xref: /linux/net/ceph/osd_client.c (revision 66a0e2d579dbec5c676cfe446234ffebb267c564)
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 	case CEPH_OSD_OP_LIST_WATCHERS:
342 		ceph_osd_data_release(&op->list_watchers.response_data);
343 		break;
344 	default:
345 		break;
346 	}
347 }
348 
349 /*
350  * Assumes @t is zero-initialized.
351  */
352 static void target_init(struct ceph_osd_request_target *t)
353 {
354 	ceph_oid_init(&t->base_oid);
355 	ceph_oloc_init(&t->base_oloc);
356 	ceph_oid_init(&t->target_oid);
357 	ceph_oloc_init(&t->target_oloc);
358 
359 	ceph_osds_init(&t->acting);
360 	ceph_osds_init(&t->up);
361 	t->size = -1;
362 	t->min_size = -1;
363 
364 	t->osd = CEPH_HOMELESS_OSD;
365 }
366 
367 static void target_copy(struct ceph_osd_request_target *dest,
368 			const struct ceph_osd_request_target *src)
369 {
370 	ceph_oid_copy(&dest->base_oid, &src->base_oid);
371 	ceph_oloc_copy(&dest->base_oloc, &src->base_oloc);
372 	ceph_oid_copy(&dest->target_oid, &src->target_oid);
373 	ceph_oloc_copy(&dest->target_oloc, &src->target_oloc);
374 
375 	dest->pgid = src->pgid; /* struct */
376 	dest->pg_num = src->pg_num;
377 	dest->pg_num_mask = src->pg_num_mask;
378 	ceph_osds_copy(&dest->acting, &src->acting);
379 	ceph_osds_copy(&dest->up, &src->up);
380 	dest->size = src->size;
381 	dest->min_size = src->min_size;
382 	dest->sort_bitwise = src->sort_bitwise;
383 
384 	dest->flags = src->flags;
385 	dest->paused = src->paused;
386 
387 	dest->osd = src->osd;
388 }
389 
390 static void target_destroy(struct ceph_osd_request_target *t)
391 {
392 	ceph_oid_destroy(&t->base_oid);
393 	ceph_oloc_destroy(&t->base_oloc);
394 	ceph_oid_destroy(&t->target_oid);
395 	ceph_oloc_destroy(&t->target_oloc);
396 }
397 
398 /*
399  * requests
400  */
401 static void request_release_checks(struct ceph_osd_request *req)
402 {
403 	WARN_ON(!RB_EMPTY_NODE(&req->r_node));
404 	WARN_ON(!RB_EMPTY_NODE(&req->r_mc_node));
405 	WARN_ON(!list_empty(&req->r_unsafe_item));
406 	WARN_ON(req->r_osd);
407 }
408 
409 static void ceph_osdc_release_request(struct kref *kref)
410 {
411 	struct ceph_osd_request *req = container_of(kref,
412 					    struct ceph_osd_request, r_kref);
413 	unsigned int which;
414 
415 	dout("%s %p (r_request %p r_reply %p)\n", __func__, req,
416 	     req->r_request, req->r_reply);
417 	request_release_checks(req);
418 
419 	if (req->r_request)
420 		ceph_msg_put(req->r_request);
421 	if (req->r_reply)
422 		ceph_msg_put(req->r_reply);
423 
424 	for (which = 0; which < req->r_num_ops; which++)
425 		osd_req_op_data_release(req, which);
426 
427 	target_destroy(&req->r_t);
428 	ceph_put_snap_context(req->r_snapc);
429 
430 	if (req->r_mempool)
431 		mempool_free(req, req->r_osdc->req_mempool);
432 	else if (req->r_num_ops <= CEPH_OSD_SLAB_OPS)
433 		kmem_cache_free(ceph_osd_request_cache, req);
434 	else
435 		kfree(req);
436 }
437 
438 void ceph_osdc_get_request(struct ceph_osd_request *req)
439 {
440 	dout("%s %p (was %d)\n", __func__, req,
441 	     atomic_read(&req->r_kref.refcount));
442 	kref_get(&req->r_kref);
443 }
444 EXPORT_SYMBOL(ceph_osdc_get_request);
445 
446 void ceph_osdc_put_request(struct ceph_osd_request *req)
447 {
448 	if (req) {
449 		dout("%s %p (was %d)\n", __func__, req,
450 		     atomic_read(&req->r_kref.refcount));
451 		kref_put(&req->r_kref, ceph_osdc_release_request);
452 	}
453 }
454 EXPORT_SYMBOL(ceph_osdc_put_request);
455 
456 static void request_init(struct ceph_osd_request *req)
457 {
458 	/* req only, each op is zeroed in _osd_req_op_init() */
459 	memset(req, 0, sizeof(*req));
460 
461 	kref_init(&req->r_kref);
462 	init_completion(&req->r_completion);
463 	init_completion(&req->r_done_completion);
464 	RB_CLEAR_NODE(&req->r_node);
465 	RB_CLEAR_NODE(&req->r_mc_node);
466 	INIT_LIST_HEAD(&req->r_unsafe_item);
467 
468 	target_init(&req->r_t);
469 }
470 
471 /*
472  * This is ugly, but it allows us to reuse linger registration and ping
473  * requests, keeping the structure of the code around send_linger{_ping}()
474  * reasonable.  Setting up a min_nr=2 mempool for each linger request
475  * and dealing with copying ops (this blasts req only, watch op remains
476  * intact) isn't any better.
477  */
478 static void request_reinit(struct ceph_osd_request *req)
479 {
480 	struct ceph_osd_client *osdc = req->r_osdc;
481 	bool mempool = req->r_mempool;
482 	unsigned int num_ops = req->r_num_ops;
483 	u64 snapid = req->r_snapid;
484 	struct ceph_snap_context *snapc = req->r_snapc;
485 	bool linger = req->r_linger;
486 	struct ceph_msg *request_msg = req->r_request;
487 	struct ceph_msg *reply_msg = req->r_reply;
488 
489 	dout("%s req %p\n", __func__, req);
490 	WARN_ON(atomic_read(&req->r_kref.refcount) != 1);
491 	request_release_checks(req);
492 
493 	WARN_ON(atomic_read(&request_msg->kref.refcount) != 1);
494 	WARN_ON(atomic_read(&reply_msg->kref.refcount) != 1);
495 	target_destroy(&req->r_t);
496 
497 	request_init(req);
498 	req->r_osdc = osdc;
499 	req->r_mempool = mempool;
500 	req->r_num_ops = num_ops;
501 	req->r_snapid = snapid;
502 	req->r_snapc = snapc;
503 	req->r_linger = linger;
504 	req->r_request = request_msg;
505 	req->r_reply = reply_msg;
506 }
507 
508 struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
509 					       struct ceph_snap_context *snapc,
510 					       unsigned int num_ops,
511 					       bool use_mempool,
512 					       gfp_t gfp_flags)
513 {
514 	struct ceph_osd_request *req;
515 
516 	if (use_mempool) {
517 		BUG_ON(num_ops > CEPH_OSD_SLAB_OPS);
518 		req = mempool_alloc(osdc->req_mempool, gfp_flags);
519 	} else if (num_ops <= CEPH_OSD_SLAB_OPS) {
520 		req = kmem_cache_alloc(ceph_osd_request_cache, gfp_flags);
521 	} else {
522 		BUG_ON(num_ops > CEPH_OSD_MAX_OPS);
523 		req = kmalloc(sizeof(*req) + num_ops * sizeof(req->r_ops[0]),
524 			      gfp_flags);
525 	}
526 	if (unlikely(!req))
527 		return NULL;
528 
529 	request_init(req);
530 	req->r_osdc = osdc;
531 	req->r_mempool = use_mempool;
532 	req->r_num_ops = num_ops;
533 	req->r_snapid = CEPH_NOSNAP;
534 	req->r_snapc = ceph_get_snap_context(snapc);
535 
536 	dout("%s req %p\n", __func__, req);
537 	return req;
538 }
539 EXPORT_SYMBOL(ceph_osdc_alloc_request);
540 
541 static int ceph_oloc_encoding_size(struct ceph_object_locator *oloc)
542 {
543 	return 8 + 4 + 4 + 4 + (oloc->pool_ns ? oloc->pool_ns->len : 0);
544 }
545 
546 int ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp)
547 {
548 	struct ceph_osd_client *osdc = req->r_osdc;
549 	struct ceph_msg *msg;
550 	int msg_size;
551 
552 	WARN_ON(ceph_oid_empty(&req->r_base_oid));
553 	WARN_ON(ceph_oloc_empty(&req->r_base_oloc));
554 
555 	/* create request message */
556 	msg_size = 4 + 4 + 4; /* client_inc, osdmap_epoch, flags */
557 	msg_size += 4 + 4 + 4 + 8; /* mtime, reassert_version */
558 	msg_size += CEPH_ENCODING_START_BLK_LEN +
559 			ceph_oloc_encoding_size(&req->r_base_oloc); /* oloc */
560 	msg_size += 1 + 8 + 4 + 4; /* pgid */
561 	msg_size += 4 + req->r_base_oid.name_len; /* oid */
562 	msg_size += 2 + req->r_num_ops * sizeof(struct ceph_osd_op);
563 	msg_size += 8; /* snapid */
564 	msg_size += 8; /* snap_seq */
565 	msg_size += 4 + 8 * (req->r_snapc ? req->r_snapc->num_snaps : 0);
566 	msg_size += 4; /* retry_attempt */
567 
568 	if (req->r_mempool)
569 		msg = ceph_msgpool_get(&osdc->msgpool_op, 0);
570 	else
571 		msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp, true);
572 	if (!msg)
573 		return -ENOMEM;
574 
575 	memset(msg->front.iov_base, 0, msg->front.iov_len);
576 	req->r_request = msg;
577 
578 	/* create reply message */
579 	msg_size = OSD_OPREPLY_FRONT_LEN;
580 	msg_size += req->r_base_oid.name_len;
581 	msg_size += req->r_num_ops * sizeof(struct ceph_osd_op);
582 
583 	if (req->r_mempool)
584 		msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0);
585 	else
586 		msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, msg_size, gfp, true);
587 	if (!msg)
588 		return -ENOMEM;
589 
590 	req->r_reply = msg;
591 
592 	return 0;
593 }
594 EXPORT_SYMBOL(ceph_osdc_alloc_messages);
595 
596 static bool osd_req_opcode_valid(u16 opcode)
597 {
598 	switch (opcode) {
599 #define GENERATE_CASE(op, opcode, str)	case CEPH_OSD_OP_##op: return true;
600 __CEPH_FORALL_OSD_OPS(GENERATE_CASE)
601 #undef GENERATE_CASE
602 	default:
603 		return false;
604 	}
605 }
606 
607 /*
608  * This is an osd op init function for opcodes that have no data or
609  * other information associated with them.  It also serves as a
610  * common init routine for all the other init functions, below.
611  */
612 static struct ceph_osd_req_op *
613 _osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which,
614 		 u16 opcode, u32 flags)
615 {
616 	struct ceph_osd_req_op *op;
617 
618 	BUG_ON(which >= osd_req->r_num_ops);
619 	BUG_ON(!osd_req_opcode_valid(opcode));
620 
621 	op = &osd_req->r_ops[which];
622 	memset(op, 0, sizeof (*op));
623 	op->op = opcode;
624 	op->flags = flags;
625 
626 	return op;
627 }
628 
629 void osd_req_op_init(struct ceph_osd_request *osd_req,
630 		     unsigned int which, u16 opcode, u32 flags)
631 {
632 	(void)_osd_req_op_init(osd_req, which, opcode, flags);
633 }
634 EXPORT_SYMBOL(osd_req_op_init);
635 
636 void osd_req_op_extent_init(struct ceph_osd_request *osd_req,
637 				unsigned int which, u16 opcode,
638 				u64 offset, u64 length,
639 				u64 truncate_size, u32 truncate_seq)
640 {
641 	struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
642 						      opcode, 0);
643 	size_t payload_len = 0;
644 
645 	BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
646 	       opcode != CEPH_OSD_OP_WRITEFULL && opcode != CEPH_OSD_OP_ZERO &&
647 	       opcode != CEPH_OSD_OP_TRUNCATE);
648 
649 	op->extent.offset = offset;
650 	op->extent.length = length;
651 	op->extent.truncate_size = truncate_size;
652 	op->extent.truncate_seq = truncate_seq;
653 	if (opcode == CEPH_OSD_OP_WRITE || opcode == CEPH_OSD_OP_WRITEFULL)
654 		payload_len += length;
655 
656 	op->indata_len = payload_len;
657 }
658 EXPORT_SYMBOL(osd_req_op_extent_init);
659 
660 void osd_req_op_extent_update(struct ceph_osd_request *osd_req,
661 				unsigned int which, u64 length)
662 {
663 	struct ceph_osd_req_op *op;
664 	u64 previous;
665 
666 	BUG_ON(which >= osd_req->r_num_ops);
667 	op = &osd_req->r_ops[which];
668 	previous = op->extent.length;
669 
670 	if (length == previous)
671 		return;		/* Nothing to do */
672 	BUG_ON(length > previous);
673 
674 	op->extent.length = length;
675 	if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL)
676 		op->indata_len -= previous - length;
677 }
678 EXPORT_SYMBOL(osd_req_op_extent_update);
679 
680 void osd_req_op_extent_dup_last(struct ceph_osd_request *osd_req,
681 				unsigned int which, u64 offset_inc)
682 {
683 	struct ceph_osd_req_op *op, *prev_op;
684 
685 	BUG_ON(which + 1 >= osd_req->r_num_ops);
686 
687 	prev_op = &osd_req->r_ops[which];
688 	op = _osd_req_op_init(osd_req, which + 1, prev_op->op, prev_op->flags);
689 	/* dup previous one */
690 	op->indata_len = prev_op->indata_len;
691 	op->outdata_len = prev_op->outdata_len;
692 	op->extent = prev_op->extent;
693 	/* adjust offset */
694 	op->extent.offset += offset_inc;
695 	op->extent.length -= offset_inc;
696 
697 	if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL)
698 		op->indata_len -= offset_inc;
699 }
700 EXPORT_SYMBOL(osd_req_op_extent_dup_last);
701 
702 void osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
703 			u16 opcode, const char *class, const char *method)
704 {
705 	struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
706 						      opcode, 0);
707 	struct ceph_pagelist *pagelist;
708 	size_t payload_len = 0;
709 	size_t size;
710 
711 	BUG_ON(opcode != CEPH_OSD_OP_CALL);
712 
713 	pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
714 	BUG_ON(!pagelist);
715 	ceph_pagelist_init(pagelist);
716 
717 	op->cls.class_name = class;
718 	size = strlen(class);
719 	BUG_ON(size > (size_t) U8_MAX);
720 	op->cls.class_len = size;
721 	ceph_pagelist_append(pagelist, class, size);
722 	payload_len += size;
723 
724 	op->cls.method_name = method;
725 	size = strlen(method);
726 	BUG_ON(size > (size_t) U8_MAX);
727 	op->cls.method_len = size;
728 	ceph_pagelist_append(pagelist, method, size);
729 	payload_len += size;
730 
731 	osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist);
732 
733 	op->indata_len = payload_len;
734 }
735 EXPORT_SYMBOL(osd_req_op_cls_init);
736 
737 int osd_req_op_xattr_init(struct ceph_osd_request *osd_req, unsigned int which,
738 			  u16 opcode, const char *name, const void *value,
739 			  size_t size, u8 cmp_op, u8 cmp_mode)
740 {
741 	struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
742 						      opcode, 0);
743 	struct ceph_pagelist *pagelist;
744 	size_t payload_len;
745 
746 	BUG_ON(opcode != CEPH_OSD_OP_SETXATTR && opcode != CEPH_OSD_OP_CMPXATTR);
747 
748 	pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
749 	if (!pagelist)
750 		return -ENOMEM;
751 
752 	ceph_pagelist_init(pagelist);
753 
754 	payload_len = strlen(name);
755 	op->xattr.name_len = payload_len;
756 	ceph_pagelist_append(pagelist, name, payload_len);
757 
758 	op->xattr.value_len = size;
759 	ceph_pagelist_append(pagelist, value, size);
760 	payload_len += size;
761 
762 	op->xattr.cmp_op = cmp_op;
763 	op->xattr.cmp_mode = cmp_mode;
764 
765 	ceph_osd_data_pagelist_init(&op->xattr.osd_data, pagelist);
766 	op->indata_len = payload_len;
767 	return 0;
768 }
769 EXPORT_SYMBOL(osd_req_op_xattr_init);
770 
771 /*
772  * @watch_opcode: CEPH_OSD_WATCH_OP_*
773  */
774 static void osd_req_op_watch_init(struct ceph_osd_request *req, int which,
775 				  u64 cookie, u8 watch_opcode)
776 {
777 	struct ceph_osd_req_op *op;
778 
779 	op = _osd_req_op_init(req, which, CEPH_OSD_OP_WATCH, 0);
780 	op->watch.cookie = cookie;
781 	op->watch.op = watch_opcode;
782 	op->watch.gen = 0;
783 }
784 
785 void osd_req_op_alloc_hint_init(struct ceph_osd_request *osd_req,
786 				unsigned int which,
787 				u64 expected_object_size,
788 				u64 expected_write_size)
789 {
790 	struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
791 						      CEPH_OSD_OP_SETALLOCHINT,
792 						      0);
793 
794 	op->alloc_hint.expected_object_size = expected_object_size;
795 	op->alloc_hint.expected_write_size = expected_write_size;
796 
797 	/*
798 	 * CEPH_OSD_OP_SETALLOCHINT op is advisory and therefore deemed
799 	 * not worth a feature bit.  Set FAILOK per-op flag to make
800 	 * sure older osds don't trip over an unsupported opcode.
801 	 */
802 	op->flags |= CEPH_OSD_OP_FLAG_FAILOK;
803 }
804 EXPORT_SYMBOL(osd_req_op_alloc_hint_init);
805 
806 static void ceph_osdc_msg_data_add(struct ceph_msg *msg,
807 				struct ceph_osd_data *osd_data)
808 {
809 	u64 length = ceph_osd_data_length(osd_data);
810 
811 	if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
812 		BUG_ON(length > (u64) SIZE_MAX);
813 		if (length)
814 			ceph_msg_data_add_pages(msg, osd_data->pages,
815 					length, osd_data->alignment);
816 	} else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
817 		BUG_ON(!length);
818 		ceph_msg_data_add_pagelist(msg, osd_data->pagelist);
819 #ifdef CONFIG_BLOCK
820 	} else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
821 		ceph_msg_data_add_bio(msg, osd_data->bio, length);
822 #endif
823 	} else {
824 		BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
825 	}
826 }
827 
828 static u32 osd_req_encode_op(struct ceph_osd_op *dst,
829 			     const struct ceph_osd_req_op *src)
830 {
831 	if (WARN_ON(!osd_req_opcode_valid(src->op))) {
832 		pr_err("unrecognized osd opcode %d\n", src->op);
833 
834 		return 0;
835 	}
836 
837 	switch (src->op) {
838 	case CEPH_OSD_OP_STAT:
839 		break;
840 	case CEPH_OSD_OP_READ:
841 	case CEPH_OSD_OP_WRITE:
842 	case CEPH_OSD_OP_WRITEFULL:
843 	case CEPH_OSD_OP_ZERO:
844 	case CEPH_OSD_OP_TRUNCATE:
845 		dst->extent.offset = cpu_to_le64(src->extent.offset);
846 		dst->extent.length = cpu_to_le64(src->extent.length);
847 		dst->extent.truncate_size =
848 			cpu_to_le64(src->extent.truncate_size);
849 		dst->extent.truncate_seq =
850 			cpu_to_le32(src->extent.truncate_seq);
851 		break;
852 	case CEPH_OSD_OP_CALL:
853 		dst->cls.class_len = src->cls.class_len;
854 		dst->cls.method_len = src->cls.method_len;
855 		dst->cls.indata_len = cpu_to_le32(src->cls.indata_len);
856 		break;
857 	case CEPH_OSD_OP_STARTSYNC:
858 		break;
859 	case CEPH_OSD_OP_WATCH:
860 		dst->watch.cookie = cpu_to_le64(src->watch.cookie);
861 		dst->watch.ver = cpu_to_le64(0);
862 		dst->watch.op = src->watch.op;
863 		dst->watch.gen = cpu_to_le32(src->watch.gen);
864 		break;
865 	case CEPH_OSD_OP_NOTIFY_ACK:
866 		break;
867 	case CEPH_OSD_OP_NOTIFY:
868 		dst->notify.cookie = cpu_to_le64(src->notify.cookie);
869 		break;
870 	case CEPH_OSD_OP_LIST_WATCHERS:
871 		break;
872 	case CEPH_OSD_OP_SETALLOCHINT:
873 		dst->alloc_hint.expected_object_size =
874 		    cpu_to_le64(src->alloc_hint.expected_object_size);
875 		dst->alloc_hint.expected_write_size =
876 		    cpu_to_le64(src->alloc_hint.expected_write_size);
877 		break;
878 	case CEPH_OSD_OP_SETXATTR:
879 	case CEPH_OSD_OP_CMPXATTR:
880 		dst->xattr.name_len = cpu_to_le32(src->xattr.name_len);
881 		dst->xattr.value_len = cpu_to_le32(src->xattr.value_len);
882 		dst->xattr.cmp_op = src->xattr.cmp_op;
883 		dst->xattr.cmp_mode = src->xattr.cmp_mode;
884 		break;
885 	case CEPH_OSD_OP_CREATE:
886 	case CEPH_OSD_OP_DELETE:
887 		break;
888 	default:
889 		pr_err("unsupported osd opcode %s\n",
890 			ceph_osd_op_name(src->op));
891 		WARN_ON(1);
892 
893 		return 0;
894 	}
895 
896 	dst->op = cpu_to_le16(src->op);
897 	dst->flags = cpu_to_le32(src->flags);
898 	dst->payload_len = cpu_to_le32(src->indata_len);
899 
900 	return src->indata_len;
901 }
902 
903 /*
904  * build new request AND message, calculate layout, and adjust file
905  * extent as needed.
906  *
907  * if the file was recently truncated, we include information about its
908  * old and new size so that the object can be updated appropriately.  (we
909  * avoid synchronously deleting truncated objects because it's slow.)
910  *
911  * if @do_sync, include a 'startsync' command so that the osd will flush
912  * data quickly.
913  */
914 struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
915 					       struct ceph_file_layout *layout,
916 					       struct ceph_vino vino,
917 					       u64 off, u64 *plen,
918 					       unsigned int which, int num_ops,
919 					       int opcode, int flags,
920 					       struct ceph_snap_context *snapc,
921 					       u32 truncate_seq,
922 					       u64 truncate_size,
923 					       bool use_mempool)
924 {
925 	struct ceph_osd_request *req;
926 	u64 objnum = 0;
927 	u64 objoff = 0;
928 	u64 objlen = 0;
929 	int r;
930 
931 	BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
932 	       opcode != CEPH_OSD_OP_ZERO && opcode != CEPH_OSD_OP_TRUNCATE &&
933 	       opcode != CEPH_OSD_OP_CREATE && opcode != CEPH_OSD_OP_DELETE);
934 
935 	req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool,
936 					GFP_NOFS);
937 	if (!req) {
938 		r = -ENOMEM;
939 		goto fail;
940 	}
941 
942 	/* calculate max write size */
943 	r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
944 	if (r)
945 		goto fail;
946 
947 	if (opcode == CEPH_OSD_OP_CREATE || opcode == CEPH_OSD_OP_DELETE) {
948 		osd_req_op_init(req, which, opcode, 0);
949 	} else {
950 		u32 object_size = layout->object_size;
951 		u32 object_base = off - objoff;
952 		if (!(truncate_seq == 1 && truncate_size == -1ULL)) {
953 			if (truncate_size <= object_base) {
954 				truncate_size = 0;
955 			} else {
956 				truncate_size -= object_base;
957 				if (truncate_size > object_size)
958 					truncate_size = object_size;
959 			}
960 		}
961 		osd_req_op_extent_init(req, which, opcode, objoff, objlen,
962 				       truncate_size, truncate_seq);
963 	}
964 
965 	req->r_flags = flags;
966 	req->r_base_oloc.pool = layout->pool_id;
967 	req->r_base_oloc.pool_ns = ceph_try_get_string(layout->pool_ns);
968 	ceph_oid_printf(&req->r_base_oid, "%llx.%08llx", vino.ino, objnum);
969 
970 	req->r_snapid = vino.snap;
971 	if (flags & CEPH_OSD_FLAG_WRITE)
972 		req->r_data_offset = off;
973 
974 	r = ceph_osdc_alloc_messages(req, GFP_NOFS);
975 	if (r)
976 		goto fail;
977 
978 	return req;
979 
980 fail:
981 	ceph_osdc_put_request(req);
982 	return ERR_PTR(r);
983 }
984 EXPORT_SYMBOL(ceph_osdc_new_request);
985 
986 /*
987  * We keep osd requests in an rbtree, sorted by ->r_tid.
988  */
989 DEFINE_RB_FUNCS(request, struct ceph_osd_request, r_tid, r_node)
990 DEFINE_RB_FUNCS(request_mc, struct ceph_osd_request, r_tid, r_mc_node)
991 
992 static bool osd_homeless(struct ceph_osd *osd)
993 {
994 	return osd->o_osd == CEPH_HOMELESS_OSD;
995 }
996 
997 static bool osd_registered(struct ceph_osd *osd)
998 {
999 	verify_osdc_locked(osd->o_osdc);
1000 
1001 	return !RB_EMPTY_NODE(&osd->o_node);
1002 }
1003 
1004 /*
1005  * Assumes @osd is zero-initialized.
1006  */
1007 static void osd_init(struct ceph_osd *osd)
1008 {
1009 	atomic_set(&osd->o_ref, 1);
1010 	RB_CLEAR_NODE(&osd->o_node);
1011 	osd->o_requests = RB_ROOT;
1012 	osd->o_linger_requests = RB_ROOT;
1013 	INIT_LIST_HEAD(&osd->o_osd_lru);
1014 	INIT_LIST_HEAD(&osd->o_keepalive_item);
1015 	osd->o_incarnation = 1;
1016 	mutex_init(&osd->lock);
1017 }
1018 
1019 static void osd_cleanup(struct ceph_osd *osd)
1020 {
1021 	WARN_ON(!RB_EMPTY_NODE(&osd->o_node));
1022 	WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests));
1023 	WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests));
1024 	WARN_ON(!list_empty(&osd->o_osd_lru));
1025 	WARN_ON(!list_empty(&osd->o_keepalive_item));
1026 
1027 	if (osd->o_auth.authorizer) {
1028 		WARN_ON(osd_homeless(osd));
1029 		ceph_auth_destroy_authorizer(osd->o_auth.authorizer);
1030 	}
1031 }
1032 
1033 /*
1034  * Track open sessions with osds.
1035  */
1036 static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
1037 {
1038 	struct ceph_osd *osd;
1039 
1040 	WARN_ON(onum == CEPH_HOMELESS_OSD);
1041 
1042 	osd = kzalloc(sizeof(*osd), GFP_NOIO | __GFP_NOFAIL);
1043 	osd_init(osd);
1044 	osd->o_osdc = osdc;
1045 	osd->o_osd = onum;
1046 
1047 	ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
1048 
1049 	return osd;
1050 }
1051 
1052 static struct ceph_osd *get_osd(struct ceph_osd *osd)
1053 {
1054 	if (atomic_inc_not_zero(&osd->o_ref)) {
1055 		dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
1056 		     atomic_read(&osd->o_ref));
1057 		return osd;
1058 	} else {
1059 		dout("get_osd %p FAIL\n", osd);
1060 		return NULL;
1061 	}
1062 }
1063 
1064 static void put_osd(struct ceph_osd *osd)
1065 {
1066 	dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
1067 	     atomic_read(&osd->o_ref) - 1);
1068 	if (atomic_dec_and_test(&osd->o_ref)) {
1069 		osd_cleanup(osd);
1070 		kfree(osd);
1071 	}
1072 }
1073 
1074 DEFINE_RB_FUNCS(osd, struct ceph_osd, o_osd, o_node)
1075 
1076 static void __move_osd_to_lru(struct ceph_osd *osd)
1077 {
1078 	struct ceph_osd_client *osdc = osd->o_osdc;
1079 
1080 	dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1081 	BUG_ON(!list_empty(&osd->o_osd_lru));
1082 
1083 	spin_lock(&osdc->osd_lru_lock);
1084 	list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
1085 	spin_unlock(&osdc->osd_lru_lock);
1086 
1087 	osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl;
1088 }
1089 
1090 static void maybe_move_osd_to_lru(struct ceph_osd *osd)
1091 {
1092 	if (RB_EMPTY_ROOT(&osd->o_requests) &&
1093 	    RB_EMPTY_ROOT(&osd->o_linger_requests))
1094 		__move_osd_to_lru(osd);
1095 }
1096 
1097 static void __remove_osd_from_lru(struct ceph_osd *osd)
1098 {
1099 	struct ceph_osd_client *osdc = osd->o_osdc;
1100 
1101 	dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1102 
1103 	spin_lock(&osdc->osd_lru_lock);
1104 	if (!list_empty(&osd->o_osd_lru))
1105 		list_del_init(&osd->o_osd_lru);
1106 	spin_unlock(&osdc->osd_lru_lock);
1107 }
1108 
1109 /*
1110  * Close the connection and assign any leftover requests to the
1111  * homeless session.
1112  */
1113 static void close_osd(struct ceph_osd *osd)
1114 {
1115 	struct ceph_osd_client *osdc = osd->o_osdc;
1116 	struct rb_node *n;
1117 
1118 	verify_osdc_wrlocked(osdc);
1119 	dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1120 
1121 	ceph_con_close(&osd->o_con);
1122 
1123 	for (n = rb_first(&osd->o_requests); n; ) {
1124 		struct ceph_osd_request *req =
1125 		    rb_entry(n, struct ceph_osd_request, r_node);
1126 
1127 		n = rb_next(n); /* unlink_request() */
1128 
1129 		dout(" reassigning req %p tid %llu\n", req, req->r_tid);
1130 		unlink_request(osd, req);
1131 		link_request(&osdc->homeless_osd, req);
1132 	}
1133 	for (n = rb_first(&osd->o_linger_requests); n; ) {
1134 		struct ceph_osd_linger_request *lreq =
1135 		    rb_entry(n, struct ceph_osd_linger_request, node);
1136 
1137 		n = rb_next(n); /* unlink_linger() */
1138 
1139 		dout(" reassigning lreq %p linger_id %llu\n", lreq,
1140 		     lreq->linger_id);
1141 		unlink_linger(osd, lreq);
1142 		link_linger(&osdc->homeless_osd, lreq);
1143 	}
1144 
1145 	__remove_osd_from_lru(osd);
1146 	erase_osd(&osdc->osds, osd);
1147 	put_osd(osd);
1148 }
1149 
1150 /*
1151  * reset osd connect
1152  */
1153 static int reopen_osd(struct ceph_osd *osd)
1154 {
1155 	struct ceph_entity_addr *peer_addr;
1156 
1157 	dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1158 
1159 	if (RB_EMPTY_ROOT(&osd->o_requests) &&
1160 	    RB_EMPTY_ROOT(&osd->o_linger_requests)) {
1161 		close_osd(osd);
1162 		return -ENODEV;
1163 	}
1164 
1165 	peer_addr = &osd->o_osdc->osdmap->osd_addr[osd->o_osd];
1166 	if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1167 			!ceph_con_opened(&osd->o_con)) {
1168 		struct rb_node *n;
1169 
1170 		dout("osd addr hasn't changed and connection never opened, "
1171 		     "letting msgr retry\n");
1172 		/* touch each r_stamp for handle_timeout()'s benfit */
1173 		for (n = rb_first(&osd->o_requests); n; n = rb_next(n)) {
1174 			struct ceph_osd_request *req =
1175 			    rb_entry(n, struct ceph_osd_request, r_node);
1176 			req->r_stamp = jiffies;
1177 		}
1178 
1179 		return -EAGAIN;
1180 	}
1181 
1182 	ceph_con_close(&osd->o_con);
1183 	ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1184 	osd->o_incarnation++;
1185 
1186 	return 0;
1187 }
1188 
1189 static struct ceph_osd *lookup_create_osd(struct ceph_osd_client *osdc, int o,
1190 					  bool wrlocked)
1191 {
1192 	struct ceph_osd *osd;
1193 
1194 	if (wrlocked)
1195 		verify_osdc_wrlocked(osdc);
1196 	else
1197 		verify_osdc_locked(osdc);
1198 
1199 	if (o != CEPH_HOMELESS_OSD)
1200 		osd = lookup_osd(&osdc->osds, o);
1201 	else
1202 		osd = &osdc->homeless_osd;
1203 	if (!osd) {
1204 		if (!wrlocked)
1205 			return ERR_PTR(-EAGAIN);
1206 
1207 		osd = create_osd(osdc, o);
1208 		insert_osd(&osdc->osds, osd);
1209 		ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd,
1210 			      &osdc->osdmap->osd_addr[osd->o_osd]);
1211 	}
1212 
1213 	dout("%s osdc %p osd%d -> osd %p\n", __func__, osdc, o, osd);
1214 	return osd;
1215 }
1216 
1217 /*
1218  * Create request <-> OSD session relation.
1219  *
1220  * @req has to be assigned a tid, @osd may be homeless.
1221  */
1222 static void link_request(struct ceph_osd *osd, struct ceph_osd_request *req)
1223 {
1224 	verify_osd_locked(osd);
1225 	WARN_ON(!req->r_tid || req->r_osd);
1226 	dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd,
1227 	     req, req->r_tid);
1228 
1229 	if (!osd_homeless(osd))
1230 		__remove_osd_from_lru(osd);
1231 	else
1232 		atomic_inc(&osd->o_osdc->num_homeless);
1233 
1234 	get_osd(osd);
1235 	insert_request(&osd->o_requests, req);
1236 	req->r_osd = osd;
1237 }
1238 
1239 static void unlink_request(struct ceph_osd *osd, struct ceph_osd_request *req)
1240 {
1241 	verify_osd_locked(osd);
1242 	WARN_ON(req->r_osd != osd);
1243 	dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd,
1244 	     req, req->r_tid);
1245 
1246 	req->r_osd = NULL;
1247 	erase_request(&osd->o_requests, req);
1248 	put_osd(osd);
1249 
1250 	if (!osd_homeless(osd))
1251 		maybe_move_osd_to_lru(osd);
1252 	else
1253 		atomic_dec(&osd->o_osdc->num_homeless);
1254 }
1255 
1256 static bool __pool_full(struct ceph_pg_pool_info *pi)
1257 {
1258 	return pi->flags & CEPH_POOL_FLAG_FULL;
1259 }
1260 
1261 static bool have_pool_full(struct ceph_osd_client *osdc)
1262 {
1263 	struct rb_node *n;
1264 
1265 	for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) {
1266 		struct ceph_pg_pool_info *pi =
1267 		    rb_entry(n, struct ceph_pg_pool_info, node);
1268 
1269 		if (__pool_full(pi))
1270 			return true;
1271 	}
1272 
1273 	return false;
1274 }
1275 
1276 static bool pool_full(struct ceph_osd_client *osdc, s64 pool_id)
1277 {
1278 	struct ceph_pg_pool_info *pi;
1279 
1280 	pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id);
1281 	if (!pi)
1282 		return false;
1283 
1284 	return __pool_full(pi);
1285 }
1286 
1287 /*
1288  * Returns whether a request should be blocked from being sent
1289  * based on the current osdmap and osd_client settings.
1290  */
1291 static bool target_should_be_paused(struct ceph_osd_client *osdc,
1292 				    const struct ceph_osd_request_target *t,
1293 				    struct ceph_pg_pool_info *pi)
1294 {
1295 	bool pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
1296 	bool pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
1297 		       ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
1298 		       __pool_full(pi);
1299 
1300 	WARN_ON(pi->id != t->base_oloc.pool);
1301 	return (t->flags & CEPH_OSD_FLAG_READ && pauserd) ||
1302 	       (t->flags & CEPH_OSD_FLAG_WRITE && pausewr);
1303 }
1304 
1305 enum calc_target_result {
1306 	CALC_TARGET_NO_ACTION = 0,
1307 	CALC_TARGET_NEED_RESEND,
1308 	CALC_TARGET_POOL_DNE,
1309 };
1310 
1311 static enum calc_target_result calc_target(struct ceph_osd_client *osdc,
1312 					   struct ceph_osd_request_target *t,
1313 					   u32 *last_force_resend,
1314 					   bool any_change)
1315 {
1316 	struct ceph_pg_pool_info *pi;
1317 	struct ceph_pg pgid, last_pgid;
1318 	struct ceph_osds up, acting;
1319 	bool force_resend = false;
1320 	bool need_check_tiering = false;
1321 	bool need_resend = false;
1322 	bool sort_bitwise = ceph_osdmap_flag(osdc, CEPH_OSDMAP_SORTBITWISE);
1323 	enum calc_target_result ct_res;
1324 	int ret;
1325 
1326 	pi = ceph_pg_pool_by_id(osdc->osdmap, t->base_oloc.pool);
1327 	if (!pi) {
1328 		t->osd = CEPH_HOMELESS_OSD;
1329 		ct_res = CALC_TARGET_POOL_DNE;
1330 		goto out;
1331 	}
1332 
1333 	if (osdc->osdmap->epoch == pi->last_force_request_resend) {
1334 		if (last_force_resend &&
1335 		    *last_force_resend < pi->last_force_request_resend) {
1336 			*last_force_resend = pi->last_force_request_resend;
1337 			force_resend = true;
1338 		} else if (!last_force_resend) {
1339 			force_resend = true;
1340 		}
1341 	}
1342 	if (ceph_oid_empty(&t->target_oid) || force_resend) {
1343 		ceph_oid_copy(&t->target_oid, &t->base_oid);
1344 		need_check_tiering = true;
1345 	}
1346 	if (ceph_oloc_empty(&t->target_oloc) || force_resend) {
1347 		ceph_oloc_copy(&t->target_oloc, &t->base_oloc);
1348 		need_check_tiering = true;
1349 	}
1350 
1351 	if (need_check_tiering &&
1352 	    (t->flags & CEPH_OSD_FLAG_IGNORE_OVERLAY) == 0) {
1353 		if (t->flags & CEPH_OSD_FLAG_READ && pi->read_tier >= 0)
1354 			t->target_oloc.pool = pi->read_tier;
1355 		if (t->flags & CEPH_OSD_FLAG_WRITE && pi->write_tier >= 0)
1356 			t->target_oloc.pool = pi->write_tier;
1357 	}
1358 
1359 	ret = ceph_object_locator_to_pg(osdc->osdmap, &t->target_oid,
1360 					&t->target_oloc, &pgid);
1361 	if (ret) {
1362 		WARN_ON(ret != -ENOENT);
1363 		t->osd = CEPH_HOMELESS_OSD;
1364 		ct_res = CALC_TARGET_POOL_DNE;
1365 		goto out;
1366 	}
1367 	last_pgid.pool = pgid.pool;
1368 	last_pgid.seed = ceph_stable_mod(pgid.seed, t->pg_num, t->pg_num_mask);
1369 
1370 	ceph_pg_to_up_acting_osds(osdc->osdmap, &pgid, &up, &acting);
1371 	if (any_change &&
1372 	    ceph_is_new_interval(&t->acting,
1373 				 &acting,
1374 				 &t->up,
1375 				 &up,
1376 				 t->size,
1377 				 pi->size,
1378 				 t->min_size,
1379 				 pi->min_size,
1380 				 t->pg_num,
1381 				 pi->pg_num,
1382 				 t->sort_bitwise,
1383 				 sort_bitwise,
1384 				 &last_pgid))
1385 		force_resend = true;
1386 
1387 	if (t->paused && !target_should_be_paused(osdc, t, pi)) {
1388 		t->paused = false;
1389 		need_resend = true;
1390 	}
1391 
1392 	if (ceph_pg_compare(&t->pgid, &pgid) ||
1393 	    ceph_osds_changed(&t->acting, &acting, any_change) ||
1394 	    force_resend) {
1395 		t->pgid = pgid; /* struct */
1396 		ceph_osds_copy(&t->acting, &acting);
1397 		ceph_osds_copy(&t->up, &up);
1398 		t->size = pi->size;
1399 		t->min_size = pi->min_size;
1400 		t->pg_num = pi->pg_num;
1401 		t->pg_num_mask = pi->pg_num_mask;
1402 		t->sort_bitwise = sort_bitwise;
1403 
1404 		t->osd = acting.primary;
1405 		need_resend = true;
1406 	}
1407 
1408 	ct_res = need_resend ? CALC_TARGET_NEED_RESEND : CALC_TARGET_NO_ACTION;
1409 out:
1410 	dout("%s t %p -> ct_res %d osd %d\n", __func__, t, ct_res, t->osd);
1411 	return ct_res;
1412 }
1413 
1414 static void setup_request_data(struct ceph_osd_request *req,
1415 			       struct ceph_msg *msg)
1416 {
1417 	u32 data_len = 0;
1418 	int i;
1419 
1420 	if (!list_empty(&msg->data))
1421 		return;
1422 
1423 	WARN_ON(msg->data_length);
1424 	for (i = 0; i < req->r_num_ops; i++) {
1425 		struct ceph_osd_req_op *op = &req->r_ops[i];
1426 
1427 		switch (op->op) {
1428 		/* request */
1429 		case CEPH_OSD_OP_WRITE:
1430 		case CEPH_OSD_OP_WRITEFULL:
1431 			WARN_ON(op->indata_len != op->extent.length);
1432 			ceph_osdc_msg_data_add(msg, &op->extent.osd_data);
1433 			break;
1434 		case CEPH_OSD_OP_SETXATTR:
1435 		case CEPH_OSD_OP_CMPXATTR:
1436 			WARN_ON(op->indata_len != op->xattr.name_len +
1437 						  op->xattr.value_len);
1438 			ceph_osdc_msg_data_add(msg, &op->xattr.osd_data);
1439 			break;
1440 		case CEPH_OSD_OP_NOTIFY_ACK:
1441 			ceph_osdc_msg_data_add(msg,
1442 					       &op->notify_ack.request_data);
1443 			break;
1444 
1445 		/* reply */
1446 		case CEPH_OSD_OP_STAT:
1447 			ceph_osdc_msg_data_add(req->r_reply,
1448 					       &op->raw_data_in);
1449 			break;
1450 		case CEPH_OSD_OP_READ:
1451 			ceph_osdc_msg_data_add(req->r_reply,
1452 					       &op->extent.osd_data);
1453 			break;
1454 		case CEPH_OSD_OP_LIST_WATCHERS:
1455 			ceph_osdc_msg_data_add(req->r_reply,
1456 					       &op->list_watchers.response_data);
1457 			break;
1458 
1459 		/* both */
1460 		case CEPH_OSD_OP_CALL:
1461 			WARN_ON(op->indata_len != op->cls.class_len +
1462 						  op->cls.method_len +
1463 						  op->cls.indata_len);
1464 			ceph_osdc_msg_data_add(msg, &op->cls.request_info);
1465 			/* optional, can be NONE */
1466 			ceph_osdc_msg_data_add(msg, &op->cls.request_data);
1467 			/* optional, can be NONE */
1468 			ceph_osdc_msg_data_add(req->r_reply,
1469 					       &op->cls.response_data);
1470 			break;
1471 		case CEPH_OSD_OP_NOTIFY:
1472 			ceph_osdc_msg_data_add(msg,
1473 					       &op->notify.request_data);
1474 			ceph_osdc_msg_data_add(req->r_reply,
1475 					       &op->notify.response_data);
1476 			break;
1477 		}
1478 
1479 		data_len += op->indata_len;
1480 	}
1481 
1482 	WARN_ON(data_len != msg->data_length);
1483 }
1484 
1485 static void encode_request(struct ceph_osd_request *req, struct ceph_msg *msg)
1486 {
1487 	void *p = msg->front.iov_base;
1488 	void *const end = p + msg->front_alloc_len;
1489 	u32 data_len = 0;
1490 	int i;
1491 
1492 	if (req->r_flags & CEPH_OSD_FLAG_WRITE) {
1493 		/* snapshots aren't writeable */
1494 		WARN_ON(req->r_snapid != CEPH_NOSNAP);
1495 	} else {
1496 		WARN_ON(req->r_mtime.tv_sec || req->r_mtime.tv_nsec ||
1497 			req->r_data_offset || req->r_snapc);
1498 	}
1499 
1500 	setup_request_data(req, msg);
1501 
1502 	ceph_encode_32(&p, 1); /* client_inc, always 1 */
1503 	ceph_encode_32(&p, req->r_osdc->osdmap->epoch);
1504 	ceph_encode_32(&p, req->r_flags);
1505 	ceph_encode_timespec(p, &req->r_mtime);
1506 	p += sizeof(struct ceph_timespec);
1507 	/* aka reassert_version */
1508 	memcpy(p, &req->r_replay_version, sizeof(req->r_replay_version));
1509 	p += sizeof(req->r_replay_version);
1510 
1511 	/* oloc */
1512 	ceph_start_encoding(&p, 5, 4,
1513 			    ceph_oloc_encoding_size(&req->r_t.target_oloc));
1514 	ceph_encode_64(&p, req->r_t.target_oloc.pool);
1515 	ceph_encode_32(&p, -1); /* preferred */
1516 	ceph_encode_32(&p, 0); /* key len */
1517 	if (req->r_t.target_oloc.pool_ns)
1518 		ceph_encode_string(&p, end, req->r_t.target_oloc.pool_ns->str,
1519 				   req->r_t.target_oloc.pool_ns->len);
1520 	else
1521 		ceph_encode_32(&p, 0);
1522 
1523 	/* pgid */
1524 	ceph_encode_8(&p, 1);
1525 	ceph_encode_64(&p, req->r_t.pgid.pool);
1526 	ceph_encode_32(&p, req->r_t.pgid.seed);
1527 	ceph_encode_32(&p, -1); /* preferred */
1528 
1529 	/* oid */
1530 	ceph_encode_32(&p, req->r_t.target_oid.name_len);
1531 	memcpy(p, req->r_t.target_oid.name, req->r_t.target_oid.name_len);
1532 	p += req->r_t.target_oid.name_len;
1533 
1534 	/* ops, can imply data */
1535 	ceph_encode_16(&p, req->r_num_ops);
1536 	for (i = 0; i < req->r_num_ops; i++) {
1537 		data_len += osd_req_encode_op(p, &req->r_ops[i]);
1538 		p += sizeof(struct ceph_osd_op);
1539 	}
1540 
1541 	ceph_encode_64(&p, req->r_snapid); /* snapid */
1542 	if (req->r_snapc) {
1543 		ceph_encode_64(&p, req->r_snapc->seq);
1544 		ceph_encode_32(&p, req->r_snapc->num_snaps);
1545 		for (i = 0; i < req->r_snapc->num_snaps; i++)
1546 			ceph_encode_64(&p, req->r_snapc->snaps[i]);
1547 	} else {
1548 		ceph_encode_64(&p, 0); /* snap_seq */
1549 		ceph_encode_32(&p, 0); /* snaps len */
1550 	}
1551 
1552 	ceph_encode_32(&p, req->r_attempts); /* retry_attempt */
1553 
1554 	BUG_ON(p > end);
1555 	msg->front.iov_len = p - msg->front.iov_base;
1556 	msg->hdr.version = cpu_to_le16(4); /* MOSDOp v4 */
1557 	msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1558 	msg->hdr.data_len = cpu_to_le32(data_len);
1559 	/*
1560 	 * The header "data_off" is a hint to the receiver allowing it
1561 	 * to align received data into its buffers such that there's no
1562 	 * need to re-copy it before writing it to disk (direct I/O).
1563 	 */
1564 	msg->hdr.data_off = cpu_to_le16(req->r_data_offset);
1565 
1566 	dout("%s req %p oid %s oid_len %d front %zu data %u\n", __func__,
1567 	     req, req->r_t.target_oid.name, req->r_t.target_oid.name_len,
1568 	     msg->front.iov_len, data_len);
1569 }
1570 
1571 /*
1572  * @req has to be assigned a tid and registered.
1573  */
1574 static void send_request(struct ceph_osd_request *req)
1575 {
1576 	struct ceph_osd *osd = req->r_osd;
1577 
1578 	verify_osd_locked(osd);
1579 	WARN_ON(osd->o_osd != req->r_t.osd);
1580 
1581 	/*
1582 	 * We may have a previously queued request message hanging
1583 	 * around.  Cancel it to avoid corrupting the msgr.
1584 	 */
1585 	if (req->r_sent)
1586 		ceph_msg_revoke(req->r_request);
1587 
1588 	req->r_flags |= CEPH_OSD_FLAG_KNOWN_REDIR;
1589 	if (req->r_attempts)
1590 		req->r_flags |= CEPH_OSD_FLAG_RETRY;
1591 	else
1592 		WARN_ON(req->r_flags & CEPH_OSD_FLAG_RETRY);
1593 
1594 	encode_request(req, req->r_request);
1595 
1596 	dout("%s req %p tid %llu to pg %llu.%x osd%d flags 0x%x attempt %d\n",
1597 	     __func__, req, req->r_tid, req->r_t.pgid.pool, req->r_t.pgid.seed,
1598 	     req->r_t.osd, req->r_flags, req->r_attempts);
1599 
1600 	req->r_t.paused = false;
1601 	req->r_stamp = jiffies;
1602 	req->r_attempts++;
1603 
1604 	req->r_sent = osd->o_incarnation;
1605 	req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
1606 	ceph_con_send(&osd->o_con, ceph_msg_get(req->r_request));
1607 }
1608 
1609 static void maybe_request_map(struct ceph_osd_client *osdc)
1610 {
1611 	bool continuous = false;
1612 
1613 	verify_osdc_locked(osdc);
1614 	WARN_ON(!osdc->osdmap->epoch);
1615 
1616 	if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
1617 	    ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD) ||
1618 	    ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) {
1619 		dout("%s osdc %p continuous\n", __func__, osdc);
1620 		continuous = true;
1621 	} else {
1622 		dout("%s osdc %p onetime\n", __func__, osdc);
1623 	}
1624 
1625 	if (ceph_monc_want_map(&osdc->client->monc, CEPH_SUB_OSDMAP,
1626 			       osdc->osdmap->epoch + 1, continuous))
1627 		ceph_monc_renew_subs(&osdc->client->monc);
1628 }
1629 
1630 static void send_map_check(struct ceph_osd_request *req);
1631 
1632 static void __submit_request(struct ceph_osd_request *req, bool wrlocked)
1633 {
1634 	struct ceph_osd_client *osdc = req->r_osdc;
1635 	struct ceph_osd *osd;
1636 	enum calc_target_result ct_res;
1637 	bool need_send = false;
1638 	bool promoted = false;
1639 
1640 	WARN_ON(req->r_tid || req->r_got_reply);
1641 	dout("%s req %p wrlocked %d\n", __func__, req, wrlocked);
1642 
1643 again:
1644 	ct_res = calc_target(osdc, &req->r_t, &req->r_last_force_resend, false);
1645 	if (ct_res == CALC_TARGET_POOL_DNE && !wrlocked)
1646 		goto promote;
1647 
1648 	osd = lookup_create_osd(osdc, req->r_t.osd, wrlocked);
1649 	if (IS_ERR(osd)) {
1650 		WARN_ON(PTR_ERR(osd) != -EAGAIN || wrlocked);
1651 		goto promote;
1652 	}
1653 
1654 	if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
1655 	    ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) {
1656 		dout("req %p pausewr\n", req);
1657 		req->r_t.paused = true;
1658 		maybe_request_map(osdc);
1659 	} else if ((req->r_flags & CEPH_OSD_FLAG_READ) &&
1660 		   ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) {
1661 		dout("req %p pauserd\n", req);
1662 		req->r_t.paused = true;
1663 		maybe_request_map(osdc);
1664 	} else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
1665 		   !(req->r_flags & (CEPH_OSD_FLAG_FULL_TRY |
1666 				     CEPH_OSD_FLAG_FULL_FORCE)) &&
1667 		   (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
1668 		    pool_full(osdc, req->r_t.base_oloc.pool))) {
1669 		dout("req %p full/pool_full\n", req);
1670 		pr_warn_ratelimited("FULL or reached pool quota\n");
1671 		req->r_t.paused = true;
1672 		maybe_request_map(osdc);
1673 	} else if (!osd_homeless(osd)) {
1674 		need_send = true;
1675 	} else {
1676 		maybe_request_map(osdc);
1677 	}
1678 
1679 	mutex_lock(&osd->lock);
1680 	/*
1681 	 * Assign the tid atomically with send_request() to protect
1682 	 * multiple writes to the same object from racing with each
1683 	 * other, resulting in out of order ops on the OSDs.
1684 	 */
1685 	req->r_tid = atomic64_inc_return(&osdc->last_tid);
1686 	link_request(osd, req);
1687 	if (need_send)
1688 		send_request(req);
1689 	mutex_unlock(&osd->lock);
1690 
1691 	if (ct_res == CALC_TARGET_POOL_DNE)
1692 		send_map_check(req);
1693 
1694 	if (promoted)
1695 		downgrade_write(&osdc->lock);
1696 	return;
1697 
1698 promote:
1699 	up_read(&osdc->lock);
1700 	down_write(&osdc->lock);
1701 	wrlocked = true;
1702 	promoted = true;
1703 	goto again;
1704 }
1705 
1706 static void account_request(struct ceph_osd_request *req)
1707 {
1708 	unsigned int mask = CEPH_OSD_FLAG_ACK | CEPH_OSD_FLAG_ONDISK;
1709 
1710 	if (req->r_flags & CEPH_OSD_FLAG_READ) {
1711 		WARN_ON(req->r_flags & mask);
1712 		req->r_flags |= CEPH_OSD_FLAG_ACK;
1713 	} else if (req->r_flags & CEPH_OSD_FLAG_WRITE)
1714 		WARN_ON(!(req->r_flags & mask));
1715 	else
1716 		WARN_ON(1);
1717 
1718 	WARN_ON(req->r_unsafe_callback && (req->r_flags & mask) != mask);
1719 	atomic_inc(&req->r_osdc->num_requests);
1720 }
1721 
1722 static void submit_request(struct ceph_osd_request *req, bool wrlocked)
1723 {
1724 	ceph_osdc_get_request(req);
1725 	account_request(req);
1726 	__submit_request(req, wrlocked);
1727 }
1728 
1729 static void finish_request(struct ceph_osd_request *req)
1730 {
1731 	struct ceph_osd_client *osdc = req->r_osdc;
1732 	struct ceph_osd *osd = req->r_osd;
1733 
1734 	verify_osd_locked(osd);
1735 	dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
1736 
1737 	WARN_ON(lookup_request_mc(&osdc->map_checks, req->r_tid));
1738 	unlink_request(osd, req);
1739 	atomic_dec(&osdc->num_requests);
1740 
1741 	/*
1742 	 * If an OSD has failed or returned and a request has been sent
1743 	 * twice, it's possible to get a reply and end up here while the
1744 	 * request message is queued for delivery.  We will ignore the
1745 	 * reply, so not a big deal, but better to try and catch it.
1746 	 */
1747 	ceph_msg_revoke(req->r_request);
1748 	ceph_msg_revoke_incoming(req->r_reply);
1749 }
1750 
1751 static void __complete_request(struct ceph_osd_request *req)
1752 {
1753 	if (req->r_callback)
1754 		req->r_callback(req);
1755 	else
1756 		complete_all(&req->r_completion);
1757 }
1758 
1759 /*
1760  * Note that this is open-coded in handle_reply(), which has to deal
1761  * with ack vs commit, dup acks, etc.
1762  */
1763 static void complete_request(struct ceph_osd_request *req, int err)
1764 {
1765 	dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err);
1766 
1767 	req->r_result = err;
1768 	finish_request(req);
1769 	__complete_request(req);
1770 	complete_all(&req->r_done_completion);
1771 	ceph_osdc_put_request(req);
1772 }
1773 
1774 static void cancel_map_check(struct ceph_osd_request *req)
1775 {
1776 	struct ceph_osd_client *osdc = req->r_osdc;
1777 	struct ceph_osd_request *lookup_req;
1778 
1779 	verify_osdc_wrlocked(osdc);
1780 
1781 	lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid);
1782 	if (!lookup_req)
1783 		return;
1784 
1785 	WARN_ON(lookup_req != req);
1786 	erase_request_mc(&osdc->map_checks, req);
1787 	ceph_osdc_put_request(req);
1788 }
1789 
1790 static void cancel_request(struct ceph_osd_request *req)
1791 {
1792 	dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
1793 
1794 	cancel_map_check(req);
1795 	finish_request(req);
1796 	complete_all(&req->r_done_completion);
1797 	ceph_osdc_put_request(req);
1798 }
1799 
1800 static void check_pool_dne(struct ceph_osd_request *req)
1801 {
1802 	struct ceph_osd_client *osdc = req->r_osdc;
1803 	struct ceph_osdmap *map = osdc->osdmap;
1804 
1805 	verify_osdc_wrlocked(osdc);
1806 	WARN_ON(!map->epoch);
1807 
1808 	if (req->r_attempts) {
1809 		/*
1810 		 * We sent a request earlier, which means that
1811 		 * previously the pool existed, and now it does not
1812 		 * (i.e., it was deleted).
1813 		 */
1814 		req->r_map_dne_bound = map->epoch;
1815 		dout("%s req %p tid %llu pool disappeared\n", __func__, req,
1816 		     req->r_tid);
1817 	} else {
1818 		dout("%s req %p tid %llu map_dne_bound %u have %u\n", __func__,
1819 		     req, req->r_tid, req->r_map_dne_bound, map->epoch);
1820 	}
1821 
1822 	if (req->r_map_dne_bound) {
1823 		if (map->epoch >= req->r_map_dne_bound) {
1824 			/* we had a new enough map */
1825 			pr_info_ratelimited("tid %llu pool does not exist\n",
1826 					    req->r_tid);
1827 			complete_request(req, -ENOENT);
1828 		}
1829 	} else {
1830 		send_map_check(req);
1831 	}
1832 }
1833 
1834 static void map_check_cb(struct ceph_mon_generic_request *greq)
1835 {
1836 	struct ceph_osd_client *osdc = &greq->monc->client->osdc;
1837 	struct ceph_osd_request *req;
1838 	u64 tid = greq->private_data;
1839 
1840 	WARN_ON(greq->result || !greq->u.newest);
1841 
1842 	down_write(&osdc->lock);
1843 	req = lookup_request_mc(&osdc->map_checks, tid);
1844 	if (!req) {
1845 		dout("%s tid %llu dne\n", __func__, tid);
1846 		goto out_unlock;
1847 	}
1848 
1849 	dout("%s req %p tid %llu map_dne_bound %u newest %llu\n", __func__,
1850 	     req, req->r_tid, req->r_map_dne_bound, greq->u.newest);
1851 	if (!req->r_map_dne_bound)
1852 		req->r_map_dne_bound = greq->u.newest;
1853 	erase_request_mc(&osdc->map_checks, req);
1854 	check_pool_dne(req);
1855 
1856 	ceph_osdc_put_request(req);
1857 out_unlock:
1858 	up_write(&osdc->lock);
1859 }
1860 
1861 static void send_map_check(struct ceph_osd_request *req)
1862 {
1863 	struct ceph_osd_client *osdc = req->r_osdc;
1864 	struct ceph_osd_request *lookup_req;
1865 	int ret;
1866 
1867 	verify_osdc_wrlocked(osdc);
1868 
1869 	lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid);
1870 	if (lookup_req) {
1871 		WARN_ON(lookup_req != req);
1872 		return;
1873 	}
1874 
1875 	ceph_osdc_get_request(req);
1876 	insert_request_mc(&osdc->map_checks, req);
1877 	ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap",
1878 					  map_check_cb, req->r_tid);
1879 	WARN_ON(ret);
1880 }
1881 
1882 /*
1883  * lingering requests, watch/notify v2 infrastructure
1884  */
1885 static void linger_release(struct kref *kref)
1886 {
1887 	struct ceph_osd_linger_request *lreq =
1888 	    container_of(kref, struct ceph_osd_linger_request, kref);
1889 
1890 	dout("%s lreq %p reg_req %p ping_req %p\n", __func__, lreq,
1891 	     lreq->reg_req, lreq->ping_req);
1892 	WARN_ON(!RB_EMPTY_NODE(&lreq->node));
1893 	WARN_ON(!RB_EMPTY_NODE(&lreq->osdc_node));
1894 	WARN_ON(!RB_EMPTY_NODE(&lreq->mc_node));
1895 	WARN_ON(!list_empty(&lreq->scan_item));
1896 	WARN_ON(!list_empty(&lreq->pending_lworks));
1897 	WARN_ON(lreq->osd);
1898 
1899 	if (lreq->reg_req)
1900 		ceph_osdc_put_request(lreq->reg_req);
1901 	if (lreq->ping_req)
1902 		ceph_osdc_put_request(lreq->ping_req);
1903 	target_destroy(&lreq->t);
1904 	kfree(lreq);
1905 }
1906 
1907 static void linger_put(struct ceph_osd_linger_request *lreq)
1908 {
1909 	if (lreq)
1910 		kref_put(&lreq->kref, linger_release);
1911 }
1912 
1913 static struct ceph_osd_linger_request *
1914 linger_get(struct ceph_osd_linger_request *lreq)
1915 {
1916 	kref_get(&lreq->kref);
1917 	return lreq;
1918 }
1919 
1920 static struct ceph_osd_linger_request *
1921 linger_alloc(struct ceph_osd_client *osdc)
1922 {
1923 	struct ceph_osd_linger_request *lreq;
1924 
1925 	lreq = kzalloc(sizeof(*lreq), GFP_NOIO);
1926 	if (!lreq)
1927 		return NULL;
1928 
1929 	kref_init(&lreq->kref);
1930 	mutex_init(&lreq->lock);
1931 	RB_CLEAR_NODE(&lreq->node);
1932 	RB_CLEAR_NODE(&lreq->osdc_node);
1933 	RB_CLEAR_NODE(&lreq->mc_node);
1934 	INIT_LIST_HEAD(&lreq->scan_item);
1935 	INIT_LIST_HEAD(&lreq->pending_lworks);
1936 	init_completion(&lreq->reg_commit_wait);
1937 	init_completion(&lreq->notify_finish_wait);
1938 
1939 	lreq->osdc = osdc;
1940 	target_init(&lreq->t);
1941 
1942 	dout("%s lreq %p\n", __func__, lreq);
1943 	return lreq;
1944 }
1945 
1946 DEFINE_RB_INSDEL_FUNCS(linger, struct ceph_osd_linger_request, linger_id, node)
1947 DEFINE_RB_FUNCS(linger_osdc, struct ceph_osd_linger_request, linger_id, osdc_node)
1948 DEFINE_RB_FUNCS(linger_mc, struct ceph_osd_linger_request, linger_id, mc_node)
1949 
1950 /*
1951  * Create linger request <-> OSD session relation.
1952  *
1953  * @lreq has to be registered, @osd may be homeless.
1954  */
1955 static void link_linger(struct ceph_osd *osd,
1956 			struct ceph_osd_linger_request *lreq)
1957 {
1958 	verify_osd_locked(osd);
1959 	WARN_ON(!lreq->linger_id || lreq->osd);
1960 	dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd,
1961 	     osd->o_osd, lreq, lreq->linger_id);
1962 
1963 	if (!osd_homeless(osd))
1964 		__remove_osd_from_lru(osd);
1965 	else
1966 		atomic_inc(&osd->o_osdc->num_homeless);
1967 
1968 	get_osd(osd);
1969 	insert_linger(&osd->o_linger_requests, lreq);
1970 	lreq->osd = osd;
1971 }
1972 
1973 static void unlink_linger(struct ceph_osd *osd,
1974 			  struct ceph_osd_linger_request *lreq)
1975 {
1976 	verify_osd_locked(osd);
1977 	WARN_ON(lreq->osd != osd);
1978 	dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd,
1979 	     osd->o_osd, lreq, lreq->linger_id);
1980 
1981 	lreq->osd = NULL;
1982 	erase_linger(&osd->o_linger_requests, lreq);
1983 	put_osd(osd);
1984 
1985 	if (!osd_homeless(osd))
1986 		maybe_move_osd_to_lru(osd);
1987 	else
1988 		atomic_dec(&osd->o_osdc->num_homeless);
1989 }
1990 
1991 static bool __linger_registered(struct ceph_osd_linger_request *lreq)
1992 {
1993 	verify_osdc_locked(lreq->osdc);
1994 
1995 	return !RB_EMPTY_NODE(&lreq->osdc_node);
1996 }
1997 
1998 static bool linger_registered(struct ceph_osd_linger_request *lreq)
1999 {
2000 	struct ceph_osd_client *osdc = lreq->osdc;
2001 	bool registered;
2002 
2003 	down_read(&osdc->lock);
2004 	registered = __linger_registered(lreq);
2005 	up_read(&osdc->lock);
2006 
2007 	return registered;
2008 }
2009 
2010 static void linger_register(struct ceph_osd_linger_request *lreq)
2011 {
2012 	struct ceph_osd_client *osdc = lreq->osdc;
2013 
2014 	verify_osdc_wrlocked(osdc);
2015 	WARN_ON(lreq->linger_id);
2016 
2017 	linger_get(lreq);
2018 	lreq->linger_id = ++osdc->last_linger_id;
2019 	insert_linger_osdc(&osdc->linger_requests, lreq);
2020 }
2021 
2022 static void linger_unregister(struct ceph_osd_linger_request *lreq)
2023 {
2024 	struct ceph_osd_client *osdc = lreq->osdc;
2025 
2026 	verify_osdc_wrlocked(osdc);
2027 
2028 	erase_linger_osdc(&osdc->linger_requests, lreq);
2029 	linger_put(lreq);
2030 }
2031 
2032 static void cancel_linger_request(struct ceph_osd_request *req)
2033 {
2034 	struct ceph_osd_linger_request *lreq = req->r_priv;
2035 
2036 	WARN_ON(!req->r_linger);
2037 	cancel_request(req);
2038 	linger_put(lreq);
2039 }
2040 
2041 struct linger_work {
2042 	struct work_struct work;
2043 	struct ceph_osd_linger_request *lreq;
2044 	struct list_head pending_item;
2045 	unsigned long queued_stamp;
2046 
2047 	union {
2048 		struct {
2049 			u64 notify_id;
2050 			u64 notifier_id;
2051 			void *payload; /* points into @msg front */
2052 			size_t payload_len;
2053 
2054 			struct ceph_msg *msg; /* for ceph_msg_put() */
2055 		} notify;
2056 		struct {
2057 			int err;
2058 		} error;
2059 	};
2060 };
2061 
2062 static struct linger_work *lwork_alloc(struct ceph_osd_linger_request *lreq,
2063 				       work_func_t workfn)
2064 {
2065 	struct linger_work *lwork;
2066 
2067 	lwork = kzalloc(sizeof(*lwork), GFP_NOIO);
2068 	if (!lwork)
2069 		return NULL;
2070 
2071 	INIT_WORK(&lwork->work, workfn);
2072 	INIT_LIST_HEAD(&lwork->pending_item);
2073 	lwork->lreq = linger_get(lreq);
2074 
2075 	return lwork;
2076 }
2077 
2078 static void lwork_free(struct linger_work *lwork)
2079 {
2080 	struct ceph_osd_linger_request *lreq = lwork->lreq;
2081 
2082 	mutex_lock(&lreq->lock);
2083 	list_del(&lwork->pending_item);
2084 	mutex_unlock(&lreq->lock);
2085 
2086 	linger_put(lreq);
2087 	kfree(lwork);
2088 }
2089 
2090 static void lwork_queue(struct linger_work *lwork)
2091 {
2092 	struct ceph_osd_linger_request *lreq = lwork->lreq;
2093 	struct ceph_osd_client *osdc = lreq->osdc;
2094 
2095 	verify_lreq_locked(lreq);
2096 	WARN_ON(!list_empty(&lwork->pending_item));
2097 
2098 	lwork->queued_stamp = jiffies;
2099 	list_add_tail(&lwork->pending_item, &lreq->pending_lworks);
2100 	queue_work(osdc->notify_wq, &lwork->work);
2101 }
2102 
2103 static void do_watch_notify(struct work_struct *w)
2104 {
2105 	struct linger_work *lwork = container_of(w, struct linger_work, work);
2106 	struct ceph_osd_linger_request *lreq = lwork->lreq;
2107 
2108 	if (!linger_registered(lreq)) {
2109 		dout("%s lreq %p not registered\n", __func__, lreq);
2110 		goto out;
2111 	}
2112 
2113 	WARN_ON(!lreq->is_watch);
2114 	dout("%s lreq %p notify_id %llu notifier_id %llu payload_len %zu\n",
2115 	     __func__, lreq, lwork->notify.notify_id, lwork->notify.notifier_id,
2116 	     lwork->notify.payload_len);
2117 	lreq->wcb(lreq->data, lwork->notify.notify_id, lreq->linger_id,
2118 		  lwork->notify.notifier_id, lwork->notify.payload,
2119 		  lwork->notify.payload_len);
2120 
2121 out:
2122 	ceph_msg_put(lwork->notify.msg);
2123 	lwork_free(lwork);
2124 }
2125 
2126 static void do_watch_error(struct work_struct *w)
2127 {
2128 	struct linger_work *lwork = container_of(w, struct linger_work, work);
2129 	struct ceph_osd_linger_request *lreq = lwork->lreq;
2130 
2131 	if (!linger_registered(lreq)) {
2132 		dout("%s lreq %p not registered\n", __func__, lreq);
2133 		goto out;
2134 	}
2135 
2136 	dout("%s lreq %p err %d\n", __func__, lreq, lwork->error.err);
2137 	lreq->errcb(lreq->data, lreq->linger_id, lwork->error.err);
2138 
2139 out:
2140 	lwork_free(lwork);
2141 }
2142 
2143 static void queue_watch_error(struct ceph_osd_linger_request *lreq)
2144 {
2145 	struct linger_work *lwork;
2146 
2147 	lwork = lwork_alloc(lreq, do_watch_error);
2148 	if (!lwork) {
2149 		pr_err("failed to allocate error-lwork\n");
2150 		return;
2151 	}
2152 
2153 	lwork->error.err = lreq->last_error;
2154 	lwork_queue(lwork);
2155 }
2156 
2157 static void linger_reg_commit_complete(struct ceph_osd_linger_request *lreq,
2158 				       int result)
2159 {
2160 	if (!completion_done(&lreq->reg_commit_wait)) {
2161 		lreq->reg_commit_error = (result <= 0 ? result : 0);
2162 		complete_all(&lreq->reg_commit_wait);
2163 	}
2164 }
2165 
2166 static void linger_commit_cb(struct ceph_osd_request *req)
2167 {
2168 	struct ceph_osd_linger_request *lreq = req->r_priv;
2169 
2170 	mutex_lock(&lreq->lock);
2171 	dout("%s lreq %p linger_id %llu result %d\n", __func__, lreq,
2172 	     lreq->linger_id, req->r_result);
2173 	WARN_ON(!__linger_registered(lreq));
2174 	linger_reg_commit_complete(lreq, req->r_result);
2175 	lreq->committed = true;
2176 
2177 	if (!lreq->is_watch) {
2178 		struct ceph_osd_data *osd_data =
2179 		    osd_req_op_data(req, 0, notify, response_data);
2180 		void *p = page_address(osd_data->pages[0]);
2181 
2182 		WARN_ON(req->r_ops[0].op != CEPH_OSD_OP_NOTIFY ||
2183 			osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
2184 
2185 		/* make note of the notify_id */
2186 		if (req->r_ops[0].outdata_len >= sizeof(u64)) {
2187 			lreq->notify_id = ceph_decode_64(&p);
2188 			dout("lreq %p notify_id %llu\n", lreq,
2189 			     lreq->notify_id);
2190 		} else {
2191 			dout("lreq %p no notify_id\n", lreq);
2192 		}
2193 	}
2194 
2195 	mutex_unlock(&lreq->lock);
2196 	linger_put(lreq);
2197 }
2198 
2199 static int normalize_watch_error(int err)
2200 {
2201 	/*
2202 	 * Translate ENOENT -> ENOTCONN so that a delete->disconnection
2203 	 * notification and a failure to reconnect because we raced with
2204 	 * the delete appear the same to the user.
2205 	 */
2206 	if (err == -ENOENT)
2207 		err = -ENOTCONN;
2208 
2209 	return err;
2210 }
2211 
2212 static void linger_reconnect_cb(struct ceph_osd_request *req)
2213 {
2214 	struct ceph_osd_linger_request *lreq = req->r_priv;
2215 
2216 	mutex_lock(&lreq->lock);
2217 	dout("%s lreq %p linger_id %llu result %d last_error %d\n", __func__,
2218 	     lreq, lreq->linger_id, req->r_result, lreq->last_error);
2219 	if (req->r_result < 0) {
2220 		if (!lreq->last_error) {
2221 			lreq->last_error = normalize_watch_error(req->r_result);
2222 			queue_watch_error(lreq);
2223 		}
2224 	}
2225 
2226 	mutex_unlock(&lreq->lock);
2227 	linger_put(lreq);
2228 }
2229 
2230 static void send_linger(struct ceph_osd_linger_request *lreq)
2231 {
2232 	struct ceph_osd_request *req = lreq->reg_req;
2233 	struct ceph_osd_req_op *op = &req->r_ops[0];
2234 
2235 	verify_osdc_wrlocked(req->r_osdc);
2236 	dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
2237 
2238 	if (req->r_osd)
2239 		cancel_linger_request(req);
2240 
2241 	request_reinit(req);
2242 	ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
2243 	ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
2244 	req->r_flags = lreq->t.flags;
2245 	req->r_mtime = lreq->mtime;
2246 
2247 	mutex_lock(&lreq->lock);
2248 	if (lreq->is_watch && lreq->committed) {
2249 		WARN_ON(op->op != CEPH_OSD_OP_WATCH ||
2250 			op->watch.cookie != lreq->linger_id);
2251 		op->watch.op = CEPH_OSD_WATCH_OP_RECONNECT;
2252 		op->watch.gen = ++lreq->register_gen;
2253 		dout("lreq %p reconnect register_gen %u\n", lreq,
2254 		     op->watch.gen);
2255 		req->r_callback = linger_reconnect_cb;
2256 	} else {
2257 		if (!lreq->is_watch)
2258 			lreq->notify_id = 0;
2259 		else
2260 			WARN_ON(op->watch.op != CEPH_OSD_WATCH_OP_WATCH);
2261 		dout("lreq %p register\n", lreq);
2262 		req->r_callback = linger_commit_cb;
2263 	}
2264 	mutex_unlock(&lreq->lock);
2265 
2266 	req->r_priv = linger_get(lreq);
2267 	req->r_linger = true;
2268 
2269 	submit_request(req, true);
2270 }
2271 
2272 static void linger_ping_cb(struct ceph_osd_request *req)
2273 {
2274 	struct ceph_osd_linger_request *lreq = req->r_priv;
2275 
2276 	mutex_lock(&lreq->lock);
2277 	dout("%s lreq %p linger_id %llu result %d ping_sent %lu last_error %d\n",
2278 	     __func__, lreq, lreq->linger_id, req->r_result, lreq->ping_sent,
2279 	     lreq->last_error);
2280 	if (lreq->register_gen == req->r_ops[0].watch.gen) {
2281 		if (!req->r_result) {
2282 			lreq->watch_valid_thru = lreq->ping_sent;
2283 		} else if (!lreq->last_error) {
2284 			lreq->last_error = normalize_watch_error(req->r_result);
2285 			queue_watch_error(lreq);
2286 		}
2287 	} else {
2288 		dout("lreq %p register_gen %u ignoring old pong %u\n", lreq,
2289 		     lreq->register_gen, req->r_ops[0].watch.gen);
2290 	}
2291 
2292 	mutex_unlock(&lreq->lock);
2293 	linger_put(lreq);
2294 }
2295 
2296 static void send_linger_ping(struct ceph_osd_linger_request *lreq)
2297 {
2298 	struct ceph_osd_client *osdc = lreq->osdc;
2299 	struct ceph_osd_request *req = lreq->ping_req;
2300 	struct ceph_osd_req_op *op = &req->r_ops[0];
2301 
2302 	if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) {
2303 		dout("%s PAUSERD\n", __func__);
2304 		return;
2305 	}
2306 
2307 	lreq->ping_sent = jiffies;
2308 	dout("%s lreq %p linger_id %llu ping_sent %lu register_gen %u\n",
2309 	     __func__, lreq, lreq->linger_id, lreq->ping_sent,
2310 	     lreq->register_gen);
2311 
2312 	if (req->r_osd)
2313 		cancel_linger_request(req);
2314 
2315 	request_reinit(req);
2316 	target_copy(&req->r_t, &lreq->t);
2317 
2318 	WARN_ON(op->op != CEPH_OSD_OP_WATCH ||
2319 		op->watch.cookie != lreq->linger_id ||
2320 		op->watch.op != CEPH_OSD_WATCH_OP_PING);
2321 	op->watch.gen = lreq->register_gen;
2322 	req->r_callback = linger_ping_cb;
2323 	req->r_priv = linger_get(lreq);
2324 	req->r_linger = true;
2325 
2326 	ceph_osdc_get_request(req);
2327 	account_request(req);
2328 	req->r_tid = atomic64_inc_return(&osdc->last_tid);
2329 	link_request(lreq->osd, req);
2330 	send_request(req);
2331 }
2332 
2333 static void linger_submit(struct ceph_osd_linger_request *lreq)
2334 {
2335 	struct ceph_osd_client *osdc = lreq->osdc;
2336 	struct ceph_osd *osd;
2337 
2338 	calc_target(osdc, &lreq->t, &lreq->last_force_resend, false);
2339 	osd = lookup_create_osd(osdc, lreq->t.osd, true);
2340 	link_linger(osd, lreq);
2341 
2342 	send_linger(lreq);
2343 }
2344 
2345 static void cancel_linger_map_check(struct ceph_osd_linger_request *lreq)
2346 {
2347 	struct ceph_osd_client *osdc = lreq->osdc;
2348 	struct ceph_osd_linger_request *lookup_lreq;
2349 
2350 	verify_osdc_wrlocked(osdc);
2351 
2352 	lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks,
2353 				       lreq->linger_id);
2354 	if (!lookup_lreq)
2355 		return;
2356 
2357 	WARN_ON(lookup_lreq != lreq);
2358 	erase_linger_mc(&osdc->linger_map_checks, lreq);
2359 	linger_put(lreq);
2360 }
2361 
2362 /*
2363  * @lreq has to be both registered and linked.
2364  */
2365 static void __linger_cancel(struct ceph_osd_linger_request *lreq)
2366 {
2367 	if (lreq->is_watch && lreq->ping_req->r_osd)
2368 		cancel_linger_request(lreq->ping_req);
2369 	if (lreq->reg_req->r_osd)
2370 		cancel_linger_request(lreq->reg_req);
2371 	cancel_linger_map_check(lreq);
2372 	unlink_linger(lreq->osd, lreq);
2373 	linger_unregister(lreq);
2374 }
2375 
2376 static void linger_cancel(struct ceph_osd_linger_request *lreq)
2377 {
2378 	struct ceph_osd_client *osdc = lreq->osdc;
2379 
2380 	down_write(&osdc->lock);
2381 	if (__linger_registered(lreq))
2382 		__linger_cancel(lreq);
2383 	up_write(&osdc->lock);
2384 }
2385 
2386 static void send_linger_map_check(struct ceph_osd_linger_request *lreq);
2387 
2388 static void check_linger_pool_dne(struct ceph_osd_linger_request *lreq)
2389 {
2390 	struct ceph_osd_client *osdc = lreq->osdc;
2391 	struct ceph_osdmap *map = osdc->osdmap;
2392 
2393 	verify_osdc_wrlocked(osdc);
2394 	WARN_ON(!map->epoch);
2395 
2396 	if (lreq->register_gen) {
2397 		lreq->map_dne_bound = map->epoch;
2398 		dout("%s lreq %p linger_id %llu pool disappeared\n", __func__,
2399 		     lreq, lreq->linger_id);
2400 	} else {
2401 		dout("%s lreq %p linger_id %llu map_dne_bound %u have %u\n",
2402 		     __func__, lreq, lreq->linger_id, lreq->map_dne_bound,
2403 		     map->epoch);
2404 	}
2405 
2406 	if (lreq->map_dne_bound) {
2407 		if (map->epoch >= lreq->map_dne_bound) {
2408 			/* we had a new enough map */
2409 			pr_info("linger_id %llu pool does not exist\n",
2410 				lreq->linger_id);
2411 			linger_reg_commit_complete(lreq, -ENOENT);
2412 			__linger_cancel(lreq);
2413 		}
2414 	} else {
2415 		send_linger_map_check(lreq);
2416 	}
2417 }
2418 
2419 static void linger_map_check_cb(struct ceph_mon_generic_request *greq)
2420 {
2421 	struct ceph_osd_client *osdc = &greq->monc->client->osdc;
2422 	struct ceph_osd_linger_request *lreq;
2423 	u64 linger_id = greq->private_data;
2424 
2425 	WARN_ON(greq->result || !greq->u.newest);
2426 
2427 	down_write(&osdc->lock);
2428 	lreq = lookup_linger_mc(&osdc->linger_map_checks, linger_id);
2429 	if (!lreq) {
2430 		dout("%s linger_id %llu dne\n", __func__, linger_id);
2431 		goto out_unlock;
2432 	}
2433 
2434 	dout("%s lreq %p linger_id %llu map_dne_bound %u newest %llu\n",
2435 	     __func__, lreq, lreq->linger_id, lreq->map_dne_bound,
2436 	     greq->u.newest);
2437 	if (!lreq->map_dne_bound)
2438 		lreq->map_dne_bound = greq->u.newest;
2439 	erase_linger_mc(&osdc->linger_map_checks, lreq);
2440 	check_linger_pool_dne(lreq);
2441 
2442 	linger_put(lreq);
2443 out_unlock:
2444 	up_write(&osdc->lock);
2445 }
2446 
2447 static void send_linger_map_check(struct ceph_osd_linger_request *lreq)
2448 {
2449 	struct ceph_osd_client *osdc = lreq->osdc;
2450 	struct ceph_osd_linger_request *lookup_lreq;
2451 	int ret;
2452 
2453 	verify_osdc_wrlocked(osdc);
2454 
2455 	lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks,
2456 				       lreq->linger_id);
2457 	if (lookup_lreq) {
2458 		WARN_ON(lookup_lreq != lreq);
2459 		return;
2460 	}
2461 
2462 	linger_get(lreq);
2463 	insert_linger_mc(&osdc->linger_map_checks, lreq);
2464 	ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap",
2465 					  linger_map_check_cb, lreq->linger_id);
2466 	WARN_ON(ret);
2467 }
2468 
2469 static int linger_reg_commit_wait(struct ceph_osd_linger_request *lreq)
2470 {
2471 	int ret;
2472 
2473 	dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
2474 	ret = wait_for_completion_interruptible(&lreq->reg_commit_wait);
2475 	return ret ?: lreq->reg_commit_error;
2476 }
2477 
2478 static int linger_notify_finish_wait(struct ceph_osd_linger_request *lreq)
2479 {
2480 	int ret;
2481 
2482 	dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
2483 	ret = wait_for_completion_interruptible(&lreq->notify_finish_wait);
2484 	return ret ?: lreq->notify_finish_error;
2485 }
2486 
2487 /*
2488  * Timeout callback, called every N seconds.  When 1 or more OSD
2489  * requests has been active for more than N seconds, we send a keepalive
2490  * (tag + timestamp) to its OSD to ensure any communications channel
2491  * reset is detected.
2492  */
2493 static void handle_timeout(struct work_struct *work)
2494 {
2495 	struct ceph_osd_client *osdc =
2496 		container_of(work, struct ceph_osd_client, timeout_work.work);
2497 	struct ceph_options *opts = osdc->client->options;
2498 	unsigned long cutoff = jiffies - opts->osd_keepalive_timeout;
2499 	LIST_HEAD(slow_osds);
2500 	struct rb_node *n, *p;
2501 
2502 	dout("%s osdc %p\n", __func__, osdc);
2503 	down_write(&osdc->lock);
2504 
2505 	/*
2506 	 * ping osds that are a bit slow.  this ensures that if there
2507 	 * is a break in the TCP connection we will notice, and reopen
2508 	 * a connection with that osd (from the fault callback).
2509 	 */
2510 	for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
2511 		struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
2512 		bool found = false;
2513 
2514 		for (p = rb_first(&osd->o_requests); p; p = rb_next(p)) {
2515 			struct ceph_osd_request *req =
2516 			    rb_entry(p, struct ceph_osd_request, r_node);
2517 
2518 			if (time_before(req->r_stamp, cutoff)) {
2519 				dout(" req %p tid %llu on osd%d is laggy\n",
2520 				     req, req->r_tid, osd->o_osd);
2521 				found = true;
2522 			}
2523 		}
2524 		for (p = rb_first(&osd->o_linger_requests); p; p = rb_next(p)) {
2525 			struct ceph_osd_linger_request *lreq =
2526 			    rb_entry(p, struct ceph_osd_linger_request, node);
2527 
2528 			dout(" lreq %p linger_id %llu is served by osd%d\n",
2529 			     lreq, lreq->linger_id, osd->o_osd);
2530 			found = true;
2531 
2532 			mutex_lock(&lreq->lock);
2533 			if (lreq->is_watch && lreq->committed && !lreq->last_error)
2534 				send_linger_ping(lreq);
2535 			mutex_unlock(&lreq->lock);
2536 		}
2537 
2538 		if (found)
2539 			list_move_tail(&osd->o_keepalive_item, &slow_osds);
2540 	}
2541 
2542 	if (atomic_read(&osdc->num_homeless) || !list_empty(&slow_osds))
2543 		maybe_request_map(osdc);
2544 
2545 	while (!list_empty(&slow_osds)) {
2546 		struct ceph_osd *osd = list_first_entry(&slow_osds,
2547 							struct ceph_osd,
2548 							o_keepalive_item);
2549 		list_del_init(&osd->o_keepalive_item);
2550 		ceph_con_keepalive(&osd->o_con);
2551 	}
2552 
2553 	up_write(&osdc->lock);
2554 	schedule_delayed_work(&osdc->timeout_work,
2555 			      osdc->client->options->osd_keepalive_timeout);
2556 }
2557 
2558 static void handle_osds_timeout(struct work_struct *work)
2559 {
2560 	struct ceph_osd_client *osdc =
2561 		container_of(work, struct ceph_osd_client,
2562 			     osds_timeout_work.work);
2563 	unsigned long delay = osdc->client->options->osd_idle_ttl / 4;
2564 	struct ceph_osd *osd, *nosd;
2565 
2566 	dout("%s osdc %p\n", __func__, osdc);
2567 	down_write(&osdc->lock);
2568 	list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
2569 		if (time_before(jiffies, osd->lru_ttl))
2570 			break;
2571 
2572 		WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests));
2573 		WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests));
2574 		close_osd(osd);
2575 	}
2576 
2577 	up_write(&osdc->lock);
2578 	schedule_delayed_work(&osdc->osds_timeout_work,
2579 			      round_jiffies_relative(delay));
2580 }
2581 
2582 static int ceph_oloc_decode(void **p, void *end,
2583 			    struct ceph_object_locator *oloc)
2584 {
2585 	u8 struct_v, struct_cv;
2586 	u32 len;
2587 	void *struct_end;
2588 	int ret = 0;
2589 
2590 	ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
2591 	struct_v = ceph_decode_8(p);
2592 	struct_cv = ceph_decode_8(p);
2593 	if (struct_v < 3) {
2594 		pr_warn("got v %d < 3 cv %d of ceph_object_locator\n",
2595 			struct_v, struct_cv);
2596 		goto e_inval;
2597 	}
2598 	if (struct_cv > 6) {
2599 		pr_warn("got v %d cv %d > 6 of ceph_object_locator\n",
2600 			struct_v, struct_cv);
2601 		goto e_inval;
2602 	}
2603 	len = ceph_decode_32(p);
2604 	ceph_decode_need(p, end, len, e_inval);
2605 	struct_end = *p + len;
2606 
2607 	oloc->pool = ceph_decode_64(p);
2608 	*p += 4; /* skip preferred */
2609 
2610 	len = ceph_decode_32(p);
2611 	if (len > 0) {
2612 		pr_warn("ceph_object_locator::key is set\n");
2613 		goto e_inval;
2614 	}
2615 
2616 	if (struct_v >= 5) {
2617 		bool changed = false;
2618 
2619 		len = ceph_decode_32(p);
2620 		if (len > 0) {
2621 			ceph_decode_need(p, end, len, e_inval);
2622 			if (!oloc->pool_ns ||
2623 			    ceph_compare_string(oloc->pool_ns, *p, len))
2624 				changed = true;
2625 			*p += len;
2626 		} else {
2627 			if (oloc->pool_ns)
2628 				changed = true;
2629 		}
2630 		if (changed) {
2631 			/* redirect changes namespace */
2632 			pr_warn("ceph_object_locator::nspace is changed\n");
2633 			goto e_inval;
2634 		}
2635 	}
2636 
2637 	if (struct_v >= 6) {
2638 		s64 hash = ceph_decode_64(p);
2639 		if (hash != -1) {
2640 			pr_warn("ceph_object_locator::hash is set\n");
2641 			goto e_inval;
2642 		}
2643 	}
2644 
2645 	/* skip the rest */
2646 	*p = struct_end;
2647 out:
2648 	return ret;
2649 
2650 e_inval:
2651 	ret = -EINVAL;
2652 	goto out;
2653 }
2654 
2655 static int ceph_redirect_decode(void **p, void *end,
2656 				struct ceph_request_redirect *redir)
2657 {
2658 	u8 struct_v, struct_cv;
2659 	u32 len;
2660 	void *struct_end;
2661 	int ret;
2662 
2663 	ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
2664 	struct_v = ceph_decode_8(p);
2665 	struct_cv = ceph_decode_8(p);
2666 	if (struct_cv > 1) {
2667 		pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n",
2668 			struct_v, struct_cv);
2669 		goto e_inval;
2670 	}
2671 	len = ceph_decode_32(p);
2672 	ceph_decode_need(p, end, len, e_inval);
2673 	struct_end = *p + len;
2674 
2675 	ret = ceph_oloc_decode(p, end, &redir->oloc);
2676 	if (ret)
2677 		goto out;
2678 
2679 	len = ceph_decode_32(p);
2680 	if (len > 0) {
2681 		pr_warn("ceph_request_redirect::object_name is set\n");
2682 		goto e_inval;
2683 	}
2684 
2685 	len = ceph_decode_32(p);
2686 	*p += len; /* skip osd_instructions */
2687 
2688 	/* skip the rest */
2689 	*p = struct_end;
2690 out:
2691 	return ret;
2692 
2693 e_inval:
2694 	ret = -EINVAL;
2695 	goto out;
2696 }
2697 
2698 struct MOSDOpReply {
2699 	struct ceph_pg pgid;
2700 	u64 flags;
2701 	int result;
2702 	u32 epoch;
2703 	int num_ops;
2704 	u32 outdata_len[CEPH_OSD_MAX_OPS];
2705 	s32 rval[CEPH_OSD_MAX_OPS];
2706 	int retry_attempt;
2707 	struct ceph_eversion replay_version;
2708 	u64 user_version;
2709 	struct ceph_request_redirect redirect;
2710 };
2711 
2712 static int decode_MOSDOpReply(const struct ceph_msg *msg, struct MOSDOpReply *m)
2713 {
2714 	void *p = msg->front.iov_base;
2715 	void *const end = p + msg->front.iov_len;
2716 	u16 version = le16_to_cpu(msg->hdr.version);
2717 	struct ceph_eversion bad_replay_version;
2718 	u8 decode_redir;
2719 	u32 len;
2720 	int ret;
2721 	int i;
2722 
2723 	ceph_decode_32_safe(&p, end, len, e_inval);
2724 	ceph_decode_need(&p, end, len, e_inval);
2725 	p += len; /* skip oid */
2726 
2727 	ret = ceph_decode_pgid(&p, end, &m->pgid);
2728 	if (ret)
2729 		return ret;
2730 
2731 	ceph_decode_64_safe(&p, end, m->flags, e_inval);
2732 	ceph_decode_32_safe(&p, end, m->result, e_inval);
2733 	ceph_decode_need(&p, end, sizeof(bad_replay_version), e_inval);
2734 	memcpy(&bad_replay_version, p, sizeof(bad_replay_version));
2735 	p += sizeof(bad_replay_version);
2736 	ceph_decode_32_safe(&p, end, m->epoch, e_inval);
2737 
2738 	ceph_decode_32_safe(&p, end, m->num_ops, e_inval);
2739 	if (m->num_ops > ARRAY_SIZE(m->outdata_len))
2740 		goto e_inval;
2741 
2742 	ceph_decode_need(&p, end, m->num_ops * sizeof(struct ceph_osd_op),
2743 			 e_inval);
2744 	for (i = 0; i < m->num_ops; i++) {
2745 		struct ceph_osd_op *op = p;
2746 
2747 		m->outdata_len[i] = le32_to_cpu(op->payload_len);
2748 		p += sizeof(*op);
2749 	}
2750 
2751 	ceph_decode_32_safe(&p, end, m->retry_attempt, e_inval);
2752 	for (i = 0; i < m->num_ops; i++)
2753 		ceph_decode_32_safe(&p, end, m->rval[i], e_inval);
2754 
2755 	if (version >= 5) {
2756 		ceph_decode_need(&p, end, sizeof(m->replay_version), e_inval);
2757 		memcpy(&m->replay_version, p, sizeof(m->replay_version));
2758 		p += sizeof(m->replay_version);
2759 		ceph_decode_64_safe(&p, end, m->user_version, e_inval);
2760 	} else {
2761 		m->replay_version = bad_replay_version; /* struct */
2762 		m->user_version = le64_to_cpu(m->replay_version.version);
2763 	}
2764 
2765 	if (version >= 6) {
2766 		if (version >= 7)
2767 			ceph_decode_8_safe(&p, end, decode_redir, e_inval);
2768 		else
2769 			decode_redir = 1;
2770 	} else {
2771 		decode_redir = 0;
2772 	}
2773 
2774 	if (decode_redir) {
2775 		ret = ceph_redirect_decode(&p, end, &m->redirect);
2776 		if (ret)
2777 			return ret;
2778 	} else {
2779 		ceph_oloc_init(&m->redirect.oloc);
2780 	}
2781 
2782 	return 0;
2783 
2784 e_inval:
2785 	return -EINVAL;
2786 }
2787 
2788 /*
2789  * We are done with @req if
2790  *   - @m is a safe reply, or
2791  *   - @m is an unsafe reply and we didn't want a safe one
2792  */
2793 static bool done_request(const struct ceph_osd_request *req,
2794 			 const struct MOSDOpReply *m)
2795 {
2796 	return (m->result < 0 ||
2797 		(m->flags & CEPH_OSD_FLAG_ONDISK) ||
2798 		!(req->r_flags & CEPH_OSD_FLAG_ONDISK));
2799 }
2800 
2801 /*
2802  * handle osd op reply.  either call the callback if it is specified,
2803  * or do the completion to wake up the waiting thread.
2804  *
2805  * ->r_unsafe_callback is set?	yes			no
2806  *
2807  * first reply is OK (needed	r_cb/r_completion,	r_cb/r_completion,
2808  * any or needed/got safe)	r_done_completion	r_done_completion
2809  *
2810  * first reply is unsafe	r_unsafe_cb(true)	(nothing)
2811  *
2812  * when we get the safe reply	r_unsafe_cb(false),	r_cb/r_completion,
2813  *				r_done_completion	r_done_completion
2814  */
2815 static void handle_reply(struct ceph_osd *osd, struct ceph_msg *msg)
2816 {
2817 	struct ceph_osd_client *osdc = osd->o_osdc;
2818 	struct ceph_osd_request *req;
2819 	struct MOSDOpReply m;
2820 	u64 tid = le64_to_cpu(msg->hdr.tid);
2821 	u32 data_len = 0;
2822 	bool already_acked;
2823 	int ret;
2824 	int i;
2825 
2826 	dout("%s msg %p tid %llu\n", __func__, msg, tid);
2827 
2828 	down_read(&osdc->lock);
2829 	if (!osd_registered(osd)) {
2830 		dout("%s osd%d unknown\n", __func__, osd->o_osd);
2831 		goto out_unlock_osdc;
2832 	}
2833 	WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num));
2834 
2835 	mutex_lock(&osd->lock);
2836 	req = lookup_request(&osd->o_requests, tid);
2837 	if (!req) {
2838 		dout("%s osd%d tid %llu unknown\n", __func__, osd->o_osd, tid);
2839 		goto out_unlock_session;
2840 	}
2841 
2842 	m.redirect.oloc.pool_ns = req->r_t.target_oloc.pool_ns;
2843 	ret = decode_MOSDOpReply(msg, &m);
2844 	m.redirect.oloc.pool_ns = NULL;
2845 	if (ret) {
2846 		pr_err("failed to decode MOSDOpReply for tid %llu: %d\n",
2847 		       req->r_tid, ret);
2848 		ceph_msg_dump(msg);
2849 		goto fail_request;
2850 	}
2851 	dout("%s req %p tid %llu flags 0x%llx pgid %llu.%x epoch %u attempt %d v %u'%llu uv %llu\n",
2852 	     __func__, req, req->r_tid, m.flags, m.pgid.pool, m.pgid.seed,
2853 	     m.epoch, m.retry_attempt, le32_to_cpu(m.replay_version.epoch),
2854 	     le64_to_cpu(m.replay_version.version), m.user_version);
2855 
2856 	if (m.retry_attempt >= 0) {
2857 		if (m.retry_attempt != req->r_attempts - 1) {
2858 			dout("req %p tid %llu retry_attempt %d != %d, ignoring\n",
2859 			     req, req->r_tid, m.retry_attempt,
2860 			     req->r_attempts - 1);
2861 			goto out_unlock_session;
2862 		}
2863 	} else {
2864 		WARN_ON(1); /* MOSDOpReply v4 is assumed */
2865 	}
2866 
2867 	if (!ceph_oloc_empty(&m.redirect.oloc)) {
2868 		dout("req %p tid %llu redirect pool %lld\n", req, req->r_tid,
2869 		     m.redirect.oloc.pool);
2870 		unlink_request(osd, req);
2871 		mutex_unlock(&osd->lock);
2872 
2873 		/*
2874 		 * Not ceph_oloc_copy() - changing pool_ns is not
2875 		 * supported.
2876 		 */
2877 		req->r_t.target_oloc.pool = m.redirect.oloc.pool;
2878 		req->r_flags |= CEPH_OSD_FLAG_REDIRECTED;
2879 		req->r_tid = 0;
2880 		__submit_request(req, false);
2881 		goto out_unlock_osdc;
2882 	}
2883 
2884 	if (m.num_ops != req->r_num_ops) {
2885 		pr_err("num_ops %d != %d for tid %llu\n", m.num_ops,
2886 		       req->r_num_ops, req->r_tid);
2887 		goto fail_request;
2888 	}
2889 	for (i = 0; i < req->r_num_ops; i++) {
2890 		dout(" req %p tid %llu op %d rval %d len %u\n", req,
2891 		     req->r_tid, i, m.rval[i], m.outdata_len[i]);
2892 		req->r_ops[i].rval = m.rval[i];
2893 		req->r_ops[i].outdata_len = m.outdata_len[i];
2894 		data_len += m.outdata_len[i];
2895 	}
2896 	if (data_len != le32_to_cpu(msg->hdr.data_len)) {
2897 		pr_err("sum of lens %u != %u for tid %llu\n", data_len,
2898 		       le32_to_cpu(msg->hdr.data_len), req->r_tid);
2899 		goto fail_request;
2900 	}
2901 	dout("%s req %p tid %llu acked %d result %d data_len %u\n", __func__,
2902 	     req, req->r_tid, req->r_got_reply, m.result, data_len);
2903 
2904 	already_acked = req->r_got_reply;
2905 	if (!already_acked) {
2906 		req->r_result = m.result ?: data_len;
2907 		req->r_replay_version = m.replay_version; /* struct */
2908 		req->r_got_reply = true;
2909 	} else if (!(m.flags & CEPH_OSD_FLAG_ONDISK)) {
2910 		dout("req %p tid %llu dup ack\n", req, req->r_tid);
2911 		goto out_unlock_session;
2912 	}
2913 
2914 	if (done_request(req, &m)) {
2915 		finish_request(req);
2916 		if (req->r_linger) {
2917 			WARN_ON(req->r_unsafe_callback);
2918 			dout("req %p tid %llu cb (locked)\n", req, req->r_tid);
2919 			__complete_request(req);
2920 		}
2921 	}
2922 
2923 	mutex_unlock(&osd->lock);
2924 	up_read(&osdc->lock);
2925 
2926 	if (done_request(req, &m)) {
2927 		if (already_acked && req->r_unsafe_callback) {
2928 			dout("req %p tid %llu safe-cb\n", req, req->r_tid);
2929 			req->r_unsafe_callback(req, false);
2930 		} else if (!req->r_linger) {
2931 			dout("req %p tid %llu cb\n", req, req->r_tid);
2932 			__complete_request(req);
2933 		}
2934 		complete_all(&req->r_done_completion);
2935 		ceph_osdc_put_request(req);
2936 	} else {
2937 		if (req->r_unsafe_callback) {
2938 			dout("req %p tid %llu unsafe-cb\n", req, req->r_tid);
2939 			req->r_unsafe_callback(req, true);
2940 		} else {
2941 			WARN_ON(1);
2942 		}
2943 	}
2944 
2945 	return;
2946 
2947 fail_request:
2948 	complete_request(req, -EIO);
2949 out_unlock_session:
2950 	mutex_unlock(&osd->lock);
2951 out_unlock_osdc:
2952 	up_read(&osdc->lock);
2953 }
2954 
2955 static void set_pool_was_full(struct ceph_osd_client *osdc)
2956 {
2957 	struct rb_node *n;
2958 
2959 	for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) {
2960 		struct ceph_pg_pool_info *pi =
2961 		    rb_entry(n, struct ceph_pg_pool_info, node);
2962 
2963 		pi->was_full = __pool_full(pi);
2964 	}
2965 }
2966 
2967 static bool pool_cleared_full(struct ceph_osd_client *osdc, s64 pool_id)
2968 {
2969 	struct ceph_pg_pool_info *pi;
2970 
2971 	pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id);
2972 	if (!pi)
2973 		return false;
2974 
2975 	return pi->was_full && !__pool_full(pi);
2976 }
2977 
2978 static enum calc_target_result
2979 recalc_linger_target(struct ceph_osd_linger_request *lreq)
2980 {
2981 	struct ceph_osd_client *osdc = lreq->osdc;
2982 	enum calc_target_result ct_res;
2983 
2984 	ct_res = calc_target(osdc, &lreq->t, &lreq->last_force_resend, true);
2985 	if (ct_res == CALC_TARGET_NEED_RESEND) {
2986 		struct ceph_osd *osd;
2987 
2988 		osd = lookup_create_osd(osdc, lreq->t.osd, true);
2989 		if (osd != lreq->osd) {
2990 			unlink_linger(lreq->osd, lreq);
2991 			link_linger(osd, lreq);
2992 		}
2993 	}
2994 
2995 	return ct_res;
2996 }
2997 
2998 /*
2999  * Requeue requests whose mapping to an OSD has changed.
3000  */
3001 static void scan_requests(struct ceph_osd *osd,
3002 			  bool force_resend,
3003 			  bool cleared_full,
3004 			  bool check_pool_cleared_full,
3005 			  struct rb_root *need_resend,
3006 			  struct list_head *need_resend_linger)
3007 {
3008 	struct ceph_osd_client *osdc = osd->o_osdc;
3009 	struct rb_node *n;
3010 	bool force_resend_writes;
3011 
3012 	for (n = rb_first(&osd->o_linger_requests); n; ) {
3013 		struct ceph_osd_linger_request *lreq =
3014 		    rb_entry(n, struct ceph_osd_linger_request, node);
3015 		enum calc_target_result ct_res;
3016 
3017 		n = rb_next(n); /* recalc_linger_target() */
3018 
3019 		dout("%s lreq %p linger_id %llu\n", __func__, lreq,
3020 		     lreq->linger_id);
3021 		ct_res = recalc_linger_target(lreq);
3022 		switch (ct_res) {
3023 		case CALC_TARGET_NO_ACTION:
3024 			force_resend_writes = cleared_full ||
3025 			    (check_pool_cleared_full &&
3026 			     pool_cleared_full(osdc, lreq->t.base_oloc.pool));
3027 			if (!force_resend && !force_resend_writes)
3028 				break;
3029 
3030 			/* fall through */
3031 		case CALC_TARGET_NEED_RESEND:
3032 			cancel_linger_map_check(lreq);
3033 			/*
3034 			 * scan_requests() for the previous epoch(s)
3035 			 * may have already added it to the list, since
3036 			 * it's not unlinked here.
3037 			 */
3038 			if (list_empty(&lreq->scan_item))
3039 				list_add_tail(&lreq->scan_item, need_resend_linger);
3040 			break;
3041 		case CALC_TARGET_POOL_DNE:
3042 			check_linger_pool_dne(lreq);
3043 			break;
3044 		}
3045 	}
3046 
3047 	for (n = rb_first(&osd->o_requests); n; ) {
3048 		struct ceph_osd_request *req =
3049 		    rb_entry(n, struct ceph_osd_request, r_node);
3050 		enum calc_target_result ct_res;
3051 
3052 		n = rb_next(n); /* unlink_request(), check_pool_dne() */
3053 
3054 		dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
3055 		ct_res = calc_target(osdc, &req->r_t,
3056 				     &req->r_last_force_resend, false);
3057 		switch (ct_res) {
3058 		case CALC_TARGET_NO_ACTION:
3059 			force_resend_writes = cleared_full ||
3060 			    (check_pool_cleared_full &&
3061 			     pool_cleared_full(osdc, req->r_t.base_oloc.pool));
3062 			if (!force_resend &&
3063 			    (!(req->r_flags & CEPH_OSD_FLAG_WRITE) ||
3064 			     !force_resend_writes))
3065 				break;
3066 
3067 			/* fall through */
3068 		case CALC_TARGET_NEED_RESEND:
3069 			cancel_map_check(req);
3070 			unlink_request(osd, req);
3071 			insert_request(need_resend, req);
3072 			break;
3073 		case CALC_TARGET_POOL_DNE:
3074 			check_pool_dne(req);
3075 			break;
3076 		}
3077 	}
3078 }
3079 
3080 static int handle_one_map(struct ceph_osd_client *osdc,
3081 			  void *p, void *end, bool incremental,
3082 			  struct rb_root *need_resend,
3083 			  struct list_head *need_resend_linger)
3084 {
3085 	struct ceph_osdmap *newmap;
3086 	struct rb_node *n;
3087 	bool skipped_map = false;
3088 	bool was_full;
3089 
3090 	was_full = ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL);
3091 	set_pool_was_full(osdc);
3092 
3093 	if (incremental)
3094 		newmap = osdmap_apply_incremental(&p, end, osdc->osdmap);
3095 	else
3096 		newmap = ceph_osdmap_decode(&p, end);
3097 	if (IS_ERR(newmap))
3098 		return PTR_ERR(newmap);
3099 
3100 	if (newmap != osdc->osdmap) {
3101 		/*
3102 		 * Preserve ->was_full before destroying the old map.
3103 		 * For pools that weren't in the old map, ->was_full
3104 		 * should be false.
3105 		 */
3106 		for (n = rb_first(&newmap->pg_pools); n; n = rb_next(n)) {
3107 			struct ceph_pg_pool_info *pi =
3108 			    rb_entry(n, struct ceph_pg_pool_info, node);
3109 			struct ceph_pg_pool_info *old_pi;
3110 
3111 			old_pi = ceph_pg_pool_by_id(osdc->osdmap, pi->id);
3112 			if (old_pi)
3113 				pi->was_full = old_pi->was_full;
3114 			else
3115 				WARN_ON(pi->was_full);
3116 		}
3117 
3118 		if (osdc->osdmap->epoch &&
3119 		    osdc->osdmap->epoch + 1 < newmap->epoch) {
3120 			WARN_ON(incremental);
3121 			skipped_map = true;
3122 		}
3123 
3124 		ceph_osdmap_destroy(osdc->osdmap);
3125 		osdc->osdmap = newmap;
3126 	}
3127 
3128 	was_full &= !ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL);
3129 	scan_requests(&osdc->homeless_osd, skipped_map, was_full, true,
3130 		      need_resend, need_resend_linger);
3131 
3132 	for (n = rb_first(&osdc->osds); n; ) {
3133 		struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
3134 
3135 		n = rb_next(n); /* close_osd() */
3136 
3137 		scan_requests(osd, skipped_map, was_full, true, need_resend,
3138 			      need_resend_linger);
3139 		if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
3140 		    memcmp(&osd->o_con.peer_addr,
3141 			   ceph_osd_addr(osdc->osdmap, osd->o_osd),
3142 			   sizeof(struct ceph_entity_addr)))
3143 			close_osd(osd);
3144 	}
3145 
3146 	return 0;
3147 }
3148 
3149 static void kick_requests(struct ceph_osd_client *osdc,
3150 			  struct rb_root *need_resend,
3151 			  struct list_head *need_resend_linger)
3152 {
3153 	struct ceph_osd_linger_request *lreq, *nlreq;
3154 	struct rb_node *n;
3155 
3156 	for (n = rb_first(need_resend); n; ) {
3157 		struct ceph_osd_request *req =
3158 		    rb_entry(n, struct ceph_osd_request, r_node);
3159 		struct ceph_osd *osd;
3160 
3161 		n = rb_next(n);
3162 		erase_request(need_resend, req); /* before link_request() */
3163 
3164 		WARN_ON(req->r_osd);
3165 		calc_target(osdc, &req->r_t, NULL, false);
3166 		osd = lookup_create_osd(osdc, req->r_t.osd, true);
3167 		link_request(osd, req);
3168 		if (!req->r_linger) {
3169 			if (!osd_homeless(osd) && !req->r_t.paused)
3170 				send_request(req);
3171 		} else {
3172 			cancel_linger_request(req);
3173 		}
3174 	}
3175 
3176 	list_for_each_entry_safe(lreq, nlreq, need_resend_linger, scan_item) {
3177 		if (!osd_homeless(lreq->osd))
3178 			send_linger(lreq);
3179 
3180 		list_del_init(&lreq->scan_item);
3181 	}
3182 }
3183 
3184 /*
3185  * Process updated osd map.
3186  *
3187  * The message contains any number of incremental and full maps, normally
3188  * indicating some sort of topology change in the cluster.  Kick requests
3189  * off to different OSDs as needed.
3190  */
3191 void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
3192 {
3193 	void *p = msg->front.iov_base;
3194 	void *const end = p + msg->front.iov_len;
3195 	u32 nr_maps, maplen;
3196 	u32 epoch;
3197 	struct ceph_fsid fsid;
3198 	struct rb_root need_resend = RB_ROOT;
3199 	LIST_HEAD(need_resend_linger);
3200 	bool handled_incremental = false;
3201 	bool was_pauserd, was_pausewr;
3202 	bool pauserd, pausewr;
3203 	int err;
3204 
3205 	dout("%s have %u\n", __func__, osdc->osdmap->epoch);
3206 	down_write(&osdc->lock);
3207 
3208 	/* verify fsid */
3209 	ceph_decode_need(&p, end, sizeof(fsid), bad);
3210 	ceph_decode_copy(&p, &fsid, sizeof(fsid));
3211 	if (ceph_check_fsid(osdc->client, &fsid) < 0)
3212 		goto bad;
3213 
3214 	was_pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
3215 	was_pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
3216 		      ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
3217 		      have_pool_full(osdc);
3218 
3219 	/* incremental maps */
3220 	ceph_decode_32_safe(&p, end, nr_maps, bad);
3221 	dout(" %d inc maps\n", nr_maps);
3222 	while (nr_maps > 0) {
3223 		ceph_decode_need(&p, end, 2*sizeof(u32), bad);
3224 		epoch = ceph_decode_32(&p);
3225 		maplen = ceph_decode_32(&p);
3226 		ceph_decode_need(&p, end, maplen, bad);
3227 		if (osdc->osdmap->epoch &&
3228 		    osdc->osdmap->epoch + 1 == epoch) {
3229 			dout("applying incremental map %u len %d\n",
3230 			     epoch, maplen);
3231 			err = handle_one_map(osdc, p, p + maplen, true,
3232 					     &need_resend, &need_resend_linger);
3233 			if (err)
3234 				goto bad;
3235 			handled_incremental = true;
3236 		} else {
3237 			dout("ignoring incremental map %u len %d\n",
3238 			     epoch, maplen);
3239 		}
3240 		p += maplen;
3241 		nr_maps--;
3242 	}
3243 	if (handled_incremental)
3244 		goto done;
3245 
3246 	/* full maps */
3247 	ceph_decode_32_safe(&p, end, nr_maps, bad);
3248 	dout(" %d full maps\n", nr_maps);
3249 	while (nr_maps) {
3250 		ceph_decode_need(&p, end, 2*sizeof(u32), bad);
3251 		epoch = ceph_decode_32(&p);
3252 		maplen = ceph_decode_32(&p);
3253 		ceph_decode_need(&p, end, maplen, bad);
3254 		if (nr_maps > 1) {
3255 			dout("skipping non-latest full map %u len %d\n",
3256 			     epoch, maplen);
3257 		} else if (osdc->osdmap->epoch >= epoch) {
3258 			dout("skipping full map %u len %d, "
3259 			     "older than our %u\n", epoch, maplen,
3260 			     osdc->osdmap->epoch);
3261 		} else {
3262 			dout("taking full map %u len %d\n", epoch, maplen);
3263 			err = handle_one_map(osdc, p, p + maplen, false,
3264 					     &need_resend, &need_resend_linger);
3265 			if (err)
3266 				goto bad;
3267 		}
3268 		p += maplen;
3269 		nr_maps--;
3270 	}
3271 
3272 done:
3273 	/*
3274 	 * subscribe to subsequent osdmap updates if full to ensure
3275 	 * we find out when we are no longer full and stop returning
3276 	 * ENOSPC.
3277 	 */
3278 	pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
3279 	pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
3280 		  ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
3281 		  have_pool_full(osdc);
3282 	if (was_pauserd || was_pausewr || pauserd || pausewr)
3283 		maybe_request_map(osdc);
3284 
3285 	kick_requests(osdc, &need_resend, &need_resend_linger);
3286 
3287 	ceph_monc_got_map(&osdc->client->monc, CEPH_SUB_OSDMAP,
3288 			  osdc->osdmap->epoch);
3289 	up_write(&osdc->lock);
3290 	wake_up_all(&osdc->client->auth_wq);
3291 	return;
3292 
3293 bad:
3294 	pr_err("osdc handle_map corrupt msg\n");
3295 	ceph_msg_dump(msg);
3296 	up_write(&osdc->lock);
3297 }
3298 
3299 /*
3300  * Resubmit requests pending on the given osd.
3301  */
3302 static void kick_osd_requests(struct ceph_osd *osd)
3303 {
3304 	struct rb_node *n;
3305 
3306 	for (n = rb_first(&osd->o_requests); n; ) {
3307 		struct ceph_osd_request *req =
3308 		    rb_entry(n, struct ceph_osd_request, r_node);
3309 
3310 		n = rb_next(n); /* cancel_linger_request() */
3311 
3312 		if (!req->r_linger) {
3313 			if (!req->r_t.paused)
3314 				send_request(req);
3315 		} else {
3316 			cancel_linger_request(req);
3317 		}
3318 	}
3319 	for (n = rb_first(&osd->o_linger_requests); n; n = rb_next(n)) {
3320 		struct ceph_osd_linger_request *lreq =
3321 		    rb_entry(n, struct ceph_osd_linger_request, node);
3322 
3323 		send_linger(lreq);
3324 	}
3325 }
3326 
3327 /*
3328  * If the osd connection drops, we need to resubmit all requests.
3329  */
3330 static void osd_fault(struct ceph_connection *con)
3331 {
3332 	struct ceph_osd *osd = con->private;
3333 	struct ceph_osd_client *osdc = osd->o_osdc;
3334 
3335 	dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
3336 
3337 	down_write(&osdc->lock);
3338 	if (!osd_registered(osd)) {
3339 		dout("%s osd%d unknown\n", __func__, osd->o_osd);
3340 		goto out_unlock;
3341 	}
3342 
3343 	if (!reopen_osd(osd))
3344 		kick_osd_requests(osd);
3345 	maybe_request_map(osdc);
3346 
3347 out_unlock:
3348 	up_write(&osdc->lock);
3349 }
3350 
3351 /*
3352  * Process osd watch notifications
3353  */
3354 static void handle_watch_notify(struct ceph_osd_client *osdc,
3355 				struct ceph_msg *msg)
3356 {
3357 	void *p = msg->front.iov_base;
3358 	void *const end = p + msg->front.iov_len;
3359 	struct ceph_osd_linger_request *lreq;
3360 	struct linger_work *lwork;
3361 	u8 proto_ver, opcode;
3362 	u64 cookie, notify_id;
3363 	u64 notifier_id = 0;
3364 	s32 return_code = 0;
3365 	void *payload = NULL;
3366 	u32 payload_len = 0;
3367 
3368 	ceph_decode_8_safe(&p, end, proto_ver, bad);
3369 	ceph_decode_8_safe(&p, end, opcode, bad);
3370 	ceph_decode_64_safe(&p, end, cookie, bad);
3371 	p += 8; /* skip ver */
3372 	ceph_decode_64_safe(&p, end, notify_id, bad);
3373 
3374 	if (proto_ver >= 1) {
3375 		ceph_decode_32_safe(&p, end, payload_len, bad);
3376 		ceph_decode_need(&p, end, payload_len, bad);
3377 		payload = p;
3378 		p += payload_len;
3379 	}
3380 
3381 	if (le16_to_cpu(msg->hdr.version) >= 2)
3382 		ceph_decode_32_safe(&p, end, return_code, bad);
3383 
3384 	if (le16_to_cpu(msg->hdr.version) >= 3)
3385 		ceph_decode_64_safe(&p, end, notifier_id, bad);
3386 
3387 	down_read(&osdc->lock);
3388 	lreq = lookup_linger_osdc(&osdc->linger_requests, cookie);
3389 	if (!lreq) {
3390 		dout("%s opcode %d cookie %llu dne\n", __func__, opcode,
3391 		     cookie);
3392 		goto out_unlock_osdc;
3393 	}
3394 
3395 	mutex_lock(&lreq->lock);
3396 	dout("%s opcode %d cookie %llu lreq %p is_watch %d\n", __func__,
3397 	     opcode, cookie, lreq, lreq->is_watch);
3398 	if (opcode == CEPH_WATCH_EVENT_DISCONNECT) {
3399 		if (!lreq->last_error) {
3400 			lreq->last_error = -ENOTCONN;
3401 			queue_watch_error(lreq);
3402 		}
3403 	} else if (!lreq->is_watch) {
3404 		/* CEPH_WATCH_EVENT_NOTIFY_COMPLETE */
3405 		if (lreq->notify_id && lreq->notify_id != notify_id) {
3406 			dout("lreq %p notify_id %llu != %llu, ignoring\n", lreq,
3407 			     lreq->notify_id, notify_id);
3408 		} else if (!completion_done(&lreq->notify_finish_wait)) {
3409 			struct ceph_msg_data *data =
3410 			    list_first_entry_or_null(&msg->data,
3411 						     struct ceph_msg_data,
3412 						     links);
3413 
3414 			if (data) {
3415 				if (lreq->preply_pages) {
3416 					WARN_ON(data->type !=
3417 							CEPH_MSG_DATA_PAGES);
3418 					*lreq->preply_pages = data->pages;
3419 					*lreq->preply_len = data->length;
3420 				} else {
3421 					ceph_release_page_vector(data->pages,
3422 					       calc_pages_for(0, data->length));
3423 				}
3424 			}
3425 			lreq->notify_finish_error = return_code;
3426 			complete_all(&lreq->notify_finish_wait);
3427 		}
3428 	} else {
3429 		/* CEPH_WATCH_EVENT_NOTIFY */
3430 		lwork = lwork_alloc(lreq, do_watch_notify);
3431 		if (!lwork) {
3432 			pr_err("failed to allocate notify-lwork\n");
3433 			goto out_unlock_lreq;
3434 		}
3435 
3436 		lwork->notify.notify_id = notify_id;
3437 		lwork->notify.notifier_id = notifier_id;
3438 		lwork->notify.payload = payload;
3439 		lwork->notify.payload_len = payload_len;
3440 		lwork->notify.msg = ceph_msg_get(msg);
3441 		lwork_queue(lwork);
3442 	}
3443 
3444 out_unlock_lreq:
3445 	mutex_unlock(&lreq->lock);
3446 out_unlock_osdc:
3447 	up_read(&osdc->lock);
3448 	return;
3449 
3450 bad:
3451 	pr_err("osdc handle_watch_notify corrupt msg\n");
3452 }
3453 
3454 /*
3455  * Register request, send initial attempt.
3456  */
3457 int ceph_osdc_start_request(struct ceph_osd_client *osdc,
3458 			    struct ceph_osd_request *req,
3459 			    bool nofail)
3460 {
3461 	down_read(&osdc->lock);
3462 	submit_request(req, false);
3463 	up_read(&osdc->lock);
3464 
3465 	return 0;
3466 }
3467 EXPORT_SYMBOL(ceph_osdc_start_request);
3468 
3469 /*
3470  * Unregister a registered request.  The request is not completed:
3471  * ->r_result isn't set and __complete_request() isn't called.
3472  */
3473 void ceph_osdc_cancel_request(struct ceph_osd_request *req)
3474 {
3475 	struct ceph_osd_client *osdc = req->r_osdc;
3476 
3477 	down_write(&osdc->lock);
3478 	if (req->r_osd)
3479 		cancel_request(req);
3480 	up_write(&osdc->lock);
3481 }
3482 EXPORT_SYMBOL(ceph_osdc_cancel_request);
3483 
3484 /*
3485  * @timeout: in jiffies, 0 means "wait forever"
3486  */
3487 static int wait_request_timeout(struct ceph_osd_request *req,
3488 				unsigned long timeout)
3489 {
3490 	long left;
3491 
3492 	dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
3493 	left = wait_for_completion_killable_timeout(&req->r_completion,
3494 						ceph_timeout_jiffies(timeout));
3495 	if (left <= 0) {
3496 		left = left ?: -ETIMEDOUT;
3497 		ceph_osdc_cancel_request(req);
3498 	} else {
3499 		left = req->r_result; /* completed */
3500 	}
3501 
3502 	return left;
3503 }
3504 
3505 /*
3506  * wait for a request to complete
3507  */
3508 int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
3509 			   struct ceph_osd_request *req)
3510 {
3511 	return wait_request_timeout(req, 0);
3512 }
3513 EXPORT_SYMBOL(ceph_osdc_wait_request);
3514 
3515 /*
3516  * sync - wait for all in-flight requests to flush.  avoid starvation.
3517  */
3518 void ceph_osdc_sync(struct ceph_osd_client *osdc)
3519 {
3520 	struct rb_node *n, *p;
3521 	u64 last_tid = atomic64_read(&osdc->last_tid);
3522 
3523 again:
3524 	down_read(&osdc->lock);
3525 	for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
3526 		struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
3527 
3528 		mutex_lock(&osd->lock);
3529 		for (p = rb_first(&osd->o_requests); p; p = rb_next(p)) {
3530 			struct ceph_osd_request *req =
3531 			    rb_entry(p, struct ceph_osd_request, r_node);
3532 
3533 			if (req->r_tid > last_tid)
3534 				break;
3535 
3536 			if (!(req->r_flags & CEPH_OSD_FLAG_WRITE))
3537 				continue;
3538 
3539 			ceph_osdc_get_request(req);
3540 			mutex_unlock(&osd->lock);
3541 			up_read(&osdc->lock);
3542 			dout("%s waiting on req %p tid %llu last_tid %llu\n",
3543 			     __func__, req, req->r_tid, last_tid);
3544 			wait_for_completion(&req->r_done_completion);
3545 			ceph_osdc_put_request(req);
3546 			goto again;
3547 		}
3548 
3549 		mutex_unlock(&osd->lock);
3550 	}
3551 
3552 	up_read(&osdc->lock);
3553 	dout("%s done last_tid %llu\n", __func__, last_tid);
3554 }
3555 EXPORT_SYMBOL(ceph_osdc_sync);
3556 
3557 static struct ceph_osd_request *
3558 alloc_linger_request(struct ceph_osd_linger_request *lreq)
3559 {
3560 	struct ceph_osd_request *req;
3561 
3562 	req = ceph_osdc_alloc_request(lreq->osdc, NULL, 1, false, GFP_NOIO);
3563 	if (!req)
3564 		return NULL;
3565 
3566 	ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
3567 	ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
3568 
3569 	if (ceph_osdc_alloc_messages(req, GFP_NOIO)) {
3570 		ceph_osdc_put_request(req);
3571 		return NULL;
3572 	}
3573 
3574 	return req;
3575 }
3576 
3577 /*
3578  * Returns a handle, caller owns a ref.
3579  */
3580 struct ceph_osd_linger_request *
3581 ceph_osdc_watch(struct ceph_osd_client *osdc,
3582 		struct ceph_object_id *oid,
3583 		struct ceph_object_locator *oloc,
3584 		rados_watchcb2_t wcb,
3585 		rados_watcherrcb_t errcb,
3586 		void *data)
3587 {
3588 	struct ceph_osd_linger_request *lreq;
3589 	int ret;
3590 
3591 	lreq = linger_alloc(osdc);
3592 	if (!lreq)
3593 		return ERR_PTR(-ENOMEM);
3594 
3595 	lreq->is_watch = true;
3596 	lreq->wcb = wcb;
3597 	lreq->errcb = errcb;
3598 	lreq->data = data;
3599 	lreq->watch_valid_thru = jiffies;
3600 
3601 	ceph_oid_copy(&lreq->t.base_oid, oid);
3602 	ceph_oloc_copy(&lreq->t.base_oloc, oloc);
3603 	lreq->t.flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
3604 	lreq->mtime = CURRENT_TIME;
3605 
3606 	lreq->reg_req = alloc_linger_request(lreq);
3607 	if (!lreq->reg_req) {
3608 		ret = -ENOMEM;
3609 		goto err_put_lreq;
3610 	}
3611 
3612 	lreq->ping_req = alloc_linger_request(lreq);
3613 	if (!lreq->ping_req) {
3614 		ret = -ENOMEM;
3615 		goto err_put_lreq;
3616 	}
3617 
3618 	down_write(&osdc->lock);
3619 	linger_register(lreq); /* before osd_req_op_* */
3620 	osd_req_op_watch_init(lreq->reg_req, 0, lreq->linger_id,
3621 			      CEPH_OSD_WATCH_OP_WATCH);
3622 	osd_req_op_watch_init(lreq->ping_req, 0, lreq->linger_id,
3623 			      CEPH_OSD_WATCH_OP_PING);
3624 	linger_submit(lreq);
3625 	up_write(&osdc->lock);
3626 
3627 	ret = linger_reg_commit_wait(lreq);
3628 	if (ret) {
3629 		linger_cancel(lreq);
3630 		goto err_put_lreq;
3631 	}
3632 
3633 	return lreq;
3634 
3635 err_put_lreq:
3636 	linger_put(lreq);
3637 	return ERR_PTR(ret);
3638 }
3639 EXPORT_SYMBOL(ceph_osdc_watch);
3640 
3641 /*
3642  * Releases a ref.
3643  *
3644  * Times out after mount_timeout to preserve rbd unmap behaviour
3645  * introduced in 2894e1d76974 ("rbd: timeout watch teardown on unmap
3646  * with mount_timeout").
3647  */
3648 int ceph_osdc_unwatch(struct ceph_osd_client *osdc,
3649 		      struct ceph_osd_linger_request *lreq)
3650 {
3651 	struct ceph_options *opts = osdc->client->options;
3652 	struct ceph_osd_request *req;
3653 	int ret;
3654 
3655 	req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
3656 	if (!req)
3657 		return -ENOMEM;
3658 
3659 	ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
3660 	ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
3661 	req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
3662 	req->r_mtime = CURRENT_TIME;
3663 	osd_req_op_watch_init(req, 0, lreq->linger_id,
3664 			      CEPH_OSD_WATCH_OP_UNWATCH);
3665 
3666 	ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
3667 	if (ret)
3668 		goto out_put_req;
3669 
3670 	ceph_osdc_start_request(osdc, req, false);
3671 	linger_cancel(lreq);
3672 	linger_put(lreq);
3673 	ret = wait_request_timeout(req, opts->mount_timeout);
3674 
3675 out_put_req:
3676 	ceph_osdc_put_request(req);
3677 	return ret;
3678 }
3679 EXPORT_SYMBOL(ceph_osdc_unwatch);
3680 
3681 static int osd_req_op_notify_ack_init(struct ceph_osd_request *req, int which,
3682 				      u64 notify_id, u64 cookie, void *payload,
3683 				      size_t payload_len)
3684 {
3685 	struct ceph_osd_req_op *op;
3686 	struct ceph_pagelist *pl;
3687 	int ret;
3688 
3689 	op = _osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY_ACK, 0);
3690 
3691 	pl = kmalloc(sizeof(*pl), GFP_NOIO);
3692 	if (!pl)
3693 		return -ENOMEM;
3694 
3695 	ceph_pagelist_init(pl);
3696 	ret = ceph_pagelist_encode_64(pl, notify_id);
3697 	ret |= ceph_pagelist_encode_64(pl, cookie);
3698 	if (payload) {
3699 		ret |= ceph_pagelist_encode_32(pl, payload_len);
3700 		ret |= ceph_pagelist_append(pl, payload, payload_len);
3701 	} else {
3702 		ret |= ceph_pagelist_encode_32(pl, 0);
3703 	}
3704 	if (ret) {
3705 		ceph_pagelist_release(pl);
3706 		return -ENOMEM;
3707 	}
3708 
3709 	ceph_osd_data_pagelist_init(&op->notify_ack.request_data, pl);
3710 	op->indata_len = pl->length;
3711 	return 0;
3712 }
3713 
3714 int ceph_osdc_notify_ack(struct ceph_osd_client *osdc,
3715 			 struct ceph_object_id *oid,
3716 			 struct ceph_object_locator *oloc,
3717 			 u64 notify_id,
3718 			 u64 cookie,
3719 			 void *payload,
3720 			 size_t payload_len)
3721 {
3722 	struct ceph_osd_request *req;
3723 	int ret;
3724 
3725 	req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
3726 	if (!req)
3727 		return -ENOMEM;
3728 
3729 	ceph_oid_copy(&req->r_base_oid, oid);
3730 	ceph_oloc_copy(&req->r_base_oloc, oloc);
3731 	req->r_flags = CEPH_OSD_FLAG_READ;
3732 
3733 	ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
3734 	if (ret)
3735 		goto out_put_req;
3736 
3737 	ret = osd_req_op_notify_ack_init(req, 0, notify_id, cookie, payload,
3738 					 payload_len);
3739 	if (ret)
3740 		goto out_put_req;
3741 
3742 	ceph_osdc_start_request(osdc, req, false);
3743 	ret = ceph_osdc_wait_request(osdc, req);
3744 
3745 out_put_req:
3746 	ceph_osdc_put_request(req);
3747 	return ret;
3748 }
3749 EXPORT_SYMBOL(ceph_osdc_notify_ack);
3750 
3751 static int osd_req_op_notify_init(struct ceph_osd_request *req, int which,
3752 				  u64 cookie, u32 prot_ver, u32 timeout,
3753 				  void *payload, size_t payload_len)
3754 {
3755 	struct ceph_osd_req_op *op;
3756 	struct ceph_pagelist *pl;
3757 	int ret;
3758 
3759 	op = _osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY, 0);
3760 	op->notify.cookie = cookie;
3761 
3762 	pl = kmalloc(sizeof(*pl), GFP_NOIO);
3763 	if (!pl)
3764 		return -ENOMEM;
3765 
3766 	ceph_pagelist_init(pl);
3767 	ret = ceph_pagelist_encode_32(pl, 1); /* prot_ver */
3768 	ret |= ceph_pagelist_encode_32(pl, timeout);
3769 	ret |= ceph_pagelist_encode_32(pl, payload_len);
3770 	ret |= ceph_pagelist_append(pl, payload, payload_len);
3771 	if (ret) {
3772 		ceph_pagelist_release(pl);
3773 		return -ENOMEM;
3774 	}
3775 
3776 	ceph_osd_data_pagelist_init(&op->notify.request_data, pl);
3777 	op->indata_len = pl->length;
3778 	return 0;
3779 }
3780 
3781 /*
3782  * @timeout: in seconds
3783  *
3784  * @preply_{pages,len} are initialized both on success and error.
3785  * The caller is responsible for:
3786  *
3787  *     ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len))
3788  */
3789 int ceph_osdc_notify(struct ceph_osd_client *osdc,
3790 		     struct ceph_object_id *oid,
3791 		     struct ceph_object_locator *oloc,
3792 		     void *payload,
3793 		     size_t payload_len,
3794 		     u32 timeout,
3795 		     struct page ***preply_pages,
3796 		     size_t *preply_len)
3797 {
3798 	struct ceph_osd_linger_request *lreq;
3799 	struct page **pages;
3800 	int ret;
3801 
3802 	WARN_ON(!timeout);
3803 	if (preply_pages) {
3804 		*preply_pages = NULL;
3805 		*preply_len = 0;
3806 	}
3807 
3808 	lreq = linger_alloc(osdc);
3809 	if (!lreq)
3810 		return -ENOMEM;
3811 
3812 	lreq->preply_pages = preply_pages;
3813 	lreq->preply_len = preply_len;
3814 
3815 	ceph_oid_copy(&lreq->t.base_oid, oid);
3816 	ceph_oloc_copy(&lreq->t.base_oloc, oloc);
3817 	lreq->t.flags = CEPH_OSD_FLAG_READ;
3818 
3819 	lreq->reg_req = alloc_linger_request(lreq);
3820 	if (!lreq->reg_req) {
3821 		ret = -ENOMEM;
3822 		goto out_put_lreq;
3823 	}
3824 
3825 	/* for notify_id */
3826 	pages = ceph_alloc_page_vector(1, GFP_NOIO);
3827 	if (IS_ERR(pages)) {
3828 		ret = PTR_ERR(pages);
3829 		goto out_put_lreq;
3830 	}
3831 
3832 	down_write(&osdc->lock);
3833 	linger_register(lreq); /* before osd_req_op_* */
3834 	ret = osd_req_op_notify_init(lreq->reg_req, 0, lreq->linger_id, 1,
3835 				     timeout, payload, payload_len);
3836 	if (ret) {
3837 		linger_unregister(lreq);
3838 		up_write(&osdc->lock);
3839 		ceph_release_page_vector(pages, 1);
3840 		goto out_put_lreq;
3841 	}
3842 	ceph_osd_data_pages_init(osd_req_op_data(lreq->reg_req, 0, notify,
3843 						 response_data),
3844 				 pages, PAGE_SIZE, 0, false, true);
3845 	linger_submit(lreq);
3846 	up_write(&osdc->lock);
3847 
3848 	ret = linger_reg_commit_wait(lreq);
3849 	if (!ret)
3850 		ret = linger_notify_finish_wait(lreq);
3851 	else
3852 		dout("lreq %p failed to initiate notify %d\n", lreq, ret);
3853 
3854 	linger_cancel(lreq);
3855 out_put_lreq:
3856 	linger_put(lreq);
3857 	return ret;
3858 }
3859 EXPORT_SYMBOL(ceph_osdc_notify);
3860 
3861 /*
3862  * Return the number of milliseconds since the watch was last
3863  * confirmed, or an error.  If there is an error, the watch is no
3864  * longer valid, and should be destroyed with ceph_osdc_unwatch().
3865  */
3866 int ceph_osdc_watch_check(struct ceph_osd_client *osdc,
3867 			  struct ceph_osd_linger_request *lreq)
3868 {
3869 	unsigned long stamp, age;
3870 	int ret;
3871 
3872 	down_read(&osdc->lock);
3873 	mutex_lock(&lreq->lock);
3874 	stamp = lreq->watch_valid_thru;
3875 	if (!list_empty(&lreq->pending_lworks)) {
3876 		struct linger_work *lwork =
3877 		    list_first_entry(&lreq->pending_lworks,
3878 				     struct linger_work,
3879 				     pending_item);
3880 
3881 		if (time_before(lwork->queued_stamp, stamp))
3882 			stamp = lwork->queued_stamp;
3883 	}
3884 	age = jiffies - stamp;
3885 	dout("%s lreq %p linger_id %llu age %lu last_error %d\n", __func__,
3886 	     lreq, lreq->linger_id, age, lreq->last_error);
3887 	/* we are truncating to msecs, so return a safe upper bound */
3888 	ret = lreq->last_error ?: 1 + jiffies_to_msecs(age);
3889 
3890 	mutex_unlock(&lreq->lock);
3891 	up_read(&osdc->lock);
3892 	return ret;
3893 }
3894 
3895 static int decode_watcher(void **p, void *end, struct ceph_watch_item *item)
3896 {
3897 	u8 struct_v;
3898 	u32 struct_len;
3899 	int ret;
3900 
3901 	ret = ceph_start_decoding(p, end, 2, "watch_item_t",
3902 				  &struct_v, &struct_len);
3903 	if (ret)
3904 		return ret;
3905 
3906 	ceph_decode_copy(p, &item->name, sizeof(item->name));
3907 	item->cookie = ceph_decode_64(p);
3908 	*p += 4; /* skip timeout_seconds */
3909 	if (struct_v >= 2) {
3910 		ceph_decode_copy(p, &item->addr, sizeof(item->addr));
3911 		ceph_decode_addr(&item->addr);
3912 	}
3913 
3914 	dout("%s %s%llu cookie %llu addr %s\n", __func__,
3915 	     ENTITY_NAME(item->name), item->cookie,
3916 	     ceph_pr_addr(&item->addr.in_addr));
3917 	return 0;
3918 }
3919 
3920 static int decode_watchers(void **p, void *end,
3921 			   struct ceph_watch_item **watchers,
3922 			   u32 *num_watchers)
3923 {
3924 	u8 struct_v;
3925 	u32 struct_len;
3926 	int i;
3927 	int ret;
3928 
3929 	ret = ceph_start_decoding(p, end, 1, "obj_list_watch_response_t",
3930 				  &struct_v, &struct_len);
3931 	if (ret)
3932 		return ret;
3933 
3934 	*num_watchers = ceph_decode_32(p);
3935 	*watchers = kcalloc(*num_watchers, sizeof(**watchers), GFP_NOIO);
3936 	if (!*watchers)
3937 		return -ENOMEM;
3938 
3939 	for (i = 0; i < *num_watchers; i++) {
3940 		ret = decode_watcher(p, end, *watchers + i);
3941 		if (ret) {
3942 			kfree(*watchers);
3943 			return ret;
3944 		}
3945 	}
3946 
3947 	return 0;
3948 }
3949 
3950 /*
3951  * On success, the caller is responsible for:
3952  *
3953  *     kfree(watchers);
3954  */
3955 int ceph_osdc_list_watchers(struct ceph_osd_client *osdc,
3956 			    struct ceph_object_id *oid,
3957 			    struct ceph_object_locator *oloc,
3958 			    struct ceph_watch_item **watchers,
3959 			    u32 *num_watchers)
3960 {
3961 	struct ceph_osd_request *req;
3962 	struct page **pages;
3963 	int ret;
3964 
3965 	req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
3966 	if (!req)
3967 		return -ENOMEM;
3968 
3969 	ceph_oid_copy(&req->r_base_oid, oid);
3970 	ceph_oloc_copy(&req->r_base_oloc, oloc);
3971 	req->r_flags = CEPH_OSD_FLAG_READ;
3972 
3973 	ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
3974 	if (ret)
3975 		goto out_put_req;
3976 
3977 	pages = ceph_alloc_page_vector(1, GFP_NOIO);
3978 	if (IS_ERR(pages)) {
3979 		ret = PTR_ERR(pages);
3980 		goto out_put_req;
3981 	}
3982 
3983 	osd_req_op_init(req, 0, CEPH_OSD_OP_LIST_WATCHERS, 0);
3984 	ceph_osd_data_pages_init(osd_req_op_data(req, 0, list_watchers,
3985 						 response_data),
3986 				 pages, PAGE_SIZE, 0, false, true);
3987 
3988 	ceph_osdc_start_request(osdc, req, false);
3989 	ret = ceph_osdc_wait_request(osdc, req);
3990 	if (ret >= 0) {
3991 		void *p = page_address(pages[0]);
3992 		void *const end = p + req->r_ops[0].outdata_len;
3993 
3994 		ret = decode_watchers(&p, end, watchers, num_watchers);
3995 	}
3996 
3997 out_put_req:
3998 	ceph_osdc_put_request(req);
3999 	return ret;
4000 }
4001 EXPORT_SYMBOL(ceph_osdc_list_watchers);
4002 
4003 /*
4004  * Call all pending notify callbacks - for use after a watch is
4005  * unregistered, to make sure no more callbacks for it will be invoked
4006  */
4007 void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc)
4008 {
4009 	dout("%s osdc %p\n", __func__, osdc);
4010 	flush_workqueue(osdc->notify_wq);
4011 }
4012 EXPORT_SYMBOL(ceph_osdc_flush_notifies);
4013 
4014 void ceph_osdc_maybe_request_map(struct ceph_osd_client *osdc)
4015 {
4016 	down_read(&osdc->lock);
4017 	maybe_request_map(osdc);
4018 	up_read(&osdc->lock);
4019 }
4020 EXPORT_SYMBOL(ceph_osdc_maybe_request_map);
4021 
4022 /*
4023  * Execute an OSD class method on an object.
4024  *
4025  * @flags: CEPH_OSD_FLAG_*
4026  * @resp_len: out param for reply length
4027  */
4028 int ceph_osdc_call(struct ceph_osd_client *osdc,
4029 		   struct ceph_object_id *oid,
4030 		   struct ceph_object_locator *oloc,
4031 		   const char *class, const char *method,
4032 		   unsigned int flags,
4033 		   struct page *req_page, size_t req_len,
4034 		   struct page *resp_page, size_t *resp_len)
4035 {
4036 	struct ceph_osd_request *req;
4037 	int ret;
4038 
4039 	req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
4040 	if (!req)
4041 		return -ENOMEM;
4042 
4043 	ceph_oid_copy(&req->r_base_oid, oid);
4044 	ceph_oloc_copy(&req->r_base_oloc, oloc);
4045 	req->r_flags = flags;
4046 
4047 	ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
4048 	if (ret)
4049 		goto out_put_req;
4050 
4051 	osd_req_op_cls_init(req, 0, CEPH_OSD_OP_CALL, class, method);
4052 	if (req_page)
4053 		osd_req_op_cls_request_data_pages(req, 0, &req_page, req_len,
4054 						  0, false, false);
4055 	if (resp_page)
4056 		osd_req_op_cls_response_data_pages(req, 0, &resp_page,
4057 						   PAGE_SIZE, 0, false, false);
4058 
4059 	ceph_osdc_start_request(osdc, req, false);
4060 	ret = ceph_osdc_wait_request(osdc, req);
4061 	if (ret >= 0) {
4062 		ret = req->r_ops[0].rval;
4063 		if (resp_page)
4064 			*resp_len = req->r_ops[0].outdata_len;
4065 	}
4066 
4067 out_put_req:
4068 	ceph_osdc_put_request(req);
4069 	return ret;
4070 }
4071 EXPORT_SYMBOL(ceph_osdc_call);
4072 
4073 /*
4074  * init, shutdown
4075  */
4076 int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
4077 {
4078 	int err;
4079 
4080 	dout("init\n");
4081 	osdc->client = client;
4082 	init_rwsem(&osdc->lock);
4083 	osdc->osds = RB_ROOT;
4084 	INIT_LIST_HEAD(&osdc->osd_lru);
4085 	spin_lock_init(&osdc->osd_lru_lock);
4086 	osd_init(&osdc->homeless_osd);
4087 	osdc->homeless_osd.o_osdc = osdc;
4088 	osdc->homeless_osd.o_osd = CEPH_HOMELESS_OSD;
4089 	osdc->last_linger_id = CEPH_LINGER_ID_START;
4090 	osdc->linger_requests = RB_ROOT;
4091 	osdc->map_checks = RB_ROOT;
4092 	osdc->linger_map_checks = RB_ROOT;
4093 	INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
4094 	INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
4095 
4096 	err = -ENOMEM;
4097 	osdc->osdmap = ceph_osdmap_alloc();
4098 	if (!osdc->osdmap)
4099 		goto out;
4100 
4101 	osdc->req_mempool = mempool_create_slab_pool(10,
4102 						     ceph_osd_request_cache);
4103 	if (!osdc->req_mempool)
4104 		goto out_map;
4105 
4106 	err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
4107 				PAGE_SIZE, 10, true, "osd_op");
4108 	if (err < 0)
4109 		goto out_mempool;
4110 	err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
4111 				PAGE_SIZE, 10, true, "osd_op_reply");
4112 	if (err < 0)
4113 		goto out_msgpool;
4114 
4115 	err = -ENOMEM;
4116 	osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
4117 	if (!osdc->notify_wq)
4118 		goto out_msgpool_reply;
4119 
4120 	schedule_delayed_work(&osdc->timeout_work,
4121 			      osdc->client->options->osd_keepalive_timeout);
4122 	schedule_delayed_work(&osdc->osds_timeout_work,
4123 	    round_jiffies_relative(osdc->client->options->osd_idle_ttl));
4124 
4125 	return 0;
4126 
4127 out_msgpool_reply:
4128 	ceph_msgpool_destroy(&osdc->msgpool_op_reply);
4129 out_msgpool:
4130 	ceph_msgpool_destroy(&osdc->msgpool_op);
4131 out_mempool:
4132 	mempool_destroy(osdc->req_mempool);
4133 out_map:
4134 	ceph_osdmap_destroy(osdc->osdmap);
4135 out:
4136 	return err;
4137 }
4138 
4139 void ceph_osdc_stop(struct ceph_osd_client *osdc)
4140 {
4141 	flush_workqueue(osdc->notify_wq);
4142 	destroy_workqueue(osdc->notify_wq);
4143 	cancel_delayed_work_sync(&osdc->timeout_work);
4144 	cancel_delayed_work_sync(&osdc->osds_timeout_work);
4145 
4146 	down_write(&osdc->lock);
4147 	while (!RB_EMPTY_ROOT(&osdc->osds)) {
4148 		struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
4149 						struct ceph_osd, o_node);
4150 		close_osd(osd);
4151 	}
4152 	up_write(&osdc->lock);
4153 	WARN_ON(atomic_read(&osdc->homeless_osd.o_ref) != 1);
4154 	osd_cleanup(&osdc->homeless_osd);
4155 
4156 	WARN_ON(!list_empty(&osdc->osd_lru));
4157 	WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_requests));
4158 	WARN_ON(!RB_EMPTY_ROOT(&osdc->map_checks));
4159 	WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_map_checks));
4160 	WARN_ON(atomic_read(&osdc->num_requests));
4161 	WARN_ON(atomic_read(&osdc->num_homeless));
4162 
4163 	ceph_osdmap_destroy(osdc->osdmap);
4164 	mempool_destroy(osdc->req_mempool);
4165 	ceph_msgpool_destroy(&osdc->msgpool_op);
4166 	ceph_msgpool_destroy(&osdc->msgpool_op_reply);
4167 }
4168 
4169 /*
4170  * Read some contiguous pages.  If we cross a stripe boundary, shorten
4171  * *plen.  Return number of bytes read, or error.
4172  */
4173 int ceph_osdc_readpages(struct ceph_osd_client *osdc,
4174 			struct ceph_vino vino, struct ceph_file_layout *layout,
4175 			u64 off, u64 *plen,
4176 			u32 truncate_seq, u64 truncate_size,
4177 			struct page **pages, int num_pages, int page_align)
4178 {
4179 	struct ceph_osd_request *req;
4180 	int rc = 0;
4181 
4182 	dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
4183 	     vino.snap, off, *plen);
4184 	req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 0, 1,
4185 				    CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
4186 				    NULL, truncate_seq, truncate_size,
4187 				    false);
4188 	if (IS_ERR(req))
4189 		return PTR_ERR(req);
4190 
4191 	/* it may be a short read due to an object boundary */
4192 	osd_req_op_extent_osd_data_pages(req, 0,
4193 				pages, *plen, page_align, false, false);
4194 
4195 	dout("readpages  final extent is %llu~%llu (%llu bytes align %d)\n",
4196 	     off, *plen, *plen, page_align);
4197 
4198 	rc = ceph_osdc_start_request(osdc, req, false);
4199 	if (!rc)
4200 		rc = ceph_osdc_wait_request(osdc, req);
4201 
4202 	ceph_osdc_put_request(req);
4203 	dout("readpages result %d\n", rc);
4204 	return rc;
4205 }
4206 EXPORT_SYMBOL(ceph_osdc_readpages);
4207 
4208 /*
4209  * do a synchronous write on N pages
4210  */
4211 int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
4212 			 struct ceph_file_layout *layout,
4213 			 struct ceph_snap_context *snapc,
4214 			 u64 off, u64 len,
4215 			 u32 truncate_seq, u64 truncate_size,
4216 			 struct timespec *mtime,
4217 			 struct page **pages, int num_pages)
4218 {
4219 	struct ceph_osd_request *req;
4220 	int rc = 0;
4221 	int page_align = off & ~PAGE_MASK;
4222 
4223 	req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 0, 1,
4224 				    CEPH_OSD_OP_WRITE,
4225 				    CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
4226 				    snapc, truncate_seq, truncate_size,
4227 				    true);
4228 	if (IS_ERR(req))
4229 		return PTR_ERR(req);
4230 
4231 	/* it may be a short write due to an object boundary */
4232 	osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align,
4233 				false, false);
4234 	dout("writepages %llu~%llu (%llu bytes)\n", off, len, len);
4235 
4236 	req->r_mtime = *mtime;
4237 	rc = ceph_osdc_start_request(osdc, req, true);
4238 	if (!rc)
4239 		rc = ceph_osdc_wait_request(osdc, req);
4240 
4241 	ceph_osdc_put_request(req);
4242 	if (rc == 0)
4243 		rc = len;
4244 	dout("writepages result %d\n", rc);
4245 	return rc;
4246 }
4247 EXPORT_SYMBOL(ceph_osdc_writepages);
4248 
4249 int ceph_osdc_setup(void)
4250 {
4251 	size_t size = sizeof(struct ceph_osd_request) +
4252 	    CEPH_OSD_SLAB_OPS * sizeof(struct ceph_osd_req_op);
4253 
4254 	BUG_ON(ceph_osd_request_cache);
4255 	ceph_osd_request_cache = kmem_cache_create("ceph_osd_request", size,
4256 						   0, 0, NULL);
4257 
4258 	return ceph_osd_request_cache ? 0 : -ENOMEM;
4259 }
4260 EXPORT_SYMBOL(ceph_osdc_setup);
4261 
4262 void ceph_osdc_cleanup(void)
4263 {
4264 	BUG_ON(!ceph_osd_request_cache);
4265 	kmem_cache_destroy(ceph_osd_request_cache);
4266 	ceph_osd_request_cache = NULL;
4267 }
4268 EXPORT_SYMBOL(ceph_osdc_cleanup);
4269 
4270 /*
4271  * handle incoming message
4272  */
4273 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
4274 {
4275 	struct ceph_osd *osd = con->private;
4276 	struct ceph_osd_client *osdc = osd->o_osdc;
4277 	int type = le16_to_cpu(msg->hdr.type);
4278 
4279 	switch (type) {
4280 	case CEPH_MSG_OSD_MAP:
4281 		ceph_osdc_handle_map(osdc, msg);
4282 		break;
4283 	case CEPH_MSG_OSD_OPREPLY:
4284 		handle_reply(osd, msg);
4285 		break;
4286 	case CEPH_MSG_WATCH_NOTIFY:
4287 		handle_watch_notify(osdc, msg);
4288 		break;
4289 
4290 	default:
4291 		pr_err("received unknown message type %d %s\n", type,
4292 		       ceph_msg_type_name(type));
4293 	}
4294 
4295 	ceph_msg_put(msg);
4296 }
4297 
4298 /*
4299  * Lookup and return message for incoming reply.  Don't try to do
4300  * anything about a larger than preallocated data portion of the
4301  * message at the moment - for now, just skip the message.
4302  */
4303 static struct ceph_msg *get_reply(struct ceph_connection *con,
4304 				  struct ceph_msg_header *hdr,
4305 				  int *skip)
4306 {
4307 	struct ceph_osd *osd = con->private;
4308 	struct ceph_osd_client *osdc = osd->o_osdc;
4309 	struct ceph_msg *m = NULL;
4310 	struct ceph_osd_request *req;
4311 	int front_len = le32_to_cpu(hdr->front_len);
4312 	int data_len = le32_to_cpu(hdr->data_len);
4313 	u64 tid = le64_to_cpu(hdr->tid);
4314 
4315 	down_read(&osdc->lock);
4316 	if (!osd_registered(osd)) {
4317 		dout("%s osd%d unknown, skipping\n", __func__, osd->o_osd);
4318 		*skip = 1;
4319 		goto out_unlock_osdc;
4320 	}
4321 	WARN_ON(osd->o_osd != le64_to_cpu(hdr->src.num));
4322 
4323 	mutex_lock(&osd->lock);
4324 	req = lookup_request(&osd->o_requests, tid);
4325 	if (!req) {
4326 		dout("%s osd%d tid %llu unknown, skipping\n", __func__,
4327 		     osd->o_osd, tid);
4328 		*skip = 1;
4329 		goto out_unlock_session;
4330 	}
4331 
4332 	ceph_msg_revoke_incoming(req->r_reply);
4333 
4334 	if (front_len > req->r_reply->front_alloc_len) {
4335 		pr_warn("%s osd%d tid %llu front %d > preallocated %d\n",
4336 			__func__, osd->o_osd, req->r_tid, front_len,
4337 			req->r_reply->front_alloc_len);
4338 		m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS,
4339 				 false);
4340 		if (!m)
4341 			goto out_unlock_session;
4342 		ceph_msg_put(req->r_reply);
4343 		req->r_reply = m;
4344 	}
4345 
4346 	if (data_len > req->r_reply->data_length) {
4347 		pr_warn("%s osd%d tid %llu data %d > preallocated %zu, skipping\n",
4348 			__func__, osd->o_osd, req->r_tid, data_len,
4349 			req->r_reply->data_length);
4350 		m = NULL;
4351 		*skip = 1;
4352 		goto out_unlock_session;
4353 	}
4354 
4355 	m = ceph_msg_get(req->r_reply);
4356 	dout("get_reply tid %lld %p\n", tid, m);
4357 
4358 out_unlock_session:
4359 	mutex_unlock(&osd->lock);
4360 out_unlock_osdc:
4361 	up_read(&osdc->lock);
4362 	return m;
4363 }
4364 
4365 /*
4366  * TODO: switch to a msg-owned pagelist
4367  */
4368 static struct ceph_msg *alloc_msg_with_page_vector(struct ceph_msg_header *hdr)
4369 {
4370 	struct ceph_msg *m;
4371 	int type = le16_to_cpu(hdr->type);
4372 	u32 front_len = le32_to_cpu(hdr->front_len);
4373 	u32 data_len = le32_to_cpu(hdr->data_len);
4374 
4375 	m = ceph_msg_new(type, front_len, GFP_NOIO, false);
4376 	if (!m)
4377 		return NULL;
4378 
4379 	if (data_len) {
4380 		struct page **pages;
4381 		struct ceph_osd_data osd_data;
4382 
4383 		pages = ceph_alloc_page_vector(calc_pages_for(0, data_len),
4384 					       GFP_NOIO);
4385 		if (IS_ERR(pages)) {
4386 			ceph_msg_put(m);
4387 			return NULL;
4388 		}
4389 
4390 		ceph_osd_data_pages_init(&osd_data, pages, data_len, 0, false,
4391 					 false);
4392 		ceph_osdc_msg_data_add(m, &osd_data);
4393 	}
4394 
4395 	return m;
4396 }
4397 
4398 static struct ceph_msg *alloc_msg(struct ceph_connection *con,
4399 				  struct ceph_msg_header *hdr,
4400 				  int *skip)
4401 {
4402 	struct ceph_osd *osd = con->private;
4403 	int type = le16_to_cpu(hdr->type);
4404 
4405 	*skip = 0;
4406 	switch (type) {
4407 	case CEPH_MSG_OSD_MAP:
4408 	case CEPH_MSG_WATCH_NOTIFY:
4409 		return alloc_msg_with_page_vector(hdr);
4410 	case CEPH_MSG_OSD_OPREPLY:
4411 		return get_reply(con, hdr, skip);
4412 	default:
4413 		pr_warn("%s osd%d unknown msg type %d, skipping\n", __func__,
4414 			osd->o_osd, type);
4415 		*skip = 1;
4416 		return NULL;
4417 	}
4418 }
4419 
4420 /*
4421  * Wrappers to refcount containing ceph_osd struct
4422  */
4423 static struct ceph_connection *get_osd_con(struct ceph_connection *con)
4424 {
4425 	struct ceph_osd *osd = con->private;
4426 	if (get_osd(osd))
4427 		return con;
4428 	return NULL;
4429 }
4430 
4431 static void put_osd_con(struct ceph_connection *con)
4432 {
4433 	struct ceph_osd *osd = con->private;
4434 	put_osd(osd);
4435 }
4436 
4437 /*
4438  * authentication
4439  */
4440 /*
4441  * Note: returned pointer is the address of a structure that's
4442  * managed separately.  Caller must *not* attempt to free it.
4443  */
4444 static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
4445 					int *proto, int force_new)
4446 {
4447 	struct ceph_osd *o = con->private;
4448 	struct ceph_osd_client *osdc = o->o_osdc;
4449 	struct ceph_auth_client *ac = osdc->client->monc.auth;
4450 	struct ceph_auth_handshake *auth = &o->o_auth;
4451 
4452 	if (force_new && auth->authorizer) {
4453 		ceph_auth_destroy_authorizer(auth->authorizer);
4454 		auth->authorizer = NULL;
4455 	}
4456 	if (!auth->authorizer) {
4457 		int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
4458 						      auth);
4459 		if (ret)
4460 			return ERR_PTR(ret);
4461 	} else {
4462 		int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
4463 						     auth);
4464 		if (ret)
4465 			return ERR_PTR(ret);
4466 	}
4467 	*proto = ac->protocol;
4468 
4469 	return auth;
4470 }
4471 
4472 
4473 static int verify_authorizer_reply(struct ceph_connection *con)
4474 {
4475 	struct ceph_osd *o = con->private;
4476 	struct ceph_osd_client *osdc = o->o_osdc;
4477 	struct ceph_auth_client *ac = osdc->client->monc.auth;
4478 
4479 	return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer);
4480 }
4481 
4482 static int invalidate_authorizer(struct ceph_connection *con)
4483 {
4484 	struct ceph_osd *o = con->private;
4485 	struct ceph_osd_client *osdc = o->o_osdc;
4486 	struct ceph_auth_client *ac = osdc->client->monc.auth;
4487 
4488 	ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
4489 	return ceph_monc_validate_auth(&osdc->client->monc);
4490 }
4491 
4492 static int osd_sign_message(struct ceph_msg *msg)
4493 {
4494 	struct ceph_osd *o = msg->con->private;
4495 	struct ceph_auth_handshake *auth = &o->o_auth;
4496 
4497 	return ceph_auth_sign_message(auth, msg);
4498 }
4499 
4500 static int osd_check_message_signature(struct ceph_msg *msg)
4501 {
4502 	struct ceph_osd *o = msg->con->private;
4503 	struct ceph_auth_handshake *auth = &o->o_auth;
4504 
4505 	return ceph_auth_check_message_signature(auth, msg);
4506 }
4507 
4508 static const struct ceph_connection_operations osd_con_ops = {
4509 	.get = get_osd_con,
4510 	.put = put_osd_con,
4511 	.dispatch = dispatch,
4512 	.get_authorizer = get_authorizer,
4513 	.verify_authorizer_reply = verify_authorizer_reply,
4514 	.invalidate_authorizer = invalidate_authorizer,
4515 	.alloc_msg = alloc_msg,
4516 	.sign_message = osd_sign_message,
4517 	.check_message_signature = osd_check_message_signature,
4518 	.fault = osd_fault,
4519 };
4520