xref: /linux/net/ceph/osd_client.c (revision 55f1b540d893da740a81200450014c45a8103f54)
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 void osd_req_op_extent_osd_data_pagelist(struct ceph_osd_request *osd_req,
224 			unsigned int which, struct ceph_pagelist *pagelist)
225 {
226 	struct ceph_osd_data *osd_data;
227 
228 	osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
229 	ceph_osd_data_pagelist_init(osd_data, pagelist);
230 }
231 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pagelist);
232 
233 #ifdef CONFIG_BLOCK
234 void osd_req_op_extent_osd_data_bio(struct ceph_osd_request *osd_req,
235 				    unsigned int which,
236 				    struct ceph_bio_iter *bio_pos,
237 				    u32 bio_length)
238 {
239 	struct ceph_osd_data *osd_data;
240 
241 	osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
242 	ceph_osd_data_bio_init(osd_data, bio_pos, bio_length);
243 }
244 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bio);
245 #endif /* CONFIG_BLOCK */
246 
247 void osd_req_op_extent_osd_data_bvecs(struct ceph_osd_request *osd_req,
248 				      unsigned int which,
249 				      struct bio_vec *bvecs, u32 num_bvecs,
250 				      u32 bytes)
251 {
252 	struct ceph_osd_data *osd_data;
253 	struct ceph_bvec_iter it = {
254 		.bvecs = bvecs,
255 		.iter = { .bi_size = bytes },
256 	};
257 
258 	osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
259 	ceph_osd_data_bvecs_init(osd_data, &it, num_bvecs);
260 }
261 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bvecs);
262 
263 void osd_req_op_extent_osd_data_bvec_pos(struct ceph_osd_request *osd_req,
264 					 unsigned int which,
265 					 struct ceph_bvec_iter *bvec_pos)
266 {
267 	struct ceph_osd_data *osd_data;
268 
269 	osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
270 	ceph_osd_data_bvecs_init(osd_data, bvec_pos, 0);
271 }
272 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bvec_pos);
273 
274 /**
275  * osd_req_op_extent_osd_iter - Set up an operation with an iterator buffer
276  * @osd_req: The request to set up
277  * @which: Index of the operation in which to set the iter
278  * @iter: The buffer iterator
279  */
280 void osd_req_op_extent_osd_iter(struct ceph_osd_request *osd_req,
281 				unsigned int which, struct iov_iter *iter)
282 {
283 	struct ceph_osd_data *osd_data;
284 
285 	osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
286 	ceph_osd_iter_init(osd_data, iter);
287 }
288 EXPORT_SYMBOL(osd_req_op_extent_osd_iter);
289 
290 static void osd_req_op_cls_request_info_pagelist(
291 			struct ceph_osd_request *osd_req,
292 			unsigned int which, struct ceph_pagelist *pagelist)
293 {
294 	struct ceph_osd_data *osd_data;
295 
296 	osd_data = osd_req_op_data(osd_req, which, cls, request_info);
297 	ceph_osd_data_pagelist_init(osd_data, pagelist);
298 }
299 
300 void osd_req_op_cls_request_data_pagelist(
301 			struct ceph_osd_request *osd_req,
302 			unsigned int which, struct ceph_pagelist *pagelist)
303 {
304 	struct ceph_osd_data *osd_data;
305 
306 	osd_data = osd_req_op_data(osd_req, which, cls, request_data);
307 	ceph_osd_data_pagelist_init(osd_data, pagelist);
308 	osd_req->r_ops[which].cls.indata_len += pagelist->length;
309 	osd_req->r_ops[which].indata_len += pagelist->length;
310 }
311 EXPORT_SYMBOL(osd_req_op_cls_request_data_pagelist);
312 
313 void osd_req_op_cls_request_data_pages(struct ceph_osd_request *osd_req,
314 			unsigned int which, struct page **pages, u64 length,
315 			u32 alignment, bool pages_from_pool, bool own_pages)
316 {
317 	struct ceph_osd_data *osd_data;
318 
319 	osd_data = osd_req_op_data(osd_req, which, cls, request_data);
320 	ceph_osd_data_pages_init(osd_data, pages, length, alignment,
321 				pages_from_pool, own_pages);
322 	osd_req->r_ops[which].cls.indata_len += length;
323 	osd_req->r_ops[which].indata_len += length;
324 }
325 EXPORT_SYMBOL(osd_req_op_cls_request_data_pages);
326 
327 void osd_req_op_cls_request_data_bvecs(struct ceph_osd_request *osd_req,
328 				       unsigned int which,
329 				       struct bio_vec *bvecs, u32 num_bvecs,
330 				       u32 bytes)
331 {
332 	struct ceph_osd_data *osd_data;
333 	struct ceph_bvec_iter it = {
334 		.bvecs = bvecs,
335 		.iter = { .bi_size = bytes },
336 	};
337 
338 	osd_data = osd_req_op_data(osd_req, which, cls, request_data);
339 	ceph_osd_data_bvecs_init(osd_data, &it, num_bvecs);
340 	osd_req->r_ops[which].cls.indata_len += bytes;
341 	osd_req->r_ops[which].indata_len += bytes;
342 }
343 EXPORT_SYMBOL(osd_req_op_cls_request_data_bvecs);
344 
345 void osd_req_op_cls_response_data_pages(struct ceph_osd_request *osd_req,
346 			unsigned int which, struct page **pages, u64 length,
347 			u32 alignment, bool pages_from_pool, bool own_pages)
348 {
349 	struct ceph_osd_data *osd_data;
350 
351 	osd_data = osd_req_op_data(osd_req, which, cls, response_data);
352 	ceph_osd_data_pages_init(osd_data, pages, length, alignment,
353 				pages_from_pool, own_pages);
354 }
355 EXPORT_SYMBOL(osd_req_op_cls_response_data_pages);
356 
357 static u64 ceph_osd_data_length(struct ceph_osd_data *osd_data)
358 {
359 	switch (osd_data->type) {
360 	case CEPH_OSD_DATA_TYPE_NONE:
361 		return 0;
362 	case CEPH_OSD_DATA_TYPE_PAGES:
363 		return osd_data->length;
364 	case CEPH_OSD_DATA_TYPE_PAGELIST:
365 		return (u64)osd_data->pagelist->length;
366 #ifdef CONFIG_BLOCK
367 	case CEPH_OSD_DATA_TYPE_BIO:
368 		return (u64)osd_data->bio_length;
369 #endif /* CONFIG_BLOCK */
370 	case CEPH_OSD_DATA_TYPE_BVECS:
371 		return osd_data->bvec_pos.iter.bi_size;
372 	case CEPH_OSD_DATA_TYPE_ITER:
373 		return iov_iter_count(&osd_data->iter);
374 	default:
375 		WARN(true, "unrecognized data type %d\n", (int)osd_data->type);
376 		return 0;
377 	}
378 }
379 
380 static void ceph_osd_data_release(struct ceph_osd_data *osd_data)
381 {
382 	if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES && osd_data->own_pages) {
383 		int num_pages;
384 
385 		num_pages = calc_pages_for((u64)osd_data->alignment,
386 						(u64)osd_data->length);
387 		ceph_release_page_vector(osd_data->pages, num_pages);
388 	} else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
389 		ceph_pagelist_release(osd_data->pagelist);
390 	}
391 	ceph_osd_data_init(osd_data);
392 }
393 
394 static void osd_req_op_data_release(struct ceph_osd_request *osd_req,
395 			unsigned int which)
396 {
397 	struct ceph_osd_req_op *op;
398 
399 	BUG_ON(which >= osd_req->r_num_ops);
400 	op = &osd_req->r_ops[which];
401 
402 	switch (op->op) {
403 	case CEPH_OSD_OP_READ:
404 	case CEPH_OSD_OP_SPARSE_READ:
405 	case CEPH_OSD_OP_WRITE:
406 	case CEPH_OSD_OP_WRITEFULL:
407 		kfree(op->extent.sparse_ext);
408 		ceph_osd_data_release(&op->extent.osd_data);
409 		break;
410 	case CEPH_OSD_OP_CALL:
411 		ceph_osd_data_release(&op->cls.request_info);
412 		ceph_osd_data_release(&op->cls.request_data);
413 		ceph_osd_data_release(&op->cls.response_data);
414 		break;
415 	case CEPH_OSD_OP_SETXATTR:
416 	case CEPH_OSD_OP_CMPXATTR:
417 		ceph_osd_data_release(&op->xattr.osd_data);
418 		break;
419 	case CEPH_OSD_OP_STAT:
420 		ceph_osd_data_release(&op->raw_data_in);
421 		break;
422 	case CEPH_OSD_OP_NOTIFY_ACK:
423 		ceph_osd_data_release(&op->notify_ack.request_data);
424 		break;
425 	case CEPH_OSD_OP_NOTIFY:
426 		ceph_osd_data_release(&op->notify.request_data);
427 		ceph_osd_data_release(&op->notify.response_data);
428 		break;
429 	case CEPH_OSD_OP_LIST_WATCHERS:
430 		ceph_osd_data_release(&op->list_watchers.response_data);
431 		break;
432 	case CEPH_OSD_OP_COPY_FROM2:
433 		ceph_osd_data_release(&op->copy_from.osd_data);
434 		break;
435 	default:
436 		break;
437 	}
438 }
439 
440 /*
441  * Assumes @t is zero-initialized.
442  */
443 static void target_init(struct ceph_osd_request_target *t)
444 {
445 	ceph_oid_init(&t->base_oid);
446 	ceph_oloc_init(&t->base_oloc);
447 	ceph_oid_init(&t->target_oid);
448 	ceph_oloc_init(&t->target_oloc);
449 
450 	ceph_osds_init(&t->acting);
451 	ceph_osds_init(&t->up);
452 	t->size = -1;
453 	t->min_size = -1;
454 
455 	t->osd = CEPH_HOMELESS_OSD;
456 }
457 
458 static void target_copy(struct ceph_osd_request_target *dest,
459 			const struct ceph_osd_request_target *src)
460 {
461 	ceph_oid_copy(&dest->base_oid, &src->base_oid);
462 	ceph_oloc_copy(&dest->base_oloc, &src->base_oloc);
463 	ceph_oid_copy(&dest->target_oid, &src->target_oid);
464 	ceph_oloc_copy(&dest->target_oloc, &src->target_oloc);
465 
466 	dest->pgid = src->pgid; /* struct */
467 	dest->spgid = src->spgid; /* struct */
468 	dest->pg_num = src->pg_num;
469 	dest->pg_num_mask = src->pg_num_mask;
470 	ceph_osds_copy(&dest->acting, &src->acting);
471 	ceph_osds_copy(&dest->up, &src->up);
472 	dest->size = src->size;
473 	dest->min_size = src->min_size;
474 	dest->sort_bitwise = src->sort_bitwise;
475 	dest->recovery_deletes = src->recovery_deletes;
476 
477 	dest->flags = src->flags;
478 	dest->used_replica = src->used_replica;
479 	dest->paused = src->paused;
480 
481 	dest->epoch = src->epoch;
482 	dest->last_force_resend = src->last_force_resend;
483 
484 	dest->osd = src->osd;
485 }
486 
487 static void target_destroy(struct ceph_osd_request_target *t)
488 {
489 	ceph_oid_destroy(&t->base_oid);
490 	ceph_oloc_destroy(&t->base_oloc);
491 	ceph_oid_destroy(&t->target_oid);
492 	ceph_oloc_destroy(&t->target_oloc);
493 }
494 
495 /*
496  * requests
497  */
498 static void request_release_checks(struct ceph_osd_request *req)
499 {
500 	WARN_ON(!RB_EMPTY_NODE(&req->r_node));
501 	WARN_ON(!RB_EMPTY_NODE(&req->r_mc_node));
502 	WARN_ON(!list_empty(&req->r_private_item));
503 	WARN_ON(req->r_osd);
504 }
505 
506 static void ceph_osdc_release_request(struct kref *kref)
507 {
508 	struct ceph_osd_request *req = container_of(kref,
509 					    struct ceph_osd_request, r_kref);
510 	unsigned int which;
511 
512 	dout("%s %p (r_request %p r_reply %p)\n", __func__, req,
513 	     req->r_request, req->r_reply);
514 	request_release_checks(req);
515 
516 	if (req->r_request)
517 		ceph_msg_put(req->r_request);
518 	if (req->r_reply)
519 		ceph_msg_put(req->r_reply);
520 
521 	for (which = 0; which < req->r_num_ops; which++)
522 		osd_req_op_data_release(req, which);
523 
524 	target_destroy(&req->r_t);
525 	ceph_put_snap_context(req->r_snapc);
526 
527 	if (req->r_mempool)
528 		mempool_free(req, req->r_osdc->req_mempool);
529 	else if (req->r_num_ops <= CEPH_OSD_SLAB_OPS)
530 		kmem_cache_free(ceph_osd_request_cache, req);
531 	else
532 		kfree(req);
533 }
534 
535 void ceph_osdc_get_request(struct ceph_osd_request *req)
536 {
537 	dout("%s %p (was %d)\n", __func__, req,
538 	     kref_read(&req->r_kref));
539 	kref_get(&req->r_kref);
540 }
541 EXPORT_SYMBOL(ceph_osdc_get_request);
542 
543 void ceph_osdc_put_request(struct ceph_osd_request *req)
544 {
545 	if (req) {
546 		dout("%s %p (was %d)\n", __func__, req,
547 		     kref_read(&req->r_kref));
548 		kref_put(&req->r_kref, ceph_osdc_release_request);
549 	}
550 }
551 EXPORT_SYMBOL(ceph_osdc_put_request);
552 
553 static void request_init(struct ceph_osd_request *req)
554 {
555 	/* req only, each op is zeroed in osd_req_op_init() */
556 	memset(req, 0, sizeof(*req));
557 
558 	kref_init(&req->r_kref);
559 	init_completion(&req->r_completion);
560 	RB_CLEAR_NODE(&req->r_node);
561 	RB_CLEAR_NODE(&req->r_mc_node);
562 	INIT_LIST_HEAD(&req->r_private_item);
563 
564 	target_init(&req->r_t);
565 }
566 
567 struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
568 					       struct ceph_snap_context *snapc,
569 					       unsigned int num_ops,
570 					       bool use_mempool,
571 					       gfp_t gfp_flags)
572 {
573 	struct ceph_osd_request *req;
574 
575 	if (use_mempool) {
576 		BUG_ON(num_ops > CEPH_OSD_SLAB_OPS);
577 		req = mempool_alloc(osdc->req_mempool, gfp_flags);
578 	} else if (num_ops <= CEPH_OSD_SLAB_OPS) {
579 		req = kmem_cache_alloc(ceph_osd_request_cache, gfp_flags);
580 	} else {
581 		BUG_ON(num_ops > CEPH_OSD_MAX_OPS);
582 		req = kmalloc(struct_size(req, r_ops, num_ops), gfp_flags);
583 	}
584 	if (unlikely(!req))
585 		return NULL;
586 
587 	request_init(req);
588 	req->r_osdc = osdc;
589 	req->r_mempool = use_mempool;
590 	req->r_num_ops = num_ops;
591 	req->r_snapid = CEPH_NOSNAP;
592 	req->r_snapc = ceph_get_snap_context(snapc);
593 
594 	dout("%s req %p\n", __func__, req);
595 	return req;
596 }
597 EXPORT_SYMBOL(ceph_osdc_alloc_request);
598 
599 static int ceph_oloc_encoding_size(const struct ceph_object_locator *oloc)
600 {
601 	return 8 + 4 + 4 + 4 + (oloc->pool_ns ? oloc->pool_ns->len : 0);
602 }
603 
604 static int __ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp,
605 				      int num_request_data_items,
606 				      int num_reply_data_items)
607 {
608 	struct ceph_osd_client *osdc = req->r_osdc;
609 	struct ceph_msg *msg;
610 	int msg_size;
611 
612 	WARN_ON(req->r_request || req->r_reply);
613 	WARN_ON(ceph_oid_empty(&req->r_base_oid));
614 	WARN_ON(ceph_oloc_empty(&req->r_base_oloc));
615 
616 	/* create request message */
617 	msg_size = CEPH_ENCODING_START_BLK_LEN +
618 			CEPH_PGID_ENCODING_LEN + 1; /* spgid */
619 	msg_size += 4 + 4 + 4; /* hash, osdmap_epoch, flags */
620 	msg_size += CEPH_ENCODING_START_BLK_LEN +
621 			sizeof(struct ceph_osd_reqid); /* reqid */
622 	msg_size += sizeof(struct ceph_blkin_trace_info); /* trace */
623 	msg_size += 4 + sizeof(struct ceph_timespec); /* client_inc, mtime */
624 	msg_size += CEPH_ENCODING_START_BLK_LEN +
625 			ceph_oloc_encoding_size(&req->r_base_oloc); /* oloc */
626 	msg_size += 4 + req->r_base_oid.name_len; /* oid */
627 	msg_size += 2 + req->r_num_ops * sizeof(struct ceph_osd_op);
628 	msg_size += 8; /* snapid */
629 	msg_size += 8; /* snap_seq */
630 	msg_size += 4 + 8 * (req->r_snapc ? req->r_snapc->num_snaps : 0);
631 	msg_size += 4 + 8; /* retry_attempt, features */
632 
633 	if (req->r_mempool)
634 		msg = ceph_msgpool_get(&osdc->msgpool_op, msg_size,
635 				       num_request_data_items);
636 	else
637 		msg = ceph_msg_new2(CEPH_MSG_OSD_OP, msg_size,
638 				    num_request_data_items, gfp, true);
639 	if (!msg)
640 		return -ENOMEM;
641 
642 	memset(msg->front.iov_base, 0, msg->front.iov_len);
643 	req->r_request = msg;
644 
645 	/* create reply message */
646 	msg_size = OSD_OPREPLY_FRONT_LEN;
647 	msg_size += req->r_base_oid.name_len;
648 	msg_size += req->r_num_ops * sizeof(struct ceph_osd_op);
649 
650 	if (req->r_mempool)
651 		msg = ceph_msgpool_get(&osdc->msgpool_op_reply, msg_size,
652 				       num_reply_data_items);
653 	else
654 		msg = ceph_msg_new2(CEPH_MSG_OSD_OPREPLY, msg_size,
655 				    num_reply_data_items, gfp, true);
656 	if (!msg)
657 		return -ENOMEM;
658 
659 	req->r_reply = msg;
660 
661 	return 0;
662 }
663 
664 static bool osd_req_opcode_valid(u16 opcode)
665 {
666 	switch (opcode) {
667 #define GENERATE_CASE(op, opcode, str)	case CEPH_OSD_OP_##op: return true;
668 __CEPH_FORALL_OSD_OPS(GENERATE_CASE)
669 #undef GENERATE_CASE
670 	default:
671 		return false;
672 	}
673 }
674 
675 static void get_num_data_items(struct ceph_osd_request *req,
676 			       int *num_request_data_items,
677 			       int *num_reply_data_items)
678 {
679 	struct ceph_osd_req_op *op;
680 
681 	*num_request_data_items = 0;
682 	*num_reply_data_items = 0;
683 
684 	for (op = req->r_ops; op != &req->r_ops[req->r_num_ops]; op++) {
685 		switch (op->op) {
686 		/* request */
687 		case CEPH_OSD_OP_WRITE:
688 		case CEPH_OSD_OP_WRITEFULL:
689 		case CEPH_OSD_OP_SETXATTR:
690 		case CEPH_OSD_OP_CMPXATTR:
691 		case CEPH_OSD_OP_NOTIFY_ACK:
692 		case CEPH_OSD_OP_COPY_FROM2:
693 			*num_request_data_items += 1;
694 			break;
695 
696 		/* reply */
697 		case CEPH_OSD_OP_STAT:
698 		case CEPH_OSD_OP_READ:
699 		case CEPH_OSD_OP_SPARSE_READ:
700 		case CEPH_OSD_OP_LIST_WATCHERS:
701 			*num_reply_data_items += 1;
702 			break;
703 
704 		/* both */
705 		case CEPH_OSD_OP_NOTIFY:
706 			*num_request_data_items += 1;
707 			*num_reply_data_items += 1;
708 			break;
709 		case CEPH_OSD_OP_CALL:
710 			*num_request_data_items += 2;
711 			*num_reply_data_items += 1;
712 			break;
713 
714 		default:
715 			WARN_ON(!osd_req_opcode_valid(op->op));
716 			break;
717 		}
718 	}
719 }
720 
721 /*
722  * oid, oloc and OSD op opcode(s) must be filled in before this function
723  * is called.
724  */
725 int ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp)
726 {
727 	int num_request_data_items, num_reply_data_items;
728 
729 	get_num_data_items(req, &num_request_data_items, &num_reply_data_items);
730 	return __ceph_osdc_alloc_messages(req, gfp, num_request_data_items,
731 					  num_reply_data_items);
732 }
733 EXPORT_SYMBOL(ceph_osdc_alloc_messages);
734 
735 /*
736  * This is an osd op init function for opcodes that have no data or
737  * other information associated with them.  It also serves as a
738  * common init routine for all the other init functions, below.
739  */
740 struct ceph_osd_req_op *
741 osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which,
742 		 u16 opcode, u32 flags)
743 {
744 	struct ceph_osd_req_op *op;
745 
746 	BUG_ON(which >= osd_req->r_num_ops);
747 	BUG_ON(!osd_req_opcode_valid(opcode));
748 
749 	op = &osd_req->r_ops[which];
750 	memset(op, 0, sizeof (*op));
751 	op->op = opcode;
752 	op->flags = flags;
753 
754 	return op;
755 }
756 EXPORT_SYMBOL(osd_req_op_init);
757 
758 void osd_req_op_extent_init(struct ceph_osd_request *osd_req,
759 				unsigned int which, u16 opcode,
760 				u64 offset, u64 length,
761 				u64 truncate_size, u32 truncate_seq)
762 {
763 	struct ceph_osd_req_op *op = osd_req_op_init(osd_req, which,
764 						     opcode, 0);
765 	size_t payload_len = 0;
766 
767 	BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
768 	       opcode != CEPH_OSD_OP_WRITEFULL && opcode != CEPH_OSD_OP_ZERO &&
769 	       opcode != CEPH_OSD_OP_TRUNCATE && opcode != CEPH_OSD_OP_SPARSE_READ);
770 
771 	op->extent.offset = offset;
772 	op->extent.length = length;
773 	op->extent.truncate_size = truncate_size;
774 	op->extent.truncate_seq = truncate_seq;
775 	if (opcode == CEPH_OSD_OP_WRITE || opcode == CEPH_OSD_OP_WRITEFULL)
776 		payload_len += length;
777 
778 	op->indata_len = payload_len;
779 }
780 EXPORT_SYMBOL(osd_req_op_extent_init);
781 
782 void osd_req_op_extent_update(struct ceph_osd_request *osd_req,
783 				unsigned int which, u64 length)
784 {
785 	struct ceph_osd_req_op *op;
786 	u64 previous;
787 
788 	BUG_ON(which >= osd_req->r_num_ops);
789 	op = &osd_req->r_ops[which];
790 	previous = op->extent.length;
791 
792 	if (length == previous)
793 		return;		/* Nothing to do */
794 	BUG_ON(length > previous);
795 
796 	op->extent.length = length;
797 	if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL)
798 		op->indata_len -= previous - length;
799 }
800 EXPORT_SYMBOL(osd_req_op_extent_update);
801 
802 void osd_req_op_extent_dup_last(struct ceph_osd_request *osd_req,
803 				unsigned int which, u64 offset_inc)
804 {
805 	struct ceph_osd_req_op *op, *prev_op;
806 
807 	BUG_ON(which + 1 >= osd_req->r_num_ops);
808 
809 	prev_op = &osd_req->r_ops[which];
810 	op = osd_req_op_init(osd_req, which + 1, prev_op->op, prev_op->flags);
811 	/* dup previous one */
812 	op->indata_len = prev_op->indata_len;
813 	op->outdata_len = prev_op->outdata_len;
814 	op->extent = prev_op->extent;
815 	/* adjust offset */
816 	op->extent.offset += offset_inc;
817 	op->extent.length -= offset_inc;
818 
819 	if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL)
820 		op->indata_len -= offset_inc;
821 }
822 EXPORT_SYMBOL(osd_req_op_extent_dup_last);
823 
824 int osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
825 			const char *class, const char *method)
826 {
827 	struct ceph_osd_req_op *op;
828 	struct ceph_pagelist *pagelist;
829 	size_t payload_len = 0;
830 	size_t size;
831 	int ret;
832 
833 	op = osd_req_op_init(osd_req, which, CEPH_OSD_OP_CALL, 0);
834 
835 	pagelist = ceph_pagelist_alloc(GFP_NOFS);
836 	if (!pagelist)
837 		return -ENOMEM;
838 
839 	op->cls.class_name = class;
840 	size = strlen(class);
841 	BUG_ON(size > (size_t) U8_MAX);
842 	op->cls.class_len = size;
843 	ret = ceph_pagelist_append(pagelist, class, size);
844 	if (ret)
845 		goto err_pagelist_free;
846 	payload_len += size;
847 
848 	op->cls.method_name = method;
849 	size = strlen(method);
850 	BUG_ON(size > (size_t) U8_MAX);
851 	op->cls.method_len = size;
852 	ret = ceph_pagelist_append(pagelist, method, size);
853 	if (ret)
854 		goto err_pagelist_free;
855 	payload_len += size;
856 
857 	osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist);
858 	op->indata_len = payload_len;
859 	return 0;
860 
861 err_pagelist_free:
862 	ceph_pagelist_release(pagelist);
863 	return ret;
864 }
865 EXPORT_SYMBOL(osd_req_op_cls_init);
866 
867 int osd_req_op_xattr_init(struct ceph_osd_request *osd_req, unsigned int which,
868 			  u16 opcode, const char *name, const void *value,
869 			  size_t size, u8 cmp_op, u8 cmp_mode)
870 {
871 	struct ceph_osd_req_op *op = osd_req_op_init(osd_req, which,
872 						     opcode, 0);
873 	struct ceph_pagelist *pagelist;
874 	size_t payload_len;
875 	int ret;
876 
877 	BUG_ON(opcode != CEPH_OSD_OP_SETXATTR && opcode != CEPH_OSD_OP_CMPXATTR);
878 
879 	pagelist = ceph_pagelist_alloc(GFP_NOFS);
880 	if (!pagelist)
881 		return -ENOMEM;
882 
883 	payload_len = strlen(name);
884 	op->xattr.name_len = payload_len;
885 	ret = ceph_pagelist_append(pagelist, name, payload_len);
886 	if (ret)
887 		goto err_pagelist_free;
888 
889 	op->xattr.value_len = size;
890 	ret = ceph_pagelist_append(pagelist, value, size);
891 	if (ret)
892 		goto err_pagelist_free;
893 	payload_len += size;
894 
895 	op->xattr.cmp_op = cmp_op;
896 	op->xattr.cmp_mode = cmp_mode;
897 
898 	ceph_osd_data_pagelist_init(&op->xattr.osd_data, pagelist);
899 	op->indata_len = payload_len;
900 	return 0;
901 
902 err_pagelist_free:
903 	ceph_pagelist_release(pagelist);
904 	return ret;
905 }
906 EXPORT_SYMBOL(osd_req_op_xattr_init);
907 
908 /*
909  * @watch_opcode: CEPH_OSD_WATCH_OP_*
910  */
911 static void osd_req_op_watch_init(struct ceph_osd_request *req, int which,
912 				  u8 watch_opcode, u64 cookie, u32 gen)
913 {
914 	struct ceph_osd_req_op *op;
915 
916 	op = osd_req_op_init(req, which, CEPH_OSD_OP_WATCH, 0);
917 	op->watch.cookie = cookie;
918 	op->watch.op = watch_opcode;
919 	op->watch.gen = gen;
920 }
921 
922 /*
923  * prot_ver, timeout and notify payload (may be empty) should already be
924  * encoded in @request_pl
925  */
926 static void osd_req_op_notify_init(struct ceph_osd_request *req, int which,
927 				   u64 cookie, struct ceph_pagelist *request_pl)
928 {
929 	struct ceph_osd_req_op *op;
930 
931 	op = osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY, 0);
932 	op->notify.cookie = cookie;
933 
934 	ceph_osd_data_pagelist_init(&op->notify.request_data, request_pl);
935 	op->indata_len = request_pl->length;
936 }
937 
938 /*
939  * @flags: CEPH_OSD_OP_ALLOC_HINT_FLAG_*
940  */
941 void osd_req_op_alloc_hint_init(struct ceph_osd_request *osd_req,
942 				unsigned int which,
943 				u64 expected_object_size,
944 				u64 expected_write_size,
945 				u32 flags)
946 {
947 	struct ceph_osd_req_op *op;
948 
949 	op = osd_req_op_init(osd_req, which, CEPH_OSD_OP_SETALLOCHINT, 0);
950 	op->alloc_hint.expected_object_size = expected_object_size;
951 	op->alloc_hint.expected_write_size = expected_write_size;
952 	op->alloc_hint.flags = flags;
953 
954 	/*
955 	 * CEPH_OSD_OP_SETALLOCHINT op is advisory and therefore deemed
956 	 * not worth a feature bit.  Set FAILOK per-op flag to make
957 	 * sure older osds don't trip over an unsupported opcode.
958 	 */
959 	op->flags |= CEPH_OSD_OP_FLAG_FAILOK;
960 }
961 EXPORT_SYMBOL(osd_req_op_alloc_hint_init);
962 
963 static void ceph_osdc_msg_data_add(struct ceph_msg *msg,
964 				struct ceph_osd_data *osd_data)
965 {
966 	u64 length = ceph_osd_data_length(osd_data);
967 
968 	if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
969 		BUG_ON(length > (u64) SIZE_MAX);
970 		if (length)
971 			ceph_msg_data_add_pages(msg, osd_data->pages,
972 					length, osd_data->alignment, false);
973 	} else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
974 		BUG_ON(!length);
975 		ceph_msg_data_add_pagelist(msg, osd_data->pagelist);
976 #ifdef CONFIG_BLOCK
977 	} else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
978 		ceph_msg_data_add_bio(msg, &osd_data->bio_pos, length);
979 #endif
980 	} else if (osd_data->type == CEPH_OSD_DATA_TYPE_BVECS) {
981 		ceph_msg_data_add_bvecs(msg, &osd_data->bvec_pos);
982 	} else if (osd_data->type == CEPH_OSD_DATA_TYPE_ITER) {
983 		ceph_msg_data_add_iter(msg, &osd_data->iter);
984 	} else {
985 		BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
986 	}
987 }
988 
989 static u32 osd_req_encode_op(struct ceph_osd_op *dst,
990 			     const struct ceph_osd_req_op *src)
991 {
992 	switch (src->op) {
993 	case CEPH_OSD_OP_STAT:
994 		break;
995 	case CEPH_OSD_OP_READ:
996 	case CEPH_OSD_OP_SPARSE_READ:
997 	case CEPH_OSD_OP_WRITE:
998 	case CEPH_OSD_OP_WRITEFULL:
999 	case CEPH_OSD_OP_ZERO:
1000 	case CEPH_OSD_OP_TRUNCATE:
1001 		dst->extent.offset = cpu_to_le64(src->extent.offset);
1002 		dst->extent.length = cpu_to_le64(src->extent.length);
1003 		dst->extent.truncate_size =
1004 			cpu_to_le64(src->extent.truncate_size);
1005 		dst->extent.truncate_seq =
1006 			cpu_to_le32(src->extent.truncate_seq);
1007 		break;
1008 	case CEPH_OSD_OP_CALL:
1009 		dst->cls.class_len = src->cls.class_len;
1010 		dst->cls.method_len = src->cls.method_len;
1011 		dst->cls.indata_len = cpu_to_le32(src->cls.indata_len);
1012 		break;
1013 	case CEPH_OSD_OP_WATCH:
1014 		dst->watch.cookie = cpu_to_le64(src->watch.cookie);
1015 		dst->watch.ver = cpu_to_le64(0);
1016 		dst->watch.op = src->watch.op;
1017 		dst->watch.gen = cpu_to_le32(src->watch.gen);
1018 		break;
1019 	case CEPH_OSD_OP_NOTIFY_ACK:
1020 		break;
1021 	case CEPH_OSD_OP_NOTIFY:
1022 		dst->notify.cookie = cpu_to_le64(src->notify.cookie);
1023 		break;
1024 	case CEPH_OSD_OP_LIST_WATCHERS:
1025 		break;
1026 	case CEPH_OSD_OP_SETALLOCHINT:
1027 		dst->alloc_hint.expected_object_size =
1028 		    cpu_to_le64(src->alloc_hint.expected_object_size);
1029 		dst->alloc_hint.expected_write_size =
1030 		    cpu_to_le64(src->alloc_hint.expected_write_size);
1031 		dst->alloc_hint.flags = cpu_to_le32(src->alloc_hint.flags);
1032 		break;
1033 	case CEPH_OSD_OP_SETXATTR:
1034 	case CEPH_OSD_OP_CMPXATTR:
1035 		dst->xattr.name_len = cpu_to_le32(src->xattr.name_len);
1036 		dst->xattr.value_len = cpu_to_le32(src->xattr.value_len);
1037 		dst->xattr.cmp_op = src->xattr.cmp_op;
1038 		dst->xattr.cmp_mode = src->xattr.cmp_mode;
1039 		break;
1040 	case CEPH_OSD_OP_CREATE:
1041 	case CEPH_OSD_OP_DELETE:
1042 		break;
1043 	case CEPH_OSD_OP_COPY_FROM2:
1044 		dst->copy_from.snapid = cpu_to_le64(src->copy_from.snapid);
1045 		dst->copy_from.src_version =
1046 			cpu_to_le64(src->copy_from.src_version);
1047 		dst->copy_from.flags = src->copy_from.flags;
1048 		dst->copy_from.src_fadvise_flags =
1049 			cpu_to_le32(src->copy_from.src_fadvise_flags);
1050 		break;
1051 	case CEPH_OSD_OP_ASSERT_VER:
1052 		dst->assert_ver.unused = cpu_to_le64(0);
1053 		dst->assert_ver.ver = cpu_to_le64(src->assert_ver.ver);
1054 		break;
1055 	default:
1056 		pr_err("unsupported osd opcode %s\n",
1057 			ceph_osd_op_name(src->op));
1058 		WARN_ON(1);
1059 
1060 		return 0;
1061 	}
1062 
1063 	dst->op = cpu_to_le16(src->op);
1064 	dst->flags = cpu_to_le32(src->flags);
1065 	dst->payload_len = cpu_to_le32(src->indata_len);
1066 
1067 	return src->indata_len;
1068 }
1069 
1070 /*
1071  * build new request AND message, calculate layout, and adjust file
1072  * extent as needed.
1073  *
1074  * if the file was recently truncated, we include information about its
1075  * old and new size so that the object can be updated appropriately.  (we
1076  * avoid synchronously deleting truncated objects because it's slow.)
1077  */
1078 struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
1079 					       struct ceph_file_layout *layout,
1080 					       struct ceph_vino vino,
1081 					       u64 off, u64 *plen,
1082 					       unsigned int which, int num_ops,
1083 					       int opcode, int flags,
1084 					       struct ceph_snap_context *snapc,
1085 					       u32 truncate_seq,
1086 					       u64 truncate_size,
1087 					       bool use_mempool)
1088 {
1089 	struct ceph_osd_request *req;
1090 	u64 objnum = 0;
1091 	u64 objoff = 0;
1092 	u64 objlen = 0;
1093 	int r;
1094 
1095 	BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
1096 	       opcode != CEPH_OSD_OP_ZERO && opcode != CEPH_OSD_OP_TRUNCATE &&
1097 	       opcode != CEPH_OSD_OP_CREATE && opcode != CEPH_OSD_OP_DELETE &&
1098 	       opcode != CEPH_OSD_OP_SPARSE_READ);
1099 
1100 	req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool,
1101 					GFP_NOFS);
1102 	if (!req) {
1103 		r = -ENOMEM;
1104 		goto fail;
1105 	}
1106 
1107 	/* calculate max write size */
1108 	r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
1109 	if (r)
1110 		goto fail;
1111 
1112 	if (opcode == CEPH_OSD_OP_CREATE || opcode == CEPH_OSD_OP_DELETE) {
1113 		osd_req_op_init(req, which, opcode, 0);
1114 	} else {
1115 		u32 object_size = layout->object_size;
1116 		u32 object_base = off - objoff;
1117 		if (!(truncate_seq == 1 && truncate_size == -1ULL)) {
1118 			if (truncate_size <= object_base) {
1119 				truncate_size = 0;
1120 			} else {
1121 				truncate_size -= object_base;
1122 				if (truncate_size > object_size)
1123 					truncate_size = object_size;
1124 			}
1125 		}
1126 		osd_req_op_extent_init(req, which, opcode, objoff, objlen,
1127 				       truncate_size, truncate_seq);
1128 	}
1129 
1130 	req->r_base_oloc.pool = layout->pool_id;
1131 	req->r_base_oloc.pool_ns = ceph_try_get_string(layout->pool_ns);
1132 	ceph_oid_printf(&req->r_base_oid, "%llx.%08llx", vino.ino, objnum);
1133 	req->r_flags = flags | osdc->client->options->read_from_replica;
1134 
1135 	req->r_snapid = vino.snap;
1136 	if (flags & CEPH_OSD_FLAG_WRITE)
1137 		req->r_data_offset = off;
1138 
1139 	if (num_ops > 1) {
1140 		int num_req_ops, num_rep_ops;
1141 
1142 		/*
1143 		 * If this is a multi-op write request, assume that we'll need
1144 		 * request ops. If it's a multi-op read then assume we'll need
1145 		 * reply ops. Anything else and call it -EINVAL.
1146 		 */
1147 		if (flags & CEPH_OSD_FLAG_WRITE) {
1148 			num_req_ops = num_ops;
1149 			num_rep_ops = 0;
1150 		} else if (flags & CEPH_OSD_FLAG_READ) {
1151 			num_req_ops = 0;
1152 			num_rep_ops = num_ops;
1153 		} else {
1154 			r = -EINVAL;
1155 			goto fail;
1156 		}
1157 
1158 		r = __ceph_osdc_alloc_messages(req, GFP_NOFS, num_req_ops,
1159 					       num_rep_ops);
1160 	} else {
1161 		r = ceph_osdc_alloc_messages(req, GFP_NOFS);
1162 	}
1163 	if (r)
1164 		goto fail;
1165 
1166 	return req;
1167 
1168 fail:
1169 	ceph_osdc_put_request(req);
1170 	return ERR_PTR(r);
1171 }
1172 EXPORT_SYMBOL(ceph_osdc_new_request);
1173 
1174 int __ceph_alloc_sparse_ext_map(struct ceph_osd_req_op *op, int cnt)
1175 {
1176 	op->extent.sparse_ext_cnt = cnt;
1177 	op->extent.sparse_ext = kmalloc_array(cnt,
1178 					      sizeof(*op->extent.sparse_ext),
1179 					      GFP_NOFS);
1180 	if (!op->extent.sparse_ext)
1181 		return -ENOMEM;
1182 	return 0;
1183 }
1184 EXPORT_SYMBOL(__ceph_alloc_sparse_ext_map);
1185 
1186 /*
1187  * We keep osd requests in an rbtree, sorted by ->r_tid.
1188  */
1189 DEFINE_RB_FUNCS(request, struct ceph_osd_request, r_tid, r_node)
1190 DEFINE_RB_FUNCS(request_mc, struct ceph_osd_request, r_tid, r_mc_node)
1191 
1192 /*
1193  * Call @fn on each OSD request as long as @fn returns 0.
1194  */
1195 static void for_each_request(struct ceph_osd_client *osdc,
1196 			int (*fn)(struct ceph_osd_request *req, void *arg),
1197 			void *arg)
1198 {
1199 	struct rb_node *n, *p;
1200 
1201 	for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
1202 		struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
1203 
1204 		for (p = rb_first(&osd->o_requests); p; ) {
1205 			struct ceph_osd_request *req =
1206 			    rb_entry(p, struct ceph_osd_request, r_node);
1207 
1208 			p = rb_next(p);
1209 			if (fn(req, arg))
1210 				return;
1211 		}
1212 	}
1213 
1214 	for (p = rb_first(&osdc->homeless_osd.o_requests); p; ) {
1215 		struct ceph_osd_request *req =
1216 		    rb_entry(p, struct ceph_osd_request, r_node);
1217 
1218 		p = rb_next(p);
1219 		if (fn(req, arg))
1220 			return;
1221 	}
1222 }
1223 
1224 static bool osd_homeless(struct ceph_osd *osd)
1225 {
1226 	return osd->o_osd == CEPH_HOMELESS_OSD;
1227 }
1228 
1229 static bool osd_registered(struct ceph_osd *osd)
1230 {
1231 	verify_osdc_locked(osd->o_osdc);
1232 
1233 	return !RB_EMPTY_NODE(&osd->o_node);
1234 }
1235 
1236 /*
1237  * Assumes @osd is zero-initialized.
1238  */
1239 static void osd_init(struct ceph_osd *osd)
1240 {
1241 	refcount_set(&osd->o_ref, 1);
1242 	RB_CLEAR_NODE(&osd->o_node);
1243 	spin_lock_init(&osd->o_requests_lock);
1244 	osd->o_requests = RB_ROOT;
1245 	osd->o_linger_requests = RB_ROOT;
1246 	osd->o_backoff_mappings = RB_ROOT;
1247 	osd->o_backoffs_by_id = RB_ROOT;
1248 	INIT_LIST_HEAD(&osd->o_osd_lru);
1249 	INIT_LIST_HEAD(&osd->o_keepalive_item);
1250 	osd->o_incarnation = 1;
1251 	mutex_init(&osd->lock);
1252 }
1253 
1254 static void ceph_init_sparse_read(struct ceph_sparse_read *sr)
1255 {
1256 	kfree(sr->sr_extent);
1257 	memset(sr, '\0', sizeof(*sr));
1258 	sr->sr_state = CEPH_SPARSE_READ_HDR;
1259 }
1260 
1261 static void osd_cleanup(struct ceph_osd *osd)
1262 {
1263 	WARN_ON(!RB_EMPTY_NODE(&osd->o_node));
1264 	WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests));
1265 	WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests));
1266 	WARN_ON(!RB_EMPTY_ROOT(&osd->o_backoff_mappings));
1267 	WARN_ON(!RB_EMPTY_ROOT(&osd->o_backoffs_by_id));
1268 	WARN_ON(!list_empty(&osd->o_osd_lru));
1269 	WARN_ON(!list_empty(&osd->o_keepalive_item));
1270 
1271 	ceph_init_sparse_read(&osd->o_sparse_read);
1272 
1273 	if (osd->o_auth.authorizer) {
1274 		WARN_ON(osd_homeless(osd));
1275 		ceph_auth_destroy_authorizer(osd->o_auth.authorizer);
1276 	}
1277 }
1278 
1279 /*
1280  * Track open sessions with osds.
1281  */
1282 static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
1283 {
1284 	struct ceph_osd *osd;
1285 
1286 	WARN_ON(onum == CEPH_HOMELESS_OSD);
1287 
1288 	osd = kzalloc(sizeof(*osd), GFP_NOIO | __GFP_NOFAIL);
1289 	osd_init(osd);
1290 	osd->o_osdc = osdc;
1291 	osd->o_osd = onum;
1292 	osd->o_sparse_op_idx = -1;
1293 
1294 	ceph_init_sparse_read(&osd->o_sparse_read);
1295 
1296 	ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
1297 
1298 	return osd;
1299 }
1300 
1301 static struct ceph_osd *get_osd(struct ceph_osd *osd)
1302 {
1303 	if (refcount_inc_not_zero(&osd->o_ref)) {
1304 		dout("get_osd %p %d -> %d\n", osd, refcount_read(&osd->o_ref)-1,
1305 		     refcount_read(&osd->o_ref));
1306 		return osd;
1307 	} else {
1308 		dout("get_osd %p FAIL\n", osd);
1309 		return NULL;
1310 	}
1311 }
1312 
1313 static void put_osd(struct ceph_osd *osd)
1314 {
1315 	dout("put_osd %p %d -> %d\n", osd, refcount_read(&osd->o_ref),
1316 	     refcount_read(&osd->o_ref) - 1);
1317 	if (refcount_dec_and_test(&osd->o_ref)) {
1318 		osd_cleanup(osd);
1319 		kfree(osd);
1320 	}
1321 }
1322 
1323 DEFINE_RB_FUNCS(osd, struct ceph_osd, o_osd, o_node)
1324 
1325 static void __move_osd_to_lru(struct ceph_osd *osd)
1326 {
1327 	struct ceph_osd_client *osdc = osd->o_osdc;
1328 
1329 	dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1330 	BUG_ON(!list_empty(&osd->o_osd_lru));
1331 
1332 	spin_lock(&osdc->osd_lru_lock);
1333 	list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
1334 	spin_unlock(&osdc->osd_lru_lock);
1335 
1336 	osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl;
1337 }
1338 
1339 static void maybe_move_osd_to_lru(struct ceph_osd *osd)
1340 {
1341 	if (RB_EMPTY_ROOT(&osd->o_requests) &&
1342 	    RB_EMPTY_ROOT(&osd->o_linger_requests))
1343 		__move_osd_to_lru(osd);
1344 }
1345 
1346 static void __remove_osd_from_lru(struct ceph_osd *osd)
1347 {
1348 	struct ceph_osd_client *osdc = osd->o_osdc;
1349 
1350 	dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1351 
1352 	spin_lock(&osdc->osd_lru_lock);
1353 	if (!list_empty(&osd->o_osd_lru))
1354 		list_del_init(&osd->o_osd_lru);
1355 	spin_unlock(&osdc->osd_lru_lock);
1356 }
1357 
1358 /*
1359  * Close the connection and assign any leftover requests to the
1360  * homeless session.
1361  */
1362 static void close_osd(struct ceph_osd *osd)
1363 {
1364 	struct ceph_osd_client *osdc = osd->o_osdc;
1365 	struct rb_node *n;
1366 
1367 	verify_osdc_wrlocked(osdc);
1368 	dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1369 
1370 	ceph_con_close(&osd->o_con);
1371 
1372 	for (n = rb_first(&osd->o_requests); n; ) {
1373 		struct ceph_osd_request *req =
1374 		    rb_entry(n, struct ceph_osd_request, r_node);
1375 
1376 		n = rb_next(n); /* unlink_request() */
1377 
1378 		dout(" reassigning req %p tid %llu\n", req, req->r_tid);
1379 		unlink_request(osd, req);
1380 		link_request(&osdc->homeless_osd, req);
1381 	}
1382 	for (n = rb_first(&osd->o_linger_requests); n; ) {
1383 		struct ceph_osd_linger_request *lreq =
1384 		    rb_entry(n, struct ceph_osd_linger_request, node);
1385 
1386 		n = rb_next(n); /* unlink_linger() */
1387 
1388 		dout(" reassigning lreq %p linger_id %llu\n", lreq,
1389 		     lreq->linger_id);
1390 		unlink_linger(osd, lreq);
1391 		link_linger(&osdc->homeless_osd, lreq);
1392 	}
1393 	clear_backoffs(osd);
1394 
1395 	__remove_osd_from_lru(osd);
1396 	erase_osd(&osdc->osds, osd);
1397 	put_osd(osd);
1398 }
1399 
1400 /*
1401  * reset osd connect
1402  */
1403 static int reopen_osd(struct ceph_osd *osd)
1404 {
1405 	struct ceph_entity_addr *peer_addr;
1406 
1407 	dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1408 
1409 	if (RB_EMPTY_ROOT(&osd->o_requests) &&
1410 	    RB_EMPTY_ROOT(&osd->o_linger_requests)) {
1411 		close_osd(osd);
1412 		return -ENODEV;
1413 	}
1414 
1415 	peer_addr = &osd->o_osdc->osdmap->osd_addr[osd->o_osd];
1416 	if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1417 			!ceph_con_opened(&osd->o_con)) {
1418 		struct rb_node *n;
1419 
1420 		dout("osd addr hasn't changed and connection never opened, "
1421 		     "letting msgr retry\n");
1422 		/* touch each r_stamp for handle_timeout()'s benfit */
1423 		for (n = rb_first(&osd->o_requests); n; n = rb_next(n)) {
1424 			struct ceph_osd_request *req =
1425 			    rb_entry(n, struct ceph_osd_request, r_node);
1426 			req->r_stamp = jiffies;
1427 		}
1428 
1429 		return -EAGAIN;
1430 	}
1431 
1432 	ceph_con_close(&osd->o_con);
1433 	ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1434 	osd->o_incarnation++;
1435 
1436 	return 0;
1437 }
1438 
1439 static struct ceph_osd *lookup_create_osd(struct ceph_osd_client *osdc, int o,
1440 					  bool wrlocked)
1441 {
1442 	struct ceph_osd *osd;
1443 
1444 	if (wrlocked)
1445 		verify_osdc_wrlocked(osdc);
1446 	else
1447 		verify_osdc_locked(osdc);
1448 
1449 	if (o != CEPH_HOMELESS_OSD)
1450 		osd = lookup_osd(&osdc->osds, o);
1451 	else
1452 		osd = &osdc->homeless_osd;
1453 	if (!osd) {
1454 		if (!wrlocked)
1455 			return ERR_PTR(-EAGAIN);
1456 
1457 		osd = create_osd(osdc, o);
1458 		insert_osd(&osdc->osds, osd);
1459 		ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd,
1460 			      &osdc->osdmap->osd_addr[osd->o_osd]);
1461 	}
1462 
1463 	dout("%s osdc %p osd%d -> osd %p\n", __func__, osdc, o, osd);
1464 	return osd;
1465 }
1466 
1467 /*
1468  * Create request <-> OSD session relation.
1469  *
1470  * @req has to be assigned a tid, @osd may be homeless.
1471  */
1472 static void link_request(struct ceph_osd *osd, struct ceph_osd_request *req)
1473 {
1474 	verify_osd_locked(osd);
1475 	WARN_ON(!req->r_tid || req->r_osd);
1476 	dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd,
1477 	     req, req->r_tid);
1478 
1479 	if (!osd_homeless(osd))
1480 		__remove_osd_from_lru(osd);
1481 	else
1482 		atomic_inc(&osd->o_osdc->num_homeless);
1483 
1484 	get_osd(osd);
1485 	spin_lock(&osd->o_requests_lock);
1486 	insert_request(&osd->o_requests, req);
1487 	spin_unlock(&osd->o_requests_lock);
1488 	req->r_osd = osd;
1489 }
1490 
1491 static void unlink_request(struct ceph_osd *osd, struct ceph_osd_request *req)
1492 {
1493 	verify_osd_locked(osd);
1494 	WARN_ON(req->r_osd != osd);
1495 	dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd,
1496 	     req, req->r_tid);
1497 
1498 	req->r_osd = NULL;
1499 	spin_lock(&osd->o_requests_lock);
1500 	erase_request(&osd->o_requests, req);
1501 	spin_unlock(&osd->o_requests_lock);
1502 	put_osd(osd);
1503 
1504 	if (!osd_homeless(osd))
1505 		maybe_move_osd_to_lru(osd);
1506 	else
1507 		atomic_dec(&osd->o_osdc->num_homeless);
1508 }
1509 
1510 static bool __pool_full(struct ceph_pg_pool_info *pi)
1511 {
1512 	return pi->flags & CEPH_POOL_FLAG_FULL;
1513 }
1514 
1515 static bool have_pool_full(struct ceph_osd_client *osdc)
1516 {
1517 	struct rb_node *n;
1518 
1519 	for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) {
1520 		struct ceph_pg_pool_info *pi =
1521 		    rb_entry(n, struct ceph_pg_pool_info, node);
1522 
1523 		if (__pool_full(pi))
1524 			return true;
1525 	}
1526 
1527 	return false;
1528 }
1529 
1530 static bool pool_full(struct ceph_osd_client *osdc, s64 pool_id)
1531 {
1532 	struct ceph_pg_pool_info *pi;
1533 
1534 	pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id);
1535 	if (!pi)
1536 		return false;
1537 
1538 	return __pool_full(pi);
1539 }
1540 
1541 /*
1542  * Returns whether a request should be blocked from being sent
1543  * based on the current osdmap and osd_client settings.
1544  */
1545 static bool target_should_be_paused(struct ceph_osd_client *osdc,
1546 				    const struct ceph_osd_request_target *t,
1547 				    struct ceph_pg_pool_info *pi)
1548 {
1549 	bool pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
1550 	bool pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
1551 		       ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
1552 		       __pool_full(pi);
1553 
1554 	WARN_ON(pi->id != t->target_oloc.pool);
1555 	return ((t->flags & CEPH_OSD_FLAG_READ) && pauserd) ||
1556 	       ((t->flags & CEPH_OSD_FLAG_WRITE) && pausewr) ||
1557 	       (osdc->osdmap->epoch < osdc->epoch_barrier);
1558 }
1559 
1560 static int pick_random_replica(const struct ceph_osds *acting)
1561 {
1562 	int i = get_random_u32_below(acting->size);
1563 
1564 	dout("%s picked osd%d, primary osd%d\n", __func__,
1565 	     acting->osds[i], acting->primary);
1566 	return i;
1567 }
1568 
1569 /*
1570  * Picks the closest replica based on client's location given by
1571  * crush_location option.  Prefers the primary if the locality is
1572  * the same.
1573  */
1574 static int pick_closest_replica(struct ceph_osd_client *osdc,
1575 				const struct ceph_osds *acting)
1576 {
1577 	struct ceph_options *opt = osdc->client->options;
1578 	int best_i, best_locality;
1579 	int i = 0, locality;
1580 
1581 	do {
1582 		locality = ceph_get_crush_locality(osdc->osdmap,
1583 						   acting->osds[i],
1584 						   &opt->crush_locs);
1585 		if (i == 0 ||
1586 		    (locality >= 0 && best_locality < 0) ||
1587 		    (locality >= 0 && best_locality >= 0 &&
1588 		     locality < best_locality)) {
1589 			best_i = i;
1590 			best_locality = locality;
1591 		}
1592 	} while (++i < acting->size);
1593 
1594 	dout("%s picked osd%d with locality %d, primary osd%d\n", __func__,
1595 	     acting->osds[best_i], best_locality, acting->primary);
1596 	return best_i;
1597 }
1598 
1599 enum calc_target_result {
1600 	CALC_TARGET_NO_ACTION = 0,
1601 	CALC_TARGET_NEED_RESEND,
1602 	CALC_TARGET_POOL_DNE,
1603 };
1604 
1605 static enum calc_target_result calc_target(struct ceph_osd_client *osdc,
1606 					   struct ceph_osd_request_target *t,
1607 					   bool any_change)
1608 {
1609 	struct ceph_pg_pool_info *pi;
1610 	struct ceph_pg pgid, last_pgid;
1611 	struct ceph_osds up, acting;
1612 	bool is_read = t->flags & CEPH_OSD_FLAG_READ;
1613 	bool is_write = t->flags & CEPH_OSD_FLAG_WRITE;
1614 	bool force_resend = false;
1615 	bool unpaused = false;
1616 	bool legacy_change = false;
1617 	bool split = false;
1618 	bool sort_bitwise = ceph_osdmap_flag(osdc, CEPH_OSDMAP_SORTBITWISE);
1619 	bool recovery_deletes = ceph_osdmap_flag(osdc,
1620 						 CEPH_OSDMAP_RECOVERY_DELETES);
1621 	enum calc_target_result ct_res;
1622 
1623 	t->epoch = osdc->osdmap->epoch;
1624 	pi = ceph_pg_pool_by_id(osdc->osdmap, t->base_oloc.pool);
1625 	if (!pi) {
1626 		t->osd = CEPH_HOMELESS_OSD;
1627 		ct_res = CALC_TARGET_POOL_DNE;
1628 		goto out;
1629 	}
1630 
1631 	if (osdc->osdmap->epoch == pi->last_force_request_resend) {
1632 		if (t->last_force_resend < pi->last_force_request_resend) {
1633 			t->last_force_resend = pi->last_force_request_resend;
1634 			force_resend = true;
1635 		} else if (t->last_force_resend == 0) {
1636 			force_resend = true;
1637 		}
1638 	}
1639 
1640 	/* apply tiering */
1641 	ceph_oid_copy(&t->target_oid, &t->base_oid);
1642 	ceph_oloc_copy(&t->target_oloc, &t->base_oloc);
1643 	if ((t->flags & CEPH_OSD_FLAG_IGNORE_OVERLAY) == 0) {
1644 		if (is_read && pi->read_tier >= 0)
1645 			t->target_oloc.pool = pi->read_tier;
1646 		if (is_write && pi->write_tier >= 0)
1647 			t->target_oloc.pool = pi->write_tier;
1648 
1649 		pi = ceph_pg_pool_by_id(osdc->osdmap, t->target_oloc.pool);
1650 		if (!pi) {
1651 			t->osd = CEPH_HOMELESS_OSD;
1652 			ct_res = CALC_TARGET_POOL_DNE;
1653 			goto out;
1654 		}
1655 	}
1656 
1657 	__ceph_object_locator_to_pg(pi, &t->target_oid, &t->target_oloc, &pgid);
1658 	last_pgid.pool = pgid.pool;
1659 	last_pgid.seed = ceph_stable_mod(pgid.seed, t->pg_num, t->pg_num_mask);
1660 
1661 	ceph_pg_to_up_acting_osds(osdc->osdmap, pi, &pgid, &up, &acting);
1662 	if (any_change &&
1663 	    ceph_is_new_interval(&t->acting,
1664 				 &acting,
1665 				 &t->up,
1666 				 &up,
1667 				 t->size,
1668 				 pi->size,
1669 				 t->min_size,
1670 				 pi->min_size,
1671 				 t->pg_num,
1672 				 pi->pg_num,
1673 				 t->sort_bitwise,
1674 				 sort_bitwise,
1675 				 t->recovery_deletes,
1676 				 recovery_deletes,
1677 				 &last_pgid))
1678 		force_resend = true;
1679 
1680 	if (t->paused && !target_should_be_paused(osdc, t, pi)) {
1681 		t->paused = false;
1682 		unpaused = true;
1683 	}
1684 	legacy_change = ceph_pg_compare(&t->pgid, &pgid) ||
1685 			ceph_osds_changed(&t->acting, &acting,
1686 					  t->used_replica || any_change);
1687 	if (t->pg_num)
1688 		split = ceph_pg_is_split(&last_pgid, t->pg_num, pi->pg_num);
1689 
1690 	if (legacy_change || force_resend || split) {
1691 		t->pgid = pgid; /* struct */
1692 		ceph_pg_to_primary_shard(osdc->osdmap, pi, &pgid, &t->spgid);
1693 		ceph_osds_copy(&t->acting, &acting);
1694 		ceph_osds_copy(&t->up, &up);
1695 		t->size = pi->size;
1696 		t->min_size = pi->min_size;
1697 		t->pg_num = pi->pg_num;
1698 		t->pg_num_mask = pi->pg_num_mask;
1699 		t->sort_bitwise = sort_bitwise;
1700 		t->recovery_deletes = recovery_deletes;
1701 
1702 		if ((t->flags & (CEPH_OSD_FLAG_BALANCE_READS |
1703 				 CEPH_OSD_FLAG_LOCALIZE_READS)) &&
1704 		    !is_write && pi->type == CEPH_POOL_TYPE_REP &&
1705 		    acting.size > 1) {
1706 			int pos;
1707 
1708 			WARN_ON(!is_read || acting.osds[0] != acting.primary);
1709 			if (t->flags & CEPH_OSD_FLAG_BALANCE_READS) {
1710 				pos = pick_random_replica(&acting);
1711 			} else {
1712 				pos = pick_closest_replica(osdc, &acting);
1713 			}
1714 			t->osd = acting.osds[pos];
1715 			t->used_replica = pos > 0;
1716 		} else {
1717 			t->osd = acting.primary;
1718 			t->used_replica = false;
1719 		}
1720 	}
1721 
1722 	if (unpaused || legacy_change || force_resend || split)
1723 		ct_res = CALC_TARGET_NEED_RESEND;
1724 	else
1725 		ct_res = CALC_TARGET_NO_ACTION;
1726 
1727 out:
1728 	dout("%s t %p -> %d%d%d%d ct_res %d osd%d\n", __func__, t, unpaused,
1729 	     legacy_change, force_resend, split, ct_res, t->osd);
1730 	return ct_res;
1731 }
1732 
1733 static struct ceph_spg_mapping *alloc_spg_mapping(void)
1734 {
1735 	struct ceph_spg_mapping *spg;
1736 
1737 	spg = kmalloc(sizeof(*spg), GFP_NOIO);
1738 	if (!spg)
1739 		return NULL;
1740 
1741 	RB_CLEAR_NODE(&spg->node);
1742 	spg->backoffs = RB_ROOT;
1743 	return spg;
1744 }
1745 
1746 static void free_spg_mapping(struct ceph_spg_mapping *spg)
1747 {
1748 	WARN_ON(!RB_EMPTY_NODE(&spg->node));
1749 	WARN_ON(!RB_EMPTY_ROOT(&spg->backoffs));
1750 
1751 	kfree(spg);
1752 }
1753 
1754 /*
1755  * rbtree of ceph_spg_mapping for handling map<spg_t, ...>, similar to
1756  * ceph_pg_mapping.  Used to track OSD backoffs -- a backoff [range] is
1757  * defined only within a specific spgid; it does not pass anything to
1758  * children on split, or to another primary.
1759  */
1760 DEFINE_RB_FUNCS2(spg_mapping, struct ceph_spg_mapping, spgid, ceph_spg_compare,
1761 		 RB_BYPTR, const struct ceph_spg *, node)
1762 
1763 static u64 hoid_get_bitwise_key(const struct ceph_hobject_id *hoid)
1764 {
1765 	return hoid->is_max ? 0x100000000ull : hoid->hash_reverse_bits;
1766 }
1767 
1768 static void hoid_get_effective_key(const struct ceph_hobject_id *hoid,
1769 				   void **pkey, size_t *pkey_len)
1770 {
1771 	if (hoid->key_len) {
1772 		*pkey = hoid->key;
1773 		*pkey_len = hoid->key_len;
1774 	} else {
1775 		*pkey = hoid->oid;
1776 		*pkey_len = hoid->oid_len;
1777 	}
1778 }
1779 
1780 static int compare_names(const void *name1, size_t name1_len,
1781 			 const void *name2, size_t name2_len)
1782 {
1783 	int ret;
1784 
1785 	ret = memcmp(name1, name2, min(name1_len, name2_len));
1786 	if (!ret) {
1787 		if (name1_len < name2_len)
1788 			ret = -1;
1789 		else if (name1_len > name2_len)
1790 			ret = 1;
1791 	}
1792 	return ret;
1793 }
1794 
1795 static int hoid_compare(const struct ceph_hobject_id *lhs,
1796 			const struct ceph_hobject_id *rhs)
1797 {
1798 	void *effective_key1, *effective_key2;
1799 	size_t effective_key1_len, effective_key2_len;
1800 	int ret;
1801 
1802 	if (lhs->is_max < rhs->is_max)
1803 		return -1;
1804 	if (lhs->is_max > rhs->is_max)
1805 		return 1;
1806 
1807 	if (lhs->pool < rhs->pool)
1808 		return -1;
1809 	if (lhs->pool > rhs->pool)
1810 		return 1;
1811 
1812 	if (hoid_get_bitwise_key(lhs) < hoid_get_bitwise_key(rhs))
1813 		return -1;
1814 	if (hoid_get_bitwise_key(lhs) > hoid_get_bitwise_key(rhs))
1815 		return 1;
1816 
1817 	ret = compare_names(lhs->nspace, lhs->nspace_len,
1818 			    rhs->nspace, rhs->nspace_len);
1819 	if (ret)
1820 		return ret;
1821 
1822 	hoid_get_effective_key(lhs, &effective_key1, &effective_key1_len);
1823 	hoid_get_effective_key(rhs, &effective_key2, &effective_key2_len);
1824 	ret = compare_names(effective_key1, effective_key1_len,
1825 			    effective_key2, effective_key2_len);
1826 	if (ret)
1827 		return ret;
1828 
1829 	ret = compare_names(lhs->oid, lhs->oid_len, rhs->oid, rhs->oid_len);
1830 	if (ret)
1831 		return ret;
1832 
1833 	if (lhs->snapid < rhs->snapid)
1834 		return -1;
1835 	if (lhs->snapid > rhs->snapid)
1836 		return 1;
1837 
1838 	return 0;
1839 }
1840 
1841 /*
1842  * For decoding ->begin and ->end of MOSDBackoff only -- no MIN/MAX
1843  * compat stuff here.
1844  *
1845  * Assumes @hoid is zero-initialized.
1846  */
1847 static int decode_hoid(void **p, void *end, struct ceph_hobject_id *hoid)
1848 {
1849 	u8 struct_v;
1850 	u32 struct_len;
1851 	int ret;
1852 
1853 	ret = ceph_start_decoding(p, end, 4, "hobject_t", &struct_v,
1854 				  &struct_len);
1855 	if (ret)
1856 		return ret;
1857 
1858 	if (struct_v < 4) {
1859 		pr_err("got struct_v %d < 4 of hobject_t\n", struct_v);
1860 		goto e_inval;
1861 	}
1862 
1863 	hoid->key = ceph_extract_encoded_string(p, end, &hoid->key_len,
1864 						GFP_NOIO);
1865 	if (IS_ERR(hoid->key)) {
1866 		ret = PTR_ERR(hoid->key);
1867 		hoid->key = NULL;
1868 		return ret;
1869 	}
1870 
1871 	hoid->oid = ceph_extract_encoded_string(p, end, &hoid->oid_len,
1872 						GFP_NOIO);
1873 	if (IS_ERR(hoid->oid)) {
1874 		ret = PTR_ERR(hoid->oid);
1875 		hoid->oid = NULL;
1876 		return ret;
1877 	}
1878 
1879 	ceph_decode_64_safe(p, end, hoid->snapid, e_inval);
1880 	ceph_decode_32_safe(p, end, hoid->hash, e_inval);
1881 	ceph_decode_8_safe(p, end, hoid->is_max, e_inval);
1882 
1883 	hoid->nspace = ceph_extract_encoded_string(p, end, &hoid->nspace_len,
1884 						   GFP_NOIO);
1885 	if (IS_ERR(hoid->nspace)) {
1886 		ret = PTR_ERR(hoid->nspace);
1887 		hoid->nspace = NULL;
1888 		return ret;
1889 	}
1890 
1891 	ceph_decode_64_safe(p, end, hoid->pool, e_inval);
1892 
1893 	ceph_hoid_build_hash_cache(hoid);
1894 	return 0;
1895 
1896 e_inval:
1897 	return -EINVAL;
1898 }
1899 
1900 static int hoid_encoding_size(const struct ceph_hobject_id *hoid)
1901 {
1902 	return 8 + 4 + 1 + 8 + /* snapid, hash, is_max, pool */
1903 	       4 + hoid->key_len + 4 + hoid->oid_len + 4 + hoid->nspace_len;
1904 }
1905 
1906 static void encode_hoid(void **p, void *end, const struct ceph_hobject_id *hoid)
1907 {
1908 	ceph_start_encoding(p, 4, 3, hoid_encoding_size(hoid));
1909 	ceph_encode_string(p, end, hoid->key, hoid->key_len);
1910 	ceph_encode_string(p, end, hoid->oid, hoid->oid_len);
1911 	ceph_encode_64(p, hoid->snapid);
1912 	ceph_encode_32(p, hoid->hash);
1913 	ceph_encode_8(p, hoid->is_max);
1914 	ceph_encode_string(p, end, hoid->nspace, hoid->nspace_len);
1915 	ceph_encode_64(p, hoid->pool);
1916 }
1917 
1918 static void free_hoid(struct ceph_hobject_id *hoid)
1919 {
1920 	if (hoid) {
1921 		kfree(hoid->key);
1922 		kfree(hoid->oid);
1923 		kfree(hoid->nspace);
1924 		kfree(hoid);
1925 	}
1926 }
1927 
1928 static struct ceph_osd_backoff *alloc_backoff(void)
1929 {
1930 	struct ceph_osd_backoff *backoff;
1931 
1932 	backoff = kzalloc(sizeof(*backoff), GFP_NOIO);
1933 	if (!backoff)
1934 		return NULL;
1935 
1936 	RB_CLEAR_NODE(&backoff->spg_node);
1937 	RB_CLEAR_NODE(&backoff->id_node);
1938 	return backoff;
1939 }
1940 
1941 static void free_backoff(struct ceph_osd_backoff *backoff)
1942 {
1943 	WARN_ON(!RB_EMPTY_NODE(&backoff->spg_node));
1944 	WARN_ON(!RB_EMPTY_NODE(&backoff->id_node));
1945 
1946 	free_hoid(backoff->begin);
1947 	free_hoid(backoff->end);
1948 	kfree(backoff);
1949 }
1950 
1951 /*
1952  * Within a specific spgid, backoffs are managed by ->begin hoid.
1953  */
1954 DEFINE_RB_INSDEL_FUNCS2(backoff, struct ceph_osd_backoff, begin, hoid_compare,
1955 			RB_BYVAL, spg_node);
1956 
1957 static struct ceph_osd_backoff *lookup_containing_backoff(struct rb_root *root,
1958 					    const struct ceph_hobject_id *hoid)
1959 {
1960 	struct rb_node *n = root->rb_node;
1961 
1962 	while (n) {
1963 		struct ceph_osd_backoff *cur =
1964 		    rb_entry(n, struct ceph_osd_backoff, spg_node);
1965 		int cmp;
1966 
1967 		cmp = hoid_compare(hoid, cur->begin);
1968 		if (cmp < 0) {
1969 			n = n->rb_left;
1970 		} else if (cmp > 0) {
1971 			if (hoid_compare(hoid, cur->end) < 0)
1972 				return cur;
1973 
1974 			n = n->rb_right;
1975 		} else {
1976 			return cur;
1977 		}
1978 	}
1979 
1980 	return NULL;
1981 }
1982 
1983 /*
1984  * Each backoff has a unique id within its OSD session.
1985  */
1986 DEFINE_RB_FUNCS(backoff_by_id, struct ceph_osd_backoff, id, id_node)
1987 
1988 static void clear_backoffs(struct ceph_osd *osd)
1989 {
1990 	while (!RB_EMPTY_ROOT(&osd->o_backoff_mappings)) {
1991 		struct ceph_spg_mapping *spg =
1992 		    rb_entry(rb_first(&osd->o_backoff_mappings),
1993 			     struct ceph_spg_mapping, node);
1994 
1995 		while (!RB_EMPTY_ROOT(&spg->backoffs)) {
1996 			struct ceph_osd_backoff *backoff =
1997 			    rb_entry(rb_first(&spg->backoffs),
1998 				     struct ceph_osd_backoff, spg_node);
1999 
2000 			erase_backoff(&spg->backoffs, backoff);
2001 			erase_backoff_by_id(&osd->o_backoffs_by_id, backoff);
2002 			free_backoff(backoff);
2003 		}
2004 		erase_spg_mapping(&osd->o_backoff_mappings, spg);
2005 		free_spg_mapping(spg);
2006 	}
2007 }
2008 
2009 /*
2010  * Set up a temporary, non-owning view into @t.
2011  */
2012 static void hoid_fill_from_target(struct ceph_hobject_id *hoid,
2013 				  const struct ceph_osd_request_target *t)
2014 {
2015 	hoid->key = NULL;
2016 	hoid->key_len = 0;
2017 	hoid->oid = t->target_oid.name;
2018 	hoid->oid_len = t->target_oid.name_len;
2019 	hoid->snapid = CEPH_NOSNAP;
2020 	hoid->hash = t->pgid.seed;
2021 	hoid->is_max = false;
2022 	if (t->target_oloc.pool_ns) {
2023 		hoid->nspace = t->target_oloc.pool_ns->str;
2024 		hoid->nspace_len = t->target_oloc.pool_ns->len;
2025 	} else {
2026 		hoid->nspace = NULL;
2027 		hoid->nspace_len = 0;
2028 	}
2029 	hoid->pool = t->target_oloc.pool;
2030 	ceph_hoid_build_hash_cache(hoid);
2031 }
2032 
2033 static bool should_plug_request(struct ceph_osd_request *req)
2034 {
2035 	struct ceph_osd *osd = req->r_osd;
2036 	struct ceph_spg_mapping *spg;
2037 	struct ceph_osd_backoff *backoff;
2038 	struct ceph_hobject_id hoid;
2039 
2040 	spg = lookup_spg_mapping(&osd->o_backoff_mappings, &req->r_t.spgid);
2041 	if (!spg)
2042 		return false;
2043 
2044 	hoid_fill_from_target(&hoid, &req->r_t);
2045 	backoff = lookup_containing_backoff(&spg->backoffs, &hoid);
2046 	if (!backoff)
2047 		return false;
2048 
2049 	dout("%s req %p tid %llu backoff osd%d spgid %llu.%xs%d id %llu\n",
2050 	     __func__, req, req->r_tid, osd->o_osd, backoff->spgid.pgid.pool,
2051 	     backoff->spgid.pgid.seed, backoff->spgid.shard, backoff->id);
2052 	return true;
2053 }
2054 
2055 /*
2056  * Keep get_num_data_items() in sync with this function.
2057  */
2058 static void setup_request_data(struct ceph_osd_request *req)
2059 {
2060 	struct ceph_msg *request_msg = req->r_request;
2061 	struct ceph_msg *reply_msg = req->r_reply;
2062 	struct ceph_osd_req_op *op;
2063 
2064 	if (req->r_request->num_data_items || req->r_reply->num_data_items)
2065 		return;
2066 
2067 	WARN_ON(request_msg->data_length || reply_msg->data_length);
2068 	for (op = req->r_ops; op != &req->r_ops[req->r_num_ops]; op++) {
2069 		switch (op->op) {
2070 		/* request */
2071 		case CEPH_OSD_OP_WRITE:
2072 		case CEPH_OSD_OP_WRITEFULL:
2073 			WARN_ON(op->indata_len != op->extent.length);
2074 			ceph_osdc_msg_data_add(request_msg,
2075 					       &op->extent.osd_data);
2076 			break;
2077 		case CEPH_OSD_OP_SETXATTR:
2078 		case CEPH_OSD_OP_CMPXATTR:
2079 			WARN_ON(op->indata_len != op->xattr.name_len +
2080 						  op->xattr.value_len);
2081 			ceph_osdc_msg_data_add(request_msg,
2082 					       &op->xattr.osd_data);
2083 			break;
2084 		case CEPH_OSD_OP_NOTIFY_ACK:
2085 			ceph_osdc_msg_data_add(request_msg,
2086 					       &op->notify_ack.request_data);
2087 			break;
2088 		case CEPH_OSD_OP_COPY_FROM2:
2089 			ceph_osdc_msg_data_add(request_msg,
2090 					       &op->copy_from.osd_data);
2091 			break;
2092 
2093 		/* reply */
2094 		case CEPH_OSD_OP_STAT:
2095 			ceph_osdc_msg_data_add(reply_msg,
2096 					       &op->raw_data_in);
2097 			break;
2098 		case CEPH_OSD_OP_READ:
2099 		case CEPH_OSD_OP_SPARSE_READ:
2100 			ceph_osdc_msg_data_add(reply_msg,
2101 					       &op->extent.osd_data);
2102 			break;
2103 		case CEPH_OSD_OP_LIST_WATCHERS:
2104 			ceph_osdc_msg_data_add(reply_msg,
2105 					       &op->list_watchers.response_data);
2106 			break;
2107 
2108 		/* both */
2109 		case CEPH_OSD_OP_CALL:
2110 			WARN_ON(op->indata_len != op->cls.class_len +
2111 						  op->cls.method_len +
2112 						  op->cls.indata_len);
2113 			ceph_osdc_msg_data_add(request_msg,
2114 					       &op->cls.request_info);
2115 			/* optional, can be NONE */
2116 			ceph_osdc_msg_data_add(request_msg,
2117 					       &op->cls.request_data);
2118 			/* optional, can be NONE */
2119 			ceph_osdc_msg_data_add(reply_msg,
2120 					       &op->cls.response_data);
2121 			break;
2122 		case CEPH_OSD_OP_NOTIFY:
2123 			ceph_osdc_msg_data_add(request_msg,
2124 					       &op->notify.request_data);
2125 			ceph_osdc_msg_data_add(reply_msg,
2126 					       &op->notify.response_data);
2127 			break;
2128 		}
2129 	}
2130 }
2131 
2132 static void encode_pgid(void **p, const struct ceph_pg *pgid)
2133 {
2134 	ceph_encode_8(p, 1);
2135 	ceph_encode_64(p, pgid->pool);
2136 	ceph_encode_32(p, pgid->seed);
2137 	ceph_encode_32(p, -1); /* preferred */
2138 }
2139 
2140 static void encode_spgid(void **p, const struct ceph_spg *spgid)
2141 {
2142 	ceph_start_encoding(p, 1, 1, CEPH_PGID_ENCODING_LEN + 1);
2143 	encode_pgid(p, &spgid->pgid);
2144 	ceph_encode_8(p, spgid->shard);
2145 }
2146 
2147 static void encode_oloc(void **p, void *end,
2148 			const struct ceph_object_locator *oloc)
2149 {
2150 	ceph_start_encoding(p, 5, 4, ceph_oloc_encoding_size(oloc));
2151 	ceph_encode_64(p, oloc->pool);
2152 	ceph_encode_32(p, -1); /* preferred */
2153 	ceph_encode_32(p, 0);  /* key len */
2154 	if (oloc->pool_ns)
2155 		ceph_encode_string(p, end, oloc->pool_ns->str,
2156 				   oloc->pool_ns->len);
2157 	else
2158 		ceph_encode_32(p, 0);
2159 }
2160 
2161 static void encode_request_partial(struct ceph_osd_request *req,
2162 				   struct ceph_msg *msg)
2163 {
2164 	void *p = msg->front.iov_base;
2165 	void *const end = p + msg->front_alloc_len;
2166 	u32 data_len = 0;
2167 	int i;
2168 
2169 	if (req->r_flags & CEPH_OSD_FLAG_WRITE) {
2170 		/* snapshots aren't writeable */
2171 		WARN_ON(req->r_snapid != CEPH_NOSNAP);
2172 	} else {
2173 		WARN_ON(req->r_mtime.tv_sec || req->r_mtime.tv_nsec ||
2174 			req->r_data_offset || req->r_snapc);
2175 	}
2176 
2177 	setup_request_data(req);
2178 
2179 	encode_spgid(&p, &req->r_t.spgid); /* actual spg */
2180 	ceph_encode_32(&p, req->r_t.pgid.seed); /* raw hash */
2181 	ceph_encode_32(&p, req->r_osdc->osdmap->epoch);
2182 	ceph_encode_32(&p, req->r_flags);
2183 
2184 	/* reqid */
2185 	ceph_start_encoding(&p, 2, 2, sizeof(struct ceph_osd_reqid));
2186 	memset(p, 0, sizeof(struct ceph_osd_reqid));
2187 	p += sizeof(struct ceph_osd_reqid);
2188 
2189 	/* trace */
2190 	memset(p, 0, sizeof(struct ceph_blkin_trace_info));
2191 	p += sizeof(struct ceph_blkin_trace_info);
2192 
2193 	ceph_encode_32(&p, 0); /* client_inc, always 0 */
2194 	ceph_encode_timespec64(p, &req->r_mtime);
2195 	p += sizeof(struct ceph_timespec);
2196 
2197 	encode_oloc(&p, end, &req->r_t.target_oloc);
2198 	ceph_encode_string(&p, end, req->r_t.target_oid.name,
2199 			   req->r_t.target_oid.name_len);
2200 
2201 	/* ops, can imply data */
2202 	ceph_encode_16(&p, req->r_num_ops);
2203 	for (i = 0; i < req->r_num_ops; i++) {
2204 		data_len += osd_req_encode_op(p, &req->r_ops[i]);
2205 		p += sizeof(struct ceph_osd_op);
2206 	}
2207 
2208 	ceph_encode_64(&p, req->r_snapid); /* snapid */
2209 	if (req->r_snapc) {
2210 		ceph_encode_64(&p, req->r_snapc->seq);
2211 		ceph_encode_32(&p, req->r_snapc->num_snaps);
2212 		for (i = 0; i < req->r_snapc->num_snaps; i++)
2213 			ceph_encode_64(&p, req->r_snapc->snaps[i]);
2214 	} else {
2215 		ceph_encode_64(&p, 0); /* snap_seq */
2216 		ceph_encode_32(&p, 0); /* snaps len */
2217 	}
2218 
2219 	ceph_encode_32(&p, req->r_attempts); /* retry_attempt */
2220 	BUG_ON(p > end - 8); /* space for features */
2221 
2222 	msg->hdr.version = cpu_to_le16(8); /* MOSDOp v8 */
2223 	/* front_len is finalized in encode_request_finish() */
2224 	msg->front.iov_len = p - msg->front.iov_base;
2225 	msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2226 	msg->hdr.data_len = cpu_to_le32(data_len);
2227 	/*
2228 	 * The header "data_off" is a hint to the receiver allowing it
2229 	 * to align received data into its buffers such that there's no
2230 	 * need to re-copy it before writing it to disk (direct I/O).
2231 	 */
2232 	msg->hdr.data_off = cpu_to_le16(req->r_data_offset);
2233 
2234 	dout("%s req %p msg %p oid %s oid_len %d\n", __func__, req, msg,
2235 	     req->r_t.target_oid.name, req->r_t.target_oid.name_len);
2236 }
2237 
2238 static void encode_request_finish(struct ceph_msg *msg)
2239 {
2240 	void *p = msg->front.iov_base;
2241 	void *const partial_end = p + msg->front.iov_len;
2242 	void *const end = p + msg->front_alloc_len;
2243 
2244 	if (CEPH_HAVE_FEATURE(msg->con->peer_features, RESEND_ON_SPLIT)) {
2245 		/* luminous OSD -- encode features and be done */
2246 		p = partial_end;
2247 		ceph_encode_64(&p, msg->con->peer_features);
2248 	} else {
2249 		struct {
2250 			char spgid[CEPH_ENCODING_START_BLK_LEN +
2251 				   CEPH_PGID_ENCODING_LEN + 1];
2252 			__le32 hash;
2253 			__le32 epoch;
2254 			__le32 flags;
2255 			char reqid[CEPH_ENCODING_START_BLK_LEN +
2256 				   sizeof(struct ceph_osd_reqid)];
2257 			char trace[sizeof(struct ceph_blkin_trace_info)];
2258 			__le32 client_inc;
2259 			struct ceph_timespec mtime;
2260 		} __packed head;
2261 		struct ceph_pg pgid;
2262 		void *oloc, *oid, *tail;
2263 		int oloc_len, oid_len, tail_len;
2264 		int len;
2265 
2266 		/*
2267 		 * Pre-luminous OSD -- reencode v8 into v4 using @head
2268 		 * as a temporary buffer.  Encode the raw PG; the rest
2269 		 * is just a matter of moving oloc, oid and tail blobs
2270 		 * around.
2271 		 */
2272 		memcpy(&head, p, sizeof(head));
2273 		p += sizeof(head);
2274 
2275 		oloc = p;
2276 		p += CEPH_ENCODING_START_BLK_LEN;
2277 		pgid.pool = ceph_decode_64(&p);
2278 		p += 4 + 4; /* preferred, key len */
2279 		len = ceph_decode_32(&p);
2280 		p += len;   /* nspace */
2281 		oloc_len = p - oloc;
2282 
2283 		oid = p;
2284 		len = ceph_decode_32(&p);
2285 		p += len;
2286 		oid_len = p - oid;
2287 
2288 		tail = p;
2289 		tail_len = partial_end - p;
2290 
2291 		p = msg->front.iov_base;
2292 		ceph_encode_copy(&p, &head.client_inc, sizeof(head.client_inc));
2293 		ceph_encode_copy(&p, &head.epoch, sizeof(head.epoch));
2294 		ceph_encode_copy(&p, &head.flags, sizeof(head.flags));
2295 		ceph_encode_copy(&p, &head.mtime, sizeof(head.mtime));
2296 
2297 		/* reassert_version */
2298 		memset(p, 0, sizeof(struct ceph_eversion));
2299 		p += sizeof(struct ceph_eversion);
2300 
2301 		BUG_ON(p >= oloc);
2302 		memmove(p, oloc, oloc_len);
2303 		p += oloc_len;
2304 
2305 		pgid.seed = le32_to_cpu(head.hash);
2306 		encode_pgid(&p, &pgid); /* raw pg */
2307 
2308 		BUG_ON(p >= oid);
2309 		memmove(p, oid, oid_len);
2310 		p += oid_len;
2311 
2312 		/* tail -- ops, snapid, snapc, retry_attempt */
2313 		BUG_ON(p >= tail);
2314 		memmove(p, tail, tail_len);
2315 		p += tail_len;
2316 
2317 		msg->hdr.version = cpu_to_le16(4); /* MOSDOp v4 */
2318 	}
2319 
2320 	BUG_ON(p > end);
2321 	msg->front.iov_len = p - msg->front.iov_base;
2322 	msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2323 
2324 	dout("%s msg %p tid %llu %u+%u+%u v%d\n", __func__, msg,
2325 	     le64_to_cpu(msg->hdr.tid), le32_to_cpu(msg->hdr.front_len),
2326 	     le32_to_cpu(msg->hdr.middle_len), le32_to_cpu(msg->hdr.data_len),
2327 	     le16_to_cpu(msg->hdr.version));
2328 }
2329 
2330 /*
2331  * @req has to be assigned a tid and registered.
2332  */
2333 static void send_request(struct ceph_osd_request *req)
2334 {
2335 	struct ceph_osd *osd = req->r_osd;
2336 
2337 	verify_osd_locked(osd);
2338 	WARN_ON(osd->o_osd != req->r_t.osd);
2339 
2340 	/* backoff? */
2341 	if (should_plug_request(req))
2342 		return;
2343 
2344 	/*
2345 	 * We may have a previously queued request message hanging
2346 	 * around.  Cancel it to avoid corrupting the msgr.
2347 	 */
2348 	if (req->r_sent)
2349 		ceph_msg_revoke(req->r_request);
2350 
2351 	req->r_flags |= CEPH_OSD_FLAG_KNOWN_REDIR;
2352 	if (req->r_attempts)
2353 		req->r_flags |= CEPH_OSD_FLAG_RETRY;
2354 	else
2355 		WARN_ON(req->r_flags & CEPH_OSD_FLAG_RETRY);
2356 
2357 	encode_request_partial(req, req->r_request);
2358 
2359 	dout("%s req %p tid %llu to pgid %llu.%x spgid %llu.%xs%d osd%d e%u flags 0x%x attempt %d\n",
2360 	     __func__, req, req->r_tid, req->r_t.pgid.pool, req->r_t.pgid.seed,
2361 	     req->r_t.spgid.pgid.pool, req->r_t.spgid.pgid.seed,
2362 	     req->r_t.spgid.shard, osd->o_osd, req->r_t.epoch, req->r_flags,
2363 	     req->r_attempts);
2364 
2365 	req->r_t.paused = false;
2366 	req->r_stamp = jiffies;
2367 	req->r_attempts++;
2368 
2369 	req->r_sent = osd->o_incarnation;
2370 	req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
2371 	ceph_con_send(&osd->o_con, ceph_msg_get(req->r_request));
2372 }
2373 
2374 static void maybe_request_map(struct ceph_osd_client *osdc)
2375 {
2376 	bool continuous = false;
2377 
2378 	verify_osdc_locked(osdc);
2379 	WARN_ON(!osdc->osdmap->epoch);
2380 
2381 	if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
2382 	    ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD) ||
2383 	    ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) {
2384 		dout("%s osdc %p continuous\n", __func__, osdc);
2385 		continuous = true;
2386 	} else {
2387 		dout("%s osdc %p onetime\n", __func__, osdc);
2388 	}
2389 
2390 	if (ceph_monc_want_map(&osdc->client->monc, CEPH_SUB_OSDMAP,
2391 			       osdc->osdmap->epoch + 1, continuous))
2392 		ceph_monc_renew_subs(&osdc->client->monc);
2393 }
2394 
2395 static void complete_request(struct ceph_osd_request *req, int err);
2396 static void send_map_check(struct ceph_osd_request *req);
2397 
2398 static void __submit_request(struct ceph_osd_request *req, bool wrlocked)
2399 {
2400 	struct ceph_osd_client *osdc = req->r_osdc;
2401 	struct ceph_osd *osd;
2402 	enum calc_target_result ct_res;
2403 	int err = 0;
2404 	bool need_send = false;
2405 	bool promoted = false;
2406 
2407 	WARN_ON(req->r_tid);
2408 	dout("%s req %p wrlocked %d\n", __func__, req, wrlocked);
2409 
2410 again:
2411 	ct_res = calc_target(osdc, &req->r_t, false);
2412 	if (ct_res == CALC_TARGET_POOL_DNE && !wrlocked)
2413 		goto promote;
2414 
2415 	osd = lookup_create_osd(osdc, req->r_t.osd, wrlocked);
2416 	if (IS_ERR(osd)) {
2417 		WARN_ON(PTR_ERR(osd) != -EAGAIN || wrlocked);
2418 		goto promote;
2419 	}
2420 
2421 	if (osdc->abort_err) {
2422 		dout("req %p abort_err %d\n", req, osdc->abort_err);
2423 		err = osdc->abort_err;
2424 	} else if (osdc->osdmap->epoch < osdc->epoch_barrier) {
2425 		dout("req %p epoch %u barrier %u\n", req, osdc->osdmap->epoch,
2426 		     osdc->epoch_barrier);
2427 		req->r_t.paused = true;
2428 		maybe_request_map(osdc);
2429 	} else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
2430 		   ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) {
2431 		dout("req %p pausewr\n", req);
2432 		req->r_t.paused = true;
2433 		maybe_request_map(osdc);
2434 	} else if ((req->r_flags & CEPH_OSD_FLAG_READ) &&
2435 		   ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) {
2436 		dout("req %p pauserd\n", req);
2437 		req->r_t.paused = true;
2438 		maybe_request_map(osdc);
2439 	} else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
2440 		   !(req->r_flags & (CEPH_OSD_FLAG_FULL_TRY |
2441 				     CEPH_OSD_FLAG_FULL_FORCE)) &&
2442 		   (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
2443 		    pool_full(osdc, req->r_t.base_oloc.pool))) {
2444 		dout("req %p full/pool_full\n", req);
2445 		if (ceph_test_opt(osdc->client, ABORT_ON_FULL)) {
2446 			err = -ENOSPC;
2447 		} else {
2448 			if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL))
2449 				pr_warn_ratelimited("cluster is full (osdmap FULL)\n");
2450 			else
2451 				pr_warn_ratelimited("pool %lld is full or reached quota\n",
2452 						    req->r_t.base_oloc.pool);
2453 			req->r_t.paused = true;
2454 			maybe_request_map(osdc);
2455 		}
2456 	} else if (!osd_homeless(osd)) {
2457 		need_send = true;
2458 	} else {
2459 		maybe_request_map(osdc);
2460 	}
2461 
2462 	mutex_lock(&osd->lock);
2463 	/*
2464 	 * Assign the tid atomically with send_request() to protect
2465 	 * multiple writes to the same object from racing with each
2466 	 * other, resulting in out of order ops on the OSDs.
2467 	 */
2468 	req->r_tid = atomic64_inc_return(&osdc->last_tid);
2469 	link_request(osd, req);
2470 	if (need_send)
2471 		send_request(req);
2472 	else if (err)
2473 		complete_request(req, err);
2474 	mutex_unlock(&osd->lock);
2475 
2476 	if (!err && ct_res == CALC_TARGET_POOL_DNE)
2477 		send_map_check(req);
2478 
2479 	if (promoted)
2480 		downgrade_write(&osdc->lock);
2481 	return;
2482 
2483 promote:
2484 	up_read(&osdc->lock);
2485 	down_write(&osdc->lock);
2486 	wrlocked = true;
2487 	promoted = true;
2488 	goto again;
2489 }
2490 
2491 static void account_request(struct ceph_osd_request *req)
2492 {
2493 	WARN_ON(req->r_flags & (CEPH_OSD_FLAG_ACK | CEPH_OSD_FLAG_ONDISK));
2494 	WARN_ON(!(req->r_flags & (CEPH_OSD_FLAG_READ | CEPH_OSD_FLAG_WRITE)));
2495 
2496 	req->r_flags |= CEPH_OSD_FLAG_ONDISK;
2497 	atomic_inc(&req->r_osdc->num_requests);
2498 
2499 	req->r_start_stamp = jiffies;
2500 	req->r_start_latency = ktime_get();
2501 }
2502 
2503 static void submit_request(struct ceph_osd_request *req, bool wrlocked)
2504 {
2505 	ceph_osdc_get_request(req);
2506 	account_request(req);
2507 	__submit_request(req, wrlocked);
2508 }
2509 
2510 static void finish_request(struct ceph_osd_request *req)
2511 {
2512 	struct ceph_osd_client *osdc = req->r_osdc;
2513 
2514 	WARN_ON(lookup_request_mc(&osdc->map_checks, req->r_tid));
2515 	dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
2516 
2517 	req->r_end_latency = ktime_get();
2518 
2519 	if (req->r_osd) {
2520 		ceph_init_sparse_read(&req->r_osd->o_sparse_read);
2521 		unlink_request(req->r_osd, req);
2522 	}
2523 	atomic_dec(&osdc->num_requests);
2524 
2525 	/*
2526 	 * If an OSD has failed or returned and a request has been sent
2527 	 * twice, it's possible to get a reply and end up here while the
2528 	 * request message is queued for delivery.  We will ignore the
2529 	 * reply, so not a big deal, but better to try and catch it.
2530 	 */
2531 	ceph_msg_revoke(req->r_request);
2532 	ceph_msg_revoke_incoming(req->r_reply);
2533 }
2534 
2535 static void __complete_request(struct ceph_osd_request *req)
2536 {
2537 	dout("%s req %p tid %llu cb %ps result %d\n", __func__, req,
2538 	     req->r_tid, req->r_callback, req->r_result);
2539 
2540 	if (req->r_callback)
2541 		req->r_callback(req);
2542 	complete_all(&req->r_completion);
2543 	ceph_osdc_put_request(req);
2544 }
2545 
2546 static void complete_request_workfn(struct work_struct *work)
2547 {
2548 	struct ceph_osd_request *req =
2549 	    container_of(work, struct ceph_osd_request, r_complete_work);
2550 
2551 	__complete_request(req);
2552 }
2553 
2554 /*
2555  * This is open-coded in handle_reply().
2556  */
2557 static void complete_request(struct ceph_osd_request *req, int err)
2558 {
2559 	dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err);
2560 
2561 	req->r_result = err;
2562 	finish_request(req);
2563 
2564 	INIT_WORK(&req->r_complete_work, complete_request_workfn);
2565 	queue_work(req->r_osdc->completion_wq, &req->r_complete_work);
2566 }
2567 
2568 static void cancel_map_check(struct ceph_osd_request *req)
2569 {
2570 	struct ceph_osd_client *osdc = req->r_osdc;
2571 	struct ceph_osd_request *lookup_req;
2572 
2573 	verify_osdc_wrlocked(osdc);
2574 
2575 	lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid);
2576 	if (!lookup_req)
2577 		return;
2578 
2579 	WARN_ON(lookup_req != req);
2580 	erase_request_mc(&osdc->map_checks, req);
2581 	ceph_osdc_put_request(req);
2582 }
2583 
2584 static void cancel_request(struct ceph_osd_request *req)
2585 {
2586 	dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
2587 
2588 	cancel_map_check(req);
2589 	finish_request(req);
2590 	complete_all(&req->r_completion);
2591 	ceph_osdc_put_request(req);
2592 }
2593 
2594 static void abort_request(struct ceph_osd_request *req, int err)
2595 {
2596 	dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err);
2597 
2598 	cancel_map_check(req);
2599 	complete_request(req, err);
2600 }
2601 
2602 static int abort_fn(struct ceph_osd_request *req, void *arg)
2603 {
2604 	int err = *(int *)arg;
2605 
2606 	abort_request(req, err);
2607 	return 0; /* continue iteration */
2608 }
2609 
2610 /*
2611  * Abort all in-flight requests with @err and arrange for all future
2612  * requests to be failed immediately.
2613  */
2614 void ceph_osdc_abort_requests(struct ceph_osd_client *osdc, int err)
2615 {
2616 	dout("%s osdc %p err %d\n", __func__, osdc, err);
2617 	down_write(&osdc->lock);
2618 	for_each_request(osdc, abort_fn, &err);
2619 	osdc->abort_err = err;
2620 	up_write(&osdc->lock);
2621 }
2622 EXPORT_SYMBOL(ceph_osdc_abort_requests);
2623 
2624 void ceph_osdc_clear_abort_err(struct ceph_osd_client *osdc)
2625 {
2626 	down_write(&osdc->lock);
2627 	osdc->abort_err = 0;
2628 	up_write(&osdc->lock);
2629 }
2630 EXPORT_SYMBOL(ceph_osdc_clear_abort_err);
2631 
2632 static void update_epoch_barrier(struct ceph_osd_client *osdc, u32 eb)
2633 {
2634 	if (likely(eb > osdc->epoch_barrier)) {
2635 		dout("updating epoch_barrier from %u to %u\n",
2636 				osdc->epoch_barrier, eb);
2637 		osdc->epoch_barrier = eb;
2638 		/* Request map if we're not to the barrier yet */
2639 		if (eb > osdc->osdmap->epoch)
2640 			maybe_request_map(osdc);
2641 	}
2642 }
2643 
2644 void ceph_osdc_update_epoch_barrier(struct ceph_osd_client *osdc, u32 eb)
2645 {
2646 	down_read(&osdc->lock);
2647 	if (unlikely(eb > osdc->epoch_barrier)) {
2648 		up_read(&osdc->lock);
2649 		down_write(&osdc->lock);
2650 		update_epoch_barrier(osdc, eb);
2651 		up_write(&osdc->lock);
2652 	} else {
2653 		up_read(&osdc->lock);
2654 	}
2655 }
2656 EXPORT_SYMBOL(ceph_osdc_update_epoch_barrier);
2657 
2658 /*
2659  * We can end up releasing caps as a result of abort_request().
2660  * In that case, we probably want to ensure that the cap release message
2661  * has an updated epoch barrier in it, so set the epoch barrier prior to
2662  * aborting the first request.
2663  */
2664 static int abort_on_full_fn(struct ceph_osd_request *req, void *arg)
2665 {
2666 	struct ceph_osd_client *osdc = req->r_osdc;
2667 	bool *victims = arg;
2668 
2669 	if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
2670 	    (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
2671 	     pool_full(osdc, req->r_t.base_oloc.pool))) {
2672 		if (!*victims) {
2673 			update_epoch_barrier(osdc, osdc->osdmap->epoch);
2674 			*victims = true;
2675 		}
2676 		abort_request(req, -ENOSPC);
2677 	}
2678 
2679 	return 0; /* continue iteration */
2680 }
2681 
2682 /*
2683  * Drop all pending requests that are stalled waiting on a full condition to
2684  * clear, and complete them with ENOSPC as the return code. Set the
2685  * osdc->epoch_barrier to the latest map epoch that we've seen if any were
2686  * cancelled.
2687  */
2688 static void ceph_osdc_abort_on_full(struct ceph_osd_client *osdc)
2689 {
2690 	bool victims = false;
2691 
2692 	if (ceph_test_opt(osdc->client, ABORT_ON_FULL) &&
2693 	    (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || have_pool_full(osdc)))
2694 		for_each_request(osdc, abort_on_full_fn, &victims);
2695 }
2696 
2697 static void check_pool_dne(struct ceph_osd_request *req)
2698 {
2699 	struct ceph_osd_client *osdc = req->r_osdc;
2700 	struct ceph_osdmap *map = osdc->osdmap;
2701 
2702 	verify_osdc_wrlocked(osdc);
2703 	WARN_ON(!map->epoch);
2704 
2705 	if (req->r_attempts) {
2706 		/*
2707 		 * We sent a request earlier, which means that
2708 		 * previously the pool existed, and now it does not
2709 		 * (i.e., it was deleted).
2710 		 */
2711 		req->r_map_dne_bound = map->epoch;
2712 		dout("%s req %p tid %llu pool disappeared\n", __func__, req,
2713 		     req->r_tid);
2714 	} else {
2715 		dout("%s req %p tid %llu map_dne_bound %u have %u\n", __func__,
2716 		     req, req->r_tid, req->r_map_dne_bound, map->epoch);
2717 	}
2718 
2719 	if (req->r_map_dne_bound) {
2720 		if (map->epoch >= req->r_map_dne_bound) {
2721 			/* we had a new enough map */
2722 			pr_info_ratelimited("tid %llu pool does not exist\n",
2723 					    req->r_tid);
2724 			complete_request(req, -ENOENT);
2725 		}
2726 	} else {
2727 		send_map_check(req);
2728 	}
2729 }
2730 
2731 static void map_check_cb(struct ceph_mon_generic_request *greq)
2732 {
2733 	struct ceph_osd_client *osdc = &greq->monc->client->osdc;
2734 	struct ceph_osd_request *req;
2735 	u64 tid = greq->private_data;
2736 
2737 	WARN_ON(greq->result || !greq->u.newest);
2738 
2739 	down_write(&osdc->lock);
2740 	req = lookup_request_mc(&osdc->map_checks, tid);
2741 	if (!req) {
2742 		dout("%s tid %llu dne\n", __func__, tid);
2743 		goto out_unlock;
2744 	}
2745 
2746 	dout("%s req %p tid %llu map_dne_bound %u newest %llu\n", __func__,
2747 	     req, req->r_tid, req->r_map_dne_bound, greq->u.newest);
2748 	if (!req->r_map_dne_bound)
2749 		req->r_map_dne_bound = greq->u.newest;
2750 	erase_request_mc(&osdc->map_checks, req);
2751 	check_pool_dne(req);
2752 
2753 	ceph_osdc_put_request(req);
2754 out_unlock:
2755 	up_write(&osdc->lock);
2756 }
2757 
2758 static void send_map_check(struct ceph_osd_request *req)
2759 {
2760 	struct ceph_osd_client *osdc = req->r_osdc;
2761 	struct ceph_osd_request *lookup_req;
2762 	int ret;
2763 
2764 	verify_osdc_wrlocked(osdc);
2765 
2766 	lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid);
2767 	if (lookup_req) {
2768 		WARN_ON(lookup_req != req);
2769 		return;
2770 	}
2771 
2772 	ceph_osdc_get_request(req);
2773 	insert_request_mc(&osdc->map_checks, req);
2774 	ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap",
2775 					  map_check_cb, req->r_tid);
2776 	WARN_ON(ret);
2777 }
2778 
2779 /*
2780  * lingering requests, watch/notify v2 infrastructure
2781  */
2782 static void linger_release(struct kref *kref)
2783 {
2784 	struct ceph_osd_linger_request *lreq =
2785 	    container_of(kref, struct ceph_osd_linger_request, kref);
2786 
2787 	dout("%s lreq %p reg_req %p ping_req %p\n", __func__, lreq,
2788 	     lreq->reg_req, lreq->ping_req);
2789 	WARN_ON(!RB_EMPTY_NODE(&lreq->node));
2790 	WARN_ON(!RB_EMPTY_NODE(&lreq->osdc_node));
2791 	WARN_ON(!RB_EMPTY_NODE(&lreq->mc_node));
2792 	WARN_ON(!list_empty(&lreq->scan_item));
2793 	WARN_ON(!list_empty(&lreq->pending_lworks));
2794 	WARN_ON(lreq->osd);
2795 
2796 	if (lreq->request_pl)
2797 		ceph_pagelist_release(lreq->request_pl);
2798 	if (lreq->notify_id_pages)
2799 		ceph_release_page_vector(lreq->notify_id_pages, 1);
2800 
2801 	ceph_osdc_put_request(lreq->reg_req);
2802 	ceph_osdc_put_request(lreq->ping_req);
2803 	target_destroy(&lreq->t);
2804 	kfree(lreq);
2805 }
2806 
2807 static void linger_put(struct ceph_osd_linger_request *lreq)
2808 {
2809 	if (lreq)
2810 		kref_put(&lreq->kref, linger_release);
2811 }
2812 
2813 static struct ceph_osd_linger_request *
2814 linger_get(struct ceph_osd_linger_request *lreq)
2815 {
2816 	kref_get(&lreq->kref);
2817 	return lreq;
2818 }
2819 
2820 static struct ceph_osd_linger_request *
2821 linger_alloc(struct ceph_osd_client *osdc)
2822 {
2823 	struct ceph_osd_linger_request *lreq;
2824 
2825 	lreq = kzalloc(sizeof(*lreq), GFP_NOIO);
2826 	if (!lreq)
2827 		return NULL;
2828 
2829 	kref_init(&lreq->kref);
2830 	mutex_init(&lreq->lock);
2831 	RB_CLEAR_NODE(&lreq->node);
2832 	RB_CLEAR_NODE(&lreq->osdc_node);
2833 	RB_CLEAR_NODE(&lreq->mc_node);
2834 	INIT_LIST_HEAD(&lreq->scan_item);
2835 	INIT_LIST_HEAD(&lreq->pending_lworks);
2836 	init_completion(&lreq->reg_commit_wait);
2837 	init_completion(&lreq->notify_finish_wait);
2838 
2839 	lreq->osdc = osdc;
2840 	target_init(&lreq->t);
2841 
2842 	dout("%s lreq %p\n", __func__, lreq);
2843 	return lreq;
2844 }
2845 
2846 DEFINE_RB_INSDEL_FUNCS(linger, struct ceph_osd_linger_request, linger_id, node)
2847 DEFINE_RB_FUNCS(linger_osdc, struct ceph_osd_linger_request, linger_id, osdc_node)
2848 DEFINE_RB_FUNCS(linger_mc, struct ceph_osd_linger_request, linger_id, mc_node)
2849 
2850 /*
2851  * Create linger request <-> OSD session relation.
2852  *
2853  * @lreq has to be registered, @osd may be homeless.
2854  */
2855 static void link_linger(struct ceph_osd *osd,
2856 			struct ceph_osd_linger_request *lreq)
2857 {
2858 	verify_osd_locked(osd);
2859 	WARN_ON(!lreq->linger_id || lreq->osd);
2860 	dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd,
2861 	     osd->o_osd, lreq, lreq->linger_id);
2862 
2863 	if (!osd_homeless(osd))
2864 		__remove_osd_from_lru(osd);
2865 	else
2866 		atomic_inc(&osd->o_osdc->num_homeless);
2867 
2868 	get_osd(osd);
2869 	insert_linger(&osd->o_linger_requests, lreq);
2870 	lreq->osd = osd;
2871 }
2872 
2873 static void unlink_linger(struct ceph_osd *osd,
2874 			  struct ceph_osd_linger_request *lreq)
2875 {
2876 	verify_osd_locked(osd);
2877 	WARN_ON(lreq->osd != osd);
2878 	dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd,
2879 	     osd->o_osd, lreq, lreq->linger_id);
2880 
2881 	lreq->osd = NULL;
2882 	erase_linger(&osd->o_linger_requests, lreq);
2883 	put_osd(osd);
2884 
2885 	if (!osd_homeless(osd))
2886 		maybe_move_osd_to_lru(osd);
2887 	else
2888 		atomic_dec(&osd->o_osdc->num_homeless);
2889 }
2890 
2891 static bool __linger_registered(struct ceph_osd_linger_request *lreq)
2892 {
2893 	verify_osdc_locked(lreq->osdc);
2894 
2895 	return !RB_EMPTY_NODE(&lreq->osdc_node);
2896 }
2897 
2898 static bool linger_registered(struct ceph_osd_linger_request *lreq)
2899 {
2900 	struct ceph_osd_client *osdc = lreq->osdc;
2901 	bool registered;
2902 
2903 	down_read(&osdc->lock);
2904 	registered = __linger_registered(lreq);
2905 	up_read(&osdc->lock);
2906 
2907 	return registered;
2908 }
2909 
2910 static void linger_register(struct ceph_osd_linger_request *lreq)
2911 {
2912 	struct ceph_osd_client *osdc = lreq->osdc;
2913 
2914 	verify_osdc_wrlocked(osdc);
2915 	WARN_ON(lreq->linger_id);
2916 
2917 	linger_get(lreq);
2918 	lreq->linger_id = ++osdc->last_linger_id;
2919 	insert_linger_osdc(&osdc->linger_requests, lreq);
2920 }
2921 
2922 static void linger_unregister(struct ceph_osd_linger_request *lreq)
2923 {
2924 	struct ceph_osd_client *osdc = lreq->osdc;
2925 
2926 	verify_osdc_wrlocked(osdc);
2927 
2928 	erase_linger_osdc(&osdc->linger_requests, lreq);
2929 	linger_put(lreq);
2930 }
2931 
2932 static void cancel_linger_request(struct ceph_osd_request *req)
2933 {
2934 	struct ceph_osd_linger_request *lreq = req->r_priv;
2935 
2936 	WARN_ON(!req->r_linger);
2937 	cancel_request(req);
2938 	linger_put(lreq);
2939 }
2940 
2941 struct linger_work {
2942 	struct work_struct work;
2943 	struct ceph_osd_linger_request *lreq;
2944 	struct list_head pending_item;
2945 	unsigned long queued_stamp;
2946 
2947 	union {
2948 		struct {
2949 			u64 notify_id;
2950 			u64 notifier_id;
2951 			void *payload; /* points into @msg front */
2952 			size_t payload_len;
2953 
2954 			struct ceph_msg *msg; /* for ceph_msg_put() */
2955 		} notify;
2956 		struct {
2957 			int err;
2958 		} error;
2959 	};
2960 };
2961 
2962 static struct linger_work *lwork_alloc(struct ceph_osd_linger_request *lreq,
2963 				       work_func_t workfn)
2964 {
2965 	struct linger_work *lwork;
2966 
2967 	lwork = kzalloc(sizeof(*lwork), GFP_NOIO);
2968 	if (!lwork)
2969 		return NULL;
2970 
2971 	INIT_WORK(&lwork->work, workfn);
2972 	INIT_LIST_HEAD(&lwork->pending_item);
2973 	lwork->lreq = linger_get(lreq);
2974 
2975 	return lwork;
2976 }
2977 
2978 static void lwork_free(struct linger_work *lwork)
2979 {
2980 	struct ceph_osd_linger_request *lreq = lwork->lreq;
2981 
2982 	mutex_lock(&lreq->lock);
2983 	list_del(&lwork->pending_item);
2984 	mutex_unlock(&lreq->lock);
2985 
2986 	linger_put(lreq);
2987 	kfree(lwork);
2988 }
2989 
2990 static void lwork_queue(struct linger_work *lwork)
2991 {
2992 	struct ceph_osd_linger_request *lreq = lwork->lreq;
2993 	struct ceph_osd_client *osdc = lreq->osdc;
2994 
2995 	verify_lreq_locked(lreq);
2996 	WARN_ON(!list_empty(&lwork->pending_item));
2997 
2998 	lwork->queued_stamp = jiffies;
2999 	list_add_tail(&lwork->pending_item, &lreq->pending_lworks);
3000 	queue_work(osdc->notify_wq, &lwork->work);
3001 }
3002 
3003 static void do_watch_notify(struct work_struct *w)
3004 {
3005 	struct linger_work *lwork = container_of(w, struct linger_work, work);
3006 	struct ceph_osd_linger_request *lreq = lwork->lreq;
3007 
3008 	if (!linger_registered(lreq)) {
3009 		dout("%s lreq %p not registered\n", __func__, lreq);
3010 		goto out;
3011 	}
3012 
3013 	WARN_ON(!lreq->is_watch);
3014 	dout("%s lreq %p notify_id %llu notifier_id %llu payload_len %zu\n",
3015 	     __func__, lreq, lwork->notify.notify_id, lwork->notify.notifier_id,
3016 	     lwork->notify.payload_len);
3017 	lreq->wcb(lreq->data, lwork->notify.notify_id, lreq->linger_id,
3018 		  lwork->notify.notifier_id, lwork->notify.payload,
3019 		  lwork->notify.payload_len);
3020 
3021 out:
3022 	ceph_msg_put(lwork->notify.msg);
3023 	lwork_free(lwork);
3024 }
3025 
3026 static void do_watch_error(struct work_struct *w)
3027 {
3028 	struct linger_work *lwork = container_of(w, struct linger_work, work);
3029 	struct ceph_osd_linger_request *lreq = lwork->lreq;
3030 
3031 	if (!linger_registered(lreq)) {
3032 		dout("%s lreq %p not registered\n", __func__, lreq);
3033 		goto out;
3034 	}
3035 
3036 	dout("%s lreq %p err %d\n", __func__, lreq, lwork->error.err);
3037 	lreq->errcb(lreq->data, lreq->linger_id, lwork->error.err);
3038 
3039 out:
3040 	lwork_free(lwork);
3041 }
3042 
3043 static void queue_watch_error(struct ceph_osd_linger_request *lreq)
3044 {
3045 	struct linger_work *lwork;
3046 
3047 	lwork = lwork_alloc(lreq, do_watch_error);
3048 	if (!lwork) {
3049 		pr_err("failed to allocate error-lwork\n");
3050 		return;
3051 	}
3052 
3053 	lwork->error.err = lreq->last_error;
3054 	lwork_queue(lwork);
3055 }
3056 
3057 static void linger_reg_commit_complete(struct ceph_osd_linger_request *lreq,
3058 				       int result)
3059 {
3060 	if (!completion_done(&lreq->reg_commit_wait)) {
3061 		lreq->reg_commit_error = (result <= 0 ? result : 0);
3062 		complete_all(&lreq->reg_commit_wait);
3063 	}
3064 }
3065 
3066 static void linger_commit_cb(struct ceph_osd_request *req)
3067 {
3068 	struct ceph_osd_linger_request *lreq = req->r_priv;
3069 
3070 	mutex_lock(&lreq->lock);
3071 	if (req != lreq->reg_req) {
3072 		dout("%s lreq %p linger_id %llu unknown req (%p != %p)\n",
3073 		     __func__, lreq, lreq->linger_id, req, lreq->reg_req);
3074 		goto out;
3075 	}
3076 
3077 	dout("%s lreq %p linger_id %llu result %d\n", __func__, lreq,
3078 	     lreq->linger_id, req->r_result);
3079 	linger_reg_commit_complete(lreq, req->r_result);
3080 	lreq->committed = true;
3081 
3082 	if (!lreq->is_watch) {
3083 		struct ceph_osd_data *osd_data =
3084 		    osd_req_op_data(req, 0, notify, response_data);
3085 		void *p = page_address(osd_data->pages[0]);
3086 
3087 		WARN_ON(req->r_ops[0].op != CEPH_OSD_OP_NOTIFY ||
3088 			osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
3089 
3090 		/* make note of the notify_id */
3091 		if (req->r_ops[0].outdata_len >= sizeof(u64)) {
3092 			lreq->notify_id = ceph_decode_64(&p);
3093 			dout("lreq %p notify_id %llu\n", lreq,
3094 			     lreq->notify_id);
3095 		} else {
3096 			dout("lreq %p no notify_id\n", lreq);
3097 		}
3098 	}
3099 
3100 out:
3101 	mutex_unlock(&lreq->lock);
3102 	linger_put(lreq);
3103 }
3104 
3105 static int normalize_watch_error(int err)
3106 {
3107 	/*
3108 	 * Translate ENOENT -> ENOTCONN so that a delete->disconnection
3109 	 * notification and a failure to reconnect because we raced with
3110 	 * the delete appear the same to the user.
3111 	 */
3112 	if (err == -ENOENT)
3113 		err = -ENOTCONN;
3114 
3115 	return err;
3116 }
3117 
3118 static void linger_reconnect_cb(struct ceph_osd_request *req)
3119 {
3120 	struct ceph_osd_linger_request *lreq = req->r_priv;
3121 
3122 	mutex_lock(&lreq->lock);
3123 	if (req != lreq->reg_req) {
3124 		dout("%s lreq %p linger_id %llu unknown req (%p != %p)\n",
3125 		     __func__, lreq, lreq->linger_id, req, lreq->reg_req);
3126 		goto out;
3127 	}
3128 
3129 	dout("%s lreq %p linger_id %llu result %d last_error %d\n", __func__,
3130 	     lreq, lreq->linger_id, req->r_result, lreq->last_error);
3131 	if (req->r_result < 0) {
3132 		if (!lreq->last_error) {
3133 			lreq->last_error = normalize_watch_error(req->r_result);
3134 			queue_watch_error(lreq);
3135 		}
3136 	}
3137 
3138 out:
3139 	mutex_unlock(&lreq->lock);
3140 	linger_put(lreq);
3141 }
3142 
3143 static void send_linger(struct ceph_osd_linger_request *lreq)
3144 {
3145 	struct ceph_osd_client *osdc = lreq->osdc;
3146 	struct ceph_osd_request *req;
3147 	int ret;
3148 
3149 	verify_osdc_wrlocked(osdc);
3150 	mutex_lock(&lreq->lock);
3151 	dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
3152 
3153 	if (lreq->reg_req) {
3154 		if (lreq->reg_req->r_osd)
3155 			cancel_linger_request(lreq->reg_req);
3156 		ceph_osdc_put_request(lreq->reg_req);
3157 	}
3158 
3159 	req = ceph_osdc_alloc_request(osdc, NULL, 1, true, GFP_NOIO);
3160 	BUG_ON(!req);
3161 
3162 	target_copy(&req->r_t, &lreq->t);
3163 	req->r_mtime = lreq->mtime;
3164 
3165 	if (lreq->is_watch && lreq->committed) {
3166 		osd_req_op_watch_init(req, 0, CEPH_OSD_WATCH_OP_RECONNECT,
3167 				      lreq->linger_id, ++lreq->register_gen);
3168 		dout("lreq %p reconnect register_gen %u\n", lreq,
3169 		     req->r_ops[0].watch.gen);
3170 		req->r_callback = linger_reconnect_cb;
3171 	} else {
3172 		if (lreq->is_watch) {
3173 			osd_req_op_watch_init(req, 0, CEPH_OSD_WATCH_OP_WATCH,
3174 					      lreq->linger_id, 0);
3175 		} else {
3176 			lreq->notify_id = 0;
3177 
3178 			refcount_inc(&lreq->request_pl->refcnt);
3179 			osd_req_op_notify_init(req, 0, lreq->linger_id,
3180 					       lreq->request_pl);
3181 			ceph_osd_data_pages_init(
3182 			    osd_req_op_data(req, 0, notify, response_data),
3183 			    lreq->notify_id_pages, PAGE_SIZE, 0, false, false);
3184 		}
3185 		dout("lreq %p register\n", lreq);
3186 		req->r_callback = linger_commit_cb;
3187 	}
3188 
3189 	ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
3190 	BUG_ON(ret);
3191 
3192 	req->r_priv = linger_get(lreq);
3193 	req->r_linger = true;
3194 	lreq->reg_req = req;
3195 	mutex_unlock(&lreq->lock);
3196 
3197 	submit_request(req, true);
3198 }
3199 
3200 static void linger_ping_cb(struct ceph_osd_request *req)
3201 {
3202 	struct ceph_osd_linger_request *lreq = req->r_priv;
3203 
3204 	mutex_lock(&lreq->lock);
3205 	if (req != lreq->ping_req) {
3206 		dout("%s lreq %p linger_id %llu unknown req (%p != %p)\n",
3207 		     __func__, lreq, lreq->linger_id, req, lreq->ping_req);
3208 		goto out;
3209 	}
3210 
3211 	dout("%s lreq %p linger_id %llu result %d ping_sent %lu last_error %d\n",
3212 	     __func__, lreq, lreq->linger_id, req->r_result, lreq->ping_sent,
3213 	     lreq->last_error);
3214 	if (lreq->register_gen == req->r_ops[0].watch.gen) {
3215 		if (!req->r_result) {
3216 			lreq->watch_valid_thru = lreq->ping_sent;
3217 		} else if (!lreq->last_error) {
3218 			lreq->last_error = normalize_watch_error(req->r_result);
3219 			queue_watch_error(lreq);
3220 		}
3221 	} else {
3222 		dout("lreq %p register_gen %u ignoring old pong %u\n", lreq,
3223 		     lreq->register_gen, req->r_ops[0].watch.gen);
3224 	}
3225 
3226 out:
3227 	mutex_unlock(&lreq->lock);
3228 	linger_put(lreq);
3229 }
3230 
3231 static void send_linger_ping(struct ceph_osd_linger_request *lreq)
3232 {
3233 	struct ceph_osd_client *osdc = lreq->osdc;
3234 	struct ceph_osd_request *req;
3235 	int ret;
3236 
3237 	if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) {
3238 		dout("%s PAUSERD\n", __func__);
3239 		return;
3240 	}
3241 
3242 	lreq->ping_sent = jiffies;
3243 	dout("%s lreq %p linger_id %llu ping_sent %lu register_gen %u\n",
3244 	     __func__, lreq, lreq->linger_id, lreq->ping_sent,
3245 	     lreq->register_gen);
3246 
3247 	if (lreq->ping_req) {
3248 		if (lreq->ping_req->r_osd)
3249 			cancel_linger_request(lreq->ping_req);
3250 		ceph_osdc_put_request(lreq->ping_req);
3251 	}
3252 
3253 	req = ceph_osdc_alloc_request(osdc, NULL, 1, true, GFP_NOIO);
3254 	BUG_ON(!req);
3255 
3256 	target_copy(&req->r_t, &lreq->t);
3257 	osd_req_op_watch_init(req, 0, CEPH_OSD_WATCH_OP_PING, lreq->linger_id,
3258 			      lreq->register_gen);
3259 	req->r_callback = linger_ping_cb;
3260 
3261 	ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
3262 	BUG_ON(ret);
3263 
3264 	req->r_priv = linger_get(lreq);
3265 	req->r_linger = true;
3266 	lreq->ping_req = req;
3267 
3268 	ceph_osdc_get_request(req);
3269 	account_request(req);
3270 	req->r_tid = atomic64_inc_return(&osdc->last_tid);
3271 	link_request(lreq->osd, req);
3272 	send_request(req);
3273 }
3274 
3275 static void linger_submit(struct ceph_osd_linger_request *lreq)
3276 {
3277 	struct ceph_osd_client *osdc = lreq->osdc;
3278 	struct ceph_osd *osd;
3279 
3280 	down_write(&osdc->lock);
3281 	linger_register(lreq);
3282 
3283 	calc_target(osdc, &lreq->t, false);
3284 	osd = lookup_create_osd(osdc, lreq->t.osd, true);
3285 	link_linger(osd, lreq);
3286 
3287 	send_linger(lreq);
3288 	up_write(&osdc->lock);
3289 }
3290 
3291 static void cancel_linger_map_check(struct ceph_osd_linger_request *lreq)
3292 {
3293 	struct ceph_osd_client *osdc = lreq->osdc;
3294 	struct ceph_osd_linger_request *lookup_lreq;
3295 
3296 	verify_osdc_wrlocked(osdc);
3297 
3298 	lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks,
3299 				       lreq->linger_id);
3300 	if (!lookup_lreq)
3301 		return;
3302 
3303 	WARN_ON(lookup_lreq != lreq);
3304 	erase_linger_mc(&osdc->linger_map_checks, lreq);
3305 	linger_put(lreq);
3306 }
3307 
3308 /*
3309  * @lreq has to be both registered and linked.
3310  */
3311 static void __linger_cancel(struct ceph_osd_linger_request *lreq)
3312 {
3313 	if (lreq->ping_req && lreq->ping_req->r_osd)
3314 		cancel_linger_request(lreq->ping_req);
3315 	if (lreq->reg_req && lreq->reg_req->r_osd)
3316 		cancel_linger_request(lreq->reg_req);
3317 	cancel_linger_map_check(lreq);
3318 	unlink_linger(lreq->osd, lreq);
3319 	linger_unregister(lreq);
3320 }
3321 
3322 static void linger_cancel(struct ceph_osd_linger_request *lreq)
3323 {
3324 	struct ceph_osd_client *osdc = lreq->osdc;
3325 
3326 	down_write(&osdc->lock);
3327 	if (__linger_registered(lreq))
3328 		__linger_cancel(lreq);
3329 	up_write(&osdc->lock);
3330 }
3331 
3332 static void send_linger_map_check(struct ceph_osd_linger_request *lreq);
3333 
3334 static void check_linger_pool_dne(struct ceph_osd_linger_request *lreq)
3335 {
3336 	struct ceph_osd_client *osdc = lreq->osdc;
3337 	struct ceph_osdmap *map = osdc->osdmap;
3338 
3339 	verify_osdc_wrlocked(osdc);
3340 	WARN_ON(!map->epoch);
3341 
3342 	if (lreq->register_gen) {
3343 		lreq->map_dne_bound = map->epoch;
3344 		dout("%s lreq %p linger_id %llu pool disappeared\n", __func__,
3345 		     lreq, lreq->linger_id);
3346 	} else {
3347 		dout("%s lreq %p linger_id %llu map_dne_bound %u have %u\n",
3348 		     __func__, lreq, lreq->linger_id, lreq->map_dne_bound,
3349 		     map->epoch);
3350 	}
3351 
3352 	if (lreq->map_dne_bound) {
3353 		if (map->epoch >= lreq->map_dne_bound) {
3354 			/* we had a new enough map */
3355 			pr_info("linger_id %llu pool does not exist\n",
3356 				lreq->linger_id);
3357 			linger_reg_commit_complete(lreq, -ENOENT);
3358 			__linger_cancel(lreq);
3359 		}
3360 	} else {
3361 		send_linger_map_check(lreq);
3362 	}
3363 }
3364 
3365 static void linger_map_check_cb(struct ceph_mon_generic_request *greq)
3366 {
3367 	struct ceph_osd_client *osdc = &greq->monc->client->osdc;
3368 	struct ceph_osd_linger_request *lreq;
3369 	u64 linger_id = greq->private_data;
3370 
3371 	WARN_ON(greq->result || !greq->u.newest);
3372 
3373 	down_write(&osdc->lock);
3374 	lreq = lookup_linger_mc(&osdc->linger_map_checks, linger_id);
3375 	if (!lreq) {
3376 		dout("%s linger_id %llu dne\n", __func__, linger_id);
3377 		goto out_unlock;
3378 	}
3379 
3380 	dout("%s lreq %p linger_id %llu map_dne_bound %u newest %llu\n",
3381 	     __func__, lreq, lreq->linger_id, lreq->map_dne_bound,
3382 	     greq->u.newest);
3383 	if (!lreq->map_dne_bound)
3384 		lreq->map_dne_bound = greq->u.newest;
3385 	erase_linger_mc(&osdc->linger_map_checks, lreq);
3386 	check_linger_pool_dne(lreq);
3387 
3388 	linger_put(lreq);
3389 out_unlock:
3390 	up_write(&osdc->lock);
3391 }
3392 
3393 static void send_linger_map_check(struct ceph_osd_linger_request *lreq)
3394 {
3395 	struct ceph_osd_client *osdc = lreq->osdc;
3396 	struct ceph_osd_linger_request *lookup_lreq;
3397 	int ret;
3398 
3399 	verify_osdc_wrlocked(osdc);
3400 
3401 	lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks,
3402 				       lreq->linger_id);
3403 	if (lookup_lreq) {
3404 		WARN_ON(lookup_lreq != lreq);
3405 		return;
3406 	}
3407 
3408 	linger_get(lreq);
3409 	insert_linger_mc(&osdc->linger_map_checks, lreq);
3410 	ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap",
3411 					  linger_map_check_cb, lreq->linger_id);
3412 	WARN_ON(ret);
3413 }
3414 
3415 static int linger_reg_commit_wait(struct ceph_osd_linger_request *lreq)
3416 {
3417 	int ret;
3418 
3419 	dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
3420 	ret = wait_for_completion_killable(&lreq->reg_commit_wait);
3421 	return ret ?: lreq->reg_commit_error;
3422 }
3423 
3424 static int linger_notify_finish_wait(struct ceph_osd_linger_request *lreq,
3425 				     unsigned long timeout)
3426 {
3427 	long left;
3428 
3429 	dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
3430 	left = wait_for_completion_killable_timeout(&lreq->notify_finish_wait,
3431 						ceph_timeout_jiffies(timeout));
3432 	if (left <= 0)
3433 		left = left ?: -ETIMEDOUT;
3434 	else
3435 		left = lreq->notify_finish_error; /* completed */
3436 
3437 	return left;
3438 }
3439 
3440 /*
3441  * Timeout callback, called every N seconds.  When 1 or more OSD
3442  * requests has been active for more than N seconds, we send a keepalive
3443  * (tag + timestamp) to its OSD to ensure any communications channel
3444  * reset is detected.
3445  */
3446 static void handle_timeout(struct work_struct *work)
3447 {
3448 	struct ceph_osd_client *osdc =
3449 		container_of(work, struct ceph_osd_client, timeout_work.work);
3450 	struct ceph_options *opts = osdc->client->options;
3451 	unsigned long cutoff = jiffies - opts->osd_keepalive_timeout;
3452 	unsigned long expiry_cutoff = jiffies - opts->osd_request_timeout;
3453 	LIST_HEAD(slow_osds);
3454 	struct rb_node *n, *p;
3455 
3456 	dout("%s osdc %p\n", __func__, osdc);
3457 	down_write(&osdc->lock);
3458 
3459 	/*
3460 	 * ping osds that are a bit slow.  this ensures that if there
3461 	 * is a break in the TCP connection we will notice, and reopen
3462 	 * a connection with that osd (from the fault callback).
3463 	 */
3464 	for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
3465 		struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
3466 		bool found = false;
3467 
3468 		for (p = rb_first(&osd->o_requests); p; ) {
3469 			struct ceph_osd_request *req =
3470 			    rb_entry(p, struct ceph_osd_request, r_node);
3471 
3472 			p = rb_next(p); /* abort_request() */
3473 
3474 			if (time_before(req->r_stamp, cutoff)) {
3475 				dout(" req %p tid %llu on osd%d is laggy\n",
3476 				     req, req->r_tid, osd->o_osd);
3477 				found = true;
3478 			}
3479 			if (opts->osd_request_timeout &&
3480 			    time_before(req->r_start_stamp, expiry_cutoff)) {
3481 				pr_err_ratelimited("tid %llu on osd%d timeout\n",
3482 				       req->r_tid, osd->o_osd);
3483 				abort_request(req, -ETIMEDOUT);
3484 			}
3485 		}
3486 		for (p = rb_first(&osd->o_linger_requests); p; p = rb_next(p)) {
3487 			struct ceph_osd_linger_request *lreq =
3488 			    rb_entry(p, struct ceph_osd_linger_request, node);
3489 
3490 			dout(" lreq %p linger_id %llu is served by osd%d\n",
3491 			     lreq, lreq->linger_id, osd->o_osd);
3492 			found = true;
3493 
3494 			mutex_lock(&lreq->lock);
3495 			if (lreq->is_watch && lreq->committed && !lreq->last_error)
3496 				send_linger_ping(lreq);
3497 			mutex_unlock(&lreq->lock);
3498 		}
3499 
3500 		if (found)
3501 			list_move_tail(&osd->o_keepalive_item, &slow_osds);
3502 	}
3503 
3504 	if (opts->osd_request_timeout) {
3505 		for (p = rb_first(&osdc->homeless_osd.o_requests); p; ) {
3506 			struct ceph_osd_request *req =
3507 			    rb_entry(p, struct ceph_osd_request, r_node);
3508 
3509 			p = rb_next(p); /* abort_request() */
3510 
3511 			if (time_before(req->r_start_stamp, expiry_cutoff)) {
3512 				pr_err_ratelimited("tid %llu on osd%d timeout\n",
3513 				       req->r_tid, osdc->homeless_osd.o_osd);
3514 				abort_request(req, -ETIMEDOUT);
3515 			}
3516 		}
3517 	}
3518 
3519 	if (atomic_read(&osdc->num_homeless) || !list_empty(&slow_osds))
3520 		maybe_request_map(osdc);
3521 
3522 	while (!list_empty(&slow_osds)) {
3523 		struct ceph_osd *osd = list_first_entry(&slow_osds,
3524 							struct ceph_osd,
3525 							o_keepalive_item);
3526 		list_del_init(&osd->o_keepalive_item);
3527 		ceph_con_keepalive(&osd->o_con);
3528 	}
3529 
3530 	up_write(&osdc->lock);
3531 	schedule_delayed_work(&osdc->timeout_work,
3532 			      osdc->client->options->osd_keepalive_timeout);
3533 }
3534 
3535 static void handle_osds_timeout(struct work_struct *work)
3536 {
3537 	struct ceph_osd_client *osdc =
3538 		container_of(work, struct ceph_osd_client,
3539 			     osds_timeout_work.work);
3540 	unsigned long delay = osdc->client->options->osd_idle_ttl / 4;
3541 	struct ceph_osd *osd, *nosd;
3542 
3543 	dout("%s osdc %p\n", __func__, osdc);
3544 	down_write(&osdc->lock);
3545 	list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
3546 		if (time_before(jiffies, osd->lru_ttl))
3547 			break;
3548 
3549 		WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests));
3550 		WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests));
3551 		close_osd(osd);
3552 	}
3553 
3554 	up_write(&osdc->lock);
3555 	schedule_delayed_work(&osdc->osds_timeout_work,
3556 			      round_jiffies_relative(delay));
3557 }
3558 
3559 static int ceph_oloc_decode(void **p, void *end,
3560 			    struct ceph_object_locator *oloc)
3561 {
3562 	u8 struct_v, struct_cv;
3563 	u32 len;
3564 	void *struct_end;
3565 	int ret = 0;
3566 
3567 	ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
3568 	struct_v = ceph_decode_8(p);
3569 	struct_cv = ceph_decode_8(p);
3570 	if (struct_v < 3) {
3571 		pr_warn("got v %d < 3 cv %d of ceph_object_locator\n",
3572 			struct_v, struct_cv);
3573 		goto e_inval;
3574 	}
3575 	if (struct_cv > 6) {
3576 		pr_warn("got v %d cv %d > 6 of ceph_object_locator\n",
3577 			struct_v, struct_cv);
3578 		goto e_inval;
3579 	}
3580 	len = ceph_decode_32(p);
3581 	ceph_decode_need(p, end, len, e_inval);
3582 	struct_end = *p + len;
3583 
3584 	oloc->pool = ceph_decode_64(p);
3585 	*p += 4; /* skip preferred */
3586 
3587 	len = ceph_decode_32(p);
3588 	if (len > 0) {
3589 		pr_warn("ceph_object_locator::key is set\n");
3590 		goto e_inval;
3591 	}
3592 
3593 	if (struct_v >= 5) {
3594 		bool changed = false;
3595 
3596 		len = ceph_decode_32(p);
3597 		if (len > 0) {
3598 			ceph_decode_need(p, end, len, e_inval);
3599 			if (!oloc->pool_ns ||
3600 			    ceph_compare_string(oloc->pool_ns, *p, len))
3601 				changed = true;
3602 			*p += len;
3603 		} else {
3604 			if (oloc->pool_ns)
3605 				changed = true;
3606 		}
3607 		if (changed) {
3608 			/* redirect changes namespace */
3609 			pr_warn("ceph_object_locator::nspace is changed\n");
3610 			goto e_inval;
3611 		}
3612 	}
3613 
3614 	if (struct_v >= 6) {
3615 		s64 hash = ceph_decode_64(p);
3616 		if (hash != -1) {
3617 			pr_warn("ceph_object_locator::hash is set\n");
3618 			goto e_inval;
3619 		}
3620 	}
3621 
3622 	/* skip the rest */
3623 	*p = struct_end;
3624 out:
3625 	return ret;
3626 
3627 e_inval:
3628 	ret = -EINVAL;
3629 	goto out;
3630 }
3631 
3632 static int ceph_redirect_decode(void **p, void *end,
3633 				struct ceph_request_redirect *redir)
3634 {
3635 	u8 struct_v, struct_cv;
3636 	u32 len;
3637 	void *struct_end;
3638 	int ret;
3639 
3640 	ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
3641 	struct_v = ceph_decode_8(p);
3642 	struct_cv = ceph_decode_8(p);
3643 	if (struct_cv > 1) {
3644 		pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n",
3645 			struct_v, struct_cv);
3646 		goto e_inval;
3647 	}
3648 	len = ceph_decode_32(p);
3649 	ceph_decode_need(p, end, len, e_inval);
3650 	struct_end = *p + len;
3651 
3652 	ret = ceph_oloc_decode(p, end, &redir->oloc);
3653 	if (ret)
3654 		goto out;
3655 
3656 	len = ceph_decode_32(p);
3657 	if (len > 0) {
3658 		pr_warn("ceph_request_redirect::object_name is set\n");
3659 		goto e_inval;
3660 	}
3661 
3662 	/* skip the rest */
3663 	*p = struct_end;
3664 out:
3665 	return ret;
3666 
3667 e_inval:
3668 	ret = -EINVAL;
3669 	goto out;
3670 }
3671 
3672 struct MOSDOpReply {
3673 	struct ceph_pg pgid;
3674 	u64 flags;
3675 	int result;
3676 	u32 epoch;
3677 	int num_ops;
3678 	u32 outdata_len[CEPH_OSD_MAX_OPS];
3679 	s32 rval[CEPH_OSD_MAX_OPS];
3680 	int retry_attempt;
3681 	struct ceph_eversion replay_version;
3682 	u64 user_version;
3683 	struct ceph_request_redirect redirect;
3684 };
3685 
3686 static int decode_MOSDOpReply(const struct ceph_msg *msg, struct MOSDOpReply *m)
3687 {
3688 	void *p = msg->front.iov_base;
3689 	void *const end = p + msg->front.iov_len;
3690 	u16 version = le16_to_cpu(msg->hdr.version);
3691 	struct ceph_eversion bad_replay_version;
3692 	u8 decode_redir;
3693 	u32 len;
3694 	int ret;
3695 	int i;
3696 
3697 	ceph_decode_32_safe(&p, end, len, e_inval);
3698 	ceph_decode_need(&p, end, len, e_inval);
3699 	p += len; /* skip oid */
3700 
3701 	ret = ceph_decode_pgid(&p, end, &m->pgid);
3702 	if (ret)
3703 		return ret;
3704 
3705 	ceph_decode_64_safe(&p, end, m->flags, e_inval);
3706 	ceph_decode_32_safe(&p, end, m->result, e_inval);
3707 	ceph_decode_need(&p, end, sizeof(bad_replay_version), e_inval);
3708 	memcpy(&bad_replay_version, p, sizeof(bad_replay_version));
3709 	p += sizeof(bad_replay_version);
3710 	ceph_decode_32_safe(&p, end, m->epoch, e_inval);
3711 
3712 	ceph_decode_32_safe(&p, end, m->num_ops, e_inval);
3713 	if (m->num_ops > ARRAY_SIZE(m->outdata_len))
3714 		goto e_inval;
3715 
3716 	ceph_decode_need(&p, end, m->num_ops * sizeof(struct ceph_osd_op),
3717 			 e_inval);
3718 	for (i = 0; i < m->num_ops; i++) {
3719 		struct ceph_osd_op *op = p;
3720 
3721 		m->outdata_len[i] = le32_to_cpu(op->payload_len);
3722 		p += sizeof(*op);
3723 	}
3724 
3725 	ceph_decode_32_safe(&p, end, m->retry_attempt, e_inval);
3726 	for (i = 0; i < m->num_ops; i++)
3727 		ceph_decode_32_safe(&p, end, m->rval[i], e_inval);
3728 
3729 	if (version >= 5) {
3730 		ceph_decode_need(&p, end, sizeof(m->replay_version), e_inval);
3731 		memcpy(&m->replay_version, p, sizeof(m->replay_version));
3732 		p += sizeof(m->replay_version);
3733 		ceph_decode_64_safe(&p, end, m->user_version, e_inval);
3734 	} else {
3735 		m->replay_version = bad_replay_version; /* struct */
3736 		m->user_version = le64_to_cpu(m->replay_version.version);
3737 	}
3738 
3739 	if (version >= 6) {
3740 		if (version >= 7)
3741 			ceph_decode_8_safe(&p, end, decode_redir, e_inval);
3742 		else
3743 			decode_redir = 1;
3744 	} else {
3745 		decode_redir = 0;
3746 	}
3747 
3748 	if (decode_redir) {
3749 		ret = ceph_redirect_decode(&p, end, &m->redirect);
3750 		if (ret)
3751 			return ret;
3752 	} else {
3753 		ceph_oloc_init(&m->redirect.oloc);
3754 	}
3755 
3756 	return 0;
3757 
3758 e_inval:
3759 	return -EINVAL;
3760 }
3761 
3762 /*
3763  * Handle MOSDOpReply.  Set ->r_result and call the callback if it is
3764  * specified.
3765  */
3766 static void handle_reply(struct ceph_osd *osd, struct ceph_msg *msg)
3767 {
3768 	struct ceph_osd_client *osdc = osd->o_osdc;
3769 	struct ceph_osd_request *req;
3770 	struct MOSDOpReply m;
3771 	u64 tid = le64_to_cpu(msg->hdr.tid);
3772 	u32 data_len = 0;
3773 	int ret;
3774 	int i;
3775 
3776 	dout("%s msg %p tid %llu\n", __func__, msg, tid);
3777 
3778 	down_read(&osdc->lock);
3779 	if (!osd_registered(osd)) {
3780 		dout("%s osd%d unknown\n", __func__, osd->o_osd);
3781 		goto out_unlock_osdc;
3782 	}
3783 	WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num));
3784 
3785 	mutex_lock(&osd->lock);
3786 	req = lookup_request(&osd->o_requests, tid);
3787 	if (!req) {
3788 		dout("%s osd%d tid %llu unknown\n", __func__, osd->o_osd, tid);
3789 		goto out_unlock_session;
3790 	}
3791 
3792 	m.redirect.oloc.pool_ns = req->r_t.target_oloc.pool_ns;
3793 	ret = decode_MOSDOpReply(msg, &m);
3794 	m.redirect.oloc.pool_ns = NULL;
3795 	if (ret) {
3796 		pr_err("failed to decode MOSDOpReply for tid %llu: %d\n",
3797 		       req->r_tid, ret);
3798 		ceph_msg_dump(msg);
3799 		goto fail_request;
3800 	}
3801 	dout("%s req %p tid %llu flags 0x%llx pgid %llu.%x epoch %u attempt %d v %u'%llu uv %llu\n",
3802 	     __func__, req, req->r_tid, m.flags, m.pgid.pool, m.pgid.seed,
3803 	     m.epoch, m.retry_attempt, le32_to_cpu(m.replay_version.epoch),
3804 	     le64_to_cpu(m.replay_version.version), m.user_version);
3805 
3806 	if (m.retry_attempt >= 0) {
3807 		if (m.retry_attempt != req->r_attempts - 1) {
3808 			dout("req %p tid %llu retry_attempt %d != %d, ignoring\n",
3809 			     req, req->r_tid, m.retry_attempt,
3810 			     req->r_attempts - 1);
3811 			goto out_unlock_session;
3812 		}
3813 	} else {
3814 		WARN_ON(1); /* MOSDOpReply v4 is assumed */
3815 	}
3816 
3817 	if (!ceph_oloc_empty(&m.redirect.oloc)) {
3818 		dout("req %p tid %llu redirect pool %lld\n", req, req->r_tid,
3819 		     m.redirect.oloc.pool);
3820 		unlink_request(osd, req);
3821 		mutex_unlock(&osd->lock);
3822 
3823 		/*
3824 		 * Not ceph_oloc_copy() - changing pool_ns is not
3825 		 * supported.
3826 		 */
3827 		req->r_t.target_oloc.pool = m.redirect.oloc.pool;
3828 		req->r_flags |= CEPH_OSD_FLAG_REDIRECTED |
3829 				CEPH_OSD_FLAG_IGNORE_OVERLAY |
3830 				CEPH_OSD_FLAG_IGNORE_CACHE;
3831 		req->r_tid = 0;
3832 		__submit_request(req, false);
3833 		goto out_unlock_osdc;
3834 	}
3835 
3836 	if (m.result == -EAGAIN) {
3837 		dout("req %p tid %llu EAGAIN\n", req, req->r_tid);
3838 		unlink_request(osd, req);
3839 		mutex_unlock(&osd->lock);
3840 
3841 		/*
3842 		 * The object is missing on the replica or not (yet)
3843 		 * readable.  Clear pgid to force a resend to the primary
3844 		 * via legacy_change.
3845 		 */
3846 		req->r_t.pgid.pool = 0;
3847 		req->r_t.pgid.seed = 0;
3848 		WARN_ON(!req->r_t.used_replica);
3849 		req->r_flags &= ~(CEPH_OSD_FLAG_BALANCE_READS |
3850 				  CEPH_OSD_FLAG_LOCALIZE_READS);
3851 		req->r_tid = 0;
3852 		__submit_request(req, false);
3853 		goto out_unlock_osdc;
3854 	}
3855 
3856 	if (m.num_ops != req->r_num_ops) {
3857 		pr_err("num_ops %d != %d for tid %llu\n", m.num_ops,
3858 		       req->r_num_ops, req->r_tid);
3859 		goto fail_request;
3860 	}
3861 	for (i = 0; i < req->r_num_ops; i++) {
3862 		dout(" req %p tid %llu op %d rval %d len %u\n", req,
3863 		     req->r_tid, i, m.rval[i], m.outdata_len[i]);
3864 		req->r_ops[i].rval = m.rval[i];
3865 		req->r_ops[i].outdata_len = m.outdata_len[i];
3866 		data_len += m.outdata_len[i];
3867 	}
3868 	if (data_len != le32_to_cpu(msg->hdr.data_len)) {
3869 		pr_err("sum of lens %u != %u for tid %llu\n", data_len,
3870 		       le32_to_cpu(msg->hdr.data_len), req->r_tid);
3871 		goto fail_request;
3872 	}
3873 	dout("%s req %p tid %llu result %d data_len %u\n", __func__,
3874 	     req, req->r_tid, m.result, data_len);
3875 
3876 	/*
3877 	 * Since we only ever request ONDISK, we should only ever get
3878 	 * one (type of) reply back.
3879 	 */
3880 	WARN_ON(!(m.flags & CEPH_OSD_FLAG_ONDISK));
3881 	req->r_version = m.user_version;
3882 	req->r_result = m.result ?: data_len;
3883 	finish_request(req);
3884 	mutex_unlock(&osd->lock);
3885 	up_read(&osdc->lock);
3886 
3887 	__complete_request(req);
3888 	return;
3889 
3890 fail_request:
3891 	complete_request(req, -EIO);
3892 out_unlock_session:
3893 	mutex_unlock(&osd->lock);
3894 out_unlock_osdc:
3895 	up_read(&osdc->lock);
3896 }
3897 
3898 static void set_pool_was_full(struct ceph_osd_client *osdc)
3899 {
3900 	struct rb_node *n;
3901 
3902 	for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) {
3903 		struct ceph_pg_pool_info *pi =
3904 		    rb_entry(n, struct ceph_pg_pool_info, node);
3905 
3906 		pi->was_full = __pool_full(pi);
3907 	}
3908 }
3909 
3910 static bool pool_cleared_full(struct ceph_osd_client *osdc, s64 pool_id)
3911 {
3912 	struct ceph_pg_pool_info *pi;
3913 
3914 	pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id);
3915 	if (!pi)
3916 		return false;
3917 
3918 	return pi->was_full && !__pool_full(pi);
3919 }
3920 
3921 static enum calc_target_result
3922 recalc_linger_target(struct ceph_osd_linger_request *lreq)
3923 {
3924 	struct ceph_osd_client *osdc = lreq->osdc;
3925 	enum calc_target_result ct_res;
3926 
3927 	ct_res = calc_target(osdc, &lreq->t, true);
3928 	if (ct_res == CALC_TARGET_NEED_RESEND) {
3929 		struct ceph_osd *osd;
3930 
3931 		osd = lookup_create_osd(osdc, lreq->t.osd, true);
3932 		if (osd != lreq->osd) {
3933 			unlink_linger(lreq->osd, lreq);
3934 			link_linger(osd, lreq);
3935 		}
3936 	}
3937 
3938 	return ct_res;
3939 }
3940 
3941 /*
3942  * Requeue requests whose mapping to an OSD has changed.
3943  */
3944 static void scan_requests(struct ceph_osd *osd,
3945 			  bool force_resend,
3946 			  bool cleared_full,
3947 			  bool check_pool_cleared_full,
3948 			  struct rb_root *need_resend,
3949 			  struct list_head *need_resend_linger)
3950 {
3951 	struct ceph_osd_client *osdc = osd->o_osdc;
3952 	struct rb_node *n;
3953 	bool force_resend_writes;
3954 
3955 	for (n = rb_first(&osd->o_linger_requests); n; ) {
3956 		struct ceph_osd_linger_request *lreq =
3957 		    rb_entry(n, struct ceph_osd_linger_request, node);
3958 		enum calc_target_result ct_res;
3959 
3960 		n = rb_next(n); /* recalc_linger_target() */
3961 
3962 		dout("%s lreq %p linger_id %llu\n", __func__, lreq,
3963 		     lreq->linger_id);
3964 		ct_res = recalc_linger_target(lreq);
3965 		switch (ct_res) {
3966 		case CALC_TARGET_NO_ACTION:
3967 			force_resend_writes = cleared_full ||
3968 			    (check_pool_cleared_full &&
3969 			     pool_cleared_full(osdc, lreq->t.base_oloc.pool));
3970 			if (!force_resend && !force_resend_writes)
3971 				break;
3972 
3973 			fallthrough;
3974 		case CALC_TARGET_NEED_RESEND:
3975 			cancel_linger_map_check(lreq);
3976 			/*
3977 			 * scan_requests() for the previous epoch(s)
3978 			 * may have already added it to the list, since
3979 			 * it's not unlinked here.
3980 			 */
3981 			if (list_empty(&lreq->scan_item))
3982 				list_add_tail(&lreq->scan_item, need_resend_linger);
3983 			break;
3984 		case CALC_TARGET_POOL_DNE:
3985 			list_del_init(&lreq->scan_item);
3986 			check_linger_pool_dne(lreq);
3987 			break;
3988 		}
3989 	}
3990 
3991 	for (n = rb_first(&osd->o_requests); n; ) {
3992 		struct ceph_osd_request *req =
3993 		    rb_entry(n, struct ceph_osd_request, r_node);
3994 		enum calc_target_result ct_res;
3995 
3996 		n = rb_next(n); /* unlink_request(), check_pool_dne() */
3997 
3998 		dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
3999 		ct_res = calc_target(osdc, &req->r_t, false);
4000 		switch (ct_res) {
4001 		case CALC_TARGET_NO_ACTION:
4002 			force_resend_writes = cleared_full ||
4003 			    (check_pool_cleared_full &&
4004 			     pool_cleared_full(osdc, req->r_t.base_oloc.pool));
4005 			if (!force_resend &&
4006 			    (!(req->r_flags & CEPH_OSD_FLAG_WRITE) ||
4007 			     !force_resend_writes))
4008 				break;
4009 
4010 			fallthrough;
4011 		case CALC_TARGET_NEED_RESEND:
4012 			cancel_map_check(req);
4013 			unlink_request(osd, req);
4014 			insert_request(need_resend, req);
4015 			break;
4016 		case CALC_TARGET_POOL_DNE:
4017 			check_pool_dne(req);
4018 			break;
4019 		}
4020 	}
4021 }
4022 
4023 static int handle_one_map(struct ceph_osd_client *osdc,
4024 			  void *p, void *end, bool incremental,
4025 			  struct rb_root *need_resend,
4026 			  struct list_head *need_resend_linger)
4027 {
4028 	struct ceph_osdmap *newmap;
4029 	struct rb_node *n;
4030 	bool skipped_map = false;
4031 	bool was_full;
4032 
4033 	was_full = ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL);
4034 	set_pool_was_full(osdc);
4035 
4036 	if (incremental)
4037 		newmap = osdmap_apply_incremental(&p, end,
4038 						  ceph_msgr2(osdc->client),
4039 						  osdc->osdmap);
4040 	else
4041 		newmap = ceph_osdmap_decode(&p, end, ceph_msgr2(osdc->client));
4042 	if (IS_ERR(newmap))
4043 		return PTR_ERR(newmap);
4044 
4045 	if (newmap != osdc->osdmap) {
4046 		/*
4047 		 * Preserve ->was_full before destroying the old map.
4048 		 * For pools that weren't in the old map, ->was_full
4049 		 * should be false.
4050 		 */
4051 		for (n = rb_first(&newmap->pg_pools); n; n = rb_next(n)) {
4052 			struct ceph_pg_pool_info *pi =
4053 			    rb_entry(n, struct ceph_pg_pool_info, node);
4054 			struct ceph_pg_pool_info *old_pi;
4055 
4056 			old_pi = ceph_pg_pool_by_id(osdc->osdmap, pi->id);
4057 			if (old_pi)
4058 				pi->was_full = old_pi->was_full;
4059 			else
4060 				WARN_ON(pi->was_full);
4061 		}
4062 
4063 		if (osdc->osdmap->epoch &&
4064 		    osdc->osdmap->epoch + 1 < newmap->epoch) {
4065 			WARN_ON(incremental);
4066 			skipped_map = true;
4067 		}
4068 
4069 		ceph_osdmap_destroy(osdc->osdmap);
4070 		osdc->osdmap = newmap;
4071 	}
4072 
4073 	was_full &= !ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL);
4074 	scan_requests(&osdc->homeless_osd, skipped_map, was_full, true,
4075 		      need_resend, need_resend_linger);
4076 
4077 	for (n = rb_first(&osdc->osds); n; ) {
4078 		struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
4079 
4080 		n = rb_next(n); /* close_osd() */
4081 
4082 		scan_requests(osd, skipped_map, was_full, true, need_resend,
4083 			      need_resend_linger);
4084 		if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
4085 		    memcmp(&osd->o_con.peer_addr,
4086 			   ceph_osd_addr(osdc->osdmap, osd->o_osd),
4087 			   sizeof(struct ceph_entity_addr)))
4088 			close_osd(osd);
4089 	}
4090 
4091 	return 0;
4092 }
4093 
4094 static void kick_requests(struct ceph_osd_client *osdc,
4095 			  struct rb_root *need_resend,
4096 			  struct list_head *need_resend_linger)
4097 {
4098 	struct ceph_osd_linger_request *lreq, *nlreq;
4099 	enum calc_target_result ct_res;
4100 	struct rb_node *n;
4101 
4102 	/* make sure need_resend targets reflect latest map */
4103 	for (n = rb_first(need_resend); n; ) {
4104 		struct ceph_osd_request *req =
4105 		    rb_entry(n, struct ceph_osd_request, r_node);
4106 
4107 		n = rb_next(n);
4108 
4109 		if (req->r_t.epoch < osdc->osdmap->epoch) {
4110 			ct_res = calc_target(osdc, &req->r_t, false);
4111 			if (ct_res == CALC_TARGET_POOL_DNE) {
4112 				erase_request(need_resend, req);
4113 				check_pool_dne(req);
4114 			}
4115 		}
4116 	}
4117 
4118 	for (n = rb_first(need_resend); n; ) {
4119 		struct ceph_osd_request *req =
4120 		    rb_entry(n, struct ceph_osd_request, r_node);
4121 		struct ceph_osd *osd;
4122 
4123 		n = rb_next(n);
4124 		erase_request(need_resend, req); /* before link_request() */
4125 
4126 		osd = lookup_create_osd(osdc, req->r_t.osd, true);
4127 		link_request(osd, req);
4128 		if (!req->r_linger) {
4129 			if (!osd_homeless(osd) && !req->r_t.paused)
4130 				send_request(req);
4131 		} else {
4132 			cancel_linger_request(req);
4133 		}
4134 	}
4135 
4136 	list_for_each_entry_safe(lreq, nlreq, need_resend_linger, scan_item) {
4137 		if (!osd_homeless(lreq->osd))
4138 			send_linger(lreq);
4139 
4140 		list_del_init(&lreq->scan_item);
4141 	}
4142 }
4143 
4144 /*
4145  * Process updated osd map.
4146  *
4147  * The message contains any number of incremental and full maps, normally
4148  * indicating some sort of topology change in the cluster.  Kick requests
4149  * off to different OSDs as needed.
4150  */
4151 void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
4152 {
4153 	void *p = msg->front.iov_base;
4154 	void *const end = p + msg->front.iov_len;
4155 	u32 nr_maps, maplen;
4156 	u32 epoch;
4157 	struct ceph_fsid fsid;
4158 	struct rb_root need_resend = RB_ROOT;
4159 	LIST_HEAD(need_resend_linger);
4160 	bool handled_incremental = false;
4161 	bool was_pauserd, was_pausewr;
4162 	bool pauserd, pausewr;
4163 	int err;
4164 
4165 	dout("%s have %u\n", __func__, osdc->osdmap->epoch);
4166 	down_write(&osdc->lock);
4167 
4168 	/* verify fsid */
4169 	ceph_decode_need(&p, end, sizeof(fsid), bad);
4170 	ceph_decode_copy(&p, &fsid, sizeof(fsid));
4171 	if (ceph_check_fsid(osdc->client, &fsid) < 0)
4172 		goto bad;
4173 
4174 	was_pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
4175 	was_pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
4176 		      ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
4177 		      have_pool_full(osdc);
4178 
4179 	/* incremental maps */
4180 	ceph_decode_32_safe(&p, end, nr_maps, bad);
4181 	dout(" %d inc maps\n", nr_maps);
4182 	while (nr_maps > 0) {
4183 		ceph_decode_need(&p, end, 2*sizeof(u32), bad);
4184 		epoch = ceph_decode_32(&p);
4185 		maplen = ceph_decode_32(&p);
4186 		ceph_decode_need(&p, end, maplen, bad);
4187 		if (osdc->osdmap->epoch &&
4188 		    osdc->osdmap->epoch + 1 == epoch) {
4189 			dout("applying incremental map %u len %d\n",
4190 			     epoch, maplen);
4191 			err = handle_one_map(osdc, p, p + maplen, true,
4192 					     &need_resend, &need_resend_linger);
4193 			if (err)
4194 				goto bad;
4195 			handled_incremental = true;
4196 		} else {
4197 			dout("ignoring incremental map %u len %d\n",
4198 			     epoch, maplen);
4199 		}
4200 		p += maplen;
4201 		nr_maps--;
4202 	}
4203 	if (handled_incremental)
4204 		goto done;
4205 
4206 	/* full maps */
4207 	ceph_decode_32_safe(&p, end, nr_maps, bad);
4208 	dout(" %d full maps\n", nr_maps);
4209 	while (nr_maps) {
4210 		ceph_decode_need(&p, end, 2*sizeof(u32), bad);
4211 		epoch = ceph_decode_32(&p);
4212 		maplen = ceph_decode_32(&p);
4213 		ceph_decode_need(&p, end, maplen, bad);
4214 		if (nr_maps > 1) {
4215 			dout("skipping non-latest full map %u len %d\n",
4216 			     epoch, maplen);
4217 		} else if (osdc->osdmap->epoch >= epoch) {
4218 			dout("skipping full map %u len %d, "
4219 			     "older than our %u\n", epoch, maplen,
4220 			     osdc->osdmap->epoch);
4221 		} else {
4222 			dout("taking full map %u len %d\n", epoch, maplen);
4223 			err = handle_one_map(osdc, p, p + maplen, false,
4224 					     &need_resend, &need_resend_linger);
4225 			if (err)
4226 				goto bad;
4227 		}
4228 		p += maplen;
4229 		nr_maps--;
4230 	}
4231 
4232 done:
4233 	/*
4234 	 * subscribe to subsequent osdmap updates if full to ensure
4235 	 * we find out when we are no longer full and stop returning
4236 	 * ENOSPC.
4237 	 */
4238 	pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
4239 	pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
4240 		  ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
4241 		  have_pool_full(osdc);
4242 	if (was_pauserd || was_pausewr || pauserd || pausewr ||
4243 	    osdc->osdmap->epoch < osdc->epoch_barrier)
4244 		maybe_request_map(osdc);
4245 
4246 	kick_requests(osdc, &need_resend, &need_resend_linger);
4247 
4248 	ceph_osdc_abort_on_full(osdc);
4249 	ceph_monc_got_map(&osdc->client->monc, CEPH_SUB_OSDMAP,
4250 			  osdc->osdmap->epoch);
4251 	up_write(&osdc->lock);
4252 	wake_up_all(&osdc->client->auth_wq);
4253 	return;
4254 
4255 bad:
4256 	pr_err("osdc handle_map corrupt msg\n");
4257 	ceph_msg_dump(msg);
4258 	up_write(&osdc->lock);
4259 }
4260 
4261 /*
4262  * Resubmit requests pending on the given osd.
4263  */
4264 static void kick_osd_requests(struct ceph_osd *osd)
4265 {
4266 	struct rb_node *n;
4267 
4268 	clear_backoffs(osd);
4269 
4270 	for (n = rb_first(&osd->o_requests); n; ) {
4271 		struct ceph_osd_request *req =
4272 		    rb_entry(n, struct ceph_osd_request, r_node);
4273 
4274 		n = rb_next(n); /* cancel_linger_request() */
4275 
4276 		if (!req->r_linger) {
4277 			if (!req->r_t.paused)
4278 				send_request(req);
4279 		} else {
4280 			cancel_linger_request(req);
4281 		}
4282 	}
4283 	for (n = rb_first(&osd->o_linger_requests); n; n = rb_next(n)) {
4284 		struct ceph_osd_linger_request *lreq =
4285 		    rb_entry(n, struct ceph_osd_linger_request, node);
4286 
4287 		send_linger(lreq);
4288 	}
4289 }
4290 
4291 /*
4292  * If the osd connection drops, we need to resubmit all requests.
4293  */
4294 static void osd_fault(struct ceph_connection *con)
4295 {
4296 	struct ceph_osd *osd = con->private;
4297 	struct ceph_osd_client *osdc = osd->o_osdc;
4298 
4299 	dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
4300 
4301 	down_write(&osdc->lock);
4302 	if (!osd_registered(osd)) {
4303 		dout("%s osd%d unknown\n", __func__, osd->o_osd);
4304 		goto out_unlock;
4305 	}
4306 
4307 	if (!reopen_osd(osd))
4308 		kick_osd_requests(osd);
4309 	maybe_request_map(osdc);
4310 
4311 out_unlock:
4312 	up_write(&osdc->lock);
4313 }
4314 
4315 struct MOSDBackoff {
4316 	struct ceph_spg spgid;
4317 	u32 map_epoch;
4318 	u8 op;
4319 	u64 id;
4320 	struct ceph_hobject_id *begin;
4321 	struct ceph_hobject_id *end;
4322 };
4323 
4324 static int decode_MOSDBackoff(const struct ceph_msg *msg, struct MOSDBackoff *m)
4325 {
4326 	void *p = msg->front.iov_base;
4327 	void *const end = p + msg->front.iov_len;
4328 	u8 struct_v;
4329 	u32 struct_len;
4330 	int ret;
4331 
4332 	ret = ceph_start_decoding(&p, end, 1, "spg_t", &struct_v, &struct_len);
4333 	if (ret)
4334 		return ret;
4335 
4336 	ret = ceph_decode_pgid(&p, end, &m->spgid.pgid);
4337 	if (ret)
4338 		return ret;
4339 
4340 	ceph_decode_8_safe(&p, end, m->spgid.shard, e_inval);
4341 	ceph_decode_32_safe(&p, end, m->map_epoch, e_inval);
4342 	ceph_decode_8_safe(&p, end, m->op, e_inval);
4343 	ceph_decode_64_safe(&p, end, m->id, e_inval);
4344 
4345 	m->begin = kzalloc(sizeof(*m->begin), GFP_NOIO);
4346 	if (!m->begin)
4347 		return -ENOMEM;
4348 
4349 	ret = decode_hoid(&p, end, m->begin);
4350 	if (ret) {
4351 		free_hoid(m->begin);
4352 		return ret;
4353 	}
4354 
4355 	m->end = kzalloc(sizeof(*m->end), GFP_NOIO);
4356 	if (!m->end) {
4357 		free_hoid(m->begin);
4358 		return -ENOMEM;
4359 	}
4360 
4361 	ret = decode_hoid(&p, end, m->end);
4362 	if (ret) {
4363 		free_hoid(m->begin);
4364 		free_hoid(m->end);
4365 		return ret;
4366 	}
4367 
4368 	return 0;
4369 
4370 e_inval:
4371 	return -EINVAL;
4372 }
4373 
4374 static struct ceph_msg *create_backoff_message(
4375 				const struct ceph_osd_backoff *backoff,
4376 				u32 map_epoch)
4377 {
4378 	struct ceph_msg *msg;
4379 	void *p, *end;
4380 	int msg_size;
4381 
4382 	msg_size = CEPH_ENCODING_START_BLK_LEN +
4383 			CEPH_PGID_ENCODING_LEN + 1; /* spgid */
4384 	msg_size += 4 + 1 + 8; /* map_epoch, op, id */
4385 	msg_size += CEPH_ENCODING_START_BLK_LEN +
4386 			hoid_encoding_size(backoff->begin);
4387 	msg_size += CEPH_ENCODING_START_BLK_LEN +
4388 			hoid_encoding_size(backoff->end);
4389 
4390 	msg = ceph_msg_new(CEPH_MSG_OSD_BACKOFF, msg_size, GFP_NOIO, true);
4391 	if (!msg)
4392 		return NULL;
4393 
4394 	p = msg->front.iov_base;
4395 	end = p + msg->front_alloc_len;
4396 
4397 	encode_spgid(&p, &backoff->spgid);
4398 	ceph_encode_32(&p, map_epoch);
4399 	ceph_encode_8(&p, CEPH_OSD_BACKOFF_OP_ACK_BLOCK);
4400 	ceph_encode_64(&p, backoff->id);
4401 	encode_hoid(&p, end, backoff->begin);
4402 	encode_hoid(&p, end, backoff->end);
4403 	BUG_ON(p != end);
4404 
4405 	msg->front.iov_len = p - msg->front.iov_base;
4406 	msg->hdr.version = cpu_to_le16(1); /* MOSDBackoff v1 */
4407 	msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
4408 
4409 	return msg;
4410 }
4411 
4412 static void handle_backoff_block(struct ceph_osd *osd, struct MOSDBackoff *m)
4413 {
4414 	struct ceph_spg_mapping *spg;
4415 	struct ceph_osd_backoff *backoff;
4416 	struct ceph_msg *msg;
4417 
4418 	dout("%s osd%d spgid %llu.%xs%d id %llu\n", __func__, osd->o_osd,
4419 	     m->spgid.pgid.pool, m->spgid.pgid.seed, m->spgid.shard, m->id);
4420 
4421 	spg = lookup_spg_mapping(&osd->o_backoff_mappings, &m->spgid);
4422 	if (!spg) {
4423 		spg = alloc_spg_mapping();
4424 		if (!spg) {
4425 			pr_err("%s failed to allocate spg\n", __func__);
4426 			return;
4427 		}
4428 		spg->spgid = m->spgid; /* struct */
4429 		insert_spg_mapping(&osd->o_backoff_mappings, spg);
4430 	}
4431 
4432 	backoff = alloc_backoff();
4433 	if (!backoff) {
4434 		pr_err("%s failed to allocate backoff\n", __func__);
4435 		return;
4436 	}
4437 	backoff->spgid = m->spgid; /* struct */
4438 	backoff->id = m->id;
4439 	backoff->begin = m->begin;
4440 	m->begin = NULL; /* backoff now owns this */
4441 	backoff->end = m->end;
4442 	m->end = NULL;   /* ditto */
4443 
4444 	insert_backoff(&spg->backoffs, backoff);
4445 	insert_backoff_by_id(&osd->o_backoffs_by_id, backoff);
4446 
4447 	/*
4448 	 * Ack with original backoff's epoch so that the OSD can
4449 	 * discard this if there was a PG split.
4450 	 */
4451 	msg = create_backoff_message(backoff, m->map_epoch);
4452 	if (!msg) {
4453 		pr_err("%s failed to allocate msg\n", __func__);
4454 		return;
4455 	}
4456 	ceph_con_send(&osd->o_con, msg);
4457 }
4458 
4459 static bool target_contained_by(const struct ceph_osd_request_target *t,
4460 				const struct ceph_hobject_id *begin,
4461 				const struct ceph_hobject_id *end)
4462 {
4463 	struct ceph_hobject_id hoid;
4464 	int cmp;
4465 
4466 	hoid_fill_from_target(&hoid, t);
4467 	cmp = hoid_compare(&hoid, begin);
4468 	return !cmp || (cmp > 0 && hoid_compare(&hoid, end) < 0);
4469 }
4470 
4471 static void handle_backoff_unblock(struct ceph_osd *osd,
4472 				   const struct MOSDBackoff *m)
4473 {
4474 	struct ceph_spg_mapping *spg;
4475 	struct ceph_osd_backoff *backoff;
4476 	struct rb_node *n;
4477 
4478 	dout("%s osd%d spgid %llu.%xs%d id %llu\n", __func__, osd->o_osd,
4479 	     m->spgid.pgid.pool, m->spgid.pgid.seed, m->spgid.shard, m->id);
4480 
4481 	backoff = lookup_backoff_by_id(&osd->o_backoffs_by_id, m->id);
4482 	if (!backoff) {
4483 		pr_err("%s osd%d spgid %llu.%xs%d id %llu backoff dne\n",
4484 		       __func__, osd->o_osd, m->spgid.pgid.pool,
4485 		       m->spgid.pgid.seed, m->spgid.shard, m->id);
4486 		return;
4487 	}
4488 
4489 	if (hoid_compare(backoff->begin, m->begin) &&
4490 	    hoid_compare(backoff->end, m->end)) {
4491 		pr_err("%s osd%d spgid %llu.%xs%d id %llu bad range?\n",
4492 		       __func__, osd->o_osd, m->spgid.pgid.pool,
4493 		       m->spgid.pgid.seed, m->spgid.shard, m->id);
4494 		/* unblock it anyway... */
4495 	}
4496 
4497 	spg = lookup_spg_mapping(&osd->o_backoff_mappings, &backoff->spgid);
4498 	BUG_ON(!spg);
4499 
4500 	erase_backoff(&spg->backoffs, backoff);
4501 	erase_backoff_by_id(&osd->o_backoffs_by_id, backoff);
4502 	free_backoff(backoff);
4503 
4504 	if (RB_EMPTY_ROOT(&spg->backoffs)) {
4505 		erase_spg_mapping(&osd->o_backoff_mappings, spg);
4506 		free_spg_mapping(spg);
4507 	}
4508 
4509 	for (n = rb_first(&osd->o_requests); n; n = rb_next(n)) {
4510 		struct ceph_osd_request *req =
4511 		    rb_entry(n, struct ceph_osd_request, r_node);
4512 
4513 		if (!ceph_spg_compare(&req->r_t.spgid, &m->spgid)) {
4514 			/*
4515 			 * Match against @m, not @backoff -- the PG may
4516 			 * have split on the OSD.
4517 			 */
4518 			if (target_contained_by(&req->r_t, m->begin, m->end)) {
4519 				/*
4520 				 * If no other installed backoff applies,
4521 				 * resend.
4522 				 */
4523 				send_request(req);
4524 			}
4525 		}
4526 	}
4527 }
4528 
4529 static void handle_backoff(struct ceph_osd *osd, struct ceph_msg *msg)
4530 {
4531 	struct ceph_osd_client *osdc = osd->o_osdc;
4532 	struct MOSDBackoff m;
4533 	int ret;
4534 
4535 	down_read(&osdc->lock);
4536 	if (!osd_registered(osd)) {
4537 		dout("%s osd%d unknown\n", __func__, osd->o_osd);
4538 		up_read(&osdc->lock);
4539 		return;
4540 	}
4541 	WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num));
4542 
4543 	mutex_lock(&osd->lock);
4544 	ret = decode_MOSDBackoff(msg, &m);
4545 	if (ret) {
4546 		pr_err("failed to decode MOSDBackoff: %d\n", ret);
4547 		ceph_msg_dump(msg);
4548 		goto out_unlock;
4549 	}
4550 
4551 	switch (m.op) {
4552 	case CEPH_OSD_BACKOFF_OP_BLOCK:
4553 		handle_backoff_block(osd, &m);
4554 		break;
4555 	case CEPH_OSD_BACKOFF_OP_UNBLOCK:
4556 		handle_backoff_unblock(osd, &m);
4557 		break;
4558 	default:
4559 		pr_err("%s osd%d unknown op %d\n", __func__, osd->o_osd, m.op);
4560 	}
4561 
4562 	free_hoid(m.begin);
4563 	free_hoid(m.end);
4564 
4565 out_unlock:
4566 	mutex_unlock(&osd->lock);
4567 	up_read(&osdc->lock);
4568 }
4569 
4570 /*
4571  * Process osd watch notifications
4572  */
4573 static void handle_watch_notify(struct ceph_osd_client *osdc,
4574 				struct ceph_msg *msg)
4575 {
4576 	void *p = msg->front.iov_base;
4577 	void *const end = p + msg->front.iov_len;
4578 	struct ceph_osd_linger_request *lreq;
4579 	struct linger_work *lwork;
4580 	u8 proto_ver, opcode;
4581 	u64 cookie, notify_id;
4582 	u64 notifier_id = 0;
4583 	s32 return_code = 0;
4584 	void *payload = NULL;
4585 	u32 payload_len = 0;
4586 
4587 	ceph_decode_8_safe(&p, end, proto_ver, bad);
4588 	ceph_decode_8_safe(&p, end, opcode, bad);
4589 	ceph_decode_64_safe(&p, end, cookie, bad);
4590 	p += 8; /* skip ver */
4591 	ceph_decode_64_safe(&p, end, notify_id, bad);
4592 
4593 	if (proto_ver >= 1) {
4594 		ceph_decode_32_safe(&p, end, payload_len, bad);
4595 		ceph_decode_need(&p, end, payload_len, bad);
4596 		payload = p;
4597 		p += payload_len;
4598 	}
4599 
4600 	if (le16_to_cpu(msg->hdr.version) >= 2)
4601 		ceph_decode_32_safe(&p, end, return_code, bad);
4602 
4603 	if (le16_to_cpu(msg->hdr.version) >= 3)
4604 		ceph_decode_64_safe(&p, end, notifier_id, bad);
4605 
4606 	down_read(&osdc->lock);
4607 	lreq = lookup_linger_osdc(&osdc->linger_requests, cookie);
4608 	if (!lreq) {
4609 		dout("%s opcode %d cookie %llu dne\n", __func__, opcode,
4610 		     cookie);
4611 		goto out_unlock_osdc;
4612 	}
4613 
4614 	mutex_lock(&lreq->lock);
4615 	dout("%s opcode %d cookie %llu lreq %p is_watch %d\n", __func__,
4616 	     opcode, cookie, lreq, lreq->is_watch);
4617 	if (opcode == CEPH_WATCH_EVENT_DISCONNECT) {
4618 		if (!lreq->last_error) {
4619 			lreq->last_error = -ENOTCONN;
4620 			queue_watch_error(lreq);
4621 		}
4622 	} else if (!lreq->is_watch) {
4623 		/* CEPH_WATCH_EVENT_NOTIFY_COMPLETE */
4624 		if (lreq->notify_id && lreq->notify_id != notify_id) {
4625 			dout("lreq %p notify_id %llu != %llu, ignoring\n", lreq,
4626 			     lreq->notify_id, notify_id);
4627 		} else if (!completion_done(&lreq->notify_finish_wait)) {
4628 			struct ceph_msg_data *data =
4629 			    msg->num_data_items ? &msg->data[0] : NULL;
4630 
4631 			if (data) {
4632 				if (lreq->preply_pages) {
4633 					WARN_ON(data->type !=
4634 							CEPH_MSG_DATA_PAGES);
4635 					*lreq->preply_pages = data->pages;
4636 					*lreq->preply_len = data->length;
4637 					data->own_pages = false;
4638 				}
4639 			}
4640 			lreq->notify_finish_error = return_code;
4641 			complete_all(&lreq->notify_finish_wait);
4642 		}
4643 	} else {
4644 		/* CEPH_WATCH_EVENT_NOTIFY */
4645 		lwork = lwork_alloc(lreq, do_watch_notify);
4646 		if (!lwork) {
4647 			pr_err("failed to allocate notify-lwork\n");
4648 			goto out_unlock_lreq;
4649 		}
4650 
4651 		lwork->notify.notify_id = notify_id;
4652 		lwork->notify.notifier_id = notifier_id;
4653 		lwork->notify.payload = payload;
4654 		lwork->notify.payload_len = payload_len;
4655 		lwork->notify.msg = ceph_msg_get(msg);
4656 		lwork_queue(lwork);
4657 	}
4658 
4659 out_unlock_lreq:
4660 	mutex_unlock(&lreq->lock);
4661 out_unlock_osdc:
4662 	up_read(&osdc->lock);
4663 	return;
4664 
4665 bad:
4666 	pr_err("osdc handle_watch_notify corrupt msg\n");
4667 }
4668 
4669 /*
4670  * Register request, send initial attempt.
4671  */
4672 void ceph_osdc_start_request(struct ceph_osd_client *osdc,
4673 			     struct ceph_osd_request *req)
4674 {
4675 	down_read(&osdc->lock);
4676 	submit_request(req, false);
4677 	up_read(&osdc->lock);
4678 }
4679 EXPORT_SYMBOL(ceph_osdc_start_request);
4680 
4681 /*
4682  * Unregister request.  If @req was registered, it isn't completed:
4683  * r_result isn't set and __complete_request() isn't invoked.
4684  *
4685  * If @req wasn't registered, this call may have raced with
4686  * handle_reply(), in which case r_result would already be set and
4687  * __complete_request() would be getting invoked, possibly even
4688  * concurrently with this call.
4689  */
4690 void ceph_osdc_cancel_request(struct ceph_osd_request *req)
4691 {
4692 	struct ceph_osd_client *osdc = req->r_osdc;
4693 
4694 	down_write(&osdc->lock);
4695 	if (req->r_osd)
4696 		cancel_request(req);
4697 	up_write(&osdc->lock);
4698 }
4699 EXPORT_SYMBOL(ceph_osdc_cancel_request);
4700 
4701 /*
4702  * @timeout: in jiffies, 0 means "wait forever"
4703  */
4704 static int wait_request_timeout(struct ceph_osd_request *req,
4705 				unsigned long timeout)
4706 {
4707 	long left;
4708 
4709 	dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
4710 	left = wait_for_completion_killable_timeout(&req->r_completion,
4711 						ceph_timeout_jiffies(timeout));
4712 	if (left <= 0) {
4713 		left = left ?: -ETIMEDOUT;
4714 		ceph_osdc_cancel_request(req);
4715 	} else {
4716 		left = req->r_result; /* completed */
4717 	}
4718 
4719 	return left;
4720 }
4721 
4722 /*
4723  * wait for a request to complete
4724  */
4725 int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
4726 			   struct ceph_osd_request *req)
4727 {
4728 	return wait_request_timeout(req, 0);
4729 }
4730 EXPORT_SYMBOL(ceph_osdc_wait_request);
4731 
4732 /*
4733  * sync - wait for all in-flight requests to flush.  avoid starvation.
4734  */
4735 void ceph_osdc_sync(struct ceph_osd_client *osdc)
4736 {
4737 	struct rb_node *n, *p;
4738 	u64 last_tid = atomic64_read(&osdc->last_tid);
4739 
4740 again:
4741 	down_read(&osdc->lock);
4742 	for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
4743 		struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
4744 
4745 		mutex_lock(&osd->lock);
4746 		for (p = rb_first(&osd->o_requests); p; p = rb_next(p)) {
4747 			struct ceph_osd_request *req =
4748 			    rb_entry(p, struct ceph_osd_request, r_node);
4749 
4750 			if (req->r_tid > last_tid)
4751 				break;
4752 
4753 			if (!(req->r_flags & CEPH_OSD_FLAG_WRITE))
4754 				continue;
4755 
4756 			ceph_osdc_get_request(req);
4757 			mutex_unlock(&osd->lock);
4758 			up_read(&osdc->lock);
4759 			dout("%s waiting on req %p tid %llu last_tid %llu\n",
4760 			     __func__, req, req->r_tid, last_tid);
4761 			wait_for_completion(&req->r_completion);
4762 			ceph_osdc_put_request(req);
4763 			goto again;
4764 		}
4765 
4766 		mutex_unlock(&osd->lock);
4767 	}
4768 
4769 	up_read(&osdc->lock);
4770 	dout("%s done last_tid %llu\n", __func__, last_tid);
4771 }
4772 EXPORT_SYMBOL(ceph_osdc_sync);
4773 
4774 /*
4775  * Returns a handle, caller owns a ref.
4776  */
4777 struct ceph_osd_linger_request *
4778 ceph_osdc_watch(struct ceph_osd_client *osdc,
4779 		struct ceph_object_id *oid,
4780 		struct ceph_object_locator *oloc,
4781 		rados_watchcb2_t wcb,
4782 		rados_watcherrcb_t errcb,
4783 		void *data)
4784 {
4785 	struct ceph_osd_linger_request *lreq;
4786 	int ret;
4787 
4788 	lreq = linger_alloc(osdc);
4789 	if (!lreq)
4790 		return ERR_PTR(-ENOMEM);
4791 
4792 	lreq->is_watch = true;
4793 	lreq->wcb = wcb;
4794 	lreq->errcb = errcb;
4795 	lreq->data = data;
4796 	lreq->watch_valid_thru = jiffies;
4797 
4798 	ceph_oid_copy(&lreq->t.base_oid, oid);
4799 	ceph_oloc_copy(&lreq->t.base_oloc, oloc);
4800 	lreq->t.flags = CEPH_OSD_FLAG_WRITE;
4801 	ktime_get_real_ts64(&lreq->mtime);
4802 
4803 	linger_submit(lreq);
4804 	ret = linger_reg_commit_wait(lreq);
4805 	if (ret) {
4806 		linger_cancel(lreq);
4807 		goto err_put_lreq;
4808 	}
4809 
4810 	return lreq;
4811 
4812 err_put_lreq:
4813 	linger_put(lreq);
4814 	return ERR_PTR(ret);
4815 }
4816 EXPORT_SYMBOL(ceph_osdc_watch);
4817 
4818 /*
4819  * Releases a ref.
4820  *
4821  * Times out after mount_timeout to preserve rbd unmap behaviour
4822  * introduced in 2894e1d76974 ("rbd: timeout watch teardown on unmap
4823  * with mount_timeout").
4824  */
4825 int ceph_osdc_unwatch(struct ceph_osd_client *osdc,
4826 		      struct ceph_osd_linger_request *lreq)
4827 {
4828 	struct ceph_options *opts = osdc->client->options;
4829 	struct ceph_osd_request *req;
4830 	int ret;
4831 
4832 	req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
4833 	if (!req)
4834 		return -ENOMEM;
4835 
4836 	ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
4837 	ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
4838 	req->r_flags = CEPH_OSD_FLAG_WRITE;
4839 	ktime_get_real_ts64(&req->r_mtime);
4840 	osd_req_op_watch_init(req, 0, CEPH_OSD_WATCH_OP_UNWATCH,
4841 			      lreq->linger_id, 0);
4842 
4843 	ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
4844 	if (ret)
4845 		goto out_put_req;
4846 
4847 	ceph_osdc_start_request(osdc, req);
4848 	linger_cancel(lreq);
4849 	linger_put(lreq);
4850 	ret = wait_request_timeout(req, opts->mount_timeout);
4851 
4852 out_put_req:
4853 	ceph_osdc_put_request(req);
4854 	return ret;
4855 }
4856 EXPORT_SYMBOL(ceph_osdc_unwatch);
4857 
4858 static int osd_req_op_notify_ack_init(struct ceph_osd_request *req, int which,
4859 				      u64 notify_id, u64 cookie, void *payload,
4860 				      u32 payload_len)
4861 {
4862 	struct ceph_osd_req_op *op;
4863 	struct ceph_pagelist *pl;
4864 	int ret;
4865 
4866 	op = osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY_ACK, 0);
4867 
4868 	pl = ceph_pagelist_alloc(GFP_NOIO);
4869 	if (!pl)
4870 		return -ENOMEM;
4871 
4872 	ret = ceph_pagelist_encode_64(pl, notify_id);
4873 	ret |= ceph_pagelist_encode_64(pl, cookie);
4874 	if (payload) {
4875 		ret |= ceph_pagelist_encode_32(pl, payload_len);
4876 		ret |= ceph_pagelist_append(pl, payload, payload_len);
4877 	} else {
4878 		ret |= ceph_pagelist_encode_32(pl, 0);
4879 	}
4880 	if (ret) {
4881 		ceph_pagelist_release(pl);
4882 		return -ENOMEM;
4883 	}
4884 
4885 	ceph_osd_data_pagelist_init(&op->notify_ack.request_data, pl);
4886 	op->indata_len = pl->length;
4887 	return 0;
4888 }
4889 
4890 int ceph_osdc_notify_ack(struct ceph_osd_client *osdc,
4891 			 struct ceph_object_id *oid,
4892 			 struct ceph_object_locator *oloc,
4893 			 u64 notify_id,
4894 			 u64 cookie,
4895 			 void *payload,
4896 			 u32 payload_len)
4897 {
4898 	struct ceph_osd_request *req;
4899 	int ret;
4900 
4901 	req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
4902 	if (!req)
4903 		return -ENOMEM;
4904 
4905 	ceph_oid_copy(&req->r_base_oid, oid);
4906 	ceph_oloc_copy(&req->r_base_oloc, oloc);
4907 	req->r_flags = CEPH_OSD_FLAG_READ;
4908 
4909 	ret = osd_req_op_notify_ack_init(req, 0, notify_id, cookie, payload,
4910 					 payload_len);
4911 	if (ret)
4912 		goto out_put_req;
4913 
4914 	ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
4915 	if (ret)
4916 		goto out_put_req;
4917 
4918 	ceph_osdc_start_request(osdc, req);
4919 	ret = ceph_osdc_wait_request(osdc, req);
4920 
4921 out_put_req:
4922 	ceph_osdc_put_request(req);
4923 	return ret;
4924 }
4925 EXPORT_SYMBOL(ceph_osdc_notify_ack);
4926 
4927 /*
4928  * @timeout: in seconds
4929  *
4930  * @preply_{pages,len} are initialized both on success and error.
4931  * The caller is responsible for:
4932  *
4933  *     ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len))
4934  */
4935 int ceph_osdc_notify(struct ceph_osd_client *osdc,
4936 		     struct ceph_object_id *oid,
4937 		     struct ceph_object_locator *oloc,
4938 		     void *payload,
4939 		     u32 payload_len,
4940 		     u32 timeout,
4941 		     struct page ***preply_pages,
4942 		     size_t *preply_len)
4943 {
4944 	struct ceph_osd_linger_request *lreq;
4945 	int ret;
4946 
4947 	WARN_ON(!timeout);
4948 	if (preply_pages) {
4949 		*preply_pages = NULL;
4950 		*preply_len = 0;
4951 	}
4952 
4953 	lreq = linger_alloc(osdc);
4954 	if (!lreq)
4955 		return -ENOMEM;
4956 
4957 	lreq->request_pl = ceph_pagelist_alloc(GFP_NOIO);
4958 	if (!lreq->request_pl) {
4959 		ret = -ENOMEM;
4960 		goto out_put_lreq;
4961 	}
4962 
4963 	ret = ceph_pagelist_encode_32(lreq->request_pl, 1); /* prot_ver */
4964 	ret |= ceph_pagelist_encode_32(lreq->request_pl, timeout);
4965 	ret |= ceph_pagelist_encode_32(lreq->request_pl, payload_len);
4966 	ret |= ceph_pagelist_append(lreq->request_pl, payload, payload_len);
4967 	if (ret) {
4968 		ret = -ENOMEM;
4969 		goto out_put_lreq;
4970 	}
4971 
4972 	/* for notify_id */
4973 	lreq->notify_id_pages = ceph_alloc_page_vector(1, GFP_NOIO);
4974 	if (IS_ERR(lreq->notify_id_pages)) {
4975 		ret = PTR_ERR(lreq->notify_id_pages);
4976 		lreq->notify_id_pages = NULL;
4977 		goto out_put_lreq;
4978 	}
4979 
4980 	lreq->preply_pages = preply_pages;
4981 	lreq->preply_len = preply_len;
4982 
4983 	ceph_oid_copy(&lreq->t.base_oid, oid);
4984 	ceph_oloc_copy(&lreq->t.base_oloc, oloc);
4985 	lreq->t.flags = CEPH_OSD_FLAG_READ;
4986 
4987 	linger_submit(lreq);
4988 	ret = linger_reg_commit_wait(lreq);
4989 	if (!ret)
4990 		ret = linger_notify_finish_wait(lreq,
4991 				 msecs_to_jiffies(2 * timeout * MSEC_PER_SEC));
4992 	else
4993 		dout("lreq %p failed to initiate notify %d\n", lreq, ret);
4994 
4995 	linger_cancel(lreq);
4996 out_put_lreq:
4997 	linger_put(lreq);
4998 	return ret;
4999 }
5000 EXPORT_SYMBOL(ceph_osdc_notify);
5001 
5002 /*
5003  * Return the number of milliseconds since the watch was last
5004  * confirmed, or an error.  If there is an error, the watch is no
5005  * longer valid, and should be destroyed with ceph_osdc_unwatch().
5006  */
5007 int ceph_osdc_watch_check(struct ceph_osd_client *osdc,
5008 			  struct ceph_osd_linger_request *lreq)
5009 {
5010 	unsigned long stamp, age;
5011 	int ret;
5012 
5013 	down_read(&osdc->lock);
5014 	mutex_lock(&lreq->lock);
5015 	stamp = lreq->watch_valid_thru;
5016 	if (!list_empty(&lreq->pending_lworks)) {
5017 		struct linger_work *lwork =
5018 		    list_first_entry(&lreq->pending_lworks,
5019 				     struct linger_work,
5020 				     pending_item);
5021 
5022 		if (time_before(lwork->queued_stamp, stamp))
5023 			stamp = lwork->queued_stamp;
5024 	}
5025 	age = jiffies - stamp;
5026 	dout("%s lreq %p linger_id %llu age %lu last_error %d\n", __func__,
5027 	     lreq, lreq->linger_id, age, lreq->last_error);
5028 	/* we are truncating to msecs, so return a safe upper bound */
5029 	ret = lreq->last_error ?: 1 + jiffies_to_msecs(age);
5030 
5031 	mutex_unlock(&lreq->lock);
5032 	up_read(&osdc->lock);
5033 	return ret;
5034 }
5035 
5036 static int decode_watcher(void **p, void *end, struct ceph_watch_item *item)
5037 {
5038 	u8 struct_v;
5039 	u32 struct_len;
5040 	int ret;
5041 
5042 	ret = ceph_start_decoding(p, end, 2, "watch_item_t",
5043 				  &struct_v, &struct_len);
5044 	if (ret)
5045 		goto bad;
5046 
5047 	ret = -EINVAL;
5048 	ceph_decode_copy_safe(p, end, &item->name, sizeof(item->name), bad);
5049 	ceph_decode_64_safe(p, end, item->cookie, bad);
5050 	ceph_decode_skip_32(p, end, bad); /* skip timeout seconds */
5051 
5052 	if (struct_v >= 2) {
5053 		ret = ceph_decode_entity_addr(p, end, &item->addr);
5054 		if (ret)
5055 			goto bad;
5056 	} else {
5057 		ret = 0;
5058 	}
5059 
5060 	dout("%s %s%llu cookie %llu addr %s\n", __func__,
5061 	     ENTITY_NAME(item->name), item->cookie,
5062 	     ceph_pr_addr(&item->addr));
5063 bad:
5064 	return ret;
5065 }
5066 
5067 static int decode_watchers(void **p, void *end,
5068 			   struct ceph_watch_item **watchers,
5069 			   u32 *num_watchers)
5070 {
5071 	u8 struct_v;
5072 	u32 struct_len;
5073 	int i;
5074 	int ret;
5075 
5076 	ret = ceph_start_decoding(p, end, 1, "obj_list_watch_response_t",
5077 				  &struct_v, &struct_len);
5078 	if (ret)
5079 		return ret;
5080 
5081 	*num_watchers = ceph_decode_32(p);
5082 	*watchers = kcalloc(*num_watchers, sizeof(**watchers), GFP_NOIO);
5083 	if (!*watchers)
5084 		return -ENOMEM;
5085 
5086 	for (i = 0; i < *num_watchers; i++) {
5087 		ret = decode_watcher(p, end, *watchers + i);
5088 		if (ret) {
5089 			kfree(*watchers);
5090 			return ret;
5091 		}
5092 	}
5093 
5094 	return 0;
5095 }
5096 
5097 /*
5098  * On success, the caller is responsible for:
5099  *
5100  *     kfree(watchers);
5101  */
5102 int ceph_osdc_list_watchers(struct ceph_osd_client *osdc,
5103 			    struct ceph_object_id *oid,
5104 			    struct ceph_object_locator *oloc,
5105 			    struct ceph_watch_item **watchers,
5106 			    u32 *num_watchers)
5107 {
5108 	struct ceph_osd_request *req;
5109 	struct page **pages;
5110 	int ret;
5111 
5112 	req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
5113 	if (!req)
5114 		return -ENOMEM;
5115 
5116 	ceph_oid_copy(&req->r_base_oid, oid);
5117 	ceph_oloc_copy(&req->r_base_oloc, oloc);
5118 	req->r_flags = CEPH_OSD_FLAG_READ;
5119 
5120 	pages = ceph_alloc_page_vector(1, GFP_NOIO);
5121 	if (IS_ERR(pages)) {
5122 		ret = PTR_ERR(pages);
5123 		goto out_put_req;
5124 	}
5125 
5126 	osd_req_op_init(req, 0, CEPH_OSD_OP_LIST_WATCHERS, 0);
5127 	ceph_osd_data_pages_init(osd_req_op_data(req, 0, list_watchers,
5128 						 response_data),
5129 				 pages, PAGE_SIZE, 0, false, true);
5130 
5131 	ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
5132 	if (ret)
5133 		goto out_put_req;
5134 
5135 	ceph_osdc_start_request(osdc, req);
5136 	ret = ceph_osdc_wait_request(osdc, req);
5137 	if (ret >= 0) {
5138 		void *p = page_address(pages[0]);
5139 		void *const end = p + req->r_ops[0].outdata_len;
5140 
5141 		ret = decode_watchers(&p, end, watchers, num_watchers);
5142 	}
5143 
5144 out_put_req:
5145 	ceph_osdc_put_request(req);
5146 	return ret;
5147 }
5148 EXPORT_SYMBOL(ceph_osdc_list_watchers);
5149 
5150 /*
5151  * Call all pending notify callbacks - for use after a watch is
5152  * unregistered, to make sure no more callbacks for it will be invoked
5153  */
5154 void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc)
5155 {
5156 	dout("%s osdc %p\n", __func__, osdc);
5157 	flush_workqueue(osdc->notify_wq);
5158 }
5159 EXPORT_SYMBOL(ceph_osdc_flush_notifies);
5160 
5161 void ceph_osdc_maybe_request_map(struct ceph_osd_client *osdc)
5162 {
5163 	down_read(&osdc->lock);
5164 	maybe_request_map(osdc);
5165 	up_read(&osdc->lock);
5166 }
5167 EXPORT_SYMBOL(ceph_osdc_maybe_request_map);
5168 
5169 /*
5170  * Execute an OSD class method on an object.
5171  *
5172  * @flags: CEPH_OSD_FLAG_*
5173  * @resp_len: in/out param for reply length
5174  */
5175 int ceph_osdc_call(struct ceph_osd_client *osdc,
5176 		   struct ceph_object_id *oid,
5177 		   struct ceph_object_locator *oloc,
5178 		   const char *class, const char *method,
5179 		   unsigned int flags,
5180 		   struct page *req_page, size_t req_len,
5181 		   struct page **resp_pages, size_t *resp_len)
5182 {
5183 	struct ceph_osd_request *req;
5184 	int ret;
5185 
5186 	if (req_len > PAGE_SIZE)
5187 		return -E2BIG;
5188 
5189 	req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
5190 	if (!req)
5191 		return -ENOMEM;
5192 
5193 	ceph_oid_copy(&req->r_base_oid, oid);
5194 	ceph_oloc_copy(&req->r_base_oloc, oloc);
5195 	req->r_flags = flags;
5196 
5197 	ret = osd_req_op_cls_init(req, 0, class, method);
5198 	if (ret)
5199 		goto out_put_req;
5200 
5201 	if (req_page)
5202 		osd_req_op_cls_request_data_pages(req, 0, &req_page, req_len,
5203 						  0, false, false);
5204 	if (resp_pages)
5205 		osd_req_op_cls_response_data_pages(req, 0, resp_pages,
5206 						   *resp_len, 0, false, false);
5207 
5208 	ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
5209 	if (ret)
5210 		goto out_put_req;
5211 
5212 	ceph_osdc_start_request(osdc, req);
5213 	ret = ceph_osdc_wait_request(osdc, req);
5214 	if (ret >= 0) {
5215 		ret = req->r_ops[0].rval;
5216 		if (resp_pages)
5217 			*resp_len = req->r_ops[0].outdata_len;
5218 	}
5219 
5220 out_put_req:
5221 	ceph_osdc_put_request(req);
5222 	return ret;
5223 }
5224 EXPORT_SYMBOL(ceph_osdc_call);
5225 
5226 /*
5227  * reset all osd connections
5228  */
5229 void ceph_osdc_reopen_osds(struct ceph_osd_client *osdc)
5230 {
5231 	struct rb_node *n;
5232 
5233 	down_write(&osdc->lock);
5234 	for (n = rb_first(&osdc->osds); n; ) {
5235 		struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
5236 
5237 		n = rb_next(n);
5238 		if (!reopen_osd(osd))
5239 			kick_osd_requests(osd);
5240 	}
5241 	up_write(&osdc->lock);
5242 }
5243 
5244 /*
5245  * init, shutdown
5246  */
5247 int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
5248 {
5249 	int err;
5250 
5251 	dout("init\n");
5252 	osdc->client = client;
5253 	init_rwsem(&osdc->lock);
5254 	osdc->osds = RB_ROOT;
5255 	INIT_LIST_HEAD(&osdc->osd_lru);
5256 	spin_lock_init(&osdc->osd_lru_lock);
5257 	osd_init(&osdc->homeless_osd);
5258 	osdc->homeless_osd.o_osdc = osdc;
5259 	osdc->homeless_osd.o_osd = CEPH_HOMELESS_OSD;
5260 	osdc->last_linger_id = CEPH_LINGER_ID_START;
5261 	osdc->linger_requests = RB_ROOT;
5262 	osdc->map_checks = RB_ROOT;
5263 	osdc->linger_map_checks = RB_ROOT;
5264 	INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
5265 	INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
5266 
5267 	err = -ENOMEM;
5268 	osdc->osdmap = ceph_osdmap_alloc();
5269 	if (!osdc->osdmap)
5270 		goto out;
5271 
5272 	osdc->req_mempool = mempool_create_slab_pool(10,
5273 						     ceph_osd_request_cache);
5274 	if (!osdc->req_mempool)
5275 		goto out_map;
5276 
5277 	err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
5278 				PAGE_SIZE, CEPH_OSD_SLAB_OPS, 10, "osd_op");
5279 	if (err < 0)
5280 		goto out_mempool;
5281 	err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
5282 				PAGE_SIZE, CEPH_OSD_SLAB_OPS, 10,
5283 				"osd_op_reply");
5284 	if (err < 0)
5285 		goto out_msgpool;
5286 
5287 	err = -ENOMEM;
5288 	osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
5289 	if (!osdc->notify_wq)
5290 		goto out_msgpool_reply;
5291 
5292 	osdc->completion_wq = create_singlethread_workqueue("ceph-completion");
5293 	if (!osdc->completion_wq)
5294 		goto out_notify_wq;
5295 
5296 	schedule_delayed_work(&osdc->timeout_work,
5297 			      osdc->client->options->osd_keepalive_timeout);
5298 	schedule_delayed_work(&osdc->osds_timeout_work,
5299 	    round_jiffies_relative(osdc->client->options->osd_idle_ttl));
5300 
5301 	return 0;
5302 
5303 out_notify_wq:
5304 	destroy_workqueue(osdc->notify_wq);
5305 out_msgpool_reply:
5306 	ceph_msgpool_destroy(&osdc->msgpool_op_reply);
5307 out_msgpool:
5308 	ceph_msgpool_destroy(&osdc->msgpool_op);
5309 out_mempool:
5310 	mempool_destroy(osdc->req_mempool);
5311 out_map:
5312 	ceph_osdmap_destroy(osdc->osdmap);
5313 out:
5314 	return err;
5315 }
5316 
5317 void ceph_osdc_stop(struct ceph_osd_client *osdc)
5318 {
5319 	destroy_workqueue(osdc->completion_wq);
5320 	destroy_workqueue(osdc->notify_wq);
5321 	cancel_delayed_work_sync(&osdc->timeout_work);
5322 	cancel_delayed_work_sync(&osdc->osds_timeout_work);
5323 
5324 	down_write(&osdc->lock);
5325 	while (!RB_EMPTY_ROOT(&osdc->osds)) {
5326 		struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
5327 						struct ceph_osd, o_node);
5328 		close_osd(osd);
5329 	}
5330 	up_write(&osdc->lock);
5331 	WARN_ON(refcount_read(&osdc->homeless_osd.o_ref) != 1);
5332 	osd_cleanup(&osdc->homeless_osd);
5333 
5334 	WARN_ON(!list_empty(&osdc->osd_lru));
5335 	WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_requests));
5336 	WARN_ON(!RB_EMPTY_ROOT(&osdc->map_checks));
5337 	WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_map_checks));
5338 	WARN_ON(atomic_read(&osdc->num_requests));
5339 	WARN_ON(atomic_read(&osdc->num_homeless));
5340 
5341 	ceph_osdmap_destroy(osdc->osdmap);
5342 	mempool_destroy(osdc->req_mempool);
5343 	ceph_msgpool_destroy(&osdc->msgpool_op);
5344 	ceph_msgpool_destroy(&osdc->msgpool_op_reply);
5345 }
5346 
5347 int osd_req_op_copy_from_init(struct ceph_osd_request *req,
5348 			      u64 src_snapid, u64 src_version,
5349 			      struct ceph_object_id *src_oid,
5350 			      struct ceph_object_locator *src_oloc,
5351 			      u32 src_fadvise_flags,
5352 			      u32 dst_fadvise_flags,
5353 			      u32 truncate_seq, u64 truncate_size,
5354 			      u8 copy_from_flags)
5355 {
5356 	struct ceph_osd_req_op *op;
5357 	struct page **pages;
5358 	void *p, *end;
5359 
5360 	pages = ceph_alloc_page_vector(1, GFP_KERNEL);
5361 	if (IS_ERR(pages))
5362 		return PTR_ERR(pages);
5363 
5364 	op = osd_req_op_init(req, 0, CEPH_OSD_OP_COPY_FROM2,
5365 			     dst_fadvise_flags);
5366 	op->copy_from.snapid = src_snapid;
5367 	op->copy_from.src_version = src_version;
5368 	op->copy_from.flags = copy_from_flags;
5369 	op->copy_from.src_fadvise_flags = src_fadvise_flags;
5370 
5371 	p = page_address(pages[0]);
5372 	end = p + PAGE_SIZE;
5373 	ceph_encode_string(&p, end, src_oid->name, src_oid->name_len);
5374 	encode_oloc(&p, end, src_oloc);
5375 	ceph_encode_32(&p, truncate_seq);
5376 	ceph_encode_64(&p, truncate_size);
5377 	op->indata_len = PAGE_SIZE - (end - p);
5378 
5379 	ceph_osd_data_pages_init(&op->copy_from.osd_data, pages,
5380 				 op->indata_len, 0, false, true);
5381 	return 0;
5382 }
5383 EXPORT_SYMBOL(osd_req_op_copy_from_init);
5384 
5385 int __init ceph_osdc_setup(void)
5386 {
5387 	size_t size = sizeof(struct ceph_osd_request) +
5388 	    CEPH_OSD_SLAB_OPS * sizeof(struct ceph_osd_req_op);
5389 
5390 	BUG_ON(ceph_osd_request_cache);
5391 	ceph_osd_request_cache = kmem_cache_create("ceph_osd_request", size,
5392 						   0, 0, NULL);
5393 
5394 	return ceph_osd_request_cache ? 0 : -ENOMEM;
5395 }
5396 
5397 void ceph_osdc_cleanup(void)
5398 {
5399 	BUG_ON(!ceph_osd_request_cache);
5400 	kmem_cache_destroy(ceph_osd_request_cache);
5401 	ceph_osd_request_cache = NULL;
5402 }
5403 
5404 /*
5405  * handle incoming message
5406  */
5407 static void osd_dispatch(struct ceph_connection *con, struct ceph_msg *msg)
5408 {
5409 	struct ceph_osd *osd = con->private;
5410 	struct ceph_osd_client *osdc = osd->o_osdc;
5411 	int type = le16_to_cpu(msg->hdr.type);
5412 
5413 	switch (type) {
5414 	case CEPH_MSG_OSD_MAP:
5415 		ceph_osdc_handle_map(osdc, msg);
5416 		break;
5417 	case CEPH_MSG_OSD_OPREPLY:
5418 		handle_reply(osd, msg);
5419 		break;
5420 	case CEPH_MSG_OSD_BACKOFF:
5421 		handle_backoff(osd, msg);
5422 		break;
5423 	case CEPH_MSG_WATCH_NOTIFY:
5424 		handle_watch_notify(osdc, msg);
5425 		break;
5426 
5427 	default:
5428 		pr_err("received unknown message type %d %s\n", type,
5429 		       ceph_msg_type_name(type));
5430 	}
5431 
5432 	ceph_msg_put(msg);
5433 }
5434 
5435 /* How much sparse data was requested? */
5436 static u64 sparse_data_requested(struct ceph_osd_request *req)
5437 {
5438 	u64 len = 0;
5439 
5440 	if (req->r_flags & CEPH_OSD_FLAG_READ) {
5441 		int i;
5442 
5443 		for (i = 0; i < req->r_num_ops; ++i) {
5444 			struct ceph_osd_req_op *op = &req->r_ops[i];
5445 
5446 			if (op->op == CEPH_OSD_OP_SPARSE_READ)
5447 				len += op->extent.length;
5448 		}
5449 	}
5450 	return len;
5451 }
5452 
5453 /*
5454  * Lookup and return message for incoming reply.  Don't try to do
5455  * anything about a larger than preallocated data portion of the
5456  * message at the moment - for now, just skip the message.
5457  */
5458 static struct ceph_msg *get_reply(struct ceph_connection *con,
5459 				  struct ceph_msg_header *hdr,
5460 				  int *skip)
5461 {
5462 	struct ceph_osd *osd = con->private;
5463 	struct ceph_osd_client *osdc = osd->o_osdc;
5464 	struct ceph_msg *m = NULL;
5465 	struct ceph_osd_request *req;
5466 	int front_len = le32_to_cpu(hdr->front_len);
5467 	int data_len = le32_to_cpu(hdr->data_len);
5468 	u64 tid = le64_to_cpu(hdr->tid);
5469 	u64 srlen;
5470 
5471 	down_read(&osdc->lock);
5472 	if (!osd_registered(osd)) {
5473 		dout("%s osd%d unknown, skipping\n", __func__, osd->o_osd);
5474 		*skip = 1;
5475 		goto out_unlock_osdc;
5476 	}
5477 	WARN_ON(osd->o_osd != le64_to_cpu(hdr->src.num));
5478 
5479 	mutex_lock(&osd->lock);
5480 	req = lookup_request(&osd->o_requests, tid);
5481 	if (!req) {
5482 		dout("%s osd%d tid %llu unknown, skipping\n", __func__,
5483 		     osd->o_osd, tid);
5484 		*skip = 1;
5485 		goto out_unlock_session;
5486 	}
5487 
5488 	ceph_msg_revoke_incoming(req->r_reply);
5489 
5490 	if (front_len > req->r_reply->front_alloc_len) {
5491 		pr_warn("%s osd%d tid %llu front %d > preallocated %d\n",
5492 			__func__, osd->o_osd, req->r_tid, front_len,
5493 			req->r_reply->front_alloc_len);
5494 		m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS,
5495 				 false);
5496 		if (!m)
5497 			goto out_unlock_session;
5498 		ceph_msg_put(req->r_reply);
5499 		req->r_reply = m;
5500 	}
5501 
5502 	srlen = sparse_data_requested(req);
5503 	if (!srlen && data_len > req->r_reply->data_length) {
5504 		pr_warn("%s osd%d tid %llu data %d > preallocated %zu, skipping\n",
5505 			__func__, osd->o_osd, req->r_tid, data_len,
5506 			req->r_reply->data_length);
5507 		m = NULL;
5508 		*skip = 1;
5509 		goto out_unlock_session;
5510 	}
5511 
5512 	m = ceph_msg_get(req->r_reply);
5513 	m->sparse_read_total = srlen;
5514 
5515 	dout("get_reply tid %lld %p\n", tid, m);
5516 
5517 out_unlock_session:
5518 	mutex_unlock(&osd->lock);
5519 out_unlock_osdc:
5520 	up_read(&osdc->lock);
5521 	return m;
5522 }
5523 
5524 static struct ceph_msg *alloc_msg_with_page_vector(struct ceph_msg_header *hdr)
5525 {
5526 	struct ceph_msg *m;
5527 	int type = le16_to_cpu(hdr->type);
5528 	u32 front_len = le32_to_cpu(hdr->front_len);
5529 	u32 data_len = le32_to_cpu(hdr->data_len);
5530 
5531 	m = ceph_msg_new2(type, front_len, 1, GFP_NOIO, false);
5532 	if (!m)
5533 		return NULL;
5534 
5535 	if (data_len) {
5536 		struct page **pages;
5537 
5538 		pages = ceph_alloc_page_vector(calc_pages_for(0, data_len),
5539 					       GFP_NOIO);
5540 		if (IS_ERR(pages)) {
5541 			ceph_msg_put(m);
5542 			return NULL;
5543 		}
5544 
5545 		ceph_msg_data_add_pages(m, pages, data_len, 0, true);
5546 	}
5547 
5548 	return m;
5549 }
5550 
5551 static struct ceph_msg *osd_alloc_msg(struct ceph_connection *con,
5552 				      struct ceph_msg_header *hdr,
5553 				      int *skip)
5554 {
5555 	struct ceph_osd *osd = con->private;
5556 	int type = le16_to_cpu(hdr->type);
5557 
5558 	*skip = 0;
5559 	switch (type) {
5560 	case CEPH_MSG_OSD_MAP:
5561 	case CEPH_MSG_OSD_BACKOFF:
5562 	case CEPH_MSG_WATCH_NOTIFY:
5563 		return alloc_msg_with_page_vector(hdr);
5564 	case CEPH_MSG_OSD_OPREPLY:
5565 		return get_reply(con, hdr, skip);
5566 	default:
5567 		pr_warn("%s osd%d unknown msg type %d, skipping\n", __func__,
5568 			osd->o_osd, type);
5569 		*skip = 1;
5570 		return NULL;
5571 	}
5572 }
5573 
5574 /*
5575  * Wrappers to refcount containing ceph_osd struct
5576  */
5577 static struct ceph_connection *osd_get_con(struct ceph_connection *con)
5578 {
5579 	struct ceph_osd *osd = con->private;
5580 	if (get_osd(osd))
5581 		return con;
5582 	return NULL;
5583 }
5584 
5585 static void osd_put_con(struct ceph_connection *con)
5586 {
5587 	struct ceph_osd *osd = con->private;
5588 	put_osd(osd);
5589 }
5590 
5591 /*
5592  * authentication
5593  */
5594 
5595 /*
5596  * Note: returned pointer is the address of a structure that's
5597  * managed separately.  Caller must *not* attempt to free it.
5598  */
5599 static struct ceph_auth_handshake *
5600 osd_get_authorizer(struct ceph_connection *con, int *proto, int force_new)
5601 {
5602 	struct ceph_osd *o = con->private;
5603 	struct ceph_osd_client *osdc = o->o_osdc;
5604 	struct ceph_auth_client *ac = osdc->client->monc.auth;
5605 	struct ceph_auth_handshake *auth = &o->o_auth;
5606 	int ret;
5607 
5608 	ret = __ceph_auth_get_authorizer(ac, auth, CEPH_ENTITY_TYPE_OSD,
5609 					 force_new, proto, NULL, NULL);
5610 	if (ret)
5611 		return ERR_PTR(ret);
5612 
5613 	return auth;
5614 }
5615 
5616 static int osd_add_authorizer_challenge(struct ceph_connection *con,
5617 				    void *challenge_buf, int challenge_buf_len)
5618 {
5619 	struct ceph_osd *o = con->private;
5620 	struct ceph_osd_client *osdc = o->o_osdc;
5621 	struct ceph_auth_client *ac = osdc->client->monc.auth;
5622 
5623 	return ceph_auth_add_authorizer_challenge(ac, o->o_auth.authorizer,
5624 					    challenge_buf, challenge_buf_len);
5625 }
5626 
5627 static int osd_verify_authorizer_reply(struct ceph_connection *con)
5628 {
5629 	struct ceph_osd *o = con->private;
5630 	struct ceph_osd_client *osdc = o->o_osdc;
5631 	struct ceph_auth_client *ac = osdc->client->monc.auth;
5632 	struct ceph_auth_handshake *auth = &o->o_auth;
5633 
5634 	return ceph_auth_verify_authorizer_reply(ac, auth->authorizer,
5635 		auth->authorizer_reply_buf, auth->authorizer_reply_buf_len,
5636 		NULL, NULL, NULL, NULL);
5637 }
5638 
5639 static int osd_invalidate_authorizer(struct ceph_connection *con)
5640 {
5641 	struct ceph_osd *o = con->private;
5642 	struct ceph_osd_client *osdc = o->o_osdc;
5643 	struct ceph_auth_client *ac = osdc->client->monc.auth;
5644 
5645 	ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
5646 	return ceph_monc_validate_auth(&osdc->client->monc);
5647 }
5648 
5649 static int osd_get_auth_request(struct ceph_connection *con,
5650 				void *buf, int *buf_len,
5651 				void **authorizer, int *authorizer_len)
5652 {
5653 	struct ceph_osd *o = con->private;
5654 	struct ceph_auth_client *ac = o->o_osdc->client->monc.auth;
5655 	struct ceph_auth_handshake *auth = &o->o_auth;
5656 	int ret;
5657 
5658 	ret = ceph_auth_get_authorizer(ac, auth, CEPH_ENTITY_TYPE_OSD,
5659 				       buf, buf_len);
5660 	if (ret)
5661 		return ret;
5662 
5663 	*authorizer = auth->authorizer_buf;
5664 	*authorizer_len = auth->authorizer_buf_len;
5665 	return 0;
5666 }
5667 
5668 static int osd_handle_auth_reply_more(struct ceph_connection *con,
5669 				      void *reply, int reply_len,
5670 				      void *buf, int *buf_len,
5671 				      void **authorizer, int *authorizer_len)
5672 {
5673 	struct ceph_osd *o = con->private;
5674 	struct ceph_auth_client *ac = o->o_osdc->client->monc.auth;
5675 	struct ceph_auth_handshake *auth = &o->o_auth;
5676 	int ret;
5677 
5678 	ret = ceph_auth_handle_svc_reply_more(ac, auth, reply, reply_len,
5679 					      buf, buf_len);
5680 	if (ret)
5681 		return ret;
5682 
5683 	*authorizer = auth->authorizer_buf;
5684 	*authorizer_len = auth->authorizer_buf_len;
5685 	return 0;
5686 }
5687 
5688 static int osd_handle_auth_done(struct ceph_connection *con,
5689 				u64 global_id, void *reply, int reply_len,
5690 				u8 *session_key, int *session_key_len,
5691 				u8 *con_secret, int *con_secret_len)
5692 {
5693 	struct ceph_osd *o = con->private;
5694 	struct ceph_auth_client *ac = o->o_osdc->client->monc.auth;
5695 	struct ceph_auth_handshake *auth = &o->o_auth;
5696 
5697 	return ceph_auth_handle_svc_reply_done(ac, auth, reply, reply_len,
5698 					       session_key, session_key_len,
5699 					       con_secret, con_secret_len);
5700 }
5701 
5702 static int osd_handle_auth_bad_method(struct ceph_connection *con,
5703 				      int used_proto, int result,
5704 				      const int *allowed_protos, int proto_cnt,
5705 				      const int *allowed_modes, int mode_cnt)
5706 {
5707 	struct ceph_osd *o = con->private;
5708 	struct ceph_mon_client *monc = &o->o_osdc->client->monc;
5709 	int ret;
5710 
5711 	if (ceph_auth_handle_bad_authorizer(monc->auth, CEPH_ENTITY_TYPE_OSD,
5712 					    used_proto, result,
5713 					    allowed_protos, proto_cnt,
5714 					    allowed_modes, mode_cnt)) {
5715 		ret = ceph_monc_validate_auth(monc);
5716 		if (ret)
5717 			return ret;
5718 	}
5719 
5720 	return -EACCES;
5721 }
5722 
5723 static void osd_reencode_message(struct ceph_msg *msg)
5724 {
5725 	int type = le16_to_cpu(msg->hdr.type);
5726 
5727 	if (type == CEPH_MSG_OSD_OP)
5728 		encode_request_finish(msg);
5729 }
5730 
5731 static int osd_sign_message(struct ceph_msg *msg)
5732 {
5733 	struct ceph_osd *o = msg->con->private;
5734 	struct ceph_auth_handshake *auth = &o->o_auth;
5735 
5736 	return ceph_auth_sign_message(auth, msg);
5737 }
5738 
5739 static int osd_check_message_signature(struct ceph_msg *msg)
5740 {
5741 	struct ceph_osd *o = msg->con->private;
5742 	struct ceph_auth_handshake *auth = &o->o_auth;
5743 
5744 	return ceph_auth_check_message_signature(auth, msg);
5745 }
5746 
5747 static void advance_cursor(struct ceph_msg_data_cursor *cursor, size_t len,
5748 			   bool zero)
5749 {
5750 	while (len) {
5751 		struct page *page;
5752 		size_t poff, plen;
5753 
5754 		page = ceph_msg_data_next(cursor, &poff, &plen);
5755 		if (plen > len)
5756 			plen = len;
5757 		if (zero)
5758 			zero_user_segment(page, poff, poff + plen);
5759 		len -= plen;
5760 		ceph_msg_data_advance(cursor, plen);
5761 	}
5762 }
5763 
5764 static int prep_next_sparse_read(struct ceph_connection *con,
5765 				 struct ceph_msg_data_cursor *cursor)
5766 {
5767 	struct ceph_osd *o = con->private;
5768 	struct ceph_sparse_read *sr = &o->o_sparse_read;
5769 	struct ceph_osd_request *req;
5770 	struct ceph_osd_req_op *op;
5771 
5772 	spin_lock(&o->o_requests_lock);
5773 	req = lookup_request(&o->o_requests, le64_to_cpu(con->in_msg->hdr.tid));
5774 	if (!req) {
5775 		spin_unlock(&o->o_requests_lock);
5776 		return -EBADR;
5777 	}
5778 
5779 	if (o->o_sparse_op_idx < 0) {
5780 		dout("%s: [%d] starting new sparse read req\n",
5781 		     __func__, o->o_osd);
5782 	} else {
5783 		u64 end;
5784 
5785 		op = &req->r_ops[o->o_sparse_op_idx];
5786 
5787 		WARN_ON_ONCE(op->extent.sparse_ext);
5788 
5789 		/* hand back buffer we took earlier */
5790 		op->extent.sparse_ext = sr->sr_extent;
5791 		sr->sr_extent = NULL;
5792 		op->extent.sparse_ext_cnt = sr->sr_count;
5793 		sr->sr_ext_len = 0;
5794 		dout("%s: [%d] completed extent array len %d cursor->resid %zd\n",
5795 		     __func__, o->o_osd, op->extent.sparse_ext_cnt, cursor->resid);
5796 		/* Advance to end of data for this operation */
5797 		end = ceph_sparse_ext_map_end(op);
5798 		if (end < sr->sr_req_len)
5799 			advance_cursor(cursor, sr->sr_req_len - end, false);
5800 	}
5801 
5802 	ceph_init_sparse_read(sr);
5803 
5804 	/* find next op in this request (if any) */
5805 	while (++o->o_sparse_op_idx < req->r_num_ops) {
5806 		op = &req->r_ops[o->o_sparse_op_idx];
5807 		if (op->op == CEPH_OSD_OP_SPARSE_READ)
5808 			goto found;
5809 	}
5810 
5811 	/* reset for next sparse read request */
5812 	spin_unlock(&o->o_requests_lock);
5813 	o->o_sparse_op_idx = -1;
5814 	return 0;
5815 found:
5816 	sr->sr_req_off = op->extent.offset;
5817 	sr->sr_req_len = op->extent.length;
5818 	sr->sr_pos = sr->sr_req_off;
5819 	dout("%s: [%d] new sparse read op at idx %d 0x%llx~0x%llx\n", __func__,
5820 	     o->o_osd, o->o_sparse_op_idx, sr->sr_req_off, sr->sr_req_len);
5821 
5822 	/* hand off request's sparse extent map buffer */
5823 	sr->sr_ext_len = op->extent.sparse_ext_cnt;
5824 	op->extent.sparse_ext_cnt = 0;
5825 	sr->sr_extent = op->extent.sparse_ext;
5826 	op->extent.sparse_ext = NULL;
5827 
5828 	spin_unlock(&o->o_requests_lock);
5829 	return 1;
5830 }
5831 
5832 #ifdef __BIG_ENDIAN
5833 static inline void convert_extent_map(struct ceph_sparse_read *sr)
5834 {
5835 	int i;
5836 
5837 	for (i = 0; i < sr->sr_count; i++) {
5838 		struct ceph_sparse_extent *ext = &sr->sr_extent[i];
5839 
5840 		ext->off = le64_to_cpu((__force __le64)ext->off);
5841 		ext->len = le64_to_cpu((__force __le64)ext->len);
5842 	}
5843 }
5844 #else
5845 static inline void convert_extent_map(struct ceph_sparse_read *sr)
5846 {
5847 }
5848 #endif
5849 
5850 static int osd_sparse_read(struct ceph_connection *con,
5851 			   struct ceph_msg_data_cursor *cursor,
5852 			   char **pbuf)
5853 {
5854 	struct ceph_osd *o = con->private;
5855 	struct ceph_sparse_read *sr = &o->o_sparse_read;
5856 	u32 count = sr->sr_count;
5857 	u64 eoff, elen, len = 0;
5858 	int i, ret;
5859 
5860 	switch (sr->sr_state) {
5861 	case CEPH_SPARSE_READ_HDR:
5862 next_op:
5863 		ret = prep_next_sparse_read(con, cursor);
5864 		if (ret <= 0)
5865 			return ret;
5866 
5867 		/* number of extents */
5868 		ret = sizeof(sr->sr_count);
5869 		*pbuf = (char *)&sr->sr_count;
5870 		sr->sr_state = CEPH_SPARSE_READ_EXTENTS;
5871 		break;
5872 	case CEPH_SPARSE_READ_EXTENTS:
5873 		/* Convert sr_count to host-endian */
5874 		count = le32_to_cpu((__force __le32)sr->sr_count);
5875 		sr->sr_count = count;
5876 		dout("[%d] got %u extents\n", o->o_osd, count);
5877 
5878 		if (count > 0) {
5879 			if (!sr->sr_extent || count > sr->sr_ext_len) {
5880 				/* no extent array provided, or too short */
5881 				kfree(sr->sr_extent);
5882 				sr->sr_extent = kmalloc_array(count,
5883 							      sizeof(*sr->sr_extent),
5884 							      GFP_NOIO);
5885 				if (!sr->sr_extent) {
5886 					pr_err("%s: failed to allocate %u extents\n",
5887 					       __func__, count);
5888 					return -ENOMEM;
5889 				}
5890 				sr->sr_ext_len = count;
5891 			}
5892 			ret = count * sizeof(*sr->sr_extent);
5893 			*pbuf = (char *)sr->sr_extent;
5894 			sr->sr_state = CEPH_SPARSE_READ_DATA_LEN;
5895 			break;
5896 		}
5897 		/* No extents? Read data len */
5898 		fallthrough;
5899 	case CEPH_SPARSE_READ_DATA_LEN:
5900 		convert_extent_map(sr);
5901 		ret = sizeof(sr->sr_datalen);
5902 		*pbuf = (char *)&sr->sr_datalen;
5903 		sr->sr_state = CEPH_SPARSE_READ_DATA_PRE;
5904 		break;
5905 	case CEPH_SPARSE_READ_DATA_PRE:
5906 		/* Convert sr_datalen to host-endian */
5907 		sr->sr_datalen = le32_to_cpu((__force __le32)sr->sr_datalen);
5908 		for (i = 0; i < count; i++)
5909 			len += sr->sr_extent[i].len;
5910 		if (sr->sr_datalen != len) {
5911 			pr_warn_ratelimited("data len %u != extent len %llu\n",
5912 					    sr->sr_datalen, len);
5913 			return -EREMOTEIO;
5914 		}
5915 		sr->sr_state = CEPH_SPARSE_READ_DATA;
5916 		fallthrough;
5917 	case CEPH_SPARSE_READ_DATA:
5918 		if (sr->sr_index >= count) {
5919 			sr->sr_state = CEPH_SPARSE_READ_HDR;
5920 			goto next_op;
5921 		}
5922 
5923 		eoff = sr->sr_extent[sr->sr_index].off;
5924 		elen = sr->sr_extent[sr->sr_index].len;
5925 
5926 		dout("[%d] ext %d off 0x%llx len 0x%llx\n",
5927 		     o->o_osd, sr->sr_index, eoff, elen);
5928 
5929 		if (elen > INT_MAX) {
5930 			dout("Sparse read extent length too long (0x%llx)\n",
5931 			     elen);
5932 			return -EREMOTEIO;
5933 		}
5934 
5935 		/* zero out anything from sr_pos to start of extent */
5936 		if (sr->sr_pos < eoff)
5937 			advance_cursor(cursor, eoff - sr->sr_pos, true);
5938 
5939 		/* Set position to end of extent */
5940 		sr->sr_pos = eoff + elen;
5941 
5942 		/* send back the new length and nullify the ptr */
5943 		cursor->sr_resid = elen;
5944 		ret = elen;
5945 		*pbuf = NULL;
5946 
5947 		/* Bump the array index */
5948 		++sr->sr_index;
5949 		break;
5950 	}
5951 	return ret;
5952 }
5953 
5954 static const struct ceph_connection_operations osd_con_ops = {
5955 	.get = osd_get_con,
5956 	.put = osd_put_con,
5957 	.sparse_read = osd_sparse_read,
5958 	.alloc_msg = osd_alloc_msg,
5959 	.dispatch = osd_dispatch,
5960 	.fault = osd_fault,
5961 	.reencode_message = osd_reencode_message,
5962 	.get_authorizer = osd_get_authorizer,
5963 	.add_authorizer_challenge = osd_add_authorizer_challenge,
5964 	.verify_authorizer_reply = osd_verify_authorizer_reply,
5965 	.invalidate_authorizer = osd_invalidate_authorizer,
5966 	.sign_message = osd_sign_message,
5967 	.check_message_signature = osd_check_message_signature,
5968 	.get_auth_request = osd_get_auth_request,
5969 	.handle_auth_reply_more = osd_handle_auth_reply_more,
5970 	.handle_auth_done = osd_handle_auth_done,
5971 	.handle_auth_bad_method = osd_handle_auth_bad_method,
5972 };
5973