xref: /openbmc/linux/net/ceph/osd_client.c (revision d2168146)
1 
2 #include <linux/ceph/ceph_debug.h>
3 
4 #include <linux/module.h>
5 #include <linux/err.h>
6 #include <linux/highmem.h>
7 #include <linux/mm.h>
8 #include <linux/pagemap.h>
9 #include <linux/slab.h>
10 #include <linux/uaccess.h>
11 #ifdef CONFIG_BLOCK
12 #include <linux/bio.h>
13 #endif
14 
15 #include <linux/ceph/libceph.h>
16 #include <linux/ceph/osd_client.h>
17 #include <linux/ceph/messenger.h>
18 #include <linux/ceph/decode.h>
19 #include <linux/ceph/auth.h>
20 #include <linux/ceph/pagelist.h>
21 
22 #define OSD_OP_FRONT_LEN	4096
23 #define OSD_OPREPLY_FRONT_LEN	512
24 
25 static struct kmem_cache	*ceph_osd_request_cache;
26 
27 static const struct ceph_connection_operations osd_con_ops;
28 
29 static void __send_queued(struct ceph_osd_client *osdc);
30 static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd);
31 static void __register_request(struct ceph_osd_client *osdc,
32 			       struct ceph_osd_request *req);
33 static void __unregister_linger_request(struct ceph_osd_client *osdc,
34 					struct ceph_osd_request *req);
35 static void __send_request(struct ceph_osd_client *osdc,
36 			   struct ceph_osd_request *req);
37 
38 /*
39  * Implement client access to distributed object storage cluster.
40  *
41  * All data objects are stored within a cluster/cloud of OSDs, or
42  * "object storage devices."  (Note that Ceph OSDs have _nothing_ to
43  * do with the T10 OSD extensions to SCSI.)  Ceph OSDs are simply
44  * remote daemons serving up and coordinating consistent and safe
45  * access to storage.
46  *
47  * Cluster membership and the mapping of data objects onto storage devices
48  * are described by the osd map.
49  *
50  * We keep track of pending OSD requests (read, write), resubmit
51  * requests to different OSDs when the cluster topology/data layout
52  * change, or retry the affected requests when the communications
53  * channel with an OSD is reset.
54  */
55 
56 /*
57  * calculate the mapping of a file extent onto an object, and fill out the
58  * request accordingly.  shorten extent as necessary if it crosses an
59  * object boundary.
60  *
61  * fill osd op in request message.
62  */
63 static int calc_layout(struct ceph_file_layout *layout, u64 off, u64 *plen,
64 			u64 *objnum, u64 *objoff, u64 *objlen)
65 {
66 	u64 orig_len = *plen;
67 	int r;
68 
69 	/* object extent? */
70 	r = ceph_calc_file_object_mapping(layout, off, orig_len, objnum,
71 					  objoff, objlen);
72 	if (r < 0)
73 		return r;
74 	if (*objlen < orig_len) {
75 		*plen = *objlen;
76 		dout(" skipping last %llu, final file extent %llu~%llu\n",
77 		     orig_len - *plen, off, *plen);
78 	}
79 
80 	dout("calc_layout objnum=%llx %llu~%llu\n", *objnum, *objoff, *objlen);
81 
82 	return 0;
83 }
84 
85 static void ceph_osd_data_init(struct ceph_osd_data *osd_data)
86 {
87 	memset(osd_data, 0, sizeof (*osd_data));
88 	osd_data->type = CEPH_OSD_DATA_TYPE_NONE;
89 }
90 
91 static void ceph_osd_data_pages_init(struct ceph_osd_data *osd_data,
92 			struct page **pages, u64 length, u32 alignment,
93 			bool pages_from_pool, bool own_pages)
94 {
95 	osd_data->type = CEPH_OSD_DATA_TYPE_PAGES;
96 	osd_data->pages = pages;
97 	osd_data->length = length;
98 	osd_data->alignment = alignment;
99 	osd_data->pages_from_pool = pages_from_pool;
100 	osd_data->own_pages = own_pages;
101 }
102 
103 static void ceph_osd_data_pagelist_init(struct ceph_osd_data *osd_data,
104 			struct ceph_pagelist *pagelist)
105 {
106 	osd_data->type = CEPH_OSD_DATA_TYPE_PAGELIST;
107 	osd_data->pagelist = pagelist;
108 }
109 
110 #ifdef CONFIG_BLOCK
111 static void ceph_osd_data_bio_init(struct ceph_osd_data *osd_data,
112 			struct bio *bio, size_t bio_length)
113 {
114 	osd_data->type = CEPH_OSD_DATA_TYPE_BIO;
115 	osd_data->bio = bio;
116 	osd_data->bio_length = bio_length;
117 }
118 #endif /* CONFIG_BLOCK */
119 
120 #define osd_req_op_data(oreq, whch, typ, fld)	\
121 	({						\
122 		BUG_ON(whch >= (oreq)->r_num_ops);	\
123 		&(oreq)->r_ops[whch].typ.fld;		\
124 	})
125 
126 static struct ceph_osd_data *
127 osd_req_op_raw_data_in(struct ceph_osd_request *osd_req, unsigned int which)
128 {
129 	BUG_ON(which >= osd_req->r_num_ops);
130 
131 	return &osd_req->r_ops[which].raw_data_in;
132 }
133 
134 struct ceph_osd_data *
135 osd_req_op_extent_osd_data(struct ceph_osd_request *osd_req,
136 			unsigned int which)
137 {
138 	return osd_req_op_data(osd_req, which, extent, osd_data);
139 }
140 EXPORT_SYMBOL(osd_req_op_extent_osd_data);
141 
142 struct ceph_osd_data *
143 osd_req_op_cls_response_data(struct ceph_osd_request *osd_req,
144 			unsigned int which)
145 {
146 	return osd_req_op_data(osd_req, which, cls, response_data);
147 }
148 EXPORT_SYMBOL(osd_req_op_cls_response_data);	/* ??? */
149 
150 void osd_req_op_raw_data_in_pages(struct ceph_osd_request *osd_req,
151 			unsigned int which, struct page **pages,
152 			u64 length, u32 alignment,
153 			bool pages_from_pool, bool own_pages)
154 {
155 	struct ceph_osd_data *osd_data;
156 
157 	osd_data = osd_req_op_raw_data_in(osd_req, which);
158 	ceph_osd_data_pages_init(osd_data, pages, length, alignment,
159 				pages_from_pool, own_pages);
160 }
161 EXPORT_SYMBOL(osd_req_op_raw_data_in_pages);
162 
163 void osd_req_op_extent_osd_data_pages(struct ceph_osd_request *osd_req,
164 			unsigned int which, struct page **pages,
165 			u64 length, u32 alignment,
166 			bool pages_from_pool, bool own_pages)
167 {
168 	struct ceph_osd_data *osd_data;
169 
170 	osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
171 	ceph_osd_data_pages_init(osd_data, pages, length, alignment,
172 				pages_from_pool, own_pages);
173 }
174 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pages);
175 
176 void osd_req_op_extent_osd_data_pagelist(struct ceph_osd_request *osd_req,
177 			unsigned int which, struct ceph_pagelist *pagelist)
178 {
179 	struct ceph_osd_data *osd_data;
180 
181 	osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
182 	ceph_osd_data_pagelist_init(osd_data, pagelist);
183 }
184 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pagelist);
185 
186 #ifdef CONFIG_BLOCK
187 void osd_req_op_extent_osd_data_bio(struct ceph_osd_request *osd_req,
188 			unsigned int which, struct bio *bio, size_t bio_length)
189 {
190 	struct ceph_osd_data *osd_data;
191 
192 	osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
193 	ceph_osd_data_bio_init(osd_data, bio, bio_length);
194 }
195 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bio);
196 #endif /* CONFIG_BLOCK */
197 
198 static void osd_req_op_cls_request_info_pagelist(
199 			struct ceph_osd_request *osd_req,
200 			unsigned int which, struct ceph_pagelist *pagelist)
201 {
202 	struct ceph_osd_data *osd_data;
203 
204 	osd_data = osd_req_op_data(osd_req, which, cls, request_info);
205 	ceph_osd_data_pagelist_init(osd_data, pagelist);
206 }
207 
208 void osd_req_op_cls_request_data_pagelist(
209 			struct ceph_osd_request *osd_req,
210 			unsigned int which, struct ceph_pagelist *pagelist)
211 {
212 	struct ceph_osd_data *osd_data;
213 
214 	osd_data = osd_req_op_data(osd_req, which, cls, request_data);
215 	ceph_osd_data_pagelist_init(osd_data, pagelist);
216 }
217 EXPORT_SYMBOL(osd_req_op_cls_request_data_pagelist);
218 
219 void osd_req_op_cls_request_data_pages(struct ceph_osd_request *osd_req,
220 			unsigned int which, struct page **pages, u64 length,
221 			u32 alignment, bool pages_from_pool, bool own_pages)
222 {
223 	struct ceph_osd_data *osd_data;
224 
225 	osd_data = osd_req_op_data(osd_req, which, cls, request_data);
226 	ceph_osd_data_pages_init(osd_data, pages, length, alignment,
227 				pages_from_pool, own_pages);
228 }
229 EXPORT_SYMBOL(osd_req_op_cls_request_data_pages);
230 
231 void osd_req_op_cls_response_data_pages(struct ceph_osd_request *osd_req,
232 			unsigned int which, struct page **pages, u64 length,
233 			u32 alignment, bool pages_from_pool, bool own_pages)
234 {
235 	struct ceph_osd_data *osd_data;
236 
237 	osd_data = osd_req_op_data(osd_req, which, cls, response_data);
238 	ceph_osd_data_pages_init(osd_data, pages, length, alignment,
239 				pages_from_pool, own_pages);
240 }
241 EXPORT_SYMBOL(osd_req_op_cls_response_data_pages);
242 
243 static u64 ceph_osd_data_length(struct ceph_osd_data *osd_data)
244 {
245 	switch (osd_data->type) {
246 	case CEPH_OSD_DATA_TYPE_NONE:
247 		return 0;
248 	case CEPH_OSD_DATA_TYPE_PAGES:
249 		return osd_data->length;
250 	case CEPH_OSD_DATA_TYPE_PAGELIST:
251 		return (u64)osd_data->pagelist->length;
252 #ifdef CONFIG_BLOCK
253 	case CEPH_OSD_DATA_TYPE_BIO:
254 		return (u64)osd_data->bio_length;
255 #endif /* CONFIG_BLOCK */
256 	default:
257 		WARN(true, "unrecognized data type %d\n", (int)osd_data->type);
258 		return 0;
259 	}
260 }
261 
262 static void ceph_osd_data_release(struct ceph_osd_data *osd_data)
263 {
264 	if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES && osd_data->own_pages) {
265 		int num_pages;
266 
267 		num_pages = calc_pages_for((u64)osd_data->alignment,
268 						(u64)osd_data->length);
269 		ceph_release_page_vector(osd_data->pages, num_pages);
270 	}
271 	ceph_osd_data_init(osd_data);
272 }
273 
274 static void osd_req_op_data_release(struct ceph_osd_request *osd_req,
275 			unsigned int which)
276 {
277 	struct ceph_osd_req_op *op;
278 
279 	BUG_ON(which >= osd_req->r_num_ops);
280 	op = &osd_req->r_ops[which];
281 
282 	switch (op->op) {
283 	case CEPH_OSD_OP_READ:
284 	case CEPH_OSD_OP_WRITE:
285 		ceph_osd_data_release(&op->extent.osd_data);
286 		break;
287 	case CEPH_OSD_OP_CALL:
288 		ceph_osd_data_release(&op->cls.request_info);
289 		ceph_osd_data_release(&op->cls.request_data);
290 		ceph_osd_data_release(&op->cls.response_data);
291 		break;
292 	default:
293 		break;
294 	}
295 }
296 
297 /*
298  * requests
299  */
300 void ceph_osdc_release_request(struct kref *kref)
301 {
302 	struct ceph_osd_request *req;
303 	unsigned int which;
304 
305 	req = container_of(kref, struct ceph_osd_request, r_kref);
306 	if (req->r_request)
307 		ceph_msg_put(req->r_request);
308 	if (req->r_reply) {
309 		ceph_msg_revoke_incoming(req->r_reply);
310 		ceph_msg_put(req->r_reply);
311 	}
312 
313 	for (which = 0; which < req->r_num_ops; which++)
314 		osd_req_op_data_release(req, which);
315 
316 	ceph_put_snap_context(req->r_snapc);
317 	if (req->r_mempool)
318 		mempool_free(req, req->r_osdc->req_mempool);
319 	else
320 		kmem_cache_free(ceph_osd_request_cache, req);
321 
322 }
323 EXPORT_SYMBOL(ceph_osdc_release_request);
324 
325 struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
326 					       struct ceph_snap_context *snapc,
327 					       unsigned int num_ops,
328 					       bool use_mempool,
329 					       gfp_t gfp_flags)
330 {
331 	struct ceph_osd_request *req;
332 	struct ceph_msg *msg;
333 	size_t msg_size;
334 
335 	BUILD_BUG_ON(CEPH_OSD_MAX_OP > U16_MAX);
336 	BUG_ON(num_ops > CEPH_OSD_MAX_OP);
337 
338 	msg_size = 4 + 4 + 8 + 8 + 4+8;
339 	msg_size += 2 + 4 + 8 + 4 + 4; /* oloc */
340 	msg_size += 1 + 8 + 4 + 4;     /* pg_t */
341 	msg_size += 4 + CEPH_MAX_OID_NAME_LEN; /* oid */
342 	msg_size += 2 + num_ops*sizeof(struct ceph_osd_op);
343 	msg_size += 8;  /* snapid */
344 	msg_size += 8;  /* snap_seq */
345 	msg_size += 8 * (snapc ? snapc->num_snaps : 0);  /* snaps */
346 	msg_size += 4;
347 
348 	if (use_mempool) {
349 		req = mempool_alloc(osdc->req_mempool, gfp_flags);
350 		memset(req, 0, sizeof(*req));
351 	} else {
352 		req = kmem_cache_zalloc(ceph_osd_request_cache, gfp_flags);
353 	}
354 	if (req == NULL)
355 		return NULL;
356 
357 	req->r_osdc = osdc;
358 	req->r_mempool = use_mempool;
359 	req->r_num_ops = num_ops;
360 
361 	kref_init(&req->r_kref);
362 	init_completion(&req->r_completion);
363 	init_completion(&req->r_safe_completion);
364 	RB_CLEAR_NODE(&req->r_node);
365 	INIT_LIST_HEAD(&req->r_unsafe_item);
366 	INIT_LIST_HEAD(&req->r_linger_item);
367 	INIT_LIST_HEAD(&req->r_linger_osd);
368 	INIT_LIST_HEAD(&req->r_req_lru_item);
369 	INIT_LIST_HEAD(&req->r_osd_item);
370 
371 	req->r_base_oloc.pool = -1;
372 	req->r_target_oloc.pool = -1;
373 
374 	/* create reply message */
375 	if (use_mempool)
376 		msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0);
377 	else
378 		msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY,
379 				   OSD_OPREPLY_FRONT_LEN, gfp_flags, true);
380 	if (!msg) {
381 		ceph_osdc_put_request(req);
382 		return NULL;
383 	}
384 	req->r_reply = msg;
385 
386 	/* create request message; allow space for oid */
387 	if (use_mempool)
388 		msg = ceph_msgpool_get(&osdc->msgpool_op, 0);
389 	else
390 		msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp_flags, true);
391 	if (!msg) {
392 		ceph_osdc_put_request(req);
393 		return NULL;
394 	}
395 
396 	memset(msg->front.iov_base, 0, msg->front.iov_len);
397 
398 	req->r_request = msg;
399 
400 	return req;
401 }
402 EXPORT_SYMBOL(ceph_osdc_alloc_request);
403 
404 static bool osd_req_opcode_valid(u16 opcode)
405 {
406 	switch (opcode) {
407 	case CEPH_OSD_OP_READ:
408 	case CEPH_OSD_OP_STAT:
409 	case CEPH_OSD_OP_MAPEXT:
410 	case CEPH_OSD_OP_MASKTRUNC:
411 	case CEPH_OSD_OP_SPARSE_READ:
412 	case CEPH_OSD_OP_NOTIFY:
413 	case CEPH_OSD_OP_NOTIFY_ACK:
414 	case CEPH_OSD_OP_ASSERT_VER:
415 	case CEPH_OSD_OP_WRITE:
416 	case CEPH_OSD_OP_WRITEFULL:
417 	case CEPH_OSD_OP_TRUNCATE:
418 	case CEPH_OSD_OP_ZERO:
419 	case CEPH_OSD_OP_DELETE:
420 	case CEPH_OSD_OP_APPEND:
421 	case CEPH_OSD_OP_STARTSYNC:
422 	case CEPH_OSD_OP_SETTRUNC:
423 	case CEPH_OSD_OP_TRIMTRUNC:
424 	case CEPH_OSD_OP_TMAPUP:
425 	case CEPH_OSD_OP_TMAPPUT:
426 	case CEPH_OSD_OP_TMAPGET:
427 	case CEPH_OSD_OP_CREATE:
428 	case CEPH_OSD_OP_ROLLBACK:
429 	case CEPH_OSD_OP_WATCH:
430 	case CEPH_OSD_OP_OMAPGETKEYS:
431 	case CEPH_OSD_OP_OMAPGETVALS:
432 	case CEPH_OSD_OP_OMAPGETHEADER:
433 	case CEPH_OSD_OP_OMAPGETVALSBYKEYS:
434 	case CEPH_OSD_OP_OMAPSETVALS:
435 	case CEPH_OSD_OP_OMAPSETHEADER:
436 	case CEPH_OSD_OP_OMAPCLEAR:
437 	case CEPH_OSD_OP_OMAPRMKEYS:
438 	case CEPH_OSD_OP_OMAP_CMP:
439 	case CEPH_OSD_OP_SETALLOCHINT:
440 	case CEPH_OSD_OP_CLONERANGE:
441 	case CEPH_OSD_OP_ASSERT_SRC_VERSION:
442 	case CEPH_OSD_OP_SRC_CMPXATTR:
443 	case CEPH_OSD_OP_GETXATTR:
444 	case CEPH_OSD_OP_GETXATTRS:
445 	case CEPH_OSD_OP_CMPXATTR:
446 	case CEPH_OSD_OP_SETXATTR:
447 	case CEPH_OSD_OP_SETXATTRS:
448 	case CEPH_OSD_OP_RESETXATTRS:
449 	case CEPH_OSD_OP_RMXATTR:
450 	case CEPH_OSD_OP_PULL:
451 	case CEPH_OSD_OP_PUSH:
452 	case CEPH_OSD_OP_BALANCEREADS:
453 	case CEPH_OSD_OP_UNBALANCEREADS:
454 	case CEPH_OSD_OP_SCRUB:
455 	case CEPH_OSD_OP_SCRUB_RESERVE:
456 	case CEPH_OSD_OP_SCRUB_UNRESERVE:
457 	case CEPH_OSD_OP_SCRUB_STOP:
458 	case CEPH_OSD_OP_SCRUB_MAP:
459 	case CEPH_OSD_OP_WRLOCK:
460 	case CEPH_OSD_OP_WRUNLOCK:
461 	case CEPH_OSD_OP_RDLOCK:
462 	case CEPH_OSD_OP_RDUNLOCK:
463 	case CEPH_OSD_OP_UPLOCK:
464 	case CEPH_OSD_OP_DNLOCK:
465 	case CEPH_OSD_OP_CALL:
466 	case CEPH_OSD_OP_PGLS:
467 	case CEPH_OSD_OP_PGLS_FILTER:
468 		return true;
469 	default:
470 		return false;
471 	}
472 }
473 
474 /*
475  * This is an osd op init function for opcodes that have no data or
476  * other information associated with them.  It also serves as a
477  * common init routine for all the other init functions, below.
478  */
479 static struct ceph_osd_req_op *
480 _osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which,
481 				u16 opcode)
482 {
483 	struct ceph_osd_req_op *op;
484 
485 	BUG_ON(which >= osd_req->r_num_ops);
486 	BUG_ON(!osd_req_opcode_valid(opcode));
487 
488 	op = &osd_req->r_ops[which];
489 	memset(op, 0, sizeof (*op));
490 	op->op = opcode;
491 
492 	return op;
493 }
494 
495 void osd_req_op_init(struct ceph_osd_request *osd_req,
496 				unsigned int which, u16 opcode)
497 {
498 	(void)_osd_req_op_init(osd_req, which, opcode);
499 }
500 EXPORT_SYMBOL(osd_req_op_init);
501 
502 void osd_req_op_extent_init(struct ceph_osd_request *osd_req,
503 				unsigned int which, u16 opcode,
504 				u64 offset, u64 length,
505 				u64 truncate_size, u32 truncate_seq)
506 {
507 	struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
508 	size_t payload_len = 0;
509 
510 	BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
511 	       opcode != CEPH_OSD_OP_DELETE && opcode != CEPH_OSD_OP_ZERO &&
512 	       opcode != CEPH_OSD_OP_TRUNCATE);
513 
514 	op->extent.offset = offset;
515 	op->extent.length = length;
516 	op->extent.truncate_size = truncate_size;
517 	op->extent.truncate_seq = truncate_seq;
518 	if (opcode == CEPH_OSD_OP_WRITE)
519 		payload_len += length;
520 
521 	op->payload_len = payload_len;
522 }
523 EXPORT_SYMBOL(osd_req_op_extent_init);
524 
525 void osd_req_op_extent_update(struct ceph_osd_request *osd_req,
526 				unsigned int which, u64 length)
527 {
528 	struct ceph_osd_req_op *op;
529 	u64 previous;
530 
531 	BUG_ON(which >= osd_req->r_num_ops);
532 	op = &osd_req->r_ops[which];
533 	previous = op->extent.length;
534 
535 	if (length == previous)
536 		return;		/* Nothing to do */
537 	BUG_ON(length > previous);
538 
539 	op->extent.length = length;
540 	op->payload_len -= previous - length;
541 }
542 EXPORT_SYMBOL(osd_req_op_extent_update);
543 
544 void osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
545 			u16 opcode, const char *class, const char *method)
546 {
547 	struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
548 	struct ceph_pagelist *pagelist;
549 	size_t payload_len = 0;
550 	size_t size;
551 
552 	BUG_ON(opcode != CEPH_OSD_OP_CALL);
553 
554 	pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
555 	BUG_ON(!pagelist);
556 	ceph_pagelist_init(pagelist);
557 
558 	op->cls.class_name = class;
559 	size = strlen(class);
560 	BUG_ON(size > (size_t) U8_MAX);
561 	op->cls.class_len = size;
562 	ceph_pagelist_append(pagelist, class, size);
563 	payload_len += size;
564 
565 	op->cls.method_name = method;
566 	size = strlen(method);
567 	BUG_ON(size > (size_t) U8_MAX);
568 	op->cls.method_len = size;
569 	ceph_pagelist_append(pagelist, method, size);
570 	payload_len += size;
571 
572 	osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist);
573 
574 	op->cls.argc = 0;	/* currently unused */
575 
576 	op->payload_len = payload_len;
577 }
578 EXPORT_SYMBOL(osd_req_op_cls_init);
579 
580 void osd_req_op_watch_init(struct ceph_osd_request *osd_req,
581 				unsigned int which, u16 opcode,
582 				u64 cookie, u64 version, int flag)
583 {
584 	struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
585 
586 	BUG_ON(opcode != CEPH_OSD_OP_NOTIFY_ACK && opcode != CEPH_OSD_OP_WATCH);
587 
588 	op->watch.cookie = cookie;
589 	op->watch.ver = version;
590 	if (opcode == CEPH_OSD_OP_WATCH && flag)
591 		op->watch.flag = (u8)1;
592 }
593 EXPORT_SYMBOL(osd_req_op_watch_init);
594 
595 void osd_req_op_alloc_hint_init(struct ceph_osd_request *osd_req,
596 				unsigned int which,
597 				u64 expected_object_size,
598 				u64 expected_write_size)
599 {
600 	struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
601 						      CEPH_OSD_OP_SETALLOCHINT);
602 
603 	op->alloc_hint.expected_object_size = expected_object_size;
604 	op->alloc_hint.expected_write_size = expected_write_size;
605 
606 	/*
607 	 * CEPH_OSD_OP_SETALLOCHINT op is advisory and therefore deemed
608 	 * not worth a feature bit.  Set FAILOK per-op flag to make
609 	 * sure older osds don't trip over an unsupported opcode.
610 	 */
611 	op->flags |= CEPH_OSD_OP_FLAG_FAILOK;
612 }
613 EXPORT_SYMBOL(osd_req_op_alloc_hint_init);
614 
615 static void ceph_osdc_msg_data_add(struct ceph_msg *msg,
616 				struct ceph_osd_data *osd_data)
617 {
618 	u64 length = ceph_osd_data_length(osd_data);
619 
620 	if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
621 		BUG_ON(length > (u64) SIZE_MAX);
622 		if (length)
623 			ceph_msg_data_add_pages(msg, osd_data->pages,
624 					length, osd_data->alignment);
625 	} else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
626 		BUG_ON(!length);
627 		ceph_msg_data_add_pagelist(msg, osd_data->pagelist);
628 #ifdef CONFIG_BLOCK
629 	} else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
630 		ceph_msg_data_add_bio(msg, osd_data->bio, length);
631 #endif
632 	} else {
633 		BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
634 	}
635 }
636 
637 static u64 osd_req_encode_op(struct ceph_osd_request *req,
638 			      struct ceph_osd_op *dst, unsigned int which)
639 {
640 	struct ceph_osd_req_op *src;
641 	struct ceph_osd_data *osd_data;
642 	u64 request_data_len = 0;
643 	u64 data_length;
644 
645 	BUG_ON(which >= req->r_num_ops);
646 	src = &req->r_ops[which];
647 	if (WARN_ON(!osd_req_opcode_valid(src->op))) {
648 		pr_err("unrecognized osd opcode %d\n", src->op);
649 
650 		return 0;
651 	}
652 
653 	switch (src->op) {
654 	case CEPH_OSD_OP_STAT:
655 		osd_data = &src->raw_data_in;
656 		ceph_osdc_msg_data_add(req->r_reply, osd_data);
657 		break;
658 	case CEPH_OSD_OP_READ:
659 	case CEPH_OSD_OP_WRITE:
660 	case CEPH_OSD_OP_ZERO:
661 	case CEPH_OSD_OP_DELETE:
662 	case CEPH_OSD_OP_TRUNCATE:
663 		if (src->op == CEPH_OSD_OP_WRITE)
664 			request_data_len = src->extent.length;
665 		dst->extent.offset = cpu_to_le64(src->extent.offset);
666 		dst->extent.length = cpu_to_le64(src->extent.length);
667 		dst->extent.truncate_size =
668 			cpu_to_le64(src->extent.truncate_size);
669 		dst->extent.truncate_seq =
670 			cpu_to_le32(src->extent.truncate_seq);
671 		osd_data = &src->extent.osd_data;
672 		if (src->op == CEPH_OSD_OP_WRITE)
673 			ceph_osdc_msg_data_add(req->r_request, osd_data);
674 		else
675 			ceph_osdc_msg_data_add(req->r_reply, osd_data);
676 		break;
677 	case CEPH_OSD_OP_CALL:
678 		dst->cls.class_len = src->cls.class_len;
679 		dst->cls.method_len = src->cls.method_len;
680 		osd_data = &src->cls.request_info;
681 		ceph_osdc_msg_data_add(req->r_request, osd_data);
682 		BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGELIST);
683 		request_data_len = osd_data->pagelist->length;
684 
685 		osd_data = &src->cls.request_data;
686 		data_length = ceph_osd_data_length(osd_data);
687 		if (data_length) {
688 			BUG_ON(osd_data->type == CEPH_OSD_DATA_TYPE_NONE);
689 			dst->cls.indata_len = cpu_to_le32(data_length);
690 			ceph_osdc_msg_data_add(req->r_request, osd_data);
691 			src->payload_len += data_length;
692 			request_data_len += data_length;
693 		}
694 		osd_data = &src->cls.response_data;
695 		ceph_osdc_msg_data_add(req->r_reply, osd_data);
696 		break;
697 	case CEPH_OSD_OP_STARTSYNC:
698 		break;
699 	case CEPH_OSD_OP_NOTIFY_ACK:
700 	case CEPH_OSD_OP_WATCH:
701 		dst->watch.cookie = cpu_to_le64(src->watch.cookie);
702 		dst->watch.ver = cpu_to_le64(src->watch.ver);
703 		dst->watch.flag = src->watch.flag;
704 		break;
705 	case CEPH_OSD_OP_SETALLOCHINT:
706 		dst->alloc_hint.expected_object_size =
707 		    cpu_to_le64(src->alloc_hint.expected_object_size);
708 		dst->alloc_hint.expected_write_size =
709 		    cpu_to_le64(src->alloc_hint.expected_write_size);
710 		break;
711 	default:
712 		pr_err("unsupported osd opcode %s\n",
713 			ceph_osd_op_name(src->op));
714 		WARN_ON(1);
715 
716 		return 0;
717 	}
718 
719 	dst->op = cpu_to_le16(src->op);
720 	dst->flags = cpu_to_le32(src->flags);
721 	dst->payload_len = cpu_to_le32(src->payload_len);
722 
723 	return request_data_len;
724 }
725 
726 /*
727  * build new request AND message, calculate layout, and adjust file
728  * extent as needed.
729  *
730  * if the file was recently truncated, we include information about its
731  * old and new size so that the object can be updated appropriately.  (we
732  * avoid synchronously deleting truncated objects because it's slow.)
733  *
734  * if @do_sync, include a 'startsync' command so that the osd will flush
735  * data quickly.
736  */
737 struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
738 					       struct ceph_file_layout *layout,
739 					       struct ceph_vino vino,
740 					       u64 off, u64 *plen, int num_ops,
741 					       int opcode, int flags,
742 					       struct ceph_snap_context *snapc,
743 					       u32 truncate_seq,
744 					       u64 truncate_size,
745 					       bool use_mempool)
746 {
747 	struct ceph_osd_request *req;
748 	u64 objnum = 0;
749 	u64 objoff = 0;
750 	u64 objlen = 0;
751 	u32 object_size;
752 	u64 object_base;
753 	int r;
754 
755 	BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
756 	       opcode != CEPH_OSD_OP_DELETE && opcode != CEPH_OSD_OP_ZERO &&
757 	       opcode != CEPH_OSD_OP_TRUNCATE);
758 
759 	req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool,
760 					GFP_NOFS);
761 	if (!req)
762 		return ERR_PTR(-ENOMEM);
763 
764 	req->r_flags = flags;
765 
766 	/* calculate max write size */
767 	r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
768 	if (r < 0) {
769 		ceph_osdc_put_request(req);
770 		return ERR_PTR(r);
771 	}
772 
773 	object_size = le32_to_cpu(layout->fl_object_size);
774 	object_base = off - objoff;
775 	if (!(truncate_seq == 1 && truncate_size == -1ULL)) {
776 		if (truncate_size <= object_base) {
777 			truncate_size = 0;
778 		} else {
779 			truncate_size -= object_base;
780 			if (truncate_size > object_size)
781 				truncate_size = object_size;
782 		}
783 	}
784 
785 	osd_req_op_extent_init(req, 0, opcode, objoff, objlen,
786 				truncate_size, truncate_seq);
787 
788 	/*
789 	 * A second op in the ops array means the caller wants to
790 	 * also issue a include a 'startsync' command so that the
791 	 * osd will flush data quickly.
792 	 */
793 	if (num_ops > 1)
794 		osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC);
795 
796 	req->r_base_oloc.pool = ceph_file_layout_pg_pool(*layout);
797 
798 	snprintf(req->r_base_oid.name, sizeof(req->r_base_oid.name),
799 		 "%llx.%08llx", vino.ino, objnum);
800 	req->r_base_oid.name_len = strlen(req->r_base_oid.name);
801 
802 	return req;
803 }
804 EXPORT_SYMBOL(ceph_osdc_new_request);
805 
806 /*
807  * We keep osd requests in an rbtree, sorted by ->r_tid.
808  */
809 static void __insert_request(struct ceph_osd_client *osdc,
810 			     struct ceph_osd_request *new)
811 {
812 	struct rb_node **p = &osdc->requests.rb_node;
813 	struct rb_node *parent = NULL;
814 	struct ceph_osd_request *req = NULL;
815 
816 	while (*p) {
817 		parent = *p;
818 		req = rb_entry(parent, struct ceph_osd_request, r_node);
819 		if (new->r_tid < req->r_tid)
820 			p = &(*p)->rb_left;
821 		else if (new->r_tid > req->r_tid)
822 			p = &(*p)->rb_right;
823 		else
824 			BUG();
825 	}
826 
827 	rb_link_node(&new->r_node, parent, p);
828 	rb_insert_color(&new->r_node, &osdc->requests);
829 }
830 
831 static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc,
832 						 u64 tid)
833 {
834 	struct ceph_osd_request *req;
835 	struct rb_node *n = osdc->requests.rb_node;
836 
837 	while (n) {
838 		req = rb_entry(n, struct ceph_osd_request, r_node);
839 		if (tid < req->r_tid)
840 			n = n->rb_left;
841 		else if (tid > req->r_tid)
842 			n = n->rb_right;
843 		else
844 			return req;
845 	}
846 	return NULL;
847 }
848 
849 static struct ceph_osd_request *
850 __lookup_request_ge(struct ceph_osd_client *osdc,
851 		    u64 tid)
852 {
853 	struct ceph_osd_request *req;
854 	struct rb_node *n = osdc->requests.rb_node;
855 
856 	while (n) {
857 		req = rb_entry(n, struct ceph_osd_request, r_node);
858 		if (tid < req->r_tid) {
859 			if (!n->rb_left)
860 				return req;
861 			n = n->rb_left;
862 		} else if (tid > req->r_tid) {
863 			n = n->rb_right;
864 		} else {
865 			return req;
866 		}
867 	}
868 	return NULL;
869 }
870 
871 /*
872  * Resubmit requests pending on the given osd.
873  */
874 static void __kick_osd_requests(struct ceph_osd_client *osdc,
875 				struct ceph_osd *osd)
876 {
877 	struct ceph_osd_request *req, *nreq;
878 	LIST_HEAD(resend);
879 	int err;
880 
881 	dout("__kick_osd_requests osd%d\n", osd->o_osd);
882 	err = __reset_osd(osdc, osd);
883 	if (err)
884 		return;
885 	/*
886 	 * Build up a list of requests to resend by traversing the
887 	 * osd's list of requests.  Requests for a given object are
888 	 * sent in tid order, and that is also the order they're
889 	 * kept on this list.  Therefore all requests that are in
890 	 * flight will be found first, followed by all requests that
891 	 * have not yet been sent.  And to resend requests while
892 	 * preserving this order we will want to put any sent
893 	 * requests back on the front of the osd client's unsent
894 	 * list.
895 	 *
896 	 * So we build a separate ordered list of already-sent
897 	 * requests for the affected osd and splice it onto the
898 	 * front of the osd client's unsent list.  Once we've seen a
899 	 * request that has not yet been sent we're done.  Those
900 	 * requests are already sitting right where they belong.
901 	 */
902 	list_for_each_entry(req, &osd->o_requests, r_osd_item) {
903 		if (!req->r_sent)
904 			break;
905 		list_move_tail(&req->r_req_lru_item, &resend);
906 		dout("requeueing %p tid %llu osd%d\n", req, req->r_tid,
907 		     osd->o_osd);
908 		if (!req->r_linger)
909 			req->r_flags |= CEPH_OSD_FLAG_RETRY;
910 	}
911 	list_splice(&resend, &osdc->req_unsent);
912 
913 	/*
914 	 * Linger requests are re-registered before sending, which
915 	 * sets up a new tid for each.  We add them to the unsent
916 	 * list at the end to keep things in tid order.
917 	 */
918 	list_for_each_entry_safe(req, nreq, &osd->o_linger_requests,
919 				 r_linger_osd) {
920 		/*
921 		 * reregister request prior to unregistering linger so
922 		 * that r_osd is preserved.
923 		 */
924 		BUG_ON(!list_empty(&req->r_req_lru_item));
925 		__register_request(osdc, req);
926 		list_add_tail(&req->r_req_lru_item, &osdc->req_unsent);
927 		list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
928 		__unregister_linger_request(osdc, req);
929 		dout("requeued lingering %p tid %llu osd%d\n", req, req->r_tid,
930 		     osd->o_osd);
931 	}
932 }
933 
934 /*
935  * If the osd connection drops, we need to resubmit all requests.
936  */
937 static void osd_reset(struct ceph_connection *con)
938 {
939 	struct ceph_osd *osd = con->private;
940 	struct ceph_osd_client *osdc;
941 
942 	if (!osd)
943 		return;
944 	dout("osd_reset osd%d\n", osd->o_osd);
945 	osdc = osd->o_osdc;
946 	down_read(&osdc->map_sem);
947 	mutex_lock(&osdc->request_mutex);
948 	__kick_osd_requests(osdc, osd);
949 	__send_queued(osdc);
950 	mutex_unlock(&osdc->request_mutex);
951 	up_read(&osdc->map_sem);
952 }
953 
954 /*
955  * Track open sessions with osds.
956  */
957 static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
958 {
959 	struct ceph_osd *osd;
960 
961 	osd = kzalloc(sizeof(*osd), GFP_NOFS);
962 	if (!osd)
963 		return NULL;
964 
965 	atomic_set(&osd->o_ref, 1);
966 	osd->o_osdc = osdc;
967 	osd->o_osd = onum;
968 	RB_CLEAR_NODE(&osd->o_node);
969 	INIT_LIST_HEAD(&osd->o_requests);
970 	INIT_LIST_HEAD(&osd->o_linger_requests);
971 	INIT_LIST_HEAD(&osd->o_osd_lru);
972 	osd->o_incarnation = 1;
973 
974 	ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
975 
976 	INIT_LIST_HEAD(&osd->o_keepalive_item);
977 	return osd;
978 }
979 
980 static struct ceph_osd *get_osd(struct ceph_osd *osd)
981 {
982 	if (atomic_inc_not_zero(&osd->o_ref)) {
983 		dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
984 		     atomic_read(&osd->o_ref));
985 		return osd;
986 	} else {
987 		dout("get_osd %p FAIL\n", osd);
988 		return NULL;
989 	}
990 }
991 
992 static void put_osd(struct ceph_osd *osd)
993 {
994 	dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
995 	     atomic_read(&osd->o_ref) - 1);
996 	if (atomic_dec_and_test(&osd->o_ref) && osd->o_auth.authorizer) {
997 		struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth;
998 
999 		ceph_auth_destroy_authorizer(ac, osd->o_auth.authorizer);
1000 		kfree(osd);
1001 	}
1002 }
1003 
1004 /*
1005  * remove an osd from our map
1006  */
1007 static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
1008 {
1009 	dout("__remove_osd %p\n", osd);
1010 	BUG_ON(!list_empty(&osd->o_requests));
1011 	rb_erase(&osd->o_node, &osdc->osds);
1012 	list_del_init(&osd->o_osd_lru);
1013 	ceph_con_close(&osd->o_con);
1014 	put_osd(osd);
1015 }
1016 
1017 static void remove_all_osds(struct ceph_osd_client *osdc)
1018 {
1019 	dout("%s %p\n", __func__, osdc);
1020 	mutex_lock(&osdc->request_mutex);
1021 	while (!RB_EMPTY_ROOT(&osdc->osds)) {
1022 		struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
1023 						struct ceph_osd, o_node);
1024 		__remove_osd(osdc, osd);
1025 	}
1026 	mutex_unlock(&osdc->request_mutex);
1027 }
1028 
1029 static void __move_osd_to_lru(struct ceph_osd_client *osdc,
1030 			      struct ceph_osd *osd)
1031 {
1032 	dout("__move_osd_to_lru %p\n", osd);
1033 	BUG_ON(!list_empty(&osd->o_osd_lru));
1034 	list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
1035 	osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl * HZ;
1036 }
1037 
1038 static void __remove_osd_from_lru(struct ceph_osd *osd)
1039 {
1040 	dout("__remove_osd_from_lru %p\n", osd);
1041 	if (!list_empty(&osd->o_osd_lru))
1042 		list_del_init(&osd->o_osd_lru);
1043 }
1044 
1045 static void remove_old_osds(struct ceph_osd_client *osdc)
1046 {
1047 	struct ceph_osd *osd, *nosd;
1048 
1049 	dout("__remove_old_osds %p\n", osdc);
1050 	mutex_lock(&osdc->request_mutex);
1051 	list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
1052 		if (time_before(jiffies, osd->lru_ttl))
1053 			break;
1054 		__remove_osd(osdc, osd);
1055 	}
1056 	mutex_unlock(&osdc->request_mutex);
1057 }
1058 
1059 /*
1060  * reset osd connect
1061  */
1062 static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
1063 {
1064 	struct ceph_entity_addr *peer_addr;
1065 
1066 	dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
1067 	if (list_empty(&osd->o_requests) &&
1068 	    list_empty(&osd->o_linger_requests)) {
1069 		__remove_osd(osdc, osd);
1070 
1071 		return -ENODEV;
1072 	}
1073 
1074 	peer_addr = &osdc->osdmap->osd_addr[osd->o_osd];
1075 	if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1076 			!ceph_con_opened(&osd->o_con)) {
1077 		struct ceph_osd_request *req;
1078 
1079 		dout("osd addr hasn't changed and connection never opened, "
1080 		     "letting msgr retry\n");
1081 		/* touch each r_stamp for handle_timeout()'s benfit */
1082 		list_for_each_entry(req, &osd->o_requests, r_osd_item)
1083 			req->r_stamp = jiffies;
1084 
1085 		return -EAGAIN;
1086 	}
1087 
1088 	ceph_con_close(&osd->o_con);
1089 	ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1090 	osd->o_incarnation++;
1091 
1092 	return 0;
1093 }
1094 
1095 static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new)
1096 {
1097 	struct rb_node **p = &osdc->osds.rb_node;
1098 	struct rb_node *parent = NULL;
1099 	struct ceph_osd *osd = NULL;
1100 
1101 	dout("__insert_osd %p osd%d\n", new, new->o_osd);
1102 	while (*p) {
1103 		parent = *p;
1104 		osd = rb_entry(parent, struct ceph_osd, o_node);
1105 		if (new->o_osd < osd->o_osd)
1106 			p = &(*p)->rb_left;
1107 		else if (new->o_osd > osd->o_osd)
1108 			p = &(*p)->rb_right;
1109 		else
1110 			BUG();
1111 	}
1112 
1113 	rb_link_node(&new->o_node, parent, p);
1114 	rb_insert_color(&new->o_node, &osdc->osds);
1115 }
1116 
1117 static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o)
1118 {
1119 	struct ceph_osd *osd;
1120 	struct rb_node *n = osdc->osds.rb_node;
1121 
1122 	while (n) {
1123 		osd = rb_entry(n, struct ceph_osd, o_node);
1124 		if (o < osd->o_osd)
1125 			n = n->rb_left;
1126 		else if (o > osd->o_osd)
1127 			n = n->rb_right;
1128 		else
1129 			return osd;
1130 	}
1131 	return NULL;
1132 }
1133 
1134 static void __schedule_osd_timeout(struct ceph_osd_client *osdc)
1135 {
1136 	schedule_delayed_work(&osdc->timeout_work,
1137 			osdc->client->options->osd_keepalive_timeout * HZ);
1138 }
1139 
1140 static void __cancel_osd_timeout(struct ceph_osd_client *osdc)
1141 {
1142 	cancel_delayed_work(&osdc->timeout_work);
1143 }
1144 
1145 /*
1146  * Register request, assign tid.  If this is the first request, set up
1147  * the timeout event.
1148  */
1149 static void __register_request(struct ceph_osd_client *osdc,
1150 			       struct ceph_osd_request *req)
1151 {
1152 	req->r_tid = ++osdc->last_tid;
1153 	req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
1154 	dout("__register_request %p tid %lld\n", req, req->r_tid);
1155 	__insert_request(osdc, req);
1156 	ceph_osdc_get_request(req);
1157 	osdc->num_requests++;
1158 	if (osdc->num_requests == 1) {
1159 		dout(" first request, scheduling timeout\n");
1160 		__schedule_osd_timeout(osdc);
1161 	}
1162 }
1163 
1164 /*
1165  * called under osdc->request_mutex
1166  */
1167 static void __unregister_request(struct ceph_osd_client *osdc,
1168 				 struct ceph_osd_request *req)
1169 {
1170 	if (RB_EMPTY_NODE(&req->r_node)) {
1171 		dout("__unregister_request %p tid %lld not registered\n",
1172 			req, req->r_tid);
1173 		return;
1174 	}
1175 
1176 	dout("__unregister_request %p tid %lld\n", req, req->r_tid);
1177 	rb_erase(&req->r_node, &osdc->requests);
1178 	osdc->num_requests--;
1179 
1180 	if (req->r_osd) {
1181 		/* make sure the original request isn't in flight. */
1182 		ceph_msg_revoke(req->r_request);
1183 
1184 		list_del_init(&req->r_osd_item);
1185 		if (list_empty(&req->r_osd->o_requests) &&
1186 		    list_empty(&req->r_osd->o_linger_requests)) {
1187 			dout("moving osd to %p lru\n", req->r_osd);
1188 			__move_osd_to_lru(osdc, req->r_osd);
1189 		}
1190 		if (list_empty(&req->r_linger_item))
1191 			req->r_osd = NULL;
1192 	}
1193 
1194 	list_del_init(&req->r_req_lru_item);
1195 	ceph_osdc_put_request(req);
1196 
1197 	if (osdc->num_requests == 0) {
1198 		dout(" no requests, canceling timeout\n");
1199 		__cancel_osd_timeout(osdc);
1200 	}
1201 }
1202 
1203 /*
1204  * Cancel a previously queued request message
1205  */
1206 static void __cancel_request(struct ceph_osd_request *req)
1207 {
1208 	if (req->r_sent && req->r_osd) {
1209 		ceph_msg_revoke(req->r_request);
1210 		req->r_sent = 0;
1211 	}
1212 }
1213 
1214 static void __register_linger_request(struct ceph_osd_client *osdc,
1215 				    struct ceph_osd_request *req)
1216 {
1217 	dout("__register_linger_request %p\n", req);
1218 	ceph_osdc_get_request(req);
1219 	list_add_tail(&req->r_linger_item, &osdc->req_linger);
1220 	if (req->r_osd)
1221 		list_add_tail(&req->r_linger_osd,
1222 			      &req->r_osd->o_linger_requests);
1223 }
1224 
1225 static void __unregister_linger_request(struct ceph_osd_client *osdc,
1226 					struct ceph_osd_request *req)
1227 {
1228 	dout("__unregister_linger_request %p\n", req);
1229 	list_del_init(&req->r_linger_item);
1230 	if (req->r_osd) {
1231 		list_del_init(&req->r_linger_osd);
1232 
1233 		if (list_empty(&req->r_osd->o_requests) &&
1234 		    list_empty(&req->r_osd->o_linger_requests)) {
1235 			dout("moving osd to %p lru\n", req->r_osd);
1236 			__move_osd_to_lru(osdc, req->r_osd);
1237 		}
1238 		if (list_empty(&req->r_osd_item))
1239 			req->r_osd = NULL;
1240 	}
1241 	ceph_osdc_put_request(req);
1242 }
1243 
1244 void ceph_osdc_unregister_linger_request(struct ceph_osd_client *osdc,
1245 					 struct ceph_osd_request *req)
1246 {
1247 	mutex_lock(&osdc->request_mutex);
1248 	if (req->r_linger) {
1249 		req->r_linger = 0;
1250 		__unregister_linger_request(osdc, req);
1251 	}
1252 	mutex_unlock(&osdc->request_mutex);
1253 }
1254 EXPORT_SYMBOL(ceph_osdc_unregister_linger_request);
1255 
1256 void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
1257 				  struct ceph_osd_request *req)
1258 {
1259 	if (!req->r_linger) {
1260 		dout("set_request_linger %p\n", req);
1261 		req->r_linger = 1;
1262 	}
1263 }
1264 EXPORT_SYMBOL(ceph_osdc_set_request_linger);
1265 
1266 /*
1267  * Returns whether a request should be blocked from being sent
1268  * based on the current osdmap and osd_client settings.
1269  *
1270  * Caller should hold map_sem for read.
1271  */
1272 static bool __req_should_be_paused(struct ceph_osd_client *osdc,
1273 				   struct ceph_osd_request *req)
1274 {
1275 	bool pauserd = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD);
1276 	bool pausewr = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR) ||
1277 		ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
1278 	return (req->r_flags & CEPH_OSD_FLAG_READ && pauserd) ||
1279 		(req->r_flags & CEPH_OSD_FLAG_WRITE && pausewr);
1280 }
1281 
1282 /*
1283  * Calculate mapping of a request to a PG.  Takes tiering into account.
1284  */
1285 static int __calc_request_pg(struct ceph_osdmap *osdmap,
1286 			     struct ceph_osd_request *req,
1287 			     struct ceph_pg *pg_out)
1288 {
1289 	bool need_check_tiering;
1290 
1291 	need_check_tiering = false;
1292 	if (req->r_target_oloc.pool == -1) {
1293 		req->r_target_oloc = req->r_base_oloc; /* struct */
1294 		need_check_tiering = true;
1295 	}
1296 	if (req->r_target_oid.name_len == 0) {
1297 		ceph_oid_copy(&req->r_target_oid, &req->r_base_oid);
1298 		need_check_tiering = true;
1299 	}
1300 
1301 	if (need_check_tiering &&
1302 	    (req->r_flags & CEPH_OSD_FLAG_IGNORE_OVERLAY) == 0) {
1303 		struct ceph_pg_pool_info *pi;
1304 
1305 		pi = ceph_pg_pool_by_id(osdmap, req->r_target_oloc.pool);
1306 		if (pi) {
1307 			if ((req->r_flags & CEPH_OSD_FLAG_READ) &&
1308 			    pi->read_tier >= 0)
1309 				req->r_target_oloc.pool = pi->read_tier;
1310 			if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
1311 			    pi->write_tier >= 0)
1312 				req->r_target_oloc.pool = pi->write_tier;
1313 		}
1314 		/* !pi is caught in ceph_oloc_oid_to_pg() */
1315 	}
1316 
1317 	return ceph_oloc_oid_to_pg(osdmap, &req->r_target_oloc,
1318 				   &req->r_target_oid, pg_out);
1319 }
1320 
1321 /*
1322  * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
1323  * (as needed), and set the request r_osd appropriately.  If there is
1324  * no up osd, set r_osd to NULL.  Move the request to the appropriate list
1325  * (unsent, homeless) or leave on in-flight lru.
1326  *
1327  * Return 0 if unchanged, 1 if changed, or negative on error.
1328  *
1329  * Caller should hold map_sem for read and request_mutex.
1330  */
1331 static int __map_request(struct ceph_osd_client *osdc,
1332 			 struct ceph_osd_request *req, int force_resend)
1333 {
1334 	struct ceph_pg pgid;
1335 	int acting[CEPH_PG_MAX_SIZE];
1336 	int num, o;
1337 	int err;
1338 	bool was_paused;
1339 
1340 	dout("map_request %p tid %lld\n", req, req->r_tid);
1341 
1342 	err = __calc_request_pg(osdc->osdmap, req, &pgid);
1343 	if (err) {
1344 		list_move(&req->r_req_lru_item, &osdc->req_notarget);
1345 		return err;
1346 	}
1347 	req->r_pgid = pgid;
1348 
1349 	num = ceph_calc_pg_acting(osdc->osdmap, pgid, acting, &o);
1350 	if (num < 0)
1351 		num = 0;
1352 
1353 	was_paused = req->r_paused;
1354 	req->r_paused = __req_should_be_paused(osdc, req);
1355 	if (was_paused && !req->r_paused)
1356 		force_resend = 1;
1357 
1358 	if ((!force_resend &&
1359 	     req->r_osd && req->r_osd->o_osd == o &&
1360 	     req->r_sent >= req->r_osd->o_incarnation &&
1361 	     req->r_num_pg_osds == num &&
1362 	     memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
1363 	    (req->r_osd == NULL && o == -1) ||
1364 	    req->r_paused)
1365 		return 0;  /* no change */
1366 
1367 	dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n",
1368 	     req->r_tid, pgid.pool, pgid.seed, o,
1369 	     req->r_osd ? req->r_osd->o_osd : -1);
1370 
1371 	/* record full pg acting set */
1372 	memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
1373 	req->r_num_pg_osds = num;
1374 
1375 	if (req->r_osd) {
1376 		__cancel_request(req);
1377 		list_del_init(&req->r_osd_item);
1378 		req->r_osd = NULL;
1379 	}
1380 
1381 	req->r_osd = __lookup_osd(osdc, o);
1382 	if (!req->r_osd && o >= 0) {
1383 		err = -ENOMEM;
1384 		req->r_osd = create_osd(osdc, o);
1385 		if (!req->r_osd) {
1386 			list_move(&req->r_req_lru_item, &osdc->req_notarget);
1387 			goto out;
1388 		}
1389 
1390 		dout("map_request osd %p is osd%d\n", req->r_osd, o);
1391 		__insert_osd(osdc, req->r_osd);
1392 
1393 		ceph_con_open(&req->r_osd->o_con,
1394 			      CEPH_ENTITY_TYPE_OSD, o,
1395 			      &osdc->osdmap->osd_addr[o]);
1396 	}
1397 
1398 	if (req->r_osd) {
1399 		__remove_osd_from_lru(req->r_osd);
1400 		list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
1401 		list_move_tail(&req->r_req_lru_item, &osdc->req_unsent);
1402 	} else {
1403 		list_move_tail(&req->r_req_lru_item, &osdc->req_notarget);
1404 	}
1405 	err = 1;   /* osd or pg changed */
1406 
1407 out:
1408 	return err;
1409 }
1410 
1411 /*
1412  * caller should hold map_sem (for read) and request_mutex
1413  */
1414 static void __send_request(struct ceph_osd_client *osdc,
1415 			   struct ceph_osd_request *req)
1416 {
1417 	void *p;
1418 
1419 	dout("send_request %p tid %llu to osd%d flags %d pg %lld.%x\n",
1420 	     req, req->r_tid, req->r_osd->o_osd, req->r_flags,
1421 	     (unsigned long long)req->r_pgid.pool, req->r_pgid.seed);
1422 
1423 	/* fill in message content that changes each time we send it */
1424 	put_unaligned_le32(osdc->osdmap->epoch, req->r_request_osdmap_epoch);
1425 	put_unaligned_le32(req->r_flags, req->r_request_flags);
1426 	put_unaligned_le64(req->r_target_oloc.pool, req->r_request_pool);
1427 	p = req->r_request_pgid;
1428 	ceph_encode_64(&p, req->r_pgid.pool);
1429 	ceph_encode_32(&p, req->r_pgid.seed);
1430 	put_unaligned_le64(1, req->r_request_attempts);  /* FIXME */
1431 	memcpy(req->r_request_reassert_version, &req->r_reassert_version,
1432 	       sizeof(req->r_reassert_version));
1433 
1434 	req->r_stamp = jiffies;
1435 	list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
1436 
1437 	ceph_msg_get(req->r_request); /* send consumes a ref */
1438 
1439 	req->r_sent = req->r_osd->o_incarnation;
1440 
1441 	ceph_con_send(&req->r_osd->o_con, req->r_request);
1442 }
1443 
1444 /*
1445  * Send any requests in the queue (req_unsent).
1446  */
1447 static void __send_queued(struct ceph_osd_client *osdc)
1448 {
1449 	struct ceph_osd_request *req, *tmp;
1450 
1451 	dout("__send_queued\n");
1452 	list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item)
1453 		__send_request(osdc, req);
1454 }
1455 
1456 /*
1457  * Caller should hold map_sem for read and request_mutex.
1458  */
1459 static int __ceph_osdc_start_request(struct ceph_osd_client *osdc,
1460 				     struct ceph_osd_request *req,
1461 				     bool nofail)
1462 {
1463 	int rc;
1464 
1465 	__register_request(osdc, req);
1466 	req->r_sent = 0;
1467 	req->r_got_reply = 0;
1468 	rc = __map_request(osdc, req, 0);
1469 	if (rc < 0) {
1470 		if (nofail) {
1471 			dout("osdc_start_request failed map, "
1472 				" will retry %lld\n", req->r_tid);
1473 			rc = 0;
1474 		} else {
1475 			__unregister_request(osdc, req);
1476 		}
1477 		return rc;
1478 	}
1479 
1480 	if (req->r_osd == NULL) {
1481 		dout("send_request %p no up osds in pg\n", req);
1482 		ceph_monc_request_next_osdmap(&osdc->client->monc);
1483 	} else {
1484 		__send_queued(osdc);
1485 	}
1486 
1487 	return 0;
1488 }
1489 
1490 /*
1491  * Timeout callback, called every N seconds when 1 or more osd
1492  * requests has been active for more than N seconds.  When this
1493  * happens, we ping all OSDs with requests who have timed out to
1494  * ensure any communications channel reset is detected.  Reset the
1495  * request timeouts another N seconds in the future as we go.
1496  * Reschedule the timeout event another N seconds in future (unless
1497  * there are no open requests).
1498  */
1499 static void handle_timeout(struct work_struct *work)
1500 {
1501 	struct ceph_osd_client *osdc =
1502 		container_of(work, struct ceph_osd_client, timeout_work.work);
1503 	struct ceph_osd_request *req;
1504 	struct ceph_osd *osd;
1505 	unsigned long keepalive =
1506 		osdc->client->options->osd_keepalive_timeout * HZ;
1507 	struct list_head slow_osds;
1508 	dout("timeout\n");
1509 	down_read(&osdc->map_sem);
1510 
1511 	ceph_monc_request_next_osdmap(&osdc->client->monc);
1512 
1513 	mutex_lock(&osdc->request_mutex);
1514 
1515 	/*
1516 	 * ping osds that are a bit slow.  this ensures that if there
1517 	 * is a break in the TCP connection we will notice, and reopen
1518 	 * a connection with that osd (from the fault callback).
1519 	 */
1520 	INIT_LIST_HEAD(&slow_osds);
1521 	list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
1522 		if (time_before(jiffies, req->r_stamp + keepalive))
1523 			break;
1524 
1525 		osd = req->r_osd;
1526 		BUG_ON(!osd);
1527 		dout(" tid %llu is slow, will send keepalive on osd%d\n",
1528 		     req->r_tid, osd->o_osd);
1529 		list_move_tail(&osd->o_keepalive_item, &slow_osds);
1530 	}
1531 	while (!list_empty(&slow_osds)) {
1532 		osd = list_entry(slow_osds.next, struct ceph_osd,
1533 				 o_keepalive_item);
1534 		list_del_init(&osd->o_keepalive_item);
1535 		ceph_con_keepalive(&osd->o_con);
1536 	}
1537 
1538 	__schedule_osd_timeout(osdc);
1539 	__send_queued(osdc);
1540 	mutex_unlock(&osdc->request_mutex);
1541 	up_read(&osdc->map_sem);
1542 }
1543 
1544 static void handle_osds_timeout(struct work_struct *work)
1545 {
1546 	struct ceph_osd_client *osdc =
1547 		container_of(work, struct ceph_osd_client,
1548 			     osds_timeout_work.work);
1549 	unsigned long delay =
1550 		osdc->client->options->osd_idle_ttl * HZ >> 2;
1551 
1552 	dout("osds timeout\n");
1553 	down_read(&osdc->map_sem);
1554 	remove_old_osds(osdc);
1555 	up_read(&osdc->map_sem);
1556 
1557 	schedule_delayed_work(&osdc->osds_timeout_work,
1558 			      round_jiffies_relative(delay));
1559 }
1560 
1561 static int ceph_oloc_decode(void **p, void *end,
1562 			    struct ceph_object_locator *oloc)
1563 {
1564 	u8 struct_v, struct_cv;
1565 	u32 len;
1566 	void *struct_end;
1567 	int ret = 0;
1568 
1569 	ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
1570 	struct_v = ceph_decode_8(p);
1571 	struct_cv = ceph_decode_8(p);
1572 	if (struct_v < 3) {
1573 		pr_warn("got v %d < 3 cv %d of ceph_object_locator\n",
1574 			struct_v, struct_cv);
1575 		goto e_inval;
1576 	}
1577 	if (struct_cv > 6) {
1578 		pr_warn("got v %d cv %d > 6 of ceph_object_locator\n",
1579 			struct_v, struct_cv);
1580 		goto e_inval;
1581 	}
1582 	len = ceph_decode_32(p);
1583 	ceph_decode_need(p, end, len, e_inval);
1584 	struct_end = *p + len;
1585 
1586 	oloc->pool = ceph_decode_64(p);
1587 	*p += 4; /* skip preferred */
1588 
1589 	len = ceph_decode_32(p);
1590 	if (len > 0) {
1591 		pr_warn("ceph_object_locator::key is set\n");
1592 		goto e_inval;
1593 	}
1594 
1595 	if (struct_v >= 5) {
1596 		len = ceph_decode_32(p);
1597 		if (len > 0) {
1598 			pr_warn("ceph_object_locator::nspace is set\n");
1599 			goto e_inval;
1600 		}
1601 	}
1602 
1603 	if (struct_v >= 6) {
1604 		s64 hash = ceph_decode_64(p);
1605 		if (hash != -1) {
1606 			pr_warn("ceph_object_locator::hash is set\n");
1607 			goto e_inval;
1608 		}
1609 	}
1610 
1611 	/* skip the rest */
1612 	*p = struct_end;
1613 out:
1614 	return ret;
1615 
1616 e_inval:
1617 	ret = -EINVAL;
1618 	goto out;
1619 }
1620 
1621 static int ceph_redirect_decode(void **p, void *end,
1622 				struct ceph_request_redirect *redir)
1623 {
1624 	u8 struct_v, struct_cv;
1625 	u32 len;
1626 	void *struct_end;
1627 	int ret;
1628 
1629 	ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
1630 	struct_v = ceph_decode_8(p);
1631 	struct_cv = ceph_decode_8(p);
1632 	if (struct_cv > 1) {
1633 		pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n",
1634 			struct_v, struct_cv);
1635 		goto e_inval;
1636 	}
1637 	len = ceph_decode_32(p);
1638 	ceph_decode_need(p, end, len, e_inval);
1639 	struct_end = *p + len;
1640 
1641 	ret = ceph_oloc_decode(p, end, &redir->oloc);
1642 	if (ret)
1643 		goto out;
1644 
1645 	len = ceph_decode_32(p);
1646 	if (len > 0) {
1647 		pr_warn("ceph_request_redirect::object_name is set\n");
1648 		goto e_inval;
1649 	}
1650 
1651 	len = ceph_decode_32(p);
1652 	*p += len; /* skip osd_instructions */
1653 
1654 	/* skip the rest */
1655 	*p = struct_end;
1656 out:
1657 	return ret;
1658 
1659 e_inval:
1660 	ret = -EINVAL;
1661 	goto out;
1662 }
1663 
1664 static void complete_request(struct ceph_osd_request *req)
1665 {
1666 	complete_all(&req->r_safe_completion);  /* fsync waiter */
1667 }
1668 
1669 /*
1670  * handle osd op reply.  either call the callback if it is specified,
1671  * or do the completion to wake up the waiting thread.
1672  */
1673 static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg,
1674 			 struct ceph_connection *con)
1675 {
1676 	void *p, *end;
1677 	struct ceph_osd_request *req;
1678 	struct ceph_request_redirect redir;
1679 	u64 tid;
1680 	int object_len;
1681 	unsigned int numops;
1682 	int payload_len, flags;
1683 	s32 result;
1684 	s32 retry_attempt;
1685 	struct ceph_pg pg;
1686 	int err;
1687 	u32 reassert_epoch;
1688 	u64 reassert_version;
1689 	u32 osdmap_epoch;
1690 	int already_completed;
1691 	u32 bytes;
1692 	unsigned int i;
1693 
1694 	tid = le64_to_cpu(msg->hdr.tid);
1695 	dout("handle_reply %p tid %llu\n", msg, tid);
1696 
1697 	p = msg->front.iov_base;
1698 	end = p + msg->front.iov_len;
1699 
1700 	ceph_decode_need(&p, end, 4, bad);
1701 	object_len = ceph_decode_32(&p);
1702 	ceph_decode_need(&p, end, object_len, bad);
1703 	p += object_len;
1704 
1705 	err = ceph_decode_pgid(&p, end, &pg);
1706 	if (err)
1707 		goto bad;
1708 
1709 	ceph_decode_need(&p, end, 8 + 4 + 4 + 8 + 4, bad);
1710 	flags = ceph_decode_64(&p);
1711 	result = ceph_decode_32(&p);
1712 	reassert_epoch = ceph_decode_32(&p);
1713 	reassert_version = ceph_decode_64(&p);
1714 	osdmap_epoch = ceph_decode_32(&p);
1715 
1716 	/* lookup */
1717 	down_read(&osdc->map_sem);
1718 	mutex_lock(&osdc->request_mutex);
1719 	req = __lookup_request(osdc, tid);
1720 	if (req == NULL) {
1721 		dout("handle_reply tid %llu dne\n", tid);
1722 		goto bad_mutex;
1723 	}
1724 	ceph_osdc_get_request(req);
1725 
1726 	dout("handle_reply %p tid %llu req %p result %d\n", msg, tid,
1727 	     req, result);
1728 
1729 	ceph_decode_need(&p, end, 4, bad_put);
1730 	numops = ceph_decode_32(&p);
1731 	if (numops > CEPH_OSD_MAX_OP)
1732 		goto bad_put;
1733 	if (numops != req->r_num_ops)
1734 		goto bad_put;
1735 	payload_len = 0;
1736 	ceph_decode_need(&p, end, numops * sizeof(struct ceph_osd_op), bad_put);
1737 	for (i = 0; i < numops; i++) {
1738 		struct ceph_osd_op *op = p;
1739 		int len;
1740 
1741 		len = le32_to_cpu(op->payload_len);
1742 		req->r_reply_op_len[i] = len;
1743 		dout(" op %d has %d bytes\n", i, len);
1744 		payload_len += len;
1745 		p += sizeof(*op);
1746 	}
1747 	bytes = le32_to_cpu(msg->hdr.data_len);
1748 	if (payload_len != bytes) {
1749 		pr_warning("sum of op payload lens %d != data_len %d",
1750 			   payload_len, bytes);
1751 		goto bad_put;
1752 	}
1753 
1754 	ceph_decode_need(&p, end, 4 + numops * 4, bad_put);
1755 	retry_attempt = ceph_decode_32(&p);
1756 	for (i = 0; i < numops; i++)
1757 		req->r_reply_op_result[i] = ceph_decode_32(&p);
1758 
1759 	if (le16_to_cpu(msg->hdr.version) >= 6) {
1760 		p += 8 + 4; /* skip replay_version */
1761 		p += 8; /* skip user_version */
1762 
1763 		err = ceph_redirect_decode(&p, end, &redir);
1764 		if (err)
1765 			goto bad_put;
1766 	} else {
1767 		redir.oloc.pool = -1;
1768 	}
1769 
1770 	if (redir.oloc.pool != -1) {
1771 		dout("redirect pool %lld\n", redir.oloc.pool);
1772 
1773 		__unregister_request(osdc, req);
1774 
1775 		req->r_target_oloc = redir.oloc; /* struct */
1776 
1777 		/*
1778 		 * Start redirect requests with nofail=true.  If
1779 		 * mapping fails, request will end up on the notarget
1780 		 * list, waiting for the new osdmap (which can take
1781 		 * a while), even though the original request mapped
1782 		 * successfully.  In the future we might want to follow
1783 		 * original request's nofail setting here.
1784 		 */
1785 		err = __ceph_osdc_start_request(osdc, req, true);
1786 		BUG_ON(err);
1787 
1788 		goto out_unlock;
1789 	}
1790 
1791 	already_completed = req->r_got_reply;
1792 	if (!req->r_got_reply) {
1793 		req->r_result = result;
1794 		dout("handle_reply result %d bytes %d\n", req->r_result,
1795 		     bytes);
1796 		if (req->r_result == 0)
1797 			req->r_result = bytes;
1798 
1799 		/* in case this is a write and we need to replay, */
1800 		req->r_reassert_version.epoch = cpu_to_le32(reassert_epoch);
1801 		req->r_reassert_version.version = cpu_to_le64(reassert_version);
1802 
1803 		req->r_got_reply = 1;
1804 	} else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) {
1805 		dout("handle_reply tid %llu dup ack\n", tid);
1806 		goto out_unlock;
1807 	}
1808 
1809 	dout("handle_reply tid %llu flags %d\n", tid, flags);
1810 
1811 	if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK))
1812 		__register_linger_request(osdc, req);
1813 
1814 	/* either this is a read, or we got the safe response */
1815 	if (result < 0 ||
1816 	    (flags & CEPH_OSD_FLAG_ONDISK) ||
1817 	    ((flags & CEPH_OSD_FLAG_WRITE) == 0))
1818 		__unregister_request(osdc, req);
1819 
1820 	mutex_unlock(&osdc->request_mutex);
1821 	up_read(&osdc->map_sem);
1822 
1823 	if (!already_completed) {
1824 		if (req->r_unsafe_callback &&
1825 		    result >= 0 && !(flags & CEPH_OSD_FLAG_ONDISK))
1826 			req->r_unsafe_callback(req, true);
1827 		if (req->r_callback)
1828 			req->r_callback(req, msg);
1829 		else
1830 			complete_all(&req->r_completion);
1831 	}
1832 
1833 	if (flags & CEPH_OSD_FLAG_ONDISK) {
1834 		if (req->r_unsafe_callback && already_completed)
1835 			req->r_unsafe_callback(req, false);
1836 		complete_request(req);
1837 	}
1838 
1839 out:
1840 	dout("req=%p req->r_linger=%d\n", req, req->r_linger);
1841 	ceph_osdc_put_request(req);
1842 	return;
1843 out_unlock:
1844 	mutex_unlock(&osdc->request_mutex);
1845 	up_read(&osdc->map_sem);
1846 	goto out;
1847 
1848 bad_put:
1849 	req->r_result = -EIO;
1850 	__unregister_request(osdc, req);
1851 	if (req->r_callback)
1852 		req->r_callback(req, msg);
1853 	else
1854 		complete_all(&req->r_completion);
1855 	complete_request(req);
1856 	ceph_osdc_put_request(req);
1857 bad_mutex:
1858 	mutex_unlock(&osdc->request_mutex);
1859 	up_read(&osdc->map_sem);
1860 bad:
1861 	pr_err("corrupt osd_op_reply got %d %d\n",
1862 	       (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len));
1863 	ceph_msg_dump(msg);
1864 }
1865 
1866 static void reset_changed_osds(struct ceph_osd_client *osdc)
1867 {
1868 	struct rb_node *p, *n;
1869 
1870 	for (p = rb_first(&osdc->osds); p; p = n) {
1871 		struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
1872 
1873 		n = rb_next(p);
1874 		if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
1875 		    memcmp(&osd->o_con.peer_addr,
1876 			   ceph_osd_addr(osdc->osdmap,
1877 					 osd->o_osd),
1878 			   sizeof(struct ceph_entity_addr)) != 0)
1879 			__reset_osd(osdc, osd);
1880 	}
1881 }
1882 
1883 /*
1884  * Requeue requests whose mapping to an OSD has changed.  If requests map to
1885  * no osd, request a new map.
1886  *
1887  * Caller should hold map_sem for read.
1888  */
1889 static void kick_requests(struct ceph_osd_client *osdc, bool force_resend,
1890 			  bool force_resend_writes)
1891 {
1892 	struct ceph_osd_request *req, *nreq;
1893 	struct rb_node *p;
1894 	int needmap = 0;
1895 	int err;
1896 	bool force_resend_req;
1897 
1898 	dout("kick_requests %s %s\n", force_resend ? " (force resend)" : "",
1899 		force_resend_writes ? " (force resend writes)" : "");
1900 	mutex_lock(&osdc->request_mutex);
1901 	for (p = rb_first(&osdc->requests); p; ) {
1902 		req = rb_entry(p, struct ceph_osd_request, r_node);
1903 		p = rb_next(p);
1904 
1905 		/*
1906 		 * For linger requests that have not yet been
1907 		 * registered, move them to the linger list; they'll
1908 		 * be sent to the osd in the loop below.  Unregister
1909 		 * the request before re-registering it as a linger
1910 		 * request to ensure the __map_request() below
1911 		 * will decide it needs to be sent.
1912 		 */
1913 		if (req->r_linger && list_empty(&req->r_linger_item)) {
1914 			dout("%p tid %llu restart on osd%d\n",
1915 			     req, req->r_tid,
1916 			     req->r_osd ? req->r_osd->o_osd : -1);
1917 			ceph_osdc_get_request(req);
1918 			__unregister_request(osdc, req);
1919 			__register_linger_request(osdc, req);
1920 			ceph_osdc_put_request(req);
1921 			continue;
1922 		}
1923 
1924 		force_resend_req = force_resend ||
1925 			(force_resend_writes &&
1926 				req->r_flags & CEPH_OSD_FLAG_WRITE);
1927 		err = __map_request(osdc, req, force_resend_req);
1928 		if (err < 0)
1929 			continue;  /* error */
1930 		if (req->r_osd == NULL) {
1931 			dout("%p tid %llu maps to no osd\n", req, req->r_tid);
1932 			needmap++;  /* request a newer map */
1933 		} else if (err > 0) {
1934 			if (!req->r_linger) {
1935 				dout("%p tid %llu requeued on osd%d\n", req,
1936 				     req->r_tid,
1937 				     req->r_osd ? req->r_osd->o_osd : -1);
1938 				req->r_flags |= CEPH_OSD_FLAG_RETRY;
1939 			}
1940 		}
1941 	}
1942 
1943 	list_for_each_entry_safe(req, nreq, &osdc->req_linger,
1944 				 r_linger_item) {
1945 		dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
1946 
1947 		err = __map_request(osdc, req,
1948 				    force_resend || force_resend_writes);
1949 		dout("__map_request returned %d\n", err);
1950 		if (err == 0)
1951 			continue;  /* no change and no osd was specified */
1952 		if (err < 0)
1953 			continue;  /* hrm! */
1954 		if (req->r_osd == NULL) {
1955 			dout("tid %llu maps to no valid osd\n", req->r_tid);
1956 			needmap++;  /* request a newer map */
1957 			continue;
1958 		}
1959 
1960 		dout("kicking lingering %p tid %llu osd%d\n", req, req->r_tid,
1961 		     req->r_osd ? req->r_osd->o_osd : -1);
1962 		__register_request(osdc, req);
1963 		__unregister_linger_request(osdc, req);
1964 	}
1965 	reset_changed_osds(osdc);
1966 	mutex_unlock(&osdc->request_mutex);
1967 
1968 	if (needmap) {
1969 		dout("%d requests for down osds, need new map\n", needmap);
1970 		ceph_monc_request_next_osdmap(&osdc->client->monc);
1971 	}
1972 }
1973 
1974 
1975 /*
1976  * Process updated osd map.
1977  *
1978  * The message contains any number of incremental and full maps, normally
1979  * indicating some sort of topology change in the cluster.  Kick requests
1980  * off to different OSDs as needed.
1981  */
1982 void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1983 {
1984 	void *p, *end, *next;
1985 	u32 nr_maps, maplen;
1986 	u32 epoch;
1987 	struct ceph_osdmap *newmap = NULL, *oldmap;
1988 	int err;
1989 	struct ceph_fsid fsid;
1990 	bool was_full;
1991 
1992 	dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
1993 	p = msg->front.iov_base;
1994 	end = p + msg->front.iov_len;
1995 
1996 	/* verify fsid */
1997 	ceph_decode_need(&p, end, sizeof(fsid), bad);
1998 	ceph_decode_copy(&p, &fsid, sizeof(fsid));
1999 	if (ceph_check_fsid(osdc->client, &fsid) < 0)
2000 		return;
2001 
2002 	down_write(&osdc->map_sem);
2003 
2004 	was_full = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
2005 
2006 	/* incremental maps */
2007 	ceph_decode_32_safe(&p, end, nr_maps, bad);
2008 	dout(" %d inc maps\n", nr_maps);
2009 	while (nr_maps > 0) {
2010 		ceph_decode_need(&p, end, 2*sizeof(u32), bad);
2011 		epoch = ceph_decode_32(&p);
2012 		maplen = ceph_decode_32(&p);
2013 		ceph_decode_need(&p, end, maplen, bad);
2014 		next = p + maplen;
2015 		if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
2016 			dout("applying incremental map %u len %d\n",
2017 			     epoch, maplen);
2018 			newmap = osdmap_apply_incremental(&p, next,
2019 							  osdc->osdmap,
2020 							  &osdc->client->msgr);
2021 			if (IS_ERR(newmap)) {
2022 				err = PTR_ERR(newmap);
2023 				goto bad;
2024 			}
2025 			BUG_ON(!newmap);
2026 			if (newmap != osdc->osdmap) {
2027 				ceph_osdmap_destroy(osdc->osdmap);
2028 				osdc->osdmap = newmap;
2029 			}
2030 			was_full = was_full ||
2031 				ceph_osdmap_flag(osdc->osdmap,
2032 						 CEPH_OSDMAP_FULL);
2033 			kick_requests(osdc, 0, was_full);
2034 		} else {
2035 			dout("ignoring incremental map %u len %d\n",
2036 			     epoch, maplen);
2037 		}
2038 		p = next;
2039 		nr_maps--;
2040 	}
2041 	if (newmap)
2042 		goto done;
2043 
2044 	/* full maps */
2045 	ceph_decode_32_safe(&p, end, nr_maps, bad);
2046 	dout(" %d full maps\n", nr_maps);
2047 	while (nr_maps) {
2048 		ceph_decode_need(&p, end, 2*sizeof(u32), bad);
2049 		epoch = ceph_decode_32(&p);
2050 		maplen = ceph_decode_32(&p);
2051 		ceph_decode_need(&p, end, maplen, bad);
2052 		if (nr_maps > 1) {
2053 			dout("skipping non-latest full map %u len %d\n",
2054 			     epoch, maplen);
2055 		} else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
2056 			dout("skipping full map %u len %d, "
2057 			     "older than our %u\n", epoch, maplen,
2058 			     osdc->osdmap->epoch);
2059 		} else {
2060 			int skipped_map = 0;
2061 
2062 			dout("taking full map %u len %d\n", epoch, maplen);
2063 			newmap = ceph_osdmap_decode(&p, p+maplen);
2064 			if (IS_ERR(newmap)) {
2065 				err = PTR_ERR(newmap);
2066 				goto bad;
2067 			}
2068 			BUG_ON(!newmap);
2069 			oldmap = osdc->osdmap;
2070 			osdc->osdmap = newmap;
2071 			if (oldmap) {
2072 				if (oldmap->epoch + 1 < newmap->epoch)
2073 					skipped_map = 1;
2074 				ceph_osdmap_destroy(oldmap);
2075 			}
2076 			was_full = was_full ||
2077 				ceph_osdmap_flag(osdc->osdmap,
2078 						 CEPH_OSDMAP_FULL);
2079 			kick_requests(osdc, skipped_map, was_full);
2080 		}
2081 		p += maplen;
2082 		nr_maps--;
2083 	}
2084 
2085 	if (!osdc->osdmap)
2086 		goto bad;
2087 done:
2088 	downgrade_write(&osdc->map_sem);
2089 	ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch);
2090 
2091 	/*
2092 	 * subscribe to subsequent osdmap updates if full to ensure
2093 	 * we find out when we are no longer full and stop returning
2094 	 * ENOSPC.
2095 	 */
2096 	if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) ||
2097 		ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD) ||
2098 		ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR))
2099 		ceph_monc_request_next_osdmap(&osdc->client->monc);
2100 
2101 	mutex_lock(&osdc->request_mutex);
2102 	__send_queued(osdc);
2103 	mutex_unlock(&osdc->request_mutex);
2104 	up_read(&osdc->map_sem);
2105 	wake_up_all(&osdc->client->auth_wq);
2106 	return;
2107 
2108 bad:
2109 	pr_err("osdc handle_map corrupt msg\n");
2110 	ceph_msg_dump(msg);
2111 	up_write(&osdc->map_sem);
2112 }
2113 
2114 /*
2115  * watch/notify callback event infrastructure
2116  *
2117  * These callbacks are used both for watch and notify operations.
2118  */
2119 static void __release_event(struct kref *kref)
2120 {
2121 	struct ceph_osd_event *event =
2122 		container_of(kref, struct ceph_osd_event, kref);
2123 
2124 	dout("__release_event %p\n", event);
2125 	kfree(event);
2126 }
2127 
2128 static void get_event(struct ceph_osd_event *event)
2129 {
2130 	kref_get(&event->kref);
2131 }
2132 
2133 void ceph_osdc_put_event(struct ceph_osd_event *event)
2134 {
2135 	kref_put(&event->kref, __release_event);
2136 }
2137 EXPORT_SYMBOL(ceph_osdc_put_event);
2138 
2139 static void __insert_event(struct ceph_osd_client *osdc,
2140 			     struct ceph_osd_event *new)
2141 {
2142 	struct rb_node **p = &osdc->event_tree.rb_node;
2143 	struct rb_node *parent = NULL;
2144 	struct ceph_osd_event *event = NULL;
2145 
2146 	while (*p) {
2147 		parent = *p;
2148 		event = rb_entry(parent, struct ceph_osd_event, node);
2149 		if (new->cookie < event->cookie)
2150 			p = &(*p)->rb_left;
2151 		else if (new->cookie > event->cookie)
2152 			p = &(*p)->rb_right;
2153 		else
2154 			BUG();
2155 	}
2156 
2157 	rb_link_node(&new->node, parent, p);
2158 	rb_insert_color(&new->node, &osdc->event_tree);
2159 }
2160 
2161 static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
2162 					        u64 cookie)
2163 {
2164 	struct rb_node **p = &osdc->event_tree.rb_node;
2165 	struct rb_node *parent = NULL;
2166 	struct ceph_osd_event *event = NULL;
2167 
2168 	while (*p) {
2169 		parent = *p;
2170 		event = rb_entry(parent, struct ceph_osd_event, node);
2171 		if (cookie < event->cookie)
2172 			p = &(*p)->rb_left;
2173 		else if (cookie > event->cookie)
2174 			p = &(*p)->rb_right;
2175 		else
2176 			return event;
2177 	}
2178 	return NULL;
2179 }
2180 
2181 static void __remove_event(struct ceph_osd_event *event)
2182 {
2183 	struct ceph_osd_client *osdc = event->osdc;
2184 
2185 	if (!RB_EMPTY_NODE(&event->node)) {
2186 		dout("__remove_event removed %p\n", event);
2187 		rb_erase(&event->node, &osdc->event_tree);
2188 		ceph_osdc_put_event(event);
2189 	} else {
2190 		dout("__remove_event didn't remove %p\n", event);
2191 	}
2192 }
2193 
2194 int ceph_osdc_create_event(struct ceph_osd_client *osdc,
2195 			   void (*event_cb)(u64, u64, u8, void *),
2196 			   void *data, struct ceph_osd_event **pevent)
2197 {
2198 	struct ceph_osd_event *event;
2199 
2200 	event = kmalloc(sizeof(*event), GFP_NOIO);
2201 	if (!event)
2202 		return -ENOMEM;
2203 
2204 	dout("create_event %p\n", event);
2205 	event->cb = event_cb;
2206 	event->one_shot = 0;
2207 	event->data = data;
2208 	event->osdc = osdc;
2209 	INIT_LIST_HEAD(&event->osd_node);
2210 	RB_CLEAR_NODE(&event->node);
2211 	kref_init(&event->kref);   /* one ref for us */
2212 	kref_get(&event->kref);    /* one ref for the caller */
2213 
2214 	spin_lock(&osdc->event_lock);
2215 	event->cookie = ++osdc->event_count;
2216 	__insert_event(osdc, event);
2217 	spin_unlock(&osdc->event_lock);
2218 
2219 	*pevent = event;
2220 	return 0;
2221 }
2222 EXPORT_SYMBOL(ceph_osdc_create_event);
2223 
2224 void ceph_osdc_cancel_event(struct ceph_osd_event *event)
2225 {
2226 	struct ceph_osd_client *osdc = event->osdc;
2227 
2228 	dout("cancel_event %p\n", event);
2229 	spin_lock(&osdc->event_lock);
2230 	__remove_event(event);
2231 	spin_unlock(&osdc->event_lock);
2232 	ceph_osdc_put_event(event); /* caller's */
2233 }
2234 EXPORT_SYMBOL(ceph_osdc_cancel_event);
2235 
2236 
2237 static void do_event_work(struct work_struct *work)
2238 {
2239 	struct ceph_osd_event_work *event_work =
2240 		container_of(work, struct ceph_osd_event_work, work);
2241 	struct ceph_osd_event *event = event_work->event;
2242 	u64 ver = event_work->ver;
2243 	u64 notify_id = event_work->notify_id;
2244 	u8 opcode = event_work->opcode;
2245 
2246 	dout("do_event_work completing %p\n", event);
2247 	event->cb(ver, notify_id, opcode, event->data);
2248 	dout("do_event_work completed %p\n", event);
2249 	ceph_osdc_put_event(event);
2250 	kfree(event_work);
2251 }
2252 
2253 
2254 /*
2255  * Process osd watch notifications
2256  */
2257 static void handle_watch_notify(struct ceph_osd_client *osdc,
2258 				struct ceph_msg *msg)
2259 {
2260 	void *p, *end;
2261 	u8 proto_ver;
2262 	u64 cookie, ver, notify_id;
2263 	u8 opcode;
2264 	struct ceph_osd_event *event;
2265 	struct ceph_osd_event_work *event_work;
2266 
2267 	p = msg->front.iov_base;
2268 	end = p + msg->front.iov_len;
2269 
2270 	ceph_decode_8_safe(&p, end, proto_ver, bad);
2271 	ceph_decode_8_safe(&p, end, opcode, bad);
2272 	ceph_decode_64_safe(&p, end, cookie, bad);
2273 	ceph_decode_64_safe(&p, end, ver, bad);
2274 	ceph_decode_64_safe(&p, end, notify_id, bad);
2275 
2276 	spin_lock(&osdc->event_lock);
2277 	event = __find_event(osdc, cookie);
2278 	if (event) {
2279 		BUG_ON(event->one_shot);
2280 		get_event(event);
2281 	}
2282 	spin_unlock(&osdc->event_lock);
2283 	dout("handle_watch_notify cookie %lld ver %lld event %p\n",
2284 	     cookie, ver, event);
2285 	if (event) {
2286 		event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
2287 		if (!event_work) {
2288 			dout("ERROR: could not allocate event_work\n");
2289 			goto done_err;
2290 		}
2291 		INIT_WORK(&event_work->work, do_event_work);
2292 		event_work->event = event;
2293 		event_work->ver = ver;
2294 		event_work->notify_id = notify_id;
2295 		event_work->opcode = opcode;
2296 		if (!queue_work(osdc->notify_wq, &event_work->work)) {
2297 			dout("WARNING: failed to queue notify event work\n");
2298 			goto done_err;
2299 		}
2300 	}
2301 
2302 	return;
2303 
2304 done_err:
2305 	ceph_osdc_put_event(event);
2306 	return;
2307 
2308 bad:
2309 	pr_err("osdc handle_watch_notify corrupt msg\n");
2310 }
2311 
2312 /*
2313  * build new request AND message
2314  *
2315  */
2316 void ceph_osdc_build_request(struct ceph_osd_request *req, u64 off,
2317 				struct ceph_snap_context *snapc, u64 snap_id,
2318 				struct timespec *mtime)
2319 {
2320 	struct ceph_msg *msg = req->r_request;
2321 	void *p;
2322 	size_t msg_size;
2323 	int flags = req->r_flags;
2324 	u64 data_len;
2325 	unsigned int i;
2326 
2327 	req->r_snapid = snap_id;
2328 	req->r_snapc = ceph_get_snap_context(snapc);
2329 
2330 	/* encode request */
2331 	msg->hdr.version = cpu_to_le16(4);
2332 
2333 	p = msg->front.iov_base;
2334 	ceph_encode_32(&p, 1);   /* client_inc  is always 1 */
2335 	req->r_request_osdmap_epoch = p;
2336 	p += 4;
2337 	req->r_request_flags = p;
2338 	p += 4;
2339 	if (req->r_flags & CEPH_OSD_FLAG_WRITE)
2340 		ceph_encode_timespec(p, mtime);
2341 	p += sizeof(struct ceph_timespec);
2342 	req->r_request_reassert_version = p;
2343 	p += sizeof(struct ceph_eversion); /* will get filled in */
2344 
2345 	/* oloc */
2346 	ceph_encode_8(&p, 4);
2347 	ceph_encode_8(&p, 4);
2348 	ceph_encode_32(&p, 8 + 4 + 4);
2349 	req->r_request_pool = p;
2350 	p += 8;
2351 	ceph_encode_32(&p, -1);  /* preferred */
2352 	ceph_encode_32(&p, 0);   /* key len */
2353 
2354 	ceph_encode_8(&p, 1);
2355 	req->r_request_pgid = p;
2356 	p += 8 + 4;
2357 	ceph_encode_32(&p, -1);  /* preferred */
2358 
2359 	/* oid */
2360 	ceph_encode_32(&p, req->r_base_oid.name_len);
2361 	memcpy(p, req->r_base_oid.name, req->r_base_oid.name_len);
2362 	dout("oid '%.*s' len %d\n", req->r_base_oid.name_len,
2363 	     req->r_base_oid.name, req->r_base_oid.name_len);
2364 	p += req->r_base_oid.name_len;
2365 
2366 	/* ops--can imply data */
2367 	ceph_encode_16(&p, (u16)req->r_num_ops);
2368 	data_len = 0;
2369 	for (i = 0; i < req->r_num_ops; i++) {
2370 		data_len += osd_req_encode_op(req, p, i);
2371 		p += sizeof(struct ceph_osd_op);
2372 	}
2373 
2374 	/* snaps */
2375 	ceph_encode_64(&p, req->r_snapid);
2376 	ceph_encode_64(&p, req->r_snapc ? req->r_snapc->seq : 0);
2377 	ceph_encode_32(&p, req->r_snapc ? req->r_snapc->num_snaps : 0);
2378 	if (req->r_snapc) {
2379 		for (i = 0; i < snapc->num_snaps; i++) {
2380 			ceph_encode_64(&p, req->r_snapc->snaps[i]);
2381 		}
2382 	}
2383 
2384 	req->r_request_attempts = p;
2385 	p += 4;
2386 
2387 	/* data */
2388 	if (flags & CEPH_OSD_FLAG_WRITE) {
2389 		u16 data_off;
2390 
2391 		/*
2392 		 * The header "data_off" is a hint to the receiver
2393 		 * allowing it to align received data into its
2394 		 * buffers such that there's no need to re-copy
2395 		 * it before writing it to disk (direct I/O).
2396 		 */
2397 		data_off = (u16) (off & 0xffff);
2398 		req->r_request->hdr.data_off = cpu_to_le16(data_off);
2399 	}
2400 	req->r_request->hdr.data_len = cpu_to_le32(data_len);
2401 
2402 	BUG_ON(p > msg->front.iov_base + msg->front.iov_len);
2403 	msg_size = p - msg->front.iov_base;
2404 	msg->front.iov_len = msg_size;
2405 	msg->hdr.front_len = cpu_to_le32(msg_size);
2406 
2407 	dout("build_request msg_size was %d\n", (int)msg_size);
2408 }
2409 EXPORT_SYMBOL(ceph_osdc_build_request);
2410 
2411 /*
2412  * Register request, send initial attempt.
2413  */
2414 int ceph_osdc_start_request(struct ceph_osd_client *osdc,
2415 			    struct ceph_osd_request *req,
2416 			    bool nofail)
2417 {
2418 	int rc;
2419 
2420 	down_read(&osdc->map_sem);
2421 	mutex_lock(&osdc->request_mutex);
2422 
2423 	rc = __ceph_osdc_start_request(osdc, req, nofail);
2424 
2425 	mutex_unlock(&osdc->request_mutex);
2426 	up_read(&osdc->map_sem);
2427 
2428 	return rc;
2429 }
2430 EXPORT_SYMBOL(ceph_osdc_start_request);
2431 
2432 /*
2433  * wait for a request to complete
2434  */
2435 int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
2436 			   struct ceph_osd_request *req)
2437 {
2438 	int rc;
2439 
2440 	rc = wait_for_completion_interruptible(&req->r_completion);
2441 	if (rc < 0) {
2442 		mutex_lock(&osdc->request_mutex);
2443 		__cancel_request(req);
2444 		__unregister_request(osdc, req);
2445 		mutex_unlock(&osdc->request_mutex);
2446 		complete_request(req);
2447 		dout("wait_request tid %llu canceled/timed out\n", req->r_tid);
2448 		return rc;
2449 	}
2450 
2451 	dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result);
2452 	return req->r_result;
2453 }
2454 EXPORT_SYMBOL(ceph_osdc_wait_request);
2455 
2456 /*
2457  * sync - wait for all in-flight requests to flush.  avoid starvation.
2458  */
2459 void ceph_osdc_sync(struct ceph_osd_client *osdc)
2460 {
2461 	struct ceph_osd_request *req;
2462 	u64 last_tid, next_tid = 0;
2463 
2464 	mutex_lock(&osdc->request_mutex);
2465 	last_tid = osdc->last_tid;
2466 	while (1) {
2467 		req = __lookup_request_ge(osdc, next_tid);
2468 		if (!req)
2469 			break;
2470 		if (req->r_tid > last_tid)
2471 			break;
2472 
2473 		next_tid = req->r_tid + 1;
2474 		if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
2475 			continue;
2476 
2477 		ceph_osdc_get_request(req);
2478 		mutex_unlock(&osdc->request_mutex);
2479 		dout("sync waiting on tid %llu (last is %llu)\n",
2480 		     req->r_tid, last_tid);
2481 		wait_for_completion(&req->r_safe_completion);
2482 		mutex_lock(&osdc->request_mutex);
2483 		ceph_osdc_put_request(req);
2484 	}
2485 	mutex_unlock(&osdc->request_mutex);
2486 	dout("sync done (thru tid %llu)\n", last_tid);
2487 }
2488 EXPORT_SYMBOL(ceph_osdc_sync);
2489 
2490 /*
2491  * Call all pending notify callbacks - for use after a watch is
2492  * unregistered, to make sure no more callbacks for it will be invoked
2493  */
2494 void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc)
2495 {
2496 	flush_workqueue(osdc->notify_wq);
2497 }
2498 EXPORT_SYMBOL(ceph_osdc_flush_notifies);
2499 
2500 
2501 /*
2502  * init, shutdown
2503  */
2504 int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
2505 {
2506 	int err;
2507 
2508 	dout("init\n");
2509 	osdc->client = client;
2510 	osdc->osdmap = NULL;
2511 	init_rwsem(&osdc->map_sem);
2512 	init_completion(&osdc->map_waiters);
2513 	osdc->last_requested_map = 0;
2514 	mutex_init(&osdc->request_mutex);
2515 	osdc->last_tid = 0;
2516 	osdc->osds = RB_ROOT;
2517 	INIT_LIST_HEAD(&osdc->osd_lru);
2518 	osdc->requests = RB_ROOT;
2519 	INIT_LIST_HEAD(&osdc->req_lru);
2520 	INIT_LIST_HEAD(&osdc->req_unsent);
2521 	INIT_LIST_HEAD(&osdc->req_notarget);
2522 	INIT_LIST_HEAD(&osdc->req_linger);
2523 	osdc->num_requests = 0;
2524 	INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
2525 	INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
2526 	spin_lock_init(&osdc->event_lock);
2527 	osdc->event_tree = RB_ROOT;
2528 	osdc->event_count = 0;
2529 
2530 	schedule_delayed_work(&osdc->osds_timeout_work,
2531 	   round_jiffies_relative(osdc->client->options->osd_idle_ttl * HZ));
2532 
2533 	err = -ENOMEM;
2534 	osdc->req_mempool = mempool_create_kmalloc_pool(10,
2535 					sizeof(struct ceph_osd_request));
2536 	if (!osdc->req_mempool)
2537 		goto out;
2538 
2539 	err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
2540 				OSD_OP_FRONT_LEN, 10, true,
2541 				"osd_op");
2542 	if (err < 0)
2543 		goto out_mempool;
2544 	err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
2545 				OSD_OPREPLY_FRONT_LEN, 10, true,
2546 				"osd_op_reply");
2547 	if (err < 0)
2548 		goto out_msgpool;
2549 
2550 	err = -ENOMEM;
2551 	osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
2552 	if (!osdc->notify_wq)
2553 		goto out_msgpool_reply;
2554 
2555 	return 0;
2556 
2557 out_msgpool_reply:
2558 	ceph_msgpool_destroy(&osdc->msgpool_op_reply);
2559 out_msgpool:
2560 	ceph_msgpool_destroy(&osdc->msgpool_op);
2561 out_mempool:
2562 	mempool_destroy(osdc->req_mempool);
2563 out:
2564 	return err;
2565 }
2566 
2567 void ceph_osdc_stop(struct ceph_osd_client *osdc)
2568 {
2569 	flush_workqueue(osdc->notify_wq);
2570 	destroy_workqueue(osdc->notify_wq);
2571 	cancel_delayed_work_sync(&osdc->timeout_work);
2572 	cancel_delayed_work_sync(&osdc->osds_timeout_work);
2573 	if (osdc->osdmap) {
2574 		ceph_osdmap_destroy(osdc->osdmap);
2575 		osdc->osdmap = NULL;
2576 	}
2577 	remove_all_osds(osdc);
2578 	mempool_destroy(osdc->req_mempool);
2579 	ceph_msgpool_destroy(&osdc->msgpool_op);
2580 	ceph_msgpool_destroy(&osdc->msgpool_op_reply);
2581 }
2582 
2583 /*
2584  * Read some contiguous pages.  If we cross a stripe boundary, shorten
2585  * *plen.  Return number of bytes read, or error.
2586  */
2587 int ceph_osdc_readpages(struct ceph_osd_client *osdc,
2588 			struct ceph_vino vino, struct ceph_file_layout *layout,
2589 			u64 off, u64 *plen,
2590 			u32 truncate_seq, u64 truncate_size,
2591 			struct page **pages, int num_pages, int page_align)
2592 {
2593 	struct ceph_osd_request *req;
2594 	int rc = 0;
2595 
2596 	dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
2597 	     vino.snap, off, *plen);
2598 	req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 1,
2599 				    CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
2600 				    NULL, truncate_seq, truncate_size,
2601 				    false);
2602 	if (IS_ERR(req))
2603 		return PTR_ERR(req);
2604 
2605 	/* it may be a short read due to an object boundary */
2606 
2607 	osd_req_op_extent_osd_data_pages(req, 0,
2608 				pages, *plen, page_align, false, false);
2609 
2610 	dout("readpages  final extent is %llu~%llu (%llu bytes align %d)\n",
2611 	     off, *plen, *plen, page_align);
2612 
2613 	ceph_osdc_build_request(req, off, NULL, vino.snap, NULL);
2614 
2615 	rc = ceph_osdc_start_request(osdc, req, false);
2616 	if (!rc)
2617 		rc = ceph_osdc_wait_request(osdc, req);
2618 
2619 	ceph_osdc_put_request(req);
2620 	dout("readpages result %d\n", rc);
2621 	return rc;
2622 }
2623 EXPORT_SYMBOL(ceph_osdc_readpages);
2624 
2625 /*
2626  * do a synchronous write on N pages
2627  */
2628 int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
2629 			 struct ceph_file_layout *layout,
2630 			 struct ceph_snap_context *snapc,
2631 			 u64 off, u64 len,
2632 			 u32 truncate_seq, u64 truncate_size,
2633 			 struct timespec *mtime,
2634 			 struct page **pages, int num_pages)
2635 {
2636 	struct ceph_osd_request *req;
2637 	int rc = 0;
2638 	int page_align = off & ~PAGE_MASK;
2639 
2640 	BUG_ON(vino.snap != CEPH_NOSNAP);	/* snapshots aren't writeable */
2641 	req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 1,
2642 				    CEPH_OSD_OP_WRITE,
2643 				    CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
2644 				    snapc, truncate_seq, truncate_size,
2645 				    true);
2646 	if (IS_ERR(req))
2647 		return PTR_ERR(req);
2648 
2649 	/* it may be a short write due to an object boundary */
2650 	osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align,
2651 				false, false);
2652 	dout("writepages %llu~%llu (%llu bytes)\n", off, len, len);
2653 
2654 	ceph_osdc_build_request(req, off, snapc, CEPH_NOSNAP, mtime);
2655 
2656 	rc = ceph_osdc_start_request(osdc, req, true);
2657 	if (!rc)
2658 		rc = ceph_osdc_wait_request(osdc, req);
2659 
2660 	ceph_osdc_put_request(req);
2661 	if (rc == 0)
2662 		rc = len;
2663 	dout("writepages result %d\n", rc);
2664 	return rc;
2665 }
2666 EXPORT_SYMBOL(ceph_osdc_writepages);
2667 
2668 int ceph_osdc_setup(void)
2669 {
2670 	BUG_ON(ceph_osd_request_cache);
2671 	ceph_osd_request_cache = kmem_cache_create("ceph_osd_request",
2672 					sizeof (struct ceph_osd_request),
2673 					__alignof__(struct ceph_osd_request),
2674 					0, NULL);
2675 
2676 	return ceph_osd_request_cache ? 0 : -ENOMEM;
2677 }
2678 EXPORT_SYMBOL(ceph_osdc_setup);
2679 
2680 void ceph_osdc_cleanup(void)
2681 {
2682 	BUG_ON(!ceph_osd_request_cache);
2683 	kmem_cache_destroy(ceph_osd_request_cache);
2684 	ceph_osd_request_cache = NULL;
2685 }
2686 EXPORT_SYMBOL(ceph_osdc_cleanup);
2687 
2688 /*
2689  * handle incoming message
2690  */
2691 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
2692 {
2693 	struct ceph_osd *osd = con->private;
2694 	struct ceph_osd_client *osdc;
2695 	int type = le16_to_cpu(msg->hdr.type);
2696 
2697 	if (!osd)
2698 		goto out;
2699 	osdc = osd->o_osdc;
2700 
2701 	switch (type) {
2702 	case CEPH_MSG_OSD_MAP:
2703 		ceph_osdc_handle_map(osdc, msg);
2704 		break;
2705 	case CEPH_MSG_OSD_OPREPLY:
2706 		handle_reply(osdc, msg, con);
2707 		break;
2708 	case CEPH_MSG_WATCH_NOTIFY:
2709 		handle_watch_notify(osdc, msg);
2710 		break;
2711 
2712 	default:
2713 		pr_err("received unknown message type %d %s\n", type,
2714 		       ceph_msg_type_name(type));
2715 	}
2716 out:
2717 	ceph_msg_put(msg);
2718 }
2719 
2720 /*
2721  * lookup and return message for incoming reply.  set up reply message
2722  * pages.
2723  */
2724 static struct ceph_msg *get_reply(struct ceph_connection *con,
2725 				  struct ceph_msg_header *hdr,
2726 				  int *skip)
2727 {
2728 	struct ceph_osd *osd = con->private;
2729 	struct ceph_osd_client *osdc = osd->o_osdc;
2730 	struct ceph_msg *m;
2731 	struct ceph_osd_request *req;
2732 	int front_len = le32_to_cpu(hdr->front_len);
2733 	int data_len = le32_to_cpu(hdr->data_len);
2734 	u64 tid;
2735 
2736 	tid = le64_to_cpu(hdr->tid);
2737 	mutex_lock(&osdc->request_mutex);
2738 	req = __lookup_request(osdc, tid);
2739 	if (!req) {
2740 		*skip = 1;
2741 		m = NULL;
2742 		dout("get_reply unknown tid %llu from osd%d\n", tid,
2743 		     osd->o_osd);
2744 		goto out;
2745 	}
2746 
2747 	if (req->r_reply->con)
2748 		dout("%s revoking msg %p from old con %p\n", __func__,
2749 		     req->r_reply, req->r_reply->con);
2750 	ceph_msg_revoke_incoming(req->r_reply);
2751 
2752 	if (front_len > req->r_reply->front_alloc_len) {
2753 		pr_warning("get_reply front %d > preallocated %d (%u#%llu)\n",
2754 			   front_len, req->r_reply->front_alloc_len,
2755 			   (unsigned int)con->peer_name.type,
2756 			   le64_to_cpu(con->peer_name.num));
2757 		m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS,
2758 				 false);
2759 		if (!m)
2760 			goto out;
2761 		ceph_msg_put(req->r_reply);
2762 		req->r_reply = m;
2763 	}
2764 	m = ceph_msg_get(req->r_reply);
2765 
2766 	if (data_len > 0) {
2767 		struct ceph_osd_data *osd_data;
2768 
2769 		/*
2770 		 * XXX This is assuming there is only one op containing
2771 		 * XXX page data.  Probably OK for reads, but this
2772 		 * XXX ought to be done more generally.
2773 		 */
2774 		osd_data = osd_req_op_extent_osd_data(req, 0);
2775 		if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
2776 			if (osd_data->pages &&
2777 				unlikely(osd_data->length < data_len)) {
2778 
2779 				pr_warning("tid %lld reply has %d bytes "
2780 					"we had only %llu bytes ready\n",
2781 					tid, data_len, osd_data->length);
2782 				*skip = 1;
2783 				ceph_msg_put(m);
2784 				m = NULL;
2785 				goto out;
2786 			}
2787 		}
2788 	}
2789 	*skip = 0;
2790 	dout("get_reply tid %lld %p\n", tid, m);
2791 
2792 out:
2793 	mutex_unlock(&osdc->request_mutex);
2794 	return m;
2795 
2796 }
2797 
2798 static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2799 				  struct ceph_msg_header *hdr,
2800 				  int *skip)
2801 {
2802 	struct ceph_osd *osd = con->private;
2803 	int type = le16_to_cpu(hdr->type);
2804 	int front = le32_to_cpu(hdr->front_len);
2805 
2806 	*skip = 0;
2807 	switch (type) {
2808 	case CEPH_MSG_OSD_MAP:
2809 	case CEPH_MSG_WATCH_NOTIFY:
2810 		return ceph_msg_new(type, front, GFP_NOFS, false);
2811 	case CEPH_MSG_OSD_OPREPLY:
2812 		return get_reply(con, hdr, skip);
2813 	default:
2814 		pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
2815 			osd->o_osd);
2816 		*skip = 1;
2817 		return NULL;
2818 	}
2819 }
2820 
2821 /*
2822  * Wrappers to refcount containing ceph_osd struct
2823  */
2824 static struct ceph_connection *get_osd_con(struct ceph_connection *con)
2825 {
2826 	struct ceph_osd *osd = con->private;
2827 	if (get_osd(osd))
2828 		return con;
2829 	return NULL;
2830 }
2831 
2832 static void put_osd_con(struct ceph_connection *con)
2833 {
2834 	struct ceph_osd *osd = con->private;
2835 	put_osd(osd);
2836 }
2837 
2838 /*
2839  * authentication
2840  */
2841 /*
2842  * Note: returned pointer is the address of a structure that's
2843  * managed separately.  Caller must *not* attempt to free it.
2844  */
2845 static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
2846 					int *proto, int force_new)
2847 {
2848 	struct ceph_osd *o = con->private;
2849 	struct ceph_osd_client *osdc = o->o_osdc;
2850 	struct ceph_auth_client *ac = osdc->client->monc.auth;
2851 	struct ceph_auth_handshake *auth = &o->o_auth;
2852 
2853 	if (force_new && auth->authorizer) {
2854 		ceph_auth_destroy_authorizer(ac, auth->authorizer);
2855 		auth->authorizer = NULL;
2856 	}
2857 	if (!auth->authorizer) {
2858 		int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2859 						      auth);
2860 		if (ret)
2861 			return ERR_PTR(ret);
2862 	} else {
2863 		int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2864 						     auth);
2865 		if (ret)
2866 			return ERR_PTR(ret);
2867 	}
2868 	*proto = ac->protocol;
2869 
2870 	return auth;
2871 }
2872 
2873 
2874 static int verify_authorizer_reply(struct ceph_connection *con, int len)
2875 {
2876 	struct ceph_osd *o = con->private;
2877 	struct ceph_osd_client *osdc = o->o_osdc;
2878 	struct ceph_auth_client *ac = osdc->client->monc.auth;
2879 
2880 	return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer, len);
2881 }
2882 
2883 static int invalidate_authorizer(struct ceph_connection *con)
2884 {
2885 	struct ceph_osd *o = con->private;
2886 	struct ceph_osd_client *osdc = o->o_osdc;
2887 	struct ceph_auth_client *ac = osdc->client->monc.auth;
2888 
2889 	ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
2890 	return ceph_monc_validate_auth(&osdc->client->monc);
2891 }
2892 
2893 static const struct ceph_connection_operations osd_con_ops = {
2894 	.get = get_osd_con,
2895 	.put = put_osd_con,
2896 	.dispatch = dispatch,
2897 	.get_authorizer = get_authorizer,
2898 	.verify_authorizer_reply = verify_authorizer_reply,
2899 	.invalidate_authorizer = invalidate_authorizer,
2900 	.alloc_msg = alloc_msg,
2901 	.fault = osd_reset,
2902 };
2903