projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
libceph: move schedule_delayed_work() in ceph_osdc_init()
[deliverable/linux.git] / net / ceph / osd_client.c
CommitLineData
a4ce40a9 1
3d14c5d2 2#include <linux/ceph/ceph_debug.h>
f24e9980 3
3d14c5d2 4#include <linux/module.h>
f24e9980
SW		 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>
68b4476b
YS		 11#ifdef CONFIG_BLOCK
12#include <linux/bio.h>
13#endif
f24e9980 14
3d14c5d2
YS		 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>
f24e9980 21
c16e7869 22#define OSD_OPREPLY_FRONT_LEN 512
0d59ab81 23
5522ae0b
AE		 24static struct kmem_cache *ceph_osd_request_cache;
25
9e32789f 26static const struct ceph_connection_operations osd_con_ops;
f24e9980 27
f9d25199 28static void __send_queued(struct ceph_osd_client *osdc);
6f6c7006 29static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd);
a40c4f10
YS		 30static void __register_request(struct ceph_osd_client *osdc,
31 struct ceph_osd_request *req);
2cc6128a
ID		 32static void __unregister_request(struct ceph_osd_client *osdc,
33 struct ceph_osd_request *req);
a40c4f10
YS		 34static void __unregister_linger_request(struct ceph_osd_client *osdc,
35 struct ceph_osd_request *req);
2cc6128a 36static void __enqueue_request(struct ceph_osd_request *req);
f24e9980
SW		 37
38/*
39 * Implement client access to distributed object storage cluster.
40 *
41 * All data objects are stored within a cluster/cloud of OSDs, or
42 * "object storage devices." (Note that Ceph OSDs have _nothing_ to
43 * do with the T10 OSD extensions to SCSI.) Ceph OSDs are simply
44 * remote daemons serving up and coordinating consistent and safe
45 * access to storage.
46 *
47 * Cluster membership and the mapping of data objects onto storage devices
48 * are described by the osd map.
49 *
50 * We keep track of pending OSD requests (read, write), resubmit
51 * requests to different OSDs when the cluster topology/data layout
52 * change, or retry the affected requests when the communications
53 * channel with an OSD is reset.
54 */
55
56/*
57 * calculate the mapping of a file extent onto an object, and fill out the
58 * request accordingly. shorten extent as necessary if it crosses an
59 * object boundary.
60 *
61 * fill osd op in request message.
62 */
dbe0fc41 63static int calc_layout(struct ceph_file_layout *layout, u64 off, u64 *plen,
a19dadfb 64 u64 *objnum, u64 *objoff, u64 *objlen)
f24e9980 65{
60e56f13 66 u64 orig_len = *plen;
d63b77f4 67 int r;
f24e9980 68
60e56f13 69 /* object extent? */
75d1c941
AE		 70 r = ceph_calc_file_object_mapping(layout, off, orig_len, objnum,
71 objoff, objlen);
d63b77f4
SW		 72 if (r < 0)
73 return r;
75d1c941
AE		 74 if (*objlen < orig_len) {
75 *plen = *objlen;
60e56f13
AE		 76 dout(" skipping last %llu, final file extent %llu~%llu\n",
77 orig_len - *plen, off, *plen);
78 }
79
75d1c941 80 dout("calc_layout objnum=%llx %llu~%llu\n", *objnum, *objoff, *objlen);
f24e9980 81
3ff5f385 82 return 0;
f24e9980
SW		 83}
84
c54d47bf
AE		 85static void ceph_osd_data_init(struct ceph_osd_data *osd_data)
86{
87 memset(osd_data, 0, sizeof (*osd_data));
88 osd_data->type = CEPH_OSD_DATA_TYPE_NONE;
89}
90
a4ce40a9 91static void ceph_osd_data_pages_init(struct ceph_osd_data *osd_data,
43bfe5de
AE		 92 struct page **pages, u64 length, u32 alignment,
93 bool pages_from_pool, bool own_pages)
94{
95 osd_data->type = CEPH_OSD_DATA_TYPE_PAGES;
96 osd_data->pages = pages;
97 osd_data->length = length;
98 osd_data->alignment = alignment;
99 osd_data->pages_from_pool = pages_from_pool;
100 osd_data->own_pages = own_pages;
101}
43bfe5de 102
a4ce40a9 103static void ceph_osd_data_pagelist_init(struct ceph_osd_data *osd_data,
43bfe5de
AE		 104 struct ceph_pagelist *pagelist)
105{
106 osd_data->type = CEPH_OSD_DATA_TYPE_PAGELIST;
107 osd_data->pagelist = pagelist;
108}
43bfe5de
AE		 109
110#ifdef CONFIG_BLOCK
a4ce40a9 111static void ceph_osd_data_bio_init(struct ceph_osd_data *osd_data,
43bfe5de
AE		 112 struct bio *bio, size_t bio_length)
113{
114 osd_data->type = CEPH_OSD_DATA_TYPE_BIO;
115 osd_data->bio = bio;
116 osd_data->bio_length = bio_length;
117}
43bfe5de
AE		 118#endif /* CONFIG_BLOCK */
119
8a703a38
IC		 120#define osd_req_op_data(oreq, whch, typ, fld) \
121({ \
122 struct ceph_osd_request *__oreq = (oreq); \
123 unsigned int __whch = (whch); \
124 BUG_ON(__whch >= __oreq->r_num_ops); \
125 &__oreq->r_ops[__whch].typ.fld; \
126})
863c7eb5 127
49719778
AE		 128static struct ceph_osd_data *
129osd_req_op_raw_data_in(struct ceph_osd_request *osd_req, unsigned int which)
130{
131 BUG_ON(which >= osd_req->r_num_ops);
132
133 return &osd_req->r_ops[which].raw_data_in;
134}
135
a4ce40a9
AE		 136struct ceph_osd_data *
137osd_req_op_extent_osd_data(struct ceph_osd_request *osd_req,
406e2c9f 138 unsigned int which)
a4ce40a9 139{
863c7eb5 140 return osd_req_op_data(osd_req, which, extent, osd_data);
a4ce40a9
AE		 141}
142EXPORT_SYMBOL(osd_req_op_extent_osd_data);
143
49719778
AE		 144void osd_req_op_raw_data_in_pages(struct ceph_osd_request *osd_req,
145 unsigned int which, struct page **pages,
146 u64 length, u32 alignment,
147 bool pages_from_pool, bool own_pages)
148{
149 struct ceph_osd_data *osd_data;
150
151 osd_data = osd_req_op_raw_data_in(osd_req, which);
152 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
153 pages_from_pool, own_pages);
154}
155EXPORT_SYMBOL(osd_req_op_raw_data_in_pages);
156
a4ce40a9 157void osd_req_op_extent_osd_data_pages(struct ceph_osd_request *osd_req,
406e2c9f
AE		 158 unsigned int which, struct page **pages,
159 u64 length, u32 alignment,
a4ce40a9
AE		 160 bool pages_from_pool, bool own_pages)
161{
162 struct ceph_osd_data *osd_data;
163
863c7eb5 164 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
a4ce40a9
AE		 165 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
166 pages_from_pool, own_pages);
a4ce40a9
AE		 167}
168EXPORT_SYMBOL(osd_req_op_extent_osd_data_pages);
169
170void osd_req_op_extent_osd_data_pagelist(struct ceph_osd_request *osd_req,
406e2c9f 171 unsigned int which, struct ceph_pagelist *pagelist)
a4ce40a9
AE		 172{
173 struct ceph_osd_data *osd_data;
174
863c7eb5 175 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
a4ce40a9 176 ceph_osd_data_pagelist_init(osd_data, pagelist);
a4ce40a9
AE		 177}
178EXPORT_SYMBOL(osd_req_op_extent_osd_data_pagelist);
179
180#ifdef CONFIG_BLOCK
181void osd_req_op_extent_osd_data_bio(struct ceph_osd_request *osd_req,
406e2c9f 182 unsigned int which, struct bio *bio, size_t bio_length)
a4ce40a9
AE		 183{
184 struct ceph_osd_data *osd_data;
863c7eb5
AE		 185
186 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
a4ce40a9 187 ceph_osd_data_bio_init(osd_data, bio, bio_length);
a4ce40a9
AE		 188}
189EXPORT_SYMBOL(osd_req_op_extent_osd_data_bio);
190#endif /* CONFIG_BLOCK */
191
192static void osd_req_op_cls_request_info_pagelist(
193 struct ceph_osd_request *osd_req,
194 unsigned int which, struct ceph_pagelist *pagelist)
195{
196 struct ceph_osd_data *osd_data;
197
863c7eb5 198 osd_data = osd_req_op_data(osd_req, which, cls, request_info);
a4ce40a9 199 ceph_osd_data_pagelist_init(osd_data, pagelist);
a4ce40a9
AE		 200}
201
04017e29
AE		 202void osd_req_op_cls_request_data_pagelist(
203 struct ceph_osd_request *osd_req,
204 unsigned int which, struct ceph_pagelist *pagelist)
205{
206 struct ceph_osd_data *osd_data;
207
863c7eb5 208 osd_data = osd_req_op_data(osd_req, which, cls, request_data);
04017e29 209 ceph_osd_data_pagelist_init(osd_data, pagelist);
bb873b53
ID		 210 osd_req->r_ops[which].cls.indata_len += pagelist->length;
211 osd_req->r_ops[which].indata_len += pagelist->length;
04017e29
AE		 212}
213EXPORT_SYMBOL(osd_req_op_cls_request_data_pagelist);
214
6c57b554
AE		 215void osd_req_op_cls_request_data_pages(struct ceph_osd_request *osd_req,
216 unsigned int which, struct page **pages, u64 length,
217 u32 alignment, bool pages_from_pool, bool own_pages)
218{
219 struct ceph_osd_data *osd_data;
220
221 osd_data = osd_req_op_data(osd_req, which, cls, request_data);
222 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
223 pages_from_pool, own_pages);
bb873b53
ID		 224 osd_req->r_ops[which].cls.indata_len += length;
225 osd_req->r_ops[which].indata_len += length;
6c57b554
AE		 226}
227EXPORT_SYMBOL(osd_req_op_cls_request_data_pages);
228
a4ce40a9
AE		 229void osd_req_op_cls_response_data_pages(struct ceph_osd_request *osd_req,
230 unsigned int which, struct page **pages, u64 length,
231 u32 alignment, bool pages_from_pool, bool own_pages)
232{
233 struct ceph_osd_data *osd_data;
234
863c7eb5 235 osd_data = osd_req_op_data(osd_req, which, cls, response_data);
a4ce40a9
AE		 236 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
237 pages_from_pool, own_pages);
a4ce40a9
AE		 238}
239EXPORT_SYMBOL(osd_req_op_cls_response_data_pages);
240
23c08a9c
AE		 241static u64 ceph_osd_data_length(struct ceph_osd_data *osd_data)
242{
243 switch (osd_data->type) {
244 case CEPH_OSD_DATA_TYPE_NONE:
245 return 0;
246 case CEPH_OSD_DATA_TYPE_PAGES:
247 return osd_data->length;
248 case CEPH_OSD_DATA_TYPE_PAGELIST:
249 return (u64)osd_data->pagelist->length;
250#ifdef CONFIG_BLOCK
251 case CEPH_OSD_DATA_TYPE_BIO:
252 return (u64)osd_data->bio_length;
253#endif /* CONFIG_BLOCK */
254 default:
255 WARN(true, "unrecognized data type %d\n", (int)osd_data->type);
256 return 0;
257 }
258}
259
c54d47bf
AE		 260static void ceph_osd_data_release(struct ceph_osd_data *osd_data)
261{
5476492f 262 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES && osd_data->own_pages) {
c54d47bf
AE		 263 int num_pages;
264
265 num_pages = calc_pages_for((u64)osd_data->alignment,
266 (u64)osd_data->length);
267 ceph_release_page_vector(osd_data->pages, num_pages);
268 }
5476492f
AE		 269 ceph_osd_data_init(osd_data);
270}
271
272static void osd_req_op_data_release(struct ceph_osd_request *osd_req,
273 unsigned int which)
274{
275 struct ceph_osd_req_op *op;
276
277 BUG_ON(which >= osd_req->r_num_ops);
278 op = &osd_req->r_ops[which];
279
280 switch (op->op) {
281 case CEPH_OSD_OP_READ:
282 case CEPH_OSD_OP_WRITE:
e30b7577 283 case CEPH_OSD_OP_WRITEFULL:
5476492f
AE		 284 ceph_osd_data_release(&op->extent.osd_data);
285 break;
286 case CEPH_OSD_OP_CALL:
287 ceph_osd_data_release(&op->cls.request_info);
04017e29 288 ceph_osd_data_release(&op->cls.request_data);
5476492f
AE		 289 ceph_osd_data_release(&op->cls.response_data);
290 break;
d74b50be
YZ		 291 case CEPH_OSD_OP_SETXATTR:
292 case CEPH_OSD_OP_CMPXATTR:
293 ceph_osd_data_release(&op->xattr.osd_data);
294 break;
66ba609f
YZ		 295 case CEPH_OSD_OP_STAT:
296 ceph_osd_data_release(&op->raw_data_in);
297 break;
5476492f
AE		 298 default:
299 break;
300 }
c54d47bf
AE		 301}
302
63244fa1
ID		 303/*
304 * Assumes @t is zero-initialized.
305 */
306static void target_init(struct ceph_osd_request_target *t)
307{
308 ceph_oid_init(&t->base_oid);
309 ceph_oloc_init(&t->base_oloc);
310 ceph_oid_init(&t->target_oid);
311 ceph_oloc_init(&t->target_oloc);
312
313 ceph_osds_init(&t->acting);
314 ceph_osds_init(&t->up);
315 t->size = -1;
316 t->min_size = -1;
317
318 t->osd = CEPH_HOMELESS_OSD;
319}
320
321static void target_destroy(struct ceph_osd_request_target *t)
322{
323 ceph_oid_destroy(&t->base_oid);
324 ceph_oid_destroy(&t->target_oid);
325}
326
f24e9980
SW		 327/*
328 * requests
329 */
9e94af20 330static void ceph_osdc_release_request(struct kref *kref)
f24e9980 331{
9e94af20
ID		 332 struct ceph_osd_request *req = container_of(kref,
333 struct ceph_osd_request, r_kref);
5476492f 334 unsigned int which;
415e49a9 335
9e94af20
ID		 336 dout("%s %p (r_request %p r_reply %p)\n", __func__, req,
337 req->r_request, req->r_reply);
6562d661
ID		 338 WARN_ON(!RB_EMPTY_NODE(&req->r_node));
339 WARN_ON(!list_empty(&req->r_req_lru_item));
340 WARN_ON(!list_empty(&req->r_osd_item));
341 WARN_ON(!list_empty(&req->r_linger_item));
342 WARN_ON(!list_empty(&req->r_linger_osd_item));
343 WARN_ON(req->r_osd);
9e94af20 344
415e49a9
SW		 345 if (req->r_request)
346 ceph_msg_put(req->r_request);
ace6d3a9 347 if (req->r_reply) {
8921d114 348 ceph_msg_revoke_incoming(req->r_reply);
ab8cb34a 349 ceph_msg_put(req->r_reply);
ace6d3a9 350 }
0fff87ec 351
5476492f
AE		 352 for (which = 0; which < req->r_num_ops; which++)
353 osd_req_op_data_release(req, which);
0fff87ec 354
a66dd383 355 target_destroy(&req->r_t);
415e49a9 356 ceph_put_snap_context(req->r_snapc);
d30291b9 357
415e49a9
SW		 358 if (req->r_mempool)
359 mempool_free(req, req->r_osdc->req_mempool);
3f1af42a 360 else if (req->r_num_ops <= CEPH_OSD_SLAB_OPS)
5522ae0b 361 kmem_cache_free(ceph_osd_request_cache, req);
3f1af42a
ID		 362 else
363 kfree(req);
f24e9980 364}
9e94af20
ID		 365
366void ceph_osdc_get_request(struct ceph_osd_request *req)
367{
368 dout("%s %p (was %d)\n", __func__, req,
369 atomic_read(&req->r_kref.refcount));
370 kref_get(&req->r_kref);
371}
372EXPORT_SYMBOL(ceph_osdc_get_request);
373
374void ceph_osdc_put_request(struct ceph_osd_request *req)
375{
3ed97d63
ID		 376 if (req) {
377 dout("%s %p (was %d)\n", __func__, req,
378 atomic_read(&req->r_kref.refcount));
379 kref_put(&req->r_kref, ceph_osdc_release_request);
380 }
9e94af20
ID		 381}
382EXPORT_SYMBOL(ceph_osdc_put_request);
68b4476b 383
3499e8a5 384struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
f24e9980 385 struct ceph_snap_context *snapc,
1b83bef2 386 unsigned int num_ops,
3499e8a5 387 bool use_mempool,
54a54007 388 gfp_t gfp_flags)
f24e9980
SW		 389{
390 struct ceph_osd_request *req;
1b83bef2 391
f24e9980 392 if (use_mempool) {
3f1af42a 393 BUG_ON(num_ops > CEPH_OSD_SLAB_OPS);
3499e8a5 394 req = mempool_alloc(osdc->req_mempool, gfp_flags);
3f1af42a
ID		 395 } else if (num_ops <= CEPH_OSD_SLAB_OPS) {
396 req = kmem_cache_alloc(ceph_osd_request_cache, gfp_flags);
f24e9980 397 } else {
3f1af42a
ID		 398 BUG_ON(num_ops > CEPH_OSD_MAX_OPS);
399 req = kmalloc(sizeof(*req) + num_ops * sizeof(req->r_ops[0]),
400 gfp_flags);
f24e9980 401 }
3f1af42a 402 if (unlikely(!req))
a79832f2 403 return NULL;
f24e9980 404
3f1af42a
ID		 405 /* req only, each op is zeroed in _osd_req_op_init() */
406 memset(req, 0, sizeof(*req));
407
f24e9980
SW		 408 req->r_osdc = osdc;
409 req->r_mempool = use_mempool;
79528734 410 req->r_num_ops = num_ops;
84127282
ID		 411 req->r_snapid = CEPH_NOSNAP;
412 req->r_snapc = ceph_get_snap_context(snapc);
68b4476b 413
415e49a9 414 kref_init(&req->r_kref);
f24e9980
SW		 415 init_completion(&req->r_completion);
416 init_completion(&req->r_safe_completion);
a978fa20 417 RB_CLEAR_NODE(&req->r_node);
f24e9980 418 INIT_LIST_HEAD(&req->r_unsafe_item);
a40c4f10 419 INIT_LIST_HEAD(&req->r_linger_item);
1d0326b1 420 INIT_LIST_HEAD(&req->r_linger_osd_item);
935b639a 421 INIT_LIST_HEAD(&req->r_req_lru_item);
cd43045c
SW		 422 INIT_LIST_HEAD(&req->r_osd_item);
423
a66dd383 424 target_init(&req->r_t);
22116525 425
13d1ad16
ID		 426 dout("%s req %p\n", __func__, req);
427 return req;
428}
429EXPORT_SYMBOL(ceph_osdc_alloc_request);
3f1af42a 430
13d1ad16
ID		 431int ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp)
432{
433 struct ceph_osd_client *osdc = req->r_osdc;
434 struct ceph_msg *msg;
435 int msg_size;
c16e7869 436
d30291b9
ID		 437 WARN_ON(ceph_oid_empty(&req->r_base_oid));
438
13d1ad16 439 /* create request message */
ae458f5a
ID		 440 msg_size = 4 + 4 + 4; /* client_inc, osdmap_epoch, flags */
441 msg_size += 4 + 4 + 4 + 8; /* mtime, reassert_version */
442 msg_size += 2 + 4 + 8 + 4 + 4; /* oloc */
443 msg_size += 1 + 8 + 4 + 4; /* pgid */
13d1ad16
ID		 444 msg_size += 4 + req->r_base_oid.name_len; /* oid */
445 msg_size += 2 + req->r_num_ops * sizeof(struct ceph_osd_op);
ae458f5a
ID		 446 msg_size += 8; /* snapid */
447 msg_size += 8; /* snap_seq */
13d1ad16 448 msg_size += 4 + 8 * (req->r_snapc ? req->r_snapc->num_snaps : 0);
ae458f5a
ID		 449 msg_size += 4; /* retry_attempt */
450
13d1ad16 451 if (req->r_mempool)
8f3bc053 452 msg = ceph_msgpool_get(&osdc->msgpool_op, 0);
f24e9980 453 else
13d1ad16
ID		 454 msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp, true);
455 if (!msg)
456 return -ENOMEM;
68b4476b 457
f24e9980 458 memset(msg->front.iov_base, 0, msg->front.iov_len);
3499e8a5 459 req->r_request = msg;
3499e8a5 460
13d1ad16
ID		 461 /* create reply message */
462 msg_size = OSD_OPREPLY_FRONT_LEN;
711da55d
ID		 463 msg_size += req->r_base_oid.name_len;
464 msg_size += req->r_num_ops * sizeof(struct ceph_osd_op);
13d1ad16
ID		 465
466 if (req->r_mempool)
467 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0);
468 else
469 msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, msg_size, gfp, true);
470 if (!msg)
471 return -ENOMEM;
472
473 req->r_reply = msg;
474
475 return 0;
3499e8a5 476}
13d1ad16 477EXPORT_SYMBOL(ceph_osdc_alloc_messages);
3499e8a5 478
a8dd0a37 479static bool osd_req_opcode_valid(u16 opcode)
68b4476b 480{
a8dd0a37 481 switch (opcode) {
70b5bfa3
ID		 482#define GENERATE_CASE(op, opcode, str) case CEPH_OSD_OP_##op: return true;
483__CEPH_FORALL_OSD_OPS(GENERATE_CASE)
484#undef GENERATE_CASE
a8dd0a37
AE		 485 default:
486 return false;
487 }
488}
489
33803f33
AE		 490/*
491 * This is an osd op init function for opcodes that have no data or
492 * other information associated with them. It also serves as a
493 * common init routine for all the other init functions, below.
494 */
c99d2d4a 495static struct ceph_osd_req_op *
49719778 496_osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which,
144cba14 497 u16 opcode, u32 flags)
33803f33 498{
c99d2d4a
AE		 499 struct ceph_osd_req_op *op;
500
501 BUG_ON(which >= osd_req->r_num_ops);
33803f33
AE		 502 BUG_ON(!osd_req_opcode_valid(opcode));
503
c99d2d4a 504 op = &osd_req->r_ops[which];
33803f33 505 memset(op, 0, sizeof (*op));
33803f33 506 op->op = opcode;
144cba14 507 op->flags = flags;
c99d2d4a
AE		 508
509 return op;
33803f33
AE		 510}
511
49719778 512void osd_req_op_init(struct ceph_osd_request *osd_req,
144cba14 513 unsigned int which, u16 opcode, u32 flags)
49719778 514{
144cba14 515 (void)_osd_req_op_init(osd_req, which, opcode, flags);
49719778
AE		 516}
517EXPORT_SYMBOL(osd_req_op_init);
518
c99d2d4a
AE		 519void osd_req_op_extent_init(struct ceph_osd_request *osd_req,
520 unsigned int which, u16 opcode,
33803f33
AE		 521 u64 offset, u64 length,
522 u64 truncate_size, u32 truncate_seq)
523{
144cba14
YZ		 524 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
525 opcode, 0);
33803f33
AE		 526 size_t payload_len = 0;
527
ad7a60de 528 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
e30b7577
ID		 529 opcode != CEPH_OSD_OP_WRITEFULL && opcode != CEPH_OSD_OP_ZERO &&
530 opcode != CEPH_OSD_OP_TRUNCATE);
33803f33 531
33803f33
AE		 532 op->extent.offset = offset;
533 op->extent.length = length;
534 op->extent.truncate_size = truncate_size;
535 op->extent.truncate_seq = truncate_seq;
e30b7577 536 if (opcode == CEPH_OSD_OP_WRITE || opcode == CEPH_OSD_OP_WRITEFULL)
33803f33
AE		 537 payload_len += length;
538
de2aa102 539 op->indata_len = payload_len;
33803f33
AE		 540}
541EXPORT_SYMBOL(osd_req_op_extent_init);
542
c99d2d4a
AE		 543void osd_req_op_extent_update(struct ceph_osd_request *osd_req,
544 unsigned int which, u64 length)
e5975c7c 545{
c99d2d4a
AE		 546 struct ceph_osd_req_op *op;
547 u64 previous;
548
549 BUG_ON(which >= osd_req->r_num_ops);
550 op = &osd_req->r_ops[which];
551 previous = op->extent.length;
e5975c7c
AE		 552
553 if (length == previous)
554 return; /* Nothing to do */
555 BUG_ON(length > previous);
556
557 op->extent.length = length;
de2aa102 558 op->indata_len -= previous - length;
e5975c7c
AE		 559}
560EXPORT_SYMBOL(osd_req_op_extent_update);
561
2c63f49a
YZ		 562void osd_req_op_extent_dup_last(struct ceph_osd_request *osd_req,
563 unsigned int which, u64 offset_inc)
564{
565 struct ceph_osd_req_op *op, *prev_op;
566
567 BUG_ON(which + 1 >= osd_req->r_num_ops);
568
569 prev_op = &osd_req->r_ops[which];
570 op = _osd_req_op_init(osd_req, which + 1, prev_op->op, prev_op->flags);
571 /* dup previous one */
572 op->indata_len = prev_op->indata_len;
573 op->outdata_len = prev_op->outdata_len;
574 op->extent = prev_op->extent;
575 /* adjust offset */
576 op->extent.offset += offset_inc;
577 op->extent.length -= offset_inc;
578
579 if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL)
580 op->indata_len -= offset_inc;
581}
582EXPORT_SYMBOL(osd_req_op_extent_dup_last);
583
c99d2d4a 584void osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
04017e29 585 u16 opcode, const char *class, const char *method)
33803f33 586{
144cba14
YZ		 587 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
588 opcode, 0);
5f562df5 589 struct ceph_pagelist *pagelist;
33803f33
AE		 590 size_t payload_len = 0;
591 size_t size;
592
593 BUG_ON(opcode != CEPH_OSD_OP_CALL);
594
5f562df5
AE		 595 pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
596 BUG_ON(!pagelist);
597 ceph_pagelist_init(pagelist);
598
33803f33
AE		 599 op->cls.class_name = class;
600 size = strlen(class);
601 BUG_ON(size > (size_t) U8_MAX);
602 op->cls.class_len = size;
5f562df5 603 ceph_pagelist_append(pagelist, class, size);
33803f33
AE		 604 payload_len += size;
605
606 op->cls.method_name = method;
607 size = strlen(method);
608 BUG_ON(size > (size_t) U8_MAX);
609 op->cls.method_len = size;
5f562df5 610 ceph_pagelist_append(pagelist, method, size);
33803f33
AE		 611 payload_len += size;
612
a4ce40a9 613 osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist);
5f562df5 614
de2aa102 615 op->indata_len = payload_len;
33803f33
AE		 616}
617EXPORT_SYMBOL(osd_req_op_cls_init);
8c042b0d 618
d74b50be
YZ		 619int osd_req_op_xattr_init(struct ceph_osd_request *osd_req, unsigned int which,
620 u16 opcode, const char *name, const void *value,
621 size_t size, u8 cmp_op, u8 cmp_mode)
622{
144cba14
YZ		 623 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
624 opcode, 0);
d74b50be
YZ		 625 struct ceph_pagelist *pagelist;
626 size_t payload_len;
627
628 BUG_ON(opcode != CEPH_OSD_OP_SETXATTR && opcode != CEPH_OSD_OP_CMPXATTR);
629
630 pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
631 if (!pagelist)
632 return -ENOMEM;
633
634 ceph_pagelist_init(pagelist);
635
636 payload_len = strlen(name);
637 op->xattr.name_len = payload_len;
638 ceph_pagelist_append(pagelist, name, payload_len);
639
640 op->xattr.value_len = size;
641 ceph_pagelist_append(pagelist, value, size);
642 payload_len += size;
643
644 op->xattr.cmp_op = cmp_op;
645 op->xattr.cmp_mode = cmp_mode;
646
647 ceph_osd_data_pagelist_init(&op->xattr.osd_data, pagelist);
de2aa102 648 op->indata_len = payload_len;
d74b50be
YZ		 649 return 0;
650}
651EXPORT_SYMBOL(osd_req_op_xattr_init);
652
c99d2d4a
AE		 653void osd_req_op_watch_init(struct ceph_osd_request *osd_req,
654 unsigned int which, u16 opcode,
33803f33
AE		 655 u64 cookie, u64 version, int flag)
656{
144cba14
YZ		 657 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
658 opcode, 0);
33803f33 659
c99d2d4a 660 BUG_ON(opcode != CEPH_OSD_OP_NOTIFY_ACK && opcode != CEPH_OSD_OP_WATCH);
33803f33
AE		 661
662 op->watch.cookie = cookie;
9ef1ee5a 663 op->watch.ver = version;
33803f33 664 if (opcode == CEPH_OSD_OP_WATCH && flag)
c99d2d4a 665 op->watch.flag = (u8)1;
33803f33
AE		 666}
667EXPORT_SYMBOL(osd_req_op_watch_init);
668
c647b8a8
ID		 669void osd_req_op_alloc_hint_init(struct ceph_osd_request *osd_req,
670 unsigned int which,
671 u64 expected_object_size,
672 u64 expected_write_size)
673{
674 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
144cba14
YZ		 675 CEPH_OSD_OP_SETALLOCHINT,
676 0);
c647b8a8
ID		 677
678 op->alloc_hint.expected_object_size = expected_object_size;
679 op->alloc_hint.expected_write_size = expected_write_size;
680
681 /*
682 * CEPH_OSD_OP_SETALLOCHINT op is advisory and therefore deemed
683 * not worth a feature bit. Set FAILOK per-op flag to make
684 * sure older osds don't trip over an unsupported opcode.
685 */
686 op->flags |= CEPH_OSD_OP_FLAG_FAILOK;
687}
688EXPORT_SYMBOL(osd_req_op_alloc_hint_init);
689
90af3602 690static void ceph_osdc_msg_data_add(struct ceph_msg *msg,
ec9123c5
AE		 691 struct ceph_osd_data *osd_data)
692{
693 u64 length = ceph_osd_data_length(osd_data);
694
695 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
696 BUG_ON(length > (u64) SIZE_MAX);
697 if (length)
90af3602 698 ceph_msg_data_add_pages(msg, osd_data->pages,
ec9123c5
AE		 699 length, osd_data->alignment);
700 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
701 BUG_ON(!length);
90af3602 702 ceph_msg_data_add_pagelist(msg, osd_data->pagelist);
ec9123c5
AE		 703#ifdef CONFIG_BLOCK
704 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
90af3602 705 ceph_msg_data_add_bio(msg, osd_data->bio, length);
ec9123c5
AE		 706#endif
707 } else {
708 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
709 }
710}
711
bb873b53
ID		 712static u32 osd_req_encode_op(struct ceph_osd_op *dst,
713 const struct ceph_osd_req_op *src)
a8dd0a37 714{
a8dd0a37
AE		 715 if (WARN_ON(!osd_req_opcode_valid(src->op))) {
716 pr_err("unrecognized osd opcode %d\n", src->op);
717
718 return 0;
719 }
720
721 switch (src->op) {
722 case CEPH_OSD_OP_STAT:
723 break;
724 case CEPH_OSD_OP_READ:
725 case CEPH_OSD_OP_WRITE:
e30b7577 726 case CEPH_OSD_OP_WRITEFULL:
ad7a60de 727 case CEPH_OSD_OP_ZERO:
ad7a60de 728 case CEPH_OSD_OP_TRUNCATE:
a8dd0a37
AE		 729 dst->extent.offset = cpu_to_le64(src->extent.offset);
730 dst->extent.length = cpu_to_le64(src->extent.length);
731 dst->extent.truncate_size =
732 cpu_to_le64(src->extent.truncate_size);
733 dst->extent.truncate_seq =
734 cpu_to_le32(src->extent.truncate_seq);
735 break;
736 case CEPH_OSD_OP_CALL:
a8dd0a37
AE		 737 dst->cls.class_len = src->cls.class_len;
738 dst->cls.method_len = src->cls.method_len;
bb873b53 739 dst->cls.indata_len = cpu_to_le32(src->cls.indata_len);
a8dd0a37
AE		 740 break;
741 case CEPH_OSD_OP_STARTSYNC:
742 break;
743 case CEPH_OSD_OP_NOTIFY_ACK:
744 case CEPH_OSD_OP_WATCH:
745 dst->watch.cookie = cpu_to_le64(src->watch.cookie);
746 dst->watch.ver = cpu_to_le64(src->watch.ver);
747 dst->watch.flag = src->watch.flag;
748 break;
c647b8a8
ID		 749 case CEPH_OSD_OP_SETALLOCHINT:
750 dst->alloc_hint.expected_object_size =
751 cpu_to_le64(src->alloc_hint.expected_object_size);
752 dst->alloc_hint.expected_write_size =
753 cpu_to_le64(src->alloc_hint.expected_write_size);
754 break;
d74b50be
YZ		 755 case CEPH_OSD_OP_SETXATTR:
756 case CEPH_OSD_OP_CMPXATTR:
757 dst->xattr.name_len = cpu_to_le32(src->xattr.name_len);
758 dst->xattr.value_len = cpu_to_le32(src->xattr.value_len);
759 dst->xattr.cmp_op = src->xattr.cmp_op;
760 dst->xattr.cmp_mode = src->xattr.cmp_mode;
d74b50be 761 break;
864e9197
YZ		 762 case CEPH_OSD_OP_CREATE:
763 case CEPH_OSD_OP_DELETE:
764 break;
a8dd0a37 765 default:
4c46459c 766 pr_err("unsupported osd opcode %s\n",
8f63ca2d 767 ceph_osd_op_name(src->op));
4c46459c 768 WARN_ON(1);
a8dd0a37
AE		 769
770 return 0;
68b4476b 771 }
7b25bf5f 772
a8dd0a37 773 dst->op = cpu_to_le16(src->op);
7b25bf5f 774 dst->flags = cpu_to_le32(src->flags);
de2aa102 775 dst->payload_len = cpu_to_le32(src->indata_len);
175face2 776
bb873b53 777 return src->indata_len;
68b4476b
YS		 778}
779
3499e8a5
YS		 780/*
781 * build new request AND message, calculate layout, and adjust file
782 * extent as needed.
783 *
784 * if the file was recently truncated, we include information about its
785 * old and new size so that the object can be updated appropriately. (we
786 * avoid synchronously deleting truncated objects because it's slow.)
787 *
788 * if @do_sync, include a 'startsync' command so that the osd will flush
789 * data quickly.
790 */
791struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
792 struct ceph_file_layout *layout,
793 struct ceph_vino vino,
715e4cd4
YZ		 794 u64 off, u64 *plen,
795 unsigned int which, int num_ops,
3499e8a5
YS		 796 int opcode, int flags,
797 struct ceph_snap_context *snapc,
3499e8a5
YS		 798 u32 truncate_seq,
799 u64 truncate_size,
153e5167 800 bool use_mempool)
3499e8a5 801{
68b4476b 802 struct ceph_osd_request *req;
75d1c941
AE		 803 u64 objnum = 0;
804 u64 objoff = 0;
805 u64 objlen = 0;
6816282d 806 int r;
68b4476b 807
ad7a60de 808 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
864e9197
YZ		 809 opcode != CEPH_OSD_OP_ZERO && opcode != CEPH_OSD_OP_TRUNCATE &&
810 opcode != CEPH_OSD_OP_CREATE && opcode != CEPH_OSD_OP_DELETE);
68b4476b 811
acead002 812 req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool,
ae7ca4a3 813 GFP_NOFS);
13d1ad16
ID		 814 if (!req) {
815 r = -ENOMEM;
816 goto fail;
817 }
79528734 818
3499e8a5 819 /* calculate max write size */
a19dadfb 820 r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
13d1ad16
ID		 821 if (r)
822 goto fail;
a19dadfb 823
864e9197 824 if (opcode == CEPH_OSD_OP_CREATE || opcode == CEPH_OSD_OP_DELETE) {
144cba14 825 osd_req_op_init(req, which, opcode, 0);
864e9197
YZ		 826 } else {
827 u32 object_size = le32_to_cpu(layout->fl_object_size);
828 u32 object_base = off - objoff;
829 if (!(truncate_seq == 1 && truncate_size == -1ULL)) {
830 if (truncate_size <= object_base) {
831 truncate_size = 0;
832 } else {
833 truncate_size -= object_base;
834 if (truncate_size > object_size)
835 truncate_size = object_size;
836 }
ccca4e37 837 }
715e4cd4 838 osd_req_op_extent_init(req, which, opcode, objoff, objlen,
864e9197
YZ		 839 truncate_size, truncate_seq);
840 }
d18d1e28 841
bb873b53 842 req->r_flags = flags;
3c972c95 843 req->r_base_oloc.pool = ceph_file_layout_pg_pool(*layout);
d30291b9 844 ceph_oid_printf(&req->r_base_oid, "%llx.%08llx", vino.ino, objnum);
dbe0fc41 845
bb873b53
ID		 846 req->r_snapid = vino.snap;
847 if (flags & CEPH_OSD_FLAG_WRITE)
848 req->r_data_offset = off;
849
13d1ad16
ID		 850 r = ceph_osdc_alloc_messages(req, GFP_NOFS);
851 if (r)
852 goto fail;
853
f24e9980 854 return req;
13d1ad16
ID		 855
856fail:
857 ceph_osdc_put_request(req);
858 return ERR_PTR(r);
f24e9980 859}
3d14c5d2 860EXPORT_SYMBOL(ceph_osdc_new_request);
f24e9980
SW		 861
862/*
863 * We keep osd requests in an rbtree, sorted by ->r_tid.
864 */
fcd00b68 865DEFINE_RB_FUNCS(request, struct ceph_osd_request, r_tid, r_node)
f24e9980
SW		 866
867static struct ceph_osd_request *
868__lookup_request_ge(struct ceph_osd_client *osdc,
869 u64 tid)
870{
871 struct ceph_osd_request *req;
872 struct rb_node *n = osdc->requests.rb_node;
873
874 while (n) {
875 req = rb_entry(n, struct ceph_osd_request, r_node);
876 if (tid < req->r_tid) {
877 if (!n->rb_left)
878 return req;
879 n = n->rb_left;
880 } else if (tid > req->r_tid) {
881 n = n->rb_right;
882 } else {
883 return req;
884 }
885 }
886 return NULL;
887}
888
2cc6128a
ID		 889static void __kick_linger_request(struct ceph_osd_request *req)
890{
891 struct ceph_osd_client *osdc = req->r_osdc;
892 struct ceph_osd *osd = req->r_osd;
893
894 /*
895 * Linger requests need to be resent with a new tid to avoid
896 * the dup op detection logic on the OSDs. Achieve this with
897 * a re-register dance instead of open-coding.
898 */
899 ceph_osdc_get_request(req);
900 if (!list_empty(&req->r_linger_item))
901 __unregister_linger_request(osdc, req);
902 else
903 __unregister_request(osdc, req);
904 __register_request(osdc, req);
905 ceph_osdc_put_request(req);
906
907 /*
908 * Unless request has been registered as both normal and
909 * lingering, __unregister{,_linger}_request clears r_osd.
910 * However, here we need to preserve r_osd to make sure we
911 * requeue on the same OSD.
912 */
913 WARN_ON(req->r_osd || !osd);
914 req->r_osd = osd;
915
916 dout("%s requeueing %p tid %llu\n", __func__, req, req->r_tid);
917 __enqueue_request(req);
918}
919
6f6c7006
SW		 920/*
921 * Resubmit requests pending on the given osd.
922 */
923static void __kick_osd_requests(struct ceph_osd_client *osdc,
924 struct ceph_osd *osd)
925{
a40c4f10 926 struct ceph_osd_request *req, *nreq;
e02493c0 927 LIST_HEAD(resend);
2cc6128a 928 LIST_HEAD(resend_linger);
6f6c7006
SW		 929 int err;
930
2cc6128a 931 dout("%s osd%d\n", __func__, osd->o_osd);
6f6c7006 932 err = __reset_osd(osdc, osd);
685a7555 933 if (err)
6f6c7006 934 return;
2cc6128a 935
e02493c0
AE		 936 /*
937 * Build up a list of requests to resend by traversing the
938 * osd's list of requests. Requests for a given object are
939 * sent in tid order, and that is also the order they're
940 * kept on this list. Therefore all requests that are in
941 * flight will be found first, followed by all requests that
942 * have not yet been sent. And to resend requests while
943 * preserving this order we will want to put any sent
944 * requests back on the front of the osd client's unsent
945 * list.
946 *
947 * So we build a separate ordered list of already-sent
948 * requests for the affected osd and splice it onto the
949 * front of the osd client's unsent list. Once we've seen a
950 * request that has not yet been sent we're done. Those
951 * requests are already sitting right where they belong.
952 */
6f6c7006 953 list_for_each_entry(req, &osd->o_requests, r_osd_item) {
e02493c0
AE		 954 if (!req->r_sent)
955 break;
2cc6128a
ID		 956
957 if (!req->r_linger) {
958 dout("%s requeueing %p tid %llu\n", __func__, req,
959 req->r_tid);
960 list_move_tail(&req->r_req_lru_item, &resend);
a40c4f10 961 req->r_flags |= CEPH_OSD_FLAG_RETRY;
2cc6128a
ID		 962 } else {
963 list_move_tail(&req->r_req_lru_item, &resend_linger);
964 }
a40c4f10 965 }
e02493c0 966 list_splice(&resend, &osdc->req_unsent);
a40c4f10 967
e02493c0 968 /*
2cc6128a
ID		 969 * Both registered and not yet registered linger requests are
970 * enqueued with a new tid on the same OSD. We add/move them
971 * to req_unsent/o_requests at the end to keep things in tid
972 * order.
e02493c0 973 */
a40c4f10 974 list_for_each_entry_safe(req, nreq, &osd->o_linger_requests,
1d0326b1 975 r_linger_osd_item) {
2cc6128a
ID		 976 WARN_ON(!list_empty(&req->r_req_lru_item));
977 __kick_linger_request(req);
6f6c7006 978 }
2cc6128a
ID		 979
980 list_for_each_entry_safe(req, nreq, &resend_linger, r_req_lru_item)
981 __kick_linger_request(req);
6f6c7006
SW		 982}
983
f24e9980 984/*
81b024e7 985 * If the osd connection drops, we need to resubmit all requests.
f24e9980
SW		 986 */
987static void osd_reset(struct ceph_connection *con)
988{
989 struct ceph_osd *osd = con->private;
990 struct ceph_osd_client *osdc;
991
992 if (!osd)
993 return;
994 dout("osd_reset osd%d\n", osd->o_osd);
995 osdc = osd->o_osdc;
f24e9980 996 down_read(&osdc->map_sem);
83aff95e
SW		 997 mutex_lock(&osdc->request_mutex);
998 __kick_osd_requests(osdc, osd);
f9d25199 999 __send_queued(osdc);
83aff95e 1000 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 1001 up_read(&osdc->map_sem);
1002}
1003
1004/*
1005 * Track open sessions with osds.
1006 */
e10006f8 1007static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
f24e9980
SW		 1008{
1009 struct ceph_osd *osd;
1010
1011 osd = kzalloc(sizeof(*osd), GFP_NOFS);
1012 if (!osd)
1013 return NULL;
1014
1015 atomic_set(&osd->o_ref, 1);
1016 osd->o_osdc = osdc;
e10006f8 1017 osd->o_osd = onum;
f407731d 1018 RB_CLEAR_NODE(&osd->o_node);
f24e9980 1019 INIT_LIST_HEAD(&osd->o_requests);
a40c4f10 1020 INIT_LIST_HEAD(&osd->o_linger_requests);
f5a2041b 1021 INIT_LIST_HEAD(&osd->o_osd_lru);
f24e9980
SW		 1022 osd->o_incarnation = 1;
1023
b7a9e5dd 1024 ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
4e7a5dcd 1025
422d2cb8 1026 INIT_LIST_HEAD(&osd->o_keepalive_item);
f24e9980
SW		 1027 return osd;
1028}
1029
1030static struct ceph_osd *get_osd(struct ceph_osd *osd)
1031{
1032 if (atomic_inc_not_zero(&osd->o_ref)) {
1033 dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
1034 atomic_read(&osd->o_ref));
1035 return osd;
1036 } else {
1037 dout("get_osd %p FAIL\n", osd);
1038 return NULL;
1039 }
1040}
1041
1042static void put_osd(struct ceph_osd *osd)
1043{
1044 dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
1045 atomic_read(&osd->o_ref) - 1);
b28ec2f3 1046 if (atomic_dec_and_test(&osd->o_ref)) {
b28ec2f3 1047 if (osd->o_auth.authorizer)
6c1ea260 1048 ceph_auth_destroy_authorizer(osd->o_auth.authorizer);
f24e9980 1049 kfree(osd);
79494d1b 1050 }
f24e9980
SW		 1051}
1052
fcd00b68
ID		 1053DEFINE_RB_FUNCS(osd, struct ceph_osd, o_osd, o_node)
1054
f24e9980
SW		 1055/*
1056 * remove an osd from our map
1057 */
f5a2041b 1058static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
f24e9980 1059{
7eb71e03 1060 dout("%s %p osd%d\n", __func__, osd, osd->o_osd);
cc9f1f51
ID		 1061 WARN_ON(!list_empty(&osd->o_requests));
1062 WARN_ON(!list_empty(&osd->o_linger_requests));
7c6e6fc5 1063
f5a2041b 1064 list_del_init(&osd->o_osd_lru);
fcd00b68 1065 erase_osd(&osdc->osds, osd);
7eb71e03
ID		 1066}
1067
1068static void remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
1069{
1070 dout("%s %p osd%d\n", __func__, osd, osd->o_osd);
1071
1072 if (!RB_EMPTY_NODE(&osd->o_node)) {
1073 ceph_con_close(&osd->o_con);
1074 __remove_osd(osdc, osd);
1075 put_osd(osd);
1076 }
f24e9980
SW		 1077}
1078
f5a2041b
YS		 1079static void __move_osd_to_lru(struct ceph_osd_client *osdc,
1080 struct ceph_osd *osd)
1081{
bbf37ec3 1082 dout("%s %p\n", __func__, osd);
f5a2041b 1083 BUG_ON(!list_empty(&osd->o_osd_lru));
bbf37ec3 1084
f5a2041b 1085 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
a319bf56 1086 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl;
f5a2041b
YS		 1087}
1088
bbf37ec3
ID		 1089static void maybe_move_osd_to_lru(struct ceph_osd_client *osdc,
1090 struct ceph_osd *osd)
1091{
1092 dout("%s %p\n", __func__, osd);
1093
1094 if (list_empty(&osd->o_requests) &&
1095 list_empty(&osd->o_linger_requests))
1096 __move_osd_to_lru(osdc, osd);
1097}
1098
f5a2041b
YS		 1099static void __remove_osd_from_lru(struct ceph_osd *osd)
1100{
1101 dout("__remove_osd_from_lru %p\n", osd);
1102 if (!list_empty(&osd->o_osd_lru))
1103 list_del_init(&osd->o_osd_lru);
1104}
1105
f24e9980
SW		 1106/*
1107 * reset osd connect
1108 */
f5a2041b 1109static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
f24e9980 1110{
c3acb181 1111 struct ceph_entity_addr *peer_addr;
f24e9980 1112
f5a2041b 1113 dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
a40c4f10
YS		 1114 if (list_empty(&osd->o_requests) &&
1115 list_empty(&osd->o_linger_requests)) {
7eb71e03 1116 remove_osd(osdc, osd);
c3acb181
AE		 1117 return -ENODEV;
1118 }
1119
1120 peer_addr = &osdc->osdmap->osd_addr[osd->o_osd];
1121 if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1122 !ceph_con_opened(&osd->o_con)) {
1123 struct ceph_osd_request *req;
1124
0b4af2e8
ID		 1125 dout("osd addr hasn't changed and connection never opened, "
1126 "letting msgr retry\n");
87b315a5
SW		 1127 /* touch each r_stamp for handle_timeout()'s benfit */
1128 list_for_each_entry(req, &osd->o_requests, r_osd_item)
1129 req->r_stamp = jiffies;
c3acb181
AE		 1130
1131 return -EAGAIN;
f24e9980 1132 }
c3acb181
AE		 1133
1134 ceph_con_close(&osd->o_con);
1135 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1136 osd->o_incarnation++;
1137
1138 return 0;
f24e9980
SW		 1139}
1140
422d2cb8
YS		 1141static void __schedule_osd_timeout(struct ceph_osd_client *osdc)
1142{
1143 schedule_delayed_work(&osdc->timeout_work,
a319bf56 1144 osdc->client->options->osd_keepalive_timeout);
422d2cb8
YS		 1145}
1146
1147static void __cancel_osd_timeout(struct ceph_osd_client *osdc)
1148{
1149 cancel_delayed_work(&osdc->timeout_work);
1150}
f24e9980
SW		 1151
1152/*
1153 * Register request, assign tid. If this is the first request, set up
1154 * the timeout event.
1155 */
a40c4f10
YS		 1156static void __register_request(struct ceph_osd_client *osdc,
1157 struct ceph_osd_request *req)
f24e9980 1158{
f24e9980 1159 req->r_tid = ++osdc->last_tid;
6df058c0 1160 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
77f38e0e 1161 dout("__register_request %p tid %lld\n", req, req->r_tid);
fcd00b68 1162 insert_request(&osdc->requests, req);
f24e9980
SW		 1163 ceph_osdc_get_request(req);
1164 osdc->num_requests++;
f24e9980 1165 if (osdc->num_requests == 1) {
422d2cb8
YS		 1166 dout(" first request, scheduling timeout\n");
1167 __schedule_osd_timeout(osdc);
f24e9980 1168 }
a40c4f10
YS		 1169}
1170
f24e9980
SW		 1171/*
1172 * called under osdc->request_mutex
1173 */
1174static void __unregister_request(struct ceph_osd_client *osdc,
1175 struct ceph_osd_request *req)
1176{
35f9f8a0
SW		 1177 if (RB_EMPTY_NODE(&req->r_node)) {
1178 dout("__unregister_request %p tid %lld not registered\n",
1179 req, req->r_tid);
1180 return;
1181 }
1182
f24e9980 1183 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
fcd00b68 1184 erase_request(&osdc->requests, req);
f24e9980
SW		 1185 osdc->num_requests--;
1186
0ba6478d
SW		 1187 if (req->r_osd) {
1188 /* make sure the original request isn't in flight. */
6740a845 1189 ceph_msg_revoke(req->r_request);
0ba6478d
SW		 1190
1191 list_del_init(&req->r_osd_item);
bbf37ec3 1192 maybe_move_osd_to_lru(osdc, req->r_osd);
4f23409e 1193 if (list_empty(&req->r_linger_osd_item))
a40c4f10 1194 req->r_osd = NULL;
0ba6478d 1195 }
f24e9980 1196
7d5f2481 1197 list_del_init(&req->r_req_lru_item);
f24e9980
SW		 1198 ceph_osdc_put_request(req);
1199
422d2cb8
YS		 1200 if (osdc->num_requests == 0) {
1201 dout(" no requests, canceling timeout\n");
1202 __cancel_osd_timeout(osdc);
f24e9980
SW		 1203 }
1204}
1205
1206/*
1207 * Cancel a previously queued request message
1208 */
1209static void __cancel_request(struct ceph_osd_request *req)
1210{
6bc18876 1211 if (req->r_sent && req->r_osd) {
6740a845 1212 ceph_msg_revoke(req->r_request);
f24e9980
SW		 1213 req->r_sent = 0;
1214 }
1215}
1216
a40c4f10
YS		 1217static void __register_linger_request(struct ceph_osd_client *osdc,
1218 struct ceph_osd_request *req)
1219{
af593064
ID		 1220 dout("%s %p tid %llu\n", __func__, req, req->r_tid);
1221 WARN_ON(!req->r_linger);
1222
96e4dac6 1223 ceph_osdc_get_request(req);
a40c4f10 1224 list_add_tail(&req->r_linger_item, &osdc->req_linger);
6194ea89 1225 if (req->r_osd)
1d0326b1 1226 list_add_tail(&req->r_linger_osd_item,
6194ea89 1227 &req->r_osd->o_linger_requests);
a40c4f10
YS		 1228}
1229
1230static void __unregister_linger_request(struct ceph_osd_client *osdc,
1231 struct ceph_osd_request *req)
1232{
af593064
ID		 1233 WARN_ON(!req->r_linger);
1234
1235 if (list_empty(&req->r_linger_item)) {
1236 dout("%s %p tid %llu not registered\n", __func__, req,
1237 req->r_tid);
1238 return;
1239 }
1240
1241 dout("%s %p tid %llu\n", __func__, req, req->r_tid);
61c74035 1242 list_del_init(&req->r_linger_item);
af593064 1243
a40c4f10 1244 if (req->r_osd) {
1d0326b1 1245 list_del_init(&req->r_linger_osd_item);
bbf37ec3 1246 maybe_move_osd_to_lru(osdc, req->r_osd);
fbdb9190
SW		 1247 if (list_empty(&req->r_osd_item))
1248 req->r_osd = NULL;
a40c4f10 1249 }
96e4dac6 1250 ceph_osdc_put_request(req);
a40c4f10
YS		 1251}
1252
a40c4f10
YS		 1253void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
1254 struct ceph_osd_request *req)
1255{
1256 if (!req->r_linger) {
1257 dout("set_request_linger %p\n", req);
1258 req->r_linger = 1;
a40c4f10
YS		 1259 }
1260}
1261EXPORT_SYMBOL(ceph_osdc_set_request_linger);
1262
63244fa1
ID		 1263static bool __pool_full(struct ceph_pg_pool_info *pi)
1264{
1265 return pi->flags & CEPH_POOL_FLAG_FULL;
1266}
1267
d29adb34
JD		 1268/*
1269 * Returns whether a request should be blocked from being sent
1270 * based on the current osdmap and osd_client settings.
1271 *
1272 * Caller should hold map_sem for read.
1273 */
63244fa1
ID		 1274static bool target_should_be_paused(struct ceph_osd_client *osdc,
1275 const struct ceph_osd_request_target *t,
1276 struct ceph_pg_pool_info *pi)
1277{
1278 bool pauserd = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD);
1279 bool pausewr = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR) ||
1280 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) ||
1281 __pool_full(pi);
1282
1283 WARN_ON(pi->id != t->base_oloc.pool);
1284 return (t->flags & CEPH_OSD_FLAG_READ && pauserd) ||
1285 (t->flags & CEPH_OSD_FLAG_WRITE && pausewr);
1286}
1287
63244fa1
ID		 1288enum calc_target_result {
1289 CALC_TARGET_NO_ACTION = 0,
1290 CALC_TARGET_NEED_RESEND,
1291 CALC_TARGET_POOL_DNE,
1292};
1293
1294static enum calc_target_result calc_target(struct ceph_osd_client *osdc,
1295 struct ceph_osd_request_target *t,
1296 u32 *last_force_resend,
1297 bool any_change)
1298{
1299 struct ceph_pg_pool_info *pi;
1300 struct ceph_pg pgid, last_pgid;
1301 struct ceph_osds up, acting;
1302 bool force_resend = false;
1303 bool need_check_tiering = false;
1304 bool need_resend = false;
1305 bool sort_bitwise = ceph_osdmap_flag(osdc->osdmap,
1306 CEPH_OSDMAP_SORTBITWISE);
1307 enum calc_target_result ct_res;
1308 int ret;
1309
1310 pi = ceph_pg_pool_by_id(osdc->osdmap, t->base_oloc.pool);
1311 if (!pi) {
1312 t->osd = CEPH_HOMELESS_OSD;
1313 ct_res = CALC_TARGET_POOL_DNE;
1314 goto out;
1315 }
1316
1317 if (osdc->osdmap->epoch == pi->last_force_request_resend) {
1318 if (last_force_resend &&
1319 *last_force_resend < pi->last_force_request_resend) {
1320 *last_force_resend = pi->last_force_request_resend;
1321 force_resend = true;
1322 } else if (!last_force_resend) {
1323 force_resend = true;
1324 }
1325 }
1326 if (ceph_oid_empty(&t->target_oid) || force_resend) {
1327 ceph_oid_copy(&t->target_oid, &t->base_oid);
1328 need_check_tiering = true;
1329 }
1330 if (ceph_oloc_empty(&t->target_oloc) || force_resend) {
1331 ceph_oloc_copy(&t->target_oloc, &t->base_oloc);
1332 need_check_tiering = true;
1333 }
1334
1335 if (need_check_tiering &&
1336 (t->flags & CEPH_OSD_FLAG_IGNORE_OVERLAY) == 0) {
1337 if (t->flags & CEPH_OSD_FLAG_READ && pi->read_tier >= 0)
1338 t->target_oloc.pool = pi->read_tier;
1339 if (t->flags & CEPH_OSD_FLAG_WRITE && pi->write_tier >= 0)
1340 t->target_oloc.pool = pi->write_tier;
1341 }
1342
1343 ret = ceph_object_locator_to_pg(osdc->osdmap, &t->target_oid,
1344 &t->target_oloc, &pgid);
1345 if (ret) {
1346 WARN_ON(ret != -ENOENT);
1347 t->osd = CEPH_HOMELESS_OSD;
1348 ct_res = CALC_TARGET_POOL_DNE;
1349 goto out;
1350 }
1351 last_pgid.pool = pgid.pool;
1352 last_pgid.seed = ceph_stable_mod(pgid.seed, t->pg_num, t->pg_num_mask);
1353
1354 ceph_pg_to_up_acting_osds(osdc->osdmap, &pgid, &up, &acting);
1355 if (any_change &&
1356 ceph_is_new_interval(&t->acting,
1357 &acting,
1358 &t->up,
1359 &up,
1360 t->size,
1361 pi->size,
1362 t->min_size,
1363 pi->min_size,
1364 t->pg_num,
1365 pi->pg_num,
1366 t->sort_bitwise,
1367 sort_bitwise,
1368 &last_pgid))
1369 force_resend = true;
1370
1371 if (t->paused && !target_should_be_paused(osdc, t, pi)) {
1372 t->paused = false;
1373 need_resend = true;
1374 }
1375
1376 if (ceph_pg_compare(&t->pgid, &pgid) ||
1377 ceph_osds_changed(&t->acting, &acting, any_change) ||
1378 force_resend) {
1379 t->pgid = pgid; /* struct */
1380 ceph_osds_copy(&t->acting, &acting);
1381 ceph_osds_copy(&t->up, &up);
1382 t->size = pi->size;
1383 t->min_size = pi->min_size;
1384 t->pg_num = pi->pg_num;
1385 t->pg_num_mask = pi->pg_num_mask;
1386 t->sort_bitwise = sort_bitwise;
1387
1388 t->osd = acting.primary;
1389 need_resend = true;
1390 }
1391
1392 ct_res = need_resend ? CALC_TARGET_NEED_RESEND : CALC_TARGET_NO_ACTION;
1393out:
1394 dout("%s t %p -> ct_res %d osd %d\n", __func__, t, ct_res, t->osd);
1395 return ct_res;
1396}
1397
f671b581
ID		 1398static void __enqueue_request(struct ceph_osd_request *req)
1399{
1400 struct ceph_osd_client *osdc = req->r_osdc;
1401
1402 dout("%s %p tid %llu to osd%d\n", __func__, req, req->r_tid,
1403 req->r_osd ? req->r_osd->o_osd : -1);
1404
1405 if (req->r_osd) {
1406 __remove_osd_from_lru(req->r_osd);
1407 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
1408 list_move_tail(&req->r_req_lru_item, &osdc->req_unsent);
1409 } else {
1410 list_move_tail(&req->r_req_lru_item, &osdc->req_notarget);
1411 }
1412}
1413
f24e9980
SW		 1414/*
1415 * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
1416 * (as needed), and set the request r_osd appropriately. If there is
25985edc 1417 * no up osd, set r_osd to NULL. Move the request to the appropriate list
6f6c7006 1418 * (unsent, homeless) or leave on in-flight lru.
f24e9980
SW		 1419 *
1420 * Return 0 if unchanged, 1 if changed, or negative on error.
1421 *
1422 * Caller should hold map_sem for read and request_mutex.
1423 */
6f6c7006 1424static int __map_request(struct ceph_osd_client *osdc,
38d6453c 1425 struct ceph_osd_request *req, int force_resend)
f24e9980 1426{
a66dd383 1427 enum calc_target_result ct_res;
f24e9980 1428 int err;
f24e9980 1429
6f6c7006 1430 dout("map_request %p tid %lld\n", req, req->r_tid);
17a13e40 1431
a66dd383
ID		 1432 ct_res = calc_target(osdc, &req->r_t, NULL, force_resend);
1433 switch (ct_res) {
1434 case CALC_TARGET_POOL_DNE:
6f6c7006 1435 list_move(&req->r_req_lru_item, &osdc->req_notarget);
a66dd383
ID		 1436 return -EIO;
1437 case CALC_TARGET_NO_ACTION:
f24e9980 1438 return 0; /* no change */
a66dd383
ID		 1439 default:
1440 BUG_ON(ct_res != CALC_TARGET_NEED_RESEND);
1441 }
f24e9980 1442
5b191d99 1443 dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n",
a66dd383 1444 req->r_tid, req->r_t.pgid.pool, req->r_t.pgid.seed, req->r_t.osd,
f24e9980
SW		 1445 req->r_osd ? req->r_osd->o_osd : -1);
1446
1447 if (req->r_osd) {
1448 __cancel_request(req);
1449 list_del_init(&req->r_osd_item);
a390de02 1450 list_del_init(&req->r_linger_osd_item);
f24e9980
SW		 1451 req->r_osd = NULL;
1452 }
1453
a66dd383
ID		 1454 req->r_osd = lookup_osd(&osdc->osds, req->r_t.osd);
1455 if (!req->r_osd && req->r_t.osd >= 0) {
c99eb1c7 1456 err = -ENOMEM;
a66dd383 1457 req->r_osd = create_osd(osdc, req->r_t.osd);
6f6c7006
SW		 1458 if (!req->r_osd) {
1459 list_move(&req->r_req_lru_item, &osdc->req_notarget);
c99eb1c7 1460 goto out;
6f6c7006 1461 }
f24e9980 1462
6f3bfd45 1463 dout("map_request osd %p is osd%d\n", req->r_osd,
a66dd383 1464 req->r_osd->o_osd);
fcd00b68 1465 insert_osd(&osdc->osds, req->r_osd);
f24e9980 1466
b7a9e5dd 1467 ceph_con_open(&req->r_osd->o_con,
a66dd383
ID		 1468 CEPH_ENTITY_TYPE_OSD, req->r_osd->o_osd,
1469 &osdc->osdmap->osd_addr[req->r_osd->o_osd]);
f24e9980
SW		 1470 }
1471
f671b581 1472 __enqueue_request(req);
d85b7056 1473 err = 1; /* osd or pg changed */
f24e9980
SW		 1474
1475out:
f24e9980
SW		 1476 return err;
1477}
1478
bb873b53
ID		 1479static void setup_request_data(struct ceph_osd_request *req,
1480 struct ceph_msg *msg)
f24e9980 1481{
bb873b53
ID		 1482 u32 data_len = 0;
1483 int i;
1484
1485 if (!list_empty(&msg->data))
1486 return;
f24e9980 1487
bb873b53
ID		 1488 WARN_ON(msg->data_length);
1489 for (i = 0; i < req->r_num_ops; i++) {
1490 struct ceph_osd_req_op *op = &req->r_ops[i];
1491
1492 switch (op->op) {
1493 /* request */
1494 case CEPH_OSD_OP_WRITE:
1495 case CEPH_OSD_OP_WRITEFULL:
1496 WARN_ON(op->indata_len != op->extent.length);
1497 ceph_osdc_msg_data_add(msg, &op->extent.osd_data);
1498 break;
1499 case CEPH_OSD_OP_SETXATTR:
1500 case CEPH_OSD_OP_CMPXATTR:
1501 WARN_ON(op->indata_len != op->xattr.name_len +
1502 op->xattr.value_len);
1503 ceph_osdc_msg_data_add(msg, &op->xattr.osd_data);
1504 break;
1505
1506 /* reply */
1507 case CEPH_OSD_OP_STAT:
1508 ceph_osdc_msg_data_add(req->r_reply,
1509 &op->raw_data_in);
1510 break;
1511 case CEPH_OSD_OP_READ:
1512 ceph_osdc_msg_data_add(req->r_reply,
1513 &op->extent.osd_data);
1514 break;
1515
1516 /* both */
1517 case CEPH_OSD_OP_CALL:
1518 WARN_ON(op->indata_len != op->cls.class_len +
1519 op->cls.method_len +
1520 op->cls.indata_len);
1521 ceph_osdc_msg_data_add(msg, &op->cls.request_info);
1522 /* optional, can be NONE */
1523 ceph_osdc_msg_data_add(msg, &op->cls.request_data);
1524 /* optional, can be NONE */
1525 ceph_osdc_msg_data_add(req->r_reply,
1526 &op->cls.response_data);
1527 break;
1528 }
1529
1530 data_len += op->indata_len;
1531 }
1b83bef2 1532
bb873b53
ID		 1533 WARN_ON(data_len != msg->data_length);
1534}
1535
1536static void encode_request(struct ceph_osd_request *req, struct ceph_msg *msg)
1537{
1538 void *p = msg->front.iov_base;
1539 void *const end = p + msg->front_alloc_len;
1540 u32 data_len = 0;
1541 int i;
1542
1543 if (req->r_flags & CEPH_OSD_FLAG_WRITE) {
1544 /* snapshots aren't writeable */
1545 WARN_ON(req->r_snapid != CEPH_NOSNAP);
1546 } else {
1547 WARN_ON(req->r_mtime.tv_sec || req->r_mtime.tv_nsec ||
1548 req->r_data_offset || req->r_snapc);
1549 }
1550
1551 setup_request_data(req, msg);
1552
1553 ceph_encode_32(&p, 1); /* client_inc, always 1 */
1554 ceph_encode_32(&p, req->r_osdc->osdmap->epoch);
1555 ceph_encode_32(&p, req->r_flags);
1556 ceph_encode_timespec(p, &req->r_mtime);
1557 p += sizeof(struct ceph_timespec);
1558 /* aka reassert_version */
1559 memcpy(p, &req->r_replay_version, sizeof(req->r_replay_version));
1560 p += sizeof(req->r_replay_version);
1561
1562 /* oloc */
1563 ceph_encode_8(&p, 4);
1564 ceph_encode_8(&p, 4);
1565 ceph_encode_32(&p, 8 + 4 + 4);
1566 ceph_encode_64(&p, req->r_t.target_oloc.pool);
1567 ceph_encode_32(&p, -1); /* preferred */
1568 ceph_encode_32(&p, 0); /* key len */
1569
1570 /* pgid */
1571 ceph_encode_8(&p, 1);
a66dd383
ID		 1572 ceph_encode_64(&p, req->r_t.pgid.pool);
1573 ceph_encode_32(&p, req->r_t.pgid.seed);
bb873b53 1574 ceph_encode_32(&p, -1); /* preferred */
2169aea6 1575
bb873b53
ID		 1576 /* oid */
1577 ceph_encode_32(&p, req->r_t.target_oid.name_len);
1578 memcpy(p, req->r_t.target_oid.name, req->r_t.target_oid.name_len);
1579 p += req->r_t.target_oid.name_len;
f24e9980 1580
bb873b53
ID		 1581 /* ops, can imply data */
1582 ceph_encode_16(&p, req->r_num_ops);
1583 for (i = 0; i < req->r_num_ops; i++) {
1584 data_len += osd_req_encode_op(p, &req->r_ops[i]);
1585 p += sizeof(struct ceph_osd_op);
1586 }
26be8808 1587
bb873b53
ID		 1588 ceph_encode_64(&p, req->r_snapid); /* snapid */
1589 if (req->r_snapc) {
1590 ceph_encode_64(&p, req->r_snapc->seq);
1591 ceph_encode_32(&p, req->r_snapc->num_snaps);
1592 for (i = 0; i < req->r_snapc->num_snaps; i++)
1593 ceph_encode_64(&p, req->r_snapc->snaps[i]);
1594 } else {
1595 ceph_encode_64(&p, 0); /* snap_seq */
1596 ceph_encode_32(&p, 0); /* snaps len */
1597 }
1598
1599 ceph_encode_32(&p, req->r_attempts); /* retry_attempt */
1600
1601 BUG_ON(p > end);
1602 msg->front.iov_len = p - msg->front.iov_base;
1603 msg->hdr.version = cpu_to_le16(4); /* MOSDOp v4 */
1604 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1605 msg->hdr.data_len = cpu_to_le32(data_len);
1606 /*
1607 * The header "data_off" is a hint to the receiver allowing it
1608 * to align received data into its buffers such that there's no
1609 * need to re-copy it before writing it to disk (direct I/O).
1610 */
1611 msg->hdr.data_off = cpu_to_le16(req->r_data_offset);
26be8808 1612
bb873b53
ID		 1613 dout("%s req %p oid %*pE oid_len %d front %zu data %u\n", __func__,
1614 req, req->r_t.target_oid.name_len, req->r_t.target_oid.name,
1615 req->r_t.target_oid.name_len, msg->front.iov_len, data_len);
1616}
1617
1618/*
1619 * @req has to be assigned a tid and registered.
1620 */
1621static void send_request(struct ceph_osd_request *req)
1622{
1623 struct ceph_osd *osd = req->r_osd;
1624
1625 WARN_ON(osd->o_osd != req->r_t.osd);
1626
1627 req->r_flags |= CEPH_OSD_FLAG_KNOWN_REDIR;
1628 if (req->r_attempts)
1629 req->r_flags |= CEPH_OSD_FLAG_RETRY;
1630 else
1631 WARN_ON(req->r_flags & CEPH_OSD_FLAG_RETRY);
1632
1633 encode_request(req, req->r_request);
1634
1635 dout("%s req %p tid %llu to pg %llu.%x osd%d flags 0x%x attempt %d\n",
1636 __func__, req, req->r_tid, req->r_t.pgid.pool, req->r_t.pgid.seed,
1637 req->r_t.osd, req->r_flags, req->r_attempts);
1638
1639 req->r_t.paused = false;
1640 req->r_stamp = jiffies;
1641 req->r_attempts++;
1642
1643 req->r_sent = osd->o_incarnation;
1644 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
1645 ceph_con_send(&osd->o_con, ceph_msg_get(req->r_request));
f24e9980
SW		 1646}
1647
6f6c7006
SW		 1648/*
1649 * Send any requests in the queue (req_unsent).
1650 */
f9d25199 1651static void __send_queued(struct ceph_osd_client *osdc)
6f6c7006
SW		 1652{
1653 struct ceph_osd_request *req, *tmp;
1654
f9d25199 1655 dout("__send_queued\n");
bb873b53
ID		 1656 list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item) {
1657 list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
1658 send_request(req);
1659 }
6f6c7006
SW		 1660}
1661
0bbfdfe8
ID		 1662/*
1663 * Caller should hold map_sem for read and request_mutex.
1664 */
1665static int __ceph_osdc_start_request(struct ceph_osd_client *osdc,
1666 struct ceph_osd_request *req,
1667 bool nofail)
1668{
1669 int rc;
1670
1671 __register_request(osdc, req);
1672 req->r_sent = 0;
1673 req->r_got_reply = 0;
1674 rc = __map_request(osdc, req, 0);
1675 if (rc < 0) {
1676 if (nofail) {
1677 dout("osdc_start_request failed map, "
1678 " will retry %lld\n", req->r_tid);
1679 rc = 0;
1680 } else {
1681 __unregister_request(osdc, req);
1682 }
1683 return rc;
1684 }
1685
1686 if (req->r_osd == NULL) {
1687 dout("send_request %p no up osds in pg\n", req);
1688 ceph_monc_request_next_osdmap(&osdc->client->monc);
1689 } else {
1690 __send_queued(osdc);
1691 }
1692
1693 return 0;
1694}
1695
fe5da05e
ID		 1696static void __complete_request(struct ceph_osd_request *req)
1697{
1698 if (req->r_callback)
1699 req->r_callback(req);
1700 else
1701 complete_all(&req->r_completion);
1702}
1703
f24e9980
SW		 1704/*
1705 * Timeout callback, called every N seconds when 1 or more osd
1706 * requests has been active for more than N seconds. When this
1707 * happens, we ping all OSDs with requests who have timed out to
1708 * ensure any communications channel reset is detected. Reset the
1709 * request timeouts another N seconds in the future as we go.
1710 * Reschedule the timeout event another N seconds in future (unless
1711 * there are no open requests).
1712 */
1713static void handle_timeout(struct work_struct *work)
1714{
1715 struct ceph_osd_client *osdc =
1716 container_of(work, struct ceph_osd_client, timeout_work.work);
a319bf56 1717 struct ceph_options *opts = osdc->client->options;
83aff95e 1718 struct ceph_osd_request *req;
f24e9980 1719 struct ceph_osd *osd;
422d2cb8 1720 struct list_head slow_osds;
f24e9980
SW		 1721 dout("timeout\n");
1722 down_read(&osdc->map_sem);
1723
1724 ceph_monc_request_next_osdmap(&osdc->client->monc);
1725
1726 mutex_lock(&osdc->request_mutex);
f24e9980 1727
422d2cb8
YS		 1728 /*
1729 * ping osds that are a bit slow. this ensures that if there
1730 * is a break in the TCP connection we will notice, and reopen
1731 * a connection with that osd (from the fault callback).
1732 */
1733 INIT_LIST_HEAD(&slow_osds);
1734 list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
a319bf56
ID		 1735 if (time_before(jiffies,
1736 req->r_stamp + opts->osd_keepalive_timeout))
422d2cb8
YS		 1737 break;
1738
1739 osd = req->r_osd;
1740 BUG_ON(!osd);
1741 dout(" tid %llu is slow, will send keepalive on osd%d\n",
f24e9980 1742 req->r_tid, osd->o_osd);
422d2cb8
YS		 1743 list_move_tail(&osd->o_keepalive_item, &slow_osds);
1744 }
1745 while (!list_empty(&slow_osds)) {
1746 osd = list_entry(slow_osds.next, struct ceph_osd,
1747 o_keepalive_item);
1748 list_del_init(&osd->o_keepalive_item);
f24e9980
SW		 1749 ceph_con_keepalive(&osd->o_con);
1750 }
1751
422d2cb8 1752 __schedule_osd_timeout(osdc);
f9d25199 1753 __send_queued(osdc);
f24e9980 1754 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 1755 up_read(&osdc->map_sem);
1756}
1757
f5a2041b
YS		 1758static void handle_osds_timeout(struct work_struct *work)
1759{
1760 struct ceph_osd_client *osdc =
1761 container_of(work, struct ceph_osd_client,
1762 osds_timeout_work.work);
a319bf56 1763 unsigned long delay = osdc->client->options->osd_idle_ttl / 4;
42a2c09f 1764 struct ceph_osd *osd, *nosd;
f5a2041b 1765
42a2c09f 1766 dout("%s osdc %p\n", __func__, osdc);
f5a2041b 1767 down_read(&osdc->map_sem);
42a2c09f
ID		 1768 mutex_lock(&osdc->request_mutex);
1769
1770 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
1771 if (time_before(jiffies, osd->lru_ttl))
1772 break;
1773
1774 remove_osd(osdc, osd);
1775 }
f5a2041b 1776
42a2c09f
ID		 1777 mutex_unlock(&osdc->request_mutex);
1778 up_read(&osdc->map_sem);
f5a2041b
YS		 1779 schedule_delayed_work(&osdc->osds_timeout_work,
1780 round_jiffies_relative(delay));
1781}
1782
205ee118
ID		 1783static int ceph_oloc_decode(void **p, void *end,
1784 struct ceph_object_locator *oloc)
1785{
1786 u8 struct_v, struct_cv;
1787 u32 len;
1788 void *struct_end;
1789 int ret = 0;
1790
1791 ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
1792 struct_v = ceph_decode_8(p);
1793 struct_cv = ceph_decode_8(p);
1794 if (struct_v < 3) {
1795 pr_warn("got v %d < 3 cv %d of ceph_object_locator\n",
1796 struct_v, struct_cv);
1797 goto e_inval;
1798 }
1799 if (struct_cv > 6) {
1800 pr_warn("got v %d cv %d > 6 of ceph_object_locator\n",
1801 struct_v, struct_cv);
1802 goto e_inval;
1803 }
1804 len = ceph_decode_32(p);
1805 ceph_decode_need(p, end, len, e_inval);
1806 struct_end = *p + len;
1807
1808 oloc->pool = ceph_decode_64(p);
1809 *p += 4; /* skip preferred */
1810
1811 len = ceph_decode_32(p);
1812 if (len > 0) {
1813 pr_warn("ceph_object_locator::key is set\n");
1814 goto e_inval;
1815 }
1816
1817 if (struct_v >= 5) {
1818 len = ceph_decode_32(p);
1819 if (len > 0) {
1820 pr_warn("ceph_object_locator::nspace is set\n");
1821 goto e_inval;
1822 }
1823 }
1824
1825 if (struct_v >= 6) {
1826 s64 hash = ceph_decode_64(p);
1827 if (hash != -1) {
1828 pr_warn("ceph_object_locator::hash is set\n");
1829 goto e_inval;
1830 }
1831 }
1832
1833 /* skip the rest */
1834 *p = struct_end;
1835out:
1836 return ret;
1837
1838e_inval:
1839 ret = -EINVAL;
1840 goto out;
1841}
1842
1843static int ceph_redirect_decode(void **p, void *end,
1844 struct ceph_request_redirect *redir)
1845{
1846 u8 struct_v, struct_cv;
1847 u32 len;
1848 void *struct_end;
1849 int ret;
1850
1851 ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
1852 struct_v = ceph_decode_8(p);
1853 struct_cv = ceph_decode_8(p);
1854 if (struct_cv > 1) {
1855 pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n",
1856 struct_v, struct_cv);
1857 goto e_inval;
1858 }
1859 len = ceph_decode_32(p);
1860 ceph_decode_need(p, end, len, e_inval);
1861 struct_end = *p + len;
1862
1863 ret = ceph_oloc_decode(p, end, &redir->oloc);
1864 if (ret)
1865 goto out;
1866
1867 len = ceph_decode_32(p);
1868 if (len > 0) {
1869 pr_warn("ceph_request_redirect::object_name is set\n");
1870 goto e_inval;
1871 }
1872
1873 len = ceph_decode_32(p);
1874 *p += len; /* skip osd_instructions */
1875
1876 /* skip the rest */
1877 *p = struct_end;
1878out:
1879 return ret;
1880
1881e_inval:
1882 ret = -EINVAL;
1883 goto out;
1884}
1885
fe5da05e
ID		 1886struct MOSDOpReply {
1887 struct ceph_pg pgid;
1888 u64 flags;
1889 int result;
1890 u32 epoch;
1891 int num_ops;
1892 u32 outdata_len[CEPH_OSD_MAX_OPS];
1893 s32 rval[CEPH_OSD_MAX_OPS];
1894 int retry_attempt;
1895 struct ceph_eversion replay_version;
1896 u64 user_version;
1897 struct ceph_request_redirect redirect;
1898};
25845472 1899
fe5da05e 1900static int decode_MOSDOpReply(const struct ceph_msg *msg, struct MOSDOpReply *m)
f24e9980 1901{
fe5da05e
ID		 1902 void *p = msg->front.iov_base;
1903 void *const end = p + msg->front.iov_len;
1904 u16 version = le16_to_cpu(msg->hdr.version);
1905 struct ceph_eversion bad_replay_version;
b0b31a8f 1906 u8 decode_redir;
fe5da05e
ID		 1907 u32 len;
1908 int ret;
1909 int i;
1b83bef2 1910
fe5da05e
ID		 1911 ceph_decode_32_safe(&p, end, len, e_inval);
1912 ceph_decode_need(&p, end, len, e_inval);
1913 p += len; /* skip oid */
1b83bef2 1914
fe5da05e
ID		 1915 ret = ceph_decode_pgid(&p, end, &m->pgid);
1916 if (ret)
1917 return ret;
1b83bef2 1918
fe5da05e
ID		 1919 ceph_decode_64_safe(&p, end, m->flags, e_inval);
1920 ceph_decode_32_safe(&p, end, m->result, e_inval);
1921 ceph_decode_need(&p, end, sizeof(bad_replay_version), e_inval);
1922 memcpy(&bad_replay_version, p, sizeof(bad_replay_version));
1923 p += sizeof(bad_replay_version);
1924 ceph_decode_32_safe(&p, end, m->epoch, e_inval);
1b83bef2 1925
fe5da05e
ID		 1926 ceph_decode_32_safe(&p, end, m->num_ops, e_inval);
1927 if (m->num_ops > ARRAY_SIZE(m->outdata_len))
1928 goto e_inval;
1b83bef2 1929
fe5da05e
ID		 1930 ceph_decode_need(&p, end, m->num_ops * sizeof(struct ceph_osd_op),
1931 e_inval);
1932 for (i = 0; i < m->num_ops; i++) {
1b83bef2 1933 struct ceph_osd_op *op = p;
1b83bef2 1934
fe5da05e 1935 m->outdata_len[i] = le32_to_cpu(op->payload_len);
1b83bef2
SW		 1936 p += sizeof(*op);
1937 }
1b83bef2 1938
fe5da05e
ID		 1939 ceph_decode_32_safe(&p, end, m->retry_attempt, e_inval);
1940 for (i = 0; i < m->num_ops; i++)
1941 ceph_decode_32_safe(&p, end, m->rval[i], e_inval);
f24e9980 1942
fe5da05e
ID		 1943 if (version >= 5) {
1944 ceph_decode_need(&p, end, sizeof(m->replay_version), e_inval);
1945 memcpy(&m->replay_version, p, sizeof(m->replay_version));
1946 p += sizeof(m->replay_version);
1947 ceph_decode_64_safe(&p, end, m->user_version, e_inval);
1948 } else {
1949 m->replay_version = bad_replay_version; /* struct */
1950 m->user_version = le64_to_cpu(m->replay_version.version);
1951 }
eb845ff1 1952
fe5da05e
ID		 1953 if (version >= 6) {
1954 if (version >= 7)
1955 ceph_decode_8_safe(&p, end, decode_redir, e_inval);
b0b31a8f
ID		 1956 else
1957 decode_redir = 1;
1958 } else {
1959 decode_redir = 0;
1960 }
1961
1962 if (decode_redir) {
fe5da05e
ID		 1963 ret = ceph_redirect_decode(&p, end, &m->redirect);
1964 if (ret)
1965 return ret;
205ee118 1966 } else {
fe5da05e 1967 ceph_oloc_init(&m->redirect.oloc);
205ee118 1968 }
f24e9980 1969
fe5da05e
ID		 1970 return 0;
1971
1972e_inval:
1973 return -EINVAL;
1974}
1975
1976/*
1977 * We are done with @req if
1978 * - @m is a safe reply, or
1979 * - @m is an unsafe reply and we didn't want a safe one
1980 */
1981static bool done_request(const struct ceph_osd_request *req,
1982 const struct MOSDOpReply *m)
1983{
1984 return (m->result < 0 ||
1985 (m->flags & CEPH_OSD_FLAG_ONDISK) ||
1986 !(req->r_flags & CEPH_OSD_FLAG_ONDISK));
1987}
205ee118 1988
fe5da05e
ID		 1989/*
1990 * handle osd op reply. either call the callback if it is specified,
1991 * or do the completion to wake up the waiting thread.
1992 *
1993 * ->r_unsafe_callback is set? yes no
1994 *
1995 * first reply is OK (needed r_cb/r_completion, r_cb/r_completion,
1996 * any or needed/got safe) r_safe_completion r_safe_completion
1997 *
1998 * first reply is unsafe r_unsafe_cb(true) (nothing)
1999 *
2000 * when we get the safe reply r_unsafe_cb(false), r_cb/r_completion,
2001 * r_safe_completion r_safe_completion
2002 */
2003static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg)
2004{
2005 struct ceph_osd_request *req;
2006 struct MOSDOpReply m;
2007 u64 tid = le64_to_cpu(msg->hdr.tid);
2008 u32 data_len = 0;
2009 bool already_acked;
2010 int ret;
2011 int i;
2012
2013 dout("%s msg %p tid %llu\n", __func__, msg, tid);
2014
2015 down_read(&osdc->map_sem);
2016 mutex_lock(&osdc->request_mutex);
2017 req = lookup_request(&osdc->requests, tid);
2018 if (!req) {
2019 dout("%s no tid %llu\n", __func__, tid);
2020 goto out_unlock;
2021 }
2022 ceph_osdc_get_request(req);
2023
2024 ret = decode_MOSDOpReply(msg, &m);
2025 if (ret) {
2026 pr_err("failed to decode MOSDOpReply for tid %llu: %d\n",
2027 req->r_tid, ret);
2028 ceph_msg_dump(msg);
2029 goto fail_request;
2030 }
2031 dout("%s req %p tid %llu flags 0x%llx pgid %llu.%x epoch %u attempt %d v %u'%llu uv %llu\n",
2032 __func__, req, req->r_tid, m.flags, m.pgid.pool, m.pgid.seed,
2033 m.epoch, m.retry_attempt, le32_to_cpu(m.replay_version.epoch),
2034 le64_to_cpu(m.replay_version.version), m.user_version);
2035
2036 if (m.retry_attempt >= 0) {
2037 if (m.retry_attempt != req->r_attempts - 1) {
2038 dout("req %p tid %llu retry_attempt %d != %d, ignoring\n",
2039 req, req->r_tid, m.retry_attempt,
2040 req->r_attempts - 1);
2041 goto out_put;
2042 }
2043 } else {
2044 WARN_ON(1); /* MOSDOpReply v4 is assumed */
2045 }
2046
2047 if (!ceph_oloc_empty(&m.redirect.oloc)) {
2048 dout("req %p tid %llu redirect pool %lld\n", req, req->r_tid,
2049 m.redirect.oloc.pool);
205ee118 2050 __unregister_request(osdc, req);
205ee118 2051
fe5da05e 2052 ceph_oloc_copy(&req->r_t.target_oloc, &m.redirect.oloc);
205ee118
ID		 2053
2054 /*
2055 * Start redirect requests with nofail=true. If
2056 * mapping fails, request will end up on the notarget
2057 * list, waiting for the new osdmap (which can take
2058 * a while), even though the original request mapped
2059 * successfully. In the future we might want to follow
2060 * original request's nofail setting here.
2061 */
fe5da05e
ID		 2062 ret = __ceph_osdc_start_request(osdc, req, true);
2063 BUG_ON(ret);
205ee118 2064
fe5da05e 2065 goto out_put;
205ee118
ID		 2066 }
2067
fe5da05e
ID		 2068 if (m.num_ops != req->r_num_ops) {
2069 pr_err("num_ops %d != %d for tid %llu\n", m.num_ops,
2070 req->r_num_ops, req->r_tid);
2071 goto fail_request;
f24e9980 2072 }
fe5da05e
ID		 2073 for (i = 0; i < req->r_num_ops; i++) {
2074 dout(" req %p tid %llu op %d rval %d len %u\n", req,
2075 req->r_tid, i, m.rval[i], m.outdata_len[i]);
2076 req->r_ops[i].rval = m.rval[i];
2077 req->r_ops[i].outdata_len = m.outdata_len[i];
2078 data_len += m.outdata_len[i];
2079 }
2080 if (data_len != le32_to_cpu(msg->hdr.data_len)) {
2081 pr_err("sum of lens %u != %u for tid %llu\n", data_len,
2082 le32_to_cpu(msg->hdr.data_len), req->r_tid);
2083 goto fail_request;
2084 }
2085 dout("%s req %p tid %llu acked %d result %d data_len %u\n", __func__,
2086 req, req->r_tid, req->r_got_reply, m.result, data_len);
2087
2088 already_acked = req->r_got_reply;
2089 if (!already_acked) {
2090 req->r_result = m.result ?: data_len;
2091 req->r_replay_version = m.replay_version; /* struct */
2092 req->r_got_reply = true;
2093 } else if (!(m.flags & CEPH_OSD_FLAG_ONDISK)) {
2094 dout("req %p tid %llu dup ack\n", req, req->r_tid);
2095 goto out_put;
2096 }
2097
2098 if (done_request(req, &m)) {
f24e9980 2099 __unregister_request(osdc, req);
fe5da05e
ID		 2100 if (req->r_linger) {
2101 WARN_ON(req->r_unsafe_callback);
2102 __register_linger_request(osdc, req);
2103 }
2104 }
f24e9980
SW		 2105
2106 mutex_unlock(&osdc->request_mutex);
ff513ace 2107 up_read(&osdc->map_sem);
f24e9980 2108
fe5da05e
ID		 2109 if (done_request(req, &m)) {
2110 if (already_acked && req->r_unsafe_callback) {
2111 dout("req %p tid %llu safe-cb\n", req, req->r_tid);
61c5d6bf 2112 req->r_unsafe_callback(req, false);
fe5da05e
ID		 2113 } else {
2114 dout("req %p tid %llu cb\n", req, req->r_tid);
2115 __complete_request(req);
2116 }
2117 } else {
2118 if (req->r_unsafe_callback) {
2119 dout("req %p tid %llu unsafe-cb\n", req, req->r_tid);
2120 req->r_unsafe_callback(req, true);
2121 } else {
2122 WARN_ON(1);
2123 }
61c5d6bf 2124 }
fe5da05e
ID		 2125 if (m.flags & CEPH_OSD_FLAG_ONDISK)
2126 complete_all(&req->r_safe_completion);
f24e9980 2127
f24e9980
SW		 2128 ceph_osdc_put_request(req);
2129 return;
2130
fe5da05e 2131fail_request:
37c89bde
LW		 2132 req->r_result = -EIO;
2133 __unregister_request(osdc, req);
fe5da05e
ID		 2134 __complete_request(req);
2135 complete_all(&req->r_safe_completion);
2136out_put:
1b83bef2 2137 ceph_osdc_put_request(req);
fe5da05e 2138out_unlock:
8058fd45 2139 mutex_unlock(&osdc->request_mutex);
ff513ace 2140 up_read(&osdc->map_sem);
f24e9980
SW		 2141}
2142
6f6c7006 2143static void reset_changed_osds(struct ceph_osd_client *osdc)
f24e9980 2144{
f24e9980 2145 struct rb_node *p, *n;
f24e9980 2146
7eb71e03 2147 dout("%s %p\n", __func__, osdc);
6f6c7006
SW		 2148 for (p = rb_first(&osdc->osds); p; p = n) {
2149 struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
f24e9980 2150
6f6c7006
SW		 2151 n = rb_next(p);
2152 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
2153 memcmp(&osd->o_con.peer_addr,
2154 ceph_osd_addr(osdc->osdmap,
2155 osd->o_osd),
2156 sizeof(struct ceph_entity_addr)) != 0)
2157 __reset_osd(osdc, osd);
f24e9980 2158 }
422d2cb8
YS		 2159}
2160
2161/*
6f6c7006
SW		 2162 * Requeue requests whose mapping to an OSD has changed. If requests map to
2163 * no osd, request a new map.
422d2cb8 2164 *
e6d50f67 2165 * Caller should hold map_sem for read.
422d2cb8 2166 */
9a1ea2db
JD		 2167static void kick_requests(struct ceph_osd_client *osdc, bool force_resend,
2168 bool force_resend_writes)
422d2cb8 2169{
a40c4f10 2170 struct ceph_osd_request *req, *nreq;
6f6c7006
SW		 2171 struct rb_node *p;
2172 int needmap = 0;
2173 int err;
9a1ea2db 2174 bool force_resend_req;
422d2cb8 2175
9a1ea2db
JD		 2176 dout("kick_requests %s %s\n", force_resend ? " (force resend)" : "",
2177 force_resend_writes ? " (force resend writes)" : "");
422d2cb8 2178 mutex_lock(&osdc->request_mutex);
6194ea89 2179 for (p = rb_first(&osdc->requests); p; ) {
6f6c7006 2180 req = rb_entry(p, struct ceph_osd_request, r_node);
6194ea89 2181 p = rb_next(p);
ab60b16d
AE		 2182
2183 /*
2184 * For linger requests that have not yet been
2185 * registered, move them to the linger list; they'll
2186 * be sent to the osd in the loop below. Unregister
2187 * the request before re-registering it as a linger
2188 * request to ensure the __map_request() below
2189 * will decide it needs to be sent.
2190 */
2191 if (req->r_linger && list_empty(&req->r_linger_item)) {
2192 dout("%p tid %llu restart on osd%d\n",
2193 req, req->r_tid,
2194 req->r_osd ? req->r_osd->o_osd : -1);
96e4dac6 2195 ceph_osdc_get_request(req);
ab60b16d
AE		 2196 __unregister_request(osdc, req);
2197 __register_linger_request(osdc, req);
96e4dac6 2198 ceph_osdc_put_request(req);
ab60b16d
AE		 2199 continue;
2200 }
2201
9a1ea2db
JD		 2202 force_resend_req = force_resend ||
2203 (force_resend_writes &&
2204 req->r_flags & CEPH_OSD_FLAG_WRITE);
2205 err = __map_request(osdc, req, force_resend_req);
6f6c7006
SW		 2206 if (err < 0)
2207 continue; /* error */
2208 if (req->r_osd == NULL) {
2209 dout("%p tid %llu maps to no osd\n", req, req->r_tid);
2210 needmap++; /* request a newer map */
2211 } else if (err > 0) {
6194ea89
SW		 2212 if (!req->r_linger) {
2213 dout("%p tid %llu requeued on osd%d\n", req,
2214 req->r_tid,
2215 req->r_osd ? req->r_osd->o_osd : -1);
a40c4f10 2216 req->r_flags |= CEPH_OSD_FLAG_RETRY;
6194ea89
SW		 2217 }
2218 }
a40c4f10
YS		 2219 }
2220
2221 list_for_each_entry_safe(req, nreq, &osdc->req_linger,
2222 r_linger_item) {
2223 dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
2224
9a1ea2db
JD		 2225 err = __map_request(osdc, req,
2226 force_resend || force_resend_writes);
ab60b16d 2227 dout("__map_request returned %d\n", err);
a40c4f10
YS		 2228 if (err < 0)
2229 continue; /* hrm! */
b0494532
ID		 2230 if (req->r_osd == NULL || err > 0) {
2231 if (req->r_osd == NULL) {
2232 dout("lingering %p tid %llu maps to no osd\n",
2233 req, req->r_tid);
2234 /*
2235 * A homeless lingering request makes
2236 * no sense, as it's job is to keep
2237 * a particular OSD connection open.
2238 * Request a newer map and kick the
2239 * request, knowing that it won't be
2240 * resent until we actually get a map
2241 * that can tell us where to send it.
2242 */
2243 needmap++;
2244 }
a40c4f10 2245
b0494532
ID		 2246 dout("kicking lingering %p tid %llu osd%d\n", req,
2247 req->r_tid, req->r_osd ? req->r_osd->o_osd : -1);
2248 __register_request(osdc, req);
2249 __unregister_linger_request(osdc, req);
2250 }
6f6c7006 2251 }
14d2f38d 2252 reset_changed_osds(osdc);
f24e9980
SW		 2253 mutex_unlock(&osdc->request_mutex);
2254
2255 if (needmap) {
2256 dout("%d requests for down osds, need new map\n", needmap);
2257 ceph_monc_request_next_osdmap(&osdc->client->monc);
2258 }
422d2cb8 2259}
6f6c7006
SW		 2260
2261
f24e9980
SW		 2262/*
2263 * Process updated osd map.
2264 *
2265 * The message contains any number of incremental and full maps, normally
2266 * indicating some sort of topology change in the cluster. Kick requests
2267 * off to different OSDs as needed.
2268 */
2269void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
2270{
2271 void *p, *end, *next;
2272 u32 nr_maps, maplen;
2273 u32 epoch;
2274 struct ceph_osdmap *newmap = NULL, *oldmap;
2275 int err;
2276 struct ceph_fsid fsid;
9a1ea2db 2277 bool was_full;
f24e9980
SW		 2278
2279 dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
2280 p = msg->front.iov_base;
2281 end = p + msg->front.iov_len;
2282
2283 /* verify fsid */
2284 ceph_decode_need(&p, end, sizeof(fsid), bad);
2285 ceph_decode_copy(&p, &fsid, sizeof(fsid));
0743304d
SW		 2286 if (ceph_check_fsid(osdc->client, &fsid) < 0)
2287 return;
f24e9980
SW		 2288
2289 down_write(&osdc->map_sem);
2290
9a1ea2db
JD		 2291 was_full = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
2292
f24e9980
SW		 2293 /* incremental maps */
2294 ceph_decode_32_safe(&p, end, nr_maps, bad);
2295 dout(" %d inc maps\n", nr_maps);
2296 while (nr_maps > 0) {
2297 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 2298 epoch = ceph_decode_32(&p);
2299 maplen = ceph_decode_32(&p);
f24e9980
SW		 2300 ceph_decode_need(&p, end, maplen, bad);
2301 next = p + maplen;
2302 if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
2303 dout("applying incremental map %u len %d\n",
2304 epoch, maplen);
2305 newmap = osdmap_apply_incremental(&p, next,
0c0a8de1 2306 osdc->osdmap);
f24e9980
SW		 2307 if (IS_ERR(newmap)) {
2308 err = PTR_ERR(newmap);
2309 goto bad;
2310 }
30dc6381 2311 BUG_ON(!newmap);
f24e9980
SW		 2312 if (newmap != osdc->osdmap) {
2313 ceph_osdmap_destroy(osdc->osdmap);
2314 osdc->osdmap = newmap;
2315 }
9a1ea2db
JD		 2316 was_full = was_full ||
2317 ceph_osdmap_flag(osdc->osdmap,
2318 CEPH_OSDMAP_FULL);
2319 kick_requests(osdc, 0, was_full);
f24e9980
SW		 2320 } else {
2321 dout("ignoring incremental map %u len %d\n",
2322 epoch, maplen);
2323 }
2324 p = next;
2325 nr_maps--;
2326 }
2327 if (newmap)
2328 goto done;
2329
2330 /* full maps */
2331 ceph_decode_32_safe(&p, end, nr_maps, bad);
2332 dout(" %d full maps\n", nr_maps);
2333 while (nr_maps) {
2334 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 2335 epoch = ceph_decode_32(&p);
2336 maplen = ceph_decode_32(&p);
f24e9980
SW		 2337 ceph_decode_need(&p, end, maplen, bad);
2338 if (nr_maps > 1) {
2339 dout("skipping non-latest full map %u len %d\n",
2340 epoch, maplen);
2341 } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
2342 dout("skipping full map %u len %d, "
2343 "older than our %u\n", epoch, maplen,
2344 osdc->osdmap->epoch);
2345 } else {
38d6453c
SW		 2346 int skipped_map = 0;
2347
f24e9980 2348 dout("taking full map %u len %d\n", epoch, maplen);
a2505d63 2349 newmap = ceph_osdmap_decode(&p, p+maplen);
f24e9980
SW		 2350 if (IS_ERR(newmap)) {
2351 err = PTR_ERR(newmap);
2352 goto bad;
2353 }
30dc6381 2354 BUG_ON(!newmap);
f24e9980
SW		 2355 oldmap = osdc->osdmap;
2356 osdc->osdmap = newmap;
38d6453c
SW		 2357 if (oldmap) {
2358 if (oldmap->epoch + 1 < newmap->epoch)
2359 skipped_map = 1;
f24e9980 2360 ceph_osdmap_destroy(oldmap);
38d6453c 2361 }
9a1ea2db
JD		 2362 was_full = was_full ||
2363 ceph_osdmap_flag(osdc->osdmap,
2364 CEPH_OSDMAP_FULL);
2365 kick_requests(osdc, skipped_map, was_full);
f24e9980
SW		 2366 }
2367 p += maplen;
2368 nr_maps--;
2369 }
2370
b72e19b9
DC		 2371 if (!osdc->osdmap)
2372 goto bad;
f24e9980
SW		 2373done:
2374 downgrade_write(&osdc->map_sem);
82dcabad
ID		 2375 ceph_monc_got_map(&osdc->client->monc, CEPH_SUB_OSDMAP,
2376 osdc->osdmap->epoch);
cd634fb6
SW		 2377
2378 /*
2379 * subscribe to subsequent osdmap updates if full to ensure
2380 * we find out when we are no longer full and stop returning
2381 * ENOSPC.
2382 */
d29adb34
JD		 2383 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) ||
2384 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD) ||
2385 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR))
cd634fb6
SW		 2386 ceph_monc_request_next_osdmap(&osdc->client->monc);
2387
f9d25199
AE		 2388 mutex_lock(&osdc->request_mutex);
2389 __send_queued(osdc);
2390 mutex_unlock(&osdc->request_mutex);
f24e9980 2391 up_read(&osdc->map_sem);
03066f23 2392 wake_up_all(&osdc->client->auth_wq);
f24e9980
SW		 2393 return;
2394
2395bad:
2396 pr_err("osdc handle_map corrupt msg\n");
9ec7cab1 2397 ceph_msg_dump(msg);
f24e9980 2398 up_write(&osdc->map_sem);
f24e9980
SW		 2399}
2400
a40c4f10
YS		 2401/*
2402 * watch/notify callback event infrastructure
2403 *
2404 * These callbacks are used both for watch and notify operations.
2405 */
2406static void __release_event(struct kref *kref)
2407{
2408 struct ceph_osd_event *event =
2409 container_of(kref, struct ceph_osd_event, kref);
2410
2411 dout("__release_event %p\n", event);
2412 kfree(event);
2413}
2414
2415static void get_event(struct ceph_osd_event *event)
2416{
2417 kref_get(&event->kref);
2418}
2419
2420void ceph_osdc_put_event(struct ceph_osd_event *event)
2421{
2422 kref_put(&event->kref, __release_event);
2423}
2424EXPORT_SYMBOL(ceph_osdc_put_event);
2425
2426static void __insert_event(struct ceph_osd_client *osdc,
2427 struct ceph_osd_event *new)
2428{
2429 struct rb_node **p = &osdc->event_tree.rb_node;
2430 struct rb_node *parent = NULL;
2431 struct ceph_osd_event *event = NULL;
2432
2433 while (*p) {
2434 parent = *p;
2435 event = rb_entry(parent, struct ceph_osd_event, node);
2436 if (new->cookie < event->cookie)
2437 p = &(*p)->rb_left;
2438 else if (new->cookie > event->cookie)
2439 p = &(*p)->rb_right;
2440 else
2441 BUG();
2442 }
2443
2444 rb_link_node(&new->node, parent, p);
2445 rb_insert_color(&new->node, &osdc->event_tree);
2446}
2447
2448static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
2449 u64 cookie)
2450{
2451 struct rb_node **p = &osdc->event_tree.rb_node;
2452 struct rb_node *parent = NULL;
2453 struct ceph_osd_event *event = NULL;
2454
2455 while (*p) {
2456 parent = *p;
2457 event = rb_entry(parent, struct ceph_osd_event, node);
2458 if (cookie < event->cookie)
2459 p = &(*p)->rb_left;
2460 else if (cookie > event->cookie)
2461 p = &(*p)->rb_right;
2462 else
2463 return event;
2464 }
2465 return NULL;
2466}
2467
2468static void __remove_event(struct ceph_osd_event *event)
2469{
2470 struct ceph_osd_client *osdc = event->osdc;
2471
2472 if (!RB_EMPTY_NODE(&event->node)) {
2473 dout("__remove_event removed %p\n", event);
2474 rb_erase(&event->node, &osdc->event_tree);
2475 ceph_osdc_put_event(event);
2476 } else {
2477 dout("__remove_event didn't remove %p\n", event);
2478 }
2479}
2480
2481int ceph_osdc_create_event(struct ceph_osd_client *osdc,
2482 void (*event_cb)(u64, u64, u8, void *),
3c663bbd 2483 void *data, struct ceph_osd_event **pevent)
a40c4f10
YS		 2484{
2485 struct ceph_osd_event *event;
2486
2487 event = kmalloc(sizeof(*event), GFP_NOIO);
2488 if (!event)
2489 return -ENOMEM;
2490
2491 dout("create_event %p\n", event);
2492 event->cb = event_cb;
3c663bbd 2493 event->one_shot = 0;
a40c4f10
YS		 2494 event->data = data;
2495 event->osdc = osdc;
2496 INIT_LIST_HEAD(&event->osd_node);
3ee5234d 2497 RB_CLEAR_NODE(&event->node);
a40c4f10
YS		 2498 kref_init(&event->kref); /* one ref for us */
2499 kref_get(&event->kref); /* one ref for the caller */
a40c4f10
YS		 2500
2501 spin_lock(&osdc->event_lock);
2502 event->cookie = ++osdc->event_count;
2503 __insert_event(osdc, event);
2504 spin_unlock(&osdc->event_lock);
2505
2506 *pevent = event;
2507 return 0;
2508}
2509EXPORT_SYMBOL(ceph_osdc_create_event);
2510
2511void ceph_osdc_cancel_event(struct ceph_osd_event *event)
2512{
2513 struct ceph_osd_client *osdc = event->osdc;
2514
2515 dout("cancel_event %p\n", event);
2516 spin_lock(&osdc->event_lock);
2517 __remove_event(event);
2518 spin_unlock(&osdc->event_lock);
2519 ceph_osdc_put_event(event); /* caller's */
2520}
2521EXPORT_SYMBOL(ceph_osdc_cancel_event);
2522
2523
2524static void do_event_work(struct work_struct *work)
2525{
2526 struct ceph_osd_event_work *event_work =
2527 container_of(work, struct ceph_osd_event_work, work);
2528 struct ceph_osd_event *event = event_work->event;
2529 u64 ver = event_work->ver;
2530 u64 notify_id = event_work->notify_id;
2531 u8 opcode = event_work->opcode;
2532
2533 dout("do_event_work completing %p\n", event);
2534 event->cb(ver, notify_id, opcode, event->data);
a40c4f10
YS		 2535 dout("do_event_work completed %p\n", event);
2536 ceph_osdc_put_event(event);
2537 kfree(event_work);
2538}
2539
2540
2541/*
2542 * Process osd watch notifications
2543 */
3c663bbd
AE		 2544static void handle_watch_notify(struct ceph_osd_client *osdc,
2545 struct ceph_msg *msg)
a40c4f10
YS		 2546{
2547 void *p, *end;
2548 u8 proto_ver;
2549 u64 cookie, ver, notify_id;
2550 u8 opcode;
2551 struct ceph_osd_event *event;
2552 struct ceph_osd_event_work *event_work;
2553
2554 p = msg->front.iov_base;
2555 end = p + msg->front.iov_len;
2556
2557 ceph_decode_8_safe(&p, end, proto_ver, bad);
2558 ceph_decode_8_safe(&p, end, opcode, bad);
2559 ceph_decode_64_safe(&p, end, cookie, bad);
2560 ceph_decode_64_safe(&p, end, ver, bad);
2561 ceph_decode_64_safe(&p, end, notify_id, bad);
2562
2563 spin_lock(&osdc->event_lock);
2564 event = __find_event(osdc, cookie);
2565 if (event) {
3c663bbd 2566 BUG_ON(event->one_shot);
a40c4f10 2567 get_event(event);
a40c4f10
YS		 2568 }
2569 spin_unlock(&osdc->event_lock);
2570 dout("handle_watch_notify cookie %lld ver %lld event %p\n",
2571 cookie, ver, event);
2572 if (event) {
2573 event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
a40c4f10 2574 if (!event_work) {
91883cd2
ID		 2575 pr_err("couldn't allocate event_work\n");
2576 ceph_osdc_put_event(event);
2577 return;
a40c4f10 2578 }
6b0ae409 2579 INIT_WORK(&event_work->work, do_event_work);
a40c4f10
YS		 2580 event_work->event = event;
2581 event_work->ver = ver;
2582 event_work->notify_id = notify_id;
2583 event_work->opcode = opcode;
a40c4f10 2584
91883cd2
ID		 2585 queue_work(osdc->notify_wq, &event_work->work);
2586 }
a40c4f10 2587
a40c4f10
YS		 2588 return;
2589
2590bad:
2591 pr_err("osdc handle_watch_notify corrupt msg\n");
a40c4f10
YS		 2592}
2593
70636773
AE		 2594/*
2595 * Register request, send initial attempt.
2596 */
2597int ceph_osdc_start_request(struct ceph_osd_client *osdc,
2598 struct ceph_osd_request *req,
2599 bool nofail)
2600{
0bbfdfe8 2601 int rc;
70636773 2602
f24e9980
SW		 2603 down_read(&osdc->map_sem);
2604 mutex_lock(&osdc->request_mutex);
0bbfdfe8
ID		 2605
2606 rc = __ceph_osdc_start_request(osdc, req, nofail);
2607
f24e9980
SW		 2608 mutex_unlock(&osdc->request_mutex);
2609 up_read(&osdc->map_sem);
0bbfdfe8 2610
f24e9980
SW		 2611 return rc;
2612}
3d14c5d2 2613EXPORT_SYMBOL(ceph_osdc_start_request);
f24e9980 2614
c9f9b93d
ID		 2615/*
2616 * Unregister a registered request. The request is not completed (i.e.
2617 * no callbacks or wakeups) - higher layers are supposed to know what
2618 * they are canceling.
2619 */
2620void ceph_osdc_cancel_request(struct ceph_osd_request *req)
2621{
2622 struct ceph_osd_client *osdc = req->r_osdc;
2623
2624 mutex_lock(&osdc->request_mutex);
2625 if (req->r_linger)
2626 __unregister_linger_request(osdc, req);
2627 __unregister_request(osdc, req);
2628 mutex_unlock(&osdc->request_mutex);
2629
2630 dout("%s %p tid %llu canceled\n", __func__, req, req->r_tid);
2631}
2632EXPORT_SYMBOL(ceph_osdc_cancel_request);
2633
f24e9980
SW		 2634/*
2635 * wait for a request to complete
2636 */
2637int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
2638 struct ceph_osd_request *req)
2639{
2640 int rc;
2641
c9f9b93d
ID		 2642 dout("%s %p tid %llu\n", __func__, req, req->r_tid);
2643
f24e9980
SW		 2644 rc = wait_for_completion_interruptible(&req->r_completion);
2645 if (rc < 0) {
c9f9b93d
ID		 2646 dout("%s %p tid %llu interrupted\n", __func__, req, req->r_tid);
2647 ceph_osdc_cancel_request(req);
fe5da05e
ID		 2648
2649 /* kludge - need to to wake ceph_osdc_sync() */
2650 complete_all(&req->r_safe_completion);
f24e9980
SW		 2651 return rc;
2652 }
2653
c9f9b93d
ID		 2654 dout("%s %p tid %llu result %d\n", __func__, req, req->r_tid,
2655 req->r_result);
f24e9980
SW		 2656 return req->r_result;
2657}
3d14c5d2 2658EXPORT_SYMBOL(ceph_osdc_wait_request);
f24e9980
SW		 2659
2660/*
2661 * sync - wait for all in-flight requests to flush. avoid starvation.
2662 */
2663void ceph_osdc_sync(struct ceph_osd_client *osdc)
2664{
2665 struct ceph_osd_request *req;
2666 u64 last_tid, next_tid = 0;
2667
2668 mutex_lock(&osdc->request_mutex);
2669 last_tid = osdc->last_tid;
2670 while (1) {
2671 req = __lookup_request_ge(osdc, next_tid);
2672 if (!req)
2673 break;
2674 if (req->r_tid > last_tid)
2675 break;
2676
2677 next_tid = req->r_tid + 1;
2678 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
2679 continue;
2680
2681 ceph_osdc_get_request(req);
2682 mutex_unlock(&osdc->request_mutex);
2683 dout("sync waiting on tid %llu (last is %llu)\n",
2684 req->r_tid, last_tid);
2685 wait_for_completion(&req->r_safe_completion);
2686 mutex_lock(&osdc->request_mutex);
2687 ceph_osdc_put_request(req);
2688 }
2689 mutex_unlock(&osdc->request_mutex);
2690 dout("sync done (thru tid %llu)\n", last_tid);
2691}
3d14c5d2 2692EXPORT_SYMBOL(ceph_osdc_sync);
f24e9980 2693
dd935f44
JD		 2694/*
2695 * Call all pending notify callbacks - for use after a watch is
2696 * unregistered, to make sure no more callbacks for it will be invoked
2697 */
f6479449 2698void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc)
dd935f44
JD		 2699{
2700 flush_workqueue(osdc->notify_wq);
2701}
2702EXPORT_SYMBOL(ceph_osdc_flush_notifies);
2703
2704
f24e9980
SW		 2705/*
2706 * init, shutdown
2707 */
2708int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
2709{
2710 int err;
2711
2712 dout("init\n");
2713 osdc->client = client;
2714 osdc->osdmap = NULL;
2715 init_rwsem(&osdc->map_sem);
f24e9980 2716 mutex_init(&osdc->request_mutex);
f24e9980
SW		 2717 osdc->last_tid = 0;
2718 osdc->osds = RB_ROOT;
f5a2041b 2719 INIT_LIST_HEAD(&osdc->osd_lru);
f24e9980 2720 osdc->requests = RB_ROOT;
422d2cb8 2721 INIT_LIST_HEAD(&osdc->req_lru);
6f6c7006
SW		 2722 INIT_LIST_HEAD(&osdc->req_unsent);
2723 INIT_LIST_HEAD(&osdc->req_notarget);
a40c4f10 2724 INIT_LIST_HEAD(&osdc->req_linger);
f24e9980
SW		 2725 osdc->num_requests = 0;
2726 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
f5a2041b 2727 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
a40c4f10
YS		 2728 spin_lock_init(&osdc->event_lock);
2729 osdc->event_tree = RB_ROOT;
2730 osdc->event_count = 0;
f5a2041b 2731
5f44f142 2732 err = -ENOMEM;
9e767adb
ID		 2733 osdc->req_mempool = mempool_create_slab_pool(10,
2734 ceph_osd_request_cache);
f24e9980 2735 if (!osdc->req_mempool)
5f44f142 2736 goto out;
f24e9980 2737
d50b409f 2738 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
711da55d 2739 PAGE_SIZE, 10, true, "osd_op");
f24e9980 2740 if (err < 0)
5f44f142 2741 goto out_mempool;
d50b409f 2742 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
711da55d 2743 PAGE_SIZE, 10, true, "osd_op_reply");
c16e7869
SW		 2744 if (err < 0)
2745 goto out_msgpool;
a40c4f10 2746
dbcae088 2747 err = -ENOMEM;
a40c4f10 2748 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
dbcae088 2749 if (!osdc->notify_wq)
c172ec5c
ID		 2750 goto out_msgpool_reply;
2751
b37ee1b9
ID		 2752 schedule_delayed_work(&osdc->osds_timeout_work,
2753 round_jiffies_relative(osdc->client->options->osd_idle_ttl));
2754
f24e9980 2755 return 0;
5f44f142 2756
c172ec5c
ID		 2757out_msgpool_reply:
2758 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
c16e7869
SW		 2759out_msgpool:
2760 ceph_msgpool_destroy(&osdc->msgpool_op);
5f44f142
SW		 2761out_mempool:
2762 mempool_destroy(osdc->req_mempool);
2763out:
2764 return err;
f24e9980
SW		 2765}
2766
2767void ceph_osdc_stop(struct ceph_osd_client *osdc)
2768{
a40c4f10
YS		 2769 flush_workqueue(osdc->notify_wq);
2770 destroy_workqueue(osdc->notify_wq);
f24e9980 2771 cancel_delayed_work_sync(&osdc->timeout_work);
f5a2041b 2772 cancel_delayed_work_sync(&osdc->osds_timeout_work);
42a2c09f
ID		 2773
2774 mutex_lock(&osdc->request_mutex);
2775 while (!RB_EMPTY_ROOT(&osdc->osds)) {
2776 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
2777 struct ceph_osd, o_node);
2778 remove_osd(osdc, osd);
2779 }
2780 mutex_unlock(&osdc->request_mutex);
2781
f24e9980
SW		 2782 if (osdc->osdmap) {
2783 ceph_osdmap_destroy(osdc->osdmap);
2784 osdc->osdmap = NULL;
2785 }
2786 mempool_destroy(osdc->req_mempool);
2787 ceph_msgpool_destroy(&osdc->msgpool_op);
c16e7869 2788 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
f24e9980
SW		 2789}
2790
2791/*
2792 * Read some contiguous pages. If we cross a stripe boundary, shorten
2793 * *plen. Return number of bytes read, or error.
2794 */
2795int ceph_osdc_readpages(struct ceph_osd_client *osdc,
2796 struct ceph_vino vino, struct ceph_file_layout *layout,
2797 u64 off, u64 *plen,
2798 u32 truncate_seq, u64 truncate_size,
b7495fc2 2799 struct page **pages, int num_pages, int page_align)
f24e9980
SW		 2800{
2801 struct ceph_osd_request *req;
2802 int rc = 0;
2803
2804 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
2805 vino.snap, off, *plen);
715e4cd4 2806 req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 0, 1,
f24e9980 2807 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
acead002 2808 NULL, truncate_seq, truncate_size,
153e5167 2809 false);
6816282d
SW		 2810 if (IS_ERR(req))
2811 return PTR_ERR(req);
f24e9980
SW		 2812
2813 /* it may be a short read due to an object boundary */
406e2c9f 2814 osd_req_op_extent_osd_data_pages(req, 0,
a4ce40a9 2815 pages, *plen, page_align, false, false);
f24e9980 2816
e0c59487 2817 dout("readpages final extent is %llu~%llu (%llu bytes align %d)\n",
43bfe5de 2818 off, *plen, *plen, page_align);
f24e9980
SW		 2819
2820 rc = ceph_osdc_start_request(osdc, req, false);
2821 if (!rc)
2822 rc = ceph_osdc_wait_request(osdc, req);
2823
2824 ceph_osdc_put_request(req);
2825 dout("readpages result %d\n", rc);
2826 return rc;
2827}
3d14c5d2 2828EXPORT_SYMBOL(ceph_osdc_readpages);
f24e9980
SW		 2829
2830/*
2831 * do a synchronous write on N pages
2832 */
2833int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
2834 struct ceph_file_layout *layout,
2835 struct ceph_snap_context *snapc,
2836 u64 off, u64 len,
2837 u32 truncate_seq, u64 truncate_size,
2838 struct timespec *mtime,
24808826 2839 struct page **pages, int num_pages)
f24e9980
SW		 2840{
2841 struct ceph_osd_request *req;
2842 int rc = 0;
b7495fc2 2843 int page_align = off & ~PAGE_MASK;
f24e9980 2844
715e4cd4 2845 req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 0, 1,
f24e9980 2846 CEPH_OSD_OP_WRITE,
24808826 2847 CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
acead002 2848 snapc, truncate_seq, truncate_size,
153e5167 2849 true);
6816282d
SW		 2850 if (IS_ERR(req))
2851 return PTR_ERR(req);
f24e9980
SW		 2852
2853 /* it may be a short write due to an object boundary */
406e2c9f 2854 osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align,
43bfe5de
AE		 2855 false, false);
2856 dout("writepages %llu~%llu (%llu bytes)\n", off, len, len);
f24e9980 2857
bb873b53 2858 req->r_mtime = *mtime;
87f979d3 2859 rc = ceph_osdc_start_request(osdc, req, true);
f24e9980
SW		 2860 if (!rc)
2861 rc = ceph_osdc_wait_request(osdc, req);
2862
2863 ceph_osdc_put_request(req);
2864 if (rc == 0)
2865 rc = len;
2866 dout("writepages result %d\n", rc);
2867 return rc;
2868}
3d14c5d2 2869EXPORT_SYMBOL(ceph_osdc_writepages);
f24e9980 2870
5522ae0b
AE		 2871int ceph_osdc_setup(void)
2872{
3f1af42a
ID		 2873 size_t size = sizeof(struct ceph_osd_request) +
2874 CEPH_OSD_SLAB_OPS * sizeof(struct ceph_osd_req_op);
2875
5522ae0b 2876 BUG_ON(ceph_osd_request_cache);
3f1af42a
ID		 2877 ceph_osd_request_cache = kmem_cache_create("ceph_osd_request", size,
2878 0, 0, NULL);
5522ae0b
AE		 2879
2880 return ceph_osd_request_cache ? 0 : -ENOMEM;
2881}
2882EXPORT_SYMBOL(ceph_osdc_setup);
2883
2884void ceph_osdc_cleanup(void)
2885{
2886 BUG_ON(!ceph_osd_request_cache);
2887 kmem_cache_destroy(ceph_osd_request_cache);
2888 ceph_osd_request_cache = NULL;
2889}
2890EXPORT_SYMBOL(ceph_osdc_cleanup);
2891
f24e9980
SW		 2892/*
2893 * handle incoming message
2894 */
2895static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
2896{
2897 struct ceph_osd *osd = con->private;
32c895e7 2898 struct ceph_osd_client *osdc;
f24e9980
SW		 2899 int type = le16_to_cpu(msg->hdr.type);
2900
2901 if (!osd)
4a32f93d 2902 goto out;
32c895e7 2903 osdc = osd->o_osdc;
f24e9980
SW		 2904
2905 switch (type) {
2906 case CEPH_MSG_OSD_MAP:
2907 ceph_osdc_handle_map(osdc, msg);
2908 break;
2909 case CEPH_MSG_OSD_OPREPLY:
70cf052d 2910 handle_reply(osdc, msg);
f24e9980 2911 break;
a40c4f10
YS		 2912 case CEPH_MSG_WATCH_NOTIFY:
2913 handle_watch_notify(osdc, msg);
2914 break;
f24e9980
SW		 2915
2916 default:
2917 pr_err("received unknown message type %d %s\n", type,
2918 ceph_msg_type_name(type));
2919 }
4a32f93d 2920out:
f24e9980
SW		 2921 ceph_msg_put(msg);
2922}
2923
5b3a4db3 2924/*
d15f9d69
ID		 2925 * Lookup and return message for incoming reply. Don't try to do
2926 * anything about a larger than preallocated data portion of the
2927 * message at the moment - for now, just skip the message.
5b3a4db3
SW		 2928 */
2929static struct ceph_msg *get_reply(struct ceph_connection *con,
2450418c
YS		 2930 struct ceph_msg_header *hdr,
2931 int *skip)
f24e9980
SW		 2932{
2933 struct ceph_osd *osd = con->private;
2934 struct ceph_osd_client *osdc = osd->o_osdc;
2450418c 2935 struct ceph_msg *m;
0547a9b3 2936 struct ceph_osd_request *req;
3f0a4ac5 2937 int front_len = le32_to_cpu(hdr->front_len);
5b3a4db3 2938 int data_len = le32_to_cpu(hdr->data_len);
0547a9b3 2939 u64 tid;
f24e9980 2940
0547a9b3
YS		 2941 tid = le64_to_cpu(hdr->tid);
2942 mutex_lock(&osdc->request_mutex);
fcd00b68 2943 req = lookup_request(&osdc->requests, tid);
0547a9b3 2944 if (!req) {
cd8140c6
ID		 2945 dout("%s osd%d tid %llu unknown, skipping\n", __func__,
2946 osd->o_osd, tid);
0547a9b3 2947 m = NULL;
d15f9d69 2948 *skip = 1;
0547a9b3
YS		 2949 goto out;
2950 }
c16e7869 2951
ace6d3a9 2952 ceph_msg_revoke_incoming(req->r_reply);
0547a9b3 2953
f2be82b0 2954 if (front_len > req->r_reply->front_alloc_len) {
d15f9d69
ID		 2955 pr_warn("%s osd%d tid %llu front %d > preallocated %d\n",
2956 __func__, osd->o_osd, req->r_tid, front_len,
2957 req->r_reply->front_alloc_len);
3f0a4ac5
ID		 2958 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS,
2959 false);
a79832f2 2960 if (!m)
c16e7869
SW		 2961 goto out;
2962 ceph_msg_put(req->r_reply);
2963 req->r_reply = m;
2964 }
0fff87ec 2965
d15f9d69
ID		 2966 if (data_len > req->r_reply->data_length) {
2967 pr_warn("%s osd%d tid %llu data %d > preallocated %zu, skipping\n",
2968 __func__, osd->o_osd, req->r_tid, data_len,
2969 req->r_reply->data_length);
2970 m = NULL;
2971 *skip = 1;
2972 goto out;
0547a9b3 2973 }
d15f9d69
ID		 2974
2975 m = ceph_msg_get(req->r_reply);
c16e7869 2976 dout("get_reply tid %lld %p\n", tid, m);
0547a9b3
YS		 2977
2978out:
2979 mutex_unlock(&osdc->request_mutex);
2450418c 2980 return m;
5b3a4db3
SW		 2981}
2982
2983static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2984 struct ceph_msg_header *hdr,
2985 int *skip)
2986{
2987 struct ceph_osd *osd = con->private;
2988 int type = le16_to_cpu(hdr->type);
2989 int front = le32_to_cpu(hdr->front_len);
2990
1c20f2d2 2991 *skip = 0;
5b3a4db3
SW		 2992 switch (type) {
2993 case CEPH_MSG_OSD_MAP:
a40c4f10 2994 case CEPH_MSG_WATCH_NOTIFY:
b61c2763 2995 return ceph_msg_new(type, front, GFP_NOFS, false);
5b3a4db3
SW		 2996 case CEPH_MSG_OSD_OPREPLY:
2997 return get_reply(con, hdr, skip);
2998 default:
2999 pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
3000 osd->o_osd);
3001 *skip = 1;
3002 return NULL;
3003 }
f24e9980
SW		 3004}
3005
3006/*
3007 * Wrappers to refcount containing ceph_osd struct
3008 */
3009static struct ceph_connection *get_osd_con(struct ceph_connection *con)
3010{
3011 struct ceph_osd *osd = con->private;
3012 if (get_osd(osd))
3013 return con;
3014 return NULL;
3015}
3016
3017static void put_osd_con(struct ceph_connection *con)
3018{
3019 struct ceph_osd *osd = con->private;
3020 put_osd(osd);
3021}
3022
4e7a5dcd
SW		 3023/*
3024 * authentication
3025 */
a3530df3
AE		 3026/*
3027 * Note: returned pointer is the address of a structure that's
3028 * managed separately. Caller must *not* attempt to free it.
3029 */
3030static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
8f43fb53 3031 int *proto, int force_new)
4e7a5dcd
SW		 3032{
3033 struct ceph_osd *o = con->private;
3034 struct ceph_osd_client *osdc = o->o_osdc;
3035 struct ceph_auth_client *ac = osdc->client->monc.auth;
74f1869f 3036 struct ceph_auth_handshake *auth = &o->o_auth;
4e7a5dcd 3037
74f1869f 3038 if (force_new && auth->authorizer) {
6c1ea260 3039 ceph_auth_destroy_authorizer(auth->authorizer);
74f1869f
AE		 3040 auth->authorizer = NULL;
3041 }
27859f97
SW		 3042 if (!auth->authorizer) {
3043 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
3044 auth);
4e7a5dcd 3045 if (ret)
a3530df3 3046 return ERR_PTR(ret);
27859f97
SW		 3047 } else {
3048 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
0bed9b5c
SW		 3049 auth);
3050 if (ret)
3051 return ERR_PTR(ret);
4e7a5dcd 3052 }
4e7a5dcd 3053 *proto = ac->protocol;
74f1869f 3054
a3530df3 3055 return auth;
4e7a5dcd
SW		 3056}
3057
3058
3059static int verify_authorizer_reply(struct ceph_connection *con, int len)
3060{
3061 struct ceph_osd *o = con->private;
3062 struct ceph_osd_client *osdc = o->o_osdc;
3063 struct ceph_auth_client *ac = osdc->client->monc.auth;
3064
27859f97 3065 return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer, len);
4e7a5dcd
SW		 3066}
3067
9bd2e6f8
SW		 3068static int invalidate_authorizer(struct ceph_connection *con)
3069{
3070 struct ceph_osd *o = con->private;
3071 struct ceph_osd_client *osdc = o->o_osdc;
3072 struct ceph_auth_client *ac = osdc->client->monc.auth;
3073
27859f97 3074 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
9bd2e6f8
SW		 3075 return ceph_monc_validate_auth(&osdc->client->monc);
3076}
4e7a5dcd 3077
79dbd1ba 3078static int osd_sign_message(struct ceph_msg *msg)
33d07337 3079{
79dbd1ba 3080 struct ceph_osd *o = msg->con->private;
33d07337 3081 struct ceph_auth_handshake *auth = &o->o_auth;
79dbd1ba 3082
33d07337
YZ		 3083 return ceph_auth_sign_message(auth, msg);
3084}
3085
79dbd1ba 3086static int osd_check_message_signature(struct ceph_msg *msg)
33d07337 3087{
79dbd1ba 3088 struct ceph_osd *o = msg->con->private;
33d07337 3089 struct ceph_auth_handshake *auth = &o->o_auth;
79dbd1ba 3090
33d07337
YZ		 3091 return ceph_auth_check_message_signature(auth, msg);
3092}
3093
9e32789f 3094static const struct ceph_connection_operations osd_con_ops = {
f24e9980
SW		 3095 .get = get_osd_con,
3096 .put = put_osd_con,
3097 .dispatch = dispatch,
4e7a5dcd
SW		 3098 .get_authorizer = get_authorizer,
3099 .verify_authorizer_reply = verify_authorizer_reply,
9bd2e6f8 3100 .invalidate_authorizer = invalidate_authorizer,
f24e9980 3101 .alloc_msg = alloc_msg,
79dbd1ba
ID		 3102 .sign_message = osd_sign_message,
3103 .check_message_signature = osd_check_message_signature,
81b024e7 3104 .fault = osd_reset,
f24e9980 3105};
Linux kernel - Deliverables
This page took 0.42272 seconds and 5 git commands to generate.