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("__remove_osd %p\n", osd);
971 BUG_ON(!list_empty(&osd->o_requests));
972 rb_erase(&osd->o_node, &osdc->osds);
973 list_del_init(&osd->o_osd_lru);
974 ceph_con_close(&osd->o_con);
978 static void remove_all_osds(struct ceph_osd_client *osdc)
980 dout("%s %p\n", __func__, osdc);
981 mutex_lock(&osdc->request_mutex);
982 while (!RB_EMPTY_ROOT(&osdc->osds)) {
983 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
984 struct ceph_osd, o_node);
985 __remove_osd(osdc, osd);
987 mutex_unlock(&osdc->request_mutex);
990 static void __move_osd_to_lru(struct ceph_osd_client *osdc,
991 struct ceph_osd *osd)
993 dout("__move_osd_to_lru %p\n", osd);
994 BUG_ON(!list_empty(&osd->o_osd_lru));
995 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
996 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl * HZ;
999 static void __remove_osd_from_lru(struct ceph_osd *osd)
1001 dout("__remove_osd_from_lru %p\n", osd);
1002 if (!list_empty(&osd->o_osd_lru))
1003 list_del_init(&osd->o_osd_lru);
1006 static void remove_old_osds(struct ceph_osd_client *osdc)
1008 struct ceph_osd *osd, *nosd;
1010 dout("__remove_old_osds %p\n", osdc);
1011 mutex_lock(&osdc->request_mutex);
1012 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
1013 if (time_before(jiffies, osd->lru_ttl))
1015 __remove_osd(osdc, osd);
1017 mutex_unlock(&osdc->request_mutex);
1023 static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
1025 struct ceph_entity_addr *peer_addr;
1027 dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
1028 if (list_empty(&osd->o_requests) &&
1029 list_empty(&osd->o_linger_requests)) {
1030 __remove_osd(osdc, osd);
1035 peer_addr = &osdc->osdmap->osd_addr[osd->o_osd];
1036 if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1037 !ceph_con_opened(&osd->o_con)) {
1038 struct ceph_osd_request *req;
1040 dout(" osd addr hasn't changed and connection never opened,"
1041 " letting msgr retry");
1042 /* touch each r_stamp for handle_timeout()'s benfit */
1043 list_for_each_entry(req, &osd->o_requests, r_osd_item)
1044 req->r_stamp = jiffies;
1049 ceph_con_close(&osd->o_con);
1050 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1051 osd->o_incarnation++;
1056 static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new)
1058 struct rb_node **p = &osdc->osds.rb_node;
1059 struct rb_node *parent = NULL;
1060 struct ceph_osd *osd = NULL;
1062 dout("__insert_osd %p osd%d\n", new, new->o_osd);
1065 osd = rb_entry(parent, struct ceph_osd, o_node);
1066 if (new->o_osd < osd->o_osd)
1068 else if (new->o_osd > osd->o_osd)
1069 p = &(*p)->rb_right;
1074 rb_link_node(&new->o_node, parent, p);
1075 rb_insert_color(&new->o_node, &osdc->osds);
1078 static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o)
1080 struct ceph_osd *osd;
1081 struct rb_node *n = osdc->osds.rb_node;
1084 osd = rb_entry(n, struct ceph_osd, o_node);
1087 else if (o > osd->o_osd)
1095 static void __schedule_osd_timeout(struct ceph_osd_client *osdc)
1097 schedule_delayed_work(&osdc->timeout_work,
1098 osdc->client->options->osd_keepalive_timeout * HZ);
1101 static void __cancel_osd_timeout(struct ceph_osd_client *osdc)
1103 cancel_delayed_work(&osdc->timeout_work);
1107 * Register request, assign tid. If this is the first request, set up
1108 * the timeout event.
1110 static void __register_request(struct ceph_osd_client *osdc,
1111 struct ceph_osd_request *req)
1113 req->r_tid = ++osdc->last_tid;
1114 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
1115 dout("__register_request %p tid %lld\n", req, req->r_tid);
1116 __insert_request(osdc, req);
1117 ceph_osdc_get_request(req);
1118 osdc->num_requests++;
1119 if (osdc->num_requests == 1) {
1120 dout(" first request, scheduling timeout\n");
1121 __schedule_osd_timeout(osdc);
1126 * called under osdc->request_mutex
1128 static void __unregister_request(struct ceph_osd_client *osdc,
1129 struct ceph_osd_request *req)
1131 if (RB_EMPTY_NODE(&req->r_node)) {
1132 dout("__unregister_request %p tid %lld not registered\n",
1137 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
1138 rb_erase(&req->r_node, &osdc->requests);
1139 osdc->num_requests--;
1142 /* make sure the original request isn't in flight. */
1143 ceph_msg_revoke(req->r_request);
1145 list_del_init(&req->r_osd_item);
1146 if (list_empty(&req->r_osd->o_requests) &&
1147 list_empty(&req->r_osd->o_linger_requests)) {
1148 dout("moving osd to %p lru\n", req->r_osd);
1149 __move_osd_to_lru(osdc, req->r_osd);
1151 if (list_empty(&req->r_linger_item))
1155 list_del_init(&req->r_req_lru_item);
1156 ceph_osdc_put_request(req);
1158 if (osdc->num_requests == 0) {
1159 dout(" no requests, canceling timeout\n");
1160 __cancel_osd_timeout(osdc);
1165 * Cancel a previously queued request message
1167 static void __cancel_request(struct ceph_osd_request *req)
1169 if (req->r_sent && req->r_osd) {
1170 ceph_msg_revoke(req->r_request);
1175 static void __register_linger_request(struct ceph_osd_client *osdc,
1176 struct ceph_osd_request *req)
1178 dout("__register_linger_request %p\n", req);
1179 ceph_osdc_get_request(req);
1180 list_add_tail(&req->r_linger_item, &osdc->req_linger);
1182 list_add_tail(&req->r_linger_osd,
1183 &req->r_osd->o_linger_requests);
1186 static void __unregister_linger_request(struct ceph_osd_client *osdc,
1187 struct ceph_osd_request *req)
1189 dout("__unregister_linger_request %p\n", req);
1190 list_del_init(&req->r_linger_item);
1192 list_del_init(&req->r_linger_osd);
1194 if (list_empty(&req->r_osd->o_requests) &&
1195 list_empty(&req->r_osd->o_linger_requests)) {
1196 dout("moving osd to %p lru\n", req->r_osd);
1197 __move_osd_to_lru(osdc, req->r_osd);
1199 if (list_empty(&req->r_osd_item))
1202 ceph_osdc_put_request(req);
1205 void ceph_osdc_unregister_linger_request(struct ceph_osd_client *osdc,
1206 struct ceph_osd_request *req)
1208 mutex_lock(&osdc->request_mutex);
1209 if (req->r_linger) {
1211 __unregister_linger_request(osdc, req);
1213 mutex_unlock(&osdc->request_mutex);
1215 EXPORT_SYMBOL(ceph_osdc_unregister_linger_request);
1217 void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
1218 struct ceph_osd_request *req)
1220 if (!req->r_linger) {
1221 dout("set_request_linger %p\n", req);
1225 EXPORT_SYMBOL(ceph_osdc_set_request_linger);
1228 * Returns whether a request should be blocked from being sent
1229 * based on the current osdmap and osd_client settings.
1231 * Caller should hold map_sem for read.
1233 static bool __req_should_be_paused(struct ceph_osd_client *osdc,
1234 struct ceph_osd_request *req)
1236 bool pauserd = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD);
1237 bool pausewr = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR) ||
1238 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
1239 return (req->r_flags & CEPH_OSD_FLAG_READ && pauserd) ||
1240 (req->r_flags & CEPH_OSD_FLAG_WRITE && pausewr);
1244 * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
1245 * (as needed), and set the request r_osd appropriately. If there is
1246 * no up osd, set r_osd to NULL. Move the request to the appropriate list
1247 * (unsent, homeless) or leave on in-flight lru.
1249 * Return 0 if unchanged, 1 if changed, or negative on error.
1251 * Caller should hold map_sem for read and request_mutex.
1253 static int __map_request(struct ceph_osd_client *osdc,
1254 struct ceph_osd_request *req, int force_resend)
1256 struct ceph_pg pgid;
1257 int acting[CEPH_PG_MAX_SIZE];
1258 int o = -1, num = 0;
1262 dout("map_request %p tid %lld\n", req, req->r_tid);
1263 err = ceph_calc_ceph_pg(&pgid, req->r_oid, osdc->osdmap,
1264 ceph_file_layout_pg_pool(req->r_file_layout));
1266 list_move(&req->r_req_lru_item, &osdc->req_notarget);
1271 err = ceph_calc_pg_acting(osdc->osdmap, pgid, acting);
1277 was_paused = req->r_paused;
1278 req->r_paused = __req_should_be_paused(osdc, req);
1279 if (was_paused && !req->r_paused)
1282 if ((!force_resend &&
1283 req->r_osd && req->r_osd->o_osd == o &&
1284 req->r_sent >= req->r_osd->o_incarnation &&
1285 req->r_num_pg_osds == num &&
1286 memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
1287 (req->r_osd == NULL && o == -1) ||
1289 return 0; /* no change */
1291 dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n",
1292 req->r_tid, pgid.pool, pgid.seed, o,
1293 req->r_osd ? req->r_osd->o_osd : -1);
1295 /* record full pg acting set */
1296 memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
1297 req->r_num_pg_osds = num;
1300 __cancel_request(req);
1301 list_del_init(&req->r_osd_item);
1305 req->r_osd = __lookup_osd(osdc, o);
1306 if (!req->r_osd && o >= 0) {
1308 req->r_osd = create_osd(osdc, o);
1310 list_move(&req->r_req_lru_item, &osdc->req_notarget);
1314 dout("map_request osd %p is osd%d\n", req->r_osd, o);
1315 __insert_osd(osdc, req->r_osd);
1317 ceph_con_open(&req->r_osd->o_con,
1318 CEPH_ENTITY_TYPE_OSD, o,
1319 &osdc->osdmap->osd_addr[o]);
1323 __remove_osd_from_lru(req->r_osd);
1324 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
1325 list_move_tail(&req->r_req_lru_item, &osdc->req_unsent);
1327 list_move_tail(&req->r_req_lru_item, &osdc->req_notarget);
1329 err = 1; /* osd or pg changed */
1336 * caller should hold map_sem (for read) and request_mutex
1338 static void __send_request(struct ceph_osd_client *osdc,
1339 struct ceph_osd_request *req)
1343 dout("send_request %p tid %llu to osd%d flags %d pg %lld.%x\n",
1344 req, req->r_tid, req->r_osd->o_osd, req->r_flags,
1345 (unsigned long long)req->r_pgid.pool, req->r_pgid.seed);
1347 /* fill in message content that changes each time we send it */
1348 put_unaligned_le32(osdc->osdmap->epoch, req->r_request_osdmap_epoch);
1349 put_unaligned_le32(req->r_flags, req->r_request_flags);
1350 put_unaligned_le64(req->r_pgid.pool, req->r_request_pool);
1351 p = req->r_request_pgid;
1352 ceph_encode_64(&p, req->r_pgid.pool);
1353 ceph_encode_32(&p, req->r_pgid.seed);
1354 put_unaligned_le64(1, req->r_request_attempts); /* FIXME */
1355 memcpy(req->r_request_reassert_version, &req->r_reassert_version,
1356 sizeof(req->r_reassert_version));
1358 req->r_stamp = jiffies;
1359 list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
1361 ceph_msg_get(req->r_request); /* send consumes a ref */
1363 req->r_sent = req->r_osd->o_incarnation;
1365 ceph_con_send(&req->r_osd->o_con, req->r_request);
1369 * Send any requests in the queue (req_unsent).
1371 static void __send_queued(struct ceph_osd_client *osdc)
1373 struct ceph_osd_request *req, *tmp;
1375 dout("__send_queued\n");
1376 list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item)
1377 __send_request(osdc, req);
1381 * Timeout callback, called every N seconds when 1 or more osd
1382 * requests has been active for more than N seconds. When this
1383 * happens, we ping all OSDs with requests who have timed out to
1384 * ensure any communications channel reset is detected. Reset the
1385 * request timeouts another N seconds in the future as we go.
1386 * Reschedule the timeout event another N seconds in future (unless
1387 * there are no open requests).
1389 static void handle_timeout(struct work_struct *work)
1391 struct ceph_osd_client *osdc =
1392 container_of(work, struct ceph_osd_client, timeout_work.work);
1393 struct ceph_osd_request *req;
1394 struct ceph_osd *osd;
1395 unsigned long keepalive =
1396 osdc->client->options->osd_keepalive_timeout * HZ;
1397 struct list_head slow_osds;
1399 down_read(&osdc->map_sem);
1401 ceph_monc_request_next_osdmap(&osdc->client->monc);
1403 mutex_lock(&osdc->request_mutex);
1406 * ping osds that are a bit slow. this ensures that if there
1407 * is a break in the TCP connection we will notice, and reopen
1408 * a connection with that osd (from the fault callback).
1410 INIT_LIST_HEAD(&slow_osds);
1411 list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
1412 if (time_before(jiffies, req->r_stamp + keepalive))
1417 dout(" tid %llu is slow, will send keepalive on osd%d\n",
1418 req->r_tid, osd->o_osd);
1419 list_move_tail(&osd->o_keepalive_item, &slow_osds);
1421 while (!list_empty(&slow_osds)) {
1422 osd = list_entry(slow_osds.next, struct ceph_osd,
1424 list_del_init(&osd->o_keepalive_item);
1425 ceph_con_keepalive(&osd->o_con);
1428 __schedule_osd_timeout(osdc);
1429 __send_queued(osdc);
1430 mutex_unlock(&osdc->request_mutex);
1431 up_read(&osdc->map_sem);
1434 static void handle_osds_timeout(struct work_struct *work)
1436 struct ceph_osd_client *osdc =
1437 container_of(work, struct ceph_osd_client,
1438 osds_timeout_work.work);
1439 unsigned long delay =
1440 osdc->client->options->osd_idle_ttl * HZ >> 2;
1442 dout("osds timeout\n");
1443 down_read(&osdc->map_sem);
1444 remove_old_osds(osdc);
1445 up_read(&osdc->map_sem);
1447 schedule_delayed_work(&osdc->osds_timeout_work,
1448 round_jiffies_relative(delay));
1451 static void complete_request(struct ceph_osd_request *req)
1453 complete_all(&req->r_safe_completion); /* fsync waiter */
1457 * handle osd op reply. either call the callback if it is specified,
1458 * or do the completion to wake up the waiting thread.
1460 static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg,
1461 struct ceph_connection *con)
1464 struct ceph_osd_request *req;
1467 unsigned int numops;
1468 int payload_len, flags;
1474 u64 reassert_version;
1476 int already_completed;
1480 tid = le64_to_cpu(msg->hdr.tid);
1481 dout("handle_reply %p tid %llu\n", msg, tid);
1483 p = msg->front.iov_base;
1484 end = p + msg->front.iov_len;
1486 ceph_decode_need(&p, end, 4, bad);
1487 object_len = ceph_decode_32(&p);
1488 ceph_decode_need(&p, end, object_len, bad);
1491 err = ceph_decode_pgid(&p, end, &pg);
1495 ceph_decode_need(&p, end, 8 + 4 + 4 + 8 + 4, bad);
1496 flags = ceph_decode_64(&p);
1497 result = ceph_decode_32(&p);
1498 reassert_epoch = ceph_decode_32(&p);
1499 reassert_version = ceph_decode_64(&p);
1500 osdmap_epoch = ceph_decode_32(&p);
1503 mutex_lock(&osdc->request_mutex);
1504 req = __lookup_request(osdc, tid);
1506 dout("handle_reply tid %llu dne\n", tid);
1509 ceph_osdc_get_request(req);
1511 dout("handle_reply %p tid %llu req %p result %d\n", msg, tid,
1514 ceph_decode_need(&p, end, 4, bad_put);
1515 numops = ceph_decode_32(&p);
1516 if (numops > CEPH_OSD_MAX_OP)
1518 if (numops != req->r_num_ops)
1521 ceph_decode_need(&p, end, numops * sizeof(struct ceph_osd_op), bad_put);
1522 for (i = 0; i < numops; i++) {
1523 struct ceph_osd_op *op = p;
1526 len = le32_to_cpu(op->payload_len);
1527 req->r_reply_op_len[i] = len;
1528 dout(" op %d has %d bytes\n", i, len);
1532 bytes = le32_to_cpu(msg->hdr.data_len);
1533 if (payload_len != bytes) {
1534 pr_warning("sum of op payload lens %d != data_len %d",
1535 payload_len, bytes);
1539 ceph_decode_need(&p, end, 4 + numops * 4, bad_put);
1540 retry_attempt = ceph_decode_32(&p);
1541 for (i = 0; i < numops; i++)
1542 req->r_reply_op_result[i] = ceph_decode_32(&p);
1544 already_completed = req->r_got_reply;
1546 if (!req->r_got_reply) {
1548 req->r_result = result;
1549 dout("handle_reply result %d bytes %d\n", req->r_result,
1551 if (req->r_result == 0)
1552 req->r_result = bytes;
1554 /* in case this is a write and we need to replay, */
1555 req->r_reassert_version.epoch = cpu_to_le32(reassert_epoch);
1556 req->r_reassert_version.version = cpu_to_le64(reassert_version);
1558 req->r_got_reply = 1;
1559 } else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) {
1560 dout("handle_reply tid %llu dup ack\n", tid);
1561 mutex_unlock(&osdc->request_mutex);
1565 dout("handle_reply tid %llu flags %d\n", tid, flags);
1567 if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK))
1568 __register_linger_request(osdc, req);
1570 /* either this is a read, or we got the safe response */
1572 (flags & CEPH_OSD_FLAG_ONDISK) ||
1573 ((flags & CEPH_OSD_FLAG_WRITE) == 0))
1574 __unregister_request(osdc, req);
1576 mutex_unlock(&osdc->request_mutex);
1578 if (!already_completed) {
1579 if (req->r_unsafe_callback &&
1580 result >= 0 && !(flags & CEPH_OSD_FLAG_ONDISK))
1581 req->r_unsafe_callback(req, true);
1582 if (req->r_callback)
1583 req->r_callback(req, msg);
1585 complete_all(&req->r_completion);
1588 if (flags & CEPH_OSD_FLAG_ONDISK) {
1589 if (req->r_unsafe_callback && already_completed)
1590 req->r_unsafe_callback(req, false);
1591 complete_request(req);
1595 dout("req=%p req->r_linger=%d\n", req, req->r_linger);
1596 ceph_osdc_put_request(req);
1600 ceph_osdc_put_request(req);
1602 mutex_unlock(&osdc->request_mutex);
1604 pr_err("corrupt osd_op_reply got %d %d\n",
1605 (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len));
1609 static void reset_changed_osds(struct ceph_osd_client *osdc)
1611 struct rb_node *p, *n;
1613 for (p = rb_first(&osdc->osds); p; p = n) {
1614 struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
1617 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
1618 memcmp(&osd->o_con.peer_addr,
1619 ceph_osd_addr(osdc->osdmap,
1621 sizeof(struct ceph_entity_addr)) != 0)
1622 __reset_osd(osdc, osd);
1627 * Requeue requests whose mapping to an OSD has changed. If requests map to
1628 * no osd, request a new map.
1630 * Caller should hold map_sem for read.
1632 static void kick_requests(struct ceph_osd_client *osdc, bool force_resend,
1633 bool force_resend_writes)
1635 struct ceph_osd_request *req, *nreq;
1639 bool force_resend_req;
1641 dout("kick_requests %s %s\n", force_resend ? " (force resend)" : "",
1642 force_resend_writes ? " (force resend writes)" : "");
1643 mutex_lock(&osdc->request_mutex);
1644 for (p = rb_first(&osdc->requests); p; ) {
1645 req = rb_entry(p, struct ceph_osd_request, r_node);
1649 * For linger requests that have not yet been
1650 * registered, move them to the linger list; they'll
1651 * be sent to the osd in the loop below. Unregister
1652 * the request before re-registering it as a linger
1653 * request to ensure the __map_request() below
1654 * will decide it needs to be sent.
1656 if (req->r_linger && list_empty(&req->r_linger_item)) {
1657 dout("%p tid %llu restart on osd%d\n",
1659 req->r_osd ? req->r_osd->o_osd : -1);
1660 ceph_osdc_get_request(req);
1661 __unregister_request(osdc, req);
1662 __register_linger_request(osdc, req);
1663 ceph_osdc_put_request(req);
1667 force_resend_req = force_resend ||
1668 (force_resend_writes &&
1669 req->r_flags & CEPH_OSD_FLAG_WRITE);
1670 err = __map_request(osdc, req, force_resend_req);
1672 continue; /* error */
1673 if (req->r_osd == NULL) {
1674 dout("%p tid %llu maps to no osd\n", req, req->r_tid);
1675 needmap++; /* request a newer map */
1676 } else if (err > 0) {
1677 if (!req->r_linger) {
1678 dout("%p tid %llu requeued on osd%d\n", req,
1680 req->r_osd ? req->r_osd->o_osd : -1);
1681 req->r_flags |= CEPH_OSD_FLAG_RETRY;
1686 list_for_each_entry_safe(req, nreq, &osdc->req_linger,
1688 dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
1690 err = __map_request(osdc, req,
1691 force_resend || force_resend_writes);
1692 dout("__map_request returned %d\n", err);
1694 continue; /* no change and no osd was specified */
1696 continue; /* hrm! */
1697 if (req->r_osd == NULL) {
1698 dout("tid %llu maps to no valid osd\n", req->r_tid);
1699 needmap++; /* request a newer map */
1703 dout("kicking lingering %p tid %llu osd%d\n", req, req->r_tid,
1704 req->r_osd ? req->r_osd->o_osd : -1);
1705 __register_request(osdc, req);
1706 __unregister_linger_request(osdc, req);
1708 reset_changed_osds(osdc);
1709 mutex_unlock(&osdc->request_mutex);
1712 dout("%d requests for down osds, need new map\n", needmap);
1713 ceph_monc_request_next_osdmap(&osdc->client->monc);
1719 * Process updated osd map.
1721 * The message contains any number of incremental and full maps, normally
1722 * indicating some sort of topology change in the cluster. Kick requests
1723 * off to different OSDs as needed.
1725 void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1727 void *p, *end, *next;
1728 u32 nr_maps, maplen;
1730 struct ceph_osdmap *newmap = NULL, *oldmap;
1732 struct ceph_fsid fsid;
1735 dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
1736 p = msg->front.iov_base;
1737 end = p + msg->front.iov_len;
1740 ceph_decode_need(&p, end, sizeof(fsid), bad);
1741 ceph_decode_copy(&p, &fsid, sizeof(fsid));
1742 if (ceph_check_fsid(osdc->client, &fsid) < 0)
1745 down_write(&osdc->map_sem);
1747 was_full = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
1749 /* incremental maps */
1750 ceph_decode_32_safe(&p, end, nr_maps, bad);
1751 dout(" %d inc maps\n", nr_maps);
1752 while (nr_maps > 0) {
1753 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
1754 epoch = ceph_decode_32(&p);
1755 maplen = ceph_decode_32(&p);
1756 ceph_decode_need(&p, end, maplen, bad);
1758 if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
1759 dout("applying incremental map %u len %d\n",
1761 newmap = osdmap_apply_incremental(&p, next,
1763 &osdc->client->msgr);
1764 if (IS_ERR(newmap)) {
1765 err = PTR_ERR(newmap);
1769 if (newmap != osdc->osdmap) {
1770 ceph_osdmap_destroy(osdc->osdmap);
1771 osdc->osdmap = newmap;
1773 was_full = was_full ||
1774 ceph_osdmap_flag(osdc->osdmap,
1776 kick_requests(osdc, 0, was_full);
1778 dout("ignoring incremental map %u len %d\n",
1788 ceph_decode_32_safe(&p, end, nr_maps, bad);
1789 dout(" %d full maps\n", nr_maps);
1791 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
1792 epoch = ceph_decode_32(&p);
1793 maplen = ceph_decode_32(&p);
1794 ceph_decode_need(&p, end, maplen, bad);
1796 dout("skipping non-latest full map %u len %d\n",
1798 } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
1799 dout("skipping full map %u len %d, "
1800 "older than our %u\n", epoch, maplen,
1801 osdc->osdmap->epoch);
1803 int skipped_map = 0;
1805 dout("taking full map %u len %d\n", epoch, maplen);
1806 newmap = osdmap_decode(&p, p+maplen);
1807 if (IS_ERR(newmap)) {
1808 err = PTR_ERR(newmap);
1812 oldmap = osdc->osdmap;
1813 osdc->osdmap = newmap;
1815 if (oldmap->epoch + 1 < newmap->epoch)
1817 ceph_osdmap_destroy(oldmap);
1819 was_full = was_full ||
1820 ceph_osdmap_flag(osdc->osdmap,
1822 kick_requests(osdc, skipped_map, was_full);
1831 downgrade_write(&osdc->map_sem);
1832 ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch);
1835 * subscribe to subsequent osdmap updates if full to ensure
1836 * we find out when we are no longer full and stop returning
1839 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) ||
1840 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD) ||
1841 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR))
1842 ceph_monc_request_next_osdmap(&osdc->client->monc);
1844 mutex_lock(&osdc->request_mutex);
1845 __send_queued(osdc);
1846 mutex_unlock(&osdc->request_mutex);
1847 up_read(&osdc->map_sem);
1848 wake_up_all(&osdc->client->auth_wq);
1852 pr_err("osdc handle_map corrupt msg\n");
1854 up_write(&osdc->map_sem);
1859 * watch/notify callback event infrastructure
1861 * These callbacks are used both for watch and notify operations.
1863 static void __release_event(struct kref *kref)
1865 struct ceph_osd_event *event =
1866 container_of(kref, struct ceph_osd_event, kref);
1868 dout("__release_event %p\n", event);
1872 static void get_event(struct ceph_osd_event *event)
1874 kref_get(&event->kref);
1877 void ceph_osdc_put_event(struct ceph_osd_event *event)
1879 kref_put(&event->kref, __release_event);
1881 EXPORT_SYMBOL(ceph_osdc_put_event);
1883 static void __insert_event(struct ceph_osd_client *osdc,
1884 struct ceph_osd_event *new)
1886 struct rb_node **p = &osdc->event_tree.rb_node;
1887 struct rb_node *parent = NULL;
1888 struct ceph_osd_event *event = NULL;
1892 event = rb_entry(parent, struct ceph_osd_event, node);
1893 if (new->cookie < event->cookie)
1895 else if (new->cookie > event->cookie)
1896 p = &(*p)->rb_right;
1901 rb_link_node(&new->node, parent, p);
1902 rb_insert_color(&new->node, &osdc->event_tree);
1905 static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
1908 struct rb_node **p = &osdc->event_tree.rb_node;
1909 struct rb_node *parent = NULL;
1910 struct ceph_osd_event *event = NULL;
1914 event = rb_entry(parent, struct ceph_osd_event, node);
1915 if (cookie < event->cookie)
1917 else if (cookie > event->cookie)
1918 p = &(*p)->rb_right;
1925 static void __remove_event(struct ceph_osd_event *event)
1927 struct ceph_osd_client *osdc = event->osdc;
1929 if (!RB_EMPTY_NODE(&event->node)) {
1930 dout("__remove_event removed %p\n", event);
1931 rb_erase(&event->node, &osdc->event_tree);
1932 ceph_osdc_put_event(event);
1934 dout("__remove_event didn't remove %p\n", event);
1938 int ceph_osdc_create_event(struct ceph_osd_client *osdc,
1939 void (*event_cb)(u64, u64, u8, void *),
1940 void *data, struct ceph_osd_event **pevent)
1942 struct ceph_osd_event *event;
1944 event = kmalloc(sizeof(*event), GFP_NOIO);
1948 dout("create_event %p\n", event);
1949 event->cb = event_cb;
1950 event->one_shot = 0;
1953 INIT_LIST_HEAD(&event->osd_node);
1954 RB_CLEAR_NODE(&event->node);
1955 kref_init(&event->kref); /* one ref for us */
1956 kref_get(&event->kref); /* one ref for the caller */
1958 spin_lock(&osdc->event_lock);
1959 event->cookie = ++osdc->event_count;
1960 __insert_event(osdc, event);
1961 spin_unlock(&osdc->event_lock);
1966 EXPORT_SYMBOL(ceph_osdc_create_event);
1968 void ceph_osdc_cancel_event(struct ceph_osd_event *event)
1970 struct ceph_osd_client *osdc = event->osdc;
1972 dout("cancel_event %p\n", event);
1973 spin_lock(&osdc->event_lock);
1974 __remove_event(event);
1975 spin_unlock(&osdc->event_lock);
1976 ceph_osdc_put_event(event); /* caller's */
1978 EXPORT_SYMBOL(ceph_osdc_cancel_event);
1981 static void do_event_work(struct work_struct *work)
1983 struct ceph_osd_event_work *event_work =
1984 container_of(work, struct ceph_osd_event_work, work);
1985 struct ceph_osd_event *event = event_work->event;
1986 u64 ver = event_work->ver;
1987 u64 notify_id = event_work->notify_id;
1988 u8 opcode = event_work->opcode;
1990 dout("do_event_work completing %p\n", event);
1991 event->cb(ver, notify_id, opcode, event->data);
1992 dout("do_event_work completed %p\n", event);
1993 ceph_osdc_put_event(event);
1999 * Process osd watch notifications
2001 static void handle_watch_notify(struct ceph_osd_client *osdc,
2002 struct ceph_msg *msg)
2006 u64 cookie, ver, notify_id;
2008 struct ceph_osd_event *event;
2009 struct ceph_osd_event_work *event_work;
2011 p = msg->front.iov_base;
2012 end = p + msg->front.iov_len;
2014 ceph_decode_8_safe(&p, end, proto_ver, bad);
2015 ceph_decode_8_safe(&p, end, opcode, bad);
2016 ceph_decode_64_safe(&p, end, cookie, bad);
2017 ceph_decode_64_safe(&p, end, ver, bad);
2018 ceph_decode_64_safe(&p, end, notify_id, bad);
2020 spin_lock(&osdc->event_lock);
2021 event = __find_event(osdc, cookie);
2023 BUG_ON(event->one_shot);
2026 spin_unlock(&osdc->event_lock);
2027 dout("handle_watch_notify cookie %lld ver %lld event %p\n",
2028 cookie, ver, event);
2030 event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
2032 dout("ERROR: could not allocate event_work\n");
2035 INIT_WORK(&event_work->work, do_event_work);
2036 event_work->event = event;
2037 event_work->ver = ver;
2038 event_work->notify_id = notify_id;
2039 event_work->opcode = opcode;
2040 if (!queue_work(osdc->notify_wq, &event_work->work)) {
2041 dout("WARNING: failed to queue notify event work\n");
2049 ceph_osdc_put_event(event);
2053 pr_err("osdc handle_watch_notify corrupt msg\n");
2058 * build new request AND message
2061 void ceph_osdc_build_request(struct ceph_osd_request *req, u64 off,
2062 struct ceph_snap_context *snapc, u64 snap_id,
2063 struct timespec *mtime)
2065 struct ceph_msg *msg = req->r_request;
2068 int flags = req->r_flags;
2072 req->r_snapid = snap_id;
2073 req->r_snapc = ceph_get_snap_context(snapc);
2075 /* encode request */
2076 msg->hdr.version = cpu_to_le16(4);
2078 p = msg->front.iov_base;
2079 ceph_encode_32(&p, 1); /* client_inc is always 1 */
2080 req->r_request_osdmap_epoch = p;
2082 req->r_request_flags = p;
2084 if (req->r_flags & CEPH_OSD_FLAG_WRITE)
2085 ceph_encode_timespec(p, mtime);
2086 p += sizeof(struct ceph_timespec);
2087 req->r_request_reassert_version = p;
2088 p += sizeof(struct ceph_eversion); /* will get filled in */
2091 ceph_encode_8(&p, 4);
2092 ceph_encode_8(&p, 4);
2093 ceph_encode_32(&p, 8 + 4 + 4);
2094 req->r_request_pool = p;
2096 ceph_encode_32(&p, -1); /* preferred */
2097 ceph_encode_32(&p, 0); /* key len */
2099 ceph_encode_8(&p, 1);
2100 req->r_request_pgid = p;
2102 ceph_encode_32(&p, -1); /* preferred */
2105 ceph_encode_32(&p, req->r_oid_len);
2106 memcpy(p, req->r_oid, req->r_oid_len);
2107 dout("oid '%.*s' len %d\n", req->r_oid_len, req->r_oid, req->r_oid_len);
2108 p += req->r_oid_len;
2110 /* ops--can imply data */
2111 ceph_encode_16(&p, (u16)req->r_num_ops);
2113 for (i = 0; i < req->r_num_ops; i++) {
2114 data_len += osd_req_encode_op(req, p, i);
2115 p += sizeof(struct ceph_osd_op);
2119 ceph_encode_64(&p, req->r_snapid);
2120 ceph_encode_64(&p, req->r_snapc ? req->r_snapc->seq : 0);
2121 ceph_encode_32(&p, req->r_snapc ? req->r_snapc->num_snaps : 0);
2123 for (i = 0; i < snapc->num_snaps; i++) {
2124 ceph_encode_64(&p, req->r_snapc->snaps[i]);
2128 req->r_request_attempts = p;
2132 if (flags & CEPH_OSD_FLAG_WRITE) {
2136 * The header "data_off" is a hint to the receiver
2137 * allowing it to align received data into its
2138 * buffers such that there's no need to re-copy
2139 * it before writing it to disk (direct I/O).
2141 data_off = (u16) (off & 0xffff);
2142 req->r_request->hdr.data_off = cpu_to_le16(data_off);
2144 req->r_request->hdr.data_len = cpu_to_le32(data_len);
2146 BUG_ON(p > msg->front.iov_base + msg->front.iov_len);
2147 msg_size = p - msg->front.iov_base;
2148 msg->front.iov_len = msg_size;
2149 msg->hdr.front_len = cpu_to_le32(msg_size);
2151 dout("build_request msg_size was %d\n", (int)msg_size);
2153 EXPORT_SYMBOL(ceph_osdc_build_request);
2156 * Register request, send initial attempt.
2158 int ceph_osdc_start_request(struct ceph_osd_client *osdc,
2159 struct ceph_osd_request *req,
2164 down_read(&osdc->map_sem);
2165 mutex_lock(&osdc->request_mutex);
2166 __register_request(osdc, req);
2168 req->r_got_reply = 0;
2169 rc = __map_request(osdc, req, 0);
2172 dout("osdc_start_request failed map, "
2173 " will retry %lld\n", req->r_tid);
2176 __unregister_request(osdc, req);
2180 if (req->r_osd == NULL) {
2181 dout("send_request %p no up osds in pg\n", req);
2182 ceph_monc_request_next_osdmap(&osdc->client->monc);
2184 __send_queued(osdc);
2188 mutex_unlock(&osdc->request_mutex);
2189 up_read(&osdc->map_sem);
2192 EXPORT_SYMBOL(ceph_osdc_start_request);
2195 * wait for a request to complete
2197 int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
2198 struct ceph_osd_request *req)
2202 rc = wait_for_completion_interruptible(&req->r_completion);
2204 mutex_lock(&osdc->request_mutex);
2205 __cancel_request(req);
2206 __unregister_request(osdc, req);
2207 mutex_unlock(&osdc->request_mutex);
2208 complete_request(req);
2209 dout("wait_request tid %llu canceled/timed out\n", req->r_tid);
2213 dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result);
2214 return req->r_result;
2216 EXPORT_SYMBOL(ceph_osdc_wait_request);
2219 * sync - wait for all in-flight requests to flush. avoid starvation.
2221 void ceph_osdc_sync(struct ceph_osd_client *osdc)
2223 struct ceph_osd_request *req;
2224 u64 last_tid, next_tid = 0;
2226 mutex_lock(&osdc->request_mutex);
2227 last_tid = osdc->last_tid;
2229 req = __lookup_request_ge(osdc, next_tid);
2232 if (req->r_tid > last_tid)
2235 next_tid = req->r_tid + 1;
2236 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
2239 ceph_osdc_get_request(req);
2240 mutex_unlock(&osdc->request_mutex);
2241 dout("sync waiting on tid %llu (last is %llu)\n",
2242 req->r_tid, last_tid);
2243 wait_for_completion(&req->r_safe_completion);
2244 mutex_lock(&osdc->request_mutex);
2245 ceph_osdc_put_request(req);
2247 mutex_unlock(&osdc->request_mutex);
2248 dout("sync done (thru tid %llu)\n", last_tid);
2250 EXPORT_SYMBOL(ceph_osdc_sync);
2253 * Call all pending notify callbacks - for use after a watch is
2254 * unregistered, to make sure no more callbacks for it will be invoked
2256 extern void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc)
2258 flush_workqueue(osdc->notify_wq);
2260 EXPORT_SYMBOL(ceph_osdc_flush_notifies);
2266 int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
2271 osdc->client = client;
2272 osdc->osdmap = NULL;
2273 init_rwsem(&osdc->map_sem);
2274 init_completion(&osdc->map_waiters);
2275 osdc->last_requested_map = 0;
2276 mutex_init(&osdc->request_mutex);
2278 osdc->osds = RB_ROOT;
2279 INIT_LIST_HEAD(&osdc->osd_lru);
2280 osdc->requests = RB_ROOT;
2281 INIT_LIST_HEAD(&osdc->req_lru);
2282 INIT_LIST_HEAD(&osdc->req_unsent);
2283 INIT_LIST_HEAD(&osdc->req_notarget);
2284 INIT_LIST_HEAD(&osdc->req_linger);
2285 osdc->num_requests = 0;
2286 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
2287 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
2288 spin_lock_init(&osdc->event_lock);
2289 osdc->event_tree = RB_ROOT;
2290 osdc->event_count = 0;
2292 schedule_delayed_work(&osdc->osds_timeout_work,
2293 round_jiffies_relative(osdc->client->options->osd_idle_ttl * HZ));
2296 osdc->req_mempool = mempool_create_kmalloc_pool(10,
2297 sizeof(struct ceph_osd_request));
2298 if (!osdc->req_mempool)
2301 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
2302 OSD_OP_FRONT_LEN, 10, true,
2306 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
2307 OSD_OPREPLY_FRONT_LEN, 10, true,
2313 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
2314 if (!osdc->notify_wq)
2319 ceph_msgpool_destroy(&osdc->msgpool_op);
2321 mempool_destroy(osdc->req_mempool);
2326 void ceph_osdc_stop(struct ceph_osd_client *osdc)
2328 flush_workqueue(osdc->notify_wq);
2329 destroy_workqueue(osdc->notify_wq);
2330 cancel_delayed_work_sync(&osdc->timeout_work);
2331 cancel_delayed_work_sync(&osdc->osds_timeout_work);
2333 ceph_osdmap_destroy(osdc->osdmap);
2334 osdc->osdmap = NULL;
2336 remove_all_osds(osdc);
2337 mempool_destroy(osdc->req_mempool);
2338 ceph_msgpool_destroy(&osdc->msgpool_op);
2339 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
2343 * Read some contiguous pages. If we cross a stripe boundary, shorten
2344 * *plen. Return number of bytes read, or error.
2346 int ceph_osdc_readpages(struct ceph_osd_client *osdc,
2347 struct ceph_vino vino, struct ceph_file_layout *layout,
2349 u32 truncate_seq, u64 truncate_size,
2350 struct page **pages, int num_pages, int page_align)
2352 struct ceph_osd_request *req;
2355 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
2356 vino.snap, off, *plen);
2357 req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 1,
2358 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
2359 NULL, truncate_seq, truncate_size,
2362 return PTR_ERR(req);
2364 /* it may be a short read due to an object boundary */
2366 osd_req_op_extent_osd_data_pages(req, 0,
2367 pages, *plen, page_align, false, false);
2369 dout("readpages final extent is %llu~%llu (%llu bytes align %d)\n",
2370 off, *plen, *plen, page_align);
2372 ceph_osdc_build_request(req, off, NULL, vino.snap, NULL);
2374 rc = ceph_osdc_start_request(osdc, req, false);
2376 rc = ceph_osdc_wait_request(osdc, req);
2378 ceph_osdc_put_request(req);
2379 dout("readpages result %d\n", rc);
2382 EXPORT_SYMBOL(ceph_osdc_readpages);
2385 * do a synchronous write on N pages
2387 int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
2388 struct ceph_file_layout *layout,
2389 struct ceph_snap_context *snapc,
2391 u32 truncate_seq, u64 truncate_size,
2392 struct timespec *mtime,
2393 struct page **pages, int num_pages)
2395 struct ceph_osd_request *req;
2397 int page_align = off & ~PAGE_MASK;
2399 BUG_ON(vino.snap != CEPH_NOSNAP); /* snapshots aren't writeable */
2400 req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 1,
2402 CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
2403 snapc, truncate_seq, truncate_size,
2406 return PTR_ERR(req);
2408 /* it may be a short write due to an object boundary */
2409 osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align,
2411 dout("writepages %llu~%llu (%llu bytes)\n", off, len, len);
2413 ceph_osdc_build_request(req, off, snapc, CEPH_NOSNAP, mtime);
2415 rc = ceph_osdc_start_request(osdc, req, true);
2417 rc = ceph_osdc_wait_request(osdc, req);
2419 ceph_osdc_put_request(req);
2422 dout("writepages result %d\n", rc);
2425 EXPORT_SYMBOL(ceph_osdc_writepages);
2427 int ceph_osdc_setup(void)
2429 BUG_ON(ceph_osd_request_cache);
2430 ceph_osd_request_cache = kmem_cache_create("ceph_osd_request",
2431 sizeof (struct ceph_osd_request),
2432 __alignof__(struct ceph_osd_request),
2435 return ceph_osd_request_cache ? 0 : -ENOMEM;
2437 EXPORT_SYMBOL(ceph_osdc_setup);
2439 void ceph_osdc_cleanup(void)
2441 BUG_ON(!ceph_osd_request_cache);
2442 kmem_cache_destroy(ceph_osd_request_cache);
2443 ceph_osd_request_cache = NULL;
2445 EXPORT_SYMBOL(ceph_osdc_cleanup);
2448 * handle incoming message
2450 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
2452 struct ceph_osd *osd = con->private;
2453 struct ceph_osd_client *osdc;
2454 int type = le16_to_cpu(msg->hdr.type);
2461 case CEPH_MSG_OSD_MAP:
2462 ceph_osdc_handle_map(osdc, msg);
2464 case CEPH_MSG_OSD_OPREPLY:
2465 handle_reply(osdc, msg, con);
2467 case CEPH_MSG_WATCH_NOTIFY:
2468 handle_watch_notify(osdc, msg);
2472 pr_err("received unknown message type %d %s\n", type,
2473 ceph_msg_type_name(type));
2480 * lookup and return message for incoming reply. set up reply message
2483 static struct ceph_msg *get_reply(struct ceph_connection *con,
2484 struct ceph_msg_header *hdr,
2487 struct ceph_osd *osd = con->private;
2488 struct ceph_osd_client *osdc = osd->o_osdc;
2490 struct ceph_osd_request *req;
2491 int front = le32_to_cpu(hdr->front_len);
2492 int data_len = le32_to_cpu(hdr->data_len);
2495 tid = le64_to_cpu(hdr->tid);
2496 mutex_lock(&osdc->request_mutex);
2497 req = __lookup_request(osdc, tid);
2501 dout("get_reply unknown tid %llu from osd%d\n", tid,
2506 if (req->r_reply->con)
2507 dout("%s revoking msg %p from old con %p\n", __func__,
2508 req->r_reply, req->r_reply->con);
2509 ceph_msg_revoke_incoming(req->r_reply);
2511 if (front > req->r_reply->front.iov_len) {
2512 pr_warning("get_reply front %d > preallocated %d\n",
2513 front, (int)req->r_reply->front.iov_len);
2514 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front, GFP_NOFS, false);
2517 ceph_msg_put(req->r_reply);
2520 m = ceph_msg_get(req->r_reply);
2523 struct ceph_osd_data *osd_data;
2526 * XXX This is assuming there is only one op containing
2527 * XXX page data. Probably OK for reads, but this
2528 * XXX ought to be done more generally.
2530 osd_data = osd_req_op_extent_osd_data(req, 0);
2531 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
2532 if (osd_data->pages &&
2533 unlikely(osd_data->length < data_len)) {
2535 pr_warning("tid %lld reply has %d bytes "
2536 "we had only %llu bytes ready\n",
2537 tid, data_len, osd_data->length);
2546 dout("get_reply tid %lld %p\n", tid, m);
2549 mutex_unlock(&osdc->request_mutex);
2554 static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2555 struct ceph_msg_header *hdr,
2558 struct ceph_osd *osd = con->private;
2559 int type = le16_to_cpu(hdr->type);
2560 int front = le32_to_cpu(hdr->front_len);
2564 case CEPH_MSG_OSD_MAP:
2565 case CEPH_MSG_WATCH_NOTIFY:
2566 return ceph_msg_new(type, front, GFP_NOFS, false);
2567 case CEPH_MSG_OSD_OPREPLY:
2568 return get_reply(con, hdr, skip);
2570 pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
2578 * Wrappers to refcount containing ceph_osd struct
2580 static struct ceph_connection *get_osd_con(struct ceph_connection *con)
2582 struct ceph_osd *osd = con->private;
2588 static void put_osd_con(struct ceph_connection *con)
2590 struct ceph_osd *osd = con->private;
2598 * Note: returned pointer is the address of a structure that's
2599 * managed separately. Caller must *not* attempt to free it.
2601 static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
2602 int *proto, int force_new)
2604 struct ceph_osd *o = con->private;
2605 struct ceph_osd_client *osdc = o->o_osdc;
2606 struct ceph_auth_client *ac = osdc->client->monc.auth;
2607 struct ceph_auth_handshake *auth = &o->o_auth;
2609 if (force_new && auth->authorizer) {
2610 ceph_auth_destroy_authorizer(ac, auth->authorizer);
2611 auth->authorizer = NULL;
2613 if (!auth->authorizer) {
2614 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2617 return ERR_PTR(ret);
2619 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2622 return ERR_PTR(ret);
2624 *proto = ac->protocol;
2630 static int verify_authorizer_reply(struct ceph_connection *con, int len)
2632 struct ceph_osd *o = con->private;
2633 struct ceph_osd_client *osdc = o->o_osdc;
2634 struct ceph_auth_client *ac = osdc->client->monc.auth;
2636 return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer, len);
2639 static int invalidate_authorizer(struct ceph_connection *con)
2641 struct ceph_osd *o = con->private;
2642 struct ceph_osd_client *osdc = o->o_osdc;
2643 struct ceph_auth_client *ac = osdc->client->monc.auth;
2645 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
2646 return ceph_monc_validate_auth(&osdc->client->monc);
2649 static const struct ceph_connection_operations osd_con_ops = {
2652 .dispatch = dispatch,
2653 .get_authorizer = get_authorizer,
2654 .verify_authorizer_reply = verify_authorizer_reply,
2655 .invalidate_authorizer = invalidate_authorizer,
2656 .alloc_msg = alloc_msg,