 __send_queued(osdc);
1441 mutex_unlock(&osdc->request_mutex);
1442 up_read(&osdc->map_sem);
1445 static void handle_osds_timeout(struct work_struct *work)
1447 struct ceph_osd_client *osdc =
1448 container_of(work, struct ceph_osd_client,
1449 osds_timeout_work.work);
1450 unsigned long delay =
1451 osdc->client->options->osd_idle_ttl * HZ >> 2;
1453 dout("osds timeout\n");
1454 down_read(&osdc->map_sem);
1455 remove_old_osds(osdc);
1456 up_read(&osdc->map_sem);
1458 schedule_delayed_work(&osdc->osds_timeout_work,
1459 round_jiffies_relative(delay));
1462 static void complete_request(struct ceph_osd_request *req)
1464 complete_all(&req->r_safe_completion); /* fsync waiter */
1468 * handle osd op reply. either call the callback if it is specified,
1469 * or do the completion to wake up the waiting thread.
1471 static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg,
1472 struct ceph_connection *con)
1475 struct ceph_osd_request *req;
1478 unsigned int numops;
1479 int payload_len, flags;
1485 u64 reassert_version;
1487 int already_completed;
1491 tid = le64_to_cpu(msg->hdr.tid);
1492 dout("handle_reply %p tid %llu\n", msg, tid);
1494 p = msg->front.iov_base;
1495 end = p + msg->front.iov_len;
1497 ceph_decode_need(&p, end, 4, bad);
1498 object_len = ceph_decode_32(&p);
1499 ceph_decode_need(&p, end, object_len, bad);
1502 err = ceph_decode_pgid(&p, end, &pg);
1506 ceph_decode_need(&p, end, 8 + 4 + 4 + 8 + 4, bad);
1507 flags = ceph_decode_64(&p);
1508 result = ceph_decode_32(&p);
1509 reassert_epoch = ceph_decode_32(&p);
1510 reassert_version = ceph_decode_64(&p);
1511 osdmap_epoch = ceph_decode_32(&p);
1514 mutex_lock(&osdc->request_mutex);
1515 req = __lookup_request(osdc, tid);
1517 dout("handle_reply tid %llu dne\n", tid);
1520 ceph_osdc_get_request(req);
1522 dout("handle_reply %p tid %llu req %p result %d\n", msg, tid,
1525 ceph_decode_need(&p, end, 4, bad_put);
1526 numops = ceph_decode_32(&p);
1527 if (numops > CEPH_OSD_MAX_OP)
1529 if (numops != req->r_num_ops)
1532 ceph_decode_need(&p, end, numops * sizeof(struct ceph_osd_op), bad_put);
1533 for (i = 0; i < numops; i++) {
1534 struct ceph_osd_op *op = p;
1537 len = le32_to_cpu(op->payload_len);
1538 req->r_reply_op_len[i] = len;
1539 dout(" op %d has %d bytes\n", i, len);
1543 bytes = le32_to_cpu(msg->hdr.data_len);
1544 if (payload_len != bytes) {
1545 pr_warning("sum of op payload lens %d != data_len %d",
1546 payload_len, bytes);
1550 ceph_decode_need(&p, end, 4 + numops * 4, bad_put);
1551 retry_attempt = ceph_decode_32(&p);
1552 for (i = 0; i < numops; i++)
1553 req->r_reply_op_result[i] = ceph_decode_32(&p);
1555 already_completed = req->r_got_reply;
1557 if (!req->r_got_reply) {
1559 req->r_result = result;
1560 dout("handle_reply result %d bytes %d\n", req->r_result,
1562 if (req->r_result == 0)
1563 req->r_result = bytes;
1565 /* in case this is a write and we need to replay, */
1566 req->r_reassert_version.epoch = cpu_to_le32(reassert_epoch);
1567 req->r_reassert_version.version = cpu_to_le64(reassert_version);
1569 req->r_got_reply = 1;
1570 } else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) {
1571 dout("handle_reply tid %llu dup ack\n", tid);
1572 mutex_unlock(&osdc->request_mutex);
1576 dout("handle_reply tid %llu flags %d\n", tid, flags);
1578 if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK))
1579 __register_linger_request(osdc, req);
1581 /* either this is a read, or we got the safe response */
1583 (flags & CEPH_OSD_FLAG_ONDISK) ||
1584 ((flags & CEPH_OSD_FLAG_WRITE) == 0))
1585 __unregister_request(osdc, req);
1587 mutex_unlock(&osdc->request_mutex);
1589 if (!already_completed) {
1590 if (req->r_unsafe_callback &&
1591 result >= 0 && !(flags & CEPH_OSD_FLAG_ONDISK))
1592 req->r_unsafe_callback(req, true);
1593 if (req->r_callback)
1594 req->r_callback(req, msg);
1596 complete_all(&req->r_completion);
1599 if (flags & CEPH_OSD_FLAG_ONDISK) {
1600 if (req->r_unsafe_callback && already_completed)
1601 req->r_unsafe_callback(req, false);
1602 complete_request(req);
1606 dout("req=%p req->r_linger=%d\n", req, req->r_linger);
1607 ceph_osdc_put_request(req);
1611 ceph_osdc_put_request(req);
1613 mutex_unlock(&osdc->request_mutex);
1615 pr_err("corrupt osd_op_reply got %d %d\n",
1616 (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len));
1620 static void reset_changed_osds(struct ceph_osd_client *osdc)
1622 struct rb_node *p, *n;
1624 dout("%s %p\n", __func__, osdc);
1625 for (p = rb_first(&osdc->osds); p; p = n) {
1626 struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
1629 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
1630 memcmp(&osd->o_con.peer_addr,
1631 ceph_osd_addr(osdc->osdmap,
1633 sizeof(struct ceph_entity_addr)) != 0)
1634 __reset_osd(osdc, osd);
1639 * Requeue requests whose mapping to an OSD has changed. If requests map to
1640 * no osd, request a new map.
1642 * Caller should hold map_sem for read.
1644 static void kick_requests(struct ceph_osd_client *osdc, bool force_resend,
1645 bool force_resend_writes)
1647 struct ceph_osd_request *req, *nreq;
1651 bool force_resend_req;
1653 dout("kick_requests %s %s\n", force_resend ? " (force resend)" : "",
1654 force_resend_writes ? " (force resend writes)" : "");
1655 mutex_lock(&osdc->request_mutex);
1656 for (p = rb_first(&osdc->requests); p; ) {
1657 req = rb_entry(p, struct ceph_osd_request, r_node);
1661 * For linger requests that have not yet been
1662 * registered, move them to the linger list; they'll
1663 * be sent to the osd in the loop below. Unregister
1664 * the request before re-registering it as a linger
1665 * request to ensure the __map_request() below
1666 * will decide it needs to be sent.
1668 if (req->r_linger && list_empty(&req->r_linger_item)) {
1669 dout("%p tid %llu restart on osd%d\n",
1671 req->r_osd ? req->r_osd->o_osd : -1);
1672 ceph_osdc_get_request(req);
1673 __unregister_request(osdc, req);
1674 __register_linger_request(osdc, req);
1675 ceph_osdc_put_request(req);
1679 force_resend_req = force_resend ||
1680 (force_resend_writes &&
1681 req->r_flags & CEPH_OSD_FLAG_WRITE);
1682 err = __map_request(osdc, req, force_resend_req);
1684 continue; /* error */
1685 if (req->r_osd == NULL) {
1686 dout("%p tid %llu maps to no osd\n", req, req->r_tid);
1687 needmap++; /* request a newer map */
1688 } else if (err > 0) {
1689 if (!req->r_linger) {
1690 dout("%p tid %llu requeued on osd%d\n", req,
1692 req->r_osd ? req->r_osd->o_osd : -1);
1693 req->r_flags |= CEPH_OSD_FLAG_RETRY;
1698 list_for_each_entry_safe(req, nreq, &osdc->req_linger,
1700 dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
1702 err = __map_request(osdc, req,
1703 force_resend || force_resend_writes);
1704 dout("__map_request returned %d\n", err);
1706 continue; /* hrm! */
1707 if (req->r_osd == NULL || err > 0) {
1708 if (req->r_osd == NULL) {
1709 dout("lingering %p tid %llu maps to no osd\n",
1712 * A homeless lingering request makes
1713 * no sense, as it's job is to keep
1714 * a particular OSD connection open.
1715 * Request a newer map and kick the
1716 * request, knowing that it won't be
1717 * resent until we actually get a map
1718 * that can tell us where to send it.
1723 dout("kicking lingering %p tid %llu osd%d\n", req,
1724 req->r_tid, req->r_osd ? req->r_osd->o_osd : -1);
1725 __register_request(osdc, req);
1726 __unregister_linger_request(osdc, req);
1729 reset_changed_osds(osdc);
1730 mutex_unlock(&osdc->request_mutex);
1733 dout("%d requests for down osds, need new map\n", needmap);
1734 ceph_monc_request_next_osdmap(&osdc->client->monc);
1740 * Process updated osd map.
1742 * The message contains any number of incremental and full maps, normally
1743 * indicating some sort of topology change in the cluster. Kick requests
1744 * off to different OSDs as needed.
1746 void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1748 void *p, *end, *next;
1749 u32 nr_maps, maplen;
1751 struct ceph_osdmap *newmap = NULL, *oldmap;
1753 struct ceph_fsid fsid;
1756 dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
1757 p = msg->front.iov_base;
1758 end = p + msg->front.iov_len;
1761 ceph_decode_need(&p, end, sizeof(fsid), bad);
1762 ceph_decode_copy(&p, &fsid, sizeof(fsid));
1763 if (ceph_check_fsid(osdc->client, &fsid) < 0)
1766 down_write(&osdc->map_sem);
1768 was_full = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
1770 /* incremental maps */
1771 ceph_decode_32_safe(&p, end, nr_maps, bad);
1772 dout(" %d inc maps\n", nr_maps);
1773 while (nr_maps > 0) {
1774 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
1775 epoch = ceph_decode_32(&p);
1776 maplen = ceph_decode_32(&p);
1777 ceph_decode_need(&p, end, maplen, bad);
1779 if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
1780 dout("applying incremental map %u len %d\n",
1782 newmap = osdmap_apply_incremental(&p, next,
1784 &osdc->client->msgr);
1785 if (IS_ERR(newmap)) {
1786 err = PTR_ERR(newmap);
1790 if (newmap != osdc->osdmap) {
1791 ceph_osdmap_destroy(osdc->osdmap);
1792 osdc->osdmap = newmap;
1794 was_full = was_full ||
1795 ceph_osdmap_flag(osdc->osdmap,
1797 kick_requests(osdc, 0, was_full);
1799 dout("ignoring incremental map %u len %d\n",
1809 ceph_decode_32_safe(&p, end, nr_maps, bad);
1810 dout(" %d full maps\n", nr_maps);
1812 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
1813 epoch = ceph_decode_32(&p);
1814 maplen = ceph_decode_32(&p);
1815 ceph_decode_need(&p, end, maplen, bad);
1817 dout("skipping non-latest full map %u len %d\n",
1819 } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
1820 dout("skipping full map %u len %d, "
1821 "older than our %u\n", epoch, maplen,
1822 osdc->osdmap->epoch);
1824 int skipped_map = 0;
1826 dout("taking full map %u len %d\n", epoch, maplen);
1827 newmap = osdmap_decode(&p, p+maplen);
1828 if (IS_ERR(newmap)) {
1829 err = PTR_ERR(newmap);
1833 oldmap = osdc->osdmap;
1834 osdc->osdmap = newmap;
1836 if (oldmap->epoch + 1 < newmap->epoch)
1838 ceph_osdmap_destroy(oldmap);
1840 was_full = was_full ||
1841 ceph_osdmap_flag(osdc->osdmap,
1843 kick_requests(osdc, skipped_map, was_full);
1852 downgrade_write(&osdc->map_sem);
1853 ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch);
1856 * subscribe to subsequent osdmap updates if full to ensure
1857 * we find out when we are no longer full and stop returning
1860 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) ||
1861 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD) ||
1862 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR))
1863 ceph_monc_request_next_osdmap(&osdc->client->monc);
1865 mutex_lock(&osdc->request_mutex);
1866 __send_queued(osdc);
1867 mutex_unlock(&osdc->request_mutex);
1868 up_read(&osdc->map_sem);
1869 wake_up_all(&osdc->client->auth_wq);
1873 pr_err("osdc handle_map corrupt msg\n");
1875 up_write(&osdc->map_sem);
1880 * watch/notify callback event infrastructure
1882 * These callbacks are used both for watch and notify operations.
1884 static void __release_event(struct kref *kref)
1886 struct ceph_osd_event *event =
1887 container_of(kref, struct ceph_osd_event, kref);
1889 dout("__release_event %p\n", event);
1893 static void get_event(struct ceph_osd_event *event)
1895 kref_get(&event->kref);
1898 void ceph_osdc_put_event(struct ceph_osd_event *event)
1900 kref_put(&event->kref, __release_event);
1902 EXPORT_SYMBOL(ceph_osdc_put_event);
1904 static void __insert_event(struct ceph_osd_client *osdc,
1905 struct ceph_osd_event *new)
1907 struct rb_node **p = &osdc->event_tree.rb_node;
1908 struct rb_node *parent = NULL;
1909 struct ceph_osd_event *event = NULL;
1913 event = rb_entry(parent, struct ceph_osd_event, node);
1914 if (new->cookie < event->cookie)
1916 else if (new->cookie > event->cookie)
1917 p = &(*p)->rb_right;
1922 rb_link_node(&new->node, parent, p);
1923 rb_insert_color(&new->node, &osdc->event_tree);
1926 static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
1929 struct rb_node **p = &osdc->event_tree.rb_node;
1930 struct rb_node *parent = NULL;
1931 struct ceph_osd_event *event = NULL;
1935 event = rb_entry(parent, struct ceph_osd_event, node);
1936 if (cookie < event->cookie)
1938 else if (cookie > event->cookie)
1939 p = &(*p)->rb_right;
1946 static void __remove_event(struct ceph_osd_event *event)
1948 struct ceph_osd_client *osdc = event->osdc;
1950 if (!RB_EMPTY_NODE(&event->node)) {
1951 dout("__remove_event removed %p\n", event);
1952 rb_erase(&event->node, &osdc->event_tree);
1953 ceph_osdc_put_event(event);
1955 dout("__remove_event didn't remove %p\n", event);
1959 int ceph_osdc_create_event(struct ceph_osd_client *osdc,
1960 void (*event_cb)(u64, u64, u8, void *),
1961 void *data, struct ceph_osd_event **pevent)
1963 struct ceph_osd_event *event;
1965 event = kmalloc(sizeof(*event), GFP_NOIO);
1969 dout("create_event %p\n", event);
1970 event->cb = event_cb;
1971 event->one_shot = 0;
1974 INIT_LIST_HEAD(&event->osd_node);
1975 RB_CLEAR_NODE(&event->node);
1976 kref_init(&event->kref); /* one ref for us */
1977 kref_get(&event->kref); /* one ref for the caller */
1979 spin_lock(&osdc->event_lock);
1980 event->cookie = ++osdc->event_count;
1981 __insert_event(osdc, event);
1982 spin_unlock(&osdc->event_lock);
1987 EXPORT_SYMBOL(ceph_osdc_create_event);
1989 void ceph_osdc_cancel_event(struct ceph_osd_event *event)
1991 struct ceph_osd_client *osdc = event->osdc;
1993 dout("cancel_event %p\n", event);
1994 spin_lock(&osdc->event_lock);
1995 __remove_event(event);
1996 spin_unlock(&osdc->event_lock);
1997 ceph_osdc_put_event(event); /* caller's */
1999 EXPORT_SYMBOL(ceph_osdc_cancel_event);
2002 static void do_event_work(struct work_struct *work)
2004 struct ceph_osd_event_work *event_work =
2005 container_of(work, struct ceph_osd_event_work, work);
2006 struct ceph_osd_event *event = event_work->event;
2007 u64 ver = event_work->ver;
2008 u64 notify_id = event_work->notify_id;
2009 u8 opcode = event_work->opcode;
2011 dout("do_event_work completing %p\n", event);
2012 event->cb(ver, notify_id, opcode, event->data);
2013 dout("do_event_work completed %p\n", event);
2014 ceph_osdc_put_event(event);
2020 * Process osd watch notifications
2022 static void handle_watch_notify(struct ceph_osd_client *osdc,
2023 struct ceph_msg *msg)
2027 u64 cookie, ver, notify_id;
2029 struct ceph_osd_event *event;
2030 struct ceph_osd_event_work *event_work;
2032 p = msg->front.iov_base;
2033 end = p + msg->front.iov_len;
2035 ceph_decode_8_safe(&p, end, proto_ver, bad);
2036 ceph_decode_8_safe(&p, end, opcode, bad);
2037 ceph_decode_64_safe(&p, end, cookie, bad);
2038 ceph_decode_64_safe(&p, end, ver, bad);
2039 ceph_decode_64_safe(&p, end, notify_id, bad);
2041 spin_lock(&osdc->event_lock);
2042 event = __find_event(osdc, cookie);
2044 BUG_ON(event->one_shot);
2047 spin_unlock(&osdc->event_lock);
2048 dout("handle_watch_notify cookie %lld ver %lld event %p\n",
2049 cookie, ver, event);
2051 event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
2053 dout("ERROR: could not allocate event_work\n");
2056 INIT_WORK(&event_work->work, do_event_work);
2057 event_work->event = event;
2058 event_work->ver = ver;
2059 event_work->notify_id = notify_id;
2060 event_work->opcode = opcode;
2061 if (!queue_work(osdc->notify_wq, &event_work->work)) {
2062 dout("WARNING: failed to queue notify event work\n");
2070 ceph_osdc_put_event(event);
2074 pr_err("osdc handle_watch_notify corrupt msg\n");
2079 * build new request AND message
2082 void ceph_osdc_build_request(struct ceph_osd_request *req, u64 off,
2083 struct ceph_snap_context *snapc, u64 snap_id,
2084 struct timespec *mtime)
2086 struct ceph_msg *msg = req->r_request;
2089 int flags = req->r_flags;
2093 req->r_snapid = snap_id;
2094 req->r_snapc = ceph_get_snap_context(snapc);
2096 /* encode request */
2097 msg->hdr.version = cpu_to_le16(4);
2099 p = msg->front.iov_base;
2100 ceph_encode_32(&p, 1); /* client_inc is always 1 */
2101 req->r_request_osdmap_epoch = p;
2103 req->r_request_flags = p;
2105 if (req->r_flags & CEPH_OSD_FLAG_WRITE)
2106 ceph_encode_timespec(p, mtime);
2107 p += sizeof(struct ceph_timespec);
2108 req->r_request_reassert_version = p;
2109 p += sizeof(struct ceph_eversion); /* will get filled in */
2112 ceph_encode_8(&p, 4);
2113 ceph_encode_8(&p, 4);
2114 ceph_encode_32(&p, 8 + 4 + 4);
2115 req->r_request_pool = p;
2117 ceph_encode_32(&p, -1); /* preferred */
2118 ceph_encode_32(&p, 0); /* key len */
2120 ceph_encode_8(&p, 1);
2121 req->r_request_pgid = p;
2123 ceph_encode_32(&p, -1); /* preferred */
2126 ceph_encode_32(&p, req->r_oid_len);
2127 memcpy(p, req->r_oid, req->r_oid_len);
2128 dout("oid '%.*s' len %d\n", req->r_oid_len, req->r_oid, req->r_oid_len);
2129 p += req->r_oid_len;
2131 /* ops--can imply data */
2132 ceph_encode_16(&p, (u16)req->r_num_ops);
2134 for (i = 0; i < req->r_num_ops; i++) {
2135 data_len += osd_req_encode_op(req, p, i);
2136 p += sizeof(struct ceph_osd_op);
2140 ceph_encode_64(&p, req->r_snapid);
2141 ceph_encode_64(&p, req->r_snapc ? req->r_snapc->seq : 0);
2142 ceph_encode_32(&p, req->r_snapc ? req->r_snapc->num_snaps : 0);
2144 for (i = 0; i < snapc->num_snaps; i++) {
2145 ceph_encode_64(&p, req->r_snapc->snaps[i]);
2149 req->r_request_attempts = p;
2153 if (flags & CEPH_OSD_FLAG_WRITE) {
2157 * The header "data_off" is a hint to the receiver
2158 * allowing it to align received data into its
2159 * buffers such that there's no need to re-copy
2160 * it before writing it to disk (direct I/O).
2162 data_off = (u16) (off & 0xffff);
2163 req->r_request->hdr.data_off = cpu_to_le16(data_off);
2165 req->r_request->hdr.data_len = cpu_to_le32(data_len);
2167 BUG_ON(p > msg->front.iov_base + msg->front.iov_len);
2168 msg_size = p - msg->front.iov_base;
2169 msg->front.iov_len = msg_size;
2170 msg->hdr.front_len = cpu_to_le32(msg_size);
2172 dout("build_request msg_size was %d\n", (int)msg_size);
2174 EXPORT_SYMBOL(ceph_osdc_build_request);
2177 * Register request, send initial attempt.
2179 int ceph_osdc_start_request(struct ceph_osd_client *osdc,
2180 struct ceph_osd_request *req,
2185 down_read(&osdc->map_sem);
2186 mutex_lock(&osdc->request_mutex);
2187 __register_request(osdc, req);
2189 req->r_got_reply = 0;
2190 rc = __map_request(osdc, req, 0);
2193 dout("osdc_start_request failed map, "
2194 " will retry %lld\n", req->r_tid);
2197 __unregister_request(osdc, req);
2201 if (req->r_osd == NULL) {
2202 dout("send_request %p no up osds in pg\n", req);
2203 ceph_monc_request_next_osdmap(&osdc->client->monc);
2205 __send_queued(osdc);
2209 mutex_unlock(&osdc->request_mutex);
2210 up_read(&osdc->map_sem);
2213 EXPORT_SYMBOL(ceph_osdc_start_request);
2216 * wait for a request to complete
2218 int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
2219 struct ceph_osd_request *req)
2223 rc = wait_for_completion_interruptible(&req->r_completion);
2225 mutex_lock(&osdc->request_mutex);
2226 __cancel_request(req);
2227 __unregister_request(osdc, req);
2228 mutex_unlock(&osdc->request_mutex);
2229 complete_request(req);
2230 dout("wait_request tid %llu canceled/timed out\n", req->r_tid);
2234 dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result);
2235 return req->r_result;
2237 EXPORT_SYMBOL(ceph_osdc_wait_request);
2240 * sync - wait for all in-flight requests to flush. avoid starvation.
2242 void ceph_osdc_sync(struct ceph_osd_client *osdc)
2244 struct ceph_osd_request *req;
2245 u64 last_tid, next_tid = 0;
2247 mutex_lock(&osdc->request_mutex);
2248 last_tid = osdc->last_tid;
2250 req = __lookup_request_ge(osdc, next_tid);
2253 if (req->r_tid > last_tid)
2256 next_tid = req->r_tid + 1;
2257 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
2260 ceph_osdc_get_request(req);
2261 mutex_unlock(&osdc->request_mutex);
2262 dout("sync waiting on tid %llu (last is %llu)\n",
2263 req->r_tid, last_tid);
2264 wait_for_completion(&req->r_safe_completion);
2265 mutex_lock(&osdc->request_mutex);
2266 ceph_osdc_put_request(req);
2268 mutex_unlock(&osdc->request_mutex);
2269 dout("sync done (thru tid %llu)\n", last_tid);
2271 EXPORT_SYMBOL(ceph_osdc_sync);
2274 * Call all pending notify callbacks - for use after a watch is
2275 * unregistered, to make sure no more callbacks for it will be invoked
2277 extern void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc)
2279 flush_workqueue(osdc->notify_wq);
2281 EXPORT_SYMBOL(ceph_osdc_flush_notifies);
2287 int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
2292 osdc->client = client;
2293 osdc->osdmap = NULL;
2294 init_rwsem(&osdc->map_sem);
2295 init_completion(&osdc->map_waiters);
2296 osdc->last_requested_map = 0;
2297 mutex_init(&osdc->request_mutex);
2299 osdc->osds = RB_ROOT;
2300 INIT_LIST_HEAD(&osdc->osd_lru);
2301 osdc->requests = RB_ROOT;
2302 INIT_LIST_HEAD(&osdc->req_lru);
2303 INIT_LIST_HEAD(&osdc->req_unsent);
2304 INIT_LIST_HEAD(&osdc->req_notarget);
2305 INIT_LIST_HEAD(&osdc->req_linger);
2306 osdc->num_requests = 0;
2307 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
2308 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
2309 spin_lock_init(&osdc->event_lock);
2310 osdc->event_tree = RB_ROOT;
2311 osdc->event_count = 0;
2313 schedule_delayed_work(&osdc->osds_timeout_work,
2314 round_jiffies_relative(osdc->client->options->osd_idle_ttl * HZ));
2317 osdc->req_mempool = mempool_create_kmalloc_pool(10,
2318 sizeof(struct ceph_osd_request));
2319 if (!osdc->req_mempool)
2322 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
2323 OSD_OP_FRONT_LEN, 10, true,
2327 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
2328 OSD_OPREPLY_FRONT_LEN, 10, true,
2334 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
2335 if (!osdc->notify_wq)
2340 ceph_msgpool_destroy(&osdc->msgpool_op);
2342 mempool_destroy(osdc->req_mempool);
2347 void ceph_osdc_stop(struct ceph_osd_client *osdc)
2349 flush_workqueue(osdc->notify_wq);
2350 destroy_workqueue(osdc->notify_wq);
2351 cancel_delayed_work_sync(&osdc->timeout_work);
2352 cancel_delayed_work_sync(&osdc->osds_timeout_work);
2354 ceph_osdmap_destroy(osdc->osdmap);
2355 osdc->osdmap = NULL;
2357 remove_all_osds(osdc);
2358 mempool_destroy(osdc->req_mempool);
2359 ceph_msgpool_destroy(&osdc->msgpool_op);
2360 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
2364 * Read some contiguous pages. If we cross a stripe boundary, shorten
2365 * *plen. Return number of bytes read, or error.
2367 int ceph_osdc_readpages(struct ceph_osd_client *osdc,
2368 struct ceph_vino vino, struct ceph_file_layout *layout,
2370 u32 truncate_seq, u64 truncate_size,
2371 struct page **pages, int num_pages, int page_align)
2373 struct ceph_osd_request *req;
2376 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
2377 vino.snap, off, *plen);
2378 req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 1,
2379 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
2380 NULL, truncate_seq, truncate_size,
2383 return PTR_ERR(req);
2385 /* it may be a short read due to an object boundary */
2387 osd_req_op_extent_osd_data_pages(req, 0,
2388 pages, *plen, page_align, false, false);
2390 dout("readpages final extent is %llu~%llu (%llu bytes align %d)\n",
2391 off, *plen, *plen, page_align);
2393 ceph_osdc_build_request(req, off, NULL, vino.snap, NULL);
2395 rc = ceph_osdc_start_request(osdc, req, false);
2397 rc = ceph_osdc_wait_request(osdc, req);
2399 ceph_osdc_put_request(req);
2400 dout("readpages result %d\n", rc);
2403 EXPORT_SYMBOL(ceph_osdc_readpages);
2406 * do a synchronous write on N pages
2408 int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
2409 struct ceph_file_layout *layout,
2410 struct ceph_snap_context *snapc,
2412 u32 truncate_seq, u64 truncate_size,
2413 struct timespec *mtime,
2414 struct page **pages, int num_pages)
2416 struct ceph_osd_request *req;
2418 int page_align = off & ~PAGE_MASK;
2420 BUG_ON(vino.snap != CEPH_NOSNAP); /* snapshots aren't writeable */
2421 req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 1,
2423 CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
2424 snapc, truncate_seq, truncate_size,
2427 return PTR_ERR(req);
2429 /* it may be a short write due to an object boundary */
2430 osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align,
2432 dout("writepages %llu~%llu (%llu bytes)\n", off, len, len);
2434 ceph_osdc_build_request(req, off, snapc, CEPH_NOSNAP, mtime);
2436 rc = ceph_osdc_start_request(osdc, req, true);
2438 rc = ceph_osdc_wait_request(osdc, req);
2440 ceph_osdc_put_request(req);
2443 dout("writepages result %d\n", rc);
2446 EXPORT_SYMBOL(ceph_osdc_writepages);
2448 int ceph_osdc_setup(void)
2450 BUG_ON(ceph_osd_request_cache);
2451 ceph_osd_request_cache = kmem_cache_create("ceph_osd_request",
2452 sizeof (struct ceph_osd_request),
2453 __alignof__(struct ceph_osd_request),
2456 return ceph_osd_request_cache ? 0 : -ENOMEM;
2458 EXPORT_SYMBOL(ceph_osdc_setup);
2460 void ceph_osdc_cleanup(void)
2462 BUG_ON(!ceph_osd_request_cache);
2463 kmem_cache_destroy(ceph_osd_request_cache);
2464 ceph_osd_request_cache = NULL;
2466 EXPORT_SYMBOL(ceph_osdc_cleanup);
2469 * handle incoming message
2471 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
2473 struct ceph_osd *osd = con->private;
2474 struct ceph_osd_client *osdc;
2475 int type = le16_to_cpu(msg->hdr.type);
2482 case CEPH_MSG_OSD_MAP:
2483 ceph_osdc_handle_map(osdc, msg);
2485 case CEPH_MSG_OSD_OPREPLY:
2486 handle_reply(osdc, msg, con);
2488 case CEPH_MSG_WATCH_NOTIFY:
2489 handle_watch_notify(osdc, msg);
2493 pr_err("received unknown message type %d %s\n", type,
2494 ceph_msg_type_name(type));
2501 * lookup and return message for incoming reply. set up reply message
2504 static struct ceph_msg *get_reply(struct ceph_connection *con,
2505 struct ceph_msg_header *hdr,
2508 struct ceph_osd *osd = con->private;
2509 struct ceph_osd_client *osdc = osd->o_osdc;
2511 struct ceph_osd_request *req;
2512 int front = le32_to_cpu(hdr->front_len);
2513 int data_len = le32_to_cpu(hdr->data_len);
2516 tid = le64_to_cpu(hdr->tid);
2517 mutex_lock(&osdc->request_mutex);
2518 req = __lookup_request(osdc, tid);
2522 dout("get_reply unknown tid %llu from osd%d\n", tid,
2527 if (req->r_reply->con)
2528 dout("%s revoking msg %p from old con %p\n", __func__,
2529 req->r_reply, req->r_reply->con);
2530 ceph_msg_revoke_incoming(req->r_reply);
2532 if (front > req->r_reply->front.iov_len) {
2533 pr_warning("get_reply front %d > preallocated %d\n",
2534 front, (int)req->r_reply->front.iov_len);
2535 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front, GFP_NOFS, false);
2538 ceph_msg_put(req->r_reply);
2541 m = ceph_msg_get(req->r_reply);
2544 struct ceph_osd_data *osd_data;
2547 * XXX This is assuming there is only one op containing
2548 * XXX page data. Probably OK for reads, but this
2549 * XXX ought to be done more generally.
2551 osd_data = osd_req_op_extent_osd_data(req, 0);
2552 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
2553 if (osd_data->pages &&
2554 unlikely(osd_data->length < data_len)) {
2556 pr_warning("tid %lld reply has %d bytes "
2557 "we had only %llu bytes ready\n",
2558 tid, data_len, osd_data->length);
2567 dout("get_reply tid %lld %p\n", tid, m);
2570 mutex_unlock(&osdc->request_mutex);
2575 static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2576 struct ceph_msg_header *hdr,
2579 struct ceph_osd *osd = con->private;
2580 int type = le16_to_cpu(hdr->type);
2581 int front = le32_to_cpu(hdr->front_len);
2585 case CEPH_MSG_OSD_MAP:
2586 case CEPH_MSG_WATCH_NOTIFY:
2587 return ceph_msg_new(type, front, GFP_NOFS, false);
2588 case CEPH_MSG_OSD_OPREPLY:
2589 return get_reply(con, hdr, skip);
2591 pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
2599 * Wrappers to refcount containing ceph_osd struct
2601 static struct ceph_connection *get_osd_con(struct ceph_connection *con)
2603 struct ceph_osd *osd = con->private;
2609 static void put_osd_con(struct ceph_connection *con)
2611 struct ceph_osd *osd = con->private;
2619 * Note: returned pointer is the address of a structure that's
2620 * managed separately. Caller must *not* attempt to free it.
2622 static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
2623 int *proto, int force_new)
2625 struct ceph_osd *o = con->private;
2626 struct ceph_osd_client *osdc = o->o_osdc;
2627 struct ceph_auth_client *ac = osdc->client->monc.auth;
2628 struct ceph_auth_handshake *auth = &o->o_auth;
2630 if (force_new && auth->authorizer) {
2631 ceph_auth_destroy_authorizer(ac, auth->authorizer);
2632 auth->authorizer = NULL;
2634 if (!auth->authorizer) {
2635 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2638 return ERR_PTR(ret);
2640 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2643 return ERR_PTR(ret);
2645 *proto = ac->protocol;
2651 static int verify_authorizer_reply(struct ceph_connection *con, int len)
2653 struct ceph_osd *o = con->private;
2654 struct ceph_osd_client *osdc = o->o_osdc;
2655 struct ceph_auth_client *ac = osdc->client->monc.auth;
2657 return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer, len);
2660 static int invalidate_authorizer(struct ceph_connection *con)
2662 struct ceph_osd *o = con->private;
2663 struct ceph_osd_client *osdc = o->o_osdc;
2664 struct ceph_auth_client *ac = osdc->client->monc.auth;
2666 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
2667 return ceph_monc_validate_auth(&osdc->client->monc);
2670 static const struct ceph_connection_operations osd_con_ops = {
2673 .dispatch = dispatch,
2674 .get_authorizer = get_authorizer,
2675 .verify_authorizer_reply = verify_authorizer_reply,
2676 .invalidate_authorizer = invalidate_authorizer,
2677 .alloc_msg = alloc_msg,