xref: /linux/net/ceph/osd_client.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
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_request(struct ceph_osd_client *osdc,
34 				 struct ceph_osd_request *req);
35 static void __unregister_linger_request(struct ceph_osd_client *osdc,
36 					struct ceph_osd_request *req);
37 static void __enqueue_request(struct ceph_osd_request *req);
38 static void __send_request(struct ceph_osd_client *osdc,
39 			   struct ceph_osd_request *req);
40 
41 /*
42  * Implement client access to distributed object storage cluster.
43  *
44  * All data objects are stored within a cluster/cloud of OSDs, or
45  * "object storage devices."  (Note that Ceph OSDs have _nothing_ to
46  * do with the T10 OSD extensions to SCSI.)  Ceph OSDs are simply
47  * remote daemons serving up and coordinating consistent and safe
48  * access to storage.
49  *
50  * Cluster membership and the mapping of data objects onto storage devices
51  * are described by the osd map.
52  *
53  * We keep track of pending OSD requests (read, write), resubmit
54  * requests to different OSDs when the cluster topology/data layout
55  * change, or retry the affected requests when the communications
56  * channel with an OSD is reset.
57  */
58 
59 /*
60  * calculate the mapping of a file extent onto an object, and fill out the
61  * request accordingly.  shorten extent as necessary if it crosses an
62  * object boundary.
63  *
64  * fill osd op in request message.
65  */
66 static int calc_layout(struct ceph_file_layout *layout, u64 off, u64 *plen,
67 			u64 *objnum, u64 *objoff, u64 *objlen)
68 {
69 	u64 orig_len = *plen;
70 	int r;
71 
72 	/* object extent? */
73 	r = ceph_calc_file_object_mapping(layout, off, orig_len, objnum,
74 					  objoff, objlen);
75 	if (r < 0)
76 		return r;
77 	if (*objlen < orig_len) {
78 		*plen = *objlen;
79 		dout(" skipping last %llu, final file extent %llu~%llu\n",
80 		     orig_len - *plen, off, *plen);
81 	}
82 
83 	dout("calc_layout objnum=%llx %llu~%llu\n", *objnum, *objoff, *objlen);
84 
85 	return 0;
86 }
87 
88 static void ceph_osd_data_init(struct ceph_osd_data *osd_data)
89 {
90 	memset(osd_data, 0, sizeof (*osd_data));
91 	osd_data->type = CEPH_OSD_DATA_TYPE_NONE;
92 }
93 
94 static void ceph_osd_data_pages_init(struct ceph_osd_data *osd_data,
95 			struct page **pages, u64 length, u32 alignment,
96 			bool pages_from_pool, bool own_pages)
97 {
98 	osd_data->type = CEPH_OSD_DATA_TYPE_PAGES;
99 	osd_data->pages = pages;
100 	osd_data->length = length;
101 	osd_data->alignment = alignment;
102 	osd_data->pages_from_pool = pages_from_pool;
103 	osd_data->own_pages = own_pages;
104 }
105 
106 static void ceph_osd_data_pagelist_init(struct ceph_osd_data *osd_data,
107 			struct ceph_pagelist *pagelist)
108 {
109 	osd_data->type = CEPH_OSD_DATA_TYPE_PAGELIST;
110 	osd_data->pagelist = pagelist;
111 }
112 
113 #ifdef CONFIG_BLOCK
114 static void ceph_osd_data_bio_init(struct ceph_osd_data *osd_data,
115 			struct bio *bio, size_t bio_length)
116 {
117 	osd_data->type = CEPH_OSD_DATA_TYPE_BIO;
118 	osd_data->bio = bio;
119 	osd_data->bio_length = bio_length;
120 }
121 #endif /* CONFIG_BLOCK */
122 
123 #define osd_req_op_data(oreq, whch, typ, fld)	\
124 	({						\
125 		BUG_ON(whch >= (oreq)->r_num_ops);	\
126 		&(oreq)->r_ops[whch].typ.fld;		\
127 	})
128 
129 static struct ceph_osd_data *
130 osd_req_op_raw_data_in(struct ceph_osd_request *osd_req, unsigned int which)
131 {
132 	BUG_ON(which >= osd_req->r_num_ops);
133 
134 	return &osd_req->r_ops[which].raw_data_in;
135 }
136 
137 struct ceph_osd_data *
138 osd_req_op_extent_osd_data(struct ceph_osd_request *osd_req,
139 			unsigned int which)
140 {
141 	return osd_req_op_data(osd_req, which, extent, osd_data);
142 }
143 EXPORT_SYMBOL(osd_req_op_extent_osd_data);
144 
145 struct ceph_osd_data *
146 osd_req_op_cls_response_data(struct ceph_osd_request *osd_req,
147 			unsigned int which)
148 {
149 	return osd_req_op_data(osd_req, which, cls, response_data);
150 }
151 EXPORT_SYMBOL(osd_req_op_cls_response_data);	/* ??? */
152 
153 void osd_req_op_raw_data_in_pages(struct ceph_osd_request *osd_req,
154 			unsigned int which, struct page **pages,
155 			u64 length, u32 alignment,
156 			bool pages_from_pool, bool own_pages)
157 {
158 	struct ceph_osd_data *osd_data;
159 
160 	osd_data = osd_req_op_raw_data_in(osd_req, which);
161 	ceph_osd_data_pages_init(osd_data, pages, length, alignment,
162 				pages_from_pool, own_pages);
163 }
164 EXPORT_SYMBOL(osd_req_op_raw_data_in_pages);
165 
166 void osd_req_op_extent_osd_data_pages(struct ceph_osd_request *osd_req,
167 			unsigned int which, struct page **pages,
168 			u64 length, u32 alignment,
169 			bool pages_from_pool, bool own_pages)
170 {
171 	struct ceph_osd_data *osd_data;
172 
173 	osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
174 	ceph_osd_data_pages_init(osd_data, pages, length, alignment,
175 				pages_from_pool, own_pages);
176 }
177 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pages);
178 
179 void osd_req_op_extent_osd_data_pagelist(struct ceph_osd_request *osd_req,
180 			unsigned int which, struct ceph_pagelist *pagelist)
181 {
182 	struct ceph_osd_data *osd_data;
183 
184 	osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
185 	ceph_osd_data_pagelist_init(osd_data, pagelist);
186 }
187 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pagelist);
188 
189 #ifdef CONFIG_BLOCK
190 void osd_req_op_extent_osd_data_bio(struct ceph_osd_request *osd_req,
191 			unsigned int which, struct bio *bio, size_t bio_length)
192 {
193 	struct ceph_osd_data *osd_data;
194 
195 	osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
196 	ceph_osd_data_bio_init(osd_data, bio, bio_length);
197 }
198 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bio);
199 #endif /* CONFIG_BLOCK */
200 
201 static void osd_req_op_cls_request_info_pagelist(
202 			struct ceph_osd_request *osd_req,
203 			unsigned int which, struct ceph_pagelist *pagelist)
204 {
205 	struct ceph_osd_data *osd_data;
206 
207 	osd_data = osd_req_op_data(osd_req, which, cls, request_info);
208 	ceph_osd_data_pagelist_init(osd_data, pagelist);
209 }
210 
211 void osd_req_op_cls_request_data_pagelist(
212 			struct ceph_osd_request *osd_req,
213 			unsigned int which, struct ceph_pagelist *pagelist)
214 {
215 	struct ceph_osd_data *osd_data;
216 
217 	osd_data = osd_req_op_data(osd_req, which, cls, request_data);
218 	ceph_osd_data_pagelist_init(osd_data, pagelist);
219 }
220 EXPORT_SYMBOL(osd_req_op_cls_request_data_pagelist);
221 
222 void osd_req_op_cls_request_data_pages(struct ceph_osd_request *osd_req,
223 			unsigned int which, struct page **pages, u64 length,
224 			u32 alignment, bool pages_from_pool, bool own_pages)
225 {
226 	struct ceph_osd_data *osd_data;
227 
228 	osd_data = osd_req_op_data(osd_req, which, cls, request_data);
229 	ceph_osd_data_pages_init(osd_data, pages, length, alignment,
230 				pages_from_pool, own_pages);
231 }
232 EXPORT_SYMBOL(osd_req_op_cls_request_data_pages);
233 
234 void osd_req_op_cls_response_data_pages(struct ceph_osd_request *osd_req,
235 			unsigned int which, struct page **pages, u64 length,
236 			u32 alignment, bool pages_from_pool, bool own_pages)
237 {
238 	struct ceph_osd_data *osd_data;
239 
240 	osd_data = osd_req_op_data(osd_req, which, cls, response_data);
241 	ceph_osd_data_pages_init(osd_data, pages, length, alignment,
242 				pages_from_pool, own_pages);
243 }
244 EXPORT_SYMBOL(osd_req_op_cls_response_data_pages);
245 
246 static u64 ceph_osd_data_length(struct ceph_osd_data *osd_data)
247 {
248 	switch (osd_data->type) {
249 	case CEPH_OSD_DATA_TYPE_NONE:
250 		return 0;
251 	case CEPH_OSD_DATA_TYPE_PAGES:
252 		return osd_data->length;
253 	case CEPH_OSD_DATA_TYPE_PAGELIST:
254 		return (u64)osd_data->pagelist->length;
255 #ifdef CONFIG_BLOCK
256 	case CEPH_OSD_DATA_TYPE_BIO:
257 		return (u64)osd_data->bio_length;
258 #endif /* CONFIG_BLOCK */
259 	default:
260 		WARN(true, "unrecognized data type %d\n", (int)osd_data->type);
261 		return 0;
262 	}
263 }
264 
265 static void ceph_osd_data_release(struct ceph_osd_data *osd_data)
266 {
267 	if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES && osd_data->own_pages) {
268 		int num_pages;
269 
270 		num_pages = calc_pages_for((u64)osd_data->alignment,
271 						(u64)osd_data->length);
272 		ceph_release_page_vector(osd_data->pages, num_pages);
273 	}
274 	ceph_osd_data_init(osd_data);
275 }
276 
277 static void osd_req_op_data_release(struct ceph_osd_request *osd_req,
278 			unsigned int which)
279 {
280 	struct ceph_osd_req_op *op;
281 
282 	BUG_ON(which >= osd_req->r_num_ops);
283 	op = &osd_req->r_ops[which];
284 
285 	switch (op->op) {
286 	case CEPH_OSD_OP_READ:
287 	case CEPH_OSD_OP_WRITE:
288 	case CEPH_OSD_OP_WRITEFULL:
289 		ceph_osd_data_release(&op->extent.osd_data);
290 		break;
291 	case CEPH_OSD_OP_CALL:
292 		ceph_osd_data_release(&op->cls.request_info);
293 		ceph_osd_data_release(&op->cls.request_data);
294 		ceph_osd_data_release(&op->cls.response_data);
295 		break;
296 	case CEPH_OSD_OP_SETXATTR:
297 	case CEPH_OSD_OP_CMPXATTR:
298 		ceph_osd_data_release(&op->xattr.osd_data);
299 		break;
300 	case CEPH_OSD_OP_STAT:
301 		ceph_osd_data_release(&op->raw_data_in);
302 		break;
303 	default:
304 		break;
305 	}
306 }
307 
308 /*
309  * requests
310  */
311 static void ceph_osdc_release_request(struct kref *kref)
312 {
313 	struct ceph_osd_request *req = container_of(kref,
314 					    struct ceph_osd_request, r_kref);
315 	unsigned int which;
316 
317 	dout("%s %p (r_request %p r_reply %p)\n", __func__, req,
318 	     req->r_request, req->r_reply);
319 	WARN_ON(!RB_EMPTY_NODE(&req->r_node));
320 	WARN_ON(!list_empty(&req->r_req_lru_item));
321 	WARN_ON(!list_empty(&req->r_osd_item));
322 	WARN_ON(!list_empty(&req->r_linger_item));
323 	WARN_ON(!list_empty(&req->r_linger_osd_item));
324 	WARN_ON(req->r_osd);
325 
326 	if (req->r_request)
327 		ceph_msg_put(req->r_request);
328 	if (req->r_reply) {
329 		ceph_msg_revoke_incoming(req->r_reply);
330 		ceph_msg_put(req->r_reply);
331 	}
332 
333 	for (which = 0; which < req->r_num_ops; which++)
334 		osd_req_op_data_release(req, which);
335 
336 	ceph_put_snap_context(req->r_snapc);
337 	if (req->r_mempool)
338 		mempool_free(req, req->r_osdc->req_mempool);
339 	else
340 		kmem_cache_free(ceph_osd_request_cache, req);
341 
342 }
343 
344 void ceph_osdc_get_request(struct ceph_osd_request *req)
345 {
346 	dout("%s %p (was %d)\n", __func__, req,
347 	     atomic_read(&req->r_kref.refcount));
348 	kref_get(&req->r_kref);
349 }
350 EXPORT_SYMBOL(ceph_osdc_get_request);
351 
352 void ceph_osdc_put_request(struct ceph_osd_request *req)
353 {
354 	dout("%s %p (was %d)\n", __func__, req,
355 	     atomic_read(&req->r_kref.refcount));
356 	kref_put(&req->r_kref, ceph_osdc_release_request);
357 }
358 EXPORT_SYMBOL(ceph_osdc_put_request);
359 
360 struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
361 					       struct ceph_snap_context *snapc,
362 					       unsigned int num_ops,
363 					       bool use_mempool,
364 					       gfp_t gfp_flags)
365 {
366 	struct ceph_osd_request *req;
367 	struct ceph_msg *msg;
368 	size_t msg_size;
369 
370 	BUILD_BUG_ON(CEPH_OSD_MAX_OP > U16_MAX);
371 	BUG_ON(num_ops > CEPH_OSD_MAX_OP);
372 
373 	msg_size = 4 + 4 + 8 + 8 + 4+8;
374 	msg_size += 2 + 4 + 8 + 4 + 4; /* oloc */
375 	msg_size += 1 + 8 + 4 + 4;     /* pg_t */
376 	msg_size += 4 + CEPH_MAX_OID_NAME_LEN; /* oid */
377 	msg_size += 2 + num_ops*sizeof(struct ceph_osd_op);
378 	msg_size += 8;  /* snapid */
379 	msg_size += 8;  /* snap_seq */
380 	msg_size += 8 * (snapc ? snapc->num_snaps : 0);  /* snaps */
381 	msg_size += 4;
382 
383 	if (use_mempool) {
384 		req = mempool_alloc(osdc->req_mempool, gfp_flags);
385 		memset(req, 0, sizeof(*req));
386 	} else {
387 		req = kmem_cache_zalloc(ceph_osd_request_cache, gfp_flags);
388 	}
389 	if (req == NULL)
390 		return NULL;
391 
392 	req->r_osdc = osdc;
393 	req->r_mempool = use_mempool;
394 	req->r_num_ops = num_ops;
395 
396 	kref_init(&req->r_kref);
397 	init_completion(&req->r_completion);
398 	init_completion(&req->r_safe_completion);
399 	RB_CLEAR_NODE(&req->r_node);
400 	INIT_LIST_HEAD(&req->r_unsafe_item);
401 	INIT_LIST_HEAD(&req->r_linger_item);
402 	INIT_LIST_HEAD(&req->r_linger_osd_item);
403 	INIT_LIST_HEAD(&req->r_req_lru_item);
404 	INIT_LIST_HEAD(&req->r_osd_item);
405 
406 	req->r_base_oloc.pool = -1;
407 	req->r_target_oloc.pool = -1;
408 
409 	/* create reply message */
410 	if (use_mempool)
411 		msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0);
412 	else
413 		msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY,
414 				   OSD_OPREPLY_FRONT_LEN, gfp_flags, true);
415 	if (!msg) {
416 		ceph_osdc_put_request(req);
417 		return NULL;
418 	}
419 	req->r_reply = msg;
420 
421 	/* create request message; allow space for oid */
422 	if (use_mempool)
423 		msg = ceph_msgpool_get(&osdc->msgpool_op, 0);
424 	else
425 		msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp_flags, true);
426 	if (!msg) {
427 		ceph_osdc_put_request(req);
428 		return NULL;
429 	}
430 
431 	memset(msg->front.iov_base, 0, msg->front.iov_len);
432 
433 	req->r_request = msg;
434 
435 	return req;
436 }
437 EXPORT_SYMBOL(ceph_osdc_alloc_request);
438 
439 static bool osd_req_opcode_valid(u16 opcode)
440 {
441 	switch (opcode) {
442 #define GENERATE_CASE(op, opcode, str)	case CEPH_OSD_OP_##op: return true;
443 __CEPH_FORALL_OSD_OPS(GENERATE_CASE)
444 #undef GENERATE_CASE
445 	default:
446 		return false;
447 	}
448 }
449 
450 /*
451  * This is an osd op init function for opcodes that have no data or
452  * other information associated with them.  It also serves as a
453  * common init routine for all the other init functions, below.
454  */
455 static struct ceph_osd_req_op *
456 _osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which,
457 		 u16 opcode, u32 flags)
458 {
459 	struct ceph_osd_req_op *op;
460 
461 	BUG_ON(which >= osd_req->r_num_ops);
462 	BUG_ON(!osd_req_opcode_valid(opcode));
463 
464 	op = &osd_req->r_ops[which];
465 	memset(op, 0, sizeof (*op));
466 	op->op = opcode;
467 	op->flags = flags;
468 
469 	return op;
470 }
471 
472 void osd_req_op_init(struct ceph_osd_request *osd_req,
473 		     unsigned int which, u16 opcode, u32 flags)
474 {
475 	(void)_osd_req_op_init(osd_req, which, opcode, flags);
476 }
477 EXPORT_SYMBOL(osd_req_op_init);
478 
479 void osd_req_op_extent_init(struct ceph_osd_request *osd_req,
480 				unsigned int which, u16 opcode,
481 				u64 offset, u64 length,
482 				u64 truncate_size, u32 truncate_seq)
483 {
484 	struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
485 						      opcode, 0);
486 	size_t payload_len = 0;
487 
488 	BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
489 	       opcode != CEPH_OSD_OP_WRITEFULL && opcode != CEPH_OSD_OP_ZERO &&
490 	       opcode != CEPH_OSD_OP_TRUNCATE);
491 
492 	op->extent.offset = offset;
493 	op->extent.length = length;
494 	op->extent.truncate_size = truncate_size;
495 	op->extent.truncate_seq = truncate_seq;
496 	if (opcode == CEPH_OSD_OP_WRITE || opcode == CEPH_OSD_OP_WRITEFULL)
497 		payload_len += length;
498 
499 	op->payload_len = payload_len;
500 }
501 EXPORT_SYMBOL(osd_req_op_extent_init);
502 
503 void osd_req_op_extent_update(struct ceph_osd_request *osd_req,
504 				unsigned int which, u64 length)
505 {
506 	struct ceph_osd_req_op *op;
507 	u64 previous;
508 
509 	BUG_ON(which >= osd_req->r_num_ops);
510 	op = &osd_req->r_ops[which];
511 	previous = op->extent.length;
512 
513 	if (length == previous)
514 		return;		/* Nothing to do */
515 	BUG_ON(length > previous);
516 
517 	op->extent.length = length;
518 	op->payload_len -= previous - length;
519 }
520 EXPORT_SYMBOL(osd_req_op_extent_update);
521 
522 void osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
523 			u16 opcode, const char *class, const char *method)
524 {
525 	struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
526 						      opcode, 0);
527 	struct ceph_pagelist *pagelist;
528 	size_t payload_len = 0;
529 	size_t size;
530 
531 	BUG_ON(opcode != CEPH_OSD_OP_CALL);
532 
533 	pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
534 	BUG_ON(!pagelist);
535 	ceph_pagelist_init(pagelist);
536 
537 	op->cls.class_name = class;
538 	size = strlen(class);
539 	BUG_ON(size > (size_t) U8_MAX);
540 	op->cls.class_len = size;
541 	ceph_pagelist_append(pagelist, class, size);
542 	payload_len += size;
543 
544 	op->cls.method_name = method;
545 	size = strlen(method);
546 	BUG_ON(size > (size_t) U8_MAX);
547 	op->cls.method_len = size;
548 	ceph_pagelist_append(pagelist, method, size);
549 	payload_len += size;
550 
551 	osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist);
552 
553 	op->cls.argc = 0;	/* currently unused */
554 
555 	op->payload_len = payload_len;
556 }
557 EXPORT_SYMBOL(osd_req_op_cls_init);
558 
559 int osd_req_op_xattr_init(struct ceph_osd_request *osd_req, unsigned int which,
560 			  u16 opcode, const char *name, const void *value,
561 			  size_t size, u8 cmp_op, u8 cmp_mode)
562 {
563 	struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
564 						      opcode, 0);
565 	struct ceph_pagelist *pagelist;
566 	size_t payload_len;
567 
568 	BUG_ON(opcode != CEPH_OSD_OP_SETXATTR && opcode != CEPH_OSD_OP_CMPXATTR);
569 
570 	pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
571 	if (!pagelist)
572 		return -ENOMEM;
573 
574 	ceph_pagelist_init(pagelist);
575 
576 	payload_len = strlen(name);
577 	op->xattr.name_len = payload_len;
578 	ceph_pagelist_append(pagelist, name, payload_len);
579 
580 	op->xattr.value_len = size;
581 	ceph_pagelist_append(pagelist, value, size);
582 	payload_len += size;
583 
584 	op->xattr.cmp_op = cmp_op;
585 	op->xattr.cmp_mode = cmp_mode;
586 
587 	ceph_osd_data_pagelist_init(&op->xattr.osd_data, pagelist);
588 	op->payload_len = payload_len;
589 	return 0;
590 }
591 EXPORT_SYMBOL(osd_req_op_xattr_init);
592 
593 void osd_req_op_watch_init(struct ceph_osd_request *osd_req,
594 				unsigned int which, u16 opcode,
595 				u64 cookie, u64 version, int flag)
596 {
597 	struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
598 						      opcode, 0);
599 
600 	BUG_ON(opcode != CEPH_OSD_OP_NOTIFY_ACK && opcode != CEPH_OSD_OP_WATCH);
601 
602 	op->watch.cookie = cookie;
603 	op->watch.ver = version;
604 	if (opcode == CEPH_OSD_OP_WATCH && flag)
605 		op->watch.flag = (u8)1;
606 }
607 EXPORT_SYMBOL(osd_req_op_watch_init);
608 
609 void osd_req_op_alloc_hint_init(struct ceph_osd_request *osd_req,
610 				unsigned int which,
611 				u64 expected_object_size,
612 				u64 expected_write_size)
613 {
614 	struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
615 						      CEPH_OSD_OP_SETALLOCHINT,
616 						      0);
617 
618 	op->alloc_hint.expected_object_size = expected_object_size;
619 	op->alloc_hint.expected_write_size = expected_write_size;
620 
621 	/*
622 	 * CEPH_OSD_OP_SETALLOCHINT op is advisory and therefore deemed
623 	 * not worth a feature bit.  Set FAILOK per-op flag to make
624 	 * sure older osds don't trip over an unsupported opcode.
625 	 */
626 	op->flags |= CEPH_OSD_OP_FLAG_FAILOK;
627 }
628 EXPORT_SYMBOL(osd_req_op_alloc_hint_init);
629 
630 static void ceph_osdc_msg_data_add(struct ceph_msg *msg,
631 				struct ceph_osd_data *osd_data)
632 {
633 	u64 length = ceph_osd_data_length(osd_data);
634 
635 	if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
636 		BUG_ON(length > (u64) SIZE_MAX);
637 		if (length)
638 			ceph_msg_data_add_pages(msg, osd_data->pages,
639 					length, osd_data->alignment);
640 	} else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
641 		BUG_ON(!length);
642 		ceph_msg_data_add_pagelist(msg, osd_data->pagelist);
643 #ifdef CONFIG_BLOCK
644 	} else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
645 		ceph_msg_data_add_bio(msg, osd_data->bio, length);
646 #endif
647 	} else {
648 		BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
649 	}
650 }
651 
652 static u64 osd_req_encode_op(struct ceph_osd_request *req,
653 			      struct ceph_osd_op *dst, unsigned int which)
654 {
655 	struct ceph_osd_req_op *src;
656 	struct ceph_osd_data *osd_data;
657 	u64 request_data_len = 0;
658 	u64 data_length;
659 
660 	BUG_ON(which >= req->r_num_ops);
661 	src = &req->r_ops[which];
662 	if (WARN_ON(!osd_req_opcode_valid(src->op))) {
663 		pr_err("unrecognized osd opcode %d\n", src->op);
664 
665 		return 0;
666 	}
667 
668 	switch (src->op) {
669 	case CEPH_OSD_OP_STAT:
670 		osd_data = &src->raw_data_in;
671 		ceph_osdc_msg_data_add(req->r_reply, osd_data);
672 		break;
673 	case CEPH_OSD_OP_READ:
674 	case CEPH_OSD_OP_WRITE:
675 	case CEPH_OSD_OP_WRITEFULL:
676 	case CEPH_OSD_OP_ZERO:
677 	case CEPH_OSD_OP_TRUNCATE:
678 		if (src->op == CEPH_OSD_OP_WRITE ||
679 		    src->op == CEPH_OSD_OP_WRITEFULL)
680 			request_data_len = src->extent.length;
681 		dst->extent.offset = cpu_to_le64(src->extent.offset);
682 		dst->extent.length = cpu_to_le64(src->extent.length);
683 		dst->extent.truncate_size =
684 			cpu_to_le64(src->extent.truncate_size);
685 		dst->extent.truncate_seq =
686 			cpu_to_le32(src->extent.truncate_seq);
687 		osd_data = &src->extent.osd_data;
688 		if (src->op == CEPH_OSD_OP_WRITE ||
689 		    src->op == CEPH_OSD_OP_WRITEFULL)
690 			ceph_osdc_msg_data_add(req->r_request, osd_data);
691 		else
692 			ceph_osdc_msg_data_add(req->r_reply, osd_data);
693 		break;
694 	case CEPH_OSD_OP_CALL:
695 		dst->cls.class_len = src->cls.class_len;
696 		dst->cls.method_len = src->cls.method_len;
697 		osd_data = &src->cls.request_info;
698 		ceph_osdc_msg_data_add(req->r_request, osd_data);
699 		BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGELIST);
700 		request_data_len = osd_data->pagelist->length;
701 
702 		osd_data = &src->cls.request_data;
703 		data_length = ceph_osd_data_length(osd_data);
704 		if (data_length) {
705 			BUG_ON(osd_data->type == CEPH_OSD_DATA_TYPE_NONE);
706 			dst->cls.indata_len = cpu_to_le32(data_length);
707 			ceph_osdc_msg_data_add(req->r_request, osd_data);
708 			src->payload_len += data_length;
709 			request_data_len += data_length;
710 		}
711 		osd_data = &src->cls.response_data;
712 		ceph_osdc_msg_data_add(req->r_reply, osd_data);
713 		break;
714 	case CEPH_OSD_OP_STARTSYNC:
715 		break;
716 	case CEPH_OSD_OP_NOTIFY_ACK:
717 	case CEPH_OSD_OP_WATCH:
718 		dst->watch.cookie = cpu_to_le64(src->watch.cookie);
719 		dst->watch.ver = cpu_to_le64(src->watch.ver);
720 		dst->watch.flag = src->watch.flag;
721 		break;
722 	case CEPH_OSD_OP_SETALLOCHINT:
723 		dst->alloc_hint.expected_object_size =
724 		    cpu_to_le64(src->alloc_hint.expected_object_size);
725 		dst->alloc_hint.expected_write_size =
726 		    cpu_to_le64(src->alloc_hint.expected_write_size);
727 		break;
728 	case CEPH_OSD_OP_SETXATTR:
729 	case CEPH_OSD_OP_CMPXATTR:
730 		dst->xattr.name_len = cpu_to_le32(src->xattr.name_len);
731 		dst->xattr.value_len = cpu_to_le32(src->xattr.value_len);
732 		dst->xattr.cmp_op = src->xattr.cmp_op;
733 		dst->xattr.cmp_mode = src->xattr.cmp_mode;
734 		osd_data = &src->xattr.osd_data;
735 		ceph_osdc_msg_data_add(req->r_request, osd_data);
736 		request_data_len = osd_data->pagelist->length;
737 		break;
738 	case CEPH_OSD_OP_CREATE:
739 	case CEPH_OSD_OP_DELETE:
740 		break;
741 	default:
742 		pr_err("unsupported osd opcode %s\n",
743 			ceph_osd_op_name(src->op));
744 		WARN_ON(1);
745 
746 		return 0;
747 	}
748 
749 	dst->op = cpu_to_le16(src->op);
750 	dst->flags = cpu_to_le32(src->flags);
751 	dst->payload_len = cpu_to_le32(src->payload_len);
752 
753 	return request_data_len;
754 }
755 
756 /*
757  * build new request AND message, calculate layout, and adjust file
758  * extent as needed.
759  *
760  * if the file was recently truncated, we include information about its
761  * old and new size so that the object can be updated appropriately.  (we
762  * avoid synchronously deleting truncated objects because it's slow.)
763  *
764  * if @do_sync, include a 'startsync' command so that the osd will flush
765  * data quickly.
766  */
767 struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
768 					       struct ceph_file_layout *layout,
769 					       struct ceph_vino vino,
770 					       u64 off, u64 *plen,
771 					       unsigned int which, int num_ops,
772 					       int opcode, int flags,
773 					       struct ceph_snap_context *snapc,
774 					       u32 truncate_seq,
775 					       u64 truncate_size,
776 					       bool use_mempool)
777 {
778 	struct ceph_osd_request *req;
779 	u64 objnum = 0;
780 	u64 objoff = 0;
781 	u64 objlen = 0;
782 	int r;
783 
784 	BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
785 	       opcode != CEPH_OSD_OP_ZERO && opcode != CEPH_OSD_OP_TRUNCATE &&
786 	       opcode != CEPH_OSD_OP_CREATE && opcode != CEPH_OSD_OP_DELETE);
787 
788 	req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool,
789 					GFP_NOFS);
790 	if (!req)
791 		return ERR_PTR(-ENOMEM);
792 
793 	req->r_flags = flags;
794 
795 	/* calculate max write size */
796 	r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
797 	if (r < 0) {
798 		ceph_osdc_put_request(req);
799 		return ERR_PTR(r);
800 	}
801 
802 	if (opcode == CEPH_OSD_OP_CREATE || opcode == CEPH_OSD_OP_DELETE) {
803 		osd_req_op_init(req, which, opcode, 0);
804 	} else {
805 		u32 object_size = le32_to_cpu(layout->fl_object_size);
806 		u32 object_base = off - objoff;
807 		if (!(truncate_seq == 1 && truncate_size == -1ULL)) {
808 			if (truncate_size <= object_base) {
809 				truncate_size = 0;
810 			} else {
811 				truncate_size -= object_base;
812 				if (truncate_size > object_size)
813 					truncate_size = object_size;
814 			}
815 		}
816 		osd_req_op_extent_init(req, which, opcode, objoff, objlen,
817 				       truncate_size, truncate_seq);
818 	}
819 
820 	req->r_base_oloc.pool = ceph_file_layout_pg_pool(*layout);
821 
822 	snprintf(req->r_base_oid.name, sizeof(req->r_base_oid.name),
823 		 "%llx.%08llx", vino.ino, objnum);
824 	req->r_base_oid.name_len = strlen(req->r_base_oid.name);
825 
826 	return req;
827 }
828 EXPORT_SYMBOL(ceph_osdc_new_request);
829 
830 /*
831  * We keep osd requests in an rbtree, sorted by ->r_tid.
832  */
833 static void __insert_request(struct ceph_osd_client *osdc,
834 			     struct ceph_osd_request *new)
835 {
836 	struct rb_node **p = &osdc->requests.rb_node;
837 	struct rb_node *parent = NULL;
838 	struct ceph_osd_request *req = NULL;
839 
840 	while (*p) {
841 		parent = *p;
842 		req = rb_entry(parent, struct ceph_osd_request, r_node);
843 		if (new->r_tid < req->r_tid)
844 			p = &(*p)->rb_left;
845 		else if (new->r_tid > req->r_tid)
846 			p = &(*p)->rb_right;
847 		else
848 			BUG();
849 	}
850 
851 	rb_link_node(&new->r_node, parent, p);
852 	rb_insert_color(&new->r_node, &osdc->requests);
853 }
854 
855 static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc,
856 						 u64 tid)
857 {
858 	struct ceph_osd_request *req;
859 	struct rb_node *n = osdc->requests.rb_node;
860 
861 	while (n) {
862 		req = rb_entry(n, struct ceph_osd_request, r_node);
863 		if (tid < req->r_tid)
864 			n = n->rb_left;
865 		else if (tid > req->r_tid)
866 			n = n->rb_right;
867 		else
868 			return req;
869 	}
870 	return NULL;
871 }
872 
873 static struct ceph_osd_request *
874 __lookup_request_ge(struct ceph_osd_client *osdc,
875 		    u64 tid)
876 {
877 	struct ceph_osd_request *req;
878 	struct rb_node *n = osdc->requests.rb_node;
879 
880 	while (n) {
881 		req = rb_entry(n, struct ceph_osd_request, r_node);
882 		if (tid < req->r_tid) {
883 			if (!n->rb_left)
884 				return req;
885 			n = n->rb_left;
886 		} else if (tid > req->r_tid) {
887 			n = n->rb_right;
888 		} else {
889 			return req;
890 		}
891 	}
892 	return NULL;
893 }
894 
895 static void __kick_linger_request(struct ceph_osd_request *req)
896 {
897 	struct ceph_osd_client *osdc = req->r_osdc;
898 	struct ceph_osd *osd = req->r_osd;
899 
900 	/*
901 	 * Linger requests need to be resent with a new tid to avoid
902 	 * the dup op detection logic on the OSDs.  Achieve this with
903 	 * a re-register dance instead of open-coding.
904 	 */
905 	ceph_osdc_get_request(req);
906 	if (!list_empty(&req->r_linger_item))
907 		__unregister_linger_request(osdc, req);
908 	else
909 		__unregister_request(osdc, req);
910 	__register_request(osdc, req);
911 	ceph_osdc_put_request(req);
912 
913 	/*
914 	 * Unless request has been registered as both normal and
915 	 * lingering, __unregister{,_linger}_request clears r_osd.
916 	 * However, here we need to preserve r_osd to make sure we
917 	 * requeue on the same OSD.
918 	 */
919 	WARN_ON(req->r_osd || !osd);
920 	req->r_osd = osd;
921 
922 	dout("%s requeueing %p tid %llu\n", __func__, req, req->r_tid);
923 	__enqueue_request(req);
924 }
925 
926 /*
927  * Resubmit requests pending on the given osd.
928  */
929 static void __kick_osd_requests(struct ceph_osd_client *osdc,
930 				struct ceph_osd *osd)
931 {
932 	struct ceph_osd_request *req, *nreq;
933 	LIST_HEAD(resend);
934 	LIST_HEAD(resend_linger);
935 	int err;
936 
937 	dout("%s osd%d\n", __func__, osd->o_osd);
938 	err = __reset_osd(osdc, osd);
939 	if (err)
940 		return;
941 
942 	/*
943 	 * Build up a list of requests to resend by traversing the
944 	 * osd's list of requests.  Requests for a given object are
945 	 * sent in tid order, and that is also the order they're
946 	 * kept on this list.  Therefore all requests that are in
947 	 * flight will be found first, followed by all requests that
948 	 * have not yet been sent.  And to resend requests while
949 	 * preserving this order we will want to put any sent
950 	 * requests back on the front of the osd client's unsent
951 	 * list.
952 	 *
953 	 * So we build a separate ordered list of already-sent
954 	 * requests for the affected osd and splice it onto the
955 	 * front of the osd client's unsent list.  Once we've seen a
956 	 * request that has not yet been sent we're done.  Those
957 	 * requests are already sitting right where they belong.
958 	 */
959 	list_for_each_entry(req, &osd->o_requests, r_osd_item) {
960 		if (!req->r_sent)
961 			break;
962 
963 		if (!req->r_linger) {
964 			dout("%s requeueing %p tid %llu\n", __func__, req,
965 			     req->r_tid);
966 			list_move_tail(&req->r_req_lru_item, &resend);
967 			req->r_flags |= CEPH_OSD_FLAG_RETRY;
968 		} else {
969 			list_move_tail(&req->r_req_lru_item, &resend_linger);
970 		}
971 	}
972 	list_splice(&resend, &osdc->req_unsent);
973 
974 	/*
975 	 * Both registered and not yet registered linger requests are
976 	 * enqueued with a new tid on the same OSD.  We add/move them
977 	 * to req_unsent/o_requests at the end to keep things in tid
978 	 * order.
979 	 */
980 	list_for_each_entry_safe(req, nreq, &osd->o_linger_requests,
981 				 r_linger_osd_item) {
982 		WARN_ON(!list_empty(&req->r_req_lru_item));
983 		__kick_linger_request(req);
984 	}
985 
986 	list_for_each_entry_safe(req, nreq, &resend_linger, r_req_lru_item)
987 		__kick_linger_request(req);
988 }
989 
990 /*
991  * If the osd connection drops, we need to resubmit all requests.
992  */
993 static void osd_reset(struct ceph_connection *con)
994 {
995 	struct ceph_osd *osd = con->private;
996 	struct ceph_osd_client *osdc;
997 
998 	if (!osd)
999 		return;
1000 	dout("osd_reset osd%d\n", osd->o_osd);
1001 	osdc = osd->o_osdc;
1002 	down_read(&osdc->map_sem);
1003 	mutex_lock(&osdc->request_mutex);
1004 	__kick_osd_requests(osdc, osd);
1005 	__send_queued(osdc);
1006 	mutex_unlock(&osdc->request_mutex);
1007 	up_read(&osdc->map_sem);
1008 }
1009 
1010 /*
1011  * Track open sessions with osds.
1012  */
1013 static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
1014 {
1015 	struct ceph_osd *osd;
1016 
1017 	osd = kzalloc(sizeof(*osd), GFP_NOFS);
1018 	if (!osd)
1019 		return NULL;
1020 
1021 	atomic_set(&osd->o_ref, 1);
1022 	osd->o_osdc = osdc;
1023 	osd->o_osd = onum;
1024 	RB_CLEAR_NODE(&osd->o_node);
1025 	INIT_LIST_HEAD(&osd->o_requests);
1026 	INIT_LIST_HEAD(&osd->o_linger_requests);
1027 	INIT_LIST_HEAD(&osd->o_osd_lru);
1028 	osd->o_incarnation = 1;
1029 
1030 	ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
1031 
1032 	INIT_LIST_HEAD(&osd->o_keepalive_item);
1033 	return osd;
1034 }
1035 
1036 static struct ceph_osd *get_osd(struct ceph_osd *osd)
1037 {
1038 	if (atomic_inc_not_zero(&osd->o_ref)) {
1039 		dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
1040 		     atomic_read(&osd->o_ref));
1041 		return osd;
1042 	} else {
1043 		dout("get_osd %p FAIL\n", osd);
1044 		return NULL;
1045 	}
1046 }
1047 
1048 static void put_osd(struct ceph_osd *osd)
1049 {
1050 	dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
1051 	     atomic_read(&osd->o_ref) - 1);
1052 	if (atomic_dec_and_test(&osd->o_ref)) {
1053 		struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth;
1054 
1055 		if (osd->o_auth.authorizer)
1056 			ceph_auth_destroy_authorizer(ac, osd->o_auth.authorizer);
1057 		kfree(osd);
1058 	}
1059 }
1060 
1061 /*
1062  * remove an osd from our map
1063  */
1064 static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
1065 {
1066 	dout("%s %p osd%d\n", __func__, osd, osd->o_osd);
1067 	WARN_ON(!list_empty(&osd->o_requests));
1068 	WARN_ON(!list_empty(&osd->o_linger_requests));
1069 
1070 	list_del_init(&osd->o_osd_lru);
1071 	rb_erase(&osd->o_node, &osdc->osds);
1072 	RB_CLEAR_NODE(&osd->o_node);
1073 }
1074 
1075 static void remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
1076 {
1077 	dout("%s %p osd%d\n", __func__, osd, osd->o_osd);
1078 
1079 	if (!RB_EMPTY_NODE(&osd->o_node)) {
1080 		ceph_con_close(&osd->o_con);
1081 		__remove_osd(osdc, osd);
1082 		put_osd(osd);
1083 	}
1084 }
1085 
1086 static void remove_all_osds(struct ceph_osd_client *osdc)
1087 {
1088 	dout("%s %p\n", __func__, osdc);
1089 	mutex_lock(&osdc->request_mutex);
1090 	while (!RB_EMPTY_ROOT(&osdc->osds)) {
1091 		struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
1092 						struct ceph_osd, o_node);
1093 		remove_osd(osdc, osd);
1094 	}
1095 	mutex_unlock(&osdc->request_mutex);
1096 }
1097 
1098 static void __move_osd_to_lru(struct ceph_osd_client *osdc,
1099 			      struct ceph_osd *osd)
1100 {
1101 	dout("%s %p\n", __func__, osd);
1102 	BUG_ON(!list_empty(&osd->o_osd_lru));
1103 
1104 	list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
1105 	osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl;
1106 }
1107 
1108 static void maybe_move_osd_to_lru(struct ceph_osd_client *osdc,
1109 				  struct ceph_osd *osd)
1110 {
1111 	dout("%s %p\n", __func__, osd);
1112 
1113 	if (list_empty(&osd->o_requests) &&
1114 	    list_empty(&osd->o_linger_requests))
1115 		__move_osd_to_lru(osdc, osd);
1116 }
1117 
1118 static void __remove_osd_from_lru(struct ceph_osd *osd)
1119 {
1120 	dout("__remove_osd_from_lru %p\n", osd);
1121 	if (!list_empty(&osd->o_osd_lru))
1122 		list_del_init(&osd->o_osd_lru);
1123 }
1124 
1125 static void remove_old_osds(struct ceph_osd_client *osdc)
1126 {
1127 	struct ceph_osd *osd, *nosd;
1128 
1129 	dout("__remove_old_osds %p\n", osdc);
1130 	mutex_lock(&osdc->request_mutex);
1131 	list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
1132 		if (time_before(jiffies, osd->lru_ttl))
1133 			break;
1134 		remove_osd(osdc, osd);
1135 	}
1136 	mutex_unlock(&osdc->request_mutex);
1137 }
1138 
1139 /*
1140  * reset osd connect
1141  */
1142 static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
1143 {
1144 	struct ceph_entity_addr *peer_addr;
1145 
1146 	dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
1147 	if (list_empty(&osd->o_requests) &&
1148 	    list_empty(&osd->o_linger_requests)) {
1149 		remove_osd(osdc, osd);
1150 		return -ENODEV;
1151 	}
1152 
1153 	peer_addr = &osdc->osdmap->osd_addr[osd->o_osd];
1154 	if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1155 			!ceph_con_opened(&osd->o_con)) {
1156 		struct ceph_osd_request *req;
1157 
1158 		dout("osd addr hasn't changed and connection never opened, "
1159 		     "letting msgr retry\n");
1160 		/* touch each r_stamp for handle_timeout()'s benfit */
1161 		list_for_each_entry(req, &osd->o_requests, r_osd_item)
1162 			req->r_stamp = jiffies;
1163 
1164 		return -EAGAIN;
1165 	}
1166 
1167 	ceph_con_close(&osd->o_con);
1168 	ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1169 	osd->o_incarnation++;
1170 
1171 	return 0;
1172 }
1173 
1174 static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new)
1175 {
1176 	struct rb_node **p = &osdc->osds.rb_node;
1177 	struct rb_node *parent = NULL;
1178 	struct ceph_osd *osd = NULL;
1179 
1180 	dout("__insert_osd %p osd%d\n", new, new->o_osd);
1181 	while (*p) {
1182 		parent = *p;
1183 		osd = rb_entry(parent, struct ceph_osd, o_node);
1184 		if (new->o_osd < osd->o_osd)
1185 			p = &(*p)->rb_left;
1186 		else if (new->o_osd > osd->o_osd)
1187 			p = &(*p)->rb_right;
1188 		else
1189 			BUG();
1190 	}
1191 
1192 	rb_link_node(&new->o_node, parent, p);
1193 	rb_insert_color(&new->o_node, &osdc->osds);
1194 }
1195 
1196 static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o)
1197 {
1198 	struct ceph_osd *osd;
1199 	struct rb_node *n = osdc->osds.rb_node;
1200 
1201 	while (n) {
1202 		osd = rb_entry(n, struct ceph_osd, o_node);
1203 		if (o < osd->o_osd)
1204 			n = n->rb_left;
1205 		else if (o > osd->o_osd)
1206 			n = n->rb_right;
1207 		else
1208 			return osd;
1209 	}
1210 	return NULL;
1211 }
1212 
1213 static void __schedule_osd_timeout(struct ceph_osd_client *osdc)
1214 {
1215 	schedule_delayed_work(&osdc->timeout_work,
1216 			      osdc->client->options->osd_keepalive_timeout);
1217 }
1218 
1219 static void __cancel_osd_timeout(struct ceph_osd_client *osdc)
1220 {
1221 	cancel_delayed_work(&osdc->timeout_work);
1222 }
1223 
1224 /*
1225  * Register request, assign tid.  If this is the first request, set up
1226  * the timeout event.
1227  */
1228 static void __register_request(struct ceph_osd_client *osdc,
1229 			       struct ceph_osd_request *req)
1230 {
1231 	req->r_tid = ++osdc->last_tid;
1232 	req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
1233 	dout("__register_request %p tid %lld\n", req, req->r_tid);
1234 	__insert_request(osdc, req);
1235 	ceph_osdc_get_request(req);
1236 	osdc->num_requests++;
1237 	if (osdc->num_requests == 1) {
1238 		dout(" first request, scheduling timeout\n");
1239 		__schedule_osd_timeout(osdc);
1240 	}
1241 }
1242 
1243 /*
1244  * called under osdc->request_mutex
1245  */
1246 static void __unregister_request(struct ceph_osd_client *osdc,
1247 				 struct ceph_osd_request *req)
1248 {
1249 	if (RB_EMPTY_NODE(&req->r_node)) {
1250 		dout("__unregister_request %p tid %lld not registered\n",
1251 			req, req->r_tid);
1252 		return;
1253 	}
1254 
1255 	dout("__unregister_request %p tid %lld\n", req, req->r_tid);
1256 	rb_erase(&req->r_node, &osdc->requests);
1257 	RB_CLEAR_NODE(&req->r_node);
1258 	osdc->num_requests--;
1259 
1260 	if (req->r_osd) {
1261 		/* make sure the original request isn't in flight. */
1262 		ceph_msg_revoke(req->r_request);
1263 
1264 		list_del_init(&req->r_osd_item);
1265 		maybe_move_osd_to_lru(osdc, req->r_osd);
1266 		if (list_empty(&req->r_linger_osd_item))
1267 			req->r_osd = NULL;
1268 	}
1269 
1270 	list_del_init(&req->r_req_lru_item);
1271 	ceph_osdc_put_request(req);
1272 
1273 	if (osdc->num_requests == 0) {
1274 		dout(" no requests, canceling timeout\n");
1275 		__cancel_osd_timeout(osdc);
1276 	}
1277 }
1278 
1279 /*
1280  * Cancel a previously queued request message
1281  */
1282 static void __cancel_request(struct ceph_osd_request *req)
1283 {
1284 	if (req->r_sent && req->r_osd) {
1285 		ceph_msg_revoke(req->r_request);
1286 		req->r_sent = 0;
1287 	}
1288 }
1289 
1290 static void __register_linger_request(struct ceph_osd_client *osdc,
1291 				    struct ceph_osd_request *req)
1292 {
1293 	dout("%s %p tid %llu\n", __func__, req, req->r_tid);
1294 	WARN_ON(!req->r_linger);
1295 
1296 	ceph_osdc_get_request(req);
1297 	list_add_tail(&req->r_linger_item, &osdc->req_linger);
1298 	if (req->r_osd)
1299 		list_add_tail(&req->r_linger_osd_item,
1300 			      &req->r_osd->o_linger_requests);
1301 }
1302 
1303 static void __unregister_linger_request(struct ceph_osd_client *osdc,
1304 					struct ceph_osd_request *req)
1305 {
1306 	WARN_ON(!req->r_linger);
1307 
1308 	if (list_empty(&req->r_linger_item)) {
1309 		dout("%s %p tid %llu not registered\n", __func__, req,
1310 		     req->r_tid);
1311 		return;
1312 	}
1313 
1314 	dout("%s %p tid %llu\n", __func__, req, req->r_tid);
1315 	list_del_init(&req->r_linger_item);
1316 
1317 	if (req->r_osd) {
1318 		list_del_init(&req->r_linger_osd_item);
1319 		maybe_move_osd_to_lru(osdc, req->r_osd);
1320 		if (list_empty(&req->r_osd_item))
1321 			req->r_osd = NULL;
1322 	}
1323 	ceph_osdc_put_request(req);
1324 }
1325 
1326 void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
1327 				  struct ceph_osd_request *req)
1328 {
1329 	if (!req->r_linger) {
1330 		dout("set_request_linger %p\n", req);
1331 		req->r_linger = 1;
1332 	}
1333 }
1334 EXPORT_SYMBOL(ceph_osdc_set_request_linger);
1335 
1336 /*
1337  * Returns whether a request should be blocked from being sent
1338  * based on the current osdmap and osd_client settings.
1339  *
1340  * Caller should hold map_sem for read.
1341  */
1342 static bool __req_should_be_paused(struct ceph_osd_client *osdc,
1343 				   struct ceph_osd_request *req)
1344 {
1345 	bool pauserd = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD);
1346 	bool pausewr = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR) ||
1347 		ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
1348 	return (req->r_flags & CEPH_OSD_FLAG_READ && pauserd) ||
1349 		(req->r_flags & CEPH_OSD_FLAG_WRITE && pausewr);
1350 }
1351 
1352 /*
1353  * Calculate mapping of a request to a PG.  Takes tiering into account.
1354  */
1355 static int __calc_request_pg(struct ceph_osdmap *osdmap,
1356 			     struct ceph_osd_request *req,
1357 			     struct ceph_pg *pg_out)
1358 {
1359 	bool need_check_tiering;
1360 
1361 	need_check_tiering = false;
1362 	if (req->r_target_oloc.pool == -1) {
1363 		req->r_target_oloc = req->r_base_oloc; /* struct */
1364 		need_check_tiering = true;
1365 	}
1366 	if (req->r_target_oid.name_len == 0) {
1367 		ceph_oid_copy(&req->r_target_oid, &req->r_base_oid);
1368 		need_check_tiering = true;
1369 	}
1370 
1371 	if (need_check_tiering &&
1372 	    (req->r_flags & CEPH_OSD_FLAG_IGNORE_OVERLAY) == 0) {
1373 		struct ceph_pg_pool_info *pi;
1374 
1375 		pi = ceph_pg_pool_by_id(osdmap, req->r_target_oloc.pool);
1376 		if (pi) {
1377 			if ((req->r_flags & CEPH_OSD_FLAG_READ) &&
1378 			    pi->read_tier >= 0)
1379 				req->r_target_oloc.pool = pi->read_tier;
1380 			if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
1381 			    pi->write_tier >= 0)
1382 				req->r_target_oloc.pool = pi->write_tier;
1383 		}
1384 		/* !pi is caught in ceph_oloc_oid_to_pg() */
1385 	}
1386 
1387 	return ceph_oloc_oid_to_pg(osdmap, &req->r_target_oloc,
1388 				   &req->r_target_oid, pg_out);
1389 }
1390 
1391 static void __enqueue_request(struct ceph_osd_request *req)
1392 {
1393 	struct ceph_osd_client *osdc = req->r_osdc;
1394 
1395 	dout("%s %p tid %llu to osd%d\n", __func__, req, req->r_tid,
1396 	     req->r_osd ? req->r_osd->o_osd : -1);
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 }
1406 
1407 /*
1408  * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
1409  * (as needed), and set the request r_osd appropriately.  If there is
1410  * no up osd, set r_osd to NULL.  Move the request to the appropriate list
1411  * (unsent, homeless) or leave on in-flight lru.
1412  *
1413  * Return 0 if unchanged, 1 if changed, or negative on error.
1414  *
1415  * Caller should hold map_sem for read and request_mutex.
1416  */
1417 static int __map_request(struct ceph_osd_client *osdc,
1418 			 struct ceph_osd_request *req, int force_resend)
1419 {
1420 	struct ceph_pg pgid;
1421 	int acting[CEPH_PG_MAX_SIZE];
1422 	int num, o;
1423 	int err;
1424 	bool was_paused;
1425 
1426 	dout("map_request %p tid %lld\n", req, req->r_tid);
1427 
1428 	err = __calc_request_pg(osdc->osdmap, req, &pgid);
1429 	if (err) {
1430 		list_move(&req->r_req_lru_item, &osdc->req_notarget);
1431 		return err;
1432 	}
1433 	req->r_pgid = pgid;
1434 
1435 	num = ceph_calc_pg_acting(osdc->osdmap, pgid, acting, &o);
1436 	if (num < 0)
1437 		num = 0;
1438 
1439 	was_paused = req->r_paused;
1440 	req->r_paused = __req_should_be_paused(osdc, req);
1441 	if (was_paused && !req->r_paused)
1442 		force_resend = 1;
1443 
1444 	if ((!force_resend &&
1445 	     req->r_osd && req->r_osd->o_osd == o &&
1446 	     req->r_sent >= req->r_osd->o_incarnation &&
1447 	     req->r_num_pg_osds == num &&
1448 	     memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
1449 	    (req->r_osd == NULL && o == -1) ||
1450 	    req->r_paused)
1451 		return 0;  /* no change */
1452 
1453 	dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n",
1454 	     req->r_tid, pgid.pool, pgid.seed, o,
1455 	     req->r_osd ? req->r_osd->o_osd : -1);
1456 
1457 	/* record full pg acting set */
1458 	memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
1459 	req->r_num_pg_osds = num;
1460 
1461 	if (req->r_osd) {
1462 		__cancel_request(req);
1463 		list_del_init(&req->r_osd_item);
1464 		list_del_init(&req->r_linger_osd_item);
1465 		req->r_osd = NULL;
1466 	}
1467 
1468 	req->r_osd = __lookup_osd(osdc, o);
1469 	if (!req->r_osd && o >= 0) {
1470 		err = -ENOMEM;
1471 		req->r_osd = create_osd(osdc, o);
1472 		if (!req->r_osd) {
1473 			list_move(&req->r_req_lru_item, &osdc->req_notarget);
1474 			goto out;
1475 		}
1476 
1477 		dout("map_request osd %p is osd%d\n", req->r_osd, o);
1478 		__insert_osd(osdc, req->r_osd);
1479 
1480 		ceph_con_open(&req->r_osd->o_con,
1481 			      CEPH_ENTITY_TYPE_OSD, o,
1482 			      &osdc->osdmap->osd_addr[o]);
1483 	}
1484 
1485 	__enqueue_request(req);
1486 	err = 1;   /* osd or pg changed */
1487 
1488 out:
1489 	return err;
1490 }
1491 
1492 /*
1493  * caller should hold map_sem (for read) and request_mutex
1494  */
1495 static void __send_request(struct ceph_osd_client *osdc,
1496 			   struct ceph_osd_request *req)
1497 {
1498 	void *p;
1499 
1500 	dout("send_request %p tid %llu to osd%d flags %d pg %lld.%x\n",
1501 	     req, req->r_tid, req->r_osd->o_osd, req->r_flags,
1502 	     (unsigned long long)req->r_pgid.pool, req->r_pgid.seed);
1503 
1504 	/* fill in message content that changes each time we send it */
1505 	put_unaligned_le32(osdc->osdmap->epoch, req->r_request_osdmap_epoch);
1506 	put_unaligned_le32(req->r_flags, req->r_request_flags);
1507 	put_unaligned_le64(req->r_target_oloc.pool, req->r_request_pool);
1508 	p = req->r_request_pgid;
1509 	ceph_encode_64(&p, req->r_pgid.pool);
1510 	ceph_encode_32(&p, req->r_pgid.seed);
1511 	put_unaligned_le64(1, req->r_request_attempts);  /* FIXME */
1512 	memcpy(req->r_request_reassert_version, &req->r_reassert_version,
1513 	       sizeof(req->r_reassert_version));
1514 
1515 	req->r_stamp = jiffies;
1516 	list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
1517 
1518 	ceph_msg_get(req->r_request); /* send consumes a ref */
1519 
1520 	req->r_sent = req->r_osd->o_incarnation;
1521 
1522 	ceph_con_send(&req->r_osd->o_con, req->r_request);
1523 }
1524 
1525 /*
1526  * Send any requests in the queue (req_unsent).
1527  */
1528 static void __send_queued(struct ceph_osd_client *osdc)
1529 {
1530 	struct ceph_osd_request *req, *tmp;
1531 
1532 	dout("__send_queued\n");
1533 	list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item)
1534 		__send_request(osdc, req);
1535 }
1536 
1537 /*
1538  * Caller should hold map_sem for read and request_mutex.
1539  */
1540 static int __ceph_osdc_start_request(struct ceph_osd_client *osdc,
1541 				     struct ceph_osd_request *req,
1542 				     bool nofail)
1543 {
1544 	int rc;
1545 
1546 	__register_request(osdc, req);
1547 	req->r_sent = 0;
1548 	req->r_got_reply = 0;
1549 	rc = __map_request(osdc, req, 0);
1550 	if (rc < 0) {
1551 		if (nofail) {
1552 			dout("osdc_start_request failed map, "
1553 				" will retry %lld\n", req->r_tid);
1554 			rc = 0;
1555 		} else {
1556 			__unregister_request(osdc, req);
1557 		}
1558 		return rc;
1559 	}
1560 
1561 	if (req->r_osd == NULL) {
1562 		dout("send_request %p no up osds in pg\n", req);
1563 		ceph_monc_request_next_osdmap(&osdc->client->monc);
1564 	} else {
1565 		__send_queued(osdc);
1566 	}
1567 
1568 	return 0;
1569 }
1570 
1571 /*
1572  * Timeout callback, called every N seconds when 1 or more osd
1573  * requests has been active for more than N seconds.  When this
1574  * happens, we ping all OSDs with requests who have timed out to
1575  * ensure any communications channel reset is detected.  Reset the
1576  * request timeouts another N seconds in the future as we go.
1577  * Reschedule the timeout event another N seconds in future (unless
1578  * there are no open requests).
1579  */
1580 static void handle_timeout(struct work_struct *work)
1581 {
1582 	struct ceph_osd_client *osdc =
1583 		container_of(work, struct ceph_osd_client, timeout_work.work);
1584 	struct ceph_options *opts = osdc->client->options;
1585 	struct ceph_osd_request *req;
1586 	struct ceph_osd *osd;
1587 	struct list_head slow_osds;
1588 	dout("timeout\n");
1589 	down_read(&osdc->map_sem);
1590 
1591 	ceph_monc_request_next_osdmap(&osdc->client->monc);
1592 
1593 	mutex_lock(&osdc->request_mutex);
1594 
1595 	/*
1596 	 * ping osds that are a bit slow.  this ensures that if there
1597 	 * is a break in the TCP connection we will notice, and reopen
1598 	 * a connection with that osd (from the fault callback).
1599 	 */
1600 	INIT_LIST_HEAD(&slow_osds);
1601 	list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
1602 		if (time_before(jiffies,
1603 				req->r_stamp + opts->osd_keepalive_timeout))
1604 			break;
1605 
1606 		osd = req->r_osd;
1607 		BUG_ON(!osd);
1608 		dout(" tid %llu is slow, will send keepalive on osd%d\n",
1609 		     req->r_tid, osd->o_osd);
1610 		list_move_tail(&osd->o_keepalive_item, &slow_osds);
1611 	}
1612 	while (!list_empty(&slow_osds)) {
1613 		osd = list_entry(slow_osds.next, struct ceph_osd,
1614 				 o_keepalive_item);
1615 		list_del_init(&osd->o_keepalive_item);
1616 		ceph_con_keepalive(&osd->o_con);
1617 	}
1618 
1619 	__schedule_osd_timeout(osdc);
1620 	__send_queued(osdc);
1621 	mutex_unlock(&osdc->request_mutex);
1622 	up_read(&osdc->map_sem);
1623 }
1624 
1625 static void handle_osds_timeout(struct work_struct *work)
1626 {
1627 	struct ceph_osd_client *osdc =
1628 		container_of(work, struct ceph_osd_client,
1629 			     osds_timeout_work.work);
1630 	unsigned long delay = osdc->client->options->osd_idle_ttl / 4;
1631 
1632 	dout("osds timeout\n");
1633 	down_read(&osdc->map_sem);
1634 	remove_old_osds(osdc);
1635 	up_read(&osdc->map_sem);
1636 
1637 	schedule_delayed_work(&osdc->osds_timeout_work,
1638 			      round_jiffies_relative(delay));
1639 }
1640 
1641 static int ceph_oloc_decode(void **p, void *end,
1642 			    struct ceph_object_locator *oloc)
1643 {
1644 	u8 struct_v, struct_cv;
1645 	u32 len;
1646 	void *struct_end;
1647 	int ret = 0;
1648 
1649 	ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
1650 	struct_v = ceph_decode_8(p);
1651 	struct_cv = ceph_decode_8(p);
1652 	if (struct_v < 3) {
1653 		pr_warn("got v %d < 3 cv %d of ceph_object_locator\n",
1654 			struct_v, struct_cv);
1655 		goto e_inval;
1656 	}
1657 	if (struct_cv > 6) {
1658 		pr_warn("got v %d cv %d > 6 of ceph_object_locator\n",
1659 			struct_v, struct_cv);
1660 		goto e_inval;
1661 	}
1662 	len = ceph_decode_32(p);
1663 	ceph_decode_need(p, end, len, e_inval);
1664 	struct_end = *p + len;
1665 
1666 	oloc->pool = ceph_decode_64(p);
1667 	*p += 4; /* skip preferred */
1668 
1669 	len = ceph_decode_32(p);
1670 	if (len > 0) {
1671 		pr_warn("ceph_object_locator::key is set\n");
1672 		goto e_inval;
1673 	}
1674 
1675 	if (struct_v >= 5) {
1676 		len = ceph_decode_32(p);
1677 		if (len > 0) {
1678 			pr_warn("ceph_object_locator::nspace is set\n");
1679 			goto e_inval;
1680 		}
1681 	}
1682 
1683 	if (struct_v >= 6) {
1684 		s64 hash = ceph_decode_64(p);
1685 		if (hash != -1) {
1686 			pr_warn("ceph_object_locator::hash is set\n");
1687 			goto e_inval;
1688 		}
1689 	}
1690 
1691 	/* skip the rest */
1692 	*p = struct_end;
1693 out:
1694 	return ret;
1695 
1696 e_inval:
1697 	ret = -EINVAL;
1698 	goto out;
1699 }
1700 
1701 static int ceph_redirect_decode(void **p, void *end,
1702 				struct ceph_request_redirect *redir)
1703 {
1704 	u8 struct_v, struct_cv;
1705 	u32 len;
1706 	void *struct_end;
1707 	int ret;
1708 
1709 	ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
1710 	struct_v = ceph_decode_8(p);
1711 	struct_cv = ceph_decode_8(p);
1712 	if (struct_cv > 1) {
1713 		pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n",
1714 			struct_v, struct_cv);
1715 		goto e_inval;
1716 	}
1717 	len = ceph_decode_32(p);
1718 	ceph_decode_need(p, end, len, e_inval);
1719 	struct_end = *p + len;
1720 
1721 	ret = ceph_oloc_decode(p, end, &redir->oloc);
1722 	if (ret)
1723 		goto out;
1724 
1725 	len = ceph_decode_32(p);
1726 	if (len > 0) {
1727 		pr_warn("ceph_request_redirect::object_name is set\n");
1728 		goto e_inval;
1729 	}
1730 
1731 	len = ceph_decode_32(p);
1732 	*p += len; /* skip osd_instructions */
1733 
1734 	/* skip the rest */
1735 	*p = struct_end;
1736 out:
1737 	return ret;
1738 
1739 e_inval:
1740 	ret = -EINVAL;
1741 	goto out;
1742 }
1743 
1744 static void complete_request(struct ceph_osd_request *req)
1745 {
1746 	complete_all(&req->r_safe_completion);  /* fsync waiter */
1747 }
1748 
1749 /*
1750  * handle osd op reply.  either call the callback if it is specified,
1751  * or do the completion to wake up the waiting thread.
1752  */
1753 static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg,
1754 			 struct ceph_connection *con)
1755 {
1756 	void *p, *end;
1757 	struct ceph_osd_request *req;
1758 	struct ceph_request_redirect redir;
1759 	u64 tid;
1760 	int object_len;
1761 	unsigned int numops;
1762 	int payload_len, flags;
1763 	s32 result;
1764 	s32 retry_attempt;
1765 	struct ceph_pg pg;
1766 	int err;
1767 	u32 reassert_epoch;
1768 	u64 reassert_version;
1769 	u32 osdmap_epoch;
1770 	int already_completed;
1771 	u32 bytes;
1772 	unsigned int i;
1773 
1774 	tid = le64_to_cpu(msg->hdr.tid);
1775 	dout("handle_reply %p tid %llu\n", msg, tid);
1776 
1777 	p = msg->front.iov_base;
1778 	end = p + msg->front.iov_len;
1779 
1780 	ceph_decode_need(&p, end, 4, bad);
1781 	object_len = ceph_decode_32(&p);
1782 	ceph_decode_need(&p, end, object_len, bad);
1783 	p += object_len;
1784 
1785 	err = ceph_decode_pgid(&p, end, &pg);
1786 	if (err)
1787 		goto bad;
1788 
1789 	ceph_decode_need(&p, end, 8 + 4 + 4 + 8 + 4, bad);
1790 	flags = ceph_decode_64(&p);
1791 	result = ceph_decode_32(&p);
1792 	reassert_epoch = ceph_decode_32(&p);
1793 	reassert_version = ceph_decode_64(&p);
1794 	osdmap_epoch = ceph_decode_32(&p);
1795 
1796 	/* lookup */
1797 	down_read(&osdc->map_sem);
1798 	mutex_lock(&osdc->request_mutex);
1799 	req = __lookup_request(osdc, tid);
1800 	if (req == NULL) {
1801 		dout("handle_reply tid %llu dne\n", tid);
1802 		goto bad_mutex;
1803 	}
1804 	ceph_osdc_get_request(req);
1805 
1806 	dout("handle_reply %p tid %llu req %p result %d\n", msg, tid,
1807 	     req, result);
1808 
1809 	ceph_decode_need(&p, end, 4, bad_put);
1810 	numops = ceph_decode_32(&p);
1811 	if (numops > CEPH_OSD_MAX_OP)
1812 		goto bad_put;
1813 	if (numops != req->r_num_ops)
1814 		goto bad_put;
1815 	payload_len = 0;
1816 	ceph_decode_need(&p, end, numops * sizeof(struct ceph_osd_op), bad_put);
1817 	for (i = 0; i < numops; i++) {
1818 		struct ceph_osd_op *op = p;
1819 		int len;
1820 
1821 		len = le32_to_cpu(op->payload_len);
1822 		req->r_reply_op_len[i] = len;
1823 		dout(" op %d has %d bytes\n", i, len);
1824 		payload_len += len;
1825 		p += sizeof(*op);
1826 	}
1827 	bytes = le32_to_cpu(msg->hdr.data_len);
1828 	if (payload_len != bytes) {
1829 		pr_warn("sum of op payload lens %d != data_len %d\n",
1830 			payload_len, bytes);
1831 		goto bad_put;
1832 	}
1833 
1834 	ceph_decode_need(&p, end, 4 + numops * 4, bad_put);
1835 	retry_attempt = ceph_decode_32(&p);
1836 	for (i = 0; i < numops; i++)
1837 		req->r_reply_op_result[i] = ceph_decode_32(&p);
1838 
1839 	if (le16_to_cpu(msg->hdr.version) >= 6) {
1840 		p += 8 + 4; /* skip replay_version */
1841 		p += 8; /* skip user_version */
1842 
1843 		err = ceph_redirect_decode(&p, end, &redir);
1844 		if (err)
1845 			goto bad_put;
1846 	} else {
1847 		redir.oloc.pool = -1;
1848 	}
1849 
1850 	if (redir.oloc.pool != -1) {
1851 		dout("redirect pool %lld\n", redir.oloc.pool);
1852 
1853 		__unregister_request(osdc, req);
1854 
1855 		req->r_target_oloc = redir.oloc; /* struct */
1856 
1857 		/*
1858 		 * Start redirect requests with nofail=true.  If
1859 		 * mapping fails, request will end up on the notarget
1860 		 * list, waiting for the new osdmap (which can take
1861 		 * a while), even though the original request mapped
1862 		 * successfully.  In the future we might want to follow
1863 		 * original request's nofail setting here.
1864 		 */
1865 		err = __ceph_osdc_start_request(osdc, req, true);
1866 		BUG_ON(err);
1867 
1868 		goto out_unlock;
1869 	}
1870 
1871 	already_completed = req->r_got_reply;
1872 	if (!req->r_got_reply) {
1873 		req->r_result = result;
1874 		dout("handle_reply result %d bytes %d\n", req->r_result,
1875 		     bytes);
1876 		if (req->r_result == 0)
1877 			req->r_result = bytes;
1878 
1879 		/* in case this is a write and we need to replay, */
1880 		req->r_reassert_version.epoch = cpu_to_le32(reassert_epoch);
1881 		req->r_reassert_version.version = cpu_to_le64(reassert_version);
1882 
1883 		req->r_got_reply = 1;
1884 	} else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) {
1885 		dout("handle_reply tid %llu dup ack\n", tid);
1886 		goto out_unlock;
1887 	}
1888 
1889 	dout("handle_reply tid %llu flags %d\n", tid, flags);
1890 
1891 	if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK))
1892 		__register_linger_request(osdc, req);
1893 
1894 	/* either this is a read, or we got the safe response */
1895 	if (result < 0 ||
1896 	    (flags & CEPH_OSD_FLAG_ONDISK) ||
1897 	    ((flags & CEPH_OSD_FLAG_WRITE) == 0))
1898 		__unregister_request(osdc, req);
1899 
1900 	mutex_unlock(&osdc->request_mutex);
1901 	up_read(&osdc->map_sem);
1902 
1903 	if (!already_completed) {
1904 		if (req->r_unsafe_callback &&
1905 		    result >= 0 && !(flags & CEPH_OSD_FLAG_ONDISK))
1906 			req->r_unsafe_callback(req, true);
1907 		if (req->r_callback)
1908 			req->r_callback(req, msg);
1909 		else
1910 			complete_all(&req->r_completion);
1911 	}
1912 
1913 	if (flags & CEPH_OSD_FLAG_ONDISK) {
1914 		if (req->r_unsafe_callback && already_completed)
1915 			req->r_unsafe_callback(req, false);
1916 		complete_request(req);
1917 	}
1918 
1919 out:
1920 	dout("req=%p req->r_linger=%d\n", req, req->r_linger);
1921 	ceph_osdc_put_request(req);
1922 	return;
1923 out_unlock:
1924 	mutex_unlock(&osdc->request_mutex);
1925 	up_read(&osdc->map_sem);
1926 	goto out;
1927 
1928 bad_put:
1929 	req->r_result = -EIO;
1930 	__unregister_request(osdc, req);
1931 	if (req->r_callback)
1932 		req->r_callback(req, msg);
1933 	else
1934 		complete_all(&req->r_completion);
1935 	complete_request(req);
1936 	ceph_osdc_put_request(req);
1937 bad_mutex:
1938 	mutex_unlock(&osdc->request_mutex);
1939 	up_read(&osdc->map_sem);
1940 bad:
1941 	pr_err("corrupt osd_op_reply got %d %d\n",
1942 	       (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len));
1943 	ceph_msg_dump(msg);
1944 }
1945 
1946 static void reset_changed_osds(struct ceph_osd_client *osdc)
1947 {
1948 	struct rb_node *p, *n;
1949 
1950 	dout("%s %p\n", __func__, osdc);
1951 	for (p = rb_first(&osdc->osds); p; p = n) {
1952 		struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
1953 
1954 		n = rb_next(p);
1955 		if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
1956 		    memcmp(&osd->o_con.peer_addr,
1957 			   ceph_osd_addr(osdc->osdmap,
1958 					 osd->o_osd),
1959 			   sizeof(struct ceph_entity_addr)) != 0)
1960 			__reset_osd(osdc, osd);
1961 	}
1962 }
1963 
1964 /*
1965  * Requeue requests whose mapping to an OSD has changed.  If requests map to
1966  * no osd, request a new map.
1967  *
1968  * Caller should hold map_sem for read.
1969  */
1970 static void kick_requests(struct ceph_osd_client *osdc, bool force_resend,
1971 			  bool force_resend_writes)
1972 {
1973 	struct ceph_osd_request *req, *nreq;
1974 	struct rb_node *p;
1975 	int needmap = 0;
1976 	int err;
1977 	bool force_resend_req;
1978 
1979 	dout("kick_requests %s %s\n", force_resend ? " (force resend)" : "",
1980 		force_resend_writes ? " (force resend writes)" : "");
1981 	mutex_lock(&osdc->request_mutex);
1982 	for (p = rb_first(&osdc->requests); p; ) {
1983 		req = rb_entry(p, struct ceph_osd_request, r_node);
1984 		p = rb_next(p);
1985 
1986 		/*
1987 		 * For linger requests that have not yet been
1988 		 * registered, move them to the linger list; they'll
1989 		 * be sent to the osd in the loop below.  Unregister
1990 		 * the request before re-registering it as a linger
1991 		 * request to ensure the __map_request() below
1992 		 * will decide it needs to be sent.
1993 		 */
1994 		if (req->r_linger && list_empty(&req->r_linger_item)) {
1995 			dout("%p tid %llu restart on osd%d\n",
1996 			     req, req->r_tid,
1997 			     req->r_osd ? req->r_osd->o_osd : -1);
1998 			ceph_osdc_get_request(req);
1999 			__unregister_request(osdc, req);
2000 			__register_linger_request(osdc, req);
2001 			ceph_osdc_put_request(req);
2002 			continue;
2003 		}
2004 
2005 		force_resend_req = force_resend ||
2006 			(force_resend_writes &&
2007 				req->r_flags & CEPH_OSD_FLAG_WRITE);
2008 		err = __map_request(osdc, req, force_resend_req);
2009 		if (err < 0)
2010 			continue;  /* error */
2011 		if (req->r_osd == NULL) {
2012 			dout("%p tid %llu maps to no osd\n", req, req->r_tid);
2013 			needmap++;  /* request a newer map */
2014 		} else if (err > 0) {
2015 			if (!req->r_linger) {
2016 				dout("%p tid %llu requeued on osd%d\n", req,
2017 				     req->r_tid,
2018 				     req->r_osd ? req->r_osd->o_osd : -1);
2019 				req->r_flags |= CEPH_OSD_FLAG_RETRY;
2020 			}
2021 		}
2022 	}
2023 
2024 	list_for_each_entry_safe(req, nreq, &osdc->req_linger,
2025 				 r_linger_item) {
2026 		dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
2027 
2028 		err = __map_request(osdc, req,
2029 				    force_resend || force_resend_writes);
2030 		dout("__map_request returned %d\n", err);
2031 		if (err < 0)
2032 			continue;  /* hrm! */
2033 		if (req->r_osd == NULL || err > 0) {
2034 			if (req->r_osd == NULL) {
2035 				dout("lingering %p tid %llu maps to no osd\n",
2036 				     req, req->r_tid);
2037 				/*
2038 				 * A homeless lingering request makes
2039 				 * no sense, as it's job is to keep
2040 				 * a particular OSD connection open.
2041 				 * Request a newer map and kick the
2042 				 * request, knowing that it won't be
2043 				 * resent until we actually get a map
2044 				 * that can tell us where to send it.
2045 				 */
2046 				needmap++;
2047 			}
2048 
2049 			dout("kicking lingering %p tid %llu osd%d\n", req,
2050 			     req->r_tid, req->r_osd ? req->r_osd->o_osd : -1);
2051 			__register_request(osdc, req);
2052 			__unregister_linger_request(osdc, req);
2053 		}
2054 	}
2055 	reset_changed_osds(osdc);
2056 	mutex_unlock(&osdc->request_mutex);
2057 
2058 	if (needmap) {
2059 		dout("%d requests for down osds, need new map\n", needmap);
2060 		ceph_monc_request_next_osdmap(&osdc->client->monc);
2061 	}
2062 }
2063 
2064 
2065 /*
2066  * Process updated osd map.
2067  *
2068  * The message contains any number of incremental and full maps, normally
2069  * indicating some sort of topology change in the cluster.  Kick requests
2070  * off to different OSDs as needed.
2071  */
2072 void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
2073 {
2074 	void *p, *end, *next;
2075 	u32 nr_maps, maplen;
2076 	u32 epoch;
2077 	struct ceph_osdmap *newmap = NULL, *oldmap;
2078 	int err;
2079 	struct ceph_fsid fsid;
2080 	bool was_full;
2081 
2082 	dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
2083 	p = msg->front.iov_base;
2084 	end = p + msg->front.iov_len;
2085 
2086 	/* verify fsid */
2087 	ceph_decode_need(&p, end, sizeof(fsid), bad);
2088 	ceph_decode_copy(&p, &fsid, sizeof(fsid));
2089 	if (ceph_check_fsid(osdc->client, &fsid) < 0)
2090 		return;
2091 
2092 	down_write(&osdc->map_sem);
2093 
2094 	was_full = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
2095 
2096 	/* incremental maps */
2097 	ceph_decode_32_safe(&p, end, nr_maps, bad);
2098 	dout(" %d inc maps\n", nr_maps);
2099 	while (nr_maps > 0) {
2100 		ceph_decode_need(&p, end, 2*sizeof(u32), bad);
2101 		epoch = ceph_decode_32(&p);
2102 		maplen = ceph_decode_32(&p);
2103 		ceph_decode_need(&p, end, maplen, bad);
2104 		next = p + maplen;
2105 		if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
2106 			dout("applying incremental map %u len %d\n",
2107 			     epoch, maplen);
2108 			newmap = osdmap_apply_incremental(&p, next,
2109 							  osdc->osdmap,
2110 							  &osdc->client->msgr);
2111 			if (IS_ERR(newmap)) {
2112 				err = PTR_ERR(newmap);
2113 				goto bad;
2114 			}
2115 			BUG_ON(!newmap);
2116 			if (newmap != osdc->osdmap) {
2117 				ceph_osdmap_destroy(osdc->osdmap);
2118 				osdc->osdmap = newmap;
2119 			}
2120 			was_full = was_full ||
2121 				ceph_osdmap_flag(osdc->osdmap,
2122 						 CEPH_OSDMAP_FULL);
2123 			kick_requests(osdc, 0, was_full);
2124 		} else {
2125 			dout("ignoring incremental map %u len %d\n",
2126 			     epoch, maplen);
2127 		}
2128 		p = next;
2129 		nr_maps--;
2130 	}
2131 	if (newmap)
2132 		goto done;
2133 
2134 	/* full maps */
2135 	ceph_decode_32_safe(&p, end, nr_maps, bad);
2136 	dout(" %d full maps\n", nr_maps);
2137 	while (nr_maps) {
2138 		ceph_decode_need(&p, end, 2*sizeof(u32), bad);
2139 		epoch = ceph_decode_32(&p);
2140 		maplen = ceph_decode_32(&p);
2141 		ceph_decode_need(&p, end, maplen, bad);
2142 		if (nr_maps > 1) {
2143 			dout("skipping non-latest full map %u len %d\n",
2144 			     epoch, maplen);
2145 		} else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
2146 			dout("skipping full map %u len %d, "
2147 			     "older than our %u\n", epoch, maplen,
2148 			     osdc->osdmap->epoch);
2149 		} else {
2150 			int skipped_map = 0;
2151 
2152 			dout("taking full map %u len %d\n", epoch, maplen);
2153 			newmap = ceph_osdmap_decode(&p, p+maplen);
2154 			if (IS_ERR(newmap)) {
2155 				err = PTR_ERR(newmap);
2156 				goto bad;
2157 			}
2158 			BUG_ON(!newmap);
2159 			oldmap = osdc->osdmap;
2160 			osdc->osdmap = newmap;
2161 			if (oldmap) {
2162 				if (oldmap->epoch + 1 < newmap->epoch)
2163 					skipped_map = 1;
2164 				ceph_osdmap_destroy(oldmap);
2165 			}
2166 			was_full = was_full ||
2167 				ceph_osdmap_flag(osdc->osdmap,
2168 						 CEPH_OSDMAP_FULL);
2169 			kick_requests(osdc, skipped_map, was_full);
2170 		}
2171 		p += maplen;
2172 		nr_maps--;
2173 	}
2174 
2175 	if (!osdc->osdmap)
2176 		goto bad;
2177 done:
2178 	downgrade_write(&osdc->map_sem);
2179 	ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch);
2180 
2181 	/*
2182 	 * subscribe to subsequent osdmap updates if full to ensure
2183 	 * we find out when we are no longer full and stop returning
2184 	 * ENOSPC.
2185 	 */
2186 	if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) ||
2187 		ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD) ||
2188 		ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR))
2189 		ceph_monc_request_next_osdmap(&osdc->client->monc);
2190 
2191 	mutex_lock(&osdc->request_mutex);
2192 	__send_queued(osdc);
2193 	mutex_unlock(&osdc->request_mutex);
2194 	up_read(&osdc->map_sem);
2195 	wake_up_all(&osdc->client->auth_wq);
2196 	return;
2197 
2198 bad:
2199 	pr_err("osdc handle_map corrupt msg\n");
2200 	ceph_msg_dump(msg);
2201 	up_write(&osdc->map_sem);
2202 }
2203 
2204 /*
2205  * watch/notify callback event infrastructure
2206  *
2207  * These callbacks are used both for watch and notify operations.
2208  */
2209 static void __release_event(struct kref *kref)
2210 {
2211 	struct ceph_osd_event *event =
2212 		container_of(kref, struct ceph_osd_event, kref);
2213 
2214 	dout("__release_event %p\n", event);
2215 	kfree(event);
2216 }
2217 
2218 static void get_event(struct ceph_osd_event *event)
2219 {
2220 	kref_get(&event->kref);
2221 }
2222 
2223 void ceph_osdc_put_event(struct ceph_osd_event *event)
2224 {
2225 	kref_put(&event->kref, __release_event);
2226 }
2227 EXPORT_SYMBOL(ceph_osdc_put_event);
2228 
2229 static void __insert_event(struct ceph_osd_client *osdc,
2230 			     struct ceph_osd_event *new)
2231 {
2232 	struct rb_node **p = &osdc->event_tree.rb_node;
2233 	struct rb_node *parent = NULL;
2234 	struct ceph_osd_event *event = NULL;
2235 
2236 	while (*p) {
2237 		parent = *p;
2238 		event = rb_entry(parent, struct ceph_osd_event, node);
2239 		if (new->cookie < event->cookie)
2240 			p = &(*p)->rb_left;
2241 		else if (new->cookie > event->cookie)
2242 			p = &(*p)->rb_right;
2243 		else
2244 			BUG();
2245 	}
2246 
2247 	rb_link_node(&new->node, parent, p);
2248 	rb_insert_color(&new->node, &osdc->event_tree);
2249 }
2250 
2251 static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
2252 					        u64 cookie)
2253 {
2254 	struct rb_node **p = &osdc->event_tree.rb_node;
2255 	struct rb_node *parent = NULL;
2256 	struct ceph_osd_event *event = NULL;
2257 
2258 	while (*p) {
2259 		parent = *p;
2260 		event = rb_entry(parent, struct ceph_osd_event, node);
2261 		if (cookie < event->cookie)
2262 			p = &(*p)->rb_left;
2263 		else if (cookie > event->cookie)
2264 			p = &(*p)->rb_right;
2265 		else
2266 			return event;
2267 	}
2268 	return NULL;
2269 }
2270 
2271 static void __remove_event(struct ceph_osd_event *event)
2272 {
2273 	struct ceph_osd_client *osdc = event->osdc;
2274 
2275 	if (!RB_EMPTY_NODE(&event->node)) {
2276 		dout("__remove_event removed %p\n", event);
2277 		rb_erase(&event->node, &osdc->event_tree);
2278 		ceph_osdc_put_event(event);
2279 	} else {
2280 		dout("__remove_event didn't remove %p\n", event);
2281 	}
2282 }
2283 
2284 int ceph_osdc_create_event(struct ceph_osd_client *osdc,
2285 			   void (*event_cb)(u64, u64, u8, void *),
2286 			   void *data, struct ceph_osd_event **pevent)
2287 {
2288 	struct ceph_osd_event *event;
2289 
2290 	event = kmalloc(sizeof(*event), GFP_NOIO);
2291 	if (!event)
2292 		return -ENOMEM;
2293 
2294 	dout("create_event %p\n", event);
2295 	event->cb = event_cb;
2296 	event->one_shot = 0;
2297 	event->data = data;
2298 	event->osdc = osdc;
2299 	INIT_LIST_HEAD(&event->osd_node);
2300 	RB_CLEAR_NODE(&event->node);
2301 	kref_init(&event->kref);   /* one ref for us */
2302 	kref_get(&event->kref);    /* one ref for the caller */
2303 
2304 	spin_lock(&osdc->event_lock);
2305 	event->cookie = ++osdc->event_count;
2306 	__insert_event(osdc, event);
2307 	spin_unlock(&osdc->event_lock);
2308 
2309 	*pevent = event;
2310 	return 0;
2311 }
2312 EXPORT_SYMBOL(ceph_osdc_create_event);
2313 
2314 void ceph_osdc_cancel_event(struct ceph_osd_event *event)
2315 {
2316 	struct ceph_osd_client *osdc = event->osdc;
2317 
2318 	dout("cancel_event %p\n", event);
2319 	spin_lock(&osdc->event_lock);
2320 	__remove_event(event);
2321 	spin_unlock(&osdc->event_lock);
2322 	ceph_osdc_put_event(event); /* caller's */
2323 }
2324 EXPORT_SYMBOL(ceph_osdc_cancel_event);
2325 
2326 
2327 static void do_event_work(struct work_struct *work)
2328 {
2329 	struct ceph_osd_event_work *event_work =
2330 		container_of(work, struct ceph_osd_event_work, work);
2331 	struct ceph_osd_event *event = event_work->event;
2332 	u64 ver = event_work->ver;
2333 	u64 notify_id = event_work->notify_id;
2334 	u8 opcode = event_work->opcode;
2335 
2336 	dout("do_event_work completing %p\n", event);
2337 	event->cb(ver, notify_id, opcode, event->data);
2338 	dout("do_event_work completed %p\n", event);
2339 	ceph_osdc_put_event(event);
2340 	kfree(event_work);
2341 }
2342 
2343 
2344 /*
2345  * Process osd watch notifications
2346  */
2347 static void handle_watch_notify(struct ceph_osd_client *osdc,
2348 				struct ceph_msg *msg)
2349 {
2350 	void *p, *end;
2351 	u8 proto_ver;
2352 	u64 cookie, ver, notify_id;
2353 	u8 opcode;
2354 	struct ceph_osd_event *event;
2355 	struct ceph_osd_event_work *event_work;
2356 
2357 	p = msg->front.iov_base;
2358 	end = p + msg->front.iov_len;
2359 
2360 	ceph_decode_8_safe(&p, end, proto_ver, bad);
2361 	ceph_decode_8_safe(&p, end, opcode, bad);
2362 	ceph_decode_64_safe(&p, end, cookie, bad);
2363 	ceph_decode_64_safe(&p, end, ver, bad);
2364 	ceph_decode_64_safe(&p, end, notify_id, bad);
2365 
2366 	spin_lock(&osdc->event_lock);
2367 	event = __find_event(osdc, cookie);
2368 	if (event) {
2369 		BUG_ON(event->one_shot);
2370 		get_event(event);
2371 	}
2372 	spin_unlock(&osdc->event_lock);
2373 	dout("handle_watch_notify cookie %lld ver %lld event %p\n",
2374 	     cookie, ver, event);
2375 	if (event) {
2376 		event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
2377 		if (!event_work) {
2378 			pr_err("couldn't allocate event_work\n");
2379 			ceph_osdc_put_event(event);
2380 			return;
2381 		}
2382 		INIT_WORK(&event_work->work, do_event_work);
2383 		event_work->event = event;
2384 		event_work->ver = ver;
2385 		event_work->notify_id = notify_id;
2386 		event_work->opcode = opcode;
2387 
2388 		queue_work(osdc->notify_wq, &event_work->work);
2389 	}
2390 
2391 	return;
2392 
2393 bad:
2394 	pr_err("osdc handle_watch_notify corrupt msg\n");
2395 }
2396 
2397 /*
2398  * build new request AND message
2399  *
2400  */
2401 void ceph_osdc_build_request(struct ceph_osd_request *req, u64 off,
2402 				struct ceph_snap_context *snapc, u64 snap_id,
2403 				struct timespec *mtime)
2404 {
2405 	struct ceph_msg *msg = req->r_request;
2406 	void *p;
2407 	size_t msg_size;
2408 	int flags = req->r_flags;
2409 	u64 data_len;
2410 	unsigned int i;
2411 
2412 	req->r_snapid = snap_id;
2413 	req->r_snapc = ceph_get_snap_context(snapc);
2414 
2415 	/* encode request */
2416 	msg->hdr.version = cpu_to_le16(4);
2417 
2418 	p = msg->front.iov_base;
2419 	ceph_encode_32(&p, 1);   /* client_inc  is always 1 */
2420 	req->r_request_osdmap_epoch = p;
2421 	p += 4;
2422 	req->r_request_flags = p;
2423 	p += 4;
2424 	if (req->r_flags & CEPH_OSD_FLAG_WRITE)
2425 		ceph_encode_timespec(p, mtime);
2426 	p += sizeof(struct ceph_timespec);
2427 	req->r_request_reassert_version = p;
2428 	p += sizeof(struct ceph_eversion); /* will get filled in */
2429 
2430 	/* oloc */
2431 	ceph_encode_8(&p, 4);
2432 	ceph_encode_8(&p, 4);
2433 	ceph_encode_32(&p, 8 + 4 + 4);
2434 	req->r_request_pool = p;
2435 	p += 8;
2436 	ceph_encode_32(&p, -1);  /* preferred */
2437 	ceph_encode_32(&p, 0);   /* key len */
2438 
2439 	ceph_encode_8(&p, 1);
2440 	req->r_request_pgid = p;
2441 	p += 8 + 4;
2442 	ceph_encode_32(&p, -1);  /* preferred */
2443 
2444 	/* oid */
2445 	ceph_encode_32(&p, req->r_base_oid.name_len);
2446 	memcpy(p, req->r_base_oid.name, req->r_base_oid.name_len);
2447 	dout("oid '%.*s' len %d\n", req->r_base_oid.name_len,
2448 	     req->r_base_oid.name, req->r_base_oid.name_len);
2449 	p += req->r_base_oid.name_len;
2450 
2451 	/* ops--can imply data */
2452 	ceph_encode_16(&p, (u16)req->r_num_ops);
2453 	data_len = 0;
2454 	for (i = 0; i < req->r_num_ops; i++) {
2455 		data_len += osd_req_encode_op(req, p, i);
2456 		p += sizeof(struct ceph_osd_op);
2457 	}
2458 
2459 	/* snaps */
2460 	ceph_encode_64(&p, req->r_snapid);
2461 	ceph_encode_64(&p, req->r_snapc ? req->r_snapc->seq : 0);
2462 	ceph_encode_32(&p, req->r_snapc ? req->r_snapc->num_snaps : 0);
2463 	if (req->r_snapc) {
2464 		for (i = 0; i < snapc->num_snaps; i++) {
2465 			ceph_encode_64(&p, req->r_snapc->snaps[i]);
2466 		}
2467 	}
2468 
2469 	req->r_request_attempts = p;
2470 	p += 4;
2471 
2472 	/* data */
2473 	if (flags & CEPH_OSD_FLAG_WRITE) {
2474 		u16 data_off;
2475 
2476 		/*
2477 		 * The header "data_off" is a hint to the receiver
2478 		 * allowing it to align received data into its
2479 		 * buffers such that there's no need to re-copy
2480 		 * it before writing it to disk (direct I/O).
2481 		 */
2482 		data_off = (u16) (off & 0xffff);
2483 		req->r_request->hdr.data_off = cpu_to_le16(data_off);
2484 	}
2485 	req->r_request->hdr.data_len = cpu_to_le32(data_len);
2486 
2487 	BUG_ON(p > msg->front.iov_base + msg->front.iov_len);
2488 	msg_size = p - msg->front.iov_base;
2489 	msg->front.iov_len = msg_size;
2490 	msg->hdr.front_len = cpu_to_le32(msg_size);
2491 
2492 	dout("build_request msg_size was %d\n", (int)msg_size);
2493 }
2494 EXPORT_SYMBOL(ceph_osdc_build_request);
2495 
2496 /*
2497  * Register request, send initial attempt.
2498  */
2499 int ceph_osdc_start_request(struct ceph_osd_client *osdc,
2500 			    struct ceph_osd_request *req,
2501 			    bool nofail)
2502 {
2503 	int rc;
2504 
2505 	down_read(&osdc->map_sem);
2506 	mutex_lock(&osdc->request_mutex);
2507 
2508 	rc = __ceph_osdc_start_request(osdc, req, nofail);
2509 
2510 	mutex_unlock(&osdc->request_mutex);
2511 	up_read(&osdc->map_sem);
2512 
2513 	return rc;
2514 }
2515 EXPORT_SYMBOL(ceph_osdc_start_request);
2516 
2517 /*
2518  * Unregister a registered request.  The request is not completed (i.e.
2519  * no callbacks or wakeups) - higher layers are supposed to know what
2520  * they are canceling.
2521  */
2522 void ceph_osdc_cancel_request(struct ceph_osd_request *req)
2523 {
2524 	struct ceph_osd_client *osdc = req->r_osdc;
2525 
2526 	mutex_lock(&osdc->request_mutex);
2527 	if (req->r_linger)
2528 		__unregister_linger_request(osdc, req);
2529 	__unregister_request(osdc, req);
2530 	mutex_unlock(&osdc->request_mutex);
2531 
2532 	dout("%s %p tid %llu canceled\n", __func__, req, req->r_tid);
2533 }
2534 EXPORT_SYMBOL(ceph_osdc_cancel_request);
2535 
2536 /*
2537  * wait for a request to complete
2538  */
2539 int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
2540 			   struct ceph_osd_request *req)
2541 {
2542 	int rc;
2543 
2544 	dout("%s %p tid %llu\n", __func__, req, req->r_tid);
2545 
2546 	rc = wait_for_completion_interruptible(&req->r_completion);
2547 	if (rc < 0) {
2548 		dout("%s %p tid %llu interrupted\n", __func__, req, req->r_tid);
2549 		ceph_osdc_cancel_request(req);
2550 		complete_request(req);
2551 		return rc;
2552 	}
2553 
2554 	dout("%s %p tid %llu result %d\n", __func__, req, req->r_tid,
2555 	     req->r_result);
2556 	return req->r_result;
2557 }
2558 EXPORT_SYMBOL(ceph_osdc_wait_request);
2559 
2560 /*
2561  * sync - wait for all in-flight requests to flush.  avoid starvation.
2562  */
2563 void ceph_osdc_sync(struct ceph_osd_client *osdc)
2564 {
2565 	struct ceph_osd_request *req;
2566 	u64 last_tid, next_tid = 0;
2567 
2568 	mutex_lock(&osdc->request_mutex);
2569 	last_tid = osdc->last_tid;
2570 	while (1) {
2571 		req = __lookup_request_ge(osdc, next_tid);
2572 		if (!req)
2573 			break;
2574 		if (req->r_tid > last_tid)
2575 			break;
2576 
2577 		next_tid = req->r_tid + 1;
2578 		if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
2579 			continue;
2580 
2581 		ceph_osdc_get_request(req);
2582 		mutex_unlock(&osdc->request_mutex);
2583 		dout("sync waiting on tid %llu (last is %llu)\n",
2584 		     req->r_tid, last_tid);
2585 		wait_for_completion(&req->r_safe_completion);
2586 		mutex_lock(&osdc->request_mutex);
2587 		ceph_osdc_put_request(req);
2588 	}
2589 	mutex_unlock(&osdc->request_mutex);
2590 	dout("sync done (thru tid %llu)\n", last_tid);
2591 }
2592 EXPORT_SYMBOL(ceph_osdc_sync);
2593 
2594 /*
2595  * Call all pending notify callbacks - for use after a watch is
2596  * unregistered, to make sure no more callbacks for it will be invoked
2597  */
2598 void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc)
2599 {
2600 	flush_workqueue(osdc->notify_wq);
2601 }
2602 EXPORT_SYMBOL(ceph_osdc_flush_notifies);
2603 
2604 
2605 /*
2606  * init, shutdown
2607  */
2608 int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
2609 {
2610 	int err;
2611 
2612 	dout("init\n");
2613 	osdc->client = client;
2614 	osdc->osdmap = NULL;
2615 	init_rwsem(&osdc->map_sem);
2616 	init_completion(&osdc->map_waiters);
2617 	osdc->last_requested_map = 0;
2618 	mutex_init(&osdc->request_mutex);
2619 	osdc->last_tid = 0;
2620 	osdc->osds = RB_ROOT;
2621 	INIT_LIST_HEAD(&osdc->osd_lru);
2622 	osdc->requests = RB_ROOT;
2623 	INIT_LIST_HEAD(&osdc->req_lru);
2624 	INIT_LIST_HEAD(&osdc->req_unsent);
2625 	INIT_LIST_HEAD(&osdc->req_notarget);
2626 	INIT_LIST_HEAD(&osdc->req_linger);
2627 	osdc->num_requests = 0;
2628 	INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
2629 	INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
2630 	spin_lock_init(&osdc->event_lock);
2631 	osdc->event_tree = RB_ROOT;
2632 	osdc->event_count = 0;
2633 
2634 	schedule_delayed_work(&osdc->osds_timeout_work,
2635 	    round_jiffies_relative(osdc->client->options->osd_idle_ttl));
2636 
2637 	err = -ENOMEM;
2638 	osdc->req_mempool = mempool_create_kmalloc_pool(10,
2639 					sizeof(struct ceph_osd_request));
2640 	if (!osdc->req_mempool)
2641 		goto out;
2642 
2643 	err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
2644 				OSD_OP_FRONT_LEN, 10, true,
2645 				"osd_op");
2646 	if (err < 0)
2647 		goto out_mempool;
2648 	err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
2649 				OSD_OPREPLY_FRONT_LEN, 10, true,
2650 				"osd_op_reply");
2651 	if (err < 0)
2652 		goto out_msgpool;
2653 
2654 	err = -ENOMEM;
2655 	osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
2656 	if (!osdc->notify_wq)
2657 		goto out_msgpool_reply;
2658 
2659 	return 0;
2660 
2661 out_msgpool_reply:
2662 	ceph_msgpool_destroy(&osdc->msgpool_op_reply);
2663 out_msgpool:
2664 	ceph_msgpool_destroy(&osdc->msgpool_op);
2665 out_mempool:
2666 	mempool_destroy(osdc->req_mempool);
2667 out:
2668 	return err;
2669 }
2670 
2671 void ceph_osdc_stop(struct ceph_osd_client *osdc)
2672 {
2673 	flush_workqueue(osdc->notify_wq);
2674 	destroy_workqueue(osdc->notify_wq);
2675 	cancel_delayed_work_sync(&osdc->timeout_work);
2676 	cancel_delayed_work_sync(&osdc->osds_timeout_work);
2677 	if (osdc->osdmap) {
2678 		ceph_osdmap_destroy(osdc->osdmap);
2679 		osdc->osdmap = NULL;
2680 	}
2681 	remove_all_osds(osdc);
2682 	mempool_destroy(osdc->req_mempool);
2683 	ceph_msgpool_destroy(&osdc->msgpool_op);
2684 	ceph_msgpool_destroy(&osdc->msgpool_op_reply);
2685 }
2686 
2687 /*
2688  * Read some contiguous pages.  If we cross a stripe boundary, shorten
2689  * *plen.  Return number of bytes read, or error.
2690  */
2691 int ceph_osdc_readpages(struct ceph_osd_client *osdc,
2692 			struct ceph_vino vino, struct ceph_file_layout *layout,
2693 			u64 off, u64 *plen,
2694 			u32 truncate_seq, u64 truncate_size,
2695 			struct page **pages, int num_pages, int page_align)
2696 {
2697 	struct ceph_osd_request *req;
2698 	int rc = 0;
2699 
2700 	dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
2701 	     vino.snap, off, *plen);
2702 	req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 0, 1,
2703 				    CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
2704 				    NULL, truncate_seq, truncate_size,
2705 				    false);
2706 	if (IS_ERR(req))
2707 		return PTR_ERR(req);
2708 
2709 	/* it may be a short read due to an object boundary */
2710 
2711 	osd_req_op_extent_osd_data_pages(req, 0,
2712 				pages, *plen, page_align, false, false);
2713 
2714 	dout("readpages  final extent is %llu~%llu (%llu bytes align %d)\n",
2715 	     off, *plen, *plen, page_align);
2716 
2717 	ceph_osdc_build_request(req, off, NULL, vino.snap, NULL);
2718 
2719 	rc = ceph_osdc_start_request(osdc, req, false);
2720 	if (!rc)
2721 		rc = ceph_osdc_wait_request(osdc, req);
2722 
2723 	ceph_osdc_put_request(req);
2724 	dout("readpages result %d\n", rc);
2725 	return rc;
2726 }
2727 EXPORT_SYMBOL(ceph_osdc_readpages);
2728 
2729 /*
2730  * do a synchronous write on N pages
2731  */
2732 int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
2733 			 struct ceph_file_layout *layout,
2734 			 struct ceph_snap_context *snapc,
2735 			 u64 off, u64 len,
2736 			 u32 truncate_seq, u64 truncate_size,
2737 			 struct timespec *mtime,
2738 			 struct page **pages, int num_pages)
2739 {
2740 	struct ceph_osd_request *req;
2741 	int rc = 0;
2742 	int page_align = off & ~PAGE_MASK;
2743 
2744 	BUG_ON(vino.snap != CEPH_NOSNAP);	/* snapshots aren't writeable */
2745 	req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 0, 1,
2746 				    CEPH_OSD_OP_WRITE,
2747 				    CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
2748 				    snapc, truncate_seq, truncate_size,
2749 				    true);
2750 	if (IS_ERR(req))
2751 		return PTR_ERR(req);
2752 
2753 	/* it may be a short write due to an object boundary */
2754 	osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align,
2755 				false, false);
2756 	dout("writepages %llu~%llu (%llu bytes)\n", off, len, len);
2757 
2758 	ceph_osdc_build_request(req, off, snapc, CEPH_NOSNAP, mtime);
2759 
2760 	rc = ceph_osdc_start_request(osdc, req, true);
2761 	if (!rc)
2762 		rc = ceph_osdc_wait_request(osdc, req);
2763 
2764 	ceph_osdc_put_request(req);
2765 	if (rc == 0)
2766 		rc = len;
2767 	dout("writepages result %d\n", rc);
2768 	return rc;
2769 }
2770 EXPORT_SYMBOL(ceph_osdc_writepages);
2771 
2772 int ceph_osdc_setup(void)
2773 {
2774 	BUG_ON(ceph_osd_request_cache);
2775 	ceph_osd_request_cache = kmem_cache_create("ceph_osd_request",
2776 					sizeof (struct ceph_osd_request),
2777 					__alignof__(struct ceph_osd_request),
2778 					0, NULL);
2779 
2780 	return ceph_osd_request_cache ? 0 : -ENOMEM;
2781 }
2782 EXPORT_SYMBOL(ceph_osdc_setup);
2783 
2784 void ceph_osdc_cleanup(void)
2785 {
2786 	BUG_ON(!ceph_osd_request_cache);
2787 	kmem_cache_destroy(ceph_osd_request_cache);
2788 	ceph_osd_request_cache = NULL;
2789 }
2790 EXPORT_SYMBOL(ceph_osdc_cleanup);
2791 
2792 /*
2793  * handle incoming message
2794  */
2795 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
2796 {
2797 	struct ceph_osd *osd = con->private;
2798 	struct ceph_osd_client *osdc;
2799 	int type = le16_to_cpu(msg->hdr.type);
2800 
2801 	if (!osd)
2802 		goto out;
2803 	osdc = osd->o_osdc;
2804 
2805 	switch (type) {
2806 	case CEPH_MSG_OSD_MAP:
2807 		ceph_osdc_handle_map(osdc, msg);
2808 		break;
2809 	case CEPH_MSG_OSD_OPREPLY:
2810 		handle_reply(osdc, msg, con);
2811 		break;
2812 	case CEPH_MSG_WATCH_NOTIFY:
2813 		handle_watch_notify(osdc, msg);
2814 		break;
2815 
2816 	default:
2817 		pr_err("received unknown message type %d %s\n", type,
2818 		       ceph_msg_type_name(type));
2819 	}
2820 out:
2821 	ceph_msg_put(msg);
2822 }
2823 
2824 /*
2825  * Lookup and return message for incoming reply.  Don't try to do
2826  * anything about a larger than preallocated data portion of the
2827  * message at the moment - for now, just skip the message.
2828  */
2829 static struct ceph_msg *get_reply(struct ceph_connection *con,
2830 				  struct ceph_msg_header *hdr,
2831 				  int *skip)
2832 {
2833 	struct ceph_osd *osd = con->private;
2834 	struct ceph_osd_client *osdc = osd->o_osdc;
2835 	struct ceph_msg *m;
2836 	struct ceph_osd_request *req;
2837 	int front_len = le32_to_cpu(hdr->front_len);
2838 	int data_len = le32_to_cpu(hdr->data_len);
2839 	u64 tid;
2840 
2841 	tid = le64_to_cpu(hdr->tid);
2842 	mutex_lock(&osdc->request_mutex);
2843 	req = __lookup_request(osdc, tid);
2844 	if (!req) {
2845 		pr_warn("%s osd%d tid %llu unknown, skipping\n",
2846 			__func__, osd->o_osd, tid);
2847 		m = NULL;
2848 		*skip = 1;
2849 		goto out;
2850 	}
2851 
2852 	if (req->r_reply->con)
2853 		dout("%s revoking msg %p from old con %p\n", __func__,
2854 		     req->r_reply, req->r_reply->con);
2855 	ceph_msg_revoke_incoming(req->r_reply);
2856 
2857 	if (front_len > req->r_reply->front_alloc_len) {
2858 		pr_warn("%s osd%d tid %llu front %d > preallocated %d\n",
2859 			__func__, osd->o_osd, req->r_tid, front_len,
2860 			req->r_reply->front_alloc_len);
2861 		m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS,
2862 				 false);
2863 		if (!m)
2864 			goto out;
2865 		ceph_msg_put(req->r_reply);
2866 		req->r_reply = m;
2867 	}
2868 
2869 	if (data_len > req->r_reply->data_length) {
2870 		pr_warn("%s osd%d tid %llu data %d > preallocated %zu, skipping\n",
2871 			__func__, osd->o_osd, req->r_tid, data_len,
2872 			req->r_reply->data_length);
2873 		m = NULL;
2874 		*skip = 1;
2875 		goto out;
2876 	}
2877 
2878 	m = ceph_msg_get(req->r_reply);
2879 	dout("get_reply tid %lld %p\n", tid, m);
2880 
2881 out:
2882 	mutex_unlock(&osdc->request_mutex);
2883 	return m;
2884 }
2885 
2886 static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2887 				  struct ceph_msg_header *hdr,
2888 				  int *skip)
2889 {
2890 	struct ceph_osd *osd = con->private;
2891 	int type = le16_to_cpu(hdr->type);
2892 	int front = le32_to_cpu(hdr->front_len);
2893 
2894 	*skip = 0;
2895 	switch (type) {
2896 	case CEPH_MSG_OSD_MAP:
2897 	case CEPH_MSG_WATCH_NOTIFY:
2898 		return ceph_msg_new(type, front, GFP_NOFS, false);
2899 	case CEPH_MSG_OSD_OPREPLY:
2900 		return get_reply(con, hdr, skip);
2901 	default:
2902 		pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
2903 			osd->o_osd);
2904 		*skip = 1;
2905 		return NULL;
2906 	}
2907 }
2908 
2909 /*
2910  * Wrappers to refcount containing ceph_osd struct
2911  */
2912 static struct ceph_connection *get_osd_con(struct ceph_connection *con)
2913 {
2914 	struct ceph_osd *osd = con->private;
2915 	if (get_osd(osd))
2916 		return con;
2917 	return NULL;
2918 }
2919 
2920 static void put_osd_con(struct ceph_connection *con)
2921 {
2922 	struct ceph_osd *osd = con->private;
2923 	put_osd(osd);
2924 }
2925 
2926 /*
2927  * authentication
2928  */
2929 /*
2930  * Note: returned pointer is the address of a structure that's
2931  * managed separately.  Caller must *not* attempt to free it.
2932  */
2933 static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
2934 					int *proto, int force_new)
2935 {
2936 	struct ceph_osd *o = con->private;
2937 	struct ceph_osd_client *osdc = o->o_osdc;
2938 	struct ceph_auth_client *ac = osdc->client->monc.auth;
2939 	struct ceph_auth_handshake *auth = &o->o_auth;
2940 
2941 	if (force_new && auth->authorizer) {
2942 		ceph_auth_destroy_authorizer(ac, auth->authorizer);
2943 		auth->authorizer = NULL;
2944 	}
2945 	if (!auth->authorizer) {
2946 		int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2947 						      auth);
2948 		if (ret)
2949 			return ERR_PTR(ret);
2950 	} else {
2951 		int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2952 						     auth);
2953 		if (ret)
2954 			return ERR_PTR(ret);
2955 	}
2956 	*proto = ac->protocol;
2957 
2958 	return auth;
2959 }
2960 
2961 
2962 static int verify_authorizer_reply(struct ceph_connection *con, int len)
2963 {
2964 	struct ceph_osd *o = con->private;
2965 	struct ceph_osd_client *osdc = o->o_osdc;
2966 	struct ceph_auth_client *ac = osdc->client->monc.auth;
2967 
2968 	return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer, len);
2969 }
2970 
2971 static int invalidate_authorizer(struct ceph_connection *con)
2972 {
2973 	struct ceph_osd *o = con->private;
2974 	struct ceph_osd_client *osdc = o->o_osdc;
2975 	struct ceph_auth_client *ac = osdc->client->monc.auth;
2976 
2977 	ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
2978 	return ceph_monc_validate_auth(&osdc->client->monc);
2979 }
2980 
2981 static int sign_message(struct ceph_connection *con, struct ceph_msg *msg)
2982 {
2983 	struct ceph_osd *o = con->private;
2984 	struct ceph_auth_handshake *auth = &o->o_auth;
2985 	return ceph_auth_sign_message(auth, msg);
2986 }
2987 
2988 static int check_message_signature(struct ceph_connection *con, struct ceph_msg *msg)
2989 {
2990 	struct ceph_osd *o = con->private;
2991 	struct ceph_auth_handshake *auth = &o->o_auth;
2992 	return ceph_auth_check_message_signature(auth, msg);
2993 }
2994 
2995 static const struct ceph_connection_operations osd_con_ops = {
2996 	.get = get_osd_con,
2997 	.put = put_osd_con,
2998 	.dispatch = dispatch,
2999 	.get_authorizer = get_authorizer,
3000 	.verify_authorizer_reply = verify_authorizer_reply,
3001 	.invalidate_authorizer = invalidate_authorizer,
3002 	.alloc_msg = alloc_msg,
3003 	.sign_message = sign_message,
3004 	.check_message_signature = check_message_signature,
3005 	.fault = osd_reset,
3006 };
3007