1 // SPDX-License-Identifier: GPL-2.0
3 #include <linux/ceph/ceph_debug.h>
5 #include <linux/module.h>
7 #include <linux/highmem.h>
9 #include <linux/pagemap.h>
10 #include <linux/slab.h>
11 #include <linux/uaccess.h>
13 #include <linux/bio.h>
16 #include <linux/ceph/ceph_features.h>
17 #include <linux/ceph/libceph.h>
18 #include <linux/ceph/osd_client.h>
19 #include <linux/ceph/messenger.h>
20 #include <linux/ceph/decode.h>
21 #include <linux/ceph/auth.h>
22 #include <linux/ceph/pagelist.h>
23 #include <linux/ceph/striper.h>
25 #define OSD_OPREPLY_FRONT_LEN 512
27 static struct kmem_cache *ceph_osd_request_cache;
29 static const struct ceph_connection_operations osd_con_ops;
32 * Implement client access to distributed object storage cluster.
34 * All data objects are stored within a cluster/cloud of OSDs, or
35 * "object storage devices." (Note that Ceph OSDs have _nothing_ to
36 * do with the T10 OSD extensions to SCSI.) Ceph OSDs are simply
37 * remote daemons serving up and coordinating consistent and safe
40 * Cluster membership and the mapping of data objects onto storage devices
41 * are described by the osd map.
43 * We keep track of pending OSD requests (read, write), resubmit
44 * requests to different OSDs when the cluster topology/data layout
45 * change, or retry the affected requests when the communications
46 * channel with an OSD is reset.
49 static void link_request(struct ceph_osd *osd, struct ceph_osd_request *req);
50 static void unlink_request(struct ceph_osd *osd, struct ceph_osd_request *req);
51 static void link_linger(struct ceph_osd *osd,
52 struct ceph_osd_linger_request *lreq);
53 static void unlink_linger(struct ceph_osd *osd,
54 struct ceph_osd_linger_request *lreq);
55 static void clear_backoffs(struct ceph_osd *osd);
58 static inline bool rwsem_is_wrlocked(struct rw_semaphore *sem)
62 if (unlikely(down_read_trylock(sem))) {
69 static inline void verify_osdc_locked(struct ceph_osd_client *osdc)
71 WARN_ON(!rwsem_is_locked(&osdc->lock));
73 static inline void verify_osdc_wrlocked(struct ceph_osd_client *osdc)
75 WARN_ON(!rwsem_is_wrlocked(&osdc->lock));
77 static inline void verify_osd_locked(struct ceph_osd *osd)
79 struct ceph_osd_client *osdc = osd->o_osdc;
81 WARN_ON(!(mutex_is_locked(&osd->lock) &&
82 rwsem_is_locked(&osdc->lock)) &&
83 !rwsem_is_wrlocked(&osdc->lock));
85 static inline void verify_lreq_locked(struct ceph_osd_linger_request *lreq)
87 WARN_ON(!mutex_is_locked(&lreq->lock));
90 static inline void verify_osdc_locked(struct ceph_osd_client *osdc) { }
91 static inline void verify_osdc_wrlocked(struct ceph_osd_client *osdc) { }
92 static inline void verify_osd_locked(struct ceph_osd *osd) { }
93 static inline void verify_lreq_locked(struct ceph_osd_linger_request *lreq) { }
97 * calculate the mapping of a file extent onto an object, and fill out the
98 * request accordingly. shorten extent as necessary if it crosses an
101 * fill osd op in request message.
103 static int calc_layout(struct ceph_file_layout *layout, u64 off, u64 *plen,
104 u64 *objnum, u64 *objoff, u64 *objlen)
106 u64 orig_len = *plen;
110 ceph_calc_file_object_mapping(layout, off, orig_len, objnum,
113 if (*objlen < orig_len) {
115 dout(" skipping last %llu, final file extent %llu~%llu\n",
116 orig_len - *plen, off, *plen);
119 dout("calc_layout objnum=%llx %llu~%llu\n", *objnum, *objoff, *objlen);
123 static void ceph_osd_data_init(struct ceph_osd_data *osd_data)
125 memset(osd_data, 0, sizeof (*osd_data));
126 osd_data->type = CEPH_OSD_DATA_TYPE_NONE;
130 * Consumes @pages if @own_pages is true.
132 static void ceph_osd_data_pages_init(struct ceph_osd_data *osd_data,
133 struct page **pages, u64 length, u32 alignment,
134 bool pages_from_pool, bool own_pages)
136 osd_data->type = CEPH_OSD_DATA_TYPE_PAGES;
137 osd_data->pages = pages;
138 osd_data->length = length;
139 osd_data->alignment = alignment;
140 osd_data->pages_from_pool = pages_from_pool;
141 osd_data->own_pages = own_pages;
145 * Consumes a ref on @pagelist.
147 static void ceph_osd_data_pagelist_init(struct ceph_osd_data *osd_data,
148 struct ceph_pagelist *pagelist)
150 osd_data->type = CEPH_OSD_DATA_TYPE_PAGELIST;
151 osd_data->pagelist = pagelist;
155 static void ceph_osd_data_bio_init(struct ceph_osd_data *osd_data,
156 struct ceph_bio_iter *bio_pos,
159 osd_data->type = CEPH_OSD_DATA_TYPE_BIO;
160 osd_data->bio_pos = *bio_pos;
161 osd_data->bio_length = bio_length;
163 #endif /* CONFIG_BLOCK */
165 static void ceph_osd_data_bvecs_init(struct ceph_osd_data *osd_data,
166 struct ceph_bvec_iter *bvec_pos,
169 osd_data->type = CEPH_OSD_DATA_TYPE_BVECS;
170 osd_data->bvec_pos = *bvec_pos;
171 osd_data->num_bvecs = num_bvecs;
174 static struct ceph_osd_data *
175 osd_req_op_raw_data_in(struct ceph_osd_request *osd_req, unsigned int which)
177 BUG_ON(which >= osd_req->r_num_ops);
179 return &osd_req->r_ops[which].raw_data_in;
182 struct ceph_osd_data *
183 osd_req_op_extent_osd_data(struct ceph_osd_request *osd_req,
186 return osd_req_op_data(osd_req, which, extent, osd_data);
188 EXPORT_SYMBOL(osd_req_op_extent_osd_data);
190 void osd_req_op_raw_data_in_pages(struct ceph_osd_request *osd_req,
191 unsigned int which, struct page **pages,
192 u64 length, u32 alignment,
193 bool pages_from_pool, bool own_pages)
195 struct ceph_osd_data *osd_data;
197 osd_data = osd_req_op_raw_data_in(osd_req, which);
198 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
199 pages_from_pool, own_pages);
201 EXPORT_SYMBOL(osd_req_op_raw_data_in_pages);
203 void osd_req_op_extent_osd_data_pages(struct ceph_osd_request *osd_req,
204 unsigned int which, struct page **pages,
205 u64 length, u32 alignment,
206 bool pages_from_pool, bool own_pages)
208 struct ceph_osd_data *osd_data;
210 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
211 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
212 pages_from_pool, own_pages);
214 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pages);
216 void osd_req_op_extent_osd_data_pagelist(struct ceph_osd_request *osd_req,
217 unsigned int which, struct ceph_pagelist *pagelist)
219 struct ceph_osd_data *osd_data;
221 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
222 ceph_osd_data_pagelist_init(osd_data, pagelist);
224 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pagelist);
227 void osd_req_op_extent_osd_data_bio(struct ceph_osd_request *osd_req,
229 struct ceph_bio_iter *bio_pos,
232 struct ceph_osd_data *osd_data;
234 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
235 ceph_osd_data_bio_init(osd_data, bio_pos, bio_length);
237 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bio);
238 #endif /* CONFIG_BLOCK */
240 void osd_req_op_extent_osd_data_bvecs(struct ceph_osd_request *osd_req,
242 struct bio_vec *bvecs, u32 num_bvecs,
245 struct ceph_osd_data *osd_data;
246 struct ceph_bvec_iter it = {
248 .iter = { .bi_size = bytes },
251 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
252 ceph_osd_data_bvecs_init(osd_data, &it, num_bvecs);
254 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bvecs);
256 void osd_req_op_extent_osd_data_bvec_pos(struct ceph_osd_request *osd_req,
258 struct ceph_bvec_iter *bvec_pos)
260 struct ceph_osd_data *osd_data;
262 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
263 ceph_osd_data_bvecs_init(osd_data, bvec_pos, 0);
265 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bvec_pos);
267 static void osd_req_op_cls_request_info_pagelist(
268 struct ceph_osd_request *osd_req,
269 unsigned int which, struct ceph_pagelist *pagelist)
271 struct ceph_osd_data *osd_data;
273 osd_data = osd_req_op_data(osd_req, which, cls, request_info);
274 ceph_osd_data_pagelist_init(osd_data, pagelist);
277 void osd_req_op_cls_request_data_pagelist(
278 struct ceph_osd_request *osd_req,
279 unsigned int which, struct ceph_pagelist *pagelist)
281 struct ceph_osd_data *osd_data;
283 osd_data = osd_req_op_data(osd_req, which, cls, request_data);
284 ceph_osd_data_pagelist_init(osd_data, pagelist);
285 osd_req->r_ops[which].cls.indata_len += pagelist->length;
286 osd_req->r_ops[which].indata_len += pagelist->length;
288 EXPORT_SYMBOL(osd_req_op_cls_request_data_pagelist);
290 void osd_req_op_cls_request_data_pages(struct ceph_osd_request *osd_req,
291 unsigned int which, struct page **pages, u64 length,
292 u32 alignment, bool pages_from_pool, bool own_pages)
294 struct ceph_osd_data *osd_data;
296 osd_data = osd_req_op_data(osd_req, which, cls, request_data);
297 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
298 pages_from_pool, own_pages);
299 osd_req->r_ops[which].cls.indata_len += length;
300 osd_req->r_ops[which].indata_len += length;
302 EXPORT_SYMBOL(osd_req_op_cls_request_data_pages);
304 void osd_req_op_cls_request_data_bvecs(struct ceph_osd_request *osd_req,
306 struct bio_vec *bvecs, u32 num_bvecs,
309 struct ceph_osd_data *osd_data;
310 struct ceph_bvec_iter it = {
312 .iter = { .bi_size = bytes },
315 osd_data = osd_req_op_data(osd_req, which, cls, request_data);
316 ceph_osd_data_bvecs_init(osd_data, &it, num_bvecs);
317 osd_req->r_ops[which].cls.indata_len += bytes;
318 osd_req->r_ops[which].indata_len += bytes;
320 EXPORT_SYMBOL(osd_req_op_cls_request_data_bvecs);
322 void osd_req_op_cls_response_data_pages(struct ceph_osd_request *osd_req,
323 unsigned int which, struct page **pages, u64 length,
324 u32 alignment, bool pages_from_pool, bool own_pages)
326 struct ceph_osd_data *osd_data;
328 osd_data = osd_req_op_data(osd_req, which, cls, response_data);
329 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
330 pages_from_pool, own_pages);
332 EXPORT_SYMBOL(osd_req_op_cls_response_data_pages);
334 static u64 ceph_osd_data_length(struct ceph_osd_data *osd_data)
336 switch (osd_data->type) {
337 case CEPH_OSD_DATA_TYPE_NONE:
339 case CEPH_OSD_DATA_TYPE_PAGES:
340 return osd_data->length;
341 case CEPH_OSD_DATA_TYPE_PAGELIST:
342 return (u64)osd_data->pagelist->length;
344 case CEPH_OSD_DATA_TYPE_BIO:
345 return (u64)osd_data->bio_length;
346 #endif /* CONFIG_BLOCK */
347 case CEPH_OSD_DATA_TYPE_BVECS:
348 return osd_data->bvec_pos.iter.bi_size;
350 WARN(true, "unrecognized data type %d\n", (int)osd_data->type);
355 static void ceph_osd_data_release(struct ceph_osd_data *osd_data)
357 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES && osd_data->own_pages) {
360 num_pages = calc_pages_for((u64)osd_data->alignment,
361 (u64)osd_data->length);
362 ceph_release_page_vector(osd_data->pages, num_pages);
363 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
364 ceph_pagelist_release(osd_data->pagelist);
366 ceph_osd_data_init(osd_data);
369 static void osd_req_op_data_release(struct ceph_osd_request *osd_req,
372 struct ceph_osd_req_op *op;
374 BUG_ON(which >= osd_req->r_num_ops);
375 op = &osd_req->r_ops[which];
378 case CEPH_OSD_OP_READ:
379 case CEPH_OSD_OP_WRITE:
380 case CEPH_OSD_OP_WRITEFULL:
381 ceph_osd_data_release(&op->extent.osd_data);
383 case CEPH_OSD_OP_CALL:
384 ceph_osd_data_release(&op->cls.request_info);
385 ceph_osd_data_release(&op->cls.request_data);
386 ceph_osd_data_release(&op->cls.response_data);
388 case CEPH_OSD_OP_SETXATTR:
389 case CEPH_OSD_OP_CMPXATTR:
390 ceph_osd_data_release(&op->xattr.osd_data);
392 case CEPH_OSD_OP_STAT:
393 ceph_osd_data_release(&op->raw_data_in);
395 case CEPH_OSD_OP_NOTIFY_ACK:
396 ceph_osd_data_release(&op->notify_ack.request_data);
398 case CEPH_OSD_OP_NOTIFY:
399 ceph_osd_data_release(&op->notify.request_data);
400 ceph_osd_data_release(&op->notify.response_data);
402 case CEPH_OSD_OP_LIST_WATCHERS:
403 ceph_osd_data_release(&op->list_watchers.response_data);
405 case CEPH_OSD_OP_COPY_FROM2:
406 ceph_osd_data_release(&op->copy_from.osd_data);
414 * Assumes @t is zero-initialized.
416 static void target_init(struct ceph_osd_request_target *t)
418 ceph_oid_init(&t->base_oid);
419 ceph_oloc_init(&t->base_oloc);
420 ceph_oid_init(&t->target_oid);
421 ceph_oloc_init(&t->target_oloc);
423 ceph_osds_init(&t->acting);
424 ceph_osds_init(&t->up);
428 t->osd = CEPH_HOMELESS_OSD;
431 static void target_copy(struct ceph_osd_request_target *dest,
432 const struct ceph_osd_request_target *src)
434 ceph_oid_copy(&dest->base_oid, &src->base_oid);
435 ceph_oloc_copy(&dest->base_oloc, &src->base_oloc);
436 ceph_oid_copy(&dest->target_oid, &src->target_oid);
437 ceph_oloc_copy(&dest->target_oloc, &src->target_oloc);
439 dest->pgid = src->pgid; /* struct */
440 dest->spgid = src->spgid; /* struct */
441 dest->pg_num = src->pg_num;
442 dest->pg_num_mask = src->pg_num_mask;
443 ceph_osds_copy(&dest->acting, &src->acting);
444 ceph_osds_copy(&dest->up, &src->up);
445 dest->size = src->size;
446 dest->min_size = src->min_size;
447 dest->sort_bitwise = src->sort_bitwise;
448 dest->recovery_deletes = src->recovery_deletes;
450 dest->flags = src->flags;
451 dest->used_replica = src->used_replica;
452 dest->paused = src->paused;
454 dest->epoch = src->epoch;
455 dest->last_force_resend = src->last_force_resend;
457 dest->osd = src->osd;
460 static void target_destroy(struct ceph_osd_request_target *t)
462 ceph_oid_destroy(&t->base_oid);
463 ceph_oloc_destroy(&t->base_oloc);
464 ceph_oid_destroy(&t->target_oid);
465 ceph_oloc_destroy(&t->target_oloc);
471 static void request_release_checks(struct ceph_osd_request *req)
473 WARN_ON(!RB_EMPTY_NODE(&req->r_node));
474 WARN_ON(!RB_EMPTY_NODE(&req->r_mc_node));
475 WARN_ON(!list_empty(&req->r_private_item));
479 static void ceph_osdc_release_request(struct kref *kref)
481 struct ceph_osd_request *req = container_of(kref,
482 struct ceph_osd_request, r_kref);
485 dout("%s %p (r_request %p r_reply %p)\n", __func__, req,
486 req->r_request, req->r_reply);
487 request_release_checks(req);
490 ceph_msg_put(req->r_request);
492 ceph_msg_put(req->r_reply);
494 for (which = 0; which < req->r_num_ops; which++)
495 osd_req_op_data_release(req, which);
497 target_destroy(&req->r_t);
498 ceph_put_snap_context(req->r_snapc);
501 mempool_free(req, req->r_osdc->req_mempool);
502 else if (req->r_num_ops <= CEPH_OSD_SLAB_OPS)
503 kmem_cache_free(ceph_osd_request_cache, req);
508 void ceph_osdc_get_request(struct ceph_osd_request *req)
510 dout("%s %p (was %d)\n", __func__, req,
511 kref_read(&req->r_kref));
512 kref_get(&req->r_kref);
514 EXPORT_SYMBOL(ceph_osdc_get_request);
516 void ceph_osdc_put_request(struct ceph_osd_request *req)
519 dout("%s %p (was %d)\n", __func__, req,
520 kref_read(&req->r_kref));
521 kref_put(&req->r_kref, ceph_osdc_release_request);
524 EXPORT_SYMBOL(ceph_osdc_put_request);
526 static void request_init(struct ceph_osd_request *req)
528 /* req only, each op is zeroed in _osd_req_op_init() */
529 memset(req, 0, sizeof(*req));
531 kref_init(&req->r_kref);
532 init_completion(&req->r_completion);
533 RB_CLEAR_NODE(&req->r_node);
534 RB_CLEAR_NODE(&req->r_mc_node);
535 INIT_LIST_HEAD(&req->r_private_item);
537 target_init(&req->r_t);
541 * This is ugly, but it allows us to reuse linger registration and ping
542 * requests, keeping the structure of the code around send_linger{_ping}()
543 * reasonable. Setting up a min_nr=2 mempool for each linger request
544 * and dealing with copying ops (this blasts req only, watch op remains
545 * intact) isn't any better.
547 static void request_reinit(struct ceph_osd_request *req)
549 struct ceph_osd_client *osdc = req->r_osdc;
550 bool mempool = req->r_mempool;
551 unsigned int num_ops = req->r_num_ops;
552 u64 snapid = req->r_snapid;
553 struct ceph_snap_context *snapc = req->r_snapc;
554 bool linger = req->r_linger;
555 struct ceph_msg *request_msg = req->r_request;
556 struct ceph_msg *reply_msg = req->r_reply;
558 dout("%s req %p\n", __func__, req);
559 WARN_ON(kref_read(&req->r_kref) != 1);
560 request_release_checks(req);
562 WARN_ON(kref_read(&request_msg->kref) != 1);
563 WARN_ON(kref_read(&reply_msg->kref) != 1);
564 target_destroy(&req->r_t);
568 req->r_mempool = mempool;
569 req->r_num_ops = num_ops;
570 req->r_snapid = snapid;
571 req->r_snapc = snapc;
572 req->r_linger = linger;
573 req->r_request = request_msg;
574 req->r_reply = reply_msg;
577 struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
578 struct ceph_snap_context *snapc,
579 unsigned int num_ops,
583 struct ceph_osd_request *req;
586 BUG_ON(num_ops > CEPH_OSD_SLAB_OPS);
587 req = mempool_alloc(osdc->req_mempool, gfp_flags);
588 } else if (num_ops <= CEPH_OSD_SLAB_OPS) {
589 req = kmem_cache_alloc(ceph_osd_request_cache, gfp_flags);
591 BUG_ON(num_ops > CEPH_OSD_MAX_OPS);
592 req = kmalloc(struct_size(req, r_ops, num_ops), gfp_flags);
599 req->r_mempool = use_mempool;
600 req->r_num_ops = num_ops;
601 req->r_snapid = CEPH_NOSNAP;
602 req->r_snapc = ceph_get_snap_context(snapc);
604 dout("%s req %p\n", __func__, req);
607 EXPORT_SYMBOL(ceph_osdc_alloc_request);
609 static int ceph_oloc_encoding_size(const struct ceph_object_locator *oloc)
611 return 8 + 4 + 4 + 4 + (oloc->pool_ns ? oloc->pool_ns->len : 0);
614 static int __ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp,
615 int num_request_data_items,
616 int num_reply_data_items)
618 struct ceph_osd_client *osdc = req->r_osdc;
619 struct ceph_msg *msg;
622 WARN_ON(req->r_request || req->r_reply);
623 WARN_ON(ceph_oid_empty(&req->r_base_oid));
624 WARN_ON(ceph_oloc_empty(&req->r_base_oloc));
626 /* create request message */
627 msg_size = CEPH_ENCODING_START_BLK_LEN +
628 CEPH_PGID_ENCODING_LEN + 1; /* spgid */
629 msg_size += 4 + 4 + 4; /* hash, osdmap_epoch, flags */
630 msg_size += CEPH_ENCODING_START_BLK_LEN +
631 sizeof(struct ceph_osd_reqid); /* reqid */
632 msg_size += sizeof(struct ceph_blkin_trace_info); /* trace */
633 msg_size += 4 + sizeof(struct ceph_timespec); /* client_inc, mtime */
634 msg_size += CEPH_ENCODING_START_BLK_LEN +
635 ceph_oloc_encoding_size(&req->r_base_oloc); /* oloc */
636 msg_size += 4 + req->r_base_oid.name_len; /* oid */
637 msg_size += 2 + req->r_num_ops * sizeof(struct ceph_osd_op);
638 msg_size += 8; /* snapid */
639 msg_size += 8; /* snap_seq */
640 msg_size += 4 + 8 * (req->r_snapc ? req->r_snapc->num_snaps : 0);
641 msg_size += 4 + 8; /* retry_attempt, features */
644 msg = ceph_msgpool_get(&osdc->msgpool_op, msg_size,
645 num_request_data_items);
647 msg = ceph_msg_new2(CEPH_MSG_OSD_OP, msg_size,
648 num_request_data_items, gfp, true);
652 memset(msg->front.iov_base, 0, msg->front.iov_len);
653 req->r_request = msg;
655 /* create reply message */
656 msg_size = OSD_OPREPLY_FRONT_LEN;
657 msg_size += req->r_base_oid.name_len;
658 msg_size += req->r_num_ops * sizeof(struct ceph_osd_op);
661 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, msg_size,
662 num_reply_data_items);
664 msg = ceph_msg_new2(CEPH_MSG_OSD_OPREPLY, msg_size,
665 num_reply_data_items, gfp, true);
674 static bool osd_req_opcode_valid(u16 opcode)
677 #define GENERATE_CASE(op, opcode, str) case CEPH_OSD_OP_##op: return true;
678 __CEPH_FORALL_OSD_OPS(GENERATE_CASE)
685 static void get_num_data_items(struct ceph_osd_request *req,
686 int *num_request_data_items,
687 int *num_reply_data_items)
689 struct ceph_osd_req_op *op;
691 *num_request_data_items = 0;
692 *num_reply_data_items = 0;
694 for (op = req->r_ops; op != &req->r_ops[req->r_num_ops]; op++) {
697 case CEPH_OSD_OP_WRITE:
698 case CEPH_OSD_OP_WRITEFULL:
699 case CEPH_OSD_OP_SETXATTR:
700 case CEPH_OSD_OP_CMPXATTR:
701 case CEPH_OSD_OP_NOTIFY_ACK:
702 case CEPH_OSD_OP_COPY_FROM2:
703 *num_request_data_items += 1;
707 case CEPH_OSD_OP_STAT:
708 case CEPH_OSD_OP_READ:
709 case CEPH_OSD_OP_LIST_WATCHERS:
710 *num_reply_data_items += 1;
714 case CEPH_OSD_OP_NOTIFY:
715 *num_request_data_items += 1;
716 *num_reply_data_items += 1;
718 case CEPH_OSD_OP_CALL:
719 *num_request_data_items += 2;
720 *num_reply_data_items += 1;
724 WARN_ON(!osd_req_opcode_valid(op->op));
731 * oid, oloc and OSD op opcode(s) must be filled in before this function
734 int ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp)
736 int num_request_data_items, num_reply_data_items;
738 get_num_data_items(req, &num_request_data_items, &num_reply_data_items);
739 return __ceph_osdc_alloc_messages(req, gfp, num_request_data_items,
740 num_reply_data_items);
742 EXPORT_SYMBOL(ceph_osdc_alloc_messages);
745 * This is an osd op init function for opcodes that have no data or
746 * other information associated with them. It also serves as a
747 * common init routine for all the other init functions, below.
749 static struct ceph_osd_req_op *
750 _osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which,
751 u16 opcode, u32 flags)
753 struct ceph_osd_req_op *op;
755 BUG_ON(which >= osd_req->r_num_ops);
756 BUG_ON(!osd_req_opcode_valid(opcode));
758 op = &osd_req->r_ops[which];
759 memset(op, 0, sizeof (*op));
766 void osd_req_op_init(struct ceph_osd_request *osd_req,
767 unsigned int which, u16 opcode, u32 flags)
769 (void)_osd_req_op_init(osd_req, which, opcode, flags);
771 EXPORT_SYMBOL(osd_req_op_init);
773 void osd_req_op_extent_init(struct ceph_osd_request *osd_req,
774 unsigned int which, u16 opcode,
775 u64 offset, u64 length,
776 u64 truncate_size, u32 truncate_seq)
778 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
780 size_t payload_len = 0;
782 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
783 opcode != CEPH_OSD_OP_WRITEFULL && opcode != CEPH_OSD_OP_ZERO &&
784 opcode != CEPH_OSD_OP_TRUNCATE);
786 op->extent.offset = offset;
787 op->extent.length = length;
788 op->extent.truncate_size = truncate_size;
789 op->extent.truncate_seq = truncate_seq;
790 if (opcode == CEPH_OSD_OP_WRITE || opcode == CEPH_OSD_OP_WRITEFULL)
791 payload_len += length;
793 op->indata_len = payload_len;
795 EXPORT_SYMBOL(osd_req_op_extent_init);
797 void osd_req_op_extent_update(struct ceph_osd_request *osd_req,
798 unsigned int which, u64 length)
800 struct ceph_osd_req_op *op;
803 BUG_ON(which >= osd_req->r_num_ops);
804 op = &osd_req->r_ops[which];
805 previous = op->extent.length;
807 if (length == previous)
808 return; /* Nothing to do */
809 BUG_ON(length > previous);
811 op->extent.length = length;
812 if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL)
813 op->indata_len -= previous - length;
815 EXPORT_SYMBOL(osd_req_op_extent_update);
817 void osd_req_op_extent_dup_last(struct ceph_osd_request *osd_req,
818 unsigned int which, u64 offset_inc)
820 struct ceph_osd_req_op *op, *prev_op;
822 BUG_ON(which + 1 >= osd_req->r_num_ops);
824 prev_op = &osd_req->r_ops[which];
825 op = _osd_req_op_init(osd_req, which + 1, prev_op->op, prev_op->flags);
826 /* dup previous one */
827 op->indata_len = prev_op->indata_len;
828 op->outdata_len = prev_op->outdata_len;
829 op->extent = prev_op->extent;
831 op->extent.offset += offset_inc;
832 op->extent.length -= offset_inc;
834 if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL)
835 op->indata_len -= offset_inc;
837 EXPORT_SYMBOL(osd_req_op_extent_dup_last);
839 int osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
840 const char *class, const char *method)
842 struct ceph_osd_req_op *op;
843 struct ceph_pagelist *pagelist;
844 size_t payload_len = 0;
848 op = _osd_req_op_init(osd_req, which, CEPH_OSD_OP_CALL, 0);
850 pagelist = ceph_pagelist_alloc(GFP_NOFS);
854 op->cls.class_name = class;
855 size = strlen(class);
856 BUG_ON(size > (size_t) U8_MAX);
857 op->cls.class_len = size;
858 ret = ceph_pagelist_append(pagelist, class, size);
860 goto err_pagelist_free;
863 op->cls.method_name = method;
864 size = strlen(method);
865 BUG_ON(size > (size_t) U8_MAX);
866 op->cls.method_len = size;
867 ret = ceph_pagelist_append(pagelist, method, size);
869 goto err_pagelist_free;
872 osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist);
873 op->indata_len = payload_len;
877 ceph_pagelist_release(pagelist);
880 EXPORT_SYMBOL(osd_req_op_cls_init);
882 int osd_req_op_xattr_init(struct ceph_osd_request *osd_req, unsigned int which,
883 u16 opcode, const char *name, const void *value,
884 size_t size, u8 cmp_op, u8 cmp_mode)
886 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
888 struct ceph_pagelist *pagelist;
892 BUG_ON(opcode != CEPH_OSD_OP_SETXATTR && opcode != CEPH_OSD_OP_CMPXATTR);
894 pagelist = ceph_pagelist_alloc(GFP_NOFS);
898 payload_len = strlen(name);
899 op->xattr.name_len = payload_len;
900 ret = ceph_pagelist_append(pagelist, name, payload_len);
902 goto err_pagelist_free;
904 op->xattr.value_len = size;
905 ret = ceph_pagelist_append(pagelist, value, size);
907 goto err_pagelist_free;
910 op->xattr.cmp_op = cmp_op;
911 op->xattr.cmp_mode = cmp_mode;
913 ceph_osd_data_pagelist_init(&op->xattr.osd_data, pagelist);
914 op->indata_len = payload_len;
918 ceph_pagelist_release(pagelist);
921 EXPORT_SYMBOL(osd_req_op_xattr_init);
924 * @watch_opcode: CEPH_OSD_WATCH_OP_*
926 static void osd_req_op_watch_init(struct ceph_osd_request *req, int which,
927 u64 cookie, u8 watch_opcode)
929 struct ceph_osd_req_op *op;
931 op = _osd_req_op_init(req, which, CEPH_OSD_OP_WATCH, 0);
932 op->watch.cookie = cookie;
933 op->watch.op = watch_opcode;
938 * @flags: CEPH_OSD_OP_ALLOC_HINT_FLAG_*
940 void osd_req_op_alloc_hint_init(struct ceph_osd_request *osd_req,
942 u64 expected_object_size,
943 u64 expected_write_size,
946 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
947 CEPH_OSD_OP_SETALLOCHINT,
950 op->alloc_hint.expected_object_size = expected_object_size;
951 op->alloc_hint.expected_write_size = expected_write_size;
952 op->alloc_hint.flags = flags;
955 * CEPH_OSD_OP_SETALLOCHINT op is advisory and therefore deemed
956 * not worth a feature bit. Set FAILOK per-op flag to make
957 * sure older osds don't trip over an unsupported opcode.
959 op->flags |= CEPH_OSD_OP_FLAG_FAILOK;
961 EXPORT_SYMBOL(osd_req_op_alloc_hint_init);
963 static void ceph_osdc_msg_data_add(struct ceph_msg *msg,
964 struct ceph_osd_data *osd_data)
966 u64 length = ceph_osd_data_length(osd_data);
968 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
969 BUG_ON(length > (u64) SIZE_MAX);
971 ceph_msg_data_add_pages(msg, osd_data->pages,
972 length, osd_data->alignment, false);
973 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
975 ceph_msg_data_add_pagelist(msg, osd_data->pagelist);
977 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
978 ceph_msg_data_add_bio(msg, &osd_data->bio_pos, length);
980 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BVECS) {
981 ceph_msg_data_add_bvecs(msg, &osd_data->bvec_pos);
983 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
987 static u32 osd_req_encode_op(struct ceph_osd_op *dst,
988 const struct ceph_osd_req_op *src)
991 case CEPH_OSD_OP_STAT:
993 case CEPH_OSD_OP_READ:
994 case CEPH_OSD_OP_WRITE:
995 case CEPH_OSD_OP_WRITEFULL:
996 case CEPH_OSD_OP_ZERO:
997 case CEPH_OSD_OP_TRUNCATE:
998 dst->extent.offset = cpu_to_le64(src->extent.offset);
999 dst->extent.length = cpu_to_le64(src->extent.length);
1000 dst->extent.truncate_size =
1001 cpu_to_le64(src->extent.truncate_size);
1002 dst->extent.truncate_seq =
1003 cpu_to_le32(src->extent.truncate_seq);
1005 case CEPH_OSD_OP_CALL:
1006 dst->cls.class_len = src->cls.class_len;
1007 dst->cls.method_len = src->cls.method_len;
1008 dst->cls.indata_len = cpu_to_le32(src->cls.indata_len);
1010 case CEPH_OSD_OP_WATCH:
1011 dst->watch.cookie = cpu_to_le64(src->watch.cookie);
1012 dst->watch.ver = cpu_to_le64(0);
1013 dst->watch.op = src->watch.op;
1014 dst->watch.gen = cpu_to_le32(src->watch.gen);
1016 case CEPH_OSD_OP_NOTIFY_ACK:
1018 case CEPH_OSD_OP_NOTIFY:
1019 dst->notify.cookie = cpu_to_le64(src->notify.cookie);
1021 case CEPH_OSD_OP_LIST_WATCHERS:
1023 case CEPH_OSD_OP_SETALLOCHINT:
1024 dst->alloc_hint.expected_object_size =
1025 cpu_to_le64(src->alloc_hint.expected_object_size);
1026 dst->alloc_hint.expected_write_size =
1027 cpu_to_le64(src->alloc_hint.expected_write_size);
1028 dst->alloc_hint.flags = cpu_to_le32(src->alloc_hint.flags);
1030 case CEPH_OSD_OP_SETXATTR:
1031 case CEPH_OSD_OP_CMPXATTR:
1032 dst->xattr.name_len = cpu_to_le32(src->xattr.name_len);
1033 dst->xattr.value_len = cpu_to_le32(src->xattr.value_len);
1034 dst->xattr.cmp_op = src->xattr.cmp_op;
1035 dst->xattr.cmp_mode = src->xattr.cmp_mode;
1037 case CEPH_OSD_OP_CREATE:
1038 case CEPH_OSD_OP_DELETE:
1040 case CEPH_OSD_OP_COPY_FROM2:
1041 dst->copy_from.snapid = cpu_to_le64(src->copy_from.snapid);
1042 dst->copy_from.src_version =
1043 cpu_to_le64(src->copy_from.src_version);
1044 dst->copy_from.flags = src->copy_from.flags;
1045 dst->copy_from.src_fadvise_flags =
1046 cpu_to_le32(src->copy_from.src_fadvise_flags);
1049 pr_err("unsupported osd opcode %s\n",
1050 ceph_osd_op_name(src->op));
1056 dst->op = cpu_to_le16(src->op);
1057 dst->flags = cpu_to_le32(src->flags);
1058 dst->payload_len = cpu_to_le32(src->indata_len);
1060 return src->indata_len;
1064 * build new request AND message, calculate layout, and adjust file
1067 * if the file was recently truncated, we include information about its
1068 * old and new size so that the object can be updated appropriately. (we
1069 * avoid synchronously deleting truncated objects because it's slow.)
1071 struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
1072 struct ceph_file_layout *layout,
1073 struct ceph_vino vino,
1075 unsigned int which, int num_ops,
1076 int opcode, int flags,
1077 struct ceph_snap_context *snapc,
1082 struct ceph_osd_request *req;
1088 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
1089 opcode != CEPH_OSD_OP_ZERO && opcode != CEPH_OSD_OP_TRUNCATE &&
1090 opcode != CEPH_OSD_OP_CREATE && opcode != CEPH_OSD_OP_DELETE);
1092 req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool,
1099 /* calculate max write size */
1100 r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
1104 if (opcode == CEPH_OSD_OP_CREATE || opcode == CEPH_OSD_OP_DELETE) {
1105 osd_req_op_init(req, which, opcode, 0);
1107 u32 object_size = layout->object_size;
1108 u32 object_base = off - objoff;
1109 if (!(truncate_seq == 1 && truncate_size == -1ULL)) {
1110 if (truncate_size <= object_base) {
1113 truncate_size -= object_base;
1114 if (truncate_size > object_size)
1115 truncate_size = object_size;
1118 osd_req_op_extent_init(req, which, opcode, objoff, objlen,
1119 truncate_size, truncate_seq);
1122 req->r_base_oloc.pool = layout->pool_id;
1123 req->r_base_oloc.pool_ns = ceph_try_get_string(layout->pool_ns);
1124 ceph_oid_printf(&req->r_base_oid, "%llx.%08llx", vino.ino, objnum);
1125 req->r_flags = flags | osdc->client->options->read_from_replica;
1127 req->r_snapid = vino.snap;
1128 if (flags & CEPH_OSD_FLAG_WRITE)
1129 req->r_data_offset = off;
1133 * This is a special case for ceph_writepages_start(), but it
1134 * also covers ceph_uninline_data(). If more multi-op request
1135 * use cases emerge, we will need a separate helper.
1137 r = __ceph_osdc_alloc_messages(req, GFP_NOFS, num_ops, 0);
1139 r = ceph_osdc_alloc_messages(req, GFP_NOFS);
1146 ceph_osdc_put_request(req);
1149 EXPORT_SYMBOL(ceph_osdc_new_request);
1152 * We keep osd requests in an rbtree, sorted by ->r_tid.
1154 DEFINE_RB_FUNCS(request, struct ceph_osd_request, r_tid, r_node)
1155 DEFINE_RB_FUNCS(request_mc, struct ceph_osd_request, r_tid, r_mc_node)
1158 * Call @fn on each OSD request as long as @fn returns 0.
1160 static void for_each_request(struct ceph_osd_client *osdc,
1161 int (*fn)(struct ceph_osd_request *req, void *arg),
1164 struct rb_node *n, *p;
1166 for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
1167 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
1169 for (p = rb_first(&osd->o_requests); p; ) {
1170 struct ceph_osd_request *req =
1171 rb_entry(p, struct ceph_osd_request, r_node);
1179 for (p = rb_first(&osdc->homeless_osd.o_requests); p; ) {
1180 struct ceph_osd_request *req =
1181 rb_entry(p, struct ceph_osd_request, r_node);
1189 static bool osd_homeless(struct ceph_osd *osd)
1191 return osd->o_osd == CEPH_HOMELESS_OSD;
1194 static bool osd_registered(struct ceph_osd *osd)
1196 verify_osdc_locked(osd->o_osdc);
1198 return !RB_EMPTY_NODE(&osd->o_node);
1202 * Assumes @osd is zero-initialized.
1204 static void osd_init(struct ceph_osd *osd)
1206 refcount_set(&osd->o_ref, 1);
1207 RB_CLEAR_NODE(&osd->o_node);
1208 osd->o_requests = RB_ROOT;
1209 osd->o_linger_requests = RB_ROOT;
1210 osd->o_backoff_mappings = RB_ROOT;
1211 osd->o_backoffs_by_id = RB_ROOT;
1212 INIT_LIST_HEAD(&osd->o_osd_lru);
1213 INIT_LIST_HEAD(&osd->o_keepalive_item);
1214 osd->o_incarnation = 1;
1215 mutex_init(&osd->lock);
1218 static void osd_cleanup(struct ceph_osd *osd)
1220 WARN_ON(!RB_EMPTY_NODE(&osd->o_node));
1221 WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests));
1222 WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests));
1223 WARN_ON(!RB_EMPTY_ROOT(&osd->o_backoff_mappings));
1224 WARN_ON(!RB_EMPTY_ROOT(&osd->o_backoffs_by_id));
1225 WARN_ON(!list_empty(&osd->o_osd_lru));
1226 WARN_ON(!list_empty(&osd->o_keepalive_item));
1228 if (osd->o_auth.authorizer) {
1229 WARN_ON(osd_homeless(osd));
1230 ceph_auth_destroy_authorizer(osd->o_auth.authorizer);
1235 * Track open sessions with osds.
1237 static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
1239 struct ceph_osd *osd;
1241 WARN_ON(onum == CEPH_HOMELESS_OSD);
1243 osd = kzalloc(sizeof(*osd), GFP_NOIO | __GFP_NOFAIL);
1248 ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
1253 static struct ceph_osd *get_osd(struct ceph_osd *osd)
1255 if (refcount_inc_not_zero(&osd->o_ref)) {
1256 dout("get_osd %p %d -> %d\n", osd, refcount_read(&osd->o_ref)-1,
1257 refcount_read(&osd->o_ref));
1260 dout("get_osd %p FAIL\n", osd);
1265 static void put_osd(struct ceph_osd *osd)
1267 dout("put_osd %p %d -> %d\n", osd, refcount_read(&osd->o_ref),
1268 refcount_read(&osd->o_ref) - 1);
1269 if (refcount_dec_and_test(&osd->o_ref)) {
1275 DEFINE_RB_FUNCS(osd, struct ceph_osd, o_osd, o_node)
1277 static void __move_osd_to_lru(struct ceph_osd *osd)
1279 struct ceph_osd_client *osdc = osd->o_osdc;
1281 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1282 BUG_ON(!list_empty(&osd->o_osd_lru));
1284 spin_lock(&osdc->osd_lru_lock);
1285 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
1286 spin_unlock(&osdc->osd_lru_lock);
1288 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl;
1291 static void maybe_move_osd_to_lru(struct ceph_osd *osd)
1293 if (RB_EMPTY_ROOT(&osd->o_requests) &&
1294 RB_EMPTY_ROOT(&osd->o_linger_requests))
1295 __move_osd_to_lru(osd);
1298 static void __remove_osd_from_lru(struct ceph_osd *osd)
1300 struct ceph_osd_client *osdc = osd->o_osdc;
1302 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1304 spin_lock(&osdc->osd_lru_lock);
1305 if (!list_empty(&osd->o_osd_lru))
1306 list_del_init(&osd->o_osd_lru);
1307 spin_unlock(&osdc->osd_lru_lock);
1311 * Close the connection and assign any leftover requests to the
1314 static void close_osd(struct ceph_osd *osd)
1316 struct ceph_osd_client *osdc = osd->o_osdc;
1319 verify_osdc_wrlocked(osdc);
1320 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1322 ceph_con_close(&osd->o_con);
1324 for (n = rb_first(&osd->o_requests); n; ) {
1325 struct ceph_osd_request *req =
1326 rb_entry(n, struct ceph_osd_request, r_node);
1328 n = rb_next(n); /* unlink_request() */
1330 dout(" reassigning req %p tid %llu\n", req, req->r_tid);
1331 unlink_request(osd, req);
1332 link_request(&osdc->homeless_osd, req);
1334 for (n = rb_first(&osd->o_linger_requests); n; ) {
1335 struct ceph_osd_linger_request *lreq =
1336 rb_entry(n, struct ceph_osd_linger_request, node);
1338 n = rb_next(n); /* unlink_linger() */
1340 dout(" reassigning lreq %p linger_id %llu\n", lreq,
1342 unlink_linger(osd, lreq);
1343 link_linger(&osdc->homeless_osd, lreq);
1345 clear_backoffs(osd);
1347 __remove_osd_from_lru(osd);
1348 erase_osd(&osdc->osds, osd);
1355 static int reopen_osd(struct ceph_osd *osd)
1357 struct ceph_entity_addr *peer_addr;
1359 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1361 if (RB_EMPTY_ROOT(&osd->o_requests) &&
1362 RB_EMPTY_ROOT(&osd->o_linger_requests)) {
1367 peer_addr = &osd->o_osdc->osdmap->osd_addr[osd->o_osd];
1368 if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1369 !ceph_con_opened(&osd->o_con)) {
1372 dout("osd addr hasn't changed and connection never opened, "
1373 "letting msgr retry\n");
1374 /* touch each r_stamp for handle_timeout()'s benfit */
1375 for (n = rb_first(&osd->o_requests); n; n = rb_next(n)) {
1376 struct ceph_osd_request *req =
1377 rb_entry(n, struct ceph_osd_request, r_node);
1378 req->r_stamp = jiffies;
1384 ceph_con_close(&osd->o_con);
1385 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1386 osd->o_incarnation++;
1391 static struct ceph_osd *lookup_create_osd(struct ceph_osd_client *osdc, int o,
1394 struct ceph_osd *osd;
1397 verify_osdc_wrlocked(osdc);
1399 verify_osdc_locked(osdc);
1401 if (o != CEPH_HOMELESS_OSD)
1402 osd = lookup_osd(&osdc->osds, o);
1404 osd = &osdc->homeless_osd;
1407 return ERR_PTR(-EAGAIN);
1409 osd = create_osd(osdc, o);
1410 insert_osd(&osdc->osds, osd);
1411 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd,
1412 &osdc->osdmap->osd_addr[osd->o_osd]);
1415 dout("%s osdc %p osd%d -> osd %p\n", __func__, osdc, o, osd);
1420 * Create request <-> OSD session relation.
1422 * @req has to be assigned a tid, @osd may be homeless.
1424 static void link_request(struct ceph_osd *osd, struct ceph_osd_request *req)
1426 verify_osd_locked(osd);
1427 WARN_ON(!req->r_tid || req->r_osd);
1428 dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd,
1431 if (!osd_homeless(osd))
1432 __remove_osd_from_lru(osd);
1434 atomic_inc(&osd->o_osdc->num_homeless);
1437 insert_request(&osd->o_requests, req);
1441 static void unlink_request(struct ceph_osd *osd, struct ceph_osd_request *req)
1443 verify_osd_locked(osd);
1444 WARN_ON(req->r_osd != osd);
1445 dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd,
1449 erase_request(&osd->o_requests, req);
1452 if (!osd_homeless(osd))
1453 maybe_move_osd_to_lru(osd);
1455 atomic_dec(&osd->o_osdc->num_homeless);
1458 static bool __pool_full(struct ceph_pg_pool_info *pi)
1460 return pi->flags & CEPH_POOL_FLAG_FULL;
1463 static bool have_pool_full(struct ceph_osd_client *osdc)
1467 for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) {
1468 struct ceph_pg_pool_info *pi =
1469 rb_entry(n, struct ceph_pg_pool_info, node);
1471 if (__pool_full(pi))
1478 static bool pool_full(struct ceph_osd_client *osdc, s64 pool_id)
1480 struct ceph_pg_pool_info *pi;
1482 pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id);
1486 return __pool_full(pi);
1490 * Returns whether a request should be blocked from being sent
1491 * based on the current osdmap and osd_client settings.
1493 static bool target_should_be_paused(struct ceph_osd_client *osdc,
1494 const struct ceph_osd_request_target *t,
1495 struct ceph_pg_pool_info *pi)
1497 bool pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
1498 bool pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
1499 ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
1502 WARN_ON(pi->id != t->target_oloc.pool);
1503 return ((t->flags & CEPH_OSD_FLAG_READ) && pauserd) ||
1504 ((t->flags & CEPH_OSD_FLAG_WRITE) && pausewr) ||
1505 (osdc->osdmap->epoch < osdc->epoch_barrier);
1508 static int pick_random_replica(const struct ceph_osds *acting)
1510 int i = prandom_u32() % acting->size;
1512 dout("%s picked osd%d, primary osd%d\n", __func__,
1513 acting->osds[i], acting->primary);
1518 * Picks the closest replica based on client's location given by
1519 * crush_location option. Prefers the primary if the locality is
1522 static int pick_closest_replica(struct ceph_osd_client *osdc,
1523 const struct ceph_osds *acting)
1525 struct ceph_options *opt = osdc->client->options;
1526 int best_i, best_locality;
1527 int i = 0, locality;
1530 locality = ceph_get_crush_locality(osdc->osdmap,
1534 (locality >= 0 && best_locality < 0) ||
1535 (locality >= 0 && best_locality >= 0 &&
1536 locality < best_locality)) {
1538 best_locality = locality;
1540 } while (++i < acting->size);
1542 dout("%s picked osd%d with locality %d, primary osd%d\n", __func__,
1543 acting->osds[best_i], best_locality, acting->primary);
1547 enum calc_target_result {
1548 CALC_TARGET_NO_ACTION = 0,
1549 CALC_TARGET_NEED_RESEND,
1550 CALC_TARGET_POOL_DNE,
1553 static enum calc_target_result calc_target(struct ceph_osd_client *osdc,
1554 struct ceph_osd_request_target *t,
1557 struct ceph_pg_pool_info *pi;
1558 struct ceph_pg pgid, last_pgid;
1559 struct ceph_osds up, acting;
1560 bool is_read = t->flags & CEPH_OSD_FLAG_READ;
1561 bool is_write = t->flags & CEPH_OSD_FLAG_WRITE;
1562 bool force_resend = false;
1563 bool unpaused = false;
1564 bool legacy_change = false;
1566 bool sort_bitwise = ceph_osdmap_flag(osdc, CEPH_OSDMAP_SORTBITWISE);
1567 bool recovery_deletes = ceph_osdmap_flag(osdc,
1568 CEPH_OSDMAP_RECOVERY_DELETES);
1569 enum calc_target_result ct_res;
1571 t->epoch = osdc->osdmap->epoch;
1572 pi = ceph_pg_pool_by_id(osdc->osdmap, t->base_oloc.pool);
1574 t->osd = CEPH_HOMELESS_OSD;
1575 ct_res = CALC_TARGET_POOL_DNE;
1579 if (osdc->osdmap->epoch == pi->last_force_request_resend) {
1580 if (t->last_force_resend < pi->last_force_request_resend) {
1581 t->last_force_resend = pi->last_force_request_resend;
1582 force_resend = true;
1583 } else if (t->last_force_resend == 0) {
1584 force_resend = true;
1589 ceph_oid_copy(&t->target_oid, &t->base_oid);
1590 ceph_oloc_copy(&t->target_oloc, &t->base_oloc);
1591 if ((t->flags & CEPH_OSD_FLAG_IGNORE_OVERLAY) == 0) {
1592 if (is_read && pi->read_tier >= 0)
1593 t->target_oloc.pool = pi->read_tier;
1594 if (is_write && pi->write_tier >= 0)
1595 t->target_oloc.pool = pi->write_tier;
1597 pi = ceph_pg_pool_by_id(osdc->osdmap, t->target_oloc.pool);
1599 t->osd = CEPH_HOMELESS_OSD;
1600 ct_res = CALC_TARGET_POOL_DNE;
1605 __ceph_object_locator_to_pg(pi, &t->target_oid, &t->target_oloc, &pgid);
1606 last_pgid.pool = pgid.pool;
1607 last_pgid.seed = ceph_stable_mod(pgid.seed, t->pg_num, t->pg_num_mask);
1609 ceph_pg_to_up_acting_osds(osdc->osdmap, pi, &pgid, &up, &acting);
1611 ceph_is_new_interval(&t->acting,
1623 t->recovery_deletes,
1626 force_resend = true;
1628 if (t->paused && !target_should_be_paused(osdc, t, pi)) {
1632 legacy_change = ceph_pg_compare(&t->pgid, &pgid) ||
1633 ceph_osds_changed(&t->acting, &acting,
1634 t->used_replica || any_change);
1636 split = ceph_pg_is_split(&last_pgid, t->pg_num, pi->pg_num);
1638 if (legacy_change || force_resend || split) {
1639 t->pgid = pgid; /* struct */
1640 ceph_pg_to_primary_shard(osdc->osdmap, pi, &pgid, &t->spgid);
1641 ceph_osds_copy(&t->acting, &acting);
1642 ceph_osds_copy(&t->up, &up);
1644 t->min_size = pi->min_size;
1645 t->pg_num = pi->pg_num;
1646 t->pg_num_mask = pi->pg_num_mask;
1647 t->sort_bitwise = sort_bitwise;
1648 t->recovery_deletes = recovery_deletes;
1650 if ((t->flags & (CEPH_OSD_FLAG_BALANCE_READS |
1651 CEPH_OSD_FLAG_LOCALIZE_READS)) &&
1652 !is_write && pi->type == CEPH_POOL_TYPE_REP &&
1656 WARN_ON(!is_read || acting.osds[0] != acting.primary);
1657 if (t->flags & CEPH_OSD_FLAG_BALANCE_READS) {
1658 pos = pick_random_replica(&acting);
1660 pos = pick_closest_replica(osdc, &acting);
1662 t->osd = acting.osds[pos];
1663 t->used_replica = pos > 0;
1665 t->osd = acting.primary;
1666 t->used_replica = false;
1670 if (unpaused || legacy_change || force_resend || split)
1671 ct_res = CALC_TARGET_NEED_RESEND;
1673 ct_res = CALC_TARGET_NO_ACTION;
1676 dout("%s t %p -> %d%d%d%d ct_res %d osd%d\n", __func__, t, unpaused,
1677 legacy_change, force_resend, split, ct_res, t->osd);
1681 static struct ceph_spg_mapping *alloc_spg_mapping(void)
1683 struct ceph_spg_mapping *spg;
1685 spg = kmalloc(sizeof(*spg), GFP_NOIO);
1689 RB_CLEAR_NODE(&spg->node);
1690 spg->backoffs = RB_ROOT;
1694 static void free_spg_mapping(struct ceph_spg_mapping *spg)
1696 WARN_ON(!RB_EMPTY_NODE(&spg->node));
1697 WARN_ON(!RB_EMPTY_ROOT(&spg->backoffs));
1703 * rbtree of ceph_spg_mapping for handling map<spg_t, ...>, similar to
1704 * ceph_pg_mapping. Used to track OSD backoffs -- a backoff [range] is
1705 * defined only within a specific spgid; it does not pass anything to
1706 * children on split, or to another primary.
1708 DEFINE_RB_FUNCS2(spg_mapping, struct ceph_spg_mapping, spgid, ceph_spg_compare,
1709 RB_BYPTR, const struct ceph_spg *, node)
1711 static u64 hoid_get_bitwise_key(const struct ceph_hobject_id *hoid)
1713 return hoid->is_max ? 0x100000000ull : hoid->hash_reverse_bits;
1716 static void hoid_get_effective_key(const struct ceph_hobject_id *hoid,
1717 void **pkey, size_t *pkey_len)
1719 if (hoid->key_len) {
1721 *pkey_len = hoid->key_len;
1724 *pkey_len = hoid->oid_len;
1728 static int compare_names(const void *name1, size_t name1_len,
1729 const void *name2, size_t name2_len)
1733 ret = memcmp(name1, name2, min(name1_len, name2_len));
1735 if (name1_len < name2_len)
1737 else if (name1_len > name2_len)
1743 static int hoid_compare(const struct ceph_hobject_id *lhs,
1744 const struct ceph_hobject_id *rhs)
1746 void *effective_key1, *effective_key2;
1747 size_t effective_key1_len, effective_key2_len;
1750 if (lhs->is_max < rhs->is_max)
1752 if (lhs->is_max > rhs->is_max)
1755 if (lhs->pool < rhs->pool)
1757 if (lhs->pool > rhs->pool)
1760 if (hoid_get_bitwise_key(lhs) < hoid_get_bitwise_key(rhs))
1762 if (hoid_get_bitwise_key(lhs) > hoid_get_bitwise_key(rhs))
1765 ret = compare_names(lhs->nspace, lhs->nspace_len,
1766 rhs->nspace, rhs->nspace_len);
1770 hoid_get_effective_key(lhs, &effective_key1, &effective_key1_len);
1771 hoid_get_effective_key(rhs, &effective_key2, &effective_key2_len);
1772 ret = compare_names(effective_key1, effective_key1_len,
1773 effective_key2, effective_key2_len);
1777 ret = compare_names(lhs->oid, lhs->oid_len, rhs->oid, rhs->oid_len);
1781 if (lhs->snapid < rhs->snapid)
1783 if (lhs->snapid > rhs->snapid)
1790 * For decoding ->begin and ->end of MOSDBackoff only -- no MIN/MAX
1791 * compat stuff here.
1793 * Assumes @hoid is zero-initialized.
1795 static int decode_hoid(void **p, void *end, struct ceph_hobject_id *hoid)
1801 ret = ceph_start_decoding(p, end, 4, "hobject_t", &struct_v,
1807 pr_err("got struct_v %d < 4 of hobject_t\n", struct_v);
1811 hoid->key = ceph_extract_encoded_string(p, end, &hoid->key_len,
1813 if (IS_ERR(hoid->key)) {
1814 ret = PTR_ERR(hoid->key);
1819 hoid->oid = ceph_extract_encoded_string(p, end, &hoid->oid_len,
1821 if (IS_ERR(hoid->oid)) {
1822 ret = PTR_ERR(hoid->oid);
1827 ceph_decode_64_safe(p, end, hoid->snapid, e_inval);
1828 ceph_decode_32_safe(p, end, hoid->hash, e_inval);
1829 ceph_decode_8_safe(p, end, hoid->is_max, e_inval);
1831 hoid->nspace = ceph_extract_encoded_string(p, end, &hoid->nspace_len,
1833 if (IS_ERR(hoid->nspace)) {
1834 ret = PTR_ERR(hoid->nspace);
1835 hoid->nspace = NULL;
1839 ceph_decode_64_safe(p, end, hoid->pool, e_inval);
1841 ceph_hoid_build_hash_cache(hoid);
1848 static int hoid_encoding_size(const struct ceph_hobject_id *hoid)
1850 return 8 + 4 + 1 + 8 + /* snapid, hash, is_max, pool */
1851 4 + hoid->key_len + 4 + hoid->oid_len + 4 + hoid->nspace_len;
1854 static void encode_hoid(void **p, void *end, const struct ceph_hobject_id *hoid)
1856 ceph_start_encoding(p, 4, 3, hoid_encoding_size(hoid));
1857 ceph_encode_string(p, end, hoid->key, hoid->key_len);
1858 ceph_encode_string(p, end, hoid->oid, hoid->oid_len);
1859 ceph_encode_64(p, hoid->snapid);
1860 ceph_encode_32(p, hoid->hash);
1861 ceph_encode_8(p, hoid->is_max);
1862 ceph_encode_string(p, end, hoid->nspace, hoid->nspace_len);
1863 ceph_encode_64(p, hoid->pool);
1866 static void free_hoid(struct ceph_hobject_id *hoid)
1871 kfree(hoid->nspace);
1876 static struct ceph_osd_backoff *alloc_backoff(void)
1878 struct ceph_osd_backoff *backoff;
1880 backoff = kzalloc(sizeof(*backoff), GFP_NOIO);
1884 RB_CLEAR_NODE(&backoff->spg_node);
1885 RB_CLEAR_NODE(&backoff->id_node);
1889 static void free_backoff(struct ceph_osd_backoff *backoff)
1891 WARN_ON(!RB_EMPTY_NODE(&backoff->spg_node));
1892 WARN_ON(!RB_EMPTY_NODE(&backoff->id_node));
1894 free_hoid(backoff->begin);
1895 free_hoid(backoff->end);
1900 * Within a specific spgid, backoffs are managed by ->begin hoid.
1902 DEFINE_RB_INSDEL_FUNCS2(backoff, struct ceph_osd_backoff, begin, hoid_compare,
1903 RB_BYVAL, spg_node);
1905 static struct ceph_osd_backoff *lookup_containing_backoff(struct rb_root *root,
1906 const struct ceph_hobject_id *hoid)
1908 struct rb_node *n = root->rb_node;
1911 struct ceph_osd_backoff *cur =
1912 rb_entry(n, struct ceph_osd_backoff, spg_node);
1915 cmp = hoid_compare(hoid, cur->begin);
1918 } else if (cmp > 0) {
1919 if (hoid_compare(hoid, cur->end) < 0)
1932 * Each backoff has a unique id within its OSD session.
1934 DEFINE_RB_FUNCS(backoff_by_id, struct ceph_osd_backoff, id, id_node)
1936 static void clear_backoffs(struct ceph_osd *osd)
1938 while (!RB_EMPTY_ROOT(&osd->o_backoff_mappings)) {
1939 struct ceph_spg_mapping *spg =
1940 rb_entry(rb_first(&osd->o_backoff_mappings),
1941 struct ceph_spg_mapping, node);
1943 while (!RB_EMPTY_ROOT(&spg->backoffs)) {
1944 struct ceph_osd_backoff *backoff =
1945 rb_entry(rb_first(&spg->backoffs),
1946 struct ceph_osd_backoff, spg_node);
1948 erase_backoff(&spg->backoffs, backoff);
1949 erase_backoff_by_id(&osd->o_backoffs_by_id, backoff);
1950 free_backoff(backoff);
1952 erase_spg_mapping(&osd->o_backoff_mappings, spg);
1953 free_spg_mapping(spg);
1958 * Set up a temporary, non-owning view into @t.
1960 static void hoid_fill_from_target(struct ceph_hobject_id *hoid,
1961 const struct ceph_osd_request_target *t)
1965 hoid->oid = t->target_oid.name;
1966 hoid->oid_len = t->target_oid.name_len;
1967 hoid->snapid = CEPH_NOSNAP;
1968 hoid->hash = t->pgid.seed;
1969 hoid->is_max = false;
1970 if (t->target_oloc.pool_ns) {
1971 hoid->nspace = t->target_oloc.pool_ns->str;
1972 hoid->nspace_len = t->target_oloc.pool_ns->len;
1974 hoid->nspace = NULL;
1975 hoid->nspace_len = 0;
1977 hoid->pool = t->target_oloc.pool;
1978 ceph_hoid_build_hash_cache(hoid);
1981 static bool should_plug_request(struct ceph_osd_request *req)
1983 struct ceph_osd *osd = req->r_osd;
1984 struct ceph_spg_mapping *spg;
1985 struct ceph_osd_backoff *backoff;
1986 struct ceph_hobject_id hoid;
1988 spg = lookup_spg_mapping(&osd->o_backoff_mappings, &req->r_t.spgid);
1992 hoid_fill_from_target(&hoid, &req->r_t);
1993 backoff = lookup_containing_backoff(&spg->backoffs, &hoid);
1997 dout("%s req %p tid %llu backoff osd%d spgid %llu.%xs%d id %llu\n",
1998 __func__, req, req->r_tid, osd->o_osd, backoff->spgid.pgid.pool,
1999 backoff->spgid.pgid.seed, backoff->spgid.shard, backoff->id);
2004 * Keep get_num_data_items() in sync with this function.
2006 static void setup_request_data(struct ceph_osd_request *req)
2008 struct ceph_msg *request_msg = req->r_request;
2009 struct ceph_msg *reply_msg = req->r_reply;
2010 struct ceph_osd_req_op *op;
2012 if (req->r_request->num_data_items || req->r_reply->num_data_items)
2015 WARN_ON(request_msg->data_length || reply_msg->data_length);
2016 for (op = req->r_ops; op != &req->r_ops[req->r_num_ops]; op++) {
2019 case CEPH_OSD_OP_WRITE:
2020 case CEPH_OSD_OP_WRITEFULL:
2021 WARN_ON(op->indata_len != op->extent.length);
2022 ceph_osdc_msg_data_add(request_msg,
2023 &op->extent.osd_data);
2025 case CEPH_OSD_OP_SETXATTR:
2026 case CEPH_OSD_OP_CMPXATTR:
2027 WARN_ON(op->indata_len != op->xattr.name_len +
2028 op->xattr.value_len);
2029 ceph_osdc_msg_data_add(request_msg,
2030 &op->xattr.osd_data);
2032 case CEPH_OSD_OP_NOTIFY_ACK:
2033 ceph_osdc_msg_data_add(request_msg,
2034 &op->notify_ack.request_data);
2036 case CEPH_OSD_OP_COPY_FROM2:
2037 ceph_osdc_msg_data_add(request_msg,
2038 &op->copy_from.osd_data);
2042 case CEPH_OSD_OP_STAT:
2043 ceph_osdc_msg_data_add(reply_msg,
2046 case CEPH_OSD_OP_READ:
2047 ceph_osdc_msg_data_add(reply_msg,
2048 &op->extent.osd_data);
2050 case CEPH_OSD_OP_LIST_WATCHERS:
2051 ceph_osdc_msg_data_add(reply_msg,
2052 &op->list_watchers.response_data);
2056 case CEPH_OSD_OP_CALL:
2057 WARN_ON(op->indata_len != op->cls.class_len +
2058 op->cls.method_len +
2059 op->cls.indata_len);
2060 ceph_osdc_msg_data_add(request_msg,
2061 &op->cls.request_info);
2062 /* optional, can be NONE */
2063 ceph_osdc_msg_data_add(request_msg,
2064 &op->cls.request_data);
2065 /* optional, can be NONE */
2066 ceph_osdc_msg_data_add(reply_msg,
2067 &op->cls.response_data);
2069 case CEPH_OSD_OP_NOTIFY:
2070 ceph_osdc_msg_data_add(request_msg,
2071 &op->notify.request_data);
2072 ceph_osdc_msg_data_add(reply_msg,
2073 &op->notify.response_data);
2079 static void encode_pgid(void **p, const struct ceph_pg *pgid)
2081 ceph_encode_8(p, 1);
2082 ceph_encode_64(p, pgid->pool);
2083 ceph_encode_32(p, pgid->seed);
2084 ceph_encode_32(p, -1); /* preferred */
2087 static void encode_spgid(void **p, const struct ceph_spg *spgid)
2089 ceph_start_encoding(p, 1, 1, CEPH_PGID_ENCODING_LEN + 1);
2090 encode_pgid(p, &spgid->pgid);
2091 ceph_encode_8(p, spgid->shard);
2094 static void encode_oloc(void **p, void *end,
2095 const struct ceph_object_locator *oloc)
2097 ceph_start_encoding(p, 5, 4, ceph_oloc_encoding_size(oloc));
2098 ceph_encode_64(p, oloc->pool);
2099 ceph_encode_32(p, -1); /* preferred */
2100 ceph_encode_32(p, 0); /* key len */
2102 ceph_encode_string(p, end, oloc->pool_ns->str,
2103 oloc->pool_ns->len);
2105 ceph_encode_32(p, 0);
2108 static void encode_request_partial(struct ceph_osd_request *req,
2109 struct ceph_msg *msg)
2111 void *p = msg->front.iov_base;
2112 void *const end = p + msg->front_alloc_len;
2116 if (req->r_flags & CEPH_OSD_FLAG_WRITE) {
2117 /* snapshots aren't writeable */
2118 WARN_ON(req->r_snapid != CEPH_NOSNAP);
2120 WARN_ON(req->r_mtime.tv_sec || req->r_mtime.tv_nsec ||
2121 req->r_data_offset || req->r_snapc);
2124 setup_request_data(req);
2126 encode_spgid(&p, &req->r_t.spgid); /* actual spg */
2127 ceph_encode_32(&p, req->r_t.pgid.seed); /* raw hash */
2128 ceph_encode_32(&p, req->r_osdc->osdmap->epoch);
2129 ceph_encode_32(&p, req->r_flags);
2132 ceph_start_encoding(&p, 2, 2, sizeof(struct ceph_osd_reqid));
2133 memset(p, 0, sizeof(struct ceph_osd_reqid));
2134 p += sizeof(struct ceph_osd_reqid);
2137 memset(p, 0, sizeof(struct ceph_blkin_trace_info));
2138 p += sizeof(struct ceph_blkin_trace_info);
2140 ceph_encode_32(&p, 0); /* client_inc, always 0 */
2141 ceph_encode_timespec64(p, &req->r_mtime);
2142 p += sizeof(struct ceph_timespec);
2144 encode_oloc(&p, end, &req->r_t.target_oloc);
2145 ceph_encode_string(&p, end, req->r_t.target_oid.name,
2146 req->r_t.target_oid.name_len);
2148 /* ops, can imply data */
2149 ceph_encode_16(&p, req->r_num_ops);
2150 for (i = 0; i < req->r_num_ops; i++) {
2151 data_len += osd_req_encode_op(p, &req->r_ops[i]);
2152 p += sizeof(struct ceph_osd_op);
2155 ceph_encode_64(&p, req->r_snapid); /* snapid */
2157 ceph_encode_64(&p, req->r_snapc->seq);
2158 ceph_encode_32(&p, req->r_snapc->num_snaps);
2159 for (i = 0; i < req->r_snapc->num_snaps; i++)
2160 ceph_encode_64(&p, req->r_snapc->snaps[i]);
2162 ceph_encode_64(&p, 0); /* snap_seq */
2163 ceph_encode_32(&p, 0); /* snaps len */
2166 ceph_encode_32(&p, req->r_attempts); /* retry_attempt */
2167 BUG_ON(p > end - 8); /* space for features */
2169 msg->hdr.version = cpu_to_le16(8); /* MOSDOp v8 */
2170 /* front_len is finalized in encode_request_finish() */
2171 msg->front.iov_len = p - msg->front.iov_base;
2172 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2173 msg->hdr.data_len = cpu_to_le32(data_len);
2175 * The header "data_off" is a hint to the receiver allowing it
2176 * to align received data into its buffers such that there's no
2177 * need to re-copy it before writing it to disk (direct I/O).
2179 msg->hdr.data_off = cpu_to_le16(req->r_data_offset);
2181 dout("%s req %p msg %p oid %s oid_len %d\n", __func__, req, msg,
2182 req->r_t.target_oid.name, req->r_t.target_oid.name_len);
2185 static void encode_request_finish(struct ceph_msg *msg)
2187 void *p = msg->front.iov_base;
2188 void *const partial_end = p + msg->front.iov_len;
2189 void *const end = p + msg->front_alloc_len;
2191 if (CEPH_HAVE_FEATURE(msg->con->peer_features, RESEND_ON_SPLIT)) {
2192 /* luminous OSD -- encode features and be done */
2194 ceph_encode_64(&p, msg->con->peer_features);
2197 char spgid[CEPH_ENCODING_START_BLK_LEN +
2198 CEPH_PGID_ENCODING_LEN + 1];
2202 char reqid[CEPH_ENCODING_START_BLK_LEN +
2203 sizeof(struct ceph_osd_reqid)];
2204 char trace[sizeof(struct ceph_blkin_trace_info)];
2206 struct ceph_timespec mtime;
2208 struct ceph_pg pgid;
2209 void *oloc, *oid, *tail;
2210 int oloc_len, oid_len, tail_len;
2214 * Pre-luminous OSD -- reencode v8 into v4 using @head
2215 * as a temporary buffer. Encode the raw PG; the rest
2216 * is just a matter of moving oloc, oid and tail blobs
2219 memcpy(&head, p, sizeof(head));
2223 p += CEPH_ENCODING_START_BLK_LEN;
2224 pgid.pool = ceph_decode_64(&p);
2225 p += 4 + 4; /* preferred, key len */
2226 len = ceph_decode_32(&p);
2227 p += len; /* nspace */
2228 oloc_len = p - oloc;
2231 len = ceph_decode_32(&p);
2236 tail_len = partial_end - p;
2238 p = msg->front.iov_base;
2239 ceph_encode_copy(&p, &head.client_inc, sizeof(head.client_inc));
2240 ceph_encode_copy(&p, &head.epoch, sizeof(head.epoch));
2241 ceph_encode_copy(&p, &head.flags, sizeof(head.flags));
2242 ceph_encode_copy(&p, &head.mtime, sizeof(head.mtime));
2244 /* reassert_version */
2245 memset(p, 0, sizeof(struct ceph_eversion));
2246 p += sizeof(struct ceph_eversion);
2249 memmove(p, oloc, oloc_len);
2252 pgid.seed = le32_to_cpu(head.hash);
2253 encode_pgid(&p, &pgid); /* raw pg */
2256 memmove(p, oid, oid_len);
2259 /* tail -- ops, snapid, snapc, retry_attempt */
2261 memmove(p, tail, tail_len);
2264 msg->hdr.version = cpu_to_le16(4); /* MOSDOp v4 */
2268 msg->front.iov_len = p - msg->front.iov_base;
2269 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2271 dout("%s msg %p tid %llu %u+%u+%u v%d\n", __func__, msg,
2272 le64_to_cpu(msg->hdr.tid), le32_to_cpu(msg->hdr.front_len),
2273 le32_to_cpu(msg->hdr.middle_len), le32_to_cpu(msg->hdr.data_len),
2274 le16_to_cpu(msg->hdr.version));
2278 * @req has to be assigned a tid and registered.
2280 static void send_request(struct ceph_osd_request *req)
2282 struct ceph_osd *osd = req->r_osd;
2284 verify_osd_locked(osd);
2285 WARN_ON(osd->o_osd != req->r_t.osd);
2288 if (should_plug_request(req))
2292 * We may have a previously queued request message hanging
2293 * around. Cancel it to avoid corrupting the msgr.
2296 ceph_msg_revoke(req->r_request);
2298 req->r_flags |= CEPH_OSD_FLAG_KNOWN_REDIR;
2299 if (req->r_attempts)
2300 req->r_flags |= CEPH_OSD_FLAG_RETRY;
2302 WARN_ON(req->r_flags & CEPH_OSD_FLAG_RETRY);
2304 encode_request_partial(req, req->r_request);
2306 dout("%s req %p tid %llu to pgid %llu.%x spgid %llu.%xs%d osd%d e%u flags 0x%x attempt %d\n",
2307 __func__, req, req->r_tid, req->r_t.pgid.pool, req->r_t.pgid.seed,
2308 req->r_t.spgid.pgid.pool, req->r_t.spgid.pgid.seed,
2309 req->r_t.spgid.shard, osd->o_osd, req->r_t.epoch, req->r_flags,
2312 req->r_t.paused = false;
2313 req->r_stamp = jiffies;
2316 req->r_sent = osd->o_incarnation;
2317 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
2318 ceph_con_send(&osd->o_con, ceph_msg_get(req->r_request));
2321 static void maybe_request_map(struct ceph_osd_client *osdc)
2323 bool continuous = false;
2325 verify_osdc_locked(osdc);
2326 WARN_ON(!osdc->osdmap->epoch);
2328 if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
2329 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD) ||
2330 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) {
2331 dout("%s osdc %p continuous\n", __func__, osdc);
2334 dout("%s osdc %p onetime\n", __func__, osdc);
2337 if (ceph_monc_want_map(&osdc->client->monc, CEPH_SUB_OSDMAP,
2338 osdc->osdmap->epoch + 1, continuous))
2339 ceph_monc_renew_subs(&osdc->client->monc);
2342 static void complete_request(struct ceph_osd_request *req, int err);
2343 static void send_map_check(struct ceph_osd_request *req);
2345 static void __submit_request(struct ceph_osd_request *req, bool wrlocked)
2347 struct ceph_osd_client *osdc = req->r_osdc;
2348 struct ceph_osd *osd;
2349 enum calc_target_result ct_res;
2351 bool need_send = false;
2352 bool promoted = false;
2354 WARN_ON(req->r_tid);
2355 dout("%s req %p wrlocked %d\n", __func__, req, wrlocked);
2358 ct_res = calc_target(osdc, &req->r_t, false);
2359 if (ct_res == CALC_TARGET_POOL_DNE && !wrlocked)
2362 osd = lookup_create_osd(osdc, req->r_t.osd, wrlocked);
2364 WARN_ON(PTR_ERR(osd) != -EAGAIN || wrlocked);
2368 if (osdc->abort_err) {
2369 dout("req %p abort_err %d\n", req, osdc->abort_err);
2370 err = osdc->abort_err;
2371 } else if (osdc->osdmap->epoch < osdc->epoch_barrier) {
2372 dout("req %p epoch %u barrier %u\n", req, osdc->osdmap->epoch,
2373 osdc->epoch_barrier);
2374 req->r_t.paused = true;
2375 maybe_request_map(osdc);
2376 } else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
2377 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) {
2378 dout("req %p pausewr\n", req);
2379 req->r_t.paused = true;
2380 maybe_request_map(osdc);
2381 } else if ((req->r_flags & CEPH_OSD_FLAG_READ) &&
2382 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) {
2383 dout("req %p pauserd\n", req);
2384 req->r_t.paused = true;
2385 maybe_request_map(osdc);
2386 } else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
2387 !(req->r_flags & (CEPH_OSD_FLAG_FULL_TRY |
2388 CEPH_OSD_FLAG_FULL_FORCE)) &&
2389 (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
2390 pool_full(osdc, req->r_t.base_oloc.pool))) {
2391 dout("req %p full/pool_full\n", req);
2392 if (ceph_test_opt(osdc->client, ABORT_ON_FULL)) {
2395 pr_warn_ratelimited("FULL or reached pool quota\n");
2396 req->r_t.paused = true;
2397 maybe_request_map(osdc);
2399 } else if (!osd_homeless(osd)) {
2402 maybe_request_map(osdc);
2405 mutex_lock(&osd->lock);
2407 * Assign the tid atomically with send_request() to protect
2408 * multiple writes to the same object from racing with each
2409 * other, resulting in out of order ops on the OSDs.
2411 req->r_tid = atomic64_inc_return(&osdc->last_tid);
2412 link_request(osd, req);
2416 complete_request(req, err);
2417 mutex_unlock(&osd->lock);
2419 if (!err && ct_res == CALC_TARGET_POOL_DNE)
2420 send_map_check(req);
2423 downgrade_write(&osdc->lock);
2427 up_read(&osdc->lock);
2428 down_write(&osdc->lock);
2434 static void account_request(struct ceph_osd_request *req)
2436 WARN_ON(req->r_flags & (CEPH_OSD_FLAG_ACK | CEPH_OSD_FLAG_ONDISK));
2437 WARN_ON(!(req->r_flags & (CEPH_OSD_FLAG_READ | CEPH_OSD_FLAG_WRITE)));
2439 req->r_flags |= CEPH_OSD_FLAG_ONDISK;
2440 atomic_inc(&req->r_osdc->num_requests);
2442 req->r_start_stamp = jiffies;
2443 req->r_start_latency = ktime_get();
2446 static void submit_request(struct ceph_osd_request *req, bool wrlocked)
2448 ceph_osdc_get_request(req);
2449 account_request(req);
2450 __submit_request(req, wrlocked);
2453 static void finish_request(struct ceph_osd_request *req)
2455 struct ceph_osd_client *osdc = req->r_osdc;
2457 WARN_ON(lookup_request_mc(&osdc->map_checks, req->r_tid));
2458 dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
2460 req->r_end_latency = ktime_get();
2463 unlink_request(req->r_osd, req);
2464 atomic_dec(&osdc->num_requests);
2467 * If an OSD has failed or returned and a request has been sent
2468 * twice, it's possible to get a reply and end up here while the
2469 * request message is queued for delivery. We will ignore the
2470 * reply, so not a big deal, but better to try and catch it.
2472 ceph_msg_revoke(req->r_request);
2473 ceph_msg_revoke_incoming(req->r_reply);
2476 static void __complete_request(struct ceph_osd_request *req)
2478 dout("%s req %p tid %llu cb %ps result %d\n", __func__, req,
2479 req->r_tid, req->r_callback, req->r_result);
2481 if (req->r_callback)
2482 req->r_callback(req);
2483 complete_all(&req->r_completion);
2484 ceph_osdc_put_request(req);
2487 static void complete_request_workfn(struct work_struct *work)
2489 struct ceph_osd_request *req =
2490 container_of(work, struct ceph_osd_request, r_complete_work);
2492 __complete_request(req);
2496 * This is open-coded in handle_reply().
2498 static void complete_request(struct ceph_osd_request *req, int err)
2500 dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err);
2502 req->r_result = err;
2503 finish_request(req);
2505 INIT_WORK(&req->r_complete_work, complete_request_workfn);
2506 queue_work(req->r_osdc->completion_wq, &req->r_complete_work);
2509 static void cancel_map_check(struct ceph_osd_request *req)
2511 struct ceph_osd_client *osdc = req->r_osdc;
2512 struct ceph_osd_request *lookup_req;
2514 verify_osdc_wrlocked(osdc);
2516 lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid);
2520 WARN_ON(lookup_req != req);
2521 erase_request_mc(&osdc->map_checks, req);
2522 ceph_osdc_put_request(req);
2525 static void cancel_request(struct ceph_osd_request *req)
2527 dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
2529 cancel_map_check(req);
2530 finish_request(req);
2531 complete_all(&req->r_completion);
2532 ceph_osdc_put_request(req);
2535 static void abort_request(struct ceph_osd_request *req, int err)
2537 dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err);
2539 cancel_map_check(req);
2540 complete_request(req, err);
2543 static int abort_fn(struct ceph_osd_request *req, void *arg)
2545 int err = *(int *)arg;
2547 abort_request(req, err);
2548 return 0; /* continue iteration */
2552 * Abort all in-flight requests with @err and arrange for all future
2553 * requests to be failed immediately.
2555 void ceph_osdc_abort_requests(struct ceph_osd_client *osdc, int err)
2557 dout("%s osdc %p err %d\n", __func__, osdc, err);
2558 down_write(&osdc->lock);
2559 for_each_request(osdc, abort_fn, &err);
2560 osdc->abort_err = err;
2561 up_write(&osdc->lock);
2563 EXPORT_SYMBOL(ceph_osdc_abort_requests);
2565 void ceph_osdc_clear_abort_err(struct ceph_osd_client *osdc)
2567 down_write(&osdc->lock);
2568 osdc->abort_err = 0;
2569 up_write(&osdc->lock);
2571 EXPORT_SYMBOL(ceph_osdc_clear_abort_err);
2573 static void update_epoch_barrier(struct ceph_osd_client *osdc, u32 eb)
2575 if (likely(eb > osdc->epoch_barrier)) {
2576 dout("updating epoch_barrier from %u to %u\n",
2577 osdc->epoch_barrier, eb);
2578 osdc->epoch_barrier = eb;
2579 /* Request map if we're not to the barrier yet */
2580 if (eb > osdc->osdmap->epoch)
2581 maybe_request_map(osdc);
2585 void ceph_osdc_update_epoch_barrier(struct ceph_osd_client *osdc, u32 eb)
2587 down_read(&osdc->lock);
2588 if (unlikely(eb > osdc->epoch_barrier)) {
2589 up_read(&osdc->lock);
2590 down_write(&osdc->lock);
2591 update_epoch_barrier(osdc, eb);
2592 up_write(&osdc->lock);
2594 up_read(&osdc->lock);
2597 EXPORT_SYMBOL(ceph_osdc_update_epoch_barrier);
2600 * We can end up releasing caps as a result of abort_request().
2601 * In that case, we probably want to ensure that the cap release message
2602 * has an updated epoch barrier in it, so set the epoch barrier prior to
2603 * aborting the first request.
2605 static int abort_on_full_fn(struct ceph_osd_request *req, void *arg)
2607 struct ceph_osd_client *osdc = req->r_osdc;
2608 bool *victims = arg;
2610 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
2611 (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
2612 pool_full(osdc, req->r_t.base_oloc.pool))) {
2614 update_epoch_barrier(osdc, osdc->osdmap->epoch);
2617 abort_request(req, -ENOSPC);
2620 return 0; /* continue iteration */
2624 * Drop all pending requests that are stalled waiting on a full condition to
2625 * clear, and complete them with ENOSPC as the return code. Set the
2626 * osdc->epoch_barrier to the latest map epoch that we've seen if any were
2629 static void ceph_osdc_abort_on_full(struct ceph_osd_client *osdc)
2631 bool victims = false;
2633 if (ceph_test_opt(osdc->client, ABORT_ON_FULL) &&
2634 (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || have_pool_full(osdc)))
2635 for_each_request(osdc, abort_on_full_fn, &victims);
2638 static void check_pool_dne(struct ceph_osd_request *req)
2640 struct ceph_osd_client *osdc = req->r_osdc;
2641 struct ceph_osdmap *map = osdc->osdmap;
2643 verify_osdc_wrlocked(osdc);
2644 WARN_ON(!map->epoch);
2646 if (req->r_attempts) {
2648 * We sent a request earlier, which means that
2649 * previously the pool existed, and now it does not
2650 * (i.e., it was deleted).
2652 req->r_map_dne_bound = map->epoch;
2653 dout("%s req %p tid %llu pool disappeared\n", __func__, req,
2656 dout("%s req %p tid %llu map_dne_bound %u have %u\n", __func__,
2657 req, req->r_tid, req->r_map_dne_bound, map->epoch);
2660 if (req->r_map_dne_bound) {
2661 if (map->epoch >= req->r_map_dne_bound) {
2662 /* we had a new enough map */
2663 pr_info_ratelimited("tid %llu pool does not exist\n",
2665 complete_request(req, -ENOENT);
2668 send_map_check(req);
2672 static void map_check_cb(struct ceph_mon_generic_request *greq)
2674 struct ceph_osd_client *osdc = &greq->monc->client->osdc;
2675 struct ceph_osd_request *req;
2676 u64 tid = greq->private_data;
2678 WARN_ON(greq->result || !greq->u.newest);
2680 down_write(&osdc->lock);
2681 req = lookup_request_mc(&osdc->map_checks, tid);
2683 dout("%s tid %llu dne\n", __func__, tid);
2687 dout("%s req %p tid %llu map_dne_bound %u newest %llu\n", __func__,
2688 req, req->r_tid, req->r_map_dne_bound, greq->u.newest);
2689 if (!req->r_map_dne_bound)
2690 req->r_map_dne_bound = greq->u.newest;
2691 erase_request_mc(&osdc->map_checks, req);
2692 check_pool_dne(req);
2694 ceph_osdc_put_request(req);
2696 up_write(&osdc->lock);
2699 static void send_map_check(struct ceph_osd_request *req)
2701 struct ceph_osd_client *osdc = req->r_osdc;
2702 struct ceph_osd_request *lookup_req;
2705 verify_osdc_wrlocked(osdc);
2707 lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid);
2709 WARN_ON(lookup_req != req);
2713 ceph_osdc_get_request(req);
2714 insert_request_mc(&osdc->map_checks, req);
2715 ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap",
2716 map_check_cb, req->r_tid);
2721 * lingering requests, watch/notify v2 infrastructure
2723 static void linger_release(struct kref *kref)
2725 struct ceph_osd_linger_request *lreq =
2726 container_of(kref, struct ceph_osd_linger_request, kref);
2728 dout("%s lreq %p reg_req %p ping_req %p\n", __func__, lreq,
2729 lreq->reg_req, lreq->ping_req);
2730 WARN_ON(!RB_EMPTY_NODE(&lreq->node));
2731 WARN_ON(!RB_EMPTY_NODE(&lreq->osdc_node));
2732 WARN_ON(!RB_EMPTY_NODE(&lreq->mc_node));
2733 WARN_ON(!list_empty(&lreq->scan_item));
2734 WARN_ON(!list_empty(&lreq->pending_lworks));
2738 ceph_osdc_put_request(lreq->reg_req);
2740 ceph_osdc_put_request(lreq->ping_req);
2741 target_destroy(&lreq->t);
2745 static void linger_put(struct ceph_osd_linger_request *lreq)
2748 kref_put(&lreq->kref, linger_release);
2751 static struct ceph_osd_linger_request *
2752 linger_get(struct ceph_osd_linger_request *lreq)
2754 kref_get(&lreq->kref);
2758 static struct ceph_osd_linger_request *
2759 linger_alloc(struct ceph_osd_client *osdc)
2761 struct ceph_osd_linger_request *lreq;
2763 lreq = kzalloc(sizeof(*lreq), GFP_NOIO);
2767 kref_init(&lreq->kref);
2768 mutex_init(&lreq->lock);
2769 RB_CLEAR_NODE(&lreq->node);
2770 RB_CLEAR_NODE(&lreq->osdc_node);
2771 RB_CLEAR_NODE(&lreq->mc_node);
2772 INIT_LIST_HEAD(&lreq->scan_item);
2773 INIT_LIST_HEAD(&lreq->pending_lworks);
2774 init_completion(&lreq->reg_commit_wait);
2775 init_completion(&lreq->notify_finish_wait);
2778 target_init(&lreq->t);
2780 dout("%s lreq %p\n", __func__, lreq);
2784 DEFINE_RB_INSDEL_FUNCS(linger, struct ceph_osd_linger_request, linger_id, node)
2785 DEFINE_RB_FUNCS(linger_osdc, struct ceph_osd_linger_request, linger_id, osdc_node)
2786 DEFINE_RB_FUNCS(linger_mc, struct ceph_osd_linger_request, linger_id, mc_node)
2789 * Create linger request <-> OSD session relation.
2791 * @lreq has to be registered, @osd may be homeless.
2793 static void link_linger(struct ceph_osd *osd,
2794 struct ceph_osd_linger_request *lreq)
2796 verify_osd_locked(osd);
2797 WARN_ON(!lreq->linger_id || lreq->osd);
2798 dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd,
2799 osd->o_osd, lreq, lreq->linger_id);
2801 if (!osd_homeless(osd))
2802 __remove_osd_from_lru(osd);
2804 atomic_inc(&osd->o_osdc->num_homeless);
2807 insert_linger(&osd->o_linger_requests, lreq);
2811 static void unlink_linger(struct ceph_osd *osd,
2812 struct ceph_osd_linger_request *lreq)
2814 verify_osd_locked(osd);
2815 WARN_ON(lreq->osd != osd);
2816 dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd,
2817 osd->o_osd, lreq, lreq->linger_id);
2820 erase_linger(&osd->o_linger_requests, lreq);
2823 if (!osd_homeless(osd))
2824 maybe_move_osd_to_lru(osd);
2826 atomic_dec(&osd->o_osdc->num_homeless);
2829 static bool __linger_registered(struct ceph_osd_linger_request *lreq)
2831 verify_osdc_locked(lreq->osdc);
2833 return !RB_EMPTY_NODE(&lreq->osdc_node);
2836 static bool linger_registered(struct ceph_osd_linger_request *lreq)
2838 struct ceph_osd_client *osdc = lreq->osdc;
2841 down_read(&osdc->lock);
2842 registered = __linger_registered(lreq);
2843 up_read(&osdc->lock);
2848 static void linger_register(struct ceph_osd_linger_request *lreq)
2850 struct ceph_osd_client *osdc = lreq->osdc;
2852 verify_osdc_wrlocked(osdc);
2853 WARN_ON(lreq->linger_id);
2856 lreq->linger_id = ++osdc->last_linger_id;
2857 insert_linger_osdc(&osdc->linger_requests, lreq);
2860 static void linger_unregister(struct ceph_osd_linger_request *lreq)
2862 struct ceph_osd_client *osdc = lreq->osdc;
2864 verify_osdc_wrlocked(osdc);
2866 erase_linger_osdc(&osdc->linger_requests, lreq);
2870 static void cancel_linger_request(struct ceph_osd_request *req)
2872 struct ceph_osd_linger_request *lreq = req->r_priv;
2874 WARN_ON(!req->r_linger);
2875 cancel_request(req);
2879 struct linger_work {
2880 struct work_struct work;
2881 struct ceph_osd_linger_request *lreq;
2882 struct list_head pending_item;
2883 unsigned long queued_stamp;
2889 void *payload; /* points into @msg front */
2892 struct ceph_msg *msg; /* for ceph_msg_put() */
2900 static struct linger_work *lwork_alloc(struct ceph_osd_linger_request *lreq,
2903 struct linger_work *lwork;
2905 lwork = kzalloc(sizeof(*lwork), GFP_NOIO);
2909 INIT_WORK(&lwork->work, workfn);
2910 INIT_LIST_HEAD(&lwork->pending_item);
2911 lwork->lreq = linger_get(lreq);
2916 static void lwork_free(struct linger_work *lwork)
2918 struct ceph_osd_linger_request *lreq = lwork->lreq;
2920 mutex_lock(&lreq->lock);
2921 list_del(&lwork->pending_item);
2922 mutex_unlock(&lreq->lock);
2928 static void lwork_queue(struct linger_work *lwork)
2930 struct ceph_osd_linger_request *lreq = lwork->lreq;
2931 struct ceph_osd_client *osdc = lreq->osdc;
2933 verify_lreq_locked(lreq);
2934 WARN_ON(!list_empty(&lwork->pending_item));
2936 lwork->queued_stamp = jiffies;
2937 list_add_tail(&lwork->pending_item, &lreq->pending_lworks);
2938 queue_work(osdc->notify_wq, &lwork->work);
2941 static void do_watch_notify(struct work_struct *w)
2943 struct linger_work *lwork = container_of(w, struct linger_work, work);
2944 struct ceph_osd_linger_request *lreq = lwork->lreq;
2946 if (!linger_registered(lreq)) {
2947 dout("%s lreq %p not registered\n", __func__, lreq);
2951 WARN_ON(!lreq->is_watch);
2952 dout("%s lreq %p notify_id %llu notifier_id %llu payload_len %zu\n",
2953 __func__, lreq, lwork->notify.notify_id, lwork->notify.notifier_id,
2954 lwork->notify.payload_len);
2955 lreq->wcb(lreq->data, lwork->notify.notify_id, lreq->linger_id,
2956 lwork->notify.notifier_id, lwork->notify.payload,
2957 lwork->notify.payload_len);
2960 ceph_msg_put(lwork->notify.msg);
2964 static void do_watch_error(struct work_struct *w)
2966 struct linger_work *lwork = container_of(w, struct linger_work, work);
2967 struct ceph_osd_linger_request *lreq = lwork->lreq;
2969 if (!linger_registered(lreq)) {
2970 dout("%s lreq %p not registered\n", __func__, lreq);
2974 dout("%s lreq %p err %d\n", __func__, lreq, lwork->error.err);
2975 lreq->errcb(lreq->data, lreq->linger_id, lwork->error.err);
2981 static void queue_watch_error(struct ceph_osd_linger_request *lreq)
2983 struct linger_work *lwork;
2985 lwork = lwork_alloc(lreq, do_watch_error);
2987 pr_err("failed to allocate error-lwork\n");
2991 lwork->error.err = lreq->last_error;
2995 static void linger_reg_commit_complete(struct ceph_osd_linger_request *lreq,
2998 if (!completion_done(&lreq->reg_commit_wait)) {
2999 lreq->reg_commit_error = (result <= 0 ? result : 0);
3000 complete_all(&lreq->reg_commit_wait);
3004 static void linger_commit_cb(struct ceph_osd_request *req)
3006 struct ceph_osd_linger_request *lreq = req->r_priv;
3008 mutex_lock(&lreq->lock);
3009 dout("%s lreq %p linger_id %llu result %d\n", __func__, lreq,
3010 lreq->linger_id, req->r_result);
3011 linger_reg_commit_complete(lreq, req->r_result);
3012 lreq->committed = true;
3014 if (!lreq->is_watch) {
3015 struct ceph_osd_data *osd_data =
3016 osd_req_op_data(req, 0, notify, response_data);
3017 void *p = page_address(osd_data->pages[0]);
3019 WARN_ON(req->r_ops[0].op != CEPH_OSD_OP_NOTIFY ||
3020 osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
3022 /* make note of the notify_id */
3023 if (req->r_ops[0].outdata_len >= sizeof(u64)) {
3024 lreq->notify_id = ceph_decode_64(&p);
3025 dout("lreq %p notify_id %llu\n", lreq,
3028 dout("lreq %p no notify_id\n", lreq);
3032 mutex_unlock(&lreq->lock);
3036 static int normalize_watch_error(int err)
3039 * Translate ENOENT -> ENOTCONN so that a delete->disconnection
3040 * notification and a failure to reconnect because we raced with
3041 * the delete appear the same to the user.
3049 static void linger_reconnect_cb(struct ceph_osd_request *req)
3051 struct ceph_osd_linger_request *lreq = req->r_priv;
3053 mutex_lock(&lreq->lock);
3054 dout("%s lreq %p linger_id %llu result %d last_error %d\n", __func__,
3055 lreq, lreq->linger_id, req->r_result, lreq->last_error);
3056 if (req->r_result < 0) {
3057 if (!lreq->last_error) {
3058 lreq->last_error = normalize_watch_error(req->r_result);
3059 queue_watch_error(lreq);
3063 mutex_unlock(&lreq->lock);
3067 static void send_linger(struct ceph_osd_linger_request *lreq)
3069 struct ceph_osd_request *req = lreq->reg_req;
3070 struct ceph_osd_req_op *op = &req->r_ops[0];
3072 verify_osdc_wrlocked(req->r_osdc);
3073 dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
3076 cancel_linger_request(req);
3078 request_reinit(req);
3079 ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
3080 ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
3081 req->r_flags = lreq->t.flags;
3082 req->r_mtime = lreq->mtime;
3084 mutex_lock(&lreq->lock);
3085 if (lreq->is_watch && lreq->committed) {
3086 WARN_ON(op->op != CEPH_OSD_OP_WATCH ||
3087 op->watch.cookie != lreq->linger_id);
3088 op->watch.op = CEPH_OSD_WATCH_OP_RECONNECT;
3089 op->watch.gen = ++lreq->register_gen;
3090 dout("lreq %p reconnect register_gen %u\n", lreq,
3092 req->r_callback = linger_reconnect_cb;
3094 if (!lreq->is_watch)
3095 lreq->notify_id = 0;
3097 WARN_ON(op->watch.op != CEPH_OSD_WATCH_OP_WATCH);
3098 dout("lreq %p register\n", lreq);
3099 req->r_callback = linger_commit_cb;
3101 mutex_unlock(&lreq->lock);
3103 req->r_priv = linger_get(lreq);
3104 req->r_linger = true;
3106 submit_request(req, true);
3109 static void linger_ping_cb(struct ceph_osd_request *req)
3111 struct ceph_osd_linger_request *lreq = req->r_priv;
3113 mutex_lock(&lreq->lock);
3114 dout("%s lreq %p linger_id %llu result %d ping_sent %lu last_error %d\n",
3115 __func__, lreq, lreq->linger_id, req->r_result, lreq->ping_sent,
3117 if (lreq->register_gen == req->r_ops[0].watch.gen) {
3118 if (!req->r_result) {
3119 lreq->watch_valid_thru = lreq->ping_sent;
3120 } else if (!lreq->last_error) {
3121 lreq->last_error = normalize_watch_error(req->r_result);
3122 queue_watch_error(lreq);
3125 dout("lreq %p register_gen %u ignoring old pong %u\n", lreq,
3126 lreq->register_gen, req->r_ops[0].watch.gen);
3129 mutex_unlock(&lreq->lock);
3133 static void send_linger_ping(struct ceph_osd_linger_request *lreq)
3135 struct ceph_osd_client *osdc = lreq->osdc;
3136 struct ceph_osd_request *req = lreq->ping_req;
3137 struct ceph_osd_req_op *op = &req->r_ops[0];
3139 if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) {
3140 dout("%s PAUSERD\n", __func__);
3144 lreq->ping_sent = jiffies;
3145 dout("%s lreq %p linger_id %llu ping_sent %lu register_gen %u\n",
3146 __func__, lreq, lreq->linger_id, lreq->ping_sent,
3147 lreq->register_gen);
3150 cancel_linger_request(req);
3152 request_reinit(req);
3153 target_copy(&req->r_t, &lreq->t);
3155 WARN_ON(op->op != CEPH_OSD_OP_WATCH ||
3156 op->watch.cookie != lreq->linger_id ||
3157 op->watch.op != CEPH_OSD_WATCH_OP_PING);
3158 op->watch.gen = lreq->register_gen;
3159 req->r_callback = linger_ping_cb;
3160 req->r_priv = linger_get(lreq);
3161 req->r_linger = true;
3163 ceph_osdc_get_request(req);
3164 account_request(req);
3165 req->r_tid = atomic64_inc_return(&osdc->last_tid);
3166 link_request(lreq->osd, req);
3170 static void linger_submit(struct ceph_osd_linger_request *lreq)
3172 struct ceph_osd_client *osdc = lreq->osdc;
3173 struct ceph_osd *osd;
3175 down_write(&osdc->lock);
3176 linger_register(lreq);
3177 if (lreq->is_watch) {
3178 lreq->reg_req->r_ops[0].watch.cookie = lreq->linger_id;
3179 lreq->ping_req->r_ops[0].watch.cookie = lreq->linger_id;
3181 lreq->reg_req->r_ops[0].notify.cookie = lreq->linger_id;
3184 calc_target(osdc, &lreq->t, false);
3185 osd = lookup_create_osd(osdc, lreq->t.osd, true);
3186 link_linger(osd, lreq);
3189 up_write(&osdc->lock);
3192 static void cancel_linger_map_check(struct ceph_osd_linger_request *lreq)
3194 struct ceph_osd_client *osdc = lreq->osdc;
3195 struct ceph_osd_linger_request *lookup_lreq;
3197 verify_osdc_wrlocked(osdc);
3199 lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks,
3204 WARN_ON(lookup_lreq != lreq);
3205 erase_linger_mc(&osdc->linger_map_checks, lreq);
3210 * @lreq has to be both registered and linked.
3212 static void __linger_cancel(struct ceph_osd_linger_request *lreq)
3214 if (lreq->is_watch && lreq->ping_req->r_osd)
3215 cancel_linger_request(lreq->ping_req);
3216 if (lreq->reg_req->r_osd)
3217 cancel_linger_request(lreq->reg_req);
3218 cancel_linger_map_check(lreq);
3219 unlink_linger(lreq->osd, lreq);
3220 linger_unregister(lreq);
3223 static void linger_cancel(struct ceph_osd_linger_request *lreq)
3225 struct ceph_osd_client *osdc = lreq->osdc;
3227 down_write(&osdc->lock);
3228 if (__linger_registered(lreq))
3229 __linger_cancel(lreq);
3230 up_write(&osdc->lock);
3233 static void send_linger_map_check(struct ceph_osd_linger_request *lreq);
3235 static void check_linger_pool_dne(struct ceph_osd_linger_request *lreq)
3237 struct ceph_osd_client *osdc = lreq->osdc;
3238 struct ceph_osdmap *map = osdc->osdmap;
3240 verify_osdc_wrlocked(osdc);
3241 WARN_ON(!map->epoch);
3243 if (lreq->register_gen) {
3244 lreq->map_dne_bound = map->epoch;
3245 dout("%s lreq %p linger_id %llu pool disappeared\n", __func__,
3246 lreq, lreq->linger_id);
3248 dout("%s lreq %p linger_id %llu map_dne_bound %u have %u\n",
3249 __func__, lreq, lreq->linger_id, lreq->map_dne_bound,
3253 if (lreq->map_dne_bound) {
3254 if (map->epoch >= lreq->map_dne_bound) {
3255 /* we had a new enough map */
3256 pr_info("linger_id %llu pool does not exist\n",
3258 linger_reg_commit_complete(lreq, -ENOENT);
3259 __linger_cancel(lreq);
3262 send_linger_map_check(lreq);
3266 static void linger_map_check_cb(struct ceph_mon_generic_request *greq)
3268 struct ceph_osd_client *osdc = &greq->monc->client->osdc;
3269 struct ceph_osd_linger_request *lreq;
3270 u64 linger_id = greq->private_data;
3272 WARN_ON(greq->result || !greq->u.newest);
3274 down_write(&osdc->lock);
3275 lreq = lookup_linger_mc(&osdc->linger_map_checks, linger_id);
3277 dout("%s linger_id %llu dne\n", __func__, linger_id);
3281 dout("%s lreq %p linger_id %llu map_dne_bound %u newest %llu\n",
3282 __func__, lreq, lreq->linger_id, lreq->map_dne_bound,
3284 if (!lreq->map_dne_bound)
3285 lreq->map_dne_bound = greq->u.newest;
3286 erase_linger_mc(&osdc->linger_map_checks, lreq);
3287 check_linger_pool_dne(lreq);
3291 up_write(&osdc->lock);
3294 static void send_linger_map_check(struct ceph_osd_linger_request *lreq)
3296 struct ceph_osd_client *osdc = lreq->osdc;
3297 struct ceph_osd_linger_request *lookup_lreq;
3300 verify_osdc_wrlocked(osdc);
3302 lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks,
3305 WARN_ON(lookup_lreq != lreq);
3310 insert_linger_mc(&osdc->linger_map_checks, lreq);
3311 ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap",
3312 linger_map_check_cb, lreq->linger_id);
3316 static int linger_reg_commit_wait(struct ceph_osd_linger_request *lreq)
3320 dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
3321 ret = wait_for_completion_interruptible(&lreq->reg_commit_wait);
3322 return ret ?: lreq->reg_commit_error;
3325 static int linger_notify_finish_wait(struct ceph_osd_linger_request *lreq)
3329 dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
3330 ret = wait_for_completion_interruptible(&lreq->notify_finish_wait);
3331 return ret ?: lreq->notify_finish_error;
3335 * Timeout callback, called every N seconds. When 1 or more OSD
3336 * requests has been active for more than N seconds, we send a keepalive
3337 * (tag + timestamp) to its OSD to ensure any communications channel
3338 * reset is detected.
3340 static void handle_timeout(struct work_struct *work)
3342 struct ceph_osd_client *osdc =
3343 container_of(work, struct ceph_osd_client, timeout_work.work);
3344 struct ceph_options *opts = osdc->client->options;
3345 unsigned long cutoff = jiffies - opts->osd_keepalive_timeout;
3346 unsigned long expiry_cutoff = jiffies - opts->osd_request_timeout;
3347 LIST_HEAD(slow_osds);
3348 struct rb_node *n, *p;
3350 dout("%s osdc %p\n", __func__, osdc);
3351 down_write(&osdc->lock);
3354 * ping osds that are a bit slow. this ensures that if there
3355 * is a break in the TCP connection we will notice, and reopen
3356 * a connection with that osd (from the fault callback).
3358 for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
3359 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
3362 for (p = rb_first(&osd->o_requests); p; ) {
3363 struct ceph_osd_request *req =
3364 rb_entry(p, struct ceph_osd_request, r_node);
3366 p = rb_next(p); /* abort_request() */
3368 if (time_before(req->r_stamp, cutoff)) {
3369 dout(" req %p tid %llu on osd%d is laggy\n",
3370 req, req->r_tid, osd->o_osd);
3373 if (opts->osd_request_timeout &&
3374 time_before(req->r_start_stamp, expiry_cutoff)) {
3375 pr_err_ratelimited("tid %llu on osd%d timeout\n",
3376 req->r_tid, osd->o_osd);
3377 abort_request(req, -ETIMEDOUT);
3380 for (p = rb_first(&osd->o_linger_requests); p; p = rb_next(p)) {
3381 struct ceph_osd_linger_request *lreq =
3382 rb_entry(p, struct ceph_osd_linger_request, node);
3384 dout(" lreq %p linger_id %llu is served by osd%d\n",
3385 lreq, lreq->linger_id, osd->o_osd);
3388 mutex_lock(&lreq->lock);
3389 if (lreq->is_watch && lreq->committed && !lreq->last_error)
3390 send_linger_ping(lreq);
3391 mutex_unlock(&lreq->lock);
3395 list_move_tail(&osd->o_keepalive_item, &slow_osds);
3398 if (opts->osd_request_timeout) {
3399 for (p = rb_first(&osdc->homeless_osd.o_requests); p; ) {
3400 struct ceph_osd_request *req =
3401 rb_entry(p, struct ceph_osd_request, r_node);
3403 p = rb_next(p); /* abort_request() */
3405 if (time_before(req->r_start_stamp, expiry_cutoff)) {
3406 pr_err_ratelimited("tid %llu on osd%d timeout\n",
3407 req->r_tid, osdc->homeless_osd.o_osd);
3408 abort_request(req, -ETIMEDOUT);
3413 if (atomic_read(&osdc->num_homeless) || !list_empty(&slow_osds))
3414 maybe_request_map(osdc);
3416 while (!list_empty(&slow_osds)) {
3417 struct ceph_osd *osd = list_first_entry(&slow_osds,
3420 list_del_init(&osd->o_keepalive_item);
3421 ceph_con_keepalive(&osd->o_con);
3424 up_write(&osdc->lock);
3425 schedule_delayed_work(&osdc->timeout_work,
3426 osdc->client->options->osd_keepalive_timeout);
3429 static void handle_osds_timeout(struct work_struct *work)
3431 struct ceph_osd_client *osdc =
3432 container_of(work, struct ceph_osd_client,
3433 osds_timeout_work.work);
3434 unsigned long delay = osdc->client->options->osd_idle_ttl / 4;
3435 struct ceph_osd *osd, *nosd;
3437 dout("%s osdc %p\n", __func__, osdc);
3438 down_write(&osdc->lock);
3439 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
3440 if (time_before(jiffies, osd->lru_ttl))
3443 WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests));
3444 WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests));
3448 up_write(&osdc->lock);
3449 schedule_delayed_work(&osdc->osds_timeout_work,
3450 round_jiffies_relative(delay));
3453 static int ceph_oloc_decode(void **p, void *end,
3454 struct ceph_object_locator *oloc)
3456 u8 struct_v, struct_cv;
3461 ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
3462 struct_v = ceph_decode_8(p);
3463 struct_cv = ceph_decode_8(p);
3465 pr_warn("got v %d < 3 cv %d of ceph_object_locator\n",
3466 struct_v, struct_cv);
3469 if (struct_cv > 6) {
3470 pr_warn("got v %d cv %d > 6 of ceph_object_locator\n",
3471 struct_v, struct_cv);
3474 len = ceph_decode_32(p);
3475 ceph_decode_need(p, end, len, e_inval);
3476 struct_end = *p + len;
3478 oloc->pool = ceph_decode_64(p);
3479 *p += 4; /* skip preferred */
3481 len = ceph_decode_32(p);
3483 pr_warn("ceph_object_locator::key is set\n");
3487 if (struct_v >= 5) {
3488 bool changed = false;
3490 len = ceph_decode_32(p);
3492 ceph_decode_need(p, end, len, e_inval);
3493 if (!oloc->pool_ns ||
3494 ceph_compare_string(oloc->pool_ns, *p, len))
3502 /* redirect changes namespace */
3503 pr_warn("ceph_object_locator::nspace is changed\n");
3508 if (struct_v >= 6) {
3509 s64 hash = ceph_decode_64(p);
3511 pr_warn("ceph_object_locator::hash is set\n");
3526 static int ceph_redirect_decode(void **p, void *end,
3527 struct ceph_request_redirect *redir)
3529 u8 struct_v, struct_cv;
3534 ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
3535 struct_v = ceph_decode_8(p);
3536 struct_cv = ceph_decode_8(p);
3537 if (struct_cv > 1) {
3538 pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n",
3539 struct_v, struct_cv);
3542 len = ceph_decode_32(p);
3543 ceph_decode_need(p, end, len, e_inval);
3544 struct_end = *p + len;
3546 ret = ceph_oloc_decode(p, end, &redir->oloc);
3550 len = ceph_decode_32(p);
3552 pr_warn("ceph_request_redirect::object_name is set\n");
3566 struct MOSDOpReply {
3567 struct ceph_pg pgid;
3572 u32 outdata_len[CEPH_OSD_MAX_OPS];
3573 s32 rval[CEPH_OSD_MAX_OPS];
3575 struct ceph_eversion replay_version;
3577 struct ceph_request_redirect redirect;
3580 static int decode_MOSDOpReply(const struct ceph_msg *msg, struct MOSDOpReply *m)
3582 void *p = msg->front.iov_base;
3583 void *const end = p + msg->front.iov_len;
3584 u16 version = le16_to_cpu(msg->hdr.version);
3585 struct ceph_eversion bad_replay_version;
3591 ceph_decode_32_safe(&p, end, len, e_inval);
3592 ceph_decode_need(&p, end, len, e_inval);
3593 p += len; /* skip oid */
3595 ret = ceph_decode_pgid(&p, end, &m->pgid);
3599 ceph_decode_64_safe(&p, end, m->flags, e_inval);
3600 ceph_decode_32_safe(&p, end, m->result, e_inval);
3601 ceph_decode_need(&p, end, sizeof(bad_replay_version), e_inval);
3602 memcpy(&bad_replay_version, p, sizeof(bad_replay_version));
3603 p += sizeof(bad_replay_version);
3604 ceph_decode_32_safe(&p, end, m->epoch, e_inval);
3606 ceph_decode_32_safe(&p, end, m->num_ops, e_inval);
3607 if (m->num_ops > ARRAY_SIZE(m->outdata_len))
3610 ceph_decode_need(&p, end, m->num_ops * sizeof(struct ceph_osd_op),
3612 for (i = 0; i < m->num_ops; i++) {
3613 struct ceph_osd_op *op = p;
3615 m->outdata_len[i] = le32_to_cpu(op->payload_len);
3619 ceph_decode_32_safe(&p, end, m->retry_attempt, e_inval);
3620 for (i = 0; i < m->num_ops; i++)
3621 ceph_decode_32_safe(&p, end, m->rval[i], e_inval);
3624 ceph_decode_need(&p, end, sizeof(m->replay_version), e_inval);
3625 memcpy(&m->replay_version, p, sizeof(m->replay_version));
3626 p += sizeof(m->replay_version);
3627 ceph_decode_64_safe(&p, end, m->user_version, e_inval);
3629 m->replay_version = bad_replay_version; /* struct */
3630 m->user_version = le64_to_cpu(m->replay_version.version);
3635 ceph_decode_8_safe(&p, end, decode_redir, e_inval);
3643 ret = ceph_redirect_decode(&p, end, &m->redirect);
3647 ceph_oloc_init(&m->redirect.oloc);
3657 * Handle MOSDOpReply. Set ->r_result and call the callback if it is
3660 static void handle_reply(struct ceph_osd *osd, struct ceph_msg *msg)
3662 struct ceph_osd_client *osdc = osd->o_osdc;
3663 struct ceph_osd_request *req;
3664 struct MOSDOpReply m;
3665 u64 tid = le64_to_cpu(msg->hdr.tid);
3670 dout("%s msg %p tid %llu\n", __func__, msg, tid);
3672 down_read(&osdc->lock);
3673 if (!osd_registered(osd)) {
3674 dout("%s osd%d unknown\n", __func__, osd->o_osd);
3675 goto out_unlock_osdc;
3677 WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num));
3679 mutex_lock(&osd->lock);
3680 req = lookup_request(&osd->o_requests, tid);
3682 dout("%s osd%d tid %llu unknown\n", __func__, osd->o_osd, tid);
3683 goto out_unlock_session;
3686 m.redirect.oloc.pool_ns = req->r_t.target_oloc.pool_ns;
3687 ret = decode_MOSDOpReply(msg, &m);
3688 m.redirect.oloc.pool_ns = NULL;
3690 pr_err("failed to decode MOSDOpReply for tid %llu: %d\n",
3695 dout("%s req %p tid %llu flags 0x%llx pgid %llu.%x epoch %u attempt %d v %u'%llu uv %llu\n",
3696 __func__, req, req->r_tid, m.flags, m.pgid.pool, m.pgid.seed,
3697 m.epoch, m.retry_attempt, le32_to_cpu(m.replay_version.epoch),
3698 le64_to_cpu(m.replay_version.version), m.user_version);
3700 if (m.retry_attempt >= 0) {
3701 if (m.retry_attempt != req->r_attempts - 1) {
3702 dout("req %p tid %llu retry_attempt %d != %d, ignoring\n",
3703 req, req->r_tid, m.retry_attempt,
3704 req->r_attempts - 1);
3705 goto out_unlock_session;
3708 WARN_ON(1); /* MOSDOpReply v4 is assumed */
3711 if (!ceph_oloc_empty(&m.redirect.oloc)) {
3712 dout("req %p tid %llu redirect pool %lld\n", req, req->r_tid,
3713 m.redirect.oloc.pool);
3714 unlink_request(osd, req);
3715 mutex_unlock(&osd->lock);
3718 * Not ceph_oloc_copy() - changing pool_ns is not
3721 req->r_t.target_oloc.pool = m.redirect.oloc.pool;
3722 req->r_flags |= CEPH_OSD_FLAG_REDIRECTED |
3723 CEPH_OSD_FLAG_IGNORE_OVERLAY |
3724 CEPH_OSD_FLAG_IGNORE_CACHE;
3726 __submit_request(req, false);
3727 goto out_unlock_osdc;
3730 if (m.result == -EAGAIN) {
3731 dout("req %p tid %llu EAGAIN\n", req, req->r_tid);
3732 unlink_request(osd, req);
3733 mutex_unlock(&osd->lock);
3736 * The object is missing on the replica or not (yet)
3737 * readable. Clear pgid to force a resend to the primary
3738 * via legacy_change.
3740 req->r_t.pgid.pool = 0;
3741 req->r_t.pgid.seed = 0;
3742 WARN_ON(!req->r_t.used_replica);
3743 req->r_flags &= ~(CEPH_OSD_FLAG_BALANCE_READS |
3744 CEPH_OSD_FLAG_LOCALIZE_READS);
3746 __submit_request(req, false);
3747 goto out_unlock_osdc;
3750 if (m.num_ops != req->r_num_ops) {
3751 pr_err("num_ops %d != %d for tid %llu\n", m.num_ops,
3752 req->r_num_ops, req->r_tid);
3755 for (i = 0; i < req->r_num_ops; i++) {
3756 dout(" req %p tid %llu op %d rval %d len %u\n", req,
3757 req->r_tid, i, m.rval[i], m.outdata_len[i]);
3758 req->r_ops[i].rval = m.rval[i];
3759 req->r_ops[i].outdata_len = m.outdata_len[i];
3760 data_len += m.outdata_len[i];
3762 if (data_len != le32_to_cpu(msg->hdr.data_len)) {
3763 pr_err("sum of lens %u != %u for tid %llu\n", data_len,
3764 le32_to_cpu(msg->hdr.data_len), req->r_tid);
3767 dout("%s req %p tid %llu result %d data_len %u\n", __func__,
3768 req, req->r_tid, m.result, data_len);
3771 * Since we only ever request ONDISK, we should only ever get
3772 * one (type of) reply back.
3774 WARN_ON(!(m.flags & CEPH_OSD_FLAG_ONDISK));
3775 req->r_result = m.result ?: data_len;
3776 finish_request(req);
3777 mutex_unlock(&osd->lock);
3778 up_read(&osdc->lock);
3780 __complete_request(req);
3784 complete_request(req, -EIO);
3786 mutex_unlock(&osd->lock);
3788 up_read(&osdc->lock);
3791 static void set_pool_was_full(struct ceph_osd_client *osdc)
3795 for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) {
3796 struct ceph_pg_pool_info *pi =
3797 rb_entry(n, struct ceph_pg_pool_info, node);
3799 pi->was_full = __pool_full(pi);
3803 static bool pool_cleared_full(struct ceph_osd_client *osdc, s64 pool_id)
3805 struct ceph_pg_pool_info *pi;
3807 pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id);
3811 return pi->was_full && !__pool_full(pi);
3814 static enum calc_target_result
3815 recalc_linger_target(struct ceph_osd_linger_request *lreq)
3817 struct ceph_osd_client *osdc = lreq->osdc;
3818 enum calc_target_result ct_res;
3820 ct_res = calc_target(osdc, &lreq->t, true);
3821 if (ct_res == CALC_TARGET_NEED_RESEND) {
3822 struct ceph_osd *osd;
3824 osd = lookup_create_osd(osdc, lreq->t.osd, true);
3825 if (osd != lreq->osd) {
3826 unlink_linger(lreq->osd, lreq);
3827 link_linger(osd, lreq);
3835 * Requeue requests whose mapping to an OSD has changed.
3837 static void scan_requests(struct ceph_osd *osd,
3840 bool check_pool_cleared_full,
3841 struct rb_root *need_resend,
3842 struct list_head *need_resend_linger)
3844 struct ceph_osd_client *osdc = osd->o_osdc;
3846 bool force_resend_writes;
3848 for (n = rb_first(&osd->o_linger_requests); n; ) {
3849 struct ceph_osd_linger_request *lreq =
3850 rb_entry(n, struct ceph_osd_linger_request, node);
3851 enum calc_target_result ct_res;
3853 n = rb_next(n); /* recalc_linger_target() */
3855 dout("%s lreq %p linger_id %llu\n", __func__, lreq,
3857 ct_res = recalc_linger_target(lreq);
3859 case CALC_TARGET_NO_ACTION:
3860 force_resend_writes = cleared_full ||
3861 (check_pool_cleared_full &&
3862 pool_cleared_full(osdc, lreq->t.base_oloc.pool));
3863 if (!force_resend && !force_resend_writes)
3867 case CALC_TARGET_NEED_RESEND:
3868 cancel_linger_map_check(lreq);
3870 * scan_requests() for the previous epoch(s)
3871 * may have already added it to the list, since
3872 * it's not unlinked here.
3874 if (list_empty(&lreq->scan_item))
3875 list_add_tail(&lreq->scan_item, need_resend_linger);
3877 case CALC_TARGET_POOL_DNE:
3878 list_del_init(&lreq->scan_item);
3879 check_linger_pool_dne(lreq);
3884 for (n = rb_first(&osd->o_requests); n; ) {
3885 struct ceph_osd_request *req =
3886 rb_entry(n, struct ceph_osd_request, r_node);
3887 enum calc_target_result ct_res;
3889 n = rb_next(n); /* unlink_request(), check_pool_dne() */
3891 dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
3892 ct_res = calc_target(osdc, &req->r_t, false);
3894 case CALC_TARGET_NO_ACTION:
3895 force_resend_writes = cleared_full ||
3896 (check_pool_cleared_full &&
3897 pool_cleared_full(osdc, req->r_t.base_oloc.pool));
3898 if (!force_resend &&
3899 (!(req->r_flags & CEPH_OSD_FLAG_WRITE) ||
3900 !force_resend_writes))
3904 case CALC_TARGET_NEED_RESEND:
3905 cancel_map_check(req);
3906 unlink_request(osd, req);
3907 insert_request(need_resend, req);
3909 case CALC_TARGET_POOL_DNE:
3910 check_pool_dne(req);
3916 static int handle_one_map(struct ceph_osd_client *osdc,
3917 void *p, void *end, bool incremental,
3918 struct rb_root *need_resend,
3919 struct list_head *need_resend_linger)
3921 struct ceph_osdmap *newmap;
3923 bool skipped_map = false;
3926 was_full = ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL);
3927 set_pool_was_full(osdc);
3930 newmap = osdmap_apply_incremental(&p, end, osdc->osdmap);
3932 newmap = ceph_osdmap_decode(&p, end);
3934 return PTR_ERR(newmap);
3936 if (newmap != osdc->osdmap) {
3938 * Preserve ->was_full before destroying the old map.
3939 * For pools that weren't in the old map, ->was_full
3942 for (n = rb_first(&newmap->pg_pools); n; n = rb_next(n)) {
3943 struct ceph_pg_pool_info *pi =
3944 rb_entry(n, struct ceph_pg_pool_info, node);
3945 struct ceph_pg_pool_info *old_pi;
3947 old_pi = ceph_pg_pool_by_id(osdc->osdmap, pi->id);
3949 pi->was_full = old_pi->was_full;
3951 WARN_ON(pi->was_full);
3954 if (osdc->osdmap->epoch &&
3955 osdc->osdmap->epoch + 1 < newmap->epoch) {
3956 WARN_ON(incremental);
3960 ceph_osdmap_destroy(osdc->osdmap);
3961 osdc->osdmap = newmap;
3964 was_full &= !ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL);
3965 scan_requests(&osdc->homeless_osd, skipped_map, was_full, true,
3966 need_resend, need_resend_linger);
3968 for (n = rb_first(&osdc->osds); n; ) {
3969 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
3971 n = rb_next(n); /* close_osd() */
3973 scan_requests(osd, skipped_map, was_full, true, need_resend,
3974 need_resend_linger);
3975 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
3976 memcmp(&osd->o_con.peer_addr,
3977 ceph_osd_addr(osdc->osdmap, osd->o_osd),
3978 sizeof(struct ceph_entity_addr)))
3985 static void kick_requests(struct ceph_osd_client *osdc,
3986 struct rb_root *need_resend,
3987 struct list_head *need_resend_linger)
3989 struct ceph_osd_linger_request *lreq, *nlreq;
3990 enum calc_target_result ct_res;
3993 /* make sure need_resend targets reflect latest map */
3994 for (n = rb_first(need_resend); n; ) {
3995 struct ceph_osd_request *req =
3996 rb_entry(n, struct ceph_osd_request, r_node);
4000 if (req->r_t.epoch < osdc->osdmap->epoch) {
4001 ct_res = calc_target(osdc, &req->r_t, false);
4002 if (ct_res == CALC_TARGET_POOL_DNE) {
4003 erase_request(need_resend, req);
4004 check_pool_dne(req);
4009 for (n = rb_first(need_resend); n; ) {
4010 struct ceph_osd_request *req =
4011 rb_entry(n, struct ceph_osd_request, r_node);
4012 struct ceph_osd *osd;
4015 erase_request(need_resend, req); /* before link_request() */
4017 osd = lookup_create_osd(osdc, req->r_t.osd, true);
4018 link_request(osd, req);
4019 if (!req->r_linger) {
4020 if (!osd_homeless(osd) && !req->r_t.paused)
4023 cancel_linger_request(req);
4027 list_for_each_entry_safe(lreq, nlreq, need_resend_linger, scan_item) {
4028 if (!osd_homeless(lreq->osd))
4031 list_del_init(&lreq->scan_item);
4036 * Process updated osd map.
4038 * The message contains any number of incremental and full maps, normally
4039 * indicating some sort of topology change in the cluster. Kick requests
4040 * off to different OSDs as needed.
4042 void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
4044 void *p = msg->front.iov_base;
4045 void *const end = p + msg->front.iov_len;
4046 u32 nr_maps, maplen;
4048 struct ceph_fsid fsid;
4049 struct rb_root need_resend = RB_ROOT;
4050 LIST_HEAD(need_resend_linger);
4051 bool handled_incremental = false;
4052 bool was_pauserd, was_pausewr;
4053 bool pauserd, pausewr;
4056 dout("%s have %u\n", __func__, osdc->osdmap->epoch);
4057 down_write(&osdc->lock);
4060 ceph_decode_need(&p, end, sizeof(fsid), bad);
4061 ceph_decode_copy(&p, &fsid, sizeof(fsid));
4062 if (ceph_check_fsid(osdc->client, &fsid) < 0)
4065 was_pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
4066 was_pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
4067 ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
4068 have_pool_full(osdc);
4070 /* incremental maps */
4071 ceph_decode_32_safe(&p, end, nr_maps, bad);
4072 dout(" %d inc maps\n", nr_maps);
4073 while (nr_maps > 0) {
4074 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
4075 epoch = ceph_decode_32(&p);
4076 maplen = ceph_decode_32(&p);
4077 ceph_decode_need(&p, end, maplen, bad);
4078 if (osdc->osdmap->epoch &&
4079 osdc->osdmap->epoch + 1 == epoch) {
4080 dout("applying incremental map %u len %d\n",
4082 err = handle_one_map(osdc, p, p + maplen, true,
4083 &need_resend, &need_resend_linger);
4086 handled_incremental = true;
4088 dout("ignoring incremental map %u len %d\n",
4094 if (handled_incremental)
4098 ceph_decode_32_safe(&p, end, nr_maps, bad);
4099 dout(" %d full maps\n", nr_maps);
4101 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
4102 epoch = ceph_decode_32(&p);
4103 maplen = ceph_decode_32(&p);
4104 ceph_decode_need(&p, end, maplen, bad);
4106 dout("skipping non-latest full map %u len %d\n",
4108 } else if (osdc->osdmap->epoch >= epoch) {
4109 dout("skipping full map %u len %d, "
4110 "older than our %u\n", epoch, maplen,
4111 osdc->osdmap->epoch);
4113 dout("taking full map %u len %d\n", epoch, maplen);
4114 err = handle_one_map(osdc, p, p + maplen, false,
4115 &need_resend, &need_resend_linger);
4125 * subscribe to subsequent osdmap updates if full to ensure
4126 * we find out when we are no longer full and stop returning
4129 pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
4130 pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
4131 ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
4132 have_pool_full(osdc);
4133 if (was_pauserd || was_pausewr || pauserd || pausewr ||
4134 osdc->osdmap->epoch < osdc->epoch_barrier)
4135 maybe_request_map(osdc);
4137 kick_requests(osdc, &need_resend, &need_resend_linger);
4139 ceph_osdc_abort_on_full(osdc);
4140 ceph_monc_got_map(&osdc->client->monc, CEPH_SUB_OSDMAP,
4141 osdc->osdmap->epoch);
4142 up_write(&osdc->lock);
4143 wake_up_all(&osdc->client->auth_wq);
4147 pr_err("osdc handle_map corrupt msg\n");
4149 up_write(&osdc->lock);
4153 * Resubmit requests pending on the given osd.
4155 static void kick_osd_requests(struct ceph_osd *osd)
4159 clear_backoffs(osd);
4161 for (n = rb_first(&osd->o_requests); n; ) {
4162 struct ceph_osd_request *req =
4163 rb_entry(n, struct ceph_osd_request, r_node);
4165 n = rb_next(n); /* cancel_linger_request() */
4167 if (!req->r_linger) {
4168 if (!req->r_t.paused)
4171 cancel_linger_request(req);
4174 for (n = rb_first(&osd->o_linger_requests); n; n = rb_next(n)) {
4175 struct ceph_osd_linger_request *lreq =
4176 rb_entry(n, struct ceph_osd_linger_request, node);
4183 * If the osd connection drops, we need to resubmit all requests.
4185 static void osd_fault(struct ceph_connection *con)
4187 struct ceph_osd *osd = con->private;
4188 struct ceph_osd_client *osdc = osd->o_osdc;
4190 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
4192 down_write(&osdc->lock);
4193 if (!osd_registered(osd)) {
4194 dout("%s osd%d unknown\n", __func__, osd->o_osd);
4198 if (!reopen_osd(osd))
4199 kick_osd_requests(osd);
4200 maybe_request_map(osdc);
4203 up_write(&osdc->lock);
4206 struct MOSDBackoff {
4207 struct ceph_spg spgid;
4211 struct ceph_hobject_id *begin;
4212 struct ceph_hobject_id *end;
4215 static int decode_MOSDBackoff(const struct ceph_msg *msg, struct MOSDBackoff *m)
4217 void *p = msg->front.iov_base;
4218 void *const end = p + msg->front.iov_len;
4223 ret = ceph_start_decoding(&p, end, 1, "spg_t", &struct_v, &struct_len);
4227 ret = ceph_decode_pgid(&p, end, &m->spgid.pgid);
4231 ceph_decode_8_safe(&p, end, m->spgid.shard, e_inval);
4232 ceph_decode_32_safe(&p, end, m->map_epoch, e_inval);
4233 ceph_decode_8_safe(&p, end, m->op, e_inval);
4234 ceph_decode_64_safe(&p, end, m->id, e_inval);
4236 m->begin = kzalloc(sizeof(*m->begin), GFP_NOIO);
4240 ret = decode_hoid(&p, end, m->begin);
4242 free_hoid(m->begin);
4246 m->end = kzalloc(sizeof(*m->end), GFP_NOIO);
4248 free_hoid(m->begin);
4252 ret = decode_hoid(&p, end, m->end);
4254 free_hoid(m->begin);
4265 static struct ceph_msg *create_backoff_message(
4266 const struct ceph_osd_backoff *backoff,
4269 struct ceph_msg *msg;
4273 msg_size = CEPH_ENCODING_START_BLK_LEN +
4274 CEPH_PGID_ENCODING_LEN + 1; /* spgid */
4275 msg_size += 4 + 1 + 8; /* map_epoch, op, id */
4276 msg_size += CEPH_ENCODING_START_BLK_LEN +
4277 hoid_encoding_size(backoff->begin);
4278 msg_size += CEPH_ENCODING_START_BLK_LEN +
4279 hoid_encoding_size(backoff->end);
4281 msg = ceph_msg_new(CEPH_MSG_OSD_BACKOFF, msg_size, GFP_NOIO, true);
4285 p = msg->front.iov_base;
4286 end = p + msg->front_alloc_len;
4288 encode_spgid(&p, &backoff->spgid);
4289 ceph_encode_32(&p, map_epoch);
4290 ceph_encode_8(&p, CEPH_OSD_BACKOFF_OP_ACK_BLOCK);
4291 ceph_encode_64(&p, backoff->id);
4292 encode_hoid(&p, end, backoff->begin);
4293 encode_hoid(&p, end, backoff->end);
4296 msg->front.iov_len = p - msg->front.iov_base;
4297 msg->hdr.version = cpu_to_le16(1); /* MOSDBackoff v1 */
4298 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
4303 static void handle_backoff_block(struct ceph_osd *osd, struct MOSDBackoff *m)
4305 struct ceph_spg_mapping *spg;
4306 struct ceph_osd_backoff *backoff;
4307 struct ceph_msg *msg;
4309 dout("%s osd%d spgid %llu.%xs%d id %llu\n", __func__, osd->o_osd,
4310 m->spgid.pgid.pool, m->spgid.pgid.seed, m->spgid.shard, m->id);
4312 spg = lookup_spg_mapping(&osd->o_backoff_mappings, &m->spgid);
4314 spg = alloc_spg_mapping();
4316 pr_err("%s failed to allocate spg\n", __func__);
4319 spg->spgid = m->spgid; /* struct */
4320 insert_spg_mapping(&osd->o_backoff_mappings, spg);
4323 backoff = alloc_backoff();
4325 pr_err("%s failed to allocate backoff\n", __func__);
4328 backoff->spgid = m->spgid; /* struct */
4329 backoff->id = m->id;
4330 backoff->begin = m->begin;
4331 m->begin = NULL; /* backoff now owns this */
4332 backoff->end = m->end;
4333 m->end = NULL; /* ditto */
4335 insert_backoff(&spg->backoffs, backoff);
4336 insert_backoff_by_id(&osd->o_backoffs_by_id, backoff);
4339 * Ack with original backoff's epoch so that the OSD can
4340 * discard this if there was a PG split.
4342 msg = create_backoff_message(backoff, m->map_epoch);
4344 pr_err("%s failed to allocate msg\n", __func__);
4347 ceph_con_send(&osd->o_con, msg);
4350 static bool target_contained_by(const struct ceph_osd_request_target *t,
4351 const struct ceph_hobject_id *begin,
4352 const struct ceph_hobject_id *end)
4354 struct ceph_hobject_id hoid;
4357 hoid_fill_from_target(&hoid, t);
4358 cmp = hoid_compare(&hoid, begin);
4359 return !cmp || (cmp > 0 && hoid_compare(&hoid, end) < 0);
4362 static void handle_backoff_unblock(struct ceph_osd *osd,
4363 const struct MOSDBackoff *m)
4365 struct ceph_spg_mapping *spg;
4366 struct ceph_osd_backoff *backoff;
4369 dout("%s osd%d spgid %llu.%xs%d id %llu\n", __func__, osd->o_osd,
4370 m->spgid.pgid.pool, m->spgid.pgid.seed, m->spgid.shard, m->id);
4372 backoff = lookup_backoff_by_id(&osd->o_backoffs_by_id, m->id);
4374 pr_err("%s osd%d spgid %llu.%xs%d id %llu backoff dne\n",
4375 __func__, osd->o_osd, m->spgid.pgid.pool,
4376 m->spgid.pgid.seed, m->spgid.shard, m->id);
4380 if (hoid_compare(backoff->begin, m->begin) &&
4381 hoid_compare(backoff->end, m->end)) {
4382 pr_err("%s osd%d spgid %llu.%xs%d id %llu bad range?\n",
4383 __func__, osd->o_osd, m->spgid.pgid.pool,
4384 m->spgid.pgid.seed, m->spgid.shard, m->id);
4385 /* unblock it anyway... */
4388 spg = lookup_spg_mapping(&osd->o_backoff_mappings, &backoff->spgid);
4391 erase_backoff(&spg->backoffs, backoff);
4392 erase_backoff_by_id(&osd->o_backoffs_by_id, backoff);
4393 free_backoff(backoff);
4395 if (RB_EMPTY_ROOT(&spg->backoffs)) {
4396 erase_spg_mapping(&osd->o_backoff_mappings, spg);
4397 free_spg_mapping(spg);
4400 for (n = rb_first(&osd->o_requests); n; n = rb_next(n)) {
4401 struct ceph_osd_request *req =
4402 rb_entry(n, struct ceph_osd_request, r_node);
4404 if (!ceph_spg_compare(&req->r_t.spgid, &m->spgid)) {
4406 * Match against @m, not @backoff -- the PG may
4407 * have split on the OSD.
4409 if (target_contained_by(&req->r_t, m->begin, m->end)) {
4411 * If no other installed backoff applies,
4420 static void handle_backoff(struct ceph_osd *osd, struct ceph_msg *msg)
4422 struct ceph_osd_client *osdc = osd->o_osdc;
4423 struct MOSDBackoff m;
4426 down_read(&osdc->lock);
4427 if (!osd_registered(osd)) {
4428 dout("%s osd%d unknown\n", __func__, osd->o_osd);
4429 up_read(&osdc->lock);
4432 WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num));
4434 mutex_lock(&osd->lock);
4435 ret = decode_MOSDBackoff(msg, &m);
4437 pr_err("failed to decode MOSDBackoff: %d\n", ret);
4443 case CEPH_OSD_BACKOFF_OP_BLOCK:
4444 handle_backoff_block(osd, &m);
4446 case CEPH_OSD_BACKOFF_OP_UNBLOCK:
4447 handle_backoff_unblock(osd, &m);
4450 pr_err("%s osd%d unknown op %d\n", __func__, osd->o_osd, m.op);
4457 mutex_unlock(&osd->lock);
4458 up_read(&osdc->lock);
4462 * Process osd watch notifications
4464 static void handle_watch_notify(struct ceph_osd_client *osdc,
4465 struct ceph_msg *msg)
4467 void *p = msg->front.iov_base;
4468 void *const end = p + msg->front.iov_len;
4469 struct ceph_osd_linger_request *lreq;
4470 struct linger_work *lwork;
4471 u8 proto_ver, opcode;
4472 u64 cookie, notify_id;
4473 u64 notifier_id = 0;
4474 s32 return_code = 0;
4475 void *payload = NULL;
4476 u32 payload_len = 0;
4478 ceph_decode_8_safe(&p, end, proto_ver, bad);
4479 ceph_decode_8_safe(&p, end, opcode, bad);
4480 ceph_decode_64_safe(&p, end, cookie, bad);
4481 p += 8; /* skip ver */
4482 ceph_decode_64_safe(&p, end, notify_id, bad);
4484 if (proto_ver >= 1) {
4485 ceph_decode_32_safe(&p, end, payload_len, bad);
4486 ceph_decode_need(&p, end, payload_len, bad);
4491 if (le16_to_cpu(msg->hdr.version) >= 2)
4492 ceph_decode_32_safe(&p, end, return_code, bad);
4494 if (le16_to_cpu(msg->hdr.version) >= 3)
4495 ceph_decode_64_safe(&p, end, notifier_id, bad);
4497 down_read(&osdc->lock);
4498 lreq = lookup_linger_osdc(&osdc->linger_requests, cookie);
4500 dout("%s opcode %d cookie %llu dne\n", __func__, opcode,
4502 goto out_unlock_osdc;
4505 mutex_lock(&lreq->lock);
4506 dout("%s opcode %d cookie %llu lreq %p is_watch %d\n", __func__,
4507 opcode, cookie, lreq, lreq->is_watch);
4508 if (opcode == CEPH_WATCH_EVENT_DISCONNECT) {
4509 if (!lreq->last_error) {
4510 lreq->last_error = -ENOTCONN;
4511 queue_watch_error(lreq);
4513 } else if (!lreq->is_watch) {
4514 /* CEPH_WATCH_EVENT_NOTIFY_COMPLETE */
4515 if (lreq->notify_id && lreq->notify_id != notify_id) {
4516 dout("lreq %p notify_id %llu != %llu, ignoring\n", lreq,
4517 lreq->notify_id, notify_id);
4518 } else if (!completion_done(&lreq->notify_finish_wait)) {
4519 struct ceph_msg_data *data =
4520 msg->num_data_items ? &msg->data[0] : NULL;
4523 if (lreq->preply_pages) {
4524 WARN_ON(data->type !=
4525 CEPH_MSG_DATA_PAGES);
4526 *lreq->preply_pages = data->pages;
4527 *lreq->preply_len = data->length;
4528 data->own_pages = false;
4531 lreq->notify_finish_error = return_code;
4532 complete_all(&lreq->notify_finish_wait);
4535 /* CEPH_WATCH_EVENT_NOTIFY */
4536 lwork = lwork_alloc(lreq, do_watch_notify);
4538 pr_err("failed to allocate notify-lwork\n");
4539 goto out_unlock_lreq;
4542 lwork->notify.notify_id = notify_id;
4543 lwork->notify.notifier_id = notifier_id;
4544 lwork->notify.payload = payload;
4545 lwork->notify.payload_len = payload_len;
4546 lwork->notify.msg = ceph_msg_get(msg);
4551 mutex_unlock(&lreq->lock);
4553 up_read(&osdc->lock);
4557 pr_err("osdc handle_watch_notify corrupt msg\n");
4561 * Register request, send initial attempt.
4563 int ceph_osdc_start_request(struct ceph_osd_client *osdc,
4564 struct ceph_osd_request *req,
4567 down_read(&osdc->lock);
4568 submit_request(req, false);
4569 up_read(&osdc->lock);
4573 EXPORT_SYMBOL(ceph_osdc_start_request);
4576 * Unregister a registered request. The request is not completed:
4577 * ->r_result isn't set and __complete_request() isn't called.
4579 void ceph_osdc_cancel_request(struct ceph_osd_request *req)
4581 struct ceph_osd_client *osdc = req->r_osdc;
4583 down_write(&osdc->lock);
4585 cancel_request(req);
4586 up_write(&osdc->lock);
4588 EXPORT_SYMBOL(ceph_osdc_cancel_request);
4591 * @timeout: in jiffies, 0 means "wait forever"
4593 static int wait_request_timeout(struct ceph_osd_request *req,
4594 unsigned long timeout)
4598 dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
4599 left = wait_for_completion_killable_timeout(&req->r_completion,
4600 ceph_timeout_jiffies(timeout));
4602 left = left ?: -ETIMEDOUT;
4603 ceph_osdc_cancel_request(req);
4605 left = req->r_result; /* completed */
4612 * wait for a request to complete
4614 int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
4615 struct ceph_osd_request *req)
4617 return wait_request_timeout(req, 0);
4619 EXPORT_SYMBOL(ceph_osdc_wait_request);
4622 * sync - wait for all in-flight requests to flush. avoid starvation.
4624 void ceph_osdc_sync(struct ceph_osd_client *osdc)
4626 struct rb_node *n, *p;
4627 u64 last_tid = atomic64_read(&osdc->last_tid);
4630 down_read(&osdc->lock);
4631 for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
4632 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
4634 mutex_lock(&osd->lock);
4635 for (p = rb_first(&osd->o_requests); p; p = rb_next(p)) {
4636 struct ceph_osd_request *req =
4637 rb_entry(p, struct ceph_osd_request, r_node);
4639 if (req->r_tid > last_tid)
4642 if (!(req->r_flags & CEPH_OSD_FLAG_WRITE))
4645 ceph_osdc_get_request(req);
4646 mutex_unlock(&osd->lock);
4647 up_read(&osdc->lock);
4648 dout("%s waiting on req %p tid %llu last_tid %llu\n",
4649 __func__, req, req->r_tid, last_tid);
4650 wait_for_completion(&req->r_completion);
4651 ceph_osdc_put_request(req);
4655 mutex_unlock(&osd->lock);
4658 up_read(&osdc->lock);
4659 dout("%s done last_tid %llu\n", __func__, last_tid);
4661 EXPORT_SYMBOL(ceph_osdc_sync);
4663 static struct ceph_osd_request *
4664 alloc_linger_request(struct ceph_osd_linger_request *lreq)
4666 struct ceph_osd_request *req;
4668 req = ceph_osdc_alloc_request(lreq->osdc, NULL, 1, false, GFP_NOIO);
4672 ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
4673 ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
4677 static struct ceph_osd_request *
4678 alloc_watch_request(struct ceph_osd_linger_request *lreq, u8 watch_opcode)
4680 struct ceph_osd_request *req;
4682 req = alloc_linger_request(lreq);
4687 * Pass 0 for cookie because we don't know it yet, it will be
4688 * filled in by linger_submit().
4690 osd_req_op_watch_init(req, 0, 0, watch_opcode);
4692 if (ceph_osdc_alloc_messages(req, GFP_NOIO)) {
4693 ceph_osdc_put_request(req);
4701 * Returns a handle, caller owns a ref.
4703 struct ceph_osd_linger_request *
4704 ceph_osdc_watch(struct ceph_osd_client *osdc,
4705 struct ceph_object_id *oid,
4706 struct ceph_object_locator *oloc,
4707 rados_watchcb2_t wcb,
4708 rados_watcherrcb_t errcb,
4711 struct ceph_osd_linger_request *lreq;
4714 lreq = linger_alloc(osdc);
4716 return ERR_PTR(-ENOMEM);
4718 lreq->is_watch = true;
4720 lreq->errcb = errcb;
4722 lreq->watch_valid_thru = jiffies;
4724 ceph_oid_copy(&lreq->t.base_oid, oid);
4725 ceph_oloc_copy(&lreq->t.base_oloc, oloc);
4726 lreq->t.flags = CEPH_OSD_FLAG_WRITE;
4727 ktime_get_real_ts64(&lreq->mtime);
4729 lreq->reg_req = alloc_watch_request(lreq, CEPH_OSD_WATCH_OP_WATCH);
4730 if (!lreq->reg_req) {
4735 lreq->ping_req = alloc_watch_request(lreq, CEPH_OSD_WATCH_OP_PING);
4736 if (!lreq->ping_req) {
4741 linger_submit(lreq);
4742 ret = linger_reg_commit_wait(lreq);
4744 linger_cancel(lreq);
4752 return ERR_PTR(ret);
4754 EXPORT_SYMBOL(ceph_osdc_watch);
4759 * Times out after mount_timeout to preserve rbd unmap behaviour
4760 * introduced in 2894e1d76974 ("rbd: timeout watch teardown on unmap
4761 * with mount_timeout").
4763 int ceph_osdc_unwatch(struct ceph_osd_client *osdc,
4764 struct ceph_osd_linger_request *lreq)
4766 struct ceph_options *opts = osdc->client->options;
4767 struct ceph_osd_request *req;
4770 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
4774 ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
4775 ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
4776 req->r_flags = CEPH_OSD_FLAG_WRITE;
4777 ktime_get_real_ts64(&req->r_mtime);
4778 osd_req_op_watch_init(req, 0, lreq->linger_id,
4779 CEPH_OSD_WATCH_OP_UNWATCH);
4781 ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
4785 ceph_osdc_start_request(osdc, req, false);
4786 linger_cancel(lreq);
4788 ret = wait_request_timeout(req, opts->mount_timeout);
4791 ceph_osdc_put_request(req);
4794 EXPORT_SYMBOL(ceph_osdc_unwatch);
4796 static int osd_req_op_notify_ack_init(struct ceph_osd_request *req, int which,
4797 u64 notify_id, u64 cookie, void *payload,
4800 struct ceph_osd_req_op *op;
4801 struct ceph_pagelist *pl;
4804 op = _osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY_ACK, 0);
4806 pl = ceph_pagelist_alloc(GFP_NOIO);
4810 ret = ceph_pagelist_encode_64(pl, notify_id);
4811 ret |= ceph_pagelist_encode_64(pl, cookie);
4813 ret |= ceph_pagelist_encode_32(pl, payload_len);
4814 ret |= ceph_pagelist_append(pl, payload, payload_len);
4816 ret |= ceph_pagelist_encode_32(pl, 0);
4819 ceph_pagelist_release(pl);
4823 ceph_osd_data_pagelist_init(&op->notify_ack.request_data, pl);
4824 op->indata_len = pl->length;
4828 int ceph_osdc_notify_ack(struct ceph_osd_client *osdc,
4829 struct ceph_object_id *oid,
4830 struct ceph_object_locator *oloc,
4836 struct ceph_osd_request *req;
4839 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
4843 ceph_oid_copy(&req->r_base_oid, oid);
4844 ceph_oloc_copy(&req->r_base_oloc, oloc);
4845 req->r_flags = CEPH_OSD_FLAG_READ;
4847 ret = osd_req_op_notify_ack_init(req, 0, notify_id, cookie, payload,
4852 ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
4856 ceph_osdc_start_request(osdc, req, false);
4857 ret = ceph_osdc_wait_request(osdc, req);
4860 ceph_osdc_put_request(req);
4863 EXPORT_SYMBOL(ceph_osdc_notify_ack);
4865 static int osd_req_op_notify_init(struct ceph_osd_request *req, int which,
4866 u64 cookie, u32 prot_ver, u32 timeout,
4867 void *payload, u32 payload_len)
4869 struct ceph_osd_req_op *op;
4870 struct ceph_pagelist *pl;
4873 op = _osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY, 0);
4874 op->notify.cookie = cookie;
4876 pl = ceph_pagelist_alloc(GFP_NOIO);
4880 ret = ceph_pagelist_encode_32(pl, 1); /* prot_ver */
4881 ret |= ceph_pagelist_encode_32(pl, timeout);
4882 ret |= ceph_pagelist_encode_32(pl, payload_len);
4883 ret |= ceph_pagelist_append(pl, payload, payload_len);
4885 ceph_pagelist_release(pl);
4889 ceph_osd_data_pagelist_init(&op->notify.request_data, pl);
4890 op->indata_len = pl->length;
4895 * @timeout: in seconds
4897 * @preply_{pages,len} are initialized both on success and error.
4898 * The caller is responsible for:
4900 * ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len))
4902 int ceph_osdc_notify(struct ceph_osd_client *osdc,
4903 struct ceph_object_id *oid,
4904 struct ceph_object_locator *oloc,
4908 struct page ***preply_pages,
4911 struct ceph_osd_linger_request *lreq;
4912 struct page **pages;
4917 *preply_pages = NULL;
4921 lreq = linger_alloc(osdc);
4925 lreq->preply_pages = preply_pages;
4926 lreq->preply_len = preply_len;
4928 ceph_oid_copy(&lreq->t.base_oid, oid);
4929 ceph_oloc_copy(&lreq->t.base_oloc, oloc);
4930 lreq->t.flags = CEPH_OSD_FLAG_READ;
4932 lreq->reg_req = alloc_linger_request(lreq);
4933 if (!lreq->reg_req) {
4939 * Pass 0 for cookie because we don't know it yet, it will be
4940 * filled in by linger_submit().
4942 ret = osd_req_op_notify_init(lreq->reg_req, 0, 0, 1, timeout,
4943 payload, payload_len);
4948 pages = ceph_alloc_page_vector(1, GFP_NOIO);
4949 if (IS_ERR(pages)) {
4950 ret = PTR_ERR(pages);
4953 ceph_osd_data_pages_init(osd_req_op_data(lreq->reg_req, 0, notify,
4955 pages, PAGE_SIZE, 0, false, true);
4957 ret = ceph_osdc_alloc_messages(lreq->reg_req, GFP_NOIO);
4961 linger_submit(lreq);
4962 ret = linger_reg_commit_wait(lreq);
4964 ret = linger_notify_finish_wait(lreq);
4966 dout("lreq %p failed to initiate notify %d\n", lreq, ret);
4968 linger_cancel(lreq);
4973 EXPORT_SYMBOL(ceph_osdc_notify);
4976 * Return the number of milliseconds since the watch was last
4977 * confirmed, or an error. If there is an error, the watch is no
4978 * longer valid, and should be destroyed with ceph_osdc_unwatch().
4980 int ceph_osdc_watch_check(struct ceph_osd_client *osdc,
4981 struct ceph_osd_linger_request *lreq)
4983 unsigned long stamp, age;
4986 down_read(&osdc->lock);
4987 mutex_lock(&lreq->lock);
4988 stamp = lreq->watch_valid_thru;
4989 if (!list_empty(&lreq->pending_lworks)) {
4990 struct linger_work *lwork =
4991 list_first_entry(&lreq->pending_lworks,
4995 if (time_before(lwork->queued_stamp, stamp))
4996 stamp = lwork->queued_stamp;
4998 age = jiffies - stamp;
4999 dout("%s lreq %p linger_id %llu age %lu last_error %d\n", __func__,
5000 lreq, lreq->linger_id, age, lreq->last_error);
5001 /* we are truncating to msecs, so return a safe upper bound */
5002 ret = lreq->last_error ?: 1 + jiffies_to_msecs(age);
5004 mutex_unlock(&lreq->lock);
5005 up_read(&osdc->lock);
5009 static int decode_watcher(void **p, void *end, struct ceph_watch_item *item)
5015 ret = ceph_start_decoding(p, end, 2, "watch_item_t",
5016 &struct_v, &struct_len);
5021 ceph_decode_copy_safe(p, end, &item->name, sizeof(item->name), bad);
5022 ceph_decode_64_safe(p, end, item->cookie, bad);
5023 ceph_decode_skip_32(p, end, bad); /* skip timeout seconds */
5025 if (struct_v >= 2) {
5026 ret = ceph_decode_entity_addr(p, end, &item->addr);
5033 dout("%s %s%llu cookie %llu addr %s\n", __func__,
5034 ENTITY_NAME(item->name), item->cookie,
5035 ceph_pr_addr(&item->addr));
5040 static int decode_watchers(void **p, void *end,
5041 struct ceph_watch_item **watchers,
5049 ret = ceph_start_decoding(p, end, 1, "obj_list_watch_response_t",
5050 &struct_v, &struct_len);
5054 *num_watchers = ceph_decode_32(p);
5055 *watchers = kcalloc(*num_watchers, sizeof(**watchers), GFP_NOIO);
5059 for (i = 0; i < *num_watchers; i++) {
5060 ret = decode_watcher(p, end, *watchers + i);
5071 * On success, the caller is responsible for:
5075 int ceph_osdc_list_watchers(struct ceph_osd_client *osdc,
5076 struct ceph_object_id *oid,
5077 struct ceph_object_locator *oloc,
5078 struct ceph_watch_item **watchers,
5081 struct ceph_osd_request *req;
5082 struct page **pages;
5085 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
5089 ceph_oid_copy(&req->r_base_oid, oid);
5090 ceph_oloc_copy(&req->r_base_oloc, oloc);
5091 req->r_flags = CEPH_OSD_FLAG_READ;
5093 pages = ceph_alloc_page_vector(1, GFP_NOIO);
5094 if (IS_ERR(pages)) {
5095 ret = PTR_ERR(pages);
5099 osd_req_op_init(req, 0, CEPH_OSD_OP_LIST_WATCHERS, 0);
5100 ceph_osd_data_pages_init(osd_req_op_data(req, 0, list_watchers,
5102 pages, PAGE_SIZE, 0, false, true);
5104 ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
5108 ceph_osdc_start_request(osdc, req, false);
5109 ret = ceph_osdc_wait_request(osdc, req);
5111 void *p = page_address(pages[0]);
5112 void *const end = p + req->r_ops[0].outdata_len;
5114 ret = decode_watchers(&p, end, watchers, num_watchers);
5118 ceph_osdc_put_request(req);
5121 EXPORT_SYMBOL(ceph_osdc_list_watchers);
5124 * Call all pending notify callbacks - for use after a watch is
5125 * unregistered, to make sure no more callbacks for it will be invoked
5127 void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc)
5129 dout("%s osdc %p\n", __func__, osdc);
5130 flush_workqueue(osdc->notify_wq);
5132 EXPORT_SYMBOL(ceph_osdc_flush_notifies);
5134 void ceph_osdc_maybe_request_map(struct ceph_osd_client *osdc)
5136 down_read(&osdc->lock);
5137 maybe_request_map(osdc);
5138 up_read(&osdc->lock);
5140 EXPORT_SYMBOL(ceph_osdc_maybe_request_map);
5143 * Execute an OSD class method on an object.
5145 * @flags: CEPH_OSD_FLAG_*
5146 * @resp_len: in/out param for reply length
5148 int ceph_osdc_call(struct ceph_osd_client *osdc,
5149 struct ceph_object_id *oid,
5150 struct ceph_object_locator *oloc,
5151 const char *class, const char *method,
5153 struct page *req_page, size_t req_len,
5154 struct page **resp_pages, size_t *resp_len)
5156 struct ceph_osd_request *req;
5159 if (req_len > PAGE_SIZE)
5162 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
5166 ceph_oid_copy(&req->r_base_oid, oid);
5167 ceph_oloc_copy(&req->r_base_oloc, oloc);
5168 req->r_flags = flags;
5170 ret = osd_req_op_cls_init(req, 0, class, method);
5175 osd_req_op_cls_request_data_pages(req, 0, &req_page, req_len,
5178 osd_req_op_cls_response_data_pages(req, 0, resp_pages,
5179 *resp_len, 0, false, false);
5181 ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
5185 ceph_osdc_start_request(osdc, req, false);
5186 ret = ceph_osdc_wait_request(osdc, req);
5188 ret = req->r_ops[0].rval;
5190 *resp_len = req->r_ops[0].outdata_len;
5194 ceph_osdc_put_request(req);
5197 EXPORT_SYMBOL(ceph_osdc_call);
5200 * reset all osd connections
5202 void ceph_osdc_reopen_osds(struct ceph_osd_client *osdc)
5206 down_write(&osdc->lock);
5207 for (n = rb_first(&osdc->osds); n; ) {
5208 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
5211 if (!reopen_osd(osd))
5212 kick_osd_requests(osd);
5214 up_write(&osdc->lock);
5220 int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
5225 osdc->client = client;
5226 init_rwsem(&osdc->lock);
5227 osdc->osds = RB_ROOT;
5228 INIT_LIST_HEAD(&osdc->osd_lru);
5229 spin_lock_init(&osdc->osd_lru_lock);
5230 osd_init(&osdc->homeless_osd);
5231 osdc->homeless_osd.o_osdc = osdc;
5232 osdc->homeless_osd.o_osd = CEPH_HOMELESS_OSD;
5233 osdc->last_linger_id = CEPH_LINGER_ID_START;
5234 osdc->linger_requests = RB_ROOT;
5235 osdc->map_checks = RB_ROOT;
5236 osdc->linger_map_checks = RB_ROOT;
5237 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
5238 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
5241 osdc->osdmap = ceph_osdmap_alloc();
5245 osdc->req_mempool = mempool_create_slab_pool(10,
5246 ceph_osd_request_cache);
5247 if (!osdc->req_mempool)
5250 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
5251 PAGE_SIZE, CEPH_OSD_SLAB_OPS, 10, "osd_op");
5254 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
5255 PAGE_SIZE, CEPH_OSD_SLAB_OPS, 10,
5261 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
5262 if (!osdc->notify_wq)
5263 goto out_msgpool_reply;
5265 osdc->completion_wq = create_singlethread_workqueue("ceph-completion");
5266 if (!osdc->completion_wq)
5269 schedule_delayed_work(&osdc->timeout_work,
5270 osdc->client->options->osd_keepalive_timeout);
5271 schedule_delayed_work(&osdc->osds_timeout_work,
5272 round_jiffies_relative(osdc->client->options->osd_idle_ttl));
5277 destroy_workqueue(osdc->notify_wq);
5279 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
5281 ceph_msgpool_destroy(&osdc->msgpool_op);
5283 mempool_destroy(osdc->req_mempool);
5285 ceph_osdmap_destroy(osdc->osdmap);
5290 void ceph_osdc_stop(struct ceph_osd_client *osdc)
5292 destroy_workqueue(osdc->completion_wq);
5293 destroy_workqueue(osdc->notify_wq);
5294 cancel_delayed_work_sync(&osdc->timeout_work);
5295 cancel_delayed_work_sync(&osdc->osds_timeout_work);
5297 down_write(&osdc->lock);
5298 while (!RB_EMPTY_ROOT(&osdc->osds)) {
5299 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
5300 struct ceph_osd, o_node);
5303 up_write(&osdc->lock);
5304 WARN_ON(refcount_read(&osdc->homeless_osd.o_ref) != 1);
5305 osd_cleanup(&osdc->homeless_osd);
5307 WARN_ON(!list_empty(&osdc->osd_lru));
5308 WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_requests));
5309 WARN_ON(!RB_EMPTY_ROOT(&osdc->map_checks));
5310 WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_map_checks));
5311 WARN_ON(atomic_read(&osdc->num_requests));
5312 WARN_ON(atomic_read(&osdc->num_homeless));
5314 ceph_osdmap_destroy(osdc->osdmap);
5315 mempool_destroy(osdc->req_mempool);
5316 ceph_msgpool_destroy(&osdc->msgpool_op);
5317 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
5320 static int osd_req_op_copy_from_init(struct ceph_osd_request *req,
5321 u64 src_snapid, u64 src_version,
5322 struct ceph_object_id *src_oid,
5323 struct ceph_object_locator *src_oloc,
5324 u32 src_fadvise_flags,
5325 u32 dst_fadvise_flags,
5326 u32 truncate_seq, u64 truncate_size,
5329 struct ceph_osd_req_op *op;
5330 struct page **pages;
5333 pages = ceph_alloc_page_vector(1, GFP_KERNEL);
5335 return PTR_ERR(pages);
5337 op = _osd_req_op_init(req, 0, CEPH_OSD_OP_COPY_FROM2,
5339 op->copy_from.snapid = src_snapid;
5340 op->copy_from.src_version = src_version;
5341 op->copy_from.flags = copy_from_flags;
5342 op->copy_from.src_fadvise_flags = src_fadvise_flags;
5344 p = page_address(pages[0]);
5345 end = p + PAGE_SIZE;
5346 ceph_encode_string(&p, end, src_oid->name, src_oid->name_len);
5347 encode_oloc(&p, end, src_oloc);
5348 ceph_encode_32(&p, truncate_seq);
5349 ceph_encode_64(&p, truncate_size);
5350 op->indata_len = PAGE_SIZE - (end - p);
5352 ceph_osd_data_pages_init(&op->copy_from.osd_data, pages,
5353 op->indata_len, 0, false, true);
5357 int ceph_osdc_copy_from(struct ceph_osd_client *osdc,
5358 u64 src_snapid, u64 src_version,
5359 struct ceph_object_id *src_oid,
5360 struct ceph_object_locator *src_oloc,
5361 u32 src_fadvise_flags,
5362 struct ceph_object_id *dst_oid,
5363 struct ceph_object_locator *dst_oloc,
5364 u32 dst_fadvise_flags,
5365 u32 truncate_seq, u64 truncate_size,
5368 struct ceph_osd_request *req;
5371 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_KERNEL);
5375 req->r_flags = CEPH_OSD_FLAG_WRITE;
5377 ceph_oloc_copy(&req->r_t.base_oloc, dst_oloc);
5378 ceph_oid_copy(&req->r_t.base_oid, dst_oid);
5380 ret = osd_req_op_copy_from_init(req, src_snapid, src_version, src_oid,
5381 src_oloc, src_fadvise_flags,
5382 dst_fadvise_flags, truncate_seq,
5383 truncate_size, copy_from_flags);
5387 ret = ceph_osdc_alloc_messages(req, GFP_KERNEL);
5391 ceph_osdc_start_request(osdc, req, false);
5392 ret = ceph_osdc_wait_request(osdc, req);
5395 ceph_osdc_put_request(req);
5398 EXPORT_SYMBOL(ceph_osdc_copy_from);
5400 int __init ceph_osdc_setup(void)
5402 size_t size = sizeof(struct ceph_osd_request) +
5403 CEPH_OSD_SLAB_OPS * sizeof(struct ceph_osd_req_op);
5405 BUG_ON(ceph_osd_request_cache);
5406 ceph_osd_request_cache = kmem_cache_create("ceph_osd_request", size,
5409 return ceph_osd_request_cache ? 0 : -ENOMEM;
5412 void ceph_osdc_cleanup(void)
5414 BUG_ON(!ceph_osd_request_cache);
5415 kmem_cache_destroy(ceph_osd_request_cache);
5416 ceph_osd_request_cache = NULL;
5420 * handle incoming message
5422 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
5424 struct ceph_osd *osd = con->private;
5425 struct ceph_osd_client *osdc = osd->o_osdc;
5426 int type = le16_to_cpu(msg->hdr.type);
5429 case CEPH_MSG_OSD_MAP:
5430 ceph_osdc_handle_map(osdc, msg);
5432 case CEPH_MSG_OSD_OPREPLY:
5433 handle_reply(osd, msg);
5435 case CEPH_MSG_OSD_BACKOFF:
5436 handle_backoff(osd, msg);
5438 case CEPH_MSG_WATCH_NOTIFY:
5439 handle_watch_notify(osdc, msg);
5443 pr_err("received unknown message type %d %s\n", type,
5444 ceph_msg_type_name(type));
5451 * Lookup and return message for incoming reply. Don't try to do
5452 * anything about a larger than preallocated data portion of the
5453 * message at the moment - for now, just skip the message.
5455 static struct ceph_msg *get_reply(struct ceph_connection *con,
5456 struct ceph_msg_header *hdr,
5459 struct ceph_osd *osd = con->private;
5460 struct ceph_osd_client *osdc = osd->o_osdc;
5461 struct ceph_msg *m = NULL;
5462 struct ceph_osd_request *req;
5463 int front_len = le32_to_cpu(hdr->front_len);
5464 int data_len = le32_to_cpu(hdr->data_len);
5465 u64 tid = le64_to_cpu(hdr->tid);
5467 down_read(&osdc->lock);
5468 if (!osd_registered(osd)) {
5469 dout("%s osd%d unknown, skipping\n", __func__, osd->o_osd);
5471 goto out_unlock_osdc;
5473 WARN_ON(osd->o_osd != le64_to_cpu(hdr->src.num));
5475 mutex_lock(&osd->lock);
5476 req = lookup_request(&osd->o_requests, tid);
5478 dout("%s osd%d tid %llu unknown, skipping\n", __func__,
5481 goto out_unlock_session;
5484 ceph_msg_revoke_incoming(req->r_reply);
5486 if (front_len > req->r_reply->front_alloc_len) {
5487 pr_warn("%s osd%d tid %llu front %d > preallocated %d\n",
5488 __func__, osd->o_osd, req->r_tid, front_len,
5489 req->r_reply->front_alloc_len);
5490 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS,
5493 goto out_unlock_session;
5494 ceph_msg_put(req->r_reply);
5498 if (data_len > req->r_reply->data_length) {
5499 pr_warn("%s osd%d tid %llu data %d > preallocated %zu, skipping\n",
5500 __func__, osd->o_osd, req->r_tid, data_len,
5501 req->r_reply->data_length);
5504 goto out_unlock_session;
5507 m = ceph_msg_get(req->r_reply);
5508 dout("get_reply tid %lld %p\n", tid, m);
5511 mutex_unlock(&osd->lock);
5513 up_read(&osdc->lock);
5517 static struct ceph_msg *alloc_msg_with_page_vector(struct ceph_msg_header *hdr)
5520 int type = le16_to_cpu(hdr->type);
5521 u32 front_len = le32_to_cpu(hdr->front_len);
5522 u32 data_len = le32_to_cpu(hdr->data_len);
5524 m = ceph_msg_new2(type, front_len, 1, GFP_NOIO, false);
5529 struct page **pages;
5531 pages = ceph_alloc_page_vector(calc_pages_for(0, data_len),
5533 if (IS_ERR(pages)) {
5538 ceph_msg_data_add_pages(m, pages, data_len, 0, true);
5544 static struct ceph_msg *alloc_msg(struct ceph_connection *con,
5545 struct ceph_msg_header *hdr,
5548 struct ceph_osd *osd = con->private;
5549 int type = le16_to_cpu(hdr->type);
5553 case CEPH_MSG_OSD_MAP:
5554 case CEPH_MSG_OSD_BACKOFF:
5555 case CEPH_MSG_WATCH_NOTIFY:
5556 return alloc_msg_with_page_vector(hdr);
5557 case CEPH_MSG_OSD_OPREPLY:
5558 return get_reply(con, hdr, skip);
5560 pr_warn("%s osd%d unknown msg type %d, skipping\n", __func__,
5568 * Wrappers to refcount containing ceph_osd struct
5570 static struct ceph_connection *get_osd_con(struct ceph_connection *con)
5572 struct ceph_osd *osd = con->private;
5578 static void put_osd_con(struct ceph_connection *con)
5580 struct ceph_osd *osd = con->private;
5588 * Note: returned pointer is the address of a structure that's
5589 * managed separately. Caller must *not* attempt to free it.
5591 static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
5592 int *proto, int force_new)
5594 struct ceph_osd *o = con->private;
5595 struct ceph_osd_client *osdc = o->o_osdc;
5596 struct ceph_auth_client *ac = osdc->client->monc.auth;
5597 struct ceph_auth_handshake *auth = &o->o_auth;
5599 if (force_new && auth->authorizer) {
5600 ceph_auth_destroy_authorizer(auth->authorizer);
5601 auth->authorizer = NULL;
5603 if (!auth->authorizer) {
5604 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
5607 return ERR_PTR(ret);
5609 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
5612 return ERR_PTR(ret);
5614 *proto = ac->protocol;
5619 static int add_authorizer_challenge(struct ceph_connection *con,
5620 void *challenge_buf, int challenge_buf_len)
5622 struct ceph_osd *o = con->private;
5623 struct ceph_osd_client *osdc = o->o_osdc;
5624 struct ceph_auth_client *ac = osdc->client->monc.auth;
5626 return ceph_auth_add_authorizer_challenge(ac, o->o_auth.authorizer,
5627 challenge_buf, challenge_buf_len);
5630 static int verify_authorizer_reply(struct ceph_connection *con)
5632 struct ceph_osd *o = con->private;
5633 struct ceph_osd_client *osdc = o->o_osdc;
5634 struct ceph_auth_client *ac = osdc->client->monc.auth;
5636 return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer);
5639 static int invalidate_authorizer(struct ceph_connection *con)
5641 struct ceph_osd *o = con->private;
5642 struct ceph_osd_client *osdc = o->o_osdc;
5643 struct ceph_auth_client *ac = osdc->client->monc.auth;
5645 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
5646 return ceph_monc_validate_auth(&osdc->client->monc);
5649 static void osd_reencode_message(struct ceph_msg *msg)
5651 int type = le16_to_cpu(msg->hdr.type);
5653 if (type == CEPH_MSG_OSD_OP)
5654 encode_request_finish(msg);
5657 static int osd_sign_message(struct ceph_msg *msg)
5659 struct ceph_osd *o = msg->con->private;
5660 struct ceph_auth_handshake *auth = &o->o_auth;
5662 return ceph_auth_sign_message(auth, msg);
5665 static int osd_check_message_signature(struct ceph_msg *msg)
5667 struct ceph_osd *o = msg->con->private;
5668 struct ceph_auth_handshake *auth = &o->o_auth;
5670 return ceph_auth_check_message_signature(auth, msg);
5673 static const struct ceph_connection_operations osd_con_ops = {
5676 .dispatch = dispatch,
5677 .get_authorizer = get_authorizer,
5678 .add_authorizer_challenge = add_authorizer_challenge,
5679 .verify_authorizer_reply = verify_authorizer_reply,
5680 .invalidate_authorizer = invalidate_authorizer,
5681 .alloc_msg = alloc_msg,
5682 .reencode_message = osd_reencode_message,
5683 .sign_message = osd_sign_message,
5684 .check_message_signature = osd_check_message_signature,