xref: /linux/net/ceph/osd_client.c (revision 2bc46b3ad3c15165f91459b07ff8682478683194)
1 
2 #include <linux/ceph/ceph_debug.h>
3 
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
14 
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>
21 
22 #define OSD_OPREPLY_FRONT_LEN	512
23 
24 static struct kmem_cache	*ceph_osd_request_cache;
25 
26 static const struct ceph_connection_operations osd_con_ops;
27 
28 /*
29  * Implement client access to distributed object storage cluster.
30  *
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.
36  *
37  * Cluster membership and the mapping of data objects onto storage devices
38  * are described by the osd map.
39  *
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.
44  */
45 
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);
52 
53 #if 1
54 static inline bool rwsem_is_wrlocked(struct rw_semaphore *sem)
55 {
56 	bool wrlocked = true;
57 
58 	if (unlikely(down_read_trylock(sem))) {
59 		wrlocked = false;
60 		up_read(sem);
61 	}
62 
63 	return wrlocked;
64 }
65 static inline void verify_osdc_locked(struct ceph_osd_client *osdc)
66 {
67 	WARN_ON(!rwsem_is_locked(&osdc->lock));
68 }
69 static inline void verify_osdc_wrlocked(struct ceph_osd_client *osdc)
70 {
71 	WARN_ON(!rwsem_is_wrlocked(&osdc->lock));
72 }
73 static inline void verify_osd_locked(struct ceph_osd *osd)
74 {
75 	struct ceph_osd_client *osdc = osd->o_osdc;
76 
77 	WARN_ON(!(mutex_is_locked(&osd->lock) &&
78 		  rwsem_is_locked(&osdc->lock)) &&
79 		!rwsem_is_wrlocked(&osdc->lock));
80 }
81 static inline void verify_lreq_locked(struct ceph_osd_linger_request *lreq)
82 {
83 	WARN_ON(!mutex_is_locked(&lreq->lock));
84 }
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
91 
92 /*
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.
96  *
97  * fill osd op in request message.
98  */
99 static int calc_layout(struct ceph_file_layout *layout, u64 off, u64 *plen,
100 			u64 *objnum, u64 *objoff, u64 *objlen)
101 {
102 	u64 orig_len = *plen;
103 	int r;
104 
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);
114 	}
115 
116 	dout("calc_layout objnum=%llx %llu~%llu\n", *objnum, *objoff, *objlen);
117 
118 	return 0;
119 }
120 
121 static void ceph_osd_data_init(struct ceph_osd_data *osd_data)
122 {
123 	memset(osd_data, 0, sizeof (*osd_data));
124 	osd_data->type = CEPH_OSD_DATA_TYPE_NONE;
125 }
126 
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)
130 {
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;
137 }
138 
139 static void ceph_osd_data_pagelist_init(struct ceph_osd_data *osd_data,
140 			struct ceph_pagelist *pagelist)
141 {
142 	osd_data->type = CEPH_OSD_DATA_TYPE_PAGELIST;
143 	osd_data->pagelist = pagelist;
144 }
145 
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)
149 {
150 	osd_data->type = CEPH_OSD_DATA_TYPE_BIO;
151 	osd_data->bio = bio;
152 	osd_data->bio_length = bio_length;
153 }
154 #endif /* CONFIG_BLOCK */
155 
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;					\
162 })
163 
164 static struct ceph_osd_data *
165 osd_req_op_raw_data_in(struct ceph_osd_request *osd_req, unsigned int which)
166 {
167 	BUG_ON(which >= osd_req->r_num_ops);
168 
169 	return &osd_req->r_ops[which].raw_data_in;
170 }
171 
172 struct ceph_osd_data *
173 osd_req_op_extent_osd_data(struct ceph_osd_request *osd_req,
174 			unsigned int which)
175 {
176 	return osd_req_op_data(osd_req, which, extent, osd_data);
177 }
178 EXPORT_SYMBOL(osd_req_op_extent_osd_data);
179 
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)
184 {
185 	struct ceph_osd_data *osd_data;
186 
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);
190 }
191 EXPORT_SYMBOL(osd_req_op_raw_data_in_pages);
192 
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)
197 {
198 	struct ceph_osd_data *osd_data;
199 
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);
203 }
204 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pages);
205 
206 void osd_req_op_extent_osd_data_pagelist(struct ceph_osd_request *osd_req,
207 			unsigned int which, struct ceph_pagelist *pagelist)
208 {
209 	struct ceph_osd_data *osd_data;
210 
211 	osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
212 	ceph_osd_data_pagelist_init(osd_data, pagelist);
213 }
214 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pagelist);
215 
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)
219 {
220 	struct ceph_osd_data *osd_data;
221 
222 	osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
223 	ceph_osd_data_bio_init(osd_data, bio, bio_length);
224 }
225 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bio);
226 #endif /* CONFIG_BLOCK */
227 
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)
231 {
232 	struct ceph_osd_data *osd_data;
233 
234 	osd_data = osd_req_op_data(osd_req, which, cls, request_info);
235 	ceph_osd_data_pagelist_init(osd_data, pagelist);
236 }
237 
238 void osd_req_op_cls_request_data_pagelist(
239 			struct ceph_osd_request *osd_req,
240 			unsigned int which, struct ceph_pagelist *pagelist)
241 {
242 	struct ceph_osd_data *osd_data;
243 
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;
248 }
249 EXPORT_SYMBOL(osd_req_op_cls_request_data_pagelist);
250 
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)
254 {
255 	struct ceph_osd_data *osd_data;
256 
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;
262 }
263 EXPORT_SYMBOL(osd_req_op_cls_request_data_pages);
264 
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)
268 {
269 	struct ceph_osd_data *osd_data;
270 
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);
274 }
275 EXPORT_SYMBOL(osd_req_op_cls_response_data_pages);
276 
277 static u64 ceph_osd_data_length(struct ceph_osd_data *osd_data)
278 {
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;
293 	}
294 }
295 
296 static void ceph_osd_data_release(struct ceph_osd_data *osd_data)
297 {
298 	if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES && osd_data->own_pages) {
299 		int num_pages;
300 
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);
304 	}
305 	ceph_osd_data_init(osd_data);
306 }
307 
308 static void osd_req_op_data_release(struct ceph_osd_request *osd_req,
309 			unsigned int which)
310 {
311 	struct ceph_osd_req_op *op;
312 
313 	BUG_ON(which >= osd_req->r_num_ops);
314 	op = &osd_req->r_ops[which];
315 
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 	default:
342 		break;
343 	}
344 }
345 
346 /*
347  * Assumes @t is zero-initialized.
348  */
349 static void target_init(struct ceph_osd_request_target *t)
350 {
351 	ceph_oid_init(&t->base_oid);
352 	ceph_oloc_init(&t->base_oloc);
353 	ceph_oid_init(&t->target_oid);
354 	ceph_oloc_init(&t->target_oloc);
355 
356 	ceph_osds_init(&t->acting);
357 	ceph_osds_init(&t->up);
358 	t->size = -1;
359 	t->min_size = -1;
360 
361 	t->osd = CEPH_HOMELESS_OSD;
362 }
363 
364 static void target_copy(struct ceph_osd_request_target *dest,
365 			const struct ceph_osd_request_target *src)
366 {
367 	ceph_oid_copy(&dest->base_oid, &src->base_oid);
368 	ceph_oloc_copy(&dest->base_oloc, &src->base_oloc);
369 	ceph_oid_copy(&dest->target_oid, &src->target_oid);
370 	ceph_oloc_copy(&dest->target_oloc, &src->target_oloc);
371 
372 	dest->pgid = src->pgid; /* struct */
373 	dest->pg_num = src->pg_num;
374 	dest->pg_num_mask = src->pg_num_mask;
375 	ceph_osds_copy(&dest->acting, &src->acting);
376 	ceph_osds_copy(&dest->up, &src->up);
377 	dest->size = src->size;
378 	dest->min_size = src->min_size;
379 	dest->sort_bitwise = src->sort_bitwise;
380 
381 	dest->flags = src->flags;
382 	dest->paused = src->paused;
383 
384 	dest->osd = src->osd;
385 }
386 
387 static void target_destroy(struct ceph_osd_request_target *t)
388 {
389 	ceph_oid_destroy(&t->base_oid);
390 	ceph_oid_destroy(&t->target_oid);
391 }
392 
393 /*
394  * requests
395  */
396 static void request_release_checks(struct ceph_osd_request *req)
397 {
398 	WARN_ON(!RB_EMPTY_NODE(&req->r_node));
399 	WARN_ON(!RB_EMPTY_NODE(&req->r_mc_node));
400 	WARN_ON(!list_empty(&req->r_unsafe_item));
401 	WARN_ON(req->r_osd);
402 }
403 
404 static void ceph_osdc_release_request(struct kref *kref)
405 {
406 	struct ceph_osd_request *req = container_of(kref,
407 					    struct ceph_osd_request, r_kref);
408 	unsigned int which;
409 
410 	dout("%s %p (r_request %p r_reply %p)\n", __func__, req,
411 	     req->r_request, req->r_reply);
412 	request_release_checks(req);
413 
414 	if (req->r_request)
415 		ceph_msg_put(req->r_request);
416 	if (req->r_reply)
417 		ceph_msg_put(req->r_reply);
418 
419 	for (which = 0; which < req->r_num_ops; which++)
420 		osd_req_op_data_release(req, which);
421 
422 	target_destroy(&req->r_t);
423 	ceph_put_snap_context(req->r_snapc);
424 
425 	if (req->r_mempool)
426 		mempool_free(req, req->r_osdc->req_mempool);
427 	else if (req->r_num_ops <= CEPH_OSD_SLAB_OPS)
428 		kmem_cache_free(ceph_osd_request_cache, req);
429 	else
430 		kfree(req);
431 }
432 
433 void ceph_osdc_get_request(struct ceph_osd_request *req)
434 {
435 	dout("%s %p (was %d)\n", __func__, req,
436 	     atomic_read(&req->r_kref.refcount));
437 	kref_get(&req->r_kref);
438 }
439 EXPORT_SYMBOL(ceph_osdc_get_request);
440 
441 void ceph_osdc_put_request(struct ceph_osd_request *req)
442 {
443 	if (req) {
444 		dout("%s %p (was %d)\n", __func__, req,
445 		     atomic_read(&req->r_kref.refcount));
446 		kref_put(&req->r_kref, ceph_osdc_release_request);
447 	}
448 }
449 EXPORT_SYMBOL(ceph_osdc_put_request);
450 
451 static void request_init(struct ceph_osd_request *req)
452 {
453 	/* req only, each op is zeroed in _osd_req_op_init() */
454 	memset(req, 0, sizeof(*req));
455 
456 	kref_init(&req->r_kref);
457 	init_completion(&req->r_completion);
458 	init_completion(&req->r_safe_completion);
459 	RB_CLEAR_NODE(&req->r_node);
460 	RB_CLEAR_NODE(&req->r_mc_node);
461 	INIT_LIST_HEAD(&req->r_unsafe_item);
462 
463 	target_init(&req->r_t);
464 }
465 
466 /*
467  * This is ugly, but it allows us to reuse linger registration and ping
468  * requests, keeping the structure of the code around send_linger{_ping}()
469  * reasonable.  Setting up a min_nr=2 mempool for each linger request
470  * and dealing with copying ops (this blasts req only, watch op remains
471  * intact) isn't any better.
472  */
473 static void request_reinit(struct ceph_osd_request *req)
474 {
475 	struct ceph_osd_client *osdc = req->r_osdc;
476 	bool mempool = req->r_mempool;
477 	unsigned int num_ops = req->r_num_ops;
478 	u64 snapid = req->r_snapid;
479 	struct ceph_snap_context *snapc = req->r_snapc;
480 	bool linger = req->r_linger;
481 	struct ceph_msg *request_msg = req->r_request;
482 	struct ceph_msg *reply_msg = req->r_reply;
483 
484 	dout("%s req %p\n", __func__, req);
485 	WARN_ON(atomic_read(&req->r_kref.refcount) != 1);
486 	request_release_checks(req);
487 
488 	WARN_ON(atomic_read(&request_msg->kref.refcount) != 1);
489 	WARN_ON(atomic_read(&reply_msg->kref.refcount) != 1);
490 	target_destroy(&req->r_t);
491 
492 	request_init(req);
493 	req->r_osdc = osdc;
494 	req->r_mempool = mempool;
495 	req->r_num_ops = num_ops;
496 	req->r_snapid = snapid;
497 	req->r_snapc = snapc;
498 	req->r_linger = linger;
499 	req->r_request = request_msg;
500 	req->r_reply = reply_msg;
501 }
502 
503 struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
504 					       struct ceph_snap_context *snapc,
505 					       unsigned int num_ops,
506 					       bool use_mempool,
507 					       gfp_t gfp_flags)
508 {
509 	struct ceph_osd_request *req;
510 
511 	if (use_mempool) {
512 		BUG_ON(num_ops > CEPH_OSD_SLAB_OPS);
513 		req = mempool_alloc(osdc->req_mempool, gfp_flags);
514 	} else if (num_ops <= CEPH_OSD_SLAB_OPS) {
515 		req = kmem_cache_alloc(ceph_osd_request_cache, gfp_flags);
516 	} else {
517 		BUG_ON(num_ops > CEPH_OSD_MAX_OPS);
518 		req = kmalloc(sizeof(*req) + num_ops * sizeof(req->r_ops[0]),
519 			      gfp_flags);
520 	}
521 	if (unlikely(!req))
522 		return NULL;
523 
524 	request_init(req);
525 	req->r_osdc = osdc;
526 	req->r_mempool = use_mempool;
527 	req->r_num_ops = num_ops;
528 	req->r_snapid = CEPH_NOSNAP;
529 	req->r_snapc = ceph_get_snap_context(snapc);
530 
531 	dout("%s req %p\n", __func__, req);
532 	return req;
533 }
534 EXPORT_SYMBOL(ceph_osdc_alloc_request);
535 
536 int ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp)
537 {
538 	struct ceph_osd_client *osdc = req->r_osdc;
539 	struct ceph_msg *msg;
540 	int msg_size;
541 
542 	WARN_ON(ceph_oid_empty(&req->r_base_oid));
543 
544 	/* create request message */
545 	msg_size = 4 + 4 + 4; /* client_inc, osdmap_epoch, flags */
546 	msg_size += 4 + 4 + 4 + 8; /* mtime, reassert_version */
547 	msg_size += 2 + 4 + 8 + 4 + 4; /* oloc */
548 	msg_size += 1 + 8 + 4 + 4; /* pgid */
549 	msg_size += 4 + req->r_base_oid.name_len; /* oid */
550 	msg_size += 2 + req->r_num_ops * sizeof(struct ceph_osd_op);
551 	msg_size += 8; /* snapid */
552 	msg_size += 8; /* snap_seq */
553 	msg_size += 4 + 8 * (req->r_snapc ? req->r_snapc->num_snaps : 0);
554 	msg_size += 4; /* retry_attempt */
555 
556 	if (req->r_mempool)
557 		msg = ceph_msgpool_get(&osdc->msgpool_op, 0);
558 	else
559 		msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp, true);
560 	if (!msg)
561 		return -ENOMEM;
562 
563 	memset(msg->front.iov_base, 0, msg->front.iov_len);
564 	req->r_request = msg;
565 
566 	/* create reply message */
567 	msg_size = OSD_OPREPLY_FRONT_LEN;
568 	msg_size += req->r_base_oid.name_len;
569 	msg_size += req->r_num_ops * sizeof(struct ceph_osd_op);
570 
571 	if (req->r_mempool)
572 		msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0);
573 	else
574 		msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, msg_size, gfp, true);
575 	if (!msg)
576 		return -ENOMEM;
577 
578 	req->r_reply = msg;
579 
580 	return 0;
581 }
582 EXPORT_SYMBOL(ceph_osdc_alloc_messages);
583 
584 static bool osd_req_opcode_valid(u16 opcode)
585 {
586 	switch (opcode) {
587 #define GENERATE_CASE(op, opcode, str)	case CEPH_OSD_OP_##op: return true;
588 __CEPH_FORALL_OSD_OPS(GENERATE_CASE)
589 #undef GENERATE_CASE
590 	default:
591 		return false;
592 	}
593 }
594 
595 /*
596  * This is an osd op init function for opcodes that have no data or
597  * other information associated with them.  It also serves as a
598  * common init routine for all the other init functions, below.
599  */
600 static struct ceph_osd_req_op *
601 _osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which,
602 		 u16 opcode, u32 flags)
603 {
604 	struct ceph_osd_req_op *op;
605 
606 	BUG_ON(which >= osd_req->r_num_ops);
607 	BUG_ON(!osd_req_opcode_valid(opcode));
608 
609 	op = &osd_req->r_ops[which];
610 	memset(op, 0, sizeof (*op));
611 	op->op = opcode;
612 	op->flags = flags;
613 
614 	return op;
615 }
616 
617 void osd_req_op_init(struct ceph_osd_request *osd_req,
618 		     unsigned int which, u16 opcode, u32 flags)
619 {
620 	(void)_osd_req_op_init(osd_req, which, opcode, flags);
621 }
622 EXPORT_SYMBOL(osd_req_op_init);
623 
624 void osd_req_op_extent_init(struct ceph_osd_request *osd_req,
625 				unsigned int which, u16 opcode,
626 				u64 offset, u64 length,
627 				u64 truncate_size, u32 truncate_seq)
628 {
629 	struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
630 						      opcode, 0);
631 	size_t payload_len = 0;
632 
633 	BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
634 	       opcode != CEPH_OSD_OP_WRITEFULL && opcode != CEPH_OSD_OP_ZERO &&
635 	       opcode != CEPH_OSD_OP_TRUNCATE);
636 
637 	op->extent.offset = offset;
638 	op->extent.length = length;
639 	op->extent.truncate_size = truncate_size;
640 	op->extent.truncate_seq = truncate_seq;
641 	if (opcode == CEPH_OSD_OP_WRITE || opcode == CEPH_OSD_OP_WRITEFULL)
642 		payload_len += length;
643 
644 	op->indata_len = payload_len;
645 }
646 EXPORT_SYMBOL(osd_req_op_extent_init);
647 
648 void osd_req_op_extent_update(struct ceph_osd_request *osd_req,
649 				unsigned int which, u64 length)
650 {
651 	struct ceph_osd_req_op *op;
652 	u64 previous;
653 
654 	BUG_ON(which >= osd_req->r_num_ops);
655 	op = &osd_req->r_ops[which];
656 	previous = op->extent.length;
657 
658 	if (length == previous)
659 		return;		/* Nothing to do */
660 	BUG_ON(length > previous);
661 
662 	op->extent.length = length;
663 	op->indata_len -= previous - length;
664 }
665 EXPORT_SYMBOL(osd_req_op_extent_update);
666 
667 void osd_req_op_extent_dup_last(struct ceph_osd_request *osd_req,
668 				unsigned int which, u64 offset_inc)
669 {
670 	struct ceph_osd_req_op *op, *prev_op;
671 
672 	BUG_ON(which + 1 >= osd_req->r_num_ops);
673 
674 	prev_op = &osd_req->r_ops[which];
675 	op = _osd_req_op_init(osd_req, which + 1, prev_op->op, prev_op->flags);
676 	/* dup previous one */
677 	op->indata_len = prev_op->indata_len;
678 	op->outdata_len = prev_op->outdata_len;
679 	op->extent = prev_op->extent;
680 	/* adjust offset */
681 	op->extent.offset += offset_inc;
682 	op->extent.length -= offset_inc;
683 
684 	if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL)
685 		op->indata_len -= offset_inc;
686 }
687 EXPORT_SYMBOL(osd_req_op_extent_dup_last);
688 
689 void osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
690 			u16 opcode, const char *class, const char *method)
691 {
692 	struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
693 						      opcode, 0);
694 	struct ceph_pagelist *pagelist;
695 	size_t payload_len = 0;
696 	size_t size;
697 
698 	BUG_ON(opcode != CEPH_OSD_OP_CALL);
699 
700 	pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
701 	BUG_ON(!pagelist);
702 	ceph_pagelist_init(pagelist);
703 
704 	op->cls.class_name = class;
705 	size = strlen(class);
706 	BUG_ON(size > (size_t) U8_MAX);
707 	op->cls.class_len = size;
708 	ceph_pagelist_append(pagelist, class, size);
709 	payload_len += size;
710 
711 	op->cls.method_name = method;
712 	size = strlen(method);
713 	BUG_ON(size > (size_t) U8_MAX);
714 	op->cls.method_len = size;
715 	ceph_pagelist_append(pagelist, method, size);
716 	payload_len += size;
717 
718 	osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist);
719 
720 	op->indata_len = payload_len;
721 }
722 EXPORT_SYMBOL(osd_req_op_cls_init);
723 
724 int osd_req_op_xattr_init(struct ceph_osd_request *osd_req, unsigned int which,
725 			  u16 opcode, const char *name, const void *value,
726 			  size_t size, u8 cmp_op, u8 cmp_mode)
727 {
728 	struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
729 						      opcode, 0);
730 	struct ceph_pagelist *pagelist;
731 	size_t payload_len;
732 
733 	BUG_ON(opcode != CEPH_OSD_OP_SETXATTR && opcode != CEPH_OSD_OP_CMPXATTR);
734 
735 	pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
736 	if (!pagelist)
737 		return -ENOMEM;
738 
739 	ceph_pagelist_init(pagelist);
740 
741 	payload_len = strlen(name);
742 	op->xattr.name_len = payload_len;
743 	ceph_pagelist_append(pagelist, name, payload_len);
744 
745 	op->xattr.value_len = size;
746 	ceph_pagelist_append(pagelist, value, size);
747 	payload_len += size;
748 
749 	op->xattr.cmp_op = cmp_op;
750 	op->xattr.cmp_mode = cmp_mode;
751 
752 	ceph_osd_data_pagelist_init(&op->xattr.osd_data, pagelist);
753 	op->indata_len = payload_len;
754 	return 0;
755 }
756 EXPORT_SYMBOL(osd_req_op_xattr_init);
757 
758 /*
759  * @watch_opcode: CEPH_OSD_WATCH_OP_*
760  */
761 static void osd_req_op_watch_init(struct ceph_osd_request *req, int which,
762 				  u64 cookie, u8 watch_opcode)
763 {
764 	struct ceph_osd_req_op *op;
765 
766 	op = _osd_req_op_init(req, which, CEPH_OSD_OP_WATCH, 0);
767 	op->watch.cookie = cookie;
768 	op->watch.op = watch_opcode;
769 	op->watch.gen = 0;
770 }
771 
772 void osd_req_op_alloc_hint_init(struct ceph_osd_request *osd_req,
773 				unsigned int which,
774 				u64 expected_object_size,
775 				u64 expected_write_size)
776 {
777 	struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
778 						      CEPH_OSD_OP_SETALLOCHINT,
779 						      0);
780 
781 	op->alloc_hint.expected_object_size = expected_object_size;
782 	op->alloc_hint.expected_write_size = expected_write_size;
783 
784 	/*
785 	 * CEPH_OSD_OP_SETALLOCHINT op is advisory and therefore deemed
786 	 * not worth a feature bit.  Set FAILOK per-op flag to make
787 	 * sure older osds don't trip over an unsupported opcode.
788 	 */
789 	op->flags |= CEPH_OSD_OP_FLAG_FAILOK;
790 }
791 EXPORT_SYMBOL(osd_req_op_alloc_hint_init);
792 
793 static void ceph_osdc_msg_data_add(struct ceph_msg *msg,
794 				struct ceph_osd_data *osd_data)
795 {
796 	u64 length = ceph_osd_data_length(osd_data);
797 
798 	if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
799 		BUG_ON(length > (u64) SIZE_MAX);
800 		if (length)
801 			ceph_msg_data_add_pages(msg, osd_data->pages,
802 					length, osd_data->alignment);
803 	} else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
804 		BUG_ON(!length);
805 		ceph_msg_data_add_pagelist(msg, osd_data->pagelist);
806 #ifdef CONFIG_BLOCK
807 	} else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
808 		ceph_msg_data_add_bio(msg, osd_data->bio, length);
809 #endif
810 	} else {
811 		BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
812 	}
813 }
814 
815 static u32 osd_req_encode_op(struct ceph_osd_op *dst,
816 			     const struct ceph_osd_req_op *src)
817 {
818 	if (WARN_ON(!osd_req_opcode_valid(src->op))) {
819 		pr_err("unrecognized osd opcode %d\n", src->op);
820 
821 		return 0;
822 	}
823 
824 	switch (src->op) {
825 	case CEPH_OSD_OP_STAT:
826 		break;
827 	case CEPH_OSD_OP_READ:
828 	case CEPH_OSD_OP_WRITE:
829 	case CEPH_OSD_OP_WRITEFULL:
830 	case CEPH_OSD_OP_ZERO:
831 	case CEPH_OSD_OP_TRUNCATE:
832 		dst->extent.offset = cpu_to_le64(src->extent.offset);
833 		dst->extent.length = cpu_to_le64(src->extent.length);
834 		dst->extent.truncate_size =
835 			cpu_to_le64(src->extent.truncate_size);
836 		dst->extent.truncate_seq =
837 			cpu_to_le32(src->extent.truncate_seq);
838 		break;
839 	case CEPH_OSD_OP_CALL:
840 		dst->cls.class_len = src->cls.class_len;
841 		dst->cls.method_len = src->cls.method_len;
842 		dst->cls.indata_len = cpu_to_le32(src->cls.indata_len);
843 		break;
844 	case CEPH_OSD_OP_STARTSYNC:
845 		break;
846 	case CEPH_OSD_OP_WATCH:
847 		dst->watch.cookie = cpu_to_le64(src->watch.cookie);
848 		dst->watch.ver = cpu_to_le64(0);
849 		dst->watch.op = src->watch.op;
850 		dst->watch.gen = cpu_to_le32(src->watch.gen);
851 		break;
852 	case CEPH_OSD_OP_NOTIFY_ACK:
853 		break;
854 	case CEPH_OSD_OP_NOTIFY:
855 		dst->notify.cookie = cpu_to_le64(src->notify.cookie);
856 		break;
857 	case CEPH_OSD_OP_SETALLOCHINT:
858 		dst->alloc_hint.expected_object_size =
859 		    cpu_to_le64(src->alloc_hint.expected_object_size);
860 		dst->alloc_hint.expected_write_size =
861 		    cpu_to_le64(src->alloc_hint.expected_write_size);
862 		break;
863 	case CEPH_OSD_OP_SETXATTR:
864 	case CEPH_OSD_OP_CMPXATTR:
865 		dst->xattr.name_len = cpu_to_le32(src->xattr.name_len);
866 		dst->xattr.value_len = cpu_to_le32(src->xattr.value_len);
867 		dst->xattr.cmp_op = src->xattr.cmp_op;
868 		dst->xattr.cmp_mode = src->xattr.cmp_mode;
869 		break;
870 	case CEPH_OSD_OP_CREATE:
871 	case CEPH_OSD_OP_DELETE:
872 		break;
873 	default:
874 		pr_err("unsupported osd opcode %s\n",
875 			ceph_osd_op_name(src->op));
876 		WARN_ON(1);
877 
878 		return 0;
879 	}
880 
881 	dst->op = cpu_to_le16(src->op);
882 	dst->flags = cpu_to_le32(src->flags);
883 	dst->payload_len = cpu_to_le32(src->indata_len);
884 
885 	return src->indata_len;
886 }
887 
888 /*
889  * build new request AND message, calculate layout, and adjust file
890  * extent as needed.
891  *
892  * if the file was recently truncated, we include information about its
893  * old and new size so that the object can be updated appropriately.  (we
894  * avoid synchronously deleting truncated objects because it's slow.)
895  *
896  * if @do_sync, include a 'startsync' command so that the osd will flush
897  * data quickly.
898  */
899 struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
900 					       struct ceph_file_layout *layout,
901 					       struct ceph_vino vino,
902 					       u64 off, u64 *plen,
903 					       unsigned int which, int num_ops,
904 					       int opcode, int flags,
905 					       struct ceph_snap_context *snapc,
906 					       u32 truncate_seq,
907 					       u64 truncate_size,
908 					       bool use_mempool)
909 {
910 	struct ceph_osd_request *req;
911 	u64 objnum = 0;
912 	u64 objoff = 0;
913 	u64 objlen = 0;
914 	int r;
915 
916 	BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
917 	       opcode != CEPH_OSD_OP_ZERO && opcode != CEPH_OSD_OP_TRUNCATE &&
918 	       opcode != CEPH_OSD_OP_CREATE && opcode != CEPH_OSD_OP_DELETE);
919 
920 	req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool,
921 					GFP_NOFS);
922 	if (!req) {
923 		r = -ENOMEM;
924 		goto fail;
925 	}
926 
927 	/* calculate max write size */
928 	r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
929 	if (r)
930 		goto fail;
931 
932 	if (opcode == CEPH_OSD_OP_CREATE || opcode == CEPH_OSD_OP_DELETE) {
933 		osd_req_op_init(req, which, opcode, 0);
934 	} else {
935 		u32 object_size = le32_to_cpu(layout->fl_object_size);
936 		u32 object_base = off - objoff;
937 		if (!(truncate_seq == 1 && truncate_size == -1ULL)) {
938 			if (truncate_size <= object_base) {
939 				truncate_size = 0;
940 			} else {
941 				truncate_size -= object_base;
942 				if (truncate_size > object_size)
943 					truncate_size = object_size;
944 			}
945 		}
946 		osd_req_op_extent_init(req, which, opcode, objoff, objlen,
947 				       truncate_size, truncate_seq);
948 	}
949 
950 	req->r_flags = flags;
951 	req->r_base_oloc.pool = ceph_file_layout_pg_pool(*layout);
952 	ceph_oid_printf(&req->r_base_oid, "%llx.%08llx", vino.ino, objnum);
953 
954 	req->r_snapid = vino.snap;
955 	if (flags & CEPH_OSD_FLAG_WRITE)
956 		req->r_data_offset = off;
957 
958 	r = ceph_osdc_alloc_messages(req, GFP_NOFS);
959 	if (r)
960 		goto fail;
961 
962 	return req;
963 
964 fail:
965 	ceph_osdc_put_request(req);
966 	return ERR_PTR(r);
967 }
968 EXPORT_SYMBOL(ceph_osdc_new_request);
969 
970 /*
971  * We keep osd requests in an rbtree, sorted by ->r_tid.
972  */
973 DEFINE_RB_FUNCS(request, struct ceph_osd_request, r_tid, r_node)
974 DEFINE_RB_FUNCS(request_mc, struct ceph_osd_request, r_tid, r_mc_node)
975 
976 static bool osd_homeless(struct ceph_osd *osd)
977 {
978 	return osd->o_osd == CEPH_HOMELESS_OSD;
979 }
980 
981 static bool osd_registered(struct ceph_osd *osd)
982 {
983 	verify_osdc_locked(osd->o_osdc);
984 
985 	return !RB_EMPTY_NODE(&osd->o_node);
986 }
987 
988 /*
989  * Assumes @osd is zero-initialized.
990  */
991 static void osd_init(struct ceph_osd *osd)
992 {
993 	atomic_set(&osd->o_ref, 1);
994 	RB_CLEAR_NODE(&osd->o_node);
995 	osd->o_requests = RB_ROOT;
996 	osd->o_linger_requests = RB_ROOT;
997 	INIT_LIST_HEAD(&osd->o_osd_lru);
998 	INIT_LIST_HEAD(&osd->o_keepalive_item);
999 	osd->o_incarnation = 1;
1000 	mutex_init(&osd->lock);
1001 }
1002 
1003 static void osd_cleanup(struct ceph_osd *osd)
1004 {
1005 	WARN_ON(!RB_EMPTY_NODE(&osd->o_node));
1006 	WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests));
1007 	WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests));
1008 	WARN_ON(!list_empty(&osd->o_osd_lru));
1009 	WARN_ON(!list_empty(&osd->o_keepalive_item));
1010 
1011 	if (osd->o_auth.authorizer) {
1012 		WARN_ON(osd_homeless(osd));
1013 		ceph_auth_destroy_authorizer(osd->o_auth.authorizer);
1014 	}
1015 }
1016 
1017 /*
1018  * Track open sessions with osds.
1019  */
1020 static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
1021 {
1022 	struct ceph_osd *osd;
1023 
1024 	WARN_ON(onum == CEPH_HOMELESS_OSD);
1025 
1026 	osd = kzalloc(sizeof(*osd), GFP_NOIO | __GFP_NOFAIL);
1027 	osd_init(osd);
1028 	osd->o_osdc = osdc;
1029 	osd->o_osd = onum;
1030 
1031 	ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
1032 
1033 	return osd;
1034 }
1035 
1036 static struct ceph_osd *get_osd(struct ceph_osd *osd)
1037 {
1038 	if (atomic_inc_not_zero(&osd->o_ref)) {
1039 		dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
1040 		     atomic_read(&osd->o_ref));
1041 		return osd;
1042 	} else {
1043 		dout("get_osd %p FAIL\n", osd);
1044 		return NULL;
1045 	}
1046 }
1047 
1048 static void put_osd(struct ceph_osd *osd)
1049 {
1050 	dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
1051 	     atomic_read(&osd->o_ref) - 1);
1052 	if (atomic_dec_and_test(&osd->o_ref)) {
1053 		osd_cleanup(osd);
1054 		kfree(osd);
1055 	}
1056 }
1057 
1058 DEFINE_RB_FUNCS(osd, struct ceph_osd, o_osd, o_node)
1059 
1060 static void __move_osd_to_lru(struct ceph_osd *osd)
1061 {
1062 	struct ceph_osd_client *osdc = osd->o_osdc;
1063 
1064 	dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1065 	BUG_ON(!list_empty(&osd->o_osd_lru));
1066 
1067 	spin_lock(&osdc->osd_lru_lock);
1068 	list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
1069 	spin_unlock(&osdc->osd_lru_lock);
1070 
1071 	osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl;
1072 }
1073 
1074 static void maybe_move_osd_to_lru(struct ceph_osd *osd)
1075 {
1076 	if (RB_EMPTY_ROOT(&osd->o_requests) &&
1077 	    RB_EMPTY_ROOT(&osd->o_linger_requests))
1078 		__move_osd_to_lru(osd);
1079 }
1080 
1081 static void __remove_osd_from_lru(struct ceph_osd *osd)
1082 {
1083 	struct ceph_osd_client *osdc = osd->o_osdc;
1084 
1085 	dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1086 
1087 	spin_lock(&osdc->osd_lru_lock);
1088 	if (!list_empty(&osd->o_osd_lru))
1089 		list_del_init(&osd->o_osd_lru);
1090 	spin_unlock(&osdc->osd_lru_lock);
1091 }
1092 
1093 /*
1094  * Close the connection and assign any leftover requests to the
1095  * homeless session.
1096  */
1097 static void close_osd(struct ceph_osd *osd)
1098 {
1099 	struct ceph_osd_client *osdc = osd->o_osdc;
1100 	struct rb_node *n;
1101 
1102 	verify_osdc_wrlocked(osdc);
1103 	dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1104 
1105 	ceph_con_close(&osd->o_con);
1106 
1107 	for (n = rb_first(&osd->o_requests); n; ) {
1108 		struct ceph_osd_request *req =
1109 		    rb_entry(n, struct ceph_osd_request, r_node);
1110 
1111 		n = rb_next(n); /* unlink_request() */
1112 
1113 		dout(" reassigning req %p tid %llu\n", req, req->r_tid);
1114 		unlink_request(osd, req);
1115 		link_request(&osdc->homeless_osd, req);
1116 	}
1117 	for (n = rb_first(&osd->o_linger_requests); n; ) {
1118 		struct ceph_osd_linger_request *lreq =
1119 		    rb_entry(n, struct ceph_osd_linger_request, node);
1120 
1121 		n = rb_next(n); /* unlink_linger() */
1122 
1123 		dout(" reassigning lreq %p linger_id %llu\n", lreq,
1124 		     lreq->linger_id);
1125 		unlink_linger(osd, lreq);
1126 		link_linger(&osdc->homeless_osd, lreq);
1127 	}
1128 
1129 	__remove_osd_from_lru(osd);
1130 	erase_osd(&osdc->osds, osd);
1131 	put_osd(osd);
1132 }
1133 
1134 /*
1135  * reset osd connect
1136  */
1137 static int reopen_osd(struct ceph_osd *osd)
1138 {
1139 	struct ceph_entity_addr *peer_addr;
1140 
1141 	dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1142 
1143 	if (RB_EMPTY_ROOT(&osd->o_requests) &&
1144 	    RB_EMPTY_ROOT(&osd->o_linger_requests)) {
1145 		close_osd(osd);
1146 		return -ENODEV;
1147 	}
1148 
1149 	peer_addr = &osd->o_osdc->osdmap->osd_addr[osd->o_osd];
1150 	if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1151 			!ceph_con_opened(&osd->o_con)) {
1152 		struct rb_node *n;
1153 
1154 		dout("osd addr hasn't changed and connection never opened, "
1155 		     "letting msgr retry\n");
1156 		/* touch each r_stamp for handle_timeout()'s benfit */
1157 		for (n = rb_first(&osd->o_requests); n; n = rb_next(n)) {
1158 			struct ceph_osd_request *req =
1159 			    rb_entry(n, struct ceph_osd_request, r_node);
1160 			req->r_stamp = jiffies;
1161 		}
1162 
1163 		return -EAGAIN;
1164 	}
1165 
1166 	ceph_con_close(&osd->o_con);
1167 	ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1168 	osd->o_incarnation++;
1169 
1170 	return 0;
1171 }
1172 
1173 static struct ceph_osd *lookup_create_osd(struct ceph_osd_client *osdc, int o,
1174 					  bool wrlocked)
1175 {
1176 	struct ceph_osd *osd;
1177 
1178 	if (wrlocked)
1179 		verify_osdc_wrlocked(osdc);
1180 	else
1181 		verify_osdc_locked(osdc);
1182 
1183 	if (o != CEPH_HOMELESS_OSD)
1184 		osd = lookup_osd(&osdc->osds, o);
1185 	else
1186 		osd = &osdc->homeless_osd;
1187 	if (!osd) {
1188 		if (!wrlocked)
1189 			return ERR_PTR(-EAGAIN);
1190 
1191 		osd = create_osd(osdc, o);
1192 		insert_osd(&osdc->osds, osd);
1193 		ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd,
1194 			      &osdc->osdmap->osd_addr[osd->o_osd]);
1195 	}
1196 
1197 	dout("%s osdc %p osd%d -> osd %p\n", __func__, osdc, o, osd);
1198 	return osd;
1199 }
1200 
1201 /*
1202  * Create request <-> OSD session relation.
1203  *
1204  * @req has to be assigned a tid, @osd may be homeless.
1205  */
1206 static void link_request(struct ceph_osd *osd, struct ceph_osd_request *req)
1207 {
1208 	verify_osd_locked(osd);
1209 	WARN_ON(!req->r_tid || req->r_osd);
1210 	dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd,
1211 	     req, req->r_tid);
1212 
1213 	if (!osd_homeless(osd))
1214 		__remove_osd_from_lru(osd);
1215 	else
1216 		atomic_inc(&osd->o_osdc->num_homeless);
1217 
1218 	get_osd(osd);
1219 	insert_request(&osd->o_requests, req);
1220 	req->r_osd = osd;
1221 }
1222 
1223 static void unlink_request(struct ceph_osd *osd, struct ceph_osd_request *req)
1224 {
1225 	verify_osd_locked(osd);
1226 	WARN_ON(req->r_osd != osd);
1227 	dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd,
1228 	     req, req->r_tid);
1229 
1230 	req->r_osd = NULL;
1231 	erase_request(&osd->o_requests, req);
1232 	put_osd(osd);
1233 
1234 	if (!osd_homeless(osd))
1235 		maybe_move_osd_to_lru(osd);
1236 	else
1237 		atomic_dec(&osd->o_osdc->num_homeless);
1238 }
1239 
1240 static bool __pool_full(struct ceph_pg_pool_info *pi)
1241 {
1242 	return pi->flags & CEPH_POOL_FLAG_FULL;
1243 }
1244 
1245 static bool have_pool_full(struct ceph_osd_client *osdc)
1246 {
1247 	struct rb_node *n;
1248 
1249 	for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) {
1250 		struct ceph_pg_pool_info *pi =
1251 		    rb_entry(n, struct ceph_pg_pool_info, node);
1252 
1253 		if (__pool_full(pi))
1254 			return true;
1255 	}
1256 
1257 	return false;
1258 }
1259 
1260 static bool pool_full(struct ceph_osd_client *osdc, s64 pool_id)
1261 {
1262 	struct ceph_pg_pool_info *pi;
1263 
1264 	pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id);
1265 	if (!pi)
1266 		return false;
1267 
1268 	return __pool_full(pi);
1269 }
1270 
1271 /*
1272  * Returns whether a request should be blocked from being sent
1273  * based on the current osdmap and osd_client settings.
1274  */
1275 static bool target_should_be_paused(struct ceph_osd_client *osdc,
1276 				    const struct ceph_osd_request_target *t,
1277 				    struct ceph_pg_pool_info *pi)
1278 {
1279 	bool pauserd = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD);
1280 	bool pausewr = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR) ||
1281 		       ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) ||
1282 		       __pool_full(pi);
1283 
1284 	WARN_ON(pi->id != t->base_oloc.pool);
1285 	return (t->flags & CEPH_OSD_FLAG_READ && pauserd) ||
1286 	       (t->flags & CEPH_OSD_FLAG_WRITE && pausewr);
1287 }
1288 
1289 enum calc_target_result {
1290 	CALC_TARGET_NO_ACTION = 0,
1291 	CALC_TARGET_NEED_RESEND,
1292 	CALC_TARGET_POOL_DNE,
1293 };
1294 
1295 static enum calc_target_result calc_target(struct ceph_osd_client *osdc,
1296 					   struct ceph_osd_request_target *t,
1297 					   u32 *last_force_resend,
1298 					   bool any_change)
1299 {
1300 	struct ceph_pg_pool_info *pi;
1301 	struct ceph_pg pgid, last_pgid;
1302 	struct ceph_osds up, acting;
1303 	bool force_resend = false;
1304 	bool need_check_tiering = false;
1305 	bool need_resend = false;
1306 	bool sort_bitwise = ceph_osdmap_flag(osdc->osdmap,
1307 					     CEPH_OSDMAP_SORTBITWISE);
1308 	enum calc_target_result ct_res;
1309 	int ret;
1310 
1311 	pi = ceph_pg_pool_by_id(osdc->osdmap, t->base_oloc.pool);
1312 	if (!pi) {
1313 		t->osd = CEPH_HOMELESS_OSD;
1314 		ct_res = CALC_TARGET_POOL_DNE;
1315 		goto out;
1316 	}
1317 
1318 	if (osdc->osdmap->epoch == pi->last_force_request_resend) {
1319 		if (last_force_resend &&
1320 		    *last_force_resend < pi->last_force_request_resend) {
1321 			*last_force_resend = pi->last_force_request_resend;
1322 			force_resend = true;
1323 		} else if (!last_force_resend) {
1324 			force_resend = true;
1325 		}
1326 	}
1327 	if (ceph_oid_empty(&t->target_oid) || force_resend) {
1328 		ceph_oid_copy(&t->target_oid, &t->base_oid);
1329 		need_check_tiering = true;
1330 	}
1331 	if (ceph_oloc_empty(&t->target_oloc) || force_resend) {
1332 		ceph_oloc_copy(&t->target_oloc, &t->base_oloc);
1333 		need_check_tiering = true;
1334 	}
1335 
1336 	if (need_check_tiering &&
1337 	    (t->flags & CEPH_OSD_FLAG_IGNORE_OVERLAY) == 0) {
1338 		if (t->flags & CEPH_OSD_FLAG_READ && pi->read_tier >= 0)
1339 			t->target_oloc.pool = pi->read_tier;
1340 		if (t->flags & CEPH_OSD_FLAG_WRITE && pi->write_tier >= 0)
1341 			t->target_oloc.pool = pi->write_tier;
1342 	}
1343 
1344 	ret = ceph_object_locator_to_pg(osdc->osdmap, &t->target_oid,
1345 					&t->target_oloc, &pgid);
1346 	if (ret) {
1347 		WARN_ON(ret != -ENOENT);
1348 		t->osd = CEPH_HOMELESS_OSD;
1349 		ct_res = CALC_TARGET_POOL_DNE;
1350 		goto out;
1351 	}
1352 	last_pgid.pool = pgid.pool;
1353 	last_pgid.seed = ceph_stable_mod(pgid.seed, t->pg_num, t->pg_num_mask);
1354 
1355 	ceph_pg_to_up_acting_osds(osdc->osdmap, &pgid, &up, &acting);
1356 	if (any_change &&
1357 	    ceph_is_new_interval(&t->acting,
1358 				 &acting,
1359 				 &t->up,
1360 				 &up,
1361 				 t->size,
1362 				 pi->size,
1363 				 t->min_size,
1364 				 pi->min_size,
1365 				 t->pg_num,
1366 				 pi->pg_num,
1367 				 t->sort_bitwise,
1368 				 sort_bitwise,
1369 				 &last_pgid))
1370 		force_resend = true;
1371 
1372 	if (t->paused && !target_should_be_paused(osdc, t, pi)) {
1373 		t->paused = false;
1374 		need_resend = true;
1375 	}
1376 
1377 	if (ceph_pg_compare(&t->pgid, &pgid) ||
1378 	    ceph_osds_changed(&t->acting, &acting, any_change) ||
1379 	    force_resend) {
1380 		t->pgid = pgid; /* struct */
1381 		ceph_osds_copy(&t->acting, &acting);
1382 		ceph_osds_copy(&t->up, &up);
1383 		t->size = pi->size;
1384 		t->min_size = pi->min_size;
1385 		t->pg_num = pi->pg_num;
1386 		t->pg_num_mask = pi->pg_num_mask;
1387 		t->sort_bitwise = sort_bitwise;
1388 
1389 		t->osd = acting.primary;
1390 		need_resend = true;
1391 	}
1392 
1393 	ct_res = need_resend ? CALC_TARGET_NEED_RESEND : CALC_TARGET_NO_ACTION;
1394 out:
1395 	dout("%s t %p -> ct_res %d osd %d\n", __func__, t, ct_res, t->osd);
1396 	return ct_res;
1397 }
1398 
1399 static void setup_request_data(struct ceph_osd_request *req,
1400 			       struct ceph_msg *msg)
1401 {
1402 	u32 data_len = 0;
1403 	int i;
1404 
1405 	if (!list_empty(&msg->data))
1406 		return;
1407 
1408 	WARN_ON(msg->data_length);
1409 	for (i = 0; i < req->r_num_ops; i++) {
1410 		struct ceph_osd_req_op *op = &req->r_ops[i];
1411 
1412 		switch (op->op) {
1413 		/* request */
1414 		case CEPH_OSD_OP_WRITE:
1415 		case CEPH_OSD_OP_WRITEFULL:
1416 			WARN_ON(op->indata_len != op->extent.length);
1417 			ceph_osdc_msg_data_add(msg, &op->extent.osd_data);
1418 			break;
1419 		case CEPH_OSD_OP_SETXATTR:
1420 		case CEPH_OSD_OP_CMPXATTR:
1421 			WARN_ON(op->indata_len != op->xattr.name_len +
1422 						  op->xattr.value_len);
1423 			ceph_osdc_msg_data_add(msg, &op->xattr.osd_data);
1424 			break;
1425 		case CEPH_OSD_OP_NOTIFY_ACK:
1426 			ceph_osdc_msg_data_add(msg,
1427 					       &op->notify_ack.request_data);
1428 			break;
1429 
1430 		/* reply */
1431 		case CEPH_OSD_OP_STAT:
1432 			ceph_osdc_msg_data_add(req->r_reply,
1433 					       &op->raw_data_in);
1434 			break;
1435 		case CEPH_OSD_OP_READ:
1436 			ceph_osdc_msg_data_add(req->r_reply,
1437 					       &op->extent.osd_data);
1438 			break;
1439 
1440 		/* both */
1441 		case CEPH_OSD_OP_CALL:
1442 			WARN_ON(op->indata_len != op->cls.class_len +
1443 						  op->cls.method_len +
1444 						  op->cls.indata_len);
1445 			ceph_osdc_msg_data_add(msg, &op->cls.request_info);
1446 			/* optional, can be NONE */
1447 			ceph_osdc_msg_data_add(msg, &op->cls.request_data);
1448 			/* optional, can be NONE */
1449 			ceph_osdc_msg_data_add(req->r_reply,
1450 					       &op->cls.response_data);
1451 			break;
1452 		case CEPH_OSD_OP_NOTIFY:
1453 			ceph_osdc_msg_data_add(msg,
1454 					       &op->notify.request_data);
1455 			ceph_osdc_msg_data_add(req->r_reply,
1456 					       &op->notify.response_data);
1457 			break;
1458 		}
1459 
1460 		data_len += op->indata_len;
1461 	}
1462 
1463 	WARN_ON(data_len != msg->data_length);
1464 }
1465 
1466 static void encode_request(struct ceph_osd_request *req, struct ceph_msg *msg)
1467 {
1468 	void *p = msg->front.iov_base;
1469 	void *const end = p + msg->front_alloc_len;
1470 	u32 data_len = 0;
1471 	int i;
1472 
1473 	if (req->r_flags & CEPH_OSD_FLAG_WRITE) {
1474 		/* snapshots aren't writeable */
1475 		WARN_ON(req->r_snapid != CEPH_NOSNAP);
1476 	} else {
1477 		WARN_ON(req->r_mtime.tv_sec || req->r_mtime.tv_nsec ||
1478 			req->r_data_offset || req->r_snapc);
1479 	}
1480 
1481 	setup_request_data(req, msg);
1482 
1483 	ceph_encode_32(&p, 1); /* client_inc, always 1 */
1484 	ceph_encode_32(&p, req->r_osdc->osdmap->epoch);
1485 	ceph_encode_32(&p, req->r_flags);
1486 	ceph_encode_timespec(p, &req->r_mtime);
1487 	p += sizeof(struct ceph_timespec);
1488 	/* aka reassert_version */
1489 	memcpy(p, &req->r_replay_version, sizeof(req->r_replay_version));
1490 	p += sizeof(req->r_replay_version);
1491 
1492 	/* oloc */
1493 	ceph_encode_8(&p, 4);
1494 	ceph_encode_8(&p, 4);
1495 	ceph_encode_32(&p, 8 + 4 + 4);
1496 	ceph_encode_64(&p, req->r_t.target_oloc.pool);
1497 	ceph_encode_32(&p, -1); /* preferred */
1498 	ceph_encode_32(&p, 0); /* key len */
1499 
1500 	/* pgid */
1501 	ceph_encode_8(&p, 1);
1502 	ceph_encode_64(&p, req->r_t.pgid.pool);
1503 	ceph_encode_32(&p, req->r_t.pgid.seed);
1504 	ceph_encode_32(&p, -1); /* preferred */
1505 
1506 	/* oid */
1507 	ceph_encode_32(&p, req->r_t.target_oid.name_len);
1508 	memcpy(p, req->r_t.target_oid.name, req->r_t.target_oid.name_len);
1509 	p += req->r_t.target_oid.name_len;
1510 
1511 	/* ops, can imply data */
1512 	ceph_encode_16(&p, req->r_num_ops);
1513 	for (i = 0; i < req->r_num_ops; i++) {
1514 		data_len += osd_req_encode_op(p, &req->r_ops[i]);
1515 		p += sizeof(struct ceph_osd_op);
1516 	}
1517 
1518 	ceph_encode_64(&p, req->r_snapid); /* snapid */
1519 	if (req->r_snapc) {
1520 		ceph_encode_64(&p, req->r_snapc->seq);
1521 		ceph_encode_32(&p, req->r_snapc->num_snaps);
1522 		for (i = 0; i < req->r_snapc->num_snaps; i++)
1523 			ceph_encode_64(&p, req->r_snapc->snaps[i]);
1524 	} else {
1525 		ceph_encode_64(&p, 0); /* snap_seq */
1526 		ceph_encode_32(&p, 0); /* snaps len */
1527 	}
1528 
1529 	ceph_encode_32(&p, req->r_attempts); /* retry_attempt */
1530 
1531 	BUG_ON(p > end);
1532 	msg->front.iov_len = p - msg->front.iov_base;
1533 	msg->hdr.version = cpu_to_le16(4); /* MOSDOp v4 */
1534 	msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1535 	msg->hdr.data_len = cpu_to_le32(data_len);
1536 	/*
1537 	 * The header "data_off" is a hint to the receiver allowing it
1538 	 * to align received data into its buffers such that there's no
1539 	 * need to re-copy it before writing it to disk (direct I/O).
1540 	 */
1541 	msg->hdr.data_off = cpu_to_le16(req->r_data_offset);
1542 
1543 	dout("%s req %p oid %*pE oid_len %d front %zu data %u\n", __func__,
1544 	     req, req->r_t.target_oid.name_len, req->r_t.target_oid.name,
1545 	     req->r_t.target_oid.name_len, msg->front.iov_len, data_len);
1546 }
1547 
1548 /*
1549  * @req has to be assigned a tid and registered.
1550  */
1551 static void send_request(struct ceph_osd_request *req)
1552 {
1553 	struct ceph_osd *osd = req->r_osd;
1554 
1555 	verify_osd_locked(osd);
1556 	WARN_ON(osd->o_osd != req->r_t.osd);
1557 
1558 	/*
1559 	 * We may have a previously queued request message hanging
1560 	 * around.  Cancel it to avoid corrupting the msgr.
1561 	 */
1562 	if (req->r_sent)
1563 		ceph_msg_revoke(req->r_request);
1564 
1565 	req->r_flags |= CEPH_OSD_FLAG_KNOWN_REDIR;
1566 	if (req->r_attempts)
1567 		req->r_flags |= CEPH_OSD_FLAG_RETRY;
1568 	else
1569 		WARN_ON(req->r_flags & CEPH_OSD_FLAG_RETRY);
1570 
1571 	encode_request(req, req->r_request);
1572 
1573 	dout("%s req %p tid %llu to pg %llu.%x osd%d flags 0x%x attempt %d\n",
1574 	     __func__, req, req->r_tid, req->r_t.pgid.pool, req->r_t.pgid.seed,
1575 	     req->r_t.osd, req->r_flags, req->r_attempts);
1576 
1577 	req->r_t.paused = false;
1578 	req->r_stamp = jiffies;
1579 	req->r_attempts++;
1580 
1581 	req->r_sent = osd->o_incarnation;
1582 	req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
1583 	ceph_con_send(&osd->o_con, ceph_msg_get(req->r_request));
1584 }
1585 
1586 static void maybe_request_map(struct ceph_osd_client *osdc)
1587 {
1588 	bool continuous = false;
1589 
1590 	verify_osdc_locked(osdc);
1591 	WARN_ON(!osdc->osdmap->epoch);
1592 
1593 	if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) ||
1594 	    ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD) ||
1595 	    ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR)) {
1596 		dout("%s osdc %p continuous\n", __func__, osdc);
1597 		continuous = true;
1598 	} else {
1599 		dout("%s osdc %p onetime\n", __func__, osdc);
1600 	}
1601 
1602 	if (ceph_monc_want_map(&osdc->client->monc, CEPH_SUB_OSDMAP,
1603 			       osdc->osdmap->epoch + 1, continuous))
1604 		ceph_monc_renew_subs(&osdc->client->monc);
1605 }
1606 
1607 static void send_map_check(struct ceph_osd_request *req);
1608 
1609 static void __submit_request(struct ceph_osd_request *req, bool wrlocked)
1610 {
1611 	struct ceph_osd_client *osdc = req->r_osdc;
1612 	struct ceph_osd *osd;
1613 	enum calc_target_result ct_res;
1614 	bool need_send = false;
1615 	bool promoted = false;
1616 
1617 	WARN_ON(req->r_tid || req->r_got_reply);
1618 	dout("%s req %p wrlocked %d\n", __func__, req, wrlocked);
1619 
1620 again:
1621 	ct_res = calc_target(osdc, &req->r_t, &req->r_last_force_resend, false);
1622 	if (ct_res == CALC_TARGET_POOL_DNE && !wrlocked)
1623 		goto promote;
1624 
1625 	osd = lookup_create_osd(osdc, req->r_t.osd, wrlocked);
1626 	if (IS_ERR(osd)) {
1627 		WARN_ON(PTR_ERR(osd) != -EAGAIN || wrlocked);
1628 		goto promote;
1629 	}
1630 
1631 	if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
1632 	    ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR)) {
1633 		dout("req %p pausewr\n", req);
1634 		req->r_t.paused = true;
1635 		maybe_request_map(osdc);
1636 	} else if ((req->r_flags & CEPH_OSD_FLAG_READ) &&
1637 		   ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD)) {
1638 		dout("req %p pauserd\n", req);
1639 		req->r_t.paused = true;
1640 		maybe_request_map(osdc);
1641 	} else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
1642 		   !(req->r_flags & (CEPH_OSD_FLAG_FULL_TRY |
1643 				     CEPH_OSD_FLAG_FULL_FORCE)) &&
1644 		   (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) ||
1645 		    pool_full(osdc, req->r_t.base_oloc.pool))) {
1646 		dout("req %p full/pool_full\n", req);
1647 		pr_warn_ratelimited("FULL or reached pool quota\n");
1648 		req->r_t.paused = true;
1649 		maybe_request_map(osdc);
1650 	} else if (!osd_homeless(osd)) {
1651 		need_send = true;
1652 	} else {
1653 		maybe_request_map(osdc);
1654 	}
1655 
1656 	mutex_lock(&osd->lock);
1657 	/*
1658 	 * Assign the tid atomically with send_request() to protect
1659 	 * multiple writes to the same object from racing with each
1660 	 * other, resulting in out of order ops on the OSDs.
1661 	 */
1662 	req->r_tid = atomic64_inc_return(&osdc->last_tid);
1663 	link_request(osd, req);
1664 	if (need_send)
1665 		send_request(req);
1666 	mutex_unlock(&osd->lock);
1667 
1668 	if (ct_res == CALC_TARGET_POOL_DNE)
1669 		send_map_check(req);
1670 
1671 	if (promoted)
1672 		downgrade_write(&osdc->lock);
1673 	return;
1674 
1675 promote:
1676 	up_read(&osdc->lock);
1677 	down_write(&osdc->lock);
1678 	wrlocked = true;
1679 	promoted = true;
1680 	goto again;
1681 }
1682 
1683 static void account_request(struct ceph_osd_request *req)
1684 {
1685 	unsigned int mask = CEPH_OSD_FLAG_ACK | CEPH_OSD_FLAG_ONDISK;
1686 
1687 	if (req->r_flags & CEPH_OSD_FLAG_READ) {
1688 		WARN_ON(req->r_flags & mask);
1689 		req->r_flags |= CEPH_OSD_FLAG_ACK;
1690 	} else if (req->r_flags & CEPH_OSD_FLAG_WRITE)
1691 		WARN_ON(!(req->r_flags & mask));
1692 	else
1693 		WARN_ON(1);
1694 
1695 	WARN_ON(req->r_unsafe_callback && (req->r_flags & mask) != mask);
1696 	atomic_inc(&req->r_osdc->num_requests);
1697 }
1698 
1699 static void submit_request(struct ceph_osd_request *req, bool wrlocked)
1700 {
1701 	ceph_osdc_get_request(req);
1702 	account_request(req);
1703 	__submit_request(req, wrlocked);
1704 }
1705 
1706 static void __finish_request(struct ceph_osd_request *req)
1707 {
1708 	struct ceph_osd_client *osdc = req->r_osdc;
1709 	struct ceph_osd *osd = req->r_osd;
1710 
1711 	verify_osd_locked(osd);
1712 	dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
1713 
1714 	WARN_ON(lookup_request_mc(&osdc->map_checks, req->r_tid));
1715 	unlink_request(osd, req);
1716 	atomic_dec(&osdc->num_requests);
1717 
1718 	/*
1719 	 * If an OSD has failed or returned and a request has been sent
1720 	 * twice, it's possible to get a reply and end up here while the
1721 	 * request message is queued for delivery.  We will ignore the
1722 	 * reply, so not a big deal, but better to try and catch it.
1723 	 */
1724 	ceph_msg_revoke(req->r_request);
1725 	ceph_msg_revoke_incoming(req->r_reply);
1726 }
1727 
1728 static void finish_request(struct ceph_osd_request *req)
1729 {
1730 	__finish_request(req);
1731 	ceph_osdc_put_request(req);
1732 }
1733 
1734 static void __complete_request(struct ceph_osd_request *req)
1735 {
1736 	if (req->r_callback)
1737 		req->r_callback(req);
1738 	else
1739 		complete_all(&req->r_completion);
1740 }
1741 
1742 /*
1743  * Note that this is open-coded in handle_reply(), which has to deal
1744  * with ack vs commit, dup acks, etc.
1745  */
1746 static void complete_request(struct ceph_osd_request *req, int err)
1747 {
1748 	dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err);
1749 
1750 	req->r_result = err;
1751 	__finish_request(req);
1752 	__complete_request(req);
1753 	complete_all(&req->r_safe_completion);
1754 	ceph_osdc_put_request(req);
1755 }
1756 
1757 static void cancel_map_check(struct ceph_osd_request *req)
1758 {
1759 	struct ceph_osd_client *osdc = req->r_osdc;
1760 	struct ceph_osd_request *lookup_req;
1761 
1762 	verify_osdc_wrlocked(osdc);
1763 
1764 	lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid);
1765 	if (!lookup_req)
1766 		return;
1767 
1768 	WARN_ON(lookup_req != req);
1769 	erase_request_mc(&osdc->map_checks, req);
1770 	ceph_osdc_put_request(req);
1771 }
1772 
1773 static void cancel_request(struct ceph_osd_request *req)
1774 {
1775 	dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
1776 
1777 	cancel_map_check(req);
1778 	finish_request(req);
1779 }
1780 
1781 static void check_pool_dne(struct ceph_osd_request *req)
1782 {
1783 	struct ceph_osd_client *osdc = req->r_osdc;
1784 	struct ceph_osdmap *map = osdc->osdmap;
1785 
1786 	verify_osdc_wrlocked(osdc);
1787 	WARN_ON(!map->epoch);
1788 
1789 	if (req->r_attempts) {
1790 		/*
1791 		 * We sent a request earlier, which means that
1792 		 * previously the pool existed, and now it does not
1793 		 * (i.e., it was deleted).
1794 		 */
1795 		req->r_map_dne_bound = map->epoch;
1796 		dout("%s req %p tid %llu pool disappeared\n", __func__, req,
1797 		     req->r_tid);
1798 	} else {
1799 		dout("%s req %p tid %llu map_dne_bound %u have %u\n", __func__,
1800 		     req, req->r_tid, req->r_map_dne_bound, map->epoch);
1801 	}
1802 
1803 	if (req->r_map_dne_bound) {
1804 		if (map->epoch >= req->r_map_dne_bound) {
1805 			/* we had a new enough map */
1806 			pr_info_ratelimited("tid %llu pool does not exist\n",
1807 					    req->r_tid);
1808 			complete_request(req, -ENOENT);
1809 		}
1810 	} else {
1811 		send_map_check(req);
1812 	}
1813 }
1814 
1815 static void map_check_cb(struct ceph_mon_generic_request *greq)
1816 {
1817 	struct ceph_osd_client *osdc = &greq->monc->client->osdc;
1818 	struct ceph_osd_request *req;
1819 	u64 tid = greq->private_data;
1820 
1821 	WARN_ON(greq->result || !greq->u.newest);
1822 
1823 	down_write(&osdc->lock);
1824 	req = lookup_request_mc(&osdc->map_checks, tid);
1825 	if (!req) {
1826 		dout("%s tid %llu dne\n", __func__, tid);
1827 		goto out_unlock;
1828 	}
1829 
1830 	dout("%s req %p tid %llu map_dne_bound %u newest %llu\n", __func__,
1831 	     req, req->r_tid, req->r_map_dne_bound, greq->u.newest);
1832 	if (!req->r_map_dne_bound)
1833 		req->r_map_dne_bound = greq->u.newest;
1834 	erase_request_mc(&osdc->map_checks, req);
1835 	check_pool_dne(req);
1836 
1837 	ceph_osdc_put_request(req);
1838 out_unlock:
1839 	up_write(&osdc->lock);
1840 }
1841 
1842 static void send_map_check(struct ceph_osd_request *req)
1843 {
1844 	struct ceph_osd_client *osdc = req->r_osdc;
1845 	struct ceph_osd_request *lookup_req;
1846 	int ret;
1847 
1848 	verify_osdc_wrlocked(osdc);
1849 
1850 	lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid);
1851 	if (lookup_req) {
1852 		WARN_ON(lookup_req != req);
1853 		return;
1854 	}
1855 
1856 	ceph_osdc_get_request(req);
1857 	insert_request_mc(&osdc->map_checks, req);
1858 	ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap",
1859 					  map_check_cb, req->r_tid);
1860 	WARN_ON(ret);
1861 }
1862 
1863 /*
1864  * lingering requests, watch/notify v2 infrastructure
1865  */
1866 static void linger_release(struct kref *kref)
1867 {
1868 	struct ceph_osd_linger_request *lreq =
1869 	    container_of(kref, struct ceph_osd_linger_request, kref);
1870 
1871 	dout("%s lreq %p reg_req %p ping_req %p\n", __func__, lreq,
1872 	     lreq->reg_req, lreq->ping_req);
1873 	WARN_ON(!RB_EMPTY_NODE(&lreq->node));
1874 	WARN_ON(!RB_EMPTY_NODE(&lreq->osdc_node));
1875 	WARN_ON(!RB_EMPTY_NODE(&lreq->mc_node));
1876 	WARN_ON(!list_empty(&lreq->scan_item));
1877 	WARN_ON(!list_empty(&lreq->pending_lworks));
1878 	WARN_ON(lreq->osd);
1879 
1880 	if (lreq->reg_req)
1881 		ceph_osdc_put_request(lreq->reg_req);
1882 	if (lreq->ping_req)
1883 		ceph_osdc_put_request(lreq->ping_req);
1884 	target_destroy(&lreq->t);
1885 	kfree(lreq);
1886 }
1887 
1888 static void linger_put(struct ceph_osd_linger_request *lreq)
1889 {
1890 	if (lreq)
1891 		kref_put(&lreq->kref, linger_release);
1892 }
1893 
1894 static struct ceph_osd_linger_request *
1895 linger_get(struct ceph_osd_linger_request *lreq)
1896 {
1897 	kref_get(&lreq->kref);
1898 	return lreq;
1899 }
1900 
1901 static struct ceph_osd_linger_request *
1902 linger_alloc(struct ceph_osd_client *osdc)
1903 {
1904 	struct ceph_osd_linger_request *lreq;
1905 
1906 	lreq = kzalloc(sizeof(*lreq), GFP_NOIO);
1907 	if (!lreq)
1908 		return NULL;
1909 
1910 	kref_init(&lreq->kref);
1911 	mutex_init(&lreq->lock);
1912 	RB_CLEAR_NODE(&lreq->node);
1913 	RB_CLEAR_NODE(&lreq->osdc_node);
1914 	RB_CLEAR_NODE(&lreq->mc_node);
1915 	INIT_LIST_HEAD(&lreq->scan_item);
1916 	INIT_LIST_HEAD(&lreq->pending_lworks);
1917 	init_completion(&lreq->reg_commit_wait);
1918 	init_completion(&lreq->notify_finish_wait);
1919 
1920 	lreq->osdc = osdc;
1921 	target_init(&lreq->t);
1922 
1923 	dout("%s lreq %p\n", __func__, lreq);
1924 	return lreq;
1925 }
1926 
1927 DEFINE_RB_INSDEL_FUNCS(linger, struct ceph_osd_linger_request, linger_id, node)
1928 DEFINE_RB_FUNCS(linger_osdc, struct ceph_osd_linger_request, linger_id, osdc_node)
1929 DEFINE_RB_FUNCS(linger_mc, struct ceph_osd_linger_request, linger_id, mc_node)
1930 
1931 /*
1932  * Create linger request <-> OSD session relation.
1933  *
1934  * @lreq has to be registered, @osd may be homeless.
1935  */
1936 static void link_linger(struct ceph_osd *osd,
1937 			struct ceph_osd_linger_request *lreq)
1938 {
1939 	verify_osd_locked(osd);
1940 	WARN_ON(!lreq->linger_id || lreq->osd);
1941 	dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd,
1942 	     osd->o_osd, lreq, lreq->linger_id);
1943 
1944 	if (!osd_homeless(osd))
1945 		__remove_osd_from_lru(osd);
1946 	else
1947 		atomic_inc(&osd->o_osdc->num_homeless);
1948 
1949 	get_osd(osd);
1950 	insert_linger(&osd->o_linger_requests, lreq);
1951 	lreq->osd = osd;
1952 }
1953 
1954 static void unlink_linger(struct ceph_osd *osd,
1955 			  struct ceph_osd_linger_request *lreq)
1956 {
1957 	verify_osd_locked(osd);
1958 	WARN_ON(lreq->osd != osd);
1959 	dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd,
1960 	     osd->o_osd, lreq, lreq->linger_id);
1961 
1962 	lreq->osd = NULL;
1963 	erase_linger(&osd->o_linger_requests, lreq);
1964 	put_osd(osd);
1965 
1966 	if (!osd_homeless(osd))
1967 		maybe_move_osd_to_lru(osd);
1968 	else
1969 		atomic_dec(&osd->o_osdc->num_homeless);
1970 }
1971 
1972 static bool __linger_registered(struct ceph_osd_linger_request *lreq)
1973 {
1974 	verify_osdc_locked(lreq->osdc);
1975 
1976 	return !RB_EMPTY_NODE(&lreq->osdc_node);
1977 }
1978 
1979 static bool linger_registered(struct ceph_osd_linger_request *lreq)
1980 {
1981 	struct ceph_osd_client *osdc = lreq->osdc;
1982 	bool registered;
1983 
1984 	down_read(&osdc->lock);
1985 	registered = __linger_registered(lreq);
1986 	up_read(&osdc->lock);
1987 
1988 	return registered;
1989 }
1990 
1991 static void linger_register(struct ceph_osd_linger_request *lreq)
1992 {
1993 	struct ceph_osd_client *osdc = lreq->osdc;
1994 
1995 	verify_osdc_wrlocked(osdc);
1996 	WARN_ON(lreq->linger_id);
1997 
1998 	linger_get(lreq);
1999 	lreq->linger_id = ++osdc->last_linger_id;
2000 	insert_linger_osdc(&osdc->linger_requests, lreq);
2001 }
2002 
2003 static void linger_unregister(struct ceph_osd_linger_request *lreq)
2004 {
2005 	struct ceph_osd_client *osdc = lreq->osdc;
2006 
2007 	verify_osdc_wrlocked(osdc);
2008 
2009 	erase_linger_osdc(&osdc->linger_requests, lreq);
2010 	linger_put(lreq);
2011 }
2012 
2013 static void cancel_linger_request(struct ceph_osd_request *req)
2014 {
2015 	struct ceph_osd_linger_request *lreq = req->r_priv;
2016 
2017 	WARN_ON(!req->r_linger);
2018 	cancel_request(req);
2019 	linger_put(lreq);
2020 }
2021 
2022 struct linger_work {
2023 	struct work_struct work;
2024 	struct ceph_osd_linger_request *lreq;
2025 	struct list_head pending_item;
2026 	unsigned long queued_stamp;
2027 
2028 	union {
2029 		struct {
2030 			u64 notify_id;
2031 			u64 notifier_id;
2032 			void *payload; /* points into @msg front */
2033 			size_t payload_len;
2034 
2035 			struct ceph_msg *msg; /* for ceph_msg_put() */
2036 		} notify;
2037 		struct {
2038 			int err;
2039 		} error;
2040 	};
2041 };
2042 
2043 static struct linger_work *lwork_alloc(struct ceph_osd_linger_request *lreq,
2044 				       work_func_t workfn)
2045 {
2046 	struct linger_work *lwork;
2047 
2048 	lwork = kzalloc(sizeof(*lwork), GFP_NOIO);
2049 	if (!lwork)
2050 		return NULL;
2051 
2052 	INIT_WORK(&lwork->work, workfn);
2053 	INIT_LIST_HEAD(&lwork->pending_item);
2054 	lwork->lreq = linger_get(lreq);
2055 
2056 	return lwork;
2057 }
2058 
2059 static void lwork_free(struct linger_work *lwork)
2060 {
2061 	struct ceph_osd_linger_request *lreq = lwork->lreq;
2062 
2063 	mutex_lock(&lreq->lock);
2064 	list_del(&lwork->pending_item);
2065 	mutex_unlock(&lreq->lock);
2066 
2067 	linger_put(lreq);
2068 	kfree(lwork);
2069 }
2070 
2071 static void lwork_queue(struct linger_work *lwork)
2072 {
2073 	struct ceph_osd_linger_request *lreq = lwork->lreq;
2074 	struct ceph_osd_client *osdc = lreq->osdc;
2075 
2076 	verify_lreq_locked(lreq);
2077 	WARN_ON(!list_empty(&lwork->pending_item));
2078 
2079 	lwork->queued_stamp = jiffies;
2080 	list_add_tail(&lwork->pending_item, &lreq->pending_lworks);
2081 	queue_work(osdc->notify_wq, &lwork->work);
2082 }
2083 
2084 static void do_watch_notify(struct work_struct *w)
2085 {
2086 	struct linger_work *lwork = container_of(w, struct linger_work, work);
2087 	struct ceph_osd_linger_request *lreq = lwork->lreq;
2088 
2089 	if (!linger_registered(lreq)) {
2090 		dout("%s lreq %p not registered\n", __func__, lreq);
2091 		goto out;
2092 	}
2093 
2094 	WARN_ON(!lreq->is_watch);
2095 	dout("%s lreq %p notify_id %llu notifier_id %llu payload_len %zu\n",
2096 	     __func__, lreq, lwork->notify.notify_id, lwork->notify.notifier_id,
2097 	     lwork->notify.payload_len);
2098 	lreq->wcb(lreq->data, lwork->notify.notify_id, lreq->linger_id,
2099 		  lwork->notify.notifier_id, lwork->notify.payload,
2100 		  lwork->notify.payload_len);
2101 
2102 out:
2103 	ceph_msg_put(lwork->notify.msg);
2104 	lwork_free(lwork);
2105 }
2106 
2107 static void do_watch_error(struct work_struct *w)
2108 {
2109 	struct linger_work *lwork = container_of(w, struct linger_work, work);
2110 	struct ceph_osd_linger_request *lreq = lwork->lreq;
2111 
2112 	if (!linger_registered(lreq)) {
2113 		dout("%s lreq %p not registered\n", __func__, lreq);
2114 		goto out;
2115 	}
2116 
2117 	dout("%s lreq %p err %d\n", __func__, lreq, lwork->error.err);
2118 	lreq->errcb(lreq->data, lreq->linger_id, lwork->error.err);
2119 
2120 out:
2121 	lwork_free(lwork);
2122 }
2123 
2124 static void queue_watch_error(struct ceph_osd_linger_request *lreq)
2125 {
2126 	struct linger_work *lwork;
2127 
2128 	lwork = lwork_alloc(lreq, do_watch_error);
2129 	if (!lwork) {
2130 		pr_err("failed to allocate error-lwork\n");
2131 		return;
2132 	}
2133 
2134 	lwork->error.err = lreq->last_error;
2135 	lwork_queue(lwork);
2136 }
2137 
2138 static void linger_reg_commit_complete(struct ceph_osd_linger_request *lreq,
2139 				       int result)
2140 {
2141 	if (!completion_done(&lreq->reg_commit_wait)) {
2142 		lreq->reg_commit_error = (result <= 0 ? result : 0);
2143 		complete_all(&lreq->reg_commit_wait);
2144 	}
2145 }
2146 
2147 static void linger_commit_cb(struct ceph_osd_request *req)
2148 {
2149 	struct ceph_osd_linger_request *lreq = req->r_priv;
2150 
2151 	mutex_lock(&lreq->lock);
2152 	dout("%s lreq %p linger_id %llu result %d\n", __func__, lreq,
2153 	     lreq->linger_id, req->r_result);
2154 	WARN_ON(!__linger_registered(lreq));
2155 	linger_reg_commit_complete(lreq, req->r_result);
2156 	lreq->committed = true;
2157 
2158 	if (!lreq->is_watch) {
2159 		struct ceph_osd_data *osd_data =
2160 		    osd_req_op_data(req, 0, notify, response_data);
2161 		void *p = page_address(osd_data->pages[0]);
2162 
2163 		WARN_ON(req->r_ops[0].op != CEPH_OSD_OP_NOTIFY ||
2164 			osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
2165 
2166 		/* make note of the notify_id */
2167 		if (req->r_ops[0].outdata_len >= sizeof(u64)) {
2168 			lreq->notify_id = ceph_decode_64(&p);
2169 			dout("lreq %p notify_id %llu\n", lreq,
2170 			     lreq->notify_id);
2171 		} else {
2172 			dout("lreq %p no notify_id\n", lreq);
2173 		}
2174 	}
2175 
2176 	mutex_unlock(&lreq->lock);
2177 	linger_put(lreq);
2178 }
2179 
2180 static int normalize_watch_error(int err)
2181 {
2182 	/*
2183 	 * Translate ENOENT -> ENOTCONN so that a delete->disconnection
2184 	 * notification and a failure to reconnect because we raced with
2185 	 * the delete appear the same to the user.
2186 	 */
2187 	if (err == -ENOENT)
2188 		err = -ENOTCONN;
2189 
2190 	return err;
2191 }
2192 
2193 static void linger_reconnect_cb(struct ceph_osd_request *req)
2194 {
2195 	struct ceph_osd_linger_request *lreq = req->r_priv;
2196 
2197 	mutex_lock(&lreq->lock);
2198 	dout("%s lreq %p linger_id %llu result %d last_error %d\n", __func__,
2199 	     lreq, lreq->linger_id, req->r_result, lreq->last_error);
2200 	if (req->r_result < 0) {
2201 		if (!lreq->last_error) {
2202 			lreq->last_error = normalize_watch_error(req->r_result);
2203 			queue_watch_error(lreq);
2204 		}
2205 	}
2206 
2207 	mutex_unlock(&lreq->lock);
2208 	linger_put(lreq);
2209 }
2210 
2211 static void send_linger(struct ceph_osd_linger_request *lreq)
2212 {
2213 	struct ceph_osd_request *req = lreq->reg_req;
2214 	struct ceph_osd_req_op *op = &req->r_ops[0];
2215 
2216 	verify_osdc_wrlocked(req->r_osdc);
2217 	dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
2218 
2219 	if (req->r_osd)
2220 		cancel_linger_request(req);
2221 
2222 	request_reinit(req);
2223 	ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
2224 	ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
2225 	req->r_flags = lreq->t.flags;
2226 	req->r_mtime = lreq->mtime;
2227 
2228 	mutex_lock(&lreq->lock);
2229 	if (lreq->is_watch && lreq->committed) {
2230 		WARN_ON(op->op != CEPH_OSD_OP_WATCH ||
2231 			op->watch.cookie != lreq->linger_id);
2232 		op->watch.op = CEPH_OSD_WATCH_OP_RECONNECT;
2233 		op->watch.gen = ++lreq->register_gen;
2234 		dout("lreq %p reconnect register_gen %u\n", lreq,
2235 		     op->watch.gen);
2236 		req->r_callback = linger_reconnect_cb;
2237 	} else {
2238 		if (!lreq->is_watch)
2239 			lreq->notify_id = 0;
2240 		else
2241 			WARN_ON(op->watch.op != CEPH_OSD_WATCH_OP_WATCH);
2242 		dout("lreq %p register\n", lreq);
2243 		req->r_callback = linger_commit_cb;
2244 	}
2245 	mutex_unlock(&lreq->lock);
2246 
2247 	req->r_priv = linger_get(lreq);
2248 	req->r_linger = true;
2249 
2250 	submit_request(req, true);
2251 }
2252 
2253 static void linger_ping_cb(struct ceph_osd_request *req)
2254 {
2255 	struct ceph_osd_linger_request *lreq = req->r_priv;
2256 
2257 	mutex_lock(&lreq->lock);
2258 	dout("%s lreq %p linger_id %llu result %d ping_sent %lu last_error %d\n",
2259 	     __func__, lreq, lreq->linger_id, req->r_result, lreq->ping_sent,
2260 	     lreq->last_error);
2261 	if (lreq->register_gen == req->r_ops[0].watch.gen) {
2262 		if (!req->r_result) {
2263 			lreq->watch_valid_thru = lreq->ping_sent;
2264 		} else if (!lreq->last_error) {
2265 			lreq->last_error = normalize_watch_error(req->r_result);
2266 			queue_watch_error(lreq);
2267 		}
2268 	} else {
2269 		dout("lreq %p register_gen %u ignoring old pong %u\n", lreq,
2270 		     lreq->register_gen, req->r_ops[0].watch.gen);
2271 	}
2272 
2273 	mutex_unlock(&lreq->lock);
2274 	linger_put(lreq);
2275 }
2276 
2277 static void send_linger_ping(struct ceph_osd_linger_request *lreq)
2278 {
2279 	struct ceph_osd_client *osdc = lreq->osdc;
2280 	struct ceph_osd_request *req = lreq->ping_req;
2281 	struct ceph_osd_req_op *op = &req->r_ops[0];
2282 
2283 	if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD)) {
2284 		dout("%s PAUSERD\n", __func__);
2285 		return;
2286 	}
2287 
2288 	lreq->ping_sent = jiffies;
2289 	dout("%s lreq %p linger_id %llu ping_sent %lu register_gen %u\n",
2290 	     __func__, lreq, lreq->linger_id, lreq->ping_sent,
2291 	     lreq->register_gen);
2292 
2293 	if (req->r_osd)
2294 		cancel_linger_request(req);
2295 
2296 	request_reinit(req);
2297 	target_copy(&req->r_t, &lreq->t);
2298 
2299 	WARN_ON(op->op != CEPH_OSD_OP_WATCH ||
2300 		op->watch.cookie != lreq->linger_id ||
2301 		op->watch.op != CEPH_OSD_WATCH_OP_PING);
2302 	op->watch.gen = lreq->register_gen;
2303 	req->r_callback = linger_ping_cb;
2304 	req->r_priv = linger_get(lreq);
2305 	req->r_linger = true;
2306 
2307 	ceph_osdc_get_request(req);
2308 	account_request(req);
2309 	req->r_tid = atomic64_inc_return(&osdc->last_tid);
2310 	link_request(lreq->osd, req);
2311 	send_request(req);
2312 }
2313 
2314 static void linger_submit(struct ceph_osd_linger_request *lreq)
2315 {
2316 	struct ceph_osd_client *osdc = lreq->osdc;
2317 	struct ceph_osd *osd;
2318 
2319 	calc_target(osdc, &lreq->t, &lreq->last_force_resend, false);
2320 	osd = lookup_create_osd(osdc, lreq->t.osd, true);
2321 	link_linger(osd, lreq);
2322 
2323 	send_linger(lreq);
2324 }
2325 
2326 static void cancel_linger_map_check(struct ceph_osd_linger_request *lreq)
2327 {
2328 	struct ceph_osd_client *osdc = lreq->osdc;
2329 	struct ceph_osd_linger_request *lookup_lreq;
2330 
2331 	verify_osdc_wrlocked(osdc);
2332 
2333 	lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks,
2334 				       lreq->linger_id);
2335 	if (!lookup_lreq)
2336 		return;
2337 
2338 	WARN_ON(lookup_lreq != lreq);
2339 	erase_linger_mc(&osdc->linger_map_checks, lreq);
2340 	linger_put(lreq);
2341 }
2342 
2343 /*
2344  * @lreq has to be both registered and linked.
2345  */
2346 static void __linger_cancel(struct ceph_osd_linger_request *lreq)
2347 {
2348 	if (lreq->is_watch && lreq->ping_req->r_osd)
2349 		cancel_linger_request(lreq->ping_req);
2350 	if (lreq->reg_req->r_osd)
2351 		cancel_linger_request(lreq->reg_req);
2352 	cancel_linger_map_check(lreq);
2353 	unlink_linger(lreq->osd, lreq);
2354 	linger_unregister(lreq);
2355 }
2356 
2357 static void linger_cancel(struct ceph_osd_linger_request *lreq)
2358 {
2359 	struct ceph_osd_client *osdc = lreq->osdc;
2360 
2361 	down_write(&osdc->lock);
2362 	if (__linger_registered(lreq))
2363 		__linger_cancel(lreq);
2364 	up_write(&osdc->lock);
2365 }
2366 
2367 static void send_linger_map_check(struct ceph_osd_linger_request *lreq);
2368 
2369 static void check_linger_pool_dne(struct ceph_osd_linger_request *lreq)
2370 {
2371 	struct ceph_osd_client *osdc = lreq->osdc;
2372 	struct ceph_osdmap *map = osdc->osdmap;
2373 
2374 	verify_osdc_wrlocked(osdc);
2375 	WARN_ON(!map->epoch);
2376 
2377 	if (lreq->register_gen) {
2378 		lreq->map_dne_bound = map->epoch;
2379 		dout("%s lreq %p linger_id %llu pool disappeared\n", __func__,
2380 		     lreq, lreq->linger_id);
2381 	} else {
2382 		dout("%s lreq %p linger_id %llu map_dne_bound %u have %u\n",
2383 		     __func__, lreq, lreq->linger_id, lreq->map_dne_bound,
2384 		     map->epoch);
2385 	}
2386 
2387 	if (lreq->map_dne_bound) {
2388 		if (map->epoch >= lreq->map_dne_bound) {
2389 			/* we had a new enough map */
2390 			pr_info("linger_id %llu pool does not exist\n",
2391 				lreq->linger_id);
2392 			linger_reg_commit_complete(lreq, -ENOENT);
2393 			__linger_cancel(lreq);
2394 		}
2395 	} else {
2396 		send_linger_map_check(lreq);
2397 	}
2398 }
2399 
2400 static void linger_map_check_cb(struct ceph_mon_generic_request *greq)
2401 {
2402 	struct ceph_osd_client *osdc = &greq->monc->client->osdc;
2403 	struct ceph_osd_linger_request *lreq;
2404 	u64 linger_id = greq->private_data;
2405 
2406 	WARN_ON(greq->result || !greq->u.newest);
2407 
2408 	down_write(&osdc->lock);
2409 	lreq = lookup_linger_mc(&osdc->linger_map_checks, linger_id);
2410 	if (!lreq) {
2411 		dout("%s linger_id %llu dne\n", __func__, linger_id);
2412 		goto out_unlock;
2413 	}
2414 
2415 	dout("%s lreq %p linger_id %llu map_dne_bound %u newest %llu\n",
2416 	     __func__, lreq, lreq->linger_id, lreq->map_dne_bound,
2417 	     greq->u.newest);
2418 	if (!lreq->map_dne_bound)
2419 		lreq->map_dne_bound = greq->u.newest;
2420 	erase_linger_mc(&osdc->linger_map_checks, lreq);
2421 	check_linger_pool_dne(lreq);
2422 
2423 	linger_put(lreq);
2424 out_unlock:
2425 	up_write(&osdc->lock);
2426 }
2427 
2428 static void send_linger_map_check(struct ceph_osd_linger_request *lreq)
2429 {
2430 	struct ceph_osd_client *osdc = lreq->osdc;
2431 	struct ceph_osd_linger_request *lookup_lreq;
2432 	int ret;
2433 
2434 	verify_osdc_wrlocked(osdc);
2435 
2436 	lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks,
2437 				       lreq->linger_id);
2438 	if (lookup_lreq) {
2439 		WARN_ON(lookup_lreq != lreq);
2440 		return;
2441 	}
2442 
2443 	linger_get(lreq);
2444 	insert_linger_mc(&osdc->linger_map_checks, lreq);
2445 	ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap",
2446 					  linger_map_check_cb, lreq->linger_id);
2447 	WARN_ON(ret);
2448 }
2449 
2450 static int linger_reg_commit_wait(struct ceph_osd_linger_request *lreq)
2451 {
2452 	int ret;
2453 
2454 	dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
2455 	ret = wait_for_completion_interruptible(&lreq->reg_commit_wait);
2456 	return ret ?: lreq->reg_commit_error;
2457 }
2458 
2459 static int linger_notify_finish_wait(struct ceph_osd_linger_request *lreq)
2460 {
2461 	int ret;
2462 
2463 	dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
2464 	ret = wait_for_completion_interruptible(&lreq->notify_finish_wait);
2465 	return ret ?: lreq->notify_finish_error;
2466 }
2467 
2468 /*
2469  * Timeout callback, called every N seconds.  When 1 or more OSD
2470  * requests has been active for more than N seconds, we send a keepalive
2471  * (tag + timestamp) to its OSD to ensure any communications channel
2472  * reset is detected.
2473  */
2474 static void handle_timeout(struct work_struct *work)
2475 {
2476 	struct ceph_osd_client *osdc =
2477 		container_of(work, struct ceph_osd_client, timeout_work.work);
2478 	struct ceph_options *opts = osdc->client->options;
2479 	unsigned long cutoff = jiffies - opts->osd_keepalive_timeout;
2480 	LIST_HEAD(slow_osds);
2481 	struct rb_node *n, *p;
2482 
2483 	dout("%s osdc %p\n", __func__, osdc);
2484 	down_write(&osdc->lock);
2485 
2486 	/*
2487 	 * ping osds that are a bit slow.  this ensures that if there
2488 	 * is a break in the TCP connection we will notice, and reopen
2489 	 * a connection with that osd (from the fault callback).
2490 	 */
2491 	for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
2492 		struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
2493 		bool found = false;
2494 
2495 		for (p = rb_first(&osd->o_requests); p; p = rb_next(p)) {
2496 			struct ceph_osd_request *req =
2497 			    rb_entry(p, struct ceph_osd_request, r_node);
2498 
2499 			if (time_before(req->r_stamp, cutoff)) {
2500 				dout(" req %p tid %llu on osd%d is laggy\n",
2501 				     req, req->r_tid, osd->o_osd);
2502 				found = true;
2503 			}
2504 		}
2505 		for (p = rb_first(&osd->o_linger_requests); p; p = rb_next(p)) {
2506 			struct ceph_osd_linger_request *lreq =
2507 			    rb_entry(p, struct ceph_osd_linger_request, node);
2508 
2509 			dout(" lreq %p linger_id %llu is served by osd%d\n",
2510 			     lreq, lreq->linger_id, osd->o_osd);
2511 			found = true;
2512 
2513 			mutex_lock(&lreq->lock);
2514 			if (lreq->is_watch && lreq->committed && !lreq->last_error)
2515 				send_linger_ping(lreq);
2516 			mutex_unlock(&lreq->lock);
2517 		}
2518 
2519 		if (found)
2520 			list_move_tail(&osd->o_keepalive_item, &slow_osds);
2521 	}
2522 
2523 	if (atomic_read(&osdc->num_homeless) || !list_empty(&slow_osds))
2524 		maybe_request_map(osdc);
2525 
2526 	while (!list_empty(&slow_osds)) {
2527 		struct ceph_osd *osd = list_first_entry(&slow_osds,
2528 							struct ceph_osd,
2529 							o_keepalive_item);
2530 		list_del_init(&osd->o_keepalive_item);
2531 		ceph_con_keepalive(&osd->o_con);
2532 	}
2533 
2534 	up_write(&osdc->lock);
2535 	schedule_delayed_work(&osdc->timeout_work,
2536 			      osdc->client->options->osd_keepalive_timeout);
2537 }
2538 
2539 static void handle_osds_timeout(struct work_struct *work)
2540 {
2541 	struct ceph_osd_client *osdc =
2542 		container_of(work, struct ceph_osd_client,
2543 			     osds_timeout_work.work);
2544 	unsigned long delay = osdc->client->options->osd_idle_ttl / 4;
2545 	struct ceph_osd *osd, *nosd;
2546 
2547 	dout("%s osdc %p\n", __func__, osdc);
2548 	down_write(&osdc->lock);
2549 	list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
2550 		if (time_before(jiffies, osd->lru_ttl))
2551 			break;
2552 
2553 		WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests));
2554 		WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests));
2555 		close_osd(osd);
2556 	}
2557 
2558 	up_write(&osdc->lock);
2559 	schedule_delayed_work(&osdc->osds_timeout_work,
2560 			      round_jiffies_relative(delay));
2561 }
2562 
2563 static int ceph_oloc_decode(void **p, void *end,
2564 			    struct ceph_object_locator *oloc)
2565 {
2566 	u8 struct_v, struct_cv;
2567 	u32 len;
2568 	void *struct_end;
2569 	int ret = 0;
2570 
2571 	ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
2572 	struct_v = ceph_decode_8(p);
2573 	struct_cv = ceph_decode_8(p);
2574 	if (struct_v < 3) {
2575 		pr_warn("got v %d < 3 cv %d of ceph_object_locator\n",
2576 			struct_v, struct_cv);
2577 		goto e_inval;
2578 	}
2579 	if (struct_cv > 6) {
2580 		pr_warn("got v %d cv %d > 6 of ceph_object_locator\n",
2581 			struct_v, struct_cv);
2582 		goto e_inval;
2583 	}
2584 	len = ceph_decode_32(p);
2585 	ceph_decode_need(p, end, len, e_inval);
2586 	struct_end = *p + len;
2587 
2588 	oloc->pool = ceph_decode_64(p);
2589 	*p += 4; /* skip preferred */
2590 
2591 	len = ceph_decode_32(p);
2592 	if (len > 0) {
2593 		pr_warn("ceph_object_locator::key is set\n");
2594 		goto e_inval;
2595 	}
2596 
2597 	if (struct_v >= 5) {
2598 		len = ceph_decode_32(p);
2599 		if (len > 0) {
2600 			pr_warn("ceph_object_locator::nspace is set\n");
2601 			goto e_inval;
2602 		}
2603 	}
2604 
2605 	if (struct_v >= 6) {
2606 		s64 hash = ceph_decode_64(p);
2607 		if (hash != -1) {
2608 			pr_warn("ceph_object_locator::hash is set\n");
2609 			goto e_inval;
2610 		}
2611 	}
2612 
2613 	/* skip the rest */
2614 	*p = struct_end;
2615 out:
2616 	return ret;
2617 
2618 e_inval:
2619 	ret = -EINVAL;
2620 	goto out;
2621 }
2622 
2623 static int ceph_redirect_decode(void **p, void *end,
2624 				struct ceph_request_redirect *redir)
2625 {
2626 	u8 struct_v, struct_cv;
2627 	u32 len;
2628 	void *struct_end;
2629 	int ret;
2630 
2631 	ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
2632 	struct_v = ceph_decode_8(p);
2633 	struct_cv = ceph_decode_8(p);
2634 	if (struct_cv > 1) {
2635 		pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n",
2636 			struct_v, struct_cv);
2637 		goto e_inval;
2638 	}
2639 	len = ceph_decode_32(p);
2640 	ceph_decode_need(p, end, len, e_inval);
2641 	struct_end = *p + len;
2642 
2643 	ret = ceph_oloc_decode(p, end, &redir->oloc);
2644 	if (ret)
2645 		goto out;
2646 
2647 	len = ceph_decode_32(p);
2648 	if (len > 0) {
2649 		pr_warn("ceph_request_redirect::object_name is set\n");
2650 		goto e_inval;
2651 	}
2652 
2653 	len = ceph_decode_32(p);
2654 	*p += len; /* skip osd_instructions */
2655 
2656 	/* skip the rest */
2657 	*p = struct_end;
2658 out:
2659 	return ret;
2660 
2661 e_inval:
2662 	ret = -EINVAL;
2663 	goto out;
2664 }
2665 
2666 struct MOSDOpReply {
2667 	struct ceph_pg pgid;
2668 	u64 flags;
2669 	int result;
2670 	u32 epoch;
2671 	int num_ops;
2672 	u32 outdata_len[CEPH_OSD_MAX_OPS];
2673 	s32 rval[CEPH_OSD_MAX_OPS];
2674 	int retry_attempt;
2675 	struct ceph_eversion replay_version;
2676 	u64 user_version;
2677 	struct ceph_request_redirect redirect;
2678 };
2679 
2680 static int decode_MOSDOpReply(const struct ceph_msg *msg, struct MOSDOpReply *m)
2681 {
2682 	void *p = msg->front.iov_base;
2683 	void *const end = p + msg->front.iov_len;
2684 	u16 version = le16_to_cpu(msg->hdr.version);
2685 	struct ceph_eversion bad_replay_version;
2686 	u8 decode_redir;
2687 	u32 len;
2688 	int ret;
2689 	int i;
2690 
2691 	ceph_decode_32_safe(&p, end, len, e_inval);
2692 	ceph_decode_need(&p, end, len, e_inval);
2693 	p += len; /* skip oid */
2694 
2695 	ret = ceph_decode_pgid(&p, end, &m->pgid);
2696 	if (ret)
2697 		return ret;
2698 
2699 	ceph_decode_64_safe(&p, end, m->flags, e_inval);
2700 	ceph_decode_32_safe(&p, end, m->result, e_inval);
2701 	ceph_decode_need(&p, end, sizeof(bad_replay_version), e_inval);
2702 	memcpy(&bad_replay_version, p, sizeof(bad_replay_version));
2703 	p += sizeof(bad_replay_version);
2704 	ceph_decode_32_safe(&p, end, m->epoch, e_inval);
2705 
2706 	ceph_decode_32_safe(&p, end, m->num_ops, e_inval);
2707 	if (m->num_ops > ARRAY_SIZE(m->outdata_len))
2708 		goto e_inval;
2709 
2710 	ceph_decode_need(&p, end, m->num_ops * sizeof(struct ceph_osd_op),
2711 			 e_inval);
2712 	for (i = 0; i < m->num_ops; i++) {
2713 		struct ceph_osd_op *op = p;
2714 
2715 		m->outdata_len[i] = le32_to_cpu(op->payload_len);
2716 		p += sizeof(*op);
2717 	}
2718 
2719 	ceph_decode_32_safe(&p, end, m->retry_attempt, e_inval);
2720 	for (i = 0; i < m->num_ops; i++)
2721 		ceph_decode_32_safe(&p, end, m->rval[i], e_inval);
2722 
2723 	if (version >= 5) {
2724 		ceph_decode_need(&p, end, sizeof(m->replay_version), e_inval);
2725 		memcpy(&m->replay_version, p, sizeof(m->replay_version));
2726 		p += sizeof(m->replay_version);
2727 		ceph_decode_64_safe(&p, end, m->user_version, e_inval);
2728 	} else {
2729 		m->replay_version = bad_replay_version; /* struct */
2730 		m->user_version = le64_to_cpu(m->replay_version.version);
2731 	}
2732 
2733 	if (version >= 6) {
2734 		if (version >= 7)
2735 			ceph_decode_8_safe(&p, end, decode_redir, e_inval);
2736 		else
2737 			decode_redir = 1;
2738 	} else {
2739 		decode_redir = 0;
2740 	}
2741 
2742 	if (decode_redir) {
2743 		ret = ceph_redirect_decode(&p, end, &m->redirect);
2744 		if (ret)
2745 			return ret;
2746 	} else {
2747 		ceph_oloc_init(&m->redirect.oloc);
2748 	}
2749 
2750 	return 0;
2751 
2752 e_inval:
2753 	return -EINVAL;
2754 }
2755 
2756 /*
2757  * We are done with @req if
2758  *   - @m is a safe reply, or
2759  *   - @m is an unsafe reply and we didn't want a safe one
2760  */
2761 static bool done_request(const struct ceph_osd_request *req,
2762 			 const struct MOSDOpReply *m)
2763 {
2764 	return (m->result < 0 ||
2765 		(m->flags & CEPH_OSD_FLAG_ONDISK) ||
2766 		!(req->r_flags & CEPH_OSD_FLAG_ONDISK));
2767 }
2768 
2769 /*
2770  * handle osd op reply.  either call the callback if it is specified,
2771  * or do the completion to wake up the waiting thread.
2772  *
2773  * ->r_unsafe_callback is set?	yes			no
2774  *
2775  * first reply is OK (needed	r_cb/r_completion,	r_cb/r_completion,
2776  * any or needed/got safe)	r_safe_completion	r_safe_completion
2777  *
2778  * first reply is unsafe	r_unsafe_cb(true)	(nothing)
2779  *
2780  * when we get the safe reply	r_unsafe_cb(false),	r_cb/r_completion,
2781  *				r_safe_completion	r_safe_completion
2782  */
2783 static void handle_reply(struct ceph_osd *osd, struct ceph_msg *msg)
2784 {
2785 	struct ceph_osd_client *osdc = osd->o_osdc;
2786 	struct ceph_osd_request *req;
2787 	struct MOSDOpReply m;
2788 	u64 tid = le64_to_cpu(msg->hdr.tid);
2789 	u32 data_len = 0;
2790 	bool already_acked;
2791 	int ret;
2792 	int i;
2793 
2794 	dout("%s msg %p tid %llu\n", __func__, msg, tid);
2795 
2796 	down_read(&osdc->lock);
2797 	if (!osd_registered(osd)) {
2798 		dout("%s osd%d unknown\n", __func__, osd->o_osd);
2799 		goto out_unlock_osdc;
2800 	}
2801 	WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num));
2802 
2803 	mutex_lock(&osd->lock);
2804 	req = lookup_request(&osd->o_requests, tid);
2805 	if (!req) {
2806 		dout("%s osd%d tid %llu unknown\n", __func__, osd->o_osd, tid);
2807 		goto out_unlock_session;
2808 	}
2809 
2810 	ret = decode_MOSDOpReply(msg, &m);
2811 	if (ret) {
2812 		pr_err("failed to decode MOSDOpReply for tid %llu: %d\n",
2813 		       req->r_tid, ret);
2814 		ceph_msg_dump(msg);
2815 		goto fail_request;
2816 	}
2817 	dout("%s req %p tid %llu flags 0x%llx pgid %llu.%x epoch %u attempt %d v %u'%llu uv %llu\n",
2818 	     __func__, req, req->r_tid, m.flags, m.pgid.pool, m.pgid.seed,
2819 	     m.epoch, m.retry_attempt, le32_to_cpu(m.replay_version.epoch),
2820 	     le64_to_cpu(m.replay_version.version), m.user_version);
2821 
2822 	if (m.retry_attempt >= 0) {
2823 		if (m.retry_attempt != req->r_attempts - 1) {
2824 			dout("req %p tid %llu retry_attempt %d != %d, ignoring\n",
2825 			     req, req->r_tid, m.retry_attempt,
2826 			     req->r_attempts - 1);
2827 			goto out_unlock_session;
2828 		}
2829 	} else {
2830 		WARN_ON(1); /* MOSDOpReply v4 is assumed */
2831 	}
2832 
2833 	if (!ceph_oloc_empty(&m.redirect.oloc)) {
2834 		dout("req %p tid %llu redirect pool %lld\n", req, req->r_tid,
2835 		     m.redirect.oloc.pool);
2836 		unlink_request(osd, req);
2837 		mutex_unlock(&osd->lock);
2838 
2839 		ceph_oloc_copy(&req->r_t.target_oloc, &m.redirect.oloc);
2840 		req->r_flags |= CEPH_OSD_FLAG_REDIRECTED;
2841 		req->r_tid = 0;
2842 		__submit_request(req, false);
2843 		goto out_unlock_osdc;
2844 	}
2845 
2846 	if (m.num_ops != req->r_num_ops) {
2847 		pr_err("num_ops %d != %d for tid %llu\n", m.num_ops,
2848 		       req->r_num_ops, req->r_tid);
2849 		goto fail_request;
2850 	}
2851 	for (i = 0; i < req->r_num_ops; i++) {
2852 		dout(" req %p tid %llu op %d rval %d len %u\n", req,
2853 		     req->r_tid, i, m.rval[i], m.outdata_len[i]);
2854 		req->r_ops[i].rval = m.rval[i];
2855 		req->r_ops[i].outdata_len = m.outdata_len[i];
2856 		data_len += m.outdata_len[i];
2857 	}
2858 	if (data_len != le32_to_cpu(msg->hdr.data_len)) {
2859 		pr_err("sum of lens %u != %u for tid %llu\n", data_len,
2860 		       le32_to_cpu(msg->hdr.data_len), req->r_tid);
2861 		goto fail_request;
2862 	}
2863 	dout("%s req %p tid %llu acked %d result %d data_len %u\n", __func__,
2864 	     req, req->r_tid, req->r_got_reply, m.result, data_len);
2865 
2866 	already_acked = req->r_got_reply;
2867 	if (!already_acked) {
2868 		req->r_result = m.result ?: data_len;
2869 		req->r_replay_version = m.replay_version; /* struct */
2870 		req->r_got_reply = true;
2871 	} else if (!(m.flags & CEPH_OSD_FLAG_ONDISK)) {
2872 		dout("req %p tid %llu dup ack\n", req, req->r_tid);
2873 		goto out_unlock_session;
2874 	}
2875 
2876 	if (done_request(req, &m)) {
2877 		__finish_request(req);
2878 		if (req->r_linger) {
2879 			WARN_ON(req->r_unsafe_callback);
2880 			dout("req %p tid %llu cb (locked)\n", req, req->r_tid);
2881 			__complete_request(req);
2882 		}
2883 	}
2884 
2885 	mutex_unlock(&osd->lock);
2886 	up_read(&osdc->lock);
2887 
2888 	if (done_request(req, &m)) {
2889 		if (already_acked && req->r_unsafe_callback) {
2890 			dout("req %p tid %llu safe-cb\n", req, req->r_tid);
2891 			req->r_unsafe_callback(req, false);
2892 		} else if (!req->r_linger) {
2893 			dout("req %p tid %llu cb\n", req, req->r_tid);
2894 			__complete_request(req);
2895 		}
2896 	} else {
2897 		if (req->r_unsafe_callback) {
2898 			dout("req %p tid %llu unsafe-cb\n", req, req->r_tid);
2899 			req->r_unsafe_callback(req, true);
2900 		} else {
2901 			WARN_ON(1);
2902 		}
2903 	}
2904 	if (m.flags & CEPH_OSD_FLAG_ONDISK)
2905 		complete_all(&req->r_safe_completion);
2906 
2907 	ceph_osdc_put_request(req);
2908 	return;
2909 
2910 fail_request:
2911 	complete_request(req, -EIO);
2912 out_unlock_session:
2913 	mutex_unlock(&osd->lock);
2914 out_unlock_osdc:
2915 	up_read(&osdc->lock);
2916 }
2917 
2918 static void set_pool_was_full(struct ceph_osd_client *osdc)
2919 {
2920 	struct rb_node *n;
2921 
2922 	for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) {
2923 		struct ceph_pg_pool_info *pi =
2924 		    rb_entry(n, struct ceph_pg_pool_info, node);
2925 
2926 		pi->was_full = __pool_full(pi);
2927 	}
2928 }
2929 
2930 static bool pool_cleared_full(struct ceph_osd_client *osdc, s64 pool_id)
2931 {
2932 	struct ceph_pg_pool_info *pi;
2933 
2934 	pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id);
2935 	if (!pi)
2936 		return false;
2937 
2938 	return pi->was_full && !__pool_full(pi);
2939 }
2940 
2941 static enum calc_target_result
2942 recalc_linger_target(struct ceph_osd_linger_request *lreq)
2943 {
2944 	struct ceph_osd_client *osdc = lreq->osdc;
2945 	enum calc_target_result ct_res;
2946 
2947 	ct_res = calc_target(osdc, &lreq->t, &lreq->last_force_resend, true);
2948 	if (ct_res == CALC_TARGET_NEED_RESEND) {
2949 		struct ceph_osd *osd;
2950 
2951 		osd = lookup_create_osd(osdc, lreq->t.osd, true);
2952 		if (osd != lreq->osd) {
2953 			unlink_linger(lreq->osd, lreq);
2954 			link_linger(osd, lreq);
2955 		}
2956 	}
2957 
2958 	return ct_res;
2959 }
2960 
2961 /*
2962  * Requeue requests whose mapping to an OSD has changed.
2963  */
2964 static void scan_requests(struct ceph_osd *osd,
2965 			  bool force_resend,
2966 			  bool cleared_full,
2967 			  bool check_pool_cleared_full,
2968 			  struct rb_root *need_resend,
2969 			  struct list_head *need_resend_linger)
2970 {
2971 	struct ceph_osd_client *osdc = osd->o_osdc;
2972 	struct rb_node *n;
2973 	bool force_resend_writes;
2974 
2975 	for (n = rb_first(&osd->o_linger_requests); n; ) {
2976 		struct ceph_osd_linger_request *lreq =
2977 		    rb_entry(n, struct ceph_osd_linger_request, node);
2978 		enum calc_target_result ct_res;
2979 
2980 		n = rb_next(n); /* recalc_linger_target() */
2981 
2982 		dout("%s lreq %p linger_id %llu\n", __func__, lreq,
2983 		     lreq->linger_id);
2984 		ct_res = recalc_linger_target(lreq);
2985 		switch (ct_res) {
2986 		case CALC_TARGET_NO_ACTION:
2987 			force_resend_writes = cleared_full ||
2988 			    (check_pool_cleared_full &&
2989 			     pool_cleared_full(osdc, lreq->t.base_oloc.pool));
2990 			if (!force_resend && !force_resend_writes)
2991 				break;
2992 
2993 			/* fall through */
2994 		case CALC_TARGET_NEED_RESEND:
2995 			cancel_linger_map_check(lreq);
2996 			/*
2997 			 * scan_requests() for the previous epoch(s)
2998 			 * may have already added it to the list, since
2999 			 * it's not unlinked here.
3000 			 */
3001 			if (list_empty(&lreq->scan_item))
3002 				list_add_tail(&lreq->scan_item, need_resend_linger);
3003 			break;
3004 		case CALC_TARGET_POOL_DNE:
3005 			check_linger_pool_dne(lreq);
3006 			break;
3007 		}
3008 	}
3009 
3010 	for (n = rb_first(&osd->o_requests); n; ) {
3011 		struct ceph_osd_request *req =
3012 		    rb_entry(n, struct ceph_osd_request, r_node);
3013 		enum calc_target_result ct_res;
3014 
3015 		n = rb_next(n); /* unlink_request(), check_pool_dne() */
3016 
3017 		dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
3018 		ct_res = calc_target(osdc, &req->r_t,
3019 				     &req->r_last_force_resend, false);
3020 		switch (ct_res) {
3021 		case CALC_TARGET_NO_ACTION:
3022 			force_resend_writes = cleared_full ||
3023 			    (check_pool_cleared_full &&
3024 			     pool_cleared_full(osdc, req->r_t.base_oloc.pool));
3025 			if (!force_resend &&
3026 			    (!(req->r_flags & CEPH_OSD_FLAG_WRITE) ||
3027 			     !force_resend_writes))
3028 				break;
3029 
3030 			/* fall through */
3031 		case CALC_TARGET_NEED_RESEND:
3032 			cancel_map_check(req);
3033 			unlink_request(osd, req);
3034 			insert_request(need_resend, req);
3035 			break;
3036 		case CALC_TARGET_POOL_DNE:
3037 			check_pool_dne(req);
3038 			break;
3039 		}
3040 	}
3041 }
3042 
3043 static int handle_one_map(struct ceph_osd_client *osdc,
3044 			  void *p, void *end, bool incremental,
3045 			  struct rb_root *need_resend,
3046 			  struct list_head *need_resend_linger)
3047 {
3048 	struct ceph_osdmap *newmap;
3049 	struct rb_node *n;
3050 	bool skipped_map = false;
3051 	bool was_full;
3052 
3053 	was_full = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
3054 	set_pool_was_full(osdc);
3055 
3056 	if (incremental)
3057 		newmap = osdmap_apply_incremental(&p, end, osdc->osdmap);
3058 	else
3059 		newmap = ceph_osdmap_decode(&p, end);
3060 	if (IS_ERR(newmap))
3061 		return PTR_ERR(newmap);
3062 
3063 	if (newmap != osdc->osdmap) {
3064 		/*
3065 		 * Preserve ->was_full before destroying the old map.
3066 		 * For pools that weren't in the old map, ->was_full
3067 		 * should be false.
3068 		 */
3069 		for (n = rb_first(&newmap->pg_pools); n; n = rb_next(n)) {
3070 			struct ceph_pg_pool_info *pi =
3071 			    rb_entry(n, struct ceph_pg_pool_info, node);
3072 			struct ceph_pg_pool_info *old_pi;
3073 
3074 			old_pi = ceph_pg_pool_by_id(osdc->osdmap, pi->id);
3075 			if (old_pi)
3076 				pi->was_full = old_pi->was_full;
3077 			else
3078 				WARN_ON(pi->was_full);
3079 		}
3080 
3081 		if (osdc->osdmap->epoch &&
3082 		    osdc->osdmap->epoch + 1 < newmap->epoch) {
3083 			WARN_ON(incremental);
3084 			skipped_map = true;
3085 		}
3086 
3087 		ceph_osdmap_destroy(osdc->osdmap);
3088 		osdc->osdmap = newmap;
3089 	}
3090 
3091 	was_full &= !ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
3092 	scan_requests(&osdc->homeless_osd, skipped_map, was_full, true,
3093 		      need_resend, need_resend_linger);
3094 
3095 	for (n = rb_first(&osdc->osds); n; ) {
3096 		struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
3097 
3098 		n = rb_next(n); /* close_osd() */
3099 
3100 		scan_requests(osd, skipped_map, was_full, true, need_resend,
3101 			      need_resend_linger);
3102 		if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
3103 		    memcmp(&osd->o_con.peer_addr,
3104 			   ceph_osd_addr(osdc->osdmap, osd->o_osd),
3105 			   sizeof(struct ceph_entity_addr)))
3106 			close_osd(osd);
3107 	}
3108 
3109 	return 0;
3110 }
3111 
3112 static void kick_requests(struct ceph_osd_client *osdc,
3113 			  struct rb_root *need_resend,
3114 			  struct list_head *need_resend_linger)
3115 {
3116 	struct ceph_osd_linger_request *lreq, *nlreq;
3117 	struct rb_node *n;
3118 
3119 	for (n = rb_first(need_resend); n; ) {
3120 		struct ceph_osd_request *req =
3121 		    rb_entry(n, struct ceph_osd_request, r_node);
3122 		struct ceph_osd *osd;
3123 
3124 		n = rb_next(n);
3125 		erase_request(need_resend, req); /* before link_request() */
3126 
3127 		WARN_ON(req->r_osd);
3128 		calc_target(osdc, &req->r_t, NULL, false);
3129 		osd = lookup_create_osd(osdc, req->r_t.osd, true);
3130 		link_request(osd, req);
3131 		if (!req->r_linger) {
3132 			if (!osd_homeless(osd) && !req->r_t.paused)
3133 				send_request(req);
3134 		} else {
3135 			cancel_linger_request(req);
3136 		}
3137 	}
3138 
3139 	list_for_each_entry_safe(lreq, nlreq, need_resend_linger, scan_item) {
3140 		if (!osd_homeless(lreq->osd))
3141 			send_linger(lreq);
3142 
3143 		list_del_init(&lreq->scan_item);
3144 	}
3145 }
3146 
3147 /*
3148  * Process updated osd map.
3149  *
3150  * The message contains any number of incremental and full maps, normally
3151  * indicating some sort of topology change in the cluster.  Kick requests
3152  * off to different OSDs as needed.
3153  */
3154 void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
3155 {
3156 	void *p = msg->front.iov_base;
3157 	void *const end = p + msg->front.iov_len;
3158 	u32 nr_maps, maplen;
3159 	u32 epoch;
3160 	struct ceph_fsid fsid;
3161 	struct rb_root need_resend = RB_ROOT;
3162 	LIST_HEAD(need_resend_linger);
3163 	bool handled_incremental = false;
3164 	bool was_pauserd, was_pausewr;
3165 	bool pauserd, pausewr;
3166 	int err;
3167 
3168 	dout("%s have %u\n", __func__, osdc->osdmap->epoch);
3169 	down_write(&osdc->lock);
3170 
3171 	/* verify fsid */
3172 	ceph_decode_need(&p, end, sizeof(fsid), bad);
3173 	ceph_decode_copy(&p, &fsid, sizeof(fsid));
3174 	if (ceph_check_fsid(osdc->client, &fsid) < 0)
3175 		goto bad;
3176 
3177 	was_pauserd = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD);
3178 	was_pausewr = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR) ||
3179 		      ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) ||
3180 		      have_pool_full(osdc);
3181 
3182 	/* incremental maps */
3183 	ceph_decode_32_safe(&p, end, nr_maps, bad);
3184 	dout(" %d inc maps\n", nr_maps);
3185 	while (nr_maps > 0) {
3186 		ceph_decode_need(&p, end, 2*sizeof(u32), bad);
3187 		epoch = ceph_decode_32(&p);
3188 		maplen = ceph_decode_32(&p);
3189 		ceph_decode_need(&p, end, maplen, bad);
3190 		if (osdc->osdmap->epoch &&
3191 		    osdc->osdmap->epoch + 1 == epoch) {
3192 			dout("applying incremental map %u len %d\n",
3193 			     epoch, maplen);
3194 			err = handle_one_map(osdc, p, p + maplen, true,
3195 					     &need_resend, &need_resend_linger);
3196 			if (err)
3197 				goto bad;
3198 			handled_incremental = true;
3199 		} else {
3200 			dout("ignoring incremental map %u len %d\n",
3201 			     epoch, maplen);
3202 		}
3203 		p += maplen;
3204 		nr_maps--;
3205 	}
3206 	if (handled_incremental)
3207 		goto done;
3208 
3209 	/* full maps */
3210 	ceph_decode_32_safe(&p, end, nr_maps, bad);
3211 	dout(" %d full maps\n", nr_maps);
3212 	while (nr_maps) {
3213 		ceph_decode_need(&p, end, 2*sizeof(u32), bad);
3214 		epoch = ceph_decode_32(&p);
3215 		maplen = ceph_decode_32(&p);
3216 		ceph_decode_need(&p, end, maplen, bad);
3217 		if (nr_maps > 1) {
3218 			dout("skipping non-latest full map %u len %d\n",
3219 			     epoch, maplen);
3220 		} else if (osdc->osdmap->epoch >= epoch) {
3221 			dout("skipping full map %u len %d, "
3222 			     "older than our %u\n", epoch, maplen,
3223 			     osdc->osdmap->epoch);
3224 		} else {
3225 			dout("taking full map %u len %d\n", epoch, maplen);
3226 			err = handle_one_map(osdc, p, p + maplen, false,
3227 					     &need_resend, &need_resend_linger);
3228 			if (err)
3229 				goto bad;
3230 		}
3231 		p += maplen;
3232 		nr_maps--;
3233 	}
3234 
3235 done:
3236 	/*
3237 	 * subscribe to subsequent osdmap updates if full to ensure
3238 	 * we find out when we are no longer full and stop returning
3239 	 * ENOSPC.
3240 	 */
3241 	pauserd = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD);
3242 	pausewr = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR) ||
3243 		  ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) ||
3244 		  have_pool_full(osdc);
3245 	if (was_pauserd || was_pausewr || pauserd || pausewr)
3246 		maybe_request_map(osdc);
3247 
3248 	kick_requests(osdc, &need_resend, &need_resend_linger);
3249 
3250 	ceph_monc_got_map(&osdc->client->monc, CEPH_SUB_OSDMAP,
3251 			  osdc->osdmap->epoch);
3252 	up_write(&osdc->lock);
3253 	wake_up_all(&osdc->client->auth_wq);
3254 	return;
3255 
3256 bad:
3257 	pr_err("osdc handle_map corrupt msg\n");
3258 	ceph_msg_dump(msg);
3259 	up_write(&osdc->lock);
3260 }
3261 
3262 /*
3263  * Resubmit requests pending on the given osd.
3264  */
3265 static void kick_osd_requests(struct ceph_osd *osd)
3266 {
3267 	struct rb_node *n;
3268 
3269 	for (n = rb_first(&osd->o_requests); n; ) {
3270 		struct ceph_osd_request *req =
3271 		    rb_entry(n, struct ceph_osd_request, r_node);
3272 
3273 		n = rb_next(n); /* cancel_linger_request() */
3274 
3275 		if (!req->r_linger) {
3276 			if (!req->r_t.paused)
3277 				send_request(req);
3278 		} else {
3279 			cancel_linger_request(req);
3280 		}
3281 	}
3282 	for (n = rb_first(&osd->o_linger_requests); n; n = rb_next(n)) {
3283 		struct ceph_osd_linger_request *lreq =
3284 		    rb_entry(n, struct ceph_osd_linger_request, node);
3285 
3286 		send_linger(lreq);
3287 	}
3288 }
3289 
3290 /*
3291  * If the osd connection drops, we need to resubmit all requests.
3292  */
3293 static void osd_fault(struct ceph_connection *con)
3294 {
3295 	struct ceph_osd *osd = con->private;
3296 	struct ceph_osd_client *osdc = osd->o_osdc;
3297 
3298 	dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
3299 
3300 	down_write(&osdc->lock);
3301 	if (!osd_registered(osd)) {
3302 		dout("%s osd%d unknown\n", __func__, osd->o_osd);
3303 		goto out_unlock;
3304 	}
3305 
3306 	if (!reopen_osd(osd))
3307 		kick_osd_requests(osd);
3308 	maybe_request_map(osdc);
3309 
3310 out_unlock:
3311 	up_write(&osdc->lock);
3312 }
3313 
3314 /*
3315  * Process osd watch notifications
3316  */
3317 static void handle_watch_notify(struct ceph_osd_client *osdc,
3318 				struct ceph_msg *msg)
3319 {
3320 	void *p = msg->front.iov_base;
3321 	void *const end = p + msg->front.iov_len;
3322 	struct ceph_osd_linger_request *lreq;
3323 	struct linger_work *lwork;
3324 	u8 proto_ver, opcode;
3325 	u64 cookie, notify_id;
3326 	u64 notifier_id = 0;
3327 	s32 return_code = 0;
3328 	void *payload = NULL;
3329 	u32 payload_len = 0;
3330 
3331 	ceph_decode_8_safe(&p, end, proto_ver, bad);
3332 	ceph_decode_8_safe(&p, end, opcode, bad);
3333 	ceph_decode_64_safe(&p, end, cookie, bad);
3334 	p += 8; /* skip ver */
3335 	ceph_decode_64_safe(&p, end, notify_id, bad);
3336 
3337 	if (proto_ver >= 1) {
3338 		ceph_decode_32_safe(&p, end, payload_len, bad);
3339 		ceph_decode_need(&p, end, payload_len, bad);
3340 		payload = p;
3341 		p += payload_len;
3342 	}
3343 
3344 	if (le16_to_cpu(msg->hdr.version) >= 2)
3345 		ceph_decode_32_safe(&p, end, return_code, bad);
3346 
3347 	if (le16_to_cpu(msg->hdr.version) >= 3)
3348 		ceph_decode_64_safe(&p, end, notifier_id, bad);
3349 
3350 	down_read(&osdc->lock);
3351 	lreq = lookup_linger_osdc(&osdc->linger_requests, cookie);
3352 	if (!lreq) {
3353 		dout("%s opcode %d cookie %llu dne\n", __func__, opcode,
3354 		     cookie);
3355 		goto out_unlock_osdc;
3356 	}
3357 
3358 	mutex_lock(&lreq->lock);
3359 	dout("%s opcode %d cookie %llu lreq %p is_watch %d\n", __func__,
3360 	     opcode, cookie, lreq, lreq->is_watch);
3361 	if (opcode == CEPH_WATCH_EVENT_DISCONNECT) {
3362 		if (!lreq->last_error) {
3363 			lreq->last_error = -ENOTCONN;
3364 			queue_watch_error(lreq);
3365 		}
3366 	} else if (!lreq->is_watch) {
3367 		/* CEPH_WATCH_EVENT_NOTIFY_COMPLETE */
3368 		if (lreq->notify_id && lreq->notify_id != notify_id) {
3369 			dout("lreq %p notify_id %llu != %llu, ignoring\n", lreq,
3370 			     lreq->notify_id, notify_id);
3371 		} else if (!completion_done(&lreq->notify_finish_wait)) {
3372 			struct ceph_msg_data *data =
3373 			    list_first_entry_or_null(&msg->data,
3374 						     struct ceph_msg_data,
3375 						     links);
3376 
3377 			if (data) {
3378 				if (lreq->preply_pages) {
3379 					WARN_ON(data->type !=
3380 							CEPH_MSG_DATA_PAGES);
3381 					*lreq->preply_pages = data->pages;
3382 					*lreq->preply_len = data->length;
3383 				} else {
3384 					ceph_release_page_vector(data->pages,
3385 					       calc_pages_for(0, data->length));
3386 				}
3387 			}
3388 			lreq->notify_finish_error = return_code;
3389 			complete_all(&lreq->notify_finish_wait);
3390 		}
3391 	} else {
3392 		/* CEPH_WATCH_EVENT_NOTIFY */
3393 		lwork = lwork_alloc(lreq, do_watch_notify);
3394 		if (!lwork) {
3395 			pr_err("failed to allocate notify-lwork\n");
3396 			goto out_unlock_lreq;
3397 		}
3398 
3399 		lwork->notify.notify_id = notify_id;
3400 		lwork->notify.notifier_id = notifier_id;
3401 		lwork->notify.payload = payload;
3402 		lwork->notify.payload_len = payload_len;
3403 		lwork->notify.msg = ceph_msg_get(msg);
3404 		lwork_queue(lwork);
3405 	}
3406 
3407 out_unlock_lreq:
3408 	mutex_unlock(&lreq->lock);
3409 out_unlock_osdc:
3410 	up_read(&osdc->lock);
3411 	return;
3412 
3413 bad:
3414 	pr_err("osdc handle_watch_notify corrupt msg\n");
3415 }
3416 
3417 /*
3418  * Register request, send initial attempt.
3419  */
3420 int ceph_osdc_start_request(struct ceph_osd_client *osdc,
3421 			    struct ceph_osd_request *req,
3422 			    bool nofail)
3423 {
3424 	down_read(&osdc->lock);
3425 	submit_request(req, false);
3426 	up_read(&osdc->lock);
3427 
3428 	return 0;
3429 }
3430 EXPORT_SYMBOL(ceph_osdc_start_request);
3431 
3432 /*
3433  * Unregister a registered request.  The request is not completed (i.e.
3434  * no callbacks or wakeups) - higher layers are supposed to know what
3435  * they are canceling.
3436  */
3437 void ceph_osdc_cancel_request(struct ceph_osd_request *req)
3438 {
3439 	struct ceph_osd_client *osdc = req->r_osdc;
3440 
3441 	down_write(&osdc->lock);
3442 	if (req->r_osd)
3443 		cancel_request(req);
3444 	up_write(&osdc->lock);
3445 }
3446 EXPORT_SYMBOL(ceph_osdc_cancel_request);
3447 
3448 /*
3449  * @timeout: in jiffies, 0 means "wait forever"
3450  */
3451 static int wait_request_timeout(struct ceph_osd_request *req,
3452 				unsigned long timeout)
3453 {
3454 	long left;
3455 
3456 	dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
3457 	left = wait_for_completion_killable_timeout(&req->r_completion,
3458 						ceph_timeout_jiffies(timeout));
3459 	if (left <= 0) {
3460 		left = left ?: -ETIMEDOUT;
3461 		ceph_osdc_cancel_request(req);
3462 
3463 		/* kludge - need to to wake ceph_osdc_sync() */
3464 		complete_all(&req->r_safe_completion);
3465 	} else {
3466 		left = req->r_result; /* completed */
3467 	}
3468 
3469 	return left;
3470 }
3471 
3472 /*
3473  * wait for a request to complete
3474  */
3475 int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
3476 			   struct ceph_osd_request *req)
3477 {
3478 	return wait_request_timeout(req, 0);
3479 }
3480 EXPORT_SYMBOL(ceph_osdc_wait_request);
3481 
3482 /*
3483  * sync - wait for all in-flight requests to flush.  avoid starvation.
3484  */
3485 void ceph_osdc_sync(struct ceph_osd_client *osdc)
3486 {
3487 	struct rb_node *n, *p;
3488 	u64 last_tid = atomic64_read(&osdc->last_tid);
3489 
3490 again:
3491 	down_read(&osdc->lock);
3492 	for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
3493 		struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
3494 
3495 		mutex_lock(&osd->lock);
3496 		for (p = rb_first(&osd->o_requests); p; p = rb_next(p)) {
3497 			struct ceph_osd_request *req =
3498 			    rb_entry(p, struct ceph_osd_request, r_node);
3499 
3500 			if (req->r_tid > last_tid)
3501 				break;
3502 
3503 			if (!(req->r_flags & CEPH_OSD_FLAG_WRITE))
3504 				continue;
3505 
3506 			ceph_osdc_get_request(req);
3507 			mutex_unlock(&osd->lock);
3508 			up_read(&osdc->lock);
3509 			dout("%s waiting on req %p tid %llu last_tid %llu\n",
3510 			     __func__, req, req->r_tid, last_tid);
3511 			wait_for_completion(&req->r_safe_completion);
3512 			ceph_osdc_put_request(req);
3513 			goto again;
3514 		}
3515 
3516 		mutex_unlock(&osd->lock);
3517 	}
3518 
3519 	up_read(&osdc->lock);
3520 	dout("%s done last_tid %llu\n", __func__, last_tid);
3521 }
3522 EXPORT_SYMBOL(ceph_osdc_sync);
3523 
3524 static struct ceph_osd_request *
3525 alloc_linger_request(struct ceph_osd_linger_request *lreq)
3526 {
3527 	struct ceph_osd_request *req;
3528 
3529 	req = ceph_osdc_alloc_request(lreq->osdc, NULL, 1, false, GFP_NOIO);
3530 	if (!req)
3531 		return NULL;
3532 
3533 	ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
3534 	ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
3535 
3536 	if (ceph_osdc_alloc_messages(req, GFP_NOIO)) {
3537 		ceph_osdc_put_request(req);
3538 		return NULL;
3539 	}
3540 
3541 	return req;
3542 }
3543 
3544 /*
3545  * Returns a handle, caller owns a ref.
3546  */
3547 struct ceph_osd_linger_request *
3548 ceph_osdc_watch(struct ceph_osd_client *osdc,
3549 		struct ceph_object_id *oid,
3550 		struct ceph_object_locator *oloc,
3551 		rados_watchcb2_t wcb,
3552 		rados_watcherrcb_t errcb,
3553 		void *data)
3554 {
3555 	struct ceph_osd_linger_request *lreq;
3556 	int ret;
3557 
3558 	lreq = linger_alloc(osdc);
3559 	if (!lreq)
3560 		return ERR_PTR(-ENOMEM);
3561 
3562 	lreq->is_watch = true;
3563 	lreq->wcb = wcb;
3564 	lreq->errcb = errcb;
3565 	lreq->data = data;
3566 	lreq->watch_valid_thru = jiffies;
3567 
3568 	ceph_oid_copy(&lreq->t.base_oid, oid);
3569 	ceph_oloc_copy(&lreq->t.base_oloc, oloc);
3570 	lreq->t.flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
3571 	lreq->mtime = CURRENT_TIME;
3572 
3573 	lreq->reg_req = alloc_linger_request(lreq);
3574 	if (!lreq->reg_req) {
3575 		ret = -ENOMEM;
3576 		goto err_put_lreq;
3577 	}
3578 
3579 	lreq->ping_req = alloc_linger_request(lreq);
3580 	if (!lreq->ping_req) {
3581 		ret = -ENOMEM;
3582 		goto err_put_lreq;
3583 	}
3584 
3585 	down_write(&osdc->lock);
3586 	linger_register(lreq); /* before osd_req_op_* */
3587 	osd_req_op_watch_init(lreq->reg_req, 0, lreq->linger_id,
3588 			      CEPH_OSD_WATCH_OP_WATCH);
3589 	osd_req_op_watch_init(lreq->ping_req, 0, lreq->linger_id,
3590 			      CEPH_OSD_WATCH_OP_PING);
3591 	linger_submit(lreq);
3592 	up_write(&osdc->lock);
3593 
3594 	ret = linger_reg_commit_wait(lreq);
3595 	if (ret) {
3596 		linger_cancel(lreq);
3597 		goto err_put_lreq;
3598 	}
3599 
3600 	return lreq;
3601 
3602 err_put_lreq:
3603 	linger_put(lreq);
3604 	return ERR_PTR(ret);
3605 }
3606 EXPORT_SYMBOL(ceph_osdc_watch);
3607 
3608 /*
3609  * Releases a ref.
3610  *
3611  * Times out after mount_timeout to preserve rbd unmap behaviour
3612  * introduced in 2894e1d76974 ("rbd: timeout watch teardown on unmap
3613  * with mount_timeout").
3614  */
3615 int ceph_osdc_unwatch(struct ceph_osd_client *osdc,
3616 		      struct ceph_osd_linger_request *lreq)
3617 {
3618 	struct ceph_options *opts = osdc->client->options;
3619 	struct ceph_osd_request *req;
3620 	int ret;
3621 
3622 	req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
3623 	if (!req)
3624 		return -ENOMEM;
3625 
3626 	ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
3627 	ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
3628 	req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
3629 	req->r_mtime = CURRENT_TIME;
3630 	osd_req_op_watch_init(req, 0, lreq->linger_id,
3631 			      CEPH_OSD_WATCH_OP_UNWATCH);
3632 
3633 	ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
3634 	if (ret)
3635 		goto out_put_req;
3636 
3637 	ceph_osdc_start_request(osdc, req, false);
3638 	linger_cancel(lreq);
3639 	linger_put(lreq);
3640 	ret = wait_request_timeout(req, opts->mount_timeout);
3641 
3642 out_put_req:
3643 	ceph_osdc_put_request(req);
3644 	return ret;
3645 }
3646 EXPORT_SYMBOL(ceph_osdc_unwatch);
3647 
3648 static int osd_req_op_notify_ack_init(struct ceph_osd_request *req, int which,
3649 				      u64 notify_id, u64 cookie, void *payload,
3650 				      size_t payload_len)
3651 {
3652 	struct ceph_osd_req_op *op;
3653 	struct ceph_pagelist *pl;
3654 	int ret;
3655 
3656 	op = _osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY_ACK, 0);
3657 
3658 	pl = kmalloc(sizeof(*pl), GFP_NOIO);
3659 	if (!pl)
3660 		return -ENOMEM;
3661 
3662 	ceph_pagelist_init(pl);
3663 	ret = ceph_pagelist_encode_64(pl, notify_id);
3664 	ret |= ceph_pagelist_encode_64(pl, cookie);
3665 	if (payload) {
3666 		ret |= ceph_pagelist_encode_32(pl, payload_len);
3667 		ret |= ceph_pagelist_append(pl, payload, payload_len);
3668 	} else {
3669 		ret |= ceph_pagelist_encode_32(pl, 0);
3670 	}
3671 	if (ret) {
3672 		ceph_pagelist_release(pl);
3673 		return -ENOMEM;
3674 	}
3675 
3676 	ceph_osd_data_pagelist_init(&op->notify_ack.request_data, pl);
3677 	op->indata_len = pl->length;
3678 	return 0;
3679 }
3680 
3681 int ceph_osdc_notify_ack(struct ceph_osd_client *osdc,
3682 			 struct ceph_object_id *oid,
3683 			 struct ceph_object_locator *oloc,
3684 			 u64 notify_id,
3685 			 u64 cookie,
3686 			 void *payload,
3687 			 size_t payload_len)
3688 {
3689 	struct ceph_osd_request *req;
3690 	int ret;
3691 
3692 	req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
3693 	if (!req)
3694 		return -ENOMEM;
3695 
3696 	ceph_oid_copy(&req->r_base_oid, oid);
3697 	ceph_oloc_copy(&req->r_base_oloc, oloc);
3698 	req->r_flags = CEPH_OSD_FLAG_READ;
3699 
3700 	ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
3701 	if (ret)
3702 		goto out_put_req;
3703 
3704 	ret = osd_req_op_notify_ack_init(req, 0, notify_id, cookie, payload,
3705 					 payload_len);
3706 	if (ret)
3707 		goto out_put_req;
3708 
3709 	ceph_osdc_start_request(osdc, req, false);
3710 	ret = ceph_osdc_wait_request(osdc, req);
3711 
3712 out_put_req:
3713 	ceph_osdc_put_request(req);
3714 	return ret;
3715 }
3716 EXPORT_SYMBOL(ceph_osdc_notify_ack);
3717 
3718 static int osd_req_op_notify_init(struct ceph_osd_request *req, int which,
3719 				  u64 cookie, u32 prot_ver, u32 timeout,
3720 				  void *payload, size_t payload_len)
3721 {
3722 	struct ceph_osd_req_op *op;
3723 	struct ceph_pagelist *pl;
3724 	int ret;
3725 
3726 	op = _osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY, 0);
3727 	op->notify.cookie = cookie;
3728 
3729 	pl = kmalloc(sizeof(*pl), GFP_NOIO);
3730 	if (!pl)
3731 		return -ENOMEM;
3732 
3733 	ceph_pagelist_init(pl);
3734 	ret = ceph_pagelist_encode_32(pl, 1); /* prot_ver */
3735 	ret |= ceph_pagelist_encode_32(pl, timeout);
3736 	ret |= ceph_pagelist_encode_32(pl, payload_len);
3737 	ret |= ceph_pagelist_append(pl, payload, payload_len);
3738 	if (ret) {
3739 		ceph_pagelist_release(pl);
3740 		return -ENOMEM;
3741 	}
3742 
3743 	ceph_osd_data_pagelist_init(&op->notify.request_data, pl);
3744 	op->indata_len = pl->length;
3745 	return 0;
3746 }
3747 
3748 /*
3749  * @timeout: in seconds
3750  *
3751  * @preply_{pages,len} are initialized both on success and error.
3752  * The caller is responsible for:
3753  *
3754  *     ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len))
3755  */
3756 int ceph_osdc_notify(struct ceph_osd_client *osdc,
3757 		     struct ceph_object_id *oid,
3758 		     struct ceph_object_locator *oloc,
3759 		     void *payload,
3760 		     size_t payload_len,
3761 		     u32 timeout,
3762 		     struct page ***preply_pages,
3763 		     size_t *preply_len)
3764 {
3765 	struct ceph_osd_linger_request *lreq;
3766 	struct page **pages;
3767 	int ret;
3768 
3769 	WARN_ON(!timeout);
3770 	if (preply_pages) {
3771 		*preply_pages = NULL;
3772 		*preply_len = 0;
3773 	}
3774 
3775 	lreq = linger_alloc(osdc);
3776 	if (!lreq)
3777 		return -ENOMEM;
3778 
3779 	lreq->preply_pages = preply_pages;
3780 	lreq->preply_len = preply_len;
3781 
3782 	ceph_oid_copy(&lreq->t.base_oid, oid);
3783 	ceph_oloc_copy(&lreq->t.base_oloc, oloc);
3784 	lreq->t.flags = CEPH_OSD_FLAG_READ;
3785 
3786 	lreq->reg_req = alloc_linger_request(lreq);
3787 	if (!lreq->reg_req) {
3788 		ret = -ENOMEM;
3789 		goto out_put_lreq;
3790 	}
3791 
3792 	/* for notify_id */
3793 	pages = ceph_alloc_page_vector(1, GFP_NOIO);
3794 	if (IS_ERR(pages)) {
3795 		ret = PTR_ERR(pages);
3796 		goto out_put_lreq;
3797 	}
3798 
3799 	down_write(&osdc->lock);
3800 	linger_register(lreq); /* before osd_req_op_* */
3801 	ret = osd_req_op_notify_init(lreq->reg_req, 0, lreq->linger_id, 1,
3802 				     timeout, payload, payload_len);
3803 	if (ret) {
3804 		linger_unregister(lreq);
3805 		up_write(&osdc->lock);
3806 		ceph_release_page_vector(pages, 1);
3807 		goto out_put_lreq;
3808 	}
3809 	ceph_osd_data_pages_init(osd_req_op_data(lreq->reg_req, 0, notify,
3810 						 response_data),
3811 				 pages, PAGE_SIZE, 0, false, true);
3812 	linger_submit(lreq);
3813 	up_write(&osdc->lock);
3814 
3815 	ret = linger_reg_commit_wait(lreq);
3816 	if (!ret)
3817 		ret = linger_notify_finish_wait(lreq);
3818 	else
3819 		dout("lreq %p failed to initiate notify %d\n", lreq, ret);
3820 
3821 	linger_cancel(lreq);
3822 out_put_lreq:
3823 	linger_put(lreq);
3824 	return ret;
3825 }
3826 EXPORT_SYMBOL(ceph_osdc_notify);
3827 
3828 /*
3829  * Return the number of milliseconds since the watch was last
3830  * confirmed, or an error.  If there is an error, the watch is no
3831  * longer valid, and should be destroyed with ceph_osdc_unwatch().
3832  */
3833 int ceph_osdc_watch_check(struct ceph_osd_client *osdc,
3834 			  struct ceph_osd_linger_request *lreq)
3835 {
3836 	unsigned long stamp, age;
3837 	int ret;
3838 
3839 	down_read(&osdc->lock);
3840 	mutex_lock(&lreq->lock);
3841 	stamp = lreq->watch_valid_thru;
3842 	if (!list_empty(&lreq->pending_lworks)) {
3843 		struct linger_work *lwork =
3844 		    list_first_entry(&lreq->pending_lworks,
3845 				     struct linger_work,
3846 				     pending_item);
3847 
3848 		if (time_before(lwork->queued_stamp, stamp))
3849 			stamp = lwork->queued_stamp;
3850 	}
3851 	age = jiffies - stamp;
3852 	dout("%s lreq %p linger_id %llu age %lu last_error %d\n", __func__,
3853 	     lreq, lreq->linger_id, age, lreq->last_error);
3854 	/* we are truncating to msecs, so return a safe upper bound */
3855 	ret = lreq->last_error ?: 1 + jiffies_to_msecs(age);
3856 
3857 	mutex_unlock(&lreq->lock);
3858 	up_read(&osdc->lock);
3859 	return ret;
3860 }
3861 
3862 /*
3863  * Call all pending notify callbacks - for use after a watch is
3864  * unregistered, to make sure no more callbacks for it will be invoked
3865  */
3866 void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc)
3867 {
3868 	flush_workqueue(osdc->notify_wq);
3869 }
3870 EXPORT_SYMBOL(ceph_osdc_flush_notifies);
3871 
3872 void ceph_osdc_maybe_request_map(struct ceph_osd_client *osdc)
3873 {
3874 	down_read(&osdc->lock);
3875 	maybe_request_map(osdc);
3876 	up_read(&osdc->lock);
3877 }
3878 EXPORT_SYMBOL(ceph_osdc_maybe_request_map);
3879 
3880 /*
3881  * init, shutdown
3882  */
3883 int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
3884 {
3885 	int err;
3886 
3887 	dout("init\n");
3888 	osdc->client = client;
3889 	init_rwsem(&osdc->lock);
3890 	osdc->osds = RB_ROOT;
3891 	INIT_LIST_HEAD(&osdc->osd_lru);
3892 	spin_lock_init(&osdc->osd_lru_lock);
3893 	osd_init(&osdc->homeless_osd);
3894 	osdc->homeless_osd.o_osdc = osdc;
3895 	osdc->homeless_osd.o_osd = CEPH_HOMELESS_OSD;
3896 	osdc->linger_requests = RB_ROOT;
3897 	osdc->map_checks = RB_ROOT;
3898 	osdc->linger_map_checks = RB_ROOT;
3899 	INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
3900 	INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
3901 
3902 	err = -ENOMEM;
3903 	osdc->osdmap = ceph_osdmap_alloc();
3904 	if (!osdc->osdmap)
3905 		goto out;
3906 
3907 	osdc->req_mempool = mempool_create_slab_pool(10,
3908 						     ceph_osd_request_cache);
3909 	if (!osdc->req_mempool)
3910 		goto out_map;
3911 
3912 	err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
3913 				PAGE_SIZE, 10, true, "osd_op");
3914 	if (err < 0)
3915 		goto out_mempool;
3916 	err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
3917 				PAGE_SIZE, 10, true, "osd_op_reply");
3918 	if (err < 0)
3919 		goto out_msgpool;
3920 
3921 	err = -ENOMEM;
3922 	osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
3923 	if (!osdc->notify_wq)
3924 		goto out_msgpool_reply;
3925 
3926 	schedule_delayed_work(&osdc->timeout_work,
3927 			      osdc->client->options->osd_keepalive_timeout);
3928 	schedule_delayed_work(&osdc->osds_timeout_work,
3929 	    round_jiffies_relative(osdc->client->options->osd_idle_ttl));
3930 
3931 	return 0;
3932 
3933 out_msgpool_reply:
3934 	ceph_msgpool_destroy(&osdc->msgpool_op_reply);
3935 out_msgpool:
3936 	ceph_msgpool_destroy(&osdc->msgpool_op);
3937 out_mempool:
3938 	mempool_destroy(osdc->req_mempool);
3939 out_map:
3940 	ceph_osdmap_destroy(osdc->osdmap);
3941 out:
3942 	return err;
3943 }
3944 
3945 void ceph_osdc_stop(struct ceph_osd_client *osdc)
3946 {
3947 	flush_workqueue(osdc->notify_wq);
3948 	destroy_workqueue(osdc->notify_wq);
3949 	cancel_delayed_work_sync(&osdc->timeout_work);
3950 	cancel_delayed_work_sync(&osdc->osds_timeout_work);
3951 
3952 	down_write(&osdc->lock);
3953 	while (!RB_EMPTY_ROOT(&osdc->osds)) {
3954 		struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
3955 						struct ceph_osd, o_node);
3956 		close_osd(osd);
3957 	}
3958 	up_write(&osdc->lock);
3959 	WARN_ON(atomic_read(&osdc->homeless_osd.o_ref) != 1);
3960 	osd_cleanup(&osdc->homeless_osd);
3961 
3962 	WARN_ON(!list_empty(&osdc->osd_lru));
3963 	WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_requests));
3964 	WARN_ON(!RB_EMPTY_ROOT(&osdc->map_checks));
3965 	WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_map_checks));
3966 	WARN_ON(atomic_read(&osdc->num_requests));
3967 	WARN_ON(atomic_read(&osdc->num_homeless));
3968 
3969 	ceph_osdmap_destroy(osdc->osdmap);
3970 	mempool_destroy(osdc->req_mempool);
3971 	ceph_msgpool_destroy(&osdc->msgpool_op);
3972 	ceph_msgpool_destroy(&osdc->msgpool_op_reply);
3973 }
3974 
3975 /*
3976  * Read some contiguous pages.  If we cross a stripe boundary, shorten
3977  * *plen.  Return number of bytes read, or error.
3978  */
3979 int ceph_osdc_readpages(struct ceph_osd_client *osdc,
3980 			struct ceph_vino vino, struct ceph_file_layout *layout,
3981 			u64 off, u64 *plen,
3982 			u32 truncate_seq, u64 truncate_size,
3983 			struct page **pages, int num_pages, int page_align)
3984 {
3985 	struct ceph_osd_request *req;
3986 	int rc = 0;
3987 
3988 	dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
3989 	     vino.snap, off, *plen);
3990 	req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 0, 1,
3991 				    CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
3992 				    NULL, truncate_seq, truncate_size,
3993 				    false);
3994 	if (IS_ERR(req))
3995 		return PTR_ERR(req);
3996 
3997 	/* it may be a short read due to an object boundary */
3998 	osd_req_op_extent_osd_data_pages(req, 0,
3999 				pages, *plen, page_align, false, false);
4000 
4001 	dout("readpages  final extent is %llu~%llu (%llu bytes align %d)\n",
4002 	     off, *plen, *plen, page_align);
4003 
4004 	rc = ceph_osdc_start_request(osdc, req, false);
4005 	if (!rc)
4006 		rc = ceph_osdc_wait_request(osdc, req);
4007 
4008 	ceph_osdc_put_request(req);
4009 	dout("readpages result %d\n", rc);
4010 	return rc;
4011 }
4012 EXPORT_SYMBOL(ceph_osdc_readpages);
4013 
4014 /*
4015  * do a synchronous write on N pages
4016  */
4017 int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
4018 			 struct ceph_file_layout *layout,
4019 			 struct ceph_snap_context *snapc,
4020 			 u64 off, u64 len,
4021 			 u32 truncate_seq, u64 truncate_size,
4022 			 struct timespec *mtime,
4023 			 struct page **pages, int num_pages)
4024 {
4025 	struct ceph_osd_request *req;
4026 	int rc = 0;
4027 	int page_align = off & ~PAGE_MASK;
4028 
4029 	req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 0, 1,
4030 				    CEPH_OSD_OP_WRITE,
4031 				    CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
4032 				    snapc, truncate_seq, truncate_size,
4033 				    true);
4034 	if (IS_ERR(req))
4035 		return PTR_ERR(req);
4036 
4037 	/* it may be a short write due to an object boundary */
4038 	osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align,
4039 				false, false);
4040 	dout("writepages %llu~%llu (%llu bytes)\n", off, len, len);
4041 
4042 	req->r_mtime = *mtime;
4043 	rc = ceph_osdc_start_request(osdc, req, true);
4044 	if (!rc)
4045 		rc = ceph_osdc_wait_request(osdc, req);
4046 
4047 	ceph_osdc_put_request(req);
4048 	if (rc == 0)
4049 		rc = len;
4050 	dout("writepages result %d\n", rc);
4051 	return rc;
4052 }
4053 EXPORT_SYMBOL(ceph_osdc_writepages);
4054 
4055 int ceph_osdc_setup(void)
4056 {
4057 	size_t size = sizeof(struct ceph_osd_request) +
4058 	    CEPH_OSD_SLAB_OPS * sizeof(struct ceph_osd_req_op);
4059 
4060 	BUG_ON(ceph_osd_request_cache);
4061 	ceph_osd_request_cache = kmem_cache_create("ceph_osd_request", size,
4062 						   0, 0, NULL);
4063 
4064 	return ceph_osd_request_cache ? 0 : -ENOMEM;
4065 }
4066 EXPORT_SYMBOL(ceph_osdc_setup);
4067 
4068 void ceph_osdc_cleanup(void)
4069 {
4070 	BUG_ON(!ceph_osd_request_cache);
4071 	kmem_cache_destroy(ceph_osd_request_cache);
4072 	ceph_osd_request_cache = NULL;
4073 }
4074 EXPORT_SYMBOL(ceph_osdc_cleanup);
4075 
4076 /*
4077  * handle incoming message
4078  */
4079 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
4080 {
4081 	struct ceph_osd *osd = con->private;
4082 	struct ceph_osd_client *osdc = osd->o_osdc;
4083 	int type = le16_to_cpu(msg->hdr.type);
4084 
4085 	switch (type) {
4086 	case CEPH_MSG_OSD_MAP:
4087 		ceph_osdc_handle_map(osdc, msg);
4088 		break;
4089 	case CEPH_MSG_OSD_OPREPLY:
4090 		handle_reply(osd, msg);
4091 		break;
4092 	case CEPH_MSG_WATCH_NOTIFY:
4093 		handle_watch_notify(osdc, msg);
4094 		break;
4095 
4096 	default:
4097 		pr_err("received unknown message type %d %s\n", type,
4098 		       ceph_msg_type_name(type));
4099 	}
4100 
4101 	ceph_msg_put(msg);
4102 }
4103 
4104 /*
4105  * Lookup and return message for incoming reply.  Don't try to do
4106  * anything about a larger than preallocated data portion of the
4107  * message at the moment - for now, just skip the message.
4108  */
4109 static struct ceph_msg *get_reply(struct ceph_connection *con,
4110 				  struct ceph_msg_header *hdr,
4111 				  int *skip)
4112 {
4113 	struct ceph_osd *osd = con->private;
4114 	struct ceph_osd_client *osdc = osd->o_osdc;
4115 	struct ceph_msg *m = NULL;
4116 	struct ceph_osd_request *req;
4117 	int front_len = le32_to_cpu(hdr->front_len);
4118 	int data_len = le32_to_cpu(hdr->data_len);
4119 	u64 tid = le64_to_cpu(hdr->tid);
4120 
4121 	down_read(&osdc->lock);
4122 	if (!osd_registered(osd)) {
4123 		dout("%s osd%d unknown, skipping\n", __func__, osd->o_osd);
4124 		*skip = 1;
4125 		goto out_unlock_osdc;
4126 	}
4127 	WARN_ON(osd->o_osd != le64_to_cpu(hdr->src.num));
4128 
4129 	mutex_lock(&osd->lock);
4130 	req = lookup_request(&osd->o_requests, tid);
4131 	if (!req) {
4132 		dout("%s osd%d tid %llu unknown, skipping\n", __func__,
4133 		     osd->o_osd, tid);
4134 		*skip = 1;
4135 		goto out_unlock_session;
4136 	}
4137 
4138 	ceph_msg_revoke_incoming(req->r_reply);
4139 
4140 	if (front_len > req->r_reply->front_alloc_len) {
4141 		pr_warn("%s osd%d tid %llu front %d > preallocated %d\n",
4142 			__func__, osd->o_osd, req->r_tid, front_len,
4143 			req->r_reply->front_alloc_len);
4144 		m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS,
4145 				 false);
4146 		if (!m)
4147 			goto out_unlock_session;
4148 		ceph_msg_put(req->r_reply);
4149 		req->r_reply = m;
4150 	}
4151 
4152 	if (data_len > req->r_reply->data_length) {
4153 		pr_warn("%s osd%d tid %llu data %d > preallocated %zu, skipping\n",
4154 			__func__, osd->o_osd, req->r_tid, data_len,
4155 			req->r_reply->data_length);
4156 		m = NULL;
4157 		*skip = 1;
4158 		goto out_unlock_session;
4159 	}
4160 
4161 	m = ceph_msg_get(req->r_reply);
4162 	dout("get_reply tid %lld %p\n", tid, m);
4163 
4164 out_unlock_session:
4165 	mutex_unlock(&osd->lock);
4166 out_unlock_osdc:
4167 	up_read(&osdc->lock);
4168 	return m;
4169 }
4170 
4171 /*
4172  * TODO: switch to a msg-owned pagelist
4173  */
4174 static struct ceph_msg *alloc_msg_with_page_vector(struct ceph_msg_header *hdr)
4175 {
4176 	struct ceph_msg *m;
4177 	int type = le16_to_cpu(hdr->type);
4178 	u32 front_len = le32_to_cpu(hdr->front_len);
4179 	u32 data_len = le32_to_cpu(hdr->data_len);
4180 
4181 	m = ceph_msg_new(type, front_len, GFP_NOIO, false);
4182 	if (!m)
4183 		return NULL;
4184 
4185 	if (data_len) {
4186 		struct page **pages;
4187 		struct ceph_osd_data osd_data;
4188 
4189 		pages = ceph_alloc_page_vector(calc_pages_for(0, data_len),
4190 					       GFP_NOIO);
4191 		if (!pages) {
4192 			ceph_msg_put(m);
4193 			return NULL;
4194 		}
4195 
4196 		ceph_osd_data_pages_init(&osd_data, pages, data_len, 0, false,
4197 					 false);
4198 		ceph_osdc_msg_data_add(m, &osd_data);
4199 	}
4200 
4201 	return m;
4202 }
4203 
4204 static struct ceph_msg *alloc_msg(struct ceph_connection *con,
4205 				  struct ceph_msg_header *hdr,
4206 				  int *skip)
4207 {
4208 	struct ceph_osd *osd = con->private;
4209 	int type = le16_to_cpu(hdr->type);
4210 
4211 	*skip = 0;
4212 	switch (type) {
4213 	case CEPH_MSG_OSD_MAP:
4214 	case CEPH_MSG_WATCH_NOTIFY:
4215 		return alloc_msg_with_page_vector(hdr);
4216 	case CEPH_MSG_OSD_OPREPLY:
4217 		return get_reply(con, hdr, skip);
4218 	default:
4219 		pr_warn("%s osd%d unknown msg type %d, skipping\n", __func__,
4220 			osd->o_osd, type);
4221 		*skip = 1;
4222 		return NULL;
4223 	}
4224 }
4225 
4226 /*
4227  * Wrappers to refcount containing ceph_osd struct
4228  */
4229 static struct ceph_connection *get_osd_con(struct ceph_connection *con)
4230 {
4231 	struct ceph_osd *osd = con->private;
4232 	if (get_osd(osd))
4233 		return con;
4234 	return NULL;
4235 }
4236 
4237 static void put_osd_con(struct ceph_connection *con)
4238 {
4239 	struct ceph_osd *osd = con->private;
4240 	put_osd(osd);
4241 }
4242 
4243 /*
4244  * authentication
4245  */
4246 /*
4247  * Note: returned pointer is the address of a structure that's
4248  * managed separately.  Caller must *not* attempt to free it.
4249  */
4250 static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
4251 					int *proto, int force_new)
4252 {
4253 	struct ceph_osd *o = con->private;
4254 	struct ceph_osd_client *osdc = o->o_osdc;
4255 	struct ceph_auth_client *ac = osdc->client->monc.auth;
4256 	struct ceph_auth_handshake *auth = &o->o_auth;
4257 
4258 	if (force_new && auth->authorizer) {
4259 		ceph_auth_destroy_authorizer(auth->authorizer);
4260 		auth->authorizer = NULL;
4261 	}
4262 	if (!auth->authorizer) {
4263 		int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
4264 						      auth);
4265 		if (ret)
4266 			return ERR_PTR(ret);
4267 	} else {
4268 		int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
4269 						     auth);
4270 		if (ret)
4271 			return ERR_PTR(ret);
4272 	}
4273 	*proto = ac->protocol;
4274 
4275 	return auth;
4276 }
4277 
4278 
4279 static int verify_authorizer_reply(struct ceph_connection *con, int len)
4280 {
4281 	struct ceph_osd *o = con->private;
4282 	struct ceph_osd_client *osdc = o->o_osdc;
4283 	struct ceph_auth_client *ac = osdc->client->monc.auth;
4284 
4285 	return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer, len);
4286 }
4287 
4288 static int invalidate_authorizer(struct ceph_connection *con)
4289 {
4290 	struct ceph_osd *o = con->private;
4291 	struct ceph_osd_client *osdc = o->o_osdc;
4292 	struct ceph_auth_client *ac = osdc->client->monc.auth;
4293 
4294 	ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
4295 	return ceph_monc_validate_auth(&osdc->client->monc);
4296 }
4297 
4298 static int osd_sign_message(struct ceph_msg *msg)
4299 {
4300 	struct ceph_osd *o = msg->con->private;
4301 	struct ceph_auth_handshake *auth = &o->o_auth;
4302 
4303 	return ceph_auth_sign_message(auth, msg);
4304 }
4305 
4306 static int osd_check_message_signature(struct ceph_msg *msg)
4307 {
4308 	struct ceph_osd *o = msg->con->private;
4309 	struct ceph_auth_handshake *auth = &o->o_auth;
4310 
4311 	return ceph_auth_check_message_signature(auth, msg);
4312 }
4313 
4314 static const struct ceph_connection_operations osd_con_ops = {
4315 	.get = get_osd_con,
4316 	.put = put_osd_con,
4317 	.dispatch = dispatch,
4318 	.get_authorizer = get_authorizer,
4319 	.verify_authorizer_reply = verify_authorizer_reply,
4320 	.invalidate_authorizer = invalidate_authorizer,
4321 	.alloc_msg = alloc_msg,
4322 	.sign_message = osd_sign_message,
4323 	.check_message_signature = osd_check_message_signature,
4324 	.fault = osd_fault,
4325 };
4326