xref: /linux/net/ceph/osd_client.c (revision 86403a92c3c5c6c395983fdbfc5e2f29dc39279b)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include <linux/ceph/ceph_debug.h>
4 
5 #include <linux/module.h>
6 #include <linux/err.h>
7 #include <linux/highmem.h>
8 #include <linux/mm.h>
9 #include <linux/pagemap.h>
10 #include <linux/slab.h>
11 #include <linux/uaccess.h>
12 #ifdef CONFIG_BLOCK
13 #include <linux/bio.h>
14 #endif
15 
16 #include <linux/ceph/ceph_features.h>
17 #include <linux/ceph/libceph.h>
18 #include <linux/ceph/osd_client.h>
19 #include <linux/ceph/messenger.h>
20 #include <linux/ceph/decode.h>
21 #include <linux/ceph/auth.h>
22 #include <linux/ceph/pagelist.h>
23 #include <linux/ceph/striper.h>
24 
25 #define OSD_OPREPLY_FRONT_LEN	512
26 
27 static struct kmem_cache	*ceph_osd_request_cache;
28 
29 static const struct ceph_connection_operations osd_con_ops;
30 
31 /*
32  * Implement client access to distributed object storage cluster.
33  *
34  * All data objects are stored within a cluster/cloud of OSDs, or
35  * "object storage devices."  (Note that Ceph OSDs have _nothing_ to
36  * do with the T10 OSD extensions to SCSI.)  Ceph OSDs are simply
37  * remote daemons serving up and coordinating consistent and safe
38  * access to storage.
39  *
40  * Cluster membership and the mapping of data objects onto storage devices
41  * are described by the osd map.
42  *
43  * We keep track of pending OSD requests (read, write), resubmit
44  * requests to different OSDs when the cluster topology/data layout
45  * change, or retry the affected requests when the communications
46  * channel with an OSD is reset.
47  */
48 
49 static void link_request(struct ceph_osd *osd, struct ceph_osd_request *req);
50 static void unlink_request(struct ceph_osd *osd, struct ceph_osd_request *req);
51 static void link_linger(struct ceph_osd *osd,
52 			struct ceph_osd_linger_request *lreq);
53 static void unlink_linger(struct ceph_osd *osd,
54 			  struct ceph_osd_linger_request *lreq);
55 static void clear_backoffs(struct ceph_osd *osd);
56 
57 #if 1
58 static inline bool rwsem_is_wrlocked(struct rw_semaphore *sem)
59 {
60 	bool wrlocked = true;
61 
62 	if (unlikely(down_read_trylock(sem))) {
63 		wrlocked = false;
64 		up_read(sem);
65 	}
66 
67 	return wrlocked;
68 }
69 static inline void verify_osdc_locked(struct ceph_osd_client *osdc)
70 {
71 	WARN_ON(!rwsem_is_locked(&osdc->lock));
72 }
73 static inline void verify_osdc_wrlocked(struct ceph_osd_client *osdc)
74 {
75 	WARN_ON(!rwsem_is_wrlocked(&osdc->lock));
76 }
77 static inline void verify_osd_locked(struct ceph_osd *osd)
78 {
79 	struct ceph_osd_client *osdc = osd->o_osdc;
80 
81 	WARN_ON(!(mutex_is_locked(&osd->lock) &&
82 		  rwsem_is_locked(&osdc->lock)) &&
83 		!rwsem_is_wrlocked(&osdc->lock));
84 }
85 static inline void verify_lreq_locked(struct ceph_osd_linger_request *lreq)
86 {
87 	WARN_ON(!mutex_is_locked(&lreq->lock));
88 }
89 #else
90 static inline void verify_osdc_locked(struct ceph_osd_client *osdc) { }
91 static inline void verify_osdc_wrlocked(struct ceph_osd_client *osdc) { }
92 static inline void verify_osd_locked(struct ceph_osd *osd) { }
93 static inline void verify_lreq_locked(struct ceph_osd_linger_request *lreq) { }
94 #endif
95 
96 /*
97  * calculate the mapping of a file extent onto an object, and fill out the
98  * request accordingly.  shorten extent as necessary if it crosses an
99  * object boundary.
100  *
101  * fill osd op in request message.
102  */
103 static int calc_layout(struct ceph_file_layout *layout, u64 off, u64 *plen,
104 			u64 *objnum, u64 *objoff, u64 *objlen)
105 {
106 	u64 orig_len = *plen;
107 	u32 xlen;
108 
109 	/* object extent? */
110 	ceph_calc_file_object_mapping(layout, off, orig_len, objnum,
111 					  objoff, &xlen);
112 	*objlen = xlen;
113 	if (*objlen < orig_len) {
114 		*plen = *objlen;
115 		dout(" skipping last %llu, final file extent %llu~%llu\n",
116 		     orig_len - *plen, off, *plen);
117 	}
118 
119 	dout("calc_layout objnum=%llx %llu~%llu\n", *objnum, *objoff, *objlen);
120 	return 0;
121 }
122 
123 static void ceph_osd_data_init(struct ceph_osd_data *osd_data)
124 {
125 	memset(osd_data, 0, sizeof (*osd_data));
126 	osd_data->type = CEPH_OSD_DATA_TYPE_NONE;
127 }
128 
129 /*
130  * Consumes @pages if @own_pages is true.
131  */
132 static void ceph_osd_data_pages_init(struct ceph_osd_data *osd_data,
133 			struct page **pages, u64 length, u32 alignment,
134 			bool pages_from_pool, bool own_pages)
135 {
136 	osd_data->type = CEPH_OSD_DATA_TYPE_PAGES;
137 	osd_data->pages = pages;
138 	osd_data->length = length;
139 	osd_data->alignment = alignment;
140 	osd_data->pages_from_pool = pages_from_pool;
141 	osd_data->own_pages = own_pages;
142 }
143 
144 /*
145  * Consumes a ref on @pagelist.
146  */
147 static void ceph_osd_data_pagelist_init(struct ceph_osd_data *osd_data,
148 			struct ceph_pagelist *pagelist)
149 {
150 	osd_data->type = CEPH_OSD_DATA_TYPE_PAGELIST;
151 	osd_data->pagelist = pagelist;
152 }
153 
154 #ifdef CONFIG_BLOCK
155 static void ceph_osd_data_bio_init(struct ceph_osd_data *osd_data,
156 				   struct ceph_bio_iter *bio_pos,
157 				   u32 bio_length)
158 {
159 	osd_data->type = CEPH_OSD_DATA_TYPE_BIO;
160 	osd_data->bio_pos = *bio_pos;
161 	osd_data->bio_length = bio_length;
162 }
163 #endif /* CONFIG_BLOCK */
164 
165 static void ceph_osd_data_bvecs_init(struct ceph_osd_data *osd_data,
166 				     struct ceph_bvec_iter *bvec_pos,
167 				     u32 num_bvecs)
168 {
169 	osd_data->type = CEPH_OSD_DATA_TYPE_BVECS;
170 	osd_data->bvec_pos = *bvec_pos;
171 	osd_data->num_bvecs = num_bvecs;
172 }
173 
174 static struct ceph_osd_data *
175 osd_req_op_raw_data_in(struct ceph_osd_request *osd_req, unsigned int which)
176 {
177 	BUG_ON(which >= osd_req->r_num_ops);
178 
179 	return &osd_req->r_ops[which].raw_data_in;
180 }
181 
182 struct ceph_osd_data *
183 osd_req_op_extent_osd_data(struct ceph_osd_request *osd_req,
184 			unsigned int which)
185 {
186 	return osd_req_op_data(osd_req, which, extent, osd_data);
187 }
188 EXPORT_SYMBOL(osd_req_op_extent_osd_data);
189 
190 void osd_req_op_raw_data_in_pages(struct ceph_osd_request *osd_req,
191 			unsigned int which, struct page **pages,
192 			u64 length, u32 alignment,
193 			bool pages_from_pool, bool own_pages)
194 {
195 	struct ceph_osd_data *osd_data;
196 
197 	osd_data = osd_req_op_raw_data_in(osd_req, which);
198 	ceph_osd_data_pages_init(osd_data, pages, length, alignment,
199 				pages_from_pool, own_pages);
200 }
201 EXPORT_SYMBOL(osd_req_op_raw_data_in_pages);
202 
203 void osd_req_op_extent_osd_data_pages(struct ceph_osd_request *osd_req,
204 			unsigned int which, struct page **pages,
205 			u64 length, u32 alignment,
206 			bool pages_from_pool, bool own_pages)
207 {
208 	struct ceph_osd_data *osd_data;
209 
210 	osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
211 	ceph_osd_data_pages_init(osd_data, pages, length, alignment,
212 				pages_from_pool, own_pages);
213 }
214 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pages);
215 
216 void osd_req_op_extent_osd_data_pagelist(struct ceph_osd_request *osd_req,
217 			unsigned int which, struct ceph_pagelist *pagelist)
218 {
219 	struct ceph_osd_data *osd_data;
220 
221 	osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
222 	ceph_osd_data_pagelist_init(osd_data, pagelist);
223 }
224 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pagelist);
225 
226 #ifdef CONFIG_BLOCK
227 void osd_req_op_extent_osd_data_bio(struct ceph_osd_request *osd_req,
228 				    unsigned int which,
229 				    struct ceph_bio_iter *bio_pos,
230 				    u32 bio_length)
231 {
232 	struct ceph_osd_data *osd_data;
233 
234 	osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
235 	ceph_osd_data_bio_init(osd_data, bio_pos, bio_length);
236 }
237 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bio);
238 #endif /* CONFIG_BLOCK */
239 
240 void osd_req_op_extent_osd_data_bvecs(struct ceph_osd_request *osd_req,
241 				      unsigned int which,
242 				      struct bio_vec *bvecs, u32 num_bvecs,
243 				      u32 bytes)
244 {
245 	struct ceph_osd_data *osd_data;
246 	struct ceph_bvec_iter it = {
247 		.bvecs = bvecs,
248 		.iter = { .bi_size = bytes },
249 	};
250 
251 	osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
252 	ceph_osd_data_bvecs_init(osd_data, &it, num_bvecs);
253 }
254 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bvecs);
255 
256 void osd_req_op_extent_osd_data_bvec_pos(struct ceph_osd_request *osd_req,
257 					 unsigned int which,
258 					 struct ceph_bvec_iter *bvec_pos)
259 {
260 	struct ceph_osd_data *osd_data;
261 
262 	osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
263 	ceph_osd_data_bvecs_init(osd_data, bvec_pos, 0);
264 }
265 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bvec_pos);
266 
267 static void osd_req_op_cls_request_info_pagelist(
268 			struct ceph_osd_request *osd_req,
269 			unsigned int which, struct ceph_pagelist *pagelist)
270 {
271 	struct ceph_osd_data *osd_data;
272 
273 	osd_data = osd_req_op_data(osd_req, which, cls, request_info);
274 	ceph_osd_data_pagelist_init(osd_data, pagelist);
275 }
276 
277 void osd_req_op_cls_request_data_pagelist(
278 			struct ceph_osd_request *osd_req,
279 			unsigned int which, struct ceph_pagelist *pagelist)
280 {
281 	struct ceph_osd_data *osd_data;
282 
283 	osd_data = osd_req_op_data(osd_req, which, cls, request_data);
284 	ceph_osd_data_pagelist_init(osd_data, pagelist);
285 	osd_req->r_ops[which].cls.indata_len += pagelist->length;
286 	osd_req->r_ops[which].indata_len += pagelist->length;
287 }
288 EXPORT_SYMBOL(osd_req_op_cls_request_data_pagelist);
289 
290 void osd_req_op_cls_request_data_pages(struct ceph_osd_request *osd_req,
291 			unsigned int which, struct page **pages, u64 length,
292 			u32 alignment, bool pages_from_pool, bool own_pages)
293 {
294 	struct ceph_osd_data *osd_data;
295 
296 	osd_data = osd_req_op_data(osd_req, which, cls, request_data);
297 	ceph_osd_data_pages_init(osd_data, pages, length, alignment,
298 				pages_from_pool, own_pages);
299 	osd_req->r_ops[which].cls.indata_len += length;
300 	osd_req->r_ops[which].indata_len += length;
301 }
302 EXPORT_SYMBOL(osd_req_op_cls_request_data_pages);
303 
304 void osd_req_op_cls_request_data_bvecs(struct ceph_osd_request *osd_req,
305 				       unsigned int which,
306 				       struct bio_vec *bvecs, u32 num_bvecs,
307 				       u32 bytes)
308 {
309 	struct ceph_osd_data *osd_data;
310 	struct ceph_bvec_iter it = {
311 		.bvecs = bvecs,
312 		.iter = { .bi_size = bytes },
313 	};
314 
315 	osd_data = osd_req_op_data(osd_req, which, cls, request_data);
316 	ceph_osd_data_bvecs_init(osd_data, &it, num_bvecs);
317 	osd_req->r_ops[which].cls.indata_len += bytes;
318 	osd_req->r_ops[which].indata_len += bytes;
319 }
320 EXPORT_SYMBOL(osd_req_op_cls_request_data_bvecs);
321 
322 void osd_req_op_cls_response_data_pages(struct ceph_osd_request *osd_req,
323 			unsigned int which, struct page **pages, u64 length,
324 			u32 alignment, bool pages_from_pool, bool own_pages)
325 {
326 	struct ceph_osd_data *osd_data;
327 
328 	osd_data = osd_req_op_data(osd_req, which, cls, response_data);
329 	ceph_osd_data_pages_init(osd_data, pages, length, alignment,
330 				pages_from_pool, own_pages);
331 }
332 EXPORT_SYMBOL(osd_req_op_cls_response_data_pages);
333 
334 static u64 ceph_osd_data_length(struct ceph_osd_data *osd_data)
335 {
336 	switch (osd_data->type) {
337 	case CEPH_OSD_DATA_TYPE_NONE:
338 		return 0;
339 	case CEPH_OSD_DATA_TYPE_PAGES:
340 		return osd_data->length;
341 	case CEPH_OSD_DATA_TYPE_PAGELIST:
342 		return (u64)osd_data->pagelist->length;
343 #ifdef CONFIG_BLOCK
344 	case CEPH_OSD_DATA_TYPE_BIO:
345 		return (u64)osd_data->bio_length;
346 #endif /* CONFIG_BLOCK */
347 	case CEPH_OSD_DATA_TYPE_BVECS:
348 		return osd_data->bvec_pos.iter.bi_size;
349 	default:
350 		WARN(true, "unrecognized data type %d\n", (int)osd_data->type);
351 		return 0;
352 	}
353 }
354 
355 static void ceph_osd_data_release(struct ceph_osd_data *osd_data)
356 {
357 	if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES && osd_data->own_pages) {
358 		int num_pages;
359 
360 		num_pages = calc_pages_for((u64)osd_data->alignment,
361 						(u64)osd_data->length);
362 		ceph_release_page_vector(osd_data->pages, num_pages);
363 	} else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
364 		ceph_pagelist_release(osd_data->pagelist);
365 	}
366 	ceph_osd_data_init(osd_data);
367 }
368 
369 static void osd_req_op_data_release(struct ceph_osd_request *osd_req,
370 			unsigned int which)
371 {
372 	struct ceph_osd_req_op *op;
373 
374 	BUG_ON(which >= osd_req->r_num_ops);
375 	op = &osd_req->r_ops[which];
376 
377 	switch (op->op) {
378 	case CEPH_OSD_OP_READ:
379 	case CEPH_OSD_OP_WRITE:
380 	case CEPH_OSD_OP_WRITEFULL:
381 		ceph_osd_data_release(&op->extent.osd_data);
382 		break;
383 	case CEPH_OSD_OP_CALL:
384 		ceph_osd_data_release(&op->cls.request_info);
385 		ceph_osd_data_release(&op->cls.request_data);
386 		ceph_osd_data_release(&op->cls.response_data);
387 		break;
388 	case CEPH_OSD_OP_SETXATTR:
389 	case CEPH_OSD_OP_CMPXATTR:
390 		ceph_osd_data_release(&op->xattr.osd_data);
391 		break;
392 	case CEPH_OSD_OP_STAT:
393 		ceph_osd_data_release(&op->raw_data_in);
394 		break;
395 	case CEPH_OSD_OP_NOTIFY_ACK:
396 		ceph_osd_data_release(&op->notify_ack.request_data);
397 		break;
398 	case CEPH_OSD_OP_NOTIFY:
399 		ceph_osd_data_release(&op->notify.request_data);
400 		ceph_osd_data_release(&op->notify.response_data);
401 		break;
402 	case CEPH_OSD_OP_LIST_WATCHERS:
403 		ceph_osd_data_release(&op->list_watchers.response_data);
404 		break;
405 	case CEPH_OSD_OP_COPY_FROM2:
406 		ceph_osd_data_release(&op->copy_from.osd_data);
407 		break;
408 	default:
409 		break;
410 	}
411 }
412 
413 /*
414  * Assumes @t is zero-initialized.
415  */
416 static void target_init(struct ceph_osd_request_target *t)
417 {
418 	ceph_oid_init(&t->base_oid);
419 	ceph_oloc_init(&t->base_oloc);
420 	ceph_oid_init(&t->target_oid);
421 	ceph_oloc_init(&t->target_oloc);
422 
423 	ceph_osds_init(&t->acting);
424 	ceph_osds_init(&t->up);
425 	t->size = -1;
426 	t->min_size = -1;
427 
428 	t->osd = CEPH_HOMELESS_OSD;
429 }
430 
431 static void target_copy(struct ceph_osd_request_target *dest,
432 			const struct ceph_osd_request_target *src)
433 {
434 	ceph_oid_copy(&dest->base_oid, &src->base_oid);
435 	ceph_oloc_copy(&dest->base_oloc, &src->base_oloc);
436 	ceph_oid_copy(&dest->target_oid, &src->target_oid);
437 	ceph_oloc_copy(&dest->target_oloc, &src->target_oloc);
438 
439 	dest->pgid = src->pgid; /* struct */
440 	dest->spgid = src->spgid; /* struct */
441 	dest->pg_num = src->pg_num;
442 	dest->pg_num_mask = src->pg_num_mask;
443 	ceph_osds_copy(&dest->acting, &src->acting);
444 	ceph_osds_copy(&dest->up, &src->up);
445 	dest->size = src->size;
446 	dest->min_size = src->min_size;
447 	dest->sort_bitwise = src->sort_bitwise;
448 
449 	dest->flags = src->flags;
450 	dest->paused = src->paused;
451 
452 	dest->epoch = src->epoch;
453 	dest->last_force_resend = src->last_force_resend;
454 
455 	dest->osd = src->osd;
456 }
457 
458 static void target_destroy(struct ceph_osd_request_target *t)
459 {
460 	ceph_oid_destroy(&t->base_oid);
461 	ceph_oloc_destroy(&t->base_oloc);
462 	ceph_oid_destroy(&t->target_oid);
463 	ceph_oloc_destroy(&t->target_oloc);
464 }
465 
466 /*
467  * requests
468  */
469 static void request_release_checks(struct ceph_osd_request *req)
470 {
471 	WARN_ON(!RB_EMPTY_NODE(&req->r_node));
472 	WARN_ON(!RB_EMPTY_NODE(&req->r_mc_node));
473 	WARN_ON(!list_empty(&req->r_private_item));
474 	WARN_ON(req->r_osd);
475 }
476 
477 static void ceph_osdc_release_request(struct kref *kref)
478 {
479 	struct ceph_osd_request *req = container_of(kref,
480 					    struct ceph_osd_request, r_kref);
481 	unsigned int which;
482 
483 	dout("%s %p (r_request %p r_reply %p)\n", __func__, req,
484 	     req->r_request, req->r_reply);
485 	request_release_checks(req);
486 
487 	if (req->r_request)
488 		ceph_msg_put(req->r_request);
489 	if (req->r_reply)
490 		ceph_msg_put(req->r_reply);
491 
492 	for (which = 0; which < req->r_num_ops; which++)
493 		osd_req_op_data_release(req, which);
494 
495 	target_destroy(&req->r_t);
496 	ceph_put_snap_context(req->r_snapc);
497 
498 	if (req->r_mempool)
499 		mempool_free(req, req->r_osdc->req_mempool);
500 	else if (req->r_num_ops <= CEPH_OSD_SLAB_OPS)
501 		kmem_cache_free(ceph_osd_request_cache, req);
502 	else
503 		kfree(req);
504 }
505 
506 void ceph_osdc_get_request(struct ceph_osd_request *req)
507 {
508 	dout("%s %p (was %d)\n", __func__, req,
509 	     kref_read(&req->r_kref));
510 	kref_get(&req->r_kref);
511 }
512 EXPORT_SYMBOL(ceph_osdc_get_request);
513 
514 void ceph_osdc_put_request(struct ceph_osd_request *req)
515 {
516 	if (req) {
517 		dout("%s %p (was %d)\n", __func__, req,
518 		     kref_read(&req->r_kref));
519 		kref_put(&req->r_kref, ceph_osdc_release_request);
520 	}
521 }
522 EXPORT_SYMBOL(ceph_osdc_put_request);
523 
524 static void request_init(struct ceph_osd_request *req)
525 {
526 	/* req only, each op is zeroed in _osd_req_op_init() */
527 	memset(req, 0, sizeof(*req));
528 
529 	kref_init(&req->r_kref);
530 	init_completion(&req->r_completion);
531 	RB_CLEAR_NODE(&req->r_node);
532 	RB_CLEAR_NODE(&req->r_mc_node);
533 	INIT_LIST_HEAD(&req->r_private_item);
534 
535 	target_init(&req->r_t);
536 }
537 
538 /*
539  * This is ugly, but it allows us to reuse linger registration and ping
540  * requests, keeping the structure of the code around send_linger{_ping}()
541  * reasonable.  Setting up a min_nr=2 mempool for each linger request
542  * and dealing with copying ops (this blasts req only, watch op remains
543  * intact) isn't any better.
544  */
545 static void request_reinit(struct ceph_osd_request *req)
546 {
547 	struct ceph_osd_client *osdc = req->r_osdc;
548 	bool mempool = req->r_mempool;
549 	unsigned int num_ops = req->r_num_ops;
550 	u64 snapid = req->r_snapid;
551 	struct ceph_snap_context *snapc = req->r_snapc;
552 	bool linger = req->r_linger;
553 	struct ceph_msg *request_msg = req->r_request;
554 	struct ceph_msg *reply_msg = req->r_reply;
555 
556 	dout("%s req %p\n", __func__, req);
557 	WARN_ON(kref_read(&req->r_kref) != 1);
558 	request_release_checks(req);
559 
560 	WARN_ON(kref_read(&request_msg->kref) != 1);
561 	WARN_ON(kref_read(&reply_msg->kref) != 1);
562 	target_destroy(&req->r_t);
563 
564 	request_init(req);
565 	req->r_osdc = osdc;
566 	req->r_mempool = mempool;
567 	req->r_num_ops = num_ops;
568 	req->r_snapid = snapid;
569 	req->r_snapc = snapc;
570 	req->r_linger = linger;
571 	req->r_request = request_msg;
572 	req->r_reply = reply_msg;
573 }
574 
575 struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
576 					       struct ceph_snap_context *snapc,
577 					       unsigned int num_ops,
578 					       bool use_mempool,
579 					       gfp_t gfp_flags)
580 {
581 	struct ceph_osd_request *req;
582 
583 	if (use_mempool) {
584 		BUG_ON(num_ops > CEPH_OSD_SLAB_OPS);
585 		req = mempool_alloc(osdc->req_mempool, gfp_flags);
586 	} else if (num_ops <= CEPH_OSD_SLAB_OPS) {
587 		req = kmem_cache_alloc(ceph_osd_request_cache, gfp_flags);
588 	} else {
589 		BUG_ON(num_ops > CEPH_OSD_MAX_OPS);
590 		req = kmalloc(struct_size(req, r_ops, num_ops), gfp_flags);
591 	}
592 	if (unlikely(!req))
593 		return NULL;
594 
595 	request_init(req);
596 	req->r_osdc = osdc;
597 	req->r_mempool = use_mempool;
598 	req->r_num_ops = num_ops;
599 	req->r_snapid = CEPH_NOSNAP;
600 	req->r_snapc = ceph_get_snap_context(snapc);
601 
602 	dout("%s req %p\n", __func__, req);
603 	return req;
604 }
605 EXPORT_SYMBOL(ceph_osdc_alloc_request);
606 
607 static int ceph_oloc_encoding_size(const struct ceph_object_locator *oloc)
608 {
609 	return 8 + 4 + 4 + 4 + (oloc->pool_ns ? oloc->pool_ns->len : 0);
610 }
611 
612 static int __ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp,
613 				      int num_request_data_items,
614 				      int num_reply_data_items)
615 {
616 	struct ceph_osd_client *osdc = req->r_osdc;
617 	struct ceph_msg *msg;
618 	int msg_size;
619 
620 	WARN_ON(req->r_request || req->r_reply);
621 	WARN_ON(ceph_oid_empty(&req->r_base_oid));
622 	WARN_ON(ceph_oloc_empty(&req->r_base_oloc));
623 
624 	/* create request message */
625 	msg_size = CEPH_ENCODING_START_BLK_LEN +
626 			CEPH_PGID_ENCODING_LEN + 1; /* spgid */
627 	msg_size += 4 + 4 + 4; /* hash, osdmap_epoch, flags */
628 	msg_size += CEPH_ENCODING_START_BLK_LEN +
629 			sizeof(struct ceph_osd_reqid); /* reqid */
630 	msg_size += sizeof(struct ceph_blkin_trace_info); /* trace */
631 	msg_size += 4 + sizeof(struct ceph_timespec); /* client_inc, mtime */
632 	msg_size += CEPH_ENCODING_START_BLK_LEN +
633 			ceph_oloc_encoding_size(&req->r_base_oloc); /* oloc */
634 	msg_size += 4 + req->r_base_oid.name_len; /* oid */
635 	msg_size += 2 + req->r_num_ops * sizeof(struct ceph_osd_op);
636 	msg_size += 8; /* snapid */
637 	msg_size += 8; /* snap_seq */
638 	msg_size += 4 + 8 * (req->r_snapc ? req->r_snapc->num_snaps : 0);
639 	msg_size += 4 + 8; /* retry_attempt, features */
640 
641 	if (req->r_mempool)
642 		msg = ceph_msgpool_get(&osdc->msgpool_op, msg_size,
643 				       num_request_data_items);
644 	else
645 		msg = ceph_msg_new2(CEPH_MSG_OSD_OP, msg_size,
646 				    num_request_data_items, gfp, true);
647 	if (!msg)
648 		return -ENOMEM;
649 
650 	memset(msg->front.iov_base, 0, msg->front.iov_len);
651 	req->r_request = msg;
652 
653 	/* create reply message */
654 	msg_size = OSD_OPREPLY_FRONT_LEN;
655 	msg_size += req->r_base_oid.name_len;
656 	msg_size += req->r_num_ops * sizeof(struct ceph_osd_op);
657 
658 	if (req->r_mempool)
659 		msg = ceph_msgpool_get(&osdc->msgpool_op_reply, msg_size,
660 				       num_reply_data_items);
661 	else
662 		msg = ceph_msg_new2(CEPH_MSG_OSD_OPREPLY, msg_size,
663 				    num_reply_data_items, gfp, true);
664 	if (!msg)
665 		return -ENOMEM;
666 
667 	req->r_reply = msg;
668 
669 	return 0;
670 }
671 
672 static bool osd_req_opcode_valid(u16 opcode)
673 {
674 	switch (opcode) {
675 #define GENERATE_CASE(op, opcode, str)	case CEPH_OSD_OP_##op: return true;
676 __CEPH_FORALL_OSD_OPS(GENERATE_CASE)
677 #undef GENERATE_CASE
678 	default:
679 		return false;
680 	}
681 }
682 
683 static void get_num_data_items(struct ceph_osd_request *req,
684 			       int *num_request_data_items,
685 			       int *num_reply_data_items)
686 {
687 	struct ceph_osd_req_op *op;
688 
689 	*num_request_data_items = 0;
690 	*num_reply_data_items = 0;
691 
692 	for (op = req->r_ops; op != &req->r_ops[req->r_num_ops]; op++) {
693 		switch (op->op) {
694 		/* request */
695 		case CEPH_OSD_OP_WRITE:
696 		case CEPH_OSD_OP_WRITEFULL:
697 		case CEPH_OSD_OP_SETXATTR:
698 		case CEPH_OSD_OP_CMPXATTR:
699 		case CEPH_OSD_OP_NOTIFY_ACK:
700 		case CEPH_OSD_OP_COPY_FROM2:
701 			*num_request_data_items += 1;
702 			break;
703 
704 		/* reply */
705 		case CEPH_OSD_OP_STAT:
706 		case CEPH_OSD_OP_READ:
707 		case CEPH_OSD_OP_LIST_WATCHERS:
708 			*num_reply_data_items += 1;
709 			break;
710 
711 		/* both */
712 		case CEPH_OSD_OP_NOTIFY:
713 			*num_request_data_items += 1;
714 			*num_reply_data_items += 1;
715 			break;
716 		case CEPH_OSD_OP_CALL:
717 			*num_request_data_items += 2;
718 			*num_reply_data_items += 1;
719 			break;
720 
721 		default:
722 			WARN_ON(!osd_req_opcode_valid(op->op));
723 			break;
724 		}
725 	}
726 }
727 
728 /*
729  * oid, oloc and OSD op opcode(s) must be filled in before this function
730  * is called.
731  */
732 int ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp)
733 {
734 	int num_request_data_items, num_reply_data_items;
735 
736 	get_num_data_items(req, &num_request_data_items, &num_reply_data_items);
737 	return __ceph_osdc_alloc_messages(req, gfp, num_request_data_items,
738 					  num_reply_data_items);
739 }
740 EXPORT_SYMBOL(ceph_osdc_alloc_messages);
741 
742 /*
743  * This is an osd op init function for opcodes that have no data or
744  * other information associated with them.  It also serves as a
745  * common init routine for all the other init functions, below.
746  */
747 static struct ceph_osd_req_op *
748 _osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which,
749 		 u16 opcode, u32 flags)
750 {
751 	struct ceph_osd_req_op *op;
752 
753 	BUG_ON(which >= osd_req->r_num_ops);
754 	BUG_ON(!osd_req_opcode_valid(opcode));
755 
756 	op = &osd_req->r_ops[which];
757 	memset(op, 0, sizeof (*op));
758 	op->op = opcode;
759 	op->flags = flags;
760 
761 	return op;
762 }
763 
764 void osd_req_op_init(struct ceph_osd_request *osd_req,
765 		     unsigned int which, u16 opcode, u32 flags)
766 {
767 	(void)_osd_req_op_init(osd_req, which, opcode, flags);
768 }
769 EXPORT_SYMBOL(osd_req_op_init);
770 
771 void osd_req_op_extent_init(struct ceph_osd_request *osd_req,
772 				unsigned int which, u16 opcode,
773 				u64 offset, u64 length,
774 				u64 truncate_size, u32 truncate_seq)
775 {
776 	struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
777 						      opcode, 0);
778 	size_t payload_len = 0;
779 
780 	BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
781 	       opcode != CEPH_OSD_OP_WRITEFULL && opcode != CEPH_OSD_OP_ZERO &&
782 	       opcode != CEPH_OSD_OP_TRUNCATE);
783 
784 	op->extent.offset = offset;
785 	op->extent.length = length;
786 	op->extent.truncate_size = truncate_size;
787 	op->extent.truncate_seq = truncate_seq;
788 	if (opcode == CEPH_OSD_OP_WRITE || opcode == CEPH_OSD_OP_WRITEFULL)
789 		payload_len += length;
790 
791 	op->indata_len = payload_len;
792 }
793 EXPORT_SYMBOL(osd_req_op_extent_init);
794 
795 void osd_req_op_extent_update(struct ceph_osd_request *osd_req,
796 				unsigned int which, u64 length)
797 {
798 	struct ceph_osd_req_op *op;
799 	u64 previous;
800 
801 	BUG_ON(which >= osd_req->r_num_ops);
802 	op = &osd_req->r_ops[which];
803 	previous = op->extent.length;
804 
805 	if (length == previous)
806 		return;		/* Nothing to do */
807 	BUG_ON(length > previous);
808 
809 	op->extent.length = length;
810 	if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL)
811 		op->indata_len -= previous - length;
812 }
813 EXPORT_SYMBOL(osd_req_op_extent_update);
814 
815 void osd_req_op_extent_dup_last(struct ceph_osd_request *osd_req,
816 				unsigned int which, u64 offset_inc)
817 {
818 	struct ceph_osd_req_op *op, *prev_op;
819 
820 	BUG_ON(which + 1 >= osd_req->r_num_ops);
821 
822 	prev_op = &osd_req->r_ops[which];
823 	op = _osd_req_op_init(osd_req, which + 1, prev_op->op, prev_op->flags);
824 	/* dup previous one */
825 	op->indata_len = prev_op->indata_len;
826 	op->outdata_len = prev_op->outdata_len;
827 	op->extent = prev_op->extent;
828 	/* adjust offset */
829 	op->extent.offset += offset_inc;
830 	op->extent.length -= offset_inc;
831 
832 	if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL)
833 		op->indata_len -= offset_inc;
834 }
835 EXPORT_SYMBOL(osd_req_op_extent_dup_last);
836 
837 int osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
838 			const char *class, const char *method)
839 {
840 	struct ceph_osd_req_op *op;
841 	struct ceph_pagelist *pagelist;
842 	size_t payload_len = 0;
843 	size_t size;
844 	int ret;
845 
846 	op = _osd_req_op_init(osd_req, which, CEPH_OSD_OP_CALL, 0);
847 
848 	pagelist = ceph_pagelist_alloc(GFP_NOFS);
849 	if (!pagelist)
850 		return -ENOMEM;
851 
852 	op->cls.class_name = class;
853 	size = strlen(class);
854 	BUG_ON(size > (size_t) U8_MAX);
855 	op->cls.class_len = size;
856 	ret = ceph_pagelist_append(pagelist, class, size);
857 	if (ret)
858 		goto err_pagelist_free;
859 	payload_len += size;
860 
861 	op->cls.method_name = method;
862 	size = strlen(method);
863 	BUG_ON(size > (size_t) U8_MAX);
864 	op->cls.method_len = size;
865 	ret = ceph_pagelist_append(pagelist, method, size);
866 	if (ret)
867 		goto err_pagelist_free;
868 	payload_len += size;
869 
870 	osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist);
871 	op->indata_len = payload_len;
872 	return 0;
873 
874 err_pagelist_free:
875 	ceph_pagelist_release(pagelist);
876 	return ret;
877 }
878 EXPORT_SYMBOL(osd_req_op_cls_init);
879 
880 int osd_req_op_xattr_init(struct ceph_osd_request *osd_req, unsigned int which,
881 			  u16 opcode, const char *name, const void *value,
882 			  size_t size, u8 cmp_op, u8 cmp_mode)
883 {
884 	struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
885 						      opcode, 0);
886 	struct ceph_pagelist *pagelist;
887 	size_t payload_len;
888 	int ret;
889 
890 	BUG_ON(opcode != CEPH_OSD_OP_SETXATTR && opcode != CEPH_OSD_OP_CMPXATTR);
891 
892 	pagelist = ceph_pagelist_alloc(GFP_NOFS);
893 	if (!pagelist)
894 		return -ENOMEM;
895 
896 	payload_len = strlen(name);
897 	op->xattr.name_len = payload_len;
898 	ret = ceph_pagelist_append(pagelist, name, payload_len);
899 	if (ret)
900 		goto err_pagelist_free;
901 
902 	op->xattr.value_len = size;
903 	ret = ceph_pagelist_append(pagelist, value, size);
904 	if (ret)
905 		goto err_pagelist_free;
906 	payload_len += size;
907 
908 	op->xattr.cmp_op = cmp_op;
909 	op->xattr.cmp_mode = cmp_mode;
910 
911 	ceph_osd_data_pagelist_init(&op->xattr.osd_data, pagelist);
912 	op->indata_len = payload_len;
913 	return 0;
914 
915 err_pagelist_free:
916 	ceph_pagelist_release(pagelist);
917 	return ret;
918 }
919 EXPORT_SYMBOL(osd_req_op_xattr_init);
920 
921 /*
922  * @watch_opcode: CEPH_OSD_WATCH_OP_*
923  */
924 static void osd_req_op_watch_init(struct ceph_osd_request *req, int which,
925 				  u64 cookie, u8 watch_opcode)
926 {
927 	struct ceph_osd_req_op *op;
928 
929 	op = _osd_req_op_init(req, which, CEPH_OSD_OP_WATCH, 0);
930 	op->watch.cookie = cookie;
931 	op->watch.op = watch_opcode;
932 	op->watch.gen = 0;
933 }
934 
935 void osd_req_op_alloc_hint_init(struct ceph_osd_request *osd_req,
936 				unsigned int which,
937 				u64 expected_object_size,
938 				u64 expected_write_size)
939 {
940 	struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
941 						      CEPH_OSD_OP_SETALLOCHINT,
942 						      0);
943 
944 	op->alloc_hint.expected_object_size = expected_object_size;
945 	op->alloc_hint.expected_write_size = expected_write_size;
946 
947 	/*
948 	 * CEPH_OSD_OP_SETALLOCHINT op is advisory and therefore deemed
949 	 * not worth a feature bit.  Set FAILOK per-op flag to make
950 	 * sure older osds don't trip over an unsupported opcode.
951 	 */
952 	op->flags |= CEPH_OSD_OP_FLAG_FAILOK;
953 }
954 EXPORT_SYMBOL(osd_req_op_alloc_hint_init);
955 
956 static void ceph_osdc_msg_data_add(struct ceph_msg *msg,
957 				struct ceph_osd_data *osd_data)
958 {
959 	u64 length = ceph_osd_data_length(osd_data);
960 
961 	if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
962 		BUG_ON(length > (u64) SIZE_MAX);
963 		if (length)
964 			ceph_msg_data_add_pages(msg, osd_data->pages,
965 					length, osd_data->alignment, false);
966 	} else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
967 		BUG_ON(!length);
968 		ceph_msg_data_add_pagelist(msg, osd_data->pagelist);
969 #ifdef CONFIG_BLOCK
970 	} else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
971 		ceph_msg_data_add_bio(msg, &osd_data->bio_pos, length);
972 #endif
973 	} else if (osd_data->type == CEPH_OSD_DATA_TYPE_BVECS) {
974 		ceph_msg_data_add_bvecs(msg, &osd_data->bvec_pos);
975 	} else {
976 		BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
977 	}
978 }
979 
980 static u32 osd_req_encode_op(struct ceph_osd_op *dst,
981 			     const struct ceph_osd_req_op *src)
982 {
983 	switch (src->op) {
984 	case CEPH_OSD_OP_STAT:
985 		break;
986 	case CEPH_OSD_OP_READ:
987 	case CEPH_OSD_OP_WRITE:
988 	case CEPH_OSD_OP_WRITEFULL:
989 	case CEPH_OSD_OP_ZERO:
990 	case CEPH_OSD_OP_TRUNCATE:
991 		dst->extent.offset = cpu_to_le64(src->extent.offset);
992 		dst->extent.length = cpu_to_le64(src->extent.length);
993 		dst->extent.truncate_size =
994 			cpu_to_le64(src->extent.truncate_size);
995 		dst->extent.truncate_seq =
996 			cpu_to_le32(src->extent.truncate_seq);
997 		break;
998 	case CEPH_OSD_OP_CALL:
999 		dst->cls.class_len = src->cls.class_len;
1000 		dst->cls.method_len = src->cls.method_len;
1001 		dst->cls.indata_len = cpu_to_le32(src->cls.indata_len);
1002 		break;
1003 	case CEPH_OSD_OP_WATCH:
1004 		dst->watch.cookie = cpu_to_le64(src->watch.cookie);
1005 		dst->watch.ver = cpu_to_le64(0);
1006 		dst->watch.op = src->watch.op;
1007 		dst->watch.gen = cpu_to_le32(src->watch.gen);
1008 		break;
1009 	case CEPH_OSD_OP_NOTIFY_ACK:
1010 		break;
1011 	case CEPH_OSD_OP_NOTIFY:
1012 		dst->notify.cookie = cpu_to_le64(src->notify.cookie);
1013 		break;
1014 	case CEPH_OSD_OP_LIST_WATCHERS:
1015 		break;
1016 	case CEPH_OSD_OP_SETALLOCHINT:
1017 		dst->alloc_hint.expected_object_size =
1018 		    cpu_to_le64(src->alloc_hint.expected_object_size);
1019 		dst->alloc_hint.expected_write_size =
1020 		    cpu_to_le64(src->alloc_hint.expected_write_size);
1021 		break;
1022 	case CEPH_OSD_OP_SETXATTR:
1023 	case CEPH_OSD_OP_CMPXATTR:
1024 		dst->xattr.name_len = cpu_to_le32(src->xattr.name_len);
1025 		dst->xattr.value_len = cpu_to_le32(src->xattr.value_len);
1026 		dst->xattr.cmp_op = src->xattr.cmp_op;
1027 		dst->xattr.cmp_mode = src->xattr.cmp_mode;
1028 		break;
1029 	case CEPH_OSD_OP_CREATE:
1030 	case CEPH_OSD_OP_DELETE:
1031 		break;
1032 	case CEPH_OSD_OP_COPY_FROM2:
1033 		dst->copy_from.snapid = cpu_to_le64(src->copy_from.snapid);
1034 		dst->copy_from.src_version =
1035 			cpu_to_le64(src->copy_from.src_version);
1036 		dst->copy_from.flags = src->copy_from.flags;
1037 		dst->copy_from.src_fadvise_flags =
1038 			cpu_to_le32(src->copy_from.src_fadvise_flags);
1039 		break;
1040 	default:
1041 		pr_err("unsupported osd opcode %s\n",
1042 			ceph_osd_op_name(src->op));
1043 		WARN_ON(1);
1044 
1045 		return 0;
1046 	}
1047 
1048 	dst->op = cpu_to_le16(src->op);
1049 	dst->flags = cpu_to_le32(src->flags);
1050 	dst->payload_len = cpu_to_le32(src->indata_len);
1051 
1052 	return src->indata_len;
1053 }
1054 
1055 /*
1056  * build new request AND message, calculate layout, and adjust file
1057  * extent as needed.
1058  *
1059  * if the file was recently truncated, we include information about its
1060  * old and new size so that the object can be updated appropriately.  (we
1061  * avoid synchronously deleting truncated objects because it's slow.)
1062  */
1063 struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
1064 					       struct ceph_file_layout *layout,
1065 					       struct ceph_vino vino,
1066 					       u64 off, u64 *plen,
1067 					       unsigned int which, int num_ops,
1068 					       int opcode, int flags,
1069 					       struct ceph_snap_context *snapc,
1070 					       u32 truncate_seq,
1071 					       u64 truncate_size,
1072 					       bool use_mempool)
1073 {
1074 	struct ceph_osd_request *req;
1075 	u64 objnum = 0;
1076 	u64 objoff = 0;
1077 	u64 objlen = 0;
1078 	int r;
1079 
1080 	BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
1081 	       opcode != CEPH_OSD_OP_ZERO && opcode != CEPH_OSD_OP_TRUNCATE &&
1082 	       opcode != CEPH_OSD_OP_CREATE && opcode != CEPH_OSD_OP_DELETE);
1083 
1084 	req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool,
1085 					GFP_NOFS);
1086 	if (!req) {
1087 		r = -ENOMEM;
1088 		goto fail;
1089 	}
1090 
1091 	/* calculate max write size */
1092 	r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
1093 	if (r)
1094 		goto fail;
1095 
1096 	if (opcode == CEPH_OSD_OP_CREATE || opcode == CEPH_OSD_OP_DELETE) {
1097 		osd_req_op_init(req, which, opcode, 0);
1098 	} else {
1099 		u32 object_size = layout->object_size;
1100 		u32 object_base = off - objoff;
1101 		if (!(truncate_seq == 1 && truncate_size == -1ULL)) {
1102 			if (truncate_size <= object_base) {
1103 				truncate_size = 0;
1104 			} else {
1105 				truncate_size -= object_base;
1106 				if (truncate_size > object_size)
1107 					truncate_size = object_size;
1108 			}
1109 		}
1110 		osd_req_op_extent_init(req, which, opcode, objoff, objlen,
1111 				       truncate_size, truncate_seq);
1112 	}
1113 
1114 	req->r_flags = flags;
1115 	req->r_base_oloc.pool = layout->pool_id;
1116 	req->r_base_oloc.pool_ns = ceph_try_get_string(layout->pool_ns);
1117 	ceph_oid_printf(&req->r_base_oid, "%llx.%08llx", vino.ino, objnum);
1118 
1119 	req->r_snapid = vino.snap;
1120 	if (flags & CEPH_OSD_FLAG_WRITE)
1121 		req->r_data_offset = off;
1122 
1123 	if (num_ops > 1)
1124 		/*
1125 		 * This is a special case for ceph_writepages_start(), but it
1126 		 * also covers ceph_uninline_data().  If more multi-op request
1127 		 * use cases emerge, we will need a separate helper.
1128 		 */
1129 		r = __ceph_osdc_alloc_messages(req, GFP_NOFS, num_ops, 0);
1130 	else
1131 		r = ceph_osdc_alloc_messages(req, GFP_NOFS);
1132 	if (r)
1133 		goto fail;
1134 
1135 	return req;
1136 
1137 fail:
1138 	ceph_osdc_put_request(req);
1139 	return ERR_PTR(r);
1140 }
1141 EXPORT_SYMBOL(ceph_osdc_new_request);
1142 
1143 /*
1144  * We keep osd requests in an rbtree, sorted by ->r_tid.
1145  */
1146 DEFINE_RB_FUNCS(request, struct ceph_osd_request, r_tid, r_node)
1147 DEFINE_RB_FUNCS(request_mc, struct ceph_osd_request, r_tid, r_mc_node)
1148 
1149 /*
1150  * Call @fn on each OSD request as long as @fn returns 0.
1151  */
1152 static void for_each_request(struct ceph_osd_client *osdc,
1153 			int (*fn)(struct ceph_osd_request *req, void *arg),
1154 			void *arg)
1155 {
1156 	struct rb_node *n, *p;
1157 
1158 	for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
1159 		struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
1160 
1161 		for (p = rb_first(&osd->o_requests); p; ) {
1162 			struct ceph_osd_request *req =
1163 			    rb_entry(p, struct ceph_osd_request, r_node);
1164 
1165 			p = rb_next(p);
1166 			if (fn(req, arg))
1167 				return;
1168 		}
1169 	}
1170 
1171 	for (p = rb_first(&osdc->homeless_osd.o_requests); p; ) {
1172 		struct ceph_osd_request *req =
1173 		    rb_entry(p, struct ceph_osd_request, r_node);
1174 
1175 		p = rb_next(p);
1176 		if (fn(req, arg))
1177 			return;
1178 	}
1179 }
1180 
1181 static bool osd_homeless(struct ceph_osd *osd)
1182 {
1183 	return osd->o_osd == CEPH_HOMELESS_OSD;
1184 }
1185 
1186 static bool osd_registered(struct ceph_osd *osd)
1187 {
1188 	verify_osdc_locked(osd->o_osdc);
1189 
1190 	return !RB_EMPTY_NODE(&osd->o_node);
1191 }
1192 
1193 /*
1194  * Assumes @osd is zero-initialized.
1195  */
1196 static void osd_init(struct ceph_osd *osd)
1197 {
1198 	refcount_set(&osd->o_ref, 1);
1199 	RB_CLEAR_NODE(&osd->o_node);
1200 	osd->o_requests = RB_ROOT;
1201 	osd->o_linger_requests = RB_ROOT;
1202 	osd->o_backoff_mappings = RB_ROOT;
1203 	osd->o_backoffs_by_id = RB_ROOT;
1204 	INIT_LIST_HEAD(&osd->o_osd_lru);
1205 	INIT_LIST_HEAD(&osd->o_keepalive_item);
1206 	osd->o_incarnation = 1;
1207 	mutex_init(&osd->lock);
1208 }
1209 
1210 static void osd_cleanup(struct ceph_osd *osd)
1211 {
1212 	WARN_ON(!RB_EMPTY_NODE(&osd->o_node));
1213 	WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests));
1214 	WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests));
1215 	WARN_ON(!RB_EMPTY_ROOT(&osd->o_backoff_mappings));
1216 	WARN_ON(!RB_EMPTY_ROOT(&osd->o_backoffs_by_id));
1217 	WARN_ON(!list_empty(&osd->o_osd_lru));
1218 	WARN_ON(!list_empty(&osd->o_keepalive_item));
1219 
1220 	if (osd->o_auth.authorizer) {
1221 		WARN_ON(osd_homeless(osd));
1222 		ceph_auth_destroy_authorizer(osd->o_auth.authorizer);
1223 	}
1224 }
1225 
1226 /*
1227  * Track open sessions with osds.
1228  */
1229 static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
1230 {
1231 	struct ceph_osd *osd;
1232 
1233 	WARN_ON(onum == CEPH_HOMELESS_OSD);
1234 
1235 	osd = kzalloc(sizeof(*osd), GFP_NOIO | __GFP_NOFAIL);
1236 	osd_init(osd);
1237 	osd->o_osdc = osdc;
1238 	osd->o_osd = onum;
1239 
1240 	ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
1241 
1242 	return osd;
1243 }
1244 
1245 static struct ceph_osd *get_osd(struct ceph_osd *osd)
1246 {
1247 	if (refcount_inc_not_zero(&osd->o_ref)) {
1248 		dout("get_osd %p %d -> %d\n", osd, refcount_read(&osd->o_ref)-1,
1249 		     refcount_read(&osd->o_ref));
1250 		return osd;
1251 	} else {
1252 		dout("get_osd %p FAIL\n", osd);
1253 		return NULL;
1254 	}
1255 }
1256 
1257 static void put_osd(struct ceph_osd *osd)
1258 {
1259 	dout("put_osd %p %d -> %d\n", osd, refcount_read(&osd->o_ref),
1260 	     refcount_read(&osd->o_ref) - 1);
1261 	if (refcount_dec_and_test(&osd->o_ref)) {
1262 		osd_cleanup(osd);
1263 		kfree(osd);
1264 	}
1265 }
1266 
1267 DEFINE_RB_FUNCS(osd, struct ceph_osd, o_osd, o_node)
1268 
1269 static void __move_osd_to_lru(struct ceph_osd *osd)
1270 {
1271 	struct ceph_osd_client *osdc = osd->o_osdc;
1272 
1273 	dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1274 	BUG_ON(!list_empty(&osd->o_osd_lru));
1275 
1276 	spin_lock(&osdc->osd_lru_lock);
1277 	list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
1278 	spin_unlock(&osdc->osd_lru_lock);
1279 
1280 	osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl;
1281 }
1282 
1283 static void maybe_move_osd_to_lru(struct ceph_osd *osd)
1284 {
1285 	if (RB_EMPTY_ROOT(&osd->o_requests) &&
1286 	    RB_EMPTY_ROOT(&osd->o_linger_requests))
1287 		__move_osd_to_lru(osd);
1288 }
1289 
1290 static void __remove_osd_from_lru(struct ceph_osd *osd)
1291 {
1292 	struct ceph_osd_client *osdc = osd->o_osdc;
1293 
1294 	dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1295 
1296 	spin_lock(&osdc->osd_lru_lock);
1297 	if (!list_empty(&osd->o_osd_lru))
1298 		list_del_init(&osd->o_osd_lru);
1299 	spin_unlock(&osdc->osd_lru_lock);
1300 }
1301 
1302 /*
1303  * Close the connection and assign any leftover requests to the
1304  * homeless session.
1305  */
1306 static void close_osd(struct ceph_osd *osd)
1307 {
1308 	struct ceph_osd_client *osdc = osd->o_osdc;
1309 	struct rb_node *n;
1310 
1311 	verify_osdc_wrlocked(osdc);
1312 	dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1313 
1314 	ceph_con_close(&osd->o_con);
1315 
1316 	for (n = rb_first(&osd->o_requests); n; ) {
1317 		struct ceph_osd_request *req =
1318 		    rb_entry(n, struct ceph_osd_request, r_node);
1319 
1320 		n = rb_next(n); /* unlink_request() */
1321 
1322 		dout(" reassigning req %p tid %llu\n", req, req->r_tid);
1323 		unlink_request(osd, req);
1324 		link_request(&osdc->homeless_osd, req);
1325 	}
1326 	for (n = rb_first(&osd->o_linger_requests); n; ) {
1327 		struct ceph_osd_linger_request *lreq =
1328 		    rb_entry(n, struct ceph_osd_linger_request, node);
1329 
1330 		n = rb_next(n); /* unlink_linger() */
1331 
1332 		dout(" reassigning lreq %p linger_id %llu\n", lreq,
1333 		     lreq->linger_id);
1334 		unlink_linger(osd, lreq);
1335 		link_linger(&osdc->homeless_osd, lreq);
1336 	}
1337 	clear_backoffs(osd);
1338 
1339 	__remove_osd_from_lru(osd);
1340 	erase_osd(&osdc->osds, osd);
1341 	put_osd(osd);
1342 }
1343 
1344 /*
1345  * reset osd connect
1346  */
1347 static int reopen_osd(struct ceph_osd *osd)
1348 {
1349 	struct ceph_entity_addr *peer_addr;
1350 
1351 	dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1352 
1353 	if (RB_EMPTY_ROOT(&osd->o_requests) &&
1354 	    RB_EMPTY_ROOT(&osd->o_linger_requests)) {
1355 		close_osd(osd);
1356 		return -ENODEV;
1357 	}
1358 
1359 	peer_addr = &osd->o_osdc->osdmap->osd_addr[osd->o_osd];
1360 	if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1361 			!ceph_con_opened(&osd->o_con)) {
1362 		struct rb_node *n;
1363 
1364 		dout("osd addr hasn't changed and connection never opened, "
1365 		     "letting msgr retry\n");
1366 		/* touch each r_stamp for handle_timeout()'s benfit */
1367 		for (n = rb_first(&osd->o_requests); n; n = rb_next(n)) {
1368 			struct ceph_osd_request *req =
1369 			    rb_entry(n, struct ceph_osd_request, r_node);
1370 			req->r_stamp = jiffies;
1371 		}
1372 
1373 		return -EAGAIN;
1374 	}
1375 
1376 	ceph_con_close(&osd->o_con);
1377 	ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1378 	osd->o_incarnation++;
1379 
1380 	return 0;
1381 }
1382 
1383 static struct ceph_osd *lookup_create_osd(struct ceph_osd_client *osdc, int o,
1384 					  bool wrlocked)
1385 {
1386 	struct ceph_osd *osd;
1387 
1388 	if (wrlocked)
1389 		verify_osdc_wrlocked(osdc);
1390 	else
1391 		verify_osdc_locked(osdc);
1392 
1393 	if (o != CEPH_HOMELESS_OSD)
1394 		osd = lookup_osd(&osdc->osds, o);
1395 	else
1396 		osd = &osdc->homeless_osd;
1397 	if (!osd) {
1398 		if (!wrlocked)
1399 			return ERR_PTR(-EAGAIN);
1400 
1401 		osd = create_osd(osdc, o);
1402 		insert_osd(&osdc->osds, osd);
1403 		ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd,
1404 			      &osdc->osdmap->osd_addr[osd->o_osd]);
1405 	}
1406 
1407 	dout("%s osdc %p osd%d -> osd %p\n", __func__, osdc, o, osd);
1408 	return osd;
1409 }
1410 
1411 /*
1412  * Create request <-> OSD session relation.
1413  *
1414  * @req has to be assigned a tid, @osd may be homeless.
1415  */
1416 static void link_request(struct ceph_osd *osd, struct ceph_osd_request *req)
1417 {
1418 	verify_osd_locked(osd);
1419 	WARN_ON(!req->r_tid || req->r_osd);
1420 	dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd,
1421 	     req, req->r_tid);
1422 
1423 	if (!osd_homeless(osd))
1424 		__remove_osd_from_lru(osd);
1425 	else
1426 		atomic_inc(&osd->o_osdc->num_homeless);
1427 
1428 	get_osd(osd);
1429 	insert_request(&osd->o_requests, req);
1430 	req->r_osd = osd;
1431 }
1432 
1433 static void unlink_request(struct ceph_osd *osd, struct ceph_osd_request *req)
1434 {
1435 	verify_osd_locked(osd);
1436 	WARN_ON(req->r_osd != osd);
1437 	dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd,
1438 	     req, req->r_tid);
1439 
1440 	req->r_osd = NULL;
1441 	erase_request(&osd->o_requests, req);
1442 	put_osd(osd);
1443 
1444 	if (!osd_homeless(osd))
1445 		maybe_move_osd_to_lru(osd);
1446 	else
1447 		atomic_dec(&osd->o_osdc->num_homeless);
1448 }
1449 
1450 static bool __pool_full(struct ceph_pg_pool_info *pi)
1451 {
1452 	return pi->flags & CEPH_POOL_FLAG_FULL;
1453 }
1454 
1455 static bool have_pool_full(struct ceph_osd_client *osdc)
1456 {
1457 	struct rb_node *n;
1458 
1459 	for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) {
1460 		struct ceph_pg_pool_info *pi =
1461 		    rb_entry(n, struct ceph_pg_pool_info, node);
1462 
1463 		if (__pool_full(pi))
1464 			return true;
1465 	}
1466 
1467 	return false;
1468 }
1469 
1470 static bool pool_full(struct ceph_osd_client *osdc, s64 pool_id)
1471 {
1472 	struct ceph_pg_pool_info *pi;
1473 
1474 	pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id);
1475 	if (!pi)
1476 		return false;
1477 
1478 	return __pool_full(pi);
1479 }
1480 
1481 /*
1482  * Returns whether a request should be blocked from being sent
1483  * based on the current osdmap and osd_client settings.
1484  */
1485 static bool target_should_be_paused(struct ceph_osd_client *osdc,
1486 				    const struct ceph_osd_request_target *t,
1487 				    struct ceph_pg_pool_info *pi)
1488 {
1489 	bool pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
1490 	bool pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
1491 		       ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
1492 		       __pool_full(pi);
1493 
1494 	WARN_ON(pi->id != t->target_oloc.pool);
1495 	return ((t->flags & CEPH_OSD_FLAG_READ) && pauserd) ||
1496 	       ((t->flags & CEPH_OSD_FLAG_WRITE) && pausewr) ||
1497 	       (osdc->osdmap->epoch < osdc->epoch_barrier);
1498 }
1499 
1500 enum calc_target_result {
1501 	CALC_TARGET_NO_ACTION = 0,
1502 	CALC_TARGET_NEED_RESEND,
1503 	CALC_TARGET_POOL_DNE,
1504 };
1505 
1506 static enum calc_target_result calc_target(struct ceph_osd_client *osdc,
1507 					   struct ceph_osd_request_target *t,
1508 					   bool any_change)
1509 {
1510 	struct ceph_pg_pool_info *pi;
1511 	struct ceph_pg pgid, last_pgid;
1512 	struct ceph_osds up, acting;
1513 	bool force_resend = false;
1514 	bool unpaused = false;
1515 	bool legacy_change = false;
1516 	bool split = false;
1517 	bool sort_bitwise = ceph_osdmap_flag(osdc, CEPH_OSDMAP_SORTBITWISE);
1518 	bool recovery_deletes = ceph_osdmap_flag(osdc,
1519 						 CEPH_OSDMAP_RECOVERY_DELETES);
1520 	enum calc_target_result ct_res;
1521 
1522 	t->epoch = osdc->osdmap->epoch;
1523 	pi = ceph_pg_pool_by_id(osdc->osdmap, t->base_oloc.pool);
1524 	if (!pi) {
1525 		t->osd = CEPH_HOMELESS_OSD;
1526 		ct_res = CALC_TARGET_POOL_DNE;
1527 		goto out;
1528 	}
1529 
1530 	if (osdc->osdmap->epoch == pi->last_force_request_resend) {
1531 		if (t->last_force_resend < pi->last_force_request_resend) {
1532 			t->last_force_resend = pi->last_force_request_resend;
1533 			force_resend = true;
1534 		} else if (t->last_force_resend == 0) {
1535 			force_resend = true;
1536 		}
1537 	}
1538 
1539 	/* apply tiering */
1540 	ceph_oid_copy(&t->target_oid, &t->base_oid);
1541 	ceph_oloc_copy(&t->target_oloc, &t->base_oloc);
1542 	if ((t->flags & CEPH_OSD_FLAG_IGNORE_OVERLAY) == 0) {
1543 		if (t->flags & CEPH_OSD_FLAG_READ && pi->read_tier >= 0)
1544 			t->target_oloc.pool = pi->read_tier;
1545 		if (t->flags & CEPH_OSD_FLAG_WRITE && pi->write_tier >= 0)
1546 			t->target_oloc.pool = pi->write_tier;
1547 
1548 		pi = ceph_pg_pool_by_id(osdc->osdmap, t->target_oloc.pool);
1549 		if (!pi) {
1550 			t->osd = CEPH_HOMELESS_OSD;
1551 			ct_res = CALC_TARGET_POOL_DNE;
1552 			goto out;
1553 		}
1554 	}
1555 
1556 	__ceph_object_locator_to_pg(pi, &t->target_oid, &t->target_oloc, &pgid);
1557 	last_pgid.pool = pgid.pool;
1558 	last_pgid.seed = ceph_stable_mod(pgid.seed, t->pg_num, t->pg_num_mask);
1559 
1560 	ceph_pg_to_up_acting_osds(osdc->osdmap, pi, &pgid, &up, &acting);
1561 	if (any_change &&
1562 	    ceph_is_new_interval(&t->acting,
1563 				 &acting,
1564 				 &t->up,
1565 				 &up,
1566 				 t->size,
1567 				 pi->size,
1568 				 t->min_size,
1569 				 pi->min_size,
1570 				 t->pg_num,
1571 				 pi->pg_num,
1572 				 t->sort_bitwise,
1573 				 sort_bitwise,
1574 				 t->recovery_deletes,
1575 				 recovery_deletes,
1576 				 &last_pgid))
1577 		force_resend = true;
1578 
1579 	if (t->paused && !target_should_be_paused(osdc, t, pi)) {
1580 		t->paused = false;
1581 		unpaused = true;
1582 	}
1583 	legacy_change = ceph_pg_compare(&t->pgid, &pgid) ||
1584 			ceph_osds_changed(&t->acting, &acting, any_change);
1585 	if (t->pg_num)
1586 		split = ceph_pg_is_split(&last_pgid, t->pg_num, pi->pg_num);
1587 
1588 	if (legacy_change || force_resend || split) {
1589 		t->pgid = pgid; /* struct */
1590 		ceph_pg_to_primary_shard(osdc->osdmap, pi, &pgid, &t->spgid);
1591 		ceph_osds_copy(&t->acting, &acting);
1592 		ceph_osds_copy(&t->up, &up);
1593 		t->size = pi->size;
1594 		t->min_size = pi->min_size;
1595 		t->pg_num = pi->pg_num;
1596 		t->pg_num_mask = pi->pg_num_mask;
1597 		t->sort_bitwise = sort_bitwise;
1598 		t->recovery_deletes = recovery_deletes;
1599 
1600 		t->osd = acting.primary;
1601 	}
1602 
1603 	if (unpaused || legacy_change || force_resend || split)
1604 		ct_res = CALC_TARGET_NEED_RESEND;
1605 	else
1606 		ct_res = CALC_TARGET_NO_ACTION;
1607 
1608 out:
1609 	dout("%s t %p -> %d%d%d%d ct_res %d osd%d\n", __func__, t, unpaused,
1610 	     legacy_change, force_resend, split, ct_res, t->osd);
1611 	return ct_res;
1612 }
1613 
1614 static struct ceph_spg_mapping *alloc_spg_mapping(void)
1615 {
1616 	struct ceph_spg_mapping *spg;
1617 
1618 	spg = kmalloc(sizeof(*spg), GFP_NOIO);
1619 	if (!spg)
1620 		return NULL;
1621 
1622 	RB_CLEAR_NODE(&spg->node);
1623 	spg->backoffs = RB_ROOT;
1624 	return spg;
1625 }
1626 
1627 static void free_spg_mapping(struct ceph_spg_mapping *spg)
1628 {
1629 	WARN_ON(!RB_EMPTY_NODE(&spg->node));
1630 	WARN_ON(!RB_EMPTY_ROOT(&spg->backoffs));
1631 
1632 	kfree(spg);
1633 }
1634 
1635 /*
1636  * rbtree of ceph_spg_mapping for handling map<spg_t, ...>, similar to
1637  * ceph_pg_mapping.  Used to track OSD backoffs -- a backoff [range] is
1638  * defined only within a specific spgid; it does not pass anything to
1639  * children on split, or to another primary.
1640  */
1641 DEFINE_RB_FUNCS2(spg_mapping, struct ceph_spg_mapping, spgid, ceph_spg_compare,
1642 		 RB_BYPTR, const struct ceph_spg *, node)
1643 
1644 static u64 hoid_get_bitwise_key(const struct ceph_hobject_id *hoid)
1645 {
1646 	return hoid->is_max ? 0x100000000ull : hoid->hash_reverse_bits;
1647 }
1648 
1649 static void hoid_get_effective_key(const struct ceph_hobject_id *hoid,
1650 				   void **pkey, size_t *pkey_len)
1651 {
1652 	if (hoid->key_len) {
1653 		*pkey = hoid->key;
1654 		*pkey_len = hoid->key_len;
1655 	} else {
1656 		*pkey = hoid->oid;
1657 		*pkey_len = hoid->oid_len;
1658 	}
1659 }
1660 
1661 static int compare_names(const void *name1, size_t name1_len,
1662 			 const void *name2, size_t name2_len)
1663 {
1664 	int ret;
1665 
1666 	ret = memcmp(name1, name2, min(name1_len, name2_len));
1667 	if (!ret) {
1668 		if (name1_len < name2_len)
1669 			ret = -1;
1670 		else if (name1_len > name2_len)
1671 			ret = 1;
1672 	}
1673 	return ret;
1674 }
1675 
1676 static int hoid_compare(const struct ceph_hobject_id *lhs,
1677 			const struct ceph_hobject_id *rhs)
1678 {
1679 	void *effective_key1, *effective_key2;
1680 	size_t effective_key1_len, effective_key2_len;
1681 	int ret;
1682 
1683 	if (lhs->is_max < rhs->is_max)
1684 		return -1;
1685 	if (lhs->is_max > rhs->is_max)
1686 		return 1;
1687 
1688 	if (lhs->pool < rhs->pool)
1689 		return -1;
1690 	if (lhs->pool > rhs->pool)
1691 		return 1;
1692 
1693 	if (hoid_get_bitwise_key(lhs) < hoid_get_bitwise_key(rhs))
1694 		return -1;
1695 	if (hoid_get_bitwise_key(lhs) > hoid_get_bitwise_key(rhs))
1696 		return 1;
1697 
1698 	ret = compare_names(lhs->nspace, lhs->nspace_len,
1699 			    rhs->nspace, rhs->nspace_len);
1700 	if (ret)
1701 		return ret;
1702 
1703 	hoid_get_effective_key(lhs, &effective_key1, &effective_key1_len);
1704 	hoid_get_effective_key(rhs, &effective_key2, &effective_key2_len);
1705 	ret = compare_names(effective_key1, effective_key1_len,
1706 			    effective_key2, effective_key2_len);
1707 	if (ret)
1708 		return ret;
1709 
1710 	ret = compare_names(lhs->oid, lhs->oid_len, rhs->oid, rhs->oid_len);
1711 	if (ret)
1712 		return ret;
1713 
1714 	if (lhs->snapid < rhs->snapid)
1715 		return -1;
1716 	if (lhs->snapid > rhs->snapid)
1717 		return 1;
1718 
1719 	return 0;
1720 }
1721 
1722 /*
1723  * For decoding ->begin and ->end of MOSDBackoff only -- no MIN/MAX
1724  * compat stuff here.
1725  *
1726  * Assumes @hoid is zero-initialized.
1727  */
1728 static int decode_hoid(void **p, void *end, struct ceph_hobject_id *hoid)
1729 {
1730 	u8 struct_v;
1731 	u32 struct_len;
1732 	int ret;
1733 
1734 	ret = ceph_start_decoding(p, end, 4, "hobject_t", &struct_v,
1735 				  &struct_len);
1736 	if (ret)
1737 		return ret;
1738 
1739 	if (struct_v < 4) {
1740 		pr_err("got struct_v %d < 4 of hobject_t\n", struct_v);
1741 		goto e_inval;
1742 	}
1743 
1744 	hoid->key = ceph_extract_encoded_string(p, end, &hoid->key_len,
1745 						GFP_NOIO);
1746 	if (IS_ERR(hoid->key)) {
1747 		ret = PTR_ERR(hoid->key);
1748 		hoid->key = NULL;
1749 		return ret;
1750 	}
1751 
1752 	hoid->oid = ceph_extract_encoded_string(p, end, &hoid->oid_len,
1753 						GFP_NOIO);
1754 	if (IS_ERR(hoid->oid)) {
1755 		ret = PTR_ERR(hoid->oid);
1756 		hoid->oid = NULL;
1757 		return ret;
1758 	}
1759 
1760 	ceph_decode_64_safe(p, end, hoid->snapid, e_inval);
1761 	ceph_decode_32_safe(p, end, hoid->hash, e_inval);
1762 	ceph_decode_8_safe(p, end, hoid->is_max, e_inval);
1763 
1764 	hoid->nspace = ceph_extract_encoded_string(p, end, &hoid->nspace_len,
1765 						   GFP_NOIO);
1766 	if (IS_ERR(hoid->nspace)) {
1767 		ret = PTR_ERR(hoid->nspace);
1768 		hoid->nspace = NULL;
1769 		return ret;
1770 	}
1771 
1772 	ceph_decode_64_safe(p, end, hoid->pool, e_inval);
1773 
1774 	ceph_hoid_build_hash_cache(hoid);
1775 	return 0;
1776 
1777 e_inval:
1778 	return -EINVAL;
1779 }
1780 
1781 static int hoid_encoding_size(const struct ceph_hobject_id *hoid)
1782 {
1783 	return 8 + 4 + 1 + 8 + /* snapid, hash, is_max, pool */
1784 	       4 + hoid->key_len + 4 + hoid->oid_len + 4 + hoid->nspace_len;
1785 }
1786 
1787 static void encode_hoid(void **p, void *end, const struct ceph_hobject_id *hoid)
1788 {
1789 	ceph_start_encoding(p, 4, 3, hoid_encoding_size(hoid));
1790 	ceph_encode_string(p, end, hoid->key, hoid->key_len);
1791 	ceph_encode_string(p, end, hoid->oid, hoid->oid_len);
1792 	ceph_encode_64(p, hoid->snapid);
1793 	ceph_encode_32(p, hoid->hash);
1794 	ceph_encode_8(p, hoid->is_max);
1795 	ceph_encode_string(p, end, hoid->nspace, hoid->nspace_len);
1796 	ceph_encode_64(p, hoid->pool);
1797 }
1798 
1799 static void free_hoid(struct ceph_hobject_id *hoid)
1800 {
1801 	if (hoid) {
1802 		kfree(hoid->key);
1803 		kfree(hoid->oid);
1804 		kfree(hoid->nspace);
1805 		kfree(hoid);
1806 	}
1807 }
1808 
1809 static struct ceph_osd_backoff *alloc_backoff(void)
1810 {
1811 	struct ceph_osd_backoff *backoff;
1812 
1813 	backoff = kzalloc(sizeof(*backoff), GFP_NOIO);
1814 	if (!backoff)
1815 		return NULL;
1816 
1817 	RB_CLEAR_NODE(&backoff->spg_node);
1818 	RB_CLEAR_NODE(&backoff->id_node);
1819 	return backoff;
1820 }
1821 
1822 static void free_backoff(struct ceph_osd_backoff *backoff)
1823 {
1824 	WARN_ON(!RB_EMPTY_NODE(&backoff->spg_node));
1825 	WARN_ON(!RB_EMPTY_NODE(&backoff->id_node));
1826 
1827 	free_hoid(backoff->begin);
1828 	free_hoid(backoff->end);
1829 	kfree(backoff);
1830 }
1831 
1832 /*
1833  * Within a specific spgid, backoffs are managed by ->begin hoid.
1834  */
1835 DEFINE_RB_INSDEL_FUNCS2(backoff, struct ceph_osd_backoff, begin, hoid_compare,
1836 			RB_BYVAL, spg_node);
1837 
1838 static struct ceph_osd_backoff *lookup_containing_backoff(struct rb_root *root,
1839 					    const struct ceph_hobject_id *hoid)
1840 {
1841 	struct rb_node *n = root->rb_node;
1842 
1843 	while (n) {
1844 		struct ceph_osd_backoff *cur =
1845 		    rb_entry(n, struct ceph_osd_backoff, spg_node);
1846 		int cmp;
1847 
1848 		cmp = hoid_compare(hoid, cur->begin);
1849 		if (cmp < 0) {
1850 			n = n->rb_left;
1851 		} else if (cmp > 0) {
1852 			if (hoid_compare(hoid, cur->end) < 0)
1853 				return cur;
1854 
1855 			n = n->rb_right;
1856 		} else {
1857 			return cur;
1858 		}
1859 	}
1860 
1861 	return NULL;
1862 }
1863 
1864 /*
1865  * Each backoff has a unique id within its OSD session.
1866  */
1867 DEFINE_RB_FUNCS(backoff_by_id, struct ceph_osd_backoff, id, id_node)
1868 
1869 static void clear_backoffs(struct ceph_osd *osd)
1870 {
1871 	while (!RB_EMPTY_ROOT(&osd->o_backoff_mappings)) {
1872 		struct ceph_spg_mapping *spg =
1873 		    rb_entry(rb_first(&osd->o_backoff_mappings),
1874 			     struct ceph_spg_mapping, node);
1875 
1876 		while (!RB_EMPTY_ROOT(&spg->backoffs)) {
1877 			struct ceph_osd_backoff *backoff =
1878 			    rb_entry(rb_first(&spg->backoffs),
1879 				     struct ceph_osd_backoff, spg_node);
1880 
1881 			erase_backoff(&spg->backoffs, backoff);
1882 			erase_backoff_by_id(&osd->o_backoffs_by_id, backoff);
1883 			free_backoff(backoff);
1884 		}
1885 		erase_spg_mapping(&osd->o_backoff_mappings, spg);
1886 		free_spg_mapping(spg);
1887 	}
1888 }
1889 
1890 /*
1891  * Set up a temporary, non-owning view into @t.
1892  */
1893 static void hoid_fill_from_target(struct ceph_hobject_id *hoid,
1894 				  const struct ceph_osd_request_target *t)
1895 {
1896 	hoid->key = NULL;
1897 	hoid->key_len = 0;
1898 	hoid->oid = t->target_oid.name;
1899 	hoid->oid_len = t->target_oid.name_len;
1900 	hoid->snapid = CEPH_NOSNAP;
1901 	hoid->hash = t->pgid.seed;
1902 	hoid->is_max = false;
1903 	if (t->target_oloc.pool_ns) {
1904 		hoid->nspace = t->target_oloc.pool_ns->str;
1905 		hoid->nspace_len = t->target_oloc.pool_ns->len;
1906 	} else {
1907 		hoid->nspace = NULL;
1908 		hoid->nspace_len = 0;
1909 	}
1910 	hoid->pool = t->target_oloc.pool;
1911 	ceph_hoid_build_hash_cache(hoid);
1912 }
1913 
1914 static bool should_plug_request(struct ceph_osd_request *req)
1915 {
1916 	struct ceph_osd *osd = req->r_osd;
1917 	struct ceph_spg_mapping *spg;
1918 	struct ceph_osd_backoff *backoff;
1919 	struct ceph_hobject_id hoid;
1920 
1921 	spg = lookup_spg_mapping(&osd->o_backoff_mappings, &req->r_t.spgid);
1922 	if (!spg)
1923 		return false;
1924 
1925 	hoid_fill_from_target(&hoid, &req->r_t);
1926 	backoff = lookup_containing_backoff(&spg->backoffs, &hoid);
1927 	if (!backoff)
1928 		return false;
1929 
1930 	dout("%s req %p tid %llu backoff osd%d spgid %llu.%xs%d id %llu\n",
1931 	     __func__, req, req->r_tid, osd->o_osd, backoff->spgid.pgid.pool,
1932 	     backoff->spgid.pgid.seed, backoff->spgid.shard, backoff->id);
1933 	return true;
1934 }
1935 
1936 /*
1937  * Keep get_num_data_items() in sync with this function.
1938  */
1939 static void setup_request_data(struct ceph_osd_request *req)
1940 {
1941 	struct ceph_msg *request_msg = req->r_request;
1942 	struct ceph_msg *reply_msg = req->r_reply;
1943 	struct ceph_osd_req_op *op;
1944 
1945 	if (req->r_request->num_data_items || req->r_reply->num_data_items)
1946 		return;
1947 
1948 	WARN_ON(request_msg->data_length || reply_msg->data_length);
1949 	for (op = req->r_ops; op != &req->r_ops[req->r_num_ops]; op++) {
1950 		switch (op->op) {
1951 		/* request */
1952 		case CEPH_OSD_OP_WRITE:
1953 		case CEPH_OSD_OP_WRITEFULL:
1954 			WARN_ON(op->indata_len != op->extent.length);
1955 			ceph_osdc_msg_data_add(request_msg,
1956 					       &op->extent.osd_data);
1957 			break;
1958 		case CEPH_OSD_OP_SETXATTR:
1959 		case CEPH_OSD_OP_CMPXATTR:
1960 			WARN_ON(op->indata_len != op->xattr.name_len +
1961 						  op->xattr.value_len);
1962 			ceph_osdc_msg_data_add(request_msg,
1963 					       &op->xattr.osd_data);
1964 			break;
1965 		case CEPH_OSD_OP_NOTIFY_ACK:
1966 			ceph_osdc_msg_data_add(request_msg,
1967 					       &op->notify_ack.request_data);
1968 			break;
1969 		case CEPH_OSD_OP_COPY_FROM2:
1970 			ceph_osdc_msg_data_add(request_msg,
1971 					       &op->copy_from.osd_data);
1972 			break;
1973 
1974 		/* reply */
1975 		case CEPH_OSD_OP_STAT:
1976 			ceph_osdc_msg_data_add(reply_msg,
1977 					       &op->raw_data_in);
1978 			break;
1979 		case CEPH_OSD_OP_READ:
1980 			ceph_osdc_msg_data_add(reply_msg,
1981 					       &op->extent.osd_data);
1982 			break;
1983 		case CEPH_OSD_OP_LIST_WATCHERS:
1984 			ceph_osdc_msg_data_add(reply_msg,
1985 					       &op->list_watchers.response_data);
1986 			break;
1987 
1988 		/* both */
1989 		case CEPH_OSD_OP_CALL:
1990 			WARN_ON(op->indata_len != op->cls.class_len +
1991 						  op->cls.method_len +
1992 						  op->cls.indata_len);
1993 			ceph_osdc_msg_data_add(request_msg,
1994 					       &op->cls.request_info);
1995 			/* optional, can be NONE */
1996 			ceph_osdc_msg_data_add(request_msg,
1997 					       &op->cls.request_data);
1998 			/* optional, can be NONE */
1999 			ceph_osdc_msg_data_add(reply_msg,
2000 					       &op->cls.response_data);
2001 			break;
2002 		case CEPH_OSD_OP_NOTIFY:
2003 			ceph_osdc_msg_data_add(request_msg,
2004 					       &op->notify.request_data);
2005 			ceph_osdc_msg_data_add(reply_msg,
2006 					       &op->notify.response_data);
2007 			break;
2008 		}
2009 	}
2010 }
2011 
2012 static void encode_pgid(void **p, const struct ceph_pg *pgid)
2013 {
2014 	ceph_encode_8(p, 1);
2015 	ceph_encode_64(p, pgid->pool);
2016 	ceph_encode_32(p, pgid->seed);
2017 	ceph_encode_32(p, -1); /* preferred */
2018 }
2019 
2020 static void encode_spgid(void **p, const struct ceph_spg *spgid)
2021 {
2022 	ceph_start_encoding(p, 1, 1, CEPH_PGID_ENCODING_LEN + 1);
2023 	encode_pgid(p, &spgid->pgid);
2024 	ceph_encode_8(p, spgid->shard);
2025 }
2026 
2027 static void encode_oloc(void **p, void *end,
2028 			const struct ceph_object_locator *oloc)
2029 {
2030 	ceph_start_encoding(p, 5, 4, ceph_oloc_encoding_size(oloc));
2031 	ceph_encode_64(p, oloc->pool);
2032 	ceph_encode_32(p, -1); /* preferred */
2033 	ceph_encode_32(p, 0);  /* key len */
2034 	if (oloc->pool_ns)
2035 		ceph_encode_string(p, end, oloc->pool_ns->str,
2036 				   oloc->pool_ns->len);
2037 	else
2038 		ceph_encode_32(p, 0);
2039 }
2040 
2041 static void encode_request_partial(struct ceph_osd_request *req,
2042 				   struct ceph_msg *msg)
2043 {
2044 	void *p = msg->front.iov_base;
2045 	void *const end = p + msg->front_alloc_len;
2046 	u32 data_len = 0;
2047 	int i;
2048 
2049 	if (req->r_flags & CEPH_OSD_FLAG_WRITE) {
2050 		/* snapshots aren't writeable */
2051 		WARN_ON(req->r_snapid != CEPH_NOSNAP);
2052 	} else {
2053 		WARN_ON(req->r_mtime.tv_sec || req->r_mtime.tv_nsec ||
2054 			req->r_data_offset || req->r_snapc);
2055 	}
2056 
2057 	setup_request_data(req);
2058 
2059 	encode_spgid(&p, &req->r_t.spgid); /* actual spg */
2060 	ceph_encode_32(&p, req->r_t.pgid.seed); /* raw hash */
2061 	ceph_encode_32(&p, req->r_osdc->osdmap->epoch);
2062 	ceph_encode_32(&p, req->r_flags);
2063 
2064 	/* reqid */
2065 	ceph_start_encoding(&p, 2, 2, sizeof(struct ceph_osd_reqid));
2066 	memset(p, 0, sizeof(struct ceph_osd_reqid));
2067 	p += sizeof(struct ceph_osd_reqid);
2068 
2069 	/* trace */
2070 	memset(p, 0, sizeof(struct ceph_blkin_trace_info));
2071 	p += sizeof(struct ceph_blkin_trace_info);
2072 
2073 	ceph_encode_32(&p, 0); /* client_inc, always 0 */
2074 	ceph_encode_timespec64(p, &req->r_mtime);
2075 	p += sizeof(struct ceph_timespec);
2076 
2077 	encode_oloc(&p, end, &req->r_t.target_oloc);
2078 	ceph_encode_string(&p, end, req->r_t.target_oid.name,
2079 			   req->r_t.target_oid.name_len);
2080 
2081 	/* ops, can imply data */
2082 	ceph_encode_16(&p, req->r_num_ops);
2083 	for (i = 0; i < req->r_num_ops; i++) {
2084 		data_len += osd_req_encode_op(p, &req->r_ops[i]);
2085 		p += sizeof(struct ceph_osd_op);
2086 	}
2087 
2088 	ceph_encode_64(&p, req->r_snapid); /* snapid */
2089 	if (req->r_snapc) {
2090 		ceph_encode_64(&p, req->r_snapc->seq);
2091 		ceph_encode_32(&p, req->r_snapc->num_snaps);
2092 		for (i = 0; i < req->r_snapc->num_snaps; i++)
2093 			ceph_encode_64(&p, req->r_snapc->snaps[i]);
2094 	} else {
2095 		ceph_encode_64(&p, 0); /* snap_seq */
2096 		ceph_encode_32(&p, 0); /* snaps len */
2097 	}
2098 
2099 	ceph_encode_32(&p, req->r_attempts); /* retry_attempt */
2100 	BUG_ON(p > end - 8); /* space for features */
2101 
2102 	msg->hdr.version = cpu_to_le16(8); /* MOSDOp v8 */
2103 	/* front_len is finalized in encode_request_finish() */
2104 	msg->front.iov_len = p - msg->front.iov_base;
2105 	msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2106 	msg->hdr.data_len = cpu_to_le32(data_len);
2107 	/*
2108 	 * The header "data_off" is a hint to the receiver allowing it
2109 	 * to align received data into its buffers such that there's no
2110 	 * need to re-copy it before writing it to disk (direct I/O).
2111 	 */
2112 	msg->hdr.data_off = cpu_to_le16(req->r_data_offset);
2113 
2114 	dout("%s req %p msg %p oid %s oid_len %d\n", __func__, req, msg,
2115 	     req->r_t.target_oid.name, req->r_t.target_oid.name_len);
2116 }
2117 
2118 static void encode_request_finish(struct ceph_msg *msg)
2119 {
2120 	void *p = msg->front.iov_base;
2121 	void *const partial_end = p + msg->front.iov_len;
2122 	void *const end = p + msg->front_alloc_len;
2123 
2124 	if (CEPH_HAVE_FEATURE(msg->con->peer_features, RESEND_ON_SPLIT)) {
2125 		/* luminous OSD -- encode features and be done */
2126 		p = partial_end;
2127 		ceph_encode_64(&p, msg->con->peer_features);
2128 	} else {
2129 		struct {
2130 			char spgid[CEPH_ENCODING_START_BLK_LEN +
2131 				   CEPH_PGID_ENCODING_LEN + 1];
2132 			__le32 hash;
2133 			__le32 epoch;
2134 			__le32 flags;
2135 			char reqid[CEPH_ENCODING_START_BLK_LEN +
2136 				   sizeof(struct ceph_osd_reqid)];
2137 			char trace[sizeof(struct ceph_blkin_trace_info)];
2138 			__le32 client_inc;
2139 			struct ceph_timespec mtime;
2140 		} __packed head;
2141 		struct ceph_pg pgid;
2142 		void *oloc, *oid, *tail;
2143 		int oloc_len, oid_len, tail_len;
2144 		int len;
2145 
2146 		/*
2147 		 * Pre-luminous OSD -- reencode v8 into v4 using @head
2148 		 * as a temporary buffer.  Encode the raw PG; the rest
2149 		 * is just a matter of moving oloc, oid and tail blobs
2150 		 * around.
2151 		 */
2152 		memcpy(&head, p, sizeof(head));
2153 		p += sizeof(head);
2154 
2155 		oloc = p;
2156 		p += CEPH_ENCODING_START_BLK_LEN;
2157 		pgid.pool = ceph_decode_64(&p);
2158 		p += 4 + 4; /* preferred, key len */
2159 		len = ceph_decode_32(&p);
2160 		p += len;   /* nspace */
2161 		oloc_len = p - oloc;
2162 
2163 		oid = p;
2164 		len = ceph_decode_32(&p);
2165 		p += len;
2166 		oid_len = p - oid;
2167 
2168 		tail = p;
2169 		tail_len = partial_end - p;
2170 
2171 		p = msg->front.iov_base;
2172 		ceph_encode_copy(&p, &head.client_inc, sizeof(head.client_inc));
2173 		ceph_encode_copy(&p, &head.epoch, sizeof(head.epoch));
2174 		ceph_encode_copy(&p, &head.flags, sizeof(head.flags));
2175 		ceph_encode_copy(&p, &head.mtime, sizeof(head.mtime));
2176 
2177 		/* reassert_version */
2178 		memset(p, 0, sizeof(struct ceph_eversion));
2179 		p += sizeof(struct ceph_eversion);
2180 
2181 		BUG_ON(p >= oloc);
2182 		memmove(p, oloc, oloc_len);
2183 		p += oloc_len;
2184 
2185 		pgid.seed = le32_to_cpu(head.hash);
2186 		encode_pgid(&p, &pgid); /* raw pg */
2187 
2188 		BUG_ON(p >= oid);
2189 		memmove(p, oid, oid_len);
2190 		p += oid_len;
2191 
2192 		/* tail -- ops, snapid, snapc, retry_attempt */
2193 		BUG_ON(p >= tail);
2194 		memmove(p, tail, tail_len);
2195 		p += tail_len;
2196 
2197 		msg->hdr.version = cpu_to_le16(4); /* MOSDOp v4 */
2198 	}
2199 
2200 	BUG_ON(p > end);
2201 	msg->front.iov_len = p - msg->front.iov_base;
2202 	msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2203 
2204 	dout("%s msg %p tid %llu %u+%u+%u v%d\n", __func__, msg,
2205 	     le64_to_cpu(msg->hdr.tid), le32_to_cpu(msg->hdr.front_len),
2206 	     le32_to_cpu(msg->hdr.middle_len), le32_to_cpu(msg->hdr.data_len),
2207 	     le16_to_cpu(msg->hdr.version));
2208 }
2209 
2210 /*
2211  * @req has to be assigned a tid and registered.
2212  */
2213 static void send_request(struct ceph_osd_request *req)
2214 {
2215 	struct ceph_osd *osd = req->r_osd;
2216 
2217 	verify_osd_locked(osd);
2218 	WARN_ON(osd->o_osd != req->r_t.osd);
2219 
2220 	/* backoff? */
2221 	if (should_plug_request(req))
2222 		return;
2223 
2224 	/*
2225 	 * We may have a previously queued request message hanging
2226 	 * around.  Cancel it to avoid corrupting the msgr.
2227 	 */
2228 	if (req->r_sent)
2229 		ceph_msg_revoke(req->r_request);
2230 
2231 	req->r_flags |= CEPH_OSD_FLAG_KNOWN_REDIR;
2232 	if (req->r_attempts)
2233 		req->r_flags |= CEPH_OSD_FLAG_RETRY;
2234 	else
2235 		WARN_ON(req->r_flags & CEPH_OSD_FLAG_RETRY);
2236 
2237 	encode_request_partial(req, req->r_request);
2238 
2239 	dout("%s req %p tid %llu to pgid %llu.%x spgid %llu.%xs%d osd%d e%u flags 0x%x attempt %d\n",
2240 	     __func__, req, req->r_tid, req->r_t.pgid.pool, req->r_t.pgid.seed,
2241 	     req->r_t.spgid.pgid.pool, req->r_t.spgid.pgid.seed,
2242 	     req->r_t.spgid.shard, osd->o_osd, req->r_t.epoch, req->r_flags,
2243 	     req->r_attempts);
2244 
2245 	req->r_t.paused = false;
2246 	req->r_stamp = jiffies;
2247 	req->r_attempts++;
2248 
2249 	req->r_sent = osd->o_incarnation;
2250 	req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
2251 	ceph_con_send(&osd->o_con, ceph_msg_get(req->r_request));
2252 }
2253 
2254 static void maybe_request_map(struct ceph_osd_client *osdc)
2255 {
2256 	bool continuous = false;
2257 
2258 	verify_osdc_locked(osdc);
2259 	WARN_ON(!osdc->osdmap->epoch);
2260 
2261 	if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
2262 	    ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD) ||
2263 	    ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) {
2264 		dout("%s osdc %p continuous\n", __func__, osdc);
2265 		continuous = true;
2266 	} else {
2267 		dout("%s osdc %p onetime\n", __func__, osdc);
2268 	}
2269 
2270 	if (ceph_monc_want_map(&osdc->client->monc, CEPH_SUB_OSDMAP,
2271 			       osdc->osdmap->epoch + 1, continuous))
2272 		ceph_monc_renew_subs(&osdc->client->monc);
2273 }
2274 
2275 static void complete_request(struct ceph_osd_request *req, int err);
2276 static void send_map_check(struct ceph_osd_request *req);
2277 
2278 static void __submit_request(struct ceph_osd_request *req, bool wrlocked)
2279 {
2280 	struct ceph_osd_client *osdc = req->r_osdc;
2281 	struct ceph_osd *osd;
2282 	enum calc_target_result ct_res;
2283 	int err = 0;
2284 	bool need_send = false;
2285 	bool promoted = false;
2286 
2287 	WARN_ON(req->r_tid);
2288 	dout("%s req %p wrlocked %d\n", __func__, req, wrlocked);
2289 
2290 again:
2291 	ct_res = calc_target(osdc, &req->r_t, false);
2292 	if (ct_res == CALC_TARGET_POOL_DNE && !wrlocked)
2293 		goto promote;
2294 
2295 	osd = lookup_create_osd(osdc, req->r_t.osd, wrlocked);
2296 	if (IS_ERR(osd)) {
2297 		WARN_ON(PTR_ERR(osd) != -EAGAIN || wrlocked);
2298 		goto promote;
2299 	}
2300 
2301 	if (osdc->abort_err) {
2302 		dout("req %p abort_err %d\n", req, osdc->abort_err);
2303 		err = osdc->abort_err;
2304 	} else if (osdc->osdmap->epoch < osdc->epoch_barrier) {
2305 		dout("req %p epoch %u barrier %u\n", req, osdc->osdmap->epoch,
2306 		     osdc->epoch_barrier);
2307 		req->r_t.paused = true;
2308 		maybe_request_map(osdc);
2309 	} else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
2310 		   ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) {
2311 		dout("req %p pausewr\n", req);
2312 		req->r_t.paused = true;
2313 		maybe_request_map(osdc);
2314 	} else if ((req->r_flags & CEPH_OSD_FLAG_READ) &&
2315 		   ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) {
2316 		dout("req %p pauserd\n", req);
2317 		req->r_t.paused = true;
2318 		maybe_request_map(osdc);
2319 	} else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
2320 		   !(req->r_flags & (CEPH_OSD_FLAG_FULL_TRY |
2321 				     CEPH_OSD_FLAG_FULL_FORCE)) &&
2322 		   (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
2323 		    pool_full(osdc, req->r_t.base_oloc.pool))) {
2324 		dout("req %p full/pool_full\n", req);
2325 		if (ceph_test_opt(osdc->client, ABORT_ON_FULL)) {
2326 			err = -ENOSPC;
2327 		} else {
2328 			pr_warn_ratelimited("FULL or reached pool quota\n");
2329 			req->r_t.paused = true;
2330 			maybe_request_map(osdc);
2331 		}
2332 	} else if (!osd_homeless(osd)) {
2333 		need_send = true;
2334 	} else {
2335 		maybe_request_map(osdc);
2336 	}
2337 
2338 	mutex_lock(&osd->lock);
2339 	/*
2340 	 * Assign the tid atomically with send_request() to protect
2341 	 * multiple writes to the same object from racing with each
2342 	 * other, resulting in out of order ops on the OSDs.
2343 	 */
2344 	req->r_tid = atomic64_inc_return(&osdc->last_tid);
2345 	link_request(osd, req);
2346 	if (need_send)
2347 		send_request(req);
2348 	else if (err)
2349 		complete_request(req, err);
2350 	mutex_unlock(&osd->lock);
2351 
2352 	if (!err && ct_res == CALC_TARGET_POOL_DNE)
2353 		send_map_check(req);
2354 
2355 	if (promoted)
2356 		downgrade_write(&osdc->lock);
2357 	return;
2358 
2359 promote:
2360 	up_read(&osdc->lock);
2361 	down_write(&osdc->lock);
2362 	wrlocked = true;
2363 	promoted = true;
2364 	goto again;
2365 }
2366 
2367 static void account_request(struct ceph_osd_request *req)
2368 {
2369 	WARN_ON(req->r_flags & (CEPH_OSD_FLAG_ACK | CEPH_OSD_FLAG_ONDISK));
2370 	WARN_ON(!(req->r_flags & (CEPH_OSD_FLAG_READ | CEPH_OSD_FLAG_WRITE)));
2371 
2372 	req->r_flags |= CEPH_OSD_FLAG_ONDISK;
2373 	atomic_inc(&req->r_osdc->num_requests);
2374 
2375 	req->r_start_stamp = jiffies;
2376 	req->r_start_latency = ktime_get();
2377 }
2378 
2379 static void submit_request(struct ceph_osd_request *req, bool wrlocked)
2380 {
2381 	ceph_osdc_get_request(req);
2382 	account_request(req);
2383 	__submit_request(req, wrlocked);
2384 }
2385 
2386 static void finish_request(struct ceph_osd_request *req)
2387 {
2388 	struct ceph_osd_client *osdc = req->r_osdc;
2389 
2390 	WARN_ON(lookup_request_mc(&osdc->map_checks, req->r_tid));
2391 	dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
2392 
2393 	req->r_end_latency = ktime_get();
2394 
2395 	if (req->r_osd)
2396 		unlink_request(req->r_osd, req);
2397 	atomic_dec(&osdc->num_requests);
2398 
2399 	/*
2400 	 * If an OSD has failed or returned and a request has been sent
2401 	 * twice, it's possible to get a reply and end up here while the
2402 	 * request message is queued for delivery.  We will ignore the
2403 	 * reply, so not a big deal, but better to try and catch it.
2404 	 */
2405 	ceph_msg_revoke(req->r_request);
2406 	ceph_msg_revoke_incoming(req->r_reply);
2407 }
2408 
2409 static void __complete_request(struct ceph_osd_request *req)
2410 {
2411 	dout("%s req %p tid %llu cb %ps result %d\n", __func__, req,
2412 	     req->r_tid, req->r_callback, req->r_result);
2413 
2414 	if (req->r_callback)
2415 		req->r_callback(req);
2416 	complete_all(&req->r_completion);
2417 	ceph_osdc_put_request(req);
2418 }
2419 
2420 static void complete_request_workfn(struct work_struct *work)
2421 {
2422 	struct ceph_osd_request *req =
2423 	    container_of(work, struct ceph_osd_request, r_complete_work);
2424 
2425 	__complete_request(req);
2426 }
2427 
2428 /*
2429  * This is open-coded in handle_reply().
2430  */
2431 static void complete_request(struct ceph_osd_request *req, int err)
2432 {
2433 	dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err);
2434 
2435 	req->r_result = err;
2436 	finish_request(req);
2437 
2438 	INIT_WORK(&req->r_complete_work, complete_request_workfn);
2439 	queue_work(req->r_osdc->completion_wq, &req->r_complete_work);
2440 }
2441 
2442 static void cancel_map_check(struct ceph_osd_request *req)
2443 {
2444 	struct ceph_osd_client *osdc = req->r_osdc;
2445 	struct ceph_osd_request *lookup_req;
2446 
2447 	verify_osdc_wrlocked(osdc);
2448 
2449 	lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid);
2450 	if (!lookup_req)
2451 		return;
2452 
2453 	WARN_ON(lookup_req != req);
2454 	erase_request_mc(&osdc->map_checks, req);
2455 	ceph_osdc_put_request(req);
2456 }
2457 
2458 static void cancel_request(struct ceph_osd_request *req)
2459 {
2460 	dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
2461 
2462 	cancel_map_check(req);
2463 	finish_request(req);
2464 	complete_all(&req->r_completion);
2465 	ceph_osdc_put_request(req);
2466 }
2467 
2468 static void abort_request(struct ceph_osd_request *req, int err)
2469 {
2470 	dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err);
2471 
2472 	cancel_map_check(req);
2473 	complete_request(req, err);
2474 }
2475 
2476 static int abort_fn(struct ceph_osd_request *req, void *arg)
2477 {
2478 	int err = *(int *)arg;
2479 
2480 	abort_request(req, err);
2481 	return 0; /* continue iteration */
2482 }
2483 
2484 /*
2485  * Abort all in-flight requests with @err and arrange for all future
2486  * requests to be failed immediately.
2487  */
2488 void ceph_osdc_abort_requests(struct ceph_osd_client *osdc, int err)
2489 {
2490 	dout("%s osdc %p err %d\n", __func__, osdc, err);
2491 	down_write(&osdc->lock);
2492 	for_each_request(osdc, abort_fn, &err);
2493 	osdc->abort_err = err;
2494 	up_write(&osdc->lock);
2495 }
2496 EXPORT_SYMBOL(ceph_osdc_abort_requests);
2497 
2498 void ceph_osdc_clear_abort_err(struct ceph_osd_client *osdc)
2499 {
2500 	down_write(&osdc->lock);
2501 	osdc->abort_err = 0;
2502 	up_write(&osdc->lock);
2503 }
2504 EXPORT_SYMBOL(ceph_osdc_clear_abort_err);
2505 
2506 static void update_epoch_barrier(struct ceph_osd_client *osdc, u32 eb)
2507 {
2508 	if (likely(eb > osdc->epoch_barrier)) {
2509 		dout("updating epoch_barrier from %u to %u\n",
2510 				osdc->epoch_barrier, eb);
2511 		osdc->epoch_barrier = eb;
2512 		/* Request map if we're not to the barrier yet */
2513 		if (eb > osdc->osdmap->epoch)
2514 			maybe_request_map(osdc);
2515 	}
2516 }
2517 
2518 void ceph_osdc_update_epoch_barrier(struct ceph_osd_client *osdc, u32 eb)
2519 {
2520 	down_read(&osdc->lock);
2521 	if (unlikely(eb > osdc->epoch_barrier)) {
2522 		up_read(&osdc->lock);
2523 		down_write(&osdc->lock);
2524 		update_epoch_barrier(osdc, eb);
2525 		up_write(&osdc->lock);
2526 	} else {
2527 		up_read(&osdc->lock);
2528 	}
2529 }
2530 EXPORT_SYMBOL(ceph_osdc_update_epoch_barrier);
2531 
2532 /*
2533  * We can end up releasing caps as a result of abort_request().
2534  * In that case, we probably want to ensure that the cap release message
2535  * has an updated epoch barrier in it, so set the epoch barrier prior to
2536  * aborting the first request.
2537  */
2538 static int abort_on_full_fn(struct ceph_osd_request *req, void *arg)
2539 {
2540 	struct ceph_osd_client *osdc = req->r_osdc;
2541 	bool *victims = arg;
2542 
2543 	if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
2544 	    (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
2545 	     pool_full(osdc, req->r_t.base_oloc.pool))) {
2546 		if (!*victims) {
2547 			update_epoch_barrier(osdc, osdc->osdmap->epoch);
2548 			*victims = true;
2549 		}
2550 		abort_request(req, -ENOSPC);
2551 	}
2552 
2553 	return 0; /* continue iteration */
2554 }
2555 
2556 /*
2557  * Drop all pending requests that are stalled waiting on a full condition to
2558  * clear, and complete them with ENOSPC as the return code. Set the
2559  * osdc->epoch_barrier to the latest map epoch that we've seen if any were
2560  * cancelled.
2561  */
2562 static void ceph_osdc_abort_on_full(struct ceph_osd_client *osdc)
2563 {
2564 	bool victims = false;
2565 
2566 	if (ceph_test_opt(osdc->client, ABORT_ON_FULL) &&
2567 	    (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || have_pool_full(osdc)))
2568 		for_each_request(osdc, abort_on_full_fn, &victims);
2569 }
2570 
2571 static void check_pool_dne(struct ceph_osd_request *req)
2572 {
2573 	struct ceph_osd_client *osdc = req->r_osdc;
2574 	struct ceph_osdmap *map = osdc->osdmap;
2575 
2576 	verify_osdc_wrlocked(osdc);
2577 	WARN_ON(!map->epoch);
2578 
2579 	if (req->r_attempts) {
2580 		/*
2581 		 * We sent a request earlier, which means that
2582 		 * previously the pool existed, and now it does not
2583 		 * (i.e., it was deleted).
2584 		 */
2585 		req->r_map_dne_bound = map->epoch;
2586 		dout("%s req %p tid %llu pool disappeared\n", __func__, req,
2587 		     req->r_tid);
2588 	} else {
2589 		dout("%s req %p tid %llu map_dne_bound %u have %u\n", __func__,
2590 		     req, req->r_tid, req->r_map_dne_bound, map->epoch);
2591 	}
2592 
2593 	if (req->r_map_dne_bound) {
2594 		if (map->epoch >= req->r_map_dne_bound) {
2595 			/* we had a new enough map */
2596 			pr_info_ratelimited("tid %llu pool does not exist\n",
2597 					    req->r_tid);
2598 			complete_request(req, -ENOENT);
2599 		}
2600 	} else {
2601 		send_map_check(req);
2602 	}
2603 }
2604 
2605 static void map_check_cb(struct ceph_mon_generic_request *greq)
2606 {
2607 	struct ceph_osd_client *osdc = &greq->monc->client->osdc;
2608 	struct ceph_osd_request *req;
2609 	u64 tid = greq->private_data;
2610 
2611 	WARN_ON(greq->result || !greq->u.newest);
2612 
2613 	down_write(&osdc->lock);
2614 	req = lookup_request_mc(&osdc->map_checks, tid);
2615 	if (!req) {
2616 		dout("%s tid %llu dne\n", __func__, tid);
2617 		goto out_unlock;
2618 	}
2619 
2620 	dout("%s req %p tid %llu map_dne_bound %u newest %llu\n", __func__,
2621 	     req, req->r_tid, req->r_map_dne_bound, greq->u.newest);
2622 	if (!req->r_map_dne_bound)
2623 		req->r_map_dne_bound = greq->u.newest;
2624 	erase_request_mc(&osdc->map_checks, req);
2625 	check_pool_dne(req);
2626 
2627 	ceph_osdc_put_request(req);
2628 out_unlock:
2629 	up_write(&osdc->lock);
2630 }
2631 
2632 static void send_map_check(struct ceph_osd_request *req)
2633 {
2634 	struct ceph_osd_client *osdc = req->r_osdc;
2635 	struct ceph_osd_request *lookup_req;
2636 	int ret;
2637 
2638 	verify_osdc_wrlocked(osdc);
2639 
2640 	lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid);
2641 	if (lookup_req) {
2642 		WARN_ON(lookup_req != req);
2643 		return;
2644 	}
2645 
2646 	ceph_osdc_get_request(req);
2647 	insert_request_mc(&osdc->map_checks, req);
2648 	ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap",
2649 					  map_check_cb, req->r_tid);
2650 	WARN_ON(ret);
2651 }
2652 
2653 /*
2654  * lingering requests, watch/notify v2 infrastructure
2655  */
2656 static void linger_release(struct kref *kref)
2657 {
2658 	struct ceph_osd_linger_request *lreq =
2659 	    container_of(kref, struct ceph_osd_linger_request, kref);
2660 
2661 	dout("%s lreq %p reg_req %p ping_req %p\n", __func__, lreq,
2662 	     lreq->reg_req, lreq->ping_req);
2663 	WARN_ON(!RB_EMPTY_NODE(&lreq->node));
2664 	WARN_ON(!RB_EMPTY_NODE(&lreq->osdc_node));
2665 	WARN_ON(!RB_EMPTY_NODE(&lreq->mc_node));
2666 	WARN_ON(!list_empty(&lreq->scan_item));
2667 	WARN_ON(!list_empty(&lreq->pending_lworks));
2668 	WARN_ON(lreq->osd);
2669 
2670 	if (lreq->reg_req)
2671 		ceph_osdc_put_request(lreq->reg_req);
2672 	if (lreq->ping_req)
2673 		ceph_osdc_put_request(lreq->ping_req);
2674 	target_destroy(&lreq->t);
2675 	kfree(lreq);
2676 }
2677 
2678 static void linger_put(struct ceph_osd_linger_request *lreq)
2679 {
2680 	if (lreq)
2681 		kref_put(&lreq->kref, linger_release);
2682 }
2683 
2684 static struct ceph_osd_linger_request *
2685 linger_get(struct ceph_osd_linger_request *lreq)
2686 {
2687 	kref_get(&lreq->kref);
2688 	return lreq;
2689 }
2690 
2691 static struct ceph_osd_linger_request *
2692 linger_alloc(struct ceph_osd_client *osdc)
2693 {
2694 	struct ceph_osd_linger_request *lreq;
2695 
2696 	lreq = kzalloc(sizeof(*lreq), GFP_NOIO);
2697 	if (!lreq)
2698 		return NULL;
2699 
2700 	kref_init(&lreq->kref);
2701 	mutex_init(&lreq->lock);
2702 	RB_CLEAR_NODE(&lreq->node);
2703 	RB_CLEAR_NODE(&lreq->osdc_node);
2704 	RB_CLEAR_NODE(&lreq->mc_node);
2705 	INIT_LIST_HEAD(&lreq->scan_item);
2706 	INIT_LIST_HEAD(&lreq->pending_lworks);
2707 	init_completion(&lreq->reg_commit_wait);
2708 	init_completion(&lreq->notify_finish_wait);
2709 
2710 	lreq->osdc = osdc;
2711 	target_init(&lreq->t);
2712 
2713 	dout("%s lreq %p\n", __func__, lreq);
2714 	return lreq;
2715 }
2716 
2717 DEFINE_RB_INSDEL_FUNCS(linger, struct ceph_osd_linger_request, linger_id, node)
2718 DEFINE_RB_FUNCS(linger_osdc, struct ceph_osd_linger_request, linger_id, osdc_node)
2719 DEFINE_RB_FUNCS(linger_mc, struct ceph_osd_linger_request, linger_id, mc_node)
2720 
2721 /*
2722  * Create linger request <-> OSD session relation.
2723  *
2724  * @lreq has to be registered, @osd may be homeless.
2725  */
2726 static void link_linger(struct ceph_osd *osd,
2727 			struct ceph_osd_linger_request *lreq)
2728 {
2729 	verify_osd_locked(osd);
2730 	WARN_ON(!lreq->linger_id || lreq->osd);
2731 	dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd,
2732 	     osd->o_osd, lreq, lreq->linger_id);
2733 
2734 	if (!osd_homeless(osd))
2735 		__remove_osd_from_lru(osd);
2736 	else
2737 		atomic_inc(&osd->o_osdc->num_homeless);
2738 
2739 	get_osd(osd);
2740 	insert_linger(&osd->o_linger_requests, lreq);
2741 	lreq->osd = osd;
2742 }
2743 
2744 static void unlink_linger(struct ceph_osd *osd,
2745 			  struct ceph_osd_linger_request *lreq)
2746 {
2747 	verify_osd_locked(osd);
2748 	WARN_ON(lreq->osd != osd);
2749 	dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd,
2750 	     osd->o_osd, lreq, lreq->linger_id);
2751 
2752 	lreq->osd = NULL;
2753 	erase_linger(&osd->o_linger_requests, lreq);
2754 	put_osd(osd);
2755 
2756 	if (!osd_homeless(osd))
2757 		maybe_move_osd_to_lru(osd);
2758 	else
2759 		atomic_dec(&osd->o_osdc->num_homeless);
2760 }
2761 
2762 static bool __linger_registered(struct ceph_osd_linger_request *lreq)
2763 {
2764 	verify_osdc_locked(lreq->osdc);
2765 
2766 	return !RB_EMPTY_NODE(&lreq->osdc_node);
2767 }
2768 
2769 static bool linger_registered(struct ceph_osd_linger_request *lreq)
2770 {
2771 	struct ceph_osd_client *osdc = lreq->osdc;
2772 	bool registered;
2773 
2774 	down_read(&osdc->lock);
2775 	registered = __linger_registered(lreq);
2776 	up_read(&osdc->lock);
2777 
2778 	return registered;
2779 }
2780 
2781 static void linger_register(struct ceph_osd_linger_request *lreq)
2782 {
2783 	struct ceph_osd_client *osdc = lreq->osdc;
2784 
2785 	verify_osdc_wrlocked(osdc);
2786 	WARN_ON(lreq->linger_id);
2787 
2788 	linger_get(lreq);
2789 	lreq->linger_id = ++osdc->last_linger_id;
2790 	insert_linger_osdc(&osdc->linger_requests, lreq);
2791 }
2792 
2793 static void linger_unregister(struct ceph_osd_linger_request *lreq)
2794 {
2795 	struct ceph_osd_client *osdc = lreq->osdc;
2796 
2797 	verify_osdc_wrlocked(osdc);
2798 
2799 	erase_linger_osdc(&osdc->linger_requests, lreq);
2800 	linger_put(lreq);
2801 }
2802 
2803 static void cancel_linger_request(struct ceph_osd_request *req)
2804 {
2805 	struct ceph_osd_linger_request *lreq = req->r_priv;
2806 
2807 	WARN_ON(!req->r_linger);
2808 	cancel_request(req);
2809 	linger_put(lreq);
2810 }
2811 
2812 struct linger_work {
2813 	struct work_struct work;
2814 	struct ceph_osd_linger_request *lreq;
2815 	struct list_head pending_item;
2816 	unsigned long queued_stamp;
2817 
2818 	union {
2819 		struct {
2820 			u64 notify_id;
2821 			u64 notifier_id;
2822 			void *payload; /* points into @msg front */
2823 			size_t payload_len;
2824 
2825 			struct ceph_msg *msg; /* for ceph_msg_put() */
2826 		} notify;
2827 		struct {
2828 			int err;
2829 		} error;
2830 	};
2831 };
2832 
2833 static struct linger_work *lwork_alloc(struct ceph_osd_linger_request *lreq,
2834 				       work_func_t workfn)
2835 {
2836 	struct linger_work *lwork;
2837 
2838 	lwork = kzalloc(sizeof(*lwork), GFP_NOIO);
2839 	if (!lwork)
2840 		return NULL;
2841 
2842 	INIT_WORK(&lwork->work, workfn);
2843 	INIT_LIST_HEAD(&lwork->pending_item);
2844 	lwork->lreq = linger_get(lreq);
2845 
2846 	return lwork;
2847 }
2848 
2849 static void lwork_free(struct linger_work *lwork)
2850 {
2851 	struct ceph_osd_linger_request *lreq = lwork->lreq;
2852 
2853 	mutex_lock(&lreq->lock);
2854 	list_del(&lwork->pending_item);
2855 	mutex_unlock(&lreq->lock);
2856 
2857 	linger_put(lreq);
2858 	kfree(lwork);
2859 }
2860 
2861 static void lwork_queue(struct linger_work *lwork)
2862 {
2863 	struct ceph_osd_linger_request *lreq = lwork->lreq;
2864 	struct ceph_osd_client *osdc = lreq->osdc;
2865 
2866 	verify_lreq_locked(lreq);
2867 	WARN_ON(!list_empty(&lwork->pending_item));
2868 
2869 	lwork->queued_stamp = jiffies;
2870 	list_add_tail(&lwork->pending_item, &lreq->pending_lworks);
2871 	queue_work(osdc->notify_wq, &lwork->work);
2872 }
2873 
2874 static void do_watch_notify(struct work_struct *w)
2875 {
2876 	struct linger_work *lwork = container_of(w, struct linger_work, work);
2877 	struct ceph_osd_linger_request *lreq = lwork->lreq;
2878 
2879 	if (!linger_registered(lreq)) {
2880 		dout("%s lreq %p not registered\n", __func__, lreq);
2881 		goto out;
2882 	}
2883 
2884 	WARN_ON(!lreq->is_watch);
2885 	dout("%s lreq %p notify_id %llu notifier_id %llu payload_len %zu\n",
2886 	     __func__, lreq, lwork->notify.notify_id, lwork->notify.notifier_id,
2887 	     lwork->notify.payload_len);
2888 	lreq->wcb(lreq->data, lwork->notify.notify_id, lreq->linger_id,
2889 		  lwork->notify.notifier_id, lwork->notify.payload,
2890 		  lwork->notify.payload_len);
2891 
2892 out:
2893 	ceph_msg_put(lwork->notify.msg);
2894 	lwork_free(lwork);
2895 }
2896 
2897 static void do_watch_error(struct work_struct *w)
2898 {
2899 	struct linger_work *lwork = container_of(w, struct linger_work, work);
2900 	struct ceph_osd_linger_request *lreq = lwork->lreq;
2901 
2902 	if (!linger_registered(lreq)) {
2903 		dout("%s lreq %p not registered\n", __func__, lreq);
2904 		goto out;
2905 	}
2906 
2907 	dout("%s lreq %p err %d\n", __func__, lreq, lwork->error.err);
2908 	lreq->errcb(lreq->data, lreq->linger_id, lwork->error.err);
2909 
2910 out:
2911 	lwork_free(lwork);
2912 }
2913 
2914 static void queue_watch_error(struct ceph_osd_linger_request *lreq)
2915 {
2916 	struct linger_work *lwork;
2917 
2918 	lwork = lwork_alloc(lreq, do_watch_error);
2919 	if (!lwork) {
2920 		pr_err("failed to allocate error-lwork\n");
2921 		return;
2922 	}
2923 
2924 	lwork->error.err = lreq->last_error;
2925 	lwork_queue(lwork);
2926 }
2927 
2928 static void linger_reg_commit_complete(struct ceph_osd_linger_request *lreq,
2929 				       int result)
2930 {
2931 	if (!completion_done(&lreq->reg_commit_wait)) {
2932 		lreq->reg_commit_error = (result <= 0 ? result : 0);
2933 		complete_all(&lreq->reg_commit_wait);
2934 	}
2935 }
2936 
2937 static void linger_commit_cb(struct ceph_osd_request *req)
2938 {
2939 	struct ceph_osd_linger_request *lreq = req->r_priv;
2940 
2941 	mutex_lock(&lreq->lock);
2942 	dout("%s lreq %p linger_id %llu result %d\n", __func__, lreq,
2943 	     lreq->linger_id, req->r_result);
2944 	linger_reg_commit_complete(lreq, req->r_result);
2945 	lreq->committed = true;
2946 
2947 	if (!lreq->is_watch) {
2948 		struct ceph_osd_data *osd_data =
2949 		    osd_req_op_data(req, 0, notify, response_data);
2950 		void *p = page_address(osd_data->pages[0]);
2951 
2952 		WARN_ON(req->r_ops[0].op != CEPH_OSD_OP_NOTIFY ||
2953 			osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
2954 
2955 		/* make note of the notify_id */
2956 		if (req->r_ops[0].outdata_len >= sizeof(u64)) {
2957 			lreq->notify_id = ceph_decode_64(&p);
2958 			dout("lreq %p notify_id %llu\n", lreq,
2959 			     lreq->notify_id);
2960 		} else {
2961 			dout("lreq %p no notify_id\n", lreq);
2962 		}
2963 	}
2964 
2965 	mutex_unlock(&lreq->lock);
2966 	linger_put(lreq);
2967 }
2968 
2969 static int normalize_watch_error(int err)
2970 {
2971 	/*
2972 	 * Translate ENOENT -> ENOTCONN so that a delete->disconnection
2973 	 * notification and a failure to reconnect because we raced with
2974 	 * the delete appear the same to the user.
2975 	 */
2976 	if (err == -ENOENT)
2977 		err = -ENOTCONN;
2978 
2979 	return err;
2980 }
2981 
2982 static void linger_reconnect_cb(struct ceph_osd_request *req)
2983 {
2984 	struct ceph_osd_linger_request *lreq = req->r_priv;
2985 
2986 	mutex_lock(&lreq->lock);
2987 	dout("%s lreq %p linger_id %llu result %d last_error %d\n", __func__,
2988 	     lreq, lreq->linger_id, req->r_result, lreq->last_error);
2989 	if (req->r_result < 0) {
2990 		if (!lreq->last_error) {
2991 			lreq->last_error = normalize_watch_error(req->r_result);
2992 			queue_watch_error(lreq);
2993 		}
2994 	}
2995 
2996 	mutex_unlock(&lreq->lock);
2997 	linger_put(lreq);
2998 }
2999 
3000 static void send_linger(struct ceph_osd_linger_request *lreq)
3001 {
3002 	struct ceph_osd_request *req = lreq->reg_req;
3003 	struct ceph_osd_req_op *op = &req->r_ops[0];
3004 
3005 	verify_osdc_wrlocked(req->r_osdc);
3006 	dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
3007 
3008 	if (req->r_osd)
3009 		cancel_linger_request(req);
3010 
3011 	request_reinit(req);
3012 	ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
3013 	ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
3014 	req->r_flags = lreq->t.flags;
3015 	req->r_mtime = lreq->mtime;
3016 
3017 	mutex_lock(&lreq->lock);
3018 	if (lreq->is_watch && lreq->committed) {
3019 		WARN_ON(op->op != CEPH_OSD_OP_WATCH ||
3020 			op->watch.cookie != lreq->linger_id);
3021 		op->watch.op = CEPH_OSD_WATCH_OP_RECONNECT;
3022 		op->watch.gen = ++lreq->register_gen;
3023 		dout("lreq %p reconnect register_gen %u\n", lreq,
3024 		     op->watch.gen);
3025 		req->r_callback = linger_reconnect_cb;
3026 	} else {
3027 		if (!lreq->is_watch)
3028 			lreq->notify_id = 0;
3029 		else
3030 			WARN_ON(op->watch.op != CEPH_OSD_WATCH_OP_WATCH);
3031 		dout("lreq %p register\n", lreq);
3032 		req->r_callback = linger_commit_cb;
3033 	}
3034 	mutex_unlock(&lreq->lock);
3035 
3036 	req->r_priv = linger_get(lreq);
3037 	req->r_linger = true;
3038 
3039 	submit_request(req, true);
3040 }
3041 
3042 static void linger_ping_cb(struct ceph_osd_request *req)
3043 {
3044 	struct ceph_osd_linger_request *lreq = req->r_priv;
3045 
3046 	mutex_lock(&lreq->lock);
3047 	dout("%s lreq %p linger_id %llu result %d ping_sent %lu last_error %d\n",
3048 	     __func__, lreq, lreq->linger_id, req->r_result, lreq->ping_sent,
3049 	     lreq->last_error);
3050 	if (lreq->register_gen == req->r_ops[0].watch.gen) {
3051 		if (!req->r_result) {
3052 			lreq->watch_valid_thru = lreq->ping_sent;
3053 		} else if (!lreq->last_error) {
3054 			lreq->last_error = normalize_watch_error(req->r_result);
3055 			queue_watch_error(lreq);
3056 		}
3057 	} else {
3058 		dout("lreq %p register_gen %u ignoring old pong %u\n", lreq,
3059 		     lreq->register_gen, req->r_ops[0].watch.gen);
3060 	}
3061 
3062 	mutex_unlock(&lreq->lock);
3063 	linger_put(lreq);
3064 }
3065 
3066 static void send_linger_ping(struct ceph_osd_linger_request *lreq)
3067 {
3068 	struct ceph_osd_client *osdc = lreq->osdc;
3069 	struct ceph_osd_request *req = lreq->ping_req;
3070 	struct ceph_osd_req_op *op = &req->r_ops[0];
3071 
3072 	if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) {
3073 		dout("%s PAUSERD\n", __func__);
3074 		return;
3075 	}
3076 
3077 	lreq->ping_sent = jiffies;
3078 	dout("%s lreq %p linger_id %llu ping_sent %lu register_gen %u\n",
3079 	     __func__, lreq, lreq->linger_id, lreq->ping_sent,
3080 	     lreq->register_gen);
3081 
3082 	if (req->r_osd)
3083 		cancel_linger_request(req);
3084 
3085 	request_reinit(req);
3086 	target_copy(&req->r_t, &lreq->t);
3087 
3088 	WARN_ON(op->op != CEPH_OSD_OP_WATCH ||
3089 		op->watch.cookie != lreq->linger_id ||
3090 		op->watch.op != CEPH_OSD_WATCH_OP_PING);
3091 	op->watch.gen = lreq->register_gen;
3092 	req->r_callback = linger_ping_cb;
3093 	req->r_priv = linger_get(lreq);
3094 	req->r_linger = true;
3095 
3096 	ceph_osdc_get_request(req);
3097 	account_request(req);
3098 	req->r_tid = atomic64_inc_return(&osdc->last_tid);
3099 	link_request(lreq->osd, req);
3100 	send_request(req);
3101 }
3102 
3103 static void linger_submit(struct ceph_osd_linger_request *lreq)
3104 {
3105 	struct ceph_osd_client *osdc = lreq->osdc;
3106 	struct ceph_osd *osd;
3107 
3108 	down_write(&osdc->lock);
3109 	linger_register(lreq);
3110 	if (lreq->is_watch) {
3111 		lreq->reg_req->r_ops[0].watch.cookie = lreq->linger_id;
3112 		lreq->ping_req->r_ops[0].watch.cookie = lreq->linger_id;
3113 	} else {
3114 		lreq->reg_req->r_ops[0].notify.cookie = lreq->linger_id;
3115 	}
3116 
3117 	calc_target(osdc, &lreq->t, false);
3118 	osd = lookup_create_osd(osdc, lreq->t.osd, true);
3119 	link_linger(osd, lreq);
3120 
3121 	send_linger(lreq);
3122 	up_write(&osdc->lock);
3123 }
3124 
3125 static void cancel_linger_map_check(struct ceph_osd_linger_request *lreq)
3126 {
3127 	struct ceph_osd_client *osdc = lreq->osdc;
3128 	struct ceph_osd_linger_request *lookup_lreq;
3129 
3130 	verify_osdc_wrlocked(osdc);
3131 
3132 	lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks,
3133 				       lreq->linger_id);
3134 	if (!lookup_lreq)
3135 		return;
3136 
3137 	WARN_ON(lookup_lreq != lreq);
3138 	erase_linger_mc(&osdc->linger_map_checks, lreq);
3139 	linger_put(lreq);
3140 }
3141 
3142 /*
3143  * @lreq has to be both registered and linked.
3144  */
3145 static void __linger_cancel(struct ceph_osd_linger_request *lreq)
3146 {
3147 	if (lreq->is_watch && lreq->ping_req->r_osd)
3148 		cancel_linger_request(lreq->ping_req);
3149 	if (lreq->reg_req->r_osd)
3150 		cancel_linger_request(lreq->reg_req);
3151 	cancel_linger_map_check(lreq);
3152 	unlink_linger(lreq->osd, lreq);
3153 	linger_unregister(lreq);
3154 }
3155 
3156 static void linger_cancel(struct ceph_osd_linger_request *lreq)
3157 {
3158 	struct ceph_osd_client *osdc = lreq->osdc;
3159 
3160 	down_write(&osdc->lock);
3161 	if (__linger_registered(lreq))
3162 		__linger_cancel(lreq);
3163 	up_write(&osdc->lock);
3164 }
3165 
3166 static void send_linger_map_check(struct ceph_osd_linger_request *lreq);
3167 
3168 static void check_linger_pool_dne(struct ceph_osd_linger_request *lreq)
3169 {
3170 	struct ceph_osd_client *osdc = lreq->osdc;
3171 	struct ceph_osdmap *map = osdc->osdmap;
3172 
3173 	verify_osdc_wrlocked(osdc);
3174 	WARN_ON(!map->epoch);
3175 
3176 	if (lreq->register_gen) {
3177 		lreq->map_dne_bound = map->epoch;
3178 		dout("%s lreq %p linger_id %llu pool disappeared\n", __func__,
3179 		     lreq, lreq->linger_id);
3180 	} else {
3181 		dout("%s lreq %p linger_id %llu map_dne_bound %u have %u\n",
3182 		     __func__, lreq, lreq->linger_id, lreq->map_dne_bound,
3183 		     map->epoch);
3184 	}
3185 
3186 	if (lreq->map_dne_bound) {
3187 		if (map->epoch >= lreq->map_dne_bound) {
3188 			/* we had a new enough map */
3189 			pr_info("linger_id %llu pool does not exist\n",
3190 				lreq->linger_id);
3191 			linger_reg_commit_complete(lreq, -ENOENT);
3192 			__linger_cancel(lreq);
3193 		}
3194 	} else {
3195 		send_linger_map_check(lreq);
3196 	}
3197 }
3198 
3199 static void linger_map_check_cb(struct ceph_mon_generic_request *greq)
3200 {
3201 	struct ceph_osd_client *osdc = &greq->monc->client->osdc;
3202 	struct ceph_osd_linger_request *lreq;
3203 	u64 linger_id = greq->private_data;
3204 
3205 	WARN_ON(greq->result || !greq->u.newest);
3206 
3207 	down_write(&osdc->lock);
3208 	lreq = lookup_linger_mc(&osdc->linger_map_checks, linger_id);
3209 	if (!lreq) {
3210 		dout("%s linger_id %llu dne\n", __func__, linger_id);
3211 		goto out_unlock;
3212 	}
3213 
3214 	dout("%s lreq %p linger_id %llu map_dne_bound %u newest %llu\n",
3215 	     __func__, lreq, lreq->linger_id, lreq->map_dne_bound,
3216 	     greq->u.newest);
3217 	if (!lreq->map_dne_bound)
3218 		lreq->map_dne_bound = greq->u.newest;
3219 	erase_linger_mc(&osdc->linger_map_checks, lreq);
3220 	check_linger_pool_dne(lreq);
3221 
3222 	linger_put(lreq);
3223 out_unlock:
3224 	up_write(&osdc->lock);
3225 }
3226 
3227 static void send_linger_map_check(struct ceph_osd_linger_request *lreq)
3228 {
3229 	struct ceph_osd_client *osdc = lreq->osdc;
3230 	struct ceph_osd_linger_request *lookup_lreq;
3231 	int ret;
3232 
3233 	verify_osdc_wrlocked(osdc);
3234 
3235 	lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks,
3236 				       lreq->linger_id);
3237 	if (lookup_lreq) {
3238 		WARN_ON(lookup_lreq != lreq);
3239 		return;
3240 	}
3241 
3242 	linger_get(lreq);
3243 	insert_linger_mc(&osdc->linger_map_checks, lreq);
3244 	ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap",
3245 					  linger_map_check_cb, lreq->linger_id);
3246 	WARN_ON(ret);
3247 }
3248 
3249 static int linger_reg_commit_wait(struct ceph_osd_linger_request *lreq)
3250 {
3251 	int ret;
3252 
3253 	dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
3254 	ret = wait_for_completion_interruptible(&lreq->reg_commit_wait);
3255 	return ret ?: lreq->reg_commit_error;
3256 }
3257 
3258 static int linger_notify_finish_wait(struct ceph_osd_linger_request *lreq)
3259 {
3260 	int ret;
3261 
3262 	dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
3263 	ret = wait_for_completion_interruptible(&lreq->notify_finish_wait);
3264 	return ret ?: lreq->notify_finish_error;
3265 }
3266 
3267 /*
3268  * Timeout callback, called every N seconds.  When 1 or more OSD
3269  * requests has been active for more than N seconds, we send a keepalive
3270  * (tag + timestamp) to its OSD to ensure any communications channel
3271  * reset is detected.
3272  */
3273 static void handle_timeout(struct work_struct *work)
3274 {
3275 	struct ceph_osd_client *osdc =
3276 		container_of(work, struct ceph_osd_client, timeout_work.work);
3277 	struct ceph_options *opts = osdc->client->options;
3278 	unsigned long cutoff = jiffies - opts->osd_keepalive_timeout;
3279 	unsigned long expiry_cutoff = jiffies - opts->osd_request_timeout;
3280 	LIST_HEAD(slow_osds);
3281 	struct rb_node *n, *p;
3282 
3283 	dout("%s osdc %p\n", __func__, osdc);
3284 	down_write(&osdc->lock);
3285 
3286 	/*
3287 	 * ping osds that are a bit slow.  this ensures that if there
3288 	 * is a break in the TCP connection we will notice, and reopen
3289 	 * a connection with that osd (from the fault callback).
3290 	 */
3291 	for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
3292 		struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
3293 		bool found = false;
3294 
3295 		for (p = rb_first(&osd->o_requests); p; ) {
3296 			struct ceph_osd_request *req =
3297 			    rb_entry(p, struct ceph_osd_request, r_node);
3298 
3299 			p = rb_next(p); /* abort_request() */
3300 
3301 			if (time_before(req->r_stamp, cutoff)) {
3302 				dout(" req %p tid %llu on osd%d is laggy\n",
3303 				     req, req->r_tid, osd->o_osd);
3304 				found = true;
3305 			}
3306 			if (opts->osd_request_timeout &&
3307 			    time_before(req->r_start_stamp, expiry_cutoff)) {
3308 				pr_err_ratelimited("tid %llu on osd%d timeout\n",
3309 				       req->r_tid, osd->o_osd);
3310 				abort_request(req, -ETIMEDOUT);
3311 			}
3312 		}
3313 		for (p = rb_first(&osd->o_linger_requests); p; p = rb_next(p)) {
3314 			struct ceph_osd_linger_request *lreq =
3315 			    rb_entry(p, struct ceph_osd_linger_request, node);
3316 
3317 			dout(" lreq %p linger_id %llu is served by osd%d\n",
3318 			     lreq, lreq->linger_id, osd->o_osd);
3319 			found = true;
3320 
3321 			mutex_lock(&lreq->lock);
3322 			if (lreq->is_watch && lreq->committed && !lreq->last_error)
3323 				send_linger_ping(lreq);
3324 			mutex_unlock(&lreq->lock);
3325 		}
3326 
3327 		if (found)
3328 			list_move_tail(&osd->o_keepalive_item, &slow_osds);
3329 	}
3330 
3331 	if (opts->osd_request_timeout) {
3332 		for (p = rb_first(&osdc->homeless_osd.o_requests); p; ) {
3333 			struct ceph_osd_request *req =
3334 			    rb_entry(p, struct ceph_osd_request, r_node);
3335 
3336 			p = rb_next(p); /* abort_request() */
3337 
3338 			if (time_before(req->r_start_stamp, expiry_cutoff)) {
3339 				pr_err_ratelimited("tid %llu on osd%d timeout\n",
3340 				       req->r_tid, osdc->homeless_osd.o_osd);
3341 				abort_request(req, -ETIMEDOUT);
3342 			}
3343 		}
3344 	}
3345 
3346 	if (atomic_read(&osdc->num_homeless) || !list_empty(&slow_osds))
3347 		maybe_request_map(osdc);
3348 
3349 	while (!list_empty(&slow_osds)) {
3350 		struct ceph_osd *osd = list_first_entry(&slow_osds,
3351 							struct ceph_osd,
3352 							o_keepalive_item);
3353 		list_del_init(&osd->o_keepalive_item);
3354 		ceph_con_keepalive(&osd->o_con);
3355 	}
3356 
3357 	up_write(&osdc->lock);
3358 	schedule_delayed_work(&osdc->timeout_work,
3359 			      osdc->client->options->osd_keepalive_timeout);
3360 }
3361 
3362 static void handle_osds_timeout(struct work_struct *work)
3363 {
3364 	struct ceph_osd_client *osdc =
3365 		container_of(work, struct ceph_osd_client,
3366 			     osds_timeout_work.work);
3367 	unsigned long delay = osdc->client->options->osd_idle_ttl / 4;
3368 	struct ceph_osd *osd, *nosd;
3369 
3370 	dout("%s osdc %p\n", __func__, osdc);
3371 	down_write(&osdc->lock);
3372 	list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
3373 		if (time_before(jiffies, osd->lru_ttl))
3374 			break;
3375 
3376 		WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests));
3377 		WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests));
3378 		close_osd(osd);
3379 	}
3380 
3381 	up_write(&osdc->lock);
3382 	schedule_delayed_work(&osdc->osds_timeout_work,
3383 			      round_jiffies_relative(delay));
3384 }
3385 
3386 static int ceph_oloc_decode(void **p, void *end,
3387 			    struct ceph_object_locator *oloc)
3388 {
3389 	u8 struct_v, struct_cv;
3390 	u32 len;
3391 	void *struct_end;
3392 	int ret = 0;
3393 
3394 	ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
3395 	struct_v = ceph_decode_8(p);
3396 	struct_cv = ceph_decode_8(p);
3397 	if (struct_v < 3) {
3398 		pr_warn("got v %d < 3 cv %d of ceph_object_locator\n",
3399 			struct_v, struct_cv);
3400 		goto e_inval;
3401 	}
3402 	if (struct_cv > 6) {
3403 		pr_warn("got v %d cv %d > 6 of ceph_object_locator\n",
3404 			struct_v, struct_cv);
3405 		goto e_inval;
3406 	}
3407 	len = ceph_decode_32(p);
3408 	ceph_decode_need(p, end, len, e_inval);
3409 	struct_end = *p + len;
3410 
3411 	oloc->pool = ceph_decode_64(p);
3412 	*p += 4; /* skip preferred */
3413 
3414 	len = ceph_decode_32(p);
3415 	if (len > 0) {
3416 		pr_warn("ceph_object_locator::key is set\n");
3417 		goto e_inval;
3418 	}
3419 
3420 	if (struct_v >= 5) {
3421 		bool changed = false;
3422 
3423 		len = ceph_decode_32(p);
3424 		if (len > 0) {
3425 			ceph_decode_need(p, end, len, e_inval);
3426 			if (!oloc->pool_ns ||
3427 			    ceph_compare_string(oloc->pool_ns, *p, len))
3428 				changed = true;
3429 			*p += len;
3430 		} else {
3431 			if (oloc->pool_ns)
3432 				changed = true;
3433 		}
3434 		if (changed) {
3435 			/* redirect changes namespace */
3436 			pr_warn("ceph_object_locator::nspace is changed\n");
3437 			goto e_inval;
3438 		}
3439 	}
3440 
3441 	if (struct_v >= 6) {
3442 		s64 hash = ceph_decode_64(p);
3443 		if (hash != -1) {
3444 			pr_warn("ceph_object_locator::hash is set\n");
3445 			goto e_inval;
3446 		}
3447 	}
3448 
3449 	/* skip the rest */
3450 	*p = struct_end;
3451 out:
3452 	return ret;
3453 
3454 e_inval:
3455 	ret = -EINVAL;
3456 	goto out;
3457 }
3458 
3459 static int ceph_redirect_decode(void **p, void *end,
3460 				struct ceph_request_redirect *redir)
3461 {
3462 	u8 struct_v, struct_cv;
3463 	u32 len;
3464 	void *struct_end;
3465 	int ret;
3466 
3467 	ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
3468 	struct_v = ceph_decode_8(p);
3469 	struct_cv = ceph_decode_8(p);
3470 	if (struct_cv > 1) {
3471 		pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n",
3472 			struct_v, struct_cv);
3473 		goto e_inval;
3474 	}
3475 	len = ceph_decode_32(p);
3476 	ceph_decode_need(p, end, len, e_inval);
3477 	struct_end = *p + len;
3478 
3479 	ret = ceph_oloc_decode(p, end, &redir->oloc);
3480 	if (ret)
3481 		goto out;
3482 
3483 	len = ceph_decode_32(p);
3484 	if (len > 0) {
3485 		pr_warn("ceph_request_redirect::object_name is set\n");
3486 		goto e_inval;
3487 	}
3488 
3489 	/* skip the rest */
3490 	*p = struct_end;
3491 out:
3492 	return ret;
3493 
3494 e_inval:
3495 	ret = -EINVAL;
3496 	goto out;
3497 }
3498 
3499 struct MOSDOpReply {
3500 	struct ceph_pg pgid;
3501 	u64 flags;
3502 	int result;
3503 	u32 epoch;
3504 	int num_ops;
3505 	u32 outdata_len[CEPH_OSD_MAX_OPS];
3506 	s32 rval[CEPH_OSD_MAX_OPS];
3507 	int retry_attempt;
3508 	struct ceph_eversion replay_version;
3509 	u64 user_version;
3510 	struct ceph_request_redirect redirect;
3511 };
3512 
3513 static int decode_MOSDOpReply(const struct ceph_msg *msg, struct MOSDOpReply *m)
3514 {
3515 	void *p = msg->front.iov_base;
3516 	void *const end = p + msg->front.iov_len;
3517 	u16 version = le16_to_cpu(msg->hdr.version);
3518 	struct ceph_eversion bad_replay_version;
3519 	u8 decode_redir;
3520 	u32 len;
3521 	int ret;
3522 	int i;
3523 
3524 	ceph_decode_32_safe(&p, end, len, e_inval);
3525 	ceph_decode_need(&p, end, len, e_inval);
3526 	p += len; /* skip oid */
3527 
3528 	ret = ceph_decode_pgid(&p, end, &m->pgid);
3529 	if (ret)
3530 		return ret;
3531 
3532 	ceph_decode_64_safe(&p, end, m->flags, e_inval);
3533 	ceph_decode_32_safe(&p, end, m->result, e_inval);
3534 	ceph_decode_need(&p, end, sizeof(bad_replay_version), e_inval);
3535 	memcpy(&bad_replay_version, p, sizeof(bad_replay_version));
3536 	p += sizeof(bad_replay_version);
3537 	ceph_decode_32_safe(&p, end, m->epoch, e_inval);
3538 
3539 	ceph_decode_32_safe(&p, end, m->num_ops, e_inval);
3540 	if (m->num_ops > ARRAY_SIZE(m->outdata_len))
3541 		goto e_inval;
3542 
3543 	ceph_decode_need(&p, end, m->num_ops * sizeof(struct ceph_osd_op),
3544 			 e_inval);
3545 	for (i = 0; i < m->num_ops; i++) {
3546 		struct ceph_osd_op *op = p;
3547 
3548 		m->outdata_len[i] = le32_to_cpu(op->payload_len);
3549 		p += sizeof(*op);
3550 	}
3551 
3552 	ceph_decode_32_safe(&p, end, m->retry_attempt, e_inval);
3553 	for (i = 0; i < m->num_ops; i++)
3554 		ceph_decode_32_safe(&p, end, m->rval[i], e_inval);
3555 
3556 	if (version >= 5) {
3557 		ceph_decode_need(&p, end, sizeof(m->replay_version), e_inval);
3558 		memcpy(&m->replay_version, p, sizeof(m->replay_version));
3559 		p += sizeof(m->replay_version);
3560 		ceph_decode_64_safe(&p, end, m->user_version, e_inval);
3561 	} else {
3562 		m->replay_version = bad_replay_version; /* struct */
3563 		m->user_version = le64_to_cpu(m->replay_version.version);
3564 	}
3565 
3566 	if (version >= 6) {
3567 		if (version >= 7)
3568 			ceph_decode_8_safe(&p, end, decode_redir, e_inval);
3569 		else
3570 			decode_redir = 1;
3571 	} else {
3572 		decode_redir = 0;
3573 	}
3574 
3575 	if (decode_redir) {
3576 		ret = ceph_redirect_decode(&p, end, &m->redirect);
3577 		if (ret)
3578 			return ret;
3579 	} else {
3580 		ceph_oloc_init(&m->redirect.oloc);
3581 	}
3582 
3583 	return 0;
3584 
3585 e_inval:
3586 	return -EINVAL;
3587 }
3588 
3589 /*
3590  * Handle MOSDOpReply.  Set ->r_result and call the callback if it is
3591  * specified.
3592  */
3593 static void handle_reply(struct ceph_osd *osd, struct ceph_msg *msg)
3594 {
3595 	struct ceph_osd_client *osdc = osd->o_osdc;
3596 	struct ceph_osd_request *req;
3597 	struct MOSDOpReply m;
3598 	u64 tid = le64_to_cpu(msg->hdr.tid);
3599 	u32 data_len = 0;
3600 	int ret;
3601 	int i;
3602 
3603 	dout("%s msg %p tid %llu\n", __func__, msg, tid);
3604 
3605 	down_read(&osdc->lock);
3606 	if (!osd_registered(osd)) {
3607 		dout("%s osd%d unknown\n", __func__, osd->o_osd);
3608 		goto out_unlock_osdc;
3609 	}
3610 	WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num));
3611 
3612 	mutex_lock(&osd->lock);
3613 	req = lookup_request(&osd->o_requests, tid);
3614 	if (!req) {
3615 		dout("%s osd%d tid %llu unknown\n", __func__, osd->o_osd, tid);
3616 		goto out_unlock_session;
3617 	}
3618 
3619 	m.redirect.oloc.pool_ns = req->r_t.target_oloc.pool_ns;
3620 	ret = decode_MOSDOpReply(msg, &m);
3621 	m.redirect.oloc.pool_ns = NULL;
3622 	if (ret) {
3623 		pr_err("failed to decode MOSDOpReply for tid %llu: %d\n",
3624 		       req->r_tid, ret);
3625 		ceph_msg_dump(msg);
3626 		goto fail_request;
3627 	}
3628 	dout("%s req %p tid %llu flags 0x%llx pgid %llu.%x epoch %u attempt %d v %u'%llu uv %llu\n",
3629 	     __func__, req, req->r_tid, m.flags, m.pgid.pool, m.pgid.seed,
3630 	     m.epoch, m.retry_attempt, le32_to_cpu(m.replay_version.epoch),
3631 	     le64_to_cpu(m.replay_version.version), m.user_version);
3632 
3633 	if (m.retry_attempt >= 0) {
3634 		if (m.retry_attempt != req->r_attempts - 1) {
3635 			dout("req %p tid %llu retry_attempt %d != %d, ignoring\n",
3636 			     req, req->r_tid, m.retry_attempt,
3637 			     req->r_attempts - 1);
3638 			goto out_unlock_session;
3639 		}
3640 	} else {
3641 		WARN_ON(1); /* MOSDOpReply v4 is assumed */
3642 	}
3643 
3644 	if (!ceph_oloc_empty(&m.redirect.oloc)) {
3645 		dout("req %p tid %llu redirect pool %lld\n", req, req->r_tid,
3646 		     m.redirect.oloc.pool);
3647 		unlink_request(osd, req);
3648 		mutex_unlock(&osd->lock);
3649 
3650 		/*
3651 		 * Not ceph_oloc_copy() - changing pool_ns is not
3652 		 * supported.
3653 		 */
3654 		req->r_t.target_oloc.pool = m.redirect.oloc.pool;
3655 		req->r_flags |= CEPH_OSD_FLAG_REDIRECTED |
3656 				CEPH_OSD_FLAG_IGNORE_OVERLAY |
3657 				CEPH_OSD_FLAG_IGNORE_CACHE;
3658 		req->r_tid = 0;
3659 		__submit_request(req, false);
3660 		goto out_unlock_osdc;
3661 	}
3662 
3663 	if (m.num_ops != req->r_num_ops) {
3664 		pr_err("num_ops %d != %d for tid %llu\n", m.num_ops,
3665 		       req->r_num_ops, req->r_tid);
3666 		goto fail_request;
3667 	}
3668 	for (i = 0; i < req->r_num_ops; i++) {
3669 		dout(" req %p tid %llu op %d rval %d len %u\n", req,
3670 		     req->r_tid, i, m.rval[i], m.outdata_len[i]);
3671 		req->r_ops[i].rval = m.rval[i];
3672 		req->r_ops[i].outdata_len = m.outdata_len[i];
3673 		data_len += m.outdata_len[i];
3674 	}
3675 	if (data_len != le32_to_cpu(msg->hdr.data_len)) {
3676 		pr_err("sum of lens %u != %u for tid %llu\n", data_len,
3677 		       le32_to_cpu(msg->hdr.data_len), req->r_tid);
3678 		goto fail_request;
3679 	}
3680 	dout("%s req %p tid %llu result %d data_len %u\n", __func__,
3681 	     req, req->r_tid, m.result, data_len);
3682 
3683 	/*
3684 	 * Since we only ever request ONDISK, we should only ever get
3685 	 * one (type of) reply back.
3686 	 */
3687 	WARN_ON(!(m.flags & CEPH_OSD_FLAG_ONDISK));
3688 	req->r_result = m.result ?: data_len;
3689 	finish_request(req);
3690 	mutex_unlock(&osd->lock);
3691 	up_read(&osdc->lock);
3692 
3693 	__complete_request(req);
3694 	return;
3695 
3696 fail_request:
3697 	complete_request(req, -EIO);
3698 out_unlock_session:
3699 	mutex_unlock(&osd->lock);
3700 out_unlock_osdc:
3701 	up_read(&osdc->lock);
3702 }
3703 
3704 static void set_pool_was_full(struct ceph_osd_client *osdc)
3705 {
3706 	struct rb_node *n;
3707 
3708 	for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) {
3709 		struct ceph_pg_pool_info *pi =
3710 		    rb_entry(n, struct ceph_pg_pool_info, node);
3711 
3712 		pi->was_full = __pool_full(pi);
3713 	}
3714 }
3715 
3716 static bool pool_cleared_full(struct ceph_osd_client *osdc, s64 pool_id)
3717 {
3718 	struct ceph_pg_pool_info *pi;
3719 
3720 	pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id);
3721 	if (!pi)
3722 		return false;
3723 
3724 	return pi->was_full && !__pool_full(pi);
3725 }
3726 
3727 static enum calc_target_result
3728 recalc_linger_target(struct ceph_osd_linger_request *lreq)
3729 {
3730 	struct ceph_osd_client *osdc = lreq->osdc;
3731 	enum calc_target_result ct_res;
3732 
3733 	ct_res = calc_target(osdc, &lreq->t, true);
3734 	if (ct_res == CALC_TARGET_NEED_RESEND) {
3735 		struct ceph_osd *osd;
3736 
3737 		osd = lookup_create_osd(osdc, lreq->t.osd, true);
3738 		if (osd != lreq->osd) {
3739 			unlink_linger(lreq->osd, lreq);
3740 			link_linger(osd, lreq);
3741 		}
3742 	}
3743 
3744 	return ct_res;
3745 }
3746 
3747 /*
3748  * Requeue requests whose mapping to an OSD has changed.
3749  */
3750 static void scan_requests(struct ceph_osd *osd,
3751 			  bool force_resend,
3752 			  bool cleared_full,
3753 			  bool check_pool_cleared_full,
3754 			  struct rb_root *need_resend,
3755 			  struct list_head *need_resend_linger)
3756 {
3757 	struct ceph_osd_client *osdc = osd->o_osdc;
3758 	struct rb_node *n;
3759 	bool force_resend_writes;
3760 
3761 	for (n = rb_first(&osd->o_linger_requests); n; ) {
3762 		struct ceph_osd_linger_request *lreq =
3763 		    rb_entry(n, struct ceph_osd_linger_request, node);
3764 		enum calc_target_result ct_res;
3765 
3766 		n = rb_next(n); /* recalc_linger_target() */
3767 
3768 		dout("%s lreq %p linger_id %llu\n", __func__, lreq,
3769 		     lreq->linger_id);
3770 		ct_res = recalc_linger_target(lreq);
3771 		switch (ct_res) {
3772 		case CALC_TARGET_NO_ACTION:
3773 			force_resend_writes = cleared_full ||
3774 			    (check_pool_cleared_full &&
3775 			     pool_cleared_full(osdc, lreq->t.base_oloc.pool));
3776 			if (!force_resend && !force_resend_writes)
3777 				break;
3778 
3779 			/* fall through */
3780 		case CALC_TARGET_NEED_RESEND:
3781 			cancel_linger_map_check(lreq);
3782 			/*
3783 			 * scan_requests() for the previous epoch(s)
3784 			 * may have already added it to the list, since
3785 			 * it's not unlinked here.
3786 			 */
3787 			if (list_empty(&lreq->scan_item))
3788 				list_add_tail(&lreq->scan_item, need_resend_linger);
3789 			break;
3790 		case CALC_TARGET_POOL_DNE:
3791 			list_del_init(&lreq->scan_item);
3792 			check_linger_pool_dne(lreq);
3793 			break;
3794 		}
3795 	}
3796 
3797 	for (n = rb_first(&osd->o_requests); n; ) {
3798 		struct ceph_osd_request *req =
3799 		    rb_entry(n, struct ceph_osd_request, r_node);
3800 		enum calc_target_result ct_res;
3801 
3802 		n = rb_next(n); /* unlink_request(), check_pool_dne() */
3803 
3804 		dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
3805 		ct_res = calc_target(osdc, &req->r_t, false);
3806 		switch (ct_res) {
3807 		case CALC_TARGET_NO_ACTION:
3808 			force_resend_writes = cleared_full ||
3809 			    (check_pool_cleared_full &&
3810 			     pool_cleared_full(osdc, req->r_t.base_oloc.pool));
3811 			if (!force_resend &&
3812 			    (!(req->r_flags & CEPH_OSD_FLAG_WRITE) ||
3813 			     !force_resend_writes))
3814 				break;
3815 
3816 			/* fall through */
3817 		case CALC_TARGET_NEED_RESEND:
3818 			cancel_map_check(req);
3819 			unlink_request(osd, req);
3820 			insert_request(need_resend, req);
3821 			break;
3822 		case CALC_TARGET_POOL_DNE:
3823 			check_pool_dne(req);
3824 			break;
3825 		}
3826 	}
3827 }
3828 
3829 static int handle_one_map(struct ceph_osd_client *osdc,
3830 			  void *p, void *end, bool incremental,
3831 			  struct rb_root *need_resend,
3832 			  struct list_head *need_resend_linger)
3833 {
3834 	struct ceph_osdmap *newmap;
3835 	struct rb_node *n;
3836 	bool skipped_map = false;
3837 	bool was_full;
3838 
3839 	was_full = ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL);
3840 	set_pool_was_full(osdc);
3841 
3842 	if (incremental)
3843 		newmap = osdmap_apply_incremental(&p, end, osdc->osdmap);
3844 	else
3845 		newmap = ceph_osdmap_decode(&p, end);
3846 	if (IS_ERR(newmap))
3847 		return PTR_ERR(newmap);
3848 
3849 	if (newmap != osdc->osdmap) {
3850 		/*
3851 		 * Preserve ->was_full before destroying the old map.
3852 		 * For pools that weren't in the old map, ->was_full
3853 		 * should be false.
3854 		 */
3855 		for (n = rb_first(&newmap->pg_pools); n; n = rb_next(n)) {
3856 			struct ceph_pg_pool_info *pi =
3857 			    rb_entry(n, struct ceph_pg_pool_info, node);
3858 			struct ceph_pg_pool_info *old_pi;
3859 
3860 			old_pi = ceph_pg_pool_by_id(osdc->osdmap, pi->id);
3861 			if (old_pi)
3862 				pi->was_full = old_pi->was_full;
3863 			else
3864 				WARN_ON(pi->was_full);
3865 		}
3866 
3867 		if (osdc->osdmap->epoch &&
3868 		    osdc->osdmap->epoch + 1 < newmap->epoch) {
3869 			WARN_ON(incremental);
3870 			skipped_map = true;
3871 		}
3872 
3873 		ceph_osdmap_destroy(osdc->osdmap);
3874 		osdc->osdmap = newmap;
3875 	}
3876 
3877 	was_full &= !ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL);
3878 	scan_requests(&osdc->homeless_osd, skipped_map, was_full, true,
3879 		      need_resend, need_resend_linger);
3880 
3881 	for (n = rb_first(&osdc->osds); n; ) {
3882 		struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
3883 
3884 		n = rb_next(n); /* close_osd() */
3885 
3886 		scan_requests(osd, skipped_map, was_full, true, need_resend,
3887 			      need_resend_linger);
3888 		if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
3889 		    memcmp(&osd->o_con.peer_addr,
3890 			   ceph_osd_addr(osdc->osdmap, osd->o_osd),
3891 			   sizeof(struct ceph_entity_addr)))
3892 			close_osd(osd);
3893 	}
3894 
3895 	return 0;
3896 }
3897 
3898 static void kick_requests(struct ceph_osd_client *osdc,
3899 			  struct rb_root *need_resend,
3900 			  struct list_head *need_resend_linger)
3901 {
3902 	struct ceph_osd_linger_request *lreq, *nlreq;
3903 	enum calc_target_result ct_res;
3904 	struct rb_node *n;
3905 
3906 	/* make sure need_resend targets reflect latest map */
3907 	for (n = rb_first(need_resend); n; ) {
3908 		struct ceph_osd_request *req =
3909 		    rb_entry(n, struct ceph_osd_request, r_node);
3910 
3911 		n = rb_next(n);
3912 
3913 		if (req->r_t.epoch < osdc->osdmap->epoch) {
3914 			ct_res = calc_target(osdc, &req->r_t, false);
3915 			if (ct_res == CALC_TARGET_POOL_DNE) {
3916 				erase_request(need_resend, req);
3917 				check_pool_dne(req);
3918 			}
3919 		}
3920 	}
3921 
3922 	for (n = rb_first(need_resend); n; ) {
3923 		struct ceph_osd_request *req =
3924 		    rb_entry(n, struct ceph_osd_request, r_node);
3925 		struct ceph_osd *osd;
3926 
3927 		n = rb_next(n);
3928 		erase_request(need_resend, req); /* before link_request() */
3929 
3930 		osd = lookup_create_osd(osdc, req->r_t.osd, true);
3931 		link_request(osd, req);
3932 		if (!req->r_linger) {
3933 			if (!osd_homeless(osd) && !req->r_t.paused)
3934 				send_request(req);
3935 		} else {
3936 			cancel_linger_request(req);
3937 		}
3938 	}
3939 
3940 	list_for_each_entry_safe(lreq, nlreq, need_resend_linger, scan_item) {
3941 		if (!osd_homeless(lreq->osd))
3942 			send_linger(lreq);
3943 
3944 		list_del_init(&lreq->scan_item);
3945 	}
3946 }
3947 
3948 /*
3949  * Process updated osd map.
3950  *
3951  * The message contains any number of incremental and full maps, normally
3952  * indicating some sort of topology change in the cluster.  Kick requests
3953  * off to different OSDs as needed.
3954  */
3955 void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
3956 {
3957 	void *p = msg->front.iov_base;
3958 	void *const end = p + msg->front.iov_len;
3959 	u32 nr_maps, maplen;
3960 	u32 epoch;
3961 	struct ceph_fsid fsid;
3962 	struct rb_root need_resend = RB_ROOT;
3963 	LIST_HEAD(need_resend_linger);
3964 	bool handled_incremental = false;
3965 	bool was_pauserd, was_pausewr;
3966 	bool pauserd, pausewr;
3967 	int err;
3968 
3969 	dout("%s have %u\n", __func__, osdc->osdmap->epoch);
3970 	down_write(&osdc->lock);
3971 
3972 	/* verify fsid */
3973 	ceph_decode_need(&p, end, sizeof(fsid), bad);
3974 	ceph_decode_copy(&p, &fsid, sizeof(fsid));
3975 	if (ceph_check_fsid(osdc->client, &fsid) < 0)
3976 		goto bad;
3977 
3978 	was_pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
3979 	was_pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
3980 		      ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
3981 		      have_pool_full(osdc);
3982 
3983 	/* incremental maps */
3984 	ceph_decode_32_safe(&p, end, nr_maps, bad);
3985 	dout(" %d inc maps\n", nr_maps);
3986 	while (nr_maps > 0) {
3987 		ceph_decode_need(&p, end, 2*sizeof(u32), bad);
3988 		epoch = ceph_decode_32(&p);
3989 		maplen = ceph_decode_32(&p);
3990 		ceph_decode_need(&p, end, maplen, bad);
3991 		if (osdc->osdmap->epoch &&
3992 		    osdc->osdmap->epoch + 1 == epoch) {
3993 			dout("applying incremental map %u len %d\n",
3994 			     epoch, maplen);
3995 			err = handle_one_map(osdc, p, p + maplen, true,
3996 					     &need_resend, &need_resend_linger);
3997 			if (err)
3998 				goto bad;
3999 			handled_incremental = true;
4000 		} else {
4001 			dout("ignoring incremental map %u len %d\n",
4002 			     epoch, maplen);
4003 		}
4004 		p += maplen;
4005 		nr_maps--;
4006 	}
4007 	if (handled_incremental)
4008 		goto done;
4009 
4010 	/* full maps */
4011 	ceph_decode_32_safe(&p, end, nr_maps, bad);
4012 	dout(" %d full maps\n", nr_maps);
4013 	while (nr_maps) {
4014 		ceph_decode_need(&p, end, 2*sizeof(u32), bad);
4015 		epoch = ceph_decode_32(&p);
4016 		maplen = ceph_decode_32(&p);
4017 		ceph_decode_need(&p, end, maplen, bad);
4018 		if (nr_maps > 1) {
4019 			dout("skipping non-latest full map %u len %d\n",
4020 			     epoch, maplen);
4021 		} else if (osdc->osdmap->epoch >= epoch) {
4022 			dout("skipping full map %u len %d, "
4023 			     "older than our %u\n", epoch, maplen,
4024 			     osdc->osdmap->epoch);
4025 		} else {
4026 			dout("taking full map %u len %d\n", epoch, maplen);
4027 			err = handle_one_map(osdc, p, p + maplen, false,
4028 					     &need_resend, &need_resend_linger);
4029 			if (err)
4030 				goto bad;
4031 		}
4032 		p += maplen;
4033 		nr_maps--;
4034 	}
4035 
4036 done:
4037 	/*
4038 	 * subscribe to subsequent osdmap updates if full to ensure
4039 	 * we find out when we are no longer full and stop returning
4040 	 * ENOSPC.
4041 	 */
4042 	pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
4043 	pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
4044 		  ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
4045 		  have_pool_full(osdc);
4046 	if (was_pauserd || was_pausewr || pauserd || pausewr ||
4047 	    osdc->osdmap->epoch < osdc->epoch_barrier)
4048 		maybe_request_map(osdc);
4049 
4050 	kick_requests(osdc, &need_resend, &need_resend_linger);
4051 
4052 	ceph_osdc_abort_on_full(osdc);
4053 	ceph_monc_got_map(&osdc->client->monc, CEPH_SUB_OSDMAP,
4054 			  osdc->osdmap->epoch);
4055 	up_write(&osdc->lock);
4056 	wake_up_all(&osdc->client->auth_wq);
4057 	return;
4058 
4059 bad:
4060 	pr_err("osdc handle_map corrupt msg\n");
4061 	ceph_msg_dump(msg);
4062 	up_write(&osdc->lock);
4063 }
4064 
4065 /*
4066  * Resubmit requests pending on the given osd.
4067  */
4068 static void kick_osd_requests(struct ceph_osd *osd)
4069 {
4070 	struct rb_node *n;
4071 
4072 	clear_backoffs(osd);
4073 
4074 	for (n = rb_first(&osd->o_requests); n; ) {
4075 		struct ceph_osd_request *req =
4076 		    rb_entry(n, struct ceph_osd_request, r_node);
4077 
4078 		n = rb_next(n); /* cancel_linger_request() */
4079 
4080 		if (!req->r_linger) {
4081 			if (!req->r_t.paused)
4082 				send_request(req);
4083 		} else {
4084 			cancel_linger_request(req);
4085 		}
4086 	}
4087 	for (n = rb_first(&osd->o_linger_requests); n; n = rb_next(n)) {
4088 		struct ceph_osd_linger_request *lreq =
4089 		    rb_entry(n, struct ceph_osd_linger_request, node);
4090 
4091 		send_linger(lreq);
4092 	}
4093 }
4094 
4095 /*
4096  * If the osd connection drops, we need to resubmit all requests.
4097  */
4098 static void osd_fault(struct ceph_connection *con)
4099 {
4100 	struct ceph_osd *osd = con->private;
4101 	struct ceph_osd_client *osdc = osd->o_osdc;
4102 
4103 	dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
4104 
4105 	down_write(&osdc->lock);
4106 	if (!osd_registered(osd)) {
4107 		dout("%s osd%d unknown\n", __func__, osd->o_osd);
4108 		goto out_unlock;
4109 	}
4110 
4111 	if (!reopen_osd(osd))
4112 		kick_osd_requests(osd);
4113 	maybe_request_map(osdc);
4114 
4115 out_unlock:
4116 	up_write(&osdc->lock);
4117 }
4118 
4119 struct MOSDBackoff {
4120 	struct ceph_spg spgid;
4121 	u32 map_epoch;
4122 	u8 op;
4123 	u64 id;
4124 	struct ceph_hobject_id *begin;
4125 	struct ceph_hobject_id *end;
4126 };
4127 
4128 static int decode_MOSDBackoff(const struct ceph_msg *msg, struct MOSDBackoff *m)
4129 {
4130 	void *p = msg->front.iov_base;
4131 	void *const end = p + msg->front.iov_len;
4132 	u8 struct_v;
4133 	u32 struct_len;
4134 	int ret;
4135 
4136 	ret = ceph_start_decoding(&p, end, 1, "spg_t", &struct_v, &struct_len);
4137 	if (ret)
4138 		return ret;
4139 
4140 	ret = ceph_decode_pgid(&p, end, &m->spgid.pgid);
4141 	if (ret)
4142 		return ret;
4143 
4144 	ceph_decode_8_safe(&p, end, m->spgid.shard, e_inval);
4145 	ceph_decode_32_safe(&p, end, m->map_epoch, e_inval);
4146 	ceph_decode_8_safe(&p, end, m->op, e_inval);
4147 	ceph_decode_64_safe(&p, end, m->id, e_inval);
4148 
4149 	m->begin = kzalloc(sizeof(*m->begin), GFP_NOIO);
4150 	if (!m->begin)
4151 		return -ENOMEM;
4152 
4153 	ret = decode_hoid(&p, end, m->begin);
4154 	if (ret) {
4155 		free_hoid(m->begin);
4156 		return ret;
4157 	}
4158 
4159 	m->end = kzalloc(sizeof(*m->end), GFP_NOIO);
4160 	if (!m->end) {
4161 		free_hoid(m->begin);
4162 		return -ENOMEM;
4163 	}
4164 
4165 	ret = decode_hoid(&p, end, m->end);
4166 	if (ret) {
4167 		free_hoid(m->begin);
4168 		free_hoid(m->end);
4169 		return ret;
4170 	}
4171 
4172 	return 0;
4173 
4174 e_inval:
4175 	return -EINVAL;
4176 }
4177 
4178 static struct ceph_msg *create_backoff_message(
4179 				const struct ceph_osd_backoff *backoff,
4180 				u32 map_epoch)
4181 {
4182 	struct ceph_msg *msg;
4183 	void *p, *end;
4184 	int msg_size;
4185 
4186 	msg_size = CEPH_ENCODING_START_BLK_LEN +
4187 			CEPH_PGID_ENCODING_LEN + 1; /* spgid */
4188 	msg_size += 4 + 1 + 8; /* map_epoch, op, id */
4189 	msg_size += CEPH_ENCODING_START_BLK_LEN +
4190 			hoid_encoding_size(backoff->begin);
4191 	msg_size += CEPH_ENCODING_START_BLK_LEN +
4192 			hoid_encoding_size(backoff->end);
4193 
4194 	msg = ceph_msg_new(CEPH_MSG_OSD_BACKOFF, msg_size, GFP_NOIO, true);
4195 	if (!msg)
4196 		return NULL;
4197 
4198 	p = msg->front.iov_base;
4199 	end = p + msg->front_alloc_len;
4200 
4201 	encode_spgid(&p, &backoff->spgid);
4202 	ceph_encode_32(&p, map_epoch);
4203 	ceph_encode_8(&p, CEPH_OSD_BACKOFF_OP_ACK_BLOCK);
4204 	ceph_encode_64(&p, backoff->id);
4205 	encode_hoid(&p, end, backoff->begin);
4206 	encode_hoid(&p, end, backoff->end);
4207 	BUG_ON(p != end);
4208 
4209 	msg->front.iov_len = p - msg->front.iov_base;
4210 	msg->hdr.version = cpu_to_le16(1); /* MOSDBackoff v1 */
4211 	msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
4212 
4213 	return msg;
4214 }
4215 
4216 static void handle_backoff_block(struct ceph_osd *osd, struct MOSDBackoff *m)
4217 {
4218 	struct ceph_spg_mapping *spg;
4219 	struct ceph_osd_backoff *backoff;
4220 	struct ceph_msg *msg;
4221 
4222 	dout("%s osd%d spgid %llu.%xs%d id %llu\n", __func__, osd->o_osd,
4223 	     m->spgid.pgid.pool, m->spgid.pgid.seed, m->spgid.shard, m->id);
4224 
4225 	spg = lookup_spg_mapping(&osd->o_backoff_mappings, &m->spgid);
4226 	if (!spg) {
4227 		spg = alloc_spg_mapping();
4228 		if (!spg) {
4229 			pr_err("%s failed to allocate spg\n", __func__);
4230 			return;
4231 		}
4232 		spg->spgid = m->spgid; /* struct */
4233 		insert_spg_mapping(&osd->o_backoff_mappings, spg);
4234 	}
4235 
4236 	backoff = alloc_backoff();
4237 	if (!backoff) {
4238 		pr_err("%s failed to allocate backoff\n", __func__);
4239 		return;
4240 	}
4241 	backoff->spgid = m->spgid; /* struct */
4242 	backoff->id = m->id;
4243 	backoff->begin = m->begin;
4244 	m->begin = NULL; /* backoff now owns this */
4245 	backoff->end = m->end;
4246 	m->end = NULL;   /* ditto */
4247 
4248 	insert_backoff(&spg->backoffs, backoff);
4249 	insert_backoff_by_id(&osd->o_backoffs_by_id, backoff);
4250 
4251 	/*
4252 	 * Ack with original backoff's epoch so that the OSD can
4253 	 * discard this if there was a PG split.
4254 	 */
4255 	msg = create_backoff_message(backoff, m->map_epoch);
4256 	if (!msg) {
4257 		pr_err("%s failed to allocate msg\n", __func__);
4258 		return;
4259 	}
4260 	ceph_con_send(&osd->o_con, msg);
4261 }
4262 
4263 static bool target_contained_by(const struct ceph_osd_request_target *t,
4264 				const struct ceph_hobject_id *begin,
4265 				const struct ceph_hobject_id *end)
4266 {
4267 	struct ceph_hobject_id hoid;
4268 	int cmp;
4269 
4270 	hoid_fill_from_target(&hoid, t);
4271 	cmp = hoid_compare(&hoid, begin);
4272 	return !cmp || (cmp > 0 && hoid_compare(&hoid, end) < 0);
4273 }
4274 
4275 static void handle_backoff_unblock(struct ceph_osd *osd,
4276 				   const struct MOSDBackoff *m)
4277 {
4278 	struct ceph_spg_mapping *spg;
4279 	struct ceph_osd_backoff *backoff;
4280 	struct rb_node *n;
4281 
4282 	dout("%s osd%d spgid %llu.%xs%d id %llu\n", __func__, osd->o_osd,
4283 	     m->spgid.pgid.pool, m->spgid.pgid.seed, m->spgid.shard, m->id);
4284 
4285 	backoff = lookup_backoff_by_id(&osd->o_backoffs_by_id, m->id);
4286 	if (!backoff) {
4287 		pr_err("%s osd%d spgid %llu.%xs%d id %llu backoff dne\n",
4288 		       __func__, osd->o_osd, m->spgid.pgid.pool,
4289 		       m->spgid.pgid.seed, m->spgid.shard, m->id);
4290 		return;
4291 	}
4292 
4293 	if (hoid_compare(backoff->begin, m->begin) &&
4294 	    hoid_compare(backoff->end, m->end)) {
4295 		pr_err("%s osd%d spgid %llu.%xs%d id %llu bad range?\n",
4296 		       __func__, osd->o_osd, m->spgid.pgid.pool,
4297 		       m->spgid.pgid.seed, m->spgid.shard, m->id);
4298 		/* unblock it anyway... */
4299 	}
4300 
4301 	spg = lookup_spg_mapping(&osd->o_backoff_mappings, &backoff->spgid);
4302 	BUG_ON(!spg);
4303 
4304 	erase_backoff(&spg->backoffs, backoff);
4305 	erase_backoff_by_id(&osd->o_backoffs_by_id, backoff);
4306 	free_backoff(backoff);
4307 
4308 	if (RB_EMPTY_ROOT(&spg->backoffs)) {
4309 		erase_spg_mapping(&osd->o_backoff_mappings, spg);
4310 		free_spg_mapping(spg);
4311 	}
4312 
4313 	for (n = rb_first(&osd->o_requests); n; n = rb_next(n)) {
4314 		struct ceph_osd_request *req =
4315 		    rb_entry(n, struct ceph_osd_request, r_node);
4316 
4317 		if (!ceph_spg_compare(&req->r_t.spgid, &m->spgid)) {
4318 			/*
4319 			 * Match against @m, not @backoff -- the PG may
4320 			 * have split on the OSD.
4321 			 */
4322 			if (target_contained_by(&req->r_t, m->begin, m->end)) {
4323 				/*
4324 				 * If no other installed backoff applies,
4325 				 * resend.
4326 				 */
4327 				send_request(req);
4328 			}
4329 		}
4330 	}
4331 }
4332 
4333 static void handle_backoff(struct ceph_osd *osd, struct ceph_msg *msg)
4334 {
4335 	struct ceph_osd_client *osdc = osd->o_osdc;
4336 	struct MOSDBackoff m;
4337 	int ret;
4338 
4339 	down_read(&osdc->lock);
4340 	if (!osd_registered(osd)) {
4341 		dout("%s osd%d unknown\n", __func__, osd->o_osd);
4342 		up_read(&osdc->lock);
4343 		return;
4344 	}
4345 	WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num));
4346 
4347 	mutex_lock(&osd->lock);
4348 	ret = decode_MOSDBackoff(msg, &m);
4349 	if (ret) {
4350 		pr_err("failed to decode MOSDBackoff: %d\n", ret);
4351 		ceph_msg_dump(msg);
4352 		goto out_unlock;
4353 	}
4354 
4355 	switch (m.op) {
4356 	case CEPH_OSD_BACKOFF_OP_BLOCK:
4357 		handle_backoff_block(osd, &m);
4358 		break;
4359 	case CEPH_OSD_BACKOFF_OP_UNBLOCK:
4360 		handle_backoff_unblock(osd, &m);
4361 		break;
4362 	default:
4363 		pr_err("%s osd%d unknown op %d\n", __func__, osd->o_osd, m.op);
4364 	}
4365 
4366 	free_hoid(m.begin);
4367 	free_hoid(m.end);
4368 
4369 out_unlock:
4370 	mutex_unlock(&osd->lock);
4371 	up_read(&osdc->lock);
4372 }
4373 
4374 /*
4375  * Process osd watch notifications
4376  */
4377 static void handle_watch_notify(struct ceph_osd_client *osdc,
4378 				struct ceph_msg *msg)
4379 {
4380 	void *p = msg->front.iov_base;
4381 	void *const end = p + msg->front.iov_len;
4382 	struct ceph_osd_linger_request *lreq;
4383 	struct linger_work *lwork;
4384 	u8 proto_ver, opcode;
4385 	u64 cookie, notify_id;
4386 	u64 notifier_id = 0;
4387 	s32 return_code = 0;
4388 	void *payload = NULL;
4389 	u32 payload_len = 0;
4390 
4391 	ceph_decode_8_safe(&p, end, proto_ver, bad);
4392 	ceph_decode_8_safe(&p, end, opcode, bad);
4393 	ceph_decode_64_safe(&p, end, cookie, bad);
4394 	p += 8; /* skip ver */
4395 	ceph_decode_64_safe(&p, end, notify_id, bad);
4396 
4397 	if (proto_ver >= 1) {
4398 		ceph_decode_32_safe(&p, end, payload_len, bad);
4399 		ceph_decode_need(&p, end, payload_len, bad);
4400 		payload = p;
4401 		p += payload_len;
4402 	}
4403 
4404 	if (le16_to_cpu(msg->hdr.version) >= 2)
4405 		ceph_decode_32_safe(&p, end, return_code, bad);
4406 
4407 	if (le16_to_cpu(msg->hdr.version) >= 3)
4408 		ceph_decode_64_safe(&p, end, notifier_id, bad);
4409 
4410 	down_read(&osdc->lock);
4411 	lreq = lookup_linger_osdc(&osdc->linger_requests, cookie);
4412 	if (!lreq) {
4413 		dout("%s opcode %d cookie %llu dne\n", __func__, opcode,
4414 		     cookie);
4415 		goto out_unlock_osdc;
4416 	}
4417 
4418 	mutex_lock(&lreq->lock);
4419 	dout("%s opcode %d cookie %llu lreq %p is_watch %d\n", __func__,
4420 	     opcode, cookie, lreq, lreq->is_watch);
4421 	if (opcode == CEPH_WATCH_EVENT_DISCONNECT) {
4422 		if (!lreq->last_error) {
4423 			lreq->last_error = -ENOTCONN;
4424 			queue_watch_error(lreq);
4425 		}
4426 	} else if (!lreq->is_watch) {
4427 		/* CEPH_WATCH_EVENT_NOTIFY_COMPLETE */
4428 		if (lreq->notify_id && lreq->notify_id != notify_id) {
4429 			dout("lreq %p notify_id %llu != %llu, ignoring\n", lreq,
4430 			     lreq->notify_id, notify_id);
4431 		} else if (!completion_done(&lreq->notify_finish_wait)) {
4432 			struct ceph_msg_data *data =
4433 			    msg->num_data_items ? &msg->data[0] : NULL;
4434 
4435 			if (data) {
4436 				if (lreq->preply_pages) {
4437 					WARN_ON(data->type !=
4438 							CEPH_MSG_DATA_PAGES);
4439 					*lreq->preply_pages = data->pages;
4440 					*lreq->preply_len = data->length;
4441 					data->own_pages = false;
4442 				}
4443 			}
4444 			lreq->notify_finish_error = return_code;
4445 			complete_all(&lreq->notify_finish_wait);
4446 		}
4447 	} else {
4448 		/* CEPH_WATCH_EVENT_NOTIFY */
4449 		lwork = lwork_alloc(lreq, do_watch_notify);
4450 		if (!lwork) {
4451 			pr_err("failed to allocate notify-lwork\n");
4452 			goto out_unlock_lreq;
4453 		}
4454 
4455 		lwork->notify.notify_id = notify_id;
4456 		lwork->notify.notifier_id = notifier_id;
4457 		lwork->notify.payload = payload;
4458 		lwork->notify.payload_len = payload_len;
4459 		lwork->notify.msg = ceph_msg_get(msg);
4460 		lwork_queue(lwork);
4461 	}
4462 
4463 out_unlock_lreq:
4464 	mutex_unlock(&lreq->lock);
4465 out_unlock_osdc:
4466 	up_read(&osdc->lock);
4467 	return;
4468 
4469 bad:
4470 	pr_err("osdc handle_watch_notify corrupt msg\n");
4471 }
4472 
4473 /*
4474  * Register request, send initial attempt.
4475  */
4476 int ceph_osdc_start_request(struct ceph_osd_client *osdc,
4477 			    struct ceph_osd_request *req,
4478 			    bool nofail)
4479 {
4480 	down_read(&osdc->lock);
4481 	submit_request(req, false);
4482 	up_read(&osdc->lock);
4483 
4484 	return 0;
4485 }
4486 EXPORT_SYMBOL(ceph_osdc_start_request);
4487 
4488 /*
4489  * Unregister a registered request.  The request is not completed:
4490  * ->r_result isn't set and __complete_request() isn't called.
4491  */
4492 void ceph_osdc_cancel_request(struct ceph_osd_request *req)
4493 {
4494 	struct ceph_osd_client *osdc = req->r_osdc;
4495 
4496 	down_write(&osdc->lock);
4497 	if (req->r_osd)
4498 		cancel_request(req);
4499 	up_write(&osdc->lock);
4500 }
4501 EXPORT_SYMBOL(ceph_osdc_cancel_request);
4502 
4503 /*
4504  * @timeout: in jiffies, 0 means "wait forever"
4505  */
4506 static int wait_request_timeout(struct ceph_osd_request *req,
4507 				unsigned long timeout)
4508 {
4509 	long left;
4510 
4511 	dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
4512 	left = wait_for_completion_killable_timeout(&req->r_completion,
4513 						ceph_timeout_jiffies(timeout));
4514 	if (left <= 0) {
4515 		left = left ?: -ETIMEDOUT;
4516 		ceph_osdc_cancel_request(req);
4517 	} else {
4518 		left = req->r_result; /* completed */
4519 	}
4520 
4521 	return left;
4522 }
4523 
4524 /*
4525  * wait for a request to complete
4526  */
4527 int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
4528 			   struct ceph_osd_request *req)
4529 {
4530 	return wait_request_timeout(req, 0);
4531 }
4532 EXPORT_SYMBOL(ceph_osdc_wait_request);
4533 
4534 /*
4535  * sync - wait for all in-flight requests to flush.  avoid starvation.
4536  */
4537 void ceph_osdc_sync(struct ceph_osd_client *osdc)
4538 {
4539 	struct rb_node *n, *p;
4540 	u64 last_tid = atomic64_read(&osdc->last_tid);
4541 
4542 again:
4543 	down_read(&osdc->lock);
4544 	for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
4545 		struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
4546 
4547 		mutex_lock(&osd->lock);
4548 		for (p = rb_first(&osd->o_requests); p; p = rb_next(p)) {
4549 			struct ceph_osd_request *req =
4550 			    rb_entry(p, struct ceph_osd_request, r_node);
4551 
4552 			if (req->r_tid > last_tid)
4553 				break;
4554 
4555 			if (!(req->r_flags & CEPH_OSD_FLAG_WRITE))
4556 				continue;
4557 
4558 			ceph_osdc_get_request(req);
4559 			mutex_unlock(&osd->lock);
4560 			up_read(&osdc->lock);
4561 			dout("%s waiting on req %p tid %llu last_tid %llu\n",
4562 			     __func__, req, req->r_tid, last_tid);
4563 			wait_for_completion(&req->r_completion);
4564 			ceph_osdc_put_request(req);
4565 			goto again;
4566 		}
4567 
4568 		mutex_unlock(&osd->lock);
4569 	}
4570 
4571 	up_read(&osdc->lock);
4572 	dout("%s done last_tid %llu\n", __func__, last_tid);
4573 }
4574 EXPORT_SYMBOL(ceph_osdc_sync);
4575 
4576 static struct ceph_osd_request *
4577 alloc_linger_request(struct ceph_osd_linger_request *lreq)
4578 {
4579 	struct ceph_osd_request *req;
4580 
4581 	req = ceph_osdc_alloc_request(lreq->osdc, NULL, 1, false, GFP_NOIO);
4582 	if (!req)
4583 		return NULL;
4584 
4585 	ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
4586 	ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
4587 	return req;
4588 }
4589 
4590 static struct ceph_osd_request *
4591 alloc_watch_request(struct ceph_osd_linger_request *lreq, u8 watch_opcode)
4592 {
4593 	struct ceph_osd_request *req;
4594 
4595 	req = alloc_linger_request(lreq);
4596 	if (!req)
4597 		return NULL;
4598 
4599 	/*
4600 	 * Pass 0 for cookie because we don't know it yet, it will be
4601 	 * filled in by linger_submit().
4602 	 */
4603 	osd_req_op_watch_init(req, 0, 0, watch_opcode);
4604 
4605 	if (ceph_osdc_alloc_messages(req, GFP_NOIO)) {
4606 		ceph_osdc_put_request(req);
4607 		return NULL;
4608 	}
4609 
4610 	return req;
4611 }
4612 
4613 /*
4614  * Returns a handle, caller owns a ref.
4615  */
4616 struct ceph_osd_linger_request *
4617 ceph_osdc_watch(struct ceph_osd_client *osdc,
4618 		struct ceph_object_id *oid,
4619 		struct ceph_object_locator *oloc,
4620 		rados_watchcb2_t wcb,
4621 		rados_watcherrcb_t errcb,
4622 		void *data)
4623 {
4624 	struct ceph_osd_linger_request *lreq;
4625 	int ret;
4626 
4627 	lreq = linger_alloc(osdc);
4628 	if (!lreq)
4629 		return ERR_PTR(-ENOMEM);
4630 
4631 	lreq->is_watch = true;
4632 	lreq->wcb = wcb;
4633 	lreq->errcb = errcb;
4634 	lreq->data = data;
4635 	lreq->watch_valid_thru = jiffies;
4636 
4637 	ceph_oid_copy(&lreq->t.base_oid, oid);
4638 	ceph_oloc_copy(&lreq->t.base_oloc, oloc);
4639 	lreq->t.flags = CEPH_OSD_FLAG_WRITE;
4640 	ktime_get_real_ts64(&lreq->mtime);
4641 
4642 	lreq->reg_req = alloc_watch_request(lreq, CEPH_OSD_WATCH_OP_WATCH);
4643 	if (!lreq->reg_req) {
4644 		ret = -ENOMEM;
4645 		goto err_put_lreq;
4646 	}
4647 
4648 	lreq->ping_req = alloc_watch_request(lreq, CEPH_OSD_WATCH_OP_PING);
4649 	if (!lreq->ping_req) {
4650 		ret = -ENOMEM;
4651 		goto err_put_lreq;
4652 	}
4653 
4654 	linger_submit(lreq);
4655 	ret = linger_reg_commit_wait(lreq);
4656 	if (ret) {
4657 		linger_cancel(lreq);
4658 		goto err_put_lreq;
4659 	}
4660 
4661 	return lreq;
4662 
4663 err_put_lreq:
4664 	linger_put(lreq);
4665 	return ERR_PTR(ret);
4666 }
4667 EXPORT_SYMBOL(ceph_osdc_watch);
4668 
4669 /*
4670  * Releases a ref.
4671  *
4672  * Times out after mount_timeout to preserve rbd unmap behaviour
4673  * introduced in 2894e1d76974 ("rbd: timeout watch teardown on unmap
4674  * with mount_timeout").
4675  */
4676 int ceph_osdc_unwatch(struct ceph_osd_client *osdc,
4677 		      struct ceph_osd_linger_request *lreq)
4678 {
4679 	struct ceph_options *opts = osdc->client->options;
4680 	struct ceph_osd_request *req;
4681 	int ret;
4682 
4683 	req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
4684 	if (!req)
4685 		return -ENOMEM;
4686 
4687 	ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
4688 	ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
4689 	req->r_flags = CEPH_OSD_FLAG_WRITE;
4690 	ktime_get_real_ts64(&req->r_mtime);
4691 	osd_req_op_watch_init(req, 0, lreq->linger_id,
4692 			      CEPH_OSD_WATCH_OP_UNWATCH);
4693 
4694 	ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
4695 	if (ret)
4696 		goto out_put_req;
4697 
4698 	ceph_osdc_start_request(osdc, req, false);
4699 	linger_cancel(lreq);
4700 	linger_put(lreq);
4701 	ret = wait_request_timeout(req, opts->mount_timeout);
4702 
4703 out_put_req:
4704 	ceph_osdc_put_request(req);
4705 	return ret;
4706 }
4707 EXPORT_SYMBOL(ceph_osdc_unwatch);
4708 
4709 static int osd_req_op_notify_ack_init(struct ceph_osd_request *req, int which,
4710 				      u64 notify_id, u64 cookie, void *payload,
4711 				      u32 payload_len)
4712 {
4713 	struct ceph_osd_req_op *op;
4714 	struct ceph_pagelist *pl;
4715 	int ret;
4716 
4717 	op = _osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY_ACK, 0);
4718 
4719 	pl = ceph_pagelist_alloc(GFP_NOIO);
4720 	if (!pl)
4721 		return -ENOMEM;
4722 
4723 	ret = ceph_pagelist_encode_64(pl, notify_id);
4724 	ret |= ceph_pagelist_encode_64(pl, cookie);
4725 	if (payload) {
4726 		ret |= ceph_pagelist_encode_32(pl, payload_len);
4727 		ret |= ceph_pagelist_append(pl, payload, payload_len);
4728 	} else {
4729 		ret |= ceph_pagelist_encode_32(pl, 0);
4730 	}
4731 	if (ret) {
4732 		ceph_pagelist_release(pl);
4733 		return -ENOMEM;
4734 	}
4735 
4736 	ceph_osd_data_pagelist_init(&op->notify_ack.request_data, pl);
4737 	op->indata_len = pl->length;
4738 	return 0;
4739 }
4740 
4741 int ceph_osdc_notify_ack(struct ceph_osd_client *osdc,
4742 			 struct ceph_object_id *oid,
4743 			 struct ceph_object_locator *oloc,
4744 			 u64 notify_id,
4745 			 u64 cookie,
4746 			 void *payload,
4747 			 u32 payload_len)
4748 {
4749 	struct ceph_osd_request *req;
4750 	int ret;
4751 
4752 	req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
4753 	if (!req)
4754 		return -ENOMEM;
4755 
4756 	ceph_oid_copy(&req->r_base_oid, oid);
4757 	ceph_oloc_copy(&req->r_base_oloc, oloc);
4758 	req->r_flags = CEPH_OSD_FLAG_READ;
4759 
4760 	ret = osd_req_op_notify_ack_init(req, 0, notify_id, cookie, payload,
4761 					 payload_len);
4762 	if (ret)
4763 		goto out_put_req;
4764 
4765 	ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
4766 	if (ret)
4767 		goto out_put_req;
4768 
4769 	ceph_osdc_start_request(osdc, req, false);
4770 	ret = ceph_osdc_wait_request(osdc, req);
4771 
4772 out_put_req:
4773 	ceph_osdc_put_request(req);
4774 	return ret;
4775 }
4776 EXPORT_SYMBOL(ceph_osdc_notify_ack);
4777 
4778 static int osd_req_op_notify_init(struct ceph_osd_request *req, int which,
4779 				  u64 cookie, u32 prot_ver, u32 timeout,
4780 				  void *payload, u32 payload_len)
4781 {
4782 	struct ceph_osd_req_op *op;
4783 	struct ceph_pagelist *pl;
4784 	int ret;
4785 
4786 	op = _osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY, 0);
4787 	op->notify.cookie = cookie;
4788 
4789 	pl = ceph_pagelist_alloc(GFP_NOIO);
4790 	if (!pl)
4791 		return -ENOMEM;
4792 
4793 	ret = ceph_pagelist_encode_32(pl, 1); /* prot_ver */
4794 	ret |= ceph_pagelist_encode_32(pl, timeout);
4795 	ret |= ceph_pagelist_encode_32(pl, payload_len);
4796 	ret |= ceph_pagelist_append(pl, payload, payload_len);
4797 	if (ret) {
4798 		ceph_pagelist_release(pl);
4799 		return -ENOMEM;
4800 	}
4801 
4802 	ceph_osd_data_pagelist_init(&op->notify.request_data, pl);
4803 	op->indata_len = pl->length;
4804 	return 0;
4805 }
4806 
4807 /*
4808  * @timeout: in seconds
4809  *
4810  * @preply_{pages,len} are initialized both on success and error.
4811  * The caller is responsible for:
4812  *
4813  *     ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len))
4814  */
4815 int ceph_osdc_notify(struct ceph_osd_client *osdc,
4816 		     struct ceph_object_id *oid,
4817 		     struct ceph_object_locator *oloc,
4818 		     void *payload,
4819 		     u32 payload_len,
4820 		     u32 timeout,
4821 		     struct page ***preply_pages,
4822 		     size_t *preply_len)
4823 {
4824 	struct ceph_osd_linger_request *lreq;
4825 	struct page **pages;
4826 	int ret;
4827 
4828 	WARN_ON(!timeout);
4829 	if (preply_pages) {
4830 		*preply_pages = NULL;
4831 		*preply_len = 0;
4832 	}
4833 
4834 	lreq = linger_alloc(osdc);
4835 	if (!lreq)
4836 		return -ENOMEM;
4837 
4838 	lreq->preply_pages = preply_pages;
4839 	lreq->preply_len = preply_len;
4840 
4841 	ceph_oid_copy(&lreq->t.base_oid, oid);
4842 	ceph_oloc_copy(&lreq->t.base_oloc, oloc);
4843 	lreq->t.flags = CEPH_OSD_FLAG_READ;
4844 
4845 	lreq->reg_req = alloc_linger_request(lreq);
4846 	if (!lreq->reg_req) {
4847 		ret = -ENOMEM;
4848 		goto out_put_lreq;
4849 	}
4850 
4851 	/*
4852 	 * Pass 0 for cookie because we don't know it yet, it will be
4853 	 * filled in by linger_submit().
4854 	 */
4855 	ret = osd_req_op_notify_init(lreq->reg_req, 0, 0, 1, timeout,
4856 				     payload, payload_len);
4857 	if (ret)
4858 		goto out_put_lreq;
4859 
4860 	/* for notify_id */
4861 	pages = ceph_alloc_page_vector(1, GFP_NOIO);
4862 	if (IS_ERR(pages)) {
4863 		ret = PTR_ERR(pages);
4864 		goto out_put_lreq;
4865 	}
4866 	ceph_osd_data_pages_init(osd_req_op_data(lreq->reg_req, 0, notify,
4867 						 response_data),
4868 				 pages, PAGE_SIZE, 0, false, true);
4869 
4870 	ret = ceph_osdc_alloc_messages(lreq->reg_req, GFP_NOIO);
4871 	if (ret)
4872 		goto out_put_lreq;
4873 
4874 	linger_submit(lreq);
4875 	ret = linger_reg_commit_wait(lreq);
4876 	if (!ret)
4877 		ret = linger_notify_finish_wait(lreq);
4878 	else
4879 		dout("lreq %p failed to initiate notify %d\n", lreq, ret);
4880 
4881 	linger_cancel(lreq);
4882 out_put_lreq:
4883 	linger_put(lreq);
4884 	return ret;
4885 }
4886 EXPORT_SYMBOL(ceph_osdc_notify);
4887 
4888 /*
4889  * Return the number of milliseconds since the watch was last
4890  * confirmed, or an error.  If there is an error, the watch is no
4891  * longer valid, and should be destroyed with ceph_osdc_unwatch().
4892  */
4893 int ceph_osdc_watch_check(struct ceph_osd_client *osdc,
4894 			  struct ceph_osd_linger_request *lreq)
4895 {
4896 	unsigned long stamp, age;
4897 	int ret;
4898 
4899 	down_read(&osdc->lock);
4900 	mutex_lock(&lreq->lock);
4901 	stamp = lreq->watch_valid_thru;
4902 	if (!list_empty(&lreq->pending_lworks)) {
4903 		struct linger_work *lwork =
4904 		    list_first_entry(&lreq->pending_lworks,
4905 				     struct linger_work,
4906 				     pending_item);
4907 
4908 		if (time_before(lwork->queued_stamp, stamp))
4909 			stamp = lwork->queued_stamp;
4910 	}
4911 	age = jiffies - stamp;
4912 	dout("%s lreq %p linger_id %llu age %lu last_error %d\n", __func__,
4913 	     lreq, lreq->linger_id, age, lreq->last_error);
4914 	/* we are truncating to msecs, so return a safe upper bound */
4915 	ret = lreq->last_error ?: 1 + jiffies_to_msecs(age);
4916 
4917 	mutex_unlock(&lreq->lock);
4918 	up_read(&osdc->lock);
4919 	return ret;
4920 }
4921 
4922 static int decode_watcher(void **p, void *end, struct ceph_watch_item *item)
4923 {
4924 	u8 struct_v;
4925 	u32 struct_len;
4926 	int ret;
4927 
4928 	ret = ceph_start_decoding(p, end, 2, "watch_item_t",
4929 				  &struct_v, &struct_len);
4930 	if (ret)
4931 		goto bad;
4932 
4933 	ret = -EINVAL;
4934 	ceph_decode_copy_safe(p, end, &item->name, sizeof(item->name), bad);
4935 	ceph_decode_64_safe(p, end, item->cookie, bad);
4936 	ceph_decode_skip_32(p, end, bad); /* skip timeout seconds */
4937 
4938 	if (struct_v >= 2) {
4939 		ret = ceph_decode_entity_addr(p, end, &item->addr);
4940 		if (ret)
4941 			goto bad;
4942 	} else {
4943 		ret = 0;
4944 	}
4945 
4946 	dout("%s %s%llu cookie %llu addr %s\n", __func__,
4947 	     ENTITY_NAME(item->name), item->cookie,
4948 	     ceph_pr_addr(&item->addr));
4949 bad:
4950 	return ret;
4951 }
4952 
4953 static int decode_watchers(void **p, void *end,
4954 			   struct ceph_watch_item **watchers,
4955 			   u32 *num_watchers)
4956 {
4957 	u8 struct_v;
4958 	u32 struct_len;
4959 	int i;
4960 	int ret;
4961 
4962 	ret = ceph_start_decoding(p, end, 1, "obj_list_watch_response_t",
4963 				  &struct_v, &struct_len);
4964 	if (ret)
4965 		return ret;
4966 
4967 	*num_watchers = ceph_decode_32(p);
4968 	*watchers = kcalloc(*num_watchers, sizeof(**watchers), GFP_NOIO);
4969 	if (!*watchers)
4970 		return -ENOMEM;
4971 
4972 	for (i = 0; i < *num_watchers; i++) {
4973 		ret = decode_watcher(p, end, *watchers + i);
4974 		if (ret) {
4975 			kfree(*watchers);
4976 			return ret;
4977 		}
4978 	}
4979 
4980 	return 0;
4981 }
4982 
4983 /*
4984  * On success, the caller is responsible for:
4985  *
4986  *     kfree(watchers);
4987  */
4988 int ceph_osdc_list_watchers(struct ceph_osd_client *osdc,
4989 			    struct ceph_object_id *oid,
4990 			    struct ceph_object_locator *oloc,
4991 			    struct ceph_watch_item **watchers,
4992 			    u32 *num_watchers)
4993 {
4994 	struct ceph_osd_request *req;
4995 	struct page **pages;
4996 	int ret;
4997 
4998 	req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
4999 	if (!req)
5000 		return -ENOMEM;
5001 
5002 	ceph_oid_copy(&req->r_base_oid, oid);
5003 	ceph_oloc_copy(&req->r_base_oloc, oloc);
5004 	req->r_flags = CEPH_OSD_FLAG_READ;
5005 
5006 	pages = ceph_alloc_page_vector(1, GFP_NOIO);
5007 	if (IS_ERR(pages)) {
5008 		ret = PTR_ERR(pages);
5009 		goto out_put_req;
5010 	}
5011 
5012 	osd_req_op_init(req, 0, CEPH_OSD_OP_LIST_WATCHERS, 0);
5013 	ceph_osd_data_pages_init(osd_req_op_data(req, 0, list_watchers,
5014 						 response_data),
5015 				 pages, PAGE_SIZE, 0, false, true);
5016 
5017 	ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
5018 	if (ret)
5019 		goto out_put_req;
5020 
5021 	ceph_osdc_start_request(osdc, req, false);
5022 	ret = ceph_osdc_wait_request(osdc, req);
5023 	if (ret >= 0) {
5024 		void *p = page_address(pages[0]);
5025 		void *const end = p + req->r_ops[0].outdata_len;
5026 
5027 		ret = decode_watchers(&p, end, watchers, num_watchers);
5028 	}
5029 
5030 out_put_req:
5031 	ceph_osdc_put_request(req);
5032 	return ret;
5033 }
5034 EXPORT_SYMBOL(ceph_osdc_list_watchers);
5035 
5036 /*
5037  * Call all pending notify callbacks - for use after a watch is
5038  * unregistered, to make sure no more callbacks for it will be invoked
5039  */
5040 void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc)
5041 {
5042 	dout("%s osdc %p\n", __func__, osdc);
5043 	flush_workqueue(osdc->notify_wq);
5044 }
5045 EXPORT_SYMBOL(ceph_osdc_flush_notifies);
5046 
5047 void ceph_osdc_maybe_request_map(struct ceph_osd_client *osdc)
5048 {
5049 	down_read(&osdc->lock);
5050 	maybe_request_map(osdc);
5051 	up_read(&osdc->lock);
5052 }
5053 EXPORT_SYMBOL(ceph_osdc_maybe_request_map);
5054 
5055 /*
5056  * Execute an OSD class method on an object.
5057  *
5058  * @flags: CEPH_OSD_FLAG_*
5059  * @resp_len: in/out param for reply length
5060  */
5061 int ceph_osdc_call(struct ceph_osd_client *osdc,
5062 		   struct ceph_object_id *oid,
5063 		   struct ceph_object_locator *oloc,
5064 		   const char *class, const char *method,
5065 		   unsigned int flags,
5066 		   struct page *req_page, size_t req_len,
5067 		   struct page **resp_pages, size_t *resp_len)
5068 {
5069 	struct ceph_osd_request *req;
5070 	int ret;
5071 
5072 	if (req_len > PAGE_SIZE)
5073 		return -E2BIG;
5074 
5075 	req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
5076 	if (!req)
5077 		return -ENOMEM;
5078 
5079 	ceph_oid_copy(&req->r_base_oid, oid);
5080 	ceph_oloc_copy(&req->r_base_oloc, oloc);
5081 	req->r_flags = flags;
5082 
5083 	ret = osd_req_op_cls_init(req, 0, class, method);
5084 	if (ret)
5085 		goto out_put_req;
5086 
5087 	if (req_page)
5088 		osd_req_op_cls_request_data_pages(req, 0, &req_page, req_len,
5089 						  0, false, false);
5090 	if (resp_pages)
5091 		osd_req_op_cls_response_data_pages(req, 0, resp_pages,
5092 						   *resp_len, 0, false, false);
5093 
5094 	ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
5095 	if (ret)
5096 		goto out_put_req;
5097 
5098 	ceph_osdc_start_request(osdc, req, false);
5099 	ret = ceph_osdc_wait_request(osdc, req);
5100 	if (ret >= 0) {
5101 		ret = req->r_ops[0].rval;
5102 		if (resp_pages)
5103 			*resp_len = req->r_ops[0].outdata_len;
5104 	}
5105 
5106 out_put_req:
5107 	ceph_osdc_put_request(req);
5108 	return ret;
5109 }
5110 EXPORT_SYMBOL(ceph_osdc_call);
5111 
5112 /*
5113  * reset all osd connections
5114  */
5115 void ceph_osdc_reopen_osds(struct ceph_osd_client *osdc)
5116 {
5117 	struct rb_node *n;
5118 
5119 	down_write(&osdc->lock);
5120 	for (n = rb_first(&osdc->osds); n; ) {
5121 		struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
5122 
5123 		n = rb_next(n);
5124 		if (!reopen_osd(osd))
5125 			kick_osd_requests(osd);
5126 	}
5127 	up_write(&osdc->lock);
5128 }
5129 
5130 /*
5131  * init, shutdown
5132  */
5133 int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
5134 {
5135 	int err;
5136 
5137 	dout("init\n");
5138 	osdc->client = client;
5139 	init_rwsem(&osdc->lock);
5140 	osdc->osds = RB_ROOT;
5141 	INIT_LIST_HEAD(&osdc->osd_lru);
5142 	spin_lock_init(&osdc->osd_lru_lock);
5143 	osd_init(&osdc->homeless_osd);
5144 	osdc->homeless_osd.o_osdc = osdc;
5145 	osdc->homeless_osd.o_osd = CEPH_HOMELESS_OSD;
5146 	osdc->last_linger_id = CEPH_LINGER_ID_START;
5147 	osdc->linger_requests = RB_ROOT;
5148 	osdc->map_checks = RB_ROOT;
5149 	osdc->linger_map_checks = RB_ROOT;
5150 	INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
5151 	INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
5152 
5153 	err = -ENOMEM;
5154 	osdc->osdmap = ceph_osdmap_alloc();
5155 	if (!osdc->osdmap)
5156 		goto out;
5157 
5158 	osdc->req_mempool = mempool_create_slab_pool(10,
5159 						     ceph_osd_request_cache);
5160 	if (!osdc->req_mempool)
5161 		goto out_map;
5162 
5163 	err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
5164 				PAGE_SIZE, CEPH_OSD_SLAB_OPS, 10, "osd_op");
5165 	if (err < 0)
5166 		goto out_mempool;
5167 	err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
5168 				PAGE_SIZE, CEPH_OSD_SLAB_OPS, 10,
5169 				"osd_op_reply");
5170 	if (err < 0)
5171 		goto out_msgpool;
5172 
5173 	err = -ENOMEM;
5174 	osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
5175 	if (!osdc->notify_wq)
5176 		goto out_msgpool_reply;
5177 
5178 	osdc->completion_wq = create_singlethread_workqueue("ceph-completion");
5179 	if (!osdc->completion_wq)
5180 		goto out_notify_wq;
5181 
5182 	schedule_delayed_work(&osdc->timeout_work,
5183 			      osdc->client->options->osd_keepalive_timeout);
5184 	schedule_delayed_work(&osdc->osds_timeout_work,
5185 	    round_jiffies_relative(osdc->client->options->osd_idle_ttl));
5186 
5187 	return 0;
5188 
5189 out_notify_wq:
5190 	destroy_workqueue(osdc->notify_wq);
5191 out_msgpool_reply:
5192 	ceph_msgpool_destroy(&osdc->msgpool_op_reply);
5193 out_msgpool:
5194 	ceph_msgpool_destroy(&osdc->msgpool_op);
5195 out_mempool:
5196 	mempool_destroy(osdc->req_mempool);
5197 out_map:
5198 	ceph_osdmap_destroy(osdc->osdmap);
5199 out:
5200 	return err;
5201 }
5202 
5203 void ceph_osdc_stop(struct ceph_osd_client *osdc)
5204 {
5205 	destroy_workqueue(osdc->completion_wq);
5206 	destroy_workqueue(osdc->notify_wq);
5207 	cancel_delayed_work_sync(&osdc->timeout_work);
5208 	cancel_delayed_work_sync(&osdc->osds_timeout_work);
5209 
5210 	down_write(&osdc->lock);
5211 	while (!RB_EMPTY_ROOT(&osdc->osds)) {
5212 		struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
5213 						struct ceph_osd, o_node);
5214 		close_osd(osd);
5215 	}
5216 	up_write(&osdc->lock);
5217 	WARN_ON(refcount_read(&osdc->homeless_osd.o_ref) != 1);
5218 	osd_cleanup(&osdc->homeless_osd);
5219 
5220 	WARN_ON(!list_empty(&osdc->osd_lru));
5221 	WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_requests));
5222 	WARN_ON(!RB_EMPTY_ROOT(&osdc->map_checks));
5223 	WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_map_checks));
5224 	WARN_ON(atomic_read(&osdc->num_requests));
5225 	WARN_ON(atomic_read(&osdc->num_homeless));
5226 
5227 	ceph_osdmap_destroy(osdc->osdmap);
5228 	mempool_destroy(osdc->req_mempool);
5229 	ceph_msgpool_destroy(&osdc->msgpool_op);
5230 	ceph_msgpool_destroy(&osdc->msgpool_op_reply);
5231 }
5232 
5233 static int osd_req_op_copy_from_init(struct ceph_osd_request *req,
5234 				     u64 src_snapid, u64 src_version,
5235 				     struct ceph_object_id *src_oid,
5236 				     struct ceph_object_locator *src_oloc,
5237 				     u32 src_fadvise_flags,
5238 				     u32 dst_fadvise_flags,
5239 				     u32 truncate_seq, u64 truncate_size,
5240 				     u8 copy_from_flags)
5241 {
5242 	struct ceph_osd_req_op *op;
5243 	struct page **pages;
5244 	void *p, *end;
5245 
5246 	pages = ceph_alloc_page_vector(1, GFP_KERNEL);
5247 	if (IS_ERR(pages))
5248 		return PTR_ERR(pages);
5249 
5250 	op = _osd_req_op_init(req, 0, CEPH_OSD_OP_COPY_FROM2,
5251 			      dst_fadvise_flags);
5252 	op->copy_from.snapid = src_snapid;
5253 	op->copy_from.src_version = src_version;
5254 	op->copy_from.flags = copy_from_flags;
5255 	op->copy_from.src_fadvise_flags = src_fadvise_flags;
5256 
5257 	p = page_address(pages[0]);
5258 	end = p + PAGE_SIZE;
5259 	ceph_encode_string(&p, end, src_oid->name, src_oid->name_len);
5260 	encode_oloc(&p, end, src_oloc);
5261 	ceph_encode_32(&p, truncate_seq);
5262 	ceph_encode_64(&p, truncate_size);
5263 	op->indata_len = PAGE_SIZE - (end - p);
5264 
5265 	ceph_osd_data_pages_init(&op->copy_from.osd_data, pages,
5266 				 op->indata_len, 0, false, true);
5267 	return 0;
5268 }
5269 
5270 int ceph_osdc_copy_from(struct ceph_osd_client *osdc,
5271 			u64 src_snapid, u64 src_version,
5272 			struct ceph_object_id *src_oid,
5273 			struct ceph_object_locator *src_oloc,
5274 			u32 src_fadvise_flags,
5275 			struct ceph_object_id *dst_oid,
5276 			struct ceph_object_locator *dst_oloc,
5277 			u32 dst_fadvise_flags,
5278 			u32 truncate_seq, u64 truncate_size,
5279 			u8 copy_from_flags)
5280 {
5281 	struct ceph_osd_request *req;
5282 	int ret;
5283 
5284 	req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_KERNEL);
5285 	if (!req)
5286 		return -ENOMEM;
5287 
5288 	req->r_flags = CEPH_OSD_FLAG_WRITE;
5289 
5290 	ceph_oloc_copy(&req->r_t.base_oloc, dst_oloc);
5291 	ceph_oid_copy(&req->r_t.base_oid, dst_oid);
5292 
5293 	ret = osd_req_op_copy_from_init(req, src_snapid, src_version, src_oid,
5294 					src_oloc, src_fadvise_flags,
5295 					dst_fadvise_flags, truncate_seq,
5296 					truncate_size, copy_from_flags);
5297 	if (ret)
5298 		goto out;
5299 
5300 	ret = ceph_osdc_alloc_messages(req, GFP_KERNEL);
5301 	if (ret)
5302 		goto out;
5303 
5304 	ceph_osdc_start_request(osdc, req, false);
5305 	ret = ceph_osdc_wait_request(osdc, req);
5306 
5307 out:
5308 	ceph_osdc_put_request(req);
5309 	return ret;
5310 }
5311 EXPORT_SYMBOL(ceph_osdc_copy_from);
5312 
5313 int __init ceph_osdc_setup(void)
5314 {
5315 	size_t size = sizeof(struct ceph_osd_request) +
5316 	    CEPH_OSD_SLAB_OPS * sizeof(struct ceph_osd_req_op);
5317 
5318 	BUG_ON(ceph_osd_request_cache);
5319 	ceph_osd_request_cache = kmem_cache_create("ceph_osd_request", size,
5320 						   0, 0, NULL);
5321 
5322 	return ceph_osd_request_cache ? 0 : -ENOMEM;
5323 }
5324 
5325 void ceph_osdc_cleanup(void)
5326 {
5327 	BUG_ON(!ceph_osd_request_cache);
5328 	kmem_cache_destroy(ceph_osd_request_cache);
5329 	ceph_osd_request_cache = NULL;
5330 }
5331 
5332 /*
5333  * handle incoming message
5334  */
5335 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
5336 {
5337 	struct ceph_osd *osd = con->private;
5338 	struct ceph_osd_client *osdc = osd->o_osdc;
5339 	int type = le16_to_cpu(msg->hdr.type);
5340 
5341 	switch (type) {
5342 	case CEPH_MSG_OSD_MAP:
5343 		ceph_osdc_handle_map(osdc, msg);
5344 		break;
5345 	case CEPH_MSG_OSD_OPREPLY:
5346 		handle_reply(osd, msg);
5347 		break;
5348 	case CEPH_MSG_OSD_BACKOFF:
5349 		handle_backoff(osd, msg);
5350 		break;
5351 	case CEPH_MSG_WATCH_NOTIFY:
5352 		handle_watch_notify(osdc, msg);
5353 		break;
5354 
5355 	default:
5356 		pr_err("received unknown message type %d %s\n", type,
5357 		       ceph_msg_type_name(type));
5358 	}
5359 
5360 	ceph_msg_put(msg);
5361 }
5362 
5363 /*
5364  * Lookup and return message for incoming reply.  Don't try to do
5365  * anything about a larger than preallocated data portion of the
5366  * message at the moment - for now, just skip the message.
5367  */
5368 static struct ceph_msg *get_reply(struct ceph_connection *con,
5369 				  struct ceph_msg_header *hdr,
5370 				  int *skip)
5371 {
5372 	struct ceph_osd *osd = con->private;
5373 	struct ceph_osd_client *osdc = osd->o_osdc;
5374 	struct ceph_msg *m = NULL;
5375 	struct ceph_osd_request *req;
5376 	int front_len = le32_to_cpu(hdr->front_len);
5377 	int data_len = le32_to_cpu(hdr->data_len);
5378 	u64 tid = le64_to_cpu(hdr->tid);
5379 
5380 	down_read(&osdc->lock);
5381 	if (!osd_registered(osd)) {
5382 		dout("%s osd%d unknown, skipping\n", __func__, osd->o_osd);
5383 		*skip = 1;
5384 		goto out_unlock_osdc;
5385 	}
5386 	WARN_ON(osd->o_osd != le64_to_cpu(hdr->src.num));
5387 
5388 	mutex_lock(&osd->lock);
5389 	req = lookup_request(&osd->o_requests, tid);
5390 	if (!req) {
5391 		dout("%s osd%d tid %llu unknown, skipping\n", __func__,
5392 		     osd->o_osd, tid);
5393 		*skip = 1;
5394 		goto out_unlock_session;
5395 	}
5396 
5397 	ceph_msg_revoke_incoming(req->r_reply);
5398 
5399 	if (front_len > req->r_reply->front_alloc_len) {
5400 		pr_warn("%s osd%d tid %llu front %d > preallocated %d\n",
5401 			__func__, osd->o_osd, req->r_tid, front_len,
5402 			req->r_reply->front_alloc_len);
5403 		m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS,
5404 				 false);
5405 		if (!m)
5406 			goto out_unlock_session;
5407 		ceph_msg_put(req->r_reply);
5408 		req->r_reply = m;
5409 	}
5410 
5411 	if (data_len > req->r_reply->data_length) {
5412 		pr_warn("%s osd%d tid %llu data %d > preallocated %zu, skipping\n",
5413 			__func__, osd->o_osd, req->r_tid, data_len,
5414 			req->r_reply->data_length);
5415 		m = NULL;
5416 		*skip = 1;
5417 		goto out_unlock_session;
5418 	}
5419 
5420 	m = ceph_msg_get(req->r_reply);
5421 	dout("get_reply tid %lld %p\n", tid, m);
5422 
5423 out_unlock_session:
5424 	mutex_unlock(&osd->lock);
5425 out_unlock_osdc:
5426 	up_read(&osdc->lock);
5427 	return m;
5428 }
5429 
5430 static struct ceph_msg *alloc_msg_with_page_vector(struct ceph_msg_header *hdr)
5431 {
5432 	struct ceph_msg *m;
5433 	int type = le16_to_cpu(hdr->type);
5434 	u32 front_len = le32_to_cpu(hdr->front_len);
5435 	u32 data_len = le32_to_cpu(hdr->data_len);
5436 
5437 	m = ceph_msg_new2(type, front_len, 1, GFP_NOIO, false);
5438 	if (!m)
5439 		return NULL;
5440 
5441 	if (data_len) {
5442 		struct page **pages;
5443 
5444 		pages = ceph_alloc_page_vector(calc_pages_for(0, data_len),
5445 					       GFP_NOIO);
5446 		if (IS_ERR(pages)) {
5447 			ceph_msg_put(m);
5448 			return NULL;
5449 		}
5450 
5451 		ceph_msg_data_add_pages(m, pages, data_len, 0, true);
5452 	}
5453 
5454 	return m;
5455 }
5456 
5457 static struct ceph_msg *alloc_msg(struct ceph_connection *con,
5458 				  struct ceph_msg_header *hdr,
5459 				  int *skip)
5460 {
5461 	struct ceph_osd *osd = con->private;
5462 	int type = le16_to_cpu(hdr->type);
5463 
5464 	*skip = 0;
5465 	switch (type) {
5466 	case CEPH_MSG_OSD_MAP:
5467 	case CEPH_MSG_OSD_BACKOFF:
5468 	case CEPH_MSG_WATCH_NOTIFY:
5469 		return alloc_msg_with_page_vector(hdr);
5470 	case CEPH_MSG_OSD_OPREPLY:
5471 		return get_reply(con, hdr, skip);
5472 	default:
5473 		pr_warn("%s osd%d unknown msg type %d, skipping\n", __func__,
5474 			osd->o_osd, type);
5475 		*skip = 1;
5476 		return NULL;
5477 	}
5478 }
5479 
5480 /*
5481  * Wrappers to refcount containing ceph_osd struct
5482  */
5483 static struct ceph_connection *get_osd_con(struct ceph_connection *con)
5484 {
5485 	struct ceph_osd *osd = con->private;
5486 	if (get_osd(osd))
5487 		return con;
5488 	return NULL;
5489 }
5490 
5491 static void put_osd_con(struct ceph_connection *con)
5492 {
5493 	struct ceph_osd *osd = con->private;
5494 	put_osd(osd);
5495 }
5496 
5497 /*
5498  * authentication
5499  */
5500 /*
5501  * Note: returned pointer is the address of a structure that's
5502  * managed separately.  Caller must *not* attempt to free it.
5503  */
5504 static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
5505 					int *proto, int force_new)
5506 {
5507 	struct ceph_osd *o = con->private;
5508 	struct ceph_osd_client *osdc = o->o_osdc;
5509 	struct ceph_auth_client *ac = osdc->client->monc.auth;
5510 	struct ceph_auth_handshake *auth = &o->o_auth;
5511 
5512 	if (force_new && auth->authorizer) {
5513 		ceph_auth_destroy_authorizer(auth->authorizer);
5514 		auth->authorizer = NULL;
5515 	}
5516 	if (!auth->authorizer) {
5517 		int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
5518 						      auth);
5519 		if (ret)
5520 			return ERR_PTR(ret);
5521 	} else {
5522 		int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
5523 						     auth);
5524 		if (ret)
5525 			return ERR_PTR(ret);
5526 	}
5527 	*proto = ac->protocol;
5528 
5529 	return auth;
5530 }
5531 
5532 static int add_authorizer_challenge(struct ceph_connection *con,
5533 				    void *challenge_buf, int challenge_buf_len)
5534 {
5535 	struct ceph_osd *o = con->private;
5536 	struct ceph_osd_client *osdc = o->o_osdc;
5537 	struct ceph_auth_client *ac = osdc->client->monc.auth;
5538 
5539 	return ceph_auth_add_authorizer_challenge(ac, o->o_auth.authorizer,
5540 					    challenge_buf, challenge_buf_len);
5541 }
5542 
5543 static int verify_authorizer_reply(struct ceph_connection *con)
5544 {
5545 	struct ceph_osd *o = con->private;
5546 	struct ceph_osd_client *osdc = o->o_osdc;
5547 	struct ceph_auth_client *ac = osdc->client->monc.auth;
5548 
5549 	return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer);
5550 }
5551 
5552 static int invalidate_authorizer(struct ceph_connection *con)
5553 {
5554 	struct ceph_osd *o = con->private;
5555 	struct ceph_osd_client *osdc = o->o_osdc;
5556 	struct ceph_auth_client *ac = osdc->client->monc.auth;
5557 
5558 	ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
5559 	return ceph_monc_validate_auth(&osdc->client->monc);
5560 }
5561 
5562 static void osd_reencode_message(struct ceph_msg *msg)
5563 {
5564 	int type = le16_to_cpu(msg->hdr.type);
5565 
5566 	if (type == CEPH_MSG_OSD_OP)
5567 		encode_request_finish(msg);
5568 }
5569 
5570 static int osd_sign_message(struct ceph_msg *msg)
5571 {
5572 	struct ceph_osd *o = msg->con->private;
5573 	struct ceph_auth_handshake *auth = &o->o_auth;
5574 
5575 	return ceph_auth_sign_message(auth, msg);
5576 }
5577 
5578 static int osd_check_message_signature(struct ceph_msg *msg)
5579 {
5580 	struct ceph_osd *o = msg->con->private;
5581 	struct ceph_auth_handshake *auth = &o->o_auth;
5582 
5583 	return ceph_auth_check_message_signature(auth, msg);
5584 }
5585 
5586 static const struct ceph_connection_operations osd_con_ops = {
5587 	.get = get_osd_con,
5588 	.put = put_osd_con,
5589 	.dispatch = dispatch,
5590 	.get_authorizer = get_authorizer,
5591 	.add_authorizer_challenge = add_authorizer_challenge,
5592 	.verify_authorizer_reply = verify_authorizer_reply,
5593 	.invalidate_authorizer = invalidate_authorizer,
5594 	.alloc_msg = alloc_msg,
5595 	.reencode_message = osd_reencode_message,
5596 	.sign_message = osd_sign_message,
5597 	.check_message_signature = osd_check_message_signature,
5598 	.fault = osd_fault,
5599 };
5600