2 #include <linux/ceph/ceph_debug.h>
4 #include <linux/module.h>
6 #include <linux/highmem.h>
8 #include <linux/pagemap.h>
9 #include <linux/slab.h>
10 #include <linux/uaccess.h>
12 #include <linux/bio.h>
15 #include <linux/ceph/libceph.h>
16 #include <linux/ceph/osd_client.h>
17 #include <linux/ceph/messenger.h>
18 #include <linux/ceph/decode.h>
19 #include <linux/ceph/auth.h>
20 #include <linux/ceph/pagelist.h>
22 #define OSD_OPREPLY_FRONT_LEN 512
24 static struct kmem_cache *ceph_osd_request_cache;
26 static const struct ceph_connection_operations osd_con_ops;
29 * Implement client access to distributed object storage cluster.
31 * All data objects are stored within a cluster/cloud of OSDs, or
32 * "object storage devices." (Note that Ceph OSDs have _nothing_ to
33 * do with the T10 OSD extensions to SCSI.) Ceph OSDs are simply
34 * remote daemons serving up and coordinating consistent and safe
37 * Cluster membership and the mapping of data objects onto storage devices
38 * are described by the osd map.
40 * We keep track of pending OSD requests (read, write), resubmit
41 * requests to different OSDs when the cluster topology/data layout
42 * change, or retry the affected requests when the communications
43 * channel with an OSD is reset.
46 static void link_request(struct ceph_osd *osd, struct ceph_osd_request *req);
47 static void unlink_request(struct ceph_osd *osd, struct ceph_osd_request *req);
48 static void link_linger(struct ceph_osd *osd,
49 struct ceph_osd_linger_request *lreq);
50 static void unlink_linger(struct ceph_osd *osd,
51 struct ceph_osd_linger_request *lreq);
54 static inline bool rwsem_is_wrlocked(struct rw_semaphore *sem)
58 if (unlikely(down_read_trylock(sem))) {
65 static inline void verify_osdc_locked(struct ceph_osd_client *osdc)
67 WARN_ON(!rwsem_is_locked(&osdc->lock));
69 static inline void verify_osdc_wrlocked(struct ceph_osd_client *osdc)
71 WARN_ON(!rwsem_is_wrlocked(&osdc->lock));
73 static inline void verify_osd_locked(struct ceph_osd *osd)
75 struct ceph_osd_client *osdc = osd->o_osdc;
77 WARN_ON(!(mutex_is_locked(&osd->lock) &&
78 rwsem_is_locked(&osdc->lock)) &&
79 !rwsem_is_wrlocked(&osdc->lock));
81 static inline void verify_lreq_locked(struct ceph_osd_linger_request *lreq)
83 WARN_ON(!mutex_is_locked(&lreq->lock));
86 static inline void verify_osdc_locked(struct ceph_osd_client *osdc) { }
87 static inline void verify_osdc_wrlocked(struct ceph_osd_client *osdc) { }
88 static inline void verify_osd_locked(struct ceph_osd *osd) { }
89 static inline void verify_lreq_locked(struct ceph_osd_linger_request *lreq) { }
93 * calculate the mapping of a file extent onto an object, and fill out the
94 * request accordingly. shorten extent as necessary if it crosses an
97 * fill osd op in request message.
99 static int calc_layout(struct ceph_file_layout *layout, u64 off, u64 *plen,
100 u64 *objnum, u64 *objoff, u64 *objlen)
102 u64 orig_len = *plen;
106 r = ceph_calc_file_object_mapping(layout, off, orig_len, objnum,
110 if (*objlen < orig_len) {
112 dout(" skipping last %llu, final file extent %llu~%llu\n",
113 orig_len - *plen, off, *plen);
116 dout("calc_layout objnum=%llx %llu~%llu\n", *objnum, *objoff, *objlen);
121 static void ceph_osd_data_init(struct ceph_osd_data *osd_data)
123 memset(osd_data, 0, sizeof (*osd_data));
124 osd_data->type = CEPH_OSD_DATA_TYPE_NONE;
127 static void ceph_osd_data_pages_init(struct ceph_osd_data *osd_data,
128 struct page **pages, u64 length, u32 alignment,
129 bool pages_from_pool, bool own_pages)
131 osd_data->type = CEPH_OSD_DATA_TYPE_PAGES;
132 osd_data->pages = pages;
133 osd_data->length = length;
134 osd_data->alignment = alignment;
135 osd_data->pages_from_pool = pages_from_pool;
136 osd_data->own_pages = own_pages;
139 static void ceph_osd_data_pagelist_init(struct ceph_osd_data *osd_data,
140 struct ceph_pagelist *pagelist)
142 osd_data->type = CEPH_OSD_DATA_TYPE_PAGELIST;
143 osd_data->pagelist = pagelist;
147 static void ceph_osd_data_bio_init(struct ceph_osd_data *osd_data,
148 struct bio *bio, size_t bio_length)
150 osd_data->type = CEPH_OSD_DATA_TYPE_BIO;
152 osd_data->bio_length = bio_length;
154 #endif /* CONFIG_BLOCK */
156 #define osd_req_op_data(oreq, whch, typ, fld) \
158 struct ceph_osd_request *__oreq = (oreq); \
159 unsigned int __whch = (whch); \
160 BUG_ON(__whch >= __oreq->r_num_ops); \
161 &__oreq->r_ops[__whch].typ.fld; \
164 static struct ceph_osd_data *
165 osd_req_op_raw_data_in(struct ceph_osd_request *osd_req, unsigned int which)
167 BUG_ON(which >= osd_req->r_num_ops);
169 return &osd_req->r_ops[which].raw_data_in;
172 struct ceph_osd_data *
173 osd_req_op_extent_osd_data(struct ceph_osd_request *osd_req,
176 return osd_req_op_data(osd_req, which, extent, osd_data);
178 EXPORT_SYMBOL(osd_req_op_extent_osd_data);
180 void osd_req_op_raw_data_in_pages(struct ceph_osd_request *osd_req,
181 unsigned int which, struct page **pages,
182 u64 length, u32 alignment,
183 bool pages_from_pool, bool own_pages)
185 struct ceph_osd_data *osd_data;
187 osd_data = osd_req_op_raw_data_in(osd_req, which);
188 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
189 pages_from_pool, own_pages);
191 EXPORT_SYMBOL(osd_req_op_raw_data_in_pages);
193 void osd_req_op_extent_osd_data_pages(struct ceph_osd_request *osd_req,
194 unsigned int which, struct page **pages,
195 u64 length, u32 alignment,
196 bool pages_from_pool, bool own_pages)
198 struct ceph_osd_data *osd_data;
200 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
201 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
202 pages_from_pool, own_pages);
204 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pages);
206 void osd_req_op_extent_osd_data_pagelist(struct ceph_osd_request *osd_req,
207 unsigned int which, struct ceph_pagelist *pagelist)
209 struct ceph_osd_data *osd_data;
211 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
212 ceph_osd_data_pagelist_init(osd_data, pagelist);
214 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pagelist);
217 void osd_req_op_extent_osd_data_bio(struct ceph_osd_request *osd_req,
218 unsigned int which, struct bio *bio, size_t bio_length)
220 struct ceph_osd_data *osd_data;
222 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
223 ceph_osd_data_bio_init(osd_data, bio, bio_length);
225 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bio);
226 #endif /* CONFIG_BLOCK */
228 static void osd_req_op_cls_request_info_pagelist(
229 struct ceph_osd_request *osd_req,
230 unsigned int which, struct ceph_pagelist *pagelist)
232 struct ceph_osd_data *osd_data;
234 osd_data = osd_req_op_data(osd_req, which, cls, request_info);
235 ceph_osd_data_pagelist_init(osd_data, pagelist);
238 void osd_req_op_cls_request_data_pagelist(
239 struct ceph_osd_request *osd_req,
240 unsigned int which, struct ceph_pagelist *pagelist)
242 struct ceph_osd_data *osd_data;
244 osd_data = osd_req_op_data(osd_req, which, cls, request_data);
245 ceph_osd_data_pagelist_init(osd_data, pagelist);
246 osd_req->r_ops[which].cls.indata_len += pagelist->length;
247 osd_req->r_ops[which].indata_len += pagelist->length;
249 EXPORT_SYMBOL(osd_req_op_cls_request_data_pagelist);
251 void osd_req_op_cls_request_data_pages(struct ceph_osd_request *osd_req,
252 unsigned int which, struct page **pages, u64 length,
253 u32 alignment, bool pages_from_pool, bool own_pages)
255 struct ceph_osd_data *osd_data;
257 osd_data = osd_req_op_data(osd_req, which, cls, request_data);
258 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
259 pages_from_pool, own_pages);
260 osd_req->r_ops[which].cls.indata_len += length;
261 osd_req->r_ops[which].indata_len += length;
263 EXPORT_SYMBOL(osd_req_op_cls_request_data_pages);
265 void osd_req_op_cls_response_data_pages(struct ceph_osd_request *osd_req,
266 unsigned int which, struct page **pages, u64 length,
267 u32 alignment, bool pages_from_pool, bool own_pages)
269 struct ceph_osd_data *osd_data;
271 osd_data = osd_req_op_data(osd_req, which, cls, response_data);
272 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
273 pages_from_pool, own_pages);
275 EXPORT_SYMBOL(osd_req_op_cls_response_data_pages);
277 static u64 ceph_osd_data_length(struct ceph_osd_data *osd_data)
279 switch (osd_data->type) {
280 case CEPH_OSD_DATA_TYPE_NONE:
282 case CEPH_OSD_DATA_TYPE_PAGES:
283 return osd_data->length;
284 case CEPH_OSD_DATA_TYPE_PAGELIST:
285 return (u64)osd_data->pagelist->length;
287 case CEPH_OSD_DATA_TYPE_BIO:
288 return (u64)osd_data->bio_length;
289 #endif /* CONFIG_BLOCK */
291 WARN(true, "unrecognized data type %d\n", (int)osd_data->type);
296 static void ceph_osd_data_release(struct ceph_osd_data *osd_data)
298 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES && osd_data->own_pages) {
301 num_pages = calc_pages_for((u64)osd_data->alignment,
302 (u64)osd_data->length);
303 ceph_release_page_vector(osd_data->pages, num_pages);
305 ceph_osd_data_init(osd_data);
308 static void osd_req_op_data_release(struct ceph_osd_request *osd_req,
311 struct ceph_osd_req_op *op;
313 BUG_ON(which >= osd_req->r_num_ops);
314 op = &osd_req->r_ops[which];
317 case CEPH_OSD_OP_READ:
318 case CEPH_OSD_OP_WRITE:
319 case CEPH_OSD_OP_WRITEFULL:
320 ceph_osd_data_release(&op->extent.osd_data);
322 case CEPH_OSD_OP_CALL:
323 ceph_osd_data_release(&op->cls.request_info);
324 ceph_osd_data_release(&op->cls.request_data);
325 ceph_osd_data_release(&op->cls.response_data);
327 case CEPH_OSD_OP_SETXATTR:
328 case CEPH_OSD_OP_CMPXATTR:
329 ceph_osd_data_release(&op->xattr.osd_data);
331 case CEPH_OSD_OP_STAT:
332 ceph_osd_data_release(&op->raw_data_in);
334 case CEPH_OSD_OP_NOTIFY_ACK:
335 ceph_osd_data_release(&op->notify_ack.request_data);
337 case CEPH_OSD_OP_NOTIFY:
338 ceph_osd_data_release(&op->notify.request_data);
339 ceph_osd_data_release(&op->notify.response_data);
341 case CEPH_OSD_OP_LIST_WATCHERS:
342 ceph_osd_data_release(&op->list_watchers.response_data);
350 * Assumes @t is zero-initialized.
352 static void target_init(struct ceph_osd_request_target *t)
354 ceph_oid_init(&t->base_oid);
355 ceph_oloc_init(&t->base_oloc);
356 ceph_oid_init(&t->target_oid);
357 ceph_oloc_init(&t->target_oloc);
359 ceph_osds_init(&t->acting);
360 ceph_osds_init(&t->up);
364 t->osd = CEPH_HOMELESS_OSD;
367 static void target_copy(struct ceph_osd_request_target *dest,
368 const struct ceph_osd_request_target *src)
370 ceph_oid_copy(&dest->base_oid, &src->base_oid);
371 ceph_oloc_copy(&dest->base_oloc, &src->base_oloc);
372 ceph_oid_copy(&dest->target_oid, &src->target_oid);
373 ceph_oloc_copy(&dest->target_oloc, &src->target_oloc);
375 dest->pgid = src->pgid; /* struct */
376 dest->pg_num = src->pg_num;
377 dest->pg_num_mask = src->pg_num_mask;
378 ceph_osds_copy(&dest->acting, &src->acting);
379 ceph_osds_copy(&dest->up, &src->up);
380 dest->size = src->size;
381 dest->min_size = src->min_size;
382 dest->sort_bitwise = src->sort_bitwise;
384 dest->flags = src->flags;
385 dest->paused = src->paused;
387 dest->osd = src->osd;
390 static void target_destroy(struct ceph_osd_request_target *t)
392 ceph_oid_destroy(&t->base_oid);
393 ceph_oloc_destroy(&t->base_oloc);
394 ceph_oid_destroy(&t->target_oid);
395 ceph_oloc_destroy(&t->target_oloc);
401 static void request_release_checks(struct ceph_osd_request *req)
403 WARN_ON(!RB_EMPTY_NODE(&req->r_node));
404 WARN_ON(!RB_EMPTY_NODE(&req->r_mc_node));
405 WARN_ON(!list_empty(&req->r_unsafe_item));
409 static void ceph_osdc_release_request(struct kref *kref)
411 struct ceph_osd_request *req = container_of(kref,
412 struct ceph_osd_request, r_kref);
415 dout("%s %p (r_request %p r_reply %p)\n", __func__, req,
416 req->r_request, req->r_reply);
417 request_release_checks(req);
420 ceph_msg_put(req->r_request);
422 ceph_msg_put(req->r_reply);
424 for (which = 0; which < req->r_num_ops; which++)
425 osd_req_op_data_release(req, which);
427 target_destroy(&req->r_t);
428 ceph_put_snap_context(req->r_snapc);
431 mempool_free(req, req->r_osdc->req_mempool);
432 else if (req->r_num_ops <= CEPH_OSD_SLAB_OPS)
433 kmem_cache_free(ceph_osd_request_cache, req);
438 void ceph_osdc_get_request(struct ceph_osd_request *req)
440 dout("%s %p (was %d)\n", __func__, req,
441 atomic_read(&req->r_kref.refcount));
442 kref_get(&req->r_kref);
444 EXPORT_SYMBOL(ceph_osdc_get_request);
446 void ceph_osdc_put_request(struct ceph_osd_request *req)
449 dout("%s %p (was %d)\n", __func__, req,
450 atomic_read(&req->r_kref.refcount));
451 kref_put(&req->r_kref, ceph_osdc_release_request);
454 EXPORT_SYMBOL(ceph_osdc_put_request);
456 static void request_init(struct ceph_osd_request *req)
458 /* req only, each op is zeroed in _osd_req_op_init() */
459 memset(req, 0, sizeof(*req));
461 kref_init(&req->r_kref);
462 init_completion(&req->r_completion);
463 init_completion(&req->r_done_completion);
464 RB_CLEAR_NODE(&req->r_node);
465 RB_CLEAR_NODE(&req->r_mc_node);
466 INIT_LIST_HEAD(&req->r_unsafe_item);
468 target_init(&req->r_t);
472 * This is ugly, but it allows us to reuse linger registration and ping
473 * requests, keeping the structure of the code around send_linger{_ping}()
474 * reasonable. Setting up a min_nr=2 mempool for each linger request
475 * and dealing with copying ops (this blasts req only, watch op remains
476 * intact) isn't any better.
478 static void request_reinit(struct ceph_osd_request *req)
480 struct ceph_osd_client *osdc = req->r_osdc;
481 bool mempool = req->r_mempool;
482 unsigned int num_ops = req->r_num_ops;
483 u64 snapid = req->r_snapid;
484 struct ceph_snap_context *snapc = req->r_snapc;
485 bool linger = req->r_linger;
486 struct ceph_msg *request_msg = req->r_request;
487 struct ceph_msg *reply_msg = req->r_reply;
489 dout("%s req %p\n", __func__, req);
490 WARN_ON(atomic_read(&req->r_kref.refcount) != 1);
491 request_release_checks(req);
493 WARN_ON(atomic_read(&request_msg->kref.refcount) != 1);
494 WARN_ON(atomic_read(&reply_msg->kref.refcount) != 1);
495 target_destroy(&req->r_t);
499 req->r_mempool = mempool;
500 req->r_num_ops = num_ops;
501 req->r_snapid = snapid;
502 req->r_snapc = snapc;
503 req->r_linger = linger;
504 req->r_request = request_msg;
505 req->r_reply = reply_msg;
508 struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
509 struct ceph_snap_context *snapc,
510 unsigned int num_ops,
514 struct ceph_osd_request *req;
517 BUG_ON(num_ops > CEPH_OSD_SLAB_OPS);
518 req = mempool_alloc(osdc->req_mempool, gfp_flags);
519 } else if (num_ops <= CEPH_OSD_SLAB_OPS) {
520 req = kmem_cache_alloc(ceph_osd_request_cache, gfp_flags);
522 BUG_ON(num_ops > CEPH_OSD_MAX_OPS);
523 req = kmalloc(sizeof(*req) + num_ops * sizeof(req->r_ops[0]),
531 req->r_mempool = use_mempool;
532 req->r_num_ops = num_ops;
533 req->r_snapid = CEPH_NOSNAP;
534 req->r_snapc = ceph_get_snap_context(snapc);
536 dout("%s req %p\n", __func__, req);
539 EXPORT_SYMBOL(ceph_osdc_alloc_request);
541 static int ceph_oloc_encoding_size(struct ceph_object_locator *oloc)
543 return 8 + 4 + 4 + 4 + (oloc->pool_ns ? oloc->pool_ns->len : 0);
546 int ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp)
548 struct ceph_osd_client *osdc = req->r_osdc;
549 struct ceph_msg *msg;
552 WARN_ON(ceph_oid_empty(&req->r_base_oid));
553 WARN_ON(ceph_oloc_empty(&req->r_base_oloc));
555 /* create request message */
556 msg_size = 4 + 4 + 4; /* client_inc, osdmap_epoch, flags */
557 msg_size += 4 + 4 + 4 + 8; /* mtime, reassert_version */
558 msg_size += CEPH_ENCODING_START_BLK_LEN +
559 ceph_oloc_encoding_size(&req->r_base_oloc); /* oloc */
560 msg_size += 1 + 8 + 4 + 4; /* pgid */
561 msg_size += 4 + req->r_base_oid.name_len; /* oid */
562 msg_size += 2 + req->r_num_ops * sizeof(struct ceph_osd_op);
563 msg_size += 8; /* snapid */
564 msg_size += 8; /* snap_seq */
565 msg_size += 4 + 8 * (req->r_snapc ? req->r_snapc->num_snaps : 0);
566 msg_size += 4; /* retry_attempt */
569 msg = ceph_msgpool_get(&osdc->msgpool_op, 0);
571 msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp, true);
575 memset(msg->front.iov_base, 0, msg->front.iov_len);
576 req->r_request = msg;
578 /* create reply message */
579 msg_size = OSD_OPREPLY_FRONT_LEN;
580 msg_size += req->r_base_oid.name_len;
581 msg_size += req->r_num_ops * sizeof(struct ceph_osd_op);
584 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0);
586 msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, msg_size, gfp, true);
594 EXPORT_SYMBOL(ceph_osdc_alloc_messages);
596 static bool osd_req_opcode_valid(u16 opcode)
599 #define GENERATE_CASE(op, opcode, str) case CEPH_OSD_OP_##op: return true;
600 __CEPH_FORALL_OSD_OPS(GENERATE_CASE)
608 * This is an osd op init function for opcodes that have no data or
609 * other information associated with them. It also serves as a
610 * common init routine for all the other init functions, below.
612 static struct ceph_osd_req_op *
613 _osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which,
614 u16 opcode, u32 flags)
616 struct ceph_osd_req_op *op;
618 BUG_ON(which >= osd_req->r_num_ops);
619 BUG_ON(!osd_req_opcode_valid(opcode));
621 op = &osd_req->r_ops[which];
622 memset(op, 0, sizeof (*op));
629 void osd_req_op_init(struct ceph_osd_request *osd_req,
630 unsigned int which, u16 opcode, u32 flags)
632 (void)_osd_req_op_init(osd_req, which, opcode, flags);
634 EXPORT_SYMBOL(osd_req_op_init);
636 void osd_req_op_extent_init(struct ceph_osd_request *osd_req,
637 unsigned int which, u16 opcode,
638 u64 offset, u64 length,
639 u64 truncate_size, u32 truncate_seq)
641 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
643 size_t payload_len = 0;
645 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
646 opcode != CEPH_OSD_OP_WRITEFULL && opcode != CEPH_OSD_OP_ZERO &&
647 opcode != CEPH_OSD_OP_TRUNCATE);
649 op->extent.offset = offset;
650 op->extent.length = length;
651 op->extent.truncate_size = truncate_size;
652 op->extent.truncate_seq = truncate_seq;
653 if (opcode == CEPH_OSD_OP_WRITE || opcode == CEPH_OSD_OP_WRITEFULL)
654 payload_len += length;
656 op->indata_len = payload_len;
658 EXPORT_SYMBOL(osd_req_op_extent_init);
660 void osd_req_op_extent_update(struct ceph_osd_request *osd_req,
661 unsigned int which, u64 length)
663 struct ceph_osd_req_op *op;
666 BUG_ON(which >= osd_req->r_num_ops);
667 op = &osd_req->r_ops[which];
668 previous = op->extent.length;
670 if (length == previous)
671 return; /* Nothing to do */
672 BUG_ON(length > previous);
674 op->extent.length = length;
675 op->indata_len -= previous - length;
677 EXPORT_SYMBOL(osd_req_op_extent_update);
679 void osd_req_op_extent_dup_last(struct ceph_osd_request *osd_req,
680 unsigned int which, u64 offset_inc)
682 struct ceph_osd_req_op *op, *prev_op;
684 BUG_ON(which + 1 >= osd_req->r_num_ops);
686 prev_op = &osd_req->r_ops[which];
687 op = _osd_req_op_init(osd_req, which + 1, prev_op->op, prev_op->flags);
688 /* dup previous one */
689 op->indata_len = prev_op->indata_len;
690 op->outdata_len = prev_op->outdata_len;
691 op->extent = prev_op->extent;
693 op->extent.offset += offset_inc;
694 op->extent.length -= offset_inc;
696 if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL)
697 op->indata_len -= offset_inc;
699 EXPORT_SYMBOL(osd_req_op_extent_dup_last);
701 void osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
702 u16 opcode, const char *class, const char *method)
704 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
706 struct ceph_pagelist *pagelist;
707 size_t payload_len = 0;
710 BUG_ON(opcode != CEPH_OSD_OP_CALL);
712 pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
714 ceph_pagelist_init(pagelist);
716 op->cls.class_name = class;
717 size = strlen(class);
718 BUG_ON(size > (size_t) U8_MAX);
719 op->cls.class_len = size;
720 ceph_pagelist_append(pagelist, class, size);
723 op->cls.method_name = method;
724 size = strlen(method);
725 BUG_ON(size > (size_t) U8_MAX);
726 op->cls.method_len = size;
727 ceph_pagelist_append(pagelist, method, size);
730 osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist);
732 op->indata_len = payload_len;
734 EXPORT_SYMBOL(osd_req_op_cls_init);
736 int osd_req_op_xattr_init(struct ceph_osd_request *osd_req, unsigned int which,
737 u16 opcode, const char *name, const void *value,
738 size_t size, u8 cmp_op, u8 cmp_mode)
740 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
742 struct ceph_pagelist *pagelist;
745 BUG_ON(opcode != CEPH_OSD_OP_SETXATTR && opcode != CEPH_OSD_OP_CMPXATTR);
747 pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
751 ceph_pagelist_init(pagelist);
753 payload_len = strlen(name);
754 op->xattr.name_len = payload_len;
755 ceph_pagelist_append(pagelist, name, payload_len);
757 op->xattr.value_len = size;
758 ceph_pagelist_append(pagelist, value, size);
761 op->xattr.cmp_op = cmp_op;
762 op->xattr.cmp_mode = cmp_mode;
764 ceph_osd_data_pagelist_init(&op->xattr.osd_data, pagelist);
765 op->indata_len = payload_len;
768 EXPORT_SYMBOL(osd_req_op_xattr_init);
771 * @watch_opcode: CEPH_OSD_WATCH_OP_*
773 static void osd_req_op_watch_init(struct ceph_osd_request *req, int which,
774 u64 cookie, u8 watch_opcode)
776 struct ceph_osd_req_op *op;
778 op = _osd_req_op_init(req, which, CEPH_OSD_OP_WATCH, 0);
779 op->watch.cookie = cookie;
780 op->watch.op = watch_opcode;
784 void osd_req_op_alloc_hint_init(struct ceph_osd_request *osd_req,
786 u64 expected_object_size,
787 u64 expected_write_size)
789 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
790 CEPH_OSD_OP_SETALLOCHINT,
793 op->alloc_hint.expected_object_size = expected_object_size;
794 op->alloc_hint.expected_write_size = expected_write_size;
797 * CEPH_OSD_OP_SETALLOCHINT op is advisory and therefore deemed
798 * not worth a feature bit. Set FAILOK per-op flag to make
799 * sure older osds don't trip over an unsupported opcode.
801 op->flags |= CEPH_OSD_OP_FLAG_FAILOK;
803 EXPORT_SYMBOL(osd_req_op_alloc_hint_init);
805 static void ceph_osdc_msg_data_add(struct ceph_msg *msg,
806 struct ceph_osd_data *osd_data)
808 u64 length = ceph_osd_data_length(osd_data);
810 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
811 BUG_ON(length > (u64) SIZE_MAX);
813 ceph_msg_data_add_pages(msg, osd_data->pages,
814 length, osd_data->alignment);
815 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
817 ceph_msg_data_add_pagelist(msg, osd_data->pagelist);
819 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
820 ceph_msg_data_add_bio(msg, osd_data->bio, length);
823 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
827 static u32 osd_req_encode_op(struct ceph_osd_op *dst,
828 const struct ceph_osd_req_op *src)
830 if (WARN_ON(!osd_req_opcode_valid(src->op))) {
831 pr_err("unrecognized osd opcode %d\n", src->op);
837 case CEPH_OSD_OP_STAT:
839 case CEPH_OSD_OP_READ:
840 case CEPH_OSD_OP_WRITE:
841 case CEPH_OSD_OP_WRITEFULL:
842 case CEPH_OSD_OP_ZERO:
843 case CEPH_OSD_OP_TRUNCATE:
844 dst->extent.offset = cpu_to_le64(src->extent.offset);
845 dst->extent.length = cpu_to_le64(src->extent.length);
846 dst->extent.truncate_size =
847 cpu_to_le64(src->extent.truncate_size);
848 dst->extent.truncate_seq =
849 cpu_to_le32(src->extent.truncate_seq);
851 case CEPH_OSD_OP_CALL:
852 dst->cls.class_len = src->cls.class_len;
853 dst->cls.method_len = src->cls.method_len;
854 dst->cls.indata_len = cpu_to_le32(src->cls.indata_len);
856 case CEPH_OSD_OP_STARTSYNC:
858 case CEPH_OSD_OP_WATCH:
859 dst->watch.cookie = cpu_to_le64(src->watch.cookie);
860 dst->watch.ver = cpu_to_le64(0);
861 dst->watch.op = src->watch.op;
862 dst->watch.gen = cpu_to_le32(src->watch.gen);
864 case CEPH_OSD_OP_NOTIFY_ACK:
866 case CEPH_OSD_OP_NOTIFY:
867 dst->notify.cookie = cpu_to_le64(src->notify.cookie);
869 case CEPH_OSD_OP_LIST_WATCHERS:
871 case CEPH_OSD_OP_SETALLOCHINT:
872 dst->alloc_hint.expected_object_size =
873 cpu_to_le64(src->alloc_hint.expected_object_size);
874 dst->alloc_hint.expected_write_size =
875 cpu_to_le64(src->alloc_hint.expected_write_size);
877 case CEPH_OSD_OP_SETXATTR:
878 case CEPH_OSD_OP_CMPXATTR:
879 dst->xattr.name_len = cpu_to_le32(src->xattr.name_len);
880 dst->xattr.value_len = cpu_to_le32(src->xattr.value_len);
881 dst->xattr.cmp_op = src->xattr.cmp_op;
882 dst->xattr.cmp_mode = src->xattr.cmp_mode;
884 case CEPH_OSD_OP_CREATE:
885 case CEPH_OSD_OP_DELETE:
888 pr_err("unsupported osd opcode %s\n",
889 ceph_osd_op_name(src->op));
895 dst->op = cpu_to_le16(src->op);
896 dst->flags = cpu_to_le32(src->flags);
897 dst->payload_len = cpu_to_le32(src->indata_len);
899 return src->indata_len;
903 * build new request AND message, calculate layout, and adjust file
906 * if the file was recently truncated, we include information about its
907 * old and new size so that the object can be updated appropriately. (we
908 * avoid synchronously deleting truncated objects because it's slow.)
910 * if @do_sync, include a 'startsync' command so that the osd will flush
913 struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
914 struct ceph_file_layout *layout,
915 struct ceph_vino vino,
917 unsigned int which, int num_ops,
918 int opcode, int flags,
919 struct ceph_snap_context *snapc,
924 struct ceph_osd_request *req;
930 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
931 opcode != CEPH_OSD_OP_ZERO && opcode != CEPH_OSD_OP_TRUNCATE &&
932 opcode != CEPH_OSD_OP_CREATE && opcode != CEPH_OSD_OP_DELETE);
934 req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool,
941 /* calculate max write size */
942 r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
946 if (opcode == CEPH_OSD_OP_CREATE || opcode == CEPH_OSD_OP_DELETE) {
947 osd_req_op_init(req, which, opcode, 0);
949 u32 object_size = layout->object_size;
950 u32 object_base = off - objoff;
951 if (!(truncate_seq == 1 && truncate_size == -1ULL)) {
952 if (truncate_size <= object_base) {
955 truncate_size -= object_base;
956 if (truncate_size > object_size)
957 truncate_size = object_size;
960 osd_req_op_extent_init(req, which, opcode, objoff, objlen,
961 truncate_size, truncate_seq);
964 req->r_flags = flags;
965 req->r_base_oloc.pool = layout->pool_id;
966 req->r_base_oloc.pool_ns = ceph_try_get_string(layout->pool_ns);
967 ceph_oid_printf(&req->r_base_oid, "%llx.%08llx", vino.ino, objnum);
969 req->r_snapid = vino.snap;
970 if (flags & CEPH_OSD_FLAG_WRITE)
971 req->r_data_offset = off;
973 r = ceph_osdc_alloc_messages(req, GFP_NOFS);
980 ceph_osdc_put_request(req);
983 EXPORT_SYMBOL(ceph_osdc_new_request);
986 * We keep osd requests in an rbtree, sorted by ->r_tid.
988 DEFINE_RB_FUNCS(request, struct ceph_osd_request, r_tid, r_node)
989 DEFINE_RB_FUNCS(request_mc, struct ceph_osd_request, r_tid, r_mc_node)
991 static bool osd_homeless(struct ceph_osd *osd)
993 return osd->o_osd == CEPH_HOMELESS_OSD;
996 static bool osd_registered(struct ceph_osd *osd)
998 verify_osdc_locked(osd->o_osdc);
1000 return !RB_EMPTY_NODE(&osd->o_node);
1004 * Assumes @osd is zero-initialized.
1006 static void osd_init(struct ceph_osd *osd)
1008 atomic_set(&osd->o_ref, 1);
1009 RB_CLEAR_NODE(&osd->o_node);
1010 osd->o_requests = RB_ROOT;
1011 osd->o_linger_requests = RB_ROOT;
1012 INIT_LIST_HEAD(&osd->o_osd_lru);
1013 INIT_LIST_HEAD(&osd->o_keepalive_item);
1014 osd->o_incarnation = 1;
1015 mutex_init(&osd->lock);
1018 static void osd_cleanup(struct ceph_osd *osd)
1020 WARN_ON(!RB_EMPTY_NODE(&osd->o_node));
1021 WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests));
1022 WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests));
1023 WARN_ON(!list_empty(&osd->o_osd_lru));
1024 WARN_ON(!list_empty(&osd->o_keepalive_item));
1026 if (osd->o_auth.authorizer) {
1027 WARN_ON(osd_homeless(osd));
1028 ceph_auth_destroy_authorizer(osd->o_auth.authorizer);
1033 * Track open sessions with osds.
1035 static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
1037 struct ceph_osd *osd;
1039 WARN_ON(onum == CEPH_HOMELESS_OSD);
1041 osd = kzalloc(sizeof(*osd), GFP_NOIO | __GFP_NOFAIL);
1046 ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
1051 static struct ceph_osd *get_osd(struct ceph_osd *osd)
1053 if (atomic_inc_not_zero(&osd->o_ref)) {
1054 dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
1055 atomic_read(&osd->o_ref));
1058 dout("get_osd %p FAIL\n", osd);
1063 static void put_osd(struct ceph_osd *osd)
1065 dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
1066 atomic_read(&osd->o_ref) - 1);
1067 if (atomic_dec_and_test(&osd->o_ref)) {
1073 DEFINE_RB_FUNCS(osd, struct ceph_osd, o_osd, o_node)
1075 static void __move_osd_to_lru(struct ceph_osd *osd)
1077 struct ceph_osd_client *osdc = osd->o_osdc;
1079 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1080 BUG_ON(!list_empty(&osd->o_osd_lru));
1082 spin_lock(&osdc->osd_lru_lock);
1083 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
1084 spin_unlock(&osdc->osd_lru_lock);
1086 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl;
1089 static void maybe_move_osd_to_lru(struct ceph_osd *osd)
1091 if (RB_EMPTY_ROOT(&osd->o_requests) &&
1092 RB_EMPTY_ROOT(&osd->o_linger_requests))
1093 __move_osd_to_lru(osd);
1096 static void __remove_osd_from_lru(struct ceph_osd *osd)
1098 struct ceph_osd_client *osdc = osd->o_osdc;
1100 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1102 spin_lock(&osdc->osd_lru_lock);
1103 if (!list_empty(&osd->o_osd_lru))
1104 list_del_init(&osd->o_osd_lru);
1105 spin_unlock(&osdc->osd_lru_lock);
1109 * Close the connection and assign any leftover requests to the
1112 static void close_osd(struct ceph_osd *osd)
1114 struct ceph_osd_client *osdc = osd->o_osdc;
1117 verify_osdc_wrlocked(osdc);
1118 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1120 ceph_con_close(&osd->o_con);
1122 for (n = rb_first(&osd->o_requests); n; ) {
1123 struct ceph_osd_request *req =
1124 rb_entry(n, struct ceph_osd_request, r_node);
1126 n = rb_next(n); /* unlink_request() */
1128 dout(" reassigning req %p tid %llu\n", req, req->r_tid);
1129 unlink_request(osd, req);
1130 link_request(&osdc->homeless_osd, req);
1132 for (n = rb_first(&osd->o_linger_requests); n; ) {
1133 struct ceph_osd_linger_request *lreq =
1134 rb_entry(n, struct ceph_osd_linger_request, node);
1136 n = rb_next(n); /* unlink_linger() */
1138 dout(" reassigning lreq %p linger_id %llu\n", lreq,
1140 unlink_linger(osd, lreq);
1141 link_linger(&osdc->homeless_osd, lreq);
1144 __remove_osd_from_lru(osd);
1145 erase_osd(&osdc->osds, osd);
1152 static int reopen_osd(struct ceph_osd *osd)
1154 struct ceph_entity_addr *peer_addr;
1156 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1158 if (RB_EMPTY_ROOT(&osd->o_requests) &&
1159 RB_EMPTY_ROOT(&osd->o_linger_requests)) {
1164 peer_addr = &osd->o_osdc->osdmap->osd_addr[osd->o_osd];
1165 if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1166 !ceph_con_opened(&osd->o_con)) {
1169 dout("osd addr hasn't changed and connection never opened, "
1170 "letting msgr retry\n");
1171 /* touch each r_stamp for handle_timeout()'s benfit */
1172 for (n = rb_first(&osd->o_requests); n; n = rb_next(n)) {
1173 struct ceph_osd_request *req =
1174 rb_entry(n, struct ceph_osd_request, r_node);
1175 req->r_stamp = jiffies;
1181 ceph_con_close(&osd->o_con);
1182 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1183 osd->o_incarnation++;
1188 static struct ceph_osd *lookup_create_osd(struct ceph_osd_client *osdc, int o,
1191 struct ceph_osd *osd;
1194 verify_osdc_wrlocked(osdc);
1196 verify_osdc_locked(osdc);
1198 if (o != CEPH_HOMELESS_OSD)
1199 osd = lookup_osd(&osdc->osds, o);
1201 osd = &osdc->homeless_osd;
1204 return ERR_PTR(-EAGAIN);
1206 osd = create_osd(osdc, o);
1207 insert_osd(&osdc->osds, osd);
1208 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd,
1209 &osdc->osdmap->osd_addr[osd->o_osd]);
1212 dout("%s osdc %p osd%d -> osd %p\n", __func__, osdc, o, osd);
1217 * Create request <-> OSD session relation.
1219 * @req has to be assigned a tid, @osd may be homeless.
1221 static void link_request(struct ceph_osd *osd, struct ceph_osd_request *req)
1223 verify_osd_locked(osd);
1224 WARN_ON(!req->r_tid || req->r_osd);
1225 dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd,
1228 if (!osd_homeless(osd))
1229 __remove_osd_from_lru(osd);
1231 atomic_inc(&osd->o_osdc->num_homeless);
1234 insert_request(&osd->o_requests, req);
1238 static void unlink_request(struct ceph_osd *osd, struct ceph_osd_request *req)
1240 verify_osd_locked(osd);
1241 WARN_ON(req->r_osd != osd);
1242 dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd,
1246 erase_request(&osd->o_requests, req);
1249 if (!osd_homeless(osd))
1250 maybe_move_osd_to_lru(osd);
1252 atomic_dec(&osd->o_osdc->num_homeless);
1255 static bool __pool_full(struct ceph_pg_pool_info *pi)
1257 return pi->flags & CEPH_POOL_FLAG_FULL;
1260 static bool have_pool_full(struct ceph_osd_client *osdc)
1264 for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) {
1265 struct ceph_pg_pool_info *pi =
1266 rb_entry(n, struct ceph_pg_pool_info, node);
1268 if (__pool_full(pi))
1275 static bool pool_full(struct ceph_osd_client *osdc, s64 pool_id)
1277 struct ceph_pg_pool_info *pi;
1279 pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id);
1283 return __pool_full(pi);
1287 * Returns whether a request should be blocked from being sent
1288 * based on the current osdmap and osd_client settings.
1290 static bool target_should_be_paused(struct ceph_osd_client *osdc,
1291 const struct ceph_osd_request_target *t,
1292 struct ceph_pg_pool_info *pi)
1294 bool pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
1295 bool pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
1296 ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
1299 WARN_ON(pi->id != t->base_oloc.pool);
1300 return (t->flags & CEPH_OSD_FLAG_READ && pauserd) ||
1301 (t->flags & CEPH_OSD_FLAG_WRITE && pausewr);
1304 enum calc_target_result {
1305 CALC_TARGET_NO_ACTION = 0,
1306 CALC_TARGET_NEED_RESEND,
1307 CALC_TARGET_POOL_DNE,
1310 static enum calc_target_result calc_target(struct ceph_osd_client *osdc,
1311 struct ceph_osd_request_target *t,
1312 u32 *last_force_resend,
1315 struct ceph_pg_pool_info *pi;
1316 struct ceph_pg pgid, last_pgid;
1317 struct ceph_osds up, acting;
1318 bool force_resend = false;
1319 bool need_check_tiering = false;
1320 bool need_resend = false;
1321 bool sort_bitwise = ceph_osdmap_flag(osdc, CEPH_OSDMAP_SORTBITWISE);
1322 enum calc_target_result ct_res;
1325 pi = ceph_pg_pool_by_id(osdc->osdmap, t->base_oloc.pool);
1327 t->osd = CEPH_HOMELESS_OSD;
1328 ct_res = CALC_TARGET_POOL_DNE;
1332 if (osdc->osdmap->epoch == pi->last_force_request_resend) {
1333 if (last_force_resend &&
1334 *last_force_resend < pi->last_force_request_resend) {
1335 *last_force_resend = pi->last_force_request_resend;
1336 force_resend = true;
1337 } else if (!last_force_resend) {
1338 force_resend = true;
1341 if (ceph_oid_empty(&t->target_oid) || force_resend) {
1342 ceph_oid_copy(&t->target_oid, &t->base_oid);
1343 need_check_tiering = true;
1345 if (ceph_oloc_empty(&t->target_oloc) || force_resend) {
1346 ceph_oloc_copy(&t->target_oloc, &t->base_oloc);
1347 need_check_tiering = true;
1350 if (need_check_tiering &&
1351 (t->flags & CEPH_OSD_FLAG_IGNORE_OVERLAY) == 0) {
1352 if (t->flags & CEPH_OSD_FLAG_READ && pi->read_tier >= 0)
1353 t->target_oloc.pool = pi->read_tier;
1354 if (t->flags & CEPH_OSD_FLAG_WRITE && pi->write_tier >= 0)
1355 t->target_oloc.pool = pi->write_tier;
1358 ret = ceph_object_locator_to_pg(osdc->osdmap, &t->target_oid,
1359 &t->target_oloc, &pgid);
1361 WARN_ON(ret != -ENOENT);
1362 t->osd = CEPH_HOMELESS_OSD;
1363 ct_res = CALC_TARGET_POOL_DNE;
1366 last_pgid.pool = pgid.pool;
1367 last_pgid.seed = ceph_stable_mod(pgid.seed, t->pg_num, t->pg_num_mask);
1369 ceph_pg_to_up_acting_osds(osdc->osdmap, &pgid, &up, &acting);
1371 ceph_is_new_interval(&t->acting,
1384 force_resend = true;
1386 if (t->paused && !target_should_be_paused(osdc, t, pi)) {
1391 if (ceph_pg_compare(&t->pgid, &pgid) ||
1392 ceph_osds_changed(&t->acting, &acting, any_change) ||
1394 t->pgid = pgid; /* struct */
1395 ceph_osds_copy(&t->acting, &acting);
1396 ceph_osds_copy(&t->up, &up);
1398 t->min_size = pi->min_size;
1399 t->pg_num = pi->pg_num;
1400 t->pg_num_mask = pi->pg_num_mask;
1401 t->sort_bitwise = sort_bitwise;
1403 t->osd = acting.primary;
1407 ct_res = need_resend ? CALC_TARGET_NEED_RESEND : CALC_TARGET_NO_ACTION;
1409 dout("%s t %p -> ct_res %d osd %d\n", __func__, t, ct_res, t->osd);
1413 static void setup_request_data(struct ceph_osd_request *req,
1414 struct ceph_msg *msg)
1419 if (!list_empty(&msg->data))
1422 WARN_ON(msg->data_length);
1423 for (i = 0; i < req->r_num_ops; i++) {
1424 struct ceph_osd_req_op *op = &req->r_ops[i];
1428 case CEPH_OSD_OP_WRITE:
1429 case CEPH_OSD_OP_WRITEFULL:
1430 WARN_ON(op->indata_len != op->extent.length);
1431 ceph_osdc_msg_data_add(msg, &op->extent.osd_data);
1433 case CEPH_OSD_OP_SETXATTR:
1434 case CEPH_OSD_OP_CMPXATTR:
1435 WARN_ON(op->indata_len != op->xattr.name_len +
1436 op->xattr.value_len);
1437 ceph_osdc_msg_data_add(msg, &op->xattr.osd_data);
1439 case CEPH_OSD_OP_NOTIFY_ACK:
1440 ceph_osdc_msg_data_add(msg,
1441 &op->notify_ack.request_data);
1445 case CEPH_OSD_OP_STAT:
1446 ceph_osdc_msg_data_add(req->r_reply,
1449 case CEPH_OSD_OP_READ:
1450 ceph_osdc_msg_data_add(req->r_reply,
1451 &op->extent.osd_data);
1453 case CEPH_OSD_OP_LIST_WATCHERS:
1454 ceph_osdc_msg_data_add(req->r_reply,
1455 &op->list_watchers.response_data);
1459 case CEPH_OSD_OP_CALL:
1460 WARN_ON(op->indata_len != op->cls.class_len +
1461 op->cls.method_len +
1462 op->cls.indata_len);
1463 ceph_osdc_msg_data_add(msg, &op->cls.request_info);
1464 /* optional, can be NONE */
1465 ceph_osdc_msg_data_add(msg, &op->cls.request_data);
1466 /* optional, can be NONE */
1467 ceph_osdc_msg_data_add(req->r_reply,
1468 &op->cls.response_data);
1470 case CEPH_OSD_OP_NOTIFY:
1471 ceph_osdc_msg_data_add(msg,
1472 &op->notify.request_data);
1473 ceph_osdc_msg_data_add(req->r_reply,
1474 &op->notify.response_data);
1478 data_len += op->indata_len;
1481 WARN_ON(data_len != msg->data_length);
1484 static void encode_request(struct ceph_osd_request *req, struct ceph_msg *msg)
1486 void *p = msg->front.iov_base;
1487 void *const end = p + msg->front_alloc_len;
1491 if (req->r_flags & CEPH_OSD_FLAG_WRITE) {
1492 /* snapshots aren't writeable */
1493 WARN_ON(req->r_snapid != CEPH_NOSNAP);
1495 WARN_ON(req->r_mtime.tv_sec || req->r_mtime.tv_nsec ||
1496 req->r_data_offset || req->r_snapc);
1499 setup_request_data(req, msg);
1501 ceph_encode_32(&p, 1); /* client_inc, always 1 */
1502 ceph_encode_32(&p, req->r_osdc->osdmap->epoch);
1503 ceph_encode_32(&p, req->r_flags);
1504 ceph_encode_timespec(p, &req->r_mtime);
1505 p += sizeof(struct ceph_timespec);
1506 /* aka reassert_version */
1507 memcpy(p, &req->r_replay_version, sizeof(req->r_replay_version));
1508 p += sizeof(req->r_replay_version);
1511 ceph_start_encoding(&p, 5, 4,
1512 ceph_oloc_encoding_size(&req->r_t.target_oloc));
1513 ceph_encode_64(&p, req->r_t.target_oloc.pool);
1514 ceph_encode_32(&p, -1); /* preferred */
1515 ceph_encode_32(&p, 0); /* key len */
1516 if (req->r_t.target_oloc.pool_ns)
1517 ceph_encode_string(&p, end, req->r_t.target_oloc.pool_ns->str,
1518 req->r_t.target_oloc.pool_ns->len);
1520 ceph_encode_32(&p, 0);
1523 ceph_encode_8(&p, 1);
1524 ceph_encode_64(&p, req->r_t.pgid.pool);
1525 ceph_encode_32(&p, req->r_t.pgid.seed);
1526 ceph_encode_32(&p, -1); /* preferred */
1529 ceph_encode_32(&p, req->r_t.target_oid.name_len);
1530 memcpy(p, req->r_t.target_oid.name, req->r_t.target_oid.name_len);
1531 p += req->r_t.target_oid.name_len;
1533 /* ops, can imply data */
1534 ceph_encode_16(&p, req->r_num_ops);
1535 for (i = 0; i < req->r_num_ops; i++) {
1536 data_len += osd_req_encode_op(p, &req->r_ops[i]);
1537 p += sizeof(struct ceph_osd_op);
1540 ceph_encode_64(&p, req->r_snapid); /* snapid */
1542 ceph_encode_64(&p, req->r_snapc->seq);
1543 ceph_encode_32(&p, req->r_snapc->num_snaps);
1544 for (i = 0; i < req->r_snapc->num_snaps; i++)
1545 ceph_encode_64(&p, req->r_snapc->snaps[i]);
1547 ceph_encode_64(&p, 0); /* snap_seq */
1548 ceph_encode_32(&p, 0); /* snaps len */
1551 ceph_encode_32(&p, req->r_attempts); /* retry_attempt */
1554 msg->front.iov_len = p - msg->front.iov_base;
1555 msg->hdr.version = cpu_to_le16(4); /* MOSDOp v4 */
1556 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1557 msg->hdr.data_len = cpu_to_le32(data_len);
1559 * The header "data_off" is a hint to the receiver allowing it
1560 * to align received data into its buffers such that there's no
1561 * need to re-copy it before writing it to disk (direct I/O).
1563 msg->hdr.data_off = cpu_to_le16(req->r_data_offset);
1565 dout("%s req %p oid %s oid_len %d front %zu data %u\n", __func__,
1566 req, req->r_t.target_oid.name, req->r_t.target_oid.name_len,
1567 msg->front.iov_len, data_len);
1571 * @req has to be assigned a tid and registered.
1573 static void send_request(struct ceph_osd_request *req)
1575 struct ceph_osd *osd = req->r_osd;
1577 verify_osd_locked(osd);
1578 WARN_ON(osd->o_osd != req->r_t.osd);
1581 * We may have a previously queued request message hanging
1582 * around. Cancel it to avoid corrupting the msgr.
1585 ceph_msg_revoke(req->r_request);
1587 req->r_flags |= CEPH_OSD_FLAG_KNOWN_REDIR;
1588 if (req->r_attempts)
1589 req->r_flags |= CEPH_OSD_FLAG_RETRY;
1591 WARN_ON(req->r_flags & CEPH_OSD_FLAG_RETRY);
1593 encode_request(req, req->r_request);
1595 dout("%s req %p tid %llu to pg %llu.%x osd%d flags 0x%x attempt %d\n",
1596 __func__, req, req->r_tid, req->r_t.pgid.pool, req->r_t.pgid.seed,
1597 req->r_t.osd, req->r_flags, req->r_attempts);
1599 req->r_t.paused = false;
1600 req->r_stamp = jiffies;
1603 req->r_sent = osd->o_incarnation;
1604 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
1605 ceph_con_send(&osd->o_con, ceph_msg_get(req->r_request));
1608 static void maybe_request_map(struct ceph_osd_client *osdc)
1610 bool continuous = false;
1612 verify_osdc_locked(osdc);
1613 WARN_ON(!osdc->osdmap->epoch);
1615 if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
1616 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD) ||
1617 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) {
1618 dout("%s osdc %p continuous\n", __func__, osdc);
1621 dout("%s osdc %p onetime\n", __func__, osdc);
1624 if (ceph_monc_want_map(&osdc->client->monc, CEPH_SUB_OSDMAP,
1625 osdc->osdmap->epoch + 1, continuous))
1626 ceph_monc_renew_subs(&osdc->client->monc);
1629 static void send_map_check(struct ceph_osd_request *req);
1631 static void __submit_request(struct ceph_osd_request *req, bool wrlocked)
1633 struct ceph_osd_client *osdc = req->r_osdc;
1634 struct ceph_osd *osd;
1635 enum calc_target_result ct_res;
1636 bool need_send = false;
1637 bool promoted = false;
1639 WARN_ON(req->r_tid || req->r_got_reply);
1640 dout("%s req %p wrlocked %d\n", __func__, req, wrlocked);
1643 ct_res = calc_target(osdc, &req->r_t, &req->r_last_force_resend, false);
1644 if (ct_res == CALC_TARGET_POOL_DNE && !wrlocked)
1647 osd = lookup_create_osd(osdc, req->r_t.osd, wrlocked);
1649 WARN_ON(PTR_ERR(osd) != -EAGAIN || wrlocked);
1653 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
1654 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) {
1655 dout("req %p pausewr\n", req);
1656 req->r_t.paused = true;
1657 maybe_request_map(osdc);
1658 } else if ((req->r_flags & CEPH_OSD_FLAG_READ) &&
1659 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) {
1660 dout("req %p pauserd\n", req);
1661 req->r_t.paused = true;
1662 maybe_request_map(osdc);
1663 } else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
1664 !(req->r_flags & (CEPH_OSD_FLAG_FULL_TRY |
1665 CEPH_OSD_FLAG_FULL_FORCE)) &&
1666 (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
1667 pool_full(osdc, req->r_t.base_oloc.pool))) {
1668 dout("req %p full/pool_full\n", req);
1669 pr_warn_ratelimited("FULL or reached pool quota\n");
1670 req->r_t.paused = true;
1671 maybe_request_map(osdc);
1672 } else if (!osd_homeless(osd)) {
1675 maybe_request_map(osdc);
1678 mutex_lock(&osd->lock);
1680 * Assign the tid atomically with send_request() to protect
1681 * multiple writes to the same object from racing with each
1682 * other, resulting in out of order ops on the OSDs.
1684 req->r_tid = atomic64_inc_return(&osdc->last_tid);
1685 link_request(osd, req);
1688 mutex_unlock(&osd->lock);
1690 if (ct_res == CALC_TARGET_POOL_DNE)
1691 send_map_check(req);
1694 downgrade_write(&osdc->lock);
1698 up_read(&osdc->lock);
1699 down_write(&osdc->lock);
1705 static void account_request(struct ceph_osd_request *req)
1707 unsigned int mask = CEPH_OSD_FLAG_ACK | CEPH_OSD_FLAG_ONDISK;
1709 if (req->r_flags & CEPH_OSD_FLAG_READ) {
1710 WARN_ON(req->r_flags & mask);
1711 req->r_flags |= CEPH_OSD_FLAG_ACK;
1712 } else if (req->r_flags & CEPH_OSD_FLAG_WRITE)
1713 WARN_ON(!(req->r_flags & mask));
1717 WARN_ON(req->r_unsafe_callback && (req->r_flags & mask) != mask);
1718 atomic_inc(&req->r_osdc->num_requests);
1721 static void submit_request(struct ceph_osd_request *req, bool wrlocked)
1723 ceph_osdc_get_request(req);
1724 account_request(req);
1725 __submit_request(req, wrlocked);
1728 static void finish_request(struct ceph_osd_request *req)
1730 struct ceph_osd_client *osdc = req->r_osdc;
1731 struct ceph_osd *osd = req->r_osd;
1733 verify_osd_locked(osd);
1734 dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
1736 WARN_ON(lookup_request_mc(&osdc->map_checks, req->r_tid));
1737 unlink_request(osd, req);
1738 atomic_dec(&osdc->num_requests);
1741 * If an OSD has failed or returned and a request has been sent
1742 * twice, it's possible to get a reply and end up here while the
1743 * request message is queued for delivery. We will ignore the
1744 * reply, so not a big deal, but better to try and catch it.
1746 ceph_msg_revoke(req->r_request);
1747 ceph_msg_revoke_incoming(req->r_reply);
1750 static void __complete_request(struct ceph_osd_request *req)
1752 if (req->r_callback)
1753 req->r_callback(req);
1755 complete_all(&req->r_completion);
1759 * Note that this is open-coded in handle_reply(), which has to deal
1760 * with ack vs commit, dup acks, etc.
1762 static void complete_request(struct ceph_osd_request *req, int err)
1764 dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err);
1766 req->r_result = err;
1767 finish_request(req);
1768 __complete_request(req);
1769 complete_all(&req->r_done_completion);
1770 ceph_osdc_put_request(req);
1773 static void cancel_map_check(struct ceph_osd_request *req)
1775 struct ceph_osd_client *osdc = req->r_osdc;
1776 struct ceph_osd_request *lookup_req;
1778 verify_osdc_wrlocked(osdc);
1780 lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid);
1784 WARN_ON(lookup_req != req);
1785 erase_request_mc(&osdc->map_checks, req);
1786 ceph_osdc_put_request(req);
1789 static void cancel_request(struct ceph_osd_request *req)
1791 dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
1793 cancel_map_check(req);
1794 finish_request(req);
1795 complete_all(&req->r_done_completion);
1796 ceph_osdc_put_request(req);
1799 static void check_pool_dne(struct ceph_osd_request *req)
1801 struct ceph_osd_client *osdc = req->r_osdc;
1802 struct ceph_osdmap *map = osdc->osdmap;
1804 verify_osdc_wrlocked(osdc);
1805 WARN_ON(!map->epoch);
1807 if (req->r_attempts) {
1809 * We sent a request earlier, which means that
1810 * previously the pool existed, and now it does not
1811 * (i.e., it was deleted).
1813 req->r_map_dne_bound = map->epoch;
1814 dout("%s req %p tid %llu pool disappeared\n", __func__, req,
1817 dout("%s req %p tid %llu map_dne_bound %u have %u\n", __func__,
1818 req, req->r_tid, req->r_map_dne_bound, map->epoch);
1821 if (req->r_map_dne_bound) {
1822 if (map->epoch >= req->r_map_dne_bound) {
1823 /* we had a new enough map */
1824 pr_info_ratelimited("tid %llu pool does not exist\n",
1826 complete_request(req, -ENOENT);
1829 send_map_check(req);
1833 static void map_check_cb(struct ceph_mon_generic_request *greq)
1835 struct ceph_osd_client *osdc = &greq->monc->client->osdc;
1836 struct ceph_osd_request *req;
1837 u64 tid = greq->private_data;
1839 WARN_ON(greq->result || !greq->u.newest);
1841 down_write(&osdc->lock);
1842 req = lookup_request_mc(&osdc->map_checks, tid);
1844 dout("%s tid %llu dne\n", __func__, tid);
1848 dout("%s req %p tid %llu map_dne_bound %u newest %llu\n", __func__,
1849 req, req->r_tid, req->r_map_dne_bound, greq->u.newest);
1850 if (!req->r_map_dne_bound)
1851 req->r_map_dne_bound = greq->u.newest;
1852 erase_request_mc(&osdc->map_checks, req);
1853 check_pool_dne(req);
1855 ceph_osdc_put_request(req);
1857 up_write(&osdc->lock);
1860 static void send_map_check(struct ceph_osd_request *req)
1862 struct ceph_osd_client *osdc = req->r_osdc;
1863 struct ceph_osd_request *lookup_req;
1866 verify_osdc_wrlocked(osdc);
1868 lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid);
1870 WARN_ON(lookup_req != req);
1874 ceph_osdc_get_request(req);
1875 insert_request_mc(&osdc->map_checks, req);
1876 ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap",
1877 map_check_cb, req->r_tid);
1882 * lingering requests, watch/notify v2 infrastructure
1884 static void linger_release(struct kref *kref)
1886 struct ceph_osd_linger_request *lreq =
1887 container_of(kref, struct ceph_osd_linger_request, kref);
1889 dout("%s lreq %p reg_req %p ping_req %p\n", __func__, lreq,
1890 lreq->reg_req, lreq->ping_req);
1891 WARN_ON(!RB_EMPTY_NODE(&lreq->node));
1892 WARN_ON(!RB_EMPTY_NODE(&lreq->osdc_node));
1893 WARN_ON(!RB_EMPTY_NODE(&lreq->mc_node));
1894 WARN_ON(!list_empty(&lreq->scan_item));
1895 WARN_ON(!list_empty(&lreq->pending_lworks));
1899 ceph_osdc_put_request(lreq->reg_req);
1901 ceph_osdc_put_request(lreq->ping_req);
1902 target_destroy(&lreq->t);
1906 static void linger_put(struct ceph_osd_linger_request *lreq)
1909 kref_put(&lreq->kref, linger_release);
1912 static struct ceph_osd_linger_request *
1913 linger_get(struct ceph_osd_linger_request *lreq)
1915 kref_get(&lreq->kref);
1919 static struct ceph_osd_linger_request *
1920 linger_alloc(struct ceph_osd_client *osdc)
1922 struct ceph_osd_linger_request *lreq;
1924 lreq = kzalloc(sizeof(*lreq), GFP_NOIO);
1928 kref_init(&lreq->kref);
1929 mutex_init(&lreq->lock);
1930 RB_CLEAR_NODE(&lreq->node);
1931 RB_CLEAR_NODE(&lreq->osdc_node);
1932 RB_CLEAR_NODE(&lreq->mc_node);
1933 INIT_LIST_HEAD(&lreq->scan_item);
1934 INIT_LIST_HEAD(&lreq->pending_lworks);
1935 init_completion(&lreq->reg_commit_wait);
1936 init_completion(&lreq->notify_finish_wait);
1939 target_init(&lreq->t);
1941 dout("%s lreq %p\n", __func__, lreq);
1945 DEFINE_RB_INSDEL_FUNCS(linger, struct ceph_osd_linger_request, linger_id, node)
1946 DEFINE_RB_FUNCS(linger_osdc, struct ceph_osd_linger_request, linger_id, osdc_node)
1947 DEFINE_RB_FUNCS(linger_mc, struct ceph_osd_linger_request, linger_id, mc_node)
1950 * Create linger request <-> OSD session relation.
1952 * @lreq has to be registered, @osd may be homeless.
1954 static void link_linger(struct ceph_osd *osd,
1955 struct ceph_osd_linger_request *lreq)
1957 verify_osd_locked(osd);
1958 WARN_ON(!lreq->linger_id || lreq->osd);
1959 dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd,
1960 osd->o_osd, lreq, lreq->linger_id);
1962 if (!osd_homeless(osd))
1963 __remove_osd_from_lru(osd);
1965 atomic_inc(&osd->o_osdc->num_homeless);
1968 insert_linger(&osd->o_linger_requests, lreq);
1972 static void unlink_linger(struct ceph_osd *osd,
1973 struct ceph_osd_linger_request *lreq)
1975 verify_osd_locked(osd);
1976 WARN_ON(lreq->osd != osd);
1977 dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd,
1978 osd->o_osd, lreq, lreq->linger_id);
1981 erase_linger(&osd->o_linger_requests, lreq);
1984 if (!osd_homeless(osd))
1985 maybe_move_osd_to_lru(osd);
1987 atomic_dec(&osd->o_osdc->num_homeless);
1990 static bool __linger_registered(struct ceph_osd_linger_request *lreq)
1992 verify_osdc_locked(lreq->osdc);
1994 return !RB_EMPTY_NODE(&lreq->osdc_node);
1997 static bool linger_registered(struct ceph_osd_linger_request *lreq)
1999 struct ceph_osd_client *osdc = lreq->osdc;
2002 down_read(&osdc->lock);
2003 registered = __linger_registered(lreq);
2004 up_read(&osdc->lock);
2009 static void linger_register(struct ceph_osd_linger_request *lreq)
2011 struct ceph_osd_client *osdc = lreq->osdc;
2013 verify_osdc_wrlocked(osdc);
2014 WARN_ON(lreq->linger_id);
2017 lreq->linger_id = ++osdc->last_linger_id;
2018 insert_linger_osdc(&osdc->linger_requests, lreq);
2021 static void linger_unregister(struct ceph_osd_linger_request *lreq)
2023 struct ceph_osd_client *osdc = lreq->osdc;
2025 verify_osdc_wrlocked(osdc);
2027 erase_linger_osdc(&osdc->linger_requests, lreq);
2031 static void cancel_linger_request(struct ceph_osd_request *req)
2033 struct ceph_osd_linger_request *lreq = req->r_priv;
2035 WARN_ON(!req->r_linger);
2036 cancel_request(req);
2040 struct linger_work {
2041 struct work_struct work;
2042 struct ceph_osd_linger_request *lreq;
2043 struct list_head pending_item;
2044 unsigned long queued_stamp;
2050 void *payload; /* points into @msg front */
2053 struct ceph_msg *msg; /* for ceph_msg_put() */
2061 static struct linger_work *lwork_alloc(struct ceph_osd_linger_request *lreq,
2064 struct linger_work *lwork;
2066 lwork = kzalloc(sizeof(*lwork), GFP_NOIO);
2070 INIT_WORK(&lwork->work, workfn);
2071 INIT_LIST_HEAD(&lwork->pending_item);
2072 lwork->lreq = linger_get(lreq);
2077 static void lwork_free(struct linger_work *lwork)
2079 struct ceph_osd_linger_request *lreq = lwork->lreq;
2081 mutex_lock(&lreq->lock);
2082 list_del(&lwork->pending_item);
2083 mutex_unlock(&lreq->lock);
2089 static void lwork_queue(struct linger_work *lwork)
2091 struct ceph_osd_linger_request *lreq = lwork->lreq;
2092 struct ceph_osd_client *osdc = lreq->osdc;
2094 verify_lreq_locked(lreq);
2095 WARN_ON(!list_empty(&lwork->pending_item));
2097 lwork->queued_stamp = jiffies;
2098 list_add_tail(&lwork->pending_item, &lreq->pending_lworks);
2099 queue_work(osdc->notify_wq, &lwork->work);
2102 static void do_watch_notify(struct work_struct *w)
2104 struct linger_work *lwork = container_of(w, struct linger_work, work);
2105 struct ceph_osd_linger_request *lreq = lwork->lreq;
2107 if (!linger_registered(lreq)) {
2108 dout("%s lreq %p not registered\n", __func__, lreq);
2112 WARN_ON(!lreq->is_watch);
2113 dout("%s lreq %p notify_id %llu notifier_id %llu payload_len %zu\n",
2114 __func__, lreq, lwork->notify.notify_id, lwork->notify.notifier_id,
2115 lwork->notify.payload_len);
2116 lreq->wcb(lreq->data, lwork->notify.notify_id, lreq->linger_id,
2117 lwork->notify.notifier_id, lwork->notify.payload,
2118 lwork->notify.payload_len);
2121 ceph_msg_put(lwork->notify.msg);
2125 static void do_watch_error(struct work_struct *w)
2127 struct linger_work *lwork = container_of(w, struct linger_work, work);
2128 struct ceph_osd_linger_request *lreq = lwork->lreq;
2130 if (!linger_registered(lreq)) {
2131 dout("%s lreq %p not registered\n", __func__, lreq);
2135 dout("%s lreq %p err %d\n", __func__, lreq, lwork->error.err);
2136 lreq->errcb(lreq->data, lreq->linger_id, lwork->error.err);
2142 static void queue_watch_error(struct ceph_osd_linger_request *lreq)
2144 struct linger_work *lwork;
2146 lwork = lwork_alloc(lreq, do_watch_error);
2148 pr_err("failed to allocate error-lwork\n");
2152 lwork->error.err = lreq->last_error;
2156 static void linger_reg_commit_complete(struct ceph_osd_linger_request *lreq,
2159 if (!completion_done(&lreq->reg_commit_wait)) {
2160 lreq->reg_commit_error = (result <= 0 ? result : 0);
2161 complete_all(&lreq->reg_commit_wait);
2165 static void linger_commit_cb(struct ceph_osd_request *req)
2167 struct ceph_osd_linger_request *lreq = req->r_priv;
2169 mutex_lock(&lreq->lock);
2170 dout("%s lreq %p linger_id %llu result %d\n", __func__, lreq,
2171 lreq->linger_id, req->r_result);
2172 WARN_ON(!__linger_registered(lreq));
2173 linger_reg_commit_complete(lreq, req->r_result);
2174 lreq->committed = true;
2176 if (!lreq->is_watch) {
2177 struct ceph_osd_data *osd_data =
2178 osd_req_op_data(req, 0, notify, response_data);
2179 void *p = page_address(osd_data->pages[0]);
2181 WARN_ON(req->r_ops[0].op != CEPH_OSD_OP_NOTIFY ||
2182 osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
2184 /* make note of the notify_id */
2185 if (req->r_ops[0].outdata_len >= sizeof(u64)) {
2186 lreq->notify_id = ceph_decode_64(&p);
2187 dout("lreq %p notify_id %llu\n", lreq,
2190 dout("lreq %p no notify_id\n", lreq);
2194 mutex_unlock(&lreq->lock);
2198 static int normalize_watch_error(int err)
2201 * Translate ENOENT -> ENOTCONN so that a delete->disconnection
2202 * notification and a failure to reconnect because we raced with
2203 * the delete appear the same to the user.
2211 static void linger_reconnect_cb(struct ceph_osd_request *req)
2213 struct ceph_osd_linger_request *lreq = req->r_priv;
2215 mutex_lock(&lreq->lock);
2216 dout("%s lreq %p linger_id %llu result %d last_error %d\n", __func__,
2217 lreq, lreq->linger_id, req->r_result, lreq->last_error);
2218 if (req->r_result < 0) {
2219 if (!lreq->last_error) {
2220 lreq->last_error = normalize_watch_error(req->r_result);
2221 queue_watch_error(lreq);
2225 mutex_unlock(&lreq->lock);
2229 static void send_linger(struct ceph_osd_linger_request *lreq)
2231 struct ceph_osd_request *req = lreq->reg_req;
2232 struct ceph_osd_req_op *op = &req->r_ops[0];
2234 verify_osdc_wrlocked(req->r_osdc);
2235 dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
2238 cancel_linger_request(req);
2240 request_reinit(req);
2241 ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
2242 ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
2243 req->r_flags = lreq->t.flags;
2244 req->r_mtime = lreq->mtime;
2246 mutex_lock(&lreq->lock);
2247 if (lreq->is_watch && lreq->committed) {
2248 WARN_ON(op->op != CEPH_OSD_OP_WATCH ||
2249 op->watch.cookie != lreq->linger_id);
2250 op->watch.op = CEPH_OSD_WATCH_OP_RECONNECT;
2251 op->watch.gen = ++lreq->register_gen;
2252 dout("lreq %p reconnect register_gen %u\n", lreq,
2254 req->r_callback = linger_reconnect_cb;
2256 if (!lreq->is_watch)
2257 lreq->notify_id = 0;
2259 WARN_ON(op->watch.op != CEPH_OSD_WATCH_OP_WATCH);
2260 dout("lreq %p register\n", lreq);
2261 req->r_callback = linger_commit_cb;
2263 mutex_unlock(&lreq->lock);
2265 req->r_priv = linger_get(lreq);
2266 req->r_linger = true;
2268 submit_request(req, true);
2271 static void linger_ping_cb(struct ceph_osd_request *req)
2273 struct ceph_osd_linger_request *lreq = req->r_priv;
2275 mutex_lock(&lreq->lock);
2276 dout("%s lreq %p linger_id %llu result %d ping_sent %lu last_error %d\n",
2277 __func__, lreq, lreq->linger_id, req->r_result, lreq->ping_sent,
2279 if (lreq->register_gen == req->r_ops[0].watch.gen) {
2280 if (!req->r_result) {
2281 lreq->watch_valid_thru = lreq->ping_sent;
2282 } else if (!lreq->last_error) {
2283 lreq->last_error = normalize_watch_error(req->r_result);
2284 queue_watch_error(lreq);
2287 dout("lreq %p register_gen %u ignoring old pong %u\n", lreq,
2288 lreq->register_gen, req->r_ops[0].watch.gen);
2291 mutex_unlock(&lreq->lock);
2295 static void send_linger_ping(struct ceph_osd_linger_request *lreq)
2297 struct ceph_osd_client *osdc = lreq->osdc;
2298 struct ceph_osd_request *req = lreq->ping_req;
2299 struct ceph_osd_req_op *op = &req->r_ops[0];
2301 if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) {
2302 dout("%s PAUSERD\n", __func__);
2306 lreq->ping_sent = jiffies;
2307 dout("%s lreq %p linger_id %llu ping_sent %lu register_gen %u\n",
2308 __func__, lreq, lreq->linger_id, lreq->ping_sent,
2309 lreq->register_gen);
2312 cancel_linger_request(req);
2314 request_reinit(req);
2315 target_copy(&req->r_t, &lreq->t);
2317 WARN_ON(op->op != CEPH_OSD_OP_WATCH ||
2318 op->watch.cookie != lreq->linger_id ||
2319 op->watch.op != CEPH_OSD_WATCH_OP_PING);
2320 op->watch.gen = lreq->register_gen;
2321 req->r_callback = linger_ping_cb;
2322 req->r_priv = linger_get(lreq);
2323 req->r_linger = true;
2325 ceph_osdc_get_request(req);
2326 account_request(req);
2327 req->r_tid = atomic64_inc_return(&osdc->last_tid);
2328 link_request(lreq->osd, req);
2332 static void linger_submit(struct ceph_osd_linger_request *lreq)
2334 struct ceph_osd_client *osdc = lreq->osdc;
2335 struct ceph_osd *osd;
2337 calc_target(osdc, &lreq->t, &lreq->last_force_resend, false);
2338 osd = lookup_create_osd(osdc, lreq->t.osd, true);
2339 link_linger(osd, lreq);
2344 static void cancel_linger_map_check(struct ceph_osd_linger_request *lreq)
2346 struct ceph_osd_client *osdc = lreq->osdc;
2347 struct ceph_osd_linger_request *lookup_lreq;
2349 verify_osdc_wrlocked(osdc);
2351 lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks,
2356 WARN_ON(lookup_lreq != lreq);
2357 erase_linger_mc(&osdc->linger_map_checks, lreq);
2362 * @lreq has to be both registered and linked.
2364 static void __linger_cancel(struct ceph_osd_linger_request *lreq)
2366 if (lreq->is_watch && lreq->ping_req->r_osd)
2367 cancel_linger_request(lreq->ping_req);
2368 if (lreq->reg_req->r_osd)
2369 cancel_linger_request(lreq->reg_req);
2370 cancel_linger_map_check(lreq);
2371 unlink_linger(lreq->osd, lreq);
2372 linger_unregister(lreq);
2375 static void linger_cancel(struct ceph_osd_linger_request *lreq)
2377 struct ceph_osd_client *osdc = lreq->osdc;
2379 down_write(&osdc->lock);
2380 if (__linger_registered(lreq))
2381 __linger_cancel(lreq);
2382 up_write(&osdc->lock);
2385 static void send_linger_map_check(struct ceph_osd_linger_request *lreq);
2387 static void check_linger_pool_dne(struct ceph_osd_linger_request *lreq)
2389 struct ceph_osd_client *osdc = lreq->osdc;
2390 struct ceph_osdmap *map = osdc->osdmap;
2392 verify_osdc_wrlocked(osdc);
2393 WARN_ON(!map->epoch);
2395 if (lreq->register_gen) {
2396 lreq->map_dne_bound = map->epoch;
2397 dout("%s lreq %p linger_id %llu pool disappeared\n", __func__,
2398 lreq, lreq->linger_id);
2400 dout("%s lreq %p linger_id %llu map_dne_bound %u have %u\n",
2401 __func__, lreq, lreq->linger_id, lreq->map_dne_bound,
2405 if (lreq->map_dne_bound) {
2406 if (map->epoch >= lreq->map_dne_bound) {
2407 /* we had a new enough map */
2408 pr_info("linger_id %llu pool does not exist\n",
2410 linger_reg_commit_complete(lreq, -ENOENT);
2411 __linger_cancel(lreq);
2414 send_linger_map_check(lreq);
2418 static void linger_map_check_cb(struct ceph_mon_generic_request *greq)
2420 struct ceph_osd_client *osdc = &greq->monc->client->osdc;
2421 struct ceph_osd_linger_request *lreq;
2422 u64 linger_id = greq->private_data;
2424 WARN_ON(greq->result || !greq->u.newest);
2426 down_write(&osdc->lock);
2427 lreq = lookup_linger_mc(&osdc->linger_map_checks, linger_id);
2429 dout("%s linger_id %llu dne\n", __func__, linger_id);
2433 dout("%s lreq %p linger_id %llu map_dne_bound %u newest %llu\n",
2434 __func__, lreq, lreq->linger_id, lreq->map_dne_bound,
2436 if (!lreq->map_dne_bound)
2437 lreq->map_dne_bound = greq->u.newest;
2438 erase_linger_mc(&osdc->linger_map_checks, lreq);
2439 check_linger_pool_dne(lreq);
2443 up_write(&osdc->lock);
2446 static void send_linger_map_check(struct ceph_osd_linger_request *lreq)
2448 struct ceph_osd_client *osdc = lreq->osdc;
2449 struct ceph_osd_linger_request *lookup_lreq;
2452 verify_osdc_wrlocked(osdc);
2454 lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks,
2457 WARN_ON(lookup_lreq != lreq);
2462 insert_linger_mc(&osdc->linger_map_checks, lreq);
2463 ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap",
2464 linger_map_check_cb, lreq->linger_id);
2468 static int linger_reg_commit_wait(struct ceph_osd_linger_request *lreq)
2472 dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
2473 ret = wait_for_completion_interruptible(&lreq->reg_commit_wait);
2474 return ret ?: lreq->reg_commit_error;
2477 static int linger_notify_finish_wait(struct ceph_osd_linger_request *lreq)
2481 dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
2482 ret = wait_for_completion_interruptible(&lreq->notify_finish_wait);
2483 return ret ?: lreq->notify_finish_error;
2487 * Timeout callback, called every N seconds. When 1 or more OSD
2488 * requests has been active for more than N seconds, we send a keepalive
2489 * (tag + timestamp) to its OSD to ensure any communications channel
2490 * reset is detected.
2492 static void handle_timeout(struct work_struct *work)
2494 struct ceph_osd_client *osdc =
2495 container_of(work, struct ceph_osd_client, timeout_work.work);
2496 struct ceph_options *opts = osdc->client->options;
2497 unsigned long cutoff = jiffies - opts->osd_keepalive_timeout;
2498 LIST_HEAD(slow_osds);
2499 struct rb_node *n, *p;
2501 dout("%s osdc %p\n", __func__, osdc);
2502 down_write(&osdc->lock);
2505 * ping osds that are a bit slow. this ensures that if there
2506 * is a break in the TCP connection we will notice, and reopen
2507 * a connection with that osd (from the fault callback).
2509 for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
2510 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
2513 for (p = rb_first(&osd->o_requests); p; p = rb_next(p)) {
2514 struct ceph_osd_request *req =
2515 rb_entry(p, struct ceph_osd_request, r_node);
2517 if (time_before(req->r_stamp, cutoff)) {
2518 dout(" req %p tid %llu on osd%d is laggy\n",
2519 req, req->r_tid, osd->o_osd);
2523 for (p = rb_first(&osd->o_linger_requests); p; p = rb_next(p)) {
2524 struct ceph_osd_linger_request *lreq =
2525 rb_entry(p, struct ceph_osd_linger_request, node);
2527 dout(" lreq %p linger_id %llu is served by osd%d\n",
2528 lreq, lreq->linger_id, osd->o_osd);
2531 mutex_lock(&lreq->lock);
2532 if (lreq->is_watch && lreq->committed && !lreq->last_error)
2533 send_linger_ping(lreq);
2534 mutex_unlock(&lreq->lock);
2538 list_move_tail(&osd->o_keepalive_item, &slow_osds);
2541 if (atomic_read(&osdc->num_homeless) || !list_empty(&slow_osds))
2542 maybe_request_map(osdc);
2544 while (!list_empty(&slow_osds)) {
2545 struct ceph_osd *osd = list_first_entry(&slow_osds,
2548 list_del_init(&osd->o_keepalive_item);
2549 ceph_con_keepalive(&osd->o_con);
2552 up_write(&osdc->lock);
2553 schedule_delayed_work(&osdc->timeout_work,
2554 osdc->client->options->osd_keepalive_timeout);
2557 static void handle_osds_timeout(struct work_struct *work)
2559 struct ceph_osd_client *osdc =
2560 container_of(work, struct ceph_osd_client,
2561 osds_timeout_work.work);
2562 unsigned long delay = osdc->client->options->osd_idle_ttl / 4;
2563 struct ceph_osd *osd, *nosd;
2565 dout("%s osdc %p\n", __func__, osdc);
2566 down_write(&osdc->lock);
2567 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
2568 if (time_before(jiffies, osd->lru_ttl))
2571 WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests));
2572 WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests));
2576 up_write(&osdc->lock);
2577 schedule_delayed_work(&osdc->osds_timeout_work,
2578 round_jiffies_relative(delay));
2581 static int ceph_oloc_decode(void **p, void *end,
2582 struct ceph_object_locator *oloc)
2584 u8 struct_v, struct_cv;
2589 ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
2590 struct_v = ceph_decode_8(p);
2591 struct_cv = ceph_decode_8(p);
2593 pr_warn("got v %d < 3 cv %d of ceph_object_locator\n",
2594 struct_v, struct_cv);
2597 if (struct_cv > 6) {
2598 pr_warn("got v %d cv %d > 6 of ceph_object_locator\n",
2599 struct_v, struct_cv);
2602 len = ceph_decode_32(p);
2603 ceph_decode_need(p, end, len, e_inval);
2604 struct_end = *p + len;
2606 oloc->pool = ceph_decode_64(p);
2607 *p += 4; /* skip preferred */
2609 len = ceph_decode_32(p);
2611 pr_warn("ceph_object_locator::key is set\n");
2615 if (struct_v >= 5) {
2616 bool changed = false;
2618 len = ceph_decode_32(p);
2620 ceph_decode_need(p, end, len, e_inval);
2621 if (!oloc->pool_ns ||
2622 ceph_compare_string(oloc->pool_ns, *p, len))
2630 /* redirect changes namespace */
2631 pr_warn("ceph_object_locator::nspace is changed\n");
2636 if (struct_v >= 6) {
2637 s64 hash = ceph_decode_64(p);
2639 pr_warn("ceph_object_locator::hash is set\n");
2654 static int ceph_redirect_decode(void **p, void *end,
2655 struct ceph_request_redirect *redir)
2657 u8 struct_v, struct_cv;
2662 ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
2663 struct_v = ceph_decode_8(p);
2664 struct_cv = ceph_decode_8(p);
2665 if (struct_cv > 1) {
2666 pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n",
2667 struct_v, struct_cv);
2670 len = ceph_decode_32(p);
2671 ceph_decode_need(p, end, len, e_inval);
2672 struct_end = *p + len;
2674 ret = ceph_oloc_decode(p, end, &redir->oloc);
2678 len = ceph_decode_32(p);
2680 pr_warn("ceph_request_redirect::object_name is set\n");
2684 len = ceph_decode_32(p);
2685 *p += len; /* skip osd_instructions */
2697 struct MOSDOpReply {
2698 struct ceph_pg pgid;
2703 u32 outdata_len[CEPH_OSD_MAX_OPS];
2704 s32 rval[CEPH_OSD_MAX_OPS];
2706 struct ceph_eversion replay_version;
2708 struct ceph_request_redirect redirect;
2711 static int decode_MOSDOpReply(const struct ceph_msg *msg, struct MOSDOpReply *m)
2713 void *p = msg->front.iov_base;
2714 void *const end = p + msg->front.iov_len;
2715 u16 version = le16_to_cpu(msg->hdr.version);
2716 struct ceph_eversion bad_replay_version;
2722 ceph_decode_32_safe(&p, end, len, e_inval);
2723 ceph_decode_need(&p, end, len, e_inval);
2724 p += len; /* skip oid */
2726 ret = ceph_decode_pgid(&p, end, &m->pgid);
2730 ceph_decode_64_safe(&p, end, m->flags, e_inval);
2731 ceph_decode_32_safe(&p, end, m->result, e_inval);
2732 ceph_decode_need(&p, end, sizeof(bad_replay_version), e_inval);
2733 memcpy(&bad_replay_version, p, sizeof(bad_replay_version));
2734 p += sizeof(bad_replay_version);
2735 ceph_decode_32_safe(&p, end, m->epoch, e_inval);
2737 ceph_decode_32_safe(&p, end, m->num_ops, e_inval);
2738 if (m->num_ops > ARRAY_SIZE(m->outdata_len))
2741 ceph_decode_need(&p, end, m->num_ops * sizeof(struct ceph_osd_op),
2743 for (i = 0; i < m->num_ops; i++) {
2744 struct ceph_osd_op *op = p;
2746 m->outdata_len[i] = le32_to_cpu(op->payload_len);
2750 ceph_decode_32_safe(&p, end, m->retry_attempt, e_inval);
2751 for (i = 0; i < m->num_ops; i++)
2752 ceph_decode_32_safe(&p, end, m->rval[i], e_inval);
2755 ceph_decode_need(&p, end, sizeof(m->replay_version), e_inval);
2756 memcpy(&m->replay_version, p, sizeof(m->replay_version));
2757 p += sizeof(m->replay_version);
2758 ceph_decode_64_safe(&p, end, m->user_version, e_inval);
2760 m->replay_version = bad_replay_version; /* struct */
2761 m->user_version = le64_to_cpu(m->replay_version.version);
2766 ceph_decode_8_safe(&p, end, decode_redir, e_inval);
2774 ret = ceph_redirect_decode(&p, end, &m->redirect);
2778 ceph_oloc_init(&m->redirect.oloc);
2788 * We are done with @req if
2789 * - @m is a safe reply, or
2790 * - @m is an unsafe reply and we didn't want a safe one
2792 static bool done_request(const struct ceph_osd_request *req,
2793 const struct MOSDOpReply *m)
2795 return (m->result < 0 ||
2796 (m->flags & CEPH_OSD_FLAG_ONDISK) ||
2797 !(req->r_flags & CEPH_OSD_FLAG_ONDISK));
2801 * handle osd op reply. either call the callback if it is specified,
2802 * or do the completion to wake up the waiting thread.
2804 * ->r_unsafe_callback is set? yes no
2806 * first reply is OK (needed r_cb/r_completion, r_cb/r_completion,
2807 * any or needed/got safe) r_done_completion r_done_completion
2809 * first reply is unsafe r_unsafe_cb(true) (nothing)
2811 * when we get the safe reply r_unsafe_cb(false), r_cb/r_completion,
2812 * r_done_completion r_done_completion
2814 static void handle_reply(struct ceph_osd *osd, struct ceph_msg *msg)
2816 struct ceph_osd_client *osdc = osd->o_osdc;
2817 struct ceph_osd_request *req;
2818 struct MOSDOpReply m;
2819 u64 tid = le64_to_cpu(msg->hdr.tid);
2825 dout("%s msg %p tid %llu\n", __func__, msg, tid);
2827 down_read(&osdc->lock);
2828 if (!osd_registered(osd)) {
2829 dout("%s osd%d unknown\n", __func__, osd->o_osd);
2830 goto out_unlock_osdc;
2832 WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num));
2834 mutex_lock(&osd->lock);
2835 req = lookup_request(&osd->o_requests, tid);
2837 dout("%s osd%d tid %llu unknown\n", __func__, osd->o_osd, tid);
2838 goto out_unlock_session;
2841 m.redirect.oloc.pool_ns = req->r_t.target_oloc.pool_ns;
2842 ret = decode_MOSDOpReply(msg, &m);
2843 m.redirect.oloc.pool_ns = NULL;
2845 pr_err("failed to decode MOSDOpReply for tid %llu: %d\n",
2850 dout("%s req %p tid %llu flags 0x%llx pgid %llu.%x epoch %u attempt %d v %u'%llu uv %llu\n",
2851 __func__, req, req->r_tid, m.flags, m.pgid.pool, m.pgid.seed,
2852 m.epoch, m.retry_attempt, le32_to_cpu(m.replay_version.epoch),
2853 le64_to_cpu(m.replay_version.version), m.user_version);
2855 if (m.retry_attempt >= 0) {
2856 if (m.retry_attempt != req->r_attempts - 1) {
2857 dout("req %p tid %llu retry_attempt %d != %d, ignoring\n",
2858 req, req->r_tid, m.retry_attempt,
2859 req->r_attempts - 1);
2860 goto out_unlock_session;
2863 WARN_ON(1); /* MOSDOpReply v4 is assumed */
2866 if (!ceph_oloc_empty(&m.redirect.oloc)) {
2867 dout("req %p tid %llu redirect pool %lld\n", req, req->r_tid,
2868 m.redirect.oloc.pool);
2869 unlink_request(osd, req);
2870 mutex_unlock(&osd->lock);
2873 * Not ceph_oloc_copy() - changing pool_ns is not
2876 req->r_t.target_oloc.pool = m.redirect.oloc.pool;
2877 req->r_flags |= CEPH_OSD_FLAG_REDIRECTED;
2879 __submit_request(req, false);
2880 goto out_unlock_osdc;
2883 if (m.num_ops != req->r_num_ops) {
2884 pr_err("num_ops %d != %d for tid %llu\n", m.num_ops,
2885 req->r_num_ops, req->r_tid);
2888 for (i = 0; i < req->r_num_ops; i++) {
2889 dout(" req %p tid %llu op %d rval %d len %u\n", req,
2890 req->r_tid, i, m.rval[i], m.outdata_len[i]);
2891 req->r_ops[i].rval = m.rval[i];
2892 req->r_ops[i].outdata_len = m.outdata_len[i];
2893 data_len += m.outdata_len[i];
2895 if (data_len != le32_to_cpu(msg->hdr.data_len)) {
2896 pr_err("sum of lens %u != %u for tid %llu\n", data_len,
2897 le32_to_cpu(msg->hdr.data_len), req->r_tid);
2900 dout("%s req %p tid %llu acked %d result %d data_len %u\n", __func__,
2901 req, req->r_tid, req->r_got_reply, m.result, data_len);
2903 already_acked = req->r_got_reply;
2904 if (!already_acked) {
2905 req->r_result = m.result ?: data_len;
2906 req->r_replay_version = m.replay_version; /* struct */
2907 req->r_got_reply = true;
2908 } else if (!(m.flags & CEPH_OSD_FLAG_ONDISK)) {
2909 dout("req %p tid %llu dup ack\n", req, req->r_tid);
2910 goto out_unlock_session;
2913 if (done_request(req, &m)) {
2914 finish_request(req);
2915 if (req->r_linger) {
2916 WARN_ON(req->r_unsafe_callback);
2917 dout("req %p tid %llu cb (locked)\n", req, req->r_tid);
2918 __complete_request(req);
2922 mutex_unlock(&osd->lock);
2923 up_read(&osdc->lock);
2925 if (done_request(req, &m)) {
2926 if (already_acked && req->r_unsafe_callback) {
2927 dout("req %p tid %llu safe-cb\n", req, req->r_tid);
2928 req->r_unsafe_callback(req, false);
2929 } else if (!req->r_linger) {
2930 dout("req %p tid %llu cb\n", req, req->r_tid);
2931 __complete_request(req);
2933 complete_all(&req->r_done_completion);
2934 ceph_osdc_put_request(req);
2936 if (req->r_unsafe_callback) {
2937 dout("req %p tid %llu unsafe-cb\n", req, req->r_tid);
2938 req->r_unsafe_callback(req, true);
2947 complete_request(req, -EIO);
2949 mutex_unlock(&osd->lock);
2951 up_read(&osdc->lock);
2954 static void set_pool_was_full(struct ceph_osd_client *osdc)
2958 for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) {
2959 struct ceph_pg_pool_info *pi =
2960 rb_entry(n, struct ceph_pg_pool_info, node);
2962 pi->was_full = __pool_full(pi);
2966 static bool pool_cleared_full(struct ceph_osd_client *osdc, s64 pool_id)
2968 struct ceph_pg_pool_info *pi;
2970 pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id);
2974 return pi->was_full && !__pool_full(pi);
2977 static enum calc_target_result
2978 recalc_linger_target(struct ceph_osd_linger_request *lreq)
2980 struct ceph_osd_client *osdc = lreq->osdc;
2981 enum calc_target_result ct_res;
2983 ct_res = calc_target(osdc, &lreq->t, &lreq->last_force_resend, true);
2984 if (ct_res == CALC_TARGET_NEED_RESEND) {
2985 struct ceph_osd *osd;
2987 osd = lookup_create_osd(osdc, lreq->t.osd, true);
2988 if (osd != lreq->osd) {
2989 unlink_linger(lreq->osd, lreq);
2990 link_linger(osd, lreq);
2998 * Requeue requests whose mapping to an OSD has changed.
3000 static void scan_requests(struct ceph_osd *osd,
3003 bool check_pool_cleared_full,
3004 struct rb_root *need_resend,
3005 struct list_head *need_resend_linger)
3007 struct ceph_osd_client *osdc = osd->o_osdc;
3009 bool force_resend_writes;
3011 for (n = rb_first(&osd->o_linger_requests); n; ) {
3012 struct ceph_osd_linger_request *lreq =
3013 rb_entry(n, struct ceph_osd_linger_request, node);
3014 enum calc_target_result ct_res;
3016 n = rb_next(n); /* recalc_linger_target() */
3018 dout("%s lreq %p linger_id %llu\n", __func__, lreq,
3020 ct_res = recalc_linger_target(lreq);
3022 case CALC_TARGET_NO_ACTION:
3023 force_resend_writes = cleared_full ||
3024 (check_pool_cleared_full &&
3025 pool_cleared_full(osdc, lreq->t.base_oloc.pool));
3026 if (!force_resend && !force_resend_writes)
3030 case CALC_TARGET_NEED_RESEND:
3031 cancel_linger_map_check(lreq);
3033 * scan_requests() for the previous epoch(s)
3034 * may have already added it to the list, since
3035 * it's not unlinked here.
3037 if (list_empty(&lreq->scan_item))
3038 list_add_tail(&lreq->scan_item, need_resend_linger);
3040 case CALC_TARGET_POOL_DNE:
3041 check_linger_pool_dne(lreq);
3046 for (n = rb_first(&osd->o_requests); n; ) {
3047 struct ceph_osd_request *req =
3048 rb_entry(n, struct ceph_osd_request, r_node);
3049 enum calc_target_result ct_res;
3051 n = rb_next(n); /* unlink_request(), check_pool_dne() */
3053 dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
3054 ct_res = calc_target(osdc, &req->r_t,
3055 &req->r_last_force_resend, false);
3057 case CALC_TARGET_NO_ACTION:
3058 force_resend_writes = cleared_full ||
3059 (check_pool_cleared_full &&
3060 pool_cleared_full(osdc, req->r_t.base_oloc.pool));
3061 if (!force_resend &&
3062 (!(req->r_flags & CEPH_OSD_FLAG_WRITE) ||
3063 !force_resend_writes))
3067 case CALC_TARGET_NEED_RESEND:
3068 cancel_map_check(req);
3069 unlink_request(osd, req);
3070 insert_request(need_resend, req);
3072 case CALC_TARGET_POOL_DNE:
3073 check_pool_dne(req);
3079 static int handle_one_map(struct ceph_osd_client *osdc,
3080 void *p, void *end, bool incremental,
3081 struct rb_root *need_resend,
3082 struct list_head *need_resend_linger)
3084 struct ceph_osdmap *newmap;
3086 bool skipped_map = false;
3089 was_full = ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL);
3090 set_pool_was_full(osdc);
3093 newmap = osdmap_apply_incremental(&p, end, osdc->osdmap);
3095 newmap = ceph_osdmap_decode(&p, end);
3097 return PTR_ERR(newmap);
3099 if (newmap != osdc->osdmap) {
3101 * Preserve ->was_full before destroying the old map.
3102 * For pools that weren't in the old map, ->was_full
3105 for (n = rb_first(&newmap->pg_pools); n; n = rb_next(n)) {
3106 struct ceph_pg_pool_info *pi =
3107 rb_entry(n, struct ceph_pg_pool_info, node);
3108 struct ceph_pg_pool_info *old_pi;
3110 old_pi = ceph_pg_pool_by_id(osdc->osdmap, pi->id);
3112 pi->was_full = old_pi->was_full;
3114 WARN_ON(pi->was_full);
3117 if (osdc->osdmap->epoch &&
3118 osdc->osdmap->epoch + 1 < newmap->epoch) {
3119 WARN_ON(incremental);
3123 ceph_osdmap_destroy(osdc->osdmap);
3124 osdc->osdmap = newmap;
3127 was_full &= !ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL);
3128 scan_requests(&osdc->homeless_osd, skipped_map, was_full, true,
3129 need_resend, need_resend_linger);
3131 for (n = rb_first(&osdc->osds); n; ) {
3132 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
3134 n = rb_next(n); /* close_osd() */
3136 scan_requests(osd, skipped_map, was_full, true, need_resend,
3137 need_resend_linger);
3138 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
3139 memcmp(&osd->o_con.peer_addr,
3140 ceph_osd_addr(osdc->osdmap, osd->o_osd),
3141 sizeof(struct ceph_entity_addr)))
3148 static void kick_requests(struct ceph_osd_client *osdc,
3149 struct rb_root *need_resend,
3150 struct list_head *need_resend_linger)
3152 struct ceph_osd_linger_request *lreq, *nlreq;
3155 for (n = rb_first(need_resend); n; ) {
3156 struct ceph_osd_request *req =
3157 rb_entry(n, struct ceph_osd_request, r_node);
3158 struct ceph_osd *osd;
3161 erase_request(need_resend, req); /* before link_request() */
3163 WARN_ON(req->r_osd);
3164 calc_target(osdc, &req->r_t, NULL, false);
3165 osd = lookup_create_osd(osdc, req->r_t.osd, true);
3166 link_request(osd, req);
3167 if (!req->r_linger) {
3168 if (!osd_homeless(osd) && !req->r_t.paused)
3171 cancel_linger_request(req);
3175 list_for_each_entry_safe(lreq, nlreq, need_resend_linger, scan_item) {
3176 if (!osd_homeless(lreq->osd))
3179 list_del_init(&lreq->scan_item);
3184 * Process updated osd map.
3186 * The message contains any number of incremental and full maps, normally
3187 * indicating some sort of topology change in the cluster. Kick requests
3188 * off to different OSDs as needed.
3190 void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
3192 void *p = msg->front.iov_base;
3193 void *const end = p + msg->front.iov_len;
3194 u32 nr_maps, maplen;
3196 struct ceph_fsid fsid;
3197 struct rb_root need_resend = RB_ROOT;
3198 LIST_HEAD(need_resend_linger);
3199 bool handled_incremental = false;
3200 bool was_pauserd, was_pausewr;
3201 bool pauserd, pausewr;
3204 dout("%s have %u\n", __func__, osdc->osdmap->epoch);
3205 down_write(&osdc->lock);
3208 ceph_decode_need(&p, end, sizeof(fsid), bad);
3209 ceph_decode_copy(&p, &fsid, sizeof(fsid));
3210 if (ceph_check_fsid(osdc->client, &fsid) < 0)
3213 was_pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
3214 was_pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
3215 ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
3216 have_pool_full(osdc);
3218 /* incremental maps */
3219 ceph_decode_32_safe(&p, end, nr_maps, bad);
3220 dout(" %d inc maps\n", nr_maps);
3221 while (nr_maps > 0) {
3222 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
3223 epoch = ceph_decode_32(&p);
3224 maplen = ceph_decode_32(&p);
3225 ceph_decode_need(&p, end, maplen, bad);
3226 if (osdc->osdmap->epoch &&
3227 osdc->osdmap->epoch + 1 == epoch) {
3228 dout("applying incremental map %u len %d\n",
3230 err = handle_one_map(osdc, p, p + maplen, true,
3231 &need_resend, &need_resend_linger);
3234 handled_incremental = true;
3236 dout("ignoring incremental map %u len %d\n",
3242 if (handled_incremental)
3246 ceph_decode_32_safe(&p, end, nr_maps, bad);
3247 dout(" %d full maps\n", nr_maps);
3249 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
3250 epoch = ceph_decode_32(&p);
3251 maplen = ceph_decode_32(&p);
3252 ceph_decode_need(&p, end, maplen, bad);
3254 dout("skipping non-latest full map %u len %d\n",
3256 } else if (osdc->osdmap->epoch >= epoch) {
3257 dout("skipping full map %u len %d, "
3258 "older than our %u\n", epoch, maplen,
3259 osdc->osdmap->epoch);
3261 dout("taking full map %u len %d\n", epoch, maplen);
3262 err = handle_one_map(osdc, p, p + maplen, false,
3263 &need_resend, &need_resend_linger);
3273 * subscribe to subsequent osdmap updates if full to ensure
3274 * we find out when we are no longer full and stop returning
3277 pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
3278 pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
3279 ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
3280 have_pool_full(osdc);
3281 if (was_pauserd || was_pausewr || pauserd || pausewr)
3282 maybe_request_map(osdc);
3284 kick_requests(osdc, &need_resend, &need_resend_linger);
3286 ceph_monc_got_map(&osdc->client->monc, CEPH_SUB_OSDMAP,
3287 osdc->osdmap->epoch);
3288 up_write(&osdc->lock);
3289 wake_up_all(&osdc->client->auth_wq);
3293 pr_err("osdc handle_map corrupt msg\n");
3295 up_write(&osdc->lock);
3299 * Resubmit requests pending on the given osd.
3301 static void kick_osd_requests(struct ceph_osd *osd)
3305 for (n = rb_first(&osd->o_requests); n; ) {
3306 struct ceph_osd_request *req =
3307 rb_entry(n, struct ceph_osd_request, r_node);
3309 n = rb_next(n); /* cancel_linger_request() */
3311 if (!req->r_linger) {
3312 if (!req->r_t.paused)
3315 cancel_linger_request(req);
3318 for (n = rb_first(&osd->o_linger_requests); n; n = rb_next(n)) {
3319 struct ceph_osd_linger_request *lreq =
3320 rb_entry(n, struct ceph_osd_linger_request, node);
3327 * If the osd connection drops, we need to resubmit all requests.
3329 static void osd_fault(struct ceph_connection *con)
3331 struct ceph_osd *osd = con->private;
3332 struct ceph_osd_client *osdc = osd->o_osdc;
3334 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
3336 down_write(&osdc->lock);
3337 if (!osd_registered(osd)) {
3338 dout("%s osd%d unknown\n", __func__, osd->o_osd);
3342 if (!reopen_osd(osd))
3343 kick_osd_requests(osd);
3344 maybe_request_map(osdc);
3347 up_write(&osdc->lock);
3351 * Process osd watch notifications
3353 static void handle_watch_notify(struct ceph_osd_client *osdc,
3354 struct ceph_msg *msg)
3356 void *p = msg->front.iov_base;
3357 void *const end = p + msg->front.iov_len;
3358 struct ceph_osd_linger_request *lreq;
3359 struct linger_work *lwork;
3360 u8 proto_ver, opcode;
3361 u64 cookie, notify_id;
3362 u64 notifier_id = 0;
3363 s32 return_code = 0;
3364 void *payload = NULL;
3365 u32 payload_len = 0;
3367 ceph_decode_8_safe(&p, end, proto_ver, bad);
3368 ceph_decode_8_safe(&p, end, opcode, bad);
3369 ceph_decode_64_safe(&p, end, cookie, bad);
3370 p += 8; /* skip ver */
3371 ceph_decode_64_safe(&p, end, notify_id, bad);
3373 if (proto_ver >= 1) {
3374 ceph_decode_32_safe(&p, end, payload_len, bad);
3375 ceph_decode_need(&p, end, payload_len, bad);
3380 if (le16_to_cpu(msg->hdr.version) >= 2)
3381 ceph_decode_32_safe(&p, end, return_code, bad);
3383 if (le16_to_cpu(msg->hdr.version) >= 3)
3384 ceph_decode_64_safe(&p, end, notifier_id, bad);
3386 down_read(&osdc->lock);
3387 lreq = lookup_linger_osdc(&osdc->linger_requests, cookie);
3389 dout("%s opcode %d cookie %llu dne\n", __func__, opcode,
3391 goto out_unlock_osdc;
3394 mutex_lock(&lreq->lock);
3395 dout("%s opcode %d cookie %llu lreq %p is_watch %d\n", __func__,
3396 opcode, cookie, lreq, lreq->is_watch);
3397 if (opcode == CEPH_WATCH_EVENT_DISCONNECT) {
3398 if (!lreq->last_error) {
3399 lreq->last_error = -ENOTCONN;
3400 queue_watch_error(lreq);
3402 } else if (!lreq->is_watch) {
3403 /* CEPH_WATCH_EVENT_NOTIFY_COMPLETE */
3404 if (lreq->notify_id && lreq->notify_id != notify_id) {
3405 dout("lreq %p notify_id %llu != %llu, ignoring\n", lreq,
3406 lreq->notify_id, notify_id);
3407 } else if (!completion_done(&lreq->notify_finish_wait)) {
3408 struct ceph_msg_data *data =
3409 list_first_entry_or_null(&msg->data,
3410 struct ceph_msg_data,
3414 if (lreq->preply_pages) {
3415 WARN_ON(data->type !=
3416 CEPH_MSG_DATA_PAGES);
3417 *lreq->preply_pages = data->pages;
3418 *lreq->preply_len = data->length;
3420 ceph_release_page_vector(data->pages,
3421 calc_pages_for(0, data->length));
3424 lreq->notify_finish_error = return_code;
3425 complete_all(&lreq->notify_finish_wait);
3428 /* CEPH_WATCH_EVENT_NOTIFY */
3429 lwork = lwork_alloc(lreq, do_watch_notify);
3431 pr_err("failed to allocate notify-lwork\n");
3432 goto out_unlock_lreq;
3435 lwork->notify.notify_id = notify_id;
3436 lwork->notify.notifier_id = notifier_id;
3437 lwork->notify.payload = payload;
3438 lwork->notify.payload_len = payload_len;
3439 lwork->notify.msg = ceph_msg_get(msg);
3444 mutex_unlock(&lreq->lock);
3446 up_read(&osdc->lock);
3450 pr_err("osdc handle_watch_notify corrupt msg\n");
3454 * Register request, send initial attempt.
3456 int ceph_osdc_start_request(struct ceph_osd_client *osdc,
3457 struct ceph_osd_request *req,
3460 down_read(&osdc->lock);
3461 submit_request(req, false);
3462 up_read(&osdc->lock);
3466 EXPORT_SYMBOL(ceph_osdc_start_request);
3469 * Unregister a registered request. The request is not completed:
3470 * ->r_result isn't set and __complete_request() isn't called.
3472 void ceph_osdc_cancel_request(struct ceph_osd_request *req)
3474 struct ceph_osd_client *osdc = req->r_osdc;
3476 down_write(&osdc->lock);
3478 cancel_request(req);
3479 up_write(&osdc->lock);
3481 EXPORT_SYMBOL(ceph_osdc_cancel_request);
3484 * @timeout: in jiffies, 0 means "wait forever"
3486 static int wait_request_timeout(struct ceph_osd_request *req,
3487 unsigned long timeout)
3491 dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
3492 left = wait_for_completion_killable_timeout(&req->r_completion,
3493 ceph_timeout_jiffies(timeout));
3495 left = left ?: -ETIMEDOUT;
3496 ceph_osdc_cancel_request(req);
3498 left = req->r_result; /* completed */
3505 * wait for a request to complete
3507 int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
3508 struct ceph_osd_request *req)
3510 return wait_request_timeout(req, 0);
3512 EXPORT_SYMBOL(ceph_osdc_wait_request);
3515 * sync - wait for all in-flight requests to flush. avoid starvation.
3517 void ceph_osdc_sync(struct ceph_osd_client *osdc)
3519 struct rb_node *n, *p;
3520 u64 last_tid = atomic64_read(&osdc->last_tid);
3523 down_read(&osdc->lock);
3524 for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
3525 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
3527 mutex_lock(&osd->lock);
3528 for (p = rb_first(&osd->o_requests); p; p = rb_next(p)) {
3529 struct ceph_osd_request *req =
3530 rb_entry(p, struct ceph_osd_request, r_node);
3532 if (req->r_tid > last_tid)
3535 if (!(req->r_flags & CEPH_OSD_FLAG_WRITE))
3538 ceph_osdc_get_request(req);
3539 mutex_unlock(&osd->lock);
3540 up_read(&osdc->lock);
3541 dout("%s waiting on req %p tid %llu last_tid %llu\n",
3542 __func__, req, req->r_tid, last_tid);
3543 wait_for_completion(&req->r_done_completion);
3544 ceph_osdc_put_request(req);
3548 mutex_unlock(&osd->lock);
3551 up_read(&osdc->lock);
3552 dout("%s done last_tid %llu\n", __func__, last_tid);
3554 EXPORT_SYMBOL(ceph_osdc_sync);
3556 static struct ceph_osd_request *
3557 alloc_linger_request(struct ceph_osd_linger_request *lreq)
3559 struct ceph_osd_request *req;
3561 req = ceph_osdc_alloc_request(lreq->osdc, NULL, 1, false, GFP_NOIO);
3565 ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
3566 ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
3568 if (ceph_osdc_alloc_messages(req, GFP_NOIO)) {
3569 ceph_osdc_put_request(req);
3577 * Returns a handle, caller owns a ref.
3579 struct ceph_osd_linger_request *
3580 ceph_osdc_watch(struct ceph_osd_client *osdc,
3581 struct ceph_object_id *oid,
3582 struct ceph_object_locator *oloc,
3583 rados_watchcb2_t wcb,
3584 rados_watcherrcb_t errcb,
3587 struct ceph_osd_linger_request *lreq;
3590 lreq = linger_alloc(osdc);
3592 return ERR_PTR(-ENOMEM);
3594 lreq->is_watch = true;
3596 lreq->errcb = errcb;
3598 lreq->watch_valid_thru = jiffies;
3600 ceph_oid_copy(&lreq->t.base_oid, oid);
3601 ceph_oloc_copy(&lreq->t.base_oloc, oloc);
3602 lreq->t.flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
3603 lreq->mtime = CURRENT_TIME;
3605 lreq->reg_req = alloc_linger_request(lreq);
3606 if (!lreq->reg_req) {
3611 lreq->ping_req = alloc_linger_request(lreq);
3612 if (!lreq->ping_req) {
3617 down_write(&osdc->lock);
3618 linger_register(lreq); /* before osd_req_op_* */
3619 osd_req_op_watch_init(lreq->reg_req, 0, lreq->linger_id,
3620 CEPH_OSD_WATCH_OP_WATCH);
3621 osd_req_op_watch_init(lreq->ping_req, 0, lreq->linger_id,
3622 CEPH_OSD_WATCH_OP_PING);
3623 linger_submit(lreq);
3624 up_write(&osdc->lock);
3626 ret = linger_reg_commit_wait(lreq);
3628 linger_cancel(lreq);
3636 return ERR_PTR(ret);
3638 EXPORT_SYMBOL(ceph_osdc_watch);
3643 * Times out after mount_timeout to preserve rbd unmap behaviour
3644 * introduced in 2894e1d76974 ("rbd: timeout watch teardown on unmap
3645 * with mount_timeout").
3647 int ceph_osdc_unwatch(struct ceph_osd_client *osdc,
3648 struct ceph_osd_linger_request *lreq)
3650 struct ceph_options *opts = osdc->client->options;
3651 struct ceph_osd_request *req;
3654 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
3658 ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
3659 ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
3660 req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
3661 req->r_mtime = CURRENT_TIME;
3662 osd_req_op_watch_init(req, 0, lreq->linger_id,
3663 CEPH_OSD_WATCH_OP_UNWATCH);
3665 ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
3669 ceph_osdc_start_request(osdc, req, false);
3670 linger_cancel(lreq);
3672 ret = wait_request_timeout(req, opts->mount_timeout);
3675 ceph_osdc_put_request(req);
3678 EXPORT_SYMBOL(ceph_osdc_unwatch);
3680 static int osd_req_op_notify_ack_init(struct ceph_osd_request *req, int which,
3681 u64 notify_id, u64 cookie, void *payload,
3684 struct ceph_osd_req_op *op;
3685 struct ceph_pagelist *pl;
3688 op = _osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY_ACK, 0);
3690 pl = kmalloc(sizeof(*pl), GFP_NOIO);
3694 ceph_pagelist_init(pl);
3695 ret = ceph_pagelist_encode_64(pl, notify_id);
3696 ret |= ceph_pagelist_encode_64(pl, cookie);
3698 ret |= ceph_pagelist_encode_32(pl, payload_len);
3699 ret |= ceph_pagelist_append(pl, payload, payload_len);
3701 ret |= ceph_pagelist_encode_32(pl, 0);
3704 ceph_pagelist_release(pl);
3708 ceph_osd_data_pagelist_init(&op->notify_ack.request_data, pl);
3709 op->indata_len = pl->length;
3713 int ceph_osdc_notify_ack(struct ceph_osd_client *osdc,
3714 struct ceph_object_id *oid,
3715 struct ceph_object_locator *oloc,
3721 struct ceph_osd_request *req;
3724 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
3728 ceph_oid_copy(&req->r_base_oid, oid);
3729 ceph_oloc_copy(&req->r_base_oloc, oloc);
3730 req->r_flags = CEPH_OSD_FLAG_READ;
3732 ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
3736 ret = osd_req_op_notify_ack_init(req, 0, notify_id, cookie, payload,
3741 ceph_osdc_start_request(osdc, req, false);
3742 ret = ceph_osdc_wait_request(osdc, req);
3745 ceph_osdc_put_request(req);
3748 EXPORT_SYMBOL(ceph_osdc_notify_ack);
3750 static int osd_req_op_notify_init(struct ceph_osd_request *req, int which,
3751 u64 cookie, u32 prot_ver, u32 timeout,
3752 void *payload, size_t payload_len)
3754 struct ceph_osd_req_op *op;
3755 struct ceph_pagelist *pl;
3758 op = _osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY, 0);
3759 op->notify.cookie = cookie;
3761 pl = kmalloc(sizeof(*pl), GFP_NOIO);
3765 ceph_pagelist_init(pl);
3766 ret = ceph_pagelist_encode_32(pl, 1); /* prot_ver */
3767 ret |= ceph_pagelist_encode_32(pl, timeout);
3768 ret |= ceph_pagelist_encode_32(pl, payload_len);
3769 ret |= ceph_pagelist_append(pl, payload, payload_len);
3771 ceph_pagelist_release(pl);
3775 ceph_osd_data_pagelist_init(&op->notify.request_data, pl);
3776 op->indata_len = pl->length;
3781 * @timeout: in seconds
3783 * @preply_{pages,len} are initialized both on success and error.
3784 * The caller is responsible for:
3786 * ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len))
3788 int ceph_osdc_notify(struct ceph_osd_client *osdc,
3789 struct ceph_object_id *oid,
3790 struct ceph_object_locator *oloc,
3794 struct page ***preply_pages,
3797 struct ceph_osd_linger_request *lreq;
3798 struct page **pages;
3803 *preply_pages = NULL;
3807 lreq = linger_alloc(osdc);
3811 lreq->preply_pages = preply_pages;
3812 lreq->preply_len = preply_len;
3814 ceph_oid_copy(&lreq->t.base_oid, oid);
3815 ceph_oloc_copy(&lreq->t.base_oloc, oloc);
3816 lreq->t.flags = CEPH_OSD_FLAG_READ;
3818 lreq->reg_req = alloc_linger_request(lreq);
3819 if (!lreq->reg_req) {
3825 pages = ceph_alloc_page_vector(1, GFP_NOIO);
3826 if (IS_ERR(pages)) {
3827 ret = PTR_ERR(pages);
3831 down_write(&osdc->lock);
3832 linger_register(lreq); /* before osd_req_op_* */
3833 ret = osd_req_op_notify_init(lreq->reg_req, 0, lreq->linger_id, 1,
3834 timeout, payload, payload_len);
3836 linger_unregister(lreq);
3837 up_write(&osdc->lock);
3838 ceph_release_page_vector(pages, 1);
3841 ceph_osd_data_pages_init(osd_req_op_data(lreq->reg_req, 0, notify,
3843 pages, PAGE_SIZE, 0, false, true);
3844 linger_submit(lreq);
3845 up_write(&osdc->lock);
3847 ret = linger_reg_commit_wait(lreq);
3849 ret = linger_notify_finish_wait(lreq);
3851 dout("lreq %p failed to initiate notify %d\n", lreq, ret);
3853 linger_cancel(lreq);
3858 EXPORT_SYMBOL(ceph_osdc_notify);
3861 * Return the number of milliseconds since the watch was last
3862 * confirmed, or an error. If there is an error, the watch is no
3863 * longer valid, and should be destroyed with ceph_osdc_unwatch().
3865 int ceph_osdc_watch_check(struct ceph_osd_client *osdc,
3866 struct ceph_osd_linger_request *lreq)
3868 unsigned long stamp, age;
3871 down_read(&osdc->lock);
3872 mutex_lock(&lreq->lock);
3873 stamp = lreq->watch_valid_thru;
3874 if (!list_empty(&lreq->pending_lworks)) {
3875 struct linger_work *lwork =
3876 list_first_entry(&lreq->pending_lworks,
3880 if (time_before(lwork->queued_stamp, stamp))
3881 stamp = lwork->queued_stamp;
3883 age = jiffies - stamp;
3884 dout("%s lreq %p linger_id %llu age %lu last_error %d\n", __func__,
3885 lreq, lreq->linger_id, age, lreq->last_error);
3886 /* we are truncating to msecs, so return a safe upper bound */
3887 ret = lreq->last_error ?: 1 + jiffies_to_msecs(age);
3889 mutex_unlock(&lreq->lock);
3890 up_read(&osdc->lock);
3894 static int decode_watcher(void **p, void *end, struct ceph_watch_item *item)
3900 ret = ceph_start_decoding(p, end, 2, "watch_item_t",
3901 &struct_v, &struct_len);
3905 ceph_decode_copy(p, &item->name, sizeof(item->name));
3906 item->cookie = ceph_decode_64(p);
3907 *p += 4; /* skip timeout_seconds */
3908 if (struct_v >= 2) {
3909 ceph_decode_copy(p, &item->addr, sizeof(item->addr));
3910 ceph_decode_addr(&item->addr);
3913 dout("%s %s%llu cookie %llu addr %s\n", __func__,
3914 ENTITY_NAME(item->name), item->cookie,
3915 ceph_pr_addr(&item->addr.in_addr));
3919 static int decode_watchers(void **p, void *end,
3920 struct ceph_watch_item **watchers,
3928 ret = ceph_start_decoding(p, end, 1, "obj_list_watch_response_t",
3929 &struct_v, &struct_len);
3933 *num_watchers = ceph_decode_32(p);
3934 *watchers = kcalloc(*num_watchers, sizeof(**watchers), GFP_NOIO);
3938 for (i = 0; i < *num_watchers; i++) {
3939 ret = decode_watcher(p, end, *watchers + i);
3950 * On success, the caller is responsible for:
3954 int ceph_osdc_list_watchers(struct ceph_osd_client *osdc,
3955 struct ceph_object_id *oid,
3956 struct ceph_object_locator *oloc,
3957 struct ceph_watch_item **watchers,
3960 struct ceph_osd_request *req;
3961 struct page **pages;
3964 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
3968 ceph_oid_copy(&req->r_base_oid, oid);
3969 ceph_oloc_copy(&req->r_base_oloc, oloc);
3970 req->r_flags = CEPH_OSD_FLAG_READ;
3972 ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
3976 pages = ceph_alloc_page_vector(1, GFP_NOIO);
3977 if (IS_ERR(pages)) {
3978 ret = PTR_ERR(pages);
3982 osd_req_op_init(req, 0, CEPH_OSD_OP_LIST_WATCHERS, 0);
3983 ceph_osd_data_pages_init(osd_req_op_data(req, 0, list_watchers,
3985 pages, PAGE_SIZE, 0, false, true);
3987 ceph_osdc_start_request(osdc, req, false);
3988 ret = ceph_osdc_wait_request(osdc, req);
3990 void *p = page_address(pages[0]);
3991 void *const end = p + req->r_ops[0].outdata_len;
3993 ret = decode_watchers(&p, end, watchers, num_watchers);
3997 ceph_osdc_put_request(req);
4000 EXPORT_SYMBOL(ceph_osdc_list_watchers);
4003 * Call all pending notify callbacks - for use after a watch is
4004 * unregistered, to make sure no more callbacks for it will be invoked
4006 void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc)
4008 dout("%s osdc %p\n", __func__, osdc);
4009 flush_workqueue(osdc->notify_wq);
4011 EXPORT_SYMBOL(ceph_osdc_flush_notifies);
4013 void ceph_osdc_maybe_request_map(struct ceph_osd_client *osdc)
4015 down_read(&osdc->lock);
4016 maybe_request_map(osdc);
4017 up_read(&osdc->lock);
4019 EXPORT_SYMBOL(ceph_osdc_maybe_request_map);
4022 * Execute an OSD class method on an object.
4024 * @flags: CEPH_OSD_FLAG_*
4025 * @resp_len: out param for reply length
4027 int ceph_osdc_call(struct ceph_osd_client *osdc,
4028 struct ceph_object_id *oid,
4029 struct ceph_object_locator *oloc,
4030 const char *class, const char *method,
4032 struct page *req_page, size_t req_len,
4033 struct page *resp_page, size_t *resp_len)
4035 struct ceph_osd_request *req;
4038 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
4042 ceph_oid_copy(&req->r_base_oid, oid);
4043 ceph_oloc_copy(&req->r_base_oloc, oloc);
4044 req->r_flags = flags;
4046 ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
4050 osd_req_op_cls_init(req, 0, CEPH_OSD_OP_CALL, class, method);
4052 osd_req_op_cls_request_data_pages(req, 0, &req_page, req_len,
4055 osd_req_op_cls_response_data_pages(req, 0, &resp_page,
4056 PAGE_SIZE, 0, false, false);
4058 ceph_osdc_start_request(osdc, req, false);
4059 ret = ceph_osdc_wait_request(osdc, req);
4061 ret = req->r_ops[0].rval;
4063 *resp_len = req->r_ops[0].outdata_len;
4067 ceph_osdc_put_request(req);
4070 EXPORT_SYMBOL(ceph_osdc_call);
4075 int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
4080 osdc->client = client;
4081 init_rwsem(&osdc->lock);
4082 osdc->osds = RB_ROOT;
4083 INIT_LIST_HEAD(&osdc->osd_lru);
4084 spin_lock_init(&osdc->osd_lru_lock);
4085 osd_init(&osdc->homeless_osd);
4086 osdc->homeless_osd.o_osdc = osdc;
4087 osdc->homeless_osd.o_osd = CEPH_HOMELESS_OSD;
4088 osdc->last_linger_id = CEPH_LINGER_ID_START;
4089 osdc->linger_requests = RB_ROOT;
4090 osdc->map_checks = RB_ROOT;
4091 osdc->linger_map_checks = RB_ROOT;
4092 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
4093 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
4096 osdc->osdmap = ceph_osdmap_alloc();
4100 osdc->req_mempool = mempool_create_slab_pool(10,
4101 ceph_osd_request_cache);
4102 if (!osdc->req_mempool)
4105 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
4106 PAGE_SIZE, 10, true, "osd_op");
4109 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
4110 PAGE_SIZE, 10, true, "osd_op_reply");
4115 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
4116 if (!osdc->notify_wq)
4117 goto out_msgpool_reply;
4119 schedule_delayed_work(&osdc->timeout_work,
4120 osdc->client->options->osd_keepalive_timeout);
4121 schedule_delayed_work(&osdc->osds_timeout_work,
4122 round_jiffies_relative(osdc->client->options->osd_idle_ttl));
4127 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
4129 ceph_msgpool_destroy(&osdc->msgpool_op);
4131 mempool_destroy(osdc->req_mempool);
4133 ceph_osdmap_destroy(osdc->osdmap);
4138 void ceph_osdc_stop(struct ceph_osd_client *osdc)
4140 flush_workqueue(osdc->notify_wq);
4141 destroy_workqueue(osdc->notify_wq);
4142 cancel_delayed_work_sync(&osdc->timeout_work);
4143 cancel_delayed_work_sync(&osdc->osds_timeout_work);
4145 down_write(&osdc->lock);
4146 while (!RB_EMPTY_ROOT(&osdc->osds)) {
4147 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
4148 struct ceph_osd, o_node);
4151 up_write(&osdc->lock);
4152 WARN_ON(atomic_read(&osdc->homeless_osd.o_ref) != 1);
4153 osd_cleanup(&osdc->homeless_osd);
4155 WARN_ON(!list_empty(&osdc->osd_lru));
4156 WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_requests));
4157 WARN_ON(!RB_EMPTY_ROOT(&osdc->map_checks));
4158 WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_map_checks));
4159 WARN_ON(atomic_read(&osdc->num_requests));
4160 WARN_ON(atomic_read(&osdc->num_homeless));
4162 ceph_osdmap_destroy(osdc->osdmap);
4163 mempool_destroy(osdc->req_mempool);
4164 ceph_msgpool_destroy(&osdc->msgpool_op);
4165 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
4169 * Read some contiguous pages. If we cross a stripe boundary, shorten
4170 * *plen. Return number of bytes read, or error.
4172 int ceph_osdc_readpages(struct ceph_osd_client *osdc,
4173 struct ceph_vino vino, struct ceph_file_layout *layout,
4175 u32 truncate_seq, u64 truncate_size,
4176 struct page **pages, int num_pages, int page_align)
4178 struct ceph_osd_request *req;
4181 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
4182 vino.snap, off, *plen);
4183 req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 0, 1,
4184 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
4185 NULL, truncate_seq, truncate_size,
4188 return PTR_ERR(req);
4190 /* it may be a short read due to an object boundary */
4191 osd_req_op_extent_osd_data_pages(req, 0,
4192 pages, *plen, page_align, false, false);
4194 dout("readpages final extent is %llu~%llu (%llu bytes align %d)\n",
4195 off, *plen, *plen, page_align);
4197 rc = ceph_osdc_start_request(osdc, req, false);
4199 rc = ceph_osdc_wait_request(osdc, req);
4201 ceph_osdc_put_request(req);
4202 dout("readpages result %d\n", rc);
4205 EXPORT_SYMBOL(ceph_osdc_readpages);
4208 * do a synchronous write on N pages
4210 int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
4211 struct ceph_file_layout *layout,
4212 struct ceph_snap_context *snapc,
4214 u32 truncate_seq, u64 truncate_size,
4215 struct timespec *mtime,
4216 struct page **pages, int num_pages)
4218 struct ceph_osd_request *req;
4220 int page_align = off & ~PAGE_MASK;
4222 req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 0, 1,
4224 CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
4225 snapc, truncate_seq, truncate_size,
4228 return PTR_ERR(req);
4230 /* it may be a short write due to an object boundary */
4231 osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align,
4233 dout("writepages %llu~%llu (%llu bytes)\n", off, len, len);
4235 req->r_mtime = *mtime;
4236 rc = ceph_osdc_start_request(osdc, req, true);
4238 rc = ceph_osdc_wait_request(osdc, req);
4240 ceph_osdc_put_request(req);
4243 dout("writepages result %d\n", rc);
4246 EXPORT_SYMBOL(ceph_osdc_writepages);
4248 int ceph_osdc_setup(void)
4250 size_t size = sizeof(struct ceph_osd_request) +
4251 CEPH_OSD_SLAB_OPS * sizeof(struct ceph_osd_req_op);
4253 BUG_ON(ceph_osd_request_cache);
4254 ceph_osd_request_cache = kmem_cache_create("ceph_osd_request", size,
4257 return ceph_osd_request_cache ? 0 : -ENOMEM;
4259 EXPORT_SYMBOL(ceph_osdc_setup);
4261 void ceph_osdc_cleanup(void)
4263 BUG_ON(!ceph_osd_request_cache);
4264 kmem_cache_destroy(ceph_osd_request_cache);
4265 ceph_osd_request_cache = NULL;
4267 EXPORT_SYMBOL(ceph_osdc_cleanup);
4270 * handle incoming message
4272 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
4274 struct ceph_osd *osd = con->private;
4275 struct ceph_osd_client *osdc = osd->o_osdc;
4276 int type = le16_to_cpu(msg->hdr.type);
4279 case CEPH_MSG_OSD_MAP:
4280 ceph_osdc_handle_map(osdc, msg);
4282 case CEPH_MSG_OSD_OPREPLY:
4283 handle_reply(osd, msg);
4285 case CEPH_MSG_WATCH_NOTIFY:
4286 handle_watch_notify(osdc, msg);
4290 pr_err("received unknown message type %d %s\n", type,
4291 ceph_msg_type_name(type));
4298 * Lookup and return message for incoming reply. Don't try to do
4299 * anything about a larger than preallocated data portion of the
4300 * message at the moment - for now, just skip the message.
4302 static struct ceph_msg *get_reply(struct ceph_connection *con,
4303 struct ceph_msg_header *hdr,
4306 struct ceph_osd *osd = con->private;
4307 struct ceph_osd_client *osdc = osd->o_osdc;
4308 struct ceph_msg *m = NULL;
4309 struct ceph_osd_request *req;
4310 int front_len = le32_to_cpu(hdr->front_len);
4311 int data_len = le32_to_cpu(hdr->data_len);
4312 u64 tid = le64_to_cpu(hdr->tid);
4314 down_read(&osdc->lock);
4315 if (!osd_registered(osd)) {
4316 dout("%s osd%d unknown, skipping\n", __func__, osd->o_osd);
4318 goto out_unlock_osdc;
4320 WARN_ON(osd->o_osd != le64_to_cpu(hdr->src.num));
4322 mutex_lock(&osd->lock);
4323 req = lookup_request(&osd->o_requests, tid);
4325 dout("%s osd%d tid %llu unknown, skipping\n", __func__,
4328 goto out_unlock_session;
4331 ceph_msg_revoke_incoming(req->r_reply);
4333 if (front_len > req->r_reply->front_alloc_len) {
4334 pr_warn("%s osd%d tid %llu front %d > preallocated %d\n",
4335 __func__, osd->o_osd, req->r_tid, front_len,
4336 req->r_reply->front_alloc_len);
4337 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS,
4340 goto out_unlock_session;
4341 ceph_msg_put(req->r_reply);
4345 if (data_len > req->r_reply->data_length) {
4346 pr_warn("%s osd%d tid %llu data %d > preallocated %zu, skipping\n",
4347 __func__, osd->o_osd, req->r_tid, data_len,
4348 req->r_reply->data_length);
4351 goto out_unlock_session;
4354 m = ceph_msg_get(req->r_reply);
4355 dout("get_reply tid %lld %p\n", tid, m);
4358 mutex_unlock(&osd->lock);
4360 up_read(&osdc->lock);
4365 * TODO: switch to a msg-owned pagelist
4367 static struct ceph_msg *alloc_msg_with_page_vector(struct ceph_msg_header *hdr)
4370 int type = le16_to_cpu(hdr->type);
4371 u32 front_len = le32_to_cpu(hdr->front_len);
4372 u32 data_len = le32_to_cpu(hdr->data_len);
4374 m = ceph_msg_new(type, front_len, GFP_NOIO, false);
4379 struct page **pages;
4380 struct ceph_osd_data osd_data;
4382 pages = ceph_alloc_page_vector(calc_pages_for(0, data_len),
4384 if (IS_ERR(pages)) {
4389 ceph_osd_data_pages_init(&osd_data, pages, data_len, 0, false,
4391 ceph_osdc_msg_data_add(m, &osd_data);
4397 static struct ceph_msg *alloc_msg(struct ceph_connection *con,
4398 struct ceph_msg_header *hdr,
4401 struct ceph_osd *osd = con->private;
4402 int type = le16_to_cpu(hdr->type);
4406 case CEPH_MSG_OSD_MAP:
4407 case CEPH_MSG_WATCH_NOTIFY:
4408 return alloc_msg_with_page_vector(hdr);
4409 case CEPH_MSG_OSD_OPREPLY:
4410 return get_reply(con, hdr, skip);
4412 pr_warn("%s osd%d unknown msg type %d, skipping\n", __func__,
4420 * Wrappers to refcount containing ceph_osd struct
4422 static struct ceph_connection *get_osd_con(struct ceph_connection *con)
4424 struct ceph_osd *osd = con->private;
4430 static void put_osd_con(struct ceph_connection *con)
4432 struct ceph_osd *osd = con->private;
4440 * Note: returned pointer is the address of a structure that's
4441 * managed separately. Caller must *not* attempt to free it.
4443 static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
4444 int *proto, int force_new)
4446 struct ceph_osd *o = con->private;
4447 struct ceph_osd_client *osdc = o->o_osdc;
4448 struct ceph_auth_client *ac = osdc->client->monc.auth;
4449 struct ceph_auth_handshake *auth = &o->o_auth;
4451 if (force_new && auth->authorizer) {
4452 ceph_auth_destroy_authorizer(auth->authorizer);
4453 auth->authorizer = NULL;
4455 if (!auth->authorizer) {
4456 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
4459 return ERR_PTR(ret);
4461 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
4464 return ERR_PTR(ret);
4466 *proto = ac->protocol;
4472 static int verify_authorizer_reply(struct ceph_connection *con)
4474 struct ceph_osd *o = con->private;
4475 struct ceph_osd_client *osdc = o->o_osdc;
4476 struct ceph_auth_client *ac = osdc->client->monc.auth;
4478 return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer);
4481 static int invalidate_authorizer(struct ceph_connection *con)
4483 struct ceph_osd *o = con->private;
4484 struct ceph_osd_client *osdc = o->o_osdc;
4485 struct ceph_auth_client *ac = osdc->client->monc.auth;
4487 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
4488 return ceph_monc_validate_auth(&osdc->client->monc);
4491 static int osd_sign_message(struct ceph_msg *msg)
4493 struct ceph_osd *o = msg->con->private;
4494 struct ceph_auth_handshake *auth = &o->o_auth;
4496 return ceph_auth_sign_message(auth, msg);
4499 static int osd_check_message_signature(struct ceph_msg *msg)
4501 struct ceph_osd *o = msg->con->private;
4502 struct ceph_auth_handshake *auth = &o->o_auth;
4504 return ceph_auth_check_message_signature(auth, msg);
4507 static const struct ceph_connection_operations osd_con_ops = {
4510 .dispatch = dispatch,
4511 .get_authorizer = get_authorizer,
4512 .verify_authorizer_reply = verify_authorizer_reply,
4513 .invalidate_authorizer = invalidate_authorizer,
4514 .alloc_msg = alloc_msg,
4515 .sign_message = osd_sign_message,
4516 .check_message_signature = osd_check_message_signature,