xref: /linux/net/ceph/osd_client.c (revision 26b0d14106954ae46d2f4f7eec3481828a210f7d)
1 #include <linux/ceph/ceph_debug.h>
2 
3 #include <linux/module.h>
4 #include <linux/err.h>
5 #include <linux/highmem.h>
6 #include <linux/mm.h>
7 #include <linux/pagemap.h>
8 #include <linux/slab.h>
9 #include <linux/uaccess.h>
10 #ifdef CONFIG_BLOCK
11 #include <linux/bio.h>
12 #endif
13 
14 #include <linux/ceph/libceph.h>
15 #include <linux/ceph/osd_client.h>
16 #include <linux/ceph/messenger.h>
17 #include <linux/ceph/decode.h>
18 #include <linux/ceph/auth.h>
19 #include <linux/ceph/pagelist.h>
20 
21 #define OSD_OP_FRONT_LEN	4096
22 #define OSD_OPREPLY_FRONT_LEN	512
23 
24 static const struct ceph_connection_operations osd_con_ops;
25 
26 static void send_queued(struct ceph_osd_client *osdc);
27 static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd);
28 static void __register_request(struct ceph_osd_client *osdc,
29 			       struct ceph_osd_request *req);
30 static void __unregister_linger_request(struct ceph_osd_client *osdc,
31 					struct ceph_osd_request *req);
32 static void __send_request(struct ceph_osd_client *osdc,
33 			   struct ceph_osd_request *req);
34 
35 static int op_needs_trail(int op)
36 {
37 	switch (op) {
38 	case CEPH_OSD_OP_GETXATTR:
39 	case CEPH_OSD_OP_SETXATTR:
40 	case CEPH_OSD_OP_CMPXATTR:
41 	case CEPH_OSD_OP_CALL:
42 	case CEPH_OSD_OP_NOTIFY:
43 		return 1;
44 	default:
45 		return 0;
46 	}
47 }
48 
49 static int op_has_extent(int op)
50 {
51 	return (op == CEPH_OSD_OP_READ ||
52 		op == CEPH_OSD_OP_WRITE);
53 }
54 
55 void ceph_calc_raw_layout(struct ceph_osd_client *osdc,
56 			struct ceph_file_layout *layout,
57 			u64 snapid,
58 			u64 off, u64 *plen, u64 *bno,
59 			struct ceph_osd_request *req,
60 			struct ceph_osd_req_op *op)
61 {
62 	struct ceph_osd_request_head *reqhead = req->r_request->front.iov_base;
63 	u64 orig_len = *plen;
64 	u64 objoff, objlen;    /* extent in object */
65 
66 	reqhead->snapid = cpu_to_le64(snapid);
67 
68 	/* object extent? */
69 	ceph_calc_file_object_mapping(layout, off, plen, bno,
70 				      &objoff, &objlen);
71 	if (*plen < orig_len)
72 		dout(" skipping last %llu, final file extent %llu~%llu\n",
73 		     orig_len - *plen, off, *plen);
74 
75 	if (op_has_extent(op->op)) {
76 		op->extent.offset = objoff;
77 		op->extent.length = objlen;
78 	}
79 	req->r_num_pages = calc_pages_for(off, *plen);
80 	req->r_page_alignment = off & ~PAGE_MASK;
81 	if (op->op == CEPH_OSD_OP_WRITE)
82 		op->payload_len = *plen;
83 
84 	dout("calc_layout bno=%llx %llu~%llu (%d pages)\n",
85 	     *bno, objoff, objlen, req->r_num_pages);
86 
87 }
88 EXPORT_SYMBOL(ceph_calc_raw_layout);
89 
90 /*
91  * Implement client access to distributed object storage cluster.
92  *
93  * All data objects are stored within a cluster/cloud of OSDs, or
94  * "object storage devices."  (Note that Ceph OSDs have _nothing_ to
95  * do with the T10 OSD extensions to SCSI.)  Ceph OSDs are simply
96  * remote daemons serving up and coordinating consistent and safe
97  * access to storage.
98  *
99  * Cluster membership and the mapping of data objects onto storage devices
100  * are described by the osd map.
101  *
102  * We keep track of pending OSD requests (read, write), resubmit
103  * requests to different OSDs when the cluster topology/data layout
104  * change, or retry the affected requests when the communications
105  * channel with an OSD is reset.
106  */
107 
108 /*
109  * calculate the mapping of a file extent onto an object, and fill out the
110  * request accordingly.  shorten extent as necessary if it crosses an
111  * object boundary.
112  *
113  * fill osd op in request message.
114  */
115 static void calc_layout(struct ceph_osd_client *osdc,
116 			struct ceph_vino vino,
117 			struct ceph_file_layout *layout,
118 			u64 off, u64 *plen,
119 			struct ceph_osd_request *req,
120 			struct ceph_osd_req_op *op)
121 {
122 	u64 bno;
123 
124 	ceph_calc_raw_layout(osdc, layout, vino.snap, off,
125 			     plen, &bno, req, op);
126 
127 	snprintf(req->r_oid, sizeof(req->r_oid), "%llx.%08llx", vino.ino, bno);
128 	req->r_oid_len = strlen(req->r_oid);
129 }
130 
131 /*
132  * requests
133  */
134 void ceph_osdc_release_request(struct kref *kref)
135 {
136 	struct ceph_osd_request *req = container_of(kref,
137 						    struct ceph_osd_request,
138 						    r_kref);
139 
140 	if (req->r_request)
141 		ceph_msg_put(req->r_request);
142 	if (req->r_con_filling_msg) {
143 		dout("release_request revoking pages %p from con %p\n",
144 		     req->r_pages, req->r_con_filling_msg);
145 		ceph_con_revoke_message(req->r_con_filling_msg,
146 				      req->r_reply);
147 		req->r_con_filling_msg->ops->put(req->r_con_filling_msg);
148 	}
149 	if (req->r_reply)
150 		ceph_msg_put(req->r_reply);
151 	if (req->r_own_pages)
152 		ceph_release_page_vector(req->r_pages,
153 					 req->r_num_pages);
154 #ifdef CONFIG_BLOCK
155 	if (req->r_bio)
156 		bio_put(req->r_bio);
157 #endif
158 	ceph_put_snap_context(req->r_snapc);
159 	if (req->r_trail) {
160 		ceph_pagelist_release(req->r_trail);
161 		kfree(req->r_trail);
162 	}
163 	if (req->r_mempool)
164 		mempool_free(req, req->r_osdc->req_mempool);
165 	else
166 		kfree(req);
167 }
168 EXPORT_SYMBOL(ceph_osdc_release_request);
169 
170 static int get_num_ops(struct ceph_osd_req_op *ops, int *needs_trail)
171 {
172 	int i = 0;
173 
174 	if (needs_trail)
175 		*needs_trail = 0;
176 	while (ops[i].op) {
177 		if (needs_trail && op_needs_trail(ops[i].op))
178 			*needs_trail = 1;
179 		i++;
180 	}
181 
182 	return i;
183 }
184 
185 struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
186 					       int flags,
187 					       struct ceph_snap_context *snapc,
188 					       struct ceph_osd_req_op *ops,
189 					       bool use_mempool,
190 					       gfp_t gfp_flags,
191 					       struct page **pages,
192 					       struct bio *bio)
193 {
194 	struct ceph_osd_request *req;
195 	struct ceph_msg *msg;
196 	int needs_trail;
197 	int num_op = get_num_ops(ops, &needs_trail);
198 	size_t msg_size = sizeof(struct ceph_osd_request_head);
199 
200 	msg_size += num_op*sizeof(struct ceph_osd_op);
201 
202 	if (use_mempool) {
203 		req = mempool_alloc(osdc->req_mempool, gfp_flags);
204 		memset(req, 0, sizeof(*req));
205 	} else {
206 		req = kzalloc(sizeof(*req), gfp_flags);
207 	}
208 	if (req == NULL)
209 		return NULL;
210 
211 	req->r_osdc = osdc;
212 	req->r_mempool = use_mempool;
213 
214 	kref_init(&req->r_kref);
215 	init_completion(&req->r_completion);
216 	init_completion(&req->r_safe_completion);
217 	INIT_LIST_HEAD(&req->r_unsafe_item);
218 	INIT_LIST_HEAD(&req->r_linger_item);
219 	INIT_LIST_HEAD(&req->r_linger_osd);
220 	INIT_LIST_HEAD(&req->r_req_lru_item);
221 	req->r_flags = flags;
222 
223 	WARN_ON((flags & (CEPH_OSD_FLAG_READ|CEPH_OSD_FLAG_WRITE)) == 0);
224 
225 	/* create reply message */
226 	if (use_mempool)
227 		msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0);
228 	else
229 		msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY,
230 				   OSD_OPREPLY_FRONT_LEN, gfp_flags, true);
231 	if (!msg) {
232 		ceph_osdc_put_request(req);
233 		return NULL;
234 	}
235 	req->r_reply = msg;
236 
237 	/* allocate space for the trailing data */
238 	if (needs_trail) {
239 		req->r_trail = kmalloc(sizeof(struct ceph_pagelist), gfp_flags);
240 		if (!req->r_trail) {
241 			ceph_osdc_put_request(req);
242 			return NULL;
243 		}
244 		ceph_pagelist_init(req->r_trail);
245 	}
246 	/* create request message; allow space for oid */
247 	msg_size += MAX_OBJ_NAME_SIZE;
248 	if (snapc)
249 		msg_size += sizeof(u64) * snapc->num_snaps;
250 	if (use_mempool)
251 		msg = ceph_msgpool_get(&osdc->msgpool_op, 0);
252 	else
253 		msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp_flags, true);
254 	if (!msg) {
255 		ceph_osdc_put_request(req);
256 		return NULL;
257 	}
258 
259 	msg->hdr.type = cpu_to_le16(CEPH_MSG_OSD_OP);
260 	memset(msg->front.iov_base, 0, msg->front.iov_len);
261 
262 	req->r_request = msg;
263 	req->r_pages = pages;
264 #ifdef CONFIG_BLOCK
265 	if (bio) {
266 		req->r_bio = bio;
267 		bio_get(req->r_bio);
268 	}
269 #endif
270 
271 	return req;
272 }
273 EXPORT_SYMBOL(ceph_osdc_alloc_request);
274 
275 static void osd_req_encode_op(struct ceph_osd_request *req,
276 			      struct ceph_osd_op *dst,
277 			      struct ceph_osd_req_op *src)
278 {
279 	dst->op = cpu_to_le16(src->op);
280 
281 	switch (src->op) {
282 	case CEPH_OSD_OP_READ:
283 	case CEPH_OSD_OP_WRITE:
284 		dst->extent.offset =
285 			cpu_to_le64(src->extent.offset);
286 		dst->extent.length =
287 			cpu_to_le64(src->extent.length);
288 		dst->extent.truncate_size =
289 			cpu_to_le64(src->extent.truncate_size);
290 		dst->extent.truncate_seq =
291 			cpu_to_le32(src->extent.truncate_seq);
292 		break;
293 
294 	case CEPH_OSD_OP_GETXATTR:
295 	case CEPH_OSD_OP_SETXATTR:
296 	case CEPH_OSD_OP_CMPXATTR:
297 		BUG_ON(!req->r_trail);
298 
299 		dst->xattr.name_len = cpu_to_le32(src->xattr.name_len);
300 		dst->xattr.value_len = cpu_to_le32(src->xattr.value_len);
301 		dst->xattr.cmp_op = src->xattr.cmp_op;
302 		dst->xattr.cmp_mode = src->xattr.cmp_mode;
303 		ceph_pagelist_append(req->r_trail, src->xattr.name,
304 				     src->xattr.name_len);
305 		ceph_pagelist_append(req->r_trail, src->xattr.val,
306 				     src->xattr.value_len);
307 		break;
308 	case CEPH_OSD_OP_CALL:
309 		BUG_ON(!req->r_trail);
310 
311 		dst->cls.class_len = src->cls.class_len;
312 		dst->cls.method_len = src->cls.method_len;
313 		dst->cls.indata_len = cpu_to_le32(src->cls.indata_len);
314 
315 		ceph_pagelist_append(req->r_trail, src->cls.class_name,
316 				     src->cls.class_len);
317 		ceph_pagelist_append(req->r_trail, src->cls.method_name,
318 				     src->cls.method_len);
319 		ceph_pagelist_append(req->r_trail, src->cls.indata,
320 				     src->cls.indata_len);
321 		break;
322 	case CEPH_OSD_OP_ROLLBACK:
323 		dst->snap.snapid = cpu_to_le64(src->snap.snapid);
324 		break;
325 	case CEPH_OSD_OP_STARTSYNC:
326 		break;
327 	case CEPH_OSD_OP_NOTIFY:
328 		{
329 			__le32 prot_ver = cpu_to_le32(src->watch.prot_ver);
330 			__le32 timeout = cpu_to_le32(src->watch.timeout);
331 
332 			BUG_ON(!req->r_trail);
333 
334 			ceph_pagelist_append(req->r_trail,
335 						&prot_ver, sizeof(prot_ver));
336 			ceph_pagelist_append(req->r_trail,
337 						&timeout, sizeof(timeout));
338 		}
339 	case CEPH_OSD_OP_NOTIFY_ACK:
340 	case CEPH_OSD_OP_WATCH:
341 		dst->watch.cookie = cpu_to_le64(src->watch.cookie);
342 		dst->watch.ver = cpu_to_le64(src->watch.ver);
343 		dst->watch.flag = src->watch.flag;
344 		break;
345 	default:
346 		pr_err("unrecognized osd opcode %d\n", dst->op);
347 		WARN_ON(1);
348 		break;
349 	}
350 	dst->payload_len = cpu_to_le32(src->payload_len);
351 }
352 
353 /*
354  * build new request AND message
355  *
356  */
357 void ceph_osdc_build_request(struct ceph_osd_request *req,
358 			     u64 off, u64 *plen,
359 			     struct ceph_osd_req_op *src_ops,
360 			     struct ceph_snap_context *snapc,
361 			     struct timespec *mtime,
362 			     const char *oid,
363 			     int oid_len)
364 {
365 	struct ceph_msg *msg = req->r_request;
366 	struct ceph_osd_request_head *head;
367 	struct ceph_osd_req_op *src_op;
368 	struct ceph_osd_op *op;
369 	void *p;
370 	int num_op = get_num_ops(src_ops, NULL);
371 	size_t msg_size = sizeof(*head) + num_op*sizeof(*op);
372 	int flags = req->r_flags;
373 	u64 data_len = 0;
374 	int i;
375 
376 	head = msg->front.iov_base;
377 	op = (void *)(head + 1);
378 	p = (void *)(op + num_op);
379 
380 	req->r_snapc = ceph_get_snap_context(snapc);
381 
382 	head->client_inc = cpu_to_le32(1); /* always, for now. */
383 	head->flags = cpu_to_le32(flags);
384 	if (flags & CEPH_OSD_FLAG_WRITE)
385 		ceph_encode_timespec(&head->mtime, mtime);
386 	head->num_ops = cpu_to_le16(num_op);
387 
388 
389 	/* fill in oid */
390 	head->object_len = cpu_to_le32(oid_len);
391 	memcpy(p, oid, oid_len);
392 	p += oid_len;
393 
394 	src_op = src_ops;
395 	while (src_op->op) {
396 		osd_req_encode_op(req, op, src_op);
397 		src_op++;
398 		op++;
399 	}
400 
401 	if (req->r_trail)
402 		data_len += req->r_trail->length;
403 
404 	if (snapc) {
405 		head->snap_seq = cpu_to_le64(snapc->seq);
406 		head->num_snaps = cpu_to_le32(snapc->num_snaps);
407 		for (i = 0; i < snapc->num_snaps; i++) {
408 			put_unaligned_le64(snapc->snaps[i], p);
409 			p += sizeof(u64);
410 		}
411 	}
412 
413 	if (flags & CEPH_OSD_FLAG_WRITE) {
414 		req->r_request->hdr.data_off = cpu_to_le16(off);
415 		req->r_request->hdr.data_len = cpu_to_le32(*plen + data_len);
416 	} else if (data_len) {
417 		req->r_request->hdr.data_off = 0;
418 		req->r_request->hdr.data_len = cpu_to_le32(data_len);
419 	}
420 
421 	req->r_request->page_alignment = req->r_page_alignment;
422 
423 	BUG_ON(p > msg->front.iov_base + msg->front.iov_len);
424 	msg_size = p - msg->front.iov_base;
425 	msg->front.iov_len = msg_size;
426 	msg->hdr.front_len = cpu_to_le32(msg_size);
427 	return;
428 }
429 EXPORT_SYMBOL(ceph_osdc_build_request);
430 
431 /*
432  * build new request AND message, calculate layout, and adjust file
433  * extent as needed.
434  *
435  * if the file was recently truncated, we include information about its
436  * old and new size so that the object can be updated appropriately.  (we
437  * avoid synchronously deleting truncated objects because it's slow.)
438  *
439  * if @do_sync, include a 'startsync' command so that the osd will flush
440  * data quickly.
441  */
442 struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
443 					       struct ceph_file_layout *layout,
444 					       struct ceph_vino vino,
445 					       u64 off, u64 *plen,
446 					       int opcode, int flags,
447 					       struct ceph_snap_context *snapc,
448 					       int do_sync,
449 					       u32 truncate_seq,
450 					       u64 truncate_size,
451 					       struct timespec *mtime,
452 					       bool use_mempool, int num_reply,
453 					       int page_align)
454 {
455 	struct ceph_osd_req_op ops[3];
456 	struct ceph_osd_request *req;
457 
458 	ops[0].op = opcode;
459 	ops[0].extent.truncate_seq = truncate_seq;
460 	ops[0].extent.truncate_size = truncate_size;
461 	ops[0].payload_len = 0;
462 
463 	if (do_sync) {
464 		ops[1].op = CEPH_OSD_OP_STARTSYNC;
465 		ops[1].payload_len = 0;
466 		ops[2].op = 0;
467 	} else
468 		ops[1].op = 0;
469 
470 	req = ceph_osdc_alloc_request(osdc, flags,
471 					 snapc, ops,
472 					 use_mempool,
473 					 GFP_NOFS, NULL, NULL);
474 	if (!req)
475 		return NULL;
476 
477 	/* calculate max write size */
478 	calc_layout(osdc, vino, layout, off, plen, req, ops);
479 	req->r_file_layout = *layout;  /* keep a copy */
480 
481 	/* in case it differs from natural (file) alignment that
482 	   calc_layout filled in for us */
483 	req->r_num_pages = calc_pages_for(page_align, *plen);
484 	req->r_page_alignment = page_align;
485 
486 	ceph_osdc_build_request(req, off, plen, ops,
487 				snapc,
488 				mtime,
489 				req->r_oid, req->r_oid_len);
490 
491 	return req;
492 }
493 EXPORT_SYMBOL(ceph_osdc_new_request);
494 
495 /*
496  * We keep osd requests in an rbtree, sorted by ->r_tid.
497  */
498 static void __insert_request(struct ceph_osd_client *osdc,
499 			     struct ceph_osd_request *new)
500 {
501 	struct rb_node **p = &osdc->requests.rb_node;
502 	struct rb_node *parent = NULL;
503 	struct ceph_osd_request *req = NULL;
504 
505 	while (*p) {
506 		parent = *p;
507 		req = rb_entry(parent, struct ceph_osd_request, r_node);
508 		if (new->r_tid < req->r_tid)
509 			p = &(*p)->rb_left;
510 		else if (new->r_tid > req->r_tid)
511 			p = &(*p)->rb_right;
512 		else
513 			BUG();
514 	}
515 
516 	rb_link_node(&new->r_node, parent, p);
517 	rb_insert_color(&new->r_node, &osdc->requests);
518 }
519 
520 static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc,
521 						 u64 tid)
522 {
523 	struct ceph_osd_request *req;
524 	struct rb_node *n = osdc->requests.rb_node;
525 
526 	while (n) {
527 		req = rb_entry(n, struct ceph_osd_request, r_node);
528 		if (tid < req->r_tid)
529 			n = n->rb_left;
530 		else if (tid > req->r_tid)
531 			n = n->rb_right;
532 		else
533 			return req;
534 	}
535 	return NULL;
536 }
537 
538 static struct ceph_osd_request *
539 __lookup_request_ge(struct ceph_osd_client *osdc,
540 		    u64 tid)
541 {
542 	struct ceph_osd_request *req;
543 	struct rb_node *n = osdc->requests.rb_node;
544 
545 	while (n) {
546 		req = rb_entry(n, struct ceph_osd_request, r_node);
547 		if (tid < req->r_tid) {
548 			if (!n->rb_left)
549 				return req;
550 			n = n->rb_left;
551 		} else if (tid > req->r_tid) {
552 			n = n->rb_right;
553 		} else {
554 			return req;
555 		}
556 	}
557 	return NULL;
558 }
559 
560 /*
561  * Resubmit requests pending on the given osd.
562  */
563 static void __kick_osd_requests(struct ceph_osd_client *osdc,
564 				struct ceph_osd *osd)
565 {
566 	struct ceph_osd_request *req, *nreq;
567 	int err;
568 
569 	dout("__kick_osd_requests osd%d\n", osd->o_osd);
570 	err = __reset_osd(osdc, osd);
571 	if (err == -EAGAIN)
572 		return;
573 
574 	list_for_each_entry(req, &osd->o_requests, r_osd_item) {
575 		list_move(&req->r_req_lru_item, &osdc->req_unsent);
576 		dout("requeued %p tid %llu osd%d\n", req, req->r_tid,
577 		     osd->o_osd);
578 		if (!req->r_linger)
579 			req->r_flags |= CEPH_OSD_FLAG_RETRY;
580 	}
581 
582 	list_for_each_entry_safe(req, nreq, &osd->o_linger_requests,
583 				 r_linger_osd) {
584 		/*
585 		 * reregister request prior to unregistering linger so
586 		 * that r_osd is preserved.
587 		 */
588 		BUG_ON(!list_empty(&req->r_req_lru_item));
589 		__register_request(osdc, req);
590 		list_add(&req->r_req_lru_item, &osdc->req_unsent);
591 		list_add(&req->r_osd_item, &req->r_osd->o_requests);
592 		__unregister_linger_request(osdc, req);
593 		dout("requeued lingering %p tid %llu osd%d\n", req, req->r_tid,
594 		     osd->o_osd);
595 	}
596 }
597 
598 static void kick_osd_requests(struct ceph_osd_client *osdc,
599 			      struct ceph_osd *kickosd)
600 {
601 	mutex_lock(&osdc->request_mutex);
602 	__kick_osd_requests(osdc, kickosd);
603 	mutex_unlock(&osdc->request_mutex);
604 }
605 
606 /*
607  * If the osd connection drops, we need to resubmit all requests.
608  */
609 static void osd_reset(struct ceph_connection *con)
610 {
611 	struct ceph_osd *osd = con->private;
612 	struct ceph_osd_client *osdc;
613 
614 	if (!osd)
615 		return;
616 	dout("osd_reset osd%d\n", osd->o_osd);
617 	osdc = osd->o_osdc;
618 	down_read(&osdc->map_sem);
619 	kick_osd_requests(osdc, osd);
620 	send_queued(osdc);
621 	up_read(&osdc->map_sem);
622 }
623 
624 /*
625  * Track open sessions with osds.
626  */
627 static struct ceph_osd *create_osd(struct ceph_osd_client *osdc)
628 {
629 	struct ceph_osd *osd;
630 
631 	osd = kzalloc(sizeof(*osd), GFP_NOFS);
632 	if (!osd)
633 		return NULL;
634 
635 	atomic_set(&osd->o_ref, 1);
636 	osd->o_osdc = osdc;
637 	INIT_LIST_HEAD(&osd->o_requests);
638 	INIT_LIST_HEAD(&osd->o_linger_requests);
639 	INIT_LIST_HEAD(&osd->o_osd_lru);
640 	osd->o_incarnation = 1;
641 
642 	ceph_con_init(osdc->client->msgr, &osd->o_con);
643 	osd->o_con.private = osd;
644 	osd->o_con.ops = &osd_con_ops;
645 	osd->o_con.peer_name.type = CEPH_ENTITY_TYPE_OSD;
646 
647 	INIT_LIST_HEAD(&osd->o_keepalive_item);
648 	return osd;
649 }
650 
651 static struct ceph_osd *get_osd(struct ceph_osd *osd)
652 {
653 	if (atomic_inc_not_zero(&osd->o_ref)) {
654 		dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
655 		     atomic_read(&osd->o_ref));
656 		return osd;
657 	} else {
658 		dout("get_osd %p FAIL\n", osd);
659 		return NULL;
660 	}
661 }
662 
663 static void put_osd(struct ceph_osd *osd)
664 {
665 	dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
666 	     atomic_read(&osd->o_ref) - 1);
667 	if (atomic_dec_and_test(&osd->o_ref) && osd->o_auth.authorizer) {
668 		struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth;
669 
670 		if (ac->ops && ac->ops->destroy_authorizer)
671 			ac->ops->destroy_authorizer(ac, osd->o_auth.authorizer);
672 		kfree(osd);
673 	}
674 }
675 
676 /*
677  * remove an osd from our map
678  */
679 static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
680 {
681 	dout("__remove_osd %p\n", osd);
682 	BUG_ON(!list_empty(&osd->o_requests));
683 	rb_erase(&osd->o_node, &osdc->osds);
684 	list_del_init(&osd->o_osd_lru);
685 	ceph_con_close(&osd->o_con);
686 	put_osd(osd);
687 }
688 
689 static void remove_all_osds(struct ceph_osd_client *osdc)
690 {
691 	dout("__remove_old_osds %p\n", osdc);
692 	mutex_lock(&osdc->request_mutex);
693 	while (!RB_EMPTY_ROOT(&osdc->osds)) {
694 		struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
695 						struct ceph_osd, o_node);
696 		__remove_osd(osdc, osd);
697 	}
698 	mutex_unlock(&osdc->request_mutex);
699 }
700 
701 static void __move_osd_to_lru(struct ceph_osd_client *osdc,
702 			      struct ceph_osd *osd)
703 {
704 	dout("__move_osd_to_lru %p\n", osd);
705 	BUG_ON(!list_empty(&osd->o_osd_lru));
706 	list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
707 	osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl * HZ;
708 }
709 
710 static void __remove_osd_from_lru(struct ceph_osd *osd)
711 {
712 	dout("__remove_osd_from_lru %p\n", osd);
713 	if (!list_empty(&osd->o_osd_lru))
714 		list_del_init(&osd->o_osd_lru);
715 }
716 
717 static void remove_old_osds(struct ceph_osd_client *osdc)
718 {
719 	struct ceph_osd *osd, *nosd;
720 
721 	dout("__remove_old_osds %p\n", osdc);
722 	mutex_lock(&osdc->request_mutex);
723 	list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
724 		if (time_before(jiffies, osd->lru_ttl))
725 			break;
726 		__remove_osd(osdc, osd);
727 	}
728 	mutex_unlock(&osdc->request_mutex);
729 }
730 
731 /*
732  * reset osd connect
733  */
734 static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
735 {
736 	struct ceph_osd_request *req;
737 	int ret = 0;
738 
739 	dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
740 	if (list_empty(&osd->o_requests) &&
741 	    list_empty(&osd->o_linger_requests)) {
742 		__remove_osd(osdc, osd);
743 	} else if (memcmp(&osdc->osdmap->osd_addr[osd->o_osd],
744 			  &osd->o_con.peer_addr,
745 			  sizeof(osd->o_con.peer_addr)) == 0 &&
746 		   !ceph_con_opened(&osd->o_con)) {
747 		dout(" osd addr hasn't changed and connection never opened,"
748 		     " letting msgr retry");
749 		/* touch each r_stamp for handle_timeout()'s benfit */
750 		list_for_each_entry(req, &osd->o_requests, r_osd_item)
751 			req->r_stamp = jiffies;
752 		ret = -EAGAIN;
753 	} else {
754 		ceph_con_close(&osd->o_con);
755 		ceph_con_open(&osd->o_con, &osdc->osdmap->osd_addr[osd->o_osd]);
756 		osd->o_incarnation++;
757 	}
758 	return ret;
759 }
760 
761 static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new)
762 {
763 	struct rb_node **p = &osdc->osds.rb_node;
764 	struct rb_node *parent = NULL;
765 	struct ceph_osd *osd = NULL;
766 
767 	dout("__insert_osd %p osd%d\n", new, new->o_osd);
768 	while (*p) {
769 		parent = *p;
770 		osd = rb_entry(parent, struct ceph_osd, o_node);
771 		if (new->o_osd < osd->o_osd)
772 			p = &(*p)->rb_left;
773 		else if (new->o_osd > osd->o_osd)
774 			p = &(*p)->rb_right;
775 		else
776 			BUG();
777 	}
778 
779 	rb_link_node(&new->o_node, parent, p);
780 	rb_insert_color(&new->o_node, &osdc->osds);
781 }
782 
783 static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o)
784 {
785 	struct ceph_osd *osd;
786 	struct rb_node *n = osdc->osds.rb_node;
787 
788 	while (n) {
789 		osd = rb_entry(n, struct ceph_osd, o_node);
790 		if (o < osd->o_osd)
791 			n = n->rb_left;
792 		else if (o > osd->o_osd)
793 			n = n->rb_right;
794 		else
795 			return osd;
796 	}
797 	return NULL;
798 }
799 
800 static void __schedule_osd_timeout(struct ceph_osd_client *osdc)
801 {
802 	schedule_delayed_work(&osdc->timeout_work,
803 			osdc->client->options->osd_keepalive_timeout * HZ);
804 }
805 
806 static void __cancel_osd_timeout(struct ceph_osd_client *osdc)
807 {
808 	cancel_delayed_work(&osdc->timeout_work);
809 }
810 
811 /*
812  * Register request, assign tid.  If this is the first request, set up
813  * the timeout event.
814  */
815 static void __register_request(struct ceph_osd_client *osdc,
816 			       struct ceph_osd_request *req)
817 {
818 	req->r_tid = ++osdc->last_tid;
819 	req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
820 	dout("__register_request %p tid %lld\n", req, req->r_tid);
821 	__insert_request(osdc, req);
822 	ceph_osdc_get_request(req);
823 	osdc->num_requests++;
824 	if (osdc->num_requests == 1) {
825 		dout(" first request, scheduling timeout\n");
826 		__schedule_osd_timeout(osdc);
827 	}
828 }
829 
830 static void register_request(struct ceph_osd_client *osdc,
831 			     struct ceph_osd_request *req)
832 {
833 	mutex_lock(&osdc->request_mutex);
834 	__register_request(osdc, req);
835 	mutex_unlock(&osdc->request_mutex);
836 }
837 
838 /*
839  * called under osdc->request_mutex
840  */
841 static void __unregister_request(struct ceph_osd_client *osdc,
842 				 struct ceph_osd_request *req)
843 {
844 	if (RB_EMPTY_NODE(&req->r_node)) {
845 		dout("__unregister_request %p tid %lld not registered\n",
846 			req, req->r_tid);
847 		return;
848 	}
849 
850 	dout("__unregister_request %p tid %lld\n", req, req->r_tid);
851 	rb_erase(&req->r_node, &osdc->requests);
852 	osdc->num_requests--;
853 
854 	if (req->r_osd) {
855 		/* make sure the original request isn't in flight. */
856 		ceph_con_revoke(&req->r_osd->o_con, req->r_request);
857 
858 		list_del_init(&req->r_osd_item);
859 		if (list_empty(&req->r_osd->o_requests) &&
860 		    list_empty(&req->r_osd->o_linger_requests)) {
861 			dout("moving osd to %p lru\n", req->r_osd);
862 			__move_osd_to_lru(osdc, req->r_osd);
863 		}
864 		if (list_empty(&req->r_linger_item))
865 			req->r_osd = NULL;
866 	}
867 
868 	ceph_osdc_put_request(req);
869 
870 	list_del_init(&req->r_req_lru_item);
871 	if (osdc->num_requests == 0) {
872 		dout(" no requests, canceling timeout\n");
873 		__cancel_osd_timeout(osdc);
874 	}
875 }
876 
877 /*
878  * Cancel a previously queued request message
879  */
880 static void __cancel_request(struct ceph_osd_request *req)
881 {
882 	if (req->r_sent && req->r_osd) {
883 		ceph_con_revoke(&req->r_osd->o_con, req->r_request);
884 		req->r_sent = 0;
885 	}
886 }
887 
888 static void __register_linger_request(struct ceph_osd_client *osdc,
889 				    struct ceph_osd_request *req)
890 {
891 	dout("__register_linger_request %p\n", req);
892 	list_add_tail(&req->r_linger_item, &osdc->req_linger);
893 	list_add_tail(&req->r_linger_osd, &req->r_osd->o_linger_requests);
894 }
895 
896 static void __unregister_linger_request(struct ceph_osd_client *osdc,
897 					struct ceph_osd_request *req)
898 {
899 	dout("__unregister_linger_request %p\n", req);
900 	if (req->r_osd) {
901 		list_del_init(&req->r_linger_item);
902 		list_del_init(&req->r_linger_osd);
903 
904 		if (list_empty(&req->r_osd->o_requests) &&
905 		    list_empty(&req->r_osd->o_linger_requests)) {
906 			dout("moving osd to %p lru\n", req->r_osd);
907 			__move_osd_to_lru(osdc, req->r_osd);
908 		}
909 		if (list_empty(&req->r_osd_item))
910 			req->r_osd = NULL;
911 	}
912 }
913 
914 void ceph_osdc_unregister_linger_request(struct ceph_osd_client *osdc,
915 					 struct ceph_osd_request *req)
916 {
917 	mutex_lock(&osdc->request_mutex);
918 	if (req->r_linger) {
919 		__unregister_linger_request(osdc, req);
920 		ceph_osdc_put_request(req);
921 	}
922 	mutex_unlock(&osdc->request_mutex);
923 }
924 EXPORT_SYMBOL(ceph_osdc_unregister_linger_request);
925 
926 void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
927 				  struct ceph_osd_request *req)
928 {
929 	if (!req->r_linger) {
930 		dout("set_request_linger %p\n", req);
931 		req->r_linger = 1;
932 		/*
933 		 * caller is now responsible for calling
934 		 * unregister_linger_request
935 		 */
936 		ceph_osdc_get_request(req);
937 	}
938 }
939 EXPORT_SYMBOL(ceph_osdc_set_request_linger);
940 
941 /*
942  * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
943  * (as needed), and set the request r_osd appropriately.  If there is
944  * no up osd, set r_osd to NULL.  Move the request to the appropriate list
945  * (unsent, homeless) or leave on in-flight lru.
946  *
947  * Return 0 if unchanged, 1 if changed, or negative on error.
948  *
949  * Caller should hold map_sem for read and request_mutex.
950  */
951 static int __map_request(struct ceph_osd_client *osdc,
952 			 struct ceph_osd_request *req, int force_resend)
953 {
954 	struct ceph_osd_request_head *reqhead = req->r_request->front.iov_base;
955 	struct ceph_pg pgid;
956 	int acting[CEPH_PG_MAX_SIZE];
957 	int o = -1, num = 0;
958 	int err;
959 
960 	dout("map_request %p tid %lld\n", req, req->r_tid);
961 	err = ceph_calc_object_layout(&reqhead->layout, req->r_oid,
962 				      &req->r_file_layout, osdc->osdmap);
963 	if (err) {
964 		list_move(&req->r_req_lru_item, &osdc->req_notarget);
965 		return err;
966 	}
967 	pgid = reqhead->layout.ol_pgid;
968 	req->r_pgid = pgid;
969 
970 	err = ceph_calc_pg_acting(osdc->osdmap, pgid, acting);
971 	if (err > 0) {
972 		o = acting[0];
973 		num = err;
974 	}
975 
976 	if ((!force_resend &&
977 	     req->r_osd && req->r_osd->o_osd == o &&
978 	     req->r_sent >= req->r_osd->o_incarnation &&
979 	     req->r_num_pg_osds == num &&
980 	     memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
981 	    (req->r_osd == NULL && o == -1))
982 		return 0;  /* no change */
983 
984 	dout("map_request tid %llu pgid %d.%x osd%d (was osd%d)\n",
985 	     req->r_tid, le32_to_cpu(pgid.pool), le16_to_cpu(pgid.ps), o,
986 	     req->r_osd ? req->r_osd->o_osd : -1);
987 
988 	/* record full pg acting set */
989 	memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
990 	req->r_num_pg_osds = num;
991 
992 	if (req->r_osd) {
993 		__cancel_request(req);
994 		list_del_init(&req->r_osd_item);
995 		req->r_osd = NULL;
996 	}
997 
998 	req->r_osd = __lookup_osd(osdc, o);
999 	if (!req->r_osd && o >= 0) {
1000 		err = -ENOMEM;
1001 		req->r_osd = create_osd(osdc);
1002 		if (!req->r_osd) {
1003 			list_move(&req->r_req_lru_item, &osdc->req_notarget);
1004 			goto out;
1005 		}
1006 
1007 		dout("map_request osd %p is osd%d\n", req->r_osd, o);
1008 		req->r_osd->o_osd = o;
1009 		req->r_osd->o_con.peer_name.num = cpu_to_le64(o);
1010 		__insert_osd(osdc, req->r_osd);
1011 
1012 		ceph_con_open(&req->r_osd->o_con, &osdc->osdmap->osd_addr[o]);
1013 	}
1014 
1015 	if (req->r_osd) {
1016 		__remove_osd_from_lru(req->r_osd);
1017 		list_add(&req->r_osd_item, &req->r_osd->o_requests);
1018 		list_move(&req->r_req_lru_item, &osdc->req_unsent);
1019 	} else {
1020 		list_move(&req->r_req_lru_item, &osdc->req_notarget);
1021 	}
1022 	err = 1;   /* osd or pg changed */
1023 
1024 out:
1025 	return err;
1026 }
1027 
1028 /*
1029  * caller should hold map_sem (for read) and request_mutex
1030  */
1031 static void __send_request(struct ceph_osd_client *osdc,
1032 			   struct ceph_osd_request *req)
1033 {
1034 	struct ceph_osd_request_head *reqhead;
1035 
1036 	dout("send_request %p tid %llu to osd%d flags %d\n",
1037 	     req, req->r_tid, req->r_osd->o_osd, req->r_flags);
1038 
1039 	reqhead = req->r_request->front.iov_base;
1040 	reqhead->osdmap_epoch = cpu_to_le32(osdc->osdmap->epoch);
1041 	reqhead->flags |= cpu_to_le32(req->r_flags);  /* e.g., RETRY */
1042 	reqhead->reassert_version = req->r_reassert_version;
1043 
1044 	req->r_stamp = jiffies;
1045 	list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
1046 
1047 	ceph_msg_get(req->r_request); /* send consumes a ref */
1048 	ceph_con_send(&req->r_osd->o_con, req->r_request);
1049 	req->r_sent = req->r_osd->o_incarnation;
1050 }
1051 
1052 /*
1053  * Send any requests in the queue (req_unsent).
1054  */
1055 static void send_queued(struct ceph_osd_client *osdc)
1056 {
1057 	struct ceph_osd_request *req, *tmp;
1058 
1059 	dout("send_queued\n");
1060 	mutex_lock(&osdc->request_mutex);
1061 	list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item) {
1062 		__send_request(osdc, req);
1063 	}
1064 	mutex_unlock(&osdc->request_mutex);
1065 }
1066 
1067 /*
1068  * Timeout callback, called every N seconds when 1 or more osd
1069  * requests has been active for more than N seconds.  When this
1070  * happens, we ping all OSDs with requests who have timed out to
1071  * ensure any communications channel reset is detected.  Reset the
1072  * request timeouts another N seconds in the future as we go.
1073  * Reschedule the timeout event another N seconds in future (unless
1074  * there are no open requests).
1075  */
1076 static void handle_timeout(struct work_struct *work)
1077 {
1078 	struct ceph_osd_client *osdc =
1079 		container_of(work, struct ceph_osd_client, timeout_work.work);
1080 	struct ceph_osd_request *req, *last_req = NULL;
1081 	struct ceph_osd *osd;
1082 	unsigned long timeout = osdc->client->options->osd_timeout * HZ;
1083 	unsigned long keepalive =
1084 		osdc->client->options->osd_keepalive_timeout * HZ;
1085 	unsigned long last_stamp = 0;
1086 	struct list_head slow_osds;
1087 	dout("timeout\n");
1088 	down_read(&osdc->map_sem);
1089 
1090 	ceph_monc_request_next_osdmap(&osdc->client->monc);
1091 
1092 	mutex_lock(&osdc->request_mutex);
1093 
1094 	/*
1095 	 * reset osds that appear to be _really_ unresponsive.  this
1096 	 * is a failsafe measure.. we really shouldn't be getting to
1097 	 * this point if the system is working properly.  the monitors
1098 	 * should mark the osd as failed and we should find out about
1099 	 * it from an updated osd map.
1100 	 */
1101 	while (timeout && !list_empty(&osdc->req_lru)) {
1102 		req = list_entry(osdc->req_lru.next, struct ceph_osd_request,
1103 				 r_req_lru_item);
1104 
1105 		/* hasn't been long enough since we sent it? */
1106 		if (time_before(jiffies, req->r_stamp + timeout))
1107 			break;
1108 
1109 		/* hasn't been long enough since it was acked? */
1110 		if (req->r_request->ack_stamp == 0 ||
1111 		    time_before(jiffies, req->r_request->ack_stamp + timeout))
1112 			break;
1113 
1114 		BUG_ON(req == last_req && req->r_stamp == last_stamp);
1115 		last_req = req;
1116 		last_stamp = req->r_stamp;
1117 
1118 		osd = req->r_osd;
1119 		BUG_ON(!osd);
1120 		pr_warning(" tid %llu timed out on osd%d, will reset osd\n",
1121 			   req->r_tid, osd->o_osd);
1122 		__kick_osd_requests(osdc, osd);
1123 	}
1124 
1125 	/*
1126 	 * ping osds that are a bit slow.  this ensures that if there
1127 	 * is a break in the TCP connection we will notice, and reopen
1128 	 * a connection with that osd (from the fault callback).
1129 	 */
1130 	INIT_LIST_HEAD(&slow_osds);
1131 	list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
1132 		if (time_before(jiffies, req->r_stamp + keepalive))
1133 			break;
1134 
1135 		osd = req->r_osd;
1136 		BUG_ON(!osd);
1137 		dout(" tid %llu is slow, will send keepalive on osd%d\n",
1138 		     req->r_tid, osd->o_osd);
1139 		list_move_tail(&osd->o_keepalive_item, &slow_osds);
1140 	}
1141 	while (!list_empty(&slow_osds)) {
1142 		osd = list_entry(slow_osds.next, struct ceph_osd,
1143 				 o_keepalive_item);
1144 		list_del_init(&osd->o_keepalive_item);
1145 		ceph_con_keepalive(&osd->o_con);
1146 	}
1147 
1148 	__schedule_osd_timeout(osdc);
1149 	mutex_unlock(&osdc->request_mutex);
1150 	send_queued(osdc);
1151 	up_read(&osdc->map_sem);
1152 }
1153 
1154 static void handle_osds_timeout(struct work_struct *work)
1155 {
1156 	struct ceph_osd_client *osdc =
1157 		container_of(work, struct ceph_osd_client,
1158 			     osds_timeout_work.work);
1159 	unsigned long delay =
1160 		osdc->client->options->osd_idle_ttl * HZ >> 2;
1161 
1162 	dout("osds timeout\n");
1163 	down_read(&osdc->map_sem);
1164 	remove_old_osds(osdc);
1165 	up_read(&osdc->map_sem);
1166 
1167 	schedule_delayed_work(&osdc->osds_timeout_work,
1168 			      round_jiffies_relative(delay));
1169 }
1170 
1171 static void complete_request(struct ceph_osd_request *req)
1172 {
1173 	if (req->r_safe_callback)
1174 		req->r_safe_callback(req, NULL);
1175 	complete_all(&req->r_safe_completion);  /* fsync waiter */
1176 }
1177 
1178 /*
1179  * handle osd op reply.  either call the callback if it is specified,
1180  * or do the completion to wake up the waiting thread.
1181  */
1182 static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg,
1183 			 struct ceph_connection *con)
1184 {
1185 	struct ceph_osd_reply_head *rhead = msg->front.iov_base;
1186 	struct ceph_osd_request *req;
1187 	u64 tid;
1188 	int numops, object_len, flags;
1189 	s32 result;
1190 
1191 	tid = le64_to_cpu(msg->hdr.tid);
1192 	if (msg->front.iov_len < sizeof(*rhead))
1193 		goto bad;
1194 	numops = le32_to_cpu(rhead->num_ops);
1195 	object_len = le32_to_cpu(rhead->object_len);
1196 	result = le32_to_cpu(rhead->result);
1197 	if (msg->front.iov_len != sizeof(*rhead) + object_len +
1198 	    numops * sizeof(struct ceph_osd_op))
1199 		goto bad;
1200 	dout("handle_reply %p tid %llu result %d\n", msg, tid, (int)result);
1201 	/* lookup */
1202 	mutex_lock(&osdc->request_mutex);
1203 	req = __lookup_request(osdc, tid);
1204 	if (req == NULL) {
1205 		dout("handle_reply tid %llu dne\n", tid);
1206 		mutex_unlock(&osdc->request_mutex);
1207 		return;
1208 	}
1209 	ceph_osdc_get_request(req);
1210 	flags = le32_to_cpu(rhead->flags);
1211 
1212 	/*
1213 	 * if this connection filled our message, drop our reference now, to
1214 	 * avoid a (safe but slower) revoke later.
1215 	 */
1216 	if (req->r_con_filling_msg == con && req->r_reply == msg) {
1217 		dout(" dropping con_filling_msg ref %p\n", con);
1218 		req->r_con_filling_msg = NULL;
1219 		con->ops->put(con);
1220 	}
1221 
1222 	if (!req->r_got_reply) {
1223 		unsigned int bytes;
1224 
1225 		req->r_result = le32_to_cpu(rhead->result);
1226 		bytes = le32_to_cpu(msg->hdr.data_len);
1227 		dout("handle_reply result %d bytes %d\n", req->r_result,
1228 		     bytes);
1229 		if (req->r_result == 0)
1230 			req->r_result = bytes;
1231 
1232 		/* in case this is a write and we need to replay, */
1233 		req->r_reassert_version = rhead->reassert_version;
1234 
1235 		req->r_got_reply = 1;
1236 	} else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) {
1237 		dout("handle_reply tid %llu dup ack\n", tid);
1238 		mutex_unlock(&osdc->request_mutex);
1239 		goto done;
1240 	}
1241 
1242 	dout("handle_reply tid %llu flags %d\n", tid, flags);
1243 
1244 	if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK))
1245 		__register_linger_request(osdc, req);
1246 
1247 	/* either this is a read, or we got the safe response */
1248 	if (result < 0 ||
1249 	    (flags & CEPH_OSD_FLAG_ONDISK) ||
1250 	    ((flags & CEPH_OSD_FLAG_WRITE) == 0))
1251 		__unregister_request(osdc, req);
1252 
1253 	mutex_unlock(&osdc->request_mutex);
1254 
1255 	if (req->r_callback)
1256 		req->r_callback(req, msg);
1257 	else
1258 		complete_all(&req->r_completion);
1259 
1260 	if (flags & CEPH_OSD_FLAG_ONDISK)
1261 		complete_request(req);
1262 
1263 done:
1264 	dout("req=%p req->r_linger=%d\n", req, req->r_linger);
1265 	ceph_osdc_put_request(req);
1266 	return;
1267 
1268 bad:
1269 	pr_err("corrupt osd_op_reply got %d %d expected %d\n",
1270 	       (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len),
1271 	       (int)sizeof(*rhead));
1272 	ceph_msg_dump(msg);
1273 }
1274 
1275 static void reset_changed_osds(struct ceph_osd_client *osdc)
1276 {
1277 	struct rb_node *p, *n;
1278 
1279 	for (p = rb_first(&osdc->osds); p; p = n) {
1280 		struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
1281 
1282 		n = rb_next(p);
1283 		if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
1284 		    memcmp(&osd->o_con.peer_addr,
1285 			   ceph_osd_addr(osdc->osdmap,
1286 					 osd->o_osd),
1287 			   sizeof(struct ceph_entity_addr)) != 0)
1288 			__reset_osd(osdc, osd);
1289 	}
1290 }
1291 
1292 /*
1293  * Requeue requests whose mapping to an OSD has changed.  If requests map to
1294  * no osd, request a new map.
1295  *
1296  * Caller should hold map_sem for read and request_mutex.
1297  */
1298 static void kick_requests(struct ceph_osd_client *osdc, int force_resend)
1299 {
1300 	struct ceph_osd_request *req, *nreq;
1301 	struct rb_node *p;
1302 	int needmap = 0;
1303 	int err;
1304 
1305 	dout("kick_requests %s\n", force_resend ? " (force resend)" : "");
1306 	mutex_lock(&osdc->request_mutex);
1307 	for (p = rb_first(&osdc->requests); p; p = rb_next(p)) {
1308 		req = rb_entry(p, struct ceph_osd_request, r_node);
1309 		err = __map_request(osdc, req, force_resend);
1310 		if (err < 0)
1311 			continue;  /* error */
1312 		if (req->r_osd == NULL) {
1313 			dout("%p tid %llu maps to no osd\n", req, req->r_tid);
1314 			needmap++;  /* request a newer map */
1315 		} else if (err > 0) {
1316 			dout("%p tid %llu requeued on osd%d\n", req, req->r_tid,
1317 			     req->r_osd ? req->r_osd->o_osd : -1);
1318 			if (!req->r_linger)
1319 				req->r_flags |= CEPH_OSD_FLAG_RETRY;
1320 		}
1321 	}
1322 
1323 	list_for_each_entry_safe(req, nreq, &osdc->req_linger,
1324 				 r_linger_item) {
1325 		dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
1326 
1327 		err = __map_request(osdc, req, force_resend);
1328 		if (err == 0)
1329 			continue;  /* no change and no osd was specified */
1330 		if (err < 0)
1331 			continue;  /* hrm! */
1332 		if (req->r_osd == NULL) {
1333 			dout("tid %llu maps to no valid osd\n", req->r_tid);
1334 			needmap++;  /* request a newer map */
1335 			continue;
1336 		}
1337 
1338 		dout("kicking lingering %p tid %llu osd%d\n", req, req->r_tid,
1339 		     req->r_osd ? req->r_osd->o_osd : -1);
1340 		__unregister_linger_request(osdc, req);
1341 		__register_request(osdc, req);
1342 	}
1343 	mutex_unlock(&osdc->request_mutex);
1344 
1345 	if (needmap) {
1346 		dout("%d requests for down osds, need new map\n", needmap);
1347 		ceph_monc_request_next_osdmap(&osdc->client->monc);
1348 	}
1349 }
1350 
1351 
1352 /*
1353  * Process updated osd map.
1354  *
1355  * The message contains any number of incremental and full maps, normally
1356  * indicating some sort of topology change in the cluster.  Kick requests
1357  * off to different OSDs as needed.
1358  */
1359 void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1360 {
1361 	void *p, *end, *next;
1362 	u32 nr_maps, maplen;
1363 	u32 epoch;
1364 	struct ceph_osdmap *newmap = NULL, *oldmap;
1365 	int err;
1366 	struct ceph_fsid fsid;
1367 
1368 	dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
1369 	p = msg->front.iov_base;
1370 	end = p + msg->front.iov_len;
1371 
1372 	/* verify fsid */
1373 	ceph_decode_need(&p, end, sizeof(fsid), bad);
1374 	ceph_decode_copy(&p, &fsid, sizeof(fsid));
1375 	if (ceph_check_fsid(osdc->client, &fsid) < 0)
1376 		return;
1377 
1378 	down_write(&osdc->map_sem);
1379 
1380 	/* incremental maps */
1381 	ceph_decode_32_safe(&p, end, nr_maps, bad);
1382 	dout(" %d inc maps\n", nr_maps);
1383 	while (nr_maps > 0) {
1384 		ceph_decode_need(&p, end, 2*sizeof(u32), bad);
1385 		epoch = ceph_decode_32(&p);
1386 		maplen = ceph_decode_32(&p);
1387 		ceph_decode_need(&p, end, maplen, bad);
1388 		next = p + maplen;
1389 		if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
1390 			dout("applying incremental map %u len %d\n",
1391 			     epoch, maplen);
1392 			newmap = osdmap_apply_incremental(&p, next,
1393 							  osdc->osdmap,
1394 							  osdc->client->msgr);
1395 			if (IS_ERR(newmap)) {
1396 				err = PTR_ERR(newmap);
1397 				goto bad;
1398 			}
1399 			BUG_ON(!newmap);
1400 			if (newmap != osdc->osdmap) {
1401 				ceph_osdmap_destroy(osdc->osdmap);
1402 				osdc->osdmap = newmap;
1403 			}
1404 			kick_requests(osdc, 0);
1405 			reset_changed_osds(osdc);
1406 		} else {
1407 			dout("ignoring incremental map %u len %d\n",
1408 			     epoch, maplen);
1409 		}
1410 		p = next;
1411 		nr_maps--;
1412 	}
1413 	if (newmap)
1414 		goto done;
1415 
1416 	/* full maps */
1417 	ceph_decode_32_safe(&p, end, nr_maps, bad);
1418 	dout(" %d full maps\n", nr_maps);
1419 	while (nr_maps) {
1420 		ceph_decode_need(&p, end, 2*sizeof(u32), bad);
1421 		epoch = ceph_decode_32(&p);
1422 		maplen = ceph_decode_32(&p);
1423 		ceph_decode_need(&p, end, maplen, bad);
1424 		if (nr_maps > 1) {
1425 			dout("skipping non-latest full map %u len %d\n",
1426 			     epoch, maplen);
1427 		} else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
1428 			dout("skipping full map %u len %d, "
1429 			     "older than our %u\n", epoch, maplen,
1430 			     osdc->osdmap->epoch);
1431 		} else {
1432 			int skipped_map = 0;
1433 
1434 			dout("taking full map %u len %d\n", epoch, maplen);
1435 			newmap = osdmap_decode(&p, p+maplen);
1436 			if (IS_ERR(newmap)) {
1437 				err = PTR_ERR(newmap);
1438 				goto bad;
1439 			}
1440 			BUG_ON(!newmap);
1441 			oldmap = osdc->osdmap;
1442 			osdc->osdmap = newmap;
1443 			if (oldmap) {
1444 				if (oldmap->epoch + 1 < newmap->epoch)
1445 					skipped_map = 1;
1446 				ceph_osdmap_destroy(oldmap);
1447 			}
1448 			kick_requests(osdc, skipped_map);
1449 		}
1450 		p += maplen;
1451 		nr_maps--;
1452 	}
1453 
1454 done:
1455 	downgrade_write(&osdc->map_sem);
1456 	ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch);
1457 
1458 	/*
1459 	 * subscribe to subsequent osdmap updates if full to ensure
1460 	 * we find out when we are no longer full and stop returning
1461 	 * ENOSPC.
1462 	 */
1463 	if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL))
1464 		ceph_monc_request_next_osdmap(&osdc->client->monc);
1465 
1466 	send_queued(osdc);
1467 	up_read(&osdc->map_sem);
1468 	wake_up_all(&osdc->client->auth_wq);
1469 	return;
1470 
1471 bad:
1472 	pr_err("osdc handle_map corrupt msg\n");
1473 	ceph_msg_dump(msg);
1474 	up_write(&osdc->map_sem);
1475 	return;
1476 }
1477 
1478 /*
1479  * watch/notify callback event infrastructure
1480  *
1481  * These callbacks are used both for watch and notify operations.
1482  */
1483 static void __release_event(struct kref *kref)
1484 {
1485 	struct ceph_osd_event *event =
1486 		container_of(kref, struct ceph_osd_event, kref);
1487 
1488 	dout("__release_event %p\n", event);
1489 	kfree(event);
1490 }
1491 
1492 static void get_event(struct ceph_osd_event *event)
1493 {
1494 	kref_get(&event->kref);
1495 }
1496 
1497 void ceph_osdc_put_event(struct ceph_osd_event *event)
1498 {
1499 	kref_put(&event->kref, __release_event);
1500 }
1501 EXPORT_SYMBOL(ceph_osdc_put_event);
1502 
1503 static void __insert_event(struct ceph_osd_client *osdc,
1504 			     struct ceph_osd_event *new)
1505 {
1506 	struct rb_node **p = &osdc->event_tree.rb_node;
1507 	struct rb_node *parent = NULL;
1508 	struct ceph_osd_event *event = NULL;
1509 
1510 	while (*p) {
1511 		parent = *p;
1512 		event = rb_entry(parent, struct ceph_osd_event, node);
1513 		if (new->cookie < event->cookie)
1514 			p = &(*p)->rb_left;
1515 		else if (new->cookie > event->cookie)
1516 			p = &(*p)->rb_right;
1517 		else
1518 			BUG();
1519 	}
1520 
1521 	rb_link_node(&new->node, parent, p);
1522 	rb_insert_color(&new->node, &osdc->event_tree);
1523 }
1524 
1525 static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
1526 					        u64 cookie)
1527 {
1528 	struct rb_node **p = &osdc->event_tree.rb_node;
1529 	struct rb_node *parent = NULL;
1530 	struct ceph_osd_event *event = NULL;
1531 
1532 	while (*p) {
1533 		parent = *p;
1534 		event = rb_entry(parent, struct ceph_osd_event, node);
1535 		if (cookie < event->cookie)
1536 			p = &(*p)->rb_left;
1537 		else if (cookie > event->cookie)
1538 			p = &(*p)->rb_right;
1539 		else
1540 			return event;
1541 	}
1542 	return NULL;
1543 }
1544 
1545 static void __remove_event(struct ceph_osd_event *event)
1546 {
1547 	struct ceph_osd_client *osdc = event->osdc;
1548 
1549 	if (!RB_EMPTY_NODE(&event->node)) {
1550 		dout("__remove_event removed %p\n", event);
1551 		rb_erase(&event->node, &osdc->event_tree);
1552 		ceph_osdc_put_event(event);
1553 	} else {
1554 		dout("__remove_event didn't remove %p\n", event);
1555 	}
1556 }
1557 
1558 int ceph_osdc_create_event(struct ceph_osd_client *osdc,
1559 			   void (*event_cb)(u64, u64, u8, void *),
1560 			   int one_shot, void *data,
1561 			   struct ceph_osd_event **pevent)
1562 {
1563 	struct ceph_osd_event *event;
1564 
1565 	event = kmalloc(sizeof(*event), GFP_NOIO);
1566 	if (!event)
1567 		return -ENOMEM;
1568 
1569 	dout("create_event %p\n", event);
1570 	event->cb = event_cb;
1571 	event->one_shot = one_shot;
1572 	event->data = data;
1573 	event->osdc = osdc;
1574 	INIT_LIST_HEAD(&event->osd_node);
1575 	kref_init(&event->kref);   /* one ref for us */
1576 	kref_get(&event->kref);    /* one ref for the caller */
1577 	init_completion(&event->completion);
1578 
1579 	spin_lock(&osdc->event_lock);
1580 	event->cookie = ++osdc->event_count;
1581 	__insert_event(osdc, event);
1582 	spin_unlock(&osdc->event_lock);
1583 
1584 	*pevent = event;
1585 	return 0;
1586 }
1587 EXPORT_SYMBOL(ceph_osdc_create_event);
1588 
1589 void ceph_osdc_cancel_event(struct ceph_osd_event *event)
1590 {
1591 	struct ceph_osd_client *osdc = event->osdc;
1592 
1593 	dout("cancel_event %p\n", event);
1594 	spin_lock(&osdc->event_lock);
1595 	__remove_event(event);
1596 	spin_unlock(&osdc->event_lock);
1597 	ceph_osdc_put_event(event); /* caller's */
1598 }
1599 EXPORT_SYMBOL(ceph_osdc_cancel_event);
1600 
1601 
1602 static void do_event_work(struct work_struct *work)
1603 {
1604 	struct ceph_osd_event_work *event_work =
1605 		container_of(work, struct ceph_osd_event_work, work);
1606 	struct ceph_osd_event *event = event_work->event;
1607 	u64 ver = event_work->ver;
1608 	u64 notify_id = event_work->notify_id;
1609 	u8 opcode = event_work->opcode;
1610 
1611 	dout("do_event_work completing %p\n", event);
1612 	event->cb(ver, notify_id, opcode, event->data);
1613 	complete(&event->completion);
1614 	dout("do_event_work completed %p\n", event);
1615 	ceph_osdc_put_event(event);
1616 	kfree(event_work);
1617 }
1618 
1619 
1620 /*
1621  * Process osd watch notifications
1622  */
1623 void handle_watch_notify(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1624 {
1625 	void *p, *end;
1626 	u8 proto_ver;
1627 	u64 cookie, ver, notify_id;
1628 	u8 opcode;
1629 	struct ceph_osd_event *event;
1630 	struct ceph_osd_event_work *event_work;
1631 
1632 	p = msg->front.iov_base;
1633 	end = p + msg->front.iov_len;
1634 
1635 	ceph_decode_8_safe(&p, end, proto_ver, bad);
1636 	ceph_decode_8_safe(&p, end, opcode, bad);
1637 	ceph_decode_64_safe(&p, end, cookie, bad);
1638 	ceph_decode_64_safe(&p, end, ver, bad);
1639 	ceph_decode_64_safe(&p, end, notify_id, bad);
1640 
1641 	spin_lock(&osdc->event_lock);
1642 	event = __find_event(osdc, cookie);
1643 	if (event) {
1644 		get_event(event);
1645 		if (event->one_shot)
1646 			__remove_event(event);
1647 	}
1648 	spin_unlock(&osdc->event_lock);
1649 	dout("handle_watch_notify cookie %lld ver %lld event %p\n",
1650 	     cookie, ver, event);
1651 	if (event) {
1652 		event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
1653 		if (!event_work) {
1654 			dout("ERROR: could not allocate event_work\n");
1655 			goto done_err;
1656 		}
1657 		INIT_WORK(&event_work->work, do_event_work);
1658 		event_work->event = event;
1659 		event_work->ver = ver;
1660 		event_work->notify_id = notify_id;
1661 		event_work->opcode = opcode;
1662 		if (!queue_work(osdc->notify_wq, &event_work->work)) {
1663 			dout("WARNING: failed to queue notify event work\n");
1664 			goto done_err;
1665 		}
1666 	}
1667 
1668 	return;
1669 
1670 done_err:
1671 	complete(&event->completion);
1672 	ceph_osdc_put_event(event);
1673 	return;
1674 
1675 bad:
1676 	pr_err("osdc handle_watch_notify corrupt msg\n");
1677 	return;
1678 }
1679 
1680 int ceph_osdc_wait_event(struct ceph_osd_event *event, unsigned long timeout)
1681 {
1682 	int err;
1683 
1684 	dout("wait_event %p\n", event);
1685 	err = wait_for_completion_interruptible_timeout(&event->completion,
1686 							timeout * HZ);
1687 	ceph_osdc_put_event(event);
1688 	if (err > 0)
1689 		err = 0;
1690 	dout("wait_event %p returns %d\n", event, err);
1691 	return err;
1692 }
1693 EXPORT_SYMBOL(ceph_osdc_wait_event);
1694 
1695 /*
1696  * Register request, send initial attempt.
1697  */
1698 int ceph_osdc_start_request(struct ceph_osd_client *osdc,
1699 			    struct ceph_osd_request *req,
1700 			    bool nofail)
1701 {
1702 	int rc = 0;
1703 
1704 	req->r_request->pages = req->r_pages;
1705 	req->r_request->nr_pages = req->r_num_pages;
1706 #ifdef CONFIG_BLOCK
1707 	req->r_request->bio = req->r_bio;
1708 #endif
1709 	req->r_request->trail = req->r_trail;
1710 
1711 	register_request(osdc, req);
1712 
1713 	down_read(&osdc->map_sem);
1714 	mutex_lock(&osdc->request_mutex);
1715 	/*
1716 	 * a racing kick_requests() may have sent the message for us
1717 	 * while we dropped request_mutex above, so only send now if
1718 	 * the request still han't been touched yet.
1719 	 */
1720 	if (req->r_sent == 0) {
1721 		rc = __map_request(osdc, req, 0);
1722 		if (rc < 0) {
1723 			if (nofail) {
1724 				dout("osdc_start_request failed map, "
1725 				     " will retry %lld\n", req->r_tid);
1726 				rc = 0;
1727 			}
1728 			goto out_unlock;
1729 		}
1730 		if (req->r_osd == NULL) {
1731 			dout("send_request %p no up osds in pg\n", req);
1732 			ceph_monc_request_next_osdmap(&osdc->client->monc);
1733 		} else {
1734 			__send_request(osdc, req);
1735 		}
1736 		rc = 0;
1737 	}
1738 
1739 out_unlock:
1740 	mutex_unlock(&osdc->request_mutex);
1741 	up_read(&osdc->map_sem);
1742 	return rc;
1743 }
1744 EXPORT_SYMBOL(ceph_osdc_start_request);
1745 
1746 /*
1747  * wait for a request to complete
1748  */
1749 int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
1750 			   struct ceph_osd_request *req)
1751 {
1752 	int rc;
1753 
1754 	rc = wait_for_completion_interruptible(&req->r_completion);
1755 	if (rc < 0) {
1756 		mutex_lock(&osdc->request_mutex);
1757 		__cancel_request(req);
1758 		__unregister_request(osdc, req);
1759 		mutex_unlock(&osdc->request_mutex);
1760 		complete_request(req);
1761 		dout("wait_request tid %llu canceled/timed out\n", req->r_tid);
1762 		return rc;
1763 	}
1764 
1765 	dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result);
1766 	return req->r_result;
1767 }
1768 EXPORT_SYMBOL(ceph_osdc_wait_request);
1769 
1770 /*
1771  * sync - wait for all in-flight requests to flush.  avoid starvation.
1772  */
1773 void ceph_osdc_sync(struct ceph_osd_client *osdc)
1774 {
1775 	struct ceph_osd_request *req;
1776 	u64 last_tid, next_tid = 0;
1777 
1778 	mutex_lock(&osdc->request_mutex);
1779 	last_tid = osdc->last_tid;
1780 	while (1) {
1781 		req = __lookup_request_ge(osdc, next_tid);
1782 		if (!req)
1783 			break;
1784 		if (req->r_tid > last_tid)
1785 			break;
1786 
1787 		next_tid = req->r_tid + 1;
1788 		if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
1789 			continue;
1790 
1791 		ceph_osdc_get_request(req);
1792 		mutex_unlock(&osdc->request_mutex);
1793 		dout("sync waiting on tid %llu (last is %llu)\n",
1794 		     req->r_tid, last_tid);
1795 		wait_for_completion(&req->r_safe_completion);
1796 		mutex_lock(&osdc->request_mutex);
1797 		ceph_osdc_put_request(req);
1798 	}
1799 	mutex_unlock(&osdc->request_mutex);
1800 	dout("sync done (thru tid %llu)\n", last_tid);
1801 }
1802 EXPORT_SYMBOL(ceph_osdc_sync);
1803 
1804 /*
1805  * init, shutdown
1806  */
1807 int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
1808 {
1809 	int err;
1810 
1811 	dout("init\n");
1812 	osdc->client = client;
1813 	osdc->osdmap = NULL;
1814 	init_rwsem(&osdc->map_sem);
1815 	init_completion(&osdc->map_waiters);
1816 	osdc->last_requested_map = 0;
1817 	mutex_init(&osdc->request_mutex);
1818 	osdc->last_tid = 0;
1819 	osdc->osds = RB_ROOT;
1820 	INIT_LIST_HEAD(&osdc->osd_lru);
1821 	osdc->requests = RB_ROOT;
1822 	INIT_LIST_HEAD(&osdc->req_lru);
1823 	INIT_LIST_HEAD(&osdc->req_unsent);
1824 	INIT_LIST_HEAD(&osdc->req_notarget);
1825 	INIT_LIST_HEAD(&osdc->req_linger);
1826 	osdc->num_requests = 0;
1827 	INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
1828 	INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
1829 	spin_lock_init(&osdc->event_lock);
1830 	osdc->event_tree = RB_ROOT;
1831 	osdc->event_count = 0;
1832 
1833 	schedule_delayed_work(&osdc->osds_timeout_work,
1834 	   round_jiffies_relative(osdc->client->options->osd_idle_ttl * HZ));
1835 
1836 	err = -ENOMEM;
1837 	osdc->req_mempool = mempool_create_kmalloc_pool(10,
1838 					sizeof(struct ceph_osd_request));
1839 	if (!osdc->req_mempool)
1840 		goto out;
1841 
1842 	err = ceph_msgpool_init(&osdc->msgpool_op, OSD_OP_FRONT_LEN, 10, true,
1843 				"osd_op");
1844 	if (err < 0)
1845 		goto out_mempool;
1846 	err = ceph_msgpool_init(&osdc->msgpool_op_reply,
1847 				OSD_OPREPLY_FRONT_LEN, 10, true,
1848 				"osd_op_reply");
1849 	if (err < 0)
1850 		goto out_msgpool;
1851 
1852 	osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
1853 	if (IS_ERR(osdc->notify_wq)) {
1854 		err = PTR_ERR(osdc->notify_wq);
1855 		osdc->notify_wq = NULL;
1856 		goto out_msgpool;
1857 	}
1858 	return 0;
1859 
1860 out_msgpool:
1861 	ceph_msgpool_destroy(&osdc->msgpool_op);
1862 out_mempool:
1863 	mempool_destroy(osdc->req_mempool);
1864 out:
1865 	return err;
1866 }
1867 EXPORT_SYMBOL(ceph_osdc_init);
1868 
1869 void ceph_osdc_stop(struct ceph_osd_client *osdc)
1870 {
1871 	flush_workqueue(osdc->notify_wq);
1872 	destroy_workqueue(osdc->notify_wq);
1873 	cancel_delayed_work_sync(&osdc->timeout_work);
1874 	cancel_delayed_work_sync(&osdc->osds_timeout_work);
1875 	if (osdc->osdmap) {
1876 		ceph_osdmap_destroy(osdc->osdmap);
1877 		osdc->osdmap = NULL;
1878 	}
1879 	remove_all_osds(osdc);
1880 	mempool_destroy(osdc->req_mempool);
1881 	ceph_msgpool_destroy(&osdc->msgpool_op);
1882 	ceph_msgpool_destroy(&osdc->msgpool_op_reply);
1883 }
1884 EXPORT_SYMBOL(ceph_osdc_stop);
1885 
1886 /*
1887  * Read some contiguous pages.  If we cross a stripe boundary, shorten
1888  * *plen.  Return number of bytes read, or error.
1889  */
1890 int ceph_osdc_readpages(struct ceph_osd_client *osdc,
1891 			struct ceph_vino vino, struct ceph_file_layout *layout,
1892 			u64 off, u64 *plen,
1893 			u32 truncate_seq, u64 truncate_size,
1894 			struct page **pages, int num_pages, int page_align)
1895 {
1896 	struct ceph_osd_request *req;
1897 	int rc = 0;
1898 
1899 	dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
1900 	     vino.snap, off, *plen);
1901 	req = ceph_osdc_new_request(osdc, layout, vino, off, plen,
1902 				    CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
1903 				    NULL, 0, truncate_seq, truncate_size, NULL,
1904 				    false, 1, page_align);
1905 	if (!req)
1906 		return -ENOMEM;
1907 
1908 	/* it may be a short read due to an object boundary */
1909 	req->r_pages = pages;
1910 
1911 	dout("readpages  final extent is %llu~%llu (%d pages align %d)\n",
1912 	     off, *plen, req->r_num_pages, page_align);
1913 
1914 	rc = ceph_osdc_start_request(osdc, req, false);
1915 	if (!rc)
1916 		rc = ceph_osdc_wait_request(osdc, req);
1917 
1918 	ceph_osdc_put_request(req);
1919 	dout("readpages result %d\n", rc);
1920 	return rc;
1921 }
1922 EXPORT_SYMBOL(ceph_osdc_readpages);
1923 
1924 /*
1925  * do a synchronous write on N pages
1926  */
1927 int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
1928 			 struct ceph_file_layout *layout,
1929 			 struct ceph_snap_context *snapc,
1930 			 u64 off, u64 len,
1931 			 u32 truncate_seq, u64 truncate_size,
1932 			 struct timespec *mtime,
1933 			 struct page **pages, int num_pages,
1934 			 int flags, int do_sync, bool nofail)
1935 {
1936 	struct ceph_osd_request *req;
1937 	int rc = 0;
1938 	int page_align = off & ~PAGE_MASK;
1939 
1940 	BUG_ON(vino.snap != CEPH_NOSNAP);
1941 	req = ceph_osdc_new_request(osdc, layout, vino, off, &len,
1942 				    CEPH_OSD_OP_WRITE,
1943 				    flags | CEPH_OSD_FLAG_ONDISK |
1944 					    CEPH_OSD_FLAG_WRITE,
1945 				    snapc, do_sync,
1946 				    truncate_seq, truncate_size, mtime,
1947 				    nofail, 1, page_align);
1948 	if (!req)
1949 		return -ENOMEM;
1950 
1951 	/* it may be a short write due to an object boundary */
1952 	req->r_pages = pages;
1953 	dout("writepages %llu~%llu (%d pages)\n", off, len,
1954 	     req->r_num_pages);
1955 
1956 	rc = ceph_osdc_start_request(osdc, req, nofail);
1957 	if (!rc)
1958 		rc = ceph_osdc_wait_request(osdc, req);
1959 
1960 	ceph_osdc_put_request(req);
1961 	if (rc == 0)
1962 		rc = len;
1963 	dout("writepages result %d\n", rc);
1964 	return rc;
1965 }
1966 EXPORT_SYMBOL(ceph_osdc_writepages);
1967 
1968 /*
1969  * handle incoming message
1970  */
1971 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
1972 {
1973 	struct ceph_osd *osd = con->private;
1974 	struct ceph_osd_client *osdc;
1975 	int type = le16_to_cpu(msg->hdr.type);
1976 
1977 	if (!osd)
1978 		goto out;
1979 	osdc = osd->o_osdc;
1980 
1981 	switch (type) {
1982 	case CEPH_MSG_OSD_MAP:
1983 		ceph_osdc_handle_map(osdc, msg);
1984 		break;
1985 	case CEPH_MSG_OSD_OPREPLY:
1986 		handle_reply(osdc, msg, con);
1987 		break;
1988 	case CEPH_MSG_WATCH_NOTIFY:
1989 		handle_watch_notify(osdc, msg);
1990 		break;
1991 
1992 	default:
1993 		pr_err("received unknown message type %d %s\n", type,
1994 		       ceph_msg_type_name(type));
1995 	}
1996 out:
1997 	ceph_msg_put(msg);
1998 }
1999 
2000 /*
2001  * lookup and return message for incoming reply.  set up reply message
2002  * pages.
2003  */
2004 static struct ceph_msg *get_reply(struct ceph_connection *con,
2005 				  struct ceph_msg_header *hdr,
2006 				  int *skip)
2007 {
2008 	struct ceph_osd *osd = con->private;
2009 	struct ceph_osd_client *osdc = osd->o_osdc;
2010 	struct ceph_msg *m;
2011 	struct ceph_osd_request *req;
2012 	int front = le32_to_cpu(hdr->front_len);
2013 	int data_len = le32_to_cpu(hdr->data_len);
2014 	u64 tid;
2015 
2016 	tid = le64_to_cpu(hdr->tid);
2017 	mutex_lock(&osdc->request_mutex);
2018 	req = __lookup_request(osdc, tid);
2019 	if (!req) {
2020 		*skip = 1;
2021 		m = NULL;
2022 		pr_info("get_reply unknown tid %llu from osd%d\n", tid,
2023 			osd->o_osd);
2024 		goto out;
2025 	}
2026 
2027 	if (req->r_con_filling_msg) {
2028 		dout("get_reply revoking msg %p from old con %p\n",
2029 		     req->r_reply, req->r_con_filling_msg);
2030 		ceph_con_revoke_message(req->r_con_filling_msg, req->r_reply);
2031 		req->r_con_filling_msg->ops->put(req->r_con_filling_msg);
2032 		req->r_con_filling_msg = NULL;
2033 	}
2034 
2035 	if (front > req->r_reply->front.iov_len) {
2036 		pr_warning("get_reply front %d > preallocated %d\n",
2037 			   front, (int)req->r_reply->front.iov_len);
2038 		m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front, GFP_NOFS, false);
2039 		if (!m)
2040 			goto out;
2041 		ceph_msg_put(req->r_reply);
2042 		req->r_reply = m;
2043 	}
2044 	m = ceph_msg_get(req->r_reply);
2045 
2046 	if (data_len > 0) {
2047 		int want = calc_pages_for(req->r_page_alignment, data_len);
2048 
2049 		if (unlikely(req->r_num_pages < want)) {
2050 			pr_warning("tid %lld reply has %d bytes %d pages, we"
2051 				   " had only %d pages ready\n", tid, data_len,
2052 				   want, req->r_num_pages);
2053 			*skip = 1;
2054 			ceph_msg_put(m);
2055 			m = NULL;
2056 			goto out;
2057 		}
2058 		m->pages = req->r_pages;
2059 		m->nr_pages = req->r_num_pages;
2060 		m->page_alignment = req->r_page_alignment;
2061 #ifdef CONFIG_BLOCK
2062 		m->bio = req->r_bio;
2063 #endif
2064 	}
2065 	*skip = 0;
2066 	req->r_con_filling_msg = con->ops->get(con);
2067 	dout("get_reply tid %lld %p\n", tid, m);
2068 
2069 out:
2070 	mutex_unlock(&osdc->request_mutex);
2071 	return m;
2072 
2073 }
2074 
2075 static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2076 				  struct ceph_msg_header *hdr,
2077 				  int *skip)
2078 {
2079 	struct ceph_osd *osd = con->private;
2080 	int type = le16_to_cpu(hdr->type);
2081 	int front = le32_to_cpu(hdr->front_len);
2082 
2083 	switch (type) {
2084 	case CEPH_MSG_OSD_MAP:
2085 	case CEPH_MSG_WATCH_NOTIFY:
2086 		return ceph_msg_new(type, front, GFP_NOFS, false);
2087 	case CEPH_MSG_OSD_OPREPLY:
2088 		return get_reply(con, hdr, skip);
2089 	default:
2090 		pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
2091 			osd->o_osd);
2092 		*skip = 1;
2093 		return NULL;
2094 	}
2095 }
2096 
2097 /*
2098  * Wrappers to refcount containing ceph_osd struct
2099  */
2100 static struct ceph_connection *get_osd_con(struct ceph_connection *con)
2101 {
2102 	struct ceph_osd *osd = con->private;
2103 	if (get_osd(osd))
2104 		return con;
2105 	return NULL;
2106 }
2107 
2108 static void put_osd_con(struct ceph_connection *con)
2109 {
2110 	struct ceph_osd *osd = con->private;
2111 	put_osd(osd);
2112 }
2113 
2114 /*
2115  * authentication
2116  */
2117 /*
2118  * Note: returned pointer is the address of a structure that's
2119  * managed separately.  Caller must *not* attempt to free it.
2120  */
2121 static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
2122 					int *proto, int force_new)
2123 {
2124 	struct ceph_osd *o = con->private;
2125 	struct ceph_osd_client *osdc = o->o_osdc;
2126 	struct ceph_auth_client *ac = osdc->client->monc.auth;
2127 	struct ceph_auth_handshake *auth = &o->o_auth;
2128 
2129 	if (force_new && auth->authorizer) {
2130 		if (ac->ops && ac->ops->destroy_authorizer)
2131 			ac->ops->destroy_authorizer(ac, auth->authorizer);
2132 		auth->authorizer = NULL;
2133 	}
2134 	if (!auth->authorizer && ac->ops && ac->ops->create_authorizer) {
2135 		int ret = ac->ops->create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2136 							auth);
2137 		if (ret)
2138 			return ERR_PTR(ret);
2139 	}
2140 	*proto = ac->protocol;
2141 
2142 	return auth;
2143 }
2144 
2145 
2146 static int verify_authorizer_reply(struct ceph_connection *con, int len)
2147 {
2148 	struct ceph_osd *o = con->private;
2149 	struct ceph_osd_client *osdc = o->o_osdc;
2150 	struct ceph_auth_client *ac = osdc->client->monc.auth;
2151 
2152 	/*
2153 	 * XXX If ac->ops or ac->ops->verify_authorizer_reply is null,
2154 	 * XXX which do we do:  succeed or fail?
2155 	 */
2156 	return ac->ops->verify_authorizer_reply(ac, o->o_auth.authorizer, len);
2157 }
2158 
2159 static int invalidate_authorizer(struct ceph_connection *con)
2160 {
2161 	struct ceph_osd *o = con->private;
2162 	struct ceph_osd_client *osdc = o->o_osdc;
2163 	struct ceph_auth_client *ac = osdc->client->monc.auth;
2164 
2165 	if (ac->ops && ac->ops->invalidate_authorizer)
2166 		ac->ops->invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
2167 
2168 	return ceph_monc_validate_auth(&osdc->client->monc);
2169 }
2170 
2171 static const struct ceph_connection_operations osd_con_ops = {
2172 	.get = get_osd_con,
2173 	.put = put_osd_con,
2174 	.dispatch = dispatch,
2175 	.get_authorizer = get_authorizer,
2176 	.verify_authorizer_reply = verify_authorizer_reply,
2177 	.invalidate_authorizer = invalidate_authorizer,
2178 	.alloc_msg = alloc_msg,
2179 	.fault = osd_reset,
2180 };
2181