xref: /openbmc/linux/net/ceph/osd_client.c (revision e02493c0)
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_has_extent(int op)
36 {
37 	return (op == CEPH_OSD_OP_READ ||
38 		op == CEPH_OSD_OP_WRITE);
39 }
40 
41 /*
42  * Implement client access to distributed object storage cluster.
43  *
44  * All data objects are stored within a cluster/cloud of OSDs, or
45  * "object storage devices."  (Note that Ceph OSDs have _nothing_ to
46  * do with the T10 OSD extensions to SCSI.)  Ceph OSDs are simply
47  * remote daemons serving up and coordinating consistent and safe
48  * access to storage.
49  *
50  * Cluster membership and the mapping of data objects onto storage devices
51  * are described by the osd map.
52  *
53  * We keep track of pending OSD requests (read, write), resubmit
54  * requests to different OSDs when the cluster topology/data layout
55  * change, or retry the affected requests when the communications
56  * channel with an OSD is reset.
57  */
58 
59 /*
60  * calculate the mapping of a file extent onto an object, and fill out the
61  * request accordingly.  shorten extent as necessary if it crosses an
62  * object boundary.
63  *
64  * fill osd op in request message.
65  */
66 static int calc_layout(struct ceph_file_layout *layout, u64 off, u64 *plen,
67 		       struct ceph_osd_req_op *op, u64 *bno)
68 {
69 	u64 orig_len = *plen;
70 	u64 objoff = 0;
71 	u64 objlen = 0;
72 	int r;
73 
74 	/* object extent? */
75 	r = ceph_calc_file_object_mapping(layout, off, orig_len, bno,
76 					  &objoff, &objlen);
77 	if (r < 0)
78 		return r;
79 	if (objlen < orig_len) {
80 		*plen = objlen;
81 		dout(" skipping last %llu, final file extent %llu~%llu\n",
82 		     orig_len - *plen, off, *plen);
83 	}
84 
85 	if (op_has_extent(op->op)) {
86 		u32 osize = le32_to_cpu(layout->fl_object_size);
87 		op->extent.offset = objoff;
88 		op->extent.length = objlen;
89 		if (op->extent.truncate_size <= off - objoff) {
90 			op->extent.truncate_size = 0;
91 		} else {
92 			op->extent.truncate_size -= off - objoff;
93 			if (op->extent.truncate_size > osize)
94 				op->extent.truncate_size = osize;
95 		}
96 	}
97 	if (op->op == CEPH_OSD_OP_WRITE)
98 		op->payload_len = *plen;
99 
100 	dout("calc_layout bno=%llx %llu~%llu\n", *bno, objoff, objlen);
101 
102 	return 0;
103 }
104 
105 /*
106  * requests
107  */
108 void ceph_osdc_release_request(struct kref *kref)
109 {
110 	int num_pages;
111 	struct ceph_osd_request *req = container_of(kref,
112 						    struct ceph_osd_request,
113 						    r_kref);
114 
115 	if (req->r_request)
116 		ceph_msg_put(req->r_request);
117 	if (req->r_con_filling_msg) {
118 		dout("%s revoking msg %p from con %p\n", __func__,
119 		     req->r_reply, req->r_con_filling_msg);
120 		ceph_msg_revoke_incoming(req->r_reply);
121 		req->r_con_filling_msg->ops->put(req->r_con_filling_msg);
122 		req->r_con_filling_msg = NULL;
123 	}
124 	if (req->r_reply)
125 		ceph_msg_put(req->r_reply);
126 
127 	if (req->r_data_in.type == CEPH_OSD_DATA_TYPE_PAGES &&
128 			req->r_data_in.own_pages) {
129 		num_pages = calc_pages_for((u64)req->r_data_in.alignment,
130 						(u64)req->r_data_in.length);
131 		ceph_release_page_vector(req->r_data_in.pages, num_pages);
132 	}
133 	if (req->r_data_out.type == CEPH_OSD_DATA_TYPE_PAGES &&
134 			req->r_data_out.own_pages) {
135 		num_pages = calc_pages_for((u64)req->r_data_out.alignment,
136 						(u64)req->r_data_out.length);
137 		ceph_release_page_vector(req->r_data_out.pages, num_pages);
138 	}
139 
140 	ceph_put_snap_context(req->r_snapc);
141 	if (req->r_mempool)
142 		mempool_free(req, req->r_osdc->req_mempool);
143 	else
144 		kfree(req);
145 }
146 EXPORT_SYMBOL(ceph_osdc_release_request);
147 
148 struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
149 					       struct ceph_snap_context *snapc,
150 					       unsigned int num_ops,
151 					       bool use_mempool,
152 					       gfp_t gfp_flags)
153 {
154 	struct ceph_osd_request *req;
155 	struct ceph_msg *msg;
156 	size_t msg_size;
157 
158 	msg_size = 4 + 4 + 8 + 8 + 4+8;
159 	msg_size += 2 + 4 + 8 + 4 + 4; /* oloc */
160 	msg_size += 1 + 8 + 4 + 4;     /* pg_t */
161 	msg_size += 4 + MAX_OBJ_NAME_SIZE;
162 	msg_size += 2 + num_ops*sizeof(struct ceph_osd_op);
163 	msg_size += 8;  /* snapid */
164 	msg_size += 8;  /* snap_seq */
165 	msg_size += 8 * (snapc ? snapc->num_snaps : 0);  /* snaps */
166 	msg_size += 4;
167 
168 	if (use_mempool) {
169 		req = mempool_alloc(osdc->req_mempool, gfp_flags);
170 		memset(req, 0, sizeof(*req));
171 	} else {
172 		req = kzalloc(sizeof(*req), gfp_flags);
173 	}
174 	if (req == NULL)
175 		return NULL;
176 
177 	req->r_osdc = osdc;
178 	req->r_mempool = use_mempool;
179 
180 	kref_init(&req->r_kref);
181 	init_completion(&req->r_completion);
182 	init_completion(&req->r_safe_completion);
183 	RB_CLEAR_NODE(&req->r_node);
184 	INIT_LIST_HEAD(&req->r_unsafe_item);
185 	INIT_LIST_HEAD(&req->r_linger_item);
186 	INIT_LIST_HEAD(&req->r_linger_osd);
187 	INIT_LIST_HEAD(&req->r_req_lru_item);
188 	INIT_LIST_HEAD(&req->r_osd_item);
189 
190 	/* create reply message */
191 	if (use_mempool)
192 		msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0);
193 	else
194 		msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY,
195 				   OSD_OPREPLY_FRONT_LEN, gfp_flags, true);
196 	if (!msg) {
197 		ceph_osdc_put_request(req);
198 		return NULL;
199 	}
200 	req->r_reply = msg;
201 
202 	req->r_data_in.type = CEPH_OSD_DATA_TYPE_NONE;
203 	req->r_data_out.type = CEPH_OSD_DATA_TYPE_NONE;
204 
205 	/* create request message; allow space for oid */
206 	if (use_mempool)
207 		msg = ceph_msgpool_get(&osdc->msgpool_op, 0);
208 	else
209 		msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp_flags, true);
210 	if (!msg) {
211 		ceph_osdc_put_request(req);
212 		return NULL;
213 	}
214 
215 	memset(msg->front.iov_base, 0, msg->front.iov_len);
216 
217 	req->r_request = msg;
218 
219 	return req;
220 }
221 EXPORT_SYMBOL(ceph_osdc_alloc_request);
222 
223 static u64 osd_req_encode_op(struct ceph_osd_request *req,
224 			      struct ceph_osd_op *dst,
225 			      struct ceph_osd_req_op *src)
226 {
227 	u64 out_data_len = 0;
228 	struct ceph_pagelist *pagelist;
229 
230 	dst->op = cpu_to_le16(src->op);
231 
232 	switch (src->op) {
233 	case CEPH_OSD_OP_STAT:
234 		break;
235 	case CEPH_OSD_OP_READ:
236 	case CEPH_OSD_OP_WRITE:
237 		if (src->op == CEPH_OSD_OP_WRITE)
238 			out_data_len = src->extent.length;
239 		dst->extent.offset = cpu_to_le64(src->extent.offset);
240 		dst->extent.length = cpu_to_le64(src->extent.length);
241 		dst->extent.truncate_size =
242 			cpu_to_le64(src->extent.truncate_size);
243 		dst->extent.truncate_seq =
244 			cpu_to_le32(src->extent.truncate_seq);
245 		break;
246 	case CEPH_OSD_OP_CALL:
247 		pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
248 		BUG_ON(!pagelist);
249 		ceph_pagelist_init(pagelist);
250 
251 		dst->cls.class_len = src->cls.class_len;
252 		dst->cls.method_len = src->cls.method_len;
253 		dst->cls.indata_len = cpu_to_le32(src->cls.indata_len);
254 		ceph_pagelist_append(pagelist, src->cls.class_name,
255 				     src->cls.class_len);
256 		ceph_pagelist_append(pagelist, src->cls.method_name,
257 				     src->cls.method_len);
258 		ceph_pagelist_append(pagelist, src->cls.indata,
259 				     src->cls.indata_len);
260 
261 		req->r_data_out.type = CEPH_OSD_DATA_TYPE_PAGELIST;
262 		req->r_data_out.pagelist = pagelist;
263 		out_data_len = pagelist->length;
264 		break;
265 	case CEPH_OSD_OP_STARTSYNC:
266 		break;
267 	case CEPH_OSD_OP_NOTIFY_ACK:
268 	case CEPH_OSD_OP_WATCH:
269 		dst->watch.cookie = cpu_to_le64(src->watch.cookie);
270 		dst->watch.ver = cpu_to_le64(src->watch.ver);
271 		dst->watch.flag = src->watch.flag;
272 		break;
273 	default:
274 		pr_err("unrecognized osd opcode %d\n", src->op);
275 		WARN_ON(1);
276 		break;
277 	case CEPH_OSD_OP_MAPEXT:
278 	case CEPH_OSD_OP_MASKTRUNC:
279 	case CEPH_OSD_OP_SPARSE_READ:
280 	case CEPH_OSD_OP_NOTIFY:
281 	case CEPH_OSD_OP_ASSERT_VER:
282 	case CEPH_OSD_OP_WRITEFULL:
283 	case CEPH_OSD_OP_TRUNCATE:
284 	case CEPH_OSD_OP_ZERO:
285 	case CEPH_OSD_OP_DELETE:
286 	case CEPH_OSD_OP_APPEND:
287 	case CEPH_OSD_OP_SETTRUNC:
288 	case CEPH_OSD_OP_TRIMTRUNC:
289 	case CEPH_OSD_OP_TMAPUP:
290 	case CEPH_OSD_OP_TMAPPUT:
291 	case CEPH_OSD_OP_TMAPGET:
292 	case CEPH_OSD_OP_CREATE:
293 	case CEPH_OSD_OP_ROLLBACK:
294 	case CEPH_OSD_OP_OMAPGETKEYS:
295 	case CEPH_OSD_OP_OMAPGETVALS:
296 	case CEPH_OSD_OP_OMAPGETHEADER:
297 	case CEPH_OSD_OP_OMAPGETVALSBYKEYS:
298 	case CEPH_OSD_OP_MODE_RD:
299 	case CEPH_OSD_OP_OMAPSETVALS:
300 	case CEPH_OSD_OP_OMAPSETHEADER:
301 	case CEPH_OSD_OP_OMAPCLEAR:
302 	case CEPH_OSD_OP_OMAPRMKEYS:
303 	case CEPH_OSD_OP_OMAP_CMP:
304 	case CEPH_OSD_OP_CLONERANGE:
305 	case CEPH_OSD_OP_ASSERT_SRC_VERSION:
306 	case CEPH_OSD_OP_SRC_CMPXATTR:
307 	case CEPH_OSD_OP_GETXATTR:
308 	case CEPH_OSD_OP_GETXATTRS:
309 	case CEPH_OSD_OP_CMPXATTR:
310 	case CEPH_OSD_OP_SETXATTR:
311 	case CEPH_OSD_OP_SETXATTRS:
312 	case CEPH_OSD_OP_RESETXATTRS:
313 	case CEPH_OSD_OP_RMXATTR:
314 	case CEPH_OSD_OP_PULL:
315 	case CEPH_OSD_OP_PUSH:
316 	case CEPH_OSD_OP_BALANCEREADS:
317 	case CEPH_OSD_OP_UNBALANCEREADS:
318 	case CEPH_OSD_OP_SCRUB:
319 	case CEPH_OSD_OP_SCRUB_RESERVE:
320 	case CEPH_OSD_OP_SCRUB_UNRESERVE:
321 	case CEPH_OSD_OP_SCRUB_STOP:
322 	case CEPH_OSD_OP_SCRUB_MAP:
323 	case CEPH_OSD_OP_WRLOCK:
324 	case CEPH_OSD_OP_WRUNLOCK:
325 	case CEPH_OSD_OP_RDLOCK:
326 	case CEPH_OSD_OP_RDUNLOCK:
327 	case CEPH_OSD_OP_UPLOCK:
328 	case CEPH_OSD_OP_DNLOCK:
329 	case CEPH_OSD_OP_PGLS:
330 	case CEPH_OSD_OP_PGLS_FILTER:
331 		pr_err("unsupported osd opcode %s\n",
332 			ceph_osd_op_name(src->op));
333 		WARN_ON(1);
334 		break;
335 	}
336 	dst->payload_len = cpu_to_le32(src->payload_len);
337 
338 	return out_data_len;
339 }
340 
341 /*
342  * build new request AND message
343  *
344  */
345 void ceph_osdc_build_request(struct ceph_osd_request *req,
346 			     u64 off, unsigned int num_ops,
347 			     struct ceph_osd_req_op *src_ops,
348 			     struct ceph_snap_context *snapc, u64 snap_id,
349 			     struct timespec *mtime)
350 {
351 	struct ceph_msg *msg = req->r_request;
352 	struct ceph_osd_req_op *src_op;
353 	void *p;
354 	size_t msg_size;
355 	int flags = req->r_flags;
356 	u64 data_len;
357 	int i;
358 
359 	req->r_num_ops = num_ops;
360 	req->r_snapid = snap_id;
361 	req->r_snapc = ceph_get_snap_context(snapc);
362 
363 	/* encode request */
364 	msg->hdr.version = cpu_to_le16(4);
365 
366 	p = msg->front.iov_base;
367 	ceph_encode_32(&p, 1);   /* client_inc  is always 1 */
368 	req->r_request_osdmap_epoch = p;
369 	p += 4;
370 	req->r_request_flags = p;
371 	p += 4;
372 	if (req->r_flags & CEPH_OSD_FLAG_WRITE)
373 		ceph_encode_timespec(p, mtime);
374 	p += sizeof(struct ceph_timespec);
375 	req->r_request_reassert_version = p;
376 	p += sizeof(struct ceph_eversion); /* will get filled in */
377 
378 	/* oloc */
379 	ceph_encode_8(&p, 4);
380 	ceph_encode_8(&p, 4);
381 	ceph_encode_32(&p, 8 + 4 + 4);
382 	req->r_request_pool = p;
383 	p += 8;
384 	ceph_encode_32(&p, -1);  /* preferred */
385 	ceph_encode_32(&p, 0);   /* key len */
386 
387 	ceph_encode_8(&p, 1);
388 	req->r_request_pgid = p;
389 	p += 8 + 4;
390 	ceph_encode_32(&p, -1);  /* preferred */
391 
392 	/* oid */
393 	ceph_encode_32(&p, req->r_oid_len);
394 	memcpy(p, req->r_oid, req->r_oid_len);
395 	dout("oid '%.*s' len %d\n", req->r_oid_len, req->r_oid, req->r_oid_len);
396 	p += req->r_oid_len;
397 
398 	/* ops--can imply data */
399 	ceph_encode_16(&p, num_ops);
400 	src_op = src_ops;
401 	req->r_request_ops = p;
402 	data_len = 0;
403 	for (i = 0; i < num_ops; i++, src_op++) {
404 		data_len += osd_req_encode_op(req, p, src_op);
405 		p += sizeof(struct ceph_osd_op);
406 	}
407 
408 	/* snaps */
409 	ceph_encode_64(&p, req->r_snapid);
410 	ceph_encode_64(&p, req->r_snapc ? req->r_snapc->seq : 0);
411 	ceph_encode_32(&p, req->r_snapc ? req->r_snapc->num_snaps : 0);
412 	if (req->r_snapc) {
413 		for (i = 0; i < snapc->num_snaps; i++) {
414 			ceph_encode_64(&p, req->r_snapc->snaps[i]);
415 		}
416 	}
417 
418 	req->r_request_attempts = p;
419 	p += 4;
420 
421 	/* data */
422 	if (flags & CEPH_OSD_FLAG_WRITE)
423 		req->r_request->hdr.data_off = cpu_to_le16(off);
424 	req->r_request->hdr.data_len = cpu_to_le32(data_len);
425 
426 	BUG_ON(p > msg->front.iov_base + msg->front.iov_len);
427 	msg_size = p - msg->front.iov_base;
428 	msg->front.iov_len = msg_size;
429 	msg->hdr.front_len = cpu_to_le32(msg_size);
430 
431 	dout("build_request msg_size was %d num_ops %d\n", (int)msg_size,
432 	     num_ops);
433 	return;
434 }
435 EXPORT_SYMBOL(ceph_osdc_build_request);
436 
437 /*
438  * build new request AND message, calculate layout, and adjust file
439  * extent as needed.
440  *
441  * if the file was recently truncated, we include information about its
442  * old and new size so that the object can be updated appropriately.  (we
443  * avoid synchronously deleting truncated objects because it's slow.)
444  *
445  * if @do_sync, include a 'startsync' command so that the osd will flush
446  * data quickly.
447  */
448 struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
449 					       struct ceph_file_layout *layout,
450 					       struct ceph_vino vino,
451 					       u64 off, u64 *plen,
452 					       int opcode, int flags,
453 					       struct ceph_snap_context *snapc,
454 					       int do_sync,
455 					       u32 truncate_seq,
456 					       u64 truncate_size,
457 					       struct timespec *mtime,
458 					       bool use_mempool)
459 {
460 	struct ceph_osd_req_op ops[2];
461 	struct ceph_osd_request *req;
462 	unsigned int num_op = 1;
463 	u64 bno = 0;
464 	int r;
465 
466 	memset(&ops, 0, sizeof ops);
467 
468 	ops[0].op = opcode;
469 	ops[0].extent.truncate_seq = truncate_seq;
470 	ops[0].extent.truncate_size = truncate_size;
471 
472 	if (do_sync) {
473 		ops[1].op = CEPH_OSD_OP_STARTSYNC;
474 		num_op++;
475 	}
476 
477 	req = ceph_osdc_alloc_request(osdc, snapc, num_op, use_mempool,
478 					GFP_NOFS);
479 	if (!req)
480 		return ERR_PTR(-ENOMEM);
481 	req->r_flags = flags;
482 
483 	/* calculate max write size */
484 	r = calc_layout(layout, off, plen, ops, &bno);
485 	if (r < 0) {
486 		ceph_osdc_put_request(req);
487 		return ERR_PTR(r);
488 	}
489 	req->r_file_layout = *layout;  /* keep a copy */
490 
491 	snprintf(req->r_oid, sizeof(req->r_oid), "%llx.%08llx", vino.ino, bno);
492 	req->r_oid_len = strlen(req->r_oid);
493 
494 	ceph_osdc_build_request(req, off, num_op, ops,
495 				snapc, vino.snap, mtime);
496 
497 	return req;
498 }
499 EXPORT_SYMBOL(ceph_osdc_new_request);
500 
501 /*
502  * We keep osd requests in an rbtree, sorted by ->r_tid.
503  */
504 static void __insert_request(struct ceph_osd_client *osdc,
505 			     struct ceph_osd_request *new)
506 {
507 	struct rb_node **p = &osdc->requests.rb_node;
508 	struct rb_node *parent = NULL;
509 	struct ceph_osd_request *req = NULL;
510 
511 	while (*p) {
512 		parent = *p;
513 		req = rb_entry(parent, struct ceph_osd_request, r_node);
514 		if (new->r_tid < req->r_tid)
515 			p = &(*p)->rb_left;
516 		else if (new->r_tid > req->r_tid)
517 			p = &(*p)->rb_right;
518 		else
519 			BUG();
520 	}
521 
522 	rb_link_node(&new->r_node, parent, p);
523 	rb_insert_color(&new->r_node, &osdc->requests);
524 }
525 
526 static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc,
527 						 u64 tid)
528 {
529 	struct ceph_osd_request *req;
530 	struct rb_node *n = osdc->requests.rb_node;
531 
532 	while (n) {
533 		req = rb_entry(n, struct ceph_osd_request, r_node);
534 		if (tid < req->r_tid)
535 			n = n->rb_left;
536 		else if (tid > req->r_tid)
537 			n = n->rb_right;
538 		else
539 			return req;
540 	}
541 	return NULL;
542 }
543 
544 static struct ceph_osd_request *
545 __lookup_request_ge(struct ceph_osd_client *osdc,
546 		    u64 tid)
547 {
548 	struct ceph_osd_request *req;
549 	struct rb_node *n = osdc->requests.rb_node;
550 
551 	while (n) {
552 		req = rb_entry(n, struct ceph_osd_request, r_node);
553 		if (tid < req->r_tid) {
554 			if (!n->rb_left)
555 				return req;
556 			n = n->rb_left;
557 		} else if (tid > req->r_tid) {
558 			n = n->rb_right;
559 		} else {
560 			return req;
561 		}
562 	}
563 	return NULL;
564 }
565 
566 /*
567  * Resubmit requests pending on the given osd.
568  */
569 static void __kick_osd_requests(struct ceph_osd_client *osdc,
570 				struct ceph_osd *osd)
571 {
572 	struct ceph_osd_request *req, *nreq;
573 	LIST_HEAD(resend);
574 	int err;
575 
576 	dout("__kick_osd_requests osd%d\n", osd->o_osd);
577 	err = __reset_osd(osdc, osd);
578 	if (err)
579 		return;
580 	/*
581 	 * Build up a list of requests to resend by traversing the
582 	 * osd's list of requests.  Requests for a given object are
583 	 * sent in tid order, and that is also the order they're
584 	 * kept on this list.  Therefore all requests that are in
585 	 * flight will be found first, followed by all requests that
586 	 * have not yet been sent.  And to resend requests while
587 	 * preserving this order we will want to put any sent
588 	 * requests back on the front of the osd client's unsent
589 	 * list.
590 	 *
591 	 * So we build a separate ordered list of already-sent
592 	 * requests for the affected osd and splice it onto the
593 	 * front of the osd client's unsent list.  Once we've seen a
594 	 * request that has not yet been sent we're done.  Those
595 	 * requests are already sitting right where they belong.
596 	 */
597 	list_for_each_entry(req, &osd->o_requests, r_osd_item) {
598 		if (!req->r_sent)
599 			break;
600 		list_move_tail(&req->r_req_lru_item, &resend);
601 		dout("requeueing %p tid %llu osd%d\n", req, req->r_tid,
602 		     osd->o_osd);
603 		if (!req->r_linger)
604 			req->r_flags |= CEPH_OSD_FLAG_RETRY;
605 	}
606 	list_splice(&resend, &osdc->req_unsent);
607 
608 	/*
609 	 * Linger requests are re-registered before sending, which
610 	 * sets up a new tid for each.  We add them to the unsent
611 	 * list at the end to keep things in tid order.
612 	 */
613 	list_for_each_entry_safe(req, nreq, &osd->o_linger_requests,
614 				 r_linger_osd) {
615 		/*
616 		 * reregister request prior to unregistering linger so
617 		 * that r_osd is preserved.
618 		 */
619 		BUG_ON(!list_empty(&req->r_req_lru_item));
620 		__register_request(osdc, req);
621 		list_add_tail(&req->r_req_lru_item, &osdc->req_unsent);
622 		list_add(&req->r_osd_item, &req->r_osd->o_requests);
623 		__unregister_linger_request(osdc, req);
624 		dout("requeued lingering %p tid %llu osd%d\n", req, req->r_tid,
625 		     osd->o_osd);
626 	}
627 }
628 
629 /*
630  * If the osd connection drops, we need to resubmit all requests.
631  */
632 static void osd_reset(struct ceph_connection *con)
633 {
634 	struct ceph_osd *osd = con->private;
635 	struct ceph_osd_client *osdc;
636 
637 	if (!osd)
638 		return;
639 	dout("osd_reset osd%d\n", osd->o_osd);
640 	osdc = osd->o_osdc;
641 	down_read(&osdc->map_sem);
642 	mutex_lock(&osdc->request_mutex);
643 	__kick_osd_requests(osdc, osd);
644 	__send_queued(osdc);
645 	mutex_unlock(&osdc->request_mutex);
646 	up_read(&osdc->map_sem);
647 }
648 
649 /*
650  * Track open sessions with osds.
651  */
652 static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
653 {
654 	struct ceph_osd *osd;
655 
656 	osd = kzalloc(sizeof(*osd), GFP_NOFS);
657 	if (!osd)
658 		return NULL;
659 
660 	atomic_set(&osd->o_ref, 1);
661 	osd->o_osdc = osdc;
662 	osd->o_osd = onum;
663 	RB_CLEAR_NODE(&osd->o_node);
664 	INIT_LIST_HEAD(&osd->o_requests);
665 	INIT_LIST_HEAD(&osd->o_linger_requests);
666 	INIT_LIST_HEAD(&osd->o_osd_lru);
667 	osd->o_incarnation = 1;
668 
669 	ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
670 
671 	INIT_LIST_HEAD(&osd->o_keepalive_item);
672 	return osd;
673 }
674 
675 static struct ceph_osd *get_osd(struct ceph_osd *osd)
676 {
677 	if (atomic_inc_not_zero(&osd->o_ref)) {
678 		dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
679 		     atomic_read(&osd->o_ref));
680 		return osd;
681 	} else {
682 		dout("get_osd %p FAIL\n", osd);
683 		return NULL;
684 	}
685 }
686 
687 static void put_osd(struct ceph_osd *osd)
688 {
689 	dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
690 	     atomic_read(&osd->o_ref) - 1);
691 	if (atomic_dec_and_test(&osd->o_ref) && osd->o_auth.authorizer) {
692 		struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth;
693 
694 		ceph_auth_destroy_authorizer(ac, osd->o_auth.authorizer);
695 		kfree(osd);
696 	}
697 }
698 
699 /*
700  * remove an osd from our map
701  */
702 static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
703 {
704 	dout("__remove_osd %p\n", osd);
705 	BUG_ON(!list_empty(&osd->o_requests));
706 	rb_erase(&osd->o_node, &osdc->osds);
707 	list_del_init(&osd->o_osd_lru);
708 	ceph_con_close(&osd->o_con);
709 	put_osd(osd);
710 }
711 
712 static void remove_all_osds(struct ceph_osd_client *osdc)
713 {
714 	dout("%s %p\n", __func__, osdc);
715 	mutex_lock(&osdc->request_mutex);
716 	while (!RB_EMPTY_ROOT(&osdc->osds)) {
717 		struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
718 						struct ceph_osd, o_node);
719 		__remove_osd(osdc, osd);
720 	}
721 	mutex_unlock(&osdc->request_mutex);
722 }
723 
724 static void __move_osd_to_lru(struct ceph_osd_client *osdc,
725 			      struct ceph_osd *osd)
726 {
727 	dout("__move_osd_to_lru %p\n", osd);
728 	BUG_ON(!list_empty(&osd->o_osd_lru));
729 	list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
730 	osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl * HZ;
731 }
732 
733 static void __remove_osd_from_lru(struct ceph_osd *osd)
734 {
735 	dout("__remove_osd_from_lru %p\n", osd);
736 	if (!list_empty(&osd->o_osd_lru))
737 		list_del_init(&osd->o_osd_lru);
738 }
739 
740 static void remove_old_osds(struct ceph_osd_client *osdc)
741 {
742 	struct ceph_osd *osd, *nosd;
743 
744 	dout("__remove_old_osds %p\n", osdc);
745 	mutex_lock(&osdc->request_mutex);
746 	list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
747 		if (time_before(jiffies, osd->lru_ttl))
748 			break;
749 		__remove_osd(osdc, osd);
750 	}
751 	mutex_unlock(&osdc->request_mutex);
752 }
753 
754 /*
755  * reset osd connect
756  */
757 static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
758 {
759 	struct ceph_entity_addr *peer_addr;
760 
761 	dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
762 	if (list_empty(&osd->o_requests) &&
763 	    list_empty(&osd->o_linger_requests)) {
764 		__remove_osd(osdc, osd);
765 
766 		return -ENODEV;
767 	}
768 
769 	peer_addr = &osdc->osdmap->osd_addr[osd->o_osd];
770 	if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
771 			!ceph_con_opened(&osd->o_con)) {
772 		struct ceph_osd_request *req;
773 
774 		dout(" osd addr hasn't changed and connection never opened,"
775 		     " letting msgr retry");
776 		/* touch each r_stamp for handle_timeout()'s benfit */
777 		list_for_each_entry(req, &osd->o_requests, r_osd_item)
778 			req->r_stamp = jiffies;
779 
780 		return -EAGAIN;
781 	}
782 
783 	ceph_con_close(&osd->o_con);
784 	ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
785 	osd->o_incarnation++;
786 
787 	return 0;
788 }
789 
790 static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new)
791 {
792 	struct rb_node **p = &osdc->osds.rb_node;
793 	struct rb_node *parent = NULL;
794 	struct ceph_osd *osd = NULL;
795 
796 	dout("__insert_osd %p osd%d\n", new, new->o_osd);
797 	while (*p) {
798 		parent = *p;
799 		osd = rb_entry(parent, struct ceph_osd, o_node);
800 		if (new->o_osd < osd->o_osd)
801 			p = &(*p)->rb_left;
802 		else if (new->o_osd > osd->o_osd)
803 			p = &(*p)->rb_right;
804 		else
805 			BUG();
806 	}
807 
808 	rb_link_node(&new->o_node, parent, p);
809 	rb_insert_color(&new->o_node, &osdc->osds);
810 }
811 
812 static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o)
813 {
814 	struct ceph_osd *osd;
815 	struct rb_node *n = osdc->osds.rb_node;
816 
817 	while (n) {
818 		osd = rb_entry(n, struct ceph_osd, o_node);
819 		if (o < osd->o_osd)
820 			n = n->rb_left;
821 		else if (o > osd->o_osd)
822 			n = n->rb_right;
823 		else
824 			return osd;
825 	}
826 	return NULL;
827 }
828 
829 static void __schedule_osd_timeout(struct ceph_osd_client *osdc)
830 {
831 	schedule_delayed_work(&osdc->timeout_work,
832 			osdc->client->options->osd_keepalive_timeout * HZ);
833 }
834 
835 static void __cancel_osd_timeout(struct ceph_osd_client *osdc)
836 {
837 	cancel_delayed_work(&osdc->timeout_work);
838 }
839 
840 /*
841  * Register request, assign tid.  If this is the first request, set up
842  * the timeout event.
843  */
844 static void __register_request(struct ceph_osd_client *osdc,
845 			       struct ceph_osd_request *req)
846 {
847 	req->r_tid = ++osdc->last_tid;
848 	req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
849 	dout("__register_request %p tid %lld\n", req, req->r_tid);
850 	__insert_request(osdc, req);
851 	ceph_osdc_get_request(req);
852 	osdc->num_requests++;
853 	if (osdc->num_requests == 1) {
854 		dout(" first request, scheduling timeout\n");
855 		__schedule_osd_timeout(osdc);
856 	}
857 }
858 
859 /*
860  * called under osdc->request_mutex
861  */
862 static void __unregister_request(struct ceph_osd_client *osdc,
863 				 struct ceph_osd_request *req)
864 {
865 	if (RB_EMPTY_NODE(&req->r_node)) {
866 		dout("__unregister_request %p tid %lld not registered\n",
867 			req, req->r_tid);
868 		return;
869 	}
870 
871 	dout("__unregister_request %p tid %lld\n", req, req->r_tid);
872 	rb_erase(&req->r_node, &osdc->requests);
873 	osdc->num_requests--;
874 
875 	if (req->r_osd) {
876 		/* make sure the original request isn't in flight. */
877 		ceph_msg_revoke(req->r_request);
878 
879 		list_del_init(&req->r_osd_item);
880 		if (list_empty(&req->r_osd->o_requests) &&
881 		    list_empty(&req->r_osd->o_linger_requests)) {
882 			dout("moving osd to %p lru\n", req->r_osd);
883 			__move_osd_to_lru(osdc, req->r_osd);
884 		}
885 		if (list_empty(&req->r_linger_item))
886 			req->r_osd = NULL;
887 	}
888 
889 	list_del_init(&req->r_req_lru_item);
890 	ceph_osdc_put_request(req);
891 
892 	if (osdc->num_requests == 0) {
893 		dout(" no requests, canceling timeout\n");
894 		__cancel_osd_timeout(osdc);
895 	}
896 }
897 
898 /*
899  * Cancel a previously queued request message
900  */
901 static void __cancel_request(struct ceph_osd_request *req)
902 {
903 	if (req->r_sent && req->r_osd) {
904 		ceph_msg_revoke(req->r_request);
905 		req->r_sent = 0;
906 	}
907 }
908 
909 static void __register_linger_request(struct ceph_osd_client *osdc,
910 				    struct ceph_osd_request *req)
911 {
912 	dout("__register_linger_request %p\n", req);
913 	list_add_tail(&req->r_linger_item, &osdc->req_linger);
914 	if (req->r_osd)
915 		list_add_tail(&req->r_linger_osd,
916 			      &req->r_osd->o_linger_requests);
917 }
918 
919 static void __unregister_linger_request(struct ceph_osd_client *osdc,
920 					struct ceph_osd_request *req)
921 {
922 	dout("__unregister_linger_request %p\n", req);
923 	list_del_init(&req->r_linger_item);
924 	if (req->r_osd) {
925 		list_del_init(&req->r_linger_osd);
926 
927 		if (list_empty(&req->r_osd->o_requests) &&
928 		    list_empty(&req->r_osd->o_linger_requests)) {
929 			dout("moving osd to %p lru\n", req->r_osd);
930 			__move_osd_to_lru(osdc, req->r_osd);
931 		}
932 		if (list_empty(&req->r_osd_item))
933 			req->r_osd = NULL;
934 	}
935 }
936 
937 void ceph_osdc_unregister_linger_request(struct ceph_osd_client *osdc,
938 					 struct ceph_osd_request *req)
939 {
940 	mutex_lock(&osdc->request_mutex);
941 	if (req->r_linger) {
942 		__unregister_linger_request(osdc, req);
943 		ceph_osdc_put_request(req);
944 	}
945 	mutex_unlock(&osdc->request_mutex);
946 }
947 EXPORT_SYMBOL(ceph_osdc_unregister_linger_request);
948 
949 void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
950 				  struct ceph_osd_request *req)
951 {
952 	if (!req->r_linger) {
953 		dout("set_request_linger %p\n", req);
954 		req->r_linger = 1;
955 		/*
956 		 * caller is now responsible for calling
957 		 * unregister_linger_request
958 		 */
959 		ceph_osdc_get_request(req);
960 	}
961 }
962 EXPORT_SYMBOL(ceph_osdc_set_request_linger);
963 
964 /*
965  * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
966  * (as needed), and set the request r_osd appropriately.  If there is
967  * no up osd, set r_osd to NULL.  Move the request to the appropriate list
968  * (unsent, homeless) or leave on in-flight lru.
969  *
970  * Return 0 if unchanged, 1 if changed, or negative on error.
971  *
972  * Caller should hold map_sem for read and request_mutex.
973  */
974 static int __map_request(struct ceph_osd_client *osdc,
975 			 struct ceph_osd_request *req, int force_resend)
976 {
977 	struct ceph_pg pgid;
978 	int acting[CEPH_PG_MAX_SIZE];
979 	int o = -1, num = 0;
980 	int err;
981 
982 	dout("map_request %p tid %lld\n", req, req->r_tid);
983 	err = ceph_calc_ceph_pg(&pgid, req->r_oid, osdc->osdmap,
984 				ceph_file_layout_pg_pool(req->r_file_layout));
985 	if (err) {
986 		list_move(&req->r_req_lru_item, &osdc->req_notarget);
987 		return err;
988 	}
989 	req->r_pgid = pgid;
990 
991 	err = ceph_calc_pg_acting(osdc->osdmap, pgid, acting);
992 	if (err > 0) {
993 		o = acting[0];
994 		num = err;
995 	}
996 
997 	if ((!force_resend &&
998 	     req->r_osd && req->r_osd->o_osd == o &&
999 	     req->r_sent >= req->r_osd->o_incarnation &&
1000 	     req->r_num_pg_osds == num &&
1001 	     memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
1002 	    (req->r_osd == NULL && o == -1))
1003 		return 0;  /* no change */
1004 
1005 	dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n",
1006 	     req->r_tid, pgid.pool, pgid.seed, o,
1007 	     req->r_osd ? req->r_osd->o_osd : -1);
1008 
1009 	/* record full pg acting set */
1010 	memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
1011 	req->r_num_pg_osds = num;
1012 
1013 	if (req->r_osd) {
1014 		__cancel_request(req);
1015 		list_del_init(&req->r_osd_item);
1016 		req->r_osd = NULL;
1017 	}
1018 
1019 	req->r_osd = __lookup_osd(osdc, o);
1020 	if (!req->r_osd && o >= 0) {
1021 		err = -ENOMEM;
1022 		req->r_osd = create_osd(osdc, o);
1023 		if (!req->r_osd) {
1024 			list_move(&req->r_req_lru_item, &osdc->req_notarget);
1025 			goto out;
1026 		}
1027 
1028 		dout("map_request osd %p is osd%d\n", req->r_osd, o);
1029 		__insert_osd(osdc, req->r_osd);
1030 
1031 		ceph_con_open(&req->r_osd->o_con,
1032 			      CEPH_ENTITY_TYPE_OSD, o,
1033 			      &osdc->osdmap->osd_addr[o]);
1034 	}
1035 
1036 	if (req->r_osd) {
1037 		__remove_osd_from_lru(req->r_osd);
1038 		list_add(&req->r_osd_item, &req->r_osd->o_requests);
1039 		list_move(&req->r_req_lru_item, &osdc->req_unsent);
1040 	} else {
1041 		list_move(&req->r_req_lru_item, &osdc->req_notarget);
1042 	}
1043 	err = 1;   /* osd or pg changed */
1044 
1045 out:
1046 	return err;
1047 }
1048 
1049 /*
1050  * caller should hold map_sem (for read) and request_mutex
1051  */
1052 static void __send_request(struct ceph_osd_client *osdc,
1053 			   struct ceph_osd_request *req)
1054 {
1055 	void *p;
1056 
1057 	dout("send_request %p tid %llu to osd%d flags %d pg %lld.%x\n",
1058 	     req, req->r_tid, req->r_osd->o_osd, req->r_flags,
1059 	     (unsigned long long)req->r_pgid.pool, req->r_pgid.seed);
1060 
1061 	/* fill in message content that changes each time we send it */
1062 	put_unaligned_le32(osdc->osdmap->epoch, req->r_request_osdmap_epoch);
1063 	put_unaligned_le32(req->r_flags, req->r_request_flags);
1064 	put_unaligned_le64(req->r_pgid.pool, req->r_request_pool);
1065 	p = req->r_request_pgid;
1066 	ceph_encode_64(&p, req->r_pgid.pool);
1067 	ceph_encode_32(&p, req->r_pgid.seed);
1068 	put_unaligned_le64(1, req->r_request_attempts);  /* FIXME */
1069 	memcpy(req->r_request_reassert_version, &req->r_reassert_version,
1070 	       sizeof(req->r_reassert_version));
1071 
1072 	req->r_stamp = jiffies;
1073 	list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
1074 
1075 	ceph_msg_get(req->r_request); /* send consumes a ref */
1076 	ceph_con_send(&req->r_osd->o_con, req->r_request);
1077 	req->r_sent = req->r_osd->o_incarnation;
1078 }
1079 
1080 /*
1081  * Send any requests in the queue (req_unsent).
1082  */
1083 static void __send_queued(struct ceph_osd_client *osdc)
1084 {
1085 	struct ceph_osd_request *req, *tmp;
1086 
1087 	dout("__send_queued\n");
1088 	list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item)
1089 		__send_request(osdc, req);
1090 }
1091 
1092 /*
1093  * Timeout callback, called every N seconds when 1 or more osd
1094  * requests has been active for more than N seconds.  When this
1095  * happens, we ping all OSDs with requests who have timed out to
1096  * ensure any communications channel reset is detected.  Reset the
1097  * request timeouts another N seconds in the future as we go.
1098  * Reschedule the timeout event another N seconds in future (unless
1099  * there are no open requests).
1100  */
1101 static void handle_timeout(struct work_struct *work)
1102 {
1103 	struct ceph_osd_client *osdc =
1104 		container_of(work, struct ceph_osd_client, timeout_work.work);
1105 	struct ceph_osd_request *req;
1106 	struct ceph_osd *osd;
1107 	unsigned long keepalive =
1108 		osdc->client->options->osd_keepalive_timeout * HZ;
1109 	struct list_head slow_osds;
1110 	dout("timeout\n");
1111 	down_read(&osdc->map_sem);
1112 
1113 	ceph_monc_request_next_osdmap(&osdc->client->monc);
1114 
1115 	mutex_lock(&osdc->request_mutex);
1116 
1117 	/*
1118 	 * ping osds that are a bit slow.  this ensures that if there
1119 	 * is a break in the TCP connection we will notice, and reopen
1120 	 * a connection with that osd (from the fault callback).
1121 	 */
1122 	INIT_LIST_HEAD(&slow_osds);
1123 	list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
1124 		if (time_before(jiffies, req->r_stamp + keepalive))
1125 			break;
1126 
1127 		osd = req->r_osd;
1128 		BUG_ON(!osd);
1129 		dout(" tid %llu is slow, will send keepalive on osd%d\n",
1130 		     req->r_tid, osd->o_osd);
1131 		list_move_tail(&osd->o_keepalive_item, &slow_osds);
1132 	}
1133 	while (!list_empty(&slow_osds)) {
1134 		osd = list_entry(slow_osds.next, struct ceph_osd,
1135 				 o_keepalive_item);
1136 		list_del_init(&osd->o_keepalive_item);
1137 		ceph_con_keepalive(&osd->o_con);
1138 	}
1139 
1140 	__schedule_osd_timeout(osdc);
1141 	__send_queued(osdc);
1142 	mutex_unlock(&osdc->request_mutex);
1143 	up_read(&osdc->map_sem);
1144 }
1145 
1146 static void handle_osds_timeout(struct work_struct *work)
1147 {
1148 	struct ceph_osd_client *osdc =
1149 		container_of(work, struct ceph_osd_client,
1150 			     osds_timeout_work.work);
1151 	unsigned long delay =
1152 		osdc->client->options->osd_idle_ttl * HZ >> 2;
1153 
1154 	dout("osds timeout\n");
1155 	down_read(&osdc->map_sem);
1156 	remove_old_osds(osdc);
1157 	up_read(&osdc->map_sem);
1158 
1159 	schedule_delayed_work(&osdc->osds_timeout_work,
1160 			      round_jiffies_relative(delay));
1161 }
1162 
1163 static void complete_request(struct ceph_osd_request *req)
1164 {
1165 	if (req->r_safe_callback)
1166 		req->r_safe_callback(req, NULL);
1167 	complete_all(&req->r_safe_completion);  /* fsync waiter */
1168 }
1169 
1170 static int __decode_pgid(void **p, void *end, struct ceph_pg *pgid)
1171 {
1172 	__u8 v;
1173 
1174 	ceph_decode_need(p, end, 1 + 8 + 4 + 4, bad);
1175 	v = ceph_decode_8(p);
1176 	if (v > 1) {
1177 		pr_warning("do not understand pg encoding %d > 1", v);
1178 		return -EINVAL;
1179 	}
1180 	pgid->pool = ceph_decode_64(p);
1181 	pgid->seed = ceph_decode_32(p);
1182 	*p += 4;
1183 	return 0;
1184 
1185 bad:
1186 	pr_warning("incomplete pg encoding");
1187 	return -EINVAL;
1188 }
1189 
1190 /*
1191  * handle osd op reply.  either call the callback if it is specified,
1192  * or do the completion to wake up the waiting thread.
1193  */
1194 static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg,
1195 			 struct ceph_connection *con)
1196 {
1197 	void *p, *end;
1198 	struct ceph_osd_request *req;
1199 	u64 tid;
1200 	int object_len;
1201 	int numops, payload_len, flags;
1202 	s32 result;
1203 	s32 retry_attempt;
1204 	struct ceph_pg pg;
1205 	int err;
1206 	u32 reassert_epoch;
1207 	u64 reassert_version;
1208 	u32 osdmap_epoch;
1209 	int already_completed;
1210 	int i;
1211 
1212 	tid = le64_to_cpu(msg->hdr.tid);
1213 	dout("handle_reply %p tid %llu\n", msg, tid);
1214 
1215 	p = msg->front.iov_base;
1216 	end = p + msg->front.iov_len;
1217 
1218 	ceph_decode_need(&p, end, 4, bad);
1219 	object_len = ceph_decode_32(&p);
1220 	ceph_decode_need(&p, end, object_len, bad);
1221 	p += object_len;
1222 
1223 	err = __decode_pgid(&p, end, &pg);
1224 	if (err)
1225 		goto bad;
1226 
1227 	ceph_decode_need(&p, end, 8 + 4 + 4 + 8 + 4, bad);
1228 	flags = ceph_decode_64(&p);
1229 	result = ceph_decode_32(&p);
1230 	reassert_epoch = ceph_decode_32(&p);
1231 	reassert_version = ceph_decode_64(&p);
1232 	osdmap_epoch = ceph_decode_32(&p);
1233 
1234 	/* lookup */
1235 	mutex_lock(&osdc->request_mutex);
1236 	req = __lookup_request(osdc, tid);
1237 	if (req == NULL) {
1238 		dout("handle_reply tid %llu dne\n", tid);
1239 		mutex_unlock(&osdc->request_mutex);
1240 		return;
1241 	}
1242 	ceph_osdc_get_request(req);
1243 
1244 	dout("handle_reply %p tid %llu req %p result %d\n", msg, tid,
1245 	     req, result);
1246 
1247 	ceph_decode_need(&p, end, 4, bad);
1248 	numops = ceph_decode_32(&p);
1249 	if (numops > CEPH_OSD_MAX_OP)
1250 		goto bad_put;
1251 	if (numops != req->r_num_ops)
1252 		goto bad_put;
1253 	payload_len = 0;
1254 	ceph_decode_need(&p, end, numops * sizeof(struct ceph_osd_op), bad);
1255 	for (i = 0; i < numops; i++) {
1256 		struct ceph_osd_op *op = p;
1257 		int len;
1258 
1259 		len = le32_to_cpu(op->payload_len);
1260 		req->r_reply_op_len[i] = len;
1261 		dout(" op %d has %d bytes\n", i, len);
1262 		payload_len += len;
1263 		p += sizeof(*op);
1264 	}
1265 	if (payload_len != le32_to_cpu(msg->hdr.data_len)) {
1266 		pr_warning("sum of op payload lens %d != data_len %d",
1267 			   payload_len, le32_to_cpu(msg->hdr.data_len));
1268 		goto bad_put;
1269 	}
1270 
1271 	ceph_decode_need(&p, end, 4 + numops * 4, bad);
1272 	retry_attempt = ceph_decode_32(&p);
1273 	for (i = 0; i < numops; i++)
1274 		req->r_reply_op_result[i] = ceph_decode_32(&p);
1275 
1276 	/*
1277 	 * if this connection filled our message, drop our reference now, to
1278 	 * avoid a (safe but slower) revoke later.
1279 	 */
1280 	if (req->r_con_filling_msg == con && req->r_reply == msg) {
1281 		dout(" dropping con_filling_msg ref %p\n", con);
1282 		req->r_con_filling_msg = NULL;
1283 		con->ops->put(con);
1284 	}
1285 
1286 	if (!req->r_got_reply) {
1287 		unsigned int bytes;
1288 
1289 		req->r_result = result;
1290 		bytes = le32_to_cpu(msg->hdr.data_len);
1291 		dout("handle_reply result %d bytes %d\n", req->r_result,
1292 		     bytes);
1293 		if (req->r_result == 0)
1294 			req->r_result = bytes;
1295 
1296 		/* in case this is a write and we need to replay, */
1297 		req->r_reassert_version.epoch = cpu_to_le32(reassert_epoch);
1298 		req->r_reassert_version.version = cpu_to_le64(reassert_version);
1299 
1300 		req->r_got_reply = 1;
1301 	} else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) {
1302 		dout("handle_reply tid %llu dup ack\n", tid);
1303 		mutex_unlock(&osdc->request_mutex);
1304 		goto done;
1305 	}
1306 
1307 	dout("handle_reply tid %llu flags %d\n", tid, flags);
1308 
1309 	if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK))
1310 		__register_linger_request(osdc, req);
1311 
1312 	/* either this is a read, or we got the safe response */
1313 	if (result < 0 ||
1314 	    (flags & CEPH_OSD_FLAG_ONDISK) ||
1315 	    ((flags & CEPH_OSD_FLAG_WRITE) == 0))
1316 		__unregister_request(osdc, req);
1317 
1318 	already_completed = req->r_completed;
1319 	req->r_completed = 1;
1320 	mutex_unlock(&osdc->request_mutex);
1321 	if (already_completed)
1322 		goto done;
1323 
1324 	if (req->r_callback)
1325 		req->r_callback(req, msg);
1326 	else
1327 		complete_all(&req->r_completion);
1328 
1329 	if (flags & CEPH_OSD_FLAG_ONDISK)
1330 		complete_request(req);
1331 
1332 done:
1333 	dout("req=%p req->r_linger=%d\n", req, req->r_linger);
1334 	ceph_osdc_put_request(req);
1335 	return;
1336 
1337 bad_put:
1338 	ceph_osdc_put_request(req);
1339 bad:
1340 	pr_err("corrupt osd_op_reply got %d %d\n",
1341 	       (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len));
1342 	ceph_msg_dump(msg);
1343 }
1344 
1345 static void reset_changed_osds(struct ceph_osd_client *osdc)
1346 {
1347 	struct rb_node *p, *n;
1348 
1349 	for (p = rb_first(&osdc->osds); p; p = n) {
1350 		struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
1351 
1352 		n = rb_next(p);
1353 		if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
1354 		    memcmp(&osd->o_con.peer_addr,
1355 			   ceph_osd_addr(osdc->osdmap,
1356 					 osd->o_osd),
1357 			   sizeof(struct ceph_entity_addr)) != 0)
1358 			__reset_osd(osdc, osd);
1359 	}
1360 }
1361 
1362 /*
1363  * Requeue requests whose mapping to an OSD has changed.  If requests map to
1364  * no osd, request a new map.
1365  *
1366  * Caller should hold map_sem for read.
1367  */
1368 static void kick_requests(struct ceph_osd_client *osdc, int force_resend)
1369 {
1370 	struct ceph_osd_request *req, *nreq;
1371 	struct rb_node *p;
1372 	int needmap = 0;
1373 	int err;
1374 
1375 	dout("kick_requests %s\n", force_resend ? " (force resend)" : "");
1376 	mutex_lock(&osdc->request_mutex);
1377 	for (p = rb_first(&osdc->requests); p; ) {
1378 		req = rb_entry(p, struct ceph_osd_request, r_node);
1379 		p = rb_next(p);
1380 
1381 		/*
1382 		 * For linger requests that have not yet been
1383 		 * registered, move them to the linger list; they'll
1384 		 * be sent to the osd in the loop below.  Unregister
1385 		 * the request before re-registering it as a linger
1386 		 * request to ensure the __map_request() below
1387 		 * will decide it needs to be sent.
1388 		 */
1389 		if (req->r_linger && list_empty(&req->r_linger_item)) {
1390 			dout("%p tid %llu restart on osd%d\n",
1391 			     req, req->r_tid,
1392 			     req->r_osd ? req->r_osd->o_osd : -1);
1393 			__unregister_request(osdc, req);
1394 			__register_linger_request(osdc, req);
1395 			continue;
1396 		}
1397 
1398 		err = __map_request(osdc, req, force_resend);
1399 		if (err < 0)
1400 			continue;  /* error */
1401 		if (req->r_osd == NULL) {
1402 			dout("%p tid %llu maps to no osd\n", req, req->r_tid);
1403 			needmap++;  /* request a newer map */
1404 		} else if (err > 0) {
1405 			if (!req->r_linger) {
1406 				dout("%p tid %llu requeued on osd%d\n", req,
1407 				     req->r_tid,
1408 				     req->r_osd ? req->r_osd->o_osd : -1);
1409 				req->r_flags |= CEPH_OSD_FLAG_RETRY;
1410 			}
1411 		}
1412 	}
1413 
1414 	list_for_each_entry_safe(req, nreq, &osdc->req_linger,
1415 				 r_linger_item) {
1416 		dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
1417 
1418 		err = __map_request(osdc, req, force_resend);
1419 		dout("__map_request returned %d\n", err);
1420 		if (err == 0)
1421 			continue;  /* no change and no osd was specified */
1422 		if (err < 0)
1423 			continue;  /* hrm! */
1424 		if (req->r_osd == NULL) {
1425 			dout("tid %llu maps to no valid osd\n", req->r_tid);
1426 			needmap++;  /* request a newer map */
1427 			continue;
1428 		}
1429 
1430 		dout("kicking lingering %p tid %llu osd%d\n", req, req->r_tid,
1431 		     req->r_osd ? req->r_osd->o_osd : -1);
1432 		__register_request(osdc, req);
1433 		__unregister_linger_request(osdc, req);
1434 	}
1435 	mutex_unlock(&osdc->request_mutex);
1436 
1437 	if (needmap) {
1438 		dout("%d requests for down osds, need new map\n", needmap);
1439 		ceph_monc_request_next_osdmap(&osdc->client->monc);
1440 	}
1441 	reset_changed_osds(osdc);
1442 }
1443 
1444 
1445 /*
1446  * Process updated osd map.
1447  *
1448  * The message contains any number of incremental and full maps, normally
1449  * indicating some sort of topology change in the cluster.  Kick requests
1450  * off to different OSDs as needed.
1451  */
1452 void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1453 {
1454 	void *p, *end, *next;
1455 	u32 nr_maps, maplen;
1456 	u32 epoch;
1457 	struct ceph_osdmap *newmap = NULL, *oldmap;
1458 	int err;
1459 	struct ceph_fsid fsid;
1460 
1461 	dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
1462 	p = msg->front.iov_base;
1463 	end = p + msg->front.iov_len;
1464 
1465 	/* verify fsid */
1466 	ceph_decode_need(&p, end, sizeof(fsid), bad);
1467 	ceph_decode_copy(&p, &fsid, sizeof(fsid));
1468 	if (ceph_check_fsid(osdc->client, &fsid) < 0)
1469 		return;
1470 
1471 	down_write(&osdc->map_sem);
1472 
1473 	/* incremental maps */
1474 	ceph_decode_32_safe(&p, end, nr_maps, bad);
1475 	dout(" %d inc maps\n", nr_maps);
1476 	while (nr_maps > 0) {
1477 		ceph_decode_need(&p, end, 2*sizeof(u32), bad);
1478 		epoch = ceph_decode_32(&p);
1479 		maplen = ceph_decode_32(&p);
1480 		ceph_decode_need(&p, end, maplen, bad);
1481 		next = p + maplen;
1482 		if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
1483 			dout("applying incremental map %u len %d\n",
1484 			     epoch, maplen);
1485 			newmap = osdmap_apply_incremental(&p, next,
1486 							  osdc->osdmap,
1487 							  &osdc->client->msgr);
1488 			if (IS_ERR(newmap)) {
1489 				err = PTR_ERR(newmap);
1490 				goto bad;
1491 			}
1492 			BUG_ON(!newmap);
1493 			if (newmap != osdc->osdmap) {
1494 				ceph_osdmap_destroy(osdc->osdmap);
1495 				osdc->osdmap = newmap;
1496 			}
1497 			kick_requests(osdc, 0);
1498 		} else {
1499 			dout("ignoring incremental map %u len %d\n",
1500 			     epoch, maplen);
1501 		}
1502 		p = next;
1503 		nr_maps--;
1504 	}
1505 	if (newmap)
1506 		goto done;
1507 
1508 	/* full maps */
1509 	ceph_decode_32_safe(&p, end, nr_maps, bad);
1510 	dout(" %d full maps\n", nr_maps);
1511 	while (nr_maps) {
1512 		ceph_decode_need(&p, end, 2*sizeof(u32), bad);
1513 		epoch = ceph_decode_32(&p);
1514 		maplen = ceph_decode_32(&p);
1515 		ceph_decode_need(&p, end, maplen, bad);
1516 		if (nr_maps > 1) {
1517 			dout("skipping non-latest full map %u len %d\n",
1518 			     epoch, maplen);
1519 		} else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
1520 			dout("skipping full map %u len %d, "
1521 			     "older than our %u\n", epoch, maplen,
1522 			     osdc->osdmap->epoch);
1523 		} else {
1524 			int skipped_map = 0;
1525 
1526 			dout("taking full map %u len %d\n", epoch, maplen);
1527 			newmap = osdmap_decode(&p, p+maplen);
1528 			if (IS_ERR(newmap)) {
1529 				err = PTR_ERR(newmap);
1530 				goto bad;
1531 			}
1532 			BUG_ON(!newmap);
1533 			oldmap = osdc->osdmap;
1534 			osdc->osdmap = newmap;
1535 			if (oldmap) {
1536 				if (oldmap->epoch + 1 < newmap->epoch)
1537 					skipped_map = 1;
1538 				ceph_osdmap_destroy(oldmap);
1539 			}
1540 			kick_requests(osdc, skipped_map);
1541 		}
1542 		p += maplen;
1543 		nr_maps--;
1544 	}
1545 
1546 done:
1547 	downgrade_write(&osdc->map_sem);
1548 	ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch);
1549 
1550 	/*
1551 	 * subscribe to subsequent osdmap updates if full to ensure
1552 	 * we find out when we are no longer full and stop returning
1553 	 * ENOSPC.
1554 	 */
1555 	if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL))
1556 		ceph_monc_request_next_osdmap(&osdc->client->monc);
1557 
1558 	mutex_lock(&osdc->request_mutex);
1559 	__send_queued(osdc);
1560 	mutex_unlock(&osdc->request_mutex);
1561 	up_read(&osdc->map_sem);
1562 	wake_up_all(&osdc->client->auth_wq);
1563 	return;
1564 
1565 bad:
1566 	pr_err("osdc handle_map corrupt msg\n");
1567 	ceph_msg_dump(msg);
1568 	up_write(&osdc->map_sem);
1569 	return;
1570 }
1571 
1572 /*
1573  * watch/notify callback event infrastructure
1574  *
1575  * These callbacks are used both for watch and notify operations.
1576  */
1577 static void __release_event(struct kref *kref)
1578 {
1579 	struct ceph_osd_event *event =
1580 		container_of(kref, struct ceph_osd_event, kref);
1581 
1582 	dout("__release_event %p\n", event);
1583 	kfree(event);
1584 }
1585 
1586 static void get_event(struct ceph_osd_event *event)
1587 {
1588 	kref_get(&event->kref);
1589 }
1590 
1591 void ceph_osdc_put_event(struct ceph_osd_event *event)
1592 {
1593 	kref_put(&event->kref, __release_event);
1594 }
1595 EXPORT_SYMBOL(ceph_osdc_put_event);
1596 
1597 static void __insert_event(struct ceph_osd_client *osdc,
1598 			     struct ceph_osd_event *new)
1599 {
1600 	struct rb_node **p = &osdc->event_tree.rb_node;
1601 	struct rb_node *parent = NULL;
1602 	struct ceph_osd_event *event = NULL;
1603 
1604 	while (*p) {
1605 		parent = *p;
1606 		event = rb_entry(parent, struct ceph_osd_event, node);
1607 		if (new->cookie < event->cookie)
1608 			p = &(*p)->rb_left;
1609 		else if (new->cookie > event->cookie)
1610 			p = &(*p)->rb_right;
1611 		else
1612 			BUG();
1613 	}
1614 
1615 	rb_link_node(&new->node, parent, p);
1616 	rb_insert_color(&new->node, &osdc->event_tree);
1617 }
1618 
1619 static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
1620 					        u64 cookie)
1621 {
1622 	struct rb_node **p = &osdc->event_tree.rb_node;
1623 	struct rb_node *parent = NULL;
1624 	struct ceph_osd_event *event = NULL;
1625 
1626 	while (*p) {
1627 		parent = *p;
1628 		event = rb_entry(parent, struct ceph_osd_event, node);
1629 		if (cookie < event->cookie)
1630 			p = &(*p)->rb_left;
1631 		else if (cookie > event->cookie)
1632 			p = &(*p)->rb_right;
1633 		else
1634 			return event;
1635 	}
1636 	return NULL;
1637 }
1638 
1639 static void __remove_event(struct ceph_osd_event *event)
1640 {
1641 	struct ceph_osd_client *osdc = event->osdc;
1642 
1643 	if (!RB_EMPTY_NODE(&event->node)) {
1644 		dout("__remove_event removed %p\n", event);
1645 		rb_erase(&event->node, &osdc->event_tree);
1646 		ceph_osdc_put_event(event);
1647 	} else {
1648 		dout("__remove_event didn't remove %p\n", event);
1649 	}
1650 }
1651 
1652 int ceph_osdc_create_event(struct ceph_osd_client *osdc,
1653 			   void (*event_cb)(u64, u64, u8, void *),
1654 			   void *data, struct ceph_osd_event **pevent)
1655 {
1656 	struct ceph_osd_event *event;
1657 
1658 	event = kmalloc(sizeof(*event), GFP_NOIO);
1659 	if (!event)
1660 		return -ENOMEM;
1661 
1662 	dout("create_event %p\n", event);
1663 	event->cb = event_cb;
1664 	event->one_shot = 0;
1665 	event->data = data;
1666 	event->osdc = osdc;
1667 	INIT_LIST_HEAD(&event->osd_node);
1668 	RB_CLEAR_NODE(&event->node);
1669 	kref_init(&event->kref);   /* one ref for us */
1670 	kref_get(&event->kref);    /* one ref for the caller */
1671 
1672 	spin_lock(&osdc->event_lock);
1673 	event->cookie = ++osdc->event_count;
1674 	__insert_event(osdc, event);
1675 	spin_unlock(&osdc->event_lock);
1676 
1677 	*pevent = event;
1678 	return 0;
1679 }
1680 EXPORT_SYMBOL(ceph_osdc_create_event);
1681 
1682 void ceph_osdc_cancel_event(struct ceph_osd_event *event)
1683 {
1684 	struct ceph_osd_client *osdc = event->osdc;
1685 
1686 	dout("cancel_event %p\n", event);
1687 	spin_lock(&osdc->event_lock);
1688 	__remove_event(event);
1689 	spin_unlock(&osdc->event_lock);
1690 	ceph_osdc_put_event(event); /* caller's */
1691 }
1692 EXPORT_SYMBOL(ceph_osdc_cancel_event);
1693 
1694 
1695 static void do_event_work(struct work_struct *work)
1696 {
1697 	struct ceph_osd_event_work *event_work =
1698 		container_of(work, struct ceph_osd_event_work, work);
1699 	struct ceph_osd_event *event = event_work->event;
1700 	u64 ver = event_work->ver;
1701 	u64 notify_id = event_work->notify_id;
1702 	u8 opcode = event_work->opcode;
1703 
1704 	dout("do_event_work completing %p\n", event);
1705 	event->cb(ver, notify_id, opcode, event->data);
1706 	dout("do_event_work completed %p\n", event);
1707 	ceph_osdc_put_event(event);
1708 	kfree(event_work);
1709 }
1710 
1711 
1712 /*
1713  * Process osd watch notifications
1714  */
1715 static void handle_watch_notify(struct ceph_osd_client *osdc,
1716 				struct ceph_msg *msg)
1717 {
1718 	void *p, *end;
1719 	u8 proto_ver;
1720 	u64 cookie, ver, notify_id;
1721 	u8 opcode;
1722 	struct ceph_osd_event *event;
1723 	struct ceph_osd_event_work *event_work;
1724 
1725 	p = msg->front.iov_base;
1726 	end = p + msg->front.iov_len;
1727 
1728 	ceph_decode_8_safe(&p, end, proto_ver, bad);
1729 	ceph_decode_8_safe(&p, end, opcode, bad);
1730 	ceph_decode_64_safe(&p, end, cookie, bad);
1731 	ceph_decode_64_safe(&p, end, ver, bad);
1732 	ceph_decode_64_safe(&p, end, notify_id, bad);
1733 
1734 	spin_lock(&osdc->event_lock);
1735 	event = __find_event(osdc, cookie);
1736 	if (event) {
1737 		BUG_ON(event->one_shot);
1738 		get_event(event);
1739 	}
1740 	spin_unlock(&osdc->event_lock);
1741 	dout("handle_watch_notify cookie %lld ver %lld event %p\n",
1742 	     cookie, ver, event);
1743 	if (event) {
1744 		event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
1745 		if (!event_work) {
1746 			dout("ERROR: could not allocate event_work\n");
1747 			goto done_err;
1748 		}
1749 		INIT_WORK(&event_work->work, do_event_work);
1750 		event_work->event = event;
1751 		event_work->ver = ver;
1752 		event_work->notify_id = notify_id;
1753 		event_work->opcode = opcode;
1754 		if (!queue_work(osdc->notify_wq, &event_work->work)) {
1755 			dout("WARNING: failed to queue notify event work\n");
1756 			goto done_err;
1757 		}
1758 	}
1759 
1760 	return;
1761 
1762 done_err:
1763 	ceph_osdc_put_event(event);
1764 	return;
1765 
1766 bad:
1767 	pr_err("osdc handle_watch_notify corrupt msg\n");
1768 	return;
1769 }
1770 
1771 static void ceph_osdc_msg_data_set(struct ceph_msg *msg,
1772 				struct ceph_osd_data *osd_data)
1773 {
1774 	if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
1775 		BUG_ON(osd_data->length > (u64) SIZE_MAX);
1776 		if (osd_data->length)
1777 			ceph_msg_data_set_pages(msg, osd_data->pages,
1778 				osd_data->length, osd_data->alignment);
1779 	} else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
1780 		BUG_ON(!osd_data->pagelist->length);
1781 		ceph_msg_data_set_pagelist(msg, osd_data->pagelist);
1782 #ifdef CONFIG_BLOCK
1783 	} else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
1784 		ceph_msg_data_set_bio(msg, osd_data->bio);
1785 #endif
1786 	} else {
1787 		BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
1788 	}
1789 }
1790 
1791 /*
1792  * Register request, send initial attempt.
1793  */
1794 int ceph_osdc_start_request(struct ceph_osd_client *osdc,
1795 			    struct ceph_osd_request *req,
1796 			    bool nofail)
1797 {
1798 	int rc = 0;
1799 
1800 	/* Set up response incoming data and request outgoing data fields */
1801 
1802 	ceph_osdc_msg_data_set(req->r_reply, &req->r_data_in);
1803 	ceph_osdc_msg_data_set(req->r_request, &req->r_data_out);
1804 
1805 	down_read(&osdc->map_sem);
1806 	mutex_lock(&osdc->request_mutex);
1807 	__register_request(osdc, req);
1808 	WARN_ON(req->r_sent);
1809 	rc = __map_request(osdc, req, 0);
1810 	if (rc < 0) {
1811 		if (nofail) {
1812 			dout("osdc_start_request failed map, "
1813 				" will retry %lld\n", req->r_tid);
1814 			rc = 0;
1815 		}
1816 		goto out_unlock;
1817 	}
1818 	if (req->r_osd == NULL) {
1819 		dout("send_request %p no up osds in pg\n", req);
1820 		ceph_monc_request_next_osdmap(&osdc->client->monc);
1821 	} else {
1822 		__send_request(osdc, req);
1823 	}
1824 	rc = 0;
1825 out_unlock:
1826 	mutex_unlock(&osdc->request_mutex);
1827 	up_read(&osdc->map_sem);
1828 	return rc;
1829 }
1830 EXPORT_SYMBOL(ceph_osdc_start_request);
1831 
1832 /*
1833  * wait for a request to complete
1834  */
1835 int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
1836 			   struct ceph_osd_request *req)
1837 {
1838 	int rc;
1839 
1840 	rc = wait_for_completion_interruptible(&req->r_completion);
1841 	if (rc < 0) {
1842 		mutex_lock(&osdc->request_mutex);
1843 		__cancel_request(req);
1844 		__unregister_request(osdc, req);
1845 		mutex_unlock(&osdc->request_mutex);
1846 		complete_request(req);
1847 		dout("wait_request tid %llu canceled/timed out\n", req->r_tid);
1848 		return rc;
1849 	}
1850 
1851 	dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result);
1852 	return req->r_result;
1853 }
1854 EXPORT_SYMBOL(ceph_osdc_wait_request);
1855 
1856 /*
1857  * sync - wait for all in-flight requests to flush.  avoid starvation.
1858  */
1859 void ceph_osdc_sync(struct ceph_osd_client *osdc)
1860 {
1861 	struct ceph_osd_request *req;
1862 	u64 last_tid, next_tid = 0;
1863 
1864 	mutex_lock(&osdc->request_mutex);
1865 	last_tid = osdc->last_tid;
1866 	while (1) {
1867 		req = __lookup_request_ge(osdc, next_tid);
1868 		if (!req)
1869 			break;
1870 		if (req->r_tid > last_tid)
1871 			break;
1872 
1873 		next_tid = req->r_tid + 1;
1874 		if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
1875 			continue;
1876 
1877 		ceph_osdc_get_request(req);
1878 		mutex_unlock(&osdc->request_mutex);
1879 		dout("sync waiting on tid %llu (last is %llu)\n",
1880 		     req->r_tid, last_tid);
1881 		wait_for_completion(&req->r_safe_completion);
1882 		mutex_lock(&osdc->request_mutex);
1883 		ceph_osdc_put_request(req);
1884 	}
1885 	mutex_unlock(&osdc->request_mutex);
1886 	dout("sync done (thru tid %llu)\n", last_tid);
1887 }
1888 EXPORT_SYMBOL(ceph_osdc_sync);
1889 
1890 /*
1891  * init, shutdown
1892  */
1893 int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
1894 {
1895 	int err;
1896 
1897 	dout("init\n");
1898 	osdc->client = client;
1899 	osdc->osdmap = NULL;
1900 	init_rwsem(&osdc->map_sem);
1901 	init_completion(&osdc->map_waiters);
1902 	osdc->last_requested_map = 0;
1903 	mutex_init(&osdc->request_mutex);
1904 	osdc->last_tid = 0;
1905 	osdc->osds = RB_ROOT;
1906 	INIT_LIST_HEAD(&osdc->osd_lru);
1907 	osdc->requests = RB_ROOT;
1908 	INIT_LIST_HEAD(&osdc->req_lru);
1909 	INIT_LIST_HEAD(&osdc->req_unsent);
1910 	INIT_LIST_HEAD(&osdc->req_notarget);
1911 	INIT_LIST_HEAD(&osdc->req_linger);
1912 	osdc->num_requests = 0;
1913 	INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
1914 	INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
1915 	spin_lock_init(&osdc->event_lock);
1916 	osdc->event_tree = RB_ROOT;
1917 	osdc->event_count = 0;
1918 
1919 	schedule_delayed_work(&osdc->osds_timeout_work,
1920 	   round_jiffies_relative(osdc->client->options->osd_idle_ttl * HZ));
1921 
1922 	err = -ENOMEM;
1923 	osdc->req_mempool = mempool_create_kmalloc_pool(10,
1924 					sizeof(struct ceph_osd_request));
1925 	if (!osdc->req_mempool)
1926 		goto out;
1927 
1928 	err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
1929 				OSD_OP_FRONT_LEN, 10, true,
1930 				"osd_op");
1931 	if (err < 0)
1932 		goto out_mempool;
1933 	err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
1934 				OSD_OPREPLY_FRONT_LEN, 10, true,
1935 				"osd_op_reply");
1936 	if (err < 0)
1937 		goto out_msgpool;
1938 
1939 	osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
1940 	if (IS_ERR(osdc->notify_wq)) {
1941 		err = PTR_ERR(osdc->notify_wq);
1942 		osdc->notify_wq = NULL;
1943 		goto out_msgpool;
1944 	}
1945 	return 0;
1946 
1947 out_msgpool:
1948 	ceph_msgpool_destroy(&osdc->msgpool_op);
1949 out_mempool:
1950 	mempool_destroy(osdc->req_mempool);
1951 out:
1952 	return err;
1953 }
1954 
1955 void ceph_osdc_stop(struct ceph_osd_client *osdc)
1956 {
1957 	flush_workqueue(osdc->notify_wq);
1958 	destroy_workqueue(osdc->notify_wq);
1959 	cancel_delayed_work_sync(&osdc->timeout_work);
1960 	cancel_delayed_work_sync(&osdc->osds_timeout_work);
1961 	if (osdc->osdmap) {
1962 		ceph_osdmap_destroy(osdc->osdmap);
1963 		osdc->osdmap = NULL;
1964 	}
1965 	remove_all_osds(osdc);
1966 	mempool_destroy(osdc->req_mempool);
1967 	ceph_msgpool_destroy(&osdc->msgpool_op);
1968 	ceph_msgpool_destroy(&osdc->msgpool_op_reply);
1969 }
1970 
1971 /*
1972  * Read some contiguous pages.  If we cross a stripe boundary, shorten
1973  * *plen.  Return number of bytes read, or error.
1974  */
1975 int ceph_osdc_readpages(struct ceph_osd_client *osdc,
1976 			struct ceph_vino vino, struct ceph_file_layout *layout,
1977 			u64 off, u64 *plen,
1978 			u32 truncate_seq, u64 truncate_size,
1979 			struct page **pages, int num_pages, int page_align)
1980 {
1981 	struct ceph_osd_request *req;
1982 	struct ceph_osd_data *osd_data;
1983 	int rc = 0;
1984 
1985 	dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
1986 	     vino.snap, off, *plen);
1987 	req = ceph_osdc_new_request(osdc, layout, vino, off, plen,
1988 				    CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
1989 				    NULL, 0, truncate_seq, truncate_size, NULL,
1990 				    false);
1991 	if (IS_ERR(req))
1992 		return PTR_ERR(req);
1993 
1994 	/* it may be a short read due to an object boundary */
1995 
1996 	osd_data = &req->r_data_in;
1997 	osd_data->type = CEPH_OSD_DATA_TYPE_PAGES;
1998 	osd_data->pages = pages;
1999 	osd_data->length = *plen;
2000 	osd_data->alignment = page_align;
2001 
2002 	dout("readpages  final extent is %llu~%llu (%llu bytes align %d)\n",
2003 	     off, *plen, osd_data->length, page_align);
2004 
2005 	rc = ceph_osdc_start_request(osdc, req, false);
2006 	if (!rc)
2007 		rc = ceph_osdc_wait_request(osdc, req);
2008 
2009 	ceph_osdc_put_request(req);
2010 	dout("readpages result %d\n", rc);
2011 	return rc;
2012 }
2013 EXPORT_SYMBOL(ceph_osdc_readpages);
2014 
2015 /*
2016  * do a synchronous write on N pages
2017  */
2018 int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
2019 			 struct ceph_file_layout *layout,
2020 			 struct ceph_snap_context *snapc,
2021 			 u64 off, u64 len,
2022 			 u32 truncate_seq, u64 truncate_size,
2023 			 struct timespec *mtime,
2024 			 struct page **pages, int num_pages)
2025 {
2026 	struct ceph_osd_request *req;
2027 	struct ceph_osd_data *osd_data;
2028 	int rc = 0;
2029 	int page_align = off & ~PAGE_MASK;
2030 
2031 	BUG_ON(vino.snap != CEPH_NOSNAP);
2032 	req = ceph_osdc_new_request(osdc, layout, vino, off, &len,
2033 				    CEPH_OSD_OP_WRITE,
2034 				    CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
2035 				    snapc, 0,
2036 				    truncate_seq, truncate_size, mtime,
2037 				    true);
2038 	if (IS_ERR(req))
2039 		return PTR_ERR(req);
2040 
2041 	/* it may be a short write due to an object boundary */
2042 	osd_data = &req->r_data_out;
2043 	osd_data->type = CEPH_OSD_DATA_TYPE_PAGES;
2044 	osd_data->pages = pages;
2045 	osd_data->length = len;
2046 	osd_data->alignment = page_align;
2047 	dout("writepages %llu~%llu (%llu bytes)\n", off, len, osd_data->length);
2048 
2049 	rc = ceph_osdc_start_request(osdc, req, true);
2050 	if (!rc)
2051 		rc = ceph_osdc_wait_request(osdc, req);
2052 
2053 	ceph_osdc_put_request(req);
2054 	if (rc == 0)
2055 		rc = len;
2056 	dout("writepages result %d\n", rc);
2057 	return rc;
2058 }
2059 EXPORT_SYMBOL(ceph_osdc_writepages);
2060 
2061 /*
2062  * handle incoming message
2063  */
2064 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
2065 {
2066 	struct ceph_osd *osd = con->private;
2067 	struct ceph_osd_client *osdc;
2068 	int type = le16_to_cpu(msg->hdr.type);
2069 
2070 	if (!osd)
2071 		goto out;
2072 	osdc = osd->o_osdc;
2073 
2074 	switch (type) {
2075 	case CEPH_MSG_OSD_MAP:
2076 		ceph_osdc_handle_map(osdc, msg);
2077 		break;
2078 	case CEPH_MSG_OSD_OPREPLY:
2079 		handle_reply(osdc, msg, con);
2080 		break;
2081 	case CEPH_MSG_WATCH_NOTIFY:
2082 		handle_watch_notify(osdc, msg);
2083 		break;
2084 
2085 	default:
2086 		pr_err("received unknown message type %d %s\n", type,
2087 		       ceph_msg_type_name(type));
2088 	}
2089 out:
2090 	ceph_msg_put(msg);
2091 }
2092 
2093 /*
2094  * lookup and return message for incoming reply.  set up reply message
2095  * pages.
2096  */
2097 static struct ceph_msg *get_reply(struct ceph_connection *con,
2098 				  struct ceph_msg_header *hdr,
2099 				  int *skip)
2100 {
2101 	struct ceph_osd *osd = con->private;
2102 	struct ceph_osd_client *osdc = osd->o_osdc;
2103 	struct ceph_msg *m;
2104 	struct ceph_osd_request *req;
2105 	int front = le32_to_cpu(hdr->front_len);
2106 	int data_len = le32_to_cpu(hdr->data_len);
2107 	u64 tid;
2108 
2109 	tid = le64_to_cpu(hdr->tid);
2110 	mutex_lock(&osdc->request_mutex);
2111 	req = __lookup_request(osdc, tid);
2112 	if (!req) {
2113 		*skip = 1;
2114 		m = NULL;
2115 		dout("get_reply unknown tid %llu from osd%d\n", tid,
2116 		     osd->o_osd);
2117 		goto out;
2118 	}
2119 
2120 	if (req->r_con_filling_msg) {
2121 		dout("%s revoking msg %p from old con %p\n", __func__,
2122 		     req->r_reply, req->r_con_filling_msg);
2123 		ceph_msg_revoke_incoming(req->r_reply);
2124 		req->r_con_filling_msg->ops->put(req->r_con_filling_msg);
2125 		req->r_con_filling_msg = NULL;
2126 	}
2127 
2128 	if (front > req->r_reply->front.iov_len) {
2129 		pr_warning("get_reply front %d > preallocated %d\n",
2130 			   front, (int)req->r_reply->front.iov_len);
2131 		m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front, GFP_NOFS, false);
2132 		if (!m)
2133 			goto out;
2134 		ceph_msg_put(req->r_reply);
2135 		req->r_reply = m;
2136 	}
2137 	m = ceph_msg_get(req->r_reply);
2138 
2139 	if (data_len > 0) {
2140 		struct ceph_osd_data *osd_data = &req->r_data_in;
2141 
2142 		if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
2143 			if (osd_data->pages &&
2144 				unlikely(osd_data->length < data_len)) {
2145 
2146 				pr_warning("tid %lld reply has %d bytes "
2147 					"we had only %llu bytes ready\n",
2148 					tid, data_len, osd_data->length);
2149 				*skip = 1;
2150 				ceph_msg_put(m);
2151 				m = NULL;
2152 				goto out;
2153 			}
2154 		}
2155 	}
2156 	*skip = 0;
2157 	req->r_con_filling_msg = con->ops->get(con);
2158 	dout("get_reply tid %lld %p\n", tid, m);
2159 
2160 out:
2161 	mutex_unlock(&osdc->request_mutex);
2162 	return m;
2163 
2164 }
2165 
2166 static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2167 				  struct ceph_msg_header *hdr,
2168 				  int *skip)
2169 {
2170 	struct ceph_osd *osd = con->private;
2171 	int type = le16_to_cpu(hdr->type);
2172 	int front = le32_to_cpu(hdr->front_len);
2173 
2174 	*skip = 0;
2175 	switch (type) {
2176 	case CEPH_MSG_OSD_MAP:
2177 	case CEPH_MSG_WATCH_NOTIFY:
2178 		return ceph_msg_new(type, front, GFP_NOFS, false);
2179 	case CEPH_MSG_OSD_OPREPLY:
2180 		return get_reply(con, hdr, skip);
2181 	default:
2182 		pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
2183 			osd->o_osd);
2184 		*skip = 1;
2185 		return NULL;
2186 	}
2187 }
2188 
2189 /*
2190  * Wrappers to refcount containing ceph_osd struct
2191  */
2192 static struct ceph_connection *get_osd_con(struct ceph_connection *con)
2193 {
2194 	struct ceph_osd *osd = con->private;
2195 	if (get_osd(osd))
2196 		return con;
2197 	return NULL;
2198 }
2199 
2200 static void put_osd_con(struct ceph_connection *con)
2201 {
2202 	struct ceph_osd *osd = con->private;
2203 	put_osd(osd);
2204 }
2205 
2206 /*
2207  * authentication
2208  */
2209 /*
2210  * Note: returned pointer is the address of a structure that's
2211  * managed separately.  Caller must *not* attempt to free it.
2212  */
2213 static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
2214 					int *proto, int force_new)
2215 {
2216 	struct ceph_osd *o = con->private;
2217 	struct ceph_osd_client *osdc = o->o_osdc;
2218 	struct ceph_auth_client *ac = osdc->client->monc.auth;
2219 	struct ceph_auth_handshake *auth = &o->o_auth;
2220 
2221 	if (force_new && auth->authorizer) {
2222 		ceph_auth_destroy_authorizer(ac, auth->authorizer);
2223 		auth->authorizer = NULL;
2224 	}
2225 	if (!auth->authorizer) {
2226 		int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2227 						      auth);
2228 		if (ret)
2229 			return ERR_PTR(ret);
2230 	} else {
2231 		int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2232 						     auth);
2233 		if (ret)
2234 			return ERR_PTR(ret);
2235 	}
2236 	*proto = ac->protocol;
2237 
2238 	return auth;
2239 }
2240 
2241 
2242 static int verify_authorizer_reply(struct ceph_connection *con, int len)
2243 {
2244 	struct ceph_osd *o = con->private;
2245 	struct ceph_osd_client *osdc = o->o_osdc;
2246 	struct ceph_auth_client *ac = osdc->client->monc.auth;
2247 
2248 	return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer, len);
2249 }
2250 
2251 static int invalidate_authorizer(struct ceph_connection *con)
2252 {
2253 	struct ceph_osd *o = con->private;
2254 	struct ceph_osd_client *osdc = o->o_osdc;
2255 	struct ceph_auth_client *ac = osdc->client->monc.auth;
2256 
2257 	ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
2258 	return ceph_monc_validate_auth(&osdc->client->monc);
2259 }
2260 
2261 static const struct ceph_connection_operations osd_con_ops = {
2262 	.get = get_osd_con,
2263 	.put = put_osd_con,
2264 	.dispatch = dispatch,
2265 	.get_authorizer = get_authorizer,
2266 	.verify_authorizer_reply = verify_authorizer_reply,
2267 	.invalidate_authorizer = invalidate_authorizer,
2268 	.alloc_msg = alloc_msg,
2269 	.fault = osd_reset,
2270 };
2271