xref: /openbmc/linux/net/ceph/osd_client.c (revision a2fb4d78)
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_CLONERANGE:
440 	case CEPH_OSD_OP_ASSERT_SRC_VERSION:
441 	case CEPH_OSD_OP_SRC_CMPXATTR:
442 	case CEPH_OSD_OP_GETXATTR:
443 	case CEPH_OSD_OP_GETXATTRS:
444 	case CEPH_OSD_OP_CMPXATTR:
445 	case CEPH_OSD_OP_SETXATTR:
446 	case CEPH_OSD_OP_SETXATTRS:
447 	case CEPH_OSD_OP_RESETXATTRS:
448 	case CEPH_OSD_OP_RMXATTR:
449 	case CEPH_OSD_OP_PULL:
450 	case CEPH_OSD_OP_PUSH:
451 	case CEPH_OSD_OP_BALANCEREADS:
452 	case CEPH_OSD_OP_UNBALANCEREADS:
453 	case CEPH_OSD_OP_SCRUB:
454 	case CEPH_OSD_OP_SCRUB_RESERVE:
455 	case CEPH_OSD_OP_SCRUB_UNRESERVE:
456 	case CEPH_OSD_OP_SCRUB_STOP:
457 	case CEPH_OSD_OP_SCRUB_MAP:
458 	case CEPH_OSD_OP_WRLOCK:
459 	case CEPH_OSD_OP_WRUNLOCK:
460 	case CEPH_OSD_OP_RDLOCK:
461 	case CEPH_OSD_OP_RDUNLOCK:
462 	case CEPH_OSD_OP_UPLOCK:
463 	case CEPH_OSD_OP_DNLOCK:
464 	case CEPH_OSD_OP_CALL:
465 	case CEPH_OSD_OP_PGLS:
466 	case CEPH_OSD_OP_PGLS_FILTER:
467 		return true;
468 	default:
469 		return false;
470 	}
471 }
472 
473 /*
474  * This is an osd op init function for opcodes that have no data or
475  * other information associated with them.  It also serves as a
476  * common init routine for all the other init functions, below.
477  */
478 static struct ceph_osd_req_op *
479 _osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which,
480 				u16 opcode)
481 {
482 	struct ceph_osd_req_op *op;
483 
484 	BUG_ON(which >= osd_req->r_num_ops);
485 	BUG_ON(!osd_req_opcode_valid(opcode));
486 
487 	op = &osd_req->r_ops[which];
488 	memset(op, 0, sizeof (*op));
489 	op->op = opcode;
490 
491 	return op;
492 }
493 
494 void osd_req_op_init(struct ceph_osd_request *osd_req,
495 				unsigned int which, u16 opcode)
496 {
497 	(void)_osd_req_op_init(osd_req, which, opcode);
498 }
499 EXPORT_SYMBOL(osd_req_op_init);
500 
501 void osd_req_op_extent_init(struct ceph_osd_request *osd_req,
502 				unsigned int which, u16 opcode,
503 				u64 offset, u64 length,
504 				u64 truncate_size, u32 truncate_seq)
505 {
506 	struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
507 	size_t payload_len = 0;
508 
509 	BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
510 	       opcode != CEPH_OSD_OP_DELETE && opcode != CEPH_OSD_OP_ZERO &&
511 	       opcode != CEPH_OSD_OP_TRUNCATE);
512 
513 	op->extent.offset = offset;
514 	op->extent.length = length;
515 	op->extent.truncate_size = truncate_size;
516 	op->extent.truncate_seq = truncate_seq;
517 	if (opcode == CEPH_OSD_OP_WRITE)
518 		payload_len += length;
519 
520 	op->payload_len = payload_len;
521 }
522 EXPORT_SYMBOL(osd_req_op_extent_init);
523 
524 void osd_req_op_extent_update(struct ceph_osd_request *osd_req,
525 				unsigned int which, u64 length)
526 {
527 	struct ceph_osd_req_op *op;
528 	u64 previous;
529 
530 	BUG_ON(which >= osd_req->r_num_ops);
531 	op = &osd_req->r_ops[which];
532 	previous = op->extent.length;
533 
534 	if (length == previous)
535 		return;		/* Nothing to do */
536 	BUG_ON(length > previous);
537 
538 	op->extent.length = length;
539 	op->payload_len -= previous - length;
540 }
541 EXPORT_SYMBOL(osd_req_op_extent_update);
542 
543 void osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
544 			u16 opcode, const char *class, const char *method)
545 {
546 	struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
547 	struct ceph_pagelist *pagelist;
548 	size_t payload_len = 0;
549 	size_t size;
550 
551 	BUG_ON(opcode != CEPH_OSD_OP_CALL);
552 
553 	pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
554 	BUG_ON(!pagelist);
555 	ceph_pagelist_init(pagelist);
556 
557 	op->cls.class_name = class;
558 	size = strlen(class);
559 	BUG_ON(size > (size_t) U8_MAX);
560 	op->cls.class_len = size;
561 	ceph_pagelist_append(pagelist, class, size);
562 	payload_len += size;
563 
564 	op->cls.method_name = method;
565 	size = strlen(method);
566 	BUG_ON(size > (size_t) U8_MAX);
567 	op->cls.method_len = size;
568 	ceph_pagelist_append(pagelist, method, size);
569 	payload_len += size;
570 
571 	osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist);
572 
573 	op->cls.argc = 0;	/* currently unused */
574 
575 	op->payload_len = payload_len;
576 }
577 EXPORT_SYMBOL(osd_req_op_cls_init);
578 
579 void osd_req_op_watch_init(struct ceph_osd_request *osd_req,
580 				unsigned int which, u16 opcode,
581 				u64 cookie, u64 version, int flag)
582 {
583 	struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
584 
585 	BUG_ON(opcode != CEPH_OSD_OP_NOTIFY_ACK && opcode != CEPH_OSD_OP_WATCH);
586 
587 	op->watch.cookie = cookie;
588 	op->watch.ver = version;
589 	if (opcode == CEPH_OSD_OP_WATCH && flag)
590 		op->watch.flag = (u8)1;
591 }
592 EXPORT_SYMBOL(osd_req_op_watch_init);
593 
594 static void ceph_osdc_msg_data_add(struct ceph_msg *msg,
595 				struct ceph_osd_data *osd_data)
596 {
597 	u64 length = ceph_osd_data_length(osd_data);
598 
599 	if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
600 		BUG_ON(length > (u64) SIZE_MAX);
601 		if (length)
602 			ceph_msg_data_add_pages(msg, osd_data->pages,
603 					length, osd_data->alignment);
604 	} else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
605 		BUG_ON(!length);
606 		ceph_msg_data_add_pagelist(msg, osd_data->pagelist);
607 #ifdef CONFIG_BLOCK
608 	} else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
609 		ceph_msg_data_add_bio(msg, osd_data->bio, length);
610 #endif
611 	} else {
612 		BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
613 	}
614 }
615 
616 static u64 osd_req_encode_op(struct ceph_osd_request *req,
617 			      struct ceph_osd_op *dst, unsigned int which)
618 {
619 	struct ceph_osd_req_op *src;
620 	struct ceph_osd_data *osd_data;
621 	u64 request_data_len = 0;
622 	u64 data_length;
623 
624 	BUG_ON(which >= req->r_num_ops);
625 	src = &req->r_ops[which];
626 	if (WARN_ON(!osd_req_opcode_valid(src->op))) {
627 		pr_err("unrecognized osd opcode %d\n", src->op);
628 
629 		return 0;
630 	}
631 
632 	switch (src->op) {
633 	case CEPH_OSD_OP_STAT:
634 		osd_data = &src->raw_data_in;
635 		ceph_osdc_msg_data_add(req->r_reply, osd_data);
636 		break;
637 	case CEPH_OSD_OP_READ:
638 	case CEPH_OSD_OP_WRITE:
639 	case CEPH_OSD_OP_ZERO:
640 	case CEPH_OSD_OP_DELETE:
641 	case CEPH_OSD_OP_TRUNCATE:
642 		if (src->op == CEPH_OSD_OP_WRITE)
643 			request_data_len = src->extent.length;
644 		dst->extent.offset = cpu_to_le64(src->extent.offset);
645 		dst->extent.length = cpu_to_le64(src->extent.length);
646 		dst->extent.truncate_size =
647 			cpu_to_le64(src->extent.truncate_size);
648 		dst->extent.truncate_seq =
649 			cpu_to_le32(src->extent.truncate_seq);
650 		osd_data = &src->extent.osd_data;
651 		if (src->op == CEPH_OSD_OP_WRITE)
652 			ceph_osdc_msg_data_add(req->r_request, osd_data);
653 		else
654 			ceph_osdc_msg_data_add(req->r_reply, osd_data);
655 		break;
656 	case CEPH_OSD_OP_CALL:
657 		dst->cls.class_len = src->cls.class_len;
658 		dst->cls.method_len = src->cls.method_len;
659 		osd_data = &src->cls.request_info;
660 		ceph_osdc_msg_data_add(req->r_request, osd_data);
661 		BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGELIST);
662 		request_data_len = osd_data->pagelist->length;
663 
664 		osd_data = &src->cls.request_data;
665 		data_length = ceph_osd_data_length(osd_data);
666 		if (data_length) {
667 			BUG_ON(osd_data->type == CEPH_OSD_DATA_TYPE_NONE);
668 			dst->cls.indata_len = cpu_to_le32(data_length);
669 			ceph_osdc_msg_data_add(req->r_request, osd_data);
670 			src->payload_len += data_length;
671 			request_data_len += data_length;
672 		}
673 		osd_data = &src->cls.response_data;
674 		ceph_osdc_msg_data_add(req->r_reply, osd_data);
675 		break;
676 	case CEPH_OSD_OP_STARTSYNC:
677 		break;
678 	case CEPH_OSD_OP_NOTIFY_ACK:
679 	case CEPH_OSD_OP_WATCH:
680 		dst->watch.cookie = cpu_to_le64(src->watch.cookie);
681 		dst->watch.ver = cpu_to_le64(src->watch.ver);
682 		dst->watch.flag = src->watch.flag;
683 		break;
684 	default:
685 		pr_err("unsupported osd opcode %s\n",
686 			ceph_osd_op_name(src->op));
687 		WARN_ON(1);
688 
689 		return 0;
690 	}
691 	dst->op = cpu_to_le16(src->op);
692 	dst->payload_len = cpu_to_le32(src->payload_len);
693 
694 	return request_data_len;
695 }
696 
697 /*
698  * build new request AND message, calculate layout, and adjust file
699  * extent as needed.
700  *
701  * if the file was recently truncated, we include information about its
702  * old and new size so that the object can be updated appropriately.  (we
703  * avoid synchronously deleting truncated objects because it's slow.)
704  *
705  * if @do_sync, include a 'startsync' command so that the osd will flush
706  * data quickly.
707  */
708 struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
709 					       struct ceph_file_layout *layout,
710 					       struct ceph_vino vino,
711 					       u64 off, u64 *plen, int num_ops,
712 					       int opcode, int flags,
713 					       struct ceph_snap_context *snapc,
714 					       u32 truncate_seq,
715 					       u64 truncate_size,
716 					       bool use_mempool)
717 {
718 	struct ceph_osd_request *req;
719 	u64 objnum = 0;
720 	u64 objoff = 0;
721 	u64 objlen = 0;
722 	u32 object_size;
723 	u64 object_base;
724 	int r;
725 
726 	BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
727 	       opcode != CEPH_OSD_OP_DELETE && opcode != CEPH_OSD_OP_ZERO &&
728 	       opcode != CEPH_OSD_OP_TRUNCATE);
729 
730 	req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool,
731 					GFP_NOFS);
732 	if (!req)
733 		return ERR_PTR(-ENOMEM);
734 
735 	req->r_flags = flags;
736 
737 	/* calculate max write size */
738 	r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
739 	if (r < 0) {
740 		ceph_osdc_put_request(req);
741 		return ERR_PTR(r);
742 	}
743 
744 	object_size = le32_to_cpu(layout->fl_object_size);
745 	object_base = off - objoff;
746 	if (!(truncate_seq == 1 && truncate_size == -1ULL)) {
747 		if (truncate_size <= object_base) {
748 			truncate_size = 0;
749 		} else {
750 			truncate_size -= object_base;
751 			if (truncate_size > object_size)
752 				truncate_size = object_size;
753 		}
754 	}
755 
756 	osd_req_op_extent_init(req, 0, opcode, objoff, objlen,
757 				truncate_size, truncate_seq);
758 
759 	/*
760 	 * A second op in the ops array means the caller wants to
761 	 * also issue a include a 'startsync' command so that the
762 	 * osd will flush data quickly.
763 	 */
764 	if (num_ops > 1)
765 		osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC);
766 
767 	req->r_base_oloc.pool = ceph_file_layout_pg_pool(*layout);
768 
769 	snprintf(req->r_base_oid.name, sizeof(req->r_base_oid.name),
770 		 "%llx.%08llx", vino.ino, objnum);
771 	req->r_base_oid.name_len = strlen(req->r_base_oid.name);
772 
773 	return req;
774 }
775 EXPORT_SYMBOL(ceph_osdc_new_request);
776 
777 /*
778  * We keep osd requests in an rbtree, sorted by ->r_tid.
779  */
780 static void __insert_request(struct ceph_osd_client *osdc,
781 			     struct ceph_osd_request *new)
782 {
783 	struct rb_node **p = &osdc->requests.rb_node;
784 	struct rb_node *parent = NULL;
785 	struct ceph_osd_request *req = NULL;
786 
787 	while (*p) {
788 		parent = *p;
789 		req = rb_entry(parent, struct ceph_osd_request, r_node);
790 		if (new->r_tid < req->r_tid)
791 			p = &(*p)->rb_left;
792 		else if (new->r_tid > req->r_tid)
793 			p = &(*p)->rb_right;
794 		else
795 			BUG();
796 	}
797 
798 	rb_link_node(&new->r_node, parent, p);
799 	rb_insert_color(&new->r_node, &osdc->requests);
800 }
801 
802 static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc,
803 						 u64 tid)
804 {
805 	struct ceph_osd_request *req;
806 	struct rb_node *n = osdc->requests.rb_node;
807 
808 	while (n) {
809 		req = rb_entry(n, struct ceph_osd_request, r_node);
810 		if (tid < req->r_tid)
811 			n = n->rb_left;
812 		else if (tid > req->r_tid)
813 			n = n->rb_right;
814 		else
815 			return req;
816 	}
817 	return NULL;
818 }
819 
820 static struct ceph_osd_request *
821 __lookup_request_ge(struct ceph_osd_client *osdc,
822 		    u64 tid)
823 {
824 	struct ceph_osd_request *req;
825 	struct rb_node *n = osdc->requests.rb_node;
826 
827 	while (n) {
828 		req = rb_entry(n, struct ceph_osd_request, r_node);
829 		if (tid < req->r_tid) {
830 			if (!n->rb_left)
831 				return req;
832 			n = n->rb_left;
833 		} else if (tid > req->r_tid) {
834 			n = n->rb_right;
835 		} else {
836 			return req;
837 		}
838 	}
839 	return NULL;
840 }
841 
842 /*
843  * Resubmit requests pending on the given osd.
844  */
845 static void __kick_osd_requests(struct ceph_osd_client *osdc,
846 				struct ceph_osd *osd)
847 {
848 	struct ceph_osd_request *req, *nreq;
849 	LIST_HEAD(resend);
850 	int err;
851 
852 	dout("__kick_osd_requests osd%d\n", osd->o_osd);
853 	err = __reset_osd(osdc, osd);
854 	if (err)
855 		return;
856 	/*
857 	 * Build up a list of requests to resend by traversing the
858 	 * osd's list of requests.  Requests for a given object are
859 	 * sent in tid order, and that is also the order they're
860 	 * kept on this list.  Therefore all requests that are in
861 	 * flight will be found first, followed by all requests that
862 	 * have not yet been sent.  And to resend requests while
863 	 * preserving this order we will want to put any sent
864 	 * requests back on the front of the osd client's unsent
865 	 * list.
866 	 *
867 	 * So we build a separate ordered list of already-sent
868 	 * requests for the affected osd and splice it onto the
869 	 * front of the osd client's unsent list.  Once we've seen a
870 	 * request that has not yet been sent we're done.  Those
871 	 * requests are already sitting right where they belong.
872 	 */
873 	list_for_each_entry(req, &osd->o_requests, r_osd_item) {
874 		if (!req->r_sent)
875 			break;
876 		list_move_tail(&req->r_req_lru_item, &resend);
877 		dout("requeueing %p tid %llu osd%d\n", req, req->r_tid,
878 		     osd->o_osd);
879 		if (!req->r_linger)
880 			req->r_flags |= CEPH_OSD_FLAG_RETRY;
881 	}
882 	list_splice(&resend, &osdc->req_unsent);
883 
884 	/*
885 	 * Linger requests are re-registered before sending, which
886 	 * sets up a new tid for each.  We add them to the unsent
887 	 * list at the end to keep things in tid order.
888 	 */
889 	list_for_each_entry_safe(req, nreq, &osd->o_linger_requests,
890 				 r_linger_osd) {
891 		/*
892 		 * reregister request prior to unregistering linger so
893 		 * that r_osd is preserved.
894 		 */
895 		BUG_ON(!list_empty(&req->r_req_lru_item));
896 		__register_request(osdc, req);
897 		list_add_tail(&req->r_req_lru_item, &osdc->req_unsent);
898 		list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
899 		__unregister_linger_request(osdc, req);
900 		dout("requeued lingering %p tid %llu osd%d\n", req, req->r_tid,
901 		     osd->o_osd);
902 	}
903 }
904 
905 /*
906  * If the osd connection drops, we need to resubmit all requests.
907  */
908 static void osd_reset(struct ceph_connection *con)
909 {
910 	struct ceph_osd *osd = con->private;
911 	struct ceph_osd_client *osdc;
912 
913 	if (!osd)
914 		return;
915 	dout("osd_reset osd%d\n", osd->o_osd);
916 	osdc = osd->o_osdc;
917 	down_read(&osdc->map_sem);
918 	mutex_lock(&osdc->request_mutex);
919 	__kick_osd_requests(osdc, osd);
920 	__send_queued(osdc);
921 	mutex_unlock(&osdc->request_mutex);
922 	up_read(&osdc->map_sem);
923 }
924 
925 /*
926  * Track open sessions with osds.
927  */
928 static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
929 {
930 	struct ceph_osd *osd;
931 
932 	osd = kzalloc(sizeof(*osd), GFP_NOFS);
933 	if (!osd)
934 		return NULL;
935 
936 	atomic_set(&osd->o_ref, 1);
937 	osd->o_osdc = osdc;
938 	osd->o_osd = onum;
939 	RB_CLEAR_NODE(&osd->o_node);
940 	INIT_LIST_HEAD(&osd->o_requests);
941 	INIT_LIST_HEAD(&osd->o_linger_requests);
942 	INIT_LIST_HEAD(&osd->o_osd_lru);
943 	osd->o_incarnation = 1;
944 
945 	ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
946 
947 	INIT_LIST_HEAD(&osd->o_keepalive_item);
948 	return osd;
949 }
950 
951 static struct ceph_osd *get_osd(struct ceph_osd *osd)
952 {
953 	if (atomic_inc_not_zero(&osd->o_ref)) {
954 		dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
955 		     atomic_read(&osd->o_ref));
956 		return osd;
957 	} else {
958 		dout("get_osd %p FAIL\n", osd);
959 		return NULL;
960 	}
961 }
962 
963 static void put_osd(struct ceph_osd *osd)
964 {
965 	dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
966 	     atomic_read(&osd->o_ref) - 1);
967 	if (atomic_dec_and_test(&osd->o_ref) && osd->o_auth.authorizer) {
968 		struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth;
969 
970 		ceph_auth_destroy_authorizer(ac, osd->o_auth.authorizer);
971 		kfree(osd);
972 	}
973 }
974 
975 /*
976  * remove an osd from our map
977  */
978 static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
979 {
980 	dout("__remove_osd %p\n", osd);
981 	BUG_ON(!list_empty(&osd->o_requests));
982 	rb_erase(&osd->o_node, &osdc->osds);
983 	list_del_init(&osd->o_osd_lru);
984 	ceph_con_close(&osd->o_con);
985 	put_osd(osd);
986 }
987 
988 static void remove_all_osds(struct ceph_osd_client *osdc)
989 {
990 	dout("%s %p\n", __func__, osdc);
991 	mutex_lock(&osdc->request_mutex);
992 	while (!RB_EMPTY_ROOT(&osdc->osds)) {
993 		struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
994 						struct ceph_osd, o_node);
995 		__remove_osd(osdc, osd);
996 	}
997 	mutex_unlock(&osdc->request_mutex);
998 }
999 
1000 static void __move_osd_to_lru(struct ceph_osd_client *osdc,
1001 			      struct ceph_osd *osd)
1002 {
1003 	dout("__move_osd_to_lru %p\n", osd);
1004 	BUG_ON(!list_empty(&osd->o_osd_lru));
1005 	list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
1006 	osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl * HZ;
1007 }
1008 
1009 static void __remove_osd_from_lru(struct ceph_osd *osd)
1010 {
1011 	dout("__remove_osd_from_lru %p\n", osd);
1012 	if (!list_empty(&osd->o_osd_lru))
1013 		list_del_init(&osd->o_osd_lru);
1014 }
1015 
1016 static void remove_old_osds(struct ceph_osd_client *osdc)
1017 {
1018 	struct ceph_osd *osd, *nosd;
1019 
1020 	dout("__remove_old_osds %p\n", osdc);
1021 	mutex_lock(&osdc->request_mutex);
1022 	list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
1023 		if (time_before(jiffies, osd->lru_ttl))
1024 			break;
1025 		__remove_osd(osdc, osd);
1026 	}
1027 	mutex_unlock(&osdc->request_mutex);
1028 }
1029 
1030 /*
1031  * reset osd connect
1032  */
1033 static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
1034 {
1035 	struct ceph_entity_addr *peer_addr;
1036 
1037 	dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
1038 	if (list_empty(&osd->o_requests) &&
1039 	    list_empty(&osd->o_linger_requests)) {
1040 		__remove_osd(osdc, osd);
1041 
1042 		return -ENODEV;
1043 	}
1044 
1045 	peer_addr = &osdc->osdmap->osd_addr[osd->o_osd];
1046 	if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1047 			!ceph_con_opened(&osd->o_con)) {
1048 		struct ceph_osd_request *req;
1049 
1050 		dout("osd addr hasn't changed and connection never opened, "
1051 		     "letting msgr retry\n");
1052 		/* touch each r_stamp for handle_timeout()'s benfit */
1053 		list_for_each_entry(req, &osd->o_requests, r_osd_item)
1054 			req->r_stamp = jiffies;
1055 
1056 		return -EAGAIN;
1057 	}
1058 
1059 	ceph_con_close(&osd->o_con);
1060 	ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1061 	osd->o_incarnation++;
1062 
1063 	return 0;
1064 }
1065 
1066 static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new)
1067 {
1068 	struct rb_node **p = &osdc->osds.rb_node;
1069 	struct rb_node *parent = NULL;
1070 	struct ceph_osd *osd = NULL;
1071 
1072 	dout("__insert_osd %p osd%d\n", new, new->o_osd);
1073 	while (*p) {
1074 		parent = *p;
1075 		osd = rb_entry(parent, struct ceph_osd, o_node);
1076 		if (new->o_osd < osd->o_osd)
1077 			p = &(*p)->rb_left;
1078 		else if (new->o_osd > osd->o_osd)
1079 			p = &(*p)->rb_right;
1080 		else
1081 			BUG();
1082 	}
1083 
1084 	rb_link_node(&new->o_node, parent, p);
1085 	rb_insert_color(&new->o_node, &osdc->osds);
1086 }
1087 
1088 static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o)
1089 {
1090 	struct ceph_osd *osd;
1091 	struct rb_node *n = osdc->osds.rb_node;
1092 
1093 	while (n) {
1094 		osd = rb_entry(n, struct ceph_osd, o_node);
1095 		if (o < osd->o_osd)
1096 			n = n->rb_left;
1097 		else if (o > osd->o_osd)
1098 			n = n->rb_right;
1099 		else
1100 			return osd;
1101 	}
1102 	return NULL;
1103 }
1104 
1105 static void __schedule_osd_timeout(struct ceph_osd_client *osdc)
1106 {
1107 	schedule_delayed_work(&osdc->timeout_work,
1108 			osdc->client->options->osd_keepalive_timeout * HZ);
1109 }
1110 
1111 static void __cancel_osd_timeout(struct ceph_osd_client *osdc)
1112 {
1113 	cancel_delayed_work(&osdc->timeout_work);
1114 }
1115 
1116 /*
1117  * Register request, assign tid.  If this is the first request, set up
1118  * the timeout event.
1119  */
1120 static void __register_request(struct ceph_osd_client *osdc,
1121 			       struct ceph_osd_request *req)
1122 {
1123 	req->r_tid = ++osdc->last_tid;
1124 	req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
1125 	dout("__register_request %p tid %lld\n", req, req->r_tid);
1126 	__insert_request(osdc, req);
1127 	ceph_osdc_get_request(req);
1128 	osdc->num_requests++;
1129 	if (osdc->num_requests == 1) {
1130 		dout(" first request, scheduling timeout\n");
1131 		__schedule_osd_timeout(osdc);
1132 	}
1133 }
1134 
1135 /*
1136  * called under osdc->request_mutex
1137  */
1138 static void __unregister_request(struct ceph_osd_client *osdc,
1139 				 struct ceph_osd_request *req)
1140 {
1141 	if (RB_EMPTY_NODE(&req->r_node)) {
1142 		dout("__unregister_request %p tid %lld not registered\n",
1143 			req, req->r_tid);
1144 		return;
1145 	}
1146 
1147 	dout("__unregister_request %p tid %lld\n", req, req->r_tid);
1148 	rb_erase(&req->r_node, &osdc->requests);
1149 	osdc->num_requests--;
1150 
1151 	if (req->r_osd) {
1152 		/* make sure the original request isn't in flight. */
1153 		ceph_msg_revoke(req->r_request);
1154 
1155 		list_del_init(&req->r_osd_item);
1156 		if (list_empty(&req->r_osd->o_requests) &&
1157 		    list_empty(&req->r_osd->o_linger_requests)) {
1158 			dout("moving osd to %p lru\n", req->r_osd);
1159 			__move_osd_to_lru(osdc, req->r_osd);
1160 		}
1161 		if (list_empty(&req->r_linger_item))
1162 			req->r_osd = NULL;
1163 	}
1164 
1165 	list_del_init(&req->r_req_lru_item);
1166 	ceph_osdc_put_request(req);
1167 
1168 	if (osdc->num_requests == 0) {
1169 		dout(" no requests, canceling timeout\n");
1170 		__cancel_osd_timeout(osdc);
1171 	}
1172 }
1173 
1174 /*
1175  * Cancel a previously queued request message
1176  */
1177 static void __cancel_request(struct ceph_osd_request *req)
1178 {
1179 	if (req->r_sent && req->r_osd) {
1180 		ceph_msg_revoke(req->r_request);
1181 		req->r_sent = 0;
1182 	}
1183 }
1184 
1185 static void __register_linger_request(struct ceph_osd_client *osdc,
1186 				    struct ceph_osd_request *req)
1187 {
1188 	dout("__register_linger_request %p\n", req);
1189 	ceph_osdc_get_request(req);
1190 	list_add_tail(&req->r_linger_item, &osdc->req_linger);
1191 	if (req->r_osd)
1192 		list_add_tail(&req->r_linger_osd,
1193 			      &req->r_osd->o_linger_requests);
1194 }
1195 
1196 static void __unregister_linger_request(struct ceph_osd_client *osdc,
1197 					struct ceph_osd_request *req)
1198 {
1199 	dout("__unregister_linger_request %p\n", req);
1200 	list_del_init(&req->r_linger_item);
1201 	if (req->r_osd) {
1202 		list_del_init(&req->r_linger_osd);
1203 
1204 		if (list_empty(&req->r_osd->o_requests) &&
1205 		    list_empty(&req->r_osd->o_linger_requests)) {
1206 			dout("moving osd to %p lru\n", req->r_osd);
1207 			__move_osd_to_lru(osdc, req->r_osd);
1208 		}
1209 		if (list_empty(&req->r_osd_item))
1210 			req->r_osd = NULL;
1211 	}
1212 	ceph_osdc_put_request(req);
1213 }
1214 
1215 void ceph_osdc_unregister_linger_request(struct ceph_osd_client *osdc,
1216 					 struct ceph_osd_request *req)
1217 {
1218 	mutex_lock(&osdc->request_mutex);
1219 	if (req->r_linger) {
1220 		req->r_linger = 0;
1221 		__unregister_linger_request(osdc, req);
1222 	}
1223 	mutex_unlock(&osdc->request_mutex);
1224 }
1225 EXPORT_SYMBOL(ceph_osdc_unregister_linger_request);
1226 
1227 void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
1228 				  struct ceph_osd_request *req)
1229 {
1230 	if (!req->r_linger) {
1231 		dout("set_request_linger %p\n", req);
1232 		req->r_linger = 1;
1233 	}
1234 }
1235 EXPORT_SYMBOL(ceph_osdc_set_request_linger);
1236 
1237 /*
1238  * Returns whether a request should be blocked from being sent
1239  * based on the current osdmap and osd_client settings.
1240  *
1241  * Caller should hold map_sem for read.
1242  */
1243 static bool __req_should_be_paused(struct ceph_osd_client *osdc,
1244 				   struct ceph_osd_request *req)
1245 {
1246 	bool pauserd = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD);
1247 	bool pausewr = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR) ||
1248 		ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
1249 	return (req->r_flags & CEPH_OSD_FLAG_READ && pauserd) ||
1250 		(req->r_flags & CEPH_OSD_FLAG_WRITE && pausewr);
1251 }
1252 
1253 /*
1254  * Calculate mapping of a request to a PG.  Takes tiering into account.
1255  */
1256 static int __calc_request_pg(struct ceph_osdmap *osdmap,
1257 			     struct ceph_osd_request *req,
1258 			     struct ceph_pg *pg_out)
1259 {
1260 	bool need_check_tiering;
1261 
1262 	need_check_tiering = false;
1263 	if (req->r_target_oloc.pool == -1) {
1264 		req->r_target_oloc = req->r_base_oloc; /* struct */
1265 		need_check_tiering = true;
1266 	}
1267 	if (req->r_target_oid.name_len == 0) {
1268 		ceph_oid_copy(&req->r_target_oid, &req->r_base_oid);
1269 		need_check_tiering = true;
1270 	}
1271 
1272 	if (need_check_tiering &&
1273 	    (req->r_flags & CEPH_OSD_FLAG_IGNORE_OVERLAY) == 0) {
1274 		struct ceph_pg_pool_info *pi;
1275 
1276 		pi = ceph_pg_pool_by_id(osdmap, req->r_target_oloc.pool);
1277 		if (pi) {
1278 			if ((req->r_flags & CEPH_OSD_FLAG_READ) &&
1279 			    pi->read_tier >= 0)
1280 				req->r_target_oloc.pool = pi->read_tier;
1281 			if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
1282 			    pi->write_tier >= 0)
1283 				req->r_target_oloc.pool = pi->write_tier;
1284 		}
1285 		/* !pi is caught in ceph_oloc_oid_to_pg() */
1286 	}
1287 
1288 	return ceph_oloc_oid_to_pg(osdmap, &req->r_target_oloc,
1289 				   &req->r_target_oid, pg_out);
1290 }
1291 
1292 /*
1293  * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
1294  * (as needed), and set the request r_osd appropriately.  If there is
1295  * no up osd, set r_osd to NULL.  Move the request to the appropriate list
1296  * (unsent, homeless) or leave on in-flight lru.
1297  *
1298  * Return 0 if unchanged, 1 if changed, or negative on error.
1299  *
1300  * Caller should hold map_sem for read and request_mutex.
1301  */
1302 static int __map_request(struct ceph_osd_client *osdc,
1303 			 struct ceph_osd_request *req, int force_resend)
1304 {
1305 	struct ceph_pg pgid;
1306 	int acting[CEPH_PG_MAX_SIZE];
1307 	int o = -1, num = 0;
1308 	int err;
1309 	bool was_paused;
1310 
1311 	dout("map_request %p tid %lld\n", req, req->r_tid);
1312 
1313 	err = __calc_request_pg(osdc->osdmap, req, &pgid);
1314 	if (err) {
1315 		list_move(&req->r_req_lru_item, &osdc->req_notarget);
1316 		return err;
1317 	}
1318 	req->r_pgid = pgid;
1319 
1320 	err = ceph_calc_pg_acting(osdc->osdmap, pgid, acting);
1321 	if (err > 0) {
1322 		o = acting[0];
1323 		num = err;
1324 	}
1325 
1326 	was_paused = req->r_paused;
1327 	req->r_paused = __req_should_be_paused(osdc, req);
1328 	if (was_paused && !req->r_paused)
1329 		force_resend = 1;
1330 
1331 	if ((!force_resend &&
1332 	     req->r_osd && req->r_osd->o_osd == o &&
1333 	     req->r_sent >= req->r_osd->o_incarnation &&
1334 	     req->r_num_pg_osds == num &&
1335 	     memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
1336 	    (req->r_osd == NULL && o == -1) ||
1337 	    req->r_paused)
1338 		return 0;  /* no change */
1339 
1340 	dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n",
1341 	     req->r_tid, pgid.pool, pgid.seed, o,
1342 	     req->r_osd ? req->r_osd->o_osd : -1);
1343 
1344 	/* record full pg acting set */
1345 	memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
1346 	req->r_num_pg_osds = num;
1347 
1348 	if (req->r_osd) {
1349 		__cancel_request(req);
1350 		list_del_init(&req->r_osd_item);
1351 		req->r_osd = NULL;
1352 	}
1353 
1354 	req->r_osd = __lookup_osd(osdc, o);
1355 	if (!req->r_osd && o >= 0) {
1356 		err = -ENOMEM;
1357 		req->r_osd = create_osd(osdc, o);
1358 		if (!req->r_osd) {
1359 			list_move(&req->r_req_lru_item, &osdc->req_notarget);
1360 			goto out;
1361 		}
1362 
1363 		dout("map_request osd %p is osd%d\n", req->r_osd, o);
1364 		__insert_osd(osdc, req->r_osd);
1365 
1366 		ceph_con_open(&req->r_osd->o_con,
1367 			      CEPH_ENTITY_TYPE_OSD, o,
1368 			      &osdc->osdmap->osd_addr[o]);
1369 	}
1370 
1371 	if (req->r_osd) {
1372 		__remove_osd_from_lru(req->r_osd);
1373 		list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
1374 		list_move_tail(&req->r_req_lru_item, &osdc->req_unsent);
1375 	} else {
1376 		list_move_tail(&req->r_req_lru_item, &osdc->req_notarget);
1377 	}
1378 	err = 1;   /* osd or pg changed */
1379 
1380 out:
1381 	return err;
1382 }
1383 
1384 /*
1385  * caller should hold map_sem (for read) and request_mutex
1386  */
1387 static void __send_request(struct ceph_osd_client *osdc,
1388 			   struct ceph_osd_request *req)
1389 {
1390 	void *p;
1391 
1392 	dout("send_request %p tid %llu to osd%d flags %d pg %lld.%x\n",
1393 	     req, req->r_tid, req->r_osd->o_osd, req->r_flags,
1394 	     (unsigned long long)req->r_pgid.pool, req->r_pgid.seed);
1395 
1396 	/* fill in message content that changes each time we send it */
1397 	put_unaligned_le32(osdc->osdmap->epoch, req->r_request_osdmap_epoch);
1398 	put_unaligned_le32(req->r_flags, req->r_request_flags);
1399 	put_unaligned_le64(req->r_target_oloc.pool, req->r_request_pool);
1400 	p = req->r_request_pgid;
1401 	ceph_encode_64(&p, req->r_pgid.pool);
1402 	ceph_encode_32(&p, req->r_pgid.seed);
1403 	put_unaligned_le64(1, req->r_request_attempts);  /* FIXME */
1404 	memcpy(req->r_request_reassert_version, &req->r_reassert_version,
1405 	       sizeof(req->r_reassert_version));
1406 
1407 	req->r_stamp = jiffies;
1408 	list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
1409 
1410 	ceph_msg_get(req->r_request); /* send consumes a ref */
1411 
1412 	req->r_sent = req->r_osd->o_incarnation;
1413 
1414 	ceph_con_send(&req->r_osd->o_con, req->r_request);
1415 }
1416 
1417 /*
1418  * Send any requests in the queue (req_unsent).
1419  */
1420 static void __send_queued(struct ceph_osd_client *osdc)
1421 {
1422 	struct ceph_osd_request *req, *tmp;
1423 
1424 	dout("__send_queued\n");
1425 	list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item)
1426 		__send_request(osdc, req);
1427 }
1428 
1429 /*
1430  * Caller should hold map_sem for read and request_mutex.
1431  */
1432 static int __ceph_osdc_start_request(struct ceph_osd_client *osdc,
1433 				     struct ceph_osd_request *req,
1434 				     bool nofail)
1435 {
1436 	int rc;
1437 
1438 	__register_request(osdc, req);
1439 	req->r_sent = 0;
1440 	req->r_got_reply = 0;
1441 	rc = __map_request(osdc, req, 0);
1442 	if (rc < 0) {
1443 		if (nofail) {
1444 			dout("osdc_start_request failed map, "
1445 				" will retry %lld\n", req->r_tid);
1446 			rc = 0;
1447 		} else {
1448 			__unregister_request(osdc, req);
1449 		}
1450 		return rc;
1451 	}
1452 
1453 	if (req->r_osd == NULL) {
1454 		dout("send_request %p no up osds in pg\n", req);
1455 		ceph_monc_request_next_osdmap(&osdc->client->monc);
1456 	} else {
1457 		__send_queued(osdc);
1458 	}
1459 
1460 	return 0;
1461 }
1462 
1463 /*
1464  * Timeout callback, called every N seconds when 1 or more osd
1465  * requests has been active for more than N seconds.  When this
1466  * happens, we ping all OSDs with requests who have timed out to
1467  * ensure any communications channel reset is detected.  Reset the
1468  * request timeouts another N seconds in the future as we go.
1469  * Reschedule the timeout event another N seconds in future (unless
1470  * there are no open requests).
1471  */
1472 static void handle_timeout(struct work_struct *work)
1473 {
1474 	struct ceph_osd_client *osdc =
1475 		container_of(work, struct ceph_osd_client, timeout_work.work);
1476 	struct ceph_osd_request *req;
1477 	struct ceph_osd *osd;
1478 	unsigned long keepalive =
1479 		osdc->client->options->osd_keepalive_timeout * HZ;
1480 	struct list_head slow_osds;
1481 	dout("timeout\n");
1482 	down_read(&osdc->map_sem);
1483 
1484 	ceph_monc_request_next_osdmap(&osdc->client->monc);
1485 
1486 	mutex_lock(&osdc->request_mutex);
1487 
1488 	/*
1489 	 * ping osds that are a bit slow.  this ensures that if there
1490 	 * is a break in the TCP connection we will notice, and reopen
1491 	 * a connection with that osd (from the fault callback).
1492 	 */
1493 	INIT_LIST_HEAD(&slow_osds);
1494 	list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
1495 		if (time_before(jiffies, req->r_stamp + keepalive))
1496 			break;
1497 
1498 		osd = req->r_osd;
1499 		BUG_ON(!osd);
1500 		dout(" tid %llu is slow, will send keepalive on osd%d\n",
1501 		     req->r_tid, osd->o_osd);
1502 		list_move_tail(&osd->o_keepalive_item, &slow_osds);
1503 	}
1504 	while (!list_empty(&slow_osds)) {
1505 		osd = list_entry(slow_osds.next, struct ceph_osd,
1506 				 o_keepalive_item);
1507 		list_del_init(&osd->o_keepalive_item);
1508 		ceph_con_keepalive(&osd->o_con);
1509 	}
1510 
1511 	__schedule_osd_timeout(osdc);
1512 	__send_queued(osdc);
1513 	mutex_unlock(&osdc->request_mutex);
1514 	up_read(&osdc->map_sem);
1515 }
1516 
1517 static void handle_osds_timeout(struct work_struct *work)
1518 {
1519 	struct ceph_osd_client *osdc =
1520 		container_of(work, struct ceph_osd_client,
1521 			     osds_timeout_work.work);
1522 	unsigned long delay =
1523 		osdc->client->options->osd_idle_ttl * HZ >> 2;
1524 
1525 	dout("osds timeout\n");
1526 	down_read(&osdc->map_sem);
1527 	remove_old_osds(osdc);
1528 	up_read(&osdc->map_sem);
1529 
1530 	schedule_delayed_work(&osdc->osds_timeout_work,
1531 			      round_jiffies_relative(delay));
1532 }
1533 
1534 static int ceph_oloc_decode(void **p, void *end,
1535 			    struct ceph_object_locator *oloc)
1536 {
1537 	u8 struct_v, struct_cv;
1538 	u32 len;
1539 	void *struct_end;
1540 	int ret = 0;
1541 
1542 	ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
1543 	struct_v = ceph_decode_8(p);
1544 	struct_cv = ceph_decode_8(p);
1545 	if (struct_v < 3) {
1546 		pr_warn("got v %d < 3 cv %d of ceph_object_locator\n",
1547 			struct_v, struct_cv);
1548 		goto e_inval;
1549 	}
1550 	if (struct_cv > 6) {
1551 		pr_warn("got v %d cv %d > 6 of ceph_object_locator\n",
1552 			struct_v, struct_cv);
1553 		goto e_inval;
1554 	}
1555 	len = ceph_decode_32(p);
1556 	ceph_decode_need(p, end, len, e_inval);
1557 	struct_end = *p + len;
1558 
1559 	oloc->pool = ceph_decode_64(p);
1560 	*p += 4; /* skip preferred */
1561 
1562 	len = ceph_decode_32(p);
1563 	if (len > 0) {
1564 		pr_warn("ceph_object_locator::key is set\n");
1565 		goto e_inval;
1566 	}
1567 
1568 	if (struct_v >= 5) {
1569 		len = ceph_decode_32(p);
1570 		if (len > 0) {
1571 			pr_warn("ceph_object_locator::nspace is set\n");
1572 			goto e_inval;
1573 		}
1574 	}
1575 
1576 	if (struct_v >= 6) {
1577 		s64 hash = ceph_decode_64(p);
1578 		if (hash != -1) {
1579 			pr_warn("ceph_object_locator::hash is set\n");
1580 			goto e_inval;
1581 		}
1582 	}
1583 
1584 	/* skip the rest */
1585 	*p = struct_end;
1586 out:
1587 	return ret;
1588 
1589 e_inval:
1590 	ret = -EINVAL;
1591 	goto out;
1592 }
1593 
1594 static int ceph_redirect_decode(void **p, void *end,
1595 				struct ceph_request_redirect *redir)
1596 {
1597 	u8 struct_v, struct_cv;
1598 	u32 len;
1599 	void *struct_end;
1600 	int ret;
1601 
1602 	ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
1603 	struct_v = ceph_decode_8(p);
1604 	struct_cv = ceph_decode_8(p);
1605 	if (struct_cv > 1) {
1606 		pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n",
1607 			struct_v, struct_cv);
1608 		goto e_inval;
1609 	}
1610 	len = ceph_decode_32(p);
1611 	ceph_decode_need(p, end, len, e_inval);
1612 	struct_end = *p + len;
1613 
1614 	ret = ceph_oloc_decode(p, end, &redir->oloc);
1615 	if (ret)
1616 		goto out;
1617 
1618 	len = ceph_decode_32(p);
1619 	if (len > 0) {
1620 		pr_warn("ceph_request_redirect::object_name is set\n");
1621 		goto e_inval;
1622 	}
1623 
1624 	len = ceph_decode_32(p);
1625 	*p += len; /* skip osd_instructions */
1626 
1627 	/* skip the rest */
1628 	*p = struct_end;
1629 out:
1630 	return ret;
1631 
1632 e_inval:
1633 	ret = -EINVAL;
1634 	goto out;
1635 }
1636 
1637 static void complete_request(struct ceph_osd_request *req)
1638 {
1639 	complete_all(&req->r_safe_completion);  /* fsync waiter */
1640 }
1641 
1642 /*
1643  * handle osd op reply.  either call the callback if it is specified,
1644  * or do the completion to wake up the waiting thread.
1645  */
1646 static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg,
1647 			 struct ceph_connection *con)
1648 {
1649 	void *p, *end;
1650 	struct ceph_osd_request *req;
1651 	struct ceph_request_redirect redir;
1652 	u64 tid;
1653 	int object_len;
1654 	unsigned int numops;
1655 	int payload_len, flags;
1656 	s32 result;
1657 	s32 retry_attempt;
1658 	struct ceph_pg pg;
1659 	int err;
1660 	u32 reassert_epoch;
1661 	u64 reassert_version;
1662 	u32 osdmap_epoch;
1663 	int already_completed;
1664 	u32 bytes;
1665 	unsigned int i;
1666 
1667 	tid = le64_to_cpu(msg->hdr.tid);
1668 	dout("handle_reply %p tid %llu\n", msg, tid);
1669 
1670 	p = msg->front.iov_base;
1671 	end = p + msg->front.iov_len;
1672 
1673 	ceph_decode_need(&p, end, 4, bad);
1674 	object_len = ceph_decode_32(&p);
1675 	ceph_decode_need(&p, end, object_len, bad);
1676 	p += object_len;
1677 
1678 	err = ceph_decode_pgid(&p, end, &pg);
1679 	if (err)
1680 		goto bad;
1681 
1682 	ceph_decode_need(&p, end, 8 + 4 + 4 + 8 + 4, bad);
1683 	flags = ceph_decode_64(&p);
1684 	result = ceph_decode_32(&p);
1685 	reassert_epoch = ceph_decode_32(&p);
1686 	reassert_version = ceph_decode_64(&p);
1687 	osdmap_epoch = ceph_decode_32(&p);
1688 
1689 	/* lookup */
1690 	down_read(&osdc->map_sem);
1691 	mutex_lock(&osdc->request_mutex);
1692 	req = __lookup_request(osdc, tid);
1693 	if (req == NULL) {
1694 		dout("handle_reply tid %llu dne\n", tid);
1695 		goto bad_mutex;
1696 	}
1697 	ceph_osdc_get_request(req);
1698 
1699 	dout("handle_reply %p tid %llu req %p result %d\n", msg, tid,
1700 	     req, result);
1701 
1702 	ceph_decode_need(&p, end, 4, bad_put);
1703 	numops = ceph_decode_32(&p);
1704 	if (numops > CEPH_OSD_MAX_OP)
1705 		goto bad_put;
1706 	if (numops != req->r_num_ops)
1707 		goto bad_put;
1708 	payload_len = 0;
1709 	ceph_decode_need(&p, end, numops * sizeof(struct ceph_osd_op), bad_put);
1710 	for (i = 0; i < numops; i++) {
1711 		struct ceph_osd_op *op = p;
1712 		int len;
1713 
1714 		len = le32_to_cpu(op->payload_len);
1715 		req->r_reply_op_len[i] = len;
1716 		dout(" op %d has %d bytes\n", i, len);
1717 		payload_len += len;
1718 		p += sizeof(*op);
1719 	}
1720 	bytes = le32_to_cpu(msg->hdr.data_len);
1721 	if (payload_len != bytes) {
1722 		pr_warning("sum of op payload lens %d != data_len %d",
1723 			   payload_len, bytes);
1724 		goto bad_put;
1725 	}
1726 
1727 	ceph_decode_need(&p, end, 4 + numops * 4, bad_put);
1728 	retry_attempt = ceph_decode_32(&p);
1729 	for (i = 0; i < numops; i++)
1730 		req->r_reply_op_result[i] = ceph_decode_32(&p);
1731 
1732 	if (le16_to_cpu(msg->hdr.version) >= 6) {
1733 		p += 8 + 4; /* skip replay_version */
1734 		p += 8; /* skip user_version */
1735 
1736 		err = ceph_redirect_decode(&p, end, &redir);
1737 		if (err)
1738 			goto bad_put;
1739 	} else {
1740 		redir.oloc.pool = -1;
1741 	}
1742 
1743 	if (redir.oloc.pool != -1) {
1744 		dout("redirect pool %lld\n", redir.oloc.pool);
1745 
1746 		__unregister_request(osdc, req);
1747 
1748 		req->r_target_oloc = redir.oloc; /* struct */
1749 
1750 		/*
1751 		 * Start redirect requests with nofail=true.  If
1752 		 * mapping fails, request will end up on the notarget
1753 		 * list, waiting for the new osdmap (which can take
1754 		 * a while), even though the original request mapped
1755 		 * successfully.  In the future we might want to follow
1756 		 * original request's nofail setting here.
1757 		 */
1758 		err = __ceph_osdc_start_request(osdc, req, true);
1759 		BUG_ON(err);
1760 
1761 		goto out_unlock;
1762 	}
1763 
1764 	already_completed = req->r_got_reply;
1765 	if (!req->r_got_reply) {
1766 		req->r_result = result;
1767 		dout("handle_reply result %d bytes %d\n", req->r_result,
1768 		     bytes);
1769 		if (req->r_result == 0)
1770 			req->r_result = bytes;
1771 
1772 		/* in case this is a write and we need to replay, */
1773 		req->r_reassert_version.epoch = cpu_to_le32(reassert_epoch);
1774 		req->r_reassert_version.version = cpu_to_le64(reassert_version);
1775 
1776 		req->r_got_reply = 1;
1777 	} else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) {
1778 		dout("handle_reply tid %llu dup ack\n", tid);
1779 		goto out_unlock;
1780 	}
1781 
1782 	dout("handle_reply tid %llu flags %d\n", tid, flags);
1783 
1784 	if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK))
1785 		__register_linger_request(osdc, req);
1786 
1787 	/* either this is a read, or we got the safe response */
1788 	if (result < 0 ||
1789 	    (flags & CEPH_OSD_FLAG_ONDISK) ||
1790 	    ((flags & CEPH_OSD_FLAG_WRITE) == 0))
1791 		__unregister_request(osdc, req);
1792 
1793 	mutex_unlock(&osdc->request_mutex);
1794 	up_read(&osdc->map_sem);
1795 
1796 	if (!already_completed) {
1797 		if (req->r_unsafe_callback &&
1798 		    result >= 0 && !(flags & CEPH_OSD_FLAG_ONDISK))
1799 			req->r_unsafe_callback(req, true);
1800 		if (req->r_callback)
1801 			req->r_callback(req, msg);
1802 		else
1803 			complete_all(&req->r_completion);
1804 	}
1805 
1806 	if (flags & CEPH_OSD_FLAG_ONDISK) {
1807 		if (req->r_unsafe_callback && already_completed)
1808 			req->r_unsafe_callback(req, false);
1809 		complete_request(req);
1810 	}
1811 
1812 out:
1813 	dout("req=%p req->r_linger=%d\n", req, req->r_linger);
1814 	ceph_osdc_put_request(req);
1815 	return;
1816 out_unlock:
1817 	mutex_unlock(&osdc->request_mutex);
1818 	up_read(&osdc->map_sem);
1819 	goto out;
1820 
1821 bad_put:
1822 	req->r_result = -EIO;
1823 	__unregister_request(osdc, req);
1824 	if (req->r_callback)
1825 		req->r_callback(req, msg);
1826 	else
1827 		complete_all(&req->r_completion);
1828 	complete_request(req);
1829 	ceph_osdc_put_request(req);
1830 bad_mutex:
1831 	mutex_unlock(&osdc->request_mutex);
1832 	up_read(&osdc->map_sem);
1833 bad:
1834 	pr_err("corrupt osd_op_reply got %d %d\n",
1835 	       (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len));
1836 	ceph_msg_dump(msg);
1837 }
1838 
1839 static void reset_changed_osds(struct ceph_osd_client *osdc)
1840 {
1841 	struct rb_node *p, *n;
1842 
1843 	for (p = rb_first(&osdc->osds); p; p = n) {
1844 		struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
1845 
1846 		n = rb_next(p);
1847 		if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
1848 		    memcmp(&osd->o_con.peer_addr,
1849 			   ceph_osd_addr(osdc->osdmap,
1850 					 osd->o_osd),
1851 			   sizeof(struct ceph_entity_addr)) != 0)
1852 			__reset_osd(osdc, osd);
1853 	}
1854 }
1855 
1856 /*
1857  * Requeue requests whose mapping to an OSD has changed.  If requests map to
1858  * no osd, request a new map.
1859  *
1860  * Caller should hold map_sem for read.
1861  */
1862 static void kick_requests(struct ceph_osd_client *osdc, bool force_resend,
1863 			  bool force_resend_writes)
1864 {
1865 	struct ceph_osd_request *req, *nreq;
1866 	struct rb_node *p;
1867 	int needmap = 0;
1868 	int err;
1869 	bool force_resend_req;
1870 
1871 	dout("kick_requests %s %s\n", force_resend ? " (force resend)" : "",
1872 		force_resend_writes ? " (force resend writes)" : "");
1873 	mutex_lock(&osdc->request_mutex);
1874 	for (p = rb_first(&osdc->requests); p; ) {
1875 		req = rb_entry(p, struct ceph_osd_request, r_node);
1876 		p = rb_next(p);
1877 
1878 		/*
1879 		 * For linger requests that have not yet been
1880 		 * registered, move them to the linger list; they'll
1881 		 * be sent to the osd in the loop below.  Unregister
1882 		 * the request before re-registering it as a linger
1883 		 * request to ensure the __map_request() below
1884 		 * will decide it needs to be sent.
1885 		 */
1886 		if (req->r_linger && list_empty(&req->r_linger_item)) {
1887 			dout("%p tid %llu restart on osd%d\n",
1888 			     req, req->r_tid,
1889 			     req->r_osd ? req->r_osd->o_osd : -1);
1890 			ceph_osdc_get_request(req);
1891 			__unregister_request(osdc, req);
1892 			__register_linger_request(osdc, req);
1893 			ceph_osdc_put_request(req);
1894 			continue;
1895 		}
1896 
1897 		force_resend_req = force_resend ||
1898 			(force_resend_writes &&
1899 				req->r_flags & CEPH_OSD_FLAG_WRITE);
1900 		err = __map_request(osdc, req, force_resend_req);
1901 		if (err < 0)
1902 			continue;  /* error */
1903 		if (req->r_osd == NULL) {
1904 			dout("%p tid %llu maps to no osd\n", req, req->r_tid);
1905 			needmap++;  /* request a newer map */
1906 		} else if (err > 0) {
1907 			if (!req->r_linger) {
1908 				dout("%p tid %llu requeued on osd%d\n", req,
1909 				     req->r_tid,
1910 				     req->r_osd ? req->r_osd->o_osd : -1);
1911 				req->r_flags |= CEPH_OSD_FLAG_RETRY;
1912 			}
1913 		}
1914 	}
1915 
1916 	list_for_each_entry_safe(req, nreq, &osdc->req_linger,
1917 				 r_linger_item) {
1918 		dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
1919 
1920 		err = __map_request(osdc, req,
1921 				    force_resend || force_resend_writes);
1922 		dout("__map_request returned %d\n", err);
1923 		if (err == 0)
1924 			continue;  /* no change and no osd was specified */
1925 		if (err < 0)
1926 			continue;  /* hrm! */
1927 		if (req->r_osd == NULL) {
1928 			dout("tid %llu maps to no valid osd\n", req->r_tid);
1929 			needmap++;  /* request a newer map */
1930 			continue;
1931 		}
1932 
1933 		dout("kicking lingering %p tid %llu osd%d\n", req, req->r_tid,
1934 		     req->r_osd ? req->r_osd->o_osd : -1);
1935 		__register_request(osdc, req);
1936 		__unregister_linger_request(osdc, req);
1937 	}
1938 	reset_changed_osds(osdc);
1939 	mutex_unlock(&osdc->request_mutex);
1940 
1941 	if (needmap) {
1942 		dout("%d requests for down osds, need new map\n", needmap);
1943 		ceph_monc_request_next_osdmap(&osdc->client->monc);
1944 	}
1945 }
1946 
1947 
1948 /*
1949  * Process updated osd map.
1950  *
1951  * The message contains any number of incremental and full maps, normally
1952  * indicating some sort of topology change in the cluster.  Kick requests
1953  * off to different OSDs as needed.
1954  */
1955 void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1956 {
1957 	void *p, *end, *next;
1958 	u32 nr_maps, maplen;
1959 	u32 epoch;
1960 	struct ceph_osdmap *newmap = NULL, *oldmap;
1961 	int err;
1962 	struct ceph_fsid fsid;
1963 	bool was_full;
1964 
1965 	dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
1966 	p = msg->front.iov_base;
1967 	end = p + msg->front.iov_len;
1968 
1969 	/* verify fsid */
1970 	ceph_decode_need(&p, end, sizeof(fsid), bad);
1971 	ceph_decode_copy(&p, &fsid, sizeof(fsid));
1972 	if (ceph_check_fsid(osdc->client, &fsid) < 0)
1973 		return;
1974 
1975 	down_write(&osdc->map_sem);
1976 
1977 	was_full = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
1978 
1979 	/* incremental maps */
1980 	ceph_decode_32_safe(&p, end, nr_maps, bad);
1981 	dout(" %d inc maps\n", nr_maps);
1982 	while (nr_maps > 0) {
1983 		ceph_decode_need(&p, end, 2*sizeof(u32), bad);
1984 		epoch = ceph_decode_32(&p);
1985 		maplen = ceph_decode_32(&p);
1986 		ceph_decode_need(&p, end, maplen, bad);
1987 		next = p + maplen;
1988 		if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
1989 			dout("applying incremental map %u len %d\n",
1990 			     epoch, maplen);
1991 			newmap = osdmap_apply_incremental(&p, next,
1992 							  osdc->osdmap,
1993 							  &osdc->client->msgr);
1994 			if (IS_ERR(newmap)) {
1995 				err = PTR_ERR(newmap);
1996 				goto bad;
1997 			}
1998 			BUG_ON(!newmap);
1999 			if (newmap != osdc->osdmap) {
2000 				ceph_osdmap_destroy(osdc->osdmap);
2001 				osdc->osdmap = newmap;
2002 			}
2003 			was_full = was_full ||
2004 				ceph_osdmap_flag(osdc->osdmap,
2005 						 CEPH_OSDMAP_FULL);
2006 			kick_requests(osdc, 0, was_full);
2007 		} else {
2008 			dout("ignoring incremental map %u len %d\n",
2009 			     epoch, maplen);
2010 		}
2011 		p = next;
2012 		nr_maps--;
2013 	}
2014 	if (newmap)
2015 		goto done;
2016 
2017 	/* full maps */
2018 	ceph_decode_32_safe(&p, end, nr_maps, bad);
2019 	dout(" %d full maps\n", nr_maps);
2020 	while (nr_maps) {
2021 		ceph_decode_need(&p, end, 2*sizeof(u32), bad);
2022 		epoch = ceph_decode_32(&p);
2023 		maplen = ceph_decode_32(&p);
2024 		ceph_decode_need(&p, end, maplen, bad);
2025 		if (nr_maps > 1) {
2026 			dout("skipping non-latest full map %u len %d\n",
2027 			     epoch, maplen);
2028 		} else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
2029 			dout("skipping full map %u len %d, "
2030 			     "older than our %u\n", epoch, maplen,
2031 			     osdc->osdmap->epoch);
2032 		} else {
2033 			int skipped_map = 0;
2034 
2035 			dout("taking full map %u len %d\n", epoch, maplen);
2036 			newmap = osdmap_decode(&p, p+maplen);
2037 			if (IS_ERR(newmap)) {
2038 				err = PTR_ERR(newmap);
2039 				goto bad;
2040 			}
2041 			BUG_ON(!newmap);
2042 			oldmap = osdc->osdmap;
2043 			osdc->osdmap = newmap;
2044 			if (oldmap) {
2045 				if (oldmap->epoch + 1 < newmap->epoch)
2046 					skipped_map = 1;
2047 				ceph_osdmap_destroy(oldmap);
2048 			}
2049 			was_full = was_full ||
2050 				ceph_osdmap_flag(osdc->osdmap,
2051 						 CEPH_OSDMAP_FULL);
2052 			kick_requests(osdc, skipped_map, was_full);
2053 		}
2054 		p += maplen;
2055 		nr_maps--;
2056 	}
2057 
2058 	if (!osdc->osdmap)
2059 		goto bad;
2060 done:
2061 	downgrade_write(&osdc->map_sem);
2062 	ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch);
2063 
2064 	/*
2065 	 * subscribe to subsequent osdmap updates if full to ensure
2066 	 * we find out when we are no longer full and stop returning
2067 	 * ENOSPC.
2068 	 */
2069 	if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) ||
2070 		ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD) ||
2071 		ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR))
2072 		ceph_monc_request_next_osdmap(&osdc->client->monc);
2073 
2074 	mutex_lock(&osdc->request_mutex);
2075 	__send_queued(osdc);
2076 	mutex_unlock(&osdc->request_mutex);
2077 	up_read(&osdc->map_sem);
2078 	wake_up_all(&osdc->client->auth_wq);
2079 	return;
2080 
2081 bad:
2082 	pr_err("osdc handle_map corrupt msg\n");
2083 	ceph_msg_dump(msg);
2084 	up_write(&osdc->map_sem);
2085 	return;
2086 }
2087 
2088 /*
2089  * watch/notify callback event infrastructure
2090  *
2091  * These callbacks are used both for watch and notify operations.
2092  */
2093 static void __release_event(struct kref *kref)
2094 {
2095 	struct ceph_osd_event *event =
2096 		container_of(kref, struct ceph_osd_event, kref);
2097 
2098 	dout("__release_event %p\n", event);
2099 	kfree(event);
2100 }
2101 
2102 static void get_event(struct ceph_osd_event *event)
2103 {
2104 	kref_get(&event->kref);
2105 }
2106 
2107 void ceph_osdc_put_event(struct ceph_osd_event *event)
2108 {
2109 	kref_put(&event->kref, __release_event);
2110 }
2111 EXPORT_SYMBOL(ceph_osdc_put_event);
2112 
2113 static void __insert_event(struct ceph_osd_client *osdc,
2114 			     struct ceph_osd_event *new)
2115 {
2116 	struct rb_node **p = &osdc->event_tree.rb_node;
2117 	struct rb_node *parent = NULL;
2118 	struct ceph_osd_event *event = NULL;
2119 
2120 	while (*p) {
2121 		parent = *p;
2122 		event = rb_entry(parent, struct ceph_osd_event, node);
2123 		if (new->cookie < event->cookie)
2124 			p = &(*p)->rb_left;
2125 		else if (new->cookie > event->cookie)
2126 			p = &(*p)->rb_right;
2127 		else
2128 			BUG();
2129 	}
2130 
2131 	rb_link_node(&new->node, parent, p);
2132 	rb_insert_color(&new->node, &osdc->event_tree);
2133 }
2134 
2135 static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
2136 					        u64 cookie)
2137 {
2138 	struct rb_node **p = &osdc->event_tree.rb_node;
2139 	struct rb_node *parent = NULL;
2140 	struct ceph_osd_event *event = NULL;
2141 
2142 	while (*p) {
2143 		parent = *p;
2144 		event = rb_entry(parent, struct ceph_osd_event, node);
2145 		if (cookie < event->cookie)
2146 			p = &(*p)->rb_left;
2147 		else if (cookie > event->cookie)
2148 			p = &(*p)->rb_right;
2149 		else
2150 			return event;
2151 	}
2152 	return NULL;
2153 }
2154 
2155 static void __remove_event(struct ceph_osd_event *event)
2156 {
2157 	struct ceph_osd_client *osdc = event->osdc;
2158 
2159 	if (!RB_EMPTY_NODE(&event->node)) {
2160 		dout("__remove_event removed %p\n", event);
2161 		rb_erase(&event->node, &osdc->event_tree);
2162 		ceph_osdc_put_event(event);
2163 	} else {
2164 		dout("__remove_event didn't remove %p\n", event);
2165 	}
2166 }
2167 
2168 int ceph_osdc_create_event(struct ceph_osd_client *osdc,
2169 			   void (*event_cb)(u64, u64, u8, void *),
2170 			   void *data, struct ceph_osd_event **pevent)
2171 {
2172 	struct ceph_osd_event *event;
2173 
2174 	event = kmalloc(sizeof(*event), GFP_NOIO);
2175 	if (!event)
2176 		return -ENOMEM;
2177 
2178 	dout("create_event %p\n", event);
2179 	event->cb = event_cb;
2180 	event->one_shot = 0;
2181 	event->data = data;
2182 	event->osdc = osdc;
2183 	INIT_LIST_HEAD(&event->osd_node);
2184 	RB_CLEAR_NODE(&event->node);
2185 	kref_init(&event->kref);   /* one ref for us */
2186 	kref_get(&event->kref);    /* one ref for the caller */
2187 
2188 	spin_lock(&osdc->event_lock);
2189 	event->cookie = ++osdc->event_count;
2190 	__insert_event(osdc, event);
2191 	spin_unlock(&osdc->event_lock);
2192 
2193 	*pevent = event;
2194 	return 0;
2195 }
2196 EXPORT_SYMBOL(ceph_osdc_create_event);
2197 
2198 void ceph_osdc_cancel_event(struct ceph_osd_event *event)
2199 {
2200 	struct ceph_osd_client *osdc = event->osdc;
2201 
2202 	dout("cancel_event %p\n", event);
2203 	spin_lock(&osdc->event_lock);
2204 	__remove_event(event);
2205 	spin_unlock(&osdc->event_lock);
2206 	ceph_osdc_put_event(event); /* caller's */
2207 }
2208 EXPORT_SYMBOL(ceph_osdc_cancel_event);
2209 
2210 
2211 static void do_event_work(struct work_struct *work)
2212 {
2213 	struct ceph_osd_event_work *event_work =
2214 		container_of(work, struct ceph_osd_event_work, work);
2215 	struct ceph_osd_event *event = event_work->event;
2216 	u64 ver = event_work->ver;
2217 	u64 notify_id = event_work->notify_id;
2218 	u8 opcode = event_work->opcode;
2219 
2220 	dout("do_event_work completing %p\n", event);
2221 	event->cb(ver, notify_id, opcode, event->data);
2222 	dout("do_event_work completed %p\n", event);
2223 	ceph_osdc_put_event(event);
2224 	kfree(event_work);
2225 }
2226 
2227 
2228 /*
2229  * Process osd watch notifications
2230  */
2231 static void handle_watch_notify(struct ceph_osd_client *osdc,
2232 				struct ceph_msg *msg)
2233 {
2234 	void *p, *end;
2235 	u8 proto_ver;
2236 	u64 cookie, ver, notify_id;
2237 	u8 opcode;
2238 	struct ceph_osd_event *event;
2239 	struct ceph_osd_event_work *event_work;
2240 
2241 	p = msg->front.iov_base;
2242 	end = p + msg->front.iov_len;
2243 
2244 	ceph_decode_8_safe(&p, end, proto_ver, bad);
2245 	ceph_decode_8_safe(&p, end, opcode, bad);
2246 	ceph_decode_64_safe(&p, end, cookie, bad);
2247 	ceph_decode_64_safe(&p, end, ver, bad);
2248 	ceph_decode_64_safe(&p, end, notify_id, bad);
2249 
2250 	spin_lock(&osdc->event_lock);
2251 	event = __find_event(osdc, cookie);
2252 	if (event) {
2253 		BUG_ON(event->one_shot);
2254 		get_event(event);
2255 	}
2256 	spin_unlock(&osdc->event_lock);
2257 	dout("handle_watch_notify cookie %lld ver %lld event %p\n",
2258 	     cookie, ver, event);
2259 	if (event) {
2260 		event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
2261 		if (!event_work) {
2262 			dout("ERROR: could not allocate event_work\n");
2263 			goto done_err;
2264 		}
2265 		INIT_WORK(&event_work->work, do_event_work);
2266 		event_work->event = event;
2267 		event_work->ver = ver;
2268 		event_work->notify_id = notify_id;
2269 		event_work->opcode = opcode;
2270 		if (!queue_work(osdc->notify_wq, &event_work->work)) {
2271 			dout("WARNING: failed to queue notify event work\n");
2272 			goto done_err;
2273 		}
2274 	}
2275 
2276 	return;
2277 
2278 done_err:
2279 	ceph_osdc_put_event(event);
2280 	return;
2281 
2282 bad:
2283 	pr_err("osdc handle_watch_notify corrupt msg\n");
2284 	return;
2285 }
2286 
2287 /*
2288  * build new request AND message
2289  *
2290  */
2291 void ceph_osdc_build_request(struct ceph_osd_request *req, u64 off,
2292 				struct ceph_snap_context *snapc, u64 snap_id,
2293 				struct timespec *mtime)
2294 {
2295 	struct ceph_msg *msg = req->r_request;
2296 	void *p;
2297 	size_t msg_size;
2298 	int flags = req->r_flags;
2299 	u64 data_len;
2300 	unsigned int i;
2301 
2302 	req->r_snapid = snap_id;
2303 	req->r_snapc = ceph_get_snap_context(snapc);
2304 
2305 	/* encode request */
2306 	msg->hdr.version = cpu_to_le16(4);
2307 
2308 	p = msg->front.iov_base;
2309 	ceph_encode_32(&p, 1);   /* client_inc  is always 1 */
2310 	req->r_request_osdmap_epoch = p;
2311 	p += 4;
2312 	req->r_request_flags = p;
2313 	p += 4;
2314 	if (req->r_flags & CEPH_OSD_FLAG_WRITE)
2315 		ceph_encode_timespec(p, mtime);
2316 	p += sizeof(struct ceph_timespec);
2317 	req->r_request_reassert_version = p;
2318 	p += sizeof(struct ceph_eversion); /* will get filled in */
2319 
2320 	/* oloc */
2321 	ceph_encode_8(&p, 4);
2322 	ceph_encode_8(&p, 4);
2323 	ceph_encode_32(&p, 8 + 4 + 4);
2324 	req->r_request_pool = p;
2325 	p += 8;
2326 	ceph_encode_32(&p, -1);  /* preferred */
2327 	ceph_encode_32(&p, 0);   /* key len */
2328 
2329 	ceph_encode_8(&p, 1);
2330 	req->r_request_pgid = p;
2331 	p += 8 + 4;
2332 	ceph_encode_32(&p, -1);  /* preferred */
2333 
2334 	/* oid */
2335 	ceph_encode_32(&p, req->r_base_oid.name_len);
2336 	memcpy(p, req->r_base_oid.name, req->r_base_oid.name_len);
2337 	dout("oid '%.*s' len %d\n", req->r_base_oid.name_len,
2338 	     req->r_base_oid.name, req->r_base_oid.name_len);
2339 	p += req->r_base_oid.name_len;
2340 
2341 	/* ops--can imply data */
2342 	ceph_encode_16(&p, (u16)req->r_num_ops);
2343 	data_len = 0;
2344 	for (i = 0; i < req->r_num_ops; i++) {
2345 		data_len += osd_req_encode_op(req, p, i);
2346 		p += sizeof(struct ceph_osd_op);
2347 	}
2348 
2349 	/* snaps */
2350 	ceph_encode_64(&p, req->r_snapid);
2351 	ceph_encode_64(&p, req->r_snapc ? req->r_snapc->seq : 0);
2352 	ceph_encode_32(&p, req->r_snapc ? req->r_snapc->num_snaps : 0);
2353 	if (req->r_snapc) {
2354 		for (i = 0; i < snapc->num_snaps; i++) {
2355 			ceph_encode_64(&p, req->r_snapc->snaps[i]);
2356 		}
2357 	}
2358 
2359 	req->r_request_attempts = p;
2360 	p += 4;
2361 
2362 	/* data */
2363 	if (flags & CEPH_OSD_FLAG_WRITE) {
2364 		u16 data_off;
2365 
2366 		/*
2367 		 * The header "data_off" is a hint to the receiver
2368 		 * allowing it to align received data into its
2369 		 * buffers such that there's no need to re-copy
2370 		 * it before writing it to disk (direct I/O).
2371 		 */
2372 		data_off = (u16) (off & 0xffff);
2373 		req->r_request->hdr.data_off = cpu_to_le16(data_off);
2374 	}
2375 	req->r_request->hdr.data_len = cpu_to_le32(data_len);
2376 
2377 	BUG_ON(p > msg->front.iov_base + msg->front.iov_len);
2378 	msg_size = p - msg->front.iov_base;
2379 	msg->front.iov_len = msg_size;
2380 	msg->hdr.front_len = cpu_to_le32(msg_size);
2381 
2382 	dout("build_request msg_size was %d\n", (int)msg_size);
2383 }
2384 EXPORT_SYMBOL(ceph_osdc_build_request);
2385 
2386 /*
2387  * Register request, send initial attempt.
2388  */
2389 int ceph_osdc_start_request(struct ceph_osd_client *osdc,
2390 			    struct ceph_osd_request *req,
2391 			    bool nofail)
2392 {
2393 	int rc;
2394 
2395 	down_read(&osdc->map_sem);
2396 	mutex_lock(&osdc->request_mutex);
2397 
2398 	rc = __ceph_osdc_start_request(osdc, req, nofail);
2399 
2400 	mutex_unlock(&osdc->request_mutex);
2401 	up_read(&osdc->map_sem);
2402 
2403 	return rc;
2404 }
2405 EXPORT_SYMBOL(ceph_osdc_start_request);
2406 
2407 /*
2408  * wait for a request to complete
2409  */
2410 int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
2411 			   struct ceph_osd_request *req)
2412 {
2413 	int rc;
2414 
2415 	rc = wait_for_completion_interruptible(&req->r_completion);
2416 	if (rc < 0) {
2417 		mutex_lock(&osdc->request_mutex);
2418 		__cancel_request(req);
2419 		__unregister_request(osdc, req);
2420 		mutex_unlock(&osdc->request_mutex);
2421 		complete_request(req);
2422 		dout("wait_request tid %llu canceled/timed out\n", req->r_tid);
2423 		return rc;
2424 	}
2425 
2426 	dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result);
2427 	return req->r_result;
2428 }
2429 EXPORT_SYMBOL(ceph_osdc_wait_request);
2430 
2431 /*
2432  * sync - wait for all in-flight requests to flush.  avoid starvation.
2433  */
2434 void ceph_osdc_sync(struct ceph_osd_client *osdc)
2435 {
2436 	struct ceph_osd_request *req;
2437 	u64 last_tid, next_tid = 0;
2438 
2439 	mutex_lock(&osdc->request_mutex);
2440 	last_tid = osdc->last_tid;
2441 	while (1) {
2442 		req = __lookup_request_ge(osdc, next_tid);
2443 		if (!req)
2444 			break;
2445 		if (req->r_tid > last_tid)
2446 			break;
2447 
2448 		next_tid = req->r_tid + 1;
2449 		if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
2450 			continue;
2451 
2452 		ceph_osdc_get_request(req);
2453 		mutex_unlock(&osdc->request_mutex);
2454 		dout("sync waiting on tid %llu (last is %llu)\n",
2455 		     req->r_tid, last_tid);
2456 		wait_for_completion(&req->r_safe_completion);
2457 		mutex_lock(&osdc->request_mutex);
2458 		ceph_osdc_put_request(req);
2459 	}
2460 	mutex_unlock(&osdc->request_mutex);
2461 	dout("sync done (thru tid %llu)\n", last_tid);
2462 }
2463 EXPORT_SYMBOL(ceph_osdc_sync);
2464 
2465 /*
2466  * Call all pending notify callbacks - for use after a watch is
2467  * unregistered, to make sure no more callbacks for it will be invoked
2468  */
2469 extern void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc)
2470 {
2471 	flush_workqueue(osdc->notify_wq);
2472 }
2473 EXPORT_SYMBOL(ceph_osdc_flush_notifies);
2474 
2475 
2476 /*
2477  * init, shutdown
2478  */
2479 int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
2480 {
2481 	int err;
2482 
2483 	dout("init\n");
2484 	osdc->client = client;
2485 	osdc->osdmap = NULL;
2486 	init_rwsem(&osdc->map_sem);
2487 	init_completion(&osdc->map_waiters);
2488 	osdc->last_requested_map = 0;
2489 	mutex_init(&osdc->request_mutex);
2490 	osdc->last_tid = 0;
2491 	osdc->osds = RB_ROOT;
2492 	INIT_LIST_HEAD(&osdc->osd_lru);
2493 	osdc->requests = RB_ROOT;
2494 	INIT_LIST_HEAD(&osdc->req_lru);
2495 	INIT_LIST_HEAD(&osdc->req_unsent);
2496 	INIT_LIST_HEAD(&osdc->req_notarget);
2497 	INIT_LIST_HEAD(&osdc->req_linger);
2498 	osdc->num_requests = 0;
2499 	INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
2500 	INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
2501 	spin_lock_init(&osdc->event_lock);
2502 	osdc->event_tree = RB_ROOT;
2503 	osdc->event_count = 0;
2504 
2505 	schedule_delayed_work(&osdc->osds_timeout_work,
2506 	   round_jiffies_relative(osdc->client->options->osd_idle_ttl * HZ));
2507 
2508 	err = -ENOMEM;
2509 	osdc->req_mempool = mempool_create_kmalloc_pool(10,
2510 					sizeof(struct ceph_osd_request));
2511 	if (!osdc->req_mempool)
2512 		goto out;
2513 
2514 	err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
2515 				OSD_OP_FRONT_LEN, 10, true,
2516 				"osd_op");
2517 	if (err < 0)
2518 		goto out_mempool;
2519 	err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
2520 				OSD_OPREPLY_FRONT_LEN, 10, true,
2521 				"osd_op_reply");
2522 	if (err < 0)
2523 		goto out_msgpool;
2524 
2525 	err = -ENOMEM;
2526 	osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
2527 	if (!osdc->notify_wq)
2528 		goto out_msgpool_reply;
2529 
2530 	return 0;
2531 
2532 out_msgpool_reply:
2533 	ceph_msgpool_destroy(&osdc->msgpool_op_reply);
2534 out_msgpool:
2535 	ceph_msgpool_destroy(&osdc->msgpool_op);
2536 out_mempool:
2537 	mempool_destroy(osdc->req_mempool);
2538 out:
2539 	return err;
2540 }
2541 
2542 void ceph_osdc_stop(struct ceph_osd_client *osdc)
2543 {
2544 	flush_workqueue(osdc->notify_wq);
2545 	destroy_workqueue(osdc->notify_wq);
2546 	cancel_delayed_work_sync(&osdc->timeout_work);
2547 	cancel_delayed_work_sync(&osdc->osds_timeout_work);
2548 	if (osdc->osdmap) {
2549 		ceph_osdmap_destroy(osdc->osdmap);
2550 		osdc->osdmap = NULL;
2551 	}
2552 	remove_all_osds(osdc);
2553 	mempool_destroy(osdc->req_mempool);
2554 	ceph_msgpool_destroy(&osdc->msgpool_op);
2555 	ceph_msgpool_destroy(&osdc->msgpool_op_reply);
2556 }
2557 
2558 /*
2559  * Read some contiguous pages.  If we cross a stripe boundary, shorten
2560  * *plen.  Return number of bytes read, or error.
2561  */
2562 int ceph_osdc_readpages(struct ceph_osd_client *osdc,
2563 			struct ceph_vino vino, struct ceph_file_layout *layout,
2564 			u64 off, u64 *plen,
2565 			u32 truncate_seq, u64 truncate_size,
2566 			struct page **pages, int num_pages, int page_align)
2567 {
2568 	struct ceph_osd_request *req;
2569 	int rc = 0;
2570 
2571 	dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
2572 	     vino.snap, off, *plen);
2573 	req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 1,
2574 				    CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
2575 				    NULL, truncate_seq, truncate_size,
2576 				    false);
2577 	if (IS_ERR(req))
2578 		return PTR_ERR(req);
2579 
2580 	/* it may be a short read due to an object boundary */
2581 
2582 	osd_req_op_extent_osd_data_pages(req, 0,
2583 				pages, *plen, page_align, false, false);
2584 
2585 	dout("readpages  final extent is %llu~%llu (%llu bytes align %d)\n",
2586 	     off, *plen, *plen, page_align);
2587 
2588 	ceph_osdc_build_request(req, off, NULL, vino.snap, NULL);
2589 
2590 	rc = ceph_osdc_start_request(osdc, req, false);
2591 	if (!rc)
2592 		rc = ceph_osdc_wait_request(osdc, req);
2593 
2594 	ceph_osdc_put_request(req);
2595 	dout("readpages result %d\n", rc);
2596 	return rc;
2597 }
2598 EXPORT_SYMBOL(ceph_osdc_readpages);
2599 
2600 /*
2601  * do a synchronous write on N pages
2602  */
2603 int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
2604 			 struct ceph_file_layout *layout,
2605 			 struct ceph_snap_context *snapc,
2606 			 u64 off, u64 len,
2607 			 u32 truncate_seq, u64 truncate_size,
2608 			 struct timespec *mtime,
2609 			 struct page **pages, int num_pages)
2610 {
2611 	struct ceph_osd_request *req;
2612 	int rc = 0;
2613 	int page_align = off & ~PAGE_MASK;
2614 
2615 	BUG_ON(vino.snap != CEPH_NOSNAP);	/* snapshots aren't writeable */
2616 	req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 1,
2617 				    CEPH_OSD_OP_WRITE,
2618 				    CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
2619 				    snapc, truncate_seq, truncate_size,
2620 				    true);
2621 	if (IS_ERR(req))
2622 		return PTR_ERR(req);
2623 
2624 	/* it may be a short write due to an object boundary */
2625 	osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align,
2626 				false, false);
2627 	dout("writepages %llu~%llu (%llu bytes)\n", off, len, len);
2628 
2629 	ceph_osdc_build_request(req, off, snapc, CEPH_NOSNAP, mtime);
2630 
2631 	rc = ceph_osdc_start_request(osdc, req, true);
2632 	if (!rc)
2633 		rc = ceph_osdc_wait_request(osdc, req);
2634 
2635 	ceph_osdc_put_request(req);
2636 	if (rc == 0)
2637 		rc = len;
2638 	dout("writepages result %d\n", rc);
2639 	return rc;
2640 }
2641 EXPORT_SYMBOL(ceph_osdc_writepages);
2642 
2643 int ceph_osdc_setup(void)
2644 {
2645 	BUG_ON(ceph_osd_request_cache);
2646 	ceph_osd_request_cache = kmem_cache_create("ceph_osd_request",
2647 					sizeof (struct ceph_osd_request),
2648 					__alignof__(struct ceph_osd_request),
2649 					0, NULL);
2650 
2651 	return ceph_osd_request_cache ? 0 : -ENOMEM;
2652 }
2653 EXPORT_SYMBOL(ceph_osdc_setup);
2654 
2655 void ceph_osdc_cleanup(void)
2656 {
2657 	BUG_ON(!ceph_osd_request_cache);
2658 	kmem_cache_destroy(ceph_osd_request_cache);
2659 	ceph_osd_request_cache = NULL;
2660 }
2661 EXPORT_SYMBOL(ceph_osdc_cleanup);
2662 
2663 /*
2664  * handle incoming message
2665  */
2666 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
2667 {
2668 	struct ceph_osd *osd = con->private;
2669 	struct ceph_osd_client *osdc;
2670 	int type = le16_to_cpu(msg->hdr.type);
2671 
2672 	if (!osd)
2673 		goto out;
2674 	osdc = osd->o_osdc;
2675 
2676 	switch (type) {
2677 	case CEPH_MSG_OSD_MAP:
2678 		ceph_osdc_handle_map(osdc, msg);
2679 		break;
2680 	case CEPH_MSG_OSD_OPREPLY:
2681 		handle_reply(osdc, msg, con);
2682 		break;
2683 	case CEPH_MSG_WATCH_NOTIFY:
2684 		handle_watch_notify(osdc, msg);
2685 		break;
2686 
2687 	default:
2688 		pr_err("received unknown message type %d %s\n", type,
2689 		       ceph_msg_type_name(type));
2690 	}
2691 out:
2692 	ceph_msg_put(msg);
2693 }
2694 
2695 /*
2696  * lookup and return message for incoming reply.  set up reply message
2697  * pages.
2698  */
2699 static struct ceph_msg *get_reply(struct ceph_connection *con,
2700 				  struct ceph_msg_header *hdr,
2701 				  int *skip)
2702 {
2703 	struct ceph_osd *osd = con->private;
2704 	struct ceph_osd_client *osdc = osd->o_osdc;
2705 	struct ceph_msg *m;
2706 	struct ceph_osd_request *req;
2707 	int front_len = le32_to_cpu(hdr->front_len);
2708 	int data_len = le32_to_cpu(hdr->data_len);
2709 	u64 tid;
2710 
2711 	tid = le64_to_cpu(hdr->tid);
2712 	mutex_lock(&osdc->request_mutex);
2713 	req = __lookup_request(osdc, tid);
2714 	if (!req) {
2715 		*skip = 1;
2716 		m = NULL;
2717 		dout("get_reply unknown tid %llu from osd%d\n", tid,
2718 		     osd->o_osd);
2719 		goto out;
2720 	}
2721 
2722 	if (req->r_reply->con)
2723 		dout("%s revoking msg %p from old con %p\n", __func__,
2724 		     req->r_reply, req->r_reply->con);
2725 	ceph_msg_revoke_incoming(req->r_reply);
2726 
2727 	if (front_len > req->r_reply->front_alloc_len) {
2728 		pr_warning("get_reply front %d > preallocated %d (%u#%llu)\n",
2729 			   front_len, req->r_reply->front_alloc_len,
2730 			   (unsigned int)con->peer_name.type,
2731 			   le64_to_cpu(con->peer_name.num));
2732 		m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS,
2733 				 false);
2734 		if (!m)
2735 			goto out;
2736 		ceph_msg_put(req->r_reply);
2737 		req->r_reply = m;
2738 	}
2739 	m = ceph_msg_get(req->r_reply);
2740 
2741 	if (data_len > 0) {
2742 		struct ceph_osd_data *osd_data;
2743 
2744 		/*
2745 		 * XXX This is assuming there is only one op containing
2746 		 * XXX page data.  Probably OK for reads, but this
2747 		 * XXX ought to be done more generally.
2748 		 */
2749 		osd_data = osd_req_op_extent_osd_data(req, 0);
2750 		if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
2751 			if (osd_data->pages &&
2752 				unlikely(osd_data->length < data_len)) {
2753 
2754 				pr_warning("tid %lld reply has %d bytes "
2755 					"we had only %llu bytes ready\n",
2756 					tid, data_len, osd_data->length);
2757 				*skip = 1;
2758 				ceph_msg_put(m);
2759 				m = NULL;
2760 				goto out;
2761 			}
2762 		}
2763 	}
2764 	*skip = 0;
2765 	dout("get_reply tid %lld %p\n", tid, m);
2766 
2767 out:
2768 	mutex_unlock(&osdc->request_mutex);
2769 	return m;
2770 
2771 }
2772 
2773 static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2774 				  struct ceph_msg_header *hdr,
2775 				  int *skip)
2776 {
2777 	struct ceph_osd *osd = con->private;
2778 	int type = le16_to_cpu(hdr->type);
2779 	int front = le32_to_cpu(hdr->front_len);
2780 
2781 	*skip = 0;
2782 	switch (type) {
2783 	case CEPH_MSG_OSD_MAP:
2784 	case CEPH_MSG_WATCH_NOTIFY:
2785 		return ceph_msg_new(type, front, GFP_NOFS, false);
2786 	case CEPH_MSG_OSD_OPREPLY:
2787 		return get_reply(con, hdr, skip);
2788 	default:
2789 		pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
2790 			osd->o_osd);
2791 		*skip = 1;
2792 		return NULL;
2793 	}
2794 }
2795 
2796 /*
2797  * Wrappers to refcount containing ceph_osd struct
2798  */
2799 static struct ceph_connection *get_osd_con(struct ceph_connection *con)
2800 {
2801 	struct ceph_osd *osd = con->private;
2802 	if (get_osd(osd))
2803 		return con;
2804 	return NULL;
2805 }
2806 
2807 static void put_osd_con(struct ceph_connection *con)
2808 {
2809 	struct ceph_osd *osd = con->private;
2810 	put_osd(osd);
2811 }
2812 
2813 /*
2814  * authentication
2815  */
2816 /*
2817  * Note: returned pointer is the address of a structure that's
2818  * managed separately.  Caller must *not* attempt to free it.
2819  */
2820 static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
2821 					int *proto, int force_new)
2822 {
2823 	struct ceph_osd *o = con->private;
2824 	struct ceph_osd_client *osdc = o->o_osdc;
2825 	struct ceph_auth_client *ac = osdc->client->monc.auth;
2826 	struct ceph_auth_handshake *auth = &o->o_auth;
2827 
2828 	if (force_new && auth->authorizer) {
2829 		ceph_auth_destroy_authorizer(ac, auth->authorizer);
2830 		auth->authorizer = NULL;
2831 	}
2832 	if (!auth->authorizer) {
2833 		int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2834 						      auth);
2835 		if (ret)
2836 			return ERR_PTR(ret);
2837 	} else {
2838 		int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2839 						     auth);
2840 		if (ret)
2841 			return ERR_PTR(ret);
2842 	}
2843 	*proto = ac->protocol;
2844 
2845 	return auth;
2846 }
2847 
2848 
2849 static int verify_authorizer_reply(struct ceph_connection *con, int len)
2850 {
2851 	struct ceph_osd *o = con->private;
2852 	struct ceph_osd_client *osdc = o->o_osdc;
2853 	struct ceph_auth_client *ac = osdc->client->monc.auth;
2854 
2855 	return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer, len);
2856 }
2857 
2858 static int invalidate_authorizer(struct ceph_connection *con)
2859 {
2860 	struct ceph_osd *o = con->private;
2861 	struct ceph_osd_client *osdc = o->o_osdc;
2862 	struct ceph_auth_client *ac = osdc->client->monc.auth;
2863 
2864 	ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
2865 	return ceph_monc_validate_auth(&osdc->client->monc);
2866 }
2867 
2868 static const struct ceph_connection_operations osd_con_ops = {
2869 	.get = get_osd_con,
2870 	.put = put_osd_con,
2871 	.dispatch = dispatch,
2872 	.get_authorizer = get_authorizer,
2873 	.verify_authorizer_reply = verify_authorizer_reply,
2874 	.invalidate_authorizer = invalidate_authorizer,
2875 	.alloc_msg = alloc_msg,
2876 	.fault = osd_reset,
2877 };
2878