xref: /linux/net/ceph/osd_client.c (revision 6a069876eb1402478900ee0eb7d7fe276bb1f4e3)
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 void ceph_osd_iter_init(struct ceph_osd_data *osd_data,
175 			       struct iov_iter *iter)
176 {
177 	osd_data->type = CEPH_OSD_DATA_TYPE_ITER;
178 	osd_data->iter = *iter;
179 }
180 
181 static struct ceph_osd_data *
182 osd_req_op_raw_data_in(struct ceph_osd_request *osd_req, unsigned int which)
183 {
184 	BUG_ON(which >= osd_req->r_num_ops);
185 
186 	return &osd_req->r_ops[which].raw_data_in;
187 }
188 
189 struct ceph_osd_data *
190 osd_req_op_extent_osd_data(struct ceph_osd_request *osd_req,
191 			unsigned int which)
192 {
193 	return osd_req_op_data(osd_req, which, extent, osd_data);
194 }
195 EXPORT_SYMBOL(osd_req_op_extent_osd_data);
196 
197 void osd_req_op_raw_data_in_pages(struct ceph_osd_request *osd_req,
198 			unsigned int which, struct page **pages,
199 			u64 length, u32 alignment,
200 			bool pages_from_pool, bool own_pages)
201 {
202 	struct ceph_osd_data *osd_data;
203 
204 	osd_data = osd_req_op_raw_data_in(osd_req, which);
205 	ceph_osd_data_pages_init(osd_data, pages, length, alignment,
206 				pages_from_pool, own_pages);
207 }
208 EXPORT_SYMBOL(osd_req_op_raw_data_in_pages);
209 
210 void osd_req_op_extent_osd_data_pages(struct ceph_osd_request *osd_req,
211 			unsigned int which, struct page **pages,
212 			u64 length, u32 alignment,
213 			bool pages_from_pool, bool own_pages)
214 {
215 	struct ceph_osd_data *osd_data;
216 
217 	osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
218 	ceph_osd_data_pages_init(osd_data, pages, length, alignment,
219 				pages_from_pool, own_pages);
220 }
221 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pages);
222 
223 #ifdef CONFIG_BLOCK
224 void osd_req_op_extent_osd_data_bio(struct ceph_osd_request *osd_req,
225 				    unsigned int which,
226 				    struct ceph_bio_iter *bio_pos,
227 				    u32 bio_length)
228 {
229 	struct ceph_osd_data *osd_data;
230 
231 	osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
232 	ceph_osd_data_bio_init(osd_data, bio_pos, bio_length);
233 }
234 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bio);
235 #endif /* CONFIG_BLOCK */
236 
237 void osd_req_op_extent_osd_data_bvecs(struct ceph_osd_request *osd_req,
238 				      unsigned int which,
239 				      struct bio_vec *bvecs, u32 num_bvecs,
240 				      u32 bytes)
241 {
242 	struct ceph_osd_data *osd_data;
243 	struct ceph_bvec_iter it = {
244 		.bvecs = bvecs,
245 		.iter = { .bi_size = bytes },
246 	};
247 
248 	osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
249 	ceph_osd_data_bvecs_init(osd_data, &it, num_bvecs);
250 }
251 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bvecs);
252 
253 void osd_req_op_extent_osd_data_bvec_pos(struct ceph_osd_request *osd_req,
254 					 unsigned int which,
255 					 struct ceph_bvec_iter *bvec_pos)
256 {
257 	struct ceph_osd_data *osd_data;
258 
259 	osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
260 	ceph_osd_data_bvecs_init(osd_data, bvec_pos, 0);
261 }
262 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bvec_pos);
263 
264 /**
265  * osd_req_op_extent_osd_iter - Set up an operation with an iterator buffer
266  * @osd_req: The request to set up
267  * @which: Index of the operation in which to set the iter
268  * @iter: The buffer iterator
269  */
270 void osd_req_op_extent_osd_iter(struct ceph_osd_request *osd_req,
271 				unsigned int which, struct iov_iter *iter)
272 {
273 	struct ceph_osd_data *osd_data;
274 
275 	osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
276 	ceph_osd_iter_init(osd_data, iter);
277 }
278 EXPORT_SYMBOL(osd_req_op_extent_osd_iter);
279 
280 static void osd_req_op_cls_request_info_pagelist(
281 			struct ceph_osd_request *osd_req,
282 			unsigned int which, struct ceph_pagelist *pagelist)
283 {
284 	struct ceph_osd_data *osd_data;
285 
286 	osd_data = osd_req_op_data(osd_req, which, cls, request_info);
287 	ceph_osd_data_pagelist_init(osd_data, pagelist);
288 }
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 	case CEPH_OSD_DATA_TYPE_ITER:
350 		return iov_iter_count(&osd_data->iter);
351 	default:
352 		WARN(true, "unrecognized data type %d\n", (int)osd_data->type);
353 		return 0;
354 	}
355 }
356 
357 static void ceph_osd_data_release(struct ceph_osd_data *osd_data)
358 {
359 	if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES && osd_data->own_pages) {
360 		int num_pages;
361 
362 		num_pages = calc_pages_for((u64)osd_data->alignment,
363 						(u64)osd_data->length);
364 		ceph_release_page_vector(osd_data->pages, num_pages);
365 	} else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
366 		ceph_pagelist_release(osd_data->pagelist);
367 	}
368 	ceph_osd_data_init(osd_data);
369 }
370 
371 static void osd_req_op_data_release(struct ceph_osd_request *osd_req,
372 			unsigned int which)
373 {
374 	struct ceph_osd_req_op *op;
375 
376 	BUG_ON(which >= osd_req->r_num_ops);
377 	op = &osd_req->r_ops[which];
378 
379 	switch (op->op) {
380 	case CEPH_OSD_OP_READ:
381 	case CEPH_OSD_OP_SPARSE_READ:
382 	case CEPH_OSD_OP_WRITE:
383 	case CEPH_OSD_OP_WRITEFULL:
384 		kfree(op->extent.sparse_ext);
385 		ceph_osd_data_release(&op->extent.osd_data);
386 		break;
387 	case CEPH_OSD_OP_CALL:
388 		ceph_osd_data_release(&op->cls.request_info);
389 		ceph_osd_data_release(&op->cls.request_data);
390 		ceph_osd_data_release(&op->cls.response_data);
391 		break;
392 	case CEPH_OSD_OP_SETXATTR:
393 	case CEPH_OSD_OP_CMPXATTR:
394 		ceph_osd_data_release(&op->xattr.osd_data);
395 		break;
396 	case CEPH_OSD_OP_STAT:
397 		ceph_osd_data_release(&op->raw_data_in);
398 		break;
399 	case CEPH_OSD_OP_NOTIFY_ACK:
400 		ceph_osd_data_release(&op->notify_ack.request_data);
401 		break;
402 	case CEPH_OSD_OP_NOTIFY:
403 		ceph_osd_data_release(&op->notify.request_data);
404 		ceph_osd_data_release(&op->notify.response_data);
405 		break;
406 	case CEPH_OSD_OP_LIST_WATCHERS:
407 		ceph_osd_data_release(&op->list_watchers.response_data);
408 		break;
409 	case CEPH_OSD_OP_COPY_FROM2:
410 		ceph_osd_data_release(&op->copy_from.osd_data);
411 		break;
412 	default:
413 		break;
414 	}
415 }
416 
417 /*
418  * Assumes @t is zero-initialized.
419  */
420 static void target_init(struct ceph_osd_request_target *t)
421 {
422 	ceph_oid_init(&t->base_oid);
423 	ceph_oloc_init(&t->base_oloc);
424 	ceph_oid_init(&t->target_oid);
425 	ceph_oloc_init(&t->target_oloc);
426 
427 	ceph_osds_init(&t->acting);
428 	ceph_osds_init(&t->up);
429 	t->size = -1;
430 	t->min_size = -1;
431 
432 	t->osd = CEPH_HOMELESS_OSD;
433 }
434 
435 static void target_copy(struct ceph_osd_request_target *dest,
436 			const struct ceph_osd_request_target *src)
437 {
438 	ceph_oid_copy(&dest->base_oid, &src->base_oid);
439 	ceph_oloc_copy(&dest->base_oloc, &src->base_oloc);
440 	ceph_oid_copy(&dest->target_oid, &src->target_oid);
441 	ceph_oloc_copy(&dest->target_oloc, &src->target_oloc);
442 
443 	dest->pgid = src->pgid; /* struct */
444 	dest->spgid = src->spgid; /* struct */
445 	dest->pg_num = src->pg_num;
446 	dest->pg_num_mask = src->pg_num_mask;
447 	ceph_osds_copy(&dest->acting, &src->acting);
448 	ceph_osds_copy(&dest->up, &src->up);
449 	dest->size = src->size;
450 	dest->min_size = src->min_size;
451 	dest->sort_bitwise = src->sort_bitwise;
452 	dest->recovery_deletes = src->recovery_deletes;
453 
454 	dest->flags = src->flags;
455 	dest->used_replica = src->used_replica;
456 	dest->paused = src->paused;
457 
458 	dest->epoch = src->epoch;
459 	dest->last_force_resend = src->last_force_resend;
460 
461 	dest->osd = src->osd;
462 }
463 
464 static void target_destroy(struct ceph_osd_request_target *t)
465 {
466 	ceph_oid_destroy(&t->base_oid);
467 	ceph_oloc_destroy(&t->base_oloc);
468 	ceph_oid_destroy(&t->target_oid);
469 	ceph_oloc_destroy(&t->target_oloc);
470 }
471 
472 /*
473  * requests
474  */
475 static void request_release_checks(struct ceph_osd_request *req)
476 {
477 	WARN_ON(!RB_EMPTY_NODE(&req->r_node));
478 	WARN_ON(!RB_EMPTY_NODE(&req->r_mc_node));
479 	WARN_ON(!list_empty(&req->r_private_item));
480 	WARN_ON(req->r_osd);
481 }
482 
483 static void ceph_osdc_release_request(struct kref *kref)
484 {
485 	struct ceph_osd_request *req = container_of(kref,
486 					    struct ceph_osd_request, r_kref);
487 	unsigned int which;
488 
489 	dout("%s %p (r_request %p r_reply %p)\n", __func__, req,
490 	     req->r_request, req->r_reply);
491 	request_release_checks(req);
492 
493 	if (req->r_request)
494 		ceph_msg_put(req->r_request);
495 	if (req->r_reply)
496 		ceph_msg_put(req->r_reply);
497 
498 	for (which = 0; which < req->r_num_ops; which++)
499 		osd_req_op_data_release(req, which);
500 
501 	target_destroy(&req->r_t);
502 	ceph_put_snap_context(req->r_snapc);
503 
504 	if (req->r_mempool)
505 		mempool_free(req, req->r_osdc->req_mempool);
506 	else if (req->r_num_ops <= CEPH_OSD_SLAB_OPS)
507 		kmem_cache_free(ceph_osd_request_cache, req);
508 	else
509 		kfree(req);
510 }
511 
512 void ceph_osdc_get_request(struct ceph_osd_request *req)
513 {
514 	dout("%s %p (was %d)\n", __func__, req,
515 	     kref_read(&req->r_kref));
516 	kref_get(&req->r_kref);
517 }
518 EXPORT_SYMBOL(ceph_osdc_get_request);
519 
520 void ceph_osdc_put_request(struct ceph_osd_request *req)
521 {
522 	if (req) {
523 		dout("%s %p (was %d)\n", __func__, req,
524 		     kref_read(&req->r_kref));
525 		kref_put(&req->r_kref, ceph_osdc_release_request);
526 	}
527 }
528 EXPORT_SYMBOL(ceph_osdc_put_request);
529 
530 static void request_init(struct ceph_osd_request *req)
531 {
532 	/* req only, each op is zeroed in osd_req_op_init() */
533 	memset(req, 0, sizeof(*req));
534 
535 	kref_init(&req->r_kref);
536 	init_completion(&req->r_completion);
537 	RB_CLEAR_NODE(&req->r_node);
538 	RB_CLEAR_NODE(&req->r_mc_node);
539 	INIT_LIST_HEAD(&req->r_private_item);
540 
541 	target_init(&req->r_t);
542 }
543 
544 struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
545 					       struct ceph_snap_context *snapc,
546 					       unsigned int num_ops,
547 					       bool use_mempool,
548 					       gfp_t gfp_flags)
549 {
550 	struct ceph_osd_request *req;
551 
552 	if (use_mempool) {
553 		BUG_ON(num_ops > CEPH_OSD_SLAB_OPS);
554 		req = mempool_alloc(osdc->req_mempool, gfp_flags);
555 	} else if (num_ops <= CEPH_OSD_SLAB_OPS) {
556 		req = kmem_cache_alloc(ceph_osd_request_cache, gfp_flags);
557 	} else {
558 		BUG_ON(num_ops > CEPH_OSD_MAX_OPS);
559 		req = kmalloc(struct_size(req, r_ops, num_ops), gfp_flags);
560 	}
561 	if (unlikely(!req))
562 		return NULL;
563 
564 	request_init(req);
565 	req->r_osdc = osdc;
566 	req->r_mempool = use_mempool;
567 	req->r_num_ops = num_ops;
568 	req->r_snapid = CEPH_NOSNAP;
569 	req->r_snapc = ceph_get_snap_context(snapc);
570 
571 	dout("%s req %p\n", __func__, req);
572 	return req;
573 }
574 EXPORT_SYMBOL(ceph_osdc_alloc_request);
575 
576 static int ceph_oloc_encoding_size(const struct ceph_object_locator *oloc)
577 {
578 	return 8 + 4 + 4 + 4 + (oloc->pool_ns ? oloc->pool_ns->len : 0);
579 }
580 
581 static int __ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp,
582 				      int num_request_data_items,
583 				      int num_reply_data_items)
584 {
585 	struct ceph_osd_client *osdc = req->r_osdc;
586 	struct ceph_msg *msg;
587 	int msg_size;
588 
589 	WARN_ON(req->r_request || req->r_reply);
590 	WARN_ON(ceph_oid_empty(&req->r_base_oid));
591 	WARN_ON(ceph_oloc_empty(&req->r_base_oloc));
592 
593 	/* create request message */
594 	msg_size = CEPH_ENCODING_START_BLK_LEN +
595 			CEPH_PGID_ENCODING_LEN + 1; /* spgid */
596 	msg_size += 4 + 4 + 4; /* hash, osdmap_epoch, flags */
597 	msg_size += CEPH_ENCODING_START_BLK_LEN +
598 			sizeof(struct ceph_osd_reqid); /* reqid */
599 	msg_size += sizeof(struct ceph_blkin_trace_info); /* trace */
600 	msg_size += 4 + sizeof(struct ceph_timespec); /* client_inc, mtime */
601 	msg_size += CEPH_ENCODING_START_BLK_LEN +
602 			ceph_oloc_encoding_size(&req->r_base_oloc); /* oloc */
603 	msg_size += 4 + req->r_base_oid.name_len; /* oid */
604 	msg_size += 2 + req->r_num_ops * sizeof(struct ceph_osd_op);
605 	msg_size += 8; /* snapid */
606 	msg_size += 8; /* snap_seq */
607 	msg_size += 4 + 8 * (req->r_snapc ? req->r_snapc->num_snaps : 0);
608 	msg_size += 4 + 8; /* retry_attempt, features */
609 
610 	if (req->r_mempool)
611 		msg = ceph_msgpool_get(&osdc->msgpool_op, msg_size,
612 				       num_request_data_items);
613 	else
614 		msg = ceph_msg_new2(CEPH_MSG_OSD_OP, msg_size,
615 				    num_request_data_items, gfp, true);
616 	if (!msg)
617 		return -ENOMEM;
618 
619 	memset(msg->front.iov_base, 0, msg->front.iov_len);
620 	req->r_request = msg;
621 
622 	/* create reply message */
623 	msg_size = OSD_OPREPLY_FRONT_LEN;
624 	msg_size += req->r_base_oid.name_len;
625 	msg_size += req->r_num_ops * sizeof(struct ceph_osd_op);
626 
627 	if (req->r_mempool)
628 		msg = ceph_msgpool_get(&osdc->msgpool_op_reply, msg_size,
629 				       num_reply_data_items);
630 	else
631 		msg = ceph_msg_new2(CEPH_MSG_OSD_OPREPLY, msg_size,
632 				    num_reply_data_items, gfp, true);
633 	if (!msg)
634 		return -ENOMEM;
635 
636 	req->r_reply = msg;
637 
638 	return 0;
639 }
640 
641 static bool osd_req_opcode_valid(u16 opcode)
642 {
643 	switch (opcode) {
644 #define GENERATE_CASE(op, opcode, str)	case CEPH_OSD_OP_##op: return true;
645 __CEPH_FORALL_OSD_OPS(GENERATE_CASE)
646 #undef GENERATE_CASE
647 	default:
648 		return false;
649 	}
650 }
651 
652 static void get_num_data_items(struct ceph_osd_request *req,
653 			       int *num_request_data_items,
654 			       int *num_reply_data_items)
655 {
656 	struct ceph_osd_req_op *op;
657 
658 	*num_request_data_items = 0;
659 	*num_reply_data_items = 0;
660 
661 	for (op = req->r_ops; op != &req->r_ops[req->r_num_ops]; op++) {
662 		switch (op->op) {
663 		/* request */
664 		case CEPH_OSD_OP_WRITE:
665 		case CEPH_OSD_OP_WRITEFULL:
666 		case CEPH_OSD_OP_SETXATTR:
667 		case CEPH_OSD_OP_CMPXATTR:
668 		case CEPH_OSD_OP_NOTIFY_ACK:
669 		case CEPH_OSD_OP_COPY_FROM2:
670 			*num_request_data_items += 1;
671 			break;
672 
673 		/* reply */
674 		case CEPH_OSD_OP_STAT:
675 		case CEPH_OSD_OP_READ:
676 		case CEPH_OSD_OP_SPARSE_READ:
677 		case CEPH_OSD_OP_LIST_WATCHERS:
678 			*num_reply_data_items += 1;
679 			break;
680 
681 		/* both */
682 		case CEPH_OSD_OP_NOTIFY:
683 			*num_request_data_items += 1;
684 			*num_reply_data_items += 1;
685 			break;
686 		case CEPH_OSD_OP_CALL:
687 			*num_request_data_items += 2;
688 			*num_reply_data_items += 1;
689 			break;
690 
691 		default:
692 			WARN_ON(!osd_req_opcode_valid(op->op));
693 			break;
694 		}
695 	}
696 }
697 
698 /*
699  * oid, oloc and OSD op opcode(s) must be filled in before this function
700  * is called.
701  */
702 int ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp)
703 {
704 	int num_request_data_items, num_reply_data_items;
705 
706 	get_num_data_items(req, &num_request_data_items, &num_reply_data_items);
707 	return __ceph_osdc_alloc_messages(req, gfp, num_request_data_items,
708 					  num_reply_data_items);
709 }
710 EXPORT_SYMBOL(ceph_osdc_alloc_messages);
711 
712 /*
713  * This is an osd op init function for opcodes that have no data or
714  * other information associated with them.  It also serves as a
715  * common init routine for all the other init functions, below.
716  */
717 struct ceph_osd_req_op *
718 osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which,
719 		 u16 opcode, u32 flags)
720 {
721 	struct ceph_osd_req_op *op;
722 
723 	BUG_ON(which >= osd_req->r_num_ops);
724 	BUG_ON(!osd_req_opcode_valid(opcode));
725 
726 	op = &osd_req->r_ops[which];
727 	memset(op, 0, sizeof (*op));
728 	op->op = opcode;
729 	op->flags = flags;
730 
731 	return op;
732 }
733 EXPORT_SYMBOL(osd_req_op_init);
734 
735 void osd_req_op_extent_init(struct ceph_osd_request *osd_req,
736 				unsigned int which, u16 opcode,
737 				u64 offset, u64 length,
738 				u64 truncate_size, u32 truncate_seq)
739 {
740 	struct ceph_osd_req_op *op = osd_req_op_init(osd_req, which,
741 						     opcode, 0);
742 	size_t payload_len = 0;
743 
744 	BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
745 	       opcode != CEPH_OSD_OP_WRITEFULL && opcode != CEPH_OSD_OP_ZERO &&
746 	       opcode != CEPH_OSD_OP_TRUNCATE && opcode != CEPH_OSD_OP_SPARSE_READ);
747 
748 	op->extent.offset = offset;
749 	op->extent.length = length;
750 	op->extent.truncate_size = truncate_size;
751 	op->extent.truncate_seq = truncate_seq;
752 	if (opcode == CEPH_OSD_OP_WRITE || opcode == CEPH_OSD_OP_WRITEFULL)
753 		payload_len += length;
754 
755 	op->indata_len = payload_len;
756 }
757 EXPORT_SYMBOL(osd_req_op_extent_init);
758 
759 void osd_req_op_extent_update(struct ceph_osd_request *osd_req,
760 				unsigned int which, u64 length)
761 {
762 	struct ceph_osd_req_op *op;
763 	u64 previous;
764 
765 	BUG_ON(which >= osd_req->r_num_ops);
766 	op = &osd_req->r_ops[which];
767 	previous = op->extent.length;
768 
769 	if (length == previous)
770 		return;		/* Nothing to do */
771 	BUG_ON(length > previous);
772 
773 	op->extent.length = length;
774 	if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL)
775 		op->indata_len -= previous - length;
776 }
777 EXPORT_SYMBOL(osd_req_op_extent_update);
778 
779 void osd_req_op_extent_dup_last(struct ceph_osd_request *osd_req,
780 				unsigned int which, u64 offset_inc)
781 {
782 	struct ceph_osd_req_op *op, *prev_op;
783 
784 	BUG_ON(which + 1 >= osd_req->r_num_ops);
785 
786 	prev_op = &osd_req->r_ops[which];
787 	op = osd_req_op_init(osd_req, which + 1, prev_op->op, prev_op->flags);
788 	/* dup previous one */
789 	op->indata_len = prev_op->indata_len;
790 	op->outdata_len = prev_op->outdata_len;
791 	op->extent = prev_op->extent;
792 	/* adjust offset */
793 	op->extent.offset += offset_inc;
794 	op->extent.length -= offset_inc;
795 
796 	if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL)
797 		op->indata_len -= offset_inc;
798 }
799 EXPORT_SYMBOL(osd_req_op_extent_dup_last);
800 
801 int osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
802 			const char *class, const char *method)
803 {
804 	struct ceph_osd_req_op *op;
805 	struct ceph_pagelist *pagelist;
806 	size_t payload_len = 0;
807 	size_t size;
808 	int ret;
809 
810 	op = osd_req_op_init(osd_req, which, CEPH_OSD_OP_CALL, 0);
811 
812 	pagelist = ceph_pagelist_alloc(GFP_NOFS);
813 	if (!pagelist)
814 		return -ENOMEM;
815 
816 	op->cls.class_name = class;
817 	size = strlen(class);
818 	BUG_ON(size > (size_t) U8_MAX);
819 	op->cls.class_len = size;
820 	ret = ceph_pagelist_append(pagelist, class, size);
821 	if (ret)
822 		goto err_pagelist_free;
823 	payload_len += size;
824 
825 	op->cls.method_name = method;
826 	size = strlen(method);
827 	BUG_ON(size > (size_t) U8_MAX);
828 	op->cls.method_len = size;
829 	ret = ceph_pagelist_append(pagelist, method, size);
830 	if (ret)
831 		goto err_pagelist_free;
832 	payload_len += size;
833 
834 	osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist);
835 	op->indata_len = payload_len;
836 	return 0;
837 
838 err_pagelist_free:
839 	ceph_pagelist_release(pagelist);
840 	return ret;
841 }
842 EXPORT_SYMBOL(osd_req_op_cls_init);
843 
844 int osd_req_op_xattr_init(struct ceph_osd_request *osd_req, unsigned int which,
845 			  u16 opcode, const char *name, const void *value,
846 			  size_t size, u8 cmp_op, u8 cmp_mode)
847 {
848 	struct ceph_osd_req_op *op = osd_req_op_init(osd_req, which,
849 						     opcode, 0);
850 	struct ceph_pagelist *pagelist;
851 	size_t payload_len;
852 	int ret;
853 
854 	BUG_ON(opcode != CEPH_OSD_OP_SETXATTR && opcode != CEPH_OSD_OP_CMPXATTR);
855 
856 	pagelist = ceph_pagelist_alloc(GFP_NOFS);
857 	if (!pagelist)
858 		return -ENOMEM;
859 
860 	payload_len = strlen(name);
861 	op->xattr.name_len = payload_len;
862 	ret = ceph_pagelist_append(pagelist, name, payload_len);
863 	if (ret)
864 		goto err_pagelist_free;
865 
866 	op->xattr.value_len = size;
867 	ret = ceph_pagelist_append(pagelist, value, size);
868 	if (ret)
869 		goto err_pagelist_free;
870 	payload_len += size;
871 
872 	op->xattr.cmp_op = cmp_op;
873 	op->xattr.cmp_mode = cmp_mode;
874 
875 	ceph_osd_data_pagelist_init(&op->xattr.osd_data, pagelist);
876 	op->indata_len = payload_len;
877 	return 0;
878 
879 err_pagelist_free:
880 	ceph_pagelist_release(pagelist);
881 	return ret;
882 }
883 EXPORT_SYMBOL(osd_req_op_xattr_init);
884 
885 /*
886  * @watch_opcode: CEPH_OSD_WATCH_OP_*
887  */
888 static void osd_req_op_watch_init(struct ceph_osd_request *req, int which,
889 				  u8 watch_opcode, u64 cookie, u32 gen)
890 {
891 	struct ceph_osd_req_op *op;
892 
893 	op = osd_req_op_init(req, which, CEPH_OSD_OP_WATCH, 0);
894 	op->watch.cookie = cookie;
895 	op->watch.op = watch_opcode;
896 	op->watch.gen = gen;
897 }
898 
899 /*
900  * prot_ver, timeout and notify payload (may be empty) should already be
901  * encoded in @request_pl
902  */
903 static void osd_req_op_notify_init(struct ceph_osd_request *req, int which,
904 				   u64 cookie, struct ceph_pagelist *request_pl)
905 {
906 	struct ceph_osd_req_op *op;
907 
908 	op = osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY, 0);
909 	op->notify.cookie = cookie;
910 
911 	ceph_osd_data_pagelist_init(&op->notify.request_data, request_pl);
912 	op->indata_len = request_pl->length;
913 }
914 
915 /*
916  * @flags: CEPH_OSD_OP_ALLOC_HINT_FLAG_*
917  */
918 void osd_req_op_alloc_hint_init(struct ceph_osd_request *osd_req,
919 				unsigned int which,
920 				u64 expected_object_size,
921 				u64 expected_write_size,
922 				u32 flags)
923 {
924 	struct ceph_osd_req_op *op;
925 
926 	op = osd_req_op_init(osd_req, which, CEPH_OSD_OP_SETALLOCHINT, 0);
927 	op->alloc_hint.expected_object_size = expected_object_size;
928 	op->alloc_hint.expected_write_size = expected_write_size;
929 	op->alloc_hint.flags = flags;
930 
931 	/*
932 	 * CEPH_OSD_OP_SETALLOCHINT op is advisory and therefore deemed
933 	 * not worth a feature bit.  Set FAILOK per-op flag to make
934 	 * sure older osds don't trip over an unsupported opcode.
935 	 */
936 	op->flags |= CEPH_OSD_OP_FLAG_FAILOK;
937 }
938 EXPORT_SYMBOL(osd_req_op_alloc_hint_init);
939 
940 static void ceph_osdc_msg_data_add(struct ceph_msg *msg,
941 				struct ceph_osd_data *osd_data)
942 {
943 	u64 length = ceph_osd_data_length(osd_data);
944 
945 	if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
946 		BUG_ON(length > (u64) SIZE_MAX);
947 		if (length)
948 			ceph_msg_data_add_pages(msg, osd_data->pages,
949 					length, osd_data->alignment, false);
950 	} else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
951 		BUG_ON(!length);
952 		ceph_msg_data_add_pagelist(msg, osd_data->pagelist);
953 #ifdef CONFIG_BLOCK
954 	} else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
955 		ceph_msg_data_add_bio(msg, &osd_data->bio_pos, length);
956 #endif
957 	} else if (osd_data->type == CEPH_OSD_DATA_TYPE_BVECS) {
958 		ceph_msg_data_add_bvecs(msg, &osd_data->bvec_pos);
959 	} else if (osd_data->type == CEPH_OSD_DATA_TYPE_ITER) {
960 		ceph_msg_data_add_iter(msg, &osd_data->iter);
961 	} else {
962 		BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
963 	}
964 }
965 
966 static u32 osd_req_encode_op(struct ceph_osd_op *dst,
967 			     const struct ceph_osd_req_op *src)
968 {
969 	switch (src->op) {
970 	case CEPH_OSD_OP_STAT:
971 		break;
972 	case CEPH_OSD_OP_READ:
973 	case CEPH_OSD_OP_SPARSE_READ:
974 	case CEPH_OSD_OP_WRITE:
975 	case CEPH_OSD_OP_WRITEFULL:
976 	case CEPH_OSD_OP_ZERO:
977 	case CEPH_OSD_OP_TRUNCATE:
978 		dst->extent.offset = cpu_to_le64(src->extent.offset);
979 		dst->extent.length = cpu_to_le64(src->extent.length);
980 		dst->extent.truncate_size =
981 			cpu_to_le64(src->extent.truncate_size);
982 		dst->extent.truncate_seq =
983 			cpu_to_le32(src->extent.truncate_seq);
984 		break;
985 	case CEPH_OSD_OP_CALL:
986 		dst->cls.class_len = src->cls.class_len;
987 		dst->cls.method_len = src->cls.method_len;
988 		dst->cls.indata_len = cpu_to_le32(src->cls.indata_len);
989 		break;
990 	case CEPH_OSD_OP_WATCH:
991 		dst->watch.cookie = cpu_to_le64(src->watch.cookie);
992 		dst->watch.ver = cpu_to_le64(0);
993 		dst->watch.op = src->watch.op;
994 		dst->watch.gen = cpu_to_le32(src->watch.gen);
995 		break;
996 	case CEPH_OSD_OP_NOTIFY_ACK:
997 		break;
998 	case CEPH_OSD_OP_NOTIFY:
999 		dst->notify.cookie = cpu_to_le64(src->notify.cookie);
1000 		break;
1001 	case CEPH_OSD_OP_LIST_WATCHERS:
1002 		break;
1003 	case CEPH_OSD_OP_SETALLOCHINT:
1004 		dst->alloc_hint.expected_object_size =
1005 		    cpu_to_le64(src->alloc_hint.expected_object_size);
1006 		dst->alloc_hint.expected_write_size =
1007 		    cpu_to_le64(src->alloc_hint.expected_write_size);
1008 		dst->alloc_hint.flags = cpu_to_le32(src->alloc_hint.flags);
1009 		break;
1010 	case CEPH_OSD_OP_SETXATTR:
1011 	case CEPH_OSD_OP_CMPXATTR:
1012 		dst->xattr.name_len = cpu_to_le32(src->xattr.name_len);
1013 		dst->xattr.value_len = cpu_to_le32(src->xattr.value_len);
1014 		dst->xattr.cmp_op = src->xattr.cmp_op;
1015 		dst->xattr.cmp_mode = src->xattr.cmp_mode;
1016 		break;
1017 	case CEPH_OSD_OP_CREATE:
1018 	case CEPH_OSD_OP_DELETE:
1019 		break;
1020 	case CEPH_OSD_OP_COPY_FROM2:
1021 		dst->copy_from.snapid = cpu_to_le64(src->copy_from.snapid);
1022 		dst->copy_from.src_version =
1023 			cpu_to_le64(src->copy_from.src_version);
1024 		dst->copy_from.flags = src->copy_from.flags;
1025 		dst->copy_from.src_fadvise_flags =
1026 			cpu_to_le32(src->copy_from.src_fadvise_flags);
1027 		break;
1028 	case CEPH_OSD_OP_ASSERT_VER:
1029 		dst->assert_ver.unused = cpu_to_le64(0);
1030 		dst->assert_ver.ver = cpu_to_le64(src->assert_ver.ver);
1031 		break;
1032 	default:
1033 		pr_err("unsupported osd opcode %s\n",
1034 			ceph_osd_op_name(src->op));
1035 		WARN_ON(1);
1036 
1037 		return 0;
1038 	}
1039 
1040 	dst->op = cpu_to_le16(src->op);
1041 	dst->flags = cpu_to_le32(src->flags);
1042 	dst->payload_len = cpu_to_le32(src->indata_len);
1043 
1044 	return src->indata_len;
1045 }
1046 
1047 /*
1048  * build new request AND message, calculate layout, and adjust file
1049  * extent as needed.
1050  *
1051  * if the file was recently truncated, we include information about its
1052  * old and new size so that the object can be updated appropriately.  (we
1053  * avoid synchronously deleting truncated objects because it's slow.)
1054  */
1055 struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
1056 					       struct ceph_file_layout *layout,
1057 					       struct ceph_vino vino,
1058 					       u64 off, u64 *plen,
1059 					       unsigned int which, int num_ops,
1060 					       int opcode, int flags,
1061 					       struct ceph_snap_context *snapc,
1062 					       u32 truncate_seq,
1063 					       u64 truncate_size,
1064 					       bool use_mempool)
1065 {
1066 	struct ceph_osd_request *req;
1067 	u64 objnum = 0;
1068 	u64 objoff = 0;
1069 	u64 objlen = 0;
1070 	int r;
1071 
1072 	BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
1073 	       opcode != CEPH_OSD_OP_ZERO && opcode != CEPH_OSD_OP_TRUNCATE &&
1074 	       opcode != CEPH_OSD_OP_CREATE && opcode != CEPH_OSD_OP_DELETE &&
1075 	       opcode != CEPH_OSD_OP_SPARSE_READ);
1076 
1077 	req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool,
1078 					GFP_NOFS);
1079 	if (!req) {
1080 		r = -ENOMEM;
1081 		goto fail;
1082 	}
1083 
1084 	/* calculate max write size */
1085 	r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
1086 	if (r)
1087 		goto fail;
1088 
1089 	if (opcode == CEPH_OSD_OP_CREATE || opcode == CEPH_OSD_OP_DELETE) {
1090 		osd_req_op_init(req, which, opcode, 0);
1091 	} else {
1092 		u32 object_size = layout->object_size;
1093 		u32 object_base = off - objoff;
1094 		if (!(truncate_seq == 1 && truncate_size == -1ULL)) {
1095 			if (truncate_size <= object_base) {
1096 				truncate_size = 0;
1097 			} else {
1098 				truncate_size -= object_base;
1099 				if (truncate_size > object_size)
1100 					truncate_size = object_size;
1101 			}
1102 		}
1103 		osd_req_op_extent_init(req, which, opcode, objoff, objlen,
1104 				       truncate_size, truncate_seq);
1105 	}
1106 
1107 	req->r_base_oloc.pool = layout->pool_id;
1108 	req->r_base_oloc.pool_ns = ceph_try_get_string(layout->pool_ns);
1109 	ceph_oid_printf(&req->r_base_oid, "%llx.%08llx", vino.ino, objnum);
1110 	req->r_flags = flags | osdc->client->options->read_from_replica;
1111 
1112 	req->r_snapid = vino.snap;
1113 	if (flags & CEPH_OSD_FLAG_WRITE)
1114 		req->r_data_offset = off;
1115 
1116 	if (num_ops > 1) {
1117 		int num_req_ops, num_rep_ops;
1118 
1119 		/*
1120 		 * If this is a multi-op write request, assume that we'll need
1121 		 * request ops. If it's a multi-op read then assume we'll need
1122 		 * reply ops. Anything else and call it -EINVAL.
1123 		 */
1124 		if (flags & CEPH_OSD_FLAG_WRITE) {
1125 			num_req_ops = num_ops;
1126 			num_rep_ops = 0;
1127 		} else if (flags & CEPH_OSD_FLAG_READ) {
1128 			num_req_ops = 0;
1129 			num_rep_ops = num_ops;
1130 		} else {
1131 			r = -EINVAL;
1132 			goto fail;
1133 		}
1134 
1135 		r = __ceph_osdc_alloc_messages(req, GFP_NOFS, num_req_ops,
1136 					       num_rep_ops);
1137 	} else {
1138 		r = ceph_osdc_alloc_messages(req, GFP_NOFS);
1139 	}
1140 	if (r)
1141 		goto fail;
1142 
1143 	return req;
1144 
1145 fail:
1146 	ceph_osdc_put_request(req);
1147 	return ERR_PTR(r);
1148 }
1149 EXPORT_SYMBOL(ceph_osdc_new_request);
1150 
1151 int __ceph_alloc_sparse_ext_map(struct ceph_osd_req_op *op, int cnt)
1152 {
1153 	WARN_ON(op->op != CEPH_OSD_OP_SPARSE_READ);
1154 
1155 	op->extent.sparse_ext_cnt = cnt;
1156 	op->extent.sparse_ext = kmalloc_array(cnt,
1157 					      sizeof(*op->extent.sparse_ext),
1158 					      GFP_NOFS);
1159 	if (!op->extent.sparse_ext)
1160 		return -ENOMEM;
1161 	return 0;
1162 }
1163 EXPORT_SYMBOL(__ceph_alloc_sparse_ext_map);
1164 
1165 /*
1166  * We keep osd requests in an rbtree, sorted by ->r_tid.
1167  */
1168 DEFINE_RB_FUNCS(request, struct ceph_osd_request, r_tid, r_node)
1169 DEFINE_RB_FUNCS(request_mc, struct ceph_osd_request, r_tid, r_mc_node)
1170 
1171 /*
1172  * Call @fn on each OSD request as long as @fn returns 0.
1173  */
1174 static void for_each_request(struct ceph_osd_client *osdc,
1175 			int (*fn)(struct ceph_osd_request *req, void *arg),
1176 			void *arg)
1177 {
1178 	struct rb_node *n, *p;
1179 
1180 	for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
1181 		struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
1182 
1183 		for (p = rb_first(&osd->o_requests); p; ) {
1184 			struct ceph_osd_request *req =
1185 			    rb_entry(p, struct ceph_osd_request, r_node);
1186 
1187 			p = rb_next(p);
1188 			if (fn(req, arg))
1189 				return;
1190 		}
1191 	}
1192 
1193 	for (p = rb_first(&osdc->homeless_osd.o_requests); p; ) {
1194 		struct ceph_osd_request *req =
1195 		    rb_entry(p, struct ceph_osd_request, r_node);
1196 
1197 		p = rb_next(p);
1198 		if (fn(req, arg))
1199 			return;
1200 	}
1201 }
1202 
1203 static bool osd_homeless(struct ceph_osd *osd)
1204 {
1205 	return osd->o_osd == CEPH_HOMELESS_OSD;
1206 }
1207 
1208 static bool osd_registered(struct ceph_osd *osd)
1209 {
1210 	verify_osdc_locked(osd->o_osdc);
1211 
1212 	return !RB_EMPTY_NODE(&osd->o_node);
1213 }
1214 
1215 /*
1216  * Assumes @osd is zero-initialized.
1217  */
1218 static void osd_init(struct ceph_osd *osd)
1219 {
1220 	refcount_set(&osd->o_ref, 1);
1221 	RB_CLEAR_NODE(&osd->o_node);
1222 	spin_lock_init(&osd->o_requests_lock);
1223 	osd->o_requests = RB_ROOT;
1224 	osd->o_linger_requests = RB_ROOT;
1225 	osd->o_backoff_mappings = RB_ROOT;
1226 	osd->o_backoffs_by_id = RB_ROOT;
1227 	INIT_LIST_HEAD(&osd->o_osd_lru);
1228 	INIT_LIST_HEAD(&osd->o_keepalive_item);
1229 	osd->o_incarnation = 1;
1230 	mutex_init(&osd->lock);
1231 }
1232 
1233 static void ceph_init_sparse_read(struct ceph_sparse_read *sr)
1234 {
1235 	kfree(sr->sr_extent);
1236 	memset(sr, '\0', sizeof(*sr));
1237 	sr->sr_state = CEPH_SPARSE_READ_HDR;
1238 }
1239 
1240 static void osd_cleanup(struct ceph_osd *osd)
1241 {
1242 	WARN_ON(!RB_EMPTY_NODE(&osd->o_node));
1243 	WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests));
1244 	WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests));
1245 	WARN_ON(!RB_EMPTY_ROOT(&osd->o_backoff_mappings));
1246 	WARN_ON(!RB_EMPTY_ROOT(&osd->o_backoffs_by_id));
1247 	WARN_ON(!list_empty(&osd->o_osd_lru));
1248 	WARN_ON(!list_empty(&osd->o_keepalive_item));
1249 
1250 	ceph_init_sparse_read(&osd->o_sparse_read);
1251 
1252 	if (osd->o_auth.authorizer) {
1253 		WARN_ON(osd_homeless(osd));
1254 		ceph_auth_destroy_authorizer(osd->o_auth.authorizer);
1255 	}
1256 }
1257 
1258 /*
1259  * Track open sessions with osds.
1260  */
1261 static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
1262 {
1263 	struct ceph_osd *osd;
1264 
1265 	WARN_ON(onum == CEPH_HOMELESS_OSD);
1266 
1267 	osd = kzalloc(sizeof(*osd), GFP_NOIO | __GFP_NOFAIL);
1268 	osd_init(osd);
1269 	osd->o_osdc = osdc;
1270 	osd->o_osd = onum;
1271 	osd->o_sparse_op_idx = -1;
1272 
1273 	ceph_init_sparse_read(&osd->o_sparse_read);
1274 
1275 	ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
1276 
1277 	return osd;
1278 }
1279 
1280 static struct ceph_osd *get_osd(struct ceph_osd *osd)
1281 {
1282 	if (refcount_inc_not_zero(&osd->o_ref)) {
1283 		dout("get_osd %p -> %d\n", osd, refcount_read(&osd->o_ref));
1284 		return osd;
1285 	} else {
1286 		dout("get_osd %p FAIL\n", osd);
1287 		return NULL;
1288 	}
1289 }
1290 
1291 static void put_osd(struct ceph_osd *osd)
1292 {
1293 	dout("put_osd %p -> %d\n", osd, refcount_read(&osd->o_ref) - 1);
1294 	if (refcount_dec_and_test(&osd->o_ref)) {
1295 		osd_cleanup(osd);
1296 		kfree(osd);
1297 	}
1298 }
1299 
1300 DEFINE_RB_FUNCS(osd, struct ceph_osd, o_osd, o_node)
1301 
1302 static void __move_osd_to_lru(struct ceph_osd *osd)
1303 {
1304 	struct ceph_osd_client *osdc = osd->o_osdc;
1305 
1306 	dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1307 	BUG_ON(!list_empty(&osd->o_osd_lru));
1308 
1309 	spin_lock(&osdc->osd_lru_lock);
1310 	list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
1311 	spin_unlock(&osdc->osd_lru_lock);
1312 
1313 	osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl;
1314 }
1315 
1316 static void maybe_move_osd_to_lru(struct ceph_osd *osd)
1317 {
1318 	if (RB_EMPTY_ROOT(&osd->o_requests) &&
1319 	    RB_EMPTY_ROOT(&osd->o_linger_requests))
1320 		__move_osd_to_lru(osd);
1321 }
1322 
1323 static void __remove_osd_from_lru(struct ceph_osd *osd)
1324 {
1325 	struct ceph_osd_client *osdc = osd->o_osdc;
1326 
1327 	dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1328 
1329 	spin_lock(&osdc->osd_lru_lock);
1330 	if (!list_empty(&osd->o_osd_lru))
1331 		list_del_init(&osd->o_osd_lru);
1332 	spin_unlock(&osdc->osd_lru_lock);
1333 }
1334 
1335 /*
1336  * Close the connection and assign any leftover requests to the
1337  * homeless session.
1338  */
1339 static void close_osd(struct ceph_osd *osd)
1340 {
1341 	struct ceph_osd_client *osdc = osd->o_osdc;
1342 	struct rb_node *n;
1343 
1344 	verify_osdc_wrlocked(osdc);
1345 	dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1346 
1347 	ceph_con_close(&osd->o_con);
1348 
1349 	for (n = rb_first(&osd->o_requests); n; ) {
1350 		struct ceph_osd_request *req =
1351 		    rb_entry(n, struct ceph_osd_request, r_node);
1352 
1353 		n = rb_next(n); /* unlink_request() */
1354 
1355 		dout(" reassigning req %p tid %llu\n", req, req->r_tid);
1356 		unlink_request(osd, req);
1357 		link_request(&osdc->homeless_osd, req);
1358 	}
1359 	for (n = rb_first(&osd->o_linger_requests); n; ) {
1360 		struct ceph_osd_linger_request *lreq =
1361 		    rb_entry(n, struct ceph_osd_linger_request, node);
1362 
1363 		n = rb_next(n); /* unlink_linger() */
1364 
1365 		dout(" reassigning lreq %p linger_id %llu\n", lreq,
1366 		     lreq->linger_id);
1367 		unlink_linger(osd, lreq);
1368 		link_linger(&osdc->homeless_osd, lreq);
1369 	}
1370 	clear_backoffs(osd);
1371 
1372 	__remove_osd_from_lru(osd);
1373 	erase_osd(&osdc->osds, osd);
1374 	put_osd(osd);
1375 }
1376 
1377 /*
1378  * reset osd connect
1379  */
1380 static int reopen_osd(struct ceph_osd *osd)
1381 {
1382 	struct ceph_entity_addr *peer_addr;
1383 
1384 	dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1385 
1386 	if (RB_EMPTY_ROOT(&osd->o_requests) &&
1387 	    RB_EMPTY_ROOT(&osd->o_linger_requests)) {
1388 		close_osd(osd);
1389 		return -ENODEV;
1390 	}
1391 
1392 	peer_addr = &osd->o_osdc->osdmap->osd_addr[osd->o_osd];
1393 	if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1394 			!ceph_con_opened(&osd->o_con)) {
1395 		struct rb_node *n;
1396 
1397 		dout("osd addr hasn't changed and connection never opened, "
1398 		     "letting msgr retry\n");
1399 		/* touch each r_stamp for handle_timeout()'s benfit */
1400 		for (n = rb_first(&osd->o_requests); n; n = rb_next(n)) {
1401 			struct ceph_osd_request *req =
1402 			    rb_entry(n, struct ceph_osd_request, r_node);
1403 			req->r_stamp = jiffies;
1404 		}
1405 
1406 		return -EAGAIN;
1407 	}
1408 
1409 	ceph_con_close(&osd->o_con);
1410 	ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1411 	osd->o_incarnation++;
1412 
1413 	return 0;
1414 }
1415 
1416 static struct ceph_osd *lookup_create_osd(struct ceph_osd_client *osdc, int o,
1417 					  bool wrlocked)
1418 {
1419 	struct ceph_osd *osd;
1420 
1421 	if (wrlocked)
1422 		verify_osdc_wrlocked(osdc);
1423 	else
1424 		verify_osdc_locked(osdc);
1425 
1426 	if (o != CEPH_HOMELESS_OSD)
1427 		osd = lookup_osd(&osdc->osds, o);
1428 	else
1429 		osd = &osdc->homeless_osd;
1430 	if (!osd) {
1431 		if (!wrlocked)
1432 			return ERR_PTR(-EAGAIN);
1433 
1434 		osd = create_osd(osdc, o);
1435 		insert_osd(&osdc->osds, osd);
1436 		ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd,
1437 			      &osdc->osdmap->osd_addr[osd->o_osd]);
1438 	}
1439 
1440 	dout("%s osdc %p osd%d -> osd %p\n", __func__, osdc, o, osd);
1441 	return osd;
1442 }
1443 
1444 /*
1445  * Create request <-> OSD session relation.
1446  *
1447  * @req has to be assigned a tid, @osd may be homeless.
1448  */
1449 static void link_request(struct ceph_osd *osd, struct ceph_osd_request *req)
1450 {
1451 	verify_osd_locked(osd);
1452 	WARN_ON(!req->r_tid || req->r_osd);
1453 	dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd,
1454 	     req, req->r_tid);
1455 
1456 	if (!osd_homeless(osd))
1457 		__remove_osd_from_lru(osd);
1458 	else
1459 		atomic_inc(&osd->o_osdc->num_homeless);
1460 
1461 	get_osd(osd);
1462 	spin_lock(&osd->o_requests_lock);
1463 	insert_request(&osd->o_requests, req);
1464 	spin_unlock(&osd->o_requests_lock);
1465 	req->r_osd = osd;
1466 }
1467 
1468 static void unlink_request(struct ceph_osd *osd, struct ceph_osd_request *req)
1469 {
1470 	verify_osd_locked(osd);
1471 	WARN_ON(req->r_osd != osd);
1472 	dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd,
1473 	     req, req->r_tid);
1474 
1475 	req->r_osd = NULL;
1476 	spin_lock(&osd->o_requests_lock);
1477 	erase_request(&osd->o_requests, req);
1478 	spin_unlock(&osd->o_requests_lock);
1479 	put_osd(osd);
1480 
1481 	if (!osd_homeless(osd))
1482 		maybe_move_osd_to_lru(osd);
1483 	else
1484 		atomic_dec(&osd->o_osdc->num_homeless);
1485 }
1486 
1487 static bool __pool_full(struct ceph_pg_pool_info *pi)
1488 {
1489 	return pi->flags & CEPH_POOL_FLAG_FULL;
1490 }
1491 
1492 static bool have_pool_full(struct ceph_osd_client *osdc)
1493 {
1494 	struct rb_node *n;
1495 
1496 	for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) {
1497 		struct ceph_pg_pool_info *pi =
1498 		    rb_entry(n, struct ceph_pg_pool_info, node);
1499 
1500 		if (__pool_full(pi))
1501 			return true;
1502 	}
1503 
1504 	return false;
1505 }
1506 
1507 static bool pool_full(struct ceph_osd_client *osdc, s64 pool_id)
1508 {
1509 	struct ceph_pg_pool_info *pi;
1510 
1511 	pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id);
1512 	if (!pi)
1513 		return false;
1514 
1515 	return __pool_full(pi);
1516 }
1517 
1518 /*
1519  * Returns whether a request should be blocked from being sent
1520  * based on the current osdmap and osd_client settings.
1521  */
1522 static bool target_should_be_paused(struct ceph_osd_client *osdc,
1523 				    const struct ceph_osd_request_target *t,
1524 				    struct ceph_pg_pool_info *pi)
1525 {
1526 	bool pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
1527 	bool pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
1528 		       ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
1529 		       __pool_full(pi);
1530 
1531 	WARN_ON(pi->id != t->target_oloc.pool);
1532 	return ((t->flags & CEPH_OSD_FLAG_READ) && pauserd) ||
1533 	       ((t->flags & CEPH_OSD_FLAG_WRITE) && pausewr) ||
1534 	       (osdc->osdmap->epoch < osdc->epoch_barrier);
1535 }
1536 
1537 static int pick_random_replica(const struct ceph_osds *acting)
1538 {
1539 	int i = get_random_u32_below(acting->size);
1540 
1541 	dout("%s picked osd%d, primary osd%d\n", __func__,
1542 	     acting->osds[i], acting->primary);
1543 	return i;
1544 }
1545 
1546 /*
1547  * Picks the closest replica based on client's location given by
1548  * crush_location option.  Prefers the primary if the locality is
1549  * the same.
1550  */
1551 static int pick_closest_replica(struct ceph_osd_client *osdc,
1552 				const struct ceph_osds *acting)
1553 {
1554 	struct ceph_options *opt = osdc->client->options;
1555 	int best_i, best_locality;
1556 	int i = 0, locality;
1557 
1558 	do {
1559 		locality = ceph_get_crush_locality(osdc->osdmap,
1560 						   acting->osds[i],
1561 						   &opt->crush_locs);
1562 		if (i == 0 ||
1563 		    (locality >= 0 && best_locality < 0) ||
1564 		    (locality >= 0 && best_locality >= 0 &&
1565 		     locality < best_locality)) {
1566 			best_i = i;
1567 			best_locality = locality;
1568 		}
1569 	} while (++i < acting->size);
1570 
1571 	dout("%s picked osd%d with locality %d, primary osd%d\n", __func__,
1572 	     acting->osds[best_i], best_locality, acting->primary);
1573 	return best_i;
1574 }
1575 
1576 enum calc_target_result {
1577 	CALC_TARGET_NO_ACTION = 0,
1578 	CALC_TARGET_NEED_RESEND,
1579 	CALC_TARGET_POOL_DNE,
1580 };
1581 
1582 static enum calc_target_result calc_target(struct ceph_osd_client *osdc,
1583 					   struct ceph_osd_request_target *t,
1584 					   bool any_change)
1585 {
1586 	struct ceph_pg_pool_info *pi;
1587 	struct ceph_pg pgid, last_pgid;
1588 	struct ceph_osds up, acting;
1589 	bool should_be_paused;
1590 	bool is_read = t->flags & CEPH_OSD_FLAG_READ;
1591 	bool is_write = t->flags & CEPH_OSD_FLAG_WRITE;
1592 	bool force_resend = false;
1593 	bool unpaused = false;
1594 	bool legacy_change = false;
1595 	bool split = false;
1596 	bool sort_bitwise = ceph_osdmap_flag(osdc, CEPH_OSDMAP_SORTBITWISE);
1597 	bool recovery_deletes = ceph_osdmap_flag(osdc,
1598 						 CEPH_OSDMAP_RECOVERY_DELETES);
1599 	enum calc_target_result ct_res;
1600 
1601 	t->epoch = osdc->osdmap->epoch;
1602 	pi = ceph_pg_pool_by_id(osdc->osdmap, t->base_oloc.pool);
1603 	if (!pi) {
1604 		t->osd = CEPH_HOMELESS_OSD;
1605 		ct_res = CALC_TARGET_POOL_DNE;
1606 		goto out;
1607 	}
1608 
1609 	if (osdc->osdmap->epoch == pi->last_force_request_resend) {
1610 		if (t->last_force_resend < pi->last_force_request_resend) {
1611 			t->last_force_resend = pi->last_force_request_resend;
1612 			force_resend = true;
1613 		} else if (t->last_force_resend == 0) {
1614 			force_resend = true;
1615 		}
1616 	}
1617 
1618 	/* apply tiering */
1619 	ceph_oid_copy(&t->target_oid, &t->base_oid);
1620 	ceph_oloc_copy(&t->target_oloc, &t->base_oloc);
1621 	if ((t->flags & CEPH_OSD_FLAG_IGNORE_OVERLAY) == 0) {
1622 		if (is_read && pi->read_tier >= 0)
1623 			t->target_oloc.pool = pi->read_tier;
1624 		if (is_write && pi->write_tier >= 0)
1625 			t->target_oloc.pool = pi->write_tier;
1626 
1627 		pi = ceph_pg_pool_by_id(osdc->osdmap, t->target_oloc.pool);
1628 		if (!pi) {
1629 			t->osd = CEPH_HOMELESS_OSD;
1630 			ct_res = CALC_TARGET_POOL_DNE;
1631 			goto out;
1632 		}
1633 	}
1634 
1635 	__ceph_object_locator_to_pg(pi, &t->target_oid, &t->target_oloc, &pgid);
1636 	last_pgid.pool = pgid.pool;
1637 	last_pgid.seed = ceph_stable_mod(pgid.seed, t->pg_num, t->pg_num_mask);
1638 
1639 	ceph_pg_to_up_acting_osds(osdc->osdmap, pi, &pgid, &up, &acting);
1640 	if (any_change &&
1641 	    ceph_is_new_interval(&t->acting,
1642 				 &acting,
1643 				 &t->up,
1644 				 &up,
1645 				 t->size,
1646 				 pi->size,
1647 				 t->min_size,
1648 				 pi->min_size,
1649 				 t->pg_num,
1650 				 pi->pg_num,
1651 				 t->sort_bitwise,
1652 				 sort_bitwise,
1653 				 t->recovery_deletes,
1654 				 recovery_deletes,
1655 				 &last_pgid))
1656 		force_resend = true;
1657 
1658 	should_be_paused = target_should_be_paused(osdc, t, pi);
1659 	if (t->paused && !should_be_paused) {
1660 		unpaused = true;
1661 	}
1662 	if (t->paused != should_be_paused) {
1663 		dout("%s t %p paused %d -> %d\n", __func__, t, t->paused,
1664 		     should_be_paused);
1665 		t->paused = should_be_paused;
1666 	}
1667 
1668 	legacy_change = ceph_pg_compare(&t->pgid, &pgid) ||
1669 			ceph_osds_changed(&t->acting, &acting,
1670 					  t->used_replica || any_change);
1671 	if (t->pg_num)
1672 		split = ceph_pg_is_split(&last_pgid, t->pg_num, pi->pg_num);
1673 
1674 	if (legacy_change || force_resend || split) {
1675 		t->pgid = pgid; /* struct */
1676 		ceph_pg_to_primary_shard(osdc->osdmap, pi, &pgid, &t->spgid);
1677 		ceph_osds_copy(&t->acting, &acting);
1678 		ceph_osds_copy(&t->up, &up);
1679 		t->size = pi->size;
1680 		t->min_size = pi->min_size;
1681 		t->pg_num = pi->pg_num;
1682 		t->pg_num_mask = pi->pg_num_mask;
1683 		t->sort_bitwise = sort_bitwise;
1684 		t->recovery_deletes = recovery_deletes;
1685 
1686 		if ((t->flags & (CEPH_OSD_FLAG_BALANCE_READS |
1687 				 CEPH_OSD_FLAG_LOCALIZE_READS)) &&
1688 		    !is_write && pi->type == CEPH_POOL_TYPE_REP &&
1689 		    acting.size > 1) {
1690 			int pos;
1691 
1692 			WARN_ON(!is_read || acting.osds[0] != acting.primary);
1693 			if (t->flags & CEPH_OSD_FLAG_BALANCE_READS) {
1694 				pos = pick_random_replica(&acting);
1695 			} else {
1696 				pos = pick_closest_replica(osdc, &acting);
1697 			}
1698 			t->osd = acting.osds[pos];
1699 			t->used_replica = pos > 0;
1700 		} else {
1701 			t->osd = acting.primary;
1702 			t->used_replica = false;
1703 		}
1704 	}
1705 
1706 	if (unpaused || legacy_change || force_resend || split)
1707 		ct_res = CALC_TARGET_NEED_RESEND;
1708 	else
1709 		ct_res = CALC_TARGET_NO_ACTION;
1710 
1711 out:
1712 	dout("%s t %p -> %d%d%d%d ct_res %d osd%d\n", __func__, t, unpaused,
1713 	     legacy_change, force_resend, split, ct_res, t->osd);
1714 	return ct_res;
1715 }
1716 
1717 static struct ceph_spg_mapping *alloc_spg_mapping(void)
1718 {
1719 	struct ceph_spg_mapping *spg;
1720 
1721 	spg = kmalloc(sizeof(*spg), GFP_NOIO);
1722 	if (!spg)
1723 		return NULL;
1724 
1725 	RB_CLEAR_NODE(&spg->node);
1726 	spg->backoffs = RB_ROOT;
1727 	return spg;
1728 }
1729 
1730 static void free_spg_mapping(struct ceph_spg_mapping *spg)
1731 {
1732 	WARN_ON(!RB_EMPTY_NODE(&spg->node));
1733 	WARN_ON(!RB_EMPTY_ROOT(&spg->backoffs));
1734 
1735 	kfree(spg);
1736 }
1737 
1738 /*
1739  * rbtree of ceph_spg_mapping for handling map<spg_t, ...>, similar to
1740  * ceph_pg_mapping.  Used to track OSD backoffs -- a backoff [range] is
1741  * defined only within a specific spgid; it does not pass anything to
1742  * children on split, or to another primary.
1743  */
1744 DEFINE_RB_FUNCS2(spg_mapping, struct ceph_spg_mapping, spgid, ceph_spg_compare,
1745 		 RB_BYPTR, const struct ceph_spg *, node)
1746 
1747 static u64 hoid_get_bitwise_key(const struct ceph_hobject_id *hoid)
1748 {
1749 	return hoid->is_max ? 0x100000000ull : hoid->hash_reverse_bits;
1750 }
1751 
1752 static void hoid_get_effective_key(const struct ceph_hobject_id *hoid,
1753 				   void **pkey, size_t *pkey_len)
1754 {
1755 	if (hoid->key_len) {
1756 		*pkey = hoid->key;
1757 		*pkey_len = hoid->key_len;
1758 	} else {
1759 		*pkey = hoid->oid;
1760 		*pkey_len = hoid->oid_len;
1761 	}
1762 }
1763 
1764 static int compare_names(const void *name1, size_t name1_len,
1765 			 const void *name2, size_t name2_len)
1766 {
1767 	int ret;
1768 
1769 	ret = memcmp(name1, name2, min(name1_len, name2_len));
1770 	if (!ret) {
1771 		if (name1_len < name2_len)
1772 			ret = -1;
1773 		else if (name1_len > name2_len)
1774 			ret = 1;
1775 	}
1776 	return ret;
1777 }
1778 
1779 static int hoid_compare(const struct ceph_hobject_id *lhs,
1780 			const struct ceph_hobject_id *rhs)
1781 {
1782 	void *effective_key1, *effective_key2;
1783 	size_t effective_key1_len, effective_key2_len;
1784 	int ret;
1785 
1786 	if (lhs->is_max < rhs->is_max)
1787 		return -1;
1788 	if (lhs->is_max > rhs->is_max)
1789 		return 1;
1790 
1791 	if (lhs->pool < rhs->pool)
1792 		return -1;
1793 	if (lhs->pool > rhs->pool)
1794 		return 1;
1795 
1796 	if (hoid_get_bitwise_key(lhs) < hoid_get_bitwise_key(rhs))
1797 		return -1;
1798 	if (hoid_get_bitwise_key(lhs) > hoid_get_bitwise_key(rhs))
1799 		return 1;
1800 
1801 	ret = compare_names(lhs->nspace, lhs->nspace_len,
1802 			    rhs->nspace, rhs->nspace_len);
1803 	if (ret)
1804 		return ret;
1805 
1806 	hoid_get_effective_key(lhs, &effective_key1, &effective_key1_len);
1807 	hoid_get_effective_key(rhs, &effective_key2, &effective_key2_len);
1808 	ret = compare_names(effective_key1, effective_key1_len,
1809 			    effective_key2, effective_key2_len);
1810 	if (ret)
1811 		return ret;
1812 
1813 	ret = compare_names(lhs->oid, lhs->oid_len, rhs->oid, rhs->oid_len);
1814 	if (ret)
1815 		return ret;
1816 
1817 	if (lhs->snapid < rhs->snapid)
1818 		return -1;
1819 	if (lhs->snapid > rhs->snapid)
1820 		return 1;
1821 
1822 	return 0;
1823 }
1824 
1825 /*
1826  * For decoding ->begin and ->end of MOSDBackoff only -- no MIN/MAX
1827  * compat stuff here.
1828  *
1829  * Assumes @hoid is zero-initialized.
1830  */
1831 static int decode_hoid(void **p, void *end, struct ceph_hobject_id *hoid)
1832 {
1833 	u8 struct_v;
1834 	u32 struct_len;
1835 	int ret;
1836 
1837 	ret = ceph_start_decoding(p, end, 4, "hobject_t", &struct_v,
1838 				  &struct_len);
1839 	if (ret)
1840 		return ret;
1841 
1842 	if (struct_v < 4) {
1843 		pr_err("got struct_v %d < 4 of hobject_t\n", struct_v);
1844 		goto e_inval;
1845 	}
1846 
1847 	hoid->key = ceph_extract_encoded_string(p, end, &hoid->key_len,
1848 						GFP_NOIO);
1849 	if (IS_ERR(hoid->key)) {
1850 		ret = PTR_ERR(hoid->key);
1851 		hoid->key = NULL;
1852 		return ret;
1853 	}
1854 
1855 	hoid->oid = ceph_extract_encoded_string(p, end, &hoid->oid_len,
1856 						GFP_NOIO);
1857 	if (IS_ERR(hoid->oid)) {
1858 		ret = PTR_ERR(hoid->oid);
1859 		hoid->oid = NULL;
1860 		return ret;
1861 	}
1862 
1863 	ceph_decode_64_safe(p, end, hoid->snapid, e_inval);
1864 	ceph_decode_32_safe(p, end, hoid->hash, e_inval);
1865 	ceph_decode_8_safe(p, end, hoid->is_max, e_inval);
1866 
1867 	hoid->nspace = ceph_extract_encoded_string(p, end, &hoid->nspace_len,
1868 						   GFP_NOIO);
1869 	if (IS_ERR(hoid->nspace)) {
1870 		ret = PTR_ERR(hoid->nspace);
1871 		hoid->nspace = NULL;
1872 		return ret;
1873 	}
1874 
1875 	ceph_decode_64_safe(p, end, hoid->pool, e_inval);
1876 
1877 	ceph_hoid_build_hash_cache(hoid);
1878 	return 0;
1879 
1880 e_inval:
1881 	return -EINVAL;
1882 }
1883 
1884 static int hoid_encoding_size(const struct ceph_hobject_id *hoid)
1885 {
1886 	return 8 + 4 + 1 + 8 + /* snapid, hash, is_max, pool */
1887 	       4 + hoid->key_len + 4 + hoid->oid_len + 4 + hoid->nspace_len;
1888 }
1889 
1890 static void encode_hoid(void **p, void *end, const struct ceph_hobject_id *hoid)
1891 {
1892 	ceph_start_encoding(p, 4, 3, hoid_encoding_size(hoid));
1893 	ceph_encode_string(p, end, hoid->key, hoid->key_len);
1894 	ceph_encode_string(p, end, hoid->oid, hoid->oid_len);
1895 	ceph_encode_64(p, hoid->snapid);
1896 	ceph_encode_32(p, hoid->hash);
1897 	ceph_encode_8(p, hoid->is_max);
1898 	ceph_encode_string(p, end, hoid->nspace, hoid->nspace_len);
1899 	ceph_encode_64(p, hoid->pool);
1900 }
1901 
1902 static void free_hoid(struct ceph_hobject_id *hoid)
1903 {
1904 	if (hoid) {
1905 		kfree(hoid->key);
1906 		kfree(hoid->oid);
1907 		kfree(hoid->nspace);
1908 		kfree(hoid);
1909 	}
1910 }
1911 
1912 static struct ceph_osd_backoff *alloc_backoff(void)
1913 {
1914 	struct ceph_osd_backoff *backoff;
1915 
1916 	backoff = kzalloc(sizeof(*backoff), GFP_NOIO);
1917 	if (!backoff)
1918 		return NULL;
1919 
1920 	RB_CLEAR_NODE(&backoff->spg_node);
1921 	RB_CLEAR_NODE(&backoff->id_node);
1922 	return backoff;
1923 }
1924 
1925 static void free_backoff(struct ceph_osd_backoff *backoff)
1926 {
1927 	WARN_ON(!RB_EMPTY_NODE(&backoff->spg_node));
1928 	WARN_ON(!RB_EMPTY_NODE(&backoff->id_node));
1929 
1930 	free_hoid(backoff->begin);
1931 	free_hoid(backoff->end);
1932 	kfree(backoff);
1933 }
1934 
1935 /*
1936  * Within a specific spgid, backoffs are managed by ->begin hoid.
1937  */
1938 DEFINE_RB_INSDEL_FUNCS2(backoff, struct ceph_osd_backoff, begin, hoid_compare,
1939 			RB_BYVAL, spg_node);
1940 
1941 static struct ceph_osd_backoff *lookup_containing_backoff(struct rb_root *root,
1942 					    const struct ceph_hobject_id *hoid)
1943 {
1944 	struct rb_node *n = root->rb_node;
1945 
1946 	while (n) {
1947 		struct ceph_osd_backoff *cur =
1948 		    rb_entry(n, struct ceph_osd_backoff, spg_node);
1949 		int cmp;
1950 
1951 		cmp = hoid_compare(hoid, cur->begin);
1952 		if (cmp < 0) {
1953 			n = n->rb_left;
1954 		} else if (cmp > 0) {
1955 			if (hoid_compare(hoid, cur->end) < 0)
1956 				return cur;
1957 
1958 			n = n->rb_right;
1959 		} else {
1960 			return cur;
1961 		}
1962 	}
1963 
1964 	return NULL;
1965 }
1966 
1967 /*
1968  * Each backoff has a unique id within its OSD session.
1969  */
1970 DEFINE_RB_FUNCS(backoff_by_id, struct ceph_osd_backoff, id, id_node)
1971 
1972 static void clear_backoffs(struct ceph_osd *osd)
1973 {
1974 	while (!RB_EMPTY_ROOT(&osd->o_backoff_mappings)) {
1975 		struct ceph_spg_mapping *spg =
1976 		    rb_entry(rb_first(&osd->o_backoff_mappings),
1977 			     struct ceph_spg_mapping, node);
1978 
1979 		while (!RB_EMPTY_ROOT(&spg->backoffs)) {
1980 			struct ceph_osd_backoff *backoff =
1981 			    rb_entry(rb_first(&spg->backoffs),
1982 				     struct ceph_osd_backoff, spg_node);
1983 
1984 			erase_backoff(&spg->backoffs, backoff);
1985 			erase_backoff_by_id(&osd->o_backoffs_by_id, backoff);
1986 			free_backoff(backoff);
1987 		}
1988 		erase_spg_mapping(&osd->o_backoff_mappings, spg);
1989 		free_spg_mapping(spg);
1990 	}
1991 }
1992 
1993 /*
1994  * Set up a temporary, non-owning view into @t.
1995  */
1996 static void hoid_fill_from_target(struct ceph_hobject_id *hoid,
1997 				  const struct ceph_osd_request_target *t)
1998 {
1999 	hoid->key = NULL;
2000 	hoid->key_len = 0;
2001 	hoid->oid = t->target_oid.name;
2002 	hoid->oid_len = t->target_oid.name_len;
2003 	hoid->snapid = CEPH_NOSNAP;
2004 	hoid->hash = t->pgid.seed;
2005 	hoid->is_max = false;
2006 	if (t->target_oloc.pool_ns) {
2007 		hoid->nspace = t->target_oloc.pool_ns->str;
2008 		hoid->nspace_len = t->target_oloc.pool_ns->len;
2009 	} else {
2010 		hoid->nspace = NULL;
2011 		hoid->nspace_len = 0;
2012 	}
2013 	hoid->pool = t->target_oloc.pool;
2014 	ceph_hoid_build_hash_cache(hoid);
2015 }
2016 
2017 static bool should_plug_request(struct ceph_osd_request *req)
2018 {
2019 	struct ceph_osd *osd = req->r_osd;
2020 	struct ceph_spg_mapping *spg;
2021 	struct ceph_osd_backoff *backoff;
2022 	struct ceph_hobject_id hoid;
2023 
2024 	spg = lookup_spg_mapping(&osd->o_backoff_mappings, &req->r_t.spgid);
2025 	if (!spg)
2026 		return false;
2027 
2028 	hoid_fill_from_target(&hoid, &req->r_t);
2029 	backoff = lookup_containing_backoff(&spg->backoffs, &hoid);
2030 	if (!backoff)
2031 		return false;
2032 
2033 	dout("%s req %p tid %llu backoff osd%d spgid %llu.%xs%d id %llu\n",
2034 	     __func__, req, req->r_tid, osd->o_osd, backoff->spgid.pgid.pool,
2035 	     backoff->spgid.pgid.seed, backoff->spgid.shard, backoff->id);
2036 	return true;
2037 }
2038 
2039 /*
2040  * Keep get_num_data_items() in sync with this function.
2041  */
2042 static void setup_request_data(struct ceph_osd_request *req)
2043 {
2044 	struct ceph_msg *request_msg = req->r_request;
2045 	struct ceph_msg *reply_msg = req->r_reply;
2046 	struct ceph_osd_req_op *op;
2047 
2048 	if (req->r_request->num_data_items || req->r_reply->num_data_items)
2049 		return;
2050 
2051 	WARN_ON(request_msg->data_length || reply_msg->data_length);
2052 	for (op = req->r_ops; op != &req->r_ops[req->r_num_ops]; op++) {
2053 		switch (op->op) {
2054 		/* request */
2055 		case CEPH_OSD_OP_WRITE:
2056 		case CEPH_OSD_OP_WRITEFULL:
2057 			WARN_ON(op->indata_len != op->extent.length);
2058 			ceph_osdc_msg_data_add(request_msg,
2059 					       &op->extent.osd_data);
2060 			break;
2061 		case CEPH_OSD_OP_SETXATTR:
2062 		case CEPH_OSD_OP_CMPXATTR:
2063 			WARN_ON(op->indata_len != op->xattr.name_len +
2064 						  op->xattr.value_len);
2065 			ceph_osdc_msg_data_add(request_msg,
2066 					       &op->xattr.osd_data);
2067 			break;
2068 		case CEPH_OSD_OP_NOTIFY_ACK:
2069 			ceph_osdc_msg_data_add(request_msg,
2070 					       &op->notify_ack.request_data);
2071 			break;
2072 		case CEPH_OSD_OP_COPY_FROM2:
2073 			ceph_osdc_msg_data_add(request_msg,
2074 					       &op->copy_from.osd_data);
2075 			break;
2076 
2077 		/* reply */
2078 		case CEPH_OSD_OP_STAT:
2079 			ceph_osdc_msg_data_add(reply_msg,
2080 					       &op->raw_data_in);
2081 			break;
2082 		case CEPH_OSD_OP_READ:
2083 		case CEPH_OSD_OP_SPARSE_READ:
2084 			ceph_osdc_msg_data_add(reply_msg,
2085 					       &op->extent.osd_data);
2086 			break;
2087 		case CEPH_OSD_OP_LIST_WATCHERS:
2088 			ceph_osdc_msg_data_add(reply_msg,
2089 					       &op->list_watchers.response_data);
2090 			break;
2091 
2092 		/* both */
2093 		case CEPH_OSD_OP_CALL:
2094 			WARN_ON(op->indata_len != op->cls.class_len +
2095 						  op->cls.method_len +
2096 						  op->cls.indata_len);
2097 			ceph_osdc_msg_data_add(request_msg,
2098 					       &op->cls.request_info);
2099 			/* optional, can be NONE */
2100 			ceph_osdc_msg_data_add(request_msg,
2101 					       &op->cls.request_data);
2102 			/* optional, can be NONE */
2103 			ceph_osdc_msg_data_add(reply_msg,
2104 					       &op->cls.response_data);
2105 			break;
2106 		case CEPH_OSD_OP_NOTIFY:
2107 			ceph_osdc_msg_data_add(request_msg,
2108 					       &op->notify.request_data);
2109 			ceph_osdc_msg_data_add(reply_msg,
2110 					       &op->notify.response_data);
2111 			break;
2112 		}
2113 	}
2114 }
2115 
2116 static void encode_pgid(void **p, const struct ceph_pg *pgid)
2117 {
2118 	ceph_encode_8(p, 1);
2119 	ceph_encode_64(p, pgid->pool);
2120 	ceph_encode_32(p, pgid->seed);
2121 	ceph_encode_32(p, -1); /* preferred */
2122 }
2123 
2124 static void encode_spgid(void **p, const struct ceph_spg *spgid)
2125 {
2126 	ceph_start_encoding(p, 1, 1, CEPH_PGID_ENCODING_LEN + 1);
2127 	encode_pgid(p, &spgid->pgid);
2128 	ceph_encode_8(p, spgid->shard);
2129 }
2130 
2131 static void encode_oloc(void **p, void *end,
2132 			const struct ceph_object_locator *oloc)
2133 {
2134 	ceph_start_encoding(p, 5, 4, ceph_oloc_encoding_size(oloc));
2135 	ceph_encode_64(p, oloc->pool);
2136 	ceph_encode_32(p, -1); /* preferred */
2137 	ceph_encode_32(p, 0);  /* key len */
2138 	if (oloc->pool_ns)
2139 		ceph_encode_string(p, end, oloc->pool_ns->str,
2140 				   oloc->pool_ns->len);
2141 	else
2142 		ceph_encode_32(p, 0);
2143 }
2144 
2145 static void encode_request_partial(struct ceph_osd_request *req,
2146 				   struct ceph_msg *msg)
2147 {
2148 	void *p = msg->front.iov_base;
2149 	void *const end = p + msg->front_alloc_len;
2150 	u32 data_len = 0;
2151 	int i;
2152 
2153 	if (req->r_flags & CEPH_OSD_FLAG_WRITE) {
2154 		/* snapshots aren't writeable */
2155 		WARN_ON(req->r_snapid != CEPH_NOSNAP);
2156 	} else {
2157 		WARN_ON(req->r_mtime.tv_sec || req->r_mtime.tv_nsec ||
2158 			req->r_data_offset || req->r_snapc);
2159 	}
2160 
2161 	setup_request_data(req);
2162 
2163 	encode_spgid(&p, &req->r_t.spgid); /* actual spg */
2164 	ceph_encode_32(&p, req->r_t.pgid.seed); /* raw hash */
2165 	ceph_encode_32(&p, req->r_osdc->osdmap->epoch);
2166 	ceph_encode_32(&p, req->r_flags);
2167 
2168 	/* reqid */
2169 	ceph_start_encoding(&p, 2, 2, sizeof(struct ceph_osd_reqid));
2170 	memset(p, 0, sizeof(struct ceph_osd_reqid));
2171 	p += sizeof(struct ceph_osd_reqid);
2172 
2173 	/* trace */
2174 	memset(p, 0, sizeof(struct ceph_blkin_trace_info));
2175 	p += sizeof(struct ceph_blkin_trace_info);
2176 
2177 	ceph_encode_32(&p, 0); /* client_inc, always 0 */
2178 	ceph_encode_timespec64(p, &req->r_mtime);
2179 	p += sizeof(struct ceph_timespec);
2180 
2181 	encode_oloc(&p, end, &req->r_t.target_oloc);
2182 	ceph_encode_string(&p, end, req->r_t.target_oid.name,
2183 			   req->r_t.target_oid.name_len);
2184 
2185 	/* ops, can imply data */
2186 	ceph_encode_16(&p, req->r_num_ops);
2187 	for (i = 0; i < req->r_num_ops; i++) {
2188 		data_len += osd_req_encode_op(p, &req->r_ops[i]);
2189 		p += sizeof(struct ceph_osd_op);
2190 	}
2191 
2192 	ceph_encode_64(&p, req->r_snapid); /* snapid */
2193 	if (req->r_snapc) {
2194 		ceph_encode_64(&p, req->r_snapc->seq);
2195 		ceph_encode_32(&p, req->r_snapc->num_snaps);
2196 		for (i = 0; i < req->r_snapc->num_snaps; i++)
2197 			ceph_encode_64(&p, req->r_snapc->snaps[i]);
2198 	} else {
2199 		ceph_encode_64(&p, 0); /* snap_seq */
2200 		ceph_encode_32(&p, 0); /* snaps len */
2201 	}
2202 
2203 	ceph_encode_32(&p, req->r_attempts); /* retry_attempt */
2204 	BUG_ON(p > end - 8); /* space for features */
2205 
2206 	msg->hdr.version = cpu_to_le16(8); /* MOSDOp v8 */
2207 	/* front_len is finalized in encode_request_finish() */
2208 	msg->front.iov_len = p - msg->front.iov_base;
2209 	msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2210 	msg->hdr.data_len = cpu_to_le32(data_len);
2211 	/*
2212 	 * The header "data_off" is a hint to the receiver allowing it
2213 	 * to align received data into its buffers such that there's no
2214 	 * need to re-copy it before writing it to disk (direct I/O).
2215 	 */
2216 	msg->hdr.data_off = cpu_to_le16(req->r_data_offset);
2217 
2218 	dout("%s req %p msg %p oid %s oid_len %d\n", __func__, req, msg,
2219 	     req->r_t.target_oid.name, req->r_t.target_oid.name_len);
2220 }
2221 
2222 static void encode_request_finish(struct ceph_msg *msg)
2223 {
2224 	void *p = msg->front.iov_base;
2225 	void *const partial_end = p + msg->front.iov_len;
2226 	void *const end = p + msg->front_alloc_len;
2227 
2228 	if (CEPH_HAVE_FEATURE(msg->con->peer_features, RESEND_ON_SPLIT)) {
2229 		/* luminous OSD -- encode features and be done */
2230 		p = partial_end;
2231 		ceph_encode_64(&p, msg->con->peer_features);
2232 	} else {
2233 		struct {
2234 			char spgid[CEPH_ENCODING_START_BLK_LEN +
2235 				   CEPH_PGID_ENCODING_LEN + 1];
2236 			__le32 hash;
2237 			__le32 epoch;
2238 			__le32 flags;
2239 			char reqid[CEPH_ENCODING_START_BLK_LEN +
2240 				   sizeof(struct ceph_osd_reqid)];
2241 			char trace[sizeof(struct ceph_blkin_trace_info)];
2242 			__le32 client_inc;
2243 			struct ceph_timespec mtime;
2244 		} __packed head;
2245 		struct ceph_pg pgid;
2246 		void *oloc, *oid, *tail;
2247 		int oloc_len, oid_len, tail_len;
2248 		int len;
2249 
2250 		/*
2251 		 * Pre-luminous OSD -- reencode v8 into v4 using @head
2252 		 * as a temporary buffer.  Encode the raw PG; the rest
2253 		 * is just a matter of moving oloc, oid and tail blobs
2254 		 * around.
2255 		 */
2256 		memcpy(&head, p, sizeof(head));
2257 		p += sizeof(head);
2258 
2259 		oloc = p;
2260 		p += CEPH_ENCODING_START_BLK_LEN;
2261 		pgid.pool = ceph_decode_64(&p);
2262 		p += 4 + 4; /* preferred, key len */
2263 		len = ceph_decode_32(&p);
2264 		p += len;   /* nspace */
2265 		oloc_len = p - oloc;
2266 
2267 		oid = p;
2268 		len = ceph_decode_32(&p);
2269 		p += len;
2270 		oid_len = p - oid;
2271 
2272 		tail = p;
2273 		tail_len = partial_end - p;
2274 
2275 		p = msg->front.iov_base;
2276 		ceph_encode_copy(&p, &head.client_inc, sizeof(head.client_inc));
2277 		ceph_encode_copy(&p, &head.epoch, sizeof(head.epoch));
2278 		ceph_encode_copy(&p, &head.flags, sizeof(head.flags));
2279 		ceph_encode_copy(&p, &head.mtime, sizeof(head.mtime));
2280 
2281 		/* reassert_version */
2282 		memset(p, 0, sizeof(struct ceph_eversion));
2283 		p += sizeof(struct ceph_eversion);
2284 
2285 		BUG_ON(p >= oloc);
2286 		memmove(p, oloc, oloc_len);
2287 		p += oloc_len;
2288 
2289 		pgid.seed = le32_to_cpu(head.hash);
2290 		encode_pgid(&p, &pgid); /* raw pg */
2291 
2292 		BUG_ON(p >= oid);
2293 		memmove(p, oid, oid_len);
2294 		p += oid_len;
2295 
2296 		/* tail -- ops, snapid, snapc, retry_attempt */
2297 		BUG_ON(p >= tail);
2298 		memmove(p, tail, tail_len);
2299 		p += tail_len;
2300 
2301 		msg->hdr.version = cpu_to_le16(4); /* MOSDOp v4 */
2302 	}
2303 
2304 	BUG_ON(p > end);
2305 	msg->front.iov_len = p - msg->front.iov_base;
2306 	msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2307 
2308 	dout("%s msg %p tid %llu %u+%u+%u v%d\n", __func__, msg,
2309 	     le64_to_cpu(msg->hdr.tid), le32_to_cpu(msg->hdr.front_len),
2310 	     le32_to_cpu(msg->hdr.middle_len), le32_to_cpu(msg->hdr.data_len),
2311 	     le16_to_cpu(msg->hdr.version));
2312 }
2313 
2314 /*
2315  * @req has to be assigned a tid and registered.
2316  */
2317 static void send_request(struct ceph_osd_request *req)
2318 {
2319 	struct ceph_osd *osd = req->r_osd;
2320 
2321 	verify_osd_locked(osd);
2322 	WARN_ON(osd->o_osd != req->r_t.osd);
2323 
2324 	/* backoff? */
2325 	if (should_plug_request(req))
2326 		return;
2327 
2328 	/*
2329 	 * We may have a previously queued request message hanging
2330 	 * around.  Cancel it to avoid corrupting the msgr.
2331 	 */
2332 	if (req->r_sent)
2333 		ceph_msg_revoke(req->r_request);
2334 
2335 	req->r_flags |= CEPH_OSD_FLAG_KNOWN_REDIR;
2336 	if (req->r_attempts)
2337 		req->r_flags |= CEPH_OSD_FLAG_RETRY;
2338 	else
2339 		WARN_ON(req->r_flags & CEPH_OSD_FLAG_RETRY);
2340 
2341 	encode_request_partial(req, req->r_request);
2342 
2343 	dout("%s req %p tid %llu to pgid %llu.%x spgid %llu.%xs%d osd%d e%u flags 0x%x attempt %d\n",
2344 	     __func__, req, req->r_tid, req->r_t.pgid.pool, req->r_t.pgid.seed,
2345 	     req->r_t.spgid.pgid.pool, req->r_t.spgid.pgid.seed,
2346 	     req->r_t.spgid.shard, osd->o_osd, req->r_t.epoch, req->r_flags,
2347 	     req->r_attempts);
2348 
2349 	req->r_t.paused = false;
2350 	req->r_stamp = jiffies;
2351 	req->r_attempts++;
2352 
2353 	req->r_sent = osd->o_incarnation;
2354 	req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
2355 	ceph_con_send(&osd->o_con, ceph_msg_get(req->r_request));
2356 }
2357 
2358 static void maybe_request_map(struct ceph_osd_client *osdc)
2359 {
2360 	bool continuous = false;
2361 
2362 	verify_osdc_locked(osdc);
2363 	WARN_ON(!osdc->osdmap->epoch);
2364 
2365 	if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
2366 	    ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD) ||
2367 	    ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) {
2368 		dout("%s osdc %p continuous\n", __func__, osdc);
2369 		continuous = true;
2370 	} else {
2371 		dout("%s osdc %p onetime\n", __func__, osdc);
2372 	}
2373 
2374 	if (ceph_monc_want_map(&osdc->client->monc, CEPH_SUB_OSDMAP,
2375 			       osdc->osdmap->epoch + 1, continuous))
2376 		ceph_monc_renew_subs(&osdc->client->monc);
2377 }
2378 
2379 static void complete_request(struct ceph_osd_request *req, int err);
2380 static void send_map_check(struct ceph_osd_request *req);
2381 
2382 static void __submit_request(struct ceph_osd_request *req, bool wrlocked)
2383 {
2384 	struct ceph_osd_client *osdc = req->r_osdc;
2385 	struct ceph_osd *osd;
2386 	enum calc_target_result ct_res;
2387 	int err = 0;
2388 	bool need_send = false;
2389 	bool promoted = false;
2390 
2391 	WARN_ON(req->r_tid);
2392 	dout("%s req %p wrlocked %d\n", __func__, req, wrlocked);
2393 
2394 again:
2395 	ct_res = calc_target(osdc, &req->r_t, false);
2396 	if (ct_res == CALC_TARGET_POOL_DNE && !wrlocked)
2397 		goto promote;
2398 
2399 	osd = lookup_create_osd(osdc, req->r_t.osd, wrlocked);
2400 	if (IS_ERR(osd)) {
2401 		WARN_ON(PTR_ERR(osd) != -EAGAIN || wrlocked);
2402 		goto promote;
2403 	}
2404 
2405 	if (osdc->abort_err) {
2406 		dout("req %p abort_err %d\n", req, osdc->abort_err);
2407 		err = osdc->abort_err;
2408 	} else if (osdc->osdmap->epoch < osdc->epoch_barrier) {
2409 		dout("req %p epoch %u barrier %u\n", req, osdc->osdmap->epoch,
2410 		     osdc->epoch_barrier);
2411 		req->r_t.paused = true;
2412 		maybe_request_map(osdc);
2413 	} else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
2414 		   ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) {
2415 		dout("req %p pausewr\n", req);
2416 		req->r_t.paused = true;
2417 		maybe_request_map(osdc);
2418 	} else if ((req->r_flags & CEPH_OSD_FLAG_READ) &&
2419 		   ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) {
2420 		dout("req %p pauserd\n", req);
2421 		req->r_t.paused = true;
2422 		maybe_request_map(osdc);
2423 	} else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
2424 		   !(req->r_flags & (CEPH_OSD_FLAG_FULL_TRY |
2425 				     CEPH_OSD_FLAG_FULL_FORCE)) &&
2426 		   (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
2427 		    pool_full(osdc, req->r_t.base_oloc.pool))) {
2428 		dout("req %p full/pool_full\n", req);
2429 		if (ceph_test_opt(osdc->client, ABORT_ON_FULL)) {
2430 			err = -ENOSPC;
2431 		} else {
2432 			if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL))
2433 				pr_warn_ratelimited("cluster is full (osdmap FULL)\n");
2434 			else
2435 				pr_warn_ratelimited("pool %lld is full or reached quota\n",
2436 						    req->r_t.base_oloc.pool);
2437 			req->r_t.paused = true;
2438 			maybe_request_map(osdc);
2439 		}
2440 	} else if (!osd_homeless(osd)) {
2441 		need_send = true;
2442 	} else {
2443 		maybe_request_map(osdc);
2444 	}
2445 
2446 	mutex_lock(&osd->lock);
2447 	/*
2448 	 * Assign the tid atomically with send_request() to protect
2449 	 * multiple writes to the same object from racing with each
2450 	 * other, resulting in out of order ops on the OSDs.
2451 	 */
2452 	req->r_tid = atomic64_inc_return(&osdc->last_tid);
2453 	link_request(osd, req);
2454 	if (need_send)
2455 		send_request(req);
2456 	else if (err)
2457 		complete_request(req, err);
2458 	mutex_unlock(&osd->lock);
2459 
2460 	if (!err && ct_res == CALC_TARGET_POOL_DNE)
2461 		send_map_check(req);
2462 
2463 	if (promoted)
2464 		downgrade_write(&osdc->lock);
2465 	return;
2466 
2467 promote:
2468 	up_read(&osdc->lock);
2469 	down_write(&osdc->lock);
2470 	wrlocked = true;
2471 	promoted = true;
2472 	goto again;
2473 }
2474 
2475 static void account_request(struct ceph_osd_request *req)
2476 {
2477 	WARN_ON(req->r_flags & (CEPH_OSD_FLAG_ACK | CEPH_OSD_FLAG_ONDISK));
2478 	WARN_ON(!(req->r_flags & (CEPH_OSD_FLAG_READ | CEPH_OSD_FLAG_WRITE)));
2479 
2480 	req->r_flags |= CEPH_OSD_FLAG_ONDISK;
2481 	atomic_inc(&req->r_osdc->num_requests);
2482 
2483 	req->r_start_stamp = jiffies;
2484 	req->r_start_latency = ktime_get();
2485 }
2486 
2487 static void submit_request(struct ceph_osd_request *req, bool wrlocked)
2488 {
2489 	ceph_osdc_get_request(req);
2490 	account_request(req);
2491 	__submit_request(req, wrlocked);
2492 }
2493 
2494 static void finish_request(struct ceph_osd_request *req)
2495 {
2496 	struct ceph_osd_client *osdc = req->r_osdc;
2497 
2498 	WARN_ON(lookup_request_mc(&osdc->map_checks, req->r_tid));
2499 	dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
2500 
2501 	req->r_end_latency = ktime_get();
2502 
2503 	if (req->r_osd) {
2504 		ceph_init_sparse_read(&req->r_osd->o_sparse_read);
2505 		unlink_request(req->r_osd, req);
2506 	}
2507 	atomic_dec(&osdc->num_requests);
2508 
2509 	/*
2510 	 * If an OSD has failed or returned and a request has been sent
2511 	 * twice, it's possible to get a reply and end up here while the
2512 	 * request message is queued for delivery.  We will ignore the
2513 	 * reply, so not a big deal, but better to try and catch it.
2514 	 */
2515 	ceph_msg_revoke(req->r_request);
2516 	ceph_msg_revoke_incoming(req->r_reply);
2517 }
2518 
2519 static void __complete_request(struct ceph_osd_request *req)
2520 {
2521 	dout("%s req %p tid %llu cb %ps result %d\n", __func__, req,
2522 	     req->r_tid, req->r_callback, req->r_result);
2523 
2524 	if (req->r_callback)
2525 		req->r_callback(req);
2526 	complete_all(&req->r_completion);
2527 	ceph_osdc_put_request(req);
2528 }
2529 
2530 static void complete_request_workfn(struct work_struct *work)
2531 {
2532 	struct ceph_osd_request *req =
2533 	    container_of(work, struct ceph_osd_request, r_complete_work);
2534 
2535 	__complete_request(req);
2536 }
2537 
2538 /*
2539  * This is open-coded in handle_reply().
2540  */
2541 static void complete_request(struct ceph_osd_request *req, int err)
2542 {
2543 	dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err);
2544 
2545 	req->r_result = err;
2546 	finish_request(req);
2547 
2548 	INIT_WORK(&req->r_complete_work, complete_request_workfn);
2549 	queue_work(req->r_osdc->completion_wq, &req->r_complete_work);
2550 }
2551 
2552 static void cancel_map_check(struct ceph_osd_request *req)
2553 {
2554 	struct ceph_osd_client *osdc = req->r_osdc;
2555 	struct ceph_osd_request *lookup_req;
2556 
2557 	verify_osdc_wrlocked(osdc);
2558 
2559 	lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid);
2560 	if (!lookup_req)
2561 		return;
2562 
2563 	WARN_ON(lookup_req != req);
2564 	erase_request_mc(&osdc->map_checks, req);
2565 	ceph_osdc_put_request(req);
2566 }
2567 
2568 static void cancel_request(struct ceph_osd_request *req)
2569 {
2570 	dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
2571 
2572 	cancel_map_check(req);
2573 	finish_request(req);
2574 	complete_all(&req->r_completion);
2575 	ceph_osdc_put_request(req);
2576 }
2577 
2578 static void abort_request(struct ceph_osd_request *req, int err)
2579 {
2580 	dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err);
2581 
2582 	cancel_map_check(req);
2583 	complete_request(req, err);
2584 }
2585 
2586 static int abort_fn(struct ceph_osd_request *req, void *arg)
2587 {
2588 	int err = *(int *)arg;
2589 
2590 	abort_request(req, err);
2591 	return 0; /* continue iteration */
2592 }
2593 
2594 /*
2595  * Abort all in-flight requests with @err and arrange for all future
2596  * requests to be failed immediately.
2597  */
2598 void ceph_osdc_abort_requests(struct ceph_osd_client *osdc, int err)
2599 {
2600 	dout("%s osdc %p err %d\n", __func__, osdc, err);
2601 	down_write(&osdc->lock);
2602 	for_each_request(osdc, abort_fn, &err);
2603 	osdc->abort_err = err;
2604 	up_write(&osdc->lock);
2605 }
2606 EXPORT_SYMBOL(ceph_osdc_abort_requests);
2607 
2608 void ceph_osdc_clear_abort_err(struct ceph_osd_client *osdc)
2609 {
2610 	down_write(&osdc->lock);
2611 	osdc->abort_err = 0;
2612 	up_write(&osdc->lock);
2613 }
2614 EXPORT_SYMBOL(ceph_osdc_clear_abort_err);
2615 
2616 static void update_epoch_barrier(struct ceph_osd_client *osdc, u32 eb)
2617 {
2618 	if (likely(eb > osdc->epoch_barrier)) {
2619 		dout("updating epoch_barrier from %u to %u\n",
2620 				osdc->epoch_barrier, eb);
2621 		osdc->epoch_barrier = eb;
2622 		/* Request map if we're not to the barrier yet */
2623 		if (eb > osdc->osdmap->epoch)
2624 			maybe_request_map(osdc);
2625 	}
2626 }
2627 
2628 void ceph_osdc_update_epoch_barrier(struct ceph_osd_client *osdc, u32 eb)
2629 {
2630 	down_read(&osdc->lock);
2631 	if (unlikely(eb > osdc->epoch_barrier)) {
2632 		up_read(&osdc->lock);
2633 		down_write(&osdc->lock);
2634 		update_epoch_barrier(osdc, eb);
2635 		up_write(&osdc->lock);
2636 	} else {
2637 		up_read(&osdc->lock);
2638 	}
2639 }
2640 EXPORT_SYMBOL(ceph_osdc_update_epoch_barrier);
2641 
2642 /*
2643  * We can end up releasing caps as a result of abort_request().
2644  * In that case, we probably want to ensure that the cap release message
2645  * has an updated epoch barrier in it, so set the epoch barrier prior to
2646  * aborting the first request.
2647  */
2648 static int abort_on_full_fn(struct ceph_osd_request *req, void *arg)
2649 {
2650 	struct ceph_osd_client *osdc = req->r_osdc;
2651 	bool *victims = arg;
2652 
2653 	if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
2654 	    (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
2655 	     pool_full(osdc, req->r_t.base_oloc.pool))) {
2656 		if (!*victims) {
2657 			update_epoch_barrier(osdc, osdc->osdmap->epoch);
2658 			*victims = true;
2659 		}
2660 		abort_request(req, -ENOSPC);
2661 	}
2662 
2663 	return 0; /* continue iteration */
2664 }
2665 
2666 /*
2667  * Drop all pending requests that are stalled waiting on a full condition to
2668  * clear, and complete them with ENOSPC as the return code. Set the
2669  * osdc->epoch_barrier to the latest map epoch that we've seen if any were
2670  * cancelled.
2671  */
2672 static void ceph_osdc_abort_on_full(struct ceph_osd_client *osdc)
2673 {
2674 	bool victims = false;
2675 
2676 	if (ceph_test_opt(osdc->client, ABORT_ON_FULL) &&
2677 	    (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || have_pool_full(osdc)))
2678 		for_each_request(osdc, abort_on_full_fn, &victims);
2679 }
2680 
2681 static void check_pool_dne(struct ceph_osd_request *req)
2682 {
2683 	struct ceph_osd_client *osdc = req->r_osdc;
2684 	struct ceph_osdmap *map = osdc->osdmap;
2685 
2686 	verify_osdc_wrlocked(osdc);
2687 	WARN_ON(!map->epoch);
2688 
2689 	if (req->r_attempts) {
2690 		/*
2691 		 * We sent a request earlier, which means that
2692 		 * previously the pool existed, and now it does not
2693 		 * (i.e., it was deleted).
2694 		 */
2695 		req->r_map_dne_bound = map->epoch;
2696 		dout("%s req %p tid %llu pool disappeared\n", __func__, req,
2697 		     req->r_tid);
2698 	} else {
2699 		dout("%s req %p tid %llu map_dne_bound %u have %u\n", __func__,
2700 		     req, req->r_tid, req->r_map_dne_bound, map->epoch);
2701 	}
2702 
2703 	if (req->r_map_dne_bound) {
2704 		if (map->epoch >= req->r_map_dne_bound) {
2705 			/* we had a new enough map */
2706 			pr_info_ratelimited("tid %llu pool does not exist\n",
2707 					    req->r_tid);
2708 			complete_request(req, -ENOENT);
2709 		}
2710 	} else {
2711 		send_map_check(req);
2712 	}
2713 }
2714 
2715 static void map_check_cb(struct ceph_mon_generic_request *greq)
2716 {
2717 	struct ceph_osd_client *osdc = &greq->monc->client->osdc;
2718 	struct ceph_osd_request *req;
2719 	u64 tid = greq->private_data;
2720 
2721 	WARN_ON(greq->result || !greq->u.newest);
2722 
2723 	down_write(&osdc->lock);
2724 	req = lookup_request_mc(&osdc->map_checks, tid);
2725 	if (!req) {
2726 		dout("%s tid %llu dne\n", __func__, tid);
2727 		goto out_unlock;
2728 	}
2729 
2730 	dout("%s req %p tid %llu map_dne_bound %u newest %llu\n", __func__,
2731 	     req, req->r_tid, req->r_map_dne_bound, greq->u.newest);
2732 	if (!req->r_map_dne_bound)
2733 		req->r_map_dne_bound = greq->u.newest;
2734 	erase_request_mc(&osdc->map_checks, req);
2735 	check_pool_dne(req);
2736 
2737 	ceph_osdc_put_request(req);
2738 out_unlock:
2739 	up_write(&osdc->lock);
2740 }
2741 
2742 static void send_map_check(struct ceph_osd_request *req)
2743 {
2744 	struct ceph_osd_client *osdc = req->r_osdc;
2745 	struct ceph_osd_request *lookup_req;
2746 	int ret;
2747 
2748 	verify_osdc_wrlocked(osdc);
2749 
2750 	lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid);
2751 	if (lookup_req) {
2752 		WARN_ON(lookup_req != req);
2753 		return;
2754 	}
2755 
2756 	ceph_osdc_get_request(req);
2757 	insert_request_mc(&osdc->map_checks, req);
2758 	ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap",
2759 					  map_check_cb, req->r_tid);
2760 	WARN_ON(ret);
2761 }
2762 
2763 /*
2764  * lingering requests, watch/notify v2 infrastructure
2765  */
2766 static void linger_release(struct kref *kref)
2767 {
2768 	struct ceph_osd_linger_request *lreq =
2769 	    container_of(kref, struct ceph_osd_linger_request, kref);
2770 
2771 	dout("%s lreq %p reg_req %p ping_req %p\n", __func__, lreq,
2772 	     lreq->reg_req, lreq->ping_req);
2773 	WARN_ON(!RB_EMPTY_NODE(&lreq->node));
2774 	WARN_ON(!RB_EMPTY_NODE(&lreq->osdc_node));
2775 	WARN_ON(!RB_EMPTY_NODE(&lreq->mc_node));
2776 	WARN_ON(!list_empty(&lreq->scan_item));
2777 	WARN_ON(!list_empty(&lreq->pending_lworks));
2778 	WARN_ON(lreq->osd);
2779 
2780 	if (lreq->request_pl)
2781 		ceph_pagelist_release(lreq->request_pl);
2782 	if (lreq->notify_id_pages)
2783 		ceph_release_page_vector(lreq->notify_id_pages, 1);
2784 
2785 	ceph_osdc_put_request(lreq->reg_req);
2786 	ceph_osdc_put_request(lreq->ping_req);
2787 	target_destroy(&lreq->t);
2788 	kfree(lreq);
2789 }
2790 
2791 static void linger_put(struct ceph_osd_linger_request *lreq)
2792 {
2793 	if (lreq)
2794 		kref_put(&lreq->kref, linger_release);
2795 }
2796 
2797 static struct ceph_osd_linger_request *
2798 linger_get(struct ceph_osd_linger_request *lreq)
2799 {
2800 	kref_get(&lreq->kref);
2801 	return lreq;
2802 }
2803 
2804 static struct ceph_osd_linger_request *
2805 linger_alloc(struct ceph_osd_client *osdc)
2806 {
2807 	struct ceph_osd_linger_request *lreq;
2808 
2809 	lreq = kzalloc(sizeof(*lreq), GFP_NOIO);
2810 	if (!lreq)
2811 		return NULL;
2812 
2813 	kref_init(&lreq->kref);
2814 	mutex_init(&lreq->lock);
2815 	RB_CLEAR_NODE(&lreq->node);
2816 	RB_CLEAR_NODE(&lreq->osdc_node);
2817 	RB_CLEAR_NODE(&lreq->mc_node);
2818 	INIT_LIST_HEAD(&lreq->scan_item);
2819 	INIT_LIST_HEAD(&lreq->pending_lworks);
2820 	init_completion(&lreq->reg_commit_wait);
2821 	init_completion(&lreq->notify_finish_wait);
2822 
2823 	lreq->osdc = osdc;
2824 	target_init(&lreq->t);
2825 
2826 	dout("%s lreq %p\n", __func__, lreq);
2827 	return lreq;
2828 }
2829 
2830 DEFINE_RB_INSDEL_FUNCS(linger, struct ceph_osd_linger_request, linger_id, node)
2831 DEFINE_RB_FUNCS(linger_osdc, struct ceph_osd_linger_request, linger_id, osdc_node)
2832 DEFINE_RB_FUNCS(linger_mc, struct ceph_osd_linger_request, linger_id, mc_node)
2833 
2834 /*
2835  * Create linger request <-> OSD session relation.
2836  *
2837  * @lreq has to be registered, @osd may be homeless.
2838  */
2839 static void link_linger(struct ceph_osd *osd,
2840 			struct ceph_osd_linger_request *lreq)
2841 {
2842 	verify_osd_locked(osd);
2843 	WARN_ON(!lreq->linger_id || lreq->osd);
2844 	dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd,
2845 	     osd->o_osd, lreq, lreq->linger_id);
2846 
2847 	if (!osd_homeless(osd))
2848 		__remove_osd_from_lru(osd);
2849 	else
2850 		atomic_inc(&osd->o_osdc->num_homeless);
2851 
2852 	get_osd(osd);
2853 	insert_linger(&osd->o_linger_requests, lreq);
2854 	lreq->osd = osd;
2855 }
2856 
2857 static void unlink_linger(struct ceph_osd *osd,
2858 			  struct ceph_osd_linger_request *lreq)
2859 {
2860 	verify_osd_locked(osd);
2861 	WARN_ON(lreq->osd != osd);
2862 	dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd,
2863 	     osd->o_osd, lreq, lreq->linger_id);
2864 
2865 	lreq->osd = NULL;
2866 	erase_linger(&osd->o_linger_requests, lreq);
2867 	put_osd(osd);
2868 
2869 	if (!osd_homeless(osd))
2870 		maybe_move_osd_to_lru(osd);
2871 	else
2872 		atomic_dec(&osd->o_osdc->num_homeless);
2873 }
2874 
2875 static bool __linger_registered(struct ceph_osd_linger_request *lreq)
2876 {
2877 	verify_osdc_locked(lreq->osdc);
2878 
2879 	return !RB_EMPTY_NODE(&lreq->osdc_node);
2880 }
2881 
2882 static bool linger_registered(struct ceph_osd_linger_request *lreq)
2883 {
2884 	struct ceph_osd_client *osdc = lreq->osdc;
2885 	bool registered;
2886 
2887 	down_read(&osdc->lock);
2888 	registered = __linger_registered(lreq);
2889 	up_read(&osdc->lock);
2890 
2891 	return registered;
2892 }
2893 
2894 static void linger_register(struct ceph_osd_linger_request *lreq)
2895 {
2896 	struct ceph_osd_client *osdc = lreq->osdc;
2897 
2898 	verify_osdc_wrlocked(osdc);
2899 	WARN_ON(lreq->linger_id);
2900 
2901 	linger_get(lreq);
2902 	lreq->linger_id = ++osdc->last_linger_id;
2903 	insert_linger_osdc(&osdc->linger_requests, lreq);
2904 }
2905 
2906 static void linger_unregister(struct ceph_osd_linger_request *lreq)
2907 {
2908 	struct ceph_osd_client *osdc = lreq->osdc;
2909 
2910 	verify_osdc_wrlocked(osdc);
2911 
2912 	erase_linger_osdc(&osdc->linger_requests, lreq);
2913 	linger_put(lreq);
2914 }
2915 
2916 static void cancel_linger_request(struct ceph_osd_request *req)
2917 {
2918 	struct ceph_osd_linger_request *lreq = req->r_priv;
2919 
2920 	WARN_ON(!req->r_linger);
2921 	cancel_request(req);
2922 	linger_put(lreq);
2923 }
2924 
2925 struct linger_work {
2926 	struct work_struct work;
2927 	struct ceph_osd_linger_request *lreq;
2928 	struct list_head pending_item;
2929 	unsigned long queued_stamp;
2930 
2931 	union {
2932 		struct {
2933 			u64 notify_id;
2934 			u64 notifier_id;
2935 			void *payload; /* points into @msg front */
2936 			size_t payload_len;
2937 
2938 			struct ceph_msg *msg; /* for ceph_msg_put() */
2939 		} notify;
2940 		struct {
2941 			int err;
2942 		} error;
2943 	};
2944 };
2945 
2946 static struct linger_work *lwork_alloc(struct ceph_osd_linger_request *lreq,
2947 				       work_func_t workfn)
2948 {
2949 	struct linger_work *lwork;
2950 
2951 	lwork = kzalloc(sizeof(*lwork), GFP_NOIO);
2952 	if (!lwork)
2953 		return NULL;
2954 
2955 	INIT_WORK(&lwork->work, workfn);
2956 	INIT_LIST_HEAD(&lwork->pending_item);
2957 	lwork->lreq = linger_get(lreq);
2958 
2959 	return lwork;
2960 }
2961 
2962 static void lwork_free(struct linger_work *lwork)
2963 {
2964 	struct ceph_osd_linger_request *lreq = lwork->lreq;
2965 
2966 	mutex_lock(&lreq->lock);
2967 	list_del(&lwork->pending_item);
2968 	mutex_unlock(&lreq->lock);
2969 
2970 	linger_put(lreq);
2971 	kfree(lwork);
2972 }
2973 
2974 static void lwork_queue(struct linger_work *lwork)
2975 {
2976 	struct ceph_osd_linger_request *lreq = lwork->lreq;
2977 	struct ceph_osd_client *osdc = lreq->osdc;
2978 
2979 	verify_lreq_locked(lreq);
2980 	WARN_ON(!list_empty(&lwork->pending_item));
2981 
2982 	lwork->queued_stamp = jiffies;
2983 	list_add_tail(&lwork->pending_item, &lreq->pending_lworks);
2984 	queue_work(osdc->notify_wq, &lwork->work);
2985 }
2986 
2987 static void do_watch_notify(struct work_struct *w)
2988 {
2989 	struct linger_work *lwork = container_of(w, struct linger_work, work);
2990 	struct ceph_osd_linger_request *lreq = lwork->lreq;
2991 
2992 	if (!linger_registered(lreq)) {
2993 		dout("%s lreq %p not registered\n", __func__, lreq);
2994 		goto out;
2995 	}
2996 
2997 	WARN_ON(!lreq->is_watch);
2998 	dout("%s lreq %p notify_id %llu notifier_id %llu payload_len %zu\n",
2999 	     __func__, lreq, lwork->notify.notify_id, lwork->notify.notifier_id,
3000 	     lwork->notify.payload_len);
3001 	lreq->wcb(lreq->data, lwork->notify.notify_id, lreq->linger_id,
3002 		  lwork->notify.notifier_id, lwork->notify.payload,
3003 		  lwork->notify.payload_len);
3004 
3005 out:
3006 	ceph_msg_put(lwork->notify.msg);
3007 	lwork_free(lwork);
3008 }
3009 
3010 static void do_watch_error(struct work_struct *w)
3011 {
3012 	struct linger_work *lwork = container_of(w, struct linger_work, work);
3013 	struct ceph_osd_linger_request *lreq = lwork->lreq;
3014 
3015 	if (!linger_registered(lreq)) {
3016 		dout("%s lreq %p not registered\n", __func__, lreq);
3017 		goto out;
3018 	}
3019 
3020 	dout("%s lreq %p err %d\n", __func__, lreq, lwork->error.err);
3021 	lreq->errcb(lreq->data, lreq->linger_id, lwork->error.err);
3022 
3023 out:
3024 	lwork_free(lwork);
3025 }
3026 
3027 static void queue_watch_error(struct ceph_osd_linger_request *lreq)
3028 {
3029 	struct linger_work *lwork;
3030 
3031 	lwork = lwork_alloc(lreq, do_watch_error);
3032 	if (!lwork) {
3033 		pr_err("failed to allocate error-lwork\n");
3034 		return;
3035 	}
3036 
3037 	lwork->error.err = lreq->last_error;
3038 	lwork_queue(lwork);
3039 }
3040 
3041 static void linger_reg_commit_complete(struct ceph_osd_linger_request *lreq,
3042 				       int result)
3043 {
3044 	if (!completion_done(&lreq->reg_commit_wait)) {
3045 		lreq->reg_commit_error = (result <= 0 ? result : 0);
3046 		complete_all(&lreq->reg_commit_wait);
3047 	}
3048 }
3049 
3050 static void linger_commit_cb(struct ceph_osd_request *req)
3051 {
3052 	struct ceph_osd_linger_request *lreq = req->r_priv;
3053 
3054 	mutex_lock(&lreq->lock);
3055 	if (req != lreq->reg_req) {
3056 		dout("%s lreq %p linger_id %llu unknown req (%p != %p)\n",
3057 		     __func__, lreq, lreq->linger_id, req, lreq->reg_req);
3058 		goto out;
3059 	}
3060 
3061 	dout("%s lreq %p linger_id %llu result %d\n", __func__, lreq,
3062 	     lreq->linger_id, req->r_result);
3063 	linger_reg_commit_complete(lreq, req->r_result);
3064 	lreq->committed = true;
3065 
3066 	if (!lreq->is_watch) {
3067 		struct ceph_osd_data *osd_data =
3068 		    osd_req_op_data(req, 0, notify, response_data);
3069 		void *p = page_address(osd_data->pages[0]);
3070 
3071 		WARN_ON(req->r_ops[0].op != CEPH_OSD_OP_NOTIFY ||
3072 			osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
3073 
3074 		/* make note of the notify_id */
3075 		if (req->r_ops[0].outdata_len >= sizeof(u64)) {
3076 			lreq->notify_id = ceph_decode_64(&p);
3077 			dout("lreq %p notify_id %llu\n", lreq,
3078 			     lreq->notify_id);
3079 		} else {
3080 			dout("lreq %p no notify_id\n", lreq);
3081 		}
3082 	}
3083 
3084 out:
3085 	mutex_unlock(&lreq->lock);
3086 	linger_put(lreq);
3087 }
3088 
3089 static int normalize_watch_error(int err)
3090 {
3091 	/*
3092 	 * Translate ENOENT -> ENOTCONN so that a delete->disconnection
3093 	 * notification and a failure to reconnect because we raced with
3094 	 * the delete appear the same to the user.
3095 	 */
3096 	if (err == -ENOENT)
3097 		err = -ENOTCONN;
3098 
3099 	return err;
3100 }
3101 
3102 static void linger_reconnect_cb(struct ceph_osd_request *req)
3103 {
3104 	struct ceph_osd_linger_request *lreq = req->r_priv;
3105 
3106 	mutex_lock(&lreq->lock);
3107 	if (req != lreq->reg_req) {
3108 		dout("%s lreq %p linger_id %llu unknown req (%p != %p)\n",
3109 		     __func__, lreq, lreq->linger_id, req, lreq->reg_req);
3110 		goto out;
3111 	}
3112 
3113 	dout("%s lreq %p linger_id %llu result %d last_error %d\n", __func__,
3114 	     lreq, lreq->linger_id, req->r_result, lreq->last_error);
3115 	if (req->r_result < 0) {
3116 		if (!lreq->last_error) {
3117 			lreq->last_error = normalize_watch_error(req->r_result);
3118 			queue_watch_error(lreq);
3119 		}
3120 	}
3121 
3122 out:
3123 	mutex_unlock(&lreq->lock);
3124 	linger_put(lreq);
3125 }
3126 
3127 static void send_linger(struct ceph_osd_linger_request *lreq)
3128 {
3129 	struct ceph_osd_client *osdc = lreq->osdc;
3130 	struct ceph_osd_request *req;
3131 	int ret;
3132 
3133 	verify_osdc_wrlocked(osdc);
3134 	mutex_lock(&lreq->lock);
3135 	dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
3136 
3137 	if (lreq->reg_req) {
3138 		if (lreq->reg_req->r_osd)
3139 			cancel_linger_request(lreq->reg_req);
3140 		ceph_osdc_put_request(lreq->reg_req);
3141 	}
3142 
3143 	req = ceph_osdc_alloc_request(osdc, NULL, 1, true, GFP_NOIO);
3144 	BUG_ON(!req);
3145 
3146 	target_copy(&req->r_t, &lreq->t);
3147 	req->r_mtime = lreq->mtime;
3148 
3149 	if (lreq->is_watch && lreq->committed) {
3150 		osd_req_op_watch_init(req, 0, CEPH_OSD_WATCH_OP_RECONNECT,
3151 				      lreq->linger_id, ++lreq->register_gen);
3152 		dout("lreq %p reconnect register_gen %u\n", lreq,
3153 		     req->r_ops[0].watch.gen);
3154 		req->r_callback = linger_reconnect_cb;
3155 	} else {
3156 		if (lreq->is_watch) {
3157 			osd_req_op_watch_init(req, 0, CEPH_OSD_WATCH_OP_WATCH,
3158 					      lreq->linger_id, 0);
3159 		} else {
3160 			lreq->notify_id = 0;
3161 
3162 			refcount_inc(&lreq->request_pl->refcnt);
3163 			osd_req_op_notify_init(req, 0, lreq->linger_id,
3164 					       lreq->request_pl);
3165 			ceph_osd_data_pages_init(
3166 			    osd_req_op_data(req, 0, notify, response_data),
3167 			    lreq->notify_id_pages, PAGE_SIZE, 0, false, false);
3168 		}
3169 		dout("lreq %p register\n", lreq);
3170 		req->r_callback = linger_commit_cb;
3171 	}
3172 
3173 	ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
3174 	BUG_ON(ret);
3175 
3176 	req->r_priv = linger_get(lreq);
3177 	req->r_linger = true;
3178 	lreq->reg_req = req;
3179 	mutex_unlock(&lreq->lock);
3180 
3181 	submit_request(req, true);
3182 }
3183 
3184 static void linger_ping_cb(struct ceph_osd_request *req)
3185 {
3186 	struct ceph_osd_linger_request *lreq = req->r_priv;
3187 
3188 	mutex_lock(&lreq->lock);
3189 	if (req != lreq->ping_req) {
3190 		dout("%s lreq %p linger_id %llu unknown req (%p != %p)\n",
3191 		     __func__, lreq, lreq->linger_id, req, lreq->ping_req);
3192 		goto out;
3193 	}
3194 
3195 	dout("%s lreq %p linger_id %llu result %d ping_sent %lu last_error %d\n",
3196 	     __func__, lreq, lreq->linger_id, req->r_result, lreq->ping_sent,
3197 	     lreq->last_error);
3198 	if (lreq->register_gen == req->r_ops[0].watch.gen) {
3199 		if (!req->r_result) {
3200 			lreq->watch_valid_thru = lreq->ping_sent;
3201 		} else if (!lreq->last_error) {
3202 			lreq->last_error = normalize_watch_error(req->r_result);
3203 			queue_watch_error(lreq);
3204 		}
3205 	} else {
3206 		dout("lreq %p register_gen %u ignoring old pong %u\n", lreq,
3207 		     lreq->register_gen, req->r_ops[0].watch.gen);
3208 	}
3209 
3210 out:
3211 	mutex_unlock(&lreq->lock);
3212 	linger_put(lreq);
3213 }
3214 
3215 static void send_linger_ping(struct ceph_osd_linger_request *lreq)
3216 {
3217 	struct ceph_osd_client *osdc = lreq->osdc;
3218 	struct ceph_osd_request *req;
3219 	int ret;
3220 
3221 	if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) {
3222 		dout("%s PAUSERD\n", __func__);
3223 		return;
3224 	}
3225 
3226 	lreq->ping_sent = jiffies;
3227 	dout("%s lreq %p linger_id %llu ping_sent %lu register_gen %u\n",
3228 	     __func__, lreq, lreq->linger_id, lreq->ping_sent,
3229 	     lreq->register_gen);
3230 
3231 	if (lreq->ping_req) {
3232 		if (lreq->ping_req->r_osd)
3233 			cancel_linger_request(lreq->ping_req);
3234 		ceph_osdc_put_request(lreq->ping_req);
3235 	}
3236 
3237 	req = ceph_osdc_alloc_request(osdc, NULL, 1, true, GFP_NOIO);
3238 	BUG_ON(!req);
3239 
3240 	target_copy(&req->r_t, &lreq->t);
3241 	osd_req_op_watch_init(req, 0, CEPH_OSD_WATCH_OP_PING, lreq->linger_id,
3242 			      lreq->register_gen);
3243 	req->r_callback = linger_ping_cb;
3244 
3245 	ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
3246 	BUG_ON(ret);
3247 
3248 	req->r_priv = linger_get(lreq);
3249 	req->r_linger = true;
3250 	lreq->ping_req = req;
3251 
3252 	ceph_osdc_get_request(req);
3253 	account_request(req);
3254 	req->r_tid = atomic64_inc_return(&osdc->last_tid);
3255 	link_request(lreq->osd, req);
3256 	send_request(req);
3257 }
3258 
3259 static void linger_submit(struct ceph_osd_linger_request *lreq)
3260 {
3261 	struct ceph_osd_client *osdc = lreq->osdc;
3262 	struct ceph_osd *osd;
3263 
3264 	down_write(&osdc->lock);
3265 	linger_register(lreq);
3266 
3267 	calc_target(osdc, &lreq->t, false);
3268 	osd = lookup_create_osd(osdc, lreq->t.osd, true);
3269 	link_linger(osd, lreq);
3270 
3271 	send_linger(lreq);
3272 	up_write(&osdc->lock);
3273 }
3274 
3275 static void cancel_linger_map_check(struct ceph_osd_linger_request *lreq)
3276 {
3277 	struct ceph_osd_client *osdc = lreq->osdc;
3278 	struct ceph_osd_linger_request *lookup_lreq;
3279 
3280 	verify_osdc_wrlocked(osdc);
3281 
3282 	lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks,
3283 				       lreq->linger_id);
3284 	if (!lookup_lreq)
3285 		return;
3286 
3287 	WARN_ON(lookup_lreq != lreq);
3288 	erase_linger_mc(&osdc->linger_map_checks, lreq);
3289 	linger_put(lreq);
3290 }
3291 
3292 /*
3293  * @lreq has to be both registered and linked.
3294  */
3295 static void __linger_cancel(struct ceph_osd_linger_request *lreq)
3296 {
3297 	if (lreq->ping_req && lreq->ping_req->r_osd)
3298 		cancel_linger_request(lreq->ping_req);
3299 	if (lreq->reg_req && lreq->reg_req->r_osd)
3300 		cancel_linger_request(lreq->reg_req);
3301 	cancel_linger_map_check(lreq);
3302 	unlink_linger(lreq->osd, lreq);
3303 	linger_unregister(lreq);
3304 }
3305 
3306 static void linger_cancel(struct ceph_osd_linger_request *lreq)
3307 {
3308 	struct ceph_osd_client *osdc = lreq->osdc;
3309 
3310 	down_write(&osdc->lock);
3311 	if (__linger_registered(lreq))
3312 		__linger_cancel(lreq);
3313 	up_write(&osdc->lock);
3314 }
3315 
3316 static void send_linger_map_check(struct ceph_osd_linger_request *lreq);
3317 
3318 static void check_linger_pool_dne(struct ceph_osd_linger_request *lreq)
3319 {
3320 	struct ceph_osd_client *osdc = lreq->osdc;
3321 	struct ceph_osdmap *map = osdc->osdmap;
3322 
3323 	verify_osdc_wrlocked(osdc);
3324 	WARN_ON(!map->epoch);
3325 
3326 	if (lreq->register_gen) {
3327 		lreq->map_dne_bound = map->epoch;
3328 		dout("%s lreq %p linger_id %llu pool disappeared\n", __func__,
3329 		     lreq, lreq->linger_id);
3330 	} else {
3331 		dout("%s lreq %p linger_id %llu map_dne_bound %u have %u\n",
3332 		     __func__, lreq, lreq->linger_id, lreq->map_dne_bound,
3333 		     map->epoch);
3334 	}
3335 
3336 	if (lreq->map_dne_bound) {
3337 		if (map->epoch >= lreq->map_dne_bound) {
3338 			/* we had a new enough map */
3339 			pr_info("linger_id %llu pool does not exist\n",
3340 				lreq->linger_id);
3341 			linger_reg_commit_complete(lreq, -ENOENT);
3342 			__linger_cancel(lreq);
3343 		}
3344 	} else {
3345 		send_linger_map_check(lreq);
3346 	}
3347 }
3348 
3349 static void linger_map_check_cb(struct ceph_mon_generic_request *greq)
3350 {
3351 	struct ceph_osd_client *osdc = &greq->monc->client->osdc;
3352 	struct ceph_osd_linger_request *lreq;
3353 	u64 linger_id = greq->private_data;
3354 
3355 	WARN_ON(greq->result || !greq->u.newest);
3356 
3357 	down_write(&osdc->lock);
3358 	lreq = lookup_linger_mc(&osdc->linger_map_checks, linger_id);
3359 	if (!lreq) {
3360 		dout("%s linger_id %llu dne\n", __func__, linger_id);
3361 		goto out_unlock;
3362 	}
3363 
3364 	dout("%s lreq %p linger_id %llu map_dne_bound %u newest %llu\n",
3365 	     __func__, lreq, lreq->linger_id, lreq->map_dne_bound,
3366 	     greq->u.newest);
3367 	if (!lreq->map_dne_bound)
3368 		lreq->map_dne_bound = greq->u.newest;
3369 	erase_linger_mc(&osdc->linger_map_checks, lreq);
3370 	check_linger_pool_dne(lreq);
3371 
3372 	linger_put(lreq);
3373 out_unlock:
3374 	up_write(&osdc->lock);
3375 }
3376 
3377 static void send_linger_map_check(struct ceph_osd_linger_request *lreq)
3378 {
3379 	struct ceph_osd_client *osdc = lreq->osdc;
3380 	struct ceph_osd_linger_request *lookup_lreq;
3381 	int ret;
3382 
3383 	verify_osdc_wrlocked(osdc);
3384 
3385 	lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks,
3386 				       lreq->linger_id);
3387 	if (lookup_lreq) {
3388 		WARN_ON(lookup_lreq != lreq);
3389 		return;
3390 	}
3391 
3392 	linger_get(lreq);
3393 	insert_linger_mc(&osdc->linger_map_checks, lreq);
3394 	ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap",
3395 					  linger_map_check_cb, lreq->linger_id);
3396 	WARN_ON(ret);
3397 }
3398 
3399 static int linger_reg_commit_wait(struct ceph_osd_linger_request *lreq)
3400 {
3401 	int ret;
3402 
3403 	dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
3404 	ret = wait_for_completion_killable(&lreq->reg_commit_wait);
3405 	return ret ?: lreq->reg_commit_error;
3406 }
3407 
3408 static int linger_notify_finish_wait(struct ceph_osd_linger_request *lreq,
3409 				     unsigned long timeout)
3410 {
3411 	long left;
3412 
3413 	dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
3414 	left = wait_for_completion_killable_timeout(&lreq->notify_finish_wait,
3415 						ceph_timeout_jiffies(timeout));
3416 	if (left <= 0)
3417 		left = left ?: -ETIMEDOUT;
3418 	else
3419 		left = lreq->notify_finish_error; /* completed */
3420 
3421 	return left;
3422 }
3423 
3424 /*
3425  * Timeout callback, called every N seconds.  When 1 or more OSD
3426  * requests has been active for more than N seconds, we send a keepalive
3427  * (tag + timestamp) to its OSD to ensure any communications channel
3428  * reset is detected.
3429  */
3430 static void handle_timeout(struct work_struct *work)
3431 {
3432 	struct ceph_osd_client *osdc =
3433 		container_of(work, struct ceph_osd_client, timeout_work.work);
3434 	struct ceph_options *opts = osdc->client->options;
3435 	unsigned long cutoff = jiffies - opts->osd_keepalive_timeout;
3436 	unsigned long expiry_cutoff = jiffies - opts->osd_request_timeout;
3437 	LIST_HEAD(slow_osds);
3438 	struct rb_node *n, *p;
3439 
3440 	dout("%s osdc %p\n", __func__, osdc);
3441 	down_write(&osdc->lock);
3442 
3443 	/*
3444 	 * ping osds that are a bit slow.  this ensures that if there
3445 	 * is a break in the TCP connection we will notice, and reopen
3446 	 * a connection with that osd (from the fault callback).
3447 	 */
3448 	for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
3449 		struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
3450 		bool found = false;
3451 
3452 		for (p = rb_first(&osd->o_requests); p; ) {
3453 			struct ceph_osd_request *req =
3454 			    rb_entry(p, struct ceph_osd_request, r_node);
3455 
3456 			p = rb_next(p); /* abort_request() */
3457 
3458 			if (time_before(req->r_stamp, cutoff)) {
3459 				dout(" req %p tid %llu on osd%d is laggy\n",
3460 				     req, req->r_tid, osd->o_osd);
3461 				found = true;
3462 			}
3463 			if (opts->osd_request_timeout &&
3464 			    time_before(req->r_start_stamp, expiry_cutoff)) {
3465 				pr_err_ratelimited("tid %llu on osd%d timeout\n",
3466 				       req->r_tid, osd->o_osd);
3467 				abort_request(req, -ETIMEDOUT);
3468 			}
3469 		}
3470 		for (p = rb_first(&osd->o_linger_requests); p; p = rb_next(p)) {
3471 			struct ceph_osd_linger_request *lreq =
3472 			    rb_entry(p, struct ceph_osd_linger_request, node);
3473 
3474 			dout(" lreq %p linger_id %llu is served by osd%d\n",
3475 			     lreq, lreq->linger_id, osd->o_osd);
3476 			found = true;
3477 
3478 			mutex_lock(&lreq->lock);
3479 			if (lreq->is_watch && lreq->committed && !lreq->last_error)
3480 				send_linger_ping(lreq);
3481 			mutex_unlock(&lreq->lock);
3482 		}
3483 
3484 		if (found)
3485 			list_move_tail(&osd->o_keepalive_item, &slow_osds);
3486 	}
3487 
3488 	if (opts->osd_request_timeout) {
3489 		for (p = rb_first(&osdc->homeless_osd.o_requests); p; ) {
3490 			struct ceph_osd_request *req =
3491 			    rb_entry(p, struct ceph_osd_request, r_node);
3492 
3493 			p = rb_next(p); /* abort_request() */
3494 
3495 			if (time_before(req->r_start_stamp, expiry_cutoff)) {
3496 				pr_err_ratelimited("tid %llu on osd%d timeout\n",
3497 				       req->r_tid, osdc->homeless_osd.o_osd);
3498 				abort_request(req, -ETIMEDOUT);
3499 			}
3500 		}
3501 	}
3502 
3503 	if (atomic_read(&osdc->num_homeless) || !list_empty(&slow_osds))
3504 		maybe_request_map(osdc);
3505 
3506 	while (!list_empty(&slow_osds)) {
3507 		struct ceph_osd *osd = list_first_entry(&slow_osds,
3508 							struct ceph_osd,
3509 							o_keepalive_item);
3510 		list_del_init(&osd->o_keepalive_item);
3511 		ceph_con_keepalive(&osd->o_con);
3512 	}
3513 
3514 	up_write(&osdc->lock);
3515 	schedule_delayed_work(&osdc->timeout_work,
3516 			      osdc->client->options->osd_keepalive_timeout);
3517 }
3518 
3519 static void handle_osds_timeout(struct work_struct *work)
3520 {
3521 	struct ceph_osd_client *osdc =
3522 		container_of(work, struct ceph_osd_client,
3523 			     osds_timeout_work.work);
3524 	unsigned long delay = osdc->client->options->osd_idle_ttl / 4;
3525 	struct ceph_osd *osd, *nosd;
3526 
3527 	dout("%s osdc %p\n", __func__, osdc);
3528 	down_write(&osdc->lock);
3529 	list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
3530 		if (time_before(jiffies, osd->lru_ttl))
3531 			break;
3532 
3533 		WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests));
3534 		WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests));
3535 		close_osd(osd);
3536 	}
3537 
3538 	up_write(&osdc->lock);
3539 	schedule_delayed_work(&osdc->osds_timeout_work,
3540 			      round_jiffies_relative(delay));
3541 }
3542 
3543 static int ceph_oloc_decode(void **p, void *end,
3544 			    struct ceph_object_locator *oloc)
3545 {
3546 	u8 struct_v, struct_cv;
3547 	u32 len;
3548 	void *struct_end;
3549 	int ret = 0;
3550 
3551 	ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
3552 	struct_v = ceph_decode_8(p);
3553 	struct_cv = ceph_decode_8(p);
3554 	if (struct_v < 3) {
3555 		pr_warn("got v %d < 3 cv %d of ceph_object_locator\n",
3556 			struct_v, struct_cv);
3557 		goto e_inval;
3558 	}
3559 	if (struct_cv > 6) {
3560 		pr_warn("got v %d cv %d > 6 of ceph_object_locator\n",
3561 			struct_v, struct_cv);
3562 		goto e_inval;
3563 	}
3564 	len = ceph_decode_32(p);
3565 	ceph_decode_need(p, end, len, e_inval);
3566 	struct_end = *p + len;
3567 
3568 	oloc->pool = ceph_decode_64(p);
3569 	*p += 4; /* skip preferred */
3570 
3571 	len = ceph_decode_32(p);
3572 	if (len > 0) {
3573 		pr_warn("ceph_object_locator::key is set\n");
3574 		goto e_inval;
3575 	}
3576 
3577 	if (struct_v >= 5) {
3578 		bool changed = false;
3579 
3580 		len = ceph_decode_32(p);
3581 		if (len > 0) {
3582 			ceph_decode_need(p, end, len, e_inval);
3583 			if (!oloc->pool_ns ||
3584 			    ceph_compare_string(oloc->pool_ns, *p, len))
3585 				changed = true;
3586 			*p += len;
3587 		} else {
3588 			if (oloc->pool_ns)
3589 				changed = true;
3590 		}
3591 		if (changed) {
3592 			/* redirect changes namespace */
3593 			pr_warn("ceph_object_locator::nspace is changed\n");
3594 			goto e_inval;
3595 		}
3596 	}
3597 
3598 	if (struct_v >= 6) {
3599 		s64 hash = ceph_decode_64(p);
3600 		if (hash != -1) {
3601 			pr_warn("ceph_object_locator::hash is set\n");
3602 			goto e_inval;
3603 		}
3604 	}
3605 
3606 	/* skip the rest */
3607 	*p = struct_end;
3608 out:
3609 	return ret;
3610 
3611 e_inval:
3612 	ret = -EINVAL;
3613 	goto out;
3614 }
3615 
3616 static int ceph_redirect_decode(void **p, void *end,
3617 				struct ceph_request_redirect *redir)
3618 {
3619 	u8 struct_v, struct_cv;
3620 	u32 len;
3621 	void *struct_end;
3622 	int ret;
3623 
3624 	ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
3625 	struct_v = ceph_decode_8(p);
3626 	struct_cv = ceph_decode_8(p);
3627 	if (struct_cv > 1) {
3628 		pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n",
3629 			struct_v, struct_cv);
3630 		goto e_inval;
3631 	}
3632 	len = ceph_decode_32(p);
3633 	ceph_decode_need(p, end, len, e_inval);
3634 	struct_end = *p + len;
3635 
3636 	ret = ceph_oloc_decode(p, end, &redir->oloc);
3637 	if (ret)
3638 		goto out;
3639 
3640 	len = ceph_decode_32(p);
3641 	if (len > 0) {
3642 		pr_warn("ceph_request_redirect::object_name is set\n");
3643 		goto e_inval;
3644 	}
3645 
3646 	/* skip the rest */
3647 	*p = struct_end;
3648 out:
3649 	return ret;
3650 
3651 e_inval:
3652 	ret = -EINVAL;
3653 	goto out;
3654 }
3655 
3656 struct MOSDOpReply {
3657 	struct ceph_pg pgid;
3658 	u64 flags;
3659 	int result;
3660 	u32 epoch;
3661 	int num_ops;
3662 	u32 outdata_len[CEPH_OSD_MAX_OPS];
3663 	s32 rval[CEPH_OSD_MAX_OPS];
3664 	int retry_attempt;
3665 	struct ceph_eversion replay_version;
3666 	u64 user_version;
3667 	struct ceph_request_redirect redirect;
3668 };
3669 
3670 static int decode_MOSDOpReply(const struct ceph_msg *msg, struct MOSDOpReply *m)
3671 {
3672 	void *p = msg->front.iov_base;
3673 	void *const end = p + msg->front.iov_len;
3674 	u16 version = le16_to_cpu(msg->hdr.version);
3675 	struct ceph_eversion bad_replay_version;
3676 	u8 decode_redir;
3677 	u32 len;
3678 	int ret;
3679 	int i;
3680 
3681 	ceph_decode_32_safe(&p, end, len, e_inval);
3682 	ceph_decode_need(&p, end, len, e_inval);
3683 	p += len; /* skip oid */
3684 
3685 	ret = ceph_decode_pgid(&p, end, &m->pgid);
3686 	if (ret)
3687 		return ret;
3688 
3689 	ceph_decode_64_safe(&p, end, m->flags, e_inval);
3690 	ceph_decode_32_safe(&p, end, m->result, e_inval);
3691 	ceph_decode_need(&p, end, sizeof(bad_replay_version), e_inval);
3692 	memcpy(&bad_replay_version, p, sizeof(bad_replay_version));
3693 	p += sizeof(bad_replay_version);
3694 	ceph_decode_32_safe(&p, end, m->epoch, e_inval);
3695 
3696 	ceph_decode_32_safe(&p, end, m->num_ops, e_inval);
3697 	if (m->num_ops > ARRAY_SIZE(m->outdata_len))
3698 		goto e_inval;
3699 
3700 	ceph_decode_need(&p, end, m->num_ops * sizeof(struct ceph_osd_op),
3701 			 e_inval);
3702 	for (i = 0; i < m->num_ops; i++) {
3703 		struct ceph_osd_op *op = p;
3704 
3705 		m->outdata_len[i] = le32_to_cpu(op->payload_len);
3706 		p += sizeof(*op);
3707 	}
3708 
3709 	ceph_decode_32_safe(&p, end, m->retry_attempt, e_inval);
3710 	for (i = 0; i < m->num_ops; i++)
3711 		ceph_decode_32_safe(&p, end, m->rval[i], e_inval);
3712 
3713 	if (version >= 5) {
3714 		ceph_decode_need(&p, end, sizeof(m->replay_version), e_inval);
3715 		memcpy(&m->replay_version, p, sizeof(m->replay_version));
3716 		p += sizeof(m->replay_version);
3717 		ceph_decode_64_safe(&p, end, m->user_version, e_inval);
3718 	} else {
3719 		m->replay_version = bad_replay_version; /* struct */
3720 		m->user_version = le64_to_cpu(m->replay_version.version);
3721 	}
3722 
3723 	if (version >= 6) {
3724 		if (version >= 7)
3725 			ceph_decode_8_safe(&p, end, decode_redir, e_inval);
3726 		else
3727 			decode_redir = 1;
3728 	} else {
3729 		decode_redir = 0;
3730 	}
3731 
3732 	if (decode_redir) {
3733 		ret = ceph_redirect_decode(&p, end, &m->redirect);
3734 		if (ret)
3735 			return ret;
3736 	} else {
3737 		ceph_oloc_init(&m->redirect.oloc);
3738 	}
3739 
3740 	return 0;
3741 
3742 e_inval:
3743 	return -EINVAL;
3744 }
3745 
3746 /*
3747  * Handle MOSDOpReply.  Set ->r_result and call the callback if it is
3748  * specified.
3749  */
3750 static void handle_reply(struct ceph_osd *osd, struct ceph_msg *msg)
3751 {
3752 	struct ceph_osd_client *osdc = osd->o_osdc;
3753 	struct ceph_osd_request *req;
3754 	struct MOSDOpReply m;
3755 	u64 tid = le64_to_cpu(msg->hdr.tid);
3756 	u32 data_len = 0;
3757 	int ret;
3758 	int i;
3759 
3760 	dout("%s msg %p tid %llu\n", __func__, msg, tid);
3761 
3762 	down_read(&osdc->lock);
3763 	if (!osd_registered(osd)) {
3764 		dout("%s osd%d unknown\n", __func__, osd->o_osd);
3765 		goto out_unlock_osdc;
3766 	}
3767 	WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num));
3768 
3769 	mutex_lock(&osd->lock);
3770 	req = lookup_request(&osd->o_requests, tid);
3771 	if (!req) {
3772 		dout("%s osd%d tid %llu unknown\n", __func__, osd->o_osd, tid);
3773 		goto out_unlock_session;
3774 	}
3775 
3776 	m.redirect.oloc.pool_ns = req->r_t.target_oloc.pool_ns;
3777 	ret = decode_MOSDOpReply(msg, &m);
3778 	m.redirect.oloc.pool_ns = NULL;
3779 	if (ret) {
3780 		pr_err("failed to decode MOSDOpReply for tid %llu: %d\n",
3781 		       req->r_tid, ret);
3782 		ceph_msg_dump(msg);
3783 		goto fail_request;
3784 	}
3785 	dout("%s req %p tid %llu flags 0x%llx pgid %llu.%x epoch %u attempt %d v %u'%llu uv %llu\n",
3786 	     __func__, req, req->r_tid, m.flags, m.pgid.pool, m.pgid.seed,
3787 	     m.epoch, m.retry_attempt, le32_to_cpu(m.replay_version.epoch),
3788 	     le64_to_cpu(m.replay_version.version), m.user_version);
3789 
3790 	if (m.retry_attempt >= 0) {
3791 		if (m.retry_attempt != req->r_attempts - 1) {
3792 			dout("req %p tid %llu retry_attempt %d != %d, ignoring\n",
3793 			     req, req->r_tid, m.retry_attempt,
3794 			     req->r_attempts - 1);
3795 			goto out_unlock_session;
3796 		}
3797 	} else {
3798 		WARN_ON(1); /* MOSDOpReply v4 is assumed */
3799 	}
3800 
3801 	if (!ceph_oloc_empty(&m.redirect.oloc)) {
3802 		dout("req %p tid %llu redirect pool %lld\n", req, req->r_tid,
3803 		     m.redirect.oloc.pool);
3804 		unlink_request(osd, req);
3805 		mutex_unlock(&osd->lock);
3806 
3807 		/*
3808 		 * Not ceph_oloc_copy() - changing pool_ns is not
3809 		 * supported.
3810 		 */
3811 		req->r_t.target_oloc.pool = m.redirect.oloc.pool;
3812 		req->r_flags |= CEPH_OSD_FLAG_REDIRECTED |
3813 				CEPH_OSD_FLAG_IGNORE_OVERLAY |
3814 				CEPH_OSD_FLAG_IGNORE_CACHE;
3815 		req->r_tid = 0;
3816 		__submit_request(req, false);
3817 		goto out_unlock_osdc;
3818 	}
3819 
3820 	if (m.result == -EAGAIN) {
3821 		dout("req %p tid %llu EAGAIN\n", req, req->r_tid);
3822 		unlink_request(osd, req);
3823 		mutex_unlock(&osd->lock);
3824 
3825 		/*
3826 		 * The object is missing on the replica or not (yet)
3827 		 * readable.  Clear pgid to force a resend to the primary
3828 		 * via legacy_change.
3829 		 */
3830 		req->r_t.pgid.pool = 0;
3831 		req->r_t.pgid.seed = 0;
3832 		WARN_ON(!req->r_t.used_replica);
3833 		req->r_flags &= ~(CEPH_OSD_FLAG_BALANCE_READS |
3834 				  CEPH_OSD_FLAG_LOCALIZE_READS);
3835 		req->r_tid = 0;
3836 		__submit_request(req, false);
3837 		goto out_unlock_osdc;
3838 	}
3839 
3840 	if (m.num_ops != req->r_num_ops) {
3841 		pr_err("num_ops %d != %d for tid %llu\n", m.num_ops,
3842 		       req->r_num_ops, req->r_tid);
3843 		goto fail_request;
3844 	}
3845 	for (i = 0; i < req->r_num_ops; i++) {
3846 		dout(" req %p tid %llu op %d rval %d len %u\n", req,
3847 		     req->r_tid, i, m.rval[i], m.outdata_len[i]);
3848 		req->r_ops[i].rval = m.rval[i];
3849 		req->r_ops[i].outdata_len = m.outdata_len[i];
3850 		data_len += m.outdata_len[i];
3851 	}
3852 	if (data_len != le32_to_cpu(msg->hdr.data_len)) {
3853 		pr_err("sum of lens %u != %u for tid %llu\n", data_len,
3854 		       le32_to_cpu(msg->hdr.data_len), req->r_tid);
3855 		goto fail_request;
3856 	}
3857 	dout("%s req %p tid %llu result %d data_len %u\n", __func__,
3858 	     req, req->r_tid, m.result, data_len);
3859 
3860 	/*
3861 	 * Since we only ever request ONDISK, we should only ever get
3862 	 * one (type of) reply back.
3863 	 */
3864 	WARN_ON(!(m.flags & CEPH_OSD_FLAG_ONDISK));
3865 	req->r_version = m.user_version;
3866 	req->r_result = m.result ?: data_len;
3867 	finish_request(req);
3868 	mutex_unlock(&osd->lock);
3869 	up_read(&osdc->lock);
3870 
3871 	__complete_request(req);
3872 	return;
3873 
3874 fail_request:
3875 	complete_request(req, -EIO);
3876 out_unlock_session:
3877 	mutex_unlock(&osd->lock);
3878 out_unlock_osdc:
3879 	up_read(&osdc->lock);
3880 }
3881 
3882 static void set_pool_was_full(struct ceph_osd_client *osdc)
3883 {
3884 	struct rb_node *n;
3885 
3886 	for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) {
3887 		struct ceph_pg_pool_info *pi =
3888 		    rb_entry(n, struct ceph_pg_pool_info, node);
3889 
3890 		pi->was_full = __pool_full(pi);
3891 	}
3892 }
3893 
3894 static bool pool_cleared_full(struct ceph_osd_client *osdc, s64 pool_id)
3895 {
3896 	struct ceph_pg_pool_info *pi;
3897 
3898 	pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id);
3899 	if (!pi)
3900 		return false;
3901 
3902 	return pi->was_full && !__pool_full(pi);
3903 }
3904 
3905 static enum calc_target_result
3906 recalc_linger_target(struct ceph_osd_linger_request *lreq)
3907 {
3908 	struct ceph_osd_client *osdc = lreq->osdc;
3909 	enum calc_target_result ct_res;
3910 
3911 	ct_res = calc_target(osdc, &lreq->t, true);
3912 	if (ct_res == CALC_TARGET_NEED_RESEND) {
3913 		struct ceph_osd *osd;
3914 
3915 		osd = lookup_create_osd(osdc, lreq->t.osd, true);
3916 		if (osd != lreq->osd) {
3917 			unlink_linger(lreq->osd, lreq);
3918 			link_linger(osd, lreq);
3919 		}
3920 	}
3921 
3922 	return ct_res;
3923 }
3924 
3925 /*
3926  * Requeue requests whose mapping to an OSD has changed.
3927  */
3928 static void scan_requests(struct ceph_osd *osd,
3929 			  bool force_resend,
3930 			  bool cleared_full,
3931 			  bool check_pool_cleared_full,
3932 			  struct rb_root *need_resend,
3933 			  struct list_head *need_resend_linger)
3934 {
3935 	struct ceph_osd_client *osdc = osd->o_osdc;
3936 	struct rb_node *n;
3937 	bool force_resend_writes;
3938 
3939 	for (n = rb_first(&osd->o_linger_requests); n; ) {
3940 		struct ceph_osd_linger_request *lreq =
3941 		    rb_entry(n, struct ceph_osd_linger_request, node);
3942 		enum calc_target_result ct_res;
3943 
3944 		n = rb_next(n); /* recalc_linger_target() */
3945 
3946 		dout("%s lreq %p linger_id %llu\n", __func__, lreq,
3947 		     lreq->linger_id);
3948 		ct_res = recalc_linger_target(lreq);
3949 		switch (ct_res) {
3950 		case CALC_TARGET_NO_ACTION:
3951 			force_resend_writes = cleared_full ||
3952 			    (check_pool_cleared_full &&
3953 			     pool_cleared_full(osdc, lreq->t.base_oloc.pool));
3954 			if (!force_resend && !force_resend_writes)
3955 				break;
3956 
3957 			fallthrough;
3958 		case CALC_TARGET_NEED_RESEND:
3959 			cancel_linger_map_check(lreq);
3960 			/*
3961 			 * scan_requests() for the previous epoch(s)
3962 			 * may have already added it to the list, since
3963 			 * it's not unlinked here.
3964 			 */
3965 			if (list_empty(&lreq->scan_item))
3966 				list_add_tail(&lreq->scan_item, need_resend_linger);
3967 			break;
3968 		case CALC_TARGET_POOL_DNE:
3969 			list_del_init(&lreq->scan_item);
3970 			check_linger_pool_dne(lreq);
3971 			break;
3972 		}
3973 	}
3974 
3975 	for (n = rb_first(&osd->o_requests); n; ) {
3976 		struct ceph_osd_request *req =
3977 		    rb_entry(n, struct ceph_osd_request, r_node);
3978 		enum calc_target_result ct_res;
3979 
3980 		n = rb_next(n); /* unlink_request(), check_pool_dne() */
3981 
3982 		dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
3983 		ct_res = calc_target(osdc, &req->r_t, false);
3984 		switch (ct_res) {
3985 		case CALC_TARGET_NO_ACTION:
3986 			force_resend_writes = cleared_full ||
3987 			    (check_pool_cleared_full &&
3988 			     pool_cleared_full(osdc, req->r_t.base_oloc.pool));
3989 			if (!force_resend &&
3990 			    (!(req->r_flags & CEPH_OSD_FLAG_WRITE) ||
3991 			     !force_resend_writes))
3992 				break;
3993 
3994 			fallthrough;
3995 		case CALC_TARGET_NEED_RESEND:
3996 			cancel_map_check(req);
3997 			unlink_request(osd, req);
3998 			insert_request(need_resend, req);
3999 			break;
4000 		case CALC_TARGET_POOL_DNE:
4001 			check_pool_dne(req);
4002 			break;
4003 		}
4004 	}
4005 }
4006 
4007 static int handle_one_map(struct ceph_osd_client *osdc,
4008 			  void *p, void *end, bool incremental,
4009 			  struct rb_root *need_resend,
4010 			  struct list_head *need_resend_linger)
4011 {
4012 	struct ceph_osdmap *newmap;
4013 	struct rb_node *n;
4014 	bool skipped_map = false;
4015 	bool was_full;
4016 
4017 	was_full = ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL);
4018 	set_pool_was_full(osdc);
4019 
4020 	if (incremental)
4021 		newmap = osdmap_apply_incremental(&p, end,
4022 						  ceph_msgr2(osdc->client),
4023 						  osdc->osdmap);
4024 	else
4025 		newmap = ceph_osdmap_decode(&p, end, ceph_msgr2(osdc->client));
4026 	if (IS_ERR(newmap))
4027 		return PTR_ERR(newmap);
4028 
4029 	if (newmap != osdc->osdmap) {
4030 		/*
4031 		 * Preserve ->was_full before destroying the old map.
4032 		 * For pools that weren't in the old map, ->was_full
4033 		 * should be false.
4034 		 */
4035 		for (n = rb_first(&newmap->pg_pools); n; n = rb_next(n)) {
4036 			struct ceph_pg_pool_info *pi =
4037 			    rb_entry(n, struct ceph_pg_pool_info, node);
4038 			struct ceph_pg_pool_info *old_pi;
4039 
4040 			old_pi = ceph_pg_pool_by_id(osdc->osdmap, pi->id);
4041 			if (old_pi)
4042 				pi->was_full = old_pi->was_full;
4043 			else
4044 				WARN_ON(pi->was_full);
4045 		}
4046 
4047 		if (osdc->osdmap->epoch &&
4048 		    osdc->osdmap->epoch + 1 < newmap->epoch) {
4049 			WARN_ON(incremental);
4050 			skipped_map = true;
4051 		}
4052 
4053 		ceph_osdmap_destroy(osdc->osdmap);
4054 		osdc->osdmap = newmap;
4055 	}
4056 
4057 	was_full &= !ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL);
4058 	scan_requests(&osdc->homeless_osd, skipped_map, was_full, true,
4059 		      need_resend, need_resend_linger);
4060 
4061 	for (n = rb_first(&osdc->osds); n; ) {
4062 		struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
4063 
4064 		n = rb_next(n); /* close_osd() */
4065 
4066 		scan_requests(osd, skipped_map, was_full, true, need_resend,
4067 			      need_resend_linger);
4068 		if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
4069 		    memcmp(&osd->o_con.peer_addr,
4070 			   ceph_osd_addr(osdc->osdmap, osd->o_osd),
4071 			   sizeof(struct ceph_entity_addr)))
4072 			close_osd(osd);
4073 	}
4074 
4075 	return 0;
4076 }
4077 
4078 static void kick_requests(struct ceph_osd_client *osdc,
4079 			  struct rb_root *need_resend,
4080 			  struct list_head *need_resend_linger)
4081 {
4082 	struct ceph_osd_linger_request *lreq, *nlreq;
4083 	enum calc_target_result ct_res;
4084 	struct rb_node *n;
4085 
4086 	/* make sure need_resend targets reflect latest map */
4087 	for (n = rb_first(need_resend); n; ) {
4088 		struct ceph_osd_request *req =
4089 		    rb_entry(n, struct ceph_osd_request, r_node);
4090 
4091 		n = rb_next(n);
4092 
4093 		if (req->r_t.epoch < osdc->osdmap->epoch) {
4094 			ct_res = calc_target(osdc, &req->r_t, false);
4095 			if (ct_res == CALC_TARGET_POOL_DNE) {
4096 				erase_request(need_resend, req);
4097 				check_pool_dne(req);
4098 			}
4099 		}
4100 	}
4101 
4102 	for (n = rb_first(need_resend); n; ) {
4103 		struct ceph_osd_request *req =
4104 		    rb_entry(n, struct ceph_osd_request, r_node);
4105 		struct ceph_osd *osd;
4106 
4107 		n = rb_next(n);
4108 		erase_request(need_resend, req); /* before link_request() */
4109 
4110 		osd = lookup_create_osd(osdc, req->r_t.osd, true);
4111 		link_request(osd, req);
4112 		if (!req->r_linger) {
4113 			if (!osd_homeless(osd) && !req->r_t.paused)
4114 				send_request(req);
4115 		} else {
4116 			cancel_linger_request(req);
4117 		}
4118 	}
4119 
4120 	list_for_each_entry_safe(lreq, nlreq, need_resend_linger, scan_item) {
4121 		if (!osd_homeless(lreq->osd))
4122 			send_linger(lreq);
4123 
4124 		list_del_init(&lreq->scan_item);
4125 	}
4126 }
4127 
4128 /*
4129  * Process updated osd map.
4130  *
4131  * The message contains any number of incremental and full maps, normally
4132  * indicating some sort of topology change in the cluster.  Kick requests
4133  * off to different OSDs as needed.
4134  */
4135 void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
4136 {
4137 	void *p = msg->front.iov_base;
4138 	void *const end = p + msg->front.iov_len;
4139 	u32 nr_maps, maplen;
4140 	u32 epoch;
4141 	struct ceph_fsid fsid;
4142 	struct rb_root need_resend = RB_ROOT;
4143 	LIST_HEAD(need_resend_linger);
4144 	bool handled_incremental = false;
4145 	bool was_pauserd, was_pausewr;
4146 	bool pauserd, pausewr;
4147 	int err;
4148 
4149 	dout("%s have %u\n", __func__, osdc->osdmap->epoch);
4150 	down_write(&osdc->lock);
4151 
4152 	/* verify fsid */
4153 	ceph_decode_need(&p, end, sizeof(fsid), bad);
4154 	ceph_decode_copy(&p, &fsid, sizeof(fsid));
4155 	if (ceph_check_fsid(osdc->client, &fsid) < 0)
4156 		goto bad;
4157 
4158 	was_pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
4159 	was_pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
4160 		      ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
4161 		      have_pool_full(osdc);
4162 
4163 	/* incremental maps */
4164 	ceph_decode_32_safe(&p, end, nr_maps, bad);
4165 	dout(" %d inc maps\n", nr_maps);
4166 	while (nr_maps > 0) {
4167 		ceph_decode_need(&p, end, 2*sizeof(u32), bad);
4168 		epoch = ceph_decode_32(&p);
4169 		maplen = ceph_decode_32(&p);
4170 		ceph_decode_need(&p, end, maplen, bad);
4171 		if (osdc->osdmap->epoch &&
4172 		    osdc->osdmap->epoch + 1 == epoch) {
4173 			dout("applying incremental map %u len %d\n",
4174 			     epoch, maplen);
4175 			err = handle_one_map(osdc, p, p + maplen, true,
4176 					     &need_resend, &need_resend_linger);
4177 			if (err)
4178 				goto bad;
4179 			handled_incremental = true;
4180 		} else {
4181 			dout("ignoring incremental map %u len %d\n",
4182 			     epoch, maplen);
4183 		}
4184 		p += maplen;
4185 		nr_maps--;
4186 	}
4187 	if (handled_incremental)
4188 		goto done;
4189 
4190 	/* full maps */
4191 	ceph_decode_32_safe(&p, end, nr_maps, bad);
4192 	dout(" %d full maps\n", nr_maps);
4193 	while (nr_maps) {
4194 		ceph_decode_need(&p, end, 2*sizeof(u32), bad);
4195 		epoch = ceph_decode_32(&p);
4196 		maplen = ceph_decode_32(&p);
4197 		ceph_decode_need(&p, end, maplen, bad);
4198 		if (nr_maps > 1) {
4199 			dout("skipping non-latest full map %u len %d\n",
4200 			     epoch, maplen);
4201 		} else if (osdc->osdmap->epoch >= epoch) {
4202 			dout("skipping full map %u len %d, "
4203 			     "older than our %u\n", epoch, maplen,
4204 			     osdc->osdmap->epoch);
4205 		} else {
4206 			dout("taking full map %u len %d\n", epoch, maplen);
4207 			err = handle_one_map(osdc, p, p + maplen, false,
4208 					     &need_resend, &need_resend_linger);
4209 			if (err)
4210 				goto bad;
4211 		}
4212 		p += maplen;
4213 		nr_maps--;
4214 	}
4215 
4216 done:
4217 	/*
4218 	 * subscribe to subsequent osdmap updates if full to ensure
4219 	 * we find out when we are no longer full and stop returning
4220 	 * ENOSPC.
4221 	 */
4222 	pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
4223 	pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
4224 		  ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
4225 		  have_pool_full(osdc);
4226 	if (was_pauserd || was_pausewr || pauserd || pausewr ||
4227 	    osdc->osdmap->epoch < osdc->epoch_barrier)
4228 		maybe_request_map(osdc);
4229 
4230 	kick_requests(osdc, &need_resend, &need_resend_linger);
4231 
4232 	ceph_osdc_abort_on_full(osdc);
4233 	ceph_monc_got_map(&osdc->client->monc, CEPH_SUB_OSDMAP,
4234 			  osdc->osdmap->epoch);
4235 	up_write(&osdc->lock);
4236 	wake_up_all(&osdc->client->auth_wq);
4237 	return;
4238 
4239 bad:
4240 	pr_err("osdc handle_map corrupt msg\n");
4241 	ceph_msg_dump(msg);
4242 	up_write(&osdc->lock);
4243 }
4244 
4245 /*
4246  * Resubmit requests pending on the given osd.
4247  */
4248 static void kick_osd_requests(struct ceph_osd *osd)
4249 {
4250 	struct rb_node *n;
4251 
4252 	clear_backoffs(osd);
4253 
4254 	for (n = rb_first(&osd->o_requests); n; ) {
4255 		struct ceph_osd_request *req =
4256 		    rb_entry(n, struct ceph_osd_request, r_node);
4257 
4258 		n = rb_next(n); /* cancel_linger_request() */
4259 
4260 		if (!req->r_linger) {
4261 			if (!req->r_t.paused)
4262 				send_request(req);
4263 		} else {
4264 			cancel_linger_request(req);
4265 		}
4266 	}
4267 	for (n = rb_first(&osd->o_linger_requests); n; n = rb_next(n)) {
4268 		struct ceph_osd_linger_request *lreq =
4269 		    rb_entry(n, struct ceph_osd_linger_request, node);
4270 
4271 		send_linger(lreq);
4272 	}
4273 }
4274 
4275 /*
4276  * If the osd connection drops, we need to resubmit all requests.
4277  */
4278 static void osd_fault(struct ceph_connection *con)
4279 {
4280 	struct ceph_osd *osd = con->private;
4281 	struct ceph_osd_client *osdc = osd->o_osdc;
4282 
4283 	dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
4284 
4285 	down_write(&osdc->lock);
4286 	if (!osd_registered(osd)) {
4287 		dout("%s osd%d unknown\n", __func__, osd->o_osd);
4288 		goto out_unlock;
4289 	}
4290 
4291 	osd->o_sparse_op_idx = -1;
4292 	ceph_init_sparse_read(&osd->o_sparse_read);
4293 
4294 	if (!reopen_osd(osd))
4295 		kick_osd_requests(osd);
4296 	maybe_request_map(osdc);
4297 
4298 out_unlock:
4299 	up_write(&osdc->lock);
4300 }
4301 
4302 struct MOSDBackoff {
4303 	struct ceph_spg spgid;
4304 	u32 map_epoch;
4305 	u8 op;
4306 	u64 id;
4307 	struct ceph_hobject_id *begin;
4308 	struct ceph_hobject_id *end;
4309 };
4310 
4311 static int decode_MOSDBackoff(const struct ceph_msg *msg, struct MOSDBackoff *m)
4312 {
4313 	void *p = msg->front.iov_base;
4314 	void *const end = p + msg->front.iov_len;
4315 	u8 struct_v;
4316 	u32 struct_len;
4317 	int ret;
4318 
4319 	ret = ceph_start_decoding(&p, end, 1, "spg_t", &struct_v, &struct_len);
4320 	if (ret)
4321 		return ret;
4322 
4323 	ret = ceph_decode_pgid(&p, end, &m->spgid.pgid);
4324 	if (ret)
4325 		return ret;
4326 
4327 	ceph_decode_8_safe(&p, end, m->spgid.shard, e_inval);
4328 	ceph_decode_32_safe(&p, end, m->map_epoch, e_inval);
4329 	ceph_decode_8_safe(&p, end, m->op, e_inval);
4330 	ceph_decode_64_safe(&p, end, m->id, e_inval);
4331 
4332 	m->begin = kzalloc(sizeof(*m->begin), GFP_NOIO);
4333 	if (!m->begin)
4334 		return -ENOMEM;
4335 
4336 	ret = decode_hoid(&p, end, m->begin);
4337 	if (ret) {
4338 		free_hoid(m->begin);
4339 		return ret;
4340 	}
4341 
4342 	m->end = kzalloc(sizeof(*m->end), GFP_NOIO);
4343 	if (!m->end) {
4344 		free_hoid(m->begin);
4345 		return -ENOMEM;
4346 	}
4347 
4348 	ret = decode_hoid(&p, end, m->end);
4349 	if (ret) {
4350 		free_hoid(m->begin);
4351 		free_hoid(m->end);
4352 		return ret;
4353 	}
4354 
4355 	return 0;
4356 
4357 e_inval:
4358 	return -EINVAL;
4359 }
4360 
4361 static struct ceph_msg *create_backoff_message(
4362 				const struct ceph_osd_backoff *backoff,
4363 				u32 map_epoch)
4364 {
4365 	struct ceph_msg *msg;
4366 	void *p, *end;
4367 	int msg_size;
4368 
4369 	msg_size = CEPH_ENCODING_START_BLK_LEN +
4370 			CEPH_PGID_ENCODING_LEN + 1; /* spgid */
4371 	msg_size += 4 + 1 + 8; /* map_epoch, op, id */
4372 	msg_size += CEPH_ENCODING_START_BLK_LEN +
4373 			hoid_encoding_size(backoff->begin);
4374 	msg_size += CEPH_ENCODING_START_BLK_LEN +
4375 			hoid_encoding_size(backoff->end);
4376 
4377 	msg = ceph_msg_new(CEPH_MSG_OSD_BACKOFF, msg_size, GFP_NOIO, true);
4378 	if (!msg)
4379 		return NULL;
4380 
4381 	p = msg->front.iov_base;
4382 	end = p + msg->front_alloc_len;
4383 
4384 	encode_spgid(&p, &backoff->spgid);
4385 	ceph_encode_32(&p, map_epoch);
4386 	ceph_encode_8(&p, CEPH_OSD_BACKOFF_OP_ACK_BLOCK);
4387 	ceph_encode_64(&p, backoff->id);
4388 	encode_hoid(&p, end, backoff->begin);
4389 	encode_hoid(&p, end, backoff->end);
4390 	BUG_ON(p != end);
4391 
4392 	msg->front.iov_len = p - msg->front.iov_base;
4393 	msg->hdr.version = cpu_to_le16(1); /* MOSDBackoff v1 */
4394 	msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
4395 
4396 	return msg;
4397 }
4398 
4399 static void handle_backoff_block(struct ceph_osd *osd, struct MOSDBackoff *m)
4400 {
4401 	struct ceph_spg_mapping *spg;
4402 	struct ceph_osd_backoff *backoff;
4403 	struct ceph_msg *msg;
4404 
4405 	dout("%s osd%d spgid %llu.%xs%d id %llu\n", __func__, osd->o_osd,
4406 	     m->spgid.pgid.pool, m->spgid.pgid.seed, m->spgid.shard, m->id);
4407 
4408 	spg = lookup_spg_mapping(&osd->o_backoff_mappings, &m->spgid);
4409 	if (!spg) {
4410 		spg = alloc_spg_mapping();
4411 		if (!spg) {
4412 			pr_err("%s failed to allocate spg\n", __func__);
4413 			return;
4414 		}
4415 		spg->spgid = m->spgid; /* struct */
4416 		insert_spg_mapping(&osd->o_backoff_mappings, spg);
4417 	}
4418 
4419 	backoff = alloc_backoff();
4420 	if (!backoff) {
4421 		pr_err("%s failed to allocate backoff\n", __func__);
4422 		return;
4423 	}
4424 	backoff->spgid = m->spgid; /* struct */
4425 	backoff->id = m->id;
4426 	backoff->begin = m->begin;
4427 	m->begin = NULL; /* backoff now owns this */
4428 	backoff->end = m->end;
4429 	m->end = NULL;   /* ditto */
4430 
4431 	insert_backoff(&spg->backoffs, backoff);
4432 	insert_backoff_by_id(&osd->o_backoffs_by_id, backoff);
4433 
4434 	/*
4435 	 * Ack with original backoff's epoch so that the OSD can
4436 	 * discard this if there was a PG split.
4437 	 */
4438 	msg = create_backoff_message(backoff, m->map_epoch);
4439 	if (!msg) {
4440 		pr_err("%s failed to allocate msg\n", __func__);
4441 		return;
4442 	}
4443 	ceph_con_send(&osd->o_con, msg);
4444 }
4445 
4446 static bool target_contained_by(const struct ceph_osd_request_target *t,
4447 				const struct ceph_hobject_id *begin,
4448 				const struct ceph_hobject_id *end)
4449 {
4450 	struct ceph_hobject_id hoid;
4451 	int cmp;
4452 
4453 	hoid_fill_from_target(&hoid, t);
4454 	cmp = hoid_compare(&hoid, begin);
4455 	return !cmp || (cmp > 0 && hoid_compare(&hoid, end) < 0);
4456 }
4457 
4458 static void handle_backoff_unblock(struct ceph_osd *osd,
4459 				   const struct MOSDBackoff *m)
4460 {
4461 	struct ceph_spg_mapping *spg;
4462 	struct ceph_osd_backoff *backoff;
4463 	struct rb_node *n;
4464 
4465 	dout("%s osd%d spgid %llu.%xs%d id %llu\n", __func__, osd->o_osd,
4466 	     m->spgid.pgid.pool, m->spgid.pgid.seed, m->spgid.shard, m->id);
4467 
4468 	backoff = lookup_backoff_by_id(&osd->o_backoffs_by_id, m->id);
4469 	if (!backoff) {
4470 		pr_err("%s osd%d spgid %llu.%xs%d id %llu backoff dne\n",
4471 		       __func__, osd->o_osd, m->spgid.pgid.pool,
4472 		       m->spgid.pgid.seed, m->spgid.shard, m->id);
4473 		return;
4474 	}
4475 
4476 	if (hoid_compare(backoff->begin, m->begin) &&
4477 	    hoid_compare(backoff->end, m->end)) {
4478 		pr_err("%s osd%d spgid %llu.%xs%d id %llu bad range?\n",
4479 		       __func__, osd->o_osd, m->spgid.pgid.pool,
4480 		       m->spgid.pgid.seed, m->spgid.shard, m->id);
4481 		/* unblock it anyway... */
4482 	}
4483 
4484 	spg = lookup_spg_mapping(&osd->o_backoff_mappings, &backoff->spgid);
4485 	BUG_ON(!spg);
4486 
4487 	erase_backoff(&spg->backoffs, backoff);
4488 	erase_backoff_by_id(&osd->o_backoffs_by_id, backoff);
4489 	free_backoff(backoff);
4490 
4491 	if (RB_EMPTY_ROOT(&spg->backoffs)) {
4492 		erase_spg_mapping(&osd->o_backoff_mappings, spg);
4493 		free_spg_mapping(spg);
4494 	}
4495 
4496 	for (n = rb_first(&osd->o_requests); n; n = rb_next(n)) {
4497 		struct ceph_osd_request *req =
4498 		    rb_entry(n, struct ceph_osd_request, r_node);
4499 
4500 		if (!ceph_spg_compare(&req->r_t.spgid, &m->spgid)) {
4501 			/*
4502 			 * Match against @m, not @backoff -- the PG may
4503 			 * have split on the OSD.
4504 			 */
4505 			if (target_contained_by(&req->r_t, m->begin, m->end)) {
4506 				/*
4507 				 * If no other installed backoff applies,
4508 				 * resend.
4509 				 */
4510 				send_request(req);
4511 			}
4512 		}
4513 	}
4514 }
4515 
4516 static void handle_backoff(struct ceph_osd *osd, struct ceph_msg *msg)
4517 {
4518 	struct ceph_osd_client *osdc = osd->o_osdc;
4519 	struct MOSDBackoff m;
4520 	int ret;
4521 
4522 	down_read(&osdc->lock);
4523 	if (!osd_registered(osd)) {
4524 		dout("%s osd%d unknown\n", __func__, osd->o_osd);
4525 		up_read(&osdc->lock);
4526 		return;
4527 	}
4528 	WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num));
4529 
4530 	mutex_lock(&osd->lock);
4531 	ret = decode_MOSDBackoff(msg, &m);
4532 	if (ret) {
4533 		pr_err("failed to decode MOSDBackoff: %d\n", ret);
4534 		ceph_msg_dump(msg);
4535 		goto out_unlock;
4536 	}
4537 
4538 	switch (m.op) {
4539 	case CEPH_OSD_BACKOFF_OP_BLOCK:
4540 		handle_backoff_block(osd, &m);
4541 		break;
4542 	case CEPH_OSD_BACKOFF_OP_UNBLOCK:
4543 		handle_backoff_unblock(osd, &m);
4544 		break;
4545 	default:
4546 		pr_err("%s osd%d unknown op %d\n", __func__, osd->o_osd, m.op);
4547 	}
4548 
4549 	free_hoid(m.begin);
4550 	free_hoid(m.end);
4551 
4552 out_unlock:
4553 	mutex_unlock(&osd->lock);
4554 	up_read(&osdc->lock);
4555 }
4556 
4557 /*
4558  * Process osd watch notifications
4559  */
4560 static void handle_watch_notify(struct ceph_osd_client *osdc,
4561 				struct ceph_msg *msg)
4562 {
4563 	void *p = msg->front.iov_base;
4564 	void *const end = p + msg->front.iov_len;
4565 	struct ceph_osd_linger_request *lreq;
4566 	struct linger_work *lwork;
4567 	u8 proto_ver, opcode;
4568 	u64 cookie, notify_id;
4569 	u64 notifier_id = 0;
4570 	s32 return_code = 0;
4571 	void *payload = NULL;
4572 	u32 payload_len = 0;
4573 
4574 	ceph_decode_8_safe(&p, end, proto_ver, bad);
4575 	ceph_decode_8_safe(&p, end, opcode, bad);
4576 	ceph_decode_64_safe(&p, end, cookie, bad);
4577 	p += 8; /* skip ver */
4578 	ceph_decode_64_safe(&p, end, notify_id, bad);
4579 
4580 	if (proto_ver >= 1) {
4581 		ceph_decode_32_safe(&p, end, payload_len, bad);
4582 		ceph_decode_need(&p, end, payload_len, bad);
4583 		payload = p;
4584 		p += payload_len;
4585 	}
4586 
4587 	if (le16_to_cpu(msg->hdr.version) >= 2)
4588 		ceph_decode_32_safe(&p, end, return_code, bad);
4589 
4590 	if (le16_to_cpu(msg->hdr.version) >= 3)
4591 		ceph_decode_64_safe(&p, end, notifier_id, bad);
4592 
4593 	down_read(&osdc->lock);
4594 	lreq = lookup_linger_osdc(&osdc->linger_requests, cookie);
4595 	if (!lreq) {
4596 		dout("%s opcode %d cookie %llu dne\n", __func__, opcode,
4597 		     cookie);
4598 		goto out_unlock_osdc;
4599 	}
4600 
4601 	mutex_lock(&lreq->lock);
4602 	dout("%s opcode %d cookie %llu lreq %p is_watch %d\n", __func__,
4603 	     opcode, cookie, lreq, lreq->is_watch);
4604 	if (opcode == CEPH_WATCH_EVENT_DISCONNECT) {
4605 		if (!lreq->last_error) {
4606 			lreq->last_error = -ENOTCONN;
4607 			queue_watch_error(lreq);
4608 		}
4609 	} else if (!lreq->is_watch) {
4610 		/* CEPH_WATCH_EVENT_NOTIFY_COMPLETE */
4611 		if (lreq->notify_id && lreq->notify_id != notify_id) {
4612 			dout("lreq %p notify_id %llu != %llu, ignoring\n", lreq,
4613 			     lreq->notify_id, notify_id);
4614 		} else if (!completion_done(&lreq->notify_finish_wait)) {
4615 			struct ceph_msg_data *data =
4616 			    msg->num_data_items ? &msg->data[0] : NULL;
4617 
4618 			if (data) {
4619 				if (lreq->preply_pages) {
4620 					WARN_ON(data->type !=
4621 							CEPH_MSG_DATA_PAGES);
4622 					*lreq->preply_pages = data->pages;
4623 					*lreq->preply_len = data->length;
4624 					data->own_pages = false;
4625 				}
4626 			}
4627 			lreq->notify_finish_error = return_code;
4628 			complete_all(&lreq->notify_finish_wait);
4629 		}
4630 	} else {
4631 		/* CEPH_WATCH_EVENT_NOTIFY */
4632 		lwork = lwork_alloc(lreq, do_watch_notify);
4633 		if (!lwork) {
4634 			pr_err("failed to allocate notify-lwork\n");
4635 			goto out_unlock_lreq;
4636 		}
4637 
4638 		lwork->notify.notify_id = notify_id;
4639 		lwork->notify.notifier_id = notifier_id;
4640 		lwork->notify.payload = payload;
4641 		lwork->notify.payload_len = payload_len;
4642 		lwork->notify.msg = ceph_msg_get(msg);
4643 		lwork_queue(lwork);
4644 	}
4645 
4646 out_unlock_lreq:
4647 	mutex_unlock(&lreq->lock);
4648 out_unlock_osdc:
4649 	up_read(&osdc->lock);
4650 	return;
4651 
4652 bad:
4653 	pr_err("osdc handle_watch_notify corrupt msg\n");
4654 }
4655 
4656 /*
4657  * Register request, send initial attempt.
4658  */
4659 void ceph_osdc_start_request(struct ceph_osd_client *osdc,
4660 			     struct ceph_osd_request *req)
4661 {
4662 	down_read(&osdc->lock);
4663 	submit_request(req, false);
4664 	up_read(&osdc->lock);
4665 }
4666 EXPORT_SYMBOL(ceph_osdc_start_request);
4667 
4668 /*
4669  * Unregister request.  If @req was registered, it isn't completed:
4670  * r_result isn't set and __complete_request() isn't invoked.
4671  *
4672  * If @req wasn't registered, this call may have raced with
4673  * handle_reply(), in which case r_result would already be set and
4674  * __complete_request() would be getting invoked, possibly even
4675  * concurrently with this call.
4676  */
4677 void ceph_osdc_cancel_request(struct ceph_osd_request *req)
4678 {
4679 	struct ceph_osd_client *osdc = req->r_osdc;
4680 
4681 	down_write(&osdc->lock);
4682 	if (req->r_osd)
4683 		cancel_request(req);
4684 	up_write(&osdc->lock);
4685 }
4686 EXPORT_SYMBOL(ceph_osdc_cancel_request);
4687 
4688 /*
4689  * @timeout: in jiffies, 0 means "wait forever"
4690  */
4691 static int wait_request_timeout(struct ceph_osd_request *req,
4692 				unsigned long timeout)
4693 {
4694 	long left;
4695 
4696 	dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
4697 	left = wait_for_completion_killable_timeout(&req->r_completion,
4698 						ceph_timeout_jiffies(timeout));
4699 	if (left <= 0) {
4700 		left = left ?: -ETIMEDOUT;
4701 		ceph_osdc_cancel_request(req);
4702 	} else {
4703 		left = req->r_result; /* completed */
4704 	}
4705 
4706 	return left;
4707 }
4708 
4709 /*
4710  * wait for a request to complete
4711  */
4712 int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
4713 			   struct ceph_osd_request *req)
4714 {
4715 	return wait_request_timeout(req, 0);
4716 }
4717 EXPORT_SYMBOL(ceph_osdc_wait_request);
4718 
4719 /*
4720  * sync - wait for all in-flight requests to flush.  avoid starvation.
4721  */
4722 void ceph_osdc_sync(struct ceph_osd_client *osdc)
4723 {
4724 	struct rb_node *n, *p;
4725 	u64 last_tid = atomic64_read(&osdc->last_tid);
4726 
4727 again:
4728 	down_read(&osdc->lock);
4729 	for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
4730 		struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
4731 
4732 		mutex_lock(&osd->lock);
4733 		for (p = rb_first(&osd->o_requests); p; p = rb_next(p)) {
4734 			struct ceph_osd_request *req =
4735 			    rb_entry(p, struct ceph_osd_request, r_node);
4736 
4737 			if (req->r_tid > last_tid)
4738 				break;
4739 
4740 			if (!(req->r_flags & CEPH_OSD_FLAG_WRITE))
4741 				continue;
4742 
4743 			ceph_osdc_get_request(req);
4744 			mutex_unlock(&osd->lock);
4745 			up_read(&osdc->lock);
4746 			dout("%s waiting on req %p tid %llu last_tid %llu\n",
4747 			     __func__, req, req->r_tid, last_tid);
4748 			wait_for_completion(&req->r_completion);
4749 			ceph_osdc_put_request(req);
4750 			goto again;
4751 		}
4752 
4753 		mutex_unlock(&osd->lock);
4754 	}
4755 
4756 	up_read(&osdc->lock);
4757 	dout("%s done last_tid %llu\n", __func__, last_tid);
4758 }
4759 EXPORT_SYMBOL(ceph_osdc_sync);
4760 
4761 /*
4762  * Returns a handle, caller owns a ref.
4763  */
4764 struct ceph_osd_linger_request *
4765 ceph_osdc_watch(struct ceph_osd_client *osdc,
4766 		struct ceph_object_id *oid,
4767 		struct ceph_object_locator *oloc,
4768 		rados_watchcb2_t wcb,
4769 		rados_watcherrcb_t errcb,
4770 		void *data)
4771 {
4772 	struct ceph_osd_linger_request *lreq;
4773 	int ret;
4774 
4775 	lreq = linger_alloc(osdc);
4776 	if (!lreq)
4777 		return ERR_PTR(-ENOMEM);
4778 
4779 	lreq->is_watch = true;
4780 	lreq->wcb = wcb;
4781 	lreq->errcb = errcb;
4782 	lreq->data = data;
4783 	lreq->watch_valid_thru = jiffies;
4784 
4785 	ceph_oid_copy(&lreq->t.base_oid, oid);
4786 	ceph_oloc_copy(&lreq->t.base_oloc, oloc);
4787 	lreq->t.flags = CEPH_OSD_FLAG_WRITE;
4788 	ktime_get_real_ts64(&lreq->mtime);
4789 
4790 	linger_submit(lreq);
4791 	ret = linger_reg_commit_wait(lreq);
4792 	if (ret) {
4793 		linger_cancel(lreq);
4794 		goto err_put_lreq;
4795 	}
4796 
4797 	return lreq;
4798 
4799 err_put_lreq:
4800 	linger_put(lreq);
4801 	return ERR_PTR(ret);
4802 }
4803 EXPORT_SYMBOL(ceph_osdc_watch);
4804 
4805 /*
4806  * Releases a ref.
4807  *
4808  * Times out after mount_timeout to preserve rbd unmap behaviour
4809  * introduced in 2894e1d76974 ("rbd: timeout watch teardown on unmap
4810  * with mount_timeout").
4811  */
4812 int ceph_osdc_unwatch(struct ceph_osd_client *osdc,
4813 		      struct ceph_osd_linger_request *lreq)
4814 {
4815 	struct ceph_options *opts = osdc->client->options;
4816 	struct ceph_osd_request *req;
4817 	int ret;
4818 
4819 	req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
4820 	if (!req)
4821 		return -ENOMEM;
4822 
4823 	ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
4824 	ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
4825 	req->r_flags = CEPH_OSD_FLAG_WRITE;
4826 	ktime_get_real_ts64(&req->r_mtime);
4827 	osd_req_op_watch_init(req, 0, CEPH_OSD_WATCH_OP_UNWATCH,
4828 			      lreq->linger_id, 0);
4829 
4830 	ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
4831 	if (ret)
4832 		goto out_put_req;
4833 
4834 	ceph_osdc_start_request(osdc, req);
4835 	linger_cancel(lreq);
4836 	linger_put(lreq);
4837 	ret = wait_request_timeout(req, opts->mount_timeout);
4838 
4839 out_put_req:
4840 	ceph_osdc_put_request(req);
4841 	return ret;
4842 }
4843 EXPORT_SYMBOL(ceph_osdc_unwatch);
4844 
4845 static int osd_req_op_notify_ack_init(struct ceph_osd_request *req, int which,
4846 				      u64 notify_id, u64 cookie, void *payload,
4847 				      u32 payload_len)
4848 {
4849 	struct ceph_osd_req_op *op;
4850 	struct ceph_pagelist *pl;
4851 	int ret;
4852 
4853 	op = osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY_ACK, 0);
4854 
4855 	pl = ceph_pagelist_alloc(GFP_NOIO);
4856 	if (!pl)
4857 		return -ENOMEM;
4858 
4859 	ret = ceph_pagelist_encode_64(pl, notify_id);
4860 	ret |= ceph_pagelist_encode_64(pl, cookie);
4861 	if (payload) {
4862 		ret |= ceph_pagelist_encode_32(pl, payload_len);
4863 		ret |= ceph_pagelist_append(pl, payload, payload_len);
4864 	} else {
4865 		ret |= ceph_pagelist_encode_32(pl, 0);
4866 	}
4867 	if (ret) {
4868 		ceph_pagelist_release(pl);
4869 		return -ENOMEM;
4870 	}
4871 
4872 	ceph_osd_data_pagelist_init(&op->notify_ack.request_data, pl);
4873 	op->indata_len = pl->length;
4874 	return 0;
4875 }
4876 
4877 int ceph_osdc_notify_ack(struct ceph_osd_client *osdc,
4878 			 struct ceph_object_id *oid,
4879 			 struct ceph_object_locator *oloc,
4880 			 u64 notify_id,
4881 			 u64 cookie,
4882 			 void *payload,
4883 			 u32 payload_len)
4884 {
4885 	struct ceph_osd_request *req;
4886 	int ret;
4887 
4888 	req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
4889 	if (!req)
4890 		return -ENOMEM;
4891 
4892 	ceph_oid_copy(&req->r_base_oid, oid);
4893 	ceph_oloc_copy(&req->r_base_oloc, oloc);
4894 	req->r_flags = CEPH_OSD_FLAG_READ;
4895 
4896 	ret = osd_req_op_notify_ack_init(req, 0, notify_id, cookie, payload,
4897 					 payload_len);
4898 	if (ret)
4899 		goto out_put_req;
4900 
4901 	ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
4902 	if (ret)
4903 		goto out_put_req;
4904 
4905 	ceph_osdc_start_request(osdc, req);
4906 	ret = ceph_osdc_wait_request(osdc, req);
4907 
4908 out_put_req:
4909 	ceph_osdc_put_request(req);
4910 	return ret;
4911 }
4912 EXPORT_SYMBOL(ceph_osdc_notify_ack);
4913 
4914 /*
4915  * @timeout: in seconds
4916  *
4917  * @preply_{pages,len} are initialized both on success and error.
4918  * The caller is responsible for:
4919  *
4920  *     ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len))
4921  */
4922 int ceph_osdc_notify(struct ceph_osd_client *osdc,
4923 		     struct ceph_object_id *oid,
4924 		     struct ceph_object_locator *oloc,
4925 		     void *payload,
4926 		     u32 payload_len,
4927 		     u32 timeout,
4928 		     struct page ***preply_pages,
4929 		     size_t *preply_len)
4930 {
4931 	struct ceph_osd_linger_request *lreq;
4932 	int ret;
4933 
4934 	WARN_ON(!timeout);
4935 	if (preply_pages) {
4936 		*preply_pages = NULL;
4937 		*preply_len = 0;
4938 	}
4939 
4940 	lreq = linger_alloc(osdc);
4941 	if (!lreq)
4942 		return -ENOMEM;
4943 
4944 	lreq->request_pl = ceph_pagelist_alloc(GFP_NOIO);
4945 	if (!lreq->request_pl) {
4946 		ret = -ENOMEM;
4947 		goto out_put_lreq;
4948 	}
4949 
4950 	ret = ceph_pagelist_encode_32(lreq->request_pl, 1); /* prot_ver */
4951 	ret |= ceph_pagelist_encode_32(lreq->request_pl, timeout);
4952 	ret |= ceph_pagelist_encode_32(lreq->request_pl, payload_len);
4953 	ret |= ceph_pagelist_append(lreq->request_pl, payload, payload_len);
4954 	if (ret) {
4955 		ret = -ENOMEM;
4956 		goto out_put_lreq;
4957 	}
4958 
4959 	/* for notify_id */
4960 	lreq->notify_id_pages = ceph_alloc_page_vector(1, GFP_NOIO);
4961 	if (IS_ERR(lreq->notify_id_pages)) {
4962 		ret = PTR_ERR(lreq->notify_id_pages);
4963 		lreq->notify_id_pages = NULL;
4964 		goto out_put_lreq;
4965 	}
4966 
4967 	lreq->preply_pages = preply_pages;
4968 	lreq->preply_len = preply_len;
4969 
4970 	ceph_oid_copy(&lreq->t.base_oid, oid);
4971 	ceph_oloc_copy(&lreq->t.base_oloc, oloc);
4972 	lreq->t.flags = CEPH_OSD_FLAG_READ;
4973 
4974 	linger_submit(lreq);
4975 	ret = linger_reg_commit_wait(lreq);
4976 	if (!ret)
4977 		ret = linger_notify_finish_wait(lreq,
4978 				 msecs_to_jiffies(2 * timeout * MSEC_PER_SEC));
4979 	else
4980 		dout("lreq %p failed to initiate notify %d\n", lreq, ret);
4981 
4982 	linger_cancel(lreq);
4983 out_put_lreq:
4984 	linger_put(lreq);
4985 	return ret;
4986 }
4987 EXPORT_SYMBOL(ceph_osdc_notify);
4988 
4989 static int decode_watcher(void **p, void *end, struct ceph_watch_item *item)
4990 {
4991 	u8 struct_v;
4992 	u32 struct_len;
4993 	int ret;
4994 
4995 	ret = ceph_start_decoding(p, end, 2, "watch_item_t",
4996 				  &struct_v, &struct_len);
4997 	if (ret)
4998 		goto bad;
4999 
5000 	ret = -EINVAL;
5001 	ceph_decode_copy_safe(p, end, &item->name, sizeof(item->name), bad);
5002 	ceph_decode_64_safe(p, end, item->cookie, bad);
5003 	ceph_decode_skip_32(p, end, bad); /* skip timeout seconds */
5004 
5005 	if (struct_v >= 2) {
5006 		ret = ceph_decode_entity_addr(p, end, &item->addr);
5007 		if (ret)
5008 			goto bad;
5009 	} else {
5010 		ret = 0;
5011 	}
5012 
5013 	dout("%s %s%llu cookie %llu addr %s\n", __func__,
5014 	     ENTITY_NAME(item->name), item->cookie,
5015 	     ceph_pr_addr(&item->addr));
5016 bad:
5017 	return ret;
5018 }
5019 
5020 static int decode_watchers(void **p, void *end,
5021 			   struct ceph_watch_item **watchers,
5022 			   u32 *num_watchers)
5023 {
5024 	u8 struct_v;
5025 	u32 struct_len;
5026 	int i;
5027 	int ret;
5028 
5029 	ret = ceph_start_decoding(p, end, 1, "obj_list_watch_response_t",
5030 				  &struct_v, &struct_len);
5031 	if (ret)
5032 		return ret;
5033 
5034 	*num_watchers = ceph_decode_32(p);
5035 	*watchers = kcalloc(*num_watchers, sizeof(**watchers), GFP_NOIO);
5036 	if (!*watchers)
5037 		return -ENOMEM;
5038 
5039 	for (i = 0; i < *num_watchers; i++) {
5040 		ret = decode_watcher(p, end, *watchers + i);
5041 		if (ret) {
5042 			kfree(*watchers);
5043 			return ret;
5044 		}
5045 	}
5046 
5047 	return 0;
5048 }
5049 
5050 /*
5051  * On success, the caller is responsible for:
5052  *
5053  *     kfree(watchers);
5054  */
5055 int ceph_osdc_list_watchers(struct ceph_osd_client *osdc,
5056 			    struct ceph_object_id *oid,
5057 			    struct ceph_object_locator *oloc,
5058 			    struct ceph_watch_item **watchers,
5059 			    u32 *num_watchers)
5060 {
5061 	struct ceph_osd_request *req;
5062 	struct page **pages;
5063 	int ret;
5064 
5065 	req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
5066 	if (!req)
5067 		return -ENOMEM;
5068 
5069 	ceph_oid_copy(&req->r_base_oid, oid);
5070 	ceph_oloc_copy(&req->r_base_oloc, oloc);
5071 	req->r_flags = CEPH_OSD_FLAG_READ;
5072 
5073 	pages = ceph_alloc_page_vector(1, GFP_NOIO);
5074 	if (IS_ERR(pages)) {
5075 		ret = PTR_ERR(pages);
5076 		goto out_put_req;
5077 	}
5078 
5079 	osd_req_op_init(req, 0, CEPH_OSD_OP_LIST_WATCHERS, 0);
5080 	ceph_osd_data_pages_init(osd_req_op_data(req, 0, list_watchers,
5081 						 response_data),
5082 				 pages, PAGE_SIZE, 0, false, true);
5083 
5084 	ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
5085 	if (ret)
5086 		goto out_put_req;
5087 
5088 	ceph_osdc_start_request(osdc, req);
5089 	ret = ceph_osdc_wait_request(osdc, req);
5090 	if (ret >= 0) {
5091 		void *p = page_address(pages[0]);
5092 		void *const end = p + req->r_ops[0].outdata_len;
5093 
5094 		ret = decode_watchers(&p, end, watchers, num_watchers);
5095 	}
5096 
5097 out_put_req:
5098 	ceph_osdc_put_request(req);
5099 	return ret;
5100 }
5101 EXPORT_SYMBOL(ceph_osdc_list_watchers);
5102 
5103 /*
5104  * Call all pending notify callbacks - for use after a watch is
5105  * unregistered, to make sure no more callbacks for it will be invoked
5106  */
5107 void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc)
5108 {
5109 	dout("%s osdc %p\n", __func__, osdc);
5110 	flush_workqueue(osdc->notify_wq);
5111 }
5112 EXPORT_SYMBOL(ceph_osdc_flush_notifies);
5113 
5114 void ceph_osdc_maybe_request_map(struct ceph_osd_client *osdc)
5115 {
5116 	down_read(&osdc->lock);
5117 	maybe_request_map(osdc);
5118 	up_read(&osdc->lock);
5119 }
5120 EXPORT_SYMBOL(ceph_osdc_maybe_request_map);
5121 
5122 /*
5123  * Execute an OSD class method on an object.
5124  *
5125  * @flags: CEPH_OSD_FLAG_*
5126  * @resp_len: in/out param for reply length
5127  */
5128 int ceph_osdc_call(struct ceph_osd_client *osdc,
5129 		   struct ceph_object_id *oid,
5130 		   struct ceph_object_locator *oloc,
5131 		   const char *class, const char *method,
5132 		   unsigned int flags,
5133 		   struct page *req_page, size_t req_len,
5134 		   struct page **resp_pages, size_t *resp_len)
5135 {
5136 	struct ceph_osd_request *req;
5137 	int ret;
5138 
5139 	if (req_len > PAGE_SIZE)
5140 		return -E2BIG;
5141 
5142 	req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
5143 	if (!req)
5144 		return -ENOMEM;
5145 
5146 	ceph_oid_copy(&req->r_base_oid, oid);
5147 	ceph_oloc_copy(&req->r_base_oloc, oloc);
5148 	req->r_flags = flags;
5149 
5150 	ret = osd_req_op_cls_init(req, 0, class, method);
5151 	if (ret)
5152 		goto out_put_req;
5153 
5154 	if (req_page)
5155 		osd_req_op_cls_request_data_pages(req, 0, &req_page, req_len,
5156 						  0, false, false);
5157 	if (resp_pages)
5158 		osd_req_op_cls_response_data_pages(req, 0, resp_pages,
5159 						   *resp_len, 0, false, false);
5160 
5161 	ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
5162 	if (ret)
5163 		goto out_put_req;
5164 
5165 	ceph_osdc_start_request(osdc, req);
5166 	ret = ceph_osdc_wait_request(osdc, req);
5167 	if (ret >= 0) {
5168 		ret = req->r_ops[0].rval;
5169 		if (resp_pages)
5170 			*resp_len = req->r_ops[0].outdata_len;
5171 	}
5172 
5173 out_put_req:
5174 	ceph_osdc_put_request(req);
5175 	return ret;
5176 }
5177 EXPORT_SYMBOL(ceph_osdc_call);
5178 
5179 /*
5180  * reset all osd connections
5181  */
5182 void ceph_osdc_reopen_osds(struct ceph_osd_client *osdc)
5183 {
5184 	struct rb_node *n;
5185 
5186 	down_write(&osdc->lock);
5187 	for (n = rb_first(&osdc->osds); n; ) {
5188 		struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
5189 
5190 		n = rb_next(n);
5191 		if (!reopen_osd(osd))
5192 			kick_osd_requests(osd);
5193 	}
5194 	up_write(&osdc->lock);
5195 }
5196 
5197 /*
5198  * init, shutdown
5199  */
5200 int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
5201 {
5202 	int err;
5203 
5204 	dout("init\n");
5205 	osdc->client = client;
5206 	init_rwsem(&osdc->lock);
5207 	osdc->osds = RB_ROOT;
5208 	INIT_LIST_HEAD(&osdc->osd_lru);
5209 	spin_lock_init(&osdc->osd_lru_lock);
5210 	osd_init(&osdc->homeless_osd);
5211 	osdc->homeless_osd.o_osdc = osdc;
5212 	osdc->homeless_osd.o_osd = CEPH_HOMELESS_OSD;
5213 	osdc->last_linger_id = CEPH_LINGER_ID_START;
5214 	osdc->linger_requests = RB_ROOT;
5215 	osdc->map_checks = RB_ROOT;
5216 	osdc->linger_map_checks = RB_ROOT;
5217 	INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
5218 	INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
5219 
5220 	err = -ENOMEM;
5221 	osdc->osdmap = ceph_osdmap_alloc();
5222 	if (!osdc->osdmap)
5223 		goto out;
5224 
5225 	osdc->req_mempool = mempool_create_slab_pool(10,
5226 						     ceph_osd_request_cache);
5227 	if (!osdc->req_mempool)
5228 		goto out_map;
5229 
5230 	err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
5231 				PAGE_SIZE, CEPH_OSD_SLAB_OPS, 10, "osd_op");
5232 	if (err < 0)
5233 		goto out_mempool;
5234 	err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
5235 				PAGE_SIZE, CEPH_OSD_SLAB_OPS, 10,
5236 				"osd_op_reply");
5237 	if (err < 0)
5238 		goto out_msgpool;
5239 
5240 	err = -ENOMEM;
5241 	osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
5242 	if (!osdc->notify_wq)
5243 		goto out_msgpool_reply;
5244 
5245 	osdc->completion_wq = create_singlethread_workqueue("ceph-completion");
5246 	if (!osdc->completion_wq)
5247 		goto out_notify_wq;
5248 
5249 	schedule_delayed_work(&osdc->timeout_work,
5250 			      osdc->client->options->osd_keepalive_timeout);
5251 	schedule_delayed_work(&osdc->osds_timeout_work,
5252 	    round_jiffies_relative(osdc->client->options->osd_idle_ttl));
5253 
5254 	return 0;
5255 
5256 out_notify_wq:
5257 	destroy_workqueue(osdc->notify_wq);
5258 out_msgpool_reply:
5259 	ceph_msgpool_destroy(&osdc->msgpool_op_reply);
5260 out_msgpool:
5261 	ceph_msgpool_destroy(&osdc->msgpool_op);
5262 out_mempool:
5263 	mempool_destroy(osdc->req_mempool);
5264 out_map:
5265 	ceph_osdmap_destroy(osdc->osdmap);
5266 out:
5267 	return err;
5268 }
5269 
5270 void ceph_osdc_stop(struct ceph_osd_client *osdc)
5271 {
5272 	destroy_workqueue(osdc->completion_wq);
5273 	destroy_workqueue(osdc->notify_wq);
5274 	cancel_delayed_work_sync(&osdc->timeout_work);
5275 	cancel_delayed_work_sync(&osdc->osds_timeout_work);
5276 
5277 	down_write(&osdc->lock);
5278 	while (!RB_EMPTY_ROOT(&osdc->osds)) {
5279 		struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
5280 						struct ceph_osd, o_node);
5281 		close_osd(osd);
5282 	}
5283 	up_write(&osdc->lock);
5284 	WARN_ON(refcount_read(&osdc->homeless_osd.o_ref) != 1);
5285 	osd_cleanup(&osdc->homeless_osd);
5286 
5287 	WARN_ON(!list_empty(&osdc->osd_lru));
5288 	WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_requests));
5289 	WARN_ON(!RB_EMPTY_ROOT(&osdc->map_checks));
5290 	WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_map_checks));
5291 	WARN_ON(atomic_read(&osdc->num_requests));
5292 	WARN_ON(atomic_read(&osdc->num_homeless));
5293 
5294 	ceph_osdmap_destroy(osdc->osdmap);
5295 	mempool_destroy(osdc->req_mempool);
5296 	ceph_msgpool_destroy(&osdc->msgpool_op);
5297 	ceph_msgpool_destroy(&osdc->msgpool_op_reply);
5298 }
5299 
5300 int osd_req_op_copy_from_init(struct ceph_osd_request *req,
5301 			      u64 src_snapid, u64 src_version,
5302 			      struct ceph_object_id *src_oid,
5303 			      struct ceph_object_locator *src_oloc,
5304 			      u32 src_fadvise_flags,
5305 			      u32 dst_fadvise_flags,
5306 			      u32 truncate_seq, u64 truncate_size,
5307 			      u8 copy_from_flags)
5308 {
5309 	struct ceph_osd_req_op *op;
5310 	struct page **pages;
5311 	void *p, *end;
5312 
5313 	pages = ceph_alloc_page_vector(1, GFP_KERNEL);
5314 	if (IS_ERR(pages))
5315 		return PTR_ERR(pages);
5316 
5317 	op = osd_req_op_init(req, 0, CEPH_OSD_OP_COPY_FROM2,
5318 			     dst_fadvise_flags);
5319 	op->copy_from.snapid = src_snapid;
5320 	op->copy_from.src_version = src_version;
5321 	op->copy_from.flags = copy_from_flags;
5322 	op->copy_from.src_fadvise_flags = src_fadvise_flags;
5323 
5324 	p = page_address(pages[0]);
5325 	end = p + PAGE_SIZE;
5326 	ceph_encode_string(&p, end, src_oid->name, src_oid->name_len);
5327 	encode_oloc(&p, end, src_oloc);
5328 	ceph_encode_32(&p, truncate_seq);
5329 	ceph_encode_64(&p, truncate_size);
5330 	op->indata_len = PAGE_SIZE - (end - p);
5331 
5332 	ceph_osd_data_pages_init(&op->copy_from.osd_data, pages,
5333 				 op->indata_len, 0, false, true);
5334 	return 0;
5335 }
5336 EXPORT_SYMBOL(osd_req_op_copy_from_init);
5337 
5338 int __init ceph_osdc_setup(void)
5339 {
5340 	size_t size = sizeof(struct ceph_osd_request) +
5341 	    CEPH_OSD_SLAB_OPS * sizeof(struct ceph_osd_req_op);
5342 
5343 	BUG_ON(ceph_osd_request_cache);
5344 	ceph_osd_request_cache = kmem_cache_create("ceph_osd_request", size,
5345 						   0, 0, NULL);
5346 
5347 	return ceph_osd_request_cache ? 0 : -ENOMEM;
5348 }
5349 
5350 void ceph_osdc_cleanup(void)
5351 {
5352 	BUG_ON(!ceph_osd_request_cache);
5353 	kmem_cache_destroy(ceph_osd_request_cache);
5354 	ceph_osd_request_cache = NULL;
5355 }
5356 
5357 /*
5358  * handle incoming message
5359  */
5360 static void osd_dispatch(struct ceph_connection *con, struct ceph_msg *msg)
5361 {
5362 	struct ceph_osd *osd = con->private;
5363 	struct ceph_osd_client *osdc = osd->o_osdc;
5364 	int type = le16_to_cpu(msg->hdr.type);
5365 
5366 	switch (type) {
5367 	case CEPH_MSG_OSD_MAP:
5368 		ceph_osdc_handle_map(osdc, msg);
5369 		break;
5370 	case CEPH_MSG_OSD_OPREPLY:
5371 		handle_reply(osd, msg);
5372 		break;
5373 	case CEPH_MSG_OSD_BACKOFF:
5374 		handle_backoff(osd, msg);
5375 		break;
5376 	case CEPH_MSG_WATCH_NOTIFY:
5377 		handle_watch_notify(osdc, msg);
5378 		break;
5379 
5380 	default:
5381 		pr_err("received unknown message type %d %s\n", type,
5382 		       ceph_msg_type_name(type));
5383 	}
5384 
5385 	ceph_msg_put(msg);
5386 }
5387 
5388 /* How much sparse data was requested? */
5389 static u64 sparse_data_requested(struct ceph_osd_request *req)
5390 {
5391 	u64 len = 0;
5392 
5393 	if (req->r_flags & CEPH_OSD_FLAG_READ) {
5394 		int i;
5395 
5396 		for (i = 0; i < req->r_num_ops; ++i) {
5397 			struct ceph_osd_req_op *op = &req->r_ops[i];
5398 
5399 			if (op->op == CEPH_OSD_OP_SPARSE_READ)
5400 				len += op->extent.length;
5401 		}
5402 	}
5403 	return len;
5404 }
5405 
5406 /*
5407  * Lookup and return message for incoming reply.  Don't try to do
5408  * anything about a larger than preallocated data portion of the
5409  * message at the moment - for now, just skip the message.
5410  */
5411 static struct ceph_msg *get_reply(struct ceph_connection *con,
5412 				  struct ceph_msg_header *hdr,
5413 				  int *skip)
5414 {
5415 	struct ceph_osd *osd = con->private;
5416 	struct ceph_osd_client *osdc = osd->o_osdc;
5417 	struct ceph_msg *m = NULL;
5418 	struct ceph_osd_request *req;
5419 	int front_len = le32_to_cpu(hdr->front_len);
5420 	int data_len = le32_to_cpu(hdr->data_len);
5421 	u64 tid = le64_to_cpu(hdr->tid);
5422 	u64 srlen;
5423 
5424 	down_read(&osdc->lock);
5425 	if (!osd_registered(osd)) {
5426 		dout("%s osd%d unknown, skipping\n", __func__, osd->o_osd);
5427 		*skip = 1;
5428 		goto out_unlock_osdc;
5429 	}
5430 	WARN_ON(osd->o_osd != le64_to_cpu(hdr->src.num));
5431 
5432 	mutex_lock(&osd->lock);
5433 	req = lookup_request(&osd->o_requests, tid);
5434 	if (!req) {
5435 		dout("%s osd%d tid %llu unknown, skipping\n", __func__,
5436 		     osd->o_osd, tid);
5437 		*skip = 1;
5438 		goto out_unlock_session;
5439 	}
5440 
5441 	ceph_msg_revoke_incoming(req->r_reply);
5442 
5443 	if (front_len > req->r_reply->front_alloc_len) {
5444 		pr_warn("%s osd%d tid %llu front %d > preallocated %d\n",
5445 			__func__, osd->o_osd, req->r_tid, front_len,
5446 			req->r_reply->front_alloc_len);
5447 		m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS,
5448 				 false);
5449 		if (!m)
5450 			goto out_unlock_session;
5451 		ceph_msg_put(req->r_reply);
5452 		req->r_reply = m;
5453 	}
5454 
5455 	srlen = sparse_data_requested(req);
5456 	if (!srlen && data_len > req->r_reply->data_length) {
5457 		pr_warn("%s osd%d tid %llu data %d > preallocated %zu, skipping\n",
5458 			__func__, osd->o_osd, req->r_tid, data_len,
5459 			req->r_reply->data_length);
5460 		m = NULL;
5461 		*skip = 1;
5462 		goto out_unlock_session;
5463 	}
5464 
5465 	m = ceph_msg_get(req->r_reply);
5466 	m->sparse_read_total = srlen;
5467 
5468 	dout("get_reply tid %lld %p\n", tid, m);
5469 
5470 out_unlock_session:
5471 	mutex_unlock(&osd->lock);
5472 out_unlock_osdc:
5473 	up_read(&osdc->lock);
5474 	return m;
5475 }
5476 
5477 static struct ceph_msg *alloc_msg_with_page_vector(struct ceph_msg_header *hdr)
5478 {
5479 	struct ceph_msg *m;
5480 	int type = le16_to_cpu(hdr->type);
5481 	u32 front_len = le32_to_cpu(hdr->front_len);
5482 	u32 data_len = le32_to_cpu(hdr->data_len);
5483 
5484 	m = ceph_msg_new2(type, front_len, 1, GFP_NOIO, false);
5485 	if (!m)
5486 		return NULL;
5487 
5488 	if (data_len) {
5489 		struct page **pages;
5490 
5491 		pages = ceph_alloc_page_vector(calc_pages_for(0, data_len),
5492 					       GFP_NOIO);
5493 		if (IS_ERR(pages)) {
5494 			ceph_msg_put(m);
5495 			return NULL;
5496 		}
5497 
5498 		ceph_msg_data_add_pages(m, pages, data_len, 0, true);
5499 	}
5500 
5501 	return m;
5502 }
5503 
5504 static struct ceph_msg *osd_alloc_msg(struct ceph_connection *con,
5505 				      struct ceph_msg_header *hdr,
5506 				      int *skip)
5507 {
5508 	struct ceph_osd *osd = con->private;
5509 	int type = le16_to_cpu(hdr->type);
5510 
5511 	*skip = 0;
5512 	switch (type) {
5513 	case CEPH_MSG_OSD_MAP:
5514 	case CEPH_MSG_OSD_BACKOFF:
5515 	case CEPH_MSG_WATCH_NOTIFY:
5516 		return alloc_msg_with_page_vector(hdr);
5517 	case CEPH_MSG_OSD_OPREPLY:
5518 		return get_reply(con, hdr, skip);
5519 	default:
5520 		pr_warn("%s osd%d unknown msg type %d, skipping\n", __func__,
5521 			osd->o_osd, type);
5522 		*skip = 1;
5523 		return NULL;
5524 	}
5525 }
5526 
5527 /*
5528  * Wrappers to refcount containing ceph_osd struct
5529  */
5530 static struct ceph_connection *osd_get_con(struct ceph_connection *con)
5531 {
5532 	struct ceph_osd *osd = con->private;
5533 	if (get_osd(osd))
5534 		return con;
5535 	return NULL;
5536 }
5537 
5538 static void osd_put_con(struct ceph_connection *con)
5539 {
5540 	struct ceph_osd *osd = con->private;
5541 	put_osd(osd);
5542 }
5543 
5544 /*
5545  * authentication
5546  */
5547 
5548 /*
5549  * Note: returned pointer is the address of a structure that's
5550  * managed separately.  Caller must *not* attempt to free it.
5551  */
5552 static struct ceph_auth_handshake *
5553 osd_get_authorizer(struct ceph_connection *con, int *proto, int force_new)
5554 {
5555 	struct ceph_osd *o = con->private;
5556 	struct ceph_osd_client *osdc = o->o_osdc;
5557 	struct ceph_auth_client *ac = osdc->client->monc.auth;
5558 	struct ceph_auth_handshake *auth = &o->o_auth;
5559 	int ret;
5560 
5561 	ret = __ceph_auth_get_authorizer(ac, auth, CEPH_ENTITY_TYPE_OSD,
5562 					 force_new, proto, NULL, NULL);
5563 	if (ret)
5564 		return ERR_PTR(ret);
5565 
5566 	return auth;
5567 }
5568 
5569 static int osd_add_authorizer_challenge(struct ceph_connection *con,
5570 				    void *challenge_buf, int challenge_buf_len)
5571 {
5572 	struct ceph_osd *o = con->private;
5573 	struct ceph_osd_client *osdc = o->o_osdc;
5574 	struct ceph_auth_client *ac = osdc->client->monc.auth;
5575 
5576 	return ceph_auth_add_authorizer_challenge(ac, o->o_auth.authorizer,
5577 					    challenge_buf, challenge_buf_len);
5578 }
5579 
5580 static int osd_verify_authorizer_reply(struct ceph_connection *con)
5581 {
5582 	struct ceph_osd *o = con->private;
5583 	struct ceph_osd_client *osdc = o->o_osdc;
5584 	struct ceph_auth_client *ac = osdc->client->monc.auth;
5585 	struct ceph_auth_handshake *auth = &o->o_auth;
5586 
5587 	return ceph_auth_verify_authorizer_reply(ac, auth->authorizer,
5588 		auth->authorizer_reply_buf, auth->authorizer_reply_buf_len,
5589 		NULL, NULL, NULL, NULL);
5590 }
5591 
5592 static int osd_invalidate_authorizer(struct ceph_connection *con)
5593 {
5594 	struct ceph_osd *o = con->private;
5595 	struct ceph_osd_client *osdc = o->o_osdc;
5596 	struct ceph_auth_client *ac = osdc->client->monc.auth;
5597 
5598 	ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
5599 	return ceph_monc_validate_auth(&osdc->client->monc);
5600 }
5601 
5602 static int osd_get_auth_request(struct ceph_connection *con,
5603 				void *buf, int *buf_len,
5604 				void **authorizer, int *authorizer_len)
5605 {
5606 	struct ceph_osd *o = con->private;
5607 	struct ceph_auth_client *ac = o->o_osdc->client->monc.auth;
5608 	struct ceph_auth_handshake *auth = &o->o_auth;
5609 	int ret;
5610 
5611 	ret = ceph_auth_get_authorizer(ac, auth, CEPH_ENTITY_TYPE_OSD,
5612 				       buf, buf_len);
5613 	if (ret)
5614 		return ret;
5615 
5616 	*authorizer = auth->authorizer_buf;
5617 	*authorizer_len = auth->authorizer_buf_len;
5618 	return 0;
5619 }
5620 
5621 static int osd_handle_auth_reply_more(struct ceph_connection *con,
5622 				      void *reply, int reply_len,
5623 				      void *buf, int *buf_len,
5624 				      void **authorizer, int *authorizer_len)
5625 {
5626 	struct ceph_osd *o = con->private;
5627 	struct ceph_auth_client *ac = o->o_osdc->client->monc.auth;
5628 	struct ceph_auth_handshake *auth = &o->o_auth;
5629 	int ret;
5630 
5631 	ret = ceph_auth_handle_svc_reply_more(ac, auth, reply, reply_len,
5632 					      buf, buf_len);
5633 	if (ret)
5634 		return ret;
5635 
5636 	*authorizer = auth->authorizer_buf;
5637 	*authorizer_len = auth->authorizer_buf_len;
5638 	return 0;
5639 }
5640 
5641 static int osd_handle_auth_done(struct ceph_connection *con,
5642 				u64 global_id, void *reply, int reply_len,
5643 				u8 *session_key, int *session_key_len,
5644 				u8 *con_secret, int *con_secret_len)
5645 {
5646 	struct ceph_osd *o = con->private;
5647 	struct ceph_auth_client *ac = o->o_osdc->client->monc.auth;
5648 	struct ceph_auth_handshake *auth = &o->o_auth;
5649 
5650 	return ceph_auth_handle_svc_reply_done(ac, auth, reply, reply_len,
5651 					       session_key, session_key_len,
5652 					       con_secret, con_secret_len);
5653 }
5654 
5655 static int osd_handle_auth_bad_method(struct ceph_connection *con,
5656 				      int used_proto, int result,
5657 				      const int *allowed_protos, int proto_cnt,
5658 				      const int *allowed_modes, int mode_cnt)
5659 {
5660 	struct ceph_osd *o = con->private;
5661 	struct ceph_mon_client *monc = &o->o_osdc->client->monc;
5662 	int ret;
5663 
5664 	if (ceph_auth_handle_bad_authorizer(monc->auth, CEPH_ENTITY_TYPE_OSD,
5665 					    used_proto, result,
5666 					    allowed_protos, proto_cnt,
5667 					    allowed_modes, mode_cnt)) {
5668 		ret = ceph_monc_validate_auth(monc);
5669 		if (ret)
5670 			return ret;
5671 	}
5672 
5673 	return -EACCES;
5674 }
5675 
5676 static void osd_reencode_message(struct ceph_msg *msg)
5677 {
5678 	int type = le16_to_cpu(msg->hdr.type);
5679 
5680 	if (type == CEPH_MSG_OSD_OP)
5681 		encode_request_finish(msg);
5682 }
5683 
5684 static int osd_sign_message(struct ceph_msg *msg)
5685 {
5686 	struct ceph_osd *o = msg->con->private;
5687 	struct ceph_auth_handshake *auth = &o->o_auth;
5688 
5689 	return ceph_auth_sign_message(auth, msg);
5690 }
5691 
5692 static int osd_check_message_signature(struct ceph_msg *msg)
5693 {
5694 	struct ceph_osd *o = msg->con->private;
5695 	struct ceph_auth_handshake *auth = &o->o_auth;
5696 
5697 	return ceph_auth_check_message_signature(auth, msg);
5698 }
5699 
5700 static void advance_cursor(struct ceph_msg_data_cursor *cursor, size_t len,
5701 			   bool zero)
5702 {
5703 	while (len) {
5704 		struct page *page;
5705 		size_t poff, plen;
5706 
5707 		page = ceph_msg_data_next(cursor, &poff, &plen);
5708 		if (plen > len)
5709 			plen = len;
5710 		if (zero)
5711 			zero_user_segment(page, poff, poff + plen);
5712 		len -= plen;
5713 		ceph_msg_data_advance(cursor, plen);
5714 	}
5715 }
5716 
5717 static int prep_next_sparse_read(struct ceph_connection *con,
5718 				 struct ceph_msg_data_cursor *cursor)
5719 {
5720 	struct ceph_osd *o = con->private;
5721 	struct ceph_sparse_read *sr = &o->o_sparse_read;
5722 	struct ceph_osd_request *req;
5723 	struct ceph_osd_req_op *op;
5724 
5725 	spin_lock(&o->o_requests_lock);
5726 	req = lookup_request(&o->o_requests, le64_to_cpu(con->in_msg->hdr.tid));
5727 	if (!req) {
5728 		spin_unlock(&o->o_requests_lock);
5729 		return -EBADR;
5730 	}
5731 
5732 	if (o->o_sparse_op_idx < 0) {
5733 		dout("%s: [%d] starting new sparse read req\n",
5734 		     __func__, o->o_osd);
5735 	} else {
5736 		u64 end;
5737 
5738 		op = &req->r_ops[o->o_sparse_op_idx];
5739 
5740 		WARN_ON_ONCE(op->extent.sparse_ext);
5741 
5742 		/* hand back buffer we took earlier */
5743 		op->extent.sparse_ext = sr->sr_extent;
5744 		sr->sr_extent = NULL;
5745 		op->extent.sparse_ext_cnt = sr->sr_count;
5746 		sr->sr_ext_len = 0;
5747 		dout("%s: [%d] completed extent array len %d cursor->resid %zd\n",
5748 		     __func__, o->o_osd, op->extent.sparse_ext_cnt, cursor->resid);
5749 		/* Advance to end of data for this operation */
5750 		end = ceph_sparse_ext_map_end(op);
5751 		if (end < sr->sr_req_len)
5752 			advance_cursor(cursor, sr->sr_req_len - end, false);
5753 	}
5754 
5755 	ceph_init_sparse_read(sr);
5756 
5757 	/* find next op in this request (if any) */
5758 	while (++o->o_sparse_op_idx < req->r_num_ops) {
5759 		op = &req->r_ops[o->o_sparse_op_idx];
5760 		if (op->op == CEPH_OSD_OP_SPARSE_READ)
5761 			goto found;
5762 	}
5763 
5764 	/* reset for next sparse read request */
5765 	spin_unlock(&o->o_requests_lock);
5766 	o->o_sparse_op_idx = -1;
5767 	return 0;
5768 found:
5769 	sr->sr_req_off = op->extent.offset;
5770 	sr->sr_req_len = op->extent.length;
5771 	sr->sr_pos = sr->sr_req_off;
5772 	dout("%s: [%d] new sparse read op at idx %d 0x%llx~0x%llx\n", __func__,
5773 	     o->o_osd, o->o_sparse_op_idx, sr->sr_req_off, sr->sr_req_len);
5774 
5775 	/* hand off request's sparse extent map buffer */
5776 	sr->sr_ext_len = op->extent.sparse_ext_cnt;
5777 	op->extent.sparse_ext_cnt = 0;
5778 	sr->sr_extent = op->extent.sparse_ext;
5779 	op->extent.sparse_ext = NULL;
5780 
5781 	spin_unlock(&o->o_requests_lock);
5782 	return 1;
5783 }
5784 
5785 #ifdef __BIG_ENDIAN
5786 static inline void convert_extent_map(struct ceph_sparse_read *sr)
5787 {
5788 	int i;
5789 
5790 	for (i = 0; i < sr->sr_count; i++) {
5791 		struct ceph_sparse_extent *ext = &sr->sr_extent[i];
5792 
5793 		ext->off = le64_to_cpu((__force __le64)ext->off);
5794 		ext->len = le64_to_cpu((__force __le64)ext->len);
5795 	}
5796 }
5797 #else
5798 static inline void convert_extent_map(struct ceph_sparse_read *sr)
5799 {
5800 }
5801 #endif
5802 
5803 static int osd_sparse_read(struct ceph_connection *con,
5804 			   struct ceph_msg_data_cursor *cursor,
5805 			   char **pbuf)
5806 {
5807 	struct ceph_osd *o = con->private;
5808 	struct ceph_sparse_read *sr = &o->o_sparse_read;
5809 	u32 count = sr->sr_count;
5810 	u64 eoff, elen, len = 0;
5811 	int i, ret;
5812 
5813 	switch (sr->sr_state) {
5814 	case CEPH_SPARSE_READ_HDR:
5815 next_op:
5816 		ret = prep_next_sparse_read(con, cursor);
5817 		if (ret <= 0)
5818 			return ret;
5819 
5820 		/* number of extents */
5821 		ret = sizeof(sr->sr_count);
5822 		*pbuf = (char *)&sr->sr_count;
5823 		sr->sr_state = CEPH_SPARSE_READ_EXTENTS;
5824 		break;
5825 	case CEPH_SPARSE_READ_EXTENTS:
5826 		/* Convert sr_count to host-endian */
5827 		count = le32_to_cpu((__force __le32)sr->sr_count);
5828 		sr->sr_count = count;
5829 		dout("[%d] got %u extents\n", o->o_osd, count);
5830 
5831 		if (count > 0) {
5832 			if (!sr->sr_extent || count > sr->sr_ext_len) {
5833 				/* no extent array provided, or too short */
5834 				kfree(sr->sr_extent);
5835 				sr->sr_extent = kmalloc_array(count,
5836 							      sizeof(*sr->sr_extent),
5837 							      GFP_NOIO);
5838 				if (!sr->sr_extent) {
5839 					pr_err("%s: failed to allocate %u extents\n",
5840 					       __func__, count);
5841 					return -ENOMEM;
5842 				}
5843 				sr->sr_ext_len = count;
5844 			}
5845 			ret = count * sizeof(*sr->sr_extent);
5846 			*pbuf = (char *)sr->sr_extent;
5847 			sr->sr_state = CEPH_SPARSE_READ_DATA_LEN;
5848 			break;
5849 		}
5850 		/* No extents? Read data len */
5851 		fallthrough;
5852 	case CEPH_SPARSE_READ_DATA_LEN:
5853 		convert_extent_map(sr);
5854 		ret = sizeof(sr->sr_datalen);
5855 		*pbuf = (char *)&sr->sr_datalen;
5856 		sr->sr_state = CEPH_SPARSE_READ_DATA_PRE;
5857 		break;
5858 	case CEPH_SPARSE_READ_DATA_PRE:
5859 		/* Convert sr_datalen to host-endian */
5860 		sr->sr_datalen = le32_to_cpu((__force __le32)sr->sr_datalen);
5861 		for (i = 0; i < count; i++)
5862 			len += sr->sr_extent[i].len;
5863 		if (sr->sr_datalen != len) {
5864 			pr_warn_ratelimited("data len %u != extent len %llu\n",
5865 					    sr->sr_datalen, len);
5866 			return -EREMOTEIO;
5867 		}
5868 		sr->sr_state = CEPH_SPARSE_READ_DATA;
5869 		fallthrough;
5870 	case CEPH_SPARSE_READ_DATA:
5871 		if (sr->sr_index >= count) {
5872 			sr->sr_state = CEPH_SPARSE_READ_HDR;
5873 			goto next_op;
5874 		}
5875 
5876 		eoff = sr->sr_extent[sr->sr_index].off;
5877 		elen = sr->sr_extent[sr->sr_index].len;
5878 
5879 		dout("[%d] ext %d off 0x%llx len 0x%llx\n",
5880 		     o->o_osd, sr->sr_index, eoff, elen);
5881 
5882 		if (elen > INT_MAX) {
5883 			dout("Sparse read extent length too long (0x%llx)\n",
5884 			     elen);
5885 			return -EREMOTEIO;
5886 		}
5887 
5888 		/* zero out anything from sr_pos to start of extent */
5889 		if (sr->sr_pos < eoff)
5890 			advance_cursor(cursor, eoff - sr->sr_pos, true);
5891 
5892 		/* Set position to end of extent */
5893 		sr->sr_pos = eoff + elen;
5894 
5895 		/* send back the new length and nullify the ptr */
5896 		cursor->sr_resid = elen;
5897 		ret = elen;
5898 		*pbuf = NULL;
5899 
5900 		/* Bump the array index */
5901 		++sr->sr_index;
5902 		break;
5903 	}
5904 	return ret;
5905 }
5906 
5907 static const struct ceph_connection_operations osd_con_ops = {
5908 	.get = osd_get_con,
5909 	.put = osd_put_con,
5910 	.sparse_read = osd_sparse_read,
5911 	.alloc_msg = osd_alloc_msg,
5912 	.dispatch = osd_dispatch,
5913 	.fault = osd_fault,
5914 	.reencode_message = osd_reencode_message,
5915 	.get_authorizer = osd_get_authorizer,
5916 	.add_authorizer_challenge = osd_add_authorizer_challenge,
5917 	.verify_authorizer_reply = osd_verify_authorizer_reply,
5918 	.invalidate_authorizer = osd_invalidate_authorizer,
5919 	.sign_message = osd_sign_message,
5920 	.check_message_signature = osd_check_message_signature,
5921 	.get_auth_request = osd_get_auth_request,
5922 	.handle_auth_reply_more = osd_handle_auth_reply_more,
5923 	.handle_auth_done = osd_handle_auth_done,
5924 	.handle_auth_bad_method = osd_handle_auth_bad_method,
5925 };
5926