]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - net/ceph/osd_client.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
libceph: msg signing callouts don't need con argument
[mirror_ubuntu-bionic-kernel.git] / net / ceph / osd_client.c
1
2 #include <linux/ceph/ceph_debug.h>
3
4 #include <linux/module.h>
5 #include <linux/err.h>
6 #include <linux/highmem.h>
7 #include <linux/mm.h>
8 #include <linux/pagemap.h>
9 #include <linux/slab.h>
10 #include <linux/uaccess.h>
11 #ifdef CONFIG_BLOCK
12 #include <linux/bio.h>
13 #endif
14
15 #include <linux/ceph/libceph.h>
16 #include <linux/ceph/osd_client.h>
17 #include <linux/ceph/messenger.h>
18 #include <linux/ceph/decode.h>
19 #include <linux/ceph/auth.h>
20 #include <linux/ceph/pagelist.h>
21
22 #define OSD_OP_FRONT_LEN 4096
23 #define OSD_OPREPLY_FRONT_LEN 512
24
25 static struct kmem_cache *ceph_osd_request_cache;
26
27 static const struct ceph_connection_operations osd_con_ops;
28
29 static void __send_queued(struct ceph_osd_client *osdc);
30 static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd);
31 static void __register_request(struct ceph_osd_client *osdc,
32 struct ceph_osd_request *req);
33 static void __unregister_request(struct ceph_osd_client *osdc,
34 struct ceph_osd_request *req);
35 static void __unregister_linger_request(struct ceph_osd_client *osdc,
36 struct ceph_osd_request *req);
37 static void __enqueue_request(struct ceph_osd_request *req);
38 static void __send_request(struct ceph_osd_client *osdc,
39 struct ceph_osd_request *req);
40
41 /*
42 * Implement client access to distributed object storage cluster.
43 *
44 * All data objects are stored within a cluster/cloud of OSDs, or
45 * "object storage devices." (Note that Ceph OSDs have _nothing_ to
46 * do with the T10 OSD extensions to SCSI.) Ceph OSDs are simply
47 * remote daemons serving up and coordinating consistent and safe
48 * access to storage.
49 *
50 * Cluster membership and the mapping of data objects onto storage devices
51 * are described by the osd map.
52 *
53 * We keep track of pending OSD requests (read, write), resubmit
54 * requests to different OSDs when the cluster topology/data layout
55 * change, or retry the affected requests when the communications
56 * channel with an OSD is reset.
57 */
58
59 /*
60 * calculate the mapping of a file extent onto an object, and fill out the
61 * request accordingly. shorten extent as necessary if it crosses an
62 * object boundary.
63 *
64 * fill osd op in request message.
65 */
66 static int calc_layout(struct ceph_file_layout *layout, u64 off, u64 *plen,
67 u64 *objnum, u64 *objoff, u64 *objlen)
68 {
69 u64 orig_len = *plen;
70 int r;
71
72 /* object extent? */
73 r = ceph_calc_file_object_mapping(layout, off, orig_len, objnum,
74 objoff, objlen);
75 if (r < 0)
76 return r;
77 if (*objlen < orig_len) {
78 *plen = *objlen;
79 dout(" skipping last %llu, final file extent %llu~%llu\n",
80 orig_len - *plen, off, *plen);
81 }
82
83 dout("calc_layout objnum=%llx %llu~%llu\n", *objnum, *objoff, *objlen);
84
85 return 0;
86 }
87
88 static void ceph_osd_data_init(struct ceph_osd_data *osd_data)
89 {
90 memset(osd_data, 0, sizeof (*osd_data));
91 osd_data->type = CEPH_OSD_DATA_TYPE_NONE;
92 }
93
94 static void ceph_osd_data_pages_init(struct ceph_osd_data *osd_data,
95 struct page **pages, u64 length, u32 alignment,
96 bool pages_from_pool, bool own_pages)
97 {
98 osd_data->type = CEPH_OSD_DATA_TYPE_PAGES;
99 osd_data->pages = pages;
100 osd_data->length = length;
101 osd_data->alignment = alignment;
102 osd_data->pages_from_pool = pages_from_pool;
103 osd_data->own_pages = own_pages;
104 }
105
106 static void ceph_osd_data_pagelist_init(struct ceph_osd_data *osd_data,
107 struct ceph_pagelist *pagelist)
108 {
109 osd_data->type = CEPH_OSD_DATA_TYPE_PAGELIST;
110 osd_data->pagelist = pagelist;
111 }
112
113 #ifdef CONFIG_BLOCK
114 static void ceph_osd_data_bio_init(struct ceph_osd_data *osd_data,
115 struct bio *bio, size_t bio_length)
116 {
117 osd_data->type = CEPH_OSD_DATA_TYPE_BIO;
118 osd_data->bio = bio;
119 osd_data->bio_length = bio_length;
120 }
121 #endif /* CONFIG_BLOCK */
122
123 #define osd_req_op_data(oreq, whch, typ, fld) \
124 ({ \
125 struct ceph_osd_request *__oreq = (oreq); \
126 unsigned int __whch = (whch); \
127 BUG_ON(__whch >= __oreq->r_num_ops); \
128 &__oreq->r_ops[__whch].typ.fld; \
129 })
130
131 static struct ceph_osd_data *
132 osd_req_op_raw_data_in(struct ceph_osd_request *osd_req, unsigned int which)
133 {
134 BUG_ON(which >= osd_req->r_num_ops);
135
136 return &osd_req->r_ops[which].raw_data_in;
137 }
138
139 struct ceph_osd_data *
140 osd_req_op_extent_osd_data(struct ceph_osd_request *osd_req,
141 unsigned int which)
142 {
143 return osd_req_op_data(osd_req, which, extent, osd_data);
144 }
145 EXPORT_SYMBOL(osd_req_op_extent_osd_data);
146
147 struct ceph_osd_data *
148 osd_req_op_cls_response_data(struct ceph_osd_request *osd_req,
149 unsigned int which)
150 {
151 return osd_req_op_data(osd_req, which, cls, response_data);
152 }
153 EXPORT_SYMBOL(osd_req_op_cls_response_data); /* ??? */
154
155 void osd_req_op_raw_data_in_pages(struct ceph_osd_request *osd_req,
156 unsigned int which, struct page **pages,
157 u64 length, u32 alignment,
158 bool pages_from_pool, bool own_pages)
159 {
160 struct ceph_osd_data *osd_data;
161
162 osd_data = osd_req_op_raw_data_in(osd_req, which);
163 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
164 pages_from_pool, own_pages);
165 }
166 EXPORT_SYMBOL(osd_req_op_raw_data_in_pages);
167
168 void osd_req_op_extent_osd_data_pages(struct ceph_osd_request *osd_req,
169 unsigned int which, struct page **pages,
170 u64 length, u32 alignment,
171 bool pages_from_pool, bool own_pages)
172 {
173 struct ceph_osd_data *osd_data;
174
175 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
176 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
177 pages_from_pool, own_pages);
178 }
179 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pages);
180
181 void osd_req_op_extent_osd_data_pagelist(struct ceph_osd_request *osd_req,
182 unsigned int which, struct ceph_pagelist *pagelist)
183 {
184 struct ceph_osd_data *osd_data;
185
186 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
187 ceph_osd_data_pagelist_init(osd_data, pagelist);
188 }
189 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pagelist);
190
191 #ifdef CONFIG_BLOCK
192 void osd_req_op_extent_osd_data_bio(struct ceph_osd_request *osd_req,
193 unsigned int which, struct bio *bio, size_t bio_length)
194 {
195 struct ceph_osd_data *osd_data;
196
197 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
198 ceph_osd_data_bio_init(osd_data, bio, bio_length);
199 }
200 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bio);
201 #endif /* CONFIG_BLOCK */
202
203 static void osd_req_op_cls_request_info_pagelist(
204 struct ceph_osd_request *osd_req,
205 unsigned int which, struct ceph_pagelist *pagelist)
206 {
207 struct ceph_osd_data *osd_data;
208
209 osd_data = osd_req_op_data(osd_req, which, cls, request_info);
210 ceph_osd_data_pagelist_init(osd_data, pagelist);
211 }
212
213 void osd_req_op_cls_request_data_pagelist(
214 struct ceph_osd_request *osd_req,
215 unsigned int which, struct ceph_pagelist *pagelist)
216 {
217 struct ceph_osd_data *osd_data;
218
219 osd_data = osd_req_op_data(osd_req, which, cls, request_data);
220 ceph_osd_data_pagelist_init(osd_data, pagelist);
221 }
222 EXPORT_SYMBOL(osd_req_op_cls_request_data_pagelist);
223
224 void osd_req_op_cls_request_data_pages(struct ceph_osd_request *osd_req,
225 unsigned int which, struct page **pages, u64 length,
226 u32 alignment, bool pages_from_pool, bool own_pages)
227 {
228 struct ceph_osd_data *osd_data;
229
230 osd_data = osd_req_op_data(osd_req, which, cls, request_data);
231 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
232 pages_from_pool, own_pages);
233 }
234 EXPORT_SYMBOL(osd_req_op_cls_request_data_pages);
235
236 void osd_req_op_cls_response_data_pages(struct ceph_osd_request *osd_req,
237 unsigned int which, struct page **pages, u64 length,
238 u32 alignment, bool pages_from_pool, bool own_pages)
239 {
240 struct ceph_osd_data *osd_data;
241
242 osd_data = osd_req_op_data(osd_req, which, cls, response_data);
243 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
244 pages_from_pool, own_pages);
245 }
246 EXPORT_SYMBOL(osd_req_op_cls_response_data_pages);
247
248 static u64 ceph_osd_data_length(struct ceph_osd_data *osd_data)
249 {
250 switch (osd_data->type) {
251 case CEPH_OSD_DATA_TYPE_NONE:
252 return 0;
253 case CEPH_OSD_DATA_TYPE_PAGES:
254 return osd_data->length;
255 case CEPH_OSD_DATA_TYPE_PAGELIST:
256 return (u64)osd_data->pagelist->length;
257 #ifdef CONFIG_BLOCK
258 case CEPH_OSD_DATA_TYPE_BIO:
259 return (u64)osd_data->bio_length;
260 #endif /* CONFIG_BLOCK */
261 default:
262 WARN(true, "unrecognized data type %d\n", (int)osd_data->type);
263 return 0;
264 }
265 }
266
267 static void ceph_osd_data_release(struct ceph_osd_data *osd_data)
268 {
269 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES && osd_data->own_pages) {
270 int num_pages;
271
272 num_pages = calc_pages_for((u64)osd_data->alignment,
273 (u64)osd_data->length);
274 ceph_release_page_vector(osd_data->pages, num_pages);
275 }
276 ceph_osd_data_init(osd_data);
277 }
278
279 static void osd_req_op_data_release(struct ceph_osd_request *osd_req,
280 unsigned int which)
281 {
282 struct ceph_osd_req_op *op;
283
284 BUG_ON(which >= osd_req->r_num_ops);
285 op = &osd_req->r_ops[which];
286
287 switch (op->op) {
288 case CEPH_OSD_OP_READ:
289 case CEPH_OSD_OP_WRITE:
290 case CEPH_OSD_OP_WRITEFULL:
291 ceph_osd_data_release(&op->extent.osd_data);
292 break;
293 case CEPH_OSD_OP_CALL:
294 ceph_osd_data_release(&op->cls.request_info);
295 ceph_osd_data_release(&op->cls.request_data);
296 ceph_osd_data_release(&op->cls.response_data);
297 break;
298 case CEPH_OSD_OP_SETXATTR:
299 case CEPH_OSD_OP_CMPXATTR:
300 ceph_osd_data_release(&op->xattr.osd_data);
301 break;
302 case CEPH_OSD_OP_STAT:
303 ceph_osd_data_release(&op->raw_data_in);
304 break;
305 default:
306 break;
307 }
308 }
309
310 /*
311 * requests
312 */
313 static void ceph_osdc_release_request(struct kref *kref)
314 {
315 struct ceph_osd_request *req = container_of(kref,
316 struct ceph_osd_request, r_kref);
317 unsigned int which;
318
319 dout("%s %p (r_request %p r_reply %p)\n", __func__, req,
320 req->r_request, req->r_reply);
321 WARN_ON(!RB_EMPTY_NODE(&req->r_node));
322 WARN_ON(!list_empty(&req->r_req_lru_item));
323 WARN_ON(!list_empty(&req->r_osd_item));
324 WARN_ON(!list_empty(&req->r_linger_item));
325 WARN_ON(!list_empty(&req->r_linger_osd_item));
326 WARN_ON(req->r_osd);
327
328 if (req->r_request)
329 ceph_msg_put(req->r_request);
330 if (req->r_reply) {
331 ceph_msg_revoke_incoming(req->r_reply);
332 ceph_msg_put(req->r_reply);
333 }
334
335 for (which = 0; which < req->r_num_ops; which++)
336 osd_req_op_data_release(req, which);
337
338 ceph_put_snap_context(req->r_snapc);
339 if (req->r_mempool)
340 mempool_free(req, req->r_osdc->req_mempool);
341 else
342 kmem_cache_free(ceph_osd_request_cache, req);
343
344 }
345
346 void ceph_osdc_get_request(struct ceph_osd_request *req)
347 {
348 dout("%s %p (was %d)\n", __func__, req,
349 atomic_read(&req->r_kref.refcount));
350 kref_get(&req->r_kref);
351 }
352 EXPORT_SYMBOL(ceph_osdc_get_request);
353
354 void ceph_osdc_put_request(struct ceph_osd_request *req)
355 {
356 dout("%s %p (was %d)\n", __func__, req,
357 atomic_read(&req->r_kref.refcount));
358 kref_put(&req->r_kref, ceph_osdc_release_request);
359 }
360 EXPORT_SYMBOL(ceph_osdc_put_request);
361
362 struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
363 struct ceph_snap_context *snapc,
364 unsigned int num_ops,
365 bool use_mempool,
366 gfp_t gfp_flags)
367 {
368 struct ceph_osd_request *req;
369 struct ceph_msg *msg;
370 size_t msg_size;
371
372 BUILD_BUG_ON(CEPH_OSD_MAX_OP > U16_MAX);
373 BUG_ON(num_ops > CEPH_OSD_MAX_OP);
374
375 msg_size = 4 + 4 + 8 + 8 + 4+8;
376 msg_size += 2 + 4 + 8 + 4 + 4; /* oloc */
377 msg_size += 1 + 8 + 4 + 4; /* pg_t */
378 msg_size += 4 + CEPH_MAX_OID_NAME_LEN; /* oid */
379 msg_size += 2 + num_ops*sizeof(struct ceph_osd_op);
380 msg_size += 8; /* snapid */
381 msg_size += 8; /* snap_seq */
382 msg_size += 8 * (snapc ? snapc->num_snaps : 0); /* snaps */
383 msg_size += 4;
384
385 if (use_mempool) {
386 req = mempool_alloc(osdc->req_mempool, gfp_flags);
387 memset(req, 0, sizeof(*req));
388 } else {
389 req = kmem_cache_zalloc(ceph_osd_request_cache, gfp_flags);
390 }
391 if (req == NULL)
392 return NULL;
393
394 req->r_osdc = osdc;
395 req->r_mempool = use_mempool;
396 req->r_num_ops = num_ops;
397
398 kref_init(&req->r_kref);
399 init_completion(&req->r_completion);
400 init_completion(&req->r_safe_completion);
401 RB_CLEAR_NODE(&req->r_node);
402 INIT_LIST_HEAD(&req->r_unsafe_item);
403 INIT_LIST_HEAD(&req->r_linger_item);
404 INIT_LIST_HEAD(&req->r_linger_osd_item);
405 INIT_LIST_HEAD(&req->r_req_lru_item);
406 INIT_LIST_HEAD(&req->r_osd_item);
407
408 req->r_base_oloc.pool = -1;
409 req->r_target_oloc.pool = -1;
410
411 /* create reply message */
412 if (use_mempool)
413 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0);
414 else
415 msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY,
416 OSD_OPREPLY_FRONT_LEN, gfp_flags, true);
417 if (!msg) {
418 ceph_osdc_put_request(req);
419 return NULL;
420 }
421 req->r_reply = msg;
422
423 /* create request message; allow space for oid */
424 if (use_mempool)
425 msg = ceph_msgpool_get(&osdc->msgpool_op, 0);
426 else
427 msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp_flags, true);
428 if (!msg) {
429 ceph_osdc_put_request(req);
430 return NULL;
431 }
432
433 memset(msg->front.iov_base, 0, msg->front.iov_len);
434
435 req->r_request = msg;
436
437 return req;
438 }
439 EXPORT_SYMBOL(ceph_osdc_alloc_request);
440
441 static bool osd_req_opcode_valid(u16 opcode)
442 {
443 switch (opcode) {
444 #define GENERATE_CASE(op, opcode, str) case CEPH_OSD_OP_##op: return true;
445 __CEPH_FORALL_OSD_OPS(GENERATE_CASE)
446 #undef GENERATE_CASE
447 default:
448 return false;
449 }
450 }
451
452 /*
453 * This is an osd op init function for opcodes that have no data or
454 * other information associated with them. It also serves as a
455 * common init routine for all the other init functions, below.
456 */
457 static struct ceph_osd_req_op *
458 _osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which,
459 u16 opcode, u32 flags)
460 {
461 struct ceph_osd_req_op *op;
462
463 BUG_ON(which >= osd_req->r_num_ops);
464 BUG_ON(!osd_req_opcode_valid(opcode));
465
466 op = &osd_req->r_ops[which];
467 memset(op, 0, sizeof (*op));
468 op->op = opcode;
469 op->flags = flags;
470
471 return op;
472 }
473
474 void osd_req_op_init(struct ceph_osd_request *osd_req,
475 unsigned int which, u16 opcode, u32 flags)
476 {
477 (void)_osd_req_op_init(osd_req, which, opcode, flags);
478 }
479 EXPORT_SYMBOL(osd_req_op_init);
480
481 void osd_req_op_extent_init(struct ceph_osd_request *osd_req,
482 unsigned int which, u16 opcode,
483 u64 offset, u64 length,
484 u64 truncate_size, u32 truncate_seq)
485 {
486 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
487 opcode, 0);
488 size_t payload_len = 0;
489
490 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
491 opcode != CEPH_OSD_OP_WRITEFULL && opcode != CEPH_OSD_OP_ZERO &&
492 opcode != CEPH_OSD_OP_TRUNCATE);
493
494 op->extent.offset = offset;
495 op->extent.length = length;
496 op->extent.truncate_size = truncate_size;
497 op->extent.truncate_seq = truncate_seq;
498 if (opcode == CEPH_OSD_OP_WRITE || opcode == CEPH_OSD_OP_WRITEFULL)
499 payload_len += length;
500
501 op->payload_len = payload_len;
502 }
503 EXPORT_SYMBOL(osd_req_op_extent_init);
504
505 void osd_req_op_extent_update(struct ceph_osd_request *osd_req,
506 unsigned int which, u64 length)
507 {
508 struct ceph_osd_req_op *op;
509 u64 previous;
510
511 BUG_ON(which >= osd_req->r_num_ops);
512 op = &osd_req->r_ops[which];
513 previous = op->extent.length;
514
515 if (length == previous)
516 return; /* Nothing to do */
517 BUG_ON(length > previous);
518
519 op->extent.length = length;
520 op->payload_len -= previous - length;
521 }
522 EXPORT_SYMBOL(osd_req_op_extent_update);
523
524 void osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
525 u16 opcode, const char *class, const char *method)
526 {
527 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
528 opcode, 0);
529 struct ceph_pagelist *pagelist;
530 size_t payload_len = 0;
531 size_t size;
532
533 BUG_ON(opcode != CEPH_OSD_OP_CALL);
534
535 pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
536 BUG_ON(!pagelist);
537 ceph_pagelist_init(pagelist);
538
539 op->cls.class_name = class;
540 size = strlen(class);
541 BUG_ON(size > (size_t) U8_MAX);
542 op->cls.class_len = size;
543 ceph_pagelist_append(pagelist, class, size);
544 payload_len += size;
545
546 op->cls.method_name = method;
547 size = strlen(method);
548 BUG_ON(size > (size_t) U8_MAX);
549 op->cls.method_len = size;
550 ceph_pagelist_append(pagelist, method, size);
551 payload_len += size;
552
553 osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist);
554
555 op->cls.argc = 0; /* currently unused */
556
557 op->payload_len = payload_len;
558 }
559 EXPORT_SYMBOL(osd_req_op_cls_init);
560
561 int osd_req_op_xattr_init(struct ceph_osd_request *osd_req, unsigned int which,
562 u16 opcode, const char *name, const void *value,
563 size_t size, u8 cmp_op, u8 cmp_mode)
564 {
565 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
566 opcode, 0);
567 struct ceph_pagelist *pagelist;
568 size_t payload_len;
569
570 BUG_ON(opcode != CEPH_OSD_OP_SETXATTR && opcode != CEPH_OSD_OP_CMPXATTR);
571
572 pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
573 if (!pagelist)
574 return -ENOMEM;
575
576 ceph_pagelist_init(pagelist);
577
578 payload_len = strlen(name);
579 op->xattr.name_len = payload_len;
580 ceph_pagelist_append(pagelist, name, payload_len);
581
582 op->xattr.value_len = size;
583 ceph_pagelist_append(pagelist, value, size);
584 payload_len += size;
585
586 op->xattr.cmp_op = cmp_op;
587 op->xattr.cmp_mode = cmp_mode;
588
589 ceph_osd_data_pagelist_init(&op->xattr.osd_data, pagelist);
590 op->payload_len = payload_len;
591 return 0;
592 }
593 EXPORT_SYMBOL(osd_req_op_xattr_init);
594
595 void osd_req_op_watch_init(struct ceph_osd_request *osd_req,
596 unsigned int which, u16 opcode,
597 u64 cookie, u64 version, int flag)
598 {
599 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
600 opcode, 0);
601
602 BUG_ON(opcode != CEPH_OSD_OP_NOTIFY_ACK && opcode != CEPH_OSD_OP_WATCH);
603
604 op->watch.cookie = cookie;
605 op->watch.ver = version;
606 if (opcode == CEPH_OSD_OP_WATCH && flag)
607 op->watch.flag = (u8)1;
608 }
609 EXPORT_SYMBOL(osd_req_op_watch_init);
610
611 void osd_req_op_alloc_hint_init(struct ceph_osd_request *osd_req,
612 unsigned int which,
613 u64 expected_object_size,
614 u64 expected_write_size)
615 {
616 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
617 CEPH_OSD_OP_SETALLOCHINT,
618 0);
619
620 op->alloc_hint.expected_object_size = expected_object_size;
621 op->alloc_hint.expected_write_size = expected_write_size;
622
623 /*
624 * CEPH_OSD_OP_SETALLOCHINT op is advisory and therefore deemed
625 * not worth a feature bit. Set FAILOK per-op flag to make
626 * sure older osds don't trip over an unsupported opcode.
627 */
628 op->flags |= CEPH_OSD_OP_FLAG_FAILOK;
629 }
630 EXPORT_SYMBOL(osd_req_op_alloc_hint_init);
631
632 static void ceph_osdc_msg_data_add(struct ceph_msg *msg,
633 struct ceph_osd_data *osd_data)
634 {
635 u64 length = ceph_osd_data_length(osd_data);
636
637 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
638 BUG_ON(length > (u64) SIZE_MAX);
639 if (length)
640 ceph_msg_data_add_pages(msg, osd_data->pages,
641 length, osd_data->alignment);
642 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
643 BUG_ON(!length);
644 ceph_msg_data_add_pagelist(msg, osd_data->pagelist);
645 #ifdef CONFIG_BLOCK
646 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
647 ceph_msg_data_add_bio(msg, osd_data->bio, length);
648 #endif
649 } else {
650 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
651 }
652 }
653
654 static u64 osd_req_encode_op(struct ceph_osd_request *req,
655 struct ceph_osd_op *dst, unsigned int which)
656 {
657 struct ceph_osd_req_op *src;
658 struct ceph_osd_data *osd_data;
659 u64 request_data_len = 0;
660 u64 data_length;
661
662 BUG_ON(which >= req->r_num_ops);
663 src = &req->r_ops[which];
664 if (WARN_ON(!osd_req_opcode_valid(src->op))) {
665 pr_err("unrecognized osd opcode %d\n", src->op);
666
667 return 0;
668 }
669
670 switch (src->op) {
671 case CEPH_OSD_OP_STAT:
672 osd_data = &src->raw_data_in;
673 ceph_osdc_msg_data_add(req->r_reply, osd_data);
674 break;
675 case CEPH_OSD_OP_READ:
676 case CEPH_OSD_OP_WRITE:
677 case CEPH_OSD_OP_WRITEFULL:
678 case CEPH_OSD_OP_ZERO:
679 case CEPH_OSD_OP_TRUNCATE:
680 if (src->op == CEPH_OSD_OP_WRITE ||
681 src->op == CEPH_OSD_OP_WRITEFULL)
682 request_data_len = src->extent.length;
683 dst->extent.offset = cpu_to_le64(src->extent.offset);
684 dst->extent.length = cpu_to_le64(src->extent.length);
685 dst->extent.truncate_size =
686 cpu_to_le64(src->extent.truncate_size);
687 dst->extent.truncate_seq =
688 cpu_to_le32(src->extent.truncate_seq);
689 osd_data = &src->extent.osd_data;
690 if (src->op == CEPH_OSD_OP_WRITE ||
691 src->op == CEPH_OSD_OP_WRITEFULL)
692 ceph_osdc_msg_data_add(req->r_request, osd_data);
693 else
694 ceph_osdc_msg_data_add(req->r_reply, osd_data);
695 break;
696 case CEPH_OSD_OP_CALL:
697 dst->cls.class_len = src->cls.class_len;
698 dst->cls.method_len = src->cls.method_len;
699 osd_data = &src->cls.request_info;
700 ceph_osdc_msg_data_add(req->r_request, osd_data);
701 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGELIST);
702 request_data_len = osd_data->pagelist->length;
703
704 osd_data = &src->cls.request_data;
705 data_length = ceph_osd_data_length(osd_data);
706 if (data_length) {
707 BUG_ON(osd_data->type == CEPH_OSD_DATA_TYPE_NONE);
708 dst->cls.indata_len = cpu_to_le32(data_length);
709 ceph_osdc_msg_data_add(req->r_request, osd_data);
710 src->payload_len += data_length;
711 request_data_len += data_length;
712 }
713 osd_data = &src->cls.response_data;
714 ceph_osdc_msg_data_add(req->r_reply, osd_data);
715 break;
716 case CEPH_OSD_OP_STARTSYNC:
717 break;
718 case CEPH_OSD_OP_NOTIFY_ACK:
719 case CEPH_OSD_OP_WATCH:
720 dst->watch.cookie = cpu_to_le64(src->watch.cookie);
721 dst->watch.ver = cpu_to_le64(src->watch.ver);
722 dst->watch.flag = src->watch.flag;
723 break;
724 case CEPH_OSD_OP_SETALLOCHINT:
725 dst->alloc_hint.expected_object_size =
726 cpu_to_le64(src->alloc_hint.expected_object_size);
727 dst->alloc_hint.expected_write_size =
728 cpu_to_le64(src->alloc_hint.expected_write_size);
729 break;
730 case CEPH_OSD_OP_SETXATTR:
731 case CEPH_OSD_OP_CMPXATTR:
732 dst->xattr.name_len = cpu_to_le32(src->xattr.name_len);
733 dst->xattr.value_len = cpu_to_le32(src->xattr.value_len);
734 dst->xattr.cmp_op = src->xattr.cmp_op;
735 dst->xattr.cmp_mode = src->xattr.cmp_mode;
736 osd_data = &src->xattr.osd_data;
737 ceph_osdc_msg_data_add(req->r_request, osd_data);
738 request_data_len = osd_data->pagelist->length;
739 break;
740 case CEPH_OSD_OP_CREATE:
741 case CEPH_OSD_OP_DELETE:
742 break;
743 default:
744 pr_err("unsupported osd opcode %s\n",
745 ceph_osd_op_name(src->op));
746 WARN_ON(1);
747
748 return 0;
749 }
750
751 dst->op = cpu_to_le16(src->op);
752 dst->flags = cpu_to_le32(src->flags);
753 dst->payload_len = cpu_to_le32(src->payload_len);
754
755 return request_data_len;
756 }
757
758 /*
759 * build new request AND message, calculate layout, and adjust file
760 * extent as needed.
761 *
762 * if the file was recently truncated, we include information about its
763 * old and new size so that the object can be updated appropriately. (we
764 * avoid synchronously deleting truncated objects because it's slow.)
765 *
766 * if @do_sync, include a 'startsync' command so that the osd will flush
767 * data quickly.
768 */
769 struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
770 struct ceph_file_layout *layout,
771 struct ceph_vino vino,
772 u64 off, u64 *plen,
773 unsigned int which, int num_ops,
774 int opcode, int flags,
775 struct ceph_snap_context *snapc,
776 u32 truncate_seq,
777 u64 truncate_size,
778 bool use_mempool)
779 {
780 struct ceph_osd_request *req;
781 u64 objnum = 0;
782 u64 objoff = 0;
783 u64 objlen = 0;
784 int r;
785
786 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
787 opcode != CEPH_OSD_OP_ZERO && opcode != CEPH_OSD_OP_TRUNCATE &&
788 opcode != CEPH_OSD_OP_CREATE && opcode != CEPH_OSD_OP_DELETE);
789
790 req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool,
791 GFP_NOFS);
792 if (!req)
793 return ERR_PTR(-ENOMEM);
794
795 req->r_flags = flags;
796
797 /* calculate max write size */
798 r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
799 if (r < 0) {
800 ceph_osdc_put_request(req);
801 return ERR_PTR(r);
802 }
803
804 if (opcode == CEPH_OSD_OP_CREATE || opcode == CEPH_OSD_OP_DELETE) {
805 osd_req_op_init(req, which, opcode, 0);
806 } else {
807 u32 object_size = le32_to_cpu(layout->fl_object_size);
808 u32 object_base = off - objoff;
809 if (!(truncate_seq == 1 && truncate_size == -1ULL)) {
810 if (truncate_size <= object_base) {
811 truncate_size = 0;
812 } else {
813 truncate_size -= object_base;
814 if (truncate_size > object_size)
815 truncate_size = object_size;
816 }
817 }
818 osd_req_op_extent_init(req, which, opcode, objoff, objlen,
819 truncate_size, truncate_seq);
820 }
821
822 req->r_base_oloc.pool = ceph_file_layout_pg_pool(*layout);
823
824 snprintf(req->r_base_oid.name, sizeof(req->r_base_oid.name),
825 "%llx.%08llx", vino.ino, objnum);
826 req->r_base_oid.name_len = strlen(req->r_base_oid.name);
827
828 return req;
829 }
830 EXPORT_SYMBOL(ceph_osdc_new_request);
831
832 /*
833 * We keep osd requests in an rbtree, sorted by ->r_tid.
834 */
835 static void __insert_request(struct ceph_osd_client *osdc,
836 struct ceph_osd_request *new)
837 {
838 struct rb_node **p = &osdc->requests.rb_node;
839 struct rb_node *parent = NULL;
840 struct ceph_osd_request *req = NULL;
841
842 while (*p) {
843 parent = *p;
844 req = rb_entry(parent, struct ceph_osd_request, r_node);
845 if (new->r_tid < req->r_tid)
846 p = &(*p)->rb_left;
847 else if (new->r_tid > req->r_tid)
848 p = &(*p)->rb_right;
849 else
850 BUG();
851 }
852
853 rb_link_node(&new->r_node, parent, p);
854 rb_insert_color(&new->r_node, &osdc->requests);
855 }
856
857 static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc,
858 u64 tid)
859 {
860 struct ceph_osd_request *req;
861 struct rb_node *n = osdc->requests.rb_node;
862
863 while (n) {
864 req = rb_entry(n, struct ceph_osd_request, r_node);
865 if (tid < req->r_tid)
866 n = n->rb_left;
867 else if (tid > req->r_tid)
868 n = n->rb_right;
869 else
870 return req;
871 }
872 return NULL;
873 }
874
875 static struct ceph_osd_request *
876 __lookup_request_ge(struct ceph_osd_client *osdc,
877 u64 tid)
878 {
879 struct ceph_osd_request *req;
880 struct rb_node *n = osdc->requests.rb_node;
881
882 while (n) {
883 req = rb_entry(n, struct ceph_osd_request, r_node);
884 if (tid < req->r_tid) {
885 if (!n->rb_left)
886 return req;
887 n = n->rb_left;
888 } else if (tid > req->r_tid) {
889 n = n->rb_right;
890 } else {
891 return req;
892 }
893 }
894 return NULL;
895 }
896
897 static void __kick_linger_request(struct ceph_osd_request *req)
898 {
899 struct ceph_osd_client *osdc = req->r_osdc;
900 struct ceph_osd *osd = req->r_osd;
901
902 /*
903 * Linger requests need to be resent with a new tid to avoid
904 * the dup op detection logic on the OSDs. Achieve this with
905 * a re-register dance instead of open-coding.
906 */
907 ceph_osdc_get_request(req);
908 if (!list_empty(&req->r_linger_item))
909 __unregister_linger_request(osdc, req);
910 else
911 __unregister_request(osdc, req);
912 __register_request(osdc, req);
913 ceph_osdc_put_request(req);
914
915 /*
916 * Unless request has been registered as both normal and
917 * lingering, __unregister{,_linger}_request clears r_osd.
918 * However, here we need to preserve r_osd to make sure we
919 * requeue on the same OSD.
920 */
921 WARN_ON(req->r_osd || !osd);
922 req->r_osd = osd;
923
924 dout("%s requeueing %p tid %llu\n", __func__, req, req->r_tid);
925 __enqueue_request(req);
926 }
927
928 /*
929 * Resubmit requests pending on the given osd.
930 */
931 static void __kick_osd_requests(struct ceph_osd_client *osdc,
932 struct ceph_osd *osd)
933 {
934 struct ceph_osd_request *req, *nreq;
935 LIST_HEAD(resend);
936 LIST_HEAD(resend_linger);
937 int err;
938
939 dout("%s osd%d\n", __func__, osd->o_osd);
940 err = __reset_osd(osdc, osd);
941 if (err)
942 return;
943
944 /*
945 * Build up a list of requests to resend by traversing the
946 * osd's list of requests. Requests for a given object are
947 * sent in tid order, and that is also the order they're
948 * kept on this list. Therefore all requests that are in
949 * flight will be found first, followed by all requests that
950 * have not yet been sent. And to resend requests while
951 * preserving this order we will want to put any sent
952 * requests back on the front of the osd client's unsent
953 * list.
954 *
955 * So we build a separate ordered list of already-sent
956 * requests for the affected osd and splice it onto the
957 * front of the osd client's unsent list. Once we've seen a
958 * request that has not yet been sent we're done. Those
959 * requests are already sitting right where they belong.
960 */
961 list_for_each_entry(req, &osd->o_requests, r_osd_item) {
962 if (!req->r_sent)
963 break;
964
965 if (!req->r_linger) {
966 dout("%s requeueing %p tid %llu\n", __func__, req,
967 req->r_tid);
968 list_move_tail(&req->r_req_lru_item, &resend);
969 req->r_flags |= CEPH_OSD_FLAG_RETRY;
970 } else {
971 list_move_tail(&req->r_req_lru_item, &resend_linger);
972 }
973 }
974 list_splice(&resend, &osdc->req_unsent);
975
976 /*
977 * Both registered and not yet registered linger requests are
978 * enqueued with a new tid on the same OSD. We add/move them
979 * to req_unsent/o_requests at the end to keep things in tid
980 * order.
981 */
982 list_for_each_entry_safe(req, nreq, &osd->o_linger_requests,
983 r_linger_osd_item) {
984 WARN_ON(!list_empty(&req->r_req_lru_item));
985 __kick_linger_request(req);
986 }
987
988 list_for_each_entry_safe(req, nreq, &resend_linger, r_req_lru_item)
989 __kick_linger_request(req);
990 }
991
992 /*
993 * If the osd connection drops, we need to resubmit all requests.
994 */
995 static void osd_reset(struct ceph_connection *con)
996 {
997 struct ceph_osd *osd = con->private;
998 struct ceph_osd_client *osdc;
999
1000 if (!osd)
1001 return;
1002 dout("osd_reset osd%d\n", osd->o_osd);
1003 osdc = osd->o_osdc;
1004 down_read(&osdc->map_sem);
1005 mutex_lock(&osdc->request_mutex);
1006 __kick_osd_requests(osdc, osd);
1007 __send_queued(osdc);
1008 mutex_unlock(&osdc->request_mutex);
1009 up_read(&osdc->map_sem);
1010 }
1011
1012 /*
1013 * Track open sessions with osds.
1014 */
1015 static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
1016 {
1017 struct ceph_osd *osd;
1018
1019 osd = kzalloc(sizeof(*osd), GFP_NOFS);
1020 if (!osd)
1021 return NULL;
1022
1023 atomic_set(&osd->o_ref, 1);
1024 osd->o_osdc = osdc;
1025 osd->o_osd = onum;
1026 RB_CLEAR_NODE(&osd->o_node);
1027 INIT_LIST_HEAD(&osd->o_requests);
1028 INIT_LIST_HEAD(&osd->o_linger_requests);
1029 INIT_LIST_HEAD(&osd->o_osd_lru);
1030 osd->o_incarnation = 1;
1031
1032 ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
1033
1034 INIT_LIST_HEAD(&osd->o_keepalive_item);
1035 return osd;
1036 }
1037
1038 static struct ceph_osd *get_osd(struct ceph_osd *osd)
1039 {
1040 if (atomic_inc_not_zero(&osd->o_ref)) {
1041 dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
1042 atomic_read(&osd->o_ref));
1043 return osd;
1044 } else {
1045 dout("get_osd %p FAIL\n", osd);
1046 return NULL;
1047 }
1048 }
1049
1050 static void put_osd(struct ceph_osd *osd)
1051 {
1052 dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
1053 atomic_read(&osd->o_ref) - 1);
1054 if (atomic_dec_and_test(&osd->o_ref)) {
1055 struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth;
1056
1057 if (osd->o_auth.authorizer)
1058 ceph_auth_destroy_authorizer(ac, osd->o_auth.authorizer);
1059 kfree(osd);
1060 }
1061 }
1062
1063 /*
1064 * remove an osd from our map
1065 */
1066 static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
1067 {
1068 dout("%s %p osd%d\n", __func__, osd, osd->o_osd);
1069 WARN_ON(!list_empty(&osd->o_requests));
1070 WARN_ON(!list_empty(&osd->o_linger_requests));
1071
1072 list_del_init(&osd->o_osd_lru);
1073 rb_erase(&osd->o_node, &osdc->osds);
1074 RB_CLEAR_NODE(&osd->o_node);
1075 }
1076
1077 static void remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
1078 {
1079 dout("%s %p osd%d\n", __func__, osd, osd->o_osd);
1080
1081 if (!RB_EMPTY_NODE(&osd->o_node)) {
1082 ceph_con_close(&osd->o_con);
1083 __remove_osd(osdc, osd);
1084 put_osd(osd);
1085 }
1086 }
1087
1088 static void remove_all_osds(struct ceph_osd_client *osdc)
1089 {
1090 dout("%s %p\n", __func__, osdc);
1091 mutex_lock(&osdc->request_mutex);
1092 while (!RB_EMPTY_ROOT(&osdc->osds)) {
1093 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
1094 struct ceph_osd, o_node);
1095 remove_osd(osdc, osd);
1096 }
1097 mutex_unlock(&osdc->request_mutex);
1098 }
1099
1100 static void __move_osd_to_lru(struct ceph_osd_client *osdc,
1101 struct ceph_osd *osd)
1102 {
1103 dout("%s %p\n", __func__, osd);
1104 BUG_ON(!list_empty(&osd->o_osd_lru));
1105
1106 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
1107 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl;
1108 }
1109
1110 static void maybe_move_osd_to_lru(struct ceph_osd_client *osdc,
1111 struct ceph_osd *osd)
1112 {
1113 dout("%s %p\n", __func__, osd);
1114
1115 if (list_empty(&osd->o_requests) &&
1116 list_empty(&osd->o_linger_requests))
1117 __move_osd_to_lru(osdc, osd);
1118 }
1119
1120 static void __remove_osd_from_lru(struct ceph_osd *osd)
1121 {
1122 dout("__remove_osd_from_lru %p\n", osd);
1123 if (!list_empty(&osd->o_osd_lru))
1124 list_del_init(&osd->o_osd_lru);
1125 }
1126
1127 static void remove_old_osds(struct ceph_osd_client *osdc)
1128 {
1129 struct ceph_osd *osd, *nosd;
1130
1131 dout("__remove_old_osds %p\n", osdc);
1132 mutex_lock(&osdc->request_mutex);
1133 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
1134 if (time_before(jiffies, osd->lru_ttl))
1135 break;
1136 remove_osd(osdc, osd);
1137 }
1138 mutex_unlock(&osdc->request_mutex);
1139 }
1140
1141 /*
1142 * reset osd connect
1143 */
1144 static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
1145 {
1146 struct ceph_entity_addr *peer_addr;
1147
1148 dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
1149 if (list_empty(&osd->o_requests) &&
1150 list_empty(&osd->o_linger_requests)) {
1151 remove_osd(osdc, osd);
1152 return -ENODEV;
1153 }
1154
1155 peer_addr = &osdc->osdmap->osd_addr[osd->o_osd];
1156 if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1157 !ceph_con_opened(&osd->o_con)) {
1158 struct ceph_osd_request *req;
1159
1160 dout("osd addr hasn't changed and connection never opened, "
1161 "letting msgr retry\n");
1162 /* touch each r_stamp for handle_timeout()'s benfit */
1163 list_for_each_entry(req, &osd->o_requests, r_osd_item)
1164 req->r_stamp = jiffies;
1165
1166 return -EAGAIN;
1167 }
1168
1169 ceph_con_close(&osd->o_con);
1170 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1171 osd->o_incarnation++;
1172
1173 return 0;
1174 }
1175
1176 static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new)
1177 {
1178 struct rb_node **p = &osdc->osds.rb_node;
1179 struct rb_node *parent = NULL;
1180 struct ceph_osd *osd = NULL;
1181
1182 dout("__insert_osd %p osd%d\n", new, new->o_osd);
1183 while (*p) {
1184 parent = *p;
1185 osd = rb_entry(parent, struct ceph_osd, o_node);
1186 if (new->o_osd < osd->o_osd)
1187 p = &(*p)->rb_left;
1188 else if (new->o_osd > osd->o_osd)
1189 p = &(*p)->rb_right;
1190 else
1191 BUG();
1192 }
1193
1194 rb_link_node(&new->o_node, parent, p);
1195 rb_insert_color(&new->o_node, &osdc->osds);
1196 }
1197
1198 static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o)
1199 {
1200 struct ceph_osd *osd;
1201 struct rb_node *n = osdc->osds.rb_node;
1202
1203 while (n) {
1204 osd = rb_entry(n, struct ceph_osd, o_node);
1205 if (o < osd->o_osd)
1206 n = n->rb_left;
1207 else if (o > osd->o_osd)
1208 n = n->rb_right;
1209 else
1210 return osd;
1211 }
1212 return NULL;
1213 }
1214
1215 static void __schedule_osd_timeout(struct ceph_osd_client *osdc)
1216 {
1217 schedule_delayed_work(&osdc->timeout_work,
1218 osdc->client->options->osd_keepalive_timeout);
1219 }
1220
1221 static void __cancel_osd_timeout(struct ceph_osd_client *osdc)
1222 {
1223 cancel_delayed_work(&osdc->timeout_work);
1224 }
1225
1226 /*
1227 * Register request, assign tid. If this is the first request, set up
1228 * the timeout event.
1229 */
1230 static void __register_request(struct ceph_osd_client *osdc,
1231 struct ceph_osd_request *req)
1232 {
1233 req->r_tid = ++osdc->last_tid;
1234 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
1235 dout("__register_request %p tid %lld\n", req, req->r_tid);
1236 __insert_request(osdc, req);
1237 ceph_osdc_get_request(req);
1238 osdc->num_requests++;
1239 if (osdc->num_requests == 1) {
1240 dout(" first request, scheduling timeout\n");
1241 __schedule_osd_timeout(osdc);
1242 }
1243 }
1244
1245 /*
1246 * called under osdc->request_mutex
1247 */
1248 static void __unregister_request(struct ceph_osd_client *osdc,
1249 struct ceph_osd_request *req)
1250 {
1251 if (RB_EMPTY_NODE(&req->r_node)) {
1252 dout("__unregister_request %p tid %lld not registered\n",
1253 req, req->r_tid);
1254 return;
1255 }
1256
1257 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
1258 rb_erase(&req->r_node, &osdc->requests);
1259 RB_CLEAR_NODE(&req->r_node);
1260 osdc->num_requests--;
1261
1262 if (req->r_osd) {
1263 /* make sure the original request isn't in flight. */
1264 ceph_msg_revoke(req->r_request);
1265
1266 list_del_init(&req->r_osd_item);
1267 maybe_move_osd_to_lru(osdc, req->r_osd);
1268 if (list_empty(&req->r_linger_osd_item))
1269 req->r_osd = NULL;
1270 }
1271
1272 list_del_init(&req->r_req_lru_item);
1273 ceph_osdc_put_request(req);
1274
1275 if (osdc->num_requests == 0) {
1276 dout(" no requests, canceling timeout\n");
1277 __cancel_osd_timeout(osdc);
1278 }
1279 }
1280
1281 /*
1282 * Cancel a previously queued request message
1283 */
1284 static void __cancel_request(struct ceph_osd_request *req)
1285 {
1286 if (req->r_sent && req->r_osd) {
1287 ceph_msg_revoke(req->r_request);
1288 req->r_sent = 0;
1289 }
1290 }
1291
1292 static void __register_linger_request(struct ceph_osd_client *osdc,
1293 struct ceph_osd_request *req)
1294 {
1295 dout("%s %p tid %llu\n", __func__, req, req->r_tid);
1296 WARN_ON(!req->r_linger);
1297
1298 ceph_osdc_get_request(req);
1299 list_add_tail(&req->r_linger_item, &osdc->req_linger);
1300 if (req->r_osd)
1301 list_add_tail(&req->r_linger_osd_item,
1302 &req->r_osd->o_linger_requests);
1303 }
1304
1305 static void __unregister_linger_request(struct ceph_osd_client *osdc,
1306 struct ceph_osd_request *req)
1307 {
1308 WARN_ON(!req->r_linger);
1309
1310 if (list_empty(&req->r_linger_item)) {
1311 dout("%s %p tid %llu not registered\n", __func__, req,
1312 req->r_tid);
1313 return;
1314 }
1315
1316 dout("%s %p tid %llu\n", __func__, req, req->r_tid);
1317 list_del_init(&req->r_linger_item);
1318
1319 if (req->r_osd) {
1320 list_del_init(&req->r_linger_osd_item);
1321 maybe_move_osd_to_lru(osdc, req->r_osd);
1322 if (list_empty(&req->r_osd_item))
1323 req->r_osd = NULL;
1324 }
1325 ceph_osdc_put_request(req);
1326 }
1327
1328 void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
1329 struct ceph_osd_request *req)
1330 {
1331 if (!req->r_linger) {
1332 dout("set_request_linger %p\n", req);
1333 req->r_linger = 1;
1334 }
1335 }
1336 EXPORT_SYMBOL(ceph_osdc_set_request_linger);
1337
1338 /*
1339 * Returns whether a request should be blocked from being sent
1340 * based on the current osdmap and osd_client settings.
1341 *
1342 * Caller should hold map_sem for read.
1343 */
1344 static bool __req_should_be_paused(struct ceph_osd_client *osdc,
1345 struct ceph_osd_request *req)
1346 {
1347 bool pauserd = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD);
1348 bool pausewr = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR) ||
1349 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
1350 return (req->r_flags & CEPH_OSD_FLAG_READ && pauserd) ||
1351 (req->r_flags & CEPH_OSD_FLAG_WRITE && pausewr);
1352 }
1353
1354 /*
1355 * Calculate mapping of a request to a PG. Takes tiering into account.
1356 */
1357 static int __calc_request_pg(struct ceph_osdmap *osdmap,
1358 struct ceph_osd_request *req,
1359 struct ceph_pg *pg_out)
1360 {
1361 bool need_check_tiering;
1362
1363 need_check_tiering = false;
1364 if (req->r_target_oloc.pool == -1) {
1365 req->r_target_oloc = req->r_base_oloc; /* struct */
1366 need_check_tiering = true;
1367 }
1368 if (req->r_target_oid.name_len == 0) {
1369 ceph_oid_copy(&req->r_target_oid, &req->r_base_oid);
1370 need_check_tiering = true;
1371 }
1372
1373 if (need_check_tiering &&
1374 (req->r_flags & CEPH_OSD_FLAG_IGNORE_OVERLAY) == 0) {
1375 struct ceph_pg_pool_info *pi;
1376
1377 pi = ceph_pg_pool_by_id(osdmap, req->r_target_oloc.pool);
1378 if (pi) {
1379 if ((req->r_flags & CEPH_OSD_FLAG_READ) &&
1380 pi->read_tier >= 0)
1381 req->r_target_oloc.pool = pi->read_tier;
1382 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
1383 pi->write_tier >= 0)
1384 req->r_target_oloc.pool = pi->write_tier;
1385 }
1386 /* !pi is caught in ceph_oloc_oid_to_pg() */
1387 }
1388
1389 return ceph_oloc_oid_to_pg(osdmap, &req->r_target_oloc,
1390 &req->r_target_oid, pg_out);
1391 }
1392
1393 static void __enqueue_request(struct ceph_osd_request *req)
1394 {
1395 struct ceph_osd_client *osdc = req->r_osdc;
1396
1397 dout("%s %p tid %llu to osd%d\n", __func__, req, req->r_tid,
1398 req->r_osd ? req->r_osd->o_osd : -1);
1399
1400 if (req->r_osd) {
1401 __remove_osd_from_lru(req->r_osd);
1402 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
1403 list_move_tail(&req->r_req_lru_item, &osdc->req_unsent);
1404 } else {
1405 list_move_tail(&req->r_req_lru_item, &osdc->req_notarget);
1406 }
1407 }
1408
1409 /*
1410 * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
1411 * (as needed), and set the request r_osd appropriately. If there is
1412 * no up osd, set r_osd to NULL. Move the request to the appropriate list
1413 * (unsent, homeless) or leave on in-flight lru.
1414 *
1415 * Return 0 if unchanged, 1 if changed, or negative on error.
1416 *
1417 * Caller should hold map_sem for read and request_mutex.
1418 */
1419 static int __map_request(struct ceph_osd_client *osdc,
1420 struct ceph_osd_request *req, int force_resend)
1421 {
1422 struct ceph_pg pgid;
1423 int acting[CEPH_PG_MAX_SIZE];
1424 int num, o;
1425 int err;
1426 bool was_paused;
1427
1428 dout("map_request %p tid %lld\n", req, req->r_tid);
1429
1430 err = __calc_request_pg(osdc->osdmap, req, &pgid);
1431 if (err) {
1432 list_move(&req->r_req_lru_item, &osdc->req_notarget);
1433 return err;
1434 }
1435 req->r_pgid = pgid;
1436
1437 num = ceph_calc_pg_acting(osdc->osdmap, pgid, acting, &o);
1438 if (num < 0)
1439 num = 0;
1440
1441 was_paused = req->r_paused;
1442 req->r_paused = __req_should_be_paused(osdc, req);
1443 if (was_paused && !req->r_paused)
1444 force_resend = 1;
1445
1446 if ((!force_resend &&
1447 req->r_osd && req->r_osd->o_osd == o &&
1448 req->r_sent >= req->r_osd->o_incarnation &&
1449 req->r_num_pg_osds == num &&
1450 memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
1451 (req->r_osd == NULL && o == -1) ||
1452 req->r_paused)
1453 return 0; /* no change */
1454
1455 dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n",
1456 req->r_tid, pgid.pool, pgid.seed, o,
1457 req->r_osd ? req->r_osd->o_osd : -1);
1458
1459 /* record full pg acting set */
1460 memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
1461 req->r_num_pg_osds = num;
1462
1463 if (req->r_osd) {
1464 __cancel_request(req);
1465 list_del_init(&req->r_osd_item);
1466 list_del_init(&req->r_linger_osd_item);
1467 req->r_osd = NULL;
1468 }
1469
1470 req->r_osd = __lookup_osd(osdc, o);
1471 if (!req->r_osd && o >= 0) {
1472 err = -ENOMEM;
1473 req->r_osd = create_osd(osdc, o);
1474 if (!req->r_osd) {
1475 list_move(&req->r_req_lru_item, &osdc->req_notarget);
1476 goto out;
1477 }
1478
1479 dout("map_request osd %p is osd%d\n", req->r_osd, o);
1480 __insert_osd(osdc, req->r_osd);
1481
1482 ceph_con_open(&req->r_osd->o_con,
1483 CEPH_ENTITY_TYPE_OSD, o,
1484 &osdc->osdmap->osd_addr[o]);
1485 }
1486
1487 __enqueue_request(req);
1488 err = 1; /* osd or pg changed */
1489
1490 out:
1491 return err;
1492 }
1493
1494 /*
1495 * caller should hold map_sem (for read) and request_mutex
1496 */
1497 static void __send_request(struct ceph_osd_client *osdc,
1498 struct ceph_osd_request *req)
1499 {
1500 void *p;
1501
1502 dout("send_request %p tid %llu to osd%d flags %d pg %lld.%x\n",
1503 req, req->r_tid, req->r_osd->o_osd, req->r_flags,
1504 (unsigned long long)req->r_pgid.pool, req->r_pgid.seed);
1505
1506 /* fill in message content that changes each time we send it */
1507 put_unaligned_le32(osdc->osdmap->epoch, req->r_request_osdmap_epoch);
1508 put_unaligned_le32(req->r_flags, req->r_request_flags);
1509 put_unaligned_le64(req->r_target_oloc.pool, req->r_request_pool);
1510 p = req->r_request_pgid;
1511 ceph_encode_64(&p, req->r_pgid.pool);
1512 ceph_encode_32(&p, req->r_pgid.seed);
1513 put_unaligned_le64(1, req->r_request_attempts); /* FIXME */
1514 memcpy(req->r_request_reassert_version, &req->r_reassert_version,
1515 sizeof(req->r_reassert_version));
1516
1517 req->r_stamp = jiffies;
1518 list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
1519
1520 ceph_msg_get(req->r_request); /* send consumes a ref */
1521
1522 req->r_sent = req->r_osd->o_incarnation;
1523
1524 ceph_con_send(&req->r_osd->o_con, req->r_request);
1525 }
1526
1527 /*
1528 * Send any requests in the queue (req_unsent).
1529 */
1530 static void __send_queued(struct ceph_osd_client *osdc)
1531 {
1532 struct ceph_osd_request *req, *tmp;
1533
1534 dout("__send_queued\n");
1535 list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item)
1536 __send_request(osdc, req);
1537 }
1538
1539 /*
1540 * Caller should hold map_sem for read and request_mutex.
1541 */
1542 static int __ceph_osdc_start_request(struct ceph_osd_client *osdc,
1543 struct ceph_osd_request *req,
1544 bool nofail)
1545 {
1546 int rc;
1547
1548 __register_request(osdc, req);
1549 req->r_sent = 0;
1550 req->r_got_reply = 0;
1551 rc = __map_request(osdc, req, 0);
1552 if (rc < 0) {
1553 if (nofail) {
1554 dout("osdc_start_request failed map, "
1555 " will retry %lld\n", req->r_tid);
1556 rc = 0;
1557 } else {
1558 __unregister_request(osdc, req);
1559 }
1560 return rc;
1561 }
1562
1563 if (req->r_osd == NULL) {
1564 dout("send_request %p no up osds in pg\n", req);
1565 ceph_monc_request_next_osdmap(&osdc->client->monc);
1566 } else {
1567 __send_queued(osdc);
1568 }
1569
1570 return 0;
1571 }
1572
1573 /*
1574 * Timeout callback, called every N seconds when 1 or more osd
1575 * requests has been active for more than N seconds. When this
1576 * happens, we ping all OSDs with requests who have timed out to
1577 * ensure any communications channel reset is detected. Reset the
1578 * request timeouts another N seconds in the future as we go.
1579 * Reschedule the timeout event another N seconds in future (unless
1580 * there are no open requests).
1581 */
1582 static void handle_timeout(struct work_struct *work)
1583 {
1584 struct ceph_osd_client *osdc =
1585 container_of(work, struct ceph_osd_client, timeout_work.work);
1586 struct ceph_options *opts = osdc->client->options;
1587 struct ceph_osd_request *req;
1588 struct ceph_osd *osd;
1589 struct list_head slow_osds;
1590 dout("timeout\n");
1591 down_read(&osdc->map_sem);
1592
1593 ceph_monc_request_next_osdmap(&osdc->client->monc);
1594
1595 mutex_lock(&osdc->request_mutex);
1596
1597 /*
1598 * ping osds that are a bit slow. this ensures that if there
1599 * is a break in the TCP connection we will notice, and reopen
1600 * a connection with that osd (from the fault callback).
1601 */
1602 INIT_LIST_HEAD(&slow_osds);
1603 list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
1604 if (time_before(jiffies,
1605 req->r_stamp + opts->osd_keepalive_timeout))
1606 break;
1607
1608 osd = req->r_osd;
1609 BUG_ON(!osd);
1610 dout(" tid %llu is slow, will send keepalive on osd%d\n",
1611 req->r_tid, osd->o_osd);
1612 list_move_tail(&osd->o_keepalive_item, &slow_osds);
1613 }
1614 while (!list_empty(&slow_osds)) {
1615 osd = list_entry(slow_osds.next, struct ceph_osd,
1616 o_keepalive_item);
1617 list_del_init(&osd->o_keepalive_item);
1618 ceph_con_keepalive(&osd->o_con);
1619 }
1620
1621 __schedule_osd_timeout(osdc);
1622 __send_queued(osdc);
1623 mutex_unlock(&osdc->request_mutex);
1624 up_read(&osdc->map_sem);
1625 }
1626
1627 static void handle_osds_timeout(struct work_struct *work)
1628 {
1629 struct ceph_osd_client *osdc =
1630 container_of(work, struct ceph_osd_client,
1631 osds_timeout_work.work);
1632 unsigned long delay = osdc->client->options->osd_idle_ttl / 4;
1633
1634 dout("osds timeout\n");
1635 down_read(&osdc->map_sem);
1636 remove_old_osds(osdc);
1637 up_read(&osdc->map_sem);
1638
1639 schedule_delayed_work(&osdc->osds_timeout_work,
1640 round_jiffies_relative(delay));
1641 }
1642
1643 static int ceph_oloc_decode(void **p, void *end,
1644 struct ceph_object_locator *oloc)
1645 {
1646 u8 struct_v, struct_cv;
1647 u32 len;
1648 void *struct_end;
1649 int ret = 0;
1650
1651 ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
1652 struct_v = ceph_decode_8(p);
1653 struct_cv = ceph_decode_8(p);
1654 if (struct_v < 3) {
1655 pr_warn("got v %d < 3 cv %d of ceph_object_locator\n",
1656 struct_v, struct_cv);
1657 goto e_inval;
1658 }
1659 if (struct_cv > 6) {
1660 pr_warn("got v %d cv %d > 6 of ceph_object_locator\n",
1661 struct_v, struct_cv);
1662 goto e_inval;
1663 }
1664 len = ceph_decode_32(p);
1665 ceph_decode_need(p, end, len, e_inval);
1666 struct_end = *p + len;
1667
1668 oloc->pool = ceph_decode_64(p);
1669 *p += 4; /* skip preferred */
1670
1671 len = ceph_decode_32(p);
1672 if (len > 0) {
1673 pr_warn("ceph_object_locator::key is set\n");
1674 goto e_inval;
1675 }
1676
1677 if (struct_v >= 5) {
1678 len = ceph_decode_32(p);
1679 if (len > 0) {
1680 pr_warn("ceph_object_locator::nspace is set\n");
1681 goto e_inval;
1682 }
1683 }
1684
1685 if (struct_v >= 6) {
1686 s64 hash = ceph_decode_64(p);
1687 if (hash != -1) {
1688 pr_warn("ceph_object_locator::hash is set\n");
1689 goto e_inval;
1690 }
1691 }
1692
1693 /* skip the rest */
1694 *p = struct_end;
1695 out:
1696 return ret;
1697
1698 e_inval:
1699 ret = -EINVAL;
1700 goto out;
1701 }
1702
1703 static int ceph_redirect_decode(void **p, void *end,
1704 struct ceph_request_redirect *redir)
1705 {
1706 u8 struct_v, struct_cv;
1707 u32 len;
1708 void *struct_end;
1709 int ret;
1710
1711 ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
1712 struct_v = ceph_decode_8(p);
1713 struct_cv = ceph_decode_8(p);
1714 if (struct_cv > 1) {
1715 pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n",
1716 struct_v, struct_cv);
1717 goto e_inval;
1718 }
1719 len = ceph_decode_32(p);
1720 ceph_decode_need(p, end, len, e_inval);
1721 struct_end = *p + len;
1722
1723 ret = ceph_oloc_decode(p, end, &redir->oloc);
1724 if (ret)
1725 goto out;
1726
1727 len = ceph_decode_32(p);
1728 if (len > 0) {
1729 pr_warn("ceph_request_redirect::object_name is set\n");
1730 goto e_inval;
1731 }
1732
1733 len = ceph_decode_32(p);
1734 *p += len; /* skip osd_instructions */
1735
1736 /* skip the rest */
1737 *p = struct_end;
1738 out:
1739 return ret;
1740
1741 e_inval:
1742 ret = -EINVAL;
1743 goto out;
1744 }
1745
1746 static void complete_request(struct ceph_osd_request *req)
1747 {
1748 complete_all(&req->r_safe_completion); /* fsync waiter */
1749 }
1750
1751 /*
1752 * handle osd op reply. either call the callback if it is specified,
1753 * or do the completion to wake up the waiting thread.
1754 */
1755 static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1756 {
1757 void *p, *end;
1758 struct ceph_osd_request *req;
1759 struct ceph_request_redirect redir;
1760 u64 tid;
1761 int object_len;
1762 unsigned int numops;
1763 int payload_len, flags;
1764 s32 result;
1765 s32 retry_attempt;
1766 struct ceph_pg pg;
1767 int err;
1768 u32 reassert_epoch;
1769 u64 reassert_version;
1770 u32 osdmap_epoch;
1771 int already_completed;
1772 u32 bytes;
1773 unsigned int i;
1774
1775 tid = le64_to_cpu(msg->hdr.tid);
1776 dout("handle_reply %p tid %llu\n", msg, tid);
1777
1778 p = msg->front.iov_base;
1779 end = p + msg->front.iov_len;
1780
1781 ceph_decode_need(&p, end, 4, bad);
1782 object_len = ceph_decode_32(&p);
1783 ceph_decode_need(&p, end, object_len, bad);
1784 p += object_len;
1785
1786 err = ceph_decode_pgid(&p, end, &pg);
1787 if (err)
1788 goto bad;
1789
1790 ceph_decode_need(&p, end, 8 + 4 + 4 + 8 + 4, bad);
1791 flags = ceph_decode_64(&p);
1792 result = ceph_decode_32(&p);
1793 reassert_epoch = ceph_decode_32(&p);
1794 reassert_version = ceph_decode_64(&p);
1795 osdmap_epoch = ceph_decode_32(&p);
1796
1797 /* lookup */
1798 down_read(&osdc->map_sem);
1799 mutex_lock(&osdc->request_mutex);
1800 req = __lookup_request(osdc, tid);
1801 if (req == NULL) {
1802 dout("handle_reply tid %llu dne\n", tid);
1803 goto bad_mutex;
1804 }
1805 ceph_osdc_get_request(req);
1806
1807 dout("handle_reply %p tid %llu req %p result %d\n", msg, tid,
1808 req, result);
1809
1810 ceph_decode_need(&p, end, 4, bad_put);
1811 numops = ceph_decode_32(&p);
1812 if (numops > CEPH_OSD_MAX_OP)
1813 goto bad_put;
1814 if (numops != req->r_num_ops)
1815 goto bad_put;
1816 payload_len = 0;
1817 ceph_decode_need(&p, end, numops * sizeof(struct ceph_osd_op), bad_put);
1818 for (i = 0; i < numops; i++) {
1819 struct ceph_osd_op *op = p;
1820 int len;
1821
1822 len = le32_to_cpu(op->payload_len);
1823 req->r_reply_op_len[i] = len;
1824 dout(" op %d has %d bytes\n", i, len);
1825 payload_len += len;
1826 p += sizeof(*op);
1827 }
1828 bytes = le32_to_cpu(msg->hdr.data_len);
1829 if (payload_len != bytes) {
1830 pr_warn("sum of op payload lens %d != data_len %d\n",
1831 payload_len, bytes);
1832 goto bad_put;
1833 }
1834
1835 ceph_decode_need(&p, end, 4 + numops * 4, bad_put);
1836 retry_attempt = ceph_decode_32(&p);
1837 for (i = 0; i < numops; i++)
1838 req->r_reply_op_result[i] = ceph_decode_32(&p);
1839
1840 if (le16_to_cpu(msg->hdr.version) >= 6) {
1841 p += 8 + 4; /* skip replay_version */
1842 p += 8; /* skip user_version */
1843
1844 err = ceph_redirect_decode(&p, end, &redir);
1845 if (err)
1846 goto bad_put;
1847 } else {
1848 redir.oloc.pool = -1;
1849 }
1850
1851 if (redir.oloc.pool != -1) {
1852 dout("redirect pool %lld\n", redir.oloc.pool);
1853
1854 __unregister_request(osdc, req);
1855
1856 req->r_target_oloc = redir.oloc; /* struct */
1857
1858 /*
1859 * Start redirect requests with nofail=true. If
1860 * mapping fails, request will end up on the notarget
1861 * list, waiting for the new osdmap (which can take
1862 * a while), even though the original request mapped
1863 * successfully. In the future we might want to follow
1864 * original request's nofail setting here.
1865 */
1866 err = __ceph_osdc_start_request(osdc, req, true);
1867 BUG_ON(err);
1868
1869 goto out_unlock;
1870 }
1871
1872 already_completed = req->r_got_reply;
1873 if (!req->r_got_reply) {
1874 req->r_result = result;
1875 dout("handle_reply result %d bytes %d\n", req->r_result,
1876 bytes);
1877 if (req->r_result == 0)
1878 req->r_result = bytes;
1879
1880 /* in case this is a write and we need to replay, */
1881 req->r_reassert_version.epoch = cpu_to_le32(reassert_epoch);
1882 req->r_reassert_version.version = cpu_to_le64(reassert_version);
1883
1884 req->r_got_reply = 1;
1885 } else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) {
1886 dout("handle_reply tid %llu dup ack\n", tid);
1887 goto out_unlock;
1888 }
1889
1890 dout("handle_reply tid %llu flags %d\n", tid, flags);
1891
1892 if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK))
1893 __register_linger_request(osdc, req);
1894
1895 /* either this is a read, or we got the safe response */
1896 if (result < 0 ||
1897 (flags & CEPH_OSD_FLAG_ONDISK) ||
1898 ((flags & CEPH_OSD_FLAG_WRITE) == 0))
1899 __unregister_request(osdc, req);
1900
1901 mutex_unlock(&osdc->request_mutex);
1902 up_read(&osdc->map_sem);
1903
1904 if (!already_completed) {
1905 if (req->r_unsafe_callback &&
1906 result >= 0 && !(flags & CEPH_OSD_FLAG_ONDISK))
1907 req->r_unsafe_callback(req, true);
1908 if (req->r_callback)
1909 req->r_callback(req, msg);
1910 else
1911 complete_all(&req->r_completion);
1912 }
1913
1914 if (flags & CEPH_OSD_FLAG_ONDISK) {
1915 if (req->r_unsafe_callback && already_completed)
1916 req->r_unsafe_callback(req, false);
1917 complete_request(req);
1918 }
1919
1920 out:
1921 dout("req=%p req->r_linger=%d\n", req, req->r_linger);
1922 ceph_osdc_put_request(req);
1923 return;
1924 out_unlock:
1925 mutex_unlock(&osdc->request_mutex);
1926 up_read(&osdc->map_sem);
1927 goto out;
1928
1929 bad_put:
1930 req->r_result = -EIO;
1931 __unregister_request(osdc, req);
1932 if (req->r_callback)
1933 req->r_callback(req, msg);
1934 else
1935 complete_all(&req->r_completion);
1936 complete_request(req);
1937 ceph_osdc_put_request(req);
1938 bad_mutex:
1939 mutex_unlock(&osdc->request_mutex);
1940 up_read(&osdc->map_sem);
1941 bad:
1942 pr_err("corrupt osd_op_reply got %d %d\n",
1943 (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len));
1944 ceph_msg_dump(msg);
1945 }
1946
1947 static void reset_changed_osds(struct ceph_osd_client *osdc)
1948 {
1949 struct rb_node *p, *n;
1950
1951 dout("%s %p\n", __func__, osdc);
1952 for (p = rb_first(&osdc->osds); p; p = n) {
1953 struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
1954
1955 n = rb_next(p);
1956 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
1957 memcmp(&osd->o_con.peer_addr,
1958 ceph_osd_addr(osdc->osdmap,
1959 osd->o_osd),
1960 sizeof(struct ceph_entity_addr)) != 0)
1961 __reset_osd(osdc, osd);
1962 }
1963 }
1964
1965 /*
1966 * Requeue requests whose mapping to an OSD has changed. If requests map to
1967 * no osd, request a new map.
1968 *
1969 * Caller should hold map_sem for read.
1970 */
1971 static void kick_requests(struct ceph_osd_client *osdc, bool force_resend,
1972 bool force_resend_writes)
1973 {
1974 struct ceph_osd_request *req, *nreq;
1975 struct rb_node *p;
1976 int needmap = 0;
1977 int err;
1978 bool force_resend_req;
1979
1980 dout("kick_requests %s %s\n", force_resend ? " (force resend)" : "",
1981 force_resend_writes ? " (force resend writes)" : "");
1982 mutex_lock(&osdc->request_mutex);
1983 for (p = rb_first(&osdc->requests); p; ) {
1984 req = rb_entry(p, struct ceph_osd_request, r_node);
1985 p = rb_next(p);
1986
1987 /*
1988 * For linger requests that have not yet been
1989 * registered, move them to the linger list; they'll
1990 * be sent to the osd in the loop below. Unregister
1991 * the request before re-registering it as a linger
1992 * request to ensure the __map_request() below
1993 * will decide it needs to be sent.
1994 */
1995 if (req->r_linger && list_empty(&req->r_linger_item)) {
1996 dout("%p tid %llu restart on osd%d\n",
1997 req, req->r_tid,
1998 req->r_osd ? req->r_osd->o_osd : -1);
1999 ceph_osdc_get_request(req);
2000 __unregister_request(osdc, req);
2001 __register_linger_request(osdc, req);
2002 ceph_osdc_put_request(req);
2003 continue;
2004 }
2005
2006 force_resend_req = force_resend ||
2007 (force_resend_writes &&
2008 req->r_flags & CEPH_OSD_FLAG_WRITE);
2009 err = __map_request(osdc, req, force_resend_req);
2010 if (err < 0)
2011 continue; /* error */
2012 if (req->r_osd == NULL) {
2013 dout("%p tid %llu maps to no osd\n", req, req->r_tid);
2014 needmap++; /* request a newer map */
2015 } else if (err > 0) {
2016 if (!req->r_linger) {
2017 dout("%p tid %llu requeued on osd%d\n", req,
2018 req->r_tid,
2019 req->r_osd ? req->r_osd->o_osd : -1);
2020 req->r_flags |= CEPH_OSD_FLAG_RETRY;
2021 }
2022 }
2023 }
2024
2025 list_for_each_entry_safe(req, nreq, &osdc->req_linger,
2026 r_linger_item) {
2027 dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
2028
2029 err = __map_request(osdc, req,
2030 force_resend || force_resend_writes);
2031 dout("__map_request returned %d\n", err);
2032 if (err < 0)
2033 continue; /* hrm! */
2034 if (req->r_osd == NULL || err > 0) {
2035 if (req->r_osd == NULL) {
2036 dout("lingering %p tid %llu maps to no osd\n",
2037 req, req->r_tid);
2038 /*
2039 * A homeless lingering request makes
2040 * no sense, as it's job is to keep
2041 * a particular OSD connection open.
2042 * Request a newer map and kick the
2043 * request, knowing that it won't be
2044 * resent until we actually get a map
2045 * that can tell us where to send it.
2046 */
2047 needmap++;
2048 }
2049
2050 dout("kicking lingering %p tid %llu osd%d\n", req,
2051 req->r_tid, req->r_osd ? req->r_osd->o_osd : -1);
2052 __register_request(osdc, req);
2053 __unregister_linger_request(osdc, req);
2054 }
2055 }
2056 reset_changed_osds(osdc);
2057 mutex_unlock(&osdc->request_mutex);
2058
2059 if (needmap) {
2060 dout("%d requests for down osds, need new map\n", needmap);
2061 ceph_monc_request_next_osdmap(&osdc->client->monc);
2062 }
2063 }
2064
2065
2066 /*
2067 * Process updated osd map.
2068 *
2069 * The message contains any number of incremental and full maps, normally
2070 * indicating some sort of topology change in the cluster. Kick requests
2071 * off to different OSDs as needed.
2072 */
2073 void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
2074 {
2075 void *p, *end, *next;
2076 u32 nr_maps, maplen;
2077 u32 epoch;
2078 struct ceph_osdmap *newmap = NULL, *oldmap;
2079 int err;
2080 struct ceph_fsid fsid;
2081 bool was_full;
2082
2083 dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
2084 p = msg->front.iov_base;
2085 end = p + msg->front.iov_len;
2086
2087 /* verify fsid */
2088 ceph_decode_need(&p, end, sizeof(fsid), bad);
2089 ceph_decode_copy(&p, &fsid, sizeof(fsid));
2090 if (ceph_check_fsid(osdc->client, &fsid) < 0)
2091 return;
2092
2093 down_write(&osdc->map_sem);
2094
2095 was_full = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
2096
2097 /* incremental maps */
2098 ceph_decode_32_safe(&p, end, nr_maps, bad);
2099 dout(" %d inc maps\n", nr_maps);
2100 while (nr_maps > 0) {
2101 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
2102 epoch = ceph_decode_32(&p);
2103 maplen = ceph_decode_32(&p);
2104 ceph_decode_need(&p, end, maplen, bad);
2105 next = p + maplen;
2106 if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
2107 dout("applying incremental map %u len %d\n",
2108 epoch, maplen);
2109 newmap = osdmap_apply_incremental(&p, next,
2110 osdc->osdmap,
2111 &osdc->client->msgr);
2112 if (IS_ERR(newmap)) {
2113 err = PTR_ERR(newmap);
2114 goto bad;
2115 }
2116 BUG_ON(!newmap);
2117 if (newmap != osdc->osdmap) {
2118 ceph_osdmap_destroy(osdc->osdmap);
2119 osdc->osdmap = newmap;
2120 }
2121 was_full = was_full ||
2122 ceph_osdmap_flag(osdc->osdmap,
2123 CEPH_OSDMAP_FULL);
2124 kick_requests(osdc, 0, was_full);
2125 } else {
2126 dout("ignoring incremental map %u len %d\n",
2127 epoch, maplen);
2128 }
2129 p = next;
2130 nr_maps--;
2131 }
2132 if (newmap)
2133 goto done;
2134
2135 /* full maps */
2136 ceph_decode_32_safe(&p, end, nr_maps, bad);
2137 dout(" %d full maps\n", nr_maps);
2138 while (nr_maps) {
2139 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
2140 epoch = ceph_decode_32(&p);
2141 maplen = ceph_decode_32(&p);
2142 ceph_decode_need(&p, end, maplen, bad);
2143 if (nr_maps > 1) {
2144 dout("skipping non-latest full map %u len %d\n",
2145 epoch, maplen);
2146 } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
2147 dout("skipping full map %u len %d, "
2148 "older than our %u\n", epoch, maplen,
2149 osdc->osdmap->epoch);
2150 } else {
2151 int skipped_map = 0;
2152
2153 dout("taking full map %u len %d\n", epoch, maplen);
2154 newmap = ceph_osdmap_decode(&p, p+maplen);
2155 if (IS_ERR(newmap)) {
2156 err = PTR_ERR(newmap);
2157 goto bad;
2158 }
2159 BUG_ON(!newmap);
2160 oldmap = osdc->osdmap;
2161 osdc->osdmap = newmap;
2162 if (oldmap) {
2163 if (oldmap->epoch + 1 < newmap->epoch)
2164 skipped_map = 1;
2165 ceph_osdmap_destroy(oldmap);
2166 }
2167 was_full = was_full ||
2168 ceph_osdmap_flag(osdc->osdmap,
2169 CEPH_OSDMAP_FULL);
2170 kick_requests(osdc, skipped_map, was_full);
2171 }
2172 p += maplen;
2173 nr_maps--;
2174 }
2175
2176 if (!osdc->osdmap)
2177 goto bad;
2178 done:
2179 downgrade_write(&osdc->map_sem);
2180 ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch);
2181
2182 /*
2183 * subscribe to subsequent osdmap updates if full to ensure
2184 * we find out when we are no longer full and stop returning
2185 * ENOSPC.
2186 */
2187 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) ||
2188 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD) ||
2189 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR))
2190 ceph_monc_request_next_osdmap(&osdc->client->monc);
2191
2192 mutex_lock(&osdc->request_mutex);
2193 __send_queued(osdc);
2194 mutex_unlock(&osdc->request_mutex);
2195 up_read(&osdc->map_sem);
2196 wake_up_all(&osdc->client->auth_wq);
2197 return;
2198
2199 bad:
2200 pr_err("osdc handle_map corrupt msg\n");
2201 ceph_msg_dump(msg);
2202 up_write(&osdc->map_sem);
2203 }
2204
2205 /*
2206 * watch/notify callback event infrastructure
2207 *
2208 * These callbacks are used both for watch and notify operations.
2209 */
2210 static void __release_event(struct kref *kref)
2211 {
2212 struct ceph_osd_event *event =
2213 container_of(kref, struct ceph_osd_event, kref);
2214
2215 dout("__release_event %p\n", event);
2216 kfree(event);
2217 }
2218
2219 static void get_event(struct ceph_osd_event *event)
2220 {
2221 kref_get(&event->kref);
2222 }
2223
2224 void ceph_osdc_put_event(struct ceph_osd_event *event)
2225 {
2226 kref_put(&event->kref, __release_event);
2227 }
2228 EXPORT_SYMBOL(ceph_osdc_put_event);
2229
2230 static void __insert_event(struct ceph_osd_client *osdc,
2231 struct ceph_osd_event *new)
2232 {
2233 struct rb_node **p = &osdc->event_tree.rb_node;
2234 struct rb_node *parent = NULL;
2235 struct ceph_osd_event *event = NULL;
2236
2237 while (*p) {
2238 parent = *p;
2239 event = rb_entry(parent, struct ceph_osd_event, node);
2240 if (new->cookie < event->cookie)
2241 p = &(*p)->rb_left;
2242 else if (new->cookie > event->cookie)
2243 p = &(*p)->rb_right;
2244 else
2245 BUG();
2246 }
2247
2248 rb_link_node(&new->node, parent, p);
2249 rb_insert_color(&new->node, &osdc->event_tree);
2250 }
2251
2252 static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
2253 u64 cookie)
2254 {
2255 struct rb_node **p = &osdc->event_tree.rb_node;
2256 struct rb_node *parent = NULL;
2257 struct ceph_osd_event *event = NULL;
2258
2259 while (*p) {
2260 parent = *p;
2261 event = rb_entry(parent, struct ceph_osd_event, node);
2262 if (cookie < event->cookie)
2263 p = &(*p)->rb_left;
2264 else if (cookie > event->cookie)
2265 p = &(*p)->rb_right;
2266 else
2267 return event;
2268 }
2269 return NULL;
2270 }
2271
2272 static void __remove_event(struct ceph_osd_event *event)
2273 {
2274 struct ceph_osd_client *osdc = event->osdc;
2275
2276 if (!RB_EMPTY_NODE(&event->node)) {
2277 dout("__remove_event removed %p\n", event);
2278 rb_erase(&event->node, &osdc->event_tree);
2279 ceph_osdc_put_event(event);
2280 } else {
2281 dout("__remove_event didn't remove %p\n", event);
2282 }
2283 }
2284
2285 int ceph_osdc_create_event(struct ceph_osd_client *osdc,
2286 void (*event_cb)(u64, u64, u8, void *),
2287 void *data, struct ceph_osd_event **pevent)
2288 {
2289 struct ceph_osd_event *event;
2290
2291 event = kmalloc(sizeof(*event), GFP_NOIO);
2292 if (!event)
2293 return -ENOMEM;
2294
2295 dout("create_event %p\n", event);
2296 event->cb = event_cb;
2297 event->one_shot = 0;
2298 event->data = data;
2299 event->osdc = osdc;
2300 INIT_LIST_HEAD(&event->osd_node);
2301 RB_CLEAR_NODE(&event->node);
2302 kref_init(&event->kref); /* one ref for us */
2303 kref_get(&event->kref); /* one ref for the caller */
2304
2305 spin_lock(&osdc->event_lock);
2306 event->cookie = ++osdc->event_count;
2307 __insert_event(osdc, event);
2308 spin_unlock(&osdc->event_lock);
2309
2310 *pevent = event;
2311 return 0;
2312 }
2313 EXPORT_SYMBOL(ceph_osdc_create_event);
2314
2315 void ceph_osdc_cancel_event(struct ceph_osd_event *event)
2316 {
2317 struct ceph_osd_client *osdc = event->osdc;
2318
2319 dout("cancel_event %p\n", event);
2320 spin_lock(&osdc->event_lock);
2321 __remove_event(event);
2322 spin_unlock(&osdc->event_lock);
2323 ceph_osdc_put_event(event); /* caller's */
2324 }
2325 EXPORT_SYMBOL(ceph_osdc_cancel_event);
2326
2327
2328 static void do_event_work(struct work_struct *work)
2329 {
2330 struct ceph_osd_event_work *event_work =
2331 container_of(work, struct ceph_osd_event_work, work);
2332 struct ceph_osd_event *event = event_work->event;
2333 u64 ver = event_work->ver;
2334 u64 notify_id = event_work->notify_id;
2335 u8 opcode = event_work->opcode;
2336
2337 dout("do_event_work completing %p\n", event);
2338 event->cb(ver, notify_id, opcode, event->data);
2339 dout("do_event_work completed %p\n", event);
2340 ceph_osdc_put_event(event);
2341 kfree(event_work);
2342 }
2343
2344
2345 /*
2346 * Process osd watch notifications
2347 */
2348 static void handle_watch_notify(struct ceph_osd_client *osdc,
2349 struct ceph_msg *msg)
2350 {
2351 void *p, *end;
2352 u8 proto_ver;
2353 u64 cookie, ver, notify_id;
2354 u8 opcode;
2355 struct ceph_osd_event *event;
2356 struct ceph_osd_event_work *event_work;
2357
2358 p = msg->front.iov_base;
2359 end = p + msg->front.iov_len;
2360
2361 ceph_decode_8_safe(&p, end, proto_ver, bad);
2362 ceph_decode_8_safe(&p, end, opcode, bad);
2363 ceph_decode_64_safe(&p, end, cookie, bad);
2364 ceph_decode_64_safe(&p, end, ver, bad);
2365 ceph_decode_64_safe(&p, end, notify_id, bad);
2366
2367 spin_lock(&osdc->event_lock);
2368 event = __find_event(osdc, cookie);
2369 if (event) {
2370 BUG_ON(event->one_shot);
2371 get_event(event);
2372 }
2373 spin_unlock(&osdc->event_lock);
2374 dout("handle_watch_notify cookie %lld ver %lld event %p\n",
2375 cookie, ver, event);
2376 if (event) {
2377 event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
2378 if (!event_work) {
2379 pr_err("couldn't allocate event_work\n");
2380 ceph_osdc_put_event(event);
2381 return;
2382 }
2383 INIT_WORK(&event_work->work, do_event_work);
2384 event_work->event = event;
2385 event_work->ver = ver;
2386 event_work->notify_id = notify_id;
2387 event_work->opcode = opcode;
2388
2389 queue_work(osdc->notify_wq, &event_work->work);
2390 }
2391
2392 return;
2393
2394 bad:
2395 pr_err("osdc handle_watch_notify corrupt msg\n");
2396 }
2397
2398 /*
2399 * build new request AND message
2400 *
2401 */
2402 void ceph_osdc_build_request(struct ceph_osd_request *req, u64 off,
2403 struct ceph_snap_context *snapc, u64 snap_id,
2404 struct timespec *mtime)
2405 {
2406 struct ceph_msg *msg = req->r_request;
2407 void *p;
2408 size_t msg_size;
2409 int flags = req->r_flags;
2410 u64 data_len;
2411 unsigned int i;
2412
2413 req->r_snapid = snap_id;
2414 req->r_snapc = ceph_get_snap_context(snapc);
2415
2416 /* encode request */
2417 msg->hdr.version = cpu_to_le16(4);
2418
2419 p = msg->front.iov_base;
2420 ceph_encode_32(&p, 1); /* client_inc is always 1 */
2421 req->r_request_osdmap_epoch = p;
2422 p += 4;
2423 req->r_request_flags = p;
2424 p += 4;
2425 if (req->r_flags & CEPH_OSD_FLAG_WRITE)
2426 ceph_encode_timespec(p, mtime);
2427 p += sizeof(struct ceph_timespec);
2428 req->r_request_reassert_version = p;
2429 p += sizeof(struct ceph_eversion); /* will get filled in */
2430
2431 /* oloc */
2432 ceph_encode_8(&p, 4);
2433 ceph_encode_8(&p, 4);
2434 ceph_encode_32(&p, 8 + 4 + 4);
2435 req->r_request_pool = p;
2436 p += 8;
2437 ceph_encode_32(&p, -1); /* preferred */
2438 ceph_encode_32(&p, 0); /* key len */
2439
2440 ceph_encode_8(&p, 1);
2441 req->r_request_pgid = p;
2442 p += 8 + 4;
2443 ceph_encode_32(&p, -1); /* preferred */
2444
2445 /* oid */
2446 ceph_encode_32(&p, req->r_base_oid.name_len);
2447 memcpy(p, req->r_base_oid.name, req->r_base_oid.name_len);
2448 dout("oid '%.*s' len %d\n", req->r_base_oid.name_len,
2449 req->r_base_oid.name, req->r_base_oid.name_len);
2450 p += req->r_base_oid.name_len;
2451
2452 /* ops--can imply data */
2453 ceph_encode_16(&p, (u16)req->r_num_ops);
2454 data_len = 0;
2455 for (i = 0; i < req->r_num_ops; i++) {
2456 data_len += osd_req_encode_op(req, p, i);
2457 p += sizeof(struct ceph_osd_op);
2458 }
2459
2460 /* snaps */
2461 ceph_encode_64(&p, req->r_snapid);
2462 ceph_encode_64(&p, req->r_snapc ? req->r_snapc->seq : 0);
2463 ceph_encode_32(&p, req->r_snapc ? req->r_snapc->num_snaps : 0);
2464 if (req->r_snapc) {
2465 for (i = 0; i < snapc->num_snaps; i++) {
2466 ceph_encode_64(&p, req->r_snapc->snaps[i]);
2467 }
2468 }
2469
2470 req->r_request_attempts = p;
2471 p += 4;
2472
2473 /* data */
2474 if (flags & CEPH_OSD_FLAG_WRITE) {
2475 u16 data_off;
2476
2477 /*
2478 * The header "data_off" is a hint to the receiver
2479 * allowing it to align received data into its
2480 * buffers such that there's no need to re-copy
2481 * it before writing it to disk (direct I/O).
2482 */
2483 data_off = (u16) (off & 0xffff);
2484 req->r_request->hdr.data_off = cpu_to_le16(data_off);
2485 }
2486 req->r_request->hdr.data_len = cpu_to_le32(data_len);
2487
2488 BUG_ON(p > msg->front.iov_base + msg->front.iov_len);
2489 msg_size = p - msg->front.iov_base;
2490 msg->front.iov_len = msg_size;
2491 msg->hdr.front_len = cpu_to_le32(msg_size);
2492
2493 dout("build_request msg_size was %d\n", (int)msg_size);
2494 }
2495 EXPORT_SYMBOL(ceph_osdc_build_request);
2496
2497 /*
2498 * Register request, send initial attempt.
2499 */
2500 int ceph_osdc_start_request(struct ceph_osd_client *osdc,
2501 struct ceph_osd_request *req,
2502 bool nofail)
2503 {
2504 int rc;
2505
2506 down_read(&osdc->map_sem);
2507 mutex_lock(&osdc->request_mutex);
2508
2509 rc = __ceph_osdc_start_request(osdc, req, nofail);
2510
2511 mutex_unlock(&osdc->request_mutex);
2512 up_read(&osdc->map_sem);
2513
2514 return rc;
2515 }
2516 EXPORT_SYMBOL(ceph_osdc_start_request);
2517
2518 /*
2519 * Unregister a registered request. The request is not completed (i.e.
2520 * no callbacks or wakeups) - higher layers are supposed to know what
2521 * they are canceling.
2522 */
2523 void ceph_osdc_cancel_request(struct ceph_osd_request *req)
2524 {
2525 struct ceph_osd_client *osdc = req->r_osdc;
2526
2527 mutex_lock(&osdc->request_mutex);
2528 if (req->r_linger)
2529 __unregister_linger_request(osdc, req);
2530 __unregister_request(osdc, req);
2531 mutex_unlock(&osdc->request_mutex);
2532
2533 dout("%s %p tid %llu canceled\n", __func__, req, req->r_tid);
2534 }
2535 EXPORT_SYMBOL(ceph_osdc_cancel_request);
2536
2537 /*
2538 * wait for a request to complete
2539 */
2540 int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
2541 struct ceph_osd_request *req)
2542 {
2543 int rc;
2544
2545 dout("%s %p tid %llu\n", __func__, req, req->r_tid);
2546
2547 rc = wait_for_completion_interruptible(&req->r_completion);
2548 if (rc < 0) {
2549 dout("%s %p tid %llu interrupted\n", __func__, req, req->r_tid);
2550 ceph_osdc_cancel_request(req);
2551 complete_request(req);
2552 return rc;
2553 }
2554
2555 dout("%s %p tid %llu result %d\n", __func__, req, req->r_tid,
2556 req->r_result);
2557 return req->r_result;
2558 }
2559 EXPORT_SYMBOL(ceph_osdc_wait_request);
2560
2561 /*
2562 * sync - wait for all in-flight requests to flush. avoid starvation.
2563 */
2564 void ceph_osdc_sync(struct ceph_osd_client *osdc)
2565 {
2566 struct ceph_osd_request *req;
2567 u64 last_tid, next_tid = 0;
2568
2569 mutex_lock(&osdc->request_mutex);
2570 last_tid = osdc->last_tid;
2571 while (1) {
2572 req = __lookup_request_ge(osdc, next_tid);
2573 if (!req)
2574 break;
2575 if (req->r_tid > last_tid)
2576 break;
2577
2578 next_tid = req->r_tid + 1;
2579 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
2580 continue;
2581
2582 ceph_osdc_get_request(req);
2583 mutex_unlock(&osdc->request_mutex);
2584 dout("sync waiting on tid %llu (last is %llu)\n",
2585 req->r_tid, last_tid);
2586 wait_for_completion(&req->r_safe_completion);
2587 mutex_lock(&osdc->request_mutex);
2588 ceph_osdc_put_request(req);
2589 }
2590 mutex_unlock(&osdc->request_mutex);
2591 dout("sync done (thru tid %llu)\n", last_tid);
2592 }
2593 EXPORT_SYMBOL(ceph_osdc_sync);
2594
2595 /*
2596 * Call all pending notify callbacks - for use after a watch is
2597 * unregistered, to make sure no more callbacks for it will be invoked
2598 */
2599 void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc)
2600 {
2601 flush_workqueue(osdc->notify_wq);
2602 }
2603 EXPORT_SYMBOL(ceph_osdc_flush_notifies);
2604
2605
2606 /*
2607 * init, shutdown
2608 */
2609 int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
2610 {
2611 int err;
2612
2613 dout("init\n");
2614 osdc->client = client;
2615 osdc->osdmap = NULL;
2616 init_rwsem(&osdc->map_sem);
2617 init_completion(&osdc->map_waiters);
2618 osdc->last_requested_map = 0;
2619 mutex_init(&osdc->request_mutex);
2620 osdc->last_tid = 0;
2621 osdc->osds = RB_ROOT;
2622 INIT_LIST_HEAD(&osdc->osd_lru);
2623 osdc->requests = RB_ROOT;
2624 INIT_LIST_HEAD(&osdc->req_lru);
2625 INIT_LIST_HEAD(&osdc->req_unsent);
2626 INIT_LIST_HEAD(&osdc->req_notarget);
2627 INIT_LIST_HEAD(&osdc->req_linger);
2628 osdc->num_requests = 0;
2629 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
2630 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
2631 spin_lock_init(&osdc->event_lock);
2632 osdc->event_tree = RB_ROOT;
2633 osdc->event_count = 0;
2634
2635 schedule_delayed_work(&osdc->osds_timeout_work,
2636 round_jiffies_relative(osdc->client->options->osd_idle_ttl));
2637
2638 err = -ENOMEM;
2639 osdc->req_mempool = mempool_create_kmalloc_pool(10,
2640 sizeof(struct ceph_osd_request));
2641 if (!osdc->req_mempool)
2642 goto out;
2643
2644 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
2645 OSD_OP_FRONT_LEN, 10, true,
2646 "osd_op");
2647 if (err < 0)
2648 goto out_mempool;
2649 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
2650 OSD_OPREPLY_FRONT_LEN, 10, true,
2651 "osd_op_reply");
2652 if (err < 0)
2653 goto out_msgpool;
2654
2655 err = -ENOMEM;
2656 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
2657 if (!osdc->notify_wq)
2658 goto out_msgpool_reply;
2659
2660 return 0;
2661
2662 out_msgpool_reply:
2663 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
2664 out_msgpool:
2665 ceph_msgpool_destroy(&osdc->msgpool_op);
2666 out_mempool:
2667 mempool_destroy(osdc->req_mempool);
2668 out:
2669 return err;
2670 }
2671
2672 void ceph_osdc_stop(struct ceph_osd_client *osdc)
2673 {
2674 flush_workqueue(osdc->notify_wq);
2675 destroy_workqueue(osdc->notify_wq);
2676 cancel_delayed_work_sync(&osdc->timeout_work);
2677 cancel_delayed_work_sync(&osdc->osds_timeout_work);
2678 if (osdc->osdmap) {
2679 ceph_osdmap_destroy(osdc->osdmap);
2680 osdc->osdmap = NULL;
2681 }
2682 remove_all_osds(osdc);
2683 mempool_destroy(osdc->req_mempool);
2684 ceph_msgpool_destroy(&osdc->msgpool_op);
2685 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
2686 }
2687
2688 /*
2689 * Read some contiguous pages. If we cross a stripe boundary, shorten
2690 * *plen. Return number of bytes read, or error.
2691 */
2692 int ceph_osdc_readpages(struct ceph_osd_client *osdc,
2693 struct ceph_vino vino, struct ceph_file_layout *layout,
2694 u64 off, u64 *plen,
2695 u32 truncate_seq, u64 truncate_size,
2696 struct page **pages, int num_pages, int page_align)
2697 {
2698 struct ceph_osd_request *req;
2699 int rc = 0;
2700
2701 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
2702 vino.snap, off, *plen);
2703 req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 0, 1,
2704 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
2705 NULL, truncate_seq, truncate_size,
2706 false);
2707 if (IS_ERR(req))
2708 return PTR_ERR(req);
2709
2710 /* it may be a short read due to an object boundary */
2711
2712 osd_req_op_extent_osd_data_pages(req, 0,
2713 pages, *plen, page_align, false, false);
2714
2715 dout("readpages final extent is %llu~%llu (%llu bytes align %d)\n",
2716 off, *plen, *plen, page_align);
2717
2718 ceph_osdc_build_request(req, off, NULL, vino.snap, NULL);
2719
2720 rc = ceph_osdc_start_request(osdc, req, false);
2721 if (!rc)
2722 rc = ceph_osdc_wait_request(osdc, req);
2723
2724 ceph_osdc_put_request(req);
2725 dout("readpages result %d\n", rc);
2726 return rc;
2727 }
2728 EXPORT_SYMBOL(ceph_osdc_readpages);
2729
2730 /*
2731 * do a synchronous write on N pages
2732 */
2733 int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
2734 struct ceph_file_layout *layout,
2735 struct ceph_snap_context *snapc,
2736 u64 off, u64 len,
2737 u32 truncate_seq, u64 truncate_size,
2738 struct timespec *mtime,
2739 struct page **pages, int num_pages)
2740 {
2741 struct ceph_osd_request *req;
2742 int rc = 0;
2743 int page_align = off & ~PAGE_MASK;
2744
2745 BUG_ON(vino.snap != CEPH_NOSNAP); /* snapshots aren't writeable */
2746 req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 0, 1,
2747 CEPH_OSD_OP_WRITE,
2748 CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
2749 snapc, truncate_seq, truncate_size,
2750 true);
2751 if (IS_ERR(req))
2752 return PTR_ERR(req);
2753
2754 /* it may be a short write due to an object boundary */
2755 osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align,
2756 false, false);
2757 dout("writepages %llu~%llu (%llu bytes)\n", off, len, len);
2758
2759 ceph_osdc_build_request(req, off, snapc, CEPH_NOSNAP, mtime);
2760
2761 rc = ceph_osdc_start_request(osdc, req, true);
2762 if (!rc)
2763 rc = ceph_osdc_wait_request(osdc, req);
2764
2765 ceph_osdc_put_request(req);
2766 if (rc == 0)
2767 rc = len;
2768 dout("writepages result %d\n", rc);
2769 return rc;
2770 }
2771 EXPORT_SYMBOL(ceph_osdc_writepages);
2772
2773 int ceph_osdc_setup(void)
2774 {
2775 BUG_ON(ceph_osd_request_cache);
2776 ceph_osd_request_cache = kmem_cache_create("ceph_osd_request",
2777 sizeof (struct ceph_osd_request),
2778 __alignof__(struct ceph_osd_request),
2779 0, NULL);
2780
2781 return ceph_osd_request_cache ? 0 : -ENOMEM;
2782 }
2783 EXPORT_SYMBOL(ceph_osdc_setup);
2784
2785 void ceph_osdc_cleanup(void)
2786 {
2787 BUG_ON(!ceph_osd_request_cache);
2788 kmem_cache_destroy(ceph_osd_request_cache);
2789 ceph_osd_request_cache = NULL;
2790 }
2791 EXPORT_SYMBOL(ceph_osdc_cleanup);
2792
2793 /*
2794 * handle incoming message
2795 */
2796 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
2797 {
2798 struct ceph_osd *osd = con->private;
2799 struct ceph_osd_client *osdc;
2800 int type = le16_to_cpu(msg->hdr.type);
2801
2802 if (!osd)
2803 goto out;
2804 osdc = osd->o_osdc;
2805
2806 switch (type) {
2807 case CEPH_MSG_OSD_MAP:
2808 ceph_osdc_handle_map(osdc, msg);
2809 break;
2810 case CEPH_MSG_OSD_OPREPLY:
2811 handle_reply(osdc, msg);
2812 break;
2813 case CEPH_MSG_WATCH_NOTIFY:
2814 handle_watch_notify(osdc, msg);
2815 break;
2816
2817 default:
2818 pr_err("received unknown message type %d %s\n", type,
2819 ceph_msg_type_name(type));
2820 }
2821 out:
2822 ceph_msg_put(msg);
2823 }
2824
2825 /*
2826 * Lookup and return message for incoming reply. Don't try to do
2827 * anything about a larger than preallocated data portion of the
2828 * message at the moment - for now, just skip the message.
2829 */
2830 static struct ceph_msg *get_reply(struct ceph_connection *con,
2831 struct ceph_msg_header *hdr,
2832 int *skip)
2833 {
2834 struct ceph_osd *osd = con->private;
2835 struct ceph_osd_client *osdc = osd->o_osdc;
2836 struct ceph_msg *m;
2837 struct ceph_osd_request *req;
2838 int front_len = le32_to_cpu(hdr->front_len);
2839 int data_len = le32_to_cpu(hdr->data_len);
2840 u64 tid;
2841
2842 tid = le64_to_cpu(hdr->tid);
2843 mutex_lock(&osdc->request_mutex);
2844 req = __lookup_request(osdc, tid);
2845 if (!req) {
2846 pr_warn("%s osd%d tid %llu unknown, skipping\n",
2847 __func__, osd->o_osd, tid);
2848 m = NULL;
2849 *skip = 1;
2850 goto out;
2851 }
2852
2853 if (req->r_reply->con)
2854 dout("%s revoking msg %p from old con %p\n", __func__,
2855 req->r_reply, req->r_reply->con);
2856 ceph_msg_revoke_incoming(req->r_reply);
2857
2858 if (front_len > req->r_reply->front_alloc_len) {
2859 pr_warn("%s osd%d tid %llu front %d > preallocated %d\n",
2860 __func__, osd->o_osd, req->r_tid, front_len,
2861 req->r_reply->front_alloc_len);
2862 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS,
2863 false);
2864 if (!m)
2865 goto out;
2866 ceph_msg_put(req->r_reply);
2867 req->r_reply = m;
2868 }
2869
2870 if (data_len > req->r_reply->data_length) {
2871 pr_warn("%s osd%d tid %llu data %d > preallocated %zu, skipping\n",
2872 __func__, osd->o_osd, req->r_tid, data_len,
2873 req->r_reply->data_length);
2874 m = NULL;
2875 *skip = 1;
2876 goto out;
2877 }
2878
2879 m = ceph_msg_get(req->r_reply);
2880 dout("get_reply tid %lld %p\n", tid, m);
2881
2882 out:
2883 mutex_unlock(&osdc->request_mutex);
2884 return m;
2885 }
2886
2887 static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2888 struct ceph_msg_header *hdr,
2889 int *skip)
2890 {
2891 struct ceph_osd *osd = con->private;
2892 int type = le16_to_cpu(hdr->type);
2893 int front = le32_to_cpu(hdr->front_len);
2894
2895 *skip = 0;
2896 switch (type) {
2897 case CEPH_MSG_OSD_MAP:
2898 case CEPH_MSG_WATCH_NOTIFY:
2899 return ceph_msg_new(type, front, GFP_NOFS, false);
2900 case CEPH_MSG_OSD_OPREPLY:
2901 return get_reply(con, hdr, skip);
2902 default:
2903 pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
2904 osd->o_osd);
2905 *skip = 1;
2906 return NULL;
2907 }
2908 }
2909
2910 /*
2911 * Wrappers to refcount containing ceph_osd struct
2912 */
2913 static struct ceph_connection *get_osd_con(struct ceph_connection *con)
2914 {
2915 struct ceph_osd *osd = con->private;
2916 if (get_osd(osd))
2917 return con;
2918 return NULL;
2919 }
2920
2921 static void put_osd_con(struct ceph_connection *con)
2922 {
2923 struct ceph_osd *osd = con->private;
2924 put_osd(osd);
2925 }
2926
2927 /*
2928 * authentication
2929 */
2930 /*
2931 * Note: returned pointer is the address of a structure that's
2932 * managed separately. Caller must *not* attempt to free it.
2933 */
2934 static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
2935 int *proto, int force_new)
2936 {
2937 struct ceph_osd *o = con->private;
2938 struct ceph_osd_client *osdc = o->o_osdc;
2939 struct ceph_auth_client *ac = osdc->client->monc.auth;
2940 struct ceph_auth_handshake *auth = &o->o_auth;
2941
2942 if (force_new && auth->authorizer) {
2943 ceph_auth_destroy_authorizer(ac, auth->authorizer);
2944 auth->authorizer = NULL;
2945 }
2946 if (!auth->authorizer) {
2947 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2948 auth);
2949 if (ret)
2950 return ERR_PTR(ret);
2951 } else {
2952 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2953 auth);
2954 if (ret)
2955 return ERR_PTR(ret);
2956 }
2957 *proto = ac->protocol;
2958
2959 return auth;
2960 }
2961
2962
2963 static int verify_authorizer_reply(struct ceph_connection *con, int len)
2964 {
2965 struct ceph_osd *o = con->private;
2966 struct ceph_osd_client *osdc = o->o_osdc;
2967 struct ceph_auth_client *ac = osdc->client->monc.auth;
2968
2969 return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer, len);
2970 }
2971
2972 static int invalidate_authorizer(struct ceph_connection *con)
2973 {
2974 struct ceph_osd *o = con->private;
2975 struct ceph_osd_client *osdc = o->o_osdc;
2976 struct ceph_auth_client *ac = osdc->client->monc.auth;
2977
2978 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
2979 return ceph_monc_validate_auth(&osdc->client->monc);
2980 }
2981
2982 static int osd_sign_message(struct ceph_msg *msg)
2983 {
2984 struct ceph_osd *o = msg->con->private;
2985 struct ceph_auth_handshake *auth = &o->o_auth;
2986
2987 return ceph_auth_sign_message(auth, msg);
2988 }
2989
2990 static int osd_check_message_signature(struct ceph_msg *msg)
2991 {
2992 struct ceph_osd *o = msg->con->private;
2993 struct ceph_auth_handshake *auth = &o->o_auth;
2994
2995 return ceph_auth_check_message_signature(auth, msg);
2996 }
2997
2998 static const struct ceph_connection_operations osd_con_ops = {
2999 .get = get_osd_con,
3000 .put = put_osd_con,
3001 .dispatch = dispatch,
3002 .get_authorizer = get_authorizer,
3003 .verify_authorizer_reply = verify_authorizer_reply,
3004 .invalidate_authorizer = invalidate_authorizer,
3005 .alloc_msg = alloc_msg,
3006 .sign_message = osd_sign_message,
3007 .check_message_signature = osd_check_message_signature,
3008 .fault = osd_reset,
3009 };
ubuntu-bionic-kernel mirror