x86/mm: Add TLB purge to free pmd/pte page interfaces
[linux/fpc-iii.git] / net / ceph / osd_client.c
blob0ffeb60cfe67d23ea5cad27607b4c477cca901a4
2 #include <linux/ceph/ceph_debug.h>
4 #include <linux/module.h>
5 #include <linux/err.h>
6 #include <linux/highmem.h>
7 #include <linux/mm.h>
8 #include <linux/pagemap.h>
9 #include <linux/slab.h>
10 #include <linux/uaccess.h>
11 #ifdef CONFIG_BLOCK
12 #include <linux/bio.h>
13 #endif
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
35 * access to storage.
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);
53 #if 1
54 static inline bool rwsem_is_wrlocked(struct rw_semaphore *sem)
56 bool wrlocked = true;
58 if (unlikely(down_read_trylock(sem))) {
59 wrlocked = false;
60 up_read(sem);
63 return wrlocked;
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));
85 #else
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) { }
90 #endif
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
95 * object boundary.
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;
103 int r;
105 /* object extent? */
106 r = ceph_calc_file_object_mapping(layout, off, orig_len, objnum,
107 objoff, objlen);
108 if (r < 0)
109 return r;
110 if (*objlen < orig_len) {
111 *plen = *objlen;
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);
118 return 0;
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;
146 #ifdef CONFIG_BLOCK
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;
151 osd_data->bio = bio;
152 osd_data->bio_length = bio_length;
154 #endif /* CONFIG_BLOCK */
156 #define osd_req_op_data(oreq, whch, typ, fld) \
157 ({ \
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,
174 unsigned int which)
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);
216 #ifdef CONFIG_BLOCK
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:
281 return 0;
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;
286 #ifdef CONFIG_BLOCK
287 case CEPH_OSD_DATA_TYPE_BIO:
288 return (u64)osd_data->bio_length;
289 #endif /* CONFIG_BLOCK */
290 default:
291 WARN(true, "unrecognized data type %d\n", (int)osd_data->type);
292 return 0;
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) {
299 int num_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,
309 unsigned int which)
311 struct ceph_osd_req_op *op;
313 BUG_ON(which >= osd_req->r_num_ops);
314 op = &osd_req->r_ops[which];
316 switch (op->op) {
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);
321 break;
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);
326 break;
327 case CEPH_OSD_OP_SETXATTR:
328 case CEPH_OSD_OP_CMPXATTR:
329 ceph_osd_data_release(&op->xattr.osd_data);
330 break;
331 case CEPH_OSD_OP_STAT:
332 ceph_osd_data_release(&op->raw_data_in);
333 break;
334 case CEPH_OSD_OP_NOTIFY_ACK:
335 ceph_osd_data_release(&op->notify_ack.request_data);
336 break;
337 case CEPH_OSD_OP_NOTIFY:
338 ceph_osd_data_release(&op->notify.request_data);
339 ceph_osd_data_release(&op->notify.response_data);
340 break;
341 case CEPH_OSD_OP_LIST_WATCHERS:
342 ceph_osd_data_release(&op->list_watchers.response_data);
343 break;
344 default:
345 break;
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);
361 t->size = -1;
362 t->min_size = -1;
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);
399 * requests
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));
406 WARN_ON(req->r_osd);
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);
413 unsigned int which;
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);
419 if (req->r_request)
420 ceph_msg_put(req->r_request);
421 if (req->r_reply)
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);
430 if (req->r_mempool)
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);
434 else
435 kfree(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)
448 if (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_safe_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);
497 request_init(req);
498 req->r_osdc = osdc;
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,
511 bool use_mempool,
512 gfp_t gfp_flags)
514 struct ceph_osd_request *req;
516 if (use_mempool) {
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);
521 } else {
522 BUG_ON(num_ops > CEPH_OSD_MAX_OPS);
523 req = kmalloc(sizeof(*req) + num_ops * sizeof(req->r_ops[0]),
524 gfp_flags);
526 if (unlikely(!req))
527 return NULL;
529 request_init(req);
530 req->r_osdc = osdc;
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);
537 return 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;
550 int msg_size;
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 */
568 if (req->r_mempool)
569 msg = ceph_msgpool_get(&osdc->msgpool_op, 0);
570 else
571 msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp, true);
572 if (!msg)
573 return -ENOMEM;
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);
583 if (req->r_mempool)
584 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0);
585 else
586 msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, msg_size, gfp, true);
587 if (!msg)
588 return -ENOMEM;
590 req->r_reply = msg;
592 return 0;
594 EXPORT_SYMBOL(ceph_osdc_alloc_messages);
596 static bool osd_req_opcode_valid(u16 opcode)
598 switch (opcode) {
599 #define GENERATE_CASE(op, opcode, str) case CEPH_OSD_OP_##op: return true;
600 __CEPH_FORALL_OSD_OPS(GENERATE_CASE)
601 #undef GENERATE_CASE
602 default:
603 return false;
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));
623 op->op = opcode;
624 op->flags = flags;
626 return 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,
642 opcode, 0);
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;
664 u64 previous;
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 if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL)
676 op->indata_len -= previous - length;
678 EXPORT_SYMBOL(osd_req_op_extent_update);
680 void osd_req_op_extent_dup_last(struct ceph_osd_request *osd_req,
681 unsigned int which, u64 offset_inc)
683 struct ceph_osd_req_op *op, *prev_op;
685 BUG_ON(which + 1 >= osd_req->r_num_ops);
687 prev_op = &osd_req->r_ops[which];
688 op = _osd_req_op_init(osd_req, which + 1, prev_op->op, prev_op->flags);
689 /* dup previous one */
690 op->indata_len = prev_op->indata_len;
691 op->outdata_len = prev_op->outdata_len;
692 op->extent = prev_op->extent;
693 /* adjust offset */
694 op->extent.offset += offset_inc;
695 op->extent.length -= offset_inc;
697 if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL)
698 op->indata_len -= offset_inc;
700 EXPORT_SYMBOL(osd_req_op_extent_dup_last);
702 void osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
703 u16 opcode, const char *class, const char *method)
705 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
706 opcode, 0);
707 struct ceph_pagelist *pagelist;
708 size_t payload_len = 0;
709 size_t size;
711 BUG_ON(opcode != CEPH_OSD_OP_CALL);
713 pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
714 BUG_ON(!pagelist);
715 ceph_pagelist_init(pagelist);
717 op->cls.class_name = class;
718 size = strlen(class);
719 BUG_ON(size > (size_t) U8_MAX);
720 op->cls.class_len = size;
721 ceph_pagelist_append(pagelist, class, size);
722 payload_len += size;
724 op->cls.method_name = method;
725 size = strlen(method);
726 BUG_ON(size > (size_t) U8_MAX);
727 op->cls.method_len = size;
728 ceph_pagelist_append(pagelist, method, size);
729 payload_len += size;
731 osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist);
733 op->indata_len = payload_len;
735 EXPORT_SYMBOL(osd_req_op_cls_init);
737 int osd_req_op_xattr_init(struct ceph_osd_request *osd_req, unsigned int which,
738 u16 opcode, const char *name, const void *value,
739 size_t size, u8 cmp_op, u8 cmp_mode)
741 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
742 opcode, 0);
743 struct ceph_pagelist *pagelist;
744 size_t payload_len;
746 BUG_ON(opcode != CEPH_OSD_OP_SETXATTR && opcode != CEPH_OSD_OP_CMPXATTR);
748 pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
749 if (!pagelist)
750 return -ENOMEM;
752 ceph_pagelist_init(pagelist);
754 payload_len = strlen(name);
755 op->xattr.name_len = payload_len;
756 ceph_pagelist_append(pagelist, name, payload_len);
758 op->xattr.value_len = size;
759 ceph_pagelist_append(pagelist, value, size);
760 payload_len += size;
762 op->xattr.cmp_op = cmp_op;
763 op->xattr.cmp_mode = cmp_mode;
765 ceph_osd_data_pagelist_init(&op->xattr.osd_data, pagelist);
766 op->indata_len = payload_len;
767 return 0;
769 EXPORT_SYMBOL(osd_req_op_xattr_init);
772 * @watch_opcode: CEPH_OSD_WATCH_OP_*
774 static void osd_req_op_watch_init(struct ceph_osd_request *req, int which,
775 u64 cookie, u8 watch_opcode)
777 struct ceph_osd_req_op *op;
779 op = _osd_req_op_init(req, which, CEPH_OSD_OP_WATCH, 0);
780 op->watch.cookie = cookie;
781 op->watch.op = watch_opcode;
782 op->watch.gen = 0;
785 void osd_req_op_alloc_hint_init(struct ceph_osd_request *osd_req,
786 unsigned int which,
787 u64 expected_object_size,
788 u64 expected_write_size)
790 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
791 CEPH_OSD_OP_SETALLOCHINT,
794 op->alloc_hint.expected_object_size = expected_object_size;
795 op->alloc_hint.expected_write_size = expected_write_size;
798 * CEPH_OSD_OP_SETALLOCHINT op is advisory and therefore deemed
799 * not worth a feature bit. Set FAILOK per-op flag to make
800 * sure older osds don't trip over an unsupported opcode.
802 op->flags |= CEPH_OSD_OP_FLAG_FAILOK;
804 EXPORT_SYMBOL(osd_req_op_alloc_hint_init);
806 static void ceph_osdc_msg_data_add(struct ceph_msg *msg,
807 struct ceph_osd_data *osd_data)
809 u64 length = ceph_osd_data_length(osd_data);
811 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
812 BUG_ON(length > (u64) SIZE_MAX);
813 if (length)
814 ceph_msg_data_add_pages(msg, osd_data->pages,
815 length, osd_data->alignment);
816 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
817 BUG_ON(!length);
818 ceph_msg_data_add_pagelist(msg, osd_data->pagelist);
819 #ifdef CONFIG_BLOCK
820 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
821 ceph_msg_data_add_bio(msg, osd_data->bio, length);
822 #endif
823 } else {
824 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
828 static u32 osd_req_encode_op(struct ceph_osd_op *dst,
829 const struct ceph_osd_req_op *src)
831 if (WARN_ON(!osd_req_opcode_valid(src->op))) {
832 pr_err("unrecognized osd opcode %d\n", src->op);
834 return 0;
837 switch (src->op) {
838 case CEPH_OSD_OP_STAT:
839 break;
840 case CEPH_OSD_OP_READ:
841 case CEPH_OSD_OP_WRITE:
842 case CEPH_OSD_OP_WRITEFULL:
843 case CEPH_OSD_OP_ZERO:
844 case CEPH_OSD_OP_TRUNCATE:
845 dst->extent.offset = cpu_to_le64(src->extent.offset);
846 dst->extent.length = cpu_to_le64(src->extent.length);
847 dst->extent.truncate_size =
848 cpu_to_le64(src->extent.truncate_size);
849 dst->extent.truncate_seq =
850 cpu_to_le32(src->extent.truncate_seq);
851 break;
852 case CEPH_OSD_OP_CALL:
853 dst->cls.class_len = src->cls.class_len;
854 dst->cls.method_len = src->cls.method_len;
855 dst->cls.indata_len = cpu_to_le32(src->cls.indata_len);
856 break;
857 case CEPH_OSD_OP_STARTSYNC:
858 break;
859 case CEPH_OSD_OP_WATCH:
860 dst->watch.cookie = cpu_to_le64(src->watch.cookie);
861 dst->watch.ver = cpu_to_le64(0);
862 dst->watch.op = src->watch.op;
863 dst->watch.gen = cpu_to_le32(src->watch.gen);
864 break;
865 case CEPH_OSD_OP_NOTIFY_ACK:
866 break;
867 case CEPH_OSD_OP_NOTIFY:
868 dst->notify.cookie = cpu_to_le64(src->notify.cookie);
869 break;
870 case CEPH_OSD_OP_LIST_WATCHERS:
871 break;
872 case CEPH_OSD_OP_SETALLOCHINT:
873 dst->alloc_hint.expected_object_size =
874 cpu_to_le64(src->alloc_hint.expected_object_size);
875 dst->alloc_hint.expected_write_size =
876 cpu_to_le64(src->alloc_hint.expected_write_size);
877 break;
878 case CEPH_OSD_OP_SETXATTR:
879 case CEPH_OSD_OP_CMPXATTR:
880 dst->xattr.name_len = cpu_to_le32(src->xattr.name_len);
881 dst->xattr.value_len = cpu_to_le32(src->xattr.value_len);
882 dst->xattr.cmp_op = src->xattr.cmp_op;
883 dst->xattr.cmp_mode = src->xattr.cmp_mode;
884 break;
885 case CEPH_OSD_OP_CREATE:
886 case CEPH_OSD_OP_DELETE:
887 break;
888 default:
889 pr_err("unsupported osd opcode %s\n",
890 ceph_osd_op_name(src->op));
891 WARN_ON(1);
893 return 0;
896 dst->op = cpu_to_le16(src->op);
897 dst->flags = cpu_to_le32(src->flags);
898 dst->payload_len = cpu_to_le32(src->indata_len);
900 return src->indata_len;
904 * build new request AND message, calculate layout, and adjust file
905 * extent as needed.
907 * if the file was recently truncated, we include information about its
908 * old and new size so that the object can be updated appropriately. (we
909 * avoid synchronously deleting truncated objects because it's slow.)
911 * if @do_sync, include a 'startsync' command so that the osd will flush
912 * data quickly.
914 struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
915 struct ceph_file_layout *layout,
916 struct ceph_vino vino,
917 u64 off, u64 *plen,
918 unsigned int which, int num_ops,
919 int opcode, int flags,
920 struct ceph_snap_context *snapc,
921 u32 truncate_seq,
922 u64 truncate_size,
923 bool use_mempool)
925 struct ceph_osd_request *req;
926 u64 objnum = 0;
927 u64 objoff = 0;
928 u64 objlen = 0;
929 int r;
931 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
932 opcode != CEPH_OSD_OP_ZERO && opcode != CEPH_OSD_OP_TRUNCATE &&
933 opcode != CEPH_OSD_OP_CREATE && opcode != CEPH_OSD_OP_DELETE);
935 req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool,
936 GFP_NOFS);
937 if (!req) {
938 r = -ENOMEM;
939 goto fail;
942 /* calculate max write size */
943 r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
944 if (r)
945 goto fail;
947 if (opcode == CEPH_OSD_OP_CREATE || opcode == CEPH_OSD_OP_DELETE) {
948 osd_req_op_init(req, which, opcode, 0);
949 } else {
950 u32 object_size = layout->object_size;
951 u32 object_base = off - objoff;
952 if (!(truncate_seq == 1 && truncate_size == -1ULL)) {
953 if (truncate_size <= object_base) {
954 truncate_size = 0;
955 } else {
956 truncate_size -= object_base;
957 if (truncate_size > object_size)
958 truncate_size = object_size;
961 osd_req_op_extent_init(req, which, opcode, objoff, objlen,
962 truncate_size, truncate_seq);
965 req->r_flags = flags;
966 req->r_base_oloc.pool = layout->pool_id;
967 req->r_base_oloc.pool_ns = ceph_try_get_string(layout->pool_ns);
968 ceph_oid_printf(&req->r_base_oid, "%llx.%08llx", vino.ino, objnum);
970 req->r_snapid = vino.snap;
971 if (flags & CEPH_OSD_FLAG_WRITE)
972 req->r_data_offset = off;
974 r = ceph_osdc_alloc_messages(req, GFP_NOFS);
975 if (r)
976 goto fail;
978 return req;
980 fail:
981 ceph_osdc_put_request(req);
982 return ERR_PTR(r);
984 EXPORT_SYMBOL(ceph_osdc_new_request);
987 * We keep osd requests in an rbtree, sorted by ->r_tid.
989 DEFINE_RB_FUNCS(request, struct ceph_osd_request, r_tid, r_node)
990 DEFINE_RB_FUNCS(request_mc, struct ceph_osd_request, r_tid, r_mc_node)
992 static bool osd_homeless(struct ceph_osd *osd)
994 return osd->o_osd == CEPH_HOMELESS_OSD;
997 static bool osd_registered(struct ceph_osd *osd)
999 verify_osdc_locked(osd->o_osdc);
1001 return !RB_EMPTY_NODE(&osd->o_node);
1005 * Assumes @osd is zero-initialized.
1007 static void osd_init(struct ceph_osd *osd)
1009 atomic_set(&osd->o_ref, 1);
1010 RB_CLEAR_NODE(&osd->o_node);
1011 osd->o_requests = RB_ROOT;
1012 osd->o_linger_requests = RB_ROOT;
1013 INIT_LIST_HEAD(&osd->o_osd_lru);
1014 INIT_LIST_HEAD(&osd->o_keepalive_item);
1015 osd->o_incarnation = 1;
1016 mutex_init(&osd->lock);
1019 static void osd_cleanup(struct ceph_osd *osd)
1021 WARN_ON(!RB_EMPTY_NODE(&osd->o_node));
1022 WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests));
1023 WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests));
1024 WARN_ON(!list_empty(&osd->o_osd_lru));
1025 WARN_ON(!list_empty(&osd->o_keepalive_item));
1027 if (osd->o_auth.authorizer) {
1028 WARN_ON(osd_homeless(osd));
1029 ceph_auth_destroy_authorizer(osd->o_auth.authorizer);
1034 * Track open sessions with osds.
1036 static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
1038 struct ceph_osd *osd;
1040 WARN_ON(onum == CEPH_HOMELESS_OSD);
1042 osd = kzalloc(sizeof(*osd), GFP_NOIO | __GFP_NOFAIL);
1043 osd_init(osd);
1044 osd->o_osdc = osdc;
1045 osd->o_osd = onum;
1047 ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
1049 return osd;
1052 static struct ceph_osd *get_osd(struct ceph_osd *osd)
1054 if (atomic_inc_not_zero(&osd->o_ref)) {
1055 dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
1056 atomic_read(&osd->o_ref));
1057 return osd;
1058 } else {
1059 dout("get_osd %p FAIL\n", osd);
1060 return NULL;
1064 static void put_osd(struct ceph_osd *osd)
1066 dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
1067 atomic_read(&osd->o_ref) - 1);
1068 if (atomic_dec_and_test(&osd->o_ref)) {
1069 osd_cleanup(osd);
1070 kfree(osd);
1074 DEFINE_RB_FUNCS(osd, struct ceph_osd, o_osd, o_node)
1076 static void __move_osd_to_lru(struct ceph_osd *osd)
1078 struct ceph_osd_client *osdc = osd->o_osdc;
1080 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1081 BUG_ON(!list_empty(&osd->o_osd_lru));
1083 spin_lock(&osdc->osd_lru_lock);
1084 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
1085 spin_unlock(&osdc->osd_lru_lock);
1087 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl;
1090 static void maybe_move_osd_to_lru(struct ceph_osd *osd)
1092 if (RB_EMPTY_ROOT(&osd->o_requests) &&
1093 RB_EMPTY_ROOT(&osd->o_linger_requests))
1094 __move_osd_to_lru(osd);
1097 static void __remove_osd_from_lru(struct ceph_osd *osd)
1099 struct ceph_osd_client *osdc = osd->o_osdc;
1101 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1103 spin_lock(&osdc->osd_lru_lock);
1104 if (!list_empty(&osd->o_osd_lru))
1105 list_del_init(&osd->o_osd_lru);
1106 spin_unlock(&osdc->osd_lru_lock);
1110 * Close the connection and assign any leftover requests to the
1111 * homeless session.
1113 static void close_osd(struct ceph_osd *osd)
1115 struct ceph_osd_client *osdc = osd->o_osdc;
1116 struct rb_node *n;
1118 verify_osdc_wrlocked(osdc);
1119 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1121 ceph_con_close(&osd->o_con);
1123 for (n = rb_first(&osd->o_requests); n; ) {
1124 struct ceph_osd_request *req =
1125 rb_entry(n, struct ceph_osd_request, r_node);
1127 n = rb_next(n); /* unlink_request() */
1129 dout(" reassigning req %p tid %llu\n", req, req->r_tid);
1130 unlink_request(osd, req);
1131 link_request(&osdc->homeless_osd, req);
1133 for (n = rb_first(&osd->o_linger_requests); n; ) {
1134 struct ceph_osd_linger_request *lreq =
1135 rb_entry(n, struct ceph_osd_linger_request, node);
1137 n = rb_next(n); /* unlink_linger() */
1139 dout(" reassigning lreq %p linger_id %llu\n", lreq,
1140 lreq->linger_id);
1141 unlink_linger(osd, lreq);
1142 link_linger(&osdc->homeless_osd, lreq);
1145 __remove_osd_from_lru(osd);
1146 erase_osd(&osdc->osds, osd);
1147 put_osd(osd);
1151 * reset osd connect
1153 static int reopen_osd(struct ceph_osd *osd)
1155 struct ceph_entity_addr *peer_addr;
1157 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1159 if (RB_EMPTY_ROOT(&osd->o_requests) &&
1160 RB_EMPTY_ROOT(&osd->o_linger_requests)) {
1161 close_osd(osd);
1162 return -ENODEV;
1165 peer_addr = &osd->o_osdc->osdmap->osd_addr[osd->o_osd];
1166 if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1167 !ceph_con_opened(&osd->o_con)) {
1168 struct rb_node *n;
1170 dout("osd addr hasn't changed and connection never opened, "
1171 "letting msgr retry\n");
1172 /* touch each r_stamp for handle_timeout()'s benfit */
1173 for (n = rb_first(&osd->o_requests); n; n = rb_next(n)) {
1174 struct ceph_osd_request *req =
1175 rb_entry(n, struct ceph_osd_request, r_node);
1176 req->r_stamp = jiffies;
1179 return -EAGAIN;
1182 ceph_con_close(&osd->o_con);
1183 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1184 osd->o_incarnation++;
1186 return 0;
1189 static struct ceph_osd *lookup_create_osd(struct ceph_osd_client *osdc, int o,
1190 bool wrlocked)
1192 struct ceph_osd *osd;
1194 if (wrlocked)
1195 verify_osdc_wrlocked(osdc);
1196 else
1197 verify_osdc_locked(osdc);
1199 if (o != CEPH_HOMELESS_OSD)
1200 osd = lookup_osd(&osdc->osds, o);
1201 else
1202 osd = &osdc->homeless_osd;
1203 if (!osd) {
1204 if (!wrlocked)
1205 return ERR_PTR(-EAGAIN);
1207 osd = create_osd(osdc, o);
1208 insert_osd(&osdc->osds, osd);
1209 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd,
1210 &osdc->osdmap->osd_addr[osd->o_osd]);
1213 dout("%s osdc %p osd%d -> osd %p\n", __func__, osdc, o, osd);
1214 return osd;
1218 * Create request <-> OSD session relation.
1220 * @req has to be assigned a tid, @osd may be homeless.
1222 static void link_request(struct ceph_osd *osd, struct ceph_osd_request *req)
1224 verify_osd_locked(osd);
1225 WARN_ON(!req->r_tid || req->r_osd);
1226 dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd,
1227 req, req->r_tid);
1229 if (!osd_homeless(osd))
1230 __remove_osd_from_lru(osd);
1231 else
1232 atomic_inc(&osd->o_osdc->num_homeless);
1234 get_osd(osd);
1235 insert_request(&osd->o_requests, req);
1236 req->r_osd = osd;
1239 static void unlink_request(struct ceph_osd *osd, struct ceph_osd_request *req)
1241 verify_osd_locked(osd);
1242 WARN_ON(req->r_osd != osd);
1243 dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd,
1244 req, req->r_tid);
1246 req->r_osd = NULL;
1247 erase_request(&osd->o_requests, req);
1248 put_osd(osd);
1250 if (!osd_homeless(osd))
1251 maybe_move_osd_to_lru(osd);
1252 else
1253 atomic_dec(&osd->o_osdc->num_homeless);
1256 static bool __pool_full(struct ceph_pg_pool_info *pi)
1258 return pi->flags & CEPH_POOL_FLAG_FULL;
1261 static bool have_pool_full(struct ceph_osd_client *osdc)
1263 struct rb_node *n;
1265 for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) {
1266 struct ceph_pg_pool_info *pi =
1267 rb_entry(n, struct ceph_pg_pool_info, node);
1269 if (__pool_full(pi))
1270 return true;
1273 return false;
1276 static bool pool_full(struct ceph_osd_client *osdc, s64 pool_id)
1278 struct ceph_pg_pool_info *pi;
1280 pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id);
1281 if (!pi)
1282 return false;
1284 return __pool_full(pi);
1288 * Returns whether a request should be blocked from being sent
1289 * based on the current osdmap and osd_client settings.
1291 static bool target_should_be_paused(struct ceph_osd_client *osdc,
1292 const struct ceph_osd_request_target *t,
1293 struct ceph_pg_pool_info *pi)
1295 bool pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
1296 bool pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
1297 ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
1298 __pool_full(pi);
1300 WARN_ON(pi->id != t->base_oloc.pool);
1301 return (t->flags & CEPH_OSD_FLAG_READ && pauserd) ||
1302 (t->flags & CEPH_OSD_FLAG_WRITE && pausewr);
1305 enum calc_target_result {
1306 CALC_TARGET_NO_ACTION = 0,
1307 CALC_TARGET_NEED_RESEND,
1308 CALC_TARGET_POOL_DNE,
1311 static enum calc_target_result calc_target(struct ceph_osd_client *osdc,
1312 struct ceph_osd_request_target *t,
1313 u32 *last_force_resend,
1314 bool any_change)
1316 struct ceph_pg_pool_info *pi;
1317 struct ceph_pg pgid, last_pgid;
1318 struct ceph_osds up, acting;
1319 bool force_resend = false;
1320 bool need_check_tiering = false;
1321 bool need_resend = false;
1322 bool sort_bitwise = ceph_osdmap_flag(osdc, CEPH_OSDMAP_SORTBITWISE);
1323 enum calc_target_result ct_res;
1324 int ret;
1326 pi = ceph_pg_pool_by_id(osdc->osdmap, t->base_oloc.pool);
1327 if (!pi) {
1328 t->osd = CEPH_HOMELESS_OSD;
1329 ct_res = CALC_TARGET_POOL_DNE;
1330 goto out;
1333 if (osdc->osdmap->epoch == pi->last_force_request_resend) {
1334 if (last_force_resend &&
1335 *last_force_resend < pi->last_force_request_resend) {
1336 *last_force_resend = pi->last_force_request_resend;
1337 force_resend = true;
1338 } else if (!last_force_resend) {
1339 force_resend = true;
1342 if (ceph_oid_empty(&t->target_oid) || force_resend) {
1343 ceph_oid_copy(&t->target_oid, &t->base_oid);
1344 need_check_tiering = true;
1346 if (ceph_oloc_empty(&t->target_oloc) || force_resend) {
1347 ceph_oloc_copy(&t->target_oloc, &t->base_oloc);
1348 need_check_tiering = true;
1351 if (need_check_tiering &&
1352 (t->flags & CEPH_OSD_FLAG_IGNORE_OVERLAY) == 0) {
1353 if (t->flags & CEPH_OSD_FLAG_READ && pi->read_tier >= 0)
1354 t->target_oloc.pool = pi->read_tier;
1355 if (t->flags & CEPH_OSD_FLAG_WRITE && pi->write_tier >= 0)
1356 t->target_oloc.pool = pi->write_tier;
1359 ret = ceph_object_locator_to_pg(osdc->osdmap, &t->target_oid,
1360 &t->target_oloc, &pgid);
1361 if (ret) {
1362 WARN_ON(ret != -ENOENT);
1363 t->osd = CEPH_HOMELESS_OSD;
1364 ct_res = CALC_TARGET_POOL_DNE;
1365 goto out;
1367 last_pgid.pool = pgid.pool;
1368 last_pgid.seed = ceph_stable_mod(pgid.seed, t->pg_num, t->pg_num_mask);
1370 ceph_pg_to_up_acting_osds(osdc->osdmap, &pgid, &up, &acting);
1371 if (any_change &&
1372 ceph_is_new_interval(&t->acting,
1373 &acting,
1374 &t->up,
1375 &up,
1376 t->size,
1377 pi->size,
1378 t->min_size,
1379 pi->min_size,
1380 t->pg_num,
1381 pi->pg_num,
1382 t->sort_bitwise,
1383 sort_bitwise,
1384 &last_pgid))
1385 force_resend = true;
1387 if (t->paused && !target_should_be_paused(osdc, t, pi)) {
1388 t->paused = false;
1389 need_resend = true;
1392 if (ceph_pg_compare(&t->pgid, &pgid) ||
1393 ceph_osds_changed(&t->acting, &acting, any_change) ||
1394 force_resend) {
1395 t->pgid = pgid; /* struct */
1396 ceph_osds_copy(&t->acting, &acting);
1397 ceph_osds_copy(&t->up, &up);
1398 t->size = pi->size;
1399 t->min_size = pi->min_size;
1400 t->pg_num = pi->pg_num;
1401 t->pg_num_mask = pi->pg_num_mask;
1402 t->sort_bitwise = sort_bitwise;
1404 t->osd = acting.primary;
1405 need_resend = true;
1408 ct_res = need_resend ? CALC_TARGET_NEED_RESEND : CALC_TARGET_NO_ACTION;
1409 out:
1410 dout("%s t %p -> ct_res %d osd %d\n", __func__, t, ct_res, t->osd);
1411 return ct_res;
1414 static void setup_request_data(struct ceph_osd_request *req,
1415 struct ceph_msg *msg)
1417 u32 data_len = 0;
1418 int i;
1420 if (!list_empty(&msg->data))
1421 return;
1423 WARN_ON(msg->data_length);
1424 for (i = 0; i < req->r_num_ops; i++) {
1425 struct ceph_osd_req_op *op = &req->r_ops[i];
1427 switch (op->op) {
1428 /* request */
1429 case CEPH_OSD_OP_WRITE:
1430 case CEPH_OSD_OP_WRITEFULL:
1431 WARN_ON(op->indata_len != op->extent.length);
1432 ceph_osdc_msg_data_add(msg, &op->extent.osd_data);
1433 break;
1434 case CEPH_OSD_OP_SETXATTR:
1435 case CEPH_OSD_OP_CMPXATTR:
1436 WARN_ON(op->indata_len != op->xattr.name_len +
1437 op->xattr.value_len);
1438 ceph_osdc_msg_data_add(msg, &op->xattr.osd_data);
1439 break;
1440 case CEPH_OSD_OP_NOTIFY_ACK:
1441 ceph_osdc_msg_data_add(msg,
1442 &op->notify_ack.request_data);
1443 break;
1445 /* reply */
1446 case CEPH_OSD_OP_STAT:
1447 ceph_osdc_msg_data_add(req->r_reply,
1448 &op->raw_data_in);
1449 break;
1450 case CEPH_OSD_OP_READ:
1451 ceph_osdc_msg_data_add(req->r_reply,
1452 &op->extent.osd_data);
1453 break;
1454 case CEPH_OSD_OP_LIST_WATCHERS:
1455 ceph_osdc_msg_data_add(req->r_reply,
1456 &op->list_watchers.response_data);
1457 break;
1459 /* both */
1460 case CEPH_OSD_OP_CALL:
1461 WARN_ON(op->indata_len != op->cls.class_len +
1462 op->cls.method_len +
1463 op->cls.indata_len);
1464 ceph_osdc_msg_data_add(msg, &op->cls.request_info);
1465 /* optional, can be NONE */
1466 ceph_osdc_msg_data_add(msg, &op->cls.request_data);
1467 /* optional, can be NONE */
1468 ceph_osdc_msg_data_add(req->r_reply,
1469 &op->cls.response_data);
1470 break;
1471 case CEPH_OSD_OP_NOTIFY:
1472 ceph_osdc_msg_data_add(msg,
1473 &op->notify.request_data);
1474 ceph_osdc_msg_data_add(req->r_reply,
1475 &op->notify.response_data);
1476 break;
1479 data_len += op->indata_len;
1482 WARN_ON(data_len != msg->data_length);
1485 static void encode_request(struct ceph_osd_request *req, struct ceph_msg *msg)
1487 void *p = msg->front.iov_base;
1488 void *const end = p + msg->front_alloc_len;
1489 u32 data_len = 0;
1490 int i;
1492 if (req->r_flags & CEPH_OSD_FLAG_WRITE) {
1493 /* snapshots aren't writeable */
1494 WARN_ON(req->r_snapid != CEPH_NOSNAP);
1495 } else {
1496 WARN_ON(req->r_mtime.tv_sec || req->r_mtime.tv_nsec ||
1497 req->r_data_offset || req->r_snapc);
1500 setup_request_data(req, msg);
1502 ceph_encode_32(&p, 1); /* client_inc, always 1 */
1503 ceph_encode_32(&p, req->r_osdc->osdmap->epoch);
1504 ceph_encode_32(&p, req->r_flags);
1505 ceph_encode_timespec(p, &req->r_mtime);
1506 p += sizeof(struct ceph_timespec);
1507 /* aka reassert_version */
1508 memcpy(p, &req->r_replay_version, sizeof(req->r_replay_version));
1509 p += sizeof(req->r_replay_version);
1511 /* oloc */
1512 ceph_start_encoding(&p, 5, 4,
1513 ceph_oloc_encoding_size(&req->r_t.target_oloc));
1514 ceph_encode_64(&p, req->r_t.target_oloc.pool);
1515 ceph_encode_32(&p, -1); /* preferred */
1516 ceph_encode_32(&p, 0); /* key len */
1517 if (req->r_t.target_oloc.pool_ns)
1518 ceph_encode_string(&p, end, req->r_t.target_oloc.pool_ns->str,
1519 req->r_t.target_oloc.pool_ns->len);
1520 else
1521 ceph_encode_32(&p, 0);
1523 /* pgid */
1524 ceph_encode_8(&p, 1);
1525 ceph_encode_64(&p, req->r_t.pgid.pool);
1526 ceph_encode_32(&p, req->r_t.pgid.seed);
1527 ceph_encode_32(&p, -1); /* preferred */
1529 /* oid */
1530 ceph_encode_32(&p, req->r_t.target_oid.name_len);
1531 memcpy(p, req->r_t.target_oid.name, req->r_t.target_oid.name_len);
1532 p += req->r_t.target_oid.name_len;
1534 /* ops, can imply data */
1535 ceph_encode_16(&p, req->r_num_ops);
1536 for (i = 0; i < req->r_num_ops; i++) {
1537 data_len += osd_req_encode_op(p, &req->r_ops[i]);
1538 p += sizeof(struct ceph_osd_op);
1541 ceph_encode_64(&p, req->r_snapid); /* snapid */
1542 if (req->r_snapc) {
1543 ceph_encode_64(&p, req->r_snapc->seq);
1544 ceph_encode_32(&p, req->r_snapc->num_snaps);
1545 for (i = 0; i < req->r_snapc->num_snaps; i++)
1546 ceph_encode_64(&p, req->r_snapc->snaps[i]);
1547 } else {
1548 ceph_encode_64(&p, 0); /* snap_seq */
1549 ceph_encode_32(&p, 0); /* snaps len */
1552 ceph_encode_32(&p, req->r_attempts); /* retry_attempt */
1554 BUG_ON(p > end);
1555 msg->front.iov_len = p - msg->front.iov_base;
1556 msg->hdr.version = cpu_to_le16(4); /* MOSDOp v4 */
1557 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1558 msg->hdr.data_len = cpu_to_le32(data_len);
1560 * The header "data_off" is a hint to the receiver allowing it
1561 * to align received data into its buffers such that there's no
1562 * need to re-copy it before writing it to disk (direct I/O).
1564 msg->hdr.data_off = cpu_to_le16(req->r_data_offset);
1566 dout("%s req %p oid %s oid_len %d front %zu data %u\n", __func__,
1567 req, req->r_t.target_oid.name, req->r_t.target_oid.name_len,
1568 msg->front.iov_len, data_len);
1572 * @req has to be assigned a tid and registered.
1574 static void send_request(struct ceph_osd_request *req)
1576 struct ceph_osd *osd = req->r_osd;
1578 verify_osd_locked(osd);
1579 WARN_ON(osd->o_osd != req->r_t.osd);
1582 * We may have a previously queued request message hanging
1583 * around. Cancel it to avoid corrupting the msgr.
1585 if (req->r_sent)
1586 ceph_msg_revoke(req->r_request);
1588 req->r_flags |= CEPH_OSD_FLAG_KNOWN_REDIR;
1589 if (req->r_attempts)
1590 req->r_flags |= CEPH_OSD_FLAG_RETRY;
1591 else
1592 WARN_ON(req->r_flags & CEPH_OSD_FLAG_RETRY);
1594 encode_request(req, req->r_request);
1596 dout("%s req %p tid %llu to pg %llu.%x osd%d flags 0x%x attempt %d\n",
1597 __func__, req, req->r_tid, req->r_t.pgid.pool, req->r_t.pgid.seed,
1598 req->r_t.osd, req->r_flags, req->r_attempts);
1600 req->r_t.paused = false;
1601 req->r_stamp = jiffies;
1602 req->r_attempts++;
1604 req->r_sent = osd->o_incarnation;
1605 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
1606 ceph_con_send(&osd->o_con, ceph_msg_get(req->r_request));
1609 static void maybe_request_map(struct ceph_osd_client *osdc)
1611 bool continuous = false;
1613 verify_osdc_locked(osdc);
1614 WARN_ON(!osdc->osdmap->epoch);
1616 if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
1617 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD) ||
1618 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) {
1619 dout("%s osdc %p continuous\n", __func__, osdc);
1620 continuous = true;
1621 } else {
1622 dout("%s osdc %p onetime\n", __func__, osdc);
1625 if (ceph_monc_want_map(&osdc->client->monc, CEPH_SUB_OSDMAP,
1626 osdc->osdmap->epoch + 1, continuous))
1627 ceph_monc_renew_subs(&osdc->client->monc);
1630 static void send_map_check(struct ceph_osd_request *req);
1632 static void __submit_request(struct ceph_osd_request *req, bool wrlocked)
1634 struct ceph_osd_client *osdc = req->r_osdc;
1635 struct ceph_osd *osd;
1636 enum calc_target_result ct_res;
1637 bool need_send = false;
1638 bool promoted = false;
1640 WARN_ON(req->r_tid || req->r_got_reply);
1641 dout("%s req %p wrlocked %d\n", __func__, req, wrlocked);
1643 again:
1644 ct_res = calc_target(osdc, &req->r_t, &req->r_last_force_resend, false);
1645 if (ct_res == CALC_TARGET_POOL_DNE && !wrlocked)
1646 goto promote;
1648 osd = lookup_create_osd(osdc, req->r_t.osd, wrlocked);
1649 if (IS_ERR(osd)) {
1650 WARN_ON(PTR_ERR(osd) != -EAGAIN || wrlocked);
1651 goto promote;
1654 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
1655 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) {
1656 dout("req %p pausewr\n", req);
1657 req->r_t.paused = true;
1658 maybe_request_map(osdc);
1659 } else if ((req->r_flags & CEPH_OSD_FLAG_READ) &&
1660 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) {
1661 dout("req %p pauserd\n", req);
1662 req->r_t.paused = true;
1663 maybe_request_map(osdc);
1664 } else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
1665 !(req->r_flags & (CEPH_OSD_FLAG_FULL_TRY |
1666 CEPH_OSD_FLAG_FULL_FORCE)) &&
1667 (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
1668 pool_full(osdc, req->r_t.base_oloc.pool))) {
1669 dout("req %p full/pool_full\n", req);
1670 pr_warn_ratelimited("FULL or reached pool quota\n");
1671 req->r_t.paused = true;
1672 maybe_request_map(osdc);
1673 } else if (!osd_homeless(osd)) {
1674 need_send = true;
1675 } else {
1676 maybe_request_map(osdc);
1679 mutex_lock(&osd->lock);
1681 * Assign the tid atomically with send_request() to protect
1682 * multiple writes to the same object from racing with each
1683 * other, resulting in out of order ops on the OSDs.
1685 req->r_tid = atomic64_inc_return(&osdc->last_tid);
1686 link_request(osd, req);
1687 if (need_send)
1688 send_request(req);
1689 mutex_unlock(&osd->lock);
1691 if (ct_res == CALC_TARGET_POOL_DNE)
1692 send_map_check(req);
1694 if (promoted)
1695 downgrade_write(&osdc->lock);
1696 return;
1698 promote:
1699 up_read(&osdc->lock);
1700 down_write(&osdc->lock);
1701 wrlocked = true;
1702 promoted = true;
1703 goto again;
1706 static void account_request(struct ceph_osd_request *req)
1708 unsigned int mask = CEPH_OSD_FLAG_ACK | CEPH_OSD_FLAG_ONDISK;
1710 if (req->r_flags & CEPH_OSD_FLAG_READ) {
1711 WARN_ON(req->r_flags & mask);
1712 req->r_flags |= CEPH_OSD_FLAG_ACK;
1713 } else if (req->r_flags & CEPH_OSD_FLAG_WRITE)
1714 WARN_ON(!(req->r_flags & mask));
1715 else
1716 WARN_ON(1);
1718 WARN_ON(req->r_unsafe_callback && (req->r_flags & mask) != mask);
1719 atomic_inc(&req->r_osdc->num_requests);
1722 static void submit_request(struct ceph_osd_request *req, bool wrlocked)
1724 ceph_osdc_get_request(req);
1725 account_request(req);
1726 __submit_request(req, wrlocked);
1729 static void __finish_request(struct ceph_osd_request *req)
1731 struct ceph_osd_client *osdc = req->r_osdc;
1732 struct ceph_osd *osd = req->r_osd;
1734 verify_osd_locked(osd);
1735 dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
1737 WARN_ON(lookup_request_mc(&osdc->map_checks, req->r_tid));
1738 unlink_request(osd, req);
1739 atomic_dec(&osdc->num_requests);
1742 * If an OSD has failed or returned and a request has been sent
1743 * twice, it's possible to get a reply and end up here while the
1744 * request message is queued for delivery. We will ignore the
1745 * reply, so not a big deal, but better to try and catch it.
1747 ceph_msg_revoke(req->r_request);
1748 ceph_msg_revoke_incoming(req->r_reply);
1751 static void finish_request(struct ceph_osd_request *req)
1753 __finish_request(req);
1754 ceph_osdc_put_request(req);
1757 static void __complete_request(struct ceph_osd_request *req)
1759 if (req->r_callback)
1760 req->r_callback(req);
1761 else
1762 complete_all(&req->r_completion);
1766 * Note that this is open-coded in handle_reply(), which has to deal
1767 * with ack vs commit, dup acks, etc.
1769 static void complete_request(struct ceph_osd_request *req, int err)
1771 dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err);
1773 req->r_result = err;
1774 __finish_request(req);
1775 __complete_request(req);
1776 complete_all(&req->r_safe_completion);
1777 ceph_osdc_put_request(req);
1780 static void cancel_map_check(struct ceph_osd_request *req)
1782 struct ceph_osd_client *osdc = req->r_osdc;
1783 struct ceph_osd_request *lookup_req;
1785 verify_osdc_wrlocked(osdc);
1787 lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid);
1788 if (!lookup_req)
1789 return;
1791 WARN_ON(lookup_req != req);
1792 erase_request_mc(&osdc->map_checks, req);
1793 ceph_osdc_put_request(req);
1796 static void cancel_request(struct ceph_osd_request *req)
1798 dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
1800 cancel_map_check(req);
1801 finish_request(req);
1804 static void check_pool_dne(struct ceph_osd_request *req)
1806 struct ceph_osd_client *osdc = req->r_osdc;
1807 struct ceph_osdmap *map = osdc->osdmap;
1809 verify_osdc_wrlocked(osdc);
1810 WARN_ON(!map->epoch);
1812 if (req->r_attempts) {
1814 * We sent a request earlier, which means that
1815 * previously the pool existed, and now it does not
1816 * (i.e., it was deleted).
1818 req->r_map_dne_bound = map->epoch;
1819 dout("%s req %p tid %llu pool disappeared\n", __func__, req,
1820 req->r_tid);
1821 } else {
1822 dout("%s req %p tid %llu map_dne_bound %u have %u\n", __func__,
1823 req, req->r_tid, req->r_map_dne_bound, map->epoch);
1826 if (req->r_map_dne_bound) {
1827 if (map->epoch >= req->r_map_dne_bound) {
1828 /* we had a new enough map */
1829 pr_info_ratelimited("tid %llu pool does not exist\n",
1830 req->r_tid);
1831 complete_request(req, -ENOENT);
1833 } else {
1834 send_map_check(req);
1838 static void map_check_cb(struct ceph_mon_generic_request *greq)
1840 struct ceph_osd_client *osdc = &greq->monc->client->osdc;
1841 struct ceph_osd_request *req;
1842 u64 tid = greq->private_data;
1844 WARN_ON(greq->result || !greq->u.newest);
1846 down_write(&osdc->lock);
1847 req = lookup_request_mc(&osdc->map_checks, tid);
1848 if (!req) {
1849 dout("%s tid %llu dne\n", __func__, tid);
1850 goto out_unlock;
1853 dout("%s req %p tid %llu map_dne_bound %u newest %llu\n", __func__,
1854 req, req->r_tid, req->r_map_dne_bound, greq->u.newest);
1855 if (!req->r_map_dne_bound)
1856 req->r_map_dne_bound = greq->u.newest;
1857 erase_request_mc(&osdc->map_checks, req);
1858 check_pool_dne(req);
1860 ceph_osdc_put_request(req);
1861 out_unlock:
1862 up_write(&osdc->lock);
1865 static void send_map_check(struct ceph_osd_request *req)
1867 struct ceph_osd_client *osdc = req->r_osdc;
1868 struct ceph_osd_request *lookup_req;
1869 int ret;
1871 verify_osdc_wrlocked(osdc);
1873 lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid);
1874 if (lookup_req) {
1875 WARN_ON(lookup_req != req);
1876 return;
1879 ceph_osdc_get_request(req);
1880 insert_request_mc(&osdc->map_checks, req);
1881 ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap",
1882 map_check_cb, req->r_tid);
1883 WARN_ON(ret);
1887 * lingering requests, watch/notify v2 infrastructure
1889 static void linger_release(struct kref *kref)
1891 struct ceph_osd_linger_request *lreq =
1892 container_of(kref, struct ceph_osd_linger_request, kref);
1894 dout("%s lreq %p reg_req %p ping_req %p\n", __func__, lreq,
1895 lreq->reg_req, lreq->ping_req);
1896 WARN_ON(!RB_EMPTY_NODE(&lreq->node));
1897 WARN_ON(!RB_EMPTY_NODE(&lreq->osdc_node));
1898 WARN_ON(!RB_EMPTY_NODE(&lreq->mc_node));
1899 WARN_ON(!list_empty(&lreq->scan_item));
1900 WARN_ON(!list_empty(&lreq->pending_lworks));
1901 WARN_ON(lreq->osd);
1903 if (lreq->reg_req)
1904 ceph_osdc_put_request(lreq->reg_req);
1905 if (lreq->ping_req)
1906 ceph_osdc_put_request(lreq->ping_req);
1907 target_destroy(&lreq->t);
1908 kfree(lreq);
1911 static void linger_put(struct ceph_osd_linger_request *lreq)
1913 if (lreq)
1914 kref_put(&lreq->kref, linger_release);
1917 static struct ceph_osd_linger_request *
1918 linger_get(struct ceph_osd_linger_request *lreq)
1920 kref_get(&lreq->kref);
1921 return lreq;
1924 static struct ceph_osd_linger_request *
1925 linger_alloc(struct ceph_osd_client *osdc)
1927 struct ceph_osd_linger_request *lreq;
1929 lreq = kzalloc(sizeof(*lreq), GFP_NOIO);
1930 if (!lreq)
1931 return NULL;
1933 kref_init(&lreq->kref);
1934 mutex_init(&lreq->lock);
1935 RB_CLEAR_NODE(&lreq->node);
1936 RB_CLEAR_NODE(&lreq->osdc_node);
1937 RB_CLEAR_NODE(&lreq->mc_node);
1938 INIT_LIST_HEAD(&lreq->scan_item);
1939 INIT_LIST_HEAD(&lreq->pending_lworks);
1940 init_completion(&lreq->reg_commit_wait);
1941 init_completion(&lreq->notify_finish_wait);
1943 lreq->osdc = osdc;
1944 target_init(&lreq->t);
1946 dout("%s lreq %p\n", __func__, lreq);
1947 return lreq;
1950 DEFINE_RB_INSDEL_FUNCS(linger, struct ceph_osd_linger_request, linger_id, node)
1951 DEFINE_RB_FUNCS(linger_osdc, struct ceph_osd_linger_request, linger_id, osdc_node)
1952 DEFINE_RB_FUNCS(linger_mc, struct ceph_osd_linger_request, linger_id, mc_node)
1955 * Create linger request <-> OSD session relation.
1957 * @lreq has to be registered, @osd may be homeless.
1959 static void link_linger(struct ceph_osd *osd,
1960 struct ceph_osd_linger_request *lreq)
1962 verify_osd_locked(osd);
1963 WARN_ON(!lreq->linger_id || lreq->osd);
1964 dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd,
1965 osd->o_osd, lreq, lreq->linger_id);
1967 if (!osd_homeless(osd))
1968 __remove_osd_from_lru(osd);
1969 else
1970 atomic_inc(&osd->o_osdc->num_homeless);
1972 get_osd(osd);
1973 insert_linger(&osd->o_linger_requests, lreq);
1974 lreq->osd = osd;
1977 static void unlink_linger(struct ceph_osd *osd,
1978 struct ceph_osd_linger_request *lreq)
1980 verify_osd_locked(osd);
1981 WARN_ON(lreq->osd != osd);
1982 dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd,
1983 osd->o_osd, lreq, lreq->linger_id);
1985 lreq->osd = NULL;
1986 erase_linger(&osd->o_linger_requests, lreq);
1987 put_osd(osd);
1989 if (!osd_homeless(osd))
1990 maybe_move_osd_to_lru(osd);
1991 else
1992 atomic_dec(&osd->o_osdc->num_homeless);
1995 static bool __linger_registered(struct ceph_osd_linger_request *lreq)
1997 verify_osdc_locked(lreq->osdc);
1999 return !RB_EMPTY_NODE(&lreq->osdc_node);
2002 static bool linger_registered(struct ceph_osd_linger_request *lreq)
2004 struct ceph_osd_client *osdc = lreq->osdc;
2005 bool registered;
2007 down_read(&osdc->lock);
2008 registered = __linger_registered(lreq);
2009 up_read(&osdc->lock);
2011 return registered;
2014 static void linger_register(struct ceph_osd_linger_request *lreq)
2016 struct ceph_osd_client *osdc = lreq->osdc;
2018 verify_osdc_wrlocked(osdc);
2019 WARN_ON(lreq->linger_id);
2021 linger_get(lreq);
2022 lreq->linger_id = ++osdc->last_linger_id;
2023 insert_linger_osdc(&osdc->linger_requests, lreq);
2026 static void linger_unregister(struct ceph_osd_linger_request *lreq)
2028 struct ceph_osd_client *osdc = lreq->osdc;
2030 verify_osdc_wrlocked(osdc);
2032 erase_linger_osdc(&osdc->linger_requests, lreq);
2033 linger_put(lreq);
2036 static void cancel_linger_request(struct ceph_osd_request *req)
2038 struct ceph_osd_linger_request *lreq = req->r_priv;
2040 WARN_ON(!req->r_linger);
2041 cancel_request(req);
2042 linger_put(lreq);
2045 struct linger_work {
2046 struct work_struct work;
2047 struct ceph_osd_linger_request *lreq;
2048 struct list_head pending_item;
2049 unsigned long queued_stamp;
2051 union {
2052 struct {
2053 u64 notify_id;
2054 u64 notifier_id;
2055 void *payload; /* points into @msg front */
2056 size_t payload_len;
2058 struct ceph_msg *msg; /* for ceph_msg_put() */
2059 } notify;
2060 struct {
2061 int err;
2062 } error;
2066 static struct linger_work *lwork_alloc(struct ceph_osd_linger_request *lreq,
2067 work_func_t workfn)
2069 struct linger_work *lwork;
2071 lwork = kzalloc(sizeof(*lwork), GFP_NOIO);
2072 if (!lwork)
2073 return NULL;
2075 INIT_WORK(&lwork->work, workfn);
2076 INIT_LIST_HEAD(&lwork->pending_item);
2077 lwork->lreq = linger_get(lreq);
2079 return lwork;
2082 static void lwork_free(struct linger_work *lwork)
2084 struct ceph_osd_linger_request *lreq = lwork->lreq;
2086 mutex_lock(&lreq->lock);
2087 list_del(&lwork->pending_item);
2088 mutex_unlock(&lreq->lock);
2090 linger_put(lreq);
2091 kfree(lwork);
2094 static void lwork_queue(struct linger_work *lwork)
2096 struct ceph_osd_linger_request *lreq = lwork->lreq;
2097 struct ceph_osd_client *osdc = lreq->osdc;
2099 verify_lreq_locked(lreq);
2100 WARN_ON(!list_empty(&lwork->pending_item));
2102 lwork->queued_stamp = jiffies;
2103 list_add_tail(&lwork->pending_item, &lreq->pending_lworks);
2104 queue_work(osdc->notify_wq, &lwork->work);
2107 static void do_watch_notify(struct work_struct *w)
2109 struct linger_work *lwork = container_of(w, struct linger_work, work);
2110 struct ceph_osd_linger_request *lreq = lwork->lreq;
2112 if (!linger_registered(lreq)) {
2113 dout("%s lreq %p not registered\n", __func__, lreq);
2114 goto out;
2117 WARN_ON(!lreq->is_watch);
2118 dout("%s lreq %p notify_id %llu notifier_id %llu payload_len %zu\n",
2119 __func__, lreq, lwork->notify.notify_id, lwork->notify.notifier_id,
2120 lwork->notify.payload_len);
2121 lreq->wcb(lreq->data, lwork->notify.notify_id, lreq->linger_id,
2122 lwork->notify.notifier_id, lwork->notify.payload,
2123 lwork->notify.payload_len);
2125 out:
2126 ceph_msg_put(lwork->notify.msg);
2127 lwork_free(lwork);
2130 static void do_watch_error(struct work_struct *w)
2132 struct linger_work *lwork = container_of(w, struct linger_work, work);
2133 struct ceph_osd_linger_request *lreq = lwork->lreq;
2135 if (!linger_registered(lreq)) {
2136 dout("%s lreq %p not registered\n", __func__, lreq);
2137 goto out;
2140 dout("%s lreq %p err %d\n", __func__, lreq, lwork->error.err);
2141 lreq->errcb(lreq->data, lreq->linger_id, lwork->error.err);
2143 out:
2144 lwork_free(lwork);
2147 static void queue_watch_error(struct ceph_osd_linger_request *lreq)
2149 struct linger_work *lwork;
2151 lwork = lwork_alloc(lreq, do_watch_error);
2152 if (!lwork) {
2153 pr_err("failed to allocate error-lwork\n");
2154 return;
2157 lwork->error.err = lreq->last_error;
2158 lwork_queue(lwork);
2161 static void linger_reg_commit_complete(struct ceph_osd_linger_request *lreq,
2162 int result)
2164 if (!completion_done(&lreq->reg_commit_wait)) {
2165 lreq->reg_commit_error = (result <= 0 ? result : 0);
2166 complete_all(&lreq->reg_commit_wait);
2170 static void linger_commit_cb(struct ceph_osd_request *req)
2172 struct ceph_osd_linger_request *lreq = req->r_priv;
2174 mutex_lock(&lreq->lock);
2175 dout("%s lreq %p linger_id %llu result %d\n", __func__, lreq,
2176 lreq->linger_id, req->r_result);
2177 WARN_ON(!__linger_registered(lreq));
2178 linger_reg_commit_complete(lreq, req->r_result);
2179 lreq->committed = true;
2181 if (!lreq->is_watch) {
2182 struct ceph_osd_data *osd_data =
2183 osd_req_op_data(req, 0, notify, response_data);
2184 void *p = page_address(osd_data->pages[0]);
2186 WARN_ON(req->r_ops[0].op != CEPH_OSD_OP_NOTIFY ||
2187 osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
2189 /* make note of the notify_id */
2190 if (req->r_ops[0].outdata_len >= sizeof(u64)) {
2191 lreq->notify_id = ceph_decode_64(&p);
2192 dout("lreq %p notify_id %llu\n", lreq,
2193 lreq->notify_id);
2194 } else {
2195 dout("lreq %p no notify_id\n", lreq);
2199 mutex_unlock(&lreq->lock);
2200 linger_put(lreq);
2203 static int normalize_watch_error(int err)
2206 * Translate ENOENT -> ENOTCONN so that a delete->disconnection
2207 * notification and a failure to reconnect because we raced with
2208 * the delete appear the same to the user.
2210 if (err == -ENOENT)
2211 err = -ENOTCONN;
2213 return err;
2216 static void linger_reconnect_cb(struct ceph_osd_request *req)
2218 struct ceph_osd_linger_request *lreq = req->r_priv;
2220 mutex_lock(&lreq->lock);
2221 dout("%s lreq %p linger_id %llu result %d last_error %d\n", __func__,
2222 lreq, lreq->linger_id, req->r_result, lreq->last_error);
2223 if (req->r_result < 0) {
2224 if (!lreq->last_error) {
2225 lreq->last_error = normalize_watch_error(req->r_result);
2226 queue_watch_error(lreq);
2230 mutex_unlock(&lreq->lock);
2231 linger_put(lreq);
2234 static void send_linger(struct ceph_osd_linger_request *lreq)
2236 struct ceph_osd_request *req = lreq->reg_req;
2237 struct ceph_osd_req_op *op = &req->r_ops[0];
2239 verify_osdc_wrlocked(req->r_osdc);
2240 dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
2242 if (req->r_osd)
2243 cancel_linger_request(req);
2245 request_reinit(req);
2246 ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
2247 ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
2248 req->r_flags = lreq->t.flags;
2249 req->r_mtime = lreq->mtime;
2251 mutex_lock(&lreq->lock);
2252 if (lreq->is_watch && lreq->committed) {
2253 WARN_ON(op->op != CEPH_OSD_OP_WATCH ||
2254 op->watch.cookie != lreq->linger_id);
2255 op->watch.op = CEPH_OSD_WATCH_OP_RECONNECT;
2256 op->watch.gen = ++lreq->register_gen;
2257 dout("lreq %p reconnect register_gen %u\n", lreq,
2258 op->watch.gen);
2259 req->r_callback = linger_reconnect_cb;
2260 } else {
2261 if (!lreq->is_watch)
2262 lreq->notify_id = 0;
2263 else
2264 WARN_ON(op->watch.op != CEPH_OSD_WATCH_OP_WATCH);
2265 dout("lreq %p register\n", lreq);
2266 req->r_callback = linger_commit_cb;
2268 mutex_unlock(&lreq->lock);
2270 req->r_priv = linger_get(lreq);
2271 req->r_linger = true;
2273 submit_request(req, true);
2276 static void linger_ping_cb(struct ceph_osd_request *req)
2278 struct ceph_osd_linger_request *lreq = req->r_priv;
2280 mutex_lock(&lreq->lock);
2281 dout("%s lreq %p linger_id %llu result %d ping_sent %lu last_error %d\n",
2282 __func__, lreq, lreq->linger_id, req->r_result, lreq->ping_sent,
2283 lreq->last_error);
2284 if (lreq->register_gen == req->r_ops[0].watch.gen) {
2285 if (!req->r_result) {
2286 lreq->watch_valid_thru = lreq->ping_sent;
2287 } else if (!lreq->last_error) {
2288 lreq->last_error = normalize_watch_error(req->r_result);
2289 queue_watch_error(lreq);
2291 } else {
2292 dout("lreq %p register_gen %u ignoring old pong %u\n", lreq,
2293 lreq->register_gen, req->r_ops[0].watch.gen);
2296 mutex_unlock(&lreq->lock);
2297 linger_put(lreq);
2300 static void send_linger_ping(struct ceph_osd_linger_request *lreq)
2302 struct ceph_osd_client *osdc = lreq->osdc;
2303 struct ceph_osd_request *req = lreq->ping_req;
2304 struct ceph_osd_req_op *op = &req->r_ops[0];
2306 if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) {
2307 dout("%s PAUSERD\n", __func__);
2308 return;
2311 lreq->ping_sent = jiffies;
2312 dout("%s lreq %p linger_id %llu ping_sent %lu register_gen %u\n",
2313 __func__, lreq, lreq->linger_id, lreq->ping_sent,
2314 lreq->register_gen);
2316 if (req->r_osd)
2317 cancel_linger_request(req);
2319 request_reinit(req);
2320 target_copy(&req->r_t, &lreq->t);
2322 WARN_ON(op->op != CEPH_OSD_OP_WATCH ||
2323 op->watch.cookie != lreq->linger_id ||
2324 op->watch.op != CEPH_OSD_WATCH_OP_PING);
2325 op->watch.gen = lreq->register_gen;
2326 req->r_callback = linger_ping_cb;
2327 req->r_priv = linger_get(lreq);
2328 req->r_linger = true;
2330 ceph_osdc_get_request(req);
2331 account_request(req);
2332 req->r_tid = atomic64_inc_return(&osdc->last_tid);
2333 link_request(lreq->osd, req);
2334 send_request(req);
2337 static void linger_submit(struct ceph_osd_linger_request *lreq)
2339 struct ceph_osd_client *osdc = lreq->osdc;
2340 struct ceph_osd *osd;
2342 calc_target(osdc, &lreq->t, &lreq->last_force_resend, false);
2343 osd = lookup_create_osd(osdc, lreq->t.osd, true);
2344 link_linger(osd, lreq);
2346 send_linger(lreq);
2349 static void cancel_linger_map_check(struct ceph_osd_linger_request *lreq)
2351 struct ceph_osd_client *osdc = lreq->osdc;
2352 struct ceph_osd_linger_request *lookup_lreq;
2354 verify_osdc_wrlocked(osdc);
2356 lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks,
2357 lreq->linger_id);
2358 if (!lookup_lreq)
2359 return;
2361 WARN_ON(lookup_lreq != lreq);
2362 erase_linger_mc(&osdc->linger_map_checks, lreq);
2363 linger_put(lreq);
2367 * @lreq has to be both registered and linked.
2369 static void __linger_cancel(struct ceph_osd_linger_request *lreq)
2371 if (lreq->is_watch && lreq->ping_req->r_osd)
2372 cancel_linger_request(lreq->ping_req);
2373 if (lreq->reg_req->r_osd)
2374 cancel_linger_request(lreq->reg_req);
2375 cancel_linger_map_check(lreq);
2376 unlink_linger(lreq->osd, lreq);
2377 linger_unregister(lreq);
2380 static void linger_cancel(struct ceph_osd_linger_request *lreq)
2382 struct ceph_osd_client *osdc = lreq->osdc;
2384 down_write(&osdc->lock);
2385 if (__linger_registered(lreq))
2386 __linger_cancel(lreq);
2387 up_write(&osdc->lock);
2390 static void send_linger_map_check(struct ceph_osd_linger_request *lreq);
2392 static void check_linger_pool_dne(struct ceph_osd_linger_request *lreq)
2394 struct ceph_osd_client *osdc = lreq->osdc;
2395 struct ceph_osdmap *map = osdc->osdmap;
2397 verify_osdc_wrlocked(osdc);
2398 WARN_ON(!map->epoch);
2400 if (lreq->register_gen) {
2401 lreq->map_dne_bound = map->epoch;
2402 dout("%s lreq %p linger_id %llu pool disappeared\n", __func__,
2403 lreq, lreq->linger_id);
2404 } else {
2405 dout("%s lreq %p linger_id %llu map_dne_bound %u have %u\n",
2406 __func__, lreq, lreq->linger_id, lreq->map_dne_bound,
2407 map->epoch);
2410 if (lreq->map_dne_bound) {
2411 if (map->epoch >= lreq->map_dne_bound) {
2412 /* we had a new enough map */
2413 pr_info("linger_id %llu pool does not exist\n",
2414 lreq->linger_id);
2415 linger_reg_commit_complete(lreq, -ENOENT);
2416 __linger_cancel(lreq);
2418 } else {
2419 send_linger_map_check(lreq);
2423 static void linger_map_check_cb(struct ceph_mon_generic_request *greq)
2425 struct ceph_osd_client *osdc = &greq->monc->client->osdc;
2426 struct ceph_osd_linger_request *lreq;
2427 u64 linger_id = greq->private_data;
2429 WARN_ON(greq->result || !greq->u.newest);
2431 down_write(&osdc->lock);
2432 lreq = lookup_linger_mc(&osdc->linger_map_checks, linger_id);
2433 if (!lreq) {
2434 dout("%s linger_id %llu dne\n", __func__, linger_id);
2435 goto out_unlock;
2438 dout("%s lreq %p linger_id %llu map_dne_bound %u newest %llu\n",
2439 __func__, lreq, lreq->linger_id, lreq->map_dne_bound,
2440 greq->u.newest);
2441 if (!lreq->map_dne_bound)
2442 lreq->map_dne_bound = greq->u.newest;
2443 erase_linger_mc(&osdc->linger_map_checks, lreq);
2444 check_linger_pool_dne(lreq);
2446 linger_put(lreq);
2447 out_unlock:
2448 up_write(&osdc->lock);
2451 static void send_linger_map_check(struct ceph_osd_linger_request *lreq)
2453 struct ceph_osd_client *osdc = lreq->osdc;
2454 struct ceph_osd_linger_request *lookup_lreq;
2455 int ret;
2457 verify_osdc_wrlocked(osdc);
2459 lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks,
2460 lreq->linger_id);
2461 if (lookup_lreq) {
2462 WARN_ON(lookup_lreq != lreq);
2463 return;
2466 linger_get(lreq);
2467 insert_linger_mc(&osdc->linger_map_checks, lreq);
2468 ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap",
2469 linger_map_check_cb, lreq->linger_id);
2470 WARN_ON(ret);
2473 static int linger_reg_commit_wait(struct ceph_osd_linger_request *lreq)
2475 int ret;
2477 dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
2478 ret = wait_for_completion_interruptible(&lreq->reg_commit_wait);
2479 return ret ?: lreq->reg_commit_error;
2482 static int linger_notify_finish_wait(struct ceph_osd_linger_request *lreq)
2484 int ret;
2486 dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
2487 ret = wait_for_completion_interruptible(&lreq->notify_finish_wait);
2488 return ret ?: lreq->notify_finish_error;
2492 * Timeout callback, called every N seconds. When 1 or more OSD
2493 * requests has been active for more than N seconds, we send a keepalive
2494 * (tag + timestamp) to its OSD to ensure any communications channel
2495 * reset is detected.
2497 static void handle_timeout(struct work_struct *work)
2499 struct ceph_osd_client *osdc =
2500 container_of(work, struct ceph_osd_client, timeout_work.work);
2501 struct ceph_options *opts = osdc->client->options;
2502 unsigned long cutoff = jiffies - opts->osd_keepalive_timeout;
2503 LIST_HEAD(slow_osds);
2504 struct rb_node *n, *p;
2506 dout("%s osdc %p\n", __func__, osdc);
2507 down_write(&osdc->lock);
2510 * ping osds that are a bit slow. this ensures that if there
2511 * is a break in the TCP connection we will notice, and reopen
2512 * a connection with that osd (from the fault callback).
2514 for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
2515 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
2516 bool found = false;
2518 for (p = rb_first(&osd->o_requests); p; p = rb_next(p)) {
2519 struct ceph_osd_request *req =
2520 rb_entry(p, struct ceph_osd_request, r_node);
2522 if (time_before(req->r_stamp, cutoff)) {
2523 dout(" req %p tid %llu on osd%d is laggy\n",
2524 req, req->r_tid, osd->o_osd);
2525 found = true;
2528 for (p = rb_first(&osd->o_linger_requests); p; p = rb_next(p)) {
2529 struct ceph_osd_linger_request *lreq =
2530 rb_entry(p, struct ceph_osd_linger_request, node);
2532 dout(" lreq %p linger_id %llu is served by osd%d\n",
2533 lreq, lreq->linger_id, osd->o_osd);
2534 found = true;
2536 mutex_lock(&lreq->lock);
2537 if (lreq->is_watch && lreq->committed && !lreq->last_error)
2538 send_linger_ping(lreq);
2539 mutex_unlock(&lreq->lock);
2542 if (found)
2543 list_move_tail(&osd->o_keepalive_item, &slow_osds);
2546 if (atomic_read(&osdc->num_homeless) || !list_empty(&slow_osds))
2547 maybe_request_map(osdc);
2549 while (!list_empty(&slow_osds)) {
2550 struct ceph_osd *osd = list_first_entry(&slow_osds,
2551 struct ceph_osd,
2552 o_keepalive_item);
2553 list_del_init(&osd->o_keepalive_item);
2554 ceph_con_keepalive(&osd->o_con);
2557 up_write(&osdc->lock);
2558 schedule_delayed_work(&osdc->timeout_work,
2559 osdc->client->options->osd_keepalive_timeout);
2562 static void handle_osds_timeout(struct work_struct *work)
2564 struct ceph_osd_client *osdc =
2565 container_of(work, struct ceph_osd_client,
2566 osds_timeout_work.work);
2567 unsigned long delay = osdc->client->options->osd_idle_ttl / 4;
2568 struct ceph_osd *osd, *nosd;
2570 dout("%s osdc %p\n", __func__, osdc);
2571 down_write(&osdc->lock);
2572 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
2573 if (time_before(jiffies, osd->lru_ttl))
2574 break;
2576 WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests));
2577 WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests));
2578 close_osd(osd);
2581 up_write(&osdc->lock);
2582 schedule_delayed_work(&osdc->osds_timeout_work,
2583 round_jiffies_relative(delay));
2586 static int ceph_oloc_decode(void **p, void *end,
2587 struct ceph_object_locator *oloc)
2589 u8 struct_v, struct_cv;
2590 u32 len;
2591 void *struct_end;
2592 int ret = 0;
2594 ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
2595 struct_v = ceph_decode_8(p);
2596 struct_cv = ceph_decode_8(p);
2597 if (struct_v < 3) {
2598 pr_warn("got v %d < 3 cv %d of ceph_object_locator\n",
2599 struct_v, struct_cv);
2600 goto e_inval;
2602 if (struct_cv > 6) {
2603 pr_warn("got v %d cv %d > 6 of ceph_object_locator\n",
2604 struct_v, struct_cv);
2605 goto e_inval;
2607 len = ceph_decode_32(p);
2608 ceph_decode_need(p, end, len, e_inval);
2609 struct_end = *p + len;
2611 oloc->pool = ceph_decode_64(p);
2612 *p += 4; /* skip preferred */
2614 len = ceph_decode_32(p);
2615 if (len > 0) {
2616 pr_warn("ceph_object_locator::key is set\n");
2617 goto e_inval;
2620 if (struct_v >= 5) {
2621 bool changed = false;
2623 len = ceph_decode_32(p);
2624 if (len > 0) {
2625 ceph_decode_need(p, end, len, e_inval);
2626 if (!oloc->pool_ns ||
2627 ceph_compare_string(oloc->pool_ns, *p, len))
2628 changed = true;
2629 *p += len;
2630 } else {
2631 if (oloc->pool_ns)
2632 changed = true;
2634 if (changed) {
2635 /* redirect changes namespace */
2636 pr_warn("ceph_object_locator::nspace is changed\n");
2637 goto e_inval;
2641 if (struct_v >= 6) {
2642 s64 hash = ceph_decode_64(p);
2643 if (hash != -1) {
2644 pr_warn("ceph_object_locator::hash is set\n");
2645 goto e_inval;
2649 /* skip the rest */
2650 *p = struct_end;
2651 out:
2652 return ret;
2654 e_inval:
2655 ret = -EINVAL;
2656 goto out;
2659 static int ceph_redirect_decode(void **p, void *end,
2660 struct ceph_request_redirect *redir)
2662 u8 struct_v, struct_cv;
2663 u32 len;
2664 void *struct_end;
2665 int ret;
2667 ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
2668 struct_v = ceph_decode_8(p);
2669 struct_cv = ceph_decode_8(p);
2670 if (struct_cv > 1) {
2671 pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n",
2672 struct_v, struct_cv);
2673 goto e_inval;
2675 len = ceph_decode_32(p);
2676 ceph_decode_need(p, end, len, e_inval);
2677 struct_end = *p + len;
2679 ret = ceph_oloc_decode(p, end, &redir->oloc);
2680 if (ret)
2681 goto out;
2683 len = ceph_decode_32(p);
2684 if (len > 0) {
2685 pr_warn("ceph_request_redirect::object_name is set\n");
2686 goto e_inval;
2689 len = ceph_decode_32(p);
2690 *p += len; /* skip osd_instructions */
2692 /* skip the rest */
2693 *p = struct_end;
2694 out:
2695 return ret;
2697 e_inval:
2698 ret = -EINVAL;
2699 goto out;
2702 struct MOSDOpReply {
2703 struct ceph_pg pgid;
2704 u64 flags;
2705 int result;
2706 u32 epoch;
2707 int num_ops;
2708 u32 outdata_len[CEPH_OSD_MAX_OPS];
2709 s32 rval[CEPH_OSD_MAX_OPS];
2710 int retry_attempt;
2711 struct ceph_eversion replay_version;
2712 u64 user_version;
2713 struct ceph_request_redirect redirect;
2716 static int decode_MOSDOpReply(const struct ceph_msg *msg, struct MOSDOpReply *m)
2718 void *p = msg->front.iov_base;
2719 void *const end = p + msg->front.iov_len;
2720 u16 version = le16_to_cpu(msg->hdr.version);
2721 struct ceph_eversion bad_replay_version;
2722 u8 decode_redir;
2723 u32 len;
2724 int ret;
2725 int i;
2727 ceph_decode_32_safe(&p, end, len, e_inval);
2728 ceph_decode_need(&p, end, len, e_inval);
2729 p += len; /* skip oid */
2731 ret = ceph_decode_pgid(&p, end, &m->pgid);
2732 if (ret)
2733 return ret;
2735 ceph_decode_64_safe(&p, end, m->flags, e_inval);
2736 ceph_decode_32_safe(&p, end, m->result, e_inval);
2737 ceph_decode_need(&p, end, sizeof(bad_replay_version), e_inval);
2738 memcpy(&bad_replay_version, p, sizeof(bad_replay_version));
2739 p += sizeof(bad_replay_version);
2740 ceph_decode_32_safe(&p, end, m->epoch, e_inval);
2742 ceph_decode_32_safe(&p, end, m->num_ops, e_inval);
2743 if (m->num_ops > ARRAY_SIZE(m->outdata_len))
2744 goto e_inval;
2746 ceph_decode_need(&p, end, m->num_ops * sizeof(struct ceph_osd_op),
2747 e_inval);
2748 for (i = 0; i < m->num_ops; i++) {
2749 struct ceph_osd_op *op = p;
2751 m->outdata_len[i] = le32_to_cpu(op->payload_len);
2752 p += sizeof(*op);
2755 ceph_decode_32_safe(&p, end, m->retry_attempt, e_inval);
2756 for (i = 0; i < m->num_ops; i++)
2757 ceph_decode_32_safe(&p, end, m->rval[i], e_inval);
2759 if (version >= 5) {
2760 ceph_decode_need(&p, end, sizeof(m->replay_version), e_inval);
2761 memcpy(&m->replay_version, p, sizeof(m->replay_version));
2762 p += sizeof(m->replay_version);
2763 ceph_decode_64_safe(&p, end, m->user_version, e_inval);
2764 } else {
2765 m->replay_version = bad_replay_version; /* struct */
2766 m->user_version = le64_to_cpu(m->replay_version.version);
2769 if (version >= 6) {
2770 if (version >= 7)
2771 ceph_decode_8_safe(&p, end, decode_redir, e_inval);
2772 else
2773 decode_redir = 1;
2774 } else {
2775 decode_redir = 0;
2778 if (decode_redir) {
2779 ret = ceph_redirect_decode(&p, end, &m->redirect);
2780 if (ret)
2781 return ret;
2782 } else {
2783 ceph_oloc_init(&m->redirect.oloc);
2786 return 0;
2788 e_inval:
2789 return -EINVAL;
2793 * We are done with @req if
2794 * - @m is a safe reply, or
2795 * - @m is an unsafe reply and we didn't want a safe one
2797 static bool done_request(const struct ceph_osd_request *req,
2798 const struct MOSDOpReply *m)
2800 return (m->result < 0 ||
2801 (m->flags & CEPH_OSD_FLAG_ONDISK) ||
2802 !(req->r_flags & CEPH_OSD_FLAG_ONDISK));
2806 * handle osd op reply. either call the callback if it is specified,
2807 * or do the completion to wake up the waiting thread.
2809 * ->r_unsafe_callback is set? yes no
2811 * first reply is OK (needed r_cb/r_completion, r_cb/r_completion,
2812 * any or needed/got safe) r_safe_completion r_safe_completion
2814 * first reply is unsafe r_unsafe_cb(true) (nothing)
2816 * when we get the safe reply r_unsafe_cb(false), r_cb/r_completion,
2817 * r_safe_completion r_safe_completion
2819 static void handle_reply(struct ceph_osd *osd, struct ceph_msg *msg)
2821 struct ceph_osd_client *osdc = osd->o_osdc;
2822 struct ceph_osd_request *req;
2823 struct MOSDOpReply m;
2824 u64 tid = le64_to_cpu(msg->hdr.tid);
2825 u32 data_len = 0;
2826 bool already_acked;
2827 int ret;
2828 int i;
2830 dout("%s msg %p tid %llu\n", __func__, msg, tid);
2832 down_read(&osdc->lock);
2833 if (!osd_registered(osd)) {
2834 dout("%s osd%d unknown\n", __func__, osd->o_osd);
2835 goto out_unlock_osdc;
2837 WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num));
2839 mutex_lock(&osd->lock);
2840 req = lookup_request(&osd->o_requests, tid);
2841 if (!req) {
2842 dout("%s osd%d tid %llu unknown\n", __func__, osd->o_osd, tid);
2843 goto out_unlock_session;
2846 m.redirect.oloc.pool_ns = req->r_t.target_oloc.pool_ns;
2847 ret = decode_MOSDOpReply(msg, &m);
2848 m.redirect.oloc.pool_ns = NULL;
2849 if (ret) {
2850 pr_err("failed to decode MOSDOpReply for tid %llu: %d\n",
2851 req->r_tid, ret);
2852 ceph_msg_dump(msg);
2853 goto fail_request;
2855 dout("%s req %p tid %llu flags 0x%llx pgid %llu.%x epoch %u attempt %d v %u'%llu uv %llu\n",
2856 __func__, req, req->r_tid, m.flags, m.pgid.pool, m.pgid.seed,
2857 m.epoch, m.retry_attempt, le32_to_cpu(m.replay_version.epoch),
2858 le64_to_cpu(m.replay_version.version), m.user_version);
2860 if (m.retry_attempt >= 0) {
2861 if (m.retry_attempt != req->r_attempts - 1) {
2862 dout("req %p tid %llu retry_attempt %d != %d, ignoring\n",
2863 req, req->r_tid, m.retry_attempt,
2864 req->r_attempts - 1);
2865 goto out_unlock_session;
2867 } else {
2868 WARN_ON(1); /* MOSDOpReply v4 is assumed */
2871 if (!ceph_oloc_empty(&m.redirect.oloc)) {
2872 dout("req %p tid %llu redirect pool %lld\n", req, req->r_tid,
2873 m.redirect.oloc.pool);
2874 unlink_request(osd, req);
2875 mutex_unlock(&osd->lock);
2878 * Not ceph_oloc_copy() - changing pool_ns is not
2879 * supported.
2881 req->r_t.target_oloc.pool = m.redirect.oloc.pool;
2882 req->r_flags |= CEPH_OSD_FLAG_REDIRECTED;
2883 req->r_tid = 0;
2884 __submit_request(req, false);
2885 goto out_unlock_osdc;
2888 if (m.num_ops != req->r_num_ops) {
2889 pr_err("num_ops %d != %d for tid %llu\n", m.num_ops,
2890 req->r_num_ops, req->r_tid);
2891 goto fail_request;
2893 for (i = 0; i < req->r_num_ops; i++) {
2894 dout(" req %p tid %llu op %d rval %d len %u\n", req,
2895 req->r_tid, i, m.rval[i], m.outdata_len[i]);
2896 req->r_ops[i].rval = m.rval[i];
2897 req->r_ops[i].outdata_len = m.outdata_len[i];
2898 data_len += m.outdata_len[i];
2900 if (data_len != le32_to_cpu(msg->hdr.data_len)) {
2901 pr_err("sum of lens %u != %u for tid %llu\n", data_len,
2902 le32_to_cpu(msg->hdr.data_len), req->r_tid);
2903 goto fail_request;
2905 dout("%s req %p tid %llu acked %d result %d data_len %u\n", __func__,
2906 req, req->r_tid, req->r_got_reply, m.result, data_len);
2908 already_acked = req->r_got_reply;
2909 if (!already_acked) {
2910 req->r_result = m.result ?: data_len;
2911 req->r_replay_version = m.replay_version; /* struct */
2912 req->r_got_reply = true;
2913 } else if (!(m.flags & CEPH_OSD_FLAG_ONDISK)) {
2914 dout("req %p tid %llu dup ack\n", req, req->r_tid);
2915 goto out_unlock_session;
2918 if (done_request(req, &m)) {
2919 __finish_request(req);
2920 if (req->r_linger) {
2921 WARN_ON(req->r_unsafe_callback);
2922 dout("req %p tid %llu cb (locked)\n", req, req->r_tid);
2923 __complete_request(req);
2927 mutex_unlock(&osd->lock);
2928 up_read(&osdc->lock);
2930 if (done_request(req, &m)) {
2931 if (already_acked && req->r_unsafe_callback) {
2932 dout("req %p tid %llu safe-cb\n", req, req->r_tid);
2933 req->r_unsafe_callback(req, false);
2934 } else if (!req->r_linger) {
2935 dout("req %p tid %llu cb\n", req, req->r_tid);
2936 __complete_request(req);
2938 if (m.flags & CEPH_OSD_FLAG_ONDISK)
2939 complete_all(&req->r_safe_completion);
2940 ceph_osdc_put_request(req);
2941 } else {
2942 if (req->r_unsafe_callback) {
2943 dout("req %p tid %llu unsafe-cb\n", req, req->r_tid);
2944 req->r_unsafe_callback(req, true);
2945 } else {
2946 WARN_ON(1);
2950 return;
2952 fail_request:
2953 complete_request(req, -EIO);
2954 out_unlock_session:
2955 mutex_unlock(&osd->lock);
2956 out_unlock_osdc:
2957 up_read(&osdc->lock);
2960 static void set_pool_was_full(struct ceph_osd_client *osdc)
2962 struct rb_node *n;
2964 for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) {
2965 struct ceph_pg_pool_info *pi =
2966 rb_entry(n, struct ceph_pg_pool_info, node);
2968 pi->was_full = __pool_full(pi);
2972 static bool pool_cleared_full(struct ceph_osd_client *osdc, s64 pool_id)
2974 struct ceph_pg_pool_info *pi;
2976 pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id);
2977 if (!pi)
2978 return false;
2980 return pi->was_full && !__pool_full(pi);
2983 static enum calc_target_result
2984 recalc_linger_target(struct ceph_osd_linger_request *lreq)
2986 struct ceph_osd_client *osdc = lreq->osdc;
2987 enum calc_target_result ct_res;
2989 ct_res = calc_target(osdc, &lreq->t, &lreq->last_force_resend, true);
2990 if (ct_res == CALC_TARGET_NEED_RESEND) {
2991 struct ceph_osd *osd;
2993 osd = lookup_create_osd(osdc, lreq->t.osd, true);
2994 if (osd != lreq->osd) {
2995 unlink_linger(lreq->osd, lreq);
2996 link_linger(osd, lreq);
3000 return ct_res;
3004 * Requeue requests whose mapping to an OSD has changed.
3006 static void scan_requests(struct ceph_osd *osd,
3007 bool force_resend,
3008 bool cleared_full,
3009 bool check_pool_cleared_full,
3010 struct rb_root *need_resend,
3011 struct list_head *need_resend_linger)
3013 struct ceph_osd_client *osdc = osd->o_osdc;
3014 struct rb_node *n;
3015 bool force_resend_writes;
3017 for (n = rb_first(&osd->o_linger_requests); n; ) {
3018 struct ceph_osd_linger_request *lreq =
3019 rb_entry(n, struct ceph_osd_linger_request, node);
3020 enum calc_target_result ct_res;
3022 n = rb_next(n); /* recalc_linger_target() */
3024 dout("%s lreq %p linger_id %llu\n", __func__, lreq,
3025 lreq->linger_id);
3026 ct_res = recalc_linger_target(lreq);
3027 switch (ct_res) {
3028 case CALC_TARGET_NO_ACTION:
3029 force_resend_writes = cleared_full ||
3030 (check_pool_cleared_full &&
3031 pool_cleared_full(osdc, lreq->t.base_oloc.pool));
3032 if (!force_resend && !force_resend_writes)
3033 break;
3035 /* fall through */
3036 case CALC_TARGET_NEED_RESEND:
3037 cancel_linger_map_check(lreq);
3039 * scan_requests() for the previous epoch(s)
3040 * may have already added it to the list, since
3041 * it's not unlinked here.
3043 if (list_empty(&lreq->scan_item))
3044 list_add_tail(&lreq->scan_item, need_resend_linger);
3045 break;
3046 case CALC_TARGET_POOL_DNE:
3047 check_linger_pool_dne(lreq);
3048 break;
3052 for (n = rb_first(&osd->o_requests); n; ) {
3053 struct ceph_osd_request *req =
3054 rb_entry(n, struct ceph_osd_request, r_node);
3055 enum calc_target_result ct_res;
3057 n = rb_next(n); /* unlink_request(), check_pool_dne() */
3059 dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
3060 ct_res = calc_target(osdc, &req->r_t,
3061 &req->r_last_force_resend, false);
3062 switch (ct_res) {
3063 case CALC_TARGET_NO_ACTION:
3064 force_resend_writes = cleared_full ||
3065 (check_pool_cleared_full &&
3066 pool_cleared_full(osdc, req->r_t.base_oloc.pool));
3067 if (!force_resend &&
3068 (!(req->r_flags & CEPH_OSD_FLAG_WRITE) ||
3069 !force_resend_writes))
3070 break;
3072 /* fall through */
3073 case CALC_TARGET_NEED_RESEND:
3074 cancel_map_check(req);
3075 unlink_request(osd, req);
3076 insert_request(need_resend, req);
3077 break;
3078 case CALC_TARGET_POOL_DNE:
3079 check_pool_dne(req);
3080 break;
3085 static int handle_one_map(struct ceph_osd_client *osdc,
3086 void *p, void *end, bool incremental,
3087 struct rb_root *need_resend,
3088 struct list_head *need_resend_linger)
3090 struct ceph_osdmap *newmap;
3091 struct rb_node *n;
3092 bool skipped_map = false;
3093 bool was_full;
3095 was_full = ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL);
3096 set_pool_was_full(osdc);
3098 if (incremental)
3099 newmap = osdmap_apply_incremental(&p, end, osdc->osdmap);
3100 else
3101 newmap = ceph_osdmap_decode(&p, end);
3102 if (IS_ERR(newmap))
3103 return PTR_ERR(newmap);
3105 if (newmap != osdc->osdmap) {
3107 * Preserve ->was_full before destroying the old map.
3108 * For pools that weren't in the old map, ->was_full
3109 * should be false.
3111 for (n = rb_first(&newmap->pg_pools); n; n = rb_next(n)) {
3112 struct ceph_pg_pool_info *pi =
3113 rb_entry(n, struct ceph_pg_pool_info, node);
3114 struct ceph_pg_pool_info *old_pi;
3116 old_pi = ceph_pg_pool_by_id(osdc->osdmap, pi->id);
3117 if (old_pi)
3118 pi->was_full = old_pi->was_full;
3119 else
3120 WARN_ON(pi->was_full);
3123 if (osdc->osdmap->epoch &&
3124 osdc->osdmap->epoch + 1 < newmap->epoch) {
3125 WARN_ON(incremental);
3126 skipped_map = true;
3129 ceph_osdmap_destroy(osdc->osdmap);
3130 osdc->osdmap = newmap;
3133 was_full &= !ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL);
3134 scan_requests(&osdc->homeless_osd, skipped_map, was_full, true,
3135 need_resend, need_resend_linger);
3137 for (n = rb_first(&osdc->osds); n; ) {
3138 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
3140 n = rb_next(n); /* close_osd() */
3142 scan_requests(osd, skipped_map, was_full, true, need_resend,
3143 need_resend_linger);
3144 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
3145 memcmp(&osd->o_con.peer_addr,
3146 ceph_osd_addr(osdc->osdmap, osd->o_osd),
3147 sizeof(struct ceph_entity_addr)))
3148 close_osd(osd);
3151 return 0;
3154 static void kick_requests(struct ceph_osd_client *osdc,
3155 struct rb_root *need_resend,
3156 struct list_head *need_resend_linger)
3158 struct ceph_osd_linger_request *lreq, *nlreq;
3159 struct rb_node *n;
3161 for (n = rb_first(need_resend); n; ) {
3162 struct ceph_osd_request *req =
3163 rb_entry(n, struct ceph_osd_request, r_node);
3164 struct ceph_osd *osd;
3166 n = rb_next(n);
3167 erase_request(need_resend, req); /* before link_request() */
3169 WARN_ON(req->r_osd);
3170 calc_target(osdc, &req->r_t, NULL, false);
3171 osd = lookup_create_osd(osdc, req->r_t.osd, true);
3172 link_request(osd, req);
3173 if (!req->r_linger) {
3174 if (!osd_homeless(osd) && !req->r_t.paused)
3175 send_request(req);
3176 } else {
3177 cancel_linger_request(req);
3181 list_for_each_entry_safe(lreq, nlreq, need_resend_linger, scan_item) {
3182 if (!osd_homeless(lreq->osd))
3183 send_linger(lreq);
3185 list_del_init(&lreq->scan_item);
3190 * Process updated osd map.
3192 * The message contains any number of incremental and full maps, normally
3193 * indicating some sort of topology change in the cluster. Kick requests
3194 * off to different OSDs as needed.
3196 void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
3198 void *p = msg->front.iov_base;
3199 void *const end = p + msg->front.iov_len;
3200 u32 nr_maps, maplen;
3201 u32 epoch;
3202 struct ceph_fsid fsid;
3203 struct rb_root need_resend = RB_ROOT;
3204 LIST_HEAD(need_resend_linger);
3205 bool handled_incremental = false;
3206 bool was_pauserd, was_pausewr;
3207 bool pauserd, pausewr;
3208 int err;
3210 dout("%s have %u\n", __func__, osdc->osdmap->epoch);
3211 down_write(&osdc->lock);
3213 /* verify fsid */
3214 ceph_decode_need(&p, end, sizeof(fsid), bad);
3215 ceph_decode_copy(&p, &fsid, sizeof(fsid));
3216 if (ceph_check_fsid(osdc->client, &fsid) < 0)
3217 goto bad;
3219 was_pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
3220 was_pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
3221 ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
3222 have_pool_full(osdc);
3224 /* incremental maps */
3225 ceph_decode_32_safe(&p, end, nr_maps, bad);
3226 dout(" %d inc maps\n", nr_maps);
3227 while (nr_maps > 0) {
3228 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
3229 epoch = ceph_decode_32(&p);
3230 maplen = ceph_decode_32(&p);
3231 ceph_decode_need(&p, end, maplen, bad);
3232 if (osdc->osdmap->epoch &&
3233 osdc->osdmap->epoch + 1 == epoch) {
3234 dout("applying incremental map %u len %d\n",
3235 epoch, maplen);
3236 err = handle_one_map(osdc, p, p + maplen, true,
3237 &need_resend, &need_resend_linger);
3238 if (err)
3239 goto bad;
3240 handled_incremental = true;
3241 } else {
3242 dout("ignoring incremental map %u len %d\n",
3243 epoch, maplen);
3245 p += maplen;
3246 nr_maps--;
3248 if (handled_incremental)
3249 goto done;
3251 /* full maps */
3252 ceph_decode_32_safe(&p, end, nr_maps, bad);
3253 dout(" %d full maps\n", nr_maps);
3254 while (nr_maps) {
3255 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
3256 epoch = ceph_decode_32(&p);
3257 maplen = ceph_decode_32(&p);
3258 ceph_decode_need(&p, end, maplen, bad);
3259 if (nr_maps > 1) {
3260 dout("skipping non-latest full map %u len %d\n",
3261 epoch, maplen);
3262 } else if (osdc->osdmap->epoch >= epoch) {
3263 dout("skipping full map %u len %d, "
3264 "older than our %u\n", epoch, maplen,
3265 osdc->osdmap->epoch);
3266 } else {
3267 dout("taking full map %u len %d\n", epoch, maplen);
3268 err = handle_one_map(osdc, p, p + maplen, false,
3269 &need_resend, &need_resend_linger);
3270 if (err)
3271 goto bad;
3273 p += maplen;
3274 nr_maps--;
3277 done:
3279 * subscribe to subsequent osdmap updates if full to ensure
3280 * we find out when we are no longer full and stop returning
3281 * ENOSPC.
3283 pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
3284 pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
3285 ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
3286 have_pool_full(osdc);
3287 if (was_pauserd || was_pausewr || pauserd || pausewr)
3288 maybe_request_map(osdc);
3290 kick_requests(osdc, &need_resend, &need_resend_linger);
3292 ceph_monc_got_map(&osdc->client->monc, CEPH_SUB_OSDMAP,
3293 osdc->osdmap->epoch);
3294 up_write(&osdc->lock);
3295 wake_up_all(&osdc->client->auth_wq);
3296 return;
3298 bad:
3299 pr_err("osdc handle_map corrupt msg\n");
3300 ceph_msg_dump(msg);
3301 up_write(&osdc->lock);
3305 * Resubmit requests pending on the given osd.
3307 static void kick_osd_requests(struct ceph_osd *osd)
3309 struct rb_node *n;
3311 for (n = rb_first(&osd->o_requests); n; ) {
3312 struct ceph_osd_request *req =
3313 rb_entry(n, struct ceph_osd_request, r_node);
3315 n = rb_next(n); /* cancel_linger_request() */
3317 if (!req->r_linger) {
3318 if (!req->r_t.paused)
3319 send_request(req);
3320 } else {
3321 cancel_linger_request(req);
3324 for (n = rb_first(&osd->o_linger_requests); n; n = rb_next(n)) {
3325 struct ceph_osd_linger_request *lreq =
3326 rb_entry(n, struct ceph_osd_linger_request, node);
3328 send_linger(lreq);
3333 * If the osd connection drops, we need to resubmit all requests.
3335 static void osd_fault(struct ceph_connection *con)
3337 struct ceph_osd *osd = con->private;
3338 struct ceph_osd_client *osdc = osd->o_osdc;
3340 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
3342 down_write(&osdc->lock);
3343 if (!osd_registered(osd)) {
3344 dout("%s osd%d unknown\n", __func__, osd->o_osd);
3345 goto out_unlock;
3348 if (!reopen_osd(osd))
3349 kick_osd_requests(osd);
3350 maybe_request_map(osdc);
3352 out_unlock:
3353 up_write(&osdc->lock);
3357 * Process osd watch notifications
3359 static void handle_watch_notify(struct ceph_osd_client *osdc,
3360 struct ceph_msg *msg)
3362 void *p = msg->front.iov_base;
3363 void *const end = p + msg->front.iov_len;
3364 struct ceph_osd_linger_request *lreq;
3365 struct linger_work *lwork;
3366 u8 proto_ver, opcode;
3367 u64 cookie, notify_id;
3368 u64 notifier_id = 0;
3369 s32 return_code = 0;
3370 void *payload = NULL;
3371 u32 payload_len = 0;
3373 ceph_decode_8_safe(&p, end, proto_ver, bad);
3374 ceph_decode_8_safe(&p, end, opcode, bad);
3375 ceph_decode_64_safe(&p, end, cookie, bad);
3376 p += 8; /* skip ver */
3377 ceph_decode_64_safe(&p, end, notify_id, bad);
3379 if (proto_ver >= 1) {
3380 ceph_decode_32_safe(&p, end, payload_len, bad);
3381 ceph_decode_need(&p, end, payload_len, bad);
3382 payload = p;
3383 p += payload_len;
3386 if (le16_to_cpu(msg->hdr.version) >= 2)
3387 ceph_decode_32_safe(&p, end, return_code, bad);
3389 if (le16_to_cpu(msg->hdr.version) >= 3)
3390 ceph_decode_64_safe(&p, end, notifier_id, bad);
3392 down_read(&osdc->lock);
3393 lreq = lookup_linger_osdc(&osdc->linger_requests, cookie);
3394 if (!lreq) {
3395 dout("%s opcode %d cookie %llu dne\n", __func__, opcode,
3396 cookie);
3397 goto out_unlock_osdc;
3400 mutex_lock(&lreq->lock);
3401 dout("%s opcode %d cookie %llu lreq %p is_watch %d\n", __func__,
3402 opcode, cookie, lreq, lreq->is_watch);
3403 if (opcode == CEPH_WATCH_EVENT_DISCONNECT) {
3404 if (!lreq->last_error) {
3405 lreq->last_error = -ENOTCONN;
3406 queue_watch_error(lreq);
3408 } else if (!lreq->is_watch) {
3409 /* CEPH_WATCH_EVENT_NOTIFY_COMPLETE */
3410 if (lreq->notify_id && lreq->notify_id != notify_id) {
3411 dout("lreq %p notify_id %llu != %llu, ignoring\n", lreq,
3412 lreq->notify_id, notify_id);
3413 } else if (!completion_done(&lreq->notify_finish_wait)) {
3414 struct ceph_msg_data *data =
3415 list_first_entry_or_null(&msg->data,
3416 struct ceph_msg_data,
3417 links);
3419 if (data) {
3420 if (lreq->preply_pages) {
3421 WARN_ON(data->type !=
3422 CEPH_MSG_DATA_PAGES);
3423 *lreq->preply_pages = data->pages;
3424 *lreq->preply_len = data->length;
3425 } else {
3426 ceph_release_page_vector(data->pages,
3427 calc_pages_for(0, data->length));
3430 lreq->notify_finish_error = return_code;
3431 complete_all(&lreq->notify_finish_wait);
3433 } else {
3434 /* CEPH_WATCH_EVENT_NOTIFY */
3435 lwork = lwork_alloc(lreq, do_watch_notify);
3436 if (!lwork) {
3437 pr_err("failed to allocate notify-lwork\n");
3438 goto out_unlock_lreq;
3441 lwork->notify.notify_id = notify_id;
3442 lwork->notify.notifier_id = notifier_id;
3443 lwork->notify.payload = payload;
3444 lwork->notify.payload_len = payload_len;
3445 lwork->notify.msg = ceph_msg_get(msg);
3446 lwork_queue(lwork);
3449 out_unlock_lreq:
3450 mutex_unlock(&lreq->lock);
3451 out_unlock_osdc:
3452 up_read(&osdc->lock);
3453 return;
3455 bad:
3456 pr_err("osdc handle_watch_notify corrupt msg\n");
3460 * Register request, send initial attempt.
3462 int ceph_osdc_start_request(struct ceph_osd_client *osdc,
3463 struct ceph_osd_request *req,
3464 bool nofail)
3466 down_read(&osdc->lock);
3467 submit_request(req, false);
3468 up_read(&osdc->lock);
3470 return 0;
3472 EXPORT_SYMBOL(ceph_osdc_start_request);
3475 * Unregister a registered request. The request is not completed (i.e.
3476 * no callbacks or wakeups) - higher layers are supposed to know what
3477 * they are canceling.
3479 void ceph_osdc_cancel_request(struct ceph_osd_request *req)
3481 struct ceph_osd_client *osdc = req->r_osdc;
3483 down_write(&osdc->lock);
3484 if (req->r_osd)
3485 cancel_request(req);
3486 up_write(&osdc->lock);
3488 EXPORT_SYMBOL(ceph_osdc_cancel_request);
3491 * @timeout: in jiffies, 0 means "wait forever"
3493 static int wait_request_timeout(struct ceph_osd_request *req,
3494 unsigned long timeout)
3496 long left;
3498 dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
3499 left = wait_for_completion_killable_timeout(&req->r_completion,
3500 ceph_timeout_jiffies(timeout));
3501 if (left <= 0) {
3502 left = left ?: -ETIMEDOUT;
3503 ceph_osdc_cancel_request(req);
3505 /* kludge - need to to wake ceph_osdc_sync() */
3506 complete_all(&req->r_safe_completion);
3507 } else {
3508 left = req->r_result; /* completed */
3511 return left;
3515 * wait for a request to complete
3517 int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
3518 struct ceph_osd_request *req)
3520 return wait_request_timeout(req, 0);
3522 EXPORT_SYMBOL(ceph_osdc_wait_request);
3525 * sync - wait for all in-flight requests to flush. avoid starvation.
3527 void ceph_osdc_sync(struct ceph_osd_client *osdc)
3529 struct rb_node *n, *p;
3530 u64 last_tid = atomic64_read(&osdc->last_tid);
3532 again:
3533 down_read(&osdc->lock);
3534 for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
3535 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
3537 mutex_lock(&osd->lock);
3538 for (p = rb_first(&osd->o_requests); p; p = rb_next(p)) {
3539 struct ceph_osd_request *req =
3540 rb_entry(p, struct ceph_osd_request, r_node);
3542 if (req->r_tid > last_tid)
3543 break;
3545 if (!(req->r_flags & CEPH_OSD_FLAG_WRITE))
3546 continue;
3548 ceph_osdc_get_request(req);
3549 mutex_unlock(&osd->lock);
3550 up_read(&osdc->lock);
3551 dout("%s waiting on req %p tid %llu last_tid %llu\n",
3552 __func__, req, req->r_tid, last_tid);
3553 wait_for_completion(&req->r_safe_completion);
3554 ceph_osdc_put_request(req);
3555 goto again;
3558 mutex_unlock(&osd->lock);
3561 up_read(&osdc->lock);
3562 dout("%s done last_tid %llu\n", __func__, last_tid);
3564 EXPORT_SYMBOL(ceph_osdc_sync);
3566 static struct ceph_osd_request *
3567 alloc_linger_request(struct ceph_osd_linger_request *lreq)
3569 struct ceph_osd_request *req;
3571 req = ceph_osdc_alloc_request(lreq->osdc, NULL, 1, false, GFP_NOIO);
3572 if (!req)
3573 return NULL;
3575 ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
3576 ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
3578 if (ceph_osdc_alloc_messages(req, GFP_NOIO)) {
3579 ceph_osdc_put_request(req);
3580 return NULL;
3583 return req;
3587 * Returns a handle, caller owns a ref.
3589 struct ceph_osd_linger_request *
3590 ceph_osdc_watch(struct ceph_osd_client *osdc,
3591 struct ceph_object_id *oid,
3592 struct ceph_object_locator *oloc,
3593 rados_watchcb2_t wcb,
3594 rados_watcherrcb_t errcb,
3595 void *data)
3597 struct ceph_osd_linger_request *lreq;
3598 int ret;
3600 lreq = linger_alloc(osdc);
3601 if (!lreq)
3602 return ERR_PTR(-ENOMEM);
3604 lreq->is_watch = true;
3605 lreq->wcb = wcb;
3606 lreq->errcb = errcb;
3607 lreq->data = data;
3608 lreq->watch_valid_thru = jiffies;
3610 ceph_oid_copy(&lreq->t.base_oid, oid);
3611 ceph_oloc_copy(&lreq->t.base_oloc, oloc);
3612 lreq->t.flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
3613 lreq->mtime = CURRENT_TIME;
3615 lreq->reg_req = alloc_linger_request(lreq);
3616 if (!lreq->reg_req) {
3617 ret = -ENOMEM;
3618 goto err_put_lreq;
3621 lreq->ping_req = alloc_linger_request(lreq);
3622 if (!lreq->ping_req) {
3623 ret = -ENOMEM;
3624 goto err_put_lreq;
3627 down_write(&osdc->lock);
3628 linger_register(lreq); /* before osd_req_op_* */
3629 osd_req_op_watch_init(lreq->reg_req, 0, lreq->linger_id,
3630 CEPH_OSD_WATCH_OP_WATCH);
3631 osd_req_op_watch_init(lreq->ping_req, 0, lreq->linger_id,
3632 CEPH_OSD_WATCH_OP_PING);
3633 linger_submit(lreq);
3634 up_write(&osdc->lock);
3636 ret = linger_reg_commit_wait(lreq);
3637 if (ret) {
3638 linger_cancel(lreq);
3639 goto err_put_lreq;
3642 return lreq;
3644 err_put_lreq:
3645 linger_put(lreq);
3646 return ERR_PTR(ret);
3648 EXPORT_SYMBOL(ceph_osdc_watch);
3651 * Releases a ref.
3653 * Times out after mount_timeout to preserve rbd unmap behaviour
3654 * introduced in 2894e1d76974 ("rbd: timeout watch teardown on unmap
3655 * with mount_timeout").
3657 int ceph_osdc_unwatch(struct ceph_osd_client *osdc,
3658 struct ceph_osd_linger_request *lreq)
3660 struct ceph_options *opts = osdc->client->options;
3661 struct ceph_osd_request *req;
3662 int ret;
3664 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
3665 if (!req)
3666 return -ENOMEM;
3668 ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
3669 ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
3670 req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
3671 req->r_mtime = CURRENT_TIME;
3672 osd_req_op_watch_init(req, 0, lreq->linger_id,
3673 CEPH_OSD_WATCH_OP_UNWATCH);
3675 ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
3676 if (ret)
3677 goto out_put_req;
3679 ceph_osdc_start_request(osdc, req, false);
3680 linger_cancel(lreq);
3681 linger_put(lreq);
3682 ret = wait_request_timeout(req, opts->mount_timeout);
3684 out_put_req:
3685 ceph_osdc_put_request(req);
3686 return ret;
3688 EXPORT_SYMBOL(ceph_osdc_unwatch);
3690 static int osd_req_op_notify_ack_init(struct ceph_osd_request *req, int which,
3691 u64 notify_id, u64 cookie, void *payload,
3692 size_t payload_len)
3694 struct ceph_osd_req_op *op;
3695 struct ceph_pagelist *pl;
3696 int ret;
3698 op = _osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY_ACK, 0);
3700 pl = kmalloc(sizeof(*pl), GFP_NOIO);
3701 if (!pl)
3702 return -ENOMEM;
3704 ceph_pagelist_init(pl);
3705 ret = ceph_pagelist_encode_64(pl, notify_id);
3706 ret |= ceph_pagelist_encode_64(pl, cookie);
3707 if (payload) {
3708 ret |= ceph_pagelist_encode_32(pl, payload_len);
3709 ret |= ceph_pagelist_append(pl, payload, payload_len);
3710 } else {
3711 ret |= ceph_pagelist_encode_32(pl, 0);
3713 if (ret) {
3714 ceph_pagelist_release(pl);
3715 return -ENOMEM;
3718 ceph_osd_data_pagelist_init(&op->notify_ack.request_data, pl);
3719 op->indata_len = pl->length;
3720 return 0;
3723 int ceph_osdc_notify_ack(struct ceph_osd_client *osdc,
3724 struct ceph_object_id *oid,
3725 struct ceph_object_locator *oloc,
3726 u64 notify_id,
3727 u64 cookie,
3728 void *payload,
3729 size_t payload_len)
3731 struct ceph_osd_request *req;
3732 int ret;
3734 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
3735 if (!req)
3736 return -ENOMEM;
3738 ceph_oid_copy(&req->r_base_oid, oid);
3739 ceph_oloc_copy(&req->r_base_oloc, oloc);
3740 req->r_flags = CEPH_OSD_FLAG_READ;
3742 ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
3743 if (ret)
3744 goto out_put_req;
3746 ret = osd_req_op_notify_ack_init(req, 0, notify_id, cookie, payload,
3747 payload_len);
3748 if (ret)
3749 goto out_put_req;
3751 ceph_osdc_start_request(osdc, req, false);
3752 ret = ceph_osdc_wait_request(osdc, req);
3754 out_put_req:
3755 ceph_osdc_put_request(req);
3756 return ret;
3758 EXPORT_SYMBOL(ceph_osdc_notify_ack);
3760 static int osd_req_op_notify_init(struct ceph_osd_request *req, int which,
3761 u64 cookie, u32 prot_ver, u32 timeout,
3762 void *payload, size_t payload_len)
3764 struct ceph_osd_req_op *op;
3765 struct ceph_pagelist *pl;
3766 int ret;
3768 op = _osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY, 0);
3769 op->notify.cookie = cookie;
3771 pl = kmalloc(sizeof(*pl), GFP_NOIO);
3772 if (!pl)
3773 return -ENOMEM;
3775 ceph_pagelist_init(pl);
3776 ret = ceph_pagelist_encode_32(pl, 1); /* prot_ver */
3777 ret |= ceph_pagelist_encode_32(pl, timeout);
3778 ret |= ceph_pagelist_encode_32(pl, payload_len);
3779 ret |= ceph_pagelist_append(pl, payload, payload_len);
3780 if (ret) {
3781 ceph_pagelist_release(pl);
3782 return -ENOMEM;
3785 ceph_osd_data_pagelist_init(&op->notify.request_data, pl);
3786 op->indata_len = pl->length;
3787 return 0;
3791 * @timeout: in seconds
3793 * @preply_{pages,len} are initialized both on success and error.
3794 * The caller is responsible for:
3796 * ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len))
3798 int ceph_osdc_notify(struct ceph_osd_client *osdc,
3799 struct ceph_object_id *oid,
3800 struct ceph_object_locator *oloc,
3801 void *payload,
3802 size_t payload_len,
3803 u32 timeout,
3804 struct page ***preply_pages,
3805 size_t *preply_len)
3807 struct ceph_osd_linger_request *lreq;
3808 struct page **pages;
3809 int ret;
3811 WARN_ON(!timeout);
3812 if (preply_pages) {
3813 *preply_pages = NULL;
3814 *preply_len = 0;
3817 lreq = linger_alloc(osdc);
3818 if (!lreq)
3819 return -ENOMEM;
3821 lreq->preply_pages = preply_pages;
3822 lreq->preply_len = preply_len;
3824 ceph_oid_copy(&lreq->t.base_oid, oid);
3825 ceph_oloc_copy(&lreq->t.base_oloc, oloc);
3826 lreq->t.flags = CEPH_OSD_FLAG_READ;
3828 lreq->reg_req = alloc_linger_request(lreq);
3829 if (!lreq->reg_req) {
3830 ret = -ENOMEM;
3831 goto out_put_lreq;
3834 /* for notify_id */
3835 pages = ceph_alloc_page_vector(1, GFP_NOIO);
3836 if (IS_ERR(pages)) {
3837 ret = PTR_ERR(pages);
3838 goto out_put_lreq;
3841 down_write(&osdc->lock);
3842 linger_register(lreq); /* before osd_req_op_* */
3843 ret = osd_req_op_notify_init(lreq->reg_req, 0, lreq->linger_id, 1,
3844 timeout, payload, payload_len);
3845 if (ret) {
3846 linger_unregister(lreq);
3847 up_write(&osdc->lock);
3848 ceph_release_page_vector(pages, 1);
3849 goto out_put_lreq;
3851 ceph_osd_data_pages_init(osd_req_op_data(lreq->reg_req, 0, notify,
3852 response_data),
3853 pages, PAGE_SIZE, 0, false, true);
3854 linger_submit(lreq);
3855 up_write(&osdc->lock);
3857 ret = linger_reg_commit_wait(lreq);
3858 if (!ret)
3859 ret = linger_notify_finish_wait(lreq);
3860 else
3861 dout("lreq %p failed to initiate notify %d\n", lreq, ret);
3863 linger_cancel(lreq);
3864 out_put_lreq:
3865 linger_put(lreq);
3866 return ret;
3868 EXPORT_SYMBOL(ceph_osdc_notify);
3871 * Return the number of milliseconds since the watch was last
3872 * confirmed, or an error. If there is an error, the watch is no
3873 * longer valid, and should be destroyed with ceph_osdc_unwatch().
3875 int ceph_osdc_watch_check(struct ceph_osd_client *osdc,
3876 struct ceph_osd_linger_request *lreq)
3878 unsigned long stamp, age;
3879 int ret;
3881 down_read(&osdc->lock);
3882 mutex_lock(&lreq->lock);
3883 stamp = lreq->watch_valid_thru;
3884 if (!list_empty(&lreq->pending_lworks)) {
3885 struct linger_work *lwork =
3886 list_first_entry(&lreq->pending_lworks,
3887 struct linger_work,
3888 pending_item);
3890 if (time_before(lwork->queued_stamp, stamp))
3891 stamp = lwork->queued_stamp;
3893 age = jiffies - stamp;
3894 dout("%s lreq %p linger_id %llu age %lu last_error %d\n", __func__,
3895 lreq, lreq->linger_id, age, lreq->last_error);
3896 /* we are truncating to msecs, so return a safe upper bound */
3897 ret = lreq->last_error ?: 1 + jiffies_to_msecs(age);
3899 mutex_unlock(&lreq->lock);
3900 up_read(&osdc->lock);
3901 return ret;
3904 static int decode_watcher(void **p, void *end, struct ceph_watch_item *item)
3906 u8 struct_v;
3907 u32 struct_len;
3908 int ret;
3910 ret = ceph_start_decoding(p, end, 2, "watch_item_t",
3911 &struct_v, &struct_len);
3912 if (ret)
3913 return ret;
3915 ceph_decode_copy(p, &item->name, sizeof(item->name));
3916 item->cookie = ceph_decode_64(p);
3917 *p += 4; /* skip timeout_seconds */
3918 if (struct_v >= 2) {
3919 ceph_decode_copy(p, &item->addr, sizeof(item->addr));
3920 ceph_decode_addr(&item->addr);
3923 dout("%s %s%llu cookie %llu addr %s\n", __func__,
3924 ENTITY_NAME(item->name), item->cookie,
3925 ceph_pr_addr(&item->addr.in_addr));
3926 return 0;
3929 static int decode_watchers(void **p, void *end,
3930 struct ceph_watch_item **watchers,
3931 u32 *num_watchers)
3933 u8 struct_v;
3934 u32 struct_len;
3935 int i;
3936 int ret;
3938 ret = ceph_start_decoding(p, end, 1, "obj_list_watch_response_t",
3939 &struct_v, &struct_len);
3940 if (ret)
3941 return ret;
3943 *num_watchers = ceph_decode_32(p);
3944 *watchers = kcalloc(*num_watchers, sizeof(**watchers), GFP_NOIO);
3945 if (!*watchers)
3946 return -ENOMEM;
3948 for (i = 0; i < *num_watchers; i++) {
3949 ret = decode_watcher(p, end, *watchers + i);
3950 if (ret) {
3951 kfree(*watchers);
3952 return ret;
3956 return 0;
3960 * On success, the caller is responsible for:
3962 * kfree(watchers);
3964 int ceph_osdc_list_watchers(struct ceph_osd_client *osdc,
3965 struct ceph_object_id *oid,
3966 struct ceph_object_locator *oloc,
3967 struct ceph_watch_item **watchers,
3968 u32 *num_watchers)
3970 struct ceph_osd_request *req;
3971 struct page **pages;
3972 int ret;
3974 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
3975 if (!req)
3976 return -ENOMEM;
3978 ceph_oid_copy(&req->r_base_oid, oid);
3979 ceph_oloc_copy(&req->r_base_oloc, oloc);
3980 req->r_flags = CEPH_OSD_FLAG_READ;
3982 ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
3983 if (ret)
3984 goto out_put_req;
3986 pages = ceph_alloc_page_vector(1, GFP_NOIO);
3987 if (IS_ERR(pages)) {
3988 ret = PTR_ERR(pages);
3989 goto out_put_req;
3992 osd_req_op_init(req, 0, CEPH_OSD_OP_LIST_WATCHERS, 0);
3993 ceph_osd_data_pages_init(osd_req_op_data(req, 0, list_watchers,
3994 response_data),
3995 pages, PAGE_SIZE, 0, false, true);
3997 ceph_osdc_start_request(osdc, req, false);
3998 ret = ceph_osdc_wait_request(osdc, req);
3999 if (ret >= 0) {
4000 void *p = page_address(pages[0]);
4001 void *const end = p + req->r_ops[0].outdata_len;
4003 ret = decode_watchers(&p, end, watchers, num_watchers);
4006 out_put_req:
4007 ceph_osdc_put_request(req);
4008 return ret;
4010 EXPORT_SYMBOL(ceph_osdc_list_watchers);
4013 * Call all pending notify callbacks - for use after a watch is
4014 * unregistered, to make sure no more callbacks for it will be invoked
4016 void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc)
4018 dout("%s osdc %p\n", __func__, osdc);
4019 flush_workqueue(osdc->notify_wq);
4021 EXPORT_SYMBOL(ceph_osdc_flush_notifies);
4023 void ceph_osdc_maybe_request_map(struct ceph_osd_client *osdc)
4025 down_read(&osdc->lock);
4026 maybe_request_map(osdc);
4027 up_read(&osdc->lock);
4029 EXPORT_SYMBOL(ceph_osdc_maybe_request_map);
4032 * Execute an OSD class method on an object.
4034 * @flags: CEPH_OSD_FLAG_*
4035 * @resp_len: out param for reply length
4037 int ceph_osdc_call(struct ceph_osd_client *osdc,
4038 struct ceph_object_id *oid,
4039 struct ceph_object_locator *oloc,
4040 const char *class, const char *method,
4041 unsigned int flags,
4042 struct page *req_page, size_t req_len,
4043 struct page *resp_page, size_t *resp_len)
4045 struct ceph_osd_request *req;
4046 int ret;
4048 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
4049 if (!req)
4050 return -ENOMEM;
4052 ceph_oid_copy(&req->r_base_oid, oid);
4053 ceph_oloc_copy(&req->r_base_oloc, oloc);
4054 req->r_flags = flags;
4056 ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
4057 if (ret)
4058 goto out_put_req;
4060 osd_req_op_cls_init(req, 0, CEPH_OSD_OP_CALL, class, method);
4061 if (req_page)
4062 osd_req_op_cls_request_data_pages(req, 0, &req_page, req_len,
4063 0, false, false);
4064 if (resp_page)
4065 osd_req_op_cls_response_data_pages(req, 0, &resp_page,
4066 PAGE_SIZE, 0, false, false);
4068 ceph_osdc_start_request(osdc, req, false);
4069 ret = ceph_osdc_wait_request(osdc, req);
4070 if (ret >= 0) {
4071 ret = req->r_ops[0].rval;
4072 if (resp_page)
4073 *resp_len = req->r_ops[0].outdata_len;
4076 out_put_req:
4077 ceph_osdc_put_request(req);
4078 return ret;
4080 EXPORT_SYMBOL(ceph_osdc_call);
4083 * init, shutdown
4085 int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
4087 int err;
4089 dout("init\n");
4090 osdc->client = client;
4091 init_rwsem(&osdc->lock);
4092 osdc->osds = RB_ROOT;
4093 INIT_LIST_HEAD(&osdc->osd_lru);
4094 spin_lock_init(&osdc->osd_lru_lock);
4095 osd_init(&osdc->homeless_osd);
4096 osdc->homeless_osd.o_osdc = osdc;
4097 osdc->homeless_osd.o_osd = CEPH_HOMELESS_OSD;
4098 osdc->last_linger_id = CEPH_LINGER_ID_START;
4099 osdc->linger_requests = RB_ROOT;
4100 osdc->map_checks = RB_ROOT;
4101 osdc->linger_map_checks = RB_ROOT;
4102 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
4103 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
4105 err = -ENOMEM;
4106 osdc->osdmap = ceph_osdmap_alloc();
4107 if (!osdc->osdmap)
4108 goto out;
4110 osdc->req_mempool = mempool_create_slab_pool(10,
4111 ceph_osd_request_cache);
4112 if (!osdc->req_mempool)
4113 goto out_map;
4115 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
4116 PAGE_SIZE, 10, true, "osd_op");
4117 if (err < 0)
4118 goto out_mempool;
4119 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
4120 PAGE_SIZE, 10, true, "osd_op_reply");
4121 if (err < 0)
4122 goto out_msgpool;
4124 err = -ENOMEM;
4125 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
4126 if (!osdc->notify_wq)
4127 goto out_msgpool_reply;
4129 schedule_delayed_work(&osdc->timeout_work,
4130 osdc->client->options->osd_keepalive_timeout);
4131 schedule_delayed_work(&osdc->osds_timeout_work,
4132 round_jiffies_relative(osdc->client->options->osd_idle_ttl));
4134 return 0;
4136 out_msgpool_reply:
4137 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
4138 out_msgpool:
4139 ceph_msgpool_destroy(&osdc->msgpool_op);
4140 out_mempool:
4141 mempool_destroy(osdc->req_mempool);
4142 out_map:
4143 ceph_osdmap_destroy(osdc->osdmap);
4144 out:
4145 return err;
4148 void ceph_osdc_stop(struct ceph_osd_client *osdc)
4150 flush_workqueue(osdc->notify_wq);
4151 destroy_workqueue(osdc->notify_wq);
4152 cancel_delayed_work_sync(&osdc->timeout_work);
4153 cancel_delayed_work_sync(&osdc->osds_timeout_work);
4155 down_write(&osdc->lock);
4156 while (!RB_EMPTY_ROOT(&osdc->osds)) {
4157 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
4158 struct ceph_osd, o_node);
4159 close_osd(osd);
4161 up_write(&osdc->lock);
4162 WARN_ON(atomic_read(&osdc->homeless_osd.o_ref) != 1);
4163 osd_cleanup(&osdc->homeless_osd);
4165 WARN_ON(!list_empty(&osdc->osd_lru));
4166 WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_requests));
4167 WARN_ON(!RB_EMPTY_ROOT(&osdc->map_checks));
4168 WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_map_checks));
4169 WARN_ON(atomic_read(&osdc->num_requests));
4170 WARN_ON(atomic_read(&osdc->num_homeless));
4172 ceph_osdmap_destroy(osdc->osdmap);
4173 mempool_destroy(osdc->req_mempool);
4174 ceph_msgpool_destroy(&osdc->msgpool_op);
4175 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
4179 * Read some contiguous pages. If we cross a stripe boundary, shorten
4180 * *plen. Return number of bytes read, or error.
4182 int ceph_osdc_readpages(struct ceph_osd_client *osdc,
4183 struct ceph_vino vino, struct ceph_file_layout *layout,
4184 u64 off, u64 *plen,
4185 u32 truncate_seq, u64 truncate_size,
4186 struct page **pages, int num_pages, int page_align)
4188 struct ceph_osd_request *req;
4189 int rc = 0;
4191 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
4192 vino.snap, off, *plen);
4193 req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 0, 1,
4194 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
4195 NULL, truncate_seq, truncate_size,
4196 false);
4197 if (IS_ERR(req))
4198 return PTR_ERR(req);
4200 /* it may be a short read due to an object boundary */
4201 osd_req_op_extent_osd_data_pages(req, 0,
4202 pages, *plen, page_align, false, false);
4204 dout("readpages final extent is %llu~%llu (%llu bytes align %d)\n",
4205 off, *plen, *plen, page_align);
4207 rc = ceph_osdc_start_request(osdc, req, false);
4208 if (!rc)
4209 rc = ceph_osdc_wait_request(osdc, req);
4211 ceph_osdc_put_request(req);
4212 dout("readpages result %d\n", rc);
4213 return rc;
4215 EXPORT_SYMBOL(ceph_osdc_readpages);
4218 * do a synchronous write on N pages
4220 int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
4221 struct ceph_file_layout *layout,
4222 struct ceph_snap_context *snapc,
4223 u64 off, u64 len,
4224 u32 truncate_seq, u64 truncate_size,
4225 struct timespec *mtime,
4226 struct page **pages, int num_pages)
4228 struct ceph_osd_request *req;
4229 int rc = 0;
4230 int page_align = off & ~PAGE_MASK;
4232 req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 0, 1,
4233 CEPH_OSD_OP_WRITE,
4234 CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
4235 snapc, truncate_seq, truncate_size,
4236 true);
4237 if (IS_ERR(req))
4238 return PTR_ERR(req);
4240 /* it may be a short write due to an object boundary */
4241 osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align,
4242 false, false);
4243 dout("writepages %llu~%llu (%llu bytes)\n", off, len, len);
4245 req->r_mtime = *mtime;
4246 rc = ceph_osdc_start_request(osdc, req, true);
4247 if (!rc)
4248 rc = ceph_osdc_wait_request(osdc, req);
4250 ceph_osdc_put_request(req);
4251 if (rc == 0)
4252 rc = len;
4253 dout("writepages result %d\n", rc);
4254 return rc;
4256 EXPORT_SYMBOL(ceph_osdc_writepages);
4258 int ceph_osdc_setup(void)
4260 size_t size = sizeof(struct ceph_osd_request) +
4261 CEPH_OSD_SLAB_OPS * sizeof(struct ceph_osd_req_op);
4263 BUG_ON(ceph_osd_request_cache);
4264 ceph_osd_request_cache = kmem_cache_create("ceph_osd_request", size,
4265 0, 0, NULL);
4267 return ceph_osd_request_cache ? 0 : -ENOMEM;
4269 EXPORT_SYMBOL(ceph_osdc_setup);
4271 void ceph_osdc_cleanup(void)
4273 BUG_ON(!ceph_osd_request_cache);
4274 kmem_cache_destroy(ceph_osd_request_cache);
4275 ceph_osd_request_cache = NULL;
4277 EXPORT_SYMBOL(ceph_osdc_cleanup);
4280 * handle incoming message
4282 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
4284 struct ceph_osd *osd = con->private;
4285 struct ceph_osd_client *osdc = osd->o_osdc;
4286 int type = le16_to_cpu(msg->hdr.type);
4288 switch (type) {
4289 case CEPH_MSG_OSD_MAP:
4290 ceph_osdc_handle_map(osdc, msg);
4291 break;
4292 case CEPH_MSG_OSD_OPREPLY:
4293 handle_reply(osd, msg);
4294 break;
4295 case CEPH_MSG_WATCH_NOTIFY:
4296 handle_watch_notify(osdc, msg);
4297 break;
4299 default:
4300 pr_err("received unknown message type %d %s\n", type,
4301 ceph_msg_type_name(type));
4304 ceph_msg_put(msg);
4308 * Lookup and return message for incoming reply. Don't try to do
4309 * anything about a larger than preallocated data portion of the
4310 * message at the moment - for now, just skip the message.
4312 static struct ceph_msg *get_reply(struct ceph_connection *con,
4313 struct ceph_msg_header *hdr,
4314 int *skip)
4316 struct ceph_osd *osd = con->private;
4317 struct ceph_osd_client *osdc = osd->o_osdc;
4318 struct ceph_msg *m = NULL;
4319 struct ceph_osd_request *req;
4320 int front_len = le32_to_cpu(hdr->front_len);
4321 int data_len = le32_to_cpu(hdr->data_len);
4322 u64 tid = le64_to_cpu(hdr->tid);
4324 down_read(&osdc->lock);
4325 if (!osd_registered(osd)) {
4326 dout("%s osd%d unknown, skipping\n", __func__, osd->o_osd);
4327 *skip = 1;
4328 goto out_unlock_osdc;
4330 WARN_ON(osd->o_osd != le64_to_cpu(hdr->src.num));
4332 mutex_lock(&osd->lock);
4333 req = lookup_request(&osd->o_requests, tid);
4334 if (!req) {
4335 dout("%s osd%d tid %llu unknown, skipping\n", __func__,
4336 osd->o_osd, tid);
4337 *skip = 1;
4338 goto out_unlock_session;
4341 ceph_msg_revoke_incoming(req->r_reply);
4343 if (front_len > req->r_reply->front_alloc_len) {
4344 pr_warn("%s osd%d tid %llu front %d > preallocated %d\n",
4345 __func__, osd->o_osd, req->r_tid, front_len,
4346 req->r_reply->front_alloc_len);
4347 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS,
4348 false);
4349 if (!m)
4350 goto out_unlock_session;
4351 ceph_msg_put(req->r_reply);
4352 req->r_reply = m;
4355 if (data_len > req->r_reply->data_length) {
4356 pr_warn("%s osd%d tid %llu data %d > preallocated %zu, skipping\n",
4357 __func__, osd->o_osd, req->r_tid, data_len,
4358 req->r_reply->data_length);
4359 m = NULL;
4360 *skip = 1;
4361 goto out_unlock_session;
4364 m = ceph_msg_get(req->r_reply);
4365 dout("get_reply tid %lld %p\n", tid, m);
4367 out_unlock_session:
4368 mutex_unlock(&osd->lock);
4369 out_unlock_osdc:
4370 up_read(&osdc->lock);
4371 return m;
4375 * TODO: switch to a msg-owned pagelist
4377 static struct ceph_msg *alloc_msg_with_page_vector(struct ceph_msg_header *hdr)
4379 struct ceph_msg *m;
4380 int type = le16_to_cpu(hdr->type);
4381 u32 front_len = le32_to_cpu(hdr->front_len);
4382 u32 data_len = le32_to_cpu(hdr->data_len);
4384 m = ceph_msg_new(type, front_len, GFP_NOIO, false);
4385 if (!m)
4386 return NULL;
4388 if (data_len) {
4389 struct page **pages;
4390 struct ceph_osd_data osd_data;
4392 pages = ceph_alloc_page_vector(calc_pages_for(0, data_len),
4393 GFP_NOIO);
4394 if (IS_ERR(pages)) {
4395 ceph_msg_put(m);
4396 return NULL;
4399 ceph_osd_data_pages_init(&osd_data, pages, data_len, 0, false,
4400 false);
4401 ceph_osdc_msg_data_add(m, &osd_data);
4404 return m;
4407 static struct ceph_msg *alloc_msg(struct ceph_connection *con,
4408 struct ceph_msg_header *hdr,
4409 int *skip)
4411 struct ceph_osd *osd = con->private;
4412 int type = le16_to_cpu(hdr->type);
4414 *skip = 0;
4415 switch (type) {
4416 case CEPH_MSG_OSD_MAP:
4417 case CEPH_MSG_WATCH_NOTIFY:
4418 return alloc_msg_with_page_vector(hdr);
4419 case CEPH_MSG_OSD_OPREPLY:
4420 return get_reply(con, hdr, skip);
4421 default:
4422 pr_warn("%s osd%d unknown msg type %d, skipping\n", __func__,
4423 osd->o_osd, type);
4424 *skip = 1;
4425 return NULL;
4430 * Wrappers to refcount containing ceph_osd struct
4432 static struct ceph_connection *get_osd_con(struct ceph_connection *con)
4434 struct ceph_osd *osd = con->private;
4435 if (get_osd(osd))
4436 return con;
4437 return NULL;
4440 static void put_osd_con(struct ceph_connection *con)
4442 struct ceph_osd *osd = con->private;
4443 put_osd(osd);
4447 * authentication
4450 * Note: returned pointer is the address of a structure that's
4451 * managed separately. Caller must *not* attempt to free it.
4453 static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
4454 int *proto, int force_new)
4456 struct ceph_osd *o = con->private;
4457 struct ceph_osd_client *osdc = o->o_osdc;
4458 struct ceph_auth_client *ac = osdc->client->monc.auth;
4459 struct ceph_auth_handshake *auth = &o->o_auth;
4461 if (force_new && auth->authorizer) {
4462 ceph_auth_destroy_authorizer(auth->authorizer);
4463 auth->authorizer = NULL;
4465 if (!auth->authorizer) {
4466 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
4467 auth);
4468 if (ret)
4469 return ERR_PTR(ret);
4470 } else {
4471 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
4472 auth);
4473 if (ret)
4474 return ERR_PTR(ret);
4476 *proto = ac->protocol;
4478 return auth;
4482 static int verify_authorizer_reply(struct ceph_connection *con, int len)
4484 struct ceph_osd *o = con->private;
4485 struct ceph_osd_client *osdc = o->o_osdc;
4486 struct ceph_auth_client *ac = osdc->client->monc.auth;
4488 return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer, len);
4491 static int invalidate_authorizer(struct ceph_connection *con)
4493 struct ceph_osd *o = con->private;
4494 struct ceph_osd_client *osdc = o->o_osdc;
4495 struct ceph_auth_client *ac = osdc->client->monc.auth;
4497 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
4498 return ceph_monc_validate_auth(&osdc->client->monc);
4501 static int osd_sign_message(struct ceph_msg *msg)
4503 struct ceph_osd *o = msg->con->private;
4504 struct ceph_auth_handshake *auth = &o->o_auth;
4506 return ceph_auth_sign_message(auth, msg);
4509 static int osd_check_message_signature(struct ceph_msg *msg)
4511 struct ceph_osd *o = msg->con->private;
4512 struct ceph_auth_handshake *auth = &o->o_auth;
4514 return ceph_auth_check_message_signature(auth, msg);
4517 static const struct ceph_connection_operations osd_con_ops = {
4518 .get = get_osd_con,
4519 .put = put_osd_con,
4520 .dispatch = dispatch,
4521 .get_authorizer = get_authorizer,
4522 .verify_authorizer_reply = verify_authorizer_reply,
4523 .invalidate_authorizer = invalidate_authorizer,
4524 .alloc_msg = alloc_msg,
4525 .sign_message = osd_sign_message,
4526 .check_message_signature = osd_check_message_signature,
4527 .fault = osd_fault,