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;
449 dest->flags = src->flags;
450 dest->paused = src->paused;
452 dest->epoch = src->epoch;
453 dest->last_force_resend = src->last_force_resend;
455 dest->osd = src->osd;
458 static void target_destroy(struct ceph_osd_request_target *t)
460 ceph_oid_destroy(&t->base_oid);
461 ceph_oloc_destroy(&t->base_oloc);
462 ceph_oid_destroy(&t->target_oid);
463 ceph_oloc_destroy(&t->target_oloc);
469 static void request_release_checks(struct ceph_osd_request *req)
471 WARN_ON(!RB_EMPTY_NODE(&req->r_node));
472 WARN_ON(!RB_EMPTY_NODE(&req->r_mc_node));
473 WARN_ON(!list_empty(&req->r_private_item));
477 static void ceph_osdc_release_request(struct kref *kref)
479 struct ceph_osd_request *req = container_of(kref,
480 struct ceph_osd_request, r_kref);
483 dout("%s %p (r_request %p r_reply %p)\n", __func__, req,
484 req->r_request, req->r_reply);
485 request_release_checks(req);
488 ceph_msg_put(req->r_request);
490 ceph_msg_put(req->r_reply);
492 for (which = 0; which < req->r_num_ops; which++)
493 osd_req_op_data_release(req, which);
495 target_destroy(&req->r_t);
496 ceph_put_snap_context(req->r_snapc);
499 mempool_free(req, req->r_osdc->req_mempool);
500 else if (req->r_num_ops <= CEPH_OSD_SLAB_OPS)
501 kmem_cache_free(ceph_osd_request_cache, req);
506 void ceph_osdc_get_request(struct ceph_osd_request *req)
508 dout("%s %p (was %d)\n", __func__, req,
509 kref_read(&req->r_kref));
510 kref_get(&req->r_kref);
512 EXPORT_SYMBOL(ceph_osdc_get_request);
514 void ceph_osdc_put_request(struct ceph_osd_request *req)
517 dout("%s %p (was %d)\n", __func__, req,
518 kref_read(&req->r_kref));
519 kref_put(&req->r_kref, ceph_osdc_release_request);
522 EXPORT_SYMBOL(ceph_osdc_put_request);
524 static void request_init(struct ceph_osd_request *req)
526 /* req only, each op is zeroed in _osd_req_op_init() */
527 memset(req, 0, sizeof(*req));
529 kref_init(&req->r_kref);
530 init_completion(&req->r_completion);
531 RB_CLEAR_NODE(&req->r_node);
532 RB_CLEAR_NODE(&req->r_mc_node);
533 INIT_LIST_HEAD(&req->r_private_item);
535 target_init(&req->r_t);
539 * This is ugly, but it allows us to reuse linger registration and ping
540 * requests, keeping the structure of the code around send_linger{_ping}()
541 * reasonable. Setting up a min_nr=2 mempool for each linger request
542 * and dealing with copying ops (this blasts req only, watch op remains
543 * intact) isn't any better.
545 static void request_reinit(struct ceph_osd_request *req)
547 struct ceph_osd_client *osdc = req->r_osdc;
548 bool mempool = req->r_mempool;
549 unsigned int num_ops = req->r_num_ops;
550 u64 snapid = req->r_snapid;
551 struct ceph_snap_context *snapc = req->r_snapc;
552 bool linger = req->r_linger;
553 struct ceph_msg *request_msg = req->r_request;
554 struct ceph_msg *reply_msg = req->r_reply;
556 dout("%s req %p\n", __func__, req);
557 WARN_ON(kref_read(&req->r_kref) != 1);
558 request_release_checks(req);
560 WARN_ON(kref_read(&request_msg->kref) != 1);
561 WARN_ON(kref_read(&reply_msg->kref) != 1);
562 target_destroy(&req->r_t);
566 req->r_mempool = mempool;
567 req->r_num_ops = num_ops;
568 req->r_snapid = snapid;
569 req->r_snapc = snapc;
570 req->r_linger = linger;
571 req->r_request = request_msg;
572 req->r_reply = reply_msg;
575 struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
576 struct ceph_snap_context *snapc,
577 unsigned int num_ops,
581 struct ceph_osd_request *req;
584 BUG_ON(num_ops > CEPH_OSD_SLAB_OPS);
585 req = mempool_alloc(osdc->req_mempool, gfp_flags);
586 } else if (num_ops <= CEPH_OSD_SLAB_OPS) {
587 req = kmem_cache_alloc(ceph_osd_request_cache, gfp_flags);
589 BUG_ON(num_ops > CEPH_OSD_MAX_OPS);
590 req = kmalloc(struct_size(req, r_ops, num_ops), gfp_flags);
597 req->r_mempool = use_mempool;
598 req->r_num_ops = num_ops;
599 req->r_snapid = CEPH_NOSNAP;
600 req->r_snapc = ceph_get_snap_context(snapc);
602 dout("%s req %p\n", __func__, req);
605 EXPORT_SYMBOL(ceph_osdc_alloc_request);
607 static int ceph_oloc_encoding_size(const struct ceph_object_locator *oloc)
609 return 8 + 4 + 4 + 4 + (oloc->pool_ns ? oloc->pool_ns->len : 0);
612 static int __ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp,
613 int num_request_data_items,
614 int num_reply_data_items)
616 struct ceph_osd_client *osdc = req->r_osdc;
617 struct ceph_msg *msg;
620 WARN_ON(req->r_request || req->r_reply);
621 WARN_ON(ceph_oid_empty(&req->r_base_oid));
622 WARN_ON(ceph_oloc_empty(&req->r_base_oloc));
624 /* create request message */
625 msg_size = CEPH_ENCODING_START_BLK_LEN +
626 CEPH_PGID_ENCODING_LEN + 1; /* spgid */
627 msg_size += 4 + 4 + 4; /* hash, osdmap_epoch, flags */
628 msg_size += CEPH_ENCODING_START_BLK_LEN +
629 sizeof(struct ceph_osd_reqid); /* reqid */
630 msg_size += sizeof(struct ceph_blkin_trace_info); /* trace */
631 msg_size += 4 + sizeof(struct ceph_timespec); /* client_inc, mtime */
632 msg_size += CEPH_ENCODING_START_BLK_LEN +
633 ceph_oloc_encoding_size(&req->r_base_oloc); /* oloc */
634 msg_size += 4 + req->r_base_oid.name_len; /* oid */
635 msg_size += 2 + req->r_num_ops * sizeof(struct ceph_osd_op);
636 msg_size += 8; /* snapid */
637 msg_size += 8; /* snap_seq */
638 msg_size += 4 + 8 * (req->r_snapc ? req->r_snapc->num_snaps : 0);
639 msg_size += 4 + 8; /* retry_attempt, features */
642 msg = ceph_msgpool_get(&osdc->msgpool_op, msg_size,
643 num_request_data_items);
645 msg = ceph_msg_new2(CEPH_MSG_OSD_OP, msg_size,
646 num_request_data_items, gfp, true);
650 memset(msg->front.iov_base, 0, msg->front.iov_len);
651 req->r_request = msg;
653 /* create reply message */
654 msg_size = OSD_OPREPLY_FRONT_LEN;
655 msg_size += req->r_base_oid.name_len;
656 msg_size += req->r_num_ops * sizeof(struct ceph_osd_op);
659 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, msg_size,
660 num_reply_data_items);
662 msg = ceph_msg_new2(CEPH_MSG_OSD_OPREPLY, msg_size,
663 num_reply_data_items, gfp, true);
672 static bool osd_req_opcode_valid(u16 opcode)
675 #define GENERATE_CASE(op, opcode, str) case CEPH_OSD_OP_##op: return true;
676 __CEPH_FORALL_OSD_OPS(GENERATE_CASE)
683 static void get_num_data_items(struct ceph_osd_request *req,
684 int *num_request_data_items,
685 int *num_reply_data_items)
687 struct ceph_osd_req_op *op;
689 *num_request_data_items = 0;
690 *num_reply_data_items = 0;
692 for (op = req->r_ops; op != &req->r_ops[req->r_num_ops]; op++) {
695 case CEPH_OSD_OP_WRITE:
696 case CEPH_OSD_OP_WRITEFULL:
697 case CEPH_OSD_OP_SETXATTR:
698 case CEPH_OSD_OP_CMPXATTR:
699 case CEPH_OSD_OP_NOTIFY_ACK:
700 case CEPH_OSD_OP_COPY_FROM2:
701 *num_request_data_items += 1;
705 case CEPH_OSD_OP_STAT:
706 case CEPH_OSD_OP_READ:
707 case CEPH_OSD_OP_LIST_WATCHERS:
708 *num_reply_data_items += 1;
712 case CEPH_OSD_OP_NOTIFY:
713 *num_request_data_items += 1;
714 *num_reply_data_items += 1;
716 case CEPH_OSD_OP_CALL:
717 *num_request_data_items += 2;
718 *num_reply_data_items += 1;
722 WARN_ON(!osd_req_opcode_valid(op->op));
729 * oid, oloc and OSD op opcode(s) must be filled in before this function
732 int ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp)
734 int num_request_data_items, num_reply_data_items;
736 get_num_data_items(req, &num_request_data_items, &num_reply_data_items);
737 return __ceph_osdc_alloc_messages(req, gfp, num_request_data_items,
738 num_reply_data_items);
740 EXPORT_SYMBOL(ceph_osdc_alloc_messages);
743 * This is an osd op init function for opcodes that have no data or
744 * other information associated with them. It also serves as a
745 * common init routine for all the other init functions, below.
747 static struct ceph_osd_req_op *
748 _osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which,
749 u16 opcode, u32 flags)
751 struct ceph_osd_req_op *op;
753 BUG_ON(which >= osd_req->r_num_ops);
754 BUG_ON(!osd_req_opcode_valid(opcode));
756 op = &osd_req->r_ops[which];
757 memset(op, 0, sizeof (*op));
764 void osd_req_op_init(struct ceph_osd_request *osd_req,
765 unsigned int which, u16 opcode, u32 flags)
767 (void)_osd_req_op_init(osd_req, which, opcode, flags);
769 EXPORT_SYMBOL(osd_req_op_init);
771 void osd_req_op_extent_init(struct ceph_osd_request *osd_req,
772 unsigned int which, u16 opcode,
773 u64 offset, u64 length,
774 u64 truncate_size, u32 truncate_seq)
776 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
778 size_t payload_len = 0;
780 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
781 opcode != CEPH_OSD_OP_WRITEFULL && opcode != CEPH_OSD_OP_ZERO &&
782 opcode != CEPH_OSD_OP_TRUNCATE);
784 op->extent.offset = offset;
785 op->extent.length = length;
786 op->extent.truncate_size = truncate_size;
787 op->extent.truncate_seq = truncate_seq;
788 if (opcode == CEPH_OSD_OP_WRITE || opcode == CEPH_OSD_OP_WRITEFULL)
789 payload_len += length;
791 op->indata_len = payload_len;
793 EXPORT_SYMBOL(osd_req_op_extent_init);
795 void osd_req_op_extent_update(struct ceph_osd_request *osd_req,
796 unsigned int which, u64 length)
798 struct ceph_osd_req_op *op;
801 BUG_ON(which >= osd_req->r_num_ops);
802 op = &osd_req->r_ops[which];
803 previous = op->extent.length;
805 if (length == previous)
806 return; /* Nothing to do */
807 BUG_ON(length > previous);
809 op->extent.length = length;
810 if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL)
811 op->indata_len -= previous - length;
813 EXPORT_SYMBOL(osd_req_op_extent_update);
815 void osd_req_op_extent_dup_last(struct ceph_osd_request *osd_req,
816 unsigned int which, u64 offset_inc)
818 struct ceph_osd_req_op *op, *prev_op;
820 BUG_ON(which + 1 >= osd_req->r_num_ops);
822 prev_op = &osd_req->r_ops[which];
823 op = _osd_req_op_init(osd_req, which + 1, prev_op->op, prev_op->flags);
824 /* dup previous one */
825 op->indata_len = prev_op->indata_len;
826 op->outdata_len = prev_op->outdata_len;
827 op->extent = prev_op->extent;
829 op->extent.offset += offset_inc;
830 op->extent.length -= offset_inc;
832 if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL)
833 op->indata_len -= offset_inc;
835 EXPORT_SYMBOL(osd_req_op_extent_dup_last);
837 int osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
838 const char *class, const char *method)
840 struct ceph_osd_req_op *op;
841 struct ceph_pagelist *pagelist;
842 size_t payload_len = 0;
846 op = _osd_req_op_init(osd_req, which, CEPH_OSD_OP_CALL, 0);
848 pagelist = ceph_pagelist_alloc(GFP_NOFS);
852 op->cls.class_name = class;
853 size = strlen(class);
854 BUG_ON(size > (size_t) U8_MAX);
855 op->cls.class_len = size;
856 ret = ceph_pagelist_append(pagelist, class, size);
858 goto err_pagelist_free;
861 op->cls.method_name = method;
862 size = strlen(method);
863 BUG_ON(size > (size_t) U8_MAX);
864 op->cls.method_len = size;
865 ret = ceph_pagelist_append(pagelist, method, size);
867 goto err_pagelist_free;
870 osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist);
871 op->indata_len = payload_len;
875 ceph_pagelist_release(pagelist);
878 EXPORT_SYMBOL(osd_req_op_cls_init);
880 int osd_req_op_xattr_init(struct ceph_osd_request *osd_req, unsigned int which,
881 u16 opcode, const char *name, const void *value,
882 size_t size, u8 cmp_op, u8 cmp_mode)
884 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
886 struct ceph_pagelist *pagelist;
890 BUG_ON(opcode != CEPH_OSD_OP_SETXATTR && opcode != CEPH_OSD_OP_CMPXATTR);
892 pagelist = ceph_pagelist_alloc(GFP_NOFS);
896 payload_len = strlen(name);
897 op->xattr.name_len = payload_len;
898 ret = ceph_pagelist_append(pagelist, name, payload_len);
900 goto err_pagelist_free;
902 op->xattr.value_len = size;
903 ret = ceph_pagelist_append(pagelist, value, size);
905 goto err_pagelist_free;
908 op->xattr.cmp_op = cmp_op;
909 op->xattr.cmp_mode = cmp_mode;
911 ceph_osd_data_pagelist_init(&op->xattr.osd_data, pagelist);
912 op->indata_len = payload_len;
916 ceph_pagelist_release(pagelist);
919 EXPORT_SYMBOL(osd_req_op_xattr_init);
922 * @watch_opcode: CEPH_OSD_WATCH_OP_*
924 static void osd_req_op_watch_init(struct ceph_osd_request *req, int which,
925 u64 cookie, u8 watch_opcode)
927 struct ceph_osd_req_op *op;
929 op = _osd_req_op_init(req, which, CEPH_OSD_OP_WATCH, 0);
930 op->watch.cookie = cookie;
931 op->watch.op = watch_opcode;
935 void osd_req_op_alloc_hint_init(struct ceph_osd_request *osd_req,
937 u64 expected_object_size,
938 u64 expected_write_size)
940 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
941 CEPH_OSD_OP_SETALLOCHINT,
944 op->alloc_hint.expected_object_size = expected_object_size;
945 op->alloc_hint.expected_write_size = expected_write_size;
948 * CEPH_OSD_OP_SETALLOCHINT op is advisory and therefore deemed
949 * not worth a feature bit. Set FAILOK per-op flag to make
950 * sure older osds don't trip over an unsupported opcode.
952 op->flags |= CEPH_OSD_OP_FLAG_FAILOK;
954 EXPORT_SYMBOL(osd_req_op_alloc_hint_init);
956 static void ceph_osdc_msg_data_add(struct ceph_msg *msg,
957 struct ceph_osd_data *osd_data)
959 u64 length = ceph_osd_data_length(osd_data);
961 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
962 BUG_ON(length > (u64) SIZE_MAX);
964 ceph_msg_data_add_pages(msg, osd_data->pages,
965 length, osd_data->alignment);
966 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
968 ceph_msg_data_add_pagelist(msg, osd_data->pagelist);
970 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
971 ceph_msg_data_add_bio(msg, &osd_data->bio_pos, length);
973 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BVECS) {
974 ceph_msg_data_add_bvecs(msg, &osd_data->bvec_pos);
976 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
980 static u32 osd_req_encode_op(struct ceph_osd_op *dst,
981 const struct ceph_osd_req_op *src)
984 case CEPH_OSD_OP_STAT:
986 case CEPH_OSD_OP_READ:
987 case CEPH_OSD_OP_WRITE:
988 case CEPH_OSD_OP_WRITEFULL:
989 case CEPH_OSD_OP_ZERO:
990 case CEPH_OSD_OP_TRUNCATE:
991 dst->extent.offset = cpu_to_le64(src->extent.offset);
992 dst->extent.length = cpu_to_le64(src->extent.length);
993 dst->extent.truncate_size =
994 cpu_to_le64(src->extent.truncate_size);
995 dst->extent.truncate_seq =
996 cpu_to_le32(src->extent.truncate_seq);
998 case CEPH_OSD_OP_CALL:
999 dst->cls.class_len = src->cls.class_len;
1000 dst->cls.method_len = src->cls.method_len;
1001 dst->cls.indata_len = cpu_to_le32(src->cls.indata_len);
1003 case CEPH_OSD_OP_WATCH:
1004 dst->watch.cookie = cpu_to_le64(src->watch.cookie);
1005 dst->watch.ver = cpu_to_le64(0);
1006 dst->watch.op = src->watch.op;
1007 dst->watch.gen = cpu_to_le32(src->watch.gen);
1009 case CEPH_OSD_OP_NOTIFY_ACK:
1011 case CEPH_OSD_OP_NOTIFY:
1012 dst->notify.cookie = cpu_to_le64(src->notify.cookie);
1014 case CEPH_OSD_OP_LIST_WATCHERS:
1016 case CEPH_OSD_OP_SETALLOCHINT:
1017 dst->alloc_hint.expected_object_size =
1018 cpu_to_le64(src->alloc_hint.expected_object_size);
1019 dst->alloc_hint.expected_write_size =
1020 cpu_to_le64(src->alloc_hint.expected_write_size);
1022 case CEPH_OSD_OP_SETXATTR:
1023 case CEPH_OSD_OP_CMPXATTR:
1024 dst->xattr.name_len = cpu_to_le32(src->xattr.name_len);
1025 dst->xattr.value_len = cpu_to_le32(src->xattr.value_len);
1026 dst->xattr.cmp_op = src->xattr.cmp_op;
1027 dst->xattr.cmp_mode = src->xattr.cmp_mode;
1029 case CEPH_OSD_OP_CREATE:
1030 case CEPH_OSD_OP_DELETE:
1032 case CEPH_OSD_OP_COPY_FROM2:
1033 dst->copy_from.snapid = cpu_to_le64(src->copy_from.snapid);
1034 dst->copy_from.src_version =
1035 cpu_to_le64(src->copy_from.src_version);
1036 dst->copy_from.flags = src->copy_from.flags;
1037 dst->copy_from.src_fadvise_flags =
1038 cpu_to_le32(src->copy_from.src_fadvise_flags);
1041 pr_err("unsupported osd opcode %s\n",
1042 ceph_osd_op_name(src->op));
1048 dst->op = cpu_to_le16(src->op);
1049 dst->flags = cpu_to_le32(src->flags);
1050 dst->payload_len = cpu_to_le32(src->indata_len);
1052 return src->indata_len;
1056 * build new request AND message, calculate layout, and adjust file
1059 * if the file was recently truncated, we include information about its
1060 * old and new size so that the object can be updated appropriately. (we
1061 * avoid synchronously deleting truncated objects because it's slow.)
1063 struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
1064 struct ceph_file_layout *layout,
1065 struct ceph_vino vino,
1067 unsigned int which, int num_ops,
1068 int opcode, int flags,
1069 struct ceph_snap_context *snapc,
1074 struct ceph_osd_request *req;
1080 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
1081 opcode != CEPH_OSD_OP_ZERO && opcode != CEPH_OSD_OP_TRUNCATE &&
1082 opcode != CEPH_OSD_OP_CREATE && opcode != CEPH_OSD_OP_DELETE);
1084 req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool,
1091 /* calculate max write size */
1092 r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
1096 if (opcode == CEPH_OSD_OP_CREATE || opcode == CEPH_OSD_OP_DELETE) {
1097 osd_req_op_init(req, which, opcode, 0);
1099 u32 object_size = layout->object_size;
1100 u32 object_base = off - objoff;
1101 if (!(truncate_seq == 1 && truncate_size == -1ULL)) {
1102 if (truncate_size <= object_base) {
1105 truncate_size -= object_base;
1106 if (truncate_size > object_size)
1107 truncate_size = object_size;
1110 osd_req_op_extent_init(req, which, opcode, objoff, objlen,
1111 truncate_size, truncate_seq);
1114 req->r_flags = flags;
1115 req->r_base_oloc.pool = layout->pool_id;
1116 req->r_base_oloc.pool_ns = ceph_try_get_string(layout->pool_ns);
1117 ceph_oid_printf(&req->r_base_oid, "%llx.%08llx", vino.ino, objnum);
1119 req->r_snapid = vino.snap;
1120 if (flags & CEPH_OSD_FLAG_WRITE)
1121 req->r_data_offset = off;
1125 * This is a special case for ceph_writepages_start(), but it
1126 * also covers ceph_uninline_data(). If more multi-op request
1127 * use cases emerge, we will need a separate helper.
1129 r = __ceph_osdc_alloc_messages(req, GFP_NOFS, num_ops, 0);
1131 r = ceph_osdc_alloc_messages(req, GFP_NOFS);
1138 ceph_osdc_put_request(req);
1141 EXPORT_SYMBOL(ceph_osdc_new_request);
1144 * We keep osd requests in an rbtree, sorted by ->r_tid.
1146 DEFINE_RB_FUNCS(request, struct ceph_osd_request, r_tid, r_node)
1147 DEFINE_RB_FUNCS(request_mc, struct ceph_osd_request, r_tid, r_mc_node)
1150 * Call @fn on each OSD request as long as @fn returns 0.
1152 static void for_each_request(struct ceph_osd_client *osdc,
1153 int (*fn)(struct ceph_osd_request *req, void *arg),
1156 struct rb_node *n, *p;
1158 for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
1159 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
1161 for (p = rb_first(&osd->o_requests); p; ) {
1162 struct ceph_osd_request *req =
1163 rb_entry(p, struct ceph_osd_request, r_node);
1171 for (p = rb_first(&osdc->homeless_osd.o_requests); p; ) {
1172 struct ceph_osd_request *req =
1173 rb_entry(p, struct ceph_osd_request, r_node);
1181 static bool osd_homeless(struct ceph_osd *osd)
1183 return osd->o_osd == CEPH_HOMELESS_OSD;
1186 static bool osd_registered(struct ceph_osd *osd)
1188 verify_osdc_locked(osd->o_osdc);
1190 return !RB_EMPTY_NODE(&osd->o_node);
1194 * Assumes @osd is zero-initialized.
1196 static void osd_init(struct ceph_osd *osd)
1198 refcount_set(&osd->o_ref, 1);
1199 RB_CLEAR_NODE(&osd->o_node);
1200 osd->o_requests = RB_ROOT;
1201 osd->o_linger_requests = RB_ROOT;
1202 osd->o_backoff_mappings = RB_ROOT;
1203 osd->o_backoffs_by_id = RB_ROOT;
1204 INIT_LIST_HEAD(&osd->o_osd_lru);
1205 INIT_LIST_HEAD(&osd->o_keepalive_item);
1206 osd->o_incarnation = 1;
1207 mutex_init(&osd->lock);
1210 static void osd_cleanup(struct ceph_osd *osd)
1212 WARN_ON(!RB_EMPTY_NODE(&osd->o_node));
1213 WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests));
1214 WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests));
1215 WARN_ON(!RB_EMPTY_ROOT(&osd->o_backoff_mappings));
1216 WARN_ON(!RB_EMPTY_ROOT(&osd->o_backoffs_by_id));
1217 WARN_ON(!list_empty(&osd->o_osd_lru));
1218 WARN_ON(!list_empty(&osd->o_keepalive_item));
1220 if (osd->o_auth.authorizer) {
1221 WARN_ON(osd_homeless(osd));
1222 ceph_auth_destroy_authorizer(osd->o_auth.authorizer);
1227 * Track open sessions with osds.
1229 static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
1231 struct ceph_osd *osd;
1233 WARN_ON(onum == CEPH_HOMELESS_OSD);
1235 osd = kzalloc(sizeof(*osd), GFP_NOIO | __GFP_NOFAIL);
1240 ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
1245 static struct ceph_osd *get_osd(struct ceph_osd *osd)
1247 if (refcount_inc_not_zero(&osd->o_ref)) {
1248 dout("get_osd %p %d -> %d\n", osd, refcount_read(&osd->o_ref)-1,
1249 refcount_read(&osd->o_ref));
1252 dout("get_osd %p FAIL\n", osd);
1257 static void put_osd(struct ceph_osd *osd)
1259 dout("put_osd %p %d -> %d\n", osd, refcount_read(&osd->o_ref),
1260 refcount_read(&osd->o_ref) - 1);
1261 if (refcount_dec_and_test(&osd->o_ref)) {
1267 DEFINE_RB_FUNCS(osd, struct ceph_osd, o_osd, o_node)
1269 static void __move_osd_to_lru(struct ceph_osd *osd)
1271 struct ceph_osd_client *osdc = osd->o_osdc;
1273 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1274 BUG_ON(!list_empty(&osd->o_osd_lru));
1276 spin_lock(&osdc->osd_lru_lock);
1277 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
1278 spin_unlock(&osdc->osd_lru_lock);
1280 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl;
1283 static void maybe_move_osd_to_lru(struct ceph_osd *osd)
1285 if (RB_EMPTY_ROOT(&osd->o_requests) &&
1286 RB_EMPTY_ROOT(&osd->o_linger_requests))
1287 __move_osd_to_lru(osd);
1290 static void __remove_osd_from_lru(struct ceph_osd *osd)
1292 struct ceph_osd_client *osdc = osd->o_osdc;
1294 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1296 spin_lock(&osdc->osd_lru_lock);
1297 if (!list_empty(&osd->o_osd_lru))
1298 list_del_init(&osd->o_osd_lru);
1299 spin_unlock(&osdc->osd_lru_lock);
1303 * Close the connection and assign any leftover requests to the
1306 static void close_osd(struct ceph_osd *osd)
1308 struct ceph_osd_client *osdc = osd->o_osdc;
1311 verify_osdc_wrlocked(osdc);
1312 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1314 ceph_con_close(&osd->o_con);
1316 for (n = rb_first(&osd->o_requests); n; ) {
1317 struct ceph_osd_request *req =
1318 rb_entry(n, struct ceph_osd_request, r_node);
1320 n = rb_next(n); /* unlink_request() */
1322 dout(" reassigning req %p tid %llu\n", req, req->r_tid);
1323 unlink_request(osd, req);
1324 link_request(&osdc->homeless_osd, req);
1326 for (n = rb_first(&osd->o_linger_requests); n; ) {
1327 struct ceph_osd_linger_request *lreq =
1328 rb_entry(n, struct ceph_osd_linger_request, node);
1330 n = rb_next(n); /* unlink_linger() */
1332 dout(" reassigning lreq %p linger_id %llu\n", lreq,
1334 unlink_linger(osd, lreq);
1335 link_linger(&osdc->homeless_osd, lreq);
1337 clear_backoffs(osd);
1339 __remove_osd_from_lru(osd);
1340 erase_osd(&osdc->osds, osd);
1347 static int reopen_osd(struct ceph_osd *osd)
1349 struct ceph_entity_addr *peer_addr;
1351 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1353 if (RB_EMPTY_ROOT(&osd->o_requests) &&
1354 RB_EMPTY_ROOT(&osd->o_linger_requests)) {
1359 peer_addr = &osd->o_osdc->osdmap->osd_addr[osd->o_osd];
1360 if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1361 !ceph_con_opened(&osd->o_con)) {
1364 dout("osd addr hasn't changed and connection never opened, "
1365 "letting msgr retry\n");
1366 /* touch each r_stamp for handle_timeout()'s benfit */
1367 for (n = rb_first(&osd->o_requests); n; n = rb_next(n)) {
1368 struct ceph_osd_request *req =
1369 rb_entry(n, struct ceph_osd_request, r_node);
1370 req->r_stamp = jiffies;
1376 ceph_con_close(&osd->o_con);
1377 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1378 osd->o_incarnation++;
1383 static struct ceph_osd *lookup_create_osd(struct ceph_osd_client *osdc, int o,
1386 struct ceph_osd *osd;
1389 verify_osdc_wrlocked(osdc);
1391 verify_osdc_locked(osdc);
1393 if (o != CEPH_HOMELESS_OSD)
1394 osd = lookup_osd(&osdc->osds, o);
1396 osd = &osdc->homeless_osd;
1399 return ERR_PTR(-EAGAIN);
1401 osd = create_osd(osdc, o);
1402 insert_osd(&osdc->osds, osd);
1403 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd,
1404 &osdc->osdmap->osd_addr[osd->o_osd]);
1407 dout("%s osdc %p osd%d -> osd %p\n", __func__, osdc, o, osd);
1412 * Create request <-> OSD session relation.
1414 * @req has to be assigned a tid, @osd may be homeless.
1416 static void link_request(struct ceph_osd *osd, struct ceph_osd_request *req)
1418 verify_osd_locked(osd);
1419 WARN_ON(!req->r_tid || req->r_osd);
1420 dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd,
1423 if (!osd_homeless(osd))
1424 __remove_osd_from_lru(osd);
1426 atomic_inc(&osd->o_osdc->num_homeless);
1429 insert_request(&osd->o_requests, req);
1433 static void unlink_request(struct ceph_osd *osd, struct ceph_osd_request *req)
1435 verify_osd_locked(osd);
1436 WARN_ON(req->r_osd != osd);
1437 dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd,
1441 erase_request(&osd->o_requests, req);
1444 if (!osd_homeless(osd))
1445 maybe_move_osd_to_lru(osd);
1447 atomic_dec(&osd->o_osdc->num_homeless);
1450 static bool __pool_full(struct ceph_pg_pool_info *pi)
1452 return pi->flags & CEPH_POOL_FLAG_FULL;
1455 static bool have_pool_full(struct ceph_osd_client *osdc)
1459 for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) {
1460 struct ceph_pg_pool_info *pi =
1461 rb_entry(n, struct ceph_pg_pool_info, node);
1463 if (__pool_full(pi))
1470 static bool pool_full(struct ceph_osd_client *osdc, s64 pool_id)
1472 struct ceph_pg_pool_info *pi;
1474 pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id);
1478 return __pool_full(pi);
1482 * Returns whether a request should be blocked from being sent
1483 * based on the current osdmap and osd_client settings.
1485 static bool target_should_be_paused(struct ceph_osd_client *osdc,
1486 const struct ceph_osd_request_target *t,
1487 struct ceph_pg_pool_info *pi)
1489 bool pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
1490 bool pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
1491 ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
1494 WARN_ON(pi->id != t->target_oloc.pool);
1495 return ((t->flags & CEPH_OSD_FLAG_READ) && pauserd) ||
1496 ((t->flags & CEPH_OSD_FLAG_WRITE) && pausewr) ||
1497 (osdc->osdmap->epoch < osdc->epoch_barrier);
1500 enum calc_target_result {
1501 CALC_TARGET_NO_ACTION = 0,
1502 CALC_TARGET_NEED_RESEND,
1503 CALC_TARGET_POOL_DNE,
1506 static enum calc_target_result calc_target(struct ceph_osd_client *osdc,
1507 struct ceph_osd_request_target *t,
1510 struct ceph_pg_pool_info *pi;
1511 struct ceph_pg pgid, last_pgid;
1512 struct ceph_osds up, acting;
1513 bool force_resend = false;
1514 bool unpaused = false;
1515 bool legacy_change = false;
1517 bool sort_bitwise = ceph_osdmap_flag(osdc, CEPH_OSDMAP_SORTBITWISE);
1518 bool recovery_deletes = ceph_osdmap_flag(osdc,
1519 CEPH_OSDMAP_RECOVERY_DELETES);
1520 enum calc_target_result ct_res;
1522 t->epoch = osdc->osdmap->epoch;
1523 pi = ceph_pg_pool_by_id(osdc->osdmap, t->base_oloc.pool);
1525 t->osd = CEPH_HOMELESS_OSD;
1526 ct_res = CALC_TARGET_POOL_DNE;
1530 if (osdc->osdmap->epoch == pi->last_force_request_resend) {
1531 if (t->last_force_resend < pi->last_force_request_resend) {
1532 t->last_force_resend = pi->last_force_request_resend;
1533 force_resend = true;
1534 } else if (t->last_force_resend == 0) {
1535 force_resend = true;
1540 ceph_oid_copy(&t->target_oid, &t->base_oid);
1541 ceph_oloc_copy(&t->target_oloc, &t->base_oloc);
1542 if ((t->flags & CEPH_OSD_FLAG_IGNORE_OVERLAY) == 0) {
1543 if (t->flags & CEPH_OSD_FLAG_READ && pi->read_tier >= 0)
1544 t->target_oloc.pool = pi->read_tier;
1545 if (t->flags & CEPH_OSD_FLAG_WRITE && pi->write_tier >= 0)
1546 t->target_oloc.pool = pi->write_tier;
1548 pi = ceph_pg_pool_by_id(osdc->osdmap, t->target_oloc.pool);
1550 t->osd = CEPH_HOMELESS_OSD;
1551 ct_res = CALC_TARGET_POOL_DNE;
1556 __ceph_object_locator_to_pg(pi, &t->target_oid, &t->target_oloc, &pgid);
1557 last_pgid.pool = pgid.pool;
1558 last_pgid.seed = ceph_stable_mod(pgid.seed, t->pg_num, t->pg_num_mask);
1560 ceph_pg_to_up_acting_osds(osdc->osdmap, pi, &pgid, &up, &acting);
1562 ceph_is_new_interval(&t->acting,
1574 t->recovery_deletes,
1577 force_resend = true;
1579 if (t->paused && !target_should_be_paused(osdc, t, pi)) {
1583 legacy_change = ceph_pg_compare(&t->pgid, &pgid) ||
1584 ceph_osds_changed(&t->acting, &acting, any_change);
1586 split = ceph_pg_is_split(&last_pgid, t->pg_num, pi->pg_num);
1588 if (legacy_change || force_resend || split) {
1589 t->pgid = pgid; /* struct */
1590 ceph_pg_to_primary_shard(osdc->osdmap, pi, &pgid, &t->spgid);
1591 ceph_osds_copy(&t->acting, &acting);
1592 ceph_osds_copy(&t->up, &up);
1594 t->min_size = pi->min_size;
1595 t->pg_num = pi->pg_num;
1596 t->pg_num_mask = pi->pg_num_mask;
1597 t->sort_bitwise = sort_bitwise;
1598 t->recovery_deletes = recovery_deletes;
1600 t->osd = acting.primary;
1603 if (unpaused || legacy_change || force_resend || split)
1604 ct_res = CALC_TARGET_NEED_RESEND;
1606 ct_res = CALC_TARGET_NO_ACTION;
1609 dout("%s t %p -> %d%d%d%d ct_res %d osd%d\n", __func__, t, unpaused,
1610 legacy_change, force_resend, split, ct_res, t->osd);
1614 static struct ceph_spg_mapping *alloc_spg_mapping(void)
1616 struct ceph_spg_mapping *spg;
1618 spg = kmalloc(sizeof(*spg), GFP_NOIO);
1622 RB_CLEAR_NODE(&spg->node);
1623 spg->backoffs = RB_ROOT;
1627 static void free_spg_mapping(struct ceph_spg_mapping *spg)
1629 WARN_ON(!RB_EMPTY_NODE(&spg->node));
1630 WARN_ON(!RB_EMPTY_ROOT(&spg->backoffs));
1636 * rbtree of ceph_spg_mapping for handling map<spg_t, ...>, similar to
1637 * ceph_pg_mapping. Used to track OSD backoffs -- a backoff [range] is
1638 * defined only within a specific spgid; it does not pass anything to
1639 * children on split, or to another primary.
1641 DEFINE_RB_FUNCS2(spg_mapping, struct ceph_spg_mapping, spgid, ceph_spg_compare,
1642 RB_BYPTR, const struct ceph_spg *, node)
1644 static u64 hoid_get_bitwise_key(const struct ceph_hobject_id *hoid)
1646 return hoid->is_max ? 0x100000000ull : hoid->hash_reverse_bits;
1649 static void hoid_get_effective_key(const struct ceph_hobject_id *hoid,
1650 void **pkey, size_t *pkey_len)
1652 if (hoid->key_len) {
1654 *pkey_len = hoid->key_len;
1657 *pkey_len = hoid->oid_len;
1661 static int compare_names(const void *name1, size_t name1_len,
1662 const void *name2, size_t name2_len)
1666 ret = memcmp(name1, name2, min(name1_len, name2_len));
1668 if (name1_len < name2_len)
1670 else if (name1_len > name2_len)
1676 static int hoid_compare(const struct ceph_hobject_id *lhs,
1677 const struct ceph_hobject_id *rhs)
1679 void *effective_key1, *effective_key2;
1680 size_t effective_key1_len, effective_key2_len;
1683 if (lhs->is_max < rhs->is_max)
1685 if (lhs->is_max > rhs->is_max)
1688 if (lhs->pool < rhs->pool)
1690 if (lhs->pool > rhs->pool)
1693 if (hoid_get_bitwise_key(lhs) < hoid_get_bitwise_key(rhs))
1695 if (hoid_get_bitwise_key(lhs) > hoid_get_bitwise_key(rhs))
1698 ret = compare_names(lhs->nspace, lhs->nspace_len,
1699 rhs->nspace, rhs->nspace_len);
1703 hoid_get_effective_key(lhs, &effective_key1, &effective_key1_len);
1704 hoid_get_effective_key(rhs, &effective_key2, &effective_key2_len);
1705 ret = compare_names(effective_key1, effective_key1_len,
1706 effective_key2, effective_key2_len);
1710 ret = compare_names(lhs->oid, lhs->oid_len, rhs->oid, rhs->oid_len);
1714 if (lhs->snapid < rhs->snapid)
1716 if (lhs->snapid > rhs->snapid)
1723 * For decoding ->begin and ->end of MOSDBackoff only -- no MIN/MAX
1724 * compat stuff here.
1726 * Assumes @hoid is zero-initialized.
1728 static int decode_hoid(void **p, void *end, struct ceph_hobject_id *hoid)
1734 ret = ceph_start_decoding(p, end, 4, "hobject_t", &struct_v,
1740 pr_err("got struct_v %d < 4 of hobject_t\n", struct_v);
1744 hoid->key = ceph_extract_encoded_string(p, end, &hoid->key_len,
1746 if (IS_ERR(hoid->key)) {
1747 ret = PTR_ERR(hoid->key);
1752 hoid->oid = ceph_extract_encoded_string(p, end, &hoid->oid_len,
1754 if (IS_ERR(hoid->oid)) {
1755 ret = PTR_ERR(hoid->oid);
1760 ceph_decode_64_safe(p, end, hoid->snapid, e_inval);
1761 ceph_decode_32_safe(p, end, hoid->hash, e_inval);
1762 ceph_decode_8_safe(p, end, hoid->is_max, e_inval);
1764 hoid->nspace = ceph_extract_encoded_string(p, end, &hoid->nspace_len,
1766 if (IS_ERR(hoid->nspace)) {
1767 ret = PTR_ERR(hoid->nspace);
1768 hoid->nspace = NULL;
1772 ceph_decode_64_safe(p, end, hoid->pool, e_inval);
1774 ceph_hoid_build_hash_cache(hoid);
1781 static int hoid_encoding_size(const struct ceph_hobject_id *hoid)
1783 return 8 + 4 + 1 + 8 + /* snapid, hash, is_max, pool */
1784 4 + hoid->key_len + 4 + hoid->oid_len + 4 + hoid->nspace_len;
1787 static void encode_hoid(void **p, void *end, const struct ceph_hobject_id *hoid)
1789 ceph_start_encoding(p, 4, 3, hoid_encoding_size(hoid));
1790 ceph_encode_string(p, end, hoid->key, hoid->key_len);
1791 ceph_encode_string(p, end, hoid->oid, hoid->oid_len);
1792 ceph_encode_64(p, hoid->snapid);
1793 ceph_encode_32(p, hoid->hash);
1794 ceph_encode_8(p, hoid->is_max);
1795 ceph_encode_string(p, end, hoid->nspace, hoid->nspace_len);
1796 ceph_encode_64(p, hoid->pool);
1799 static void free_hoid(struct ceph_hobject_id *hoid)
1804 kfree(hoid->nspace);
1809 static struct ceph_osd_backoff *alloc_backoff(void)
1811 struct ceph_osd_backoff *backoff;
1813 backoff = kzalloc(sizeof(*backoff), GFP_NOIO);
1817 RB_CLEAR_NODE(&backoff->spg_node);
1818 RB_CLEAR_NODE(&backoff->id_node);
1822 static void free_backoff(struct ceph_osd_backoff *backoff)
1824 WARN_ON(!RB_EMPTY_NODE(&backoff->spg_node));
1825 WARN_ON(!RB_EMPTY_NODE(&backoff->id_node));
1827 free_hoid(backoff->begin);
1828 free_hoid(backoff->end);
1833 * Within a specific spgid, backoffs are managed by ->begin hoid.
1835 DEFINE_RB_INSDEL_FUNCS2(backoff, struct ceph_osd_backoff, begin, hoid_compare,
1836 RB_BYVAL, spg_node);
1838 static struct ceph_osd_backoff *lookup_containing_backoff(struct rb_root *root,
1839 const struct ceph_hobject_id *hoid)
1841 struct rb_node *n = root->rb_node;
1844 struct ceph_osd_backoff *cur =
1845 rb_entry(n, struct ceph_osd_backoff, spg_node);
1848 cmp = hoid_compare(hoid, cur->begin);
1851 } else if (cmp > 0) {
1852 if (hoid_compare(hoid, cur->end) < 0)
1865 * Each backoff has a unique id within its OSD session.
1867 DEFINE_RB_FUNCS(backoff_by_id, struct ceph_osd_backoff, id, id_node)
1869 static void clear_backoffs(struct ceph_osd *osd)
1871 while (!RB_EMPTY_ROOT(&osd->o_backoff_mappings)) {
1872 struct ceph_spg_mapping *spg =
1873 rb_entry(rb_first(&osd->o_backoff_mappings),
1874 struct ceph_spg_mapping, node);
1876 while (!RB_EMPTY_ROOT(&spg->backoffs)) {
1877 struct ceph_osd_backoff *backoff =
1878 rb_entry(rb_first(&spg->backoffs),
1879 struct ceph_osd_backoff, spg_node);
1881 erase_backoff(&spg->backoffs, backoff);
1882 erase_backoff_by_id(&osd->o_backoffs_by_id, backoff);
1883 free_backoff(backoff);
1885 erase_spg_mapping(&osd->o_backoff_mappings, spg);
1886 free_spg_mapping(spg);
1891 * Set up a temporary, non-owning view into @t.
1893 static void hoid_fill_from_target(struct ceph_hobject_id *hoid,
1894 const struct ceph_osd_request_target *t)
1898 hoid->oid = t->target_oid.name;
1899 hoid->oid_len = t->target_oid.name_len;
1900 hoid->snapid = CEPH_NOSNAP;
1901 hoid->hash = t->pgid.seed;
1902 hoid->is_max = false;
1903 if (t->target_oloc.pool_ns) {
1904 hoid->nspace = t->target_oloc.pool_ns->str;
1905 hoid->nspace_len = t->target_oloc.pool_ns->len;
1907 hoid->nspace = NULL;
1908 hoid->nspace_len = 0;
1910 hoid->pool = t->target_oloc.pool;
1911 ceph_hoid_build_hash_cache(hoid);
1914 static bool should_plug_request(struct ceph_osd_request *req)
1916 struct ceph_osd *osd = req->r_osd;
1917 struct ceph_spg_mapping *spg;
1918 struct ceph_osd_backoff *backoff;
1919 struct ceph_hobject_id hoid;
1921 spg = lookup_spg_mapping(&osd->o_backoff_mappings, &req->r_t.spgid);
1925 hoid_fill_from_target(&hoid, &req->r_t);
1926 backoff = lookup_containing_backoff(&spg->backoffs, &hoid);
1930 dout("%s req %p tid %llu backoff osd%d spgid %llu.%xs%d id %llu\n",
1931 __func__, req, req->r_tid, osd->o_osd, backoff->spgid.pgid.pool,
1932 backoff->spgid.pgid.seed, backoff->spgid.shard, backoff->id);
1937 * Keep get_num_data_items() in sync with this function.
1939 static void setup_request_data(struct ceph_osd_request *req)
1941 struct ceph_msg *request_msg = req->r_request;
1942 struct ceph_msg *reply_msg = req->r_reply;
1943 struct ceph_osd_req_op *op;
1945 if (req->r_request->num_data_items || req->r_reply->num_data_items)
1948 WARN_ON(request_msg->data_length || reply_msg->data_length);
1949 for (op = req->r_ops; op != &req->r_ops[req->r_num_ops]; op++) {
1952 case CEPH_OSD_OP_WRITE:
1953 case CEPH_OSD_OP_WRITEFULL:
1954 WARN_ON(op->indata_len != op->extent.length);
1955 ceph_osdc_msg_data_add(request_msg,
1956 &op->extent.osd_data);
1958 case CEPH_OSD_OP_SETXATTR:
1959 case CEPH_OSD_OP_CMPXATTR:
1960 WARN_ON(op->indata_len != op->xattr.name_len +
1961 op->xattr.value_len);
1962 ceph_osdc_msg_data_add(request_msg,
1963 &op->xattr.osd_data);
1965 case CEPH_OSD_OP_NOTIFY_ACK:
1966 ceph_osdc_msg_data_add(request_msg,
1967 &op->notify_ack.request_data);
1969 case CEPH_OSD_OP_COPY_FROM2:
1970 ceph_osdc_msg_data_add(request_msg,
1971 &op->copy_from.osd_data);
1975 case CEPH_OSD_OP_STAT:
1976 ceph_osdc_msg_data_add(reply_msg,
1979 case CEPH_OSD_OP_READ:
1980 ceph_osdc_msg_data_add(reply_msg,
1981 &op->extent.osd_data);
1983 case CEPH_OSD_OP_LIST_WATCHERS:
1984 ceph_osdc_msg_data_add(reply_msg,
1985 &op->list_watchers.response_data);
1989 case CEPH_OSD_OP_CALL:
1990 WARN_ON(op->indata_len != op->cls.class_len +
1991 op->cls.method_len +
1992 op->cls.indata_len);
1993 ceph_osdc_msg_data_add(request_msg,
1994 &op->cls.request_info);
1995 /* optional, can be NONE */
1996 ceph_osdc_msg_data_add(request_msg,
1997 &op->cls.request_data);
1998 /* optional, can be NONE */
1999 ceph_osdc_msg_data_add(reply_msg,
2000 &op->cls.response_data);
2002 case CEPH_OSD_OP_NOTIFY:
2003 ceph_osdc_msg_data_add(request_msg,
2004 &op->notify.request_data);
2005 ceph_osdc_msg_data_add(reply_msg,
2006 &op->notify.response_data);
2012 static void encode_pgid(void **p, const struct ceph_pg *pgid)
2014 ceph_encode_8(p, 1);
2015 ceph_encode_64(p, pgid->pool);
2016 ceph_encode_32(p, pgid->seed);
2017 ceph_encode_32(p, -1); /* preferred */
2020 static void encode_spgid(void **p, const struct ceph_spg *spgid)
2022 ceph_start_encoding(p, 1, 1, CEPH_PGID_ENCODING_LEN + 1);
2023 encode_pgid(p, &spgid->pgid);
2024 ceph_encode_8(p, spgid->shard);
2027 static void encode_oloc(void **p, void *end,
2028 const struct ceph_object_locator *oloc)
2030 ceph_start_encoding(p, 5, 4, ceph_oloc_encoding_size(oloc));
2031 ceph_encode_64(p, oloc->pool);
2032 ceph_encode_32(p, -1); /* preferred */
2033 ceph_encode_32(p, 0); /* key len */
2035 ceph_encode_string(p, end, oloc->pool_ns->str,
2036 oloc->pool_ns->len);
2038 ceph_encode_32(p, 0);
2041 static void encode_request_partial(struct ceph_osd_request *req,
2042 struct ceph_msg *msg)
2044 void *p = msg->front.iov_base;
2045 void *const end = p + msg->front_alloc_len;
2049 if (req->r_flags & CEPH_OSD_FLAG_WRITE) {
2050 /* snapshots aren't writeable */
2051 WARN_ON(req->r_snapid != CEPH_NOSNAP);
2053 WARN_ON(req->r_mtime.tv_sec || req->r_mtime.tv_nsec ||
2054 req->r_data_offset || req->r_snapc);
2057 setup_request_data(req);
2059 encode_spgid(&p, &req->r_t.spgid); /* actual spg */
2060 ceph_encode_32(&p, req->r_t.pgid.seed); /* raw hash */
2061 ceph_encode_32(&p, req->r_osdc->osdmap->epoch);
2062 ceph_encode_32(&p, req->r_flags);
2065 ceph_start_encoding(&p, 2, 2, sizeof(struct ceph_osd_reqid));
2066 memset(p, 0, sizeof(struct ceph_osd_reqid));
2067 p += sizeof(struct ceph_osd_reqid);
2070 memset(p, 0, sizeof(struct ceph_blkin_trace_info));
2071 p += sizeof(struct ceph_blkin_trace_info);
2073 ceph_encode_32(&p, 0); /* client_inc, always 0 */
2074 ceph_encode_timespec64(p, &req->r_mtime);
2075 p += sizeof(struct ceph_timespec);
2077 encode_oloc(&p, end, &req->r_t.target_oloc);
2078 ceph_encode_string(&p, end, req->r_t.target_oid.name,
2079 req->r_t.target_oid.name_len);
2081 /* ops, can imply data */
2082 ceph_encode_16(&p, req->r_num_ops);
2083 for (i = 0; i < req->r_num_ops; i++) {
2084 data_len += osd_req_encode_op(p, &req->r_ops[i]);
2085 p += sizeof(struct ceph_osd_op);
2088 ceph_encode_64(&p, req->r_snapid); /* snapid */
2090 ceph_encode_64(&p, req->r_snapc->seq);
2091 ceph_encode_32(&p, req->r_snapc->num_snaps);
2092 for (i = 0; i < req->r_snapc->num_snaps; i++)
2093 ceph_encode_64(&p, req->r_snapc->snaps[i]);
2095 ceph_encode_64(&p, 0); /* snap_seq */
2096 ceph_encode_32(&p, 0); /* snaps len */
2099 ceph_encode_32(&p, req->r_attempts); /* retry_attempt */
2100 BUG_ON(p > end - 8); /* space for features */
2102 msg->hdr.version = cpu_to_le16(8); /* MOSDOp v8 */
2103 /* front_len is finalized in encode_request_finish() */
2104 msg->front.iov_len = p - msg->front.iov_base;
2105 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2106 msg->hdr.data_len = cpu_to_le32(data_len);
2108 * The header "data_off" is a hint to the receiver allowing it
2109 * to align received data into its buffers such that there's no
2110 * need to re-copy it before writing it to disk (direct I/O).
2112 msg->hdr.data_off = cpu_to_le16(req->r_data_offset);
2114 dout("%s req %p msg %p oid %s oid_len %d\n", __func__, req, msg,
2115 req->r_t.target_oid.name, req->r_t.target_oid.name_len);
2118 static void encode_request_finish(struct ceph_msg *msg)
2120 void *p = msg->front.iov_base;
2121 void *const partial_end = p + msg->front.iov_len;
2122 void *const end = p + msg->front_alloc_len;
2124 if (CEPH_HAVE_FEATURE(msg->con->peer_features, RESEND_ON_SPLIT)) {
2125 /* luminous OSD -- encode features and be done */
2127 ceph_encode_64(&p, msg->con->peer_features);
2130 char spgid[CEPH_ENCODING_START_BLK_LEN +
2131 CEPH_PGID_ENCODING_LEN + 1];
2135 char reqid[CEPH_ENCODING_START_BLK_LEN +
2136 sizeof(struct ceph_osd_reqid)];
2137 char trace[sizeof(struct ceph_blkin_trace_info)];
2139 struct ceph_timespec mtime;
2141 struct ceph_pg pgid;
2142 void *oloc, *oid, *tail;
2143 int oloc_len, oid_len, tail_len;
2147 * Pre-luminous OSD -- reencode v8 into v4 using @head
2148 * as a temporary buffer. Encode the raw PG; the rest
2149 * is just a matter of moving oloc, oid and tail blobs
2152 memcpy(&head, p, sizeof(head));
2156 p += CEPH_ENCODING_START_BLK_LEN;
2157 pgid.pool = ceph_decode_64(&p);
2158 p += 4 + 4; /* preferred, key len */
2159 len = ceph_decode_32(&p);
2160 p += len; /* nspace */
2161 oloc_len = p - oloc;
2164 len = ceph_decode_32(&p);
2169 tail_len = partial_end - p;
2171 p = msg->front.iov_base;
2172 ceph_encode_copy(&p, &head.client_inc, sizeof(head.client_inc));
2173 ceph_encode_copy(&p, &head.epoch, sizeof(head.epoch));
2174 ceph_encode_copy(&p, &head.flags, sizeof(head.flags));
2175 ceph_encode_copy(&p, &head.mtime, sizeof(head.mtime));
2177 /* reassert_version */
2178 memset(p, 0, sizeof(struct ceph_eversion));
2179 p += sizeof(struct ceph_eversion);
2182 memmove(p, oloc, oloc_len);
2185 pgid.seed = le32_to_cpu(head.hash);
2186 encode_pgid(&p, &pgid); /* raw pg */
2189 memmove(p, oid, oid_len);
2192 /* tail -- ops, snapid, snapc, retry_attempt */
2194 memmove(p, tail, tail_len);
2197 msg->hdr.version = cpu_to_le16(4); /* MOSDOp v4 */
2201 msg->front.iov_len = p - msg->front.iov_base;
2202 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2204 dout("%s msg %p tid %llu %u+%u+%u v%d\n", __func__, msg,
2205 le64_to_cpu(msg->hdr.tid), le32_to_cpu(msg->hdr.front_len),
2206 le32_to_cpu(msg->hdr.middle_len), le32_to_cpu(msg->hdr.data_len),
2207 le16_to_cpu(msg->hdr.version));
2211 * @req has to be assigned a tid and registered.
2213 static void send_request(struct ceph_osd_request *req)
2215 struct ceph_osd *osd = req->r_osd;
2217 verify_osd_locked(osd);
2218 WARN_ON(osd->o_osd != req->r_t.osd);
2221 if (should_plug_request(req))
2225 * We may have a previously queued request message hanging
2226 * around. Cancel it to avoid corrupting the msgr.
2229 ceph_msg_revoke(req->r_request);
2231 req->r_flags |= CEPH_OSD_FLAG_KNOWN_REDIR;
2232 if (req->r_attempts)
2233 req->r_flags |= CEPH_OSD_FLAG_RETRY;
2235 WARN_ON(req->r_flags & CEPH_OSD_FLAG_RETRY);
2237 encode_request_partial(req, req->r_request);
2239 dout("%s req %p tid %llu to pgid %llu.%x spgid %llu.%xs%d osd%d e%u flags 0x%x attempt %d\n",
2240 __func__, req, req->r_tid, req->r_t.pgid.pool, req->r_t.pgid.seed,
2241 req->r_t.spgid.pgid.pool, req->r_t.spgid.pgid.seed,
2242 req->r_t.spgid.shard, osd->o_osd, req->r_t.epoch, req->r_flags,
2245 req->r_t.paused = false;
2246 req->r_stamp = jiffies;
2249 req->r_sent = osd->o_incarnation;
2250 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
2251 ceph_con_send(&osd->o_con, ceph_msg_get(req->r_request));
2254 static void maybe_request_map(struct ceph_osd_client *osdc)
2256 bool continuous = false;
2258 verify_osdc_locked(osdc);
2259 WARN_ON(!osdc->osdmap->epoch);
2261 if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
2262 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD) ||
2263 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) {
2264 dout("%s osdc %p continuous\n", __func__, osdc);
2267 dout("%s osdc %p onetime\n", __func__, osdc);
2270 if (ceph_monc_want_map(&osdc->client->monc, CEPH_SUB_OSDMAP,
2271 osdc->osdmap->epoch + 1, continuous))
2272 ceph_monc_renew_subs(&osdc->client->monc);
2275 static void complete_request(struct ceph_osd_request *req, int err);
2276 static void send_map_check(struct ceph_osd_request *req);
2278 static void __submit_request(struct ceph_osd_request *req, bool wrlocked)
2280 struct ceph_osd_client *osdc = req->r_osdc;
2281 struct ceph_osd *osd;
2282 enum calc_target_result ct_res;
2284 bool need_send = false;
2285 bool promoted = false;
2287 WARN_ON(req->r_tid);
2288 dout("%s req %p wrlocked %d\n", __func__, req, wrlocked);
2291 ct_res = calc_target(osdc, &req->r_t, false);
2292 if (ct_res == CALC_TARGET_POOL_DNE && !wrlocked)
2295 osd = lookup_create_osd(osdc, req->r_t.osd, wrlocked);
2297 WARN_ON(PTR_ERR(osd) != -EAGAIN || wrlocked);
2301 if (osdc->abort_err) {
2302 dout("req %p abort_err %d\n", req, osdc->abort_err);
2303 err = osdc->abort_err;
2304 } else if (osdc->osdmap->epoch < osdc->epoch_barrier) {
2305 dout("req %p epoch %u barrier %u\n", req, osdc->osdmap->epoch,
2306 osdc->epoch_barrier);
2307 req->r_t.paused = true;
2308 maybe_request_map(osdc);
2309 } else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
2310 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) {
2311 dout("req %p pausewr\n", req);
2312 req->r_t.paused = true;
2313 maybe_request_map(osdc);
2314 } else if ((req->r_flags & CEPH_OSD_FLAG_READ) &&
2315 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) {
2316 dout("req %p pauserd\n", req);
2317 req->r_t.paused = true;
2318 maybe_request_map(osdc);
2319 } else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
2320 !(req->r_flags & (CEPH_OSD_FLAG_FULL_TRY |
2321 CEPH_OSD_FLAG_FULL_FORCE)) &&
2322 (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
2323 pool_full(osdc, req->r_t.base_oloc.pool))) {
2324 dout("req %p full/pool_full\n", req);
2325 if (ceph_test_opt(osdc->client, ABORT_ON_FULL)) {
2328 pr_warn_ratelimited("FULL or reached pool quota\n");
2329 req->r_t.paused = true;
2330 maybe_request_map(osdc);
2332 } else if (!osd_homeless(osd)) {
2335 maybe_request_map(osdc);
2338 mutex_lock(&osd->lock);
2340 * Assign the tid atomically with send_request() to protect
2341 * multiple writes to the same object from racing with each
2342 * other, resulting in out of order ops on the OSDs.
2344 req->r_tid = atomic64_inc_return(&osdc->last_tid);
2345 link_request(osd, req);
2349 complete_request(req, err);
2350 mutex_unlock(&osd->lock);
2352 if (!err && ct_res == CALC_TARGET_POOL_DNE)
2353 send_map_check(req);
2356 downgrade_write(&osdc->lock);
2360 up_read(&osdc->lock);
2361 down_write(&osdc->lock);
2367 static void account_request(struct ceph_osd_request *req)
2369 WARN_ON(req->r_flags & (CEPH_OSD_FLAG_ACK | CEPH_OSD_FLAG_ONDISK));
2370 WARN_ON(!(req->r_flags & (CEPH_OSD_FLAG_READ | CEPH_OSD_FLAG_WRITE)));
2372 req->r_flags |= CEPH_OSD_FLAG_ONDISK;
2373 atomic_inc(&req->r_osdc->num_requests);
2375 req->r_start_stamp = jiffies;
2378 static void submit_request(struct ceph_osd_request *req, bool wrlocked)
2380 ceph_osdc_get_request(req);
2381 account_request(req);
2382 __submit_request(req, wrlocked);
2385 static void finish_request(struct ceph_osd_request *req)
2387 struct ceph_osd_client *osdc = req->r_osdc;
2389 WARN_ON(lookup_request_mc(&osdc->map_checks, req->r_tid));
2390 dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
2393 unlink_request(req->r_osd, req);
2394 atomic_dec(&osdc->num_requests);
2397 * If an OSD has failed or returned and a request has been sent
2398 * twice, it's possible to get a reply and end up here while the
2399 * request message is queued for delivery. We will ignore the
2400 * reply, so not a big deal, but better to try and catch it.
2402 ceph_msg_revoke(req->r_request);
2403 ceph_msg_revoke_incoming(req->r_reply);
2406 static void __complete_request(struct ceph_osd_request *req)
2408 dout("%s req %p tid %llu cb %ps result %d\n", __func__, req,
2409 req->r_tid, req->r_callback, req->r_result);
2411 if (req->r_callback)
2412 req->r_callback(req);
2413 complete_all(&req->r_completion);
2414 ceph_osdc_put_request(req);
2417 static void complete_request_workfn(struct work_struct *work)
2419 struct ceph_osd_request *req =
2420 container_of(work, struct ceph_osd_request, r_complete_work);
2422 __complete_request(req);
2426 * This is open-coded in handle_reply().
2428 static void complete_request(struct ceph_osd_request *req, int err)
2430 dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err);
2432 req->r_result = err;
2433 finish_request(req);
2435 INIT_WORK(&req->r_complete_work, complete_request_workfn);
2436 queue_work(req->r_osdc->completion_wq, &req->r_complete_work);
2439 static void cancel_map_check(struct ceph_osd_request *req)
2441 struct ceph_osd_client *osdc = req->r_osdc;
2442 struct ceph_osd_request *lookup_req;
2444 verify_osdc_wrlocked(osdc);
2446 lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid);
2450 WARN_ON(lookup_req != req);
2451 erase_request_mc(&osdc->map_checks, req);
2452 ceph_osdc_put_request(req);
2455 static void cancel_request(struct ceph_osd_request *req)
2457 dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
2459 cancel_map_check(req);
2460 finish_request(req);
2461 complete_all(&req->r_completion);
2462 ceph_osdc_put_request(req);
2465 static void abort_request(struct ceph_osd_request *req, int err)
2467 dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err);
2469 cancel_map_check(req);
2470 complete_request(req, err);
2473 static int abort_fn(struct ceph_osd_request *req, void *arg)
2475 int err = *(int *)arg;
2477 abort_request(req, err);
2478 return 0; /* continue iteration */
2482 * Abort all in-flight requests with @err and arrange for all future
2483 * requests to be failed immediately.
2485 void ceph_osdc_abort_requests(struct ceph_osd_client *osdc, int err)
2487 dout("%s osdc %p err %d\n", __func__, osdc, err);
2488 down_write(&osdc->lock);
2489 for_each_request(osdc, abort_fn, &err);
2490 osdc->abort_err = err;
2491 up_write(&osdc->lock);
2493 EXPORT_SYMBOL(ceph_osdc_abort_requests);
2495 void ceph_osdc_clear_abort_err(struct ceph_osd_client *osdc)
2497 down_write(&osdc->lock);
2498 osdc->abort_err = 0;
2499 up_write(&osdc->lock);
2501 EXPORT_SYMBOL(ceph_osdc_clear_abort_err);
2503 static void update_epoch_barrier(struct ceph_osd_client *osdc, u32 eb)
2505 if (likely(eb > osdc->epoch_barrier)) {
2506 dout("updating epoch_barrier from %u to %u\n",
2507 osdc->epoch_barrier, eb);
2508 osdc->epoch_barrier = eb;
2509 /* Request map if we're not to the barrier yet */
2510 if (eb > osdc->osdmap->epoch)
2511 maybe_request_map(osdc);
2515 void ceph_osdc_update_epoch_barrier(struct ceph_osd_client *osdc, u32 eb)
2517 down_read(&osdc->lock);
2518 if (unlikely(eb > osdc->epoch_barrier)) {
2519 up_read(&osdc->lock);
2520 down_write(&osdc->lock);
2521 update_epoch_barrier(osdc, eb);
2522 up_write(&osdc->lock);
2524 up_read(&osdc->lock);
2527 EXPORT_SYMBOL(ceph_osdc_update_epoch_barrier);
2530 * We can end up releasing caps as a result of abort_request().
2531 * In that case, we probably want to ensure that the cap release message
2532 * has an updated epoch barrier in it, so set the epoch barrier prior to
2533 * aborting the first request.
2535 static int abort_on_full_fn(struct ceph_osd_request *req, void *arg)
2537 struct ceph_osd_client *osdc = req->r_osdc;
2538 bool *victims = arg;
2540 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
2541 (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
2542 pool_full(osdc, req->r_t.base_oloc.pool))) {
2544 update_epoch_barrier(osdc, osdc->osdmap->epoch);
2547 abort_request(req, -ENOSPC);
2550 return 0; /* continue iteration */
2554 * Drop all pending requests that are stalled waiting on a full condition to
2555 * clear, and complete them with ENOSPC as the return code. Set the
2556 * osdc->epoch_barrier to the latest map epoch that we've seen if any were
2559 static void ceph_osdc_abort_on_full(struct ceph_osd_client *osdc)
2561 bool victims = false;
2563 if (ceph_test_opt(osdc->client, ABORT_ON_FULL) &&
2564 (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || have_pool_full(osdc)))
2565 for_each_request(osdc, abort_on_full_fn, &victims);
2568 static void check_pool_dne(struct ceph_osd_request *req)
2570 struct ceph_osd_client *osdc = req->r_osdc;
2571 struct ceph_osdmap *map = osdc->osdmap;
2573 verify_osdc_wrlocked(osdc);
2574 WARN_ON(!map->epoch);
2576 if (req->r_attempts) {
2578 * We sent a request earlier, which means that
2579 * previously the pool existed, and now it does not
2580 * (i.e., it was deleted).
2582 req->r_map_dne_bound = map->epoch;
2583 dout("%s req %p tid %llu pool disappeared\n", __func__, req,
2586 dout("%s req %p tid %llu map_dne_bound %u have %u\n", __func__,
2587 req, req->r_tid, req->r_map_dne_bound, map->epoch);
2590 if (req->r_map_dne_bound) {
2591 if (map->epoch >= req->r_map_dne_bound) {
2592 /* we had a new enough map */
2593 pr_info_ratelimited("tid %llu pool does not exist\n",
2595 complete_request(req, -ENOENT);
2598 send_map_check(req);
2602 static void map_check_cb(struct ceph_mon_generic_request *greq)
2604 struct ceph_osd_client *osdc = &greq->monc->client->osdc;
2605 struct ceph_osd_request *req;
2606 u64 tid = greq->private_data;
2608 WARN_ON(greq->result || !greq->u.newest);
2610 down_write(&osdc->lock);
2611 req = lookup_request_mc(&osdc->map_checks, tid);
2613 dout("%s tid %llu dne\n", __func__, tid);
2617 dout("%s req %p tid %llu map_dne_bound %u newest %llu\n", __func__,
2618 req, req->r_tid, req->r_map_dne_bound, greq->u.newest);
2619 if (!req->r_map_dne_bound)
2620 req->r_map_dne_bound = greq->u.newest;
2621 erase_request_mc(&osdc->map_checks, req);
2622 check_pool_dne(req);
2624 ceph_osdc_put_request(req);
2626 up_write(&osdc->lock);
2629 static void send_map_check(struct ceph_osd_request *req)
2631 struct ceph_osd_client *osdc = req->r_osdc;
2632 struct ceph_osd_request *lookup_req;
2635 verify_osdc_wrlocked(osdc);
2637 lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid);
2639 WARN_ON(lookup_req != req);
2643 ceph_osdc_get_request(req);
2644 insert_request_mc(&osdc->map_checks, req);
2645 ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap",
2646 map_check_cb, req->r_tid);
2651 * lingering requests, watch/notify v2 infrastructure
2653 static void linger_release(struct kref *kref)
2655 struct ceph_osd_linger_request *lreq =
2656 container_of(kref, struct ceph_osd_linger_request, kref);
2658 dout("%s lreq %p reg_req %p ping_req %p\n", __func__, lreq,
2659 lreq->reg_req, lreq->ping_req);
2660 WARN_ON(!RB_EMPTY_NODE(&lreq->node));
2661 WARN_ON(!RB_EMPTY_NODE(&lreq->osdc_node));
2662 WARN_ON(!RB_EMPTY_NODE(&lreq->mc_node));
2663 WARN_ON(!list_empty(&lreq->scan_item));
2664 WARN_ON(!list_empty(&lreq->pending_lworks));
2668 ceph_osdc_put_request(lreq->reg_req);
2670 ceph_osdc_put_request(lreq->ping_req);
2671 target_destroy(&lreq->t);
2675 static void linger_put(struct ceph_osd_linger_request *lreq)
2678 kref_put(&lreq->kref, linger_release);
2681 static struct ceph_osd_linger_request *
2682 linger_get(struct ceph_osd_linger_request *lreq)
2684 kref_get(&lreq->kref);
2688 static struct ceph_osd_linger_request *
2689 linger_alloc(struct ceph_osd_client *osdc)
2691 struct ceph_osd_linger_request *lreq;
2693 lreq = kzalloc(sizeof(*lreq), GFP_NOIO);
2697 kref_init(&lreq->kref);
2698 mutex_init(&lreq->lock);
2699 RB_CLEAR_NODE(&lreq->node);
2700 RB_CLEAR_NODE(&lreq->osdc_node);
2701 RB_CLEAR_NODE(&lreq->mc_node);
2702 INIT_LIST_HEAD(&lreq->scan_item);
2703 INIT_LIST_HEAD(&lreq->pending_lworks);
2704 init_completion(&lreq->reg_commit_wait);
2705 init_completion(&lreq->notify_finish_wait);
2708 target_init(&lreq->t);
2710 dout("%s lreq %p\n", __func__, lreq);
2714 DEFINE_RB_INSDEL_FUNCS(linger, struct ceph_osd_linger_request, linger_id, node)
2715 DEFINE_RB_FUNCS(linger_osdc, struct ceph_osd_linger_request, linger_id, osdc_node)
2716 DEFINE_RB_FUNCS(linger_mc, struct ceph_osd_linger_request, linger_id, mc_node)
2719 * Create linger request <-> OSD session relation.
2721 * @lreq has to be registered, @osd may be homeless.
2723 static void link_linger(struct ceph_osd *osd,
2724 struct ceph_osd_linger_request *lreq)
2726 verify_osd_locked(osd);
2727 WARN_ON(!lreq->linger_id || lreq->osd);
2728 dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd,
2729 osd->o_osd, lreq, lreq->linger_id);
2731 if (!osd_homeless(osd))
2732 __remove_osd_from_lru(osd);
2734 atomic_inc(&osd->o_osdc->num_homeless);
2737 insert_linger(&osd->o_linger_requests, lreq);
2741 static void unlink_linger(struct ceph_osd *osd,
2742 struct ceph_osd_linger_request *lreq)
2744 verify_osd_locked(osd);
2745 WARN_ON(lreq->osd != osd);
2746 dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd,
2747 osd->o_osd, lreq, lreq->linger_id);
2750 erase_linger(&osd->o_linger_requests, lreq);
2753 if (!osd_homeless(osd))
2754 maybe_move_osd_to_lru(osd);
2756 atomic_dec(&osd->o_osdc->num_homeless);
2759 static bool __linger_registered(struct ceph_osd_linger_request *lreq)
2761 verify_osdc_locked(lreq->osdc);
2763 return !RB_EMPTY_NODE(&lreq->osdc_node);
2766 static bool linger_registered(struct ceph_osd_linger_request *lreq)
2768 struct ceph_osd_client *osdc = lreq->osdc;
2771 down_read(&osdc->lock);
2772 registered = __linger_registered(lreq);
2773 up_read(&osdc->lock);
2778 static void linger_register(struct ceph_osd_linger_request *lreq)
2780 struct ceph_osd_client *osdc = lreq->osdc;
2782 verify_osdc_wrlocked(osdc);
2783 WARN_ON(lreq->linger_id);
2786 lreq->linger_id = ++osdc->last_linger_id;
2787 insert_linger_osdc(&osdc->linger_requests, lreq);
2790 static void linger_unregister(struct ceph_osd_linger_request *lreq)
2792 struct ceph_osd_client *osdc = lreq->osdc;
2794 verify_osdc_wrlocked(osdc);
2796 erase_linger_osdc(&osdc->linger_requests, lreq);
2800 static void cancel_linger_request(struct ceph_osd_request *req)
2802 struct ceph_osd_linger_request *lreq = req->r_priv;
2804 WARN_ON(!req->r_linger);
2805 cancel_request(req);
2809 struct linger_work {
2810 struct work_struct work;
2811 struct ceph_osd_linger_request *lreq;
2812 struct list_head pending_item;
2813 unsigned long queued_stamp;
2819 void *payload; /* points into @msg front */
2822 struct ceph_msg *msg; /* for ceph_msg_put() */
2830 static struct linger_work *lwork_alloc(struct ceph_osd_linger_request *lreq,
2833 struct linger_work *lwork;
2835 lwork = kzalloc(sizeof(*lwork), GFP_NOIO);
2839 INIT_WORK(&lwork->work, workfn);
2840 INIT_LIST_HEAD(&lwork->pending_item);
2841 lwork->lreq = linger_get(lreq);
2846 static void lwork_free(struct linger_work *lwork)
2848 struct ceph_osd_linger_request *lreq = lwork->lreq;
2850 mutex_lock(&lreq->lock);
2851 list_del(&lwork->pending_item);
2852 mutex_unlock(&lreq->lock);
2858 static void lwork_queue(struct linger_work *lwork)
2860 struct ceph_osd_linger_request *lreq = lwork->lreq;
2861 struct ceph_osd_client *osdc = lreq->osdc;
2863 verify_lreq_locked(lreq);
2864 WARN_ON(!list_empty(&lwork->pending_item));
2866 lwork->queued_stamp = jiffies;
2867 list_add_tail(&lwork->pending_item, &lreq->pending_lworks);
2868 queue_work(osdc->notify_wq, &lwork->work);
2871 static void do_watch_notify(struct work_struct *w)
2873 struct linger_work *lwork = container_of(w, struct linger_work, work);
2874 struct ceph_osd_linger_request *lreq = lwork->lreq;
2876 if (!linger_registered(lreq)) {
2877 dout("%s lreq %p not registered\n", __func__, lreq);
2881 WARN_ON(!lreq->is_watch);
2882 dout("%s lreq %p notify_id %llu notifier_id %llu payload_len %zu\n",
2883 __func__, lreq, lwork->notify.notify_id, lwork->notify.notifier_id,
2884 lwork->notify.payload_len);
2885 lreq->wcb(lreq->data, lwork->notify.notify_id, lreq->linger_id,
2886 lwork->notify.notifier_id, lwork->notify.payload,
2887 lwork->notify.payload_len);
2890 ceph_msg_put(lwork->notify.msg);
2894 static void do_watch_error(struct work_struct *w)
2896 struct linger_work *lwork = container_of(w, struct linger_work, work);
2897 struct ceph_osd_linger_request *lreq = lwork->lreq;
2899 if (!linger_registered(lreq)) {
2900 dout("%s lreq %p not registered\n", __func__, lreq);
2904 dout("%s lreq %p err %d\n", __func__, lreq, lwork->error.err);
2905 lreq->errcb(lreq->data, lreq->linger_id, lwork->error.err);
2911 static void queue_watch_error(struct ceph_osd_linger_request *lreq)
2913 struct linger_work *lwork;
2915 lwork = lwork_alloc(lreq, do_watch_error);
2917 pr_err("failed to allocate error-lwork\n");
2921 lwork->error.err = lreq->last_error;
2925 static void linger_reg_commit_complete(struct ceph_osd_linger_request *lreq,
2928 if (!completion_done(&lreq->reg_commit_wait)) {
2929 lreq->reg_commit_error = (result <= 0 ? result : 0);
2930 complete_all(&lreq->reg_commit_wait);
2934 static void linger_commit_cb(struct ceph_osd_request *req)
2936 struct ceph_osd_linger_request *lreq = req->r_priv;
2938 mutex_lock(&lreq->lock);
2939 dout("%s lreq %p linger_id %llu result %d\n", __func__, lreq,
2940 lreq->linger_id, req->r_result);
2941 linger_reg_commit_complete(lreq, req->r_result);
2942 lreq->committed = true;
2944 if (!lreq->is_watch) {
2945 struct ceph_osd_data *osd_data =
2946 osd_req_op_data(req, 0, notify, response_data);
2947 void *p = page_address(osd_data->pages[0]);
2949 WARN_ON(req->r_ops[0].op != CEPH_OSD_OP_NOTIFY ||
2950 osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
2952 /* make note of the notify_id */
2953 if (req->r_ops[0].outdata_len >= sizeof(u64)) {
2954 lreq->notify_id = ceph_decode_64(&p);
2955 dout("lreq %p notify_id %llu\n", lreq,
2958 dout("lreq %p no notify_id\n", lreq);
2962 mutex_unlock(&lreq->lock);
2966 static int normalize_watch_error(int err)
2969 * Translate ENOENT -> ENOTCONN so that a delete->disconnection
2970 * notification and a failure to reconnect because we raced with
2971 * the delete appear the same to the user.
2979 static void linger_reconnect_cb(struct ceph_osd_request *req)
2981 struct ceph_osd_linger_request *lreq = req->r_priv;
2983 mutex_lock(&lreq->lock);
2984 dout("%s lreq %p linger_id %llu result %d last_error %d\n", __func__,
2985 lreq, lreq->linger_id, req->r_result, lreq->last_error);
2986 if (req->r_result < 0) {
2987 if (!lreq->last_error) {
2988 lreq->last_error = normalize_watch_error(req->r_result);
2989 queue_watch_error(lreq);
2993 mutex_unlock(&lreq->lock);
2997 static void send_linger(struct ceph_osd_linger_request *lreq)
2999 struct ceph_osd_request *req = lreq->reg_req;
3000 struct ceph_osd_req_op *op = &req->r_ops[0];
3002 verify_osdc_wrlocked(req->r_osdc);
3003 dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
3006 cancel_linger_request(req);
3008 request_reinit(req);
3009 ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
3010 ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
3011 req->r_flags = lreq->t.flags;
3012 req->r_mtime = lreq->mtime;
3014 mutex_lock(&lreq->lock);
3015 if (lreq->is_watch && lreq->committed) {
3016 WARN_ON(op->op != CEPH_OSD_OP_WATCH ||
3017 op->watch.cookie != lreq->linger_id);
3018 op->watch.op = CEPH_OSD_WATCH_OP_RECONNECT;
3019 op->watch.gen = ++lreq->register_gen;
3020 dout("lreq %p reconnect register_gen %u\n", lreq,
3022 req->r_callback = linger_reconnect_cb;
3024 if (!lreq->is_watch)
3025 lreq->notify_id = 0;
3027 WARN_ON(op->watch.op != CEPH_OSD_WATCH_OP_WATCH);
3028 dout("lreq %p register\n", lreq);
3029 req->r_callback = linger_commit_cb;
3031 mutex_unlock(&lreq->lock);
3033 req->r_priv = linger_get(lreq);
3034 req->r_linger = true;
3036 submit_request(req, true);
3039 static void linger_ping_cb(struct ceph_osd_request *req)
3041 struct ceph_osd_linger_request *lreq = req->r_priv;
3043 mutex_lock(&lreq->lock);
3044 dout("%s lreq %p linger_id %llu result %d ping_sent %lu last_error %d\n",
3045 __func__, lreq, lreq->linger_id, req->r_result, lreq->ping_sent,
3047 if (lreq->register_gen == req->r_ops[0].watch.gen) {
3048 if (!req->r_result) {
3049 lreq->watch_valid_thru = lreq->ping_sent;
3050 } else if (!lreq->last_error) {
3051 lreq->last_error = normalize_watch_error(req->r_result);
3052 queue_watch_error(lreq);
3055 dout("lreq %p register_gen %u ignoring old pong %u\n", lreq,
3056 lreq->register_gen, req->r_ops[0].watch.gen);
3059 mutex_unlock(&lreq->lock);
3063 static void send_linger_ping(struct ceph_osd_linger_request *lreq)
3065 struct ceph_osd_client *osdc = lreq->osdc;
3066 struct ceph_osd_request *req = lreq->ping_req;
3067 struct ceph_osd_req_op *op = &req->r_ops[0];
3069 if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) {
3070 dout("%s PAUSERD\n", __func__);
3074 lreq->ping_sent = jiffies;
3075 dout("%s lreq %p linger_id %llu ping_sent %lu register_gen %u\n",
3076 __func__, lreq, lreq->linger_id, lreq->ping_sent,
3077 lreq->register_gen);
3080 cancel_linger_request(req);
3082 request_reinit(req);
3083 target_copy(&req->r_t, &lreq->t);
3085 WARN_ON(op->op != CEPH_OSD_OP_WATCH ||
3086 op->watch.cookie != lreq->linger_id ||
3087 op->watch.op != CEPH_OSD_WATCH_OP_PING);
3088 op->watch.gen = lreq->register_gen;
3089 req->r_callback = linger_ping_cb;
3090 req->r_priv = linger_get(lreq);
3091 req->r_linger = true;
3093 ceph_osdc_get_request(req);
3094 account_request(req);
3095 req->r_tid = atomic64_inc_return(&osdc->last_tid);
3096 link_request(lreq->osd, req);
3100 static void linger_submit(struct ceph_osd_linger_request *lreq)
3102 struct ceph_osd_client *osdc = lreq->osdc;
3103 struct ceph_osd *osd;
3105 down_write(&osdc->lock);
3106 linger_register(lreq);
3107 if (lreq->is_watch) {
3108 lreq->reg_req->r_ops[0].watch.cookie = lreq->linger_id;
3109 lreq->ping_req->r_ops[0].watch.cookie = lreq->linger_id;
3111 lreq->reg_req->r_ops[0].notify.cookie = lreq->linger_id;
3114 calc_target(osdc, &lreq->t, false);
3115 osd = lookup_create_osd(osdc, lreq->t.osd, true);
3116 link_linger(osd, lreq);
3119 up_write(&osdc->lock);
3122 static void cancel_linger_map_check(struct ceph_osd_linger_request *lreq)
3124 struct ceph_osd_client *osdc = lreq->osdc;
3125 struct ceph_osd_linger_request *lookup_lreq;
3127 verify_osdc_wrlocked(osdc);
3129 lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks,
3134 WARN_ON(lookup_lreq != lreq);
3135 erase_linger_mc(&osdc->linger_map_checks, lreq);
3140 * @lreq has to be both registered and linked.
3142 static void __linger_cancel(struct ceph_osd_linger_request *lreq)
3144 if (lreq->is_watch && lreq->ping_req->r_osd)
3145 cancel_linger_request(lreq->ping_req);
3146 if (lreq->reg_req->r_osd)
3147 cancel_linger_request(lreq->reg_req);
3148 cancel_linger_map_check(lreq);
3149 unlink_linger(lreq->osd, lreq);
3150 linger_unregister(lreq);
3153 static void linger_cancel(struct ceph_osd_linger_request *lreq)
3155 struct ceph_osd_client *osdc = lreq->osdc;
3157 down_write(&osdc->lock);
3158 if (__linger_registered(lreq))
3159 __linger_cancel(lreq);
3160 up_write(&osdc->lock);
3163 static void send_linger_map_check(struct ceph_osd_linger_request *lreq);
3165 static void check_linger_pool_dne(struct ceph_osd_linger_request *lreq)
3167 struct ceph_osd_client *osdc = lreq->osdc;
3168 struct ceph_osdmap *map = osdc->osdmap;
3170 verify_osdc_wrlocked(osdc);
3171 WARN_ON(!map->epoch);
3173 if (lreq->register_gen) {
3174 lreq->map_dne_bound = map->epoch;
3175 dout("%s lreq %p linger_id %llu pool disappeared\n", __func__,
3176 lreq, lreq->linger_id);
3178 dout("%s lreq %p linger_id %llu map_dne_bound %u have %u\n",
3179 __func__, lreq, lreq->linger_id, lreq->map_dne_bound,
3183 if (lreq->map_dne_bound) {
3184 if (map->epoch >= lreq->map_dne_bound) {
3185 /* we had a new enough map */
3186 pr_info("linger_id %llu pool does not exist\n",
3188 linger_reg_commit_complete(lreq, -ENOENT);
3189 __linger_cancel(lreq);
3192 send_linger_map_check(lreq);
3196 static void linger_map_check_cb(struct ceph_mon_generic_request *greq)
3198 struct ceph_osd_client *osdc = &greq->monc->client->osdc;
3199 struct ceph_osd_linger_request *lreq;
3200 u64 linger_id = greq->private_data;
3202 WARN_ON(greq->result || !greq->u.newest);
3204 down_write(&osdc->lock);
3205 lreq = lookup_linger_mc(&osdc->linger_map_checks, linger_id);
3207 dout("%s linger_id %llu dne\n", __func__, linger_id);
3211 dout("%s lreq %p linger_id %llu map_dne_bound %u newest %llu\n",
3212 __func__, lreq, lreq->linger_id, lreq->map_dne_bound,
3214 if (!lreq->map_dne_bound)
3215 lreq->map_dne_bound = greq->u.newest;
3216 erase_linger_mc(&osdc->linger_map_checks, lreq);
3217 check_linger_pool_dne(lreq);
3221 up_write(&osdc->lock);
3224 static void send_linger_map_check(struct ceph_osd_linger_request *lreq)
3226 struct ceph_osd_client *osdc = lreq->osdc;
3227 struct ceph_osd_linger_request *lookup_lreq;
3230 verify_osdc_wrlocked(osdc);
3232 lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks,
3235 WARN_ON(lookup_lreq != lreq);
3240 insert_linger_mc(&osdc->linger_map_checks, lreq);
3241 ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap",
3242 linger_map_check_cb, lreq->linger_id);
3246 static int linger_reg_commit_wait(struct ceph_osd_linger_request *lreq)
3250 dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
3251 ret = wait_for_completion_interruptible(&lreq->reg_commit_wait);
3252 return ret ?: lreq->reg_commit_error;
3255 static int linger_notify_finish_wait(struct ceph_osd_linger_request *lreq)
3259 dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
3260 ret = wait_for_completion_interruptible(&lreq->notify_finish_wait);
3261 return ret ?: lreq->notify_finish_error;
3265 * Timeout callback, called every N seconds. When 1 or more OSD
3266 * requests has been active for more than N seconds, we send a keepalive
3267 * (tag + timestamp) to its OSD to ensure any communications channel
3268 * reset is detected.
3270 static void handle_timeout(struct work_struct *work)
3272 struct ceph_osd_client *osdc =
3273 container_of(work, struct ceph_osd_client, timeout_work.work);
3274 struct ceph_options *opts = osdc->client->options;
3275 unsigned long cutoff = jiffies - opts->osd_keepalive_timeout;
3276 unsigned long expiry_cutoff = jiffies - opts->osd_request_timeout;
3277 LIST_HEAD(slow_osds);
3278 struct rb_node *n, *p;
3280 dout("%s osdc %p\n", __func__, osdc);
3281 down_write(&osdc->lock);
3284 * ping osds that are a bit slow. this ensures that if there
3285 * is a break in the TCP connection we will notice, and reopen
3286 * a connection with that osd (from the fault callback).
3288 for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
3289 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
3292 for (p = rb_first(&osd->o_requests); p; ) {
3293 struct ceph_osd_request *req =
3294 rb_entry(p, struct ceph_osd_request, r_node);
3296 p = rb_next(p); /* abort_request() */
3298 if (time_before(req->r_stamp, cutoff)) {
3299 dout(" req %p tid %llu on osd%d is laggy\n",
3300 req, req->r_tid, osd->o_osd);
3303 if (opts->osd_request_timeout &&
3304 time_before(req->r_start_stamp, expiry_cutoff)) {
3305 pr_err_ratelimited("tid %llu on osd%d timeout\n",
3306 req->r_tid, osd->o_osd);
3307 abort_request(req, -ETIMEDOUT);
3310 for (p = rb_first(&osd->o_linger_requests); p; p = rb_next(p)) {
3311 struct ceph_osd_linger_request *lreq =
3312 rb_entry(p, struct ceph_osd_linger_request, node);
3314 dout(" lreq %p linger_id %llu is served by osd%d\n",
3315 lreq, lreq->linger_id, osd->o_osd);
3318 mutex_lock(&lreq->lock);
3319 if (lreq->is_watch && lreq->committed && !lreq->last_error)
3320 send_linger_ping(lreq);
3321 mutex_unlock(&lreq->lock);
3325 list_move_tail(&osd->o_keepalive_item, &slow_osds);
3328 if (opts->osd_request_timeout) {
3329 for (p = rb_first(&osdc->homeless_osd.o_requests); p; ) {
3330 struct ceph_osd_request *req =
3331 rb_entry(p, struct ceph_osd_request, r_node);
3333 p = rb_next(p); /* abort_request() */
3335 if (time_before(req->r_start_stamp, expiry_cutoff)) {
3336 pr_err_ratelimited("tid %llu on osd%d timeout\n",
3337 req->r_tid, osdc->homeless_osd.o_osd);
3338 abort_request(req, -ETIMEDOUT);
3343 if (atomic_read(&osdc->num_homeless) || !list_empty(&slow_osds))
3344 maybe_request_map(osdc);
3346 while (!list_empty(&slow_osds)) {
3347 struct ceph_osd *osd = list_first_entry(&slow_osds,
3350 list_del_init(&osd->o_keepalive_item);
3351 ceph_con_keepalive(&osd->o_con);
3354 up_write(&osdc->lock);
3355 schedule_delayed_work(&osdc->timeout_work,
3356 osdc->client->options->osd_keepalive_timeout);
3359 static void handle_osds_timeout(struct work_struct *work)
3361 struct ceph_osd_client *osdc =
3362 container_of(work, struct ceph_osd_client,
3363 osds_timeout_work.work);
3364 unsigned long delay = osdc->client->options->osd_idle_ttl / 4;
3365 struct ceph_osd *osd, *nosd;
3367 dout("%s osdc %p\n", __func__, osdc);
3368 down_write(&osdc->lock);
3369 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
3370 if (time_before(jiffies, osd->lru_ttl))
3373 WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests));
3374 WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests));
3378 up_write(&osdc->lock);
3379 schedule_delayed_work(&osdc->osds_timeout_work,
3380 round_jiffies_relative(delay));
3383 static int ceph_oloc_decode(void **p, void *end,
3384 struct ceph_object_locator *oloc)
3386 u8 struct_v, struct_cv;
3391 ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
3392 struct_v = ceph_decode_8(p);
3393 struct_cv = ceph_decode_8(p);
3395 pr_warn("got v %d < 3 cv %d of ceph_object_locator\n",
3396 struct_v, struct_cv);
3399 if (struct_cv > 6) {
3400 pr_warn("got v %d cv %d > 6 of ceph_object_locator\n",
3401 struct_v, struct_cv);
3404 len = ceph_decode_32(p);
3405 ceph_decode_need(p, end, len, e_inval);
3406 struct_end = *p + len;
3408 oloc->pool = ceph_decode_64(p);
3409 *p += 4; /* skip preferred */
3411 len = ceph_decode_32(p);
3413 pr_warn("ceph_object_locator::key is set\n");
3417 if (struct_v >= 5) {
3418 bool changed = false;
3420 len = ceph_decode_32(p);
3422 ceph_decode_need(p, end, len, e_inval);
3423 if (!oloc->pool_ns ||
3424 ceph_compare_string(oloc->pool_ns, *p, len))
3432 /* redirect changes namespace */
3433 pr_warn("ceph_object_locator::nspace is changed\n");
3438 if (struct_v >= 6) {
3439 s64 hash = ceph_decode_64(p);
3441 pr_warn("ceph_object_locator::hash is set\n");
3456 static int ceph_redirect_decode(void **p, void *end,
3457 struct ceph_request_redirect *redir)
3459 u8 struct_v, struct_cv;
3464 ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
3465 struct_v = ceph_decode_8(p);
3466 struct_cv = ceph_decode_8(p);
3467 if (struct_cv > 1) {
3468 pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n",
3469 struct_v, struct_cv);
3472 len = ceph_decode_32(p);
3473 ceph_decode_need(p, end, len, e_inval);
3474 struct_end = *p + len;
3476 ret = ceph_oloc_decode(p, end, &redir->oloc);
3480 len = ceph_decode_32(p);
3482 pr_warn("ceph_request_redirect::object_name is set\n");
3486 len = ceph_decode_32(p);
3487 *p += len; /* skip osd_instructions */
3499 struct MOSDOpReply {
3500 struct ceph_pg pgid;
3505 u32 outdata_len[CEPH_OSD_MAX_OPS];
3506 s32 rval[CEPH_OSD_MAX_OPS];
3508 struct ceph_eversion replay_version;
3510 struct ceph_request_redirect redirect;
3513 static int decode_MOSDOpReply(const struct ceph_msg *msg, struct MOSDOpReply *m)
3515 void *p = msg->front.iov_base;
3516 void *const end = p + msg->front.iov_len;
3517 u16 version = le16_to_cpu(msg->hdr.version);
3518 struct ceph_eversion bad_replay_version;
3524 ceph_decode_32_safe(&p, end, len, e_inval);
3525 ceph_decode_need(&p, end, len, e_inval);
3526 p += len; /* skip oid */
3528 ret = ceph_decode_pgid(&p, end, &m->pgid);
3532 ceph_decode_64_safe(&p, end, m->flags, e_inval);
3533 ceph_decode_32_safe(&p, end, m->result, e_inval);
3534 ceph_decode_need(&p, end, sizeof(bad_replay_version), e_inval);
3535 memcpy(&bad_replay_version, p, sizeof(bad_replay_version));
3536 p += sizeof(bad_replay_version);
3537 ceph_decode_32_safe(&p, end, m->epoch, e_inval);
3539 ceph_decode_32_safe(&p, end, m->num_ops, e_inval);
3540 if (m->num_ops > ARRAY_SIZE(m->outdata_len))
3543 ceph_decode_need(&p, end, m->num_ops * sizeof(struct ceph_osd_op),
3545 for (i = 0; i < m->num_ops; i++) {
3546 struct ceph_osd_op *op = p;
3548 m->outdata_len[i] = le32_to_cpu(op->payload_len);
3552 ceph_decode_32_safe(&p, end, m->retry_attempt, e_inval);
3553 for (i = 0; i < m->num_ops; i++)
3554 ceph_decode_32_safe(&p, end, m->rval[i], e_inval);
3557 ceph_decode_need(&p, end, sizeof(m->replay_version), e_inval);
3558 memcpy(&m->replay_version, p, sizeof(m->replay_version));
3559 p += sizeof(m->replay_version);
3560 ceph_decode_64_safe(&p, end, m->user_version, e_inval);
3562 m->replay_version = bad_replay_version; /* struct */
3563 m->user_version = le64_to_cpu(m->replay_version.version);
3568 ceph_decode_8_safe(&p, end, decode_redir, e_inval);
3576 ret = ceph_redirect_decode(&p, end, &m->redirect);
3580 ceph_oloc_init(&m->redirect.oloc);
3590 * Handle MOSDOpReply. Set ->r_result and call the callback if it is
3593 static void handle_reply(struct ceph_osd *osd, struct ceph_msg *msg)
3595 struct ceph_osd_client *osdc = osd->o_osdc;
3596 struct ceph_osd_request *req;
3597 struct MOSDOpReply m;
3598 u64 tid = le64_to_cpu(msg->hdr.tid);
3603 dout("%s msg %p tid %llu\n", __func__, msg, tid);
3605 down_read(&osdc->lock);
3606 if (!osd_registered(osd)) {
3607 dout("%s osd%d unknown\n", __func__, osd->o_osd);
3608 goto out_unlock_osdc;
3610 WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num));
3612 mutex_lock(&osd->lock);
3613 req = lookup_request(&osd->o_requests, tid);
3615 dout("%s osd%d tid %llu unknown\n", __func__, osd->o_osd, tid);
3616 goto out_unlock_session;
3619 m.redirect.oloc.pool_ns = req->r_t.target_oloc.pool_ns;
3620 ret = decode_MOSDOpReply(msg, &m);
3621 m.redirect.oloc.pool_ns = NULL;
3623 pr_err("failed to decode MOSDOpReply for tid %llu: %d\n",
3628 dout("%s req %p tid %llu flags 0x%llx pgid %llu.%x epoch %u attempt %d v %u'%llu uv %llu\n",
3629 __func__, req, req->r_tid, m.flags, m.pgid.pool, m.pgid.seed,
3630 m.epoch, m.retry_attempt, le32_to_cpu(m.replay_version.epoch),
3631 le64_to_cpu(m.replay_version.version), m.user_version);
3633 if (m.retry_attempt >= 0) {
3634 if (m.retry_attempt != req->r_attempts - 1) {
3635 dout("req %p tid %llu retry_attempt %d != %d, ignoring\n",
3636 req, req->r_tid, m.retry_attempt,
3637 req->r_attempts - 1);
3638 goto out_unlock_session;
3641 WARN_ON(1); /* MOSDOpReply v4 is assumed */
3644 if (!ceph_oloc_empty(&m.redirect.oloc)) {
3645 dout("req %p tid %llu redirect pool %lld\n", req, req->r_tid,
3646 m.redirect.oloc.pool);
3647 unlink_request(osd, req);
3648 mutex_unlock(&osd->lock);
3651 * Not ceph_oloc_copy() - changing pool_ns is not
3654 req->r_t.target_oloc.pool = m.redirect.oloc.pool;
3655 req->r_flags |= CEPH_OSD_FLAG_REDIRECTED;
3657 __submit_request(req, false);
3658 goto out_unlock_osdc;
3661 if (m.num_ops != req->r_num_ops) {
3662 pr_err("num_ops %d != %d for tid %llu\n", m.num_ops,
3663 req->r_num_ops, req->r_tid);
3666 for (i = 0; i < req->r_num_ops; i++) {
3667 dout(" req %p tid %llu op %d rval %d len %u\n", req,
3668 req->r_tid, i, m.rval[i], m.outdata_len[i]);
3669 req->r_ops[i].rval = m.rval[i];
3670 req->r_ops[i].outdata_len = m.outdata_len[i];
3671 data_len += m.outdata_len[i];
3673 if (data_len != le32_to_cpu(msg->hdr.data_len)) {
3674 pr_err("sum of lens %u != %u for tid %llu\n", data_len,
3675 le32_to_cpu(msg->hdr.data_len), req->r_tid);
3678 dout("%s req %p tid %llu result %d data_len %u\n", __func__,
3679 req, req->r_tid, m.result, data_len);
3682 * Since we only ever request ONDISK, we should only ever get
3683 * one (type of) reply back.
3685 WARN_ON(!(m.flags & CEPH_OSD_FLAG_ONDISK));
3686 req->r_result = m.result ?: data_len;
3687 finish_request(req);
3688 mutex_unlock(&osd->lock);
3689 up_read(&osdc->lock);
3691 __complete_request(req);
3695 complete_request(req, -EIO);
3697 mutex_unlock(&osd->lock);
3699 up_read(&osdc->lock);
3702 static void set_pool_was_full(struct ceph_osd_client *osdc)
3706 for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) {
3707 struct ceph_pg_pool_info *pi =
3708 rb_entry(n, struct ceph_pg_pool_info, node);
3710 pi->was_full = __pool_full(pi);
3714 static bool pool_cleared_full(struct ceph_osd_client *osdc, s64 pool_id)
3716 struct ceph_pg_pool_info *pi;
3718 pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id);
3722 return pi->was_full && !__pool_full(pi);
3725 static enum calc_target_result
3726 recalc_linger_target(struct ceph_osd_linger_request *lreq)
3728 struct ceph_osd_client *osdc = lreq->osdc;
3729 enum calc_target_result ct_res;
3731 ct_res = calc_target(osdc, &lreq->t, true);
3732 if (ct_res == CALC_TARGET_NEED_RESEND) {
3733 struct ceph_osd *osd;
3735 osd = lookup_create_osd(osdc, lreq->t.osd, true);
3736 if (osd != lreq->osd) {
3737 unlink_linger(lreq->osd, lreq);
3738 link_linger(osd, lreq);
3746 * Requeue requests whose mapping to an OSD has changed.
3748 static void scan_requests(struct ceph_osd *osd,
3751 bool check_pool_cleared_full,
3752 struct rb_root *need_resend,
3753 struct list_head *need_resend_linger)
3755 struct ceph_osd_client *osdc = osd->o_osdc;
3757 bool force_resend_writes;
3759 for (n = rb_first(&osd->o_linger_requests); n; ) {
3760 struct ceph_osd_linger_request *lreq =
3761 rb_entry(n, struct ceph_osd_linger_request, node);
3762 enum calc_target_result ct_res;
3764 n = rb_next(n); /* recalc_linger_target() */
3766 dout("%s lreq %p linger_id %llu\n", __func__, lreq,
3768 ct_res = recalc_linger_target(lreq);
3770 case CALC_TARGET_NO_ACTION:
3771 force_resend_writes = cleared_full ||
3772 (check_pool_cleared_full &&
3773 pool_cleared_full(osdc, lreq->t.base_oloc.pool));
3774 if (!force_resend && !force_resend_writes)
3778 case CALC_TARGET_NEED_RESEND:
3779 cancel_linger_map_check(lreq);
3781 * scan_requests() for the previous epoch(s)
3782 * may have already added it to the list, since
3783 * it's not unlinked here.
3785 if (list_empty(&lreq->scan_item))
3786 list_add_tail(&lreq->scan_item, need_resend_linger);
3788 case CALC_TARGET_POOL_DNE:
3789 list_del_init(&lreq->scan_item);
3790 check_linger_pool_dne(lreq);
3795 for (n = rb_first(&osd->o_requests); n; ) {
3796 struct ceph_osd_request *req =
3797 rb_entry(n, struct ceph_osd_request, r_node);
3798 enum calc_target_result ct_res;
3800 n = rb_next(n); /* unlink_request(), check_pool_dne() */
3802 dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
3803 ct_res = calc_target(osdc, &req->r_t, false);
3805 case CALC_TARGET_NO_ACTION:
3806 force_resend_writes = cleared_full ||
3807 (check_pool_cleared_full &&
3808 pool_cleared_full(osdc, req->r_t.base_oloc.pool));
3809 if (!force_resend &&
3810 (!(req->r_flags & CEPH_OSD_FLAG_WRITE) ||
3811 !force_resend_writes))
3815 case CALC_TARGET_NEED_RESEND:
3816 cancel_map_check(req);
3817 unlink_request(osd, req);
3818 insert_request(need_resend, req);
3820 case CALC_TARGET_POOL_DNE:
3821 check_pool_dne(req);
3827 static int handle_one_map(struct ceph_osd_client *osdc,
3828 void *p, void *end, bool incremental,
3829 struct rb_root *need_resend,
3830 struct list_head *need_resend_linger)
3832 struct ceph_osdmap *newmap;
3834 bool skipped_map = false;
3837 was_full = ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL);
3838 set_pool_was_full(osdc);
3841 newmap = osdmap_apply_incremental(&p, end, osdc->osdmap);
3843 newmap = ceph_osdmap_decode(&p, end);
3845 return PTR_ERR(newmap);
3847 if (newmap != osdc->osdmap) {
3849 * Preserve ->was_full before destroying the old map.
3850 * For pools that weren't in the old map, ->was_full
3853 for (n = rb_first(&newmap->pg_pools); n; n = rb_next(n)) {
3854 struct ceph_pg_pool_info *pi =
3855 rb_entry(n, struct ceph_pg_pool_info, node);
3856 struct ceph_pg_pool_info *old_pi;
3858 old_pi = ceph_pg_pool_by_id(osdc->osdmap, pi->id);
3860 pi->was_full = old_pi->was_full;
3862 WARN_ON(pi->was_full);
3865 if (osdc->osdmap->epoch &&
3866 osdc->osdmap->epoch + 1 < newmap->epoch) {
3867 WARN_ON(incremental);
3871 ceph_osdmap_destroy(osdc->osdmap);
3872 osdc->osdmap = newmap;
3875 was_full &= !ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL);
3876 scan_requests(&osdc->homeless_osd, skipped_map, was_full, true,
3877 need_resend, need_resend_linger);
3879 for (n = rb_first(&osdc->osds); n; ) {
3880 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
3882 n = rb_next(n); /* close_osd() */
3884 scan_requests(osd, skipped_map, was_full, true, need_resend,
3885 need_resend_linger);
3886 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
3887 memcmp(&osd->o_con.peer_addr,
3888 ceph_osd_addr(osdc->osdmap, osd->o_osd),
3889 sizeof(struct ceph_entity_addr)))
3896 static void kick_requests(struct ceph_osd_client *osdc,
3897 struct rb_root *need_resend,
3898 struct list_head *need_resend_linger)
3900 struct ceph_osd_linger_request *lreq, *nlreq;
3901 enum calc_target_result ct_res;
3904 /* make sure need_resend targets reflect latest map */
3905 for (n = rb_first(need_resend); n; ) {
3906 struct ceph_osd_request *req =
3907 rb_entry(n, struct ceph_osd_request, r_node);
3911 if (req->r_t.epoch < osdc->osdmap->epoch) {
3912 ct_res = calc_target(osdc, &req->r_t, false);
3913 if (ct_res == CALC_TARGET_POOL_DNE) {
3914 erase_request(need_resend, req);
3915 check_pool_dne(req);
3920 for (n = rb_first(need_resend); n; ) {
3921 struct ceph_osd_request *req =
3922 rb_entry(n, struct ceph_osd_request, r_node);
3923 struct ceph_osd *osd;
3926 erase_request(need_resend, req); /* before link_request() */
3928 osd = lookup_create_osd(osdc, req->r_t.osd, true);
3929 link_request(osd, req);
3930 if (!req->r_linger) {
3931 if (!osd_homeless(osd) && !req->r_t.paused)
3934 cancel_linger_request(req);
3938 list_for_each_entry_safe(lreq, nlreq, need_resend_linger, scan_item) {
3939 if (!osd_homeless(lreq->osd))
3942 list_del_init(&lreq->scan_item);
3947 * Process updated osd map.
3949 * The message contains any number of incremental and full maps, normally
3950 * indicating some sort of topology change in the cluster. Kick requests
3951 * off to different OSDs as needed.
3953 void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
3955 void *p = msg->front.iov_base;
3956 void *const end = p + msg->front.iov_len;
3957 u32 nr_maps, maplen;
3959 struct ceph_fsid fsid;
3960 struct rb_root need_resend = RB_ROOT;
3961 LIST_HEAD(need_resend_linger);
3962 bool handled_incremental = false;
3963 bool was_pauserd, was_pausewr;
3964 bool pauserd, pausewr;
3967 dout("%s have %u\n", __func__, osdc->osdmap->epoch);
3968 down_write(&osdc->lock);
3971 ceph_decode_need(&p, end, sizeof(fsid), bad);
3972 ceph_decode_copy(&p, &fsid, sizeof(fsid));
3973 if (ceph_check_fsid(osdc->client, &fsid) < 0)
3976 was_pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
3977 was_pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
3978 ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
3979 have_pool_full(osdc);
3981 /* incremental maps */
3982 ceph_decode_32_safe(&p, end, nr_maps, bad);
3983 dout(" %d inc maps\n", nr_maps);
3984 while (nr_maps > 0) {
3985 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
3986 epoch = ceph_decode_32(&p);
3987 maplen = ceph_decode_32(&p);
3988 ceph_decode_need(&p, end, maplen, bad);
3989 if (osdc->osdmap->epoch &&
3990 osdc->osdmap->epoch + 1 == epoch) {
3991 dout("applying incremental map %u len %d\n",
3993 err = handle_one_map(osdc, p, p + maplen, true,
3994 &need_resend, &need_resend_linger);
3997 handled_incremental = true;
3999 dout("ignoring incremental map %u len %d\n",
4005 if (handled_incremental)
4009 ceph_decode_32_safe(&p, end, nr_maps, bad);
4010 dout(" %d full maps\n", nr_maps);
4012 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
4013 epoch = ceph_decode_32(&p);
4014 maplen = ceph_decode_32(&p);
4015 ceph_decode_need(&p, end, maplen, bad);
4017 dout("skipping non-latest full map %u len %d\n",
4019 } else if (osdc->osdmap->epoch >= epoch) {
4020 dout("skipping full map %u len %d, "
4021 "older than our %u\n", epoch, maplen,
4022 osdc->osdmap->epoch);
4024 dout("taking full map %u len %d\n", epoch, maplen);
4025 err = handle_one_map(osdc, p, p + maplen, false,
4026 &need_resend, &need_resend_linger);
4036 * subscribe to subsequent osdmap updates if full to ensure
4037 * we find out when we are no longer full and stop returning
4040 pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
4041 pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
4042 ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
4043 have_pool_full(osdc);
4044 if (was_pauserd || was_pausewr || pauserd || pausewr ||
4045 osdc->osdmap->epoch < osdc->epoch_barrier)
4046 maybe_request_map(osdc);
4048 kick_requests(osdc, &need_resend, &need_resend_linger);
4050 ceph_osdc_abort_on_full(osdc);
4051 ceph_monc_got_map(&osdc->client->monc, CEPH_SUB_OSDMAP,
4052 osdc->osdmap->epoch);
4053 up_write(&osdc->lock);
4054 wake_up_all(&osdc->client->auth_wq);
4058 pr_err("osdc handle_map corrupt msg\n");
4060 up_write(&osdc->lock);
4064 * Resubmit requests pending on the given osd.
4066 static void kick_osd_requests(struct ceph_osd *osd)
4070 clear_backoffs(osd);
4072 for (n = rb_first(&osd->o_requests); n; ) {
4073 struct ceph_osd_request *req =
4074 rb_entry(n, struct ceph_osd_request, r_node);
4076 n = rb_next(n); /* cancel_linger_request() */
4078 if (!req->r_linger) {
4079 if (!req->r_t.paused)
4082 cancel_linger_request(req);
4085 for (n = rb_first(&osd->o_linger_requests); n; n = rb_next(n)) {
4086 struct ceph_osd_linger_request *lreq =
4087 rb_entry(n, struct ceph_osd_linger_request, node);
4094 * If the osd connection drops, we need to resubmit all requests.
4096 static void osd_fault(struct ceph_connection *con)
4098 struct ceph_osd *osd = con->private;
4099 struct ceph_osd_client *osdc = osd->o_osdc;
4101 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
4103 down_write(&osdc->lock);
4104 if (!osd_registered(osd)) {
4105 dout("%s osd%d unknown\n", __func__, osd->o_osd);
4109 if (!reopen_osd(osd))
4110 kick_osd_requests(osd);
4111 maybe_request_map(osdc);
4114 up_write(&osdc->lock);
4117 struct MOSDBackoff {
4118 struct ceph_spg spgid;
4122 struct ceph_hobject_id *begin;
4123 struct ceph_hobject_id *end;
4126 static int decode_MOSDBackoff(const struct ceph_msg *msg, struct MOSDBackoff *m)
4128 void *p = msg->front.iov_base;
4129 void *const end = p + msg->front.iov_len;
4134 ret = ceph_start_decoding(&p, end, 1, "spg_t", &struct_v, &struct_len);
4138 ret = ceph_decode_pgid(&p, end, &m->spgid.pgid);
4142 ceph_decode_8_safe(&p, end, m->spgid.shard, e_inval);
4143 ceph_decode_32_safe(&p, end, m->map_epoch, e_inval);
4144 ceph_decode_8_safe(&p, end, m->op, e_inval);
4145 ceph_decode_64_safe(&p, end, m->id, e_inval);
4147 m->begin = kzalloc(sizeof(*m->begin), GFP_NOIO);
4151 ret = decode_hoid(&p, end, m->begin);
4153 free_hoid(m->begin);
4157 m->end = kzalloc(sizeof(*m->end), GFP_NOIO);
4159 free_hoid(m->begin);
4163 ret = decode_hoid(&p, end, m->end);
4165 free_hoid(m->begin);
4176 static struct ceph_msg *create_backoff_message(
4177 const struct ceph_osd_backoff *backoff,
4180 struct ceph_msg *msg;
4184 msg_size = CEPH_ENCODING_START_BLK_LEN +
4185 CEPH_PGID_ENCODING_LEN + 1; /* spgid */
4186 msg_size += 4 + 1 + 8; /* map_epoch, op, id */
4187 msg_size += CEPH_ENCODING_START_BLK_LEN +
4188 hoid_encoding_size(backoff->begin);
4189 msg_size += CEPH_ENCODING_START_BLK_LEN +
4190 hoid_encoding_size(backoff->end);
4192 msg = ceph_msg_new(CEPH_MSG_OSD_BACKOFF, msg_size, GFP_NOIO, true);
4196 p = msg->front.iov_base;
4197 end = p + msg->front_alloc_len;
4199 encode_spgid(&p, &backoff->spgid);
4200 ceph_encode_32(&p, map_epoch);
4201 ceph_encode_8(&p, CEPH_OSD_BACKOFF_OP_ACK_BLOCK);
4202 ceph_encode_64(&p, backoff->id);
4203 encode_hoid(&p, end, backoff->begin);
4204 encode_hoid(&p, end, backoff->end);
4207 msg->front.iov_len = p - msg->front.iov_base;
4208 msg->hdr.version = cpu_to_le16(1); /* MOSDBackoff v1 */
4209 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
4214 static void handle_backoff_block(struct ceph_osd *osd, struct MOSDBackoff *m)
4216 struct ceph_spg_mapping *spg;
4217 struct ceph_osd_backoff *backoff;
4218 struct ceph_msg *msg;
4220 dout("%s osd%d spgid %llu.%xs%d id %llu\n", __func__, osd->o_osd,
4221 m->spgid.pgid.pool, m->spgid.pgid.seed, m->spgid.shard, m->id);
4223 spg = lookup_spg_mapping(&osd->o_backoff_mappings, &m->spgid);
4225 spg = alloc_spg_mapping();
4227 pr_err("%s failed to allocate spg\n", __func__);
4230 spg->spgid = m->spgid; /* struct */
4231 insert_spg_mapping(&osd->o_backoff_mappings, spg);
4234 backoff = alloc_backoff();
4236 pr_err("%s failed to allocate backoff\n", __func__);
4239 backoff->spgid = m->spgid; /* struct */
4240 backoff->id = m->id;
4241 backoff->begin = m->begin;
4242 m->begin = NULL; /* backoff now owns this */
4243 backoff->end = m->end;
4244 m->end = NULL; /* ditto */
4246 insert_backoff(&spg->backoffs, backoff);
4247 insert_backoff_by_id(&osd->o_backoffs_by_id, backoff);
4250 * Ack with original backoff's epoch so that the OSD can
4251 * discard this if there was a PG split.
4253 msg = create_backoff_message(backoff, m->map_epoch);
4255 pr_err("%s failed to allocate msg\n", __func__);
4258 ceph_con_send(&osd->o_con, msg);
4261 static bool target_contained_by(const struct ceph_osd_request_target *t,
4262 const struct ceph_hobject_id *begin,
4263 const struct ceph_hobject_id *end)
4265 struct ceph_hobject_id hoid;
4268 hoid_fill_from_target(&hoid, t);
4269 cmp = hoid_compare(&hoid, begin);
4270 return !cmp || (cmp > 0 && hoid_compare(&hoid, end) < 0);
4273 static void handle_backoff_unblock(struct ceph_osd *osd,
4274 const struct MOSDBackoff *m)
4276 struct ceph_spg_mapping *spg;
4277 struct ceph_osd_backoff *backoff;
4280 dout("%s osd%d spgid %llu.%xs%d id %llu\n", __func__, osd->o_osd,
4281 m->spgid.pgid.pool, m->spgid.pgid.seed, m->spgid.shard, m->id);
4283 backoff = lookup_backoff_by_id(&osd->o_backoffs_by_id, m->id);
4285 pr_err("%s osd%d spgid %llu.%xs%d id %llu backoff dne\n",
4286 __func__, osd->o_osd, m->spgid.pgid.pool,
4287 m->spgid.pgid.seed, m->spgid.shard, m->id);
4291 if (hoid_compare(backoff->begin, m->begin) &&
4292 hoid_compare(backoff->end, m->end)) {
4293 pr_err("%s osd%d spgid %llu.%xs%d id %llu bad range?\n",
4294 __func__, osd->o_osd, m->spgid.pgid.pool,
4295 m->spgid.pgid.seed, m->spgid.shard, m->id);
4296 /* unblock it anyway... */
4299 spg = lookup_spg_mapping(&osd->o_backoff_mappings, &backoff->spgid);
4302 erase_backoff(&spg->backoffs, backoff);
4303 erase_backoff_by_id(&osd->o_backoffs_by_id, backoff);
4304 free_backoff(backoff);
4306 if (RB_EMPTY_ROOT(&spg->backoffs)) {
4307 erase_spg_mapping(&osd->o_backoff_mappings, spg);
4308 free_spg_mapping(spg);
4311 for (n = rb_first(&osd->o_requests); n; n = rb_next(n)) {
4312 struct ceph_osd_request *req =
4313 rb_entry(n, struct ceph_osd_request, r_node);
4315 if (!ceph_spg_compare(&req->r_t.spgid, &m->spgid)) {
4317 * Match against @m, not @backoff -- the PG may
4318 * have split on the OSD.
4320 if (target_contained_by(&req->r_t, m->begin, m->end)) {
4322 * If no other installed backoff applies,
4331 static void handle_backoff(struct ceph_osd *osd, struct ceph_msg *msg)
4333 struct ceph_osd_client *osdc = osd->o_osdc;
4334 struct MOSDBackoff m;
4337 down_read(&osdc->lock);
4338 if (!osd_registered(osd)) {
4339 dout("%s osd%d unknown\n", __func__, osd->o_osd);
4340 up_read(&osdc->lock);
4343 WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num));
4345 mutex_lock(&osd->lock);
4346 ret = decode_MOSDBackoff(msg, &m);
4348 pr_err("failed to decode MOSDBackoff: %d\n", ret);
4354 case CEPH_OSD_BACKOFF_OP_BLOCK:
4355 handle_backoff_block(osd, &m);
4357 case CEPH_OSD_BACKOFF_OP_UNBLOCK:
4358 handle_backoff_unblock(osd, &m);
4361 pr_err("%s osd%d unknown op %d\n", __func__, osd->o_osd, m.op);
4368 mutex_unlock(&osd->lock);
4369 up_read(&osdc->lock);
4373 * Process osd watch notifications
4375 static void handle_watch_notify(struct ceph_osd_client *osdc,
4376 struct ceph_msg *msg)
4378 void *p = msg->front.iov_base;
4379 void *const end = p + msg->front.iov_len;
4380 struct ceph_osd_linger_request *lreq;
4381 struct linger_work *lwork;
4382 u8 proto_ver, opcode;
4383 u64 cookie, notify_id;
4384 u64 notifier_id = 0;
4385 s32 return_code = 0;
4386 void *payload = NULL;
4387 u32 payload_len = 0;
4389 ceph_decode_8_safe(&p, end, proto_ver, bad);
4390 ceph_decode_8_safe(&p, end, opcode, bad);
4391 ceph_decode_64_safe(&p, end, cookie, bad);
4392 p += 8; /* skip ver */
4393 ceph_decode_64_safe(&p, end, notify_id, bad);
4395 if (proto_ver >= 1) {
4396 ceph_decode_32_safe(&p, end, payload_len, bad);
4397 ceph_decode_need(&p, end, payload_len, bad);
4402 if (le16_to_cpu(msg->hdr.version) >= 2)
4403 ceph_decode_32_safe(&p, end, return_code, bad);
4405 if (le16_to_cpu(msg->hdr.version) >= 3)
4406 ceph_decode_64_safe(&p, end, notifier_id, bad);
4408 down_read(&osdc->lock);
4409 lreq = lookup_linger_osdc(&osdc->linger_requests, cookie);
4411 dout("%s opcode %d cookie %llu dne\n", __func__, opcode,
4413 goto out_unlock_osdc;
4416 mutex_lock(&lreq->lock);
4417 dout("%s opcode %d cookie %llu lreq %p is_watch %d\n", __func__,
4418 opcode, cookie, lreq, lreq->is_watch);
4419 if (opcode == CEPH_WATCH_EVENT_DISCONNECT) {
4420 if (!lreq->last_error) {
4421 lreq->last_error = -ENOTCONN;
4422 queue_watch_error(lreq);
4424 } else if (!lreq->is_watch) {
4425 /* CEPH_WATCH_EVENT_NOTIFY_COMPLETE */
4426 if (lreq->notify_id && lreq->notify_id != notify_id) {
4427 dout("lreq %p notify_id %llu != %llu, ignoring\n", lreq,
4428 lreq->notify_id, notify_id);
4429 } else if (!completion_done(&lreq->notify_finish_wait)) {
4430 struct ceph_msg_data *data =
4431 msg->num_data_items ? &msg->data[0] : NULL;
4434 if (lreq->preply_pages) {
4435 WARN_ON(data->type !=
4436 CEPH_MSG_DATA_PAGES);
4437 *lreq->preply_pages = data->pages;
4438 *lreq->preply_len = data->length;
4440 ceph_release_page_vector(data->pages,
4441 calc_pages_for(0, data->length));
4444 lreq->notify_finish_error = return_code;
4445 complete_all(&lreq->notify_finish_wait);
4448 /* CEPH_WATCH_EVENT_NOTIFY */
4449 lwork = lwork_alloc(lreq, do_watch_notify);
4451 pr_err("failed to allocate notify-lwork\n");
4452 goto out_unlock_lreq;
4455 lwork->notify.notify_id = notify_id;
4456 lwork->notify.notifier_id = notifier_id;
4457 lwork->notify.payload = payload;
4458 lwork->notify.payload_len = payload_len;
4459 lwork->notify.msg = ceph_msg_get(msg);
4464 mutex_unlock(&lreq->lock);
4466 up_read(&osdc->lock);
4470 pr_err("osdc handle_watch_notify corrupt msg\n");
4474 * Register request, send initial attempt.
4476 int ceph_osdc_start_request(struct ceph_osd_client *osdc,
4477 struct ceph_osd_request *req,
4480 down_read(&osdc->lock);
4481 submit_request(req, false);
4482 up_read(&osdc->lock);
4486 EXPORT_SYMBOL(ceph_osdc_start_request);
4489 * Unregister a registered request. The request is not completed:
4490 * ->r_result isn't set and __complete_request() isn't called.
4492 void ceph_osdc_cancel_request(struct ceph_osd_request *req)
4494 struct ceph_osd_client *osdc = req->r_osdc;
4496 down_write(&osdc->lock);
4498 cancel_request(req);
4499 up_write(&osdc->lock);
4501 EXPORT_SYMBOL(ceph_osdc_cancel_request);
4504 * @timeout: in jiffies, 0 means "wait forever"
4506 static int wait_request_timeout(struct ceph_osd_request *req,
4507 unsigned long timeout)
4511 dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
4512 left = wait_for_completion_killable_timeout(&req->r_completion,
4513 ceph_timeout_jiffies(timeout));
4515 left = left ?: -ETIMEDOUT;
4516 ceph_osdc_cancel_request(req);
4518 left = req->r_result; /* completed */
4525 * wait for a request to complete
4527 int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
4528 struct ceph_osd_request *req)
4530 return wait_request_timeout(req, 0);
4532 EXPORT_SYMBOL(ceph_osdc_wait_request);
4535 * sync - wait for all in-flight requests to flush. avoid starvation.
4537 void ceph_osdc_sync(struct ceph_osd_client *osdc)
4539 struct rb_node *n, *p;
4540 u64 last_tid = atomic64_read(&osdc->last_tid);
4543 down_read(&osdc->lock);
4544 for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
4545 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
4547 mutex_lock(&osd->lock);
4548 for (p = rb_first(&osd->o_requests); p; p = rb_next(p)) {
4549 struct ceph_osd_request *req =
4550 rb_entry(p, struct ceph_osd_request, r_node);
4552 if (req->r_tid > last_tid)
4555 if (!(req->r_flags & CEPH_OSD_FLAG_WRITE))
4558 ceph_osdc_get_request(req);
4559 mutex_unlock(&osd->lock);
4560 up_read(&osdc->lock);
4561 dout("%s waiting on req %p tid %llu last_tid %llu\n",
4562 __func__, req, req->r_tid, last_tid);
4563 wait_for_completion(&req->r_completion);
4564 ceph_osdc_put_request(req);
4568 mutex_unlock(&osd->lock);
4571 up_read(&osdc->lock);
4572 dout("%s done last_tid %llu\n", __func__, last_tid);
4574 EXPORT_SYMBOL(ceph_osdc_sync);
4576 static struct ceph_osd_request *
4577 alloc_linger_request(struct ceph_osd_linger_request *lreq)
4579 struct ceph_osd_request *req;
4581 req = ceph_osdc_alloc_request(lreq->osdc, NULL, 1, false, GFP_NOIO);
4585 ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
4586 ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
4590 static struct ceph_osd_request *
4591 alloc_watch_request(struct ceph_osd_linger_request *lreq, u8 watch_opcode)
4593 struct ceph_osd_request *req;
4595 req = alloc_linger_request(lreq);
4600 * Pass 0 for cookie because we don't know it yet, it will be
4601 * filled in by linger_submit().
4603 osd_req_op_watch_init(req, 0, 0, watch_opcode);
4605 if (ceph_osdc_alloc_messages(req, GFP_NOIO)) {
4606 ceph_osdc_put_request(req);
4614 * Returns a handle, caller owns a ref.
4616 struct ceph_osd_linger_request *
4617 ceph_osdc_watch(struct ceph_osd_client *osdc,
4618 struct ceph_object_id *oid,
4619 struct ceph_object_locator *oloc,
4620 rados_watchcb2_t wcb,
4621 rados_watcherrcb_t errcb,
4624 struct ceph_osd_linger_request *lreq;
4627 lreq = linger_alloc(osdc);
4629 return ERR_PTR(-ENOMEM);
4631 lreq->is_watch = true;
4633 lreq->errcb = errcb;
4635 lreq->watch_valid_thru = jiffies;
4637 ceph_oid_copy(&lreq->t.base_oid, oid);
4638 ceph_oloc_copy(&lreq->t.base_oloc, oloc);
4639 lreq->t.flags = CEPH_OSD_FLAG_WRITE;
4640 ktime_get_real_ts64(&lreq->mtime);
4642 lreq->reg_req = alloc_watch_request(lreq, CEPH_OSD_WATCH_OP_WATCH);
4643 if (!lreq->reg_req) {
4648 lreq->ping_req = alloc_watch_request(lreq, CEPH_OSD_WATCH_OP_PING);
4649 if (!lreq->ping_req) {
4654 linger_submit(lreq);
4655 ret = linger_reg_commit_wait(lreq);
4657 linger_cancel(lreq);
4665 return ERR_PTR(ret);
4667 EXPORT_SYMBOL(ceph_osdc_watch);
4672 * Times out after mount_timeout to preserve rbd unmap behaviour
4673 * introduced in 2894e1d76974 ("rbd: timeout watch teardown on unmap
4674 * with mount_timeout").
4676 int ceph_osdc_unwatch(struct ceph_osd_client *osdc,
4677 struct ceph_osd_linger_request *lreq)
4679 struct ceph_options *opts = osdc->client->options;
4680 struct ceph_osd_request *req;
4683 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
4687 ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
4688 ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
4689 req->r_flags = CEPH_OSD_FLAG_WRITE;
4690 ktime_get_real_ts64(&req->r_mtime);
4691 osd_req_op_watch_init(req, 0, lreq->linger_id,
4692 CEPH_OSD_WATCH_OP_UNWATCH);
4694 ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
4698 ceph_osdc_start_request(osdc, req, false);
4699 linger_cancel(lreq);
4701 ret = wait_request_timeout(req, opts->mount_timeout);
4704 ceph_osdc_put_request(req);
4707 EXPORT_SYMBOL(ceph_osdc_unwatch);
4709 static int osd_req_op_notify_ack_init(struct ceph_osd_request *req, int which,
4710 u64 notify_id, u64 cookie, void *payload,
4713 struct ceph_osd_req_op *op;
4714 struct ceph_pagelist *pl;
4717 op = _osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY_ACK, 0);
4719 pl = ceph_pagelist_alloc(GFP_NOIO);
4723 ret = ceph_pagelist_encode_64(pl, notify_id);
4724 ret |= ceph_pagelist_encode_64(pl, cookie);
4726 ret |= ceph_pagelist_encode_32(pl, payload_len);
4727 ret |= ceph_pagelist_append(pl, payload, payload_len);
4729 ret |= ceph_pagelist_encode_32(pl, 0);
4732 ceph_pagelist_release(pl);
4736 ceph_osd_data_pagelist_init(&op->notify_ack.request_data, pl);
4737 op->indata_len = pl->length;
4741 int ceph_osdc_notify_ack(struct ceph_osd_client *osdc,
4742 struct ceph_object_id *oid,
4743 struct ceph_object_locator *oloc,
4749 struct ceph_osd_request *req;
4752 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
4756 ceph_oid_copy(&req->r_base_oid, oid);
4757 ceph_oloc_copy(&req->r_base_oloc, oloc);
4758 req->r_flags = CEPH_OSD_FLAG_READ;
4760 ret = osd_req_op_notify_ack_init(req, 0, notify_id, cookie, payload,
4765 ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
4769 ceph_osdc_start_request(osdc, req, false);
4770 ret = ceph_osdc_wait_request(osdc, req);
4773 ceph_osdc_put_request(req);
4776 EXPORT_SYMBOL(ceph_osdc_notify_ack);
4778 static int osd_req_op_notify_init(struct ceph_osd_request *req, int which,
4779 u64 cookie, u32 prot_ver, u32 timeout,
4780 void *payload, u32 payload_len)
4782 struct ceph_osd_req_op *op;
4783 struct ceph_pagelist *pl;
4786 op = _osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY, 0);
4787 op->notify.cookie = cookie;
4789 pl = ceph_pagelist_alloc(GFP_NOIO);
4793 ret = ceph_pagelist_encode_32(pl, 1); /* prot_ver */
4794 ret |= ceph_pagelist_encode_32(pl, timeout);
4795 ret |= ceph_pagelist_encode_32(pl, payload_len);
4796 ret |= ceph_pagelist_append(pl, payload, payload_len);
4798 ceph_pagelist_release(pl);
4802 ceph_osd_data_pagelist_init(&op->notify.request_data, pl);
4803 op->indata_len = pl->length;
4808 * @timeout: in seconds
4810 * @preply_{pages,len} are initialized both on success and error.
4811 * The caller is responsible for:
4813 * ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len))
4815 int ceph_osdc_notify(struct ceph_osd_client *osdc,
4816 struct ceph_object_id *oid,
4817 struct ceph_object_locator *oloc,
4821 struct page ***preply_pages,
4824 struct ceph_osd_linger_request *lreq;
4825 struct page **pages;
4830 *preply_pages = NULL;
4834 lreq = linger_alloc(osdc);
4838 lreq->preply_pages = preply_pages;
4839 lreq->preply_len = preply_len;
4841 ceph_oid_copy(&lreq->t.base_oid, oid);
4842 ceph_oloc_copy(&lreq->t.base_oloc, oloc);
4843 lreq->t.flags = CEPH_OSD_FLAG_READ;
4845 lreq->reg_req = alloc_linger_request(lreq);
4846 if (!lreq->reg_req) {
4852 * Pass 0 for cookie because we don't know it yet, it will be
4853 * filled in by linger_submit().
4855 ret = osd_req_op_notify_init(lreq->reg_req, 0, 0, 1, timeout,
4856 payload, payload_len);
4861 pages = ceph_alloc_page_vector(1, GFP_NOIO);
4862 if (IS_ERR(pages)) {
4863 ret = PTR_ERR(pages);
4866 ceph_osd_data_pages_init(osd_req_op_data(lreq->reg_req, 0, notify,
4868 pages, PAGE_SIZE, 0, false, true);
4870 ret = ceph_osdc_alloc_messages(lreq->reg_req, GFP_NOIO);
4874 linger_submit(lreq);
4875 ret = linger_reg_commit_wait(lreq);
4877 ret = linger_notify_finish_wait(lreq);
4879 dout("lreq %p failed to initiate notify %d\n", lreq, ret);
4881 linger_cancel(lreq);
4886 EXPORT_SYMBOL(ceph_osdc_notify);
4889 * Return the number of milliseconds since the watch was last
4890 * confirmed, or an error. If there is an error, the watch is no
4891 * longer valid, and should be destroyed with ceph_osdc_unwatch().
4893 int ceph_osdc_watch_check(struct ceph_osd_client *osdc,
4894 struct ceph_osd_linger_request *lreq)
4896 unsigned long stamp, age;
4899 down_read(&osdc->lock);
4900 mutex_lock(&lreq->lock);
4901 stamp = lreq->watch_valid_thru;
4902 if (!list_empty(&lreq->pending_lworks)) {
4903 struct linger_work *lwork =
4904 list_first_entry(&lreq->pending_lworks,
4908 if (time_before(lwork->queued_stamp, stamp))
4909 stamp = lwork->queued_stamp;
4911 age = jiffies - stamp;
4912 dout("%s lreq %p linger_id %llu age %lu last_error %d\n", __func__,
4913 lreq, lreq->linger_id, age, lreq->last_error);
4914 /* we are truncating to msecs, so return a safe upper bound */
4915 ret = lreq->last_error ?: 1 + jiffies_to_msecs(age);
4917 mutex_unlock(&lreq->lock);
4918 up_read(&osdc->lock);
4922 static int decode_watcher(void **p, void *end, struct ceph_watch_item *item)
4928 ret = ceph_start_decoding(p, end, 2, "watch_item_t",
4929 &struct_v, &struct_len);
4934 ceph_decode_copy_safe(p, end, &item->name, sizeof(item->name), bad);
4935 ceph_decode_64_safe(p, end, item->cookie, bad);
4936 ceph_decode_skip_32(p, end, bad); /* skip timeout seconds */
4938 if (struct_v >= 2) {
4939 ret = ceph_decode_entity_addr(p, end, &item->addr);
4946 dout("%s %s%llu cookie %llu addr %s\n", __func__,
4947 ENTITY_NAME(item->name), item->cookie,
4948 ceph_pr_addr(&item->addr));
4953 static int decode_watchers(void **p, void *end,
4954 struct ceph_watch_item **watchers,
4962 ret = ceph_start_decoding(p, end, 1, "obj_list_watch_response_t",
4963 &struct_v, &struct_len);
4967 *num_watchers = ceph_decode_32(p);
4968 *watchers = kcalloc(*num_watchers, sizeof(**watchers), GFP_NOIO);
4972 for (i = 0; i < *num_watchers; i++) {
4973 ret = decode_watcher(p, end, *watchers + i);
4984 * On success, the caller is responsible for:
4988 int ceph_osdc_list_watchers(struct ceph_osd_client *osdc,
4989 struct ceph_object_id *oid,
4990 struct ceph_object_locator *oloc,
4991 struct ceph_watch_item **watchers,
4994 struct ceph_osd_request *req;
4995 struct page **pages;
4998 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
5002 ceph_oid_copy(&req->r_base_oid, oid);
5003 ceph_oloc_copy(&req->r_base_oloc, oloc);
5004 req->r_flags = CEPH_OSD_FLAG_READ;
5006 pages = ceph_alloc_page_vector(1, GFP_NOIO);
5007 if (IS_ERR(pages)) {
5008 ret = PTR_ERR(pages);
5012 osd_req_op_init(req, 0, CEPH_OSD_OP_LIST_WATCHERS, 0);
5013 ceph_osd_data_pages_init(osd_req_op_data(req, 0, list_watchers,
5015 pages, PAGE_SIZE, 0, false, true);
5017 ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
5021 ceph_osdc_start_request(osdc, req, false);
5022 ret = ceph_osdc_wait_request(osdc, req);
5024 void *p = page_address(pages[0]);
5025 void *const end = p + req->r_ops[0].outdata_len;
5027 ret = decode_watchers(&p, end, watchers, num_watchers);
5031 ceph_osdc_put_request(req);
5034 EXPORT_SYMBOL(ceph_osdc_list_watchers);
5037 * Call all pending notify callbacks - for use after a watch is
5038 * unregistered, to make sure no more callbacks for it will be invoked
5040 void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc)
5042 dout("%s osdc %p\n", __func__, osdc);
5043 flush_workqueue(osdc->notify_wq);
5045 EXPORT_SYMBOL(ceph_osdc_flush_notifies);
5047 void ceph_osdc_maybe_request_map(struct ceph_osd_client *osdc)
5049 down_read(&osdc->lock);
5050 maybe_request_map(osdc);
5051 up_read(&osdc->lock);
5053 EXPORT_SYMBOL(ceph_osdc_maybe_request_map);
5056 * Execute an OSD class method on an object.
5058 * @flags: CEPH_OSD_FLAG_*
5059 * @resp_len: in/out param for reply length
5061 int ceph_osdc_call(struct ceph_osd_client *osdc,
5062 struct ceph_object_id *oid,
5063 struct ceph_object_locator *oloc,
5064 const char *class, const char *method,
5066 struct page *req_page, size_t req_len,
5067 struct page **resp_pages, size_t *resp_len)
5069 struct ceph_osd_request *req;
5072 if (req_len > PAGE_SIZE)
5075 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
5079 ceph_oid_copy(&req->r_base_oid, oid);
5080 ceph_oloc_copy(&req->r_base_oloc, oloc);
5081 req->r_flags = flags;
5083 ret = osd_req_op_cls_init(req, 0, class, method);
5088 osd_req_op_cls_request_data_pages(req, 0, &req_page, req_len,
5091 osd_req_op_cls_response_data_pages(req, 0, resp_pages,
5092 *resp_len, 0, false, false);
5094 ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
5098 ceph_osdc_start_request(osdc, req, false);
5099 ret = ceph_osdc_wait_request(osdc, req);
5101 ret = req->r_ops[0].rval;
5103 *resp_len = req->r_ops[0].outdata_len;
5107 ceph_osdc_put_request(req);
5110 EXPORT_SYMBOL(ceph_osdc_call);
5113 * reset all osd connections
5115 void ceph_osdc_reopen_osds(struct ceph_osd_client *osdc)
5119 down_write(&osdc->lock);
5120 for (n = rb_first(&osdc->osds); n; ) {
5121 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
5124 if (!reopen_osd(osd))
5125 kick_osd_requests(osd);
5127 up_write(&osdc->lock);
5133 int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
5138 osdc->client = client;
5139 init_rwsem(&osdc->lock);
5140 osdc->osds = RB_ROOT;
5141 INIT_LIST_HEAD(&osdc->osd_lru);
5142 spin_lock_init(&osdc->osd_lru_lock);
5143 osd_init(&osdc->homeless_osd);
5144 osdc->homeless_osd.o_osdc = osdc;
5145 osdc->homeless_osd.o_osd = CEPH_HOMELESS_OSD;
5146 osdc->last_linger_id = CEPH_LINGER_ID_START;
5147 osdc->linger_requests = RB_ROOT;
5148 osdc->map_checks = RB_ROOT;
5149 osdc->linger_map_checks = RB_ROOT;
5150 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
5151 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
5154 osdc->osdmap = ceph_osdmap_alloc();
5158 osdc->req_mempool = mempool_create_slab_pool(10,
5159 ceph_osd_request_cache);
5160 if (!osdc->req_mempool)
5163 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
5164 PAGE_SIZE, CEPH_OSD_SLAB_OPS, 10, "osd_op");
5167 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
5168 PAGE_SIZE, CEPH_OSD_SLAB_OPS, 10,
5174 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
5175 if (!osdc->notify_wq)
5176 goto out_msgpool_reply;
5178 osdc->completion_wq = create_singlethread_workqueue("ceph-completion");
5179 if (!osdc->completion_wq)
5182 schedule_delayed_work(&osdc->timeout_work,
5183 osdc->client->options->osd_keepalive_timeout);
5184 schedule_delayed_work(&osdc->osds_timeout_work,
5185 round_jiffies_relative(osdc->client->options->osd_idle_ttl));
5190 destroy_workqueue(osdc->notify_wq);
5192 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
5194 ceph_msgpool_destroy(&osdc->msgpool_op);
5196 mempool_destroy(osdc->req_mempool);
5198 ceph_osdmap_destroy(osdc->osdmap);
5203 void ceph_osdc_stop(struct ceph_osd_client *osdc)
5205 destroy_workqueue(osdc->completion_wq);
5206 destroy_workqueue(osdc->notify_wq);
5207 cancel_delayed_work_sync(&osdc->timeout_work);
5208 cancel_delayed_work_sync(&osdc->osds_timeout_work);
5210 down_write(&osdc->lock);
5211 while (!RB_EMPTY_ROOT(&osdc->osds)) {
5212 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
5213 struct ceph_osd, o_node);
5216 up_write(&osdc->lock);
5217 WARN_ON(refcount_read(&osdc->homeless_osd.o_ref) != 1);
5218 osd_cleanup(&osdc->homeless_osd);
5220 WARN_ON(!list_empty(&osdc->osd_lru));
5221 WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_requests));
5222 WARN_ON(!RB_EMPTY_ROOT(&osdc->map_checks));
5223 WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_map_checks));
5224 WARN_ON(atomic_read(&osdc->num_requests));
5225 WARN_ON(atomic_read(&osdc->num_homeless));
5227 ceph_osdmap_destroy(osdc->osdmap);
5228 mempool_destroy(osdc->req_mempool);
5229 ceph_msgpool_destroy(&osdc->msgpool_op);
5230 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
5234 * Read some contiguous pages. If we cross a stripe boundary, shorten
5235 * *plen. Return number of bytes read, or error.
5237 int ceph_osdc_readpages(struct ceph_osd_client *osdc,
5238 struct ceph_vino vino, struct ceph_file_layout *layout,
5240 u32 truncate_seq, u64 truncate_size,
5241 struct page **pages, int num_pages, int page_align)
5243 struct ceph_osd_request *req;
5246 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
5247 vino.snap, off, *plen);
5248 req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 0, 1,
5249 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
5250 NULL, truncate_seq, truncate_size,
5253 return PTR_ERR(req);
5255 /* it may be a short read due to an object boundary */
5256 osd_req_op_extent_osd_data_pages(req, 0,
5257 pages, *plen, page_align, false, false);
5259 dout("readpages final extent is %llu~%llu (%llu bytes align %d)\n",
5260 off, *plen, *plen, page_align);
5262 rc = ceph_osdc_start_request(osdc, req, false);
5264 rc = ceph_osdc_wait_request(osdc, req);
5266 ceph_osdc_put_request(req);
5267 dout("readpages result %d\n", rc);
5270 EXPORT_SYMBOL(ceph_osdc_readpages);
5273 * do a synchronous write on N pages
5275 int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
5276 struct ceph_file_layout *layout,
5277 struct ceph_snap_context *snapc,
5279 u32 truncate_seq, u64 truncate_size,
5280 struct timespec64 *mtime,
5281 struct page **pages, int num_pages)
5283 struct ceph_osd_request *req;
5285 int page_align = off & ~PAGE_MASK;
5287 req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 0, 1,
5288 CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE,
5289 snapc, truncate_seq, truncate_size,
5292 return PTR_ERR(req);
5294 /* it may be a short write due to an object boundary */
5295 osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align,
5297 dout("writepages %llu~%llu (%llu bytes)\n", off, len, len);
5299 req->r_mtime = *mtime;
5300 rc = ceph_osdc_start_request(osdc, req, true);
5302 rc = ceph_osdc_wait_request(osdc, req);
5304 ceph_osdc_put_request(req);
5307 dout("writepages result %d\n", rc);
5310 EXPORT_SYMBOL(ceph_osdc_writepages);
5312 static int osd_req_op_copy_from_init(struct ceph_osd_request *req,
5313 u64 src_snapid, u64 src_version,
5314 struct ceph_object_id *src_oid,
5315 struct ceph_object_locator *src_oloc,
5316 u32 src_fadvise_flags,
5317 u32 dst_fadvise_flags,
5318 u32 truncate_seq, u64 truncate_size,
5321 struct ceph_osd_req_op *op;
5322 struct page **pages;
5325 pages = ceph_alloc_page_vector(1, GFP_KERNEL);
5327 return PTR_ERR(pages);
5329 op = _osd_req_op_init(req, 0, CEPH_OSD_OP_COPY_FROM2,
5331 op->copy_from.snapid = src_snapid;
5332 op->copy_from.src_version = src_version;
5333 op->copy_from.flags = copy_from_flags;
5334 op->copy_from.src_fadvise_flags = src_fadvise_flags;
5336 p = page_address(pages[0]);
5337 end = p + PAGE_SIZE;
5338 ceph_encode_string(&p, end, src_oid->name, src_oid->name_len);
5339 encode_oloc(&p, end, src_oloc);
5340 ceph_encode_32(&p, truncate_seq);
5341 ceph_encode_64(&p, truncate_size);
5342 op->indata_len = PAGE_SIZE - (end - p);
5344 ceph_osd_data_pages_init(&op->copy_from.osd_data, pages,
5345 op->indata_len, 0, false, true);
5349 int ceph_osdc_copy_from(struct ceph_osd_client *osdc,
5350 u64 src_snapid, u64 src_version,
5351 struct ceph_object_id *src_oid,
5352 struct ceph_object_locator *src_oloc,
5353 u32 src_fadvise_flags,
5354 struct ceph_object_id *dst_oid,
5355 struct ceph_object_locator *dst_oloc,
5356 u32 dst_fadvise_flags,
5357 u32 truncate_seq, u64 truncate_size,
5360 struct ceph_osd_request *req;
5363 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_KERNEL);
5367 req->r_flags = CEPH_OSD_FLAG_WRITE;
5369 ceph_oloc_copy(&req->r_t.base_oloc, dst_oloc);
5370 ceph_oid_copy(&req->r_t.base_oid, dst_oid);
5372 ret = osd_req_op_copy_from_init(req, src_snapid, src_version, src_oid,
5373 src_oloc, src_fadvise_flags,
5374 dst_fadvise_flags, truncate_seq,
5375 truncate_size, copy_from_flags);
5379 ret = ceph_osdc_alloc_messages(req, GFP_KERNEL);
5383 ceph_osdc_start_request(osdc, req, false);
5384 ret = ceph_osdc_wait_request(osdc, req);
5387 ceph_osdc_put_request(req);
5390 EXPORT_SYMBOL(ceph_osdc_copy_from);
5392 int __init ceph_osdc_setup(void)
5394 size_t size = sizeof(struct ceph_osd_request) +
5395 CEPH_OSD_SLAB_OPS * sizeof(struct ceph_osd_req_op);
5397 BUG_ON(ceph_osd_request_cache);
5398 ceph_osd_request_cache = kmem_cache_create("ceph_osd_request", size,
5401 return ceph_osd_request_cache ? 0 : -ENOMEM;
5404 void ceph_osdc_cleanup(void)
5406 BUG_ON(!ceph_osd_request_cache);
5407 kmem_cache_destroy(ceph_osd_request_cache);
5408 ceph_osd_request_cache = NULL;
5412 * handle incoming message
5414 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
5416 struct ceph_osd *osd = con->private;
5417 struct ceph_osd_client *osdc = osd->o_osdc;
5418 int type = le16_to_cpu(msg->hdr.type);
5421 case CEPH_MSG_OSD_MAP:
5422 ceph_osdc_handle_map(osdc, msg);
5424 case CEPH_MSG_OSD_OPREPLY:
5425 handle_reply(osd, msg);
5427 case CEPH_MSG_OSD_BACKOFF:
5428 handle_backoff(osd, msg);
5430 case CEPH_MSG_WATCH_NOTIFY:
5431 handle_watch_notify(osdc, msg);
5435 pr_err("received unknown message type %d %s\n", type,
5436 ceph_msg_type_name(type));
5443 * Lookup and return message for incoming reply. Don't try to do
5444 * anything about a larger than preallocated data portion of the
5445 * message at the moment - for now, just skip the message.
5447 static struct ceph_msg *get_reply(struct ceph_connection *con,
5448 struct ceph_msg_header *hdr,
5451 struct ceph_osd *osd = con->private;
5452 struct ceph_osd_client *osdc = osd->o_osdc;
5453 struct ceph_msg *m = NULL;
5454 struct ceph_osd_request *req;
5455 int front_len = le32_to_cpu(hdr->front_len);
5456 int data_len = le32_to_cpu(hdr->data_len);
5457 u64 tid = le64_to_cpu(hdr->tid);
5459 down_read(&osdc->lock);
5460 if (!osd_registered(osd)) {
5461 dout("%s osd%d unknown, skipping\n", __func__, osd->o_osd);
5463 goto out_unlock_osdc;
5465 WARN_ON(osd->o_osd != le64_to_cpu(hdr->src.num));
5467 mutex_lock(&osd->lock);
5468 req = lookup_request(&osd->o_requests, tid);
5470 dout("%s osd%d tid %llu unknown, skipping\n", __func__,
5473 goto out_unlock_session;
5476 ceph_msg_revoke_incoming(req->r_reply);
5478 if (front_len > req->r_reply->front_alloc_len) {
5479 pr_warn("%s osd%d tid %llu front %d > preallocated %d\n",
5480 __func__, osd->o_osd, req->r_tid, front_len,
5481 req->r_reply->front_alloc_len);
5482 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS,
5485 goto out_unlock_session;
5486 ceph_msg_put(req->r_reply);
5490 if (data_len > req->r_reply->data_length) {
5491 pr_warn("%s osd%d tid %llu data %d > preallocated %zu, skipping\n",
5492 __func__, osd->o_osd, req->r_tid, data_len,
5493 req->r_reply->data_length);
5496 goto out_unlock_session;
5499 m = ceph_msg_get(req->r_reply);
5500 dout("get_reply tid %lld %p\n", tid, m);
5503 mutex_unlock(&osd->lock);
5505 up_read(&osdc->lock);
5510 * TODO: switch to a msg-owned pagelist
5512 static struct ceph_msg *alloc_msg_with_page_vector(struct ceph_msg_header *hdr)
5515 int type = le16_to_cpu(hdr->type);
5516 u32 front_len = le32_to_cpu(hdr->front_len);
5517 u32 data_len = le32_to_cpu(hdr->data_len);
5519 m = ceph_msg_new2(type, front_len, 1, GFP_NOIO, false);
5524 struct page **pages;
5525 struct ceph_osd_data osd_data;
5527 pages = ceph_alloc_page_vector(calc_pages_for(0, data_len),
5529 if (IS_ERR(pages)) {
5534 ceph_osd_data_pages_init(&osd_data, pages, data_len, 0, false,
5536 ceph_osdc_msg_data_add(m, &osd_data);
5542 static struct ceph_msg *alloc_msg(struct ceph_connection *con,
5543 struct ceph_msg_header *hdr,
5546 struct ceph_osd *osd = con->private;
5547 int type = le16_to_cpu(hdr->type);
5551 case CEPH_MSG_OSD_MAP:
5552 case CEPH_MSG_OSD_BACKOFF:
5553 case CEPH_MSG_WATCH_NOTIFY:
5554 return alloc_msg_with_page_vector(hdr);
5555 case CEPH_MSG_OSD_OPREPLY:
5556 return get_reply(con, hdr, skip);
5558 pr_warn("%s osd%d unknown msg type %d, skipping\n", __func__,
5566 * Wrappers to refcount containing ceph_osd struct
5568 static struct ceph_connection *get_osd_con(struct ceph_connection *con)
5570 struct ceph_osd *osd = con->private;
5576 static void put_osd_con(struct ceph_connection *con)
5578 struct ceph_osd *osd = con->private;
5586 * Note: returned pointer is the address of a structure that's
5587 * managed separately. Caller must *not* attempt to free it.
5589 static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
5590 int *proto, int force_new)
5592 struct ceph_osd *o = con->private;
5593 struct ceph_osd_client *osdc = o->o_osdc;
5594 struct ceph_auth_client *ac = osdc->client->monc.auth;
5595 struct ceph_auth_handshake *auth = &o->o_auth;
5597 if (force_new && auth->authorizer) {
5598 ceph_auth_destroy_authorizer(auth->authorizer);
5599 auth->authorizer = NULL;
5601 if (!auth->authorizer) {
5602 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
5605 return ERR_PTR(ret);
5607 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
5610 return ERR_PTR(ret);
5612 *proto = ac->protocol;
5617 static int add_authorizer_challenge(struct ceph_connection *con,
5618 void *challenge_buf, int challenge_buf_len)
5620 struct ceph_osd *o = con->private;
5621 struct ceph_osd_client *osdc = o->o_osdc;
5622 struct ceph_auth_client *ac = osdc->client->monc.auth;
5624 return ceph_auth_add_authorizer_challenge(ac, o->o_auth.authorizer,
5625 challenge_buf, challenge_buf_len);
5628 static int verify_authorizer_reply(struct ceph_connection *con)
5630 struct ceph_osd *o = con->private;
5631 struct ceph_osd_client *osdc = o->o_osdc;
5632 struct ceph_auth_client *ac = osdc->client->monc.auth;
5634 return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer);
5637 static int invalidate_authorizer(struct ceph_connection *con)
5639 struct ceph_osd *o = con->private;
5640 struct ceph_osd_client *osdc = o->o_osdc;
5641 struct ceph_auth_client *ac = osdc->client->monc.auth;
5643 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
5644 return ceph_monc_validate_auth(&osdc->client->monc);
5647 static void osd_reencode_message(struct ceph_msg *msg)
5649 int type = le16_to_cpu(msg->hdr.type);
5651 if (type == CEPH_MSG_OSD_OP)
5652 encode_request_finish(msg);
5655 static int osd_sign_message(struct ceph_msg *msg)
5657 struct ceph_osd *o = msg->con->private;
5658 struct ceph_auth_handshake *auth = &o->o_auth;
5660 return ceph_auth_sign_message(auth, msg);
5663 static int osd_check_message_signature(struct ceph_msg *msg)
5665 struct ceph_osd *o = msg->con->private;
5666 struct ceph_auth_handshake *auth = &o->o_auth;
5668 return ceph_auth_check_message_signature(auth, msg);
5671 static const struct ceph_connection_operations osd_con_ops = {
5674 .dispatch = dispatch,
5675 .get_authorizer = get_authorizer,
5676 .add_authorizer_challenge = add_authorizer_challenge,
5677 .verify_authorizer_reply = verify_authorizer_reply,
5678 .invalidate_authorizer = invalidate_authorizer,
5679 .alloc_msg = alloc_msg,
5680 .reencode_message = osd_reencode_message,
5681 .sign_message = osd_sign_message,
5682 .check_message_signature = osd_check_message_signature,