xref: /linux/net/ceph/osd_client.c (revision 2c1ba398ac9da3305815f6ae8e95ae2b9fd3b5ff)
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 int __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_reply)
143 		ceph_msg_put(req->r_reply);
144 	if (req->r_con_filling_msg) {
145 		dout("release_request revoking pages %p from con %p\n",
146 		     req->r_pages, req->r_con_filling_msg);
147 		ceph_con_revoke_message(req->r_con_filling_msg,
148 				      req->r_reply);
149 		ceph_con_put(req->r_con_filling_msg);
150 	}
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 	req->r_flags = flags;
221 
222 	WARN_ON((flags & (CEPH_OSD_FLAG_READ|CEPH_OSD_FLAG_WRITE)) == 0);
223 
224 	/* create reply message */
225 	if (use_mempool)
226 		msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0);
227 	else
228 		msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY,
229 				   OSD_OPREPLY_FRONT_LEN, gfp_flags);
230 	if (!msg) {
231 		ceph_osdc_put_request(req);
232 		return NULL;
233 	}
234 	req->r_reply = msg;
235 
236 	/* allocate space for the trailing data */
237 	if (needs_trail) {
238 		req->r_trail = kmalloc(sizeof(struct ceph_pagelist), gfp_flags);
239 		if (!req->r_trail) {
240 			ceph_osdc_put_request(req);
241 			return NULL;
242 		}
243 		ceph_pagelist_init(req->r_trail);
244 	}
245 	/* create request message; allow space for oid */
246 	msg_size += 40;
247 	if (snapc)
248 		msg_size += sizeof(u64) * snapc->num_snaps;
249 	if (use_mempool)
250 		msg = ceph_msgpool_get(&osdc->msgpool_op, 0);
251 	else
252 		msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp_flags);
253 	if (!msg) {
254 		ceph_osdc_put_request(req);
255 		return NULL;
256 	}
257 
258 	msg->hdr.type = cpu_to_le16(CEPH_MSG_OSD_OP);
259 	memset(msg->front.iov_base, 0, msg->front.iov_len);
260 
261 	req->r_request = msg;
262 	req->r_pages = pages;
263 #ifdef CONFIG_BLOCK
264 	if (bio) {
265 		req->r_bio = bio;
266 		bio_get(req->r_bio);
267 	}
268 #endif
269 
270 	return req;
271 }
272 EXPORT_SYMBOL(ceph_osdc_alloc_request);
273 
274 static void osd_req_encode_op(struct ceph_osd_request *req,
275 			      struct ceph_osd_op *dst,
276 			      struct ceph_osd_req_op *src)
277 {
278 	dst->op = cpu_to_le16(src->op);
279 
280 	switch (dst->op) {
281 	case CEPH_OSD_OP_READ:
282 	case CEPH_OSD_OP_WRITE:
283 		dst->extent.offset =
284 			cpu_to_le64(src->extent.offset);
285 		dst->extent.length =
286 			cpu_to_le64(src->extent.length);
287 		dst->extent.truncate_size =
288 			cpu_to_le64(src->extent.truncate_size);
289 		dst->extent.truncate_seq =
290 			cpu_to_le32(src->extent.truncate_seq);
291 		break;
292 
293 	case CEPH_OSD_OP_GETXATTR:
294 	case CEPH_OSD_OP_SETXATTR:
295 	case CEPH_OSD_OP_CMPXATTR:
296 		BUG_ON(!req->r_trail);
297 
298 		dst->xattr.name_len = cpu_to_le32(src->xattr.name_len);
299 		dst->xattr.value_len = cpu_to_le32(src->xattr.value_len);
300 		dst->xattr.cmp_op = src->xattr.cmp_op;
301 		dst->xattr.cmp_mode = src->xattr.cmp_mode;
302 		ceph_pagelist_append(req->r_trail, src->xattr.name,
303 				     src->xattr.name_len);
304 		ceph_pagelist_append(req->r_trail, src->xattr.val,
305 				     src->xattr.value_len);
306 		break;
307 	case CEPH_OSD_OP_CALL:
308 		BUG_ON(!req->r_trail);
309 
310 		dst->cls.class_len = src->cls.class_len;
311 		dst->cls.method_len = src->cls.method_len;
312 		dst->cls.indata_len = cpu_to_le32(src->cls.indata_len);
313 
314 		ceph_pagelist_append(req->r_trail, src->cls.class_name,
315 				     src->cls.class_len);
316 		ceph_pagelist_append(req->r_trail, src->cls.method_name,
317 				     src->cls.method_len);
318 		ceph_pagelist_append(req->r_trail, src->cls.indata,
319 				     src->cls.indata_len);
320 		break;
321 	case CEPH_OSD_OP_ROLLBACK:
322 		dst->snap.snapid = cpu_to_le64(src->snap.snapid);
323 		break;
324 	case CEPH_OSD_OP_STARTSYNC:
325 		break;
326 	case CEPH_OSD_OP_NOTIFY:
327 		{
328 			__le32 prot_ver = cpu_to_le32(src->watch.prot_ver);
329 			__le32 timeout = cpu_to_le32(src->watch.timeout);
330 
331 			BUG_ON(!req->r_trail);
332 
333 			ceph_pagelist_append(req->r_trail,
334 						&prot_ver, sizeof(prot_ver));
335 			ceph_pagelist_append(req->r_trail,
336 						&timeout, sizeof(timeout));
337 		}
338 	case CEPH_OSD_OP_NOTIFY_ACK:
339 	case CEPH_OSD_OP_WATCH:
340 		dst->watch.cookie = cpu_to_le64(src->watch.cookie);
341 		dst->watch.ver = cpu_to_le64(src->watch.ver);
342 		dst->watch.flag = src->watch.flag;
343 		break;
344 	default:
345 		pr_err("unrecognized osd opcode %d\n", dst->op);
346 		WARN_ON(1);
347 		break;
348 	}
349 	dst->payload_len = cpu_to_le32(src->payload_len);
350 }
351 
352 /*
353  * build new request AND message
354  *
355  */
356 void ceph_osdc_build_request(struct ceph_osd_request *req,
357 			     u64 off, u64 *plen,
358 			     struct ceph_osd_req_op *src_ops,
359 			     struct ceph_snap_context *snapc,
360 			     struct timespec *mtime,
361 			     const char *oid,
362 			     int oid_len)
363 {
364 	struct ceph_msg *msg = req->r_request;
365 	struct ceph_osd_request_head *head;
366 	struct ceph_osd_req_op *src_op;
367 	struct ceph_osd_op *op;
368 	void *p;
369 	int num_op = get_num_ops(src_ops, NULL);
370 	size_t msg_size = sizeof(*head) + num_op*sizeof(*op);
371 	int flags = req->r_flags;
372 	u64 data_len = 0;
373 	int i;
374 
375 	head = msg->front.iov_base;
376 	op = (void *)(head + 1);
377 	p = (void *)(op + num_op);
378 
379 	req->r_snapc = ceph_get_snap_context(snapc);
380 
381 	head->client_inc = cpu_to_le32(1); /* always, for now. */
382 	head->flags = cpu_to_le32(flags);
383 	if (flags & CEPH_OSD_FLAG_WRITE)
384 		ceph_encode_timespec(&head->mtime, mtime);
385 	head->num_ops = cpu_to_le16(num_op);
386 
387 
388 	/* fill in oid */
389 	head->object_len = cpu_to_le32(oid_len);
390 	memcpy(p, oid, oid_len);
391 	p += oid_len;
392 
393 	src_op = src_ops;
394 	while (src_op->op) {
395 		osd_req_encode_op(req, op, src_op);
396 		src_op++;
397 		op++;
398 	}
399 
400 	if (req->r_trail)
401 		data_len += req->r_trail->length;
402 
403 	if (snapc) {
404 		head->snap_seq = cpu_to_le64(snapc->seq);
405 		head->num_snaps = cpu_to_le32(snapc->num_snaps);
406 		for (i = 0; i < snapc->num_snaps; i++) {
407 			put_unaligned_le64(snapc->snaps[i], p);
408 			p += sizeof(u64);
409 		}
410 	}
411 
412 	if (flags & CEPH_OSD_FLAG_WRITE) {
413 		req->r_request->hdr.data_off = cpu_to_le16(off);
414 		req->r_request->hdr.data_len = cpu_to_le32(*plen + data_len);
415 	} else if (data_len) {
416 		req->r_request->hdr.data_off = 0;
417 		req->r_request->hdr.data_len = cpu_to_le32(data_len);
418 	}
419 
420 	req->r_request->page_alignment = req->r_page_alignment;
421 
422 	BUG_ON(p > msg->front.iov_base + msg->front.iov_len);
423 	msg_size = p - msg->front.iov_base;
424 	msg->front.iov_len = msg_size;
425 	msg->hdr.front_len = cpu_to_le32(msg_size);
426 	return;
427 }
428 EXPORT_SYMBOL(ceph_osdc_build_request);
429 
430 /*
431  * build new request AND message, calculate layout, and adjust file
432  * extent as needed.
433  *
434  * if the file was recently truncated, we include information about its
435  * old and new size so that the object can be updated appropriately.  (we
436  * avoid synchronously deleting truncated objects because it's slow.)
437  *
438  * if @do_sync, include a 'startsync' command so that the osd will flush
439  * data quickly.
440  */
441 struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
442 					       struct ceph_file_layout *layout,
443 					       struct ceph_vino vino,
444 					       u64 off, u64 *plen,
445 					       int opcode, int flags,
446 					       struct ceph_snap_context *snapc,
447 					       int do_sync,
448 					       u32 truncate_seq,
449 					       u64 truncate_size,
450 					       struct timespec *mtime,
451 					       bool use_mempool, int num_reply,
452 					       int page_align)
453 {
454 	struct ceph_osd_req_op ops[3];
455 	struct ceph_osd_request *req;
456 
457 	ops[0].op = opcode;
458 	ops[0].extent.truncate_seq = truncate_seq;
459 	ops[0].extent.truncate_size = truncate_size;
460 	ops[0].payload_len = 0;
461 
462 	if (do_sync) {
463 		ops[1].op = CEPH_OSD_OP_STARTSYNC;
464 		ops[1].payload_len = 0;
465 		ops[2].op = 0;
466 	} else
467 		ops[1].op = 0;
468 
469 	req = ceph_osdc_alloc_request(osdc, flags,
470 					 snapc, ops,
471 					 use_mempool,
472 					 GFP_NOFS, NULL, NULL);
473 	if (!req)
474 		return NULL;
475 
476 	/* calculate max write size */
477 	calc_layout(osdc, vino, layout, off, plen, req, ops);
478 	req->r_file_layout = *layout;  /* keep a copy */
479 
480 	/* in case it differs from natural (file) alignment that
481 	   calc_layout filled in for us */
482 	req->r_num_pages = calc_pages_for(page_align, *plen);
483 	req->r_page_alignment = page_align;
484 
485 	ceph_osdc_build_request(req, off, plen, ops,
486 				snapc,
487 				mtime,
488 				req->r_oid, req->r_oid_len);
489 
490 	return req;
491 }
492 EXPORT_SYMBOL(ceph_osdc_new_request);
493 
494 /*
495  * We keep osd requests in an rbtree, sorted by ->r_tid.
496  */
497 static void __insert_request(struct ceph_osd_client *osdc,
498 			     struct ceph_osd_request *new)
499 {
500 	struct rb_node **p = &osdc->requests.rb_node;
501 	struct rb_node *parent = NULL;
502 	struct ceph_osd_request *req = NULL;
503 
504 	while (*p) {
505 		parent = *p;
506 		req = rb_entry(parent, struct ceph_osd_request, r_node);
507 		if (new->r_tid < req->r_tid)
508 			p = &(*p)->rb_left;
509 		else if (new->r_tid > req->r_tid)
510 			p = &(*p)->rb_right;
511 		else
512 			BUG();
513 	}
514 
515 	rb_link_node(&new->r_node, parent, p);
516 	rb_insert_color(&new->r_node, &osdc->requests);
517 }
518 
519 static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc,
520 						 u64 tid)
521 {
522 	struct ceph_osd_request *req;
523 	struct rb_node *n = osdc->requests.rb_node;
524 
525 	while (n) {
526 		req = rb_entry(n, struct ceph_osd_request, r_node);
527 		if (tid < req->r_tid)
528 			n = n->rb_left;
529 		else if (tid > req->r_tid)
530 			n = n->rb_right;
531 		else
532 			return req;
533 	}
534 	return NULL;
535 }
536 
537 static struct ceph_osd_request *
538 __lookup_request_ge(struct ceph_osd_client *osdc,
539 		    u64 tid)
540 {
541 	struct ceph_osd_request *req;
542 	struct rb_node *n = osdc->requests.rb_node;
543 
544 	while (n) {
545 		req = rb_entry(n, struct ceph_osd_request, r_node);
546 		if (tid < req->r_tid) {
547 			if (!n->rb_left)
548 				return req;
549 			n = n->rb_left;
550 		} else if (tid > req->r_tid) {
551 			n = n->rb_right;
552 		} else {
553 			return req;
554 		}
555 	}
556 	return NULL;
557 }
558 
559 /*
560  * Resubmit requests pending on the given osd.
561  */
562 static void __kick_osd_requests(struct ceph_osd_client *osdc,
563 				struct ceph_osd *osd)
564 {
565 	struct ceph_osd_request *req, *nreq;
566 	int err;
567 
568 	dout("__kick_osd_requests osd%d\n", osd->o_osd);
569 	err = __reset_osd(osdc, osd);
570 	if (err == -EAGAIN)
571 		return;
572 
573 	list_for_each_entry(req, &osd->o_requests, r_osd_item) {
574 		list_move(&req->r_req_lru_item, &osdc->req_unsent);
575 		dout("requeued %p tid %llu osd%d\n", req, req->r_tid,
576 		     osd->o_osd);
577 		if (!req->r_linger)
578 			req->r_flags |= CEPH_OSD_FLAG_RETRY;
579 	}
580 
581 	list_for_each_entry_safe(req, nreq, &osd->o_linger_requests,
582 				 r_linger_osd) {
583 		/*
584 		 * reregister request prior to unregistering linger so
585 		 * that r_osd is preserved.
586 		 */
587 		BUG_ON(!list_empty(&req->r_req_lru_item));
588 		__register_request(osdc, req);
589 		list_add(&req->r_req_lru_item, &osdc->req_unsent);
590 		list_add(&req->r_osd_item, &req->r_osd->o_requests);
591 		__unregister_linger_request(osdc, req);
592 		dout("requeued lingering %p tid %llu osd%d\n", req, req->r_tid,
593 		     osd->o_osd);
594 	}
595 }
596 
597 static void kick_osd_requests(struct ceph_osd_client *osdc,
598 			      struct ceph_osd *kickosd)
599 {
600 	mutex_lock(&osdc->request_mutex);
601 	__kick_osd_requests(osdc, kickosd);
602 	mutex_unlock(&osdc->request_mutex);
603 }
604 
605 /*
606  * If the osd connection drops, we need to resubmit all requests.
607  */
608 static void osd_reset(struct ceph_connection *con)
609 {
610 	struct ceph_osd *osd = con->private;
611 	struct ceph_osd_client *osdc;
612 
613 	if (!osd)
614 		return;
615 	dout("osd_reset osd%d\n", osd->o_osd);
616 	osdc = osd->o_osdc;
617 	down_read(&osdc->map_sem);
618 	kick_osd_requests(osdc, osd);
619 	send_queued(osdc);
620 	up_read(&osdc->map_sem);
621 }
622 
623 /*
624  * Track open sessions with osds.
625  */
626 static struct ceph_osd *create_osd(struct ceph_osd_client *osdc)
627 {
628 	struct ceph_osd *osd;
629 
630 	osd = kzalloc(sizeof(*osd), GFP_NOFS);
631 	if (!osd)
632 		return NULL;
633 
634 	atomic_set(&osd->o_ref, 1);
635 	osd->o_osdc = osdc;
636 	INIT_LIST_HEAD(&osd->o_requests);
637 	INIT_LIST_HEAD(&osd->o_linger_requests);
638 	INIT_LIST_HEAD(&osd->o_osd_lru);
639 	osd->o_incarnation = 1;
640 
641 	ceph_con_init(osdc->client->msgr, &osd->o_con);
642 	osd->o_con.private = osd;
643 	osd->o_con.ops = &osd_con_ops;
644 	osd->o_con.peer_name.type = CEPH_ENTITY_TYPE_OSD;
645 
646 	INIT_LIST_HEAD(&osd->o_keepalive_item);
647 	return osd;
648 }
649 
650 static struct ceph_osd *get_osd(struct ceph_osd *osd)
651 {
652 	if (atomic_inc_not_zero(&osd->o_ref)) {
653 		dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
654 		     atomic_read(&osd->o_ref));
655 		return osd;
656 	} else {
657 		dout("get_osd %p FAIL\n", osd);
658 		return NULL;
659 	}
660 }
661 
662 static void put_osd(struct ceph_osd *osd)
663 {
664 	dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
665 	     atomic_read(&osd->o_ref) - 1);
666 	if (atomic_dec_and_test(&osd->o_ref)) {
667 		struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth;
668 
669 		if (osd->o_authorizer)
670 			ac->ops->destroy_authorizer(ac, osd->o_authorizer);
671 		kfree(osd);
672 	}
673 }
674 
675 /*
676  * remove an osd from our map
677  */
678 static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
679 {
680 	dout("__remove_osd %p\n", osd);
681 	BUG_ON(!list_empty(&osd->o_requests));
682 	rb_erase(&osd->o_node, &osdc->osds);
683 	list_del_init(&osd->o_osd_lru);
684 	ceph_con_close(&osd->o_con);
685 	put_osd(osd);
686 }
687 
688 static void __move_osd_to_lru(struct ceph_osd_client *osdc,
689 			      struct ceph_osd *osd)
690 {
691 	dout("__move_osd_to_lru %p\n", osd);
692 	BUG_ON(!list_empty(&osd->o_osd_lru));
693 	list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
694 	osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl * HZ;
695 }
696 
697 static void __remove_osd_from_lru(struct ceph_osd *osd)
698 {
699 	dout("__remove_osd_from_lru %p\n", osd);
700 	if (!list_empty(&osd->o_osd_lru))
701 		list_del_init(&osd->o_osd_lru);
702 }
703 
704 static void remove_old_osds(struct ceph_osd_client *osdc, int remove_all)
705 {
706 	struct ceph_osd *osd, *nosd;
707 
708 	dout("__remove_old_osds %p\n", osdc);
709 	mutex_lock(&osdc->request_mutex);
710 	list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
711 		if (!remove_all && time_before(jiffies, osd->lru_ttl))
712 			break;
713 		__remove_osd(osdc, osd);
714 	}
715 	mutex_unlock(&osdc->request_mutex);
716 }
717 
718 /*
719  * reset osd connect
720  */
721 static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
722 {
723 	struct ceph_osd_request *req;
724 	int ret = 0;
725 
726 	dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
727 	if (list_empty(&osd->o_requests) &&
728 	    list_empty(&osd->o_linger_requests)) {
729 		__remove_osd(osdc, osd);
730 	} else if (memcmp(&osdc->osdmap->osd_addr[osd->o_osd],
731 			  &osd->o_con.peer_addr,
732 			  sizeof(osd->o_con.peer_addr)) == 0 &&
733 		   !ceph_con_opened(&osd->o_con)) {
734 		dout(" osd addr hasn't changed and connection never opened,"
735 		     " letting msgr retry");
736 		/* touch each r_stamp for handle_timeout()'s benfit */
737 		list_for_each_entry(req, &osd->o_requests, r_osd_item)
738 			req->r_stamp = jiffies;
739 		ret = -EAGAIN;
740 	} else {
741 		ceph_con_close(&osd->o_con);
742 		ceph_con_open(&osd->o_con, &osdc->osdmap->osd_addr[osd->o_osd]);
743 		osd->o_incarnation++;
744 	}
745 	return ret;
746 }
747 
748 static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new)
749 {
750 	struct rb_node **p = &osdc->osds.rb_node;
751 	struct rb_node *parent = NULL;
752 	struct ceph_osd *osd = NULL;
753 
754 	while (*p) {
755 		parent = *p;
756 		osd = rb_entry(parent, struct ceph_osd, o_node);
757 		if (new->o_osd < osd->o_osd)
758 			p = &(*p)->rb_left;
759 		else if (new->o_osd > osd->o_osd)
760 			p = &(*p)->rb_right;
761 		else
762 			BUG();
763 	}
764 
765 	rb_link_node(&new->o_node, parent, p);
766 	rb_insert_color(&new->o_node, &osdc->osds);
767 }
768 
769 static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o)
770 {
771 	struct ceph_osd *osd;
772 	struct rb_node *n = osdc->osds.rb_node;
773 
774 	while (n) {
775 		osd = rb_entry(n, struct ceph_osd, o_node);
776 		if (o < osd->o_osd)
777 			n = n->rb_left;
778 		else if (o > osd->o_osd)
779 			n = n->rb_right;
780 		else
781 			return osd;
782 	}
783 	return NULL;
784 }
785 
786 static void __schedule_osd_timeout(struct ceph_osd_client *osdc)
787 {
788 	schedule_delayed_work(&osdc->timeout_work,
789 			osdc->client->options->osd_keepalive_timeout * HZ);
790 }
791 
792 static void __cancel_osd_timeout(struct ceph_osd_client *osdc)
793 {
794 	cancel_delayed_work(&osdc->timeout_work);
795 }
796 
797 /*
798  * Register request, assign tid.  If this is the first request, set up
799  * the timeout event.
800  */
801 static void __register_request(struct ceph_osd_client *osdc,
802 			       struct ceph_osd_request *req)
803 {
804 	req->r_tid = ++osdc->last_tid;
805 	req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
806 	INIT_LIST_HEAD(&req->r_req_lru_item);
807 
808 	dout("__register_request %p tid %lld\n", req, req->r_tid);
809 	__insert_request(osdc, req);
810 	ceph_osdc_get_request(req);
811 	osdc->num_requests++;
812 
813 	if (osdc->num_requests == 1) {
814 		dout(" first request, scheduling timeout\n");
815 		__schedule_osd_timeout(osdc);
816 	}
817 }
818 
819 static void register_request(struct ceph_osd_client *osdc,
820 			     struct ceph_osd_request *req)
821 {
822 	mutex_lock(&osdc->request_mutex);
823 	__register_request(osdc, req);
824 	mutex_unlock(&osdc->request_mutex);
825 }
826 
827 /*
828  * called under osdc->request_mutex
829  */
830 static void __unregister_request(struct ceph_osd_client *osdc,
831 				 struct ceph_osd_request *req)
832 {
833 	dout("__unregister_request %p tid %lld\n", req, req->r_tid);
834 	rb_erase(&req->r_node, &osdc->requests);
835 	osdc->num_requests--;
836 
837 	if (req->r_osd) {
838 		/* make sure the original request isn't in flight. */
839 		ceph_con_revoke(&req->r_osd->o_con, req->r_request);
840 
841 		list_del_init(&req->r_osd_item);
842 		if (list_empty(&req->r_osd->o_requests) &&
843 		    list_empty(&req->r_osd->o_linger_requests)) {
844 			dout("moving osd to %p lru\n", req->r_osd);
845 			__move_osd_to_lru(osdc, req->r_osd);
846 		}
847 		if (list_empty(&req->r_linger_item))
848 			req->r_osd = NULL;
849 	}
850 
851 	ceph_osdc_put_request(req);
852 
853 	list_del_init(&req->r_req_lru_item);
854 	if (osdc->num_requests == 0) {
855 		dout(" no requests, canceling timeout\n");
856 		__cancel_osd_timeout(osdc);
857 	}
858 }
859 
860 /*
861  * Cancel a previously queued request message
862  */
863 static void __cancel_request(struct ceph_osd_request *req)
864 {
865 	if (req->r_sent && req->r_osd) {
866 		ceph_con_revoke(&req->r_osd->o_con, req->r_request);
867 		req->r_sent = 0;
868 	}
869 }
870 
871 static void __register_linger_request(struct ceph_osd_client *osdc,
872 				    struct ceph_osd_request *req)
873 {
874 	dout("__register_linger_request %p\n", req);
875 	list_add_tail(&req->r_linger_item, &osdc->req_linger);
876 	list_add_tail(&req->r_linger_osd, &req->r_osd->o_linger_requests);
877 }
878 
879 static void __unregister_linger_request(struct ceph_osd_client *osdc,
880 					struct ceph_osd_request *req)
881 {
882 	dout("__unregister_linger_request %p\n", req);
883 	if (req->r_osd) {
884 		list_del_init(&req->r_linger_item);
885 		list_del_init(&req->r_linger_osd);
886 
887 		if (list_empty(&req->r_osd->o_requests) &&
888 		    list_empty(&req->r_osd->o_linger_requests)) {
889 			dout("moving osd to %p lru\n", req->r_osd);
890 			__move_osd_to_lru(osdc, req->r_osd);
891 		}
892 		if (list_empty(&req->r_osd_item))
893 			req->r_osd = NULL;
894 	}
895 }
896 
897 void ceph_osdc_unregister_linger_request(struct ceph_osd_client *osdc,
898 					 struct ceph_osd_request *req)
899 {
900 	mutex_lock(&osdc->request_mutex);
901 	if (req->r_linger) {
902 		__unregister_linger_request(osdc, req);
903 		ceph_osdc_put_request(req);
904 	}
905 	mutex_unlock(&osdc->request_mutex);
906 }
907 EXPORT_SYMBOL(ceph_osdc_unregister_linger_request);
908 
909 void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
910 				  struct ceph_osd_request *req)
911 {
912 	if (!req->r_linger) {
913 		dout("set_request_linger %p\n", req);
914 		req->r_linger = 1;
915 		/*
916 		 * caller is now responsible for calling
917 		 * unregister_linger_request
918 		 */
919 		ceph_osdc_get_request(req);
920 	}
921 }
922 EXPORT_SYMBOL(ceph_osdc_set_request_linger);
923 
924 /*
925  * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
926  * (as needed), and set the request r_osd appropriately.  If there is
927  * no up osd, set r_osd to NULL.  Move the request to the appropriate list
928  * (unsent, homeless) or leave on in-flight lru.
929  *
930  * Return 0 if unchanged, 1 if changed, or negative on error.
931  *
932  * Caller should hold map_sem for read and request_mutex.
933  */
934 static int __map_request(struct ceph_osd_client *osdc,
935 			 struct ceph_osd_request *req)
936 {
937 	struct ceph_osd_request_head *reqhead = req->r_request->front.iov_base;
938 	struct ceph_pg pgid;
939 	int acting[CEPH_PG_MAX_SIZE];
940 	int o = -1, num = 0;
941 	int err;
942 
943 	dout("map_request %p tid %lld\n", req, req->r_tid);
944 	err = ceph_calc_object_layout(&reqhead->layout, req->r_oid,
945 				      &req->r_file_layout, osdc->osdmap);
946 	if (err) {
947 		list_move(&req->r_req_lru_item, &osdc->req_notarget);
948 		return err;
949 	}
950 	pgid = reqhead->layout.ol_pgid;
951 	req->r_pgid = pgid;
952 
953 	err = ceph_calc_pg_acting(osdc->osdmap, pgid, acting);
954 	if (err > 0) {
955 		o = acting[0];
956 		num = err;
957 	}
958 
959 	if ((req->r_osd && req->r_osd->o_osd == o &&
960 	     req->r_sent >= req->r_osd->o_incarnation &&
961 	     req->r_num_pg_osds == num &&
962 	     memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
963 	    (req->r_osd == NULL && o == -1))
964 		return 0;  /* no change */
965 
966 	dout("map_request tid %llu pgid %d.%x osd%d (was osd%d)\n",
967 	     req->r_tid, le32_to_cpu(pgid.pool), le16_to_cpu(pgid.ps), o,
968 	     req->r_osd ? req->r_osd->o_osd : -1);
969 
970 	/* record full pg acting set */
971 	memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
972 	req->r_num_pg_osds = num;
973 
974 	if (req->r_osd) {
975 		__cancel_request(req);
976 		list_del_init(&req->r_osd_item);
977 		req->r_osd = NULL;
978 	}
979 
980 	req->r_osd = __lookup_osd(osdc, o);
981 	if (!req->r_osd && o >= 0) {
982 		err = -ENOMEM;
983 		req->r_osd = create_osd(osdc);
984 		if (!req->r_osd) {
985 			list_move(&req->r_req_lru_item, &osdc->req_notarget);
986 			goto out;
987 		}
988 
989 		dout("map_request osd %p is osd%d\n", req->r_osd, o);
990 		req->r_osd->o_osd = o;
991 		req->r_osd->o_con.peer_name.num = cpu_to_le64(o);
992 		__insert_osd(osdc, req->r_osd);
993 
994 		ceph_con_open(&req->r_osd->o_con, &osdc->osdmap->osd_addr[o]);
995 	}
996 
997 	if (req->r_osd) {
998 		__remove_osd_from_lru(req->r_osd);
999 		list_add(&req->r_osd_item, &req->r_osd->o_requests);
1000 		list_move(&req->r_req_lru_item, &osdc->req_unsent);
1001 	} else {
1002 		list_move(&req->r_req_lru_item, &osdc->req_notarget);
1003 	}
1004 	err = 1;   /* osd or pg changed */
1005 
1006 out:
1007 	return err;
1008 }
1009 
1010 /*
1011  * caller should hold map_sem (for read) and request_mutex
1012  */
1013 static int __send_request(struct ceph_osd_client *osdc,
1014 			  struct ceph_osd_request *req)
1015 {
1016 	struct ceph_osd_request_head *reqhead;
1017 
1018 	dout("send_request %p tid %llu to osd%d flags %d\n",
1019 	     req, req->r_tid, req->r_osd->o_osd, req->r_flags);
1020 
1021 	reqhead = req->r_request->front.iov_base;
1022 	reqhead->osdmap_epoch = cpu_to_le32(osdc->osdmap->epoch);
1023 	reqhead->flags |= cpu_to_le32(req->r_flags);  /* e.g., RETRY */
1024 	reqhead->reassert_version = req->r_reassert_version;
1025 
1026 	req->r_stamp = jiffies;
1027 	list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
1028 
1029 	ceph_msg_get(req->r_request); /* send consumes a ref */
1030 	ceph_con_send(&req->r_osd->o_con, req->r_request);
1031 	req->r_sent = req->r_osd->o_incarnation;
1032 	return 0;
1033 }
1034 
1035 /*
1036  * Send any requests in the queue (req_unsent).
1037  */
1038 static void send_queued(struct ceph_osd_client *osdc)
1039 {
1040 	struct ceph_osd_request *req, *tmp;
1041 
1042 	dout("send_queued\n");
1043 	mutex_lock(&osdc->request_mutex);
1044 	list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item) {
1045 		__send_request(osdc, req);
1046 	}
1047 	mutex_unlock(&osdc->request_mutex);
1048 }
1049 
1050 /*
1051  * Timeout callback, called every N seconds when 1 or more osd
1052  * requests has been active for more than N seconds.  When this
1053  * happens, we ping all OSDs with requests who have timed out to
1054  * ensure any communications channel reset is detected.  Reset the
1055  * request timeouts another N seconds in the future as we go.
1056  * Reschedule the timeout event another N seconds in future (unless
1057  * there are no open requests).
1058  */
1059 static void handle_timeout(struct work_struct *work)
1060 {
1061 	struct ceph_osd_client *osdc =
1062 		container_of(work, struct ceph_osd_client, timeout_work.work);
1063 	struct ceph_osd_request *req, *last_req = NULL;
1064 	struct ceph_osd *osd;
1065 	unsigned long timeout = osdc->client->options->osd_timeout * HZ;
1066 	unsigned long keepalive =
1067 		osdc->client->options->osd_keepalive_timeout * HZ;
1068 	unsigned long last_stamp = 0;
1069 	struct list_head slow_osds;
1070 	dout("timeout\n");
1071 	down_read(&osdc->map_sem);
1072 
1073 	ceph_monc_request_next_osdmap(&osdc->client->monc);
1074 
1075 	mutex_lock(&osdc->request_mutex);
1076 
1077 	/*
1078 	 * reset osds that appear to be _really_ unresponsive.  this
1079 	 * is a failsafe measure.. we really shouldn't be getting to
1080 	 * this point if the system is working properly.  the monitors
1081 	 * should mark the osd as failed and we should find out about
1082 	 * it from an updated osd map.
1083 	 */
1084 	while (timeout && !list_empty(&osdc->req_lru)) {
1085 		req = list_entry(osdc->req_lru.next, struct ceph_osd_request,
1086 				 r_req_lru_item);
1087 
1088 		/* hasn't been long enough since we sent it? */
1089 		if (time_before(jiffies, req->r_stamp + timeout))
1090 			break;
1091 
1092 		/* hasn't been long enough since it was acked? */
1093 		if (req->r_request->ack_stamp == 0 ||
1094 		    time_before(jiffies, req->r_request->ack_stamp + timeout))
1095 			break;
1096 
1097 		BUG_ON(req == last_req && req->r_stamp == last_stamp);
1098 		last_req = req;
1099 		last_stamp = req->r_stamp;
1100 
1101 		osd = req->r_osd;
1102 		BUG_ON(!osd);
1103 		pr_warning(" tid %llu timed out on osd%d, will reset osd\n",
1104 			   req->r_tid, osd->o_osd);
1105 		__kick_osd_requests(osdc, osd);
1106 	}
1107 
1108 	/*
1109 	 * ping osds that are a bit slow.  this ensures that if there
1110 	 * is a break in the TCP connection we will notice, and reopen
1111 	 * a connection with that osd (from the fault callback).
1112 	 */
1113 	INIT_LIST_HEAD(&slow_osds);
1114 	list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
1115 		if (time_before(jiffies, req->r_stamp + keepalive))
1116 			break;
1117 
1118 		osd = req->r_osd;
1119 		BUG_ON(!osd);
1120 		dout(" tid %llu is slow, will send keepalive on osd%d\n",
1121 		     req->r_tid, osd->o_osd);
1122 		list_move_tail(&osd->o_keepalive_item, &slow_osds);
1123 	}
1124 	while (!list_empty(&slow_osds)) {
1125 		osd = list_entry(slow_osds.next, struct ceph_osd,
1126 				 o_keepalive_item);
1127 		list_del_init(&osd->o_keepalive_item);
1128 		ceph_con_keepalive(&osd->o_con);
1129 	}
1130 
1131 	__schedule_osd_timeout(osdc);
1132 	mutex_unlock(&osdc->request_mutex);
1133 	send_queued(osdc);
1134 	up_read(&osdc->map_sem);
1135 }
1136 
1137 static void handle_osds_timeout(struct work_struct *work)
1138 {
1139 	struct ceph_osd_client *osdc =
1140 		container_of(work, struct ceph_osd_client,
1141 			     osds_timeout_work.work);
1142 	unsigned long delay =
1143 		osdc->client->options->osd_idle_ttl * HZ >> 2;
1144 
1145 	dout("osds timeout\n");
1146 	down_read(&osdc->map_sem);
1147 	remove_old_osds(osdc, 0);
1148 	up_read(&osdc->map_sem);
1149 
1150 	schedule_delayed_work(&osdc->osds_timeout_work,
1151 			      round_jiffies_relative(delay));
1152 }
1153 
1154 static void complete_request(struct ceph_osd_request *req)
1155 {
1156 	if (req->r_safe_callback)
1157 		req->r_safe_callback(req, NULL);
1158 	complete_all(&req->r_safe_completion);  /* fsync waiter */
1159 }
1160 
1161 /*
1162  * handle osd op reply.  either call the callback if it is specified,
1163  * or do the completion to wake up the waiting thread.
1164  */
1165 static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg,
1166 			 struct ceph_connection *con)
1167 {
1168 	struct ceph_osd_reply_head *rhead = msg->front.iov_base;
1169 	struct ceph_osd_request *req;
1170 	u64 tid;
1171 	int numops, object_len, flags;
1172 	s32 result;
1173 
1174 	tid = le64_to_cpu(msg->hdr.tid);
1175 	if (msg->front.iov_len < sizeof(*rhead))
1176 		goto bad;
1177 	numops = le32_to_cpu(rhead->num_ops);
1178 	object_len = le32_to_cpu(rhead->object_len);
1179 	result = le32_to_cpu(rhead->result);
1180 	if (msg->front.iov_len != sizeof(*rhead) + object_len +
1181 	    numops * sizeof(struct ceph_osd_op))
1182 		goto bad;
1183 	dout("handle_reply %p tid %llu result %d\n", msg, tid, (int)result);
1184 	/* lookup */
1185 	mutex_lock(&osdc->request_mutex);
1186 	req = __lookup_request(osdc, tid);
1187 	if (req == NULL) {
1188 		dout("handle_reply tid %llu dne\n", tid);
1189 		mutex_unlock(&osdc->request_mutex);
1190 		return;
1191 	}
1192 	ceph_osdc_get_request(req);
1193 	flags = le32_to_cpu(rhead->flags);
1194 
1195 	/*
1196 	 * if this connection filled our message, drop our reference now, to
1197 	 * avoid a (safe but slower) revoke later.
1198 	 */
1199 	if (req->r_con_filling_msg == con && req->r_reply == msg) {
1200 		dout(" dropping con_filling_msg ref %p\n", con);
1201 		req->r_con_filling_msg = NULL;
1202 		ceph_con_put(con);
1203 	}
1204 
1205 	if (!req->r_got_reply) {
1206 		unsigned bytes;
1207 
1208 		req->r_result = le32_to_cpu(rhead->result);
1209 		bytes = le32_to_cpu(msg->hdr.data_len);
1210 		dout("handle_reply result %d bytes %d\n", req->r_result,
1211 		     bytes);
1212 		if (req->r_result == 0)
1213 			req->r_result = bytes;
1214 
1215 		/* in case this is a write and we need to replay, */
1216 		req->r_reassert_version = rhead->reassert_version;
1217 
1218 		req->r_got_reply = 1;
1219 	} else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) {
1220 		dout("handle_reply tid %llu dup ack\n", tid);
1221 		mutex_unlock(&osdc->request_mutex);
1222 		goto done;
1223 	}
1224 
1225 	dout("handle_reply tid %llu flags %d\n", tid, flags);
1226 
1227 	if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK))
1228 		__register_linger_request(osdc, req);
1229 
1230 	/* either this is a read, or we got the safe response */
1231 	if (result < 0 ||
1232 	    (flags & CEPH_OSD_FLAG_ONDISK) ||
1233 	    ((flags & CEPH_OSD_FLAG_WRITE) == 0))
1234 		__unregister_request(osdc, req);
1235 
1236 	mutex_unlock(&osdc->request_mutex);
1237 
1238 	if (req->r_callback)
1239 		req->r_callback(req, msg);
1240 	else
1241 		complete_all(&req->r_completion);
1242 
1243 	if (flags & CEPH_OSD_FLAG_ONDISK)
1244 		complete_request(req);
1245 
1246 done:
1247 	dout("req=%p req->r_linger=%d\n", req, req->r_linger);
1248 	ceph_osdc_put_request(req);
1249 	return;
1250 
1251 bad:
1252 	pr_err("corrupt osd_op_reply got %d %d expected %d\n",
1253 	       (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len),
1254 	       (int)sizeof(*rhead));
1255 	ceph_msg_dump(msg);
1256 }
1257 
1258 static void reset_changed_osds(struct ceph_osd_client *osdc)
1259 {
1260 	struct rb_node *p, *n;
1261 
1262 	for (p = rb_first(&osdc->osds); p; p = n) {
1263 		struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
1264 
1265 		n = rb_next(p);
1266 		if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
1267 		    memcmp(&osd->o_con.peer_addr,
1268 			   ceph_osd_addr(osdc->osdmap,
1269 					 osd->o_osd),
1270 			   sizeof(struct ceph_entity_addr)) != 0)
1271 			__reset_osd(osdc, osd);
1272 	}
1273 }
1274 
1275 /*
1276  * Requeue requests whose mapping to an OSD has changed.  If requests map to
1277  * no osd, request a new map.
1278  *
1279  * Caller should hold map_sem for read and request_mutex.
1280  */
1281 static void kick_requests(struct ceph_osd_client *osdc)
1282 {
1283 	struct ceph_osd_request *req, *nreq;
1284 	struct rb_node *p;
1285 	int needmap = 0;
1286 	int err;
1287 
1288 	dout("kick_requests\n");
1289 	mutex_lock(&osdc->request_mutex);
1290 	for (p = rb_first(&osdc->requests); p; p = rb_next(p)) {
1291 		req = rb_entry(p, struct ceph_osd_request, r_node);
1292 		err = __map_request(osdc, req);
1293 		if (err < 0)
1294 			continue;  /* error */
1295 		if (req->r_osd == NULL) {
1296 			dout("%p tid %llu maps to no osd\n", req, req->r_tid);
1297 			needmap++;  /* request a newer map */
1298 		} else if (err > 0) {
1299 			dout("%p tid %llu requeued on osd%d\n", req, req->r_tid,
1300 			     req->r_osd ? req->r_osd->o_osd : -1);
1301 			if (!req->r_linger)
1302 				req->r_flags |= CEPH_OSD_FLAG_RETRY;
1303 		}
1304 	}
1305 
1306 	list_for_each_entry_safe(req, nreq, &osdc->req_linger,
1307 				 r_linger_item) {
1308 		dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
1309 
1310 		err = __map_request(osdc, req);
1311 		if (err == 0)
1312 			continue;  /* no change and no osd was specified */
1313 		if (err < 0)
1314 			continue;  /* hrm! */
1315 		if (req->r_osd == NULL) {
1316 			dout("tid %llu maps to no valid osd\n", req->r_tid);
1317 			needmap++;  /* request a newer map */
1318 			continue;
1319 		}
1320 
1321 		dout("kicking lingering %p tid %llu osd%d\n", req, req->r_tid,
1322 		     req->r_osd ? req->r_osd->o_osd : -1);
1323 		__unregister_linger_request(osdc, req);
1324 		__register_request(osdc, req);
1325 	}
1326 	mutex_unlock(&osdc->request_mutex);
1327 
1328 	if (needmap) {
1329 		dout("%d requests for down osds, need new map\n", needmap);
1330 		ceph_monc_request_next_osdmap(&osdc->client->monc);
1331 	}
1332 }
1333 
1334 
1335 /*
1336  * Process updated osd map.
1337  *
1338  * The message contains any number of incremental and full maps, normally
1339  * indicating some sort of topology change in the cluster.  Kick requests
1340  * off to different OSDs as needed.
1341  */
1342 void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1343 {
1344 	void *p, *end, *next;
1345 	u32 nr_maps, maplen;
1346 	u32 epoch;
1347 	struct ceph_osdmap *newmap = NULL, *oldmap;
1348 	int err;
1349 	struct ceph_fsid fsid;
1350 
1351 	dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
1352 	p = msg->front.iov_base;
1353 	end = p + msg->front.iov_len;
1354 
1355 	/* verify fsid */
1356 	ceph_decode_need(&p, end, sizeof(fsid), bad);
1357 	ceph_decode_copy(&p, &fsid, sizeof(fsid));
1358 	if (ceph_check_fsid(osdc->client, &fsid) < 0)
1359 		return;
1360 
1361 	down_write(&osdc->map_sem);
1362 
1363 	/* incremental maps */
1364 	ceph_decode_32_safe(&p, end, nr_maps, bad);
1365 	dout(" %d inc maps\n", nr_maps);
1366 	while (nr_maps > 0) {
1367 		ceph_decode_need(&p, end, 2*sizeof(u32), bad);
1368 		epoch = ceph_decode_32(&p);
1369 		maplen = ceph_decode_32(&p);
1370 		ceph_decode_need(&p, end, maplen, bad);
1371 		next = p + maplen;
1372 		if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
1373 			dout("applying incremental map %u len %d\n",
1374 			     epoch, maplen);
1375 			newmap = osdmap_apply_incremental(&p, next,
1376 							  osdc->osdmap,
1377 							  osdc->client->msgr);
1378 			if (IS_ERR(newmap)) {
1379 				err = PTR_ERR(newmap);
1380 				goto bad;
1381 			}
1382 			BUG_ON(!newmap);
1383 			if (newmap != osdc->osdmap) {
1384 				ceph_osdmap_destroy(osdc->osdmap);
1385 				osdc->osdmap = newmap;
1386 			}
1387 			kick_requests(osdc);
1388 			reset_changed_osds(osdc);
1389 		} else {
1390 			dout("ignoring incremental map %u len %d\n",
1391 			     epoch, maplen);
1392 		}
1393 		p = next;
1394 		nr_maps--;
1395 	}
1396 	if (newmap)
1397 		goto done;
1398 
1399 	/* full maps */
1400 	ceph_decode_32_safe(&p, end, nr_maps, bad);
1401 	dout(" %d full maps\n", nr_maps);
1402 	while (nr_maps) {
1403 		ceph_decode_need(&p, end, 2*sizeof(u32), bad);
1404 		epoch = ceph_decode_32(&p);
1405 		maplen = ceph_decode_32(&p);
1406 		ceph_decode_need(&p, end, maplen, bad);
1407 		if (nr_maps > 1) {
1408 			dout("skipping non-latest full map %u len %d\n",
1409 			     epoch, maplen);
1410 		} else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
1411 			dout("skipping full map %u len %d, "
1412 			     "older than our %u\n", epoch, maplen,
1413 			     osdc->osdmap->epoch);
1414 		} else {
1415 			dout("taking full map %u len %d\n", epoch, maplen);
1416 			newmap = osdmap_decode(&p, p+maplen);
1417 			if (IS_ERR(newmap)) {
1418 				err = PTR_ERR(newmap);
1419 				goto bad;
1420 			}
1421 			BUG_ON(!newmap);
1422 			oldmap = osdc->osdmap;
1423 			osdc->osdmap = newmap;
1424 			if (oldmap)
1425 				ceph_osdmap_destroy(oldmap);
1426 			kick_requests(osdc);
1427 		}
1428 		p += maplen;
1429 		nr_maps--;
1430 	}
1431 
1432 done:
1433 	downgrade_write(&osdc->map_sem);
1434 	ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch);
1435 
1436 	/*
1437 	 * subscribe to subsequent osdmap updates if full to ensure
1438 	 * we find out when we are no longer full and stop returning
1439 	 * ENOSPC.
1440 	 */
1441 	if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL))
1442 		ceph_monc_request_next_osdmap(&osdc->client->monc);
1443 
1444 	send_queued(osdc);
1445 	up_read(&osdc->map_sem);
1446 	wake_up_all(&osdc->client->auth_wq);
1447 	return;
1448 
1449 bad:
1450 	pr_err("osdc handle_map corrupt msg\n");
1451 	ceph_msg_dump(msg);
1452 	up_write(&osdc->map_sem);
1453 	return;
1454 }
1455 
1456 /*
1457  * watch/notify callback event infrastructure
1458  *
1459  * These callbacks are used both for watch and notify operations.
1460  */
1461 static void __release_event(struct kref *kref)
1462 {
1463 	struct ceph_osd_event *event =
1464 		container_of(kref, struct ceph_osd_event, kref);
1465 
1466 	dout("__release_event %p\n", event);
1467 	kfree(event);
1468 }
1469 
1470 static void get_event(struct ceph_osd_event *event)
1471 {
1472 	kref_get(&event->kref);
1473 }
1474 
1475 void ceph_osdc_put_event(struct ceph_osd_event *event)
1476 {
1477 	kref_put(&event->kref, __release_event);
1478 }
1479 EXPORT_SYMBOL(ceph_osdc_put_event);
1480 
1481 static void __insert_event(struct ceph_osd_client *osdc,
1482 			     struct ceph_osd_event *new)
1483 {
1484 	struct rb_node **p = &osdc->event_tree.rb_node;
1485 	struct rb_node *parent = NULL;
1486 	struct ceph_osd_event *event = NULL;
1487 
1488 	while (*p) {
1489 		parent = *p;
1490 		event = rb_entry(parent, struct ceph_osd_event, node);
1491 		if (new->cookie < event->cookie)
1492 			p = &(*p)->rb_left;
1493 		else if (new->cookie > event->cookie)
1494 			p = &(*p)->rb_right;
1495 		else
1496 			BUG();
1497 	}
1498 
1499 	rb_link_node(&new->node, parent, p);
1500 	rb_insert_color(&new->node, &osdc->event_tree);
1501 }
1502 
1503 static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
1504 					        u64 cookie)
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 (cookie < event->cookie)
1514 			p = &(*p)->rb_left;
1515 		else if (cookie > event->cookie)
1516 			p = &(*p)->rb_right;
1517 		else
1518 			return event;
1519 	}
1520 	return NULL;
1521 }
1522 
1523 static void __remove_event(struct ceph_osd_event *event)
1524 {
1525 	struct ceph_osd_client *osdc = event->osdc;
1526 
1527 	if (!RB_EMPTY_NODE(&event->node)) {
1528 		dout("__remove_event removed %p\n", event);
1529 		rb_erase(&event->node, &osdc->event_tree);
1530 		ceph_osdc_put_event(event);
1531 	} else {
1532 		dout("__remove_event didn't remove %p\n", event);
1533 	}
1534 }
1535 
1536 int ceph_osdc_create_event(struct ceph_osd_client *osdc,
1537 			   void (*event_cb)(u64, u64, u8, void *),
1538 			   int one_shot, void *data,
1539 			   struct ceph_osd_event **pevent)
1540 {
1541 	struct ceph_osd_event *event;
1542 
1543 	event = kmalloc(sizeof(*event), GFP_NOIO);
1544 	if (!event)
1545 		return -ENOMEM;
1546 
1547 	dout("create_event %p\n", event);
1548 	event->cb = event_cb;
1549 	event->one_shot = one_shot;
1550 	event->data = data;
1551 	event->osdc = osdc;
1552 	INIT_LIST_HEAD(&event->osd_node);
1553 	kref_init(&event->kref);   /* one ref for us */
1554 	kref_get(&event->kref);    /* one ref for the caller */
1555 	init_completion(&event->completion);
1556 
1557 	spin_lock(&osdc->event_lock);
1558 	event->cookie = ++osdc->event_count;
1559 	__insert_event(osdc, event);
1560 	spin_unlock(&osdc->event_lock);
1561 
1562 	*pevent = event;
1563 	return 0;
1564 }
1565 EXPORT_SYMBOL(ceph_osdc_create_event);
1566 
1567 void ceph_osdc_cancel_event(struct ceph_osd_event *event)
1568 {
1569 	struct ceph_osd_client *osdc = event->osdc;
1570 
1571 	dout("cancel_event %p\n", event);
1572 	spin_lock(&osdc->event_lock);
1573 	__remove_event(event);
1574 	spin_unlock(&osdc->event_lock);
1575 	ceph_osdc_put_event(event); /* caller's */
1576 }
1577 EXPORT_SYMBOL(ceph_osdc_cancel_event);
1578 
1579 
1580 static void do_event_work(struct work_struct *work)
1581 {
1582 	struct ceph_osd_event_work *event_work =
1583 		container_of(work, struct ceph_osd_event_work, work);
1584 	struct ceph_osd_event *event = event_work->event;
1585 	u64 ver = event_work->ver;
1586 	u64 notify_id = event_work->notify_id;
1587 	u8 opcode = event_work->opcode;
1588 
1589 	dout("do_event_work completing %p\n", event);
1590 	event->cb(ver, notify_id, opcode, event->data);
1591 	complete(&event->completion);
1592 	dout("do_event_work completed %p\n", event);
1593 	ceph_osdc_put_event(event);
1594 	kfree(event_work);
1595 }
1596 
1597 
1598 /*
1599  * Process osd watch notifications
1600  */
1601 void handle_watch_notify(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1602 {
1603 	void *p, *end;
1604 	u8 proto_ver;
1605 	u64 cookie, ver, notify_id;
1606 	u8 opcode;
1607 	struct ceph_osd_event *event;
1608 	struct ceph_osd_event_work *event_work;
1609 
1610 	p = msg->front.iov_base;
1611 	end = p + msg->front.iov_len;
1612 
1613 	ceph_decode_8_safe(&p, end, proto_ver, bad);
1614 	ceph_decode_8_safe(&p, end, opcode, bad);
1615 	ceph_decode_64_safe(&p, end, cookie, bad);
1616 	ceph_decode_64_safe(&p, end, ver, bad);
1617 	ceph_decode_64_safe(&p, end, notify_id, bad);
1618 
1619 	spin_lock(&osdc->event_lock);
1620 	event = __find_event(osdc, cookie);
1621 	if (event) {
1622 		get_event(event);
1623 		if (event->one_shot)
1624 			__remove_event(event);
1625 	}
1626 	spin_unlock(&osdc->event_lock);
1627 	dout("handle_watch_notify cookie %lld ver %lld event %p\n",
1628 	     cookie, ver, event);
1629 	if (event) {
1630 		event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
1631 		if (!event_work) {
1632 			dout("ERROR: could not allocate event_work\n");
1633 			goto done_err;
1634 		}
1635 		INIT_WORK(&event_work->work, do_event_work);
1636 		event_work->event = event;
1637 		event_work->ver = ver;
1638 		event_work->notify_id = notify_id;
1639 		event_work->opcode = opcode;
1640 		if (!queue_work(osdc->notify_wq, &event_work->work)) {
1641 			dout("WARNING: failed to queue notify event work\n");
1642 			goto done_err;
1643 		}
1644 	}
1645 
1646 	return;
1647 
1648 done_err:
1649 	complete(&event->completion);
1650 	ceph_osdc_put_event(event);
1651 	return;
1652 
1653 bad:
1654 	pr_err("osdc handle_watch_notify corrupt msg\n");
1655 	return;
1656 }
1657 
1658 int ceph_osdc_wait_event(struct ceph_osd_event *event, unsigned long timeout)
1659 {
1660 	int err;
1661 
1662 	dout("wait_event %p\n", event);
1663 	err = wait_for_completion_interruptible_timeout(&event->completion,
1664 							timeout * HZ);
1665 	ceph_osdc_put_event(event);
1666 	if (err > 0)
1667 		err = 0;
1668 	dout("wait_event %p returns %d\n", event, err);
1669 	return err;
1670 }
1671 EXPORT_SYMBOL(ceph_osdc_wait_event);
1672 
1673 /*
1674  * Register request, send initial attempt.
1675  */
1676 int ceph_osdc_start_request(struct ceph_osd_client *osdc,
1677 			    struct ceph_osd_request *req,
1678 			    bool nofail)
1679 {
1680 	int rc = 0;
1681 
1682 	req->r_request->pages = req->r_pages;
1683 	req->r_request->nr_pages = req->r_num_pages;
1684 #ifdef CONFIG_BLOCK
1685 	req->r_request->bio = req->r_bio;
1686 #endif
1687 	req->r_request->trail = req->r_trail;
1688 
1689 	register_request(osdc, req);
1690 
1691 	down_read(&osdc->map_sem);
1692 	mutex_lock(&osdc->request_mutex);
1693 	/*
1694 	 * a racing kick_requests() may have sent the message for us
1695 	 * while we dropped request_mutex above, so only send now if
1696 	 * the request still han't been touched yet.
1697 	 */
1698 	if (req->r_sent == 0) {
1699 		rc = __map_request(osdc, req);
1700 		if (rc < 0) {
1701 			if (nofail) {
1702 				dout("osdc_start_request failed map, "
1703 				     " will retry %lld\n", req->r_tid);
1704 				rc = 0;
1705 			}
1706 			goto out_unlock;
1707 		}
1708 		if (req->r_osd == NULL) {
1709 			dout("send_request %p no up osds in pg\n", req);
1710 			ceph_monc_request_next_osdmap(&osdc->client->monc);
1711 		} else {
1712 			rc = __send_request(osdc, req);
1713 			if (rc) {
1714 				if (nofail) {
1715 					dout("osdc_start_request failed send, "
1716 					     " will retry %lld\n", req->r_tid);
1717 					rc = 0;
1718 				} else {
1719 					__unregister_request(osdc, req);
1720 				}
1721 			}
1722 		}
1723 	}
1724 
1725 out_unlock:
1726 	mutex_unlock(&osdc->request_mutex);
1727 	up_read(&osdc->map_sem);
1728 	return rc;
1729 }
1730 EXPORT_SYMBOL(ceph_osdc_start_request);
1731 
1732 /*
1733  * wait for a request to complete
1734  */
1735 int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
1736 			   struct ceph_osd_request *req)
1737 {
1738 	int rc;
1739 
1740 	rc = wait_for_completion_interruptible(&req->r_completion);
1741 	if (rc < 0) {
1742 		mutex_lock(&osdc->request_mutex);
1743 		__cancel_request(req);
1744 		__unregister_request(osdc, req);
1745 		mutex_unlock(&osdc->request_mutex);
1746 		complete_request(req);
1747 		dout("wait_request tid %llu canceled/timed out\n", req->r_tid);
1748 		return rc;
1749 	}
1750 
1751 	dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result);
1752 	return req->r_result;
1753 }
1754 EXPORT_SYMBOL(ceph_osdc_wait_request);
1755 
1756 /*
1757  * sync - wait for all in-flight requests to flush.  avoid starvation.
1758  */
1759 void ceph_osdc_sync(struct ceph_osd_client *osdc)
1760 {
1761 	struct ceph_osd_request *req;
1762 	u64 last_tid, next_tid = 0;
1763 
1764 	mutex_lock(&osdc->request_mutex);
1765 	last_tid = osdc->last_tid;
1766 	while (1) {
1767 		req = __lookup_request_ge(osdc, next_tid);
1768 		if (!req)
1769 			break;
1770 		if (req->r_tid > last_tid)
1771 			break;
1772 
1773 		next_tid = req->r_tid + 1;
1774 		if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
1775 			continue;
1776 
1777 		ceph_osdc_get_request(req);
1778 		mutex_unlock(&osdc->request_mutex);
1779 		dout("sync waiting on tid %llu (last is %llu)\n",
1780 		     req->r_tid, last_tid);
1781 		wait_for_completion(&req->r_safe_completion);
1782 		mutex_lock(&osdc->request_mutex);
1783 		ceph_osdc_put_request(req);
1784 	}
1785 	mutex_unlock(&osdc->request_mutex);
1786 	dout("sync done (thru tid %llu)\n", last_tid);
1787 }
1788 EXPORT_SYMBOL(ceph_osdc_sync);
1789 
1790 /*
1791  * init, shutdown
1792  */
1793 int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
1794 {
1795 	int err;
1796 
1797 	dout("init\n");
1798 	osdc->client = client;
1799 	osdc->osdmap = NULL;
1800 	init_rwsem(&osdc->map_sem);
1801 	init_completion(&osdc->map_waiters);
1802 	osdc->last_requested_map = 0;
1803 	mutex_init(&osdc->request_mutex);
1804 	osdc->last_tid = 0;
1805 	osdc->osds = RB_ROOT;
1806 	INIT_LIST_HEAD(&osdc->osd_lru);
1807 	osdc->requests = RB_ROOT;
1808 	INIT_LIST_HEAD(&osdc->req_lru);
1809 	INIT_LIST_HEAD(&osdc->req_unsent);
1810 	INIT_LIST_HEAD(&osdc->req_notarget);
1811 	INIT_LIST_HEAD(&osdc->req_linger);
1812 	osdc->num_requests = 0;
1813 	INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
1814 	INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
1815 	spin_lock_init(&osdc->event_lock);
1816 	osdc->event_tree = RB_ROOT;
1817 	osdc->event_count = 0;
1818 
1819 	schedule_delayed_work(&osdc->osds_timeout_work,
1820 	   round_jiffies_relative(osdc->client->options->osd_idle_ttl * HZ));
1821 
1822 	err = -ENOMEM;
1823 	osdc->req_mempool = mempool_create_kmalloc_pool(10,
1824 					sizeof(struct ceph_osd_request));
1825 	if (!osdc->req_mempool)
1826 		goto out;
1827 
1828 	err = ceph_msgpool_init(&osdc->msgpool_op, OSD_OP_FRONT_LEN, 10, true,
1829 				"osd_op");
1830 	if (err < 0)
1831 		goto out_mempool;
1832 	err = ceph_msgpool_init(&osdc->msgpool_op_reply,
1833 				OSD_OPREPLY_FRONT_LEN, 10, true,
1834 				"osd_op_reply");
1835 	if (err < 0)
1836 		goto out_msgpool;
1837 
1838 	osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
1839 	if (IS_ERR(osdc->notify_wq)) {
1840 		err = PTR_ERR(osdc->notify_wq);
1841 		osdc->notify_wq = NULL;
1842 		goto out_msgpool;
1843 	}
1844 	return 0;
1845 
1846 out_msgpool:
1847 	ceph_msgpool_destroy(&osdc->msgpool_op);
1848 out_mempool:
1849 	mempool_destroy(osdc->req_mempool);
1850 out:
1851 	return err;
1852 }
1853 EXPORT_SYMBOL(ceph_osdc_init);
1854 
1855 void ceph_osdc_stop(struct ceph_osd_client *osdc)
1856 {
1857 	flush_workqueue(osdc->notify_wq);
1858 	destroy_workqueue(osdc->notify_wq);
1859 	cancel_delayed_work_sync(&osdc->timeout_work);
1860 	cancel_delayed_work_sync(&osdc->osds_timeout_work);
1861 	if (osdc->osdmap) {
1862 		ceph_osdmap_destroy(osdc->osdmap);
1863 		osdc->osdmap = NULL;
1864 	}
1865 	remove_old_osds(osdc, 1);
1866 	WARN_ON(!RB_EMPTY_ROOT(&osdc->osds));
1867 	mempool_destroy(osdc->req_mempool);
1868 	ceph_msgpool_destroy(&osdc->msgpool_op);
1869 	ceph_msgpool_destroy(&osdc->msgpool_op_reply);
1870 }
1871 EXPORT_SYMBOL(ceph_osdc_stop);
1872 
1873 /*
1874  * Read some contiguous pages.  If we cross a stripe boundary, shorten
1875  * *plen.  Return number of bytes read, or error.
1876  */
1877 int ceph_osdc_readpages(struct ceph_osd_client *osdc,
1878 			struct ceph_vino vino, struct ceph_file_layout *layout,
1879 			u64 off, u64 *plen,
1880 			u32 truncate_seq, u64 truncate_size,
1881 			struct page **pages, int num_pages, int page_align)
1882 {
1883 	struct ceph_osd_request *req;
1884 	int rc = 0;
1885 
1886 	dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
1887 	     vino.snap, off, *plen);
1888 	req = ceph_osdc_new_request(osdc, layout, vino, off, plen,
1889 				    CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
1890 				    NULL, 0, truncate_seq, truncate_size, NULL,
1891 				    false, 1, page_align);
1892 	if (!req)
1893 		return -ENOMEM;
1894 
1895 	/* it may be a short read due to an object boundary */
1896 	req->r_pages = pages;
1897 
1898 	dout("readpages  final extent is %llu~%llu (%d pages align %d)\n",
1899 	     off, *plen, req->r_num_pages, page_align);
1900 
1901 	rc = ceph_osdc_start_request(osdc, req, false);
1902 	if (!rc)
1903 		rc = ceph_osdc_wait_request(osdc, req);
1904 
1905 	ceph_osdc_put_request(req);
1906 	dout("readpages result %d\n", rc);
1907 	return rc;
1908 }
1909 EXPORT_SYMBOL(ceph_osdc_readpages);
1910 
1911 /*
1912  * do a synchronous write on N pages
1913  */
1914 int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
1915 			 struct ceph_file_layout *layout,
1916 			 struct ceph_snap_context *snapc,
1917 			 u64 off, u64 len,
1918 			 u32 truncate_seq, u64 truncate_size,
1919 			 struct timespec *mtime,
1920 			 struct page **pages, int num_pages,
1921 			 int flags, int do_sync, bool nofail)
1922 {
1923 	struct ceph_osd_request *req;
1924 	int rc = 0;
1925 	int page_align = off & ~PAGE_MASK;
1926 
1927 	BUG_ON(vino.snap != CEPH_NOSNAP);
1928 	req = ceph_osdc_new_request(osdc, layout, vino, off, &len,
1929 				    CEPH_OSD_OP_WRITE,
1930 				    flags | CEPH_OSD_FLAG_ONDISK |
1931 					    CEPH_OSD_FLAG_WRITE,
1932 				    snapc, do_sync,
1933 				    truncate_seq, truncate_size, mtime,
1934 				    nofail, 1, page_align);
1935 	if (!req)
1936 		return -ENOMEM;
1937 
1938 	/* it may be a short write due to an object boundary */
1939 	req->r_pages = pages;
1940 	dout("writepages %llu~%llu (%d pages)\n", off, len,
1941 	     req->r_num_pages);
1942 
1943 	rc = ceph_osdc_start_request(osdc, req, nofail);
1944 	if (!rc)
1945 		rc = ceph_osdc_wait_request(osdc, req);
1946 
1947 	ceph_osdc_put_request(req);
1948 	if (rc == 0)
1949 		rc = len;
1950 	dout("writepages result %d\n", rc);
1951 	return rc;
1952 }
1953 EXPORT_SYMBOL(ceph_osdc_writepages);
1954 
1955 /*
1956  * handle incoming message
1957  */
1958 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
1959 {
1960 	struct ceph_osd *osd = con->private;
1961 	struct ceph_osd_client *osdc;
1962 	int type = le16_to_cpu(msg->hdr.type);
1963 
1964 	if (!osd)
1965 		goto out;
1966 	osdc = osd->o_osdc;
1967 
1968 	switch (type) {
1969 	case CEPH_MSG_OSD_MAP:
1970 		ceph_osdc_handle_map(osdc, msg);
1971 		break;
1972 	case CEPH_MSG_OSD_OPREPLY:
1973 		handle_reply(osdc, msg, con);
1974 		break;
1975 	case CEPH_MSG_WATCH_NOTIFY:
1976 		handle_watch_notify(osdc, msg);
1977 		break;
1978 
1979 	default:
1980 		pr_err("received unknown message type %d %s\n", type,
1981 		       ceph_msg_type_name(type));
1982 	}
1983 out:
1984 	ceph_msg_put(msg);
1985 }
1986 
1987 /*
1988  * lookup and return message for incoming reply.  set up reply message
1989  * pages.
1990  */
1991 static struct ceph_msg *get_reply(struct ceph_connection *con,
1992 				  struct ceph_msg_header *hdr,
1993 				  int *skip)
1994 {
1995 	struct ceph_osd *osd = con->private;
1996 	struct ceph_osd_client *osdc = osd->o_osdc;
1997 	struct ceph_msg *m;
1998 	struct ceph_osd_request *req;
1999 	int front = le32_to_cpu(hdr->front_len);
2000 	int data_len = le32_to_cpu(hdr->data_len);
2001 	u64 tid;
2002 
2003 	tid = le64_to_cpu(hdr->tid);
2004 	mutex_lock(&osdc->request_mutex);
2005 	req = __lookup_request(osdc, tid);
2006 	if (!req) {
2007 		*skip = 1;
2008 		m = NULL;
2009 		pr_info("get_reply unknown tid %llu from osd%d\n", tid,
2010 			osd->o_osd);
2011 		goto out;
2012 	}
2013 
2014 	if (req->r_con_filling_msg) {
2015 		dout("get_reply revoking msg %p from old con %p\n",
2016 		     req->r_reply, req->r_con_filling_msg);
2017 		ceph_con_revoke_message(req->r_con_filling_msg, req->r_reply);
2018 		ceph_con_put(req->r_con_filling_msg);
2019 		req->r_con_filling_msg = NULL;
2020 	}
2021 
2022 	if (front > req->r_reply->front.iov_len) {
2023 		pr_warning("get_reply front %d > preallocated %d\n",
2024 			   front, (int)req->r_reply->front.iov_len);
2025 		m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front, GFP_NOFS);
2026 		if (!m)
2027 			goto out;
2028 		ceph_msg_put(req->r_reply);
2029 		req->r_reply = m;
2030 	}
2031 	m = ceph_msg_get(req->r_reply);
2032 
2033 	if (data_len > 0) {
2034 		int want = calc_pages_for(req->r_page_alignment, data_len);
2035 
2036 		if (unlikely(req->r_num_pages < want)) {
2037 			pr_warning("tid %lld reply has %d bytes %d pages, we"
2038 				   " had only %d pages ready\n", tid, data_len,
2039 				   want, req->r_num_pages);
2040 			*skip = 1;
2041 			ceph_msg_put(m);
2042 			m = NULL;
2043 			goto out;
2044 		}
2045 		m->pages = req->r_pages;
2046 		m->nr_pages = req->r_num_pages;
2047 		m->page_alignment = req->r_page_alignment;
2048 #ifdef CONFIG_BLOCK
2049 		m->bio = req->r_bio;
2050 #endif
2051 	}
2052 	*skip = 0;
2053 	req->r_con_filling_msg = ceph_con_get(con);
2054 	dout("get_reply tid %lld %p\n", tid, m);
2055 
2056 out:
2057 	mutex_unlock(&osdc->request_mutex);
2058 	return m;
2059 
2060 }
2061 
2062 static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2063 				  struct ceph_msg_header *hdr,
2064 				  int *skip)
2065 {
2066 	struct ceph_osd *osd = con->private;
2067 	int type = le16_to_cpu(hdr->type);
2068 	int front = le32_to_cpu(hdr->front_len);
2069 
2070 	switch (type) {
2071 	case CEPH_MSG_OSD_MAP:
2072 	case CEPH_MSG_WATCH_NOTIFY:
2073 		return ceph_msg_new(type, front, GFP_NOFS);
2074 	case CEPH_MSG_OSD_OPREPLY:
2075 		return get_reply(con, hdr, skip);
2076 	default:
2077 		pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
2078 			osd->o_osd);
2079 		*skip = 1;
2080 		return NULL;
2081 	}
2082 }
2083 
2084 /*
2085  * Wrappers to refcount containing ceph_osd struct
2086  */
2087 static struct ceph_connection *get_osd_con(struct ceph_connection *con)
2088 {
2089 	struct ceph_osd *osd = con->private;
2090 	if (get_osd(osd))
2091 		return con;
2092 	return NULL;
2093 }
2094 
2095 static void put_osd_con(struct ceph_connection *con)
2096 {
2097 	struct ceph_osd *osd = con->private;
2098 	put_osd(osd);
2099 }
2100 
2101 /*
2102  * authentication
2103  */
2104 static int get_authorizer(struct ceph_connection *con,
2105 			  void **buf, int *len, int *proto,
2106 			  void **reply_buf, int *reply_len, int force_new)
2107 {
2108 	struct ceph_osd *o = con->private;
2109 	struct ceph_osd_client *osdc = o->o_osdc;
2110 	struct ceph_auth_client *ac = osdc->client->monc.auth;
2111 	int ret = 0;
2112 
2113 	if (force_new && o->o_authorizer) {
2114 		ac->ops->destroy_authorizer(ac, o->o_authorizer);
2115 		o->o_authorizer = NULL;
2116 	}
2117 	if (o->o_authorizer == NULL) {
2118 		ret = ac->ops->create_authorizer(
2119 			ac, CEPH_ENTITY_TYPE_OSD,
2120 			&o->o_authorizer,
2121 			&o->o_authorizer_buf,
2122 			&o->o_authorizer_buf_len,
2123 			&o->o_authorizer_reply_buf,
2124 			&o->o_authorizer_reply_buf_len);
2125 		if (ret)
2126 			return ret;
2127 	}
2128 
2129 	*proto = ac->protocol;
2130 	*buf = o->o_authorizer_buf;
2131 	*len = o->o_authorizer_buf_len;
2132 	*reply_buf = o->o_authorizer_reply_buf;
2133 	*reply_len = o->o_authorizer_reply_buf_len;
2134 	return 0;
2135 }
2136 
2137 
2138 static int verify_authorizer_reply(struct ceph_connection *con, int len)
2139 {
2140 	struct ceph_osd *o = con->private;
2141 	struct ceph_osd_client *osdc = o->o_osdc;
2142 	struct ceph_auth_client *ac = osdc->client->monc.auth;
2143 
2144 	return ac->ops->verify_authorizer_reply(ac, o->o_authorizer, len);
2145 }
2146 
2147 static int invalidate_authorizer(struct ceph_connection *con)
2148 {
2149 	struct ceph_osd *o = con->private;
2150 	struct ceph_osd_client *osdc = o->o_osdc;
2151 	struct ceph_auth_client *ac = osdc->client->monc.auth;
2152 
2153 	if (ac->ops->invalidate_authorizer)
2154 		ac->ops->invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
2155 
2156 	return ceph_monc_validate_auth(&osdc->client->monc);
2157 }
2158 
2159 static const struct ceph_connection_operations osd_con_ops = {
2160 	.get = get_osd_con,
2161 	.put = put_osd_con,
2162 	.dispatch = dispatch,
2163 	.get_authorizer = get_authorizer,
2164 	.verify_authorizer_reply = verify_authorizer_reply,
2165 	.invalidate_authorizer = invalidate_authorizer,
2166 	.alloc_msg = alloc_msg,
2167 	.fault = osd_reset,
2168 };
2169