projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
libceph: osd_init() and osd_cleanup()
[linux-2.6-block.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 866
0247a0cf
ID		 867static bool osd_homeless(struct ceph_osd *osd)
868{
869 return osd->o_osd == CEPH_HOMELESS_OSD;
870}
871
f24e9980
SW		 872static struct ceph_osd_request *
873__lookup_request_ge(struct ceph_osd_client *osdc,
874 u64 tid)
875{
876 struct ceph_osd_request *req;
877 struct rb_node *n = osdc->requests.rb_node;
878
879 while (n) {
880 req = rb_entry(n, struct ceph_osd_request, r_node);
881 if (tid < req->r_tid) {
882 if (!n->rb_left)
883 return req;
884 n = n->rb_left;
885 } else if (tid > req->r_tid) {
886 n = n->rb_right;
887 } else {
888 return req;
889 }
890 }
891 return NULL;
892}
893
2cc6128a
ID		 894static void __kick_linger_request(struct ceph_osd_request *req)
895{
896 struct ceph_osd_client *osdc = req->r_osdc;
897 struct ceph_osd *osd = req->r_osd;
898
899 /*
900 * Linger requests need to be resent with a new tid to avoid
901 * the dup op detection logic on the OSDs. Achieve this with
902 * a re-register dance instead of open-coding.
903 */
904 ceph_osdc_get_request(req);
905 if (!list_empty(&req->r_linger_item))
906 __unregister_linger_request(osdc, req);
907 else
908 __unregister_request(osdc, req);
909 __register_request(osdc, req);
910 ceph_osdc_put_request(req);
911
912 /*
913 * Unless request has been registered as both normal and
914 * lingering, __unregister{,_linger}_request clears r_osd.
915 * However, here we need to preserve r_osd to make sure we
916 * requeue on the same OSD.
917 */
918 WARN_ON(req->r_osd || !osd);
919 req->r_osd = osd;
920
921 dout("%s requeueing %p tid %llu\n", __func__, req, req->r_tid);
922 __enqueue_request(req);
923}
924
6f6c7006
SW		 925/*
926 * Resubmit requests pending on the given osd.
927 */
928static void __kick_osd_requests(struct ceph_osd_client *osdc,
929 struct ceph_osd *osd)
930{
a40c4f10 931 struct ceph_osd_request *req, *nreq;
e02493c0 932 LIST_HEAD(resend);
2cc6128a 933 LIST_HEAD(resend_linger);
6f6c7006
SW		 934 int err;
935
2cc6128a 936 dout("%s osd%d\n", __func__, osd->o_osd);
6f6c7006 937 err = __reset_osd(osdc, osd);
685a7555 938 if (err)
6f6c7006 939 return;
2cc6128a 940
e02493c0
AE		 941 /*
942 * Build up a list of requests to resend by traversing the
943 * osd's list of requests. Requests for a given object are
944 * sent in tid order, and that is also the order they're
945 * kept on this list. Therefore all requests that are in
946 * flight will be found first, followed by all requests that
947 * have not yet been sent. And to resend requests while
948 * preserving this order we will want to put any sent
949 * requests back on the front of the osd client's unsent
950 * list.
951 *
952 * So we build a separate ordered list of already-sent
953 * requests for the affected osd and splice it onto the
954 * front of the osd client's unsent list. Once we've seen a
955 * request that has not yet been sent we're done. Those
956 * requests are already sitting right where they belong.
957 */
6f6c7006 958 list_for_each_entry(req, &osd->o_requests, r_osd_item) {
e02493c0
AE		 959 if (!req->r_sent)
960 break;
2cc6128a
ID		 961
962 if (!req->r_linger) {
963 dout("%s requeueing %p tid %llu\n", __func__, req,
964 req->r_tid);
965 list_move_tail(&req->r_req_lru_item, &resend);
a40c4f10 966 req->r_flags |= CEPH_OSD_FLAG_RETRY;
2cc6128a
ID		 967 } else {
968 list_move_tail(&req->r_req_lru_item, &resend_linger);
969 }
a40c4f10 970 }
e02493c0 971 list_splice(&resend, &osdc->req_unsent);
a40c4f10 972
e02493c0 973 /*
2cc6128a
ID		 974 * Both registered and not yet registered linger requests are
975 * enqueued with a new tid on the same OSD. We add/move them
976 * to req_unsent/o_requests at the end to keep things in tid
977 * order.
e02493c0 978 */
a40c4f10 979 list_for_each_entry_safe(req, nreq, &osd->o_linger_requests,
1d0326b1 980 r_linger_osd_item) {
2cc6128a
ID		 981 WARN_ON(!list_empty(&req->r_req_lru_item));
982 __kick_linger_request(req);
6f6c7006 983 }
2cc6128a
ID		 984
985 list_for_each_entry_safe(req, nreq, &resend_linger, r_req_lru_item)
986 __kick_linger_request(req);
6f6c7006
SW		 987}
988
f24e9980 989/*
81b024e7 990 * If the osd connection drops, we need to resubmit all requests.
f24e9980
SW		 991 */
992static void osd_reset(struct ceph_connection *con)
993{
994 struct ceph_osd *osd = con->private;
995 struct ceph_osd_client *osdc;
996
997 if (!osd)
998 return;
999 dout("osd_reset osd%d\n", osd->o_osd);
1000 osdc = osd->o_osdc;
f24e9980 1001 down_read(&osdc->map_sem);
83aff95e
SW		 1002 mutex_lock(&osdc->request_mutex);
1003 __kick_osd_requests(osdc, osd);
f9d25199 1004 __send_queued(osdc);
83aff95e 1005 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 1006 up_read(&osdc->map_sem);
1007}
1008
0247a0cf
ID		 1009/*
1010 * Assumes @osd is zero-initialized.
1011 */
1012static void osd_init(struct ceph_osd *osd)
1013{
1014 atomic_set(&osd->o_ref, 1);
1015 RB_CLEAR_NODE(&osd->o_node);
1016 INIT_LIST_HEAD(&osd->o_requests);
1017 INIT_LIST_HEAD(&osd->o_linger_requests);
1018 INIT_LIST_HEAD(&osd->o_osd_lru);
1019 INIT_LIST_HEAD(&osd->o_keepalive_item);
1020 osd->o_incarnation = 1;
1021}
1022
1023static void osd_cleanup(struct ceph_osd *osd)
1024{
1025 WARN_ON(!RB_EMPTY_NODE(&osd->o_node));
1026 WARN_ON(!list_empty(&osd->o_requests));
1027 WARN_ON(!list_empty(&osd->o_linger_requests));
1028 WARN_ON(!list_empty(&osd->o_osd_lru));
1029 WARN_ON(!list_empty(&osd->o_keepalive_item));
1030
1031 if (osd->o_auth.authorizer) {
1032 WARN_ON(osd_homeless(osd));
1033 ceph_auth_destroy_authorizer(osd->o_auth.authorizer);
1034 }
1035}
1036
f24e9980
SW		 1037/*
1038 * Track open sessions with osds.
1039 */
e10006f8 1040static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
f24e9980
SW		 1041{
1042 struct ceph_osd *osd;
1043
0247a0cf
ID		 1044 WARN_ON(onum == CEPH_HOMELESS_OSD);
1045
f24e9980
SW		 1046 osd = kzalloc(sizeof(*osd), GFP_NOFS);
1047 if (!osd)
1048 return NULL;
1049
0247a0cf 1050 osd_init(osd);
f24e9980 1051 osd->o_osdc = osdc;
e10006f8 1052 osd->o_osd = onum;
f24e9980 1053
b7a9e5dd 1054 ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
4e7a5dcd 1055
f24e9980
SW		 1056 return osd;
1057}
1058
1059static struct ceph_osd *get_osd(struct ceph_osd *osd)
1060{
1061 if (atomic_inc_not_zero(&osd->o_ref)) {
1062 dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
1063 atomic_read(&osd->o_ref));
1064 return osd;
1065 } else {
1066 dout("get_osd %p FAIL\n", osd);
1067 return NULL;
1068 }
1069}
1070
1071static void put_osd(struct ceph_osd *osd)
1072{
1073 dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
1074 atomic_read(&osd->o_ref) - 1);
b28ec2f3 1075 if (atomic_dec_and_test(&osd->o_ref)) {
0247a0cf 1076 osd_cleanup(osd);
f24e9980 1077 kfree(osd);
79494d1b 1078 }
f24e9980
SW		 1079}
1080
fcd00b68
ID		 1081DEFINE_RB_FUNCS(osd, struct ceph_osd, o_osd, o_node)
1082
f24e9980
SW		 1083/*
1084 * remove an osd from our map
1085 */
f5a2041b 1086static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
f24e9980 1087{
7eb71e03 1088 dout("%s %p osd%d\n", __func__, osd, osd->o_osd);
cc9f1f51
ID		 1089 WARN_ON(!list_empty(&osd->o_requests));
1090 WARN_ON(!list_empty(&osd->o_linger_requests));
7c6e6fc5 1091
f5a2041b 1092 list_del_init(&osd->o_osd_lru);
fcd00b68 1093 erase_osd(&osdc->osds, osd);
7eb71e03
ID		 1094}
1095
1096static void remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
1097{
1098 dout("%s %p osd%d\n", __func__, osd, osd->o_osd);
1099
1100 if (!RB_EMPTY_NODE(&osd->o_node)) {
1101 ceph_con_close(&osd->o_con);
1102 __remove_osd(osdc, osd);
1103 put_osd(osd);
1104 }
f24e9980
SW		 1105}
1106
f5a2041b
YS		 1107static void __move_osd_to_lru(struct ceph_osd_client *osdc,
1108 struct ceph_osd *osd)
1109{
bbf37ec3 1110 dout("%s %p\n", __func__, osd);
f5a2041b 1111 BUG_ON(!list_empty(&osd->o_osd_lru));
bbf37ec3 1112
f5a2041b 1113 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
a319bf56 1114 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl;
f5a2041b
YS		 1115}
1116
bbf37ec3
ID		 1117static void maybe_move_osd_to_lru(struct ceph_osd_client *osdc,
1118 struct ceph_osd *osd)
1119{
1120 dout("%s %p\n", __func__, osd);
1121
1122 if (list_empty(&osd->o_requests) &&
1123 list_empty(&osd->o_linger_requests))
1124 __move_osd_to_lru(osdc, osd);
1125}
1126
f5a2041b
YS		 1127static void __remove_osd_from_lru(struct ceph_osd *osd)
1128{
1129 dout("__remove_osd_from_lru %p\n", osd);
1130 if (!list_empty(&osd->o_osd_lru))
1131 list_del_init(&osd->o_osd_lru);
1132}
1133
f24e9980
SW		 1134/*
1135 * reset osd connect
1136 */
f5a2041b 1137static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
f24e9980 1138{
c3acb181 1139 struct ceph_entity_addr *peer_addr;
f24e9980 1140
f5a2041b 1141 dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
a40c4f10
YS		 1142 if (list_empty(&osd->o_requests) &&
1143 list_empty(&osd->o_linger_requests)) {
7eb71e03 1144 remove_osd(osdc, osd);
c3acb181
AE		 1145 return -ENODEV;
1146 }
1147
1148 peer_addr = &osdc->osdmap->osd_addr[osd->o_osd];
1149 if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1150 !ceph_con_opened(&osd->o_con)) {
1151 struct ceph_osd_request *req;
1152
0b4af2e8
ID		 1153 dout("osd addr hasn't changed and connection never opened, "
1154 "letting msgr retry\n");
87b315a5
SW		 1155 /* touch each r_stamp for handle_timeout()'s benfit */
1156 list_for_each_entry(req, &osd->o_requests, r_osd_item)
1157 req->r_stamp = jiffies;
c3acb181
AE		 1158
1159 return -EAGAIN;
f24e9980 1160 }
c3acb181
AE		 1161
1162 ceph_con_close(&osd->o_con);
1163 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1164 osd->o_incarnation++;
1165
1166 return 0;
f24e9980
SW		 1167}
1168
f24e9980
SW		 1169/*
1170 * Register request, assign tid. If this is the first request, set up
1171 * the timeout event.
1172 */
a40c4f10
YS		 1173static void __register_request(struct ceph_osd_client *osdc,
1174 struct ceph_osd_request *req)
f24e9980 1175{
f24e9980 1176 req->r_tid = ++osdc->last_tid;
6df058c0 1177 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
77f38e0e 1178 dout("__register_request %p tid %lld\n", req, req->r_tid);
fcd00b68 1179 insert_request(&osdc->requests, req);
f24e9980
SW		 1180 ceph_osdc_get_request(req);
1181 osdc->num_requests++;
a40c4f10
YS		 1182}
1183
f24e9980
SW		 1184/*
1185 * called under osdc->request_mutex
1186 */
1187static void __unregister_request(struct ceph_osd_client *osdc,
1188 struct ceph_osd_request *req)
1189{
35f9f8a0
SW		 1190 if (RB_EMPTY_NODE(&req->r_node)) {
1191 dout("__unregister_request %p tid %lld not registered\n",
1192 req, req->r_tid);
1193 return;
1194 }
1195
f24e9980 1196 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
fcd00b68 1197 erase_request(&osdc->requests, req);
f24e9980
SW		 1198 osdc->num_requests--;
1199
0ba6478d
SW		 1200 if (req->r_osd) {
1201 /* make sure the original request isn't in flight. */
6740a845 1202 ceph_msg_revoke(req->r_request);
0ba6478d
SW		 1203
1204 list_del_init(&req->r_osd_item);
bbf37ec3 1205 maybe_move_osd_to_lru(osdc, req->r_osd);
4f23409e 1206 if (list_empty(&req->r_linger_osd_item))
a40c4f10 1207 req->r_osd = NULL;
0ba6478d 1208 }
f24e9980 1209
7d5f2481 1210 list_del_init(&req->r_req_lru_item);
f24e9980 1211 ceph_osdc_put_request(req);
f24e9980
SW		 1212}
1213
1214/*
1215 * Cancel a previously queued request message
1216 */
1217static void __cancel_request(struct ceph_osd_request *req)
1218{
6bc18876 1219 if (req->r_sent && req->r_osd) {
6740a845 1220 ceph_msg_revoke(req->r_request);
f24e9980
SW		 1221 req->r_sent = 0;
1222 }
1223}
1224
a40c4f10
YS		 1225static void __register_linger_request(struct ceph_osd_client *osdc,
1226 struct ceph_osd_request *req)
1227{
af593064
ID		 1228 dout("%s %p tid %llu\n", __func__, req, req->r_tid);
1229 WARN_ON(!req->r_linger);
1230
96e4dac6 1231 ceph_osdc_get_request(req);
a40c4f10 1232 list_add_tail(&req->r_linger_item, &osdc->req_linger);
6194ea89 1233 if (req->r_osd)
1d0326b1 1234 list_add_tail(&req->r_linger_osd_item,
6194ea89 1235 &req->r_osd->o_linger_requests);
a40c4f10
YS		 1236}
1237
1238static void __unregister_linger_request(struct ceph_osd_client *osdc,
1239 struct ceph_osd_request *req)
1240{
af593064
ID		 1241 WARN_ON(!req->r_linger);
1242
1243 if (list_empty(&req->r_linger_item)) {
1244 dout("%s %p tid %llu not registered\n", __func__, req,
1245 req->r_tid);
1246 return;
1247 }
1248
1249 dout("%s %p tid %llu\n", __func__, req, req->r_tid);
61c74035 1250 list_del_init(&req->r_linger_item);
af593064 1251
a40c4f10 1252 if (req->r_osd) {
1d0326b1 1253 list_del_init(&req->r_linger_osd_item);
bbf37ec3 1254 maybe_move_osd_to_lru(osdc, req->r_osd);
fbdb9190
SW		 1255 if (list_empty(&req->r_osd_item))
1256 req->r_osd = NULL;
a40c4f10 1257 }
96e4dac6 1258 ceph_osdc_put_request(req);
a40c4f10
YS		 1259}
1260
a40c4f10
YS		 1261void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
1262 struct ceph_osd_request *req)
1263{
1264 if (!req->r_linger) {
1265 dout("set_request_linger %p\n", req);
1266 req->r_linger = 1;
a40c4f10
YS		 1267 }
1268}
1269EXPORT_SYMBOL(ceph_osdc_set_request_linger);
1270
63244fa1
ID		 1271static bool __pool_full(struct ceph_pg_pool_info *pi)
1272{
1273 return pi->flags & CEPH_POOL_FLAG_FULL;
1274}
1275
42c1b124
ID		 1276static bool have_pool_full(struct ceph_osd_client *osdc)
1277{
1278 struct rb_node *n;
1279
1280 for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) {
1281 struct ceph_pg_pool_info *pi =
1282 rb_entry(n, struct ceph_pg_pool_info, node);
1283
1284 if (__pool_full(pi))
1285 return true;
1286 }
1287
1288 return false;
1289}
1290
d29adb34
JD		 1291/*
1292 * Returns whether a request should be blocked from being sent
1293 * based on the current osdmap and osd_client settings.
1294 *
1295 * Caller should hold map_sem for read.
1296 */
63244fa1
ID		 1297static bool target_should_be_paused(struct ceph_osd_client *osdc,
1298 const struct ceph_osd_request_target *t,
1299 struct ceph_pg_pool_info *pi)
1300{
1301 bool pauserd = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD);
1302 bool pausewr = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR) ||
1303 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) ||
1304 __pool_full(pi);
1305
1306 WARN_ON(pi->id != t->base_oloc.pool);
1307 return (t->flags & CEPH_OSD_FLAG_READ && pauserd) ||
1308 (t->flags & CEPH_OSD_FLAG_WRITE && pausewr);
1309}
1310
63244fa1
ID		 1311enum calc_target_result {
1312 CALC_TARGET_NO_ACTION = 0,
1313 CALC_TARGET_NEED_RESEND,
1314 CALC_TARGET_POOL_DNE,
1315};
1316
1317static enum calc_target_result calc_target(struct ceph_osd_client *osdc,
1318 struct ceph_osd_request_target *t,
1319 u32 *last_force_resend,
1320 bool any_change)
1321{
1322 struct ceph_pg_pool_info *pi;
1323 struct ceph_pg pgid, last_pgid;
1324 struct ceph_osds up, acting;
1325 bool force_resend = false;
1326 bool need_check_tiering = false;
1327 bool need_resend = false;
1328 bool sort_bitwise = ceph_osdmap_flag(osdc->osdmap,
1329 CEPH_OSDMAP_SORTBITWISE);
1330 enum calc_target_result ct_res;
1331 int ret;
1332
1333 pi = ceph_pg_pool_by_id(osdc->osdmap, t->base_oloc.pool);
1334 if (!pi) {
1335 t->osd = CEPH_HOMELESS_OSD;
1336 ct_res = CALC_TARGET_POOL_DNE;
1337 goto out;
1338 }
1339
1340 if (osdc->osdmap->epoch == pi->last_force_request_resend) {
1341 if (last_force_resend &&
1342 *last_force_resend < pi->last_force_request_resend) {
1343 *last_force_resend = pi->last_force_request_resend;
1344 force_resend = true;
1345 } else if (!last_force_resend) {
1346 force_resend = true;
1347 }
1348 }
1349 if (ceph_oid_empty(&t->target_oid) || force_resend) {
1350 ceph_oid_copy(&t->target_oid, &t->base_oid);
1351 need_check_tiering = true;
1352 }
1353 if (ceph_oloc_empty(&t->target_oloc) || force_resend) {
1354 ceph_oloc_copy(&t->target_oloc, &t->base_oloc);
1355 need_check_tiering = true;
1356 }
1357
1358 if (need_check_tiering &&
1359 (t->flags & CEPH_OSD_FLAG_IGNORE_OVERLAY) == 0) {
1360 if (t->flags & CEPH_OSD_FLAG_READ && pi->read_tier >= 0)
1361 t->target_oloc.pool = pi->read_tier;
1362 if (t->flags & CEPH_OSD_FLAG_WRITE && pi->write_tier >= 0)
1363 t->target_oloc.pool = pi->write_tier;
1364 }
1365
1366 ret = ceph_object_locator_to_pg(osdc->osdmap, &t->target_oid,
1367 &t->target_oloc, &pgid);
1368 if (ret) {
1369 WARN_ON(ret != -ENOENT);
1370 t->osd = CEPH_HOMELESS_OSD;
1371 ct_res = CALC_TARGET_POOL_DNE;
1372 goto out;
1373 }
1374 last_pgid.pool = pgid.pool;
1375 last_pgid.seed = ceph_stable_mod(pgid.seed, t->pg_num, t->pg_num_mask);
1376
1377 ceph_pg_to_up_acting_osds(osdc->osdmap, &pgid, &up, &acting);
1378 if (any_change &&
1379 ceph_is_new_interval(&t->acting,
1380 &acting,
1381 &t->up,
1382 &up,
1383 t->size,
1384 pi->size,
1385 t->min_size,
1386 pi->min_size,
1387 t->pg_num,
1388 pi->pg_num,
1389 t->sort_bitwise,
1390 sort_bitwise,
1391 &last_pgid))
1392 force_resend = true;
1393
1394 if (t->paused && !target_should_be_paused(osdc, t, pi)) {
1395 t->paused = false;
1396 need_resend = true;
1397 }
1398
1399 if (ceph_pg_compare(&t->pgid, &pgid) ||
1400 ceph_osds_changed(&t->acting, &acting, any_change) ||
1401 force_resend) {
1402 t->pgid = pgid; /* struct */
1403 ceph_osds_copy(&t->acting, &acting);
1404 ceph_osds_copy(&t->up, &up);
1405 t->size = pi->size;
1406 t->min_size = pi->min_size;
1407 t->pg_num = pi->pg_num;
1408 t->pg_num_mask = pi->pg_num_mask;
1409 t->sort_bitwise = sort_bitwise;
1410
1411 t->osd = acting.primary;
1412 need_resend = true;
1413 }
1414
1415 ct_res = need_resend ? CALC_TARGET_NEED_RESEND : CALC_TARGET_NO_ACTION;
1416out:
1417 dout("%s t %p -> ct_res %d osd %d\n", __func__, t, ct_res, t->osd);
1418 return ct_res;
1419}
1420
f671b581
ID		 1421static void __enqueue_request(struct ceph_osd_request *req)
1422{
1423 struct ceph_osd_client *osdc = req->r_osdc;
1424
1425 dout("%s %p tid %llu to osd%d\n", __func__, req, req->r_tid,
1426 req->r_osd ? req->r_osd->o_osd : -1);
1427
1428 if (req->r_osd) {
1429 __remove_osd_from_lru(req->r_osd);
1430 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
1431 list_move_tail(&req->r_req_lru_item, &osdc->req_unsent);
1432 } else {
1433 list_move_tail(&req->r_req_lru_item, &osdc->req_notarget);
1434 }
1435}
1436
f24e9980
SW		 1437/*
1438 * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
1439 * (as needed), and set the request r_osd appropriately. If there is
25985edc 1440 * no up osd, set r_osd to NULL. Move the request to the appropriate list
6f6c7006 1441 * (unsent, homeless) or leave on in-flight lru.
f24e9980
SW		 1442 *
1443 * Return 0 if unchanged, 1 if changed, or negative on error.
1444 *
1445 * Caller should hold map_sem for read and request_mutex.
1446 */
6f6c7006 1447static int __map_request(struct ceph_osd_client *osdc,
38d6453c 1448 struct ceph_osd_request *req, int force_resend)
f24e9980 1449{
a66dd383 1450 enum calc_target_result ct_res;
f24e9980 1451 int err;
f24e9980 1452
6f6c7006 1453 dout("map_request %p tid %lld\n", req, req->r_tid);
17a13e40 1454
a66dd383
ID		 1455 ct_res = calc_target(osdc, &req->r_t, NULL, force_resend);
1456 switch (ct_res) {
1457 case CALC_TARGET_POOL_DNE:
6f6c7006 1458 list_move(&req->r_req_lru_item, &osdc->req_notarget);
a66dd383
ID		 1459 return -EIO;
1460 case CALC_TARGET_NO_ACTION:
f24e9980 1461 return 0; /* no change */
a66dd383
ID		 1462 default:
1463 BUG_ON(ct_res != CALC_TARGET_NEED_RESEND);
1464 }
f24e9980 1465
5b191d99 1466 dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n",
a66dd383 1467 req->r_tid, req->r_t.pgid.pool, req->r_t.pgid.seed, req->r_t.osd,
f24e9980
SW		 1468 req->r_osd ? req->r_osd->o_osd : -1);
1469
1470 if (req->r_osd) {
1471 __cancel_request(req);
1472 list_del_init(&req->r_osd_item);
a390de02 1473 list_del_init(&req->r_linger_osd_item);
f24e9980
SW		 1474 req->r_osd = NULL;
1475 }
1476
a66dd383
ID		 1477 req->r_osd = lookup_osd(&osdc->osds, req->r_t.osd);
1478 if (!req->r_osd && req->r_t.osd >= 0) {
c99eb1c7 1479 err = -ENOMEM;
a66dd383 1480 req->r_osd = create_osd(osdc, req->r_t.osd);
6f6c7006
SW		 1481 if (!req->r_osd) {
1482 list_move(&req->r_req_lru_item, &osdc->req_notarget);
c99eb1c7 1483 goto out;
6f6c7006 1484 }
f24e9980 1485
6f3bfd45 1486 dout("map_request osd %p is osd%d\n", req->r_osd,
a66dd383 1487 req->r_osd->o_osd);
fcd00b68 1488 insert_osd(&osdc->osds, req->r_osd);
f24e9980 1489
b7a9e5dd 1490 ceph_con_open(&req->r_osd->o_con,
a66dd383
ID		 1491 CEPH_ENTITY_TYPE_OSD, req->r_osd->o_osd,
1492 &osdc->osdmap->osd_addr[req->r_osd->o_osd]);
f24e9980
SW		 1493 }
1494
f671b581 1495 __enqueue_request(req);
d85b7056 1496 err = 1; /* osd or pg changed */
f24e9980
SW		 1497
1498out:
f24e9980
SW		 1499 return err;
1500}
1501
bb873b53
ID		 1502static void setup_request_data(struct ceph_osd_request *req,
1503 struct ceph_msg *msg)
f24e9980 1504{
bb873b53
ID		 1505 u32 data_len = 0;
1506 int i;
1507
1508 if (!list_empty(&msg->data))
1509 return;
f24e9980 1510
bb873b53
ID		 1511 WARN_ON(msg->data_length);
1512 for (i = 0; i < req->r_num_ops; i++) {
1513 struct ceph_osd_req_op *op = &req->r_ops[i];
1514
1515 switch (op->op) {
1516 /* request */
1517 case CEPH_OSD_OP_WRITE:
1518 case CEPH_OSD_OP_WRITEFULL:
1519 WARN_ON(op->indata_len != op->extent.length);
1520 ceph_osdc_msg_data_add(msg, &op->extent.osd_data);
1521 break;
1522 case CEPH_OSD_OP_SETXATTR:
1523 case CEPH_OSD_OP_CMPXATTR:
1524 WARN_ON(op->indata_len != op->xattr.name_len +
1525 op->xattr.value_len);
1526 ceph_osdc_msg_data_add(msg, &op->xattr.osd_data);
1527 break;
1528
1529 /* reply */
1530 case CEPH_OSD_OP_STAT:
1531 ceph_osdc_msg_data_add(req->r_reply,
1532 &op->raw_data_in);
1533 break;
1534 case CEPH_OSD_OP_READ:
1535 ceph_osdc_msg_data_add(req->r_reply,
1536 &op->extent.osd_data);
1537 break;
1538
1539 /* both */
1540 case CEPH_OSD_OP_CALL:
1541 WARN_ON(op->indata_len != op->cls.class_len +
1542 op->cls.method_len +
1543 op->cls.indata_len);
1544 ceph_osdc_msg_data_add(msg, &op->cls.request_info);
1545 /* optional, can be NONE */
1546 ceph_osdc_msg_data_add(msg, &op->cls.request_data);
1547 /* optional, can be NONE */
1548 ceph_osdc_msg_data_add(req->r_reply,
1549 &op->cls.response_data);
1550 break;
1551 }
1552
1553 data_len += op->indata_len;
1554 }
1b83bef2 1555
bb873b53
ID		 1556 WARN_ON(data_len != msg->data_length);
1557}
1558
1559static void encode_request(struct ceph_osd_request *req, struct ceph_msg *msg)
1560{
1561 void *p = msg->front.iov_base;
1562 void *const end = p + msg->front_alloc_len;
1563 u32 data_len = 0;
1564 int i;
1565
1566 if (req->r_flags & CEPH_OSD_FLAG_WRITE) {
1567 /* snapshots aren't writeable */
1568 WARN_ON(req->r_snapid != CEPH_NOSNAP);
1569 } else {
1570 WARN_ON(req->r_mtime.tv_sec || req->r_mtime.tv_nsec ||
1571 req->r_data_offset || req->r_snapc);
1572 }
1573
1574 setup_request_data(req, msg);
1575
1576 ceph_encode_32(&p, 1); /* client_inc, always 1 */
1577 ceph_encode_32(&p, req->r_osdc->osdmap->epoch);
1578 ceph_encode_32(&p, req->r_flags);
1579 ceph_encode_timespec(p, &req->r_mtime);
1580 p += sizeof(struct ceph_timespec);
1581 /* aka reassert_version */
1582 memcpy(p, &req->r_replay_version, sizeof(req->r_replay_version));
1583 p += sizeof(req->r_replay_version);
1584
1585 /* oloc */
1586 ceph_encode_8(&p, 4);
1587 ceph_encode_8(&p, 4);
1588 ceph_encode_32(&p, 8 + 4 + 4);
1589 ceph_encode_64(&p, req->r_t.target_oloc.pool);
1590 ceph_encode_32(&p, -1); /* preferred */
1591 ceph_encode_32(&p, 0); /* key len */
1592
1593 /* pgid */
1594 ceph_encode_8(&p, 1);
a66dd383
ID		 1595 ceph_encode_64(&p, req->r_t.pgid.pool);
1596 ceph_encode_32(&p, req->r_t.pgid.seed);
bb873b53 1597 ceph_encode_32(&p, -1); /* preferred */
2169aea6 1598
bb873b53
ID		 1599 /* oid */
1600 ceph_encode_32(&p, req->r_t.target_oid.name_len);
1601 memcpy(p, req->r_t.target_oid.name, req->r_t.target_oid.name_len);
1602 p += req->r_t.target_oid.name_len;
f24e9980 1603
bb873b53
ID		 1604 /* ops, can imply data */
1605 ceph_encode_16(&p, req->r_num_ops);
1606 for (i = 0; i < req->r_num_ops; i++) {
1607 data_len += osd_req_encode_op(p, &req->r_ops[i]);
1608 p += sizeof(struct ceph_osd_op);
1609 }
26be8808 1610
bb873b53
ID		 1611 ceph_encode_64(&p, req->r_snapid); /* snapid */
1612 if (req->r_snapc) {
1613 ceph_encode_64(&p, req->r_snapc->seq);
1614 ceph_encode_32(&p, req->r_snapc->num_snaps);
1615 for (i = 0; i < req->r_snapc->num_snaps; i++)
1616 ceph_encode_64(&p, req->r_snapc->snaps[i]);
1617 } else {
1618 ceph_encode_64(&p, 0); /* snap_seq */
1619 ceph_encode_32(&p, 0); /* snaps len */
1620 }
1621
1622 ceph_encode_32(&p, req->r_attempts); /* retry_attempt */
1623
1624 BUG_ON(p > end);
1625 msg->front.iov_len = p - msg->front.iov_base;
1626 msg->hdr.version = cpu_to_le16(4); /* MOSDOp v4 */
1627 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1628 msg->hdr.data_len = cpu_to_le32(data_len);
1629 /*
1630 * The header "data_off" is a hint to the receiver allowing it
1631 * to align received data into its buffers such that there's no
1632 * need to re-copy it before writing it to disk (direct I/O).
1633 */
1634 msg->hdr.data_off = cpu_to_le16(req->r_data_offset);
26be8808 1635
bb873b53
ID		 1636 dout("%s req %p oid %*pE oid_len %d front %zu data %u\n", __func__,
1637 req, req->r_t.target_oid.name_len, req->r_t.target_oid.name,
1638 req->r_t.target_oid.name_len, msg->front.iov_len, data_len);
1639}
1640
1641/*
1642 * @req has to be assigned a tid and registered.
1643 */
1644static void send_request(struct ceph_osd_request *req)
1645{
1646 struct ceph_osd *osd = req->r_osd;
1647
1648 WARN_ON(osd->o_osd != req->r_t.osd);
1649
1650 req->r_flags |= CEPH_OSD_FLAG_KNOWN_REDIR;
1651 if (req->r_attempts)
1652 req->r_flags |= CEPH_OSD_FLAG_RETRY;
1653 else
1654 WARN_ON(req->r_flags & CEPH_OSD_FLAG_RETRY);
1655
1656 encode_request(req, req->r_request);
1657
1658 dout("%s req %p tid %llu to pg %llu.%x osd%d flags 0x%x attempt %d\n",
1659 __func__, req, req->r_tid, req->r_t.pgid.pool, req->r_t.pgid.seed,
1660 req->r_t.osd, req->r_flags, req->r_attempts);
1661
1662 req->r_t.paused = false;
1663 req->r_stamp = jiffies;
1664 req->r_attempts++;
1665
1666 req->r_sent = osd->o_incarnation;
1667 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
1668 ceph_con_send(&osd->o_con, ceph_msg_get(req->r_request));
f24e9980
SW		 1669}
1670
6f6c7006
SW		 1671/*
1672 * Send any requests in the queue (req_unsent).
1673 */
f9d25199 1674static void __send_queued(struct ceph_osd_client *osdc)
6f6c7006
SW		 1675{
1676 struct ceph_osd_request *req, *tmp;
1677
f9d25199 1678 dout("__send_queued\n");
bb873b53
ID		 1679 list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item) {
1680 list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
1681 send_request(req);
1682 }
6f6c7006
SW		 1683}
1684
42c1b124
ID		 1685static void maybe_request_map(struct ceph_osd_client *osdc)
1686{
1687 bool continuous = false;
1688
1689 WARN_ON(!osdc->osdmap->epoch);
1690
1691 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) ||
1692 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD) ||
1693 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR)) {
1694 dout("%s osdc %p continuous\n", __func__, osdc);
1695 continuous = true;
1696 } else {
1697 dout("%s osdc %p onetime\n", __func__, osdc);
1698 }
1699
1700 if (ceph_monc_want_map(&osdc->client->monc, CEPH_SUB_OSDMAP,
1701 osdc->osdmap->epoch + 1, continuous))
1702 ceph_monc_renew_subs(&osdc->client->monc);
1703}
1704
0bbfdfe8
ID		 1705/*
1706 * Caller should hold map_sem for read and request_mutex.
1707 */
1708static int __ceph_osdc_start_request(struct ceph_osd_client *osdc,
1709 struct ceph_osd_request *req,
1710 bool nofail)
1711{
1712 int rc;
1713
1714 __register_request(osdc, req);
1715 req->r_sent = 0;
1716 req->r_got_reply = 0;
1717 rc = __map_request(osdc, req, 0);
1718 if (rc < 0) {
1719 if (nofail) {
1720 dout("osdc_start_request failed map, "
1721 " will retry %lld\n", req->r_tid);
1722 rc = 0;
1723 } else {
1724 __unregister_request(osdc, req);
1725 }
1726 return rc;
1727 }
1728
1729 if (req->r_osd == NULL) {
1730 dout("send_request %p no up osds in pg\n", req);
1731 ceph_monc_request_next_osdmap(&osdc->client->monc);
1732 } else {
1733 __send_queued(osdc);
1734 }
1735
1736 return 0;
1737}
1738
fe5da05e
ID		 1739static void __complete_request(struct ceph_osd_request *req)
1740{
1741 if (req->r_callback)
1742 req->r_callback(req);
1743 else
1744 complete_all(&req->r_completion);
1745}
1746
f24e9980 1747/*
fbca9635
ID		 1748 * Timeout callback, called every N seconds. When 1 or more OSD
1749 * requests has been active for more than N seconds, we send a keepalive
1750 * (tag + timestamp) to its OSD to ensure any communications channel
1751 * reset is detected.
f24e9980
SW		 1752 */
1753static void handle_timeout(struct work_struct *work)
1754{
1755 struct ceph_osd_client *osdc =
1756 container_of(work, struct ceph_osd_client, timeout_work.work);
a319bf56 1757 struct ceph_options *opts = osdc->client->options;
83aff95e 1758 struct ceph_osd_request *req;
f24e9980 1759 struct ceph_osd *osd;
422d2cb8 1760 struct list_head slow_osds;
f24e9980
SW		 1761 dout("timeout\n");
1762 down_read(&osdc->map_sem);
1763
1764 ceph_monc_request_next_osdmap(&osdc->client->monc);
1765
1766 mutex_lock(&osdc->request_mutex);
f24e9980 1767
422d2cb8
YS		 1768 /*
1769 * ping osds that are a bit slow. this ensures that if there
1770 * is a break in the TCP connection we will notice, and reopen
1771 * a connection with that osd (from the fault callback).
1772 */
1773 INIT_LIST_HEAD(&slow_osds);
1774 list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
a319bf56
ID		 1775 if (time_before(jiffies,
1776 req->r_stamp + opts->osd_keepalive_timeout))
422d2cb8
YS		 1777 break;
1778
1779 osd = req->r_osd;
1780 BUG_ON(!osd);
1781 dout(" tid %llu is slow, will send keepalive on osd%d\n",
f24e9980 1782 req->r_tid, osd->o_osd);
422d2cb8
YS		 1783 list_move_tail(&osd->o_keepalive_item, &slow_osds);
1784 }
1785 while (!list_empty(&slow_osds)) {
1786 osd = list_entry(slow_osds.next, struct ceph_osd,
1787 o_keepalive_item);
1788 list_del_init(&osd->o_keepalive_item);
f24e9980
SW		 1789 ceph_con_keepalive(&osd->o_con);
1790 }
1791
f9d25199 1792 __send_queued(osdc);
f24e9980 1793 mutex_unlock(&osdc->request_mutex);
f24e9980 1794 up_read(&osdc->map_sem);
fbca9635
ID		 1795
1796 schedule_delayed_work(&osdc->timeout_work,
1797 osdc->client->options->osd_keepalive_timeout);
f24e9980
SW		 1798}
1799
f5a2041b
YS		 1800static void handle_osds_timeout(struct work_struct *work)
1801{
1802 struct ceph_osd_client *osdc =
1803 container_of(work, struct ceph_osd_client,
1804 osds_timeout_work.work);
a319bf56 1805 unsigned long delay = osdc->client->options->osd_idle_ttl / 4;
42a2c09f 1806 struct ceph_osd *osd, *nosd;
f5a2041b 1807
42a2c09f 1808 dout("%s osdc %p\n", __func__, osdc);
f5a2041b 1809 down_read(&osdc->map_sem);
42a2c09f
ID		 1810 mutex_lock(&osdc->request_mutex);
1811
1812 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
1813 if (time_before(jiffies, osd->lru_ttl))
1814 break;
1815
1816 remove_osd(osdc, osd);
1817 }
f5a2041b 1818
42a2c09f
ID		 1819 mutex_unlock(&osdc->request_mutex);
1820 up_read(&osdc->map_sem);
f5a2041b
YS		 1821 schedule_delayed_work(&osdc->osds_timeout_work,
1822 round_jiffies_relative(delay));
1823}
1824
205ee118
ID		 1825static int ceph_oloc_decode(void **p, void *end,
1826 struct ceph_object_locator *oloc)
1827{
1828 u8 struct_v, struct_cv;
1829 u32 len;
1830 void *struct_end;
1831 int ret = 0;
1832
1833 ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
1834 struct_v = ceph_decode_8(p);
1835 struct_cv = ceph_decode_8(p);
1836 if (struct_v < 3) {
1837 pr_warn("got v %d < 3 cv %d of ceph_object_locator\n",
1838 struct_v, struct_cv);
1839 goto e_inval;
1840 }
1841 if (struct_cv > 6) {
1842 pr_warn("got v %d cv %d > 6 of ceph_object_locator\n",
1843 struct_v, struct_cv);
1844 goto e_inval;
1845 }
1846 len = ceph_decode_32(p);
1847 ceph_decode_need(p, end, len, e_inval);
1848 struct_end = *p + len;
1849
1850 oloc->pool = ceph_decode_64(p);
1851 *p += 4; /* skip preferred */
1852
1853 len = ceph_decode_32(p);
1854 if (len > 0) {
1855 pr_warn("ceph_object_locator::key is set\n");
1856 goto e_inval;
1857 }
1858
1859 if (struct_v >= 5) {
1860 len = ceph_decode_32(p);
1861 if (len > 0) {
1862 pr_warn("ceph_object_locator::nspace is set\n");
1863 goto e_inval;
1864 }
1865 }
1866
1867 if (struct_v >= 6) {
1868 s64 hash = ceph_decode_64(p);
1869 if (hash != -1) {
1870 pr_warn("ceph_object_locator::hash is set\n");
1871 goto e_inval;
1872 }
1873 }
1874
1875 /* skip the rest */
1876 *p = struct_end;
1877out:
1878 return ret;
1879
1880e_inval:
1881 ret = -EINVAL;
1882 goto out;
1883}
1884
1885static int ceph_redirect_decode(void **p, void *end,
1886 struct ceph_request_redirect *redir)
1887{
1888 u8 struct_v, struct_cv;
1889 u32 len;
1890 void *struct_end;
1891 int ret;
1892
1893 ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
1894 struct_v = ceph_decode_8(p);
1895 struct_cv = ceph_decode_8(p);
1896 if (struct_cv > 1) {
1897 pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n",
1898 struct_v, struct_cv);
1899 goto e_inval;
1900 }
1901 len = ceph_decode_32(p);
1902 ceph_decode_need(p, end, len, e_inval);
1903 struct_end = *p + len;
1904
1905 ret = ceph_oloc_decode(p, end, &redir->oloc);
1906 if (ret)
1907 goto out;
1908
1909 len = ceph_decode_32(p);
1910 if (len > 0) {
1911 pr_warn("ceph_request_redirect::object_name is set\n");
1912 goto e_inval;
1913 }
1914
1915 len = ceph_decode_32(p);
1916 *p += len; /* skip osd_instructions */
1917
1918 /* skip the rest */
1919 *p = struct_end;
1920out:
1921 return ret;
1922
1923e_inval:
1924 ret = -EINVAL;
1925 goto out;
1926}
1927
fe5da05e
ID		 1928struct MOSDOpReply {
1929 struct ceph_pg pgid;
1930 u64 flags;
1931 int result;
1932 u32 epoch;
1933 int num_ops;
1934 u32 outdata_len[CEPH_OSD_MAX_OPS];
1935 s32 rval[CEPH_OSD_MAX_OPS];
1936 int retry_attempt;
1937 struct ceph_eversion replay_version;
1938 u64 user_version;
1939 struct ceph_request_redirect redirect;
1940};
25845472 1941
fe5da05e 1942static int decode_MOSDOpReply(const struct ceph_msg *msg, struct MOSDOpReply *m)
f24e9980 1943{
fe5da05e
ID		 1944 void *p = msg->front.iov_base;
1945 void *const end = p + msg->front.iov_len;
1946 u16 version = le16_to_cpu(msg->hdr.version);
1947 struct ceph_eversion bad_replay_version;
b0b31a8f 1948 u8 decode_redir;
fe5da05e
ID		 1949 u32 len;
1950 int ret;
1951 int i;
1b83bef2 1952
fe5da05e
ID		 1953 ceph_decode_32_safe(&p, end, len, e_inval);
1954 ceph_decode_need(&p, end, len, e_inval);
1955 p += len; /* skip oid */
1b83bef2 1956
fe5da05e
ID		 1957 ret = ceph_decode_pgid(&p, end, &m->pgid);
1958 if (ret)
1959 return ret;
1b83bef2 1960
fe5da05e
ID		 1961 ceph_decode_64_safe(&p, end, m->flags, e_inval);
1962 ceph_decode_32_safe(&p, end, m->result, e_inval);
1963 ceph_decode_need(&p, end, sizeof(bad_replay_version), e_inval);
1964 memcpy(&bad_replay_version, p, sizeof(bad_replay_version));
1965 p += sizeof(bad_replay_version);
1966 ceph_decode_32_safe(&p, end, m->epoch, e_inval);
1b83bef2 1967
fe5da05e
ID		 1968 ceph_decode_32_safe(&p, end, m->num_ops, e_inval);
1969 if (m->num_ops > ARRAY_SIZE(m->outdata_len))
1970 goto e_inval;
1b83bef2 1971
fe5da05e
ID		 1972 ceph_decode_need(&p, end, m->num_ops * sizeof(struct ceph_osd_op),
1973 e_inval);
1974 for (i = 0; i < m->num_ops; i++) {
1b83bef2 1975 struct ceph_osd_op *op = p;
1b83bef2 1976
fe5da05e 1977 m->outdata_len[i] = le32_to_cpu(op->payload_len);
1b83bef2
SW		 1978 p += sizeof(*op);
1979 }
1b83bef2 1980
fe5da05e
ID		 1981 ceph_decode_32_safe(&p, end, m->retry_attempt, e_inval);
1982 for (i = 0; i < m->num_ops; i++)
1983 ceph_decode_32_safe(&p, end, m->rval[i], e_inval);
f24e9980 1984
fe5da05e
ID		 1985 if (version >= 5) {
1986 ceph_decode_need(&p, end, sizeof(m->replay_version), e_inval);
1987 memcpy(&m->replay_version, p, sizeof(m->replay_version));
1988 p += sizeof(m->replay_version);
1989 ceph_decode_64_safe(&p, end, m->user_version, e_inval);
1990 } else {
1991 m->replay_version = bad_replay_version; /* struct */
1992 m->user_version = le64_to_cpu(m->replay_version.version);
1993 }
eb845ff1 1994
fe5da05e
ID		 1995 if (version >= 6) {
1996 if (version >= 7)
1997 ceph_decode_8_safe(&p, end, decode_redir, e_inval);
b0b31a8f
ID		 1998 else
1999 decode_redir = 1;
2000 } else {
2001 decode_redir = 0;
2002 }
2003
2004 if (decode_redir) {
fe5da05e
ID		 2005 ret = ceph_redirect_decode(&p, end, &m->redirect);
2006 if (ret)
2007 return ret;
205ee118 2008 } else {
fe5da05e 2009 ceph_oloc_init(&m->redirect.oloc);
205ee118 2010 }
f24e9980 2011
fe5da05e
ID		 2012 return 0;
2013
2014e_inval:
2015 return -EINVAL;
2016}
2017
2018/*
2019 * We are done with @req if
2020 * - @m is a safe reply, or
2021 * - @m is an unsafe reply and we didn't want a safe one
2022 */
2023static bool done_request(const struct ceph_osd_request *req,
2024 const struct MOSDOpReply *m)
2025{
2026 return (m->result < 0 ||
2027 (m->flags & CEPH_OSD_FLAG_ONDISK) ||
2028 !(req->r_flags & CEPH_OSD_FLAG_ONDISK));
2029}
205ee118 2030
fe5da05e
ID		 2031/*
2032 * handle osd op reply. either call the callback if it is specified,
2033 * or do the completion to wake up the waiting thread.
2034 *
2035 * ->r_unsafe_callback is set? yes no
2036 *
2037 * first reply is OK (needed r_cb/r_completion, r_cb/r_completion,
2038 * any or needed/got safe) r_safe_completion r_safe_completion
2039 *
2040 * first reply is unsafe r_unsafe_cb(true) (nothing)
2041 *
2042 * when we get the safe reply r_unsafe_cb(false), r_cb/r_completion,
2043 * r_safe_completion r_safe_completion
2044 */
2045static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg)
2046{
2047 struct ceph_osd_request *req;
2048 struct MOSDOpReply m;
2049 u64 tid = le64_to_cpu(msg->hdr.tid);
2050 u32 data_len = 0;
2051 bool already_acked;
2052 int ret;
2053 int i;
2054
2055 dout("%s msg %p tid %llu\n", __func__, msg, tid);
2056
2057 down_read(&osdc->map_sem);
2058 mutex_lock(&osdc->request_mutex);
2059 req = lookup_request(&osdc->requests, tid);
2060 if (!req) {
2061 dout("%s no tid %llu\n", __func__, tid);
2062 goto out_unlock;
2063 }
2064 ceph_osdc_get_request(req);
2065
2066 ret = decode_MOSDOpReply(msg, &m);
2067 if (ret) {
2068 pr_err("failed to decode MOSDOpReply for tid %llu: %d\n",
2069 req->r_tid, ret);
2070 ceph_msg_dump(msg);
2071 goto fail_request;
2072 }
2073 dout("%s req %p tid %llu flags 0x%llx pgid %llu.%x epoch %u attempt %d v %u'%llu uv %llu\n",
2074 __func__, req, req->r_tid, m.flags, m.pgid.pool, m.pgid.seed,
2075 m.epoch, m.retry_attempt, le32_to_cpu(m.replay_version.epoch),
2076 le64_to_cpu(m.replay_version.version), m.user_version);
2077
2078 if (m.retry_attempt >= 0) {
2079 if (m.retry_attempt != req->r_attempts - 1) {
2080 dout("req %p tid %llu retry_attempt %d != %d, ignoring\n",
2081 req, req->r_tid, m.retry_attempt,
2082 req->r_attempts - 1);
2083 goto out_put;
2084 }
2085 } else {
2086 WARN_ON(1); /* MOSDOpReply v4 is assumed */
2087 }
2088
2089 if (!ceph_oloc_empty(&m.redirect.oloc)) {
2090 dout("req %p tid %llu redirect pool %lld\n", req, req->r_tid,
2091 m.redirect.oloc.pool);
205ee118 2092 __unregister_request(osdc, req);
205ee118 2093
fe5da05e 2094 ceph_oloc_copy(&req->r_t.target_oloc, &m.redirect.oloc);
205ee118
ID		 2095
2096 /*
2097 * Start redirect requests with nofail=true. If
2098 * mapping fails, request will end up on the notarget
2099 * list, waiting for the new osdmap (which can take
2100 * a while), even though the original request mapped
2101 * successfully. In the future we might want to follow
2102 * original request's nofail setting here.
2103 */
fe5da05e
ID		 2104 ret = __ceph_osdc_start_request(osdc, req, true);
2105 BUG_ON(ret);
205ee118 2106
fe5da05e 2107 goto out_put;
205ee118
ID		 2108 }
2109
fe5da05e
ID		 2110 if (m.num_ops != req->r_num_ops) {
2111 pr_err("num_ops %d != %d for tid %llu\n", m.num_ops,
2112 req->r_num_ops, req->r_tid);
2113 goto fail_request;
f24e9980 2114 }
fe5da05e
ID		 2115 for (i = 0; i < req->r_num_ops; i++) {
2116 dout(" req %p tid %llu op %d rval %d len %u\n", req,
2117 req->r_tid, i, m.rval[i], m.outdata_len[i]);
2118 req->r_ops[i].rval = m.rval[i];
2119 req->r_ops[i].outdata_len = m.outdata_len[i];
2120 data_len += m.outdata_len[i];
2121 }
2122 if (data_len != le32_to_cpu(msg->hdr.data_len)) {
2123 pr_err("sum of lens %u != %u for tid %llu\n", data_len,
2124 le32_to_cpu(msg->hdr.data_len), req->r_tid);
2125 goto fail_request;
2126 }
2127 dout("%s req %p tid %llu acked %d result %d data_len %u\n", __func__,
2128 req, req->r_tid, req->r_got_reply, m.result, data_len);
2129
2130 already_acked = req->r_got_reply;
2131 if (!already_acked) {
2132 req->r_result = m.result ?: data_len;
2133 req->r_replay_version = m.replay_version; /* struct */
2134 req->r_got_reply = true;
2135 } else if (!(m.flags & CEPH_OSD_FLAG_ONDISK)) {
2136 dout("req %p tid %llu dup ack\n", req, req->r_tid);
2137 goto out_put;
2138 }
2139
2140 if (done_request(req, &m)) {
f24e9980 2141 __unregister_request(osdc, req);
fe5da05e
ID		 2142 if (req->r_linger) {
2143 WARN_ON(req->r_unsafe_callback);
2144 __register_linger_request(osdc, req);
2145 }
2146 }
f24e9980
SW		 2147
2148 mutex_unlock(&osdc->request_mutex);
ff513ace 2149 up_read(&osdc->map_sem);
f24e9980 2150
fe5da05e
ID		 2151 if (done_request(req, &m)) {
2152 if (already_acked && req->r_unsafe_callback) {
2153 dout("req %p tid %llu safe-cb\n", req, req->r_tid);
61c5d6bf 2154 req->r_unsafe_callback(req, false);
fe5da05e
ID		 2155 } else {
2156 dout("req %p tid %llu cb\n", req, req->r_tid);
2157 __complete_request(req);
2158 }
2159 } else {
2160 if (req->r_unsafe_callback) {
2161 dout("req %p tid %llu unsafe-cb\n", req, req->r_tid);
2162 req->r_unsafe_callback(req, true);
2163 } else {
2164 WARN_ON(1);
2165 }
61c5d6bf 2166 }
fe5da05e
ID		 2167 if (m.flags & CEPH_OSD_FLAG_ONDISK)
2168 complete_all(&req->r_safe_completion);
f24e9980 2169
f24e9980
SW		 2170 ceph_osdc_put_request(req);
2171 return;
2172
fe5da05e 2173fail_request:
37c89bde
LW		 2174 req->r_result = -EIO;
2175 __unregister_request(osdc, req);
fe5da05e
ID		 2176 __complete_request(req);
2177 complete_all(&req->r_safe_completion);
2178out_put:
1b83bef2 2179 ceph_osdc_put_request(req);
fe5da05e 2180out_unlock:
8058fd45 2181 mutex_unlock(&osdc->request_mutex);
ff513ace 2182 up_read(&osdc->map_sem);
f24e9980
SW		 2183}
2184
42c1b124
ID		 2185static void set_pool_was_full(struct ceph_osd_client *osdc)
2186{
2187 struct rb_node *n;
2188
2189 for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) {
2190 struct ceph_pg_pool_info *pi =
2191 rb_entry(n, struct ceph_pg_pool_info, node);
2192
2193 pi->was_full = __pool_full(pi);
2194 }
2195}
2196
6f6c7006 2197static void reset_changed_osds(struct ceph_osd_client *osdc)
f24e9980 2198{
f24e9980 2199 struct rb_node *p, *n;
f24e9980 2200
7eb71e03 2201 dout("%s %p\n", __func__, osdc);
6f6c7006
SW		 2202 for (p = rb_first(&osdc->osds); p; p = n) {
2203 struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
f24e9980 2204
6f6c7006
SW		 2205 n = rb_next(p);
2206 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
2207 memcmp(&osd->o_con.peer_addr,
2208 ceph_osd_addr(osdc->osdmap,
2209 osd->o_osd),
2210 sizeof(struct ceph_entity_addr)) != 0)
2211 __reset_osd(osdc, osd);
f24e9980 2212 }
422d2cb8
YS		 2213}
2214
2215/*
6f6c7006
SW		 2216 * Requeue requests whose mapping to an OSD has changed. If requests map to
2217 * no osd, request a new map.
422d2cb8 2218 *
e6d50f67 2219 * Caller should hold map_sem for read.
422d2cb8 2220 */
9a1ea2db
JD		 2221static void kick_requests(struct ceph_osd_client *osdc, bool force_resend,
2222 bool force_resend_writes)
422d2cb8 2223{
a40c4f10 2224 struct ceph_osd_request *req, *nreq;
6f6c7006
SW		 2225 struct rb_node *p;
2226 int needmap = 0;
2227 int err;
9a1ea2db 2228 bool force_resend_req;
422d2cb8 2229
9a1ea2db
JD		 2230 dout("kick_requests %s %s\n", force_resend ? " (force resend)" : "",
2231 force_resend_writes ? " (force resend writes)" : "");
422d2cb8 2232 mutex_lock(&osdc->request_mutex);
6194ea89 2233 for (p = rb_first(&osdc->requests); p; ) {
6f6c7006 2234 req = rb_entry(p, struct ceph_osd_request, r_node);
6194ea89 2235 p = rb_next(p);
ab60b16d
AE		 2236
2237 /*
2238 * For linger requests that have not yet been
2239 * registered, move them to the linger list; they'll
2240 * be sent to the osd in the loop below. Unregister
2241 * the request before re-registering it as a linger
2242 * request to ensure the __map_request() below
2243 * will decide it needs to be sent.
2244 */
2245 if (req->r_linger && list_empty(&req->r_linger_item)) {
2246 dout("%p tid %llu restart on osd%d\n",
2247 req, req->r_tid,
2248 req->r_osd ? req->r_osd->o_osd : -1);
96e4dac6 2249 ceph_osdc_get_request(req);
ab60b16d
AE		 2250 __unregister_request(osdc, req);
2251 __register_linger_request(osdc, req);
96e4dac6 2252 ceph_osdc_put_request(req);
ab60b16d
AE		 2253 continue;
2254 }
2255
9a1ea2db
JD		 2256 force_resend_req = force_resend ||
2257 (force_resend_writes &&
2258 req->r_flags & CEPH_OSD_FLAG_WRITE);
2259 err = __map_request(osdc, req, force_resend_req);
6f6c7006
SW		 2260 if (err < 0)
2261 continue; /* error */
2262 if (req->r_osd == NULL) {
2263 dout("%p tid %llu maps to no osd\n", req, req->r_tid);
2264 needmap++; /* request a newer map */
2265 } else if (err > 0) {
6194ea89
SW		 2266 if (!req->r_linger) {
2267 dout("%p tid %llu requeued on osd%d\n", req,
2268 req->r_tid,
2269 req->r_osd ? req->r_osd->o_osd : -1);
a40c4f10 2270 req->r_flags |= CEPH_OSD_FLAG_RETRY;
6194ea89
SW		 2271 }
2272 }
a40c4f10
YS		 2273 }
2274
2275 list_for_each_entry_safe(req, nreq, &osdc->req_linger,
2276 r_linger_item) {
2277 dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
2278
9a1ea2db
JD		 2279 err = __map_request(osdc, req,
2280 force_resend || force_resend_writes);
ab60b16d 2281 dout("__map_request returned %d\n", err);
a40c4f10
YS		 2282 if (err < 0)
2283 continue; /* hrm! */
b0494532
ID		 2284 if (req->r_osd == NULL || err > 0) {
2285 if (req->r_osd == NULL) {
2286 dout("lingering %p tid %llu maps to no osd\n",
2287 req, req->r_tid);
2288 /*
2289 * A homeless lingering request makes
2290 * no sense, as it's job is to keep
2291 * a particular OSD connection open.
2292 * Request a newer map and kick the
2293 * request, knowing that it won't be
2294 * resent until we actually get a map
2295 * that can tell us where to send it.
2296 */
2297 needmap++;
2298 }
a40c4f10 2299
b0494532
ID		 2300 dout("kicking lingering %p tid %llu osd%d\n", req,
2301 req->r_tid, req->r_osd ? req->r_osd->o_osd : -1);
2302 __register_request(osdc, req);
2303 __unregister_linger_request(osdc, req);
2304 }
6f6c7006 2305 }
14d2f38d 2306 reset_changed_osds(osdc);
f24e9980
SW		 2307 mutex_unlock(&osdc->request_mutex);
2308
2309 if (needmap) {
2310 dout("%d requests for down osds, need new map\n", needmap);
2311 ceph_monc_request_next_osdmap(&osdc->client->monc);
2312 }
422d2cb8 2313}
6f6c7006 2314
42c1b124
ID		 2315static int handle_one_map(struct ceph_osd_client *osdc,
2316 void *p, void *end, bool incremental)
2317{
2318 struct ceph_osdmap *newmap;
2319 struct rb_node *n;
2320 bool skipped_map = false;
2321 bool was_full;
2322
2323 was_full = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
2324 set_pool_was_full(osdc);
2325
2326 if (incremental)
2327 newmap = osdmap_apply_incremental(&p, end, osdc->osdmap);
2328 else
2329 newmap = ceph_osdmap_decode(&p, end);
2330 if (IS_ERR(newmap))
2331 return PTR_ERR(newmap);
2332
2333 if (newmap != osdc->osdmap) {
2334 /*
2335 * Preserve ->was_full before destroying the old map.
2336 * For pools that weren't in the old map, ->was_full
2337 * should be false.
2338 */
2339 for (n = rb_first(&newmap->pg_pools); n; n = rb_next(n)) {
2340 struct ceph_pg_pool_info *pi =
2341 rb_entry(n, struct ceph_pg_pool_info, node);
2342 struct ceph_pg_pool_info *old_pi;
2343
2344 old_pi = ceph_pg_pool_by_id(osdc->osdmap, pi->id);
2345 if (old_pi)
2346 pi->was_full = old_pi->was_full;
2347 else
2348 WARN_ON(pi->was_full);
2349 }
2350
2351 if (osdc->osdmap->epoch &&
2352 osdc->osdmap->epoch + 1 < newmap->epoch) {
2353 WARN_ON(incremental);
2354 skipped_map = true;
2355 }
2356
2357 ceph_osdmap_destroy(osdc->osdmap);
2358 osdc->osdmap = newmap;
2359 }
2360
2361 was_full &= !ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
2362 kick_requests(osdc, skipped_map, was_full);
2363
2364 return 0;
2365}
6f6c7006 2366
f24e9980
SW		 2367/*
2368 * Process updated osd map.
2369 *
2370 * The message contains any number of incremental and full maps, normally
2371 * indicating some sort of topology change in the cluster. Kick requests
2372 * off to different OSDs as needed.
2373 */
2374void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
2375{
42c1b124
ID		 2376 void *p = msg->front.iov_base;
2377 void *const end = p + msg->front.iov_len;
f24e9980
SW		 2378 u32 nr_maps, maplen;
2379 u32 epoch;
f24e9980 2380 struct ceph_fsid fsid;
42c1b124
ID		 2381 bool handled_incremental = false;
2382 bool was_pauserd, was_pausewr;
2383 bool pauserd, pausewr;
2384 int err;
f24e9980 2385
42c1b124
ID		 2386 dout("%s have %u\n", __func__, osdc->osdmap->epoch);
2387 down_write(&osdc->map_sem);
f24e9980
SW		 2388
2389 /* verify fsid */
2390 ceph_decode_need(&p, end, sizeof(fsid), bad);
2391 ceph_decode_copy(&p, &fsid, sizeof(fsid));
0743304d 2392 if (ceph_check_fsid(osdc->client, &fsid) < 0)
42c1b124 2393 goto bad;
f24e9980 2394
42c1b124
ID		 2395 was_pauserd = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD);
2396 was_pausewr = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR) ||
2397 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) ||
2398 have_pool_full(osdc);
9a1ea2db 2399
f24e9980
SW		 2400 /* incremental maps */
2401 ceph_decode_32_safe(&p, end, nr_maps, bad);
2402 dout(" %d inc maps\n", nr_maps);
2403 while (nr_maps > 0) {
2404 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 2405 epoch = ceph_decode_32(&p);
2406 maplen = ceph_decode_32(&p);
f24e9980 2407 ceph_decode_need(&p, end, maplen, bad);
42c1b124
ID		 2408 if (osdc->osdmap->epoch &&
2409 osdc->osdmap->epoch + 1 == epoch) {
f24e9980
SW		 2410 dout("applying incremental map %u len %d\n",
2411 epoch, maplen);
42c1b124
ID		 2412 err = handle_one_map(osdc, p, p + maplen, true);
2413 if (err)
f24e9980 2414 goto bad;
42c1b124 2415 handled_incremental = true;
f24e9980
SW		 2416 } else {
2417 dout("ignoring incremental map %u len %d\n",
2418 epoch, maplen);
2419 }
42c1b124 2420 p += maplen;
f24e9980
SW		 2421 nr_maps--;
2422 }
42c1b124 2423 if (handled_incremental)
f24e9980
SW		 2424 goto done;
2425
2426 /* full maps */
2427 ceph_decode_32_safe(&p, end, nr_maps, bad);
2428 dout(" %d full maps\n", nr_maps);
2429 while (nr_maps) {
2430 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 2431 epoch = ceph_decode_32(&p);
2432 maplen = ceph_decode_32(&p);
f24e9980
SW		 2433 ceph_decode_need(&p, end, maplen, bad);
2434 if (nr_maps > 1) {
2435 dout("skipping non-latest full map %u len %d\n",
2436 epoch, maplen);
e5253a7b 2437 } else if (osdc->osdmap->epoch >= epoch) {
f24e9980
SW		 2438 dout("skipping full map %u len %d, "
2439 "older than our %u\n", epoch, maplen,
2440 osdc->osdmap->epoch);
2441 } else {
2442 dout("taking full map %u len %d\n", epoch, maplen);
42c1b124
ID		 2443 err = handle_one_map(osdc, p, p + maplen, false);
2444 if (err)
f24e9980 2445 goto bad;
f24e9980
SW		 2446 }
2447 p += maplen;
2448 nr_maps--;
2449 }
2450
2451done:
cd634fb6
SW		 2452 /*
2453 * subscribe to subsequent osdmap updates if full to ensure
2454 * we find out when we are no longer full and stop returning
2455 * ENOSPC.
2456 */
42c1b124
ID		 2457 pauserd = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD);
2458 pausewr = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR) ||
2459 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) ||
2460 have_pool_full(osdc);
2461 if (was_pauserd || was_pausewr || pauserd || pausewr)
2462 maybe_request_map(osdc);
cd634fb6 2463
f9d25199
AE		 2464 mutex_lock(&osdc->request_mutex);
2465 __send_queued(osdc);
2466 mutex_unlock(&osdc->request_mutex);
42c1b124
ID		 2467
2468 ceph_monc_got_map(&osdc->client->monc, CEPH_SUB_OSDMAP,
2469 osdc->osdmap->epoch);
2470 up_write(&osdc->map_sem);
03066f23 2471 wake_up_all(&osdc->client->auth_wq);
f24e9980
SW		 2472 return;
2473
2474bad:
2475 pr_err("osdc handle_map corrupt msg\n");
9ec7cab1 2476 ceph_msg_dump(msg);
f24e9980 2477 up_write(&osdc->map_sem);
f24e9980
SW		 2478}
2479
a40c4f10
YS		 2480/*
2481 * watch/notify callback event infrastructure
2482 *
2483 * These callbacks are used both for watch and notify operations.
2484 */
2485static void __release_event(struct kref *kref)
2486{
2487 struct ceph_osd_event *event =
2488 container_of(kref, struct ceph_osd_event, kref);
2489
2490 dout("__release_event %p\n", event);
2491 kfree(event);
2492}
2493
2494static void get_event(struct ceph_osd_event *event)
2495{
2496 kref_get(&event->kref);
2497}
2498
2499void ceph_osdc_put_event(struct ceph_osd_event *event)
2500{
2501 kref_put(&event->kref, __release_event);
2502}
2503EXPORT_SYMBOL(ceph_osdc_put_event);
2504
2505static void __insert_event(struct ceph_osd_client *osdc,
2506 struct ceph_osd_event *new)
2507{
2508 struct rb_node **p = &osdc->event_tree.rb_node;
2509 struct rb_node *parent = NULL;
2510 struct ceph_osd_event *event = NULL;
2511
2512 while (*p) {
2513 parent = *p;
2514 event = rb_entry(parent, struct ceph_osd_event, node);
2515 if (new->cookie < event->cookie)
2516 p = &(*p)->rb_left;
2517 else if (new->cookie > event->cookie)
2518 p = &(*p)->rb_right;
2519 else
2520 BUG();
2521 }
2522
2523 rb_link_node(&new->node, parent, p);
2524 rb_insert_color(&new->node, &osdc->event_tree);
2525}
2526
2527static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
2528 u64 cookie)
2529{
2530 struct rb_node **p = &osdc->event_tree.rb_node;
2531 struct rb_node *parent = NULL;
2532 struct ceph_osd_event *event = NULL;
2533
2534 while (*p) {
2535 parent = *p;
2536 event = rb_entry(parent, struct ceph_osd_event, node);
2537 if (cookie < event->cookie)
2538 p = &(*p)->rb_left;
2539 else if (cookie > event->cookie)
2540 p = &(*p)->rb_right;
2541 else
2542 return event;
2543 }
2544 return NULL;
2545}
2546
2547static void __remove_event(struct ceph_osd_event *event)
2548{
2549 struct ceph_osd_client *osdc = event->osdc;
2550
2551 if (!RB_EMPTY_NODE(&event->node)) {
2552 dout("__remove_event removed %p\n", event);
2553 rb_erase(&event->node, &osdc->event_tree);
2554 ceph_osdc_put_event(event);
2555 } else {
2556 dout("__remove_event didn't remove %p\n", event);
2557 }
2558}
2559
2560int ceph_osdc_create_event(struct ceph_osd_client *osdc,
2561 void (*event_cb)(u64, u64, u8, void *),
3c663bbd 2562 void *data, struct ceph_osd_event **pevent)
a40c4f10
YS		 2563{
2564 struct ceph_osd_event *event;
2565
2566 event = kmalloc(sizeof(*event), GFP_NOIO);
2567 if (!event)
2568 return -ENOMEM;
2569
2570 dout("create_event %p\n", event);
2571 event->cb = event_cb;
3c663bbd 2572 event->one_shot = 0;
a40c4f10
YS		 2573 event->data = data;
2574 event->osdc = osdc;
2575 INIT_LIST_HEAD(&event->osd_node);
3ee5234d 2576 RB_CLEAR_NODE(&event->node);
a40c4f10
YS		 2577 kref_init(&event->kref); /* one ref for us */
2578 kref_get(&event->kref); /* one ref for the caller */
a40c4f10
YS		 2579
2580 spin_lock(&osdc->event_lock);
2581 event->cookie = ++osdc->event_count;
2582 __insert_event(osdc, event);
2583 spin_unlock(&osdc->event_lock);
2584
2585 *pevent = event;
2586 return 0;
2587}
2588EXPORT_SYMBOL(ceph_osdc_create_event);
2589
2590void ceph_osdc_cancel_event(struct ceph_osd_event *event)
2591{
2592 struct ceph_osd_client *osdc = event->osdc;
2593
2594 dout("cancel_event %p\n", event);
2595 spin_lock(&osdc->event_lock);
2596 __remove_event(event);
2597 spin_unlock(&osdc->event_lock);
2598 ceph_osdc_put_event(event); /* caller's */
2599}
2600EXPORT_SYMBOL(ceph_osdc_cancel_event);
2601
2602
2603static void do_event_work(struct work_struct *work)
2604{
2605 struct ceph_osd_event_work *event_work =
2606 container_of(work, struct ceph_osd_event_work, work);
2607 struct ceph_osd_event *event = event_work->event;
2608 u64 ver = event_work->ver;
2609 u64 notify_id = event_work->notify_id;
2610 u8 opcode = event_work->opcode;
2611
2612 dout("do_event_work completing %p\n", event);
2613 event->cb(ver, notify_id, opcode, event->data);
a40c4f10
YS		 2614 dout("do_event_work completed %p\n", event);
2615 ceph_osdc_put_event(event);
2616 kfree(event_work);
2617}
2618
2619
2620/*
2621 * Process osd watch notifications
2622 */
3c663bbd
AE		 2623static void handle_watch_notify(struct ceph_osd_client *osdc,
2624 struct ceph_msg *msg)
a40c4f10
YS		 2625{
2626 void *p, *end;
2627 u8 proto_ver;
2628 u64 cookie, ver, notify_id;
2629 u8 opcode;
2630 struct ceph_osd_event *event;
2631 struct ceph_osd_event_work *event_work;
2632
2633 p = msg->front.iov_base;
2634 end = p + msg->front.iov_len;
2635
2636 ceph_decode_8_safe(&p, end, proto_ver, bad);
2637 ceph_decode_8_safe(&p, end, opcode, bad);
2638 ceph_decode_64_safe(&p, end, cookie, bad);
2639 ceph_decode_64_safe(&p, end, ver, bad);
2640 ceph_decode_64_safe(&p, end, notify_id, bad);
2641
2642 spin_lock(&osdc->event_lock);
2643 event = __find_event(osdc, cookie);
2644 if (event) {
3c663bbd 2645 BUG_ON(event->one_shot);
a40c4f10 2646 get_event(event);
a40c4f10
YS		 2647 }
2648 spin_unlock(&osdc->event_lock);
2649 dout("handle_watch_notify cookie %lld ver %lld event %p\n",
2650 cookie, ver, event);
2651 if (event) {
2652 event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
a40c4f10 2653 if (!event_work) {
91883cd2
ID		 2654 pr_err("couldn't allocate event_work\n");
2655 ceph_osdc_put_event(event);
2656 return;
a40c4f10 2657 }
6b0ae409 2658 INIT_WORK(&event_work->work, do_event_work);
a40c4f10
YS		 2659 event_work->event = event;
2660 event_work->ver = ver;
2661 event_work->notify_id = notify_id;
2662 event_work->opcode = opcode;
a40c4f10 2663
91883cd2
ID		 2664 queue_work(osdc->notify_wq, &event_work->work);
2665 }
a40c4f10 2666
a40c4f10
YS		 2667 return;
2668
2669bad:
2670 pr_err("osdc handle_watch_notify corrupt msg\n");
a40c4f10
YS		 2671}
2672
70636773
AE		 2673/*
2674 * Register request, send initial attempt.
2675 */
2676int ceph_osdc_start_request(struct ceph_osd_client *osdc,
2677 struct ceph_osd_request *req,
2678 bool nofail)
2679{
0bbfdfe8 2680 int rc;
70636773 2681
f24e9980
SW		 2682 down_read(&osdc->map_sem);
2683 mutex_lock(&osdc->request_mutex);
0bbfdfe8
ID		 2684
2685 rc = __ceph_osdc_start_request(osdc, req, nofail);
2686
f24e9980
SW		 2687 mutex_unlock(&osdc->request_mutex);
2688 up_read(&osdc->map_sem);
0bbfdfe8 2689
f24e9980
SW		 2690 return rc;
2691}
3d14c5d2 2692EXPORT_SYMBOL(ceph_osdc_start_request);
f24e9980 2693
c9f9b93d
ID		 2694/*
2695 * Unregister a registered request. The request is not completed (i.e.
2696 * no callbacks or wakeups) - higher layers are supposed to know what
2697 * they are canceling.
2698 */
2699void ceph_osdc_cancel_request(struct ceph_osd_request *req)
2700{
2701 struct ceph_osd_client *osdc = req->r_osdc;
2702
2703 mutex_lock(&osdc->request_mutex);
2704 if (req->r_linger)
2705 __unregister_linger_request(osdc, req);
2706 __unregister_request(osdc, req);
2707 mutex_unlock(&osdc->request_mutex);
2708
2709 dout("%s %p tid %llu canceled\n", __func__, req, req->r_tid);
2710}
2711EXPORT_SYMBOL(ceph_osdc_cancel_request);
2712
f24e9980
SW		 2713/*
2714 * wait for a request to complete
2715 */
2716int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
2717 struct ceph_osd_request *req)
2718{
2719 int rc;
2720
c9f9b93d
ID		 2721 dout("%s %p tid %llu\n", __func__, req, req->r_tid);
2722
f24e9980
SW		 2723 rc = wait_for_completion_interruptible(&req->r_completion);
2724 if (rc < 0) {
c9f9b93d
ID		 2725 dout("%s %p tid %llu interrupted\n", __func__, req, req->r_tid);
2726 ceph_osdc_cancel_request(req);
fe5da05e
ID		 2727
2728 /* kludge - need to to wake ceph_osdc_sync() */
2729 complete_all(&req->r_safe_completion);
f24e9980
SW		 2730 return rc;
2731 }
2732
c9f9b93d
ID		 2733 dout("%s %p tid %llu result %d\n", __func__, req, req->r_tid,
2734 req->r_result);
f24e9980
SW		 2735 return req->r_result;
2736}
3d14c5d2 2737EXPORT_SYMBOL(ceph_osdc_wait_request);
f24e9980
SW		 2738
2739/*
2740 * sync - wait for all in-flight requests to flush. avoid starvation.
2741 */
2742void ceph_osdc_sync(struct ceph_osd_client *osdc)
2743{
2744 struct ceph_osd_request *req;
2745 u64 last_tid, next_tid = 0;
2746
2747 mutex_lock(&osdc->request_mutex);
2748 last_tid = osdc->last_tid;
2749 while (1) {
2750 req = __lookup_request_ge(osdc, next_tid);
2751 if (!req)
2752 break;
2753 if (req->r_tid > last_tid)
2754 break;
2755
2756 next_tid = req->r_tid + 1;
2757 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
2758 continue;
2759
2760 ceph_osdc_get_request(req);
2761 mutex_unlock(&osdc->request_mutex);
2762 dout("sync waiting on tid %llu (last is %llu)\n",
2763 req->r_tid, last_tid);
2764 wait_for_completion(&req->r_safe_completion);
2765 mutex_lock(&osdc->request_mutex);
2766 ceph_osdc_put_request(req);
2767 }
2768 mutex_unlock(&osdc->request_mutex);
2769 dout("sync done (thru tid %llu)\n", last_tid);
2770}
3d14c5d2 2771EXPORT_SYMBOL(ceph_osdc_sync);
f24e9980 2772
dd935f44
JD		 2773/*
2774 * Call all pending notify callbacks - for use after a watch is
2775 * unregistered, to make sure no more callbacks for it will be invoked
2776 */
f6479449 2777void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc)
dd935f44
JD		 2778{
2779 flush_workqueue(osdc->notify_wq);
2780}
2781EXPORT_SYMBOL(ceph_osdc_flush_notifies);
2782
2783
f24e9980
SW		 2784/*
2785 * init, shutdown
2786 */
2787int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
2788{
2789 int err;
2790
2791 dout("init\n");
2792 osdc->client = client;
f24e9980 2793 init_rwsem(&osdc->map_sem);
f24e9980 2794 mutex_init(&osdc->request_mutex);
f24e9980
SW		 2795 osdc->last_tid = 0;
2796 osdc->osds = RB_ROOT;
f5a2041b 2797 INIT_LIST_HEAD(&osdc->osd_lru);
f24e9980 2798 osdc->requests = RB_ROOT;
422d2cb8 2799 INIT_LIST_HEAD(&osdc->req_lru);
6f6c7006
SW		 2800 INIT_LIST_HEAD(&osdc->req_unsent);
2801 INIT_LIST_HEAD(&osdc->req_notarget);
a40c4f10 2802 INIT_LIST_HEAD(&osdc->req_linger);
f24e9980
SW		 2803 osdc->num_requests = 0;
2804 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
f5a2041b 2805 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
a40c4f10
YS		 2806 spin_lock_init(&osdc->event_lock);
2807 osdc->event_tree = RB_ROOT;
2808 osdc->event_count = 0;
f5a2041b 2809
5f44f142 2810 err = -ENOMEM;
e5253a7b
ID		 2811 osdc->osdmap = ceph_osdmap_alloc();
2812 if (!osdc->osdmap)
2813 goto out;
2814
9e767adb
ID		 2815 osdc->req_mempool = mempool_create_slab_pool(10,
2816 ceph_osd_request_cache);
f24e9980 2817 if (!osdc->req_mempool)
e5253a7b 2818 goto out_map;
f24e9980 2819
d50b409f 2820 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
711da55d 2821 PAGE_SIZE, 10, true, "osd_op");
f24e9980 2822 if (err < 0)
5f44f142 2823 goto out_mempool;
d50b409f 2824 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
711da55d 2825 PAGE_SIZE, 10, true, "osd_op_reply");
c16e7869
SW		 2826 if (err < 0)
2827 goto out_msgpool;
a40c4f10 2828
dbcae088 2829 err = -ENOMEM;
a40c4f10 2830 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
dbcae088 2831 if (!osdc->notify_wq)
c172ec5c
ID		 2832 goto out_msgpool_reply;
2833
fbca9635
ID		 2834 schedule_delayed_work(&osdc->timeout_work,
2835 osdc->client->options->osd_keepalive_timeout);
b37ee1b9
ID		 2836 schedule_delayed_work(&osdc->osds_timeout_work,
2837 round_jiffies_relative(osdc->client->options->osd_idle_ttl));
2838
f24e9980 2839 return 0;
5f44f142 2840
c172ec5c
ID		 2841out_msgpool_reply:
2842 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
c16e7869
SW		 2843out_msgpool:
2844 ceph_msgpool_destroy(&osdc->msgpool_op);
5f44f142
SW		 2845out_mempool:
2846 mempool_destroy(osdc->req_mempool);
e5253a7b
ID		 2847out_map:
2848 ceph_osdmap_destroy(osdc->osdmap);
5f44f142
SW		 2849out:
2850 return err;
f24e9980
SW		 2851}
2852
2853void ceph_osdc_stop(struct ceph_osd_client *osdc)
2854{
a40c4f10
YS		 2855 flush_workqueue(osdc->notify_wq);
2856 destroy_workqueue(osdc->notify_wq);
f24e9980 2857 cancel_delayed_work_sync(&osdc->timeout_work);
f5a2041b 2858 cancel_delayed_work_sync(&osdc->osds_timeout_work);
42a2c09f
ID		 2859
2860 mutex_lock(&osdc->request_mutex);
2861 while (!RB_EMPTY_ROOT(&osdc->osds)) {
2862 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
2863 struct ceph_osd, o_node);
2864 remove_osd(osdc, osd);
2865 }
2866 mutex_unlock(&osdc->request_mutex);
2867
e5253a7b 2868 ceph_osdmap_destroy(osdc->osdmap);
f24e9980
SW		 2869 mempool_destroy(osdc->req_mempool);
2870 ceph_msgpool_destroy(&osdc->msgpool_op);
c16e7869 2871 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
f24e9980
SW		 2872}
2873
2874/*
2875 * Read some contiguous pages. If we cross a stripe boundary, shorten
2876 * *plen. Return number of bytes read, or error.
2877 */
2878int ceph_osdc_readpages(struct ceph_osd_client *osdc,
2879 struct ceph_vino vino, struct ceph_file_layout *layout,
2880 u64 off, u64 *plen,
2881 u32 truncate_seq, u64 truncate_size,
b7495fc2 2882 struct page **pages, int num_pages, int page_align)
f24e9980
SW		 2883{
2884 struct ceph_osd_request *req;
2885 int rc = 0;
2886
2887 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
2888 vino.snap, off, *plen);
715e4cd4 2889 req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 0, 1,
f24e9980 2890 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
acead002 2891 NULL, truncate_seq, truncate_size,
153e5167 2892 false);
6816282d
SW		 2893 if (IS_ERR(req))
2894 return PTR_ERR(req);
f24e9980
SW		 2895
2896 /* it may be a short read due to an object boundary */
406e2c9f 2897 osd_req_op_extent_osd_data_pages(req, 0,
a4ce40a9 2898 pages, *plen, page_align, false, false);
f24e9980 2899
e0c59487 2900 dout("readpages final extent is %llu~%llu (%llu bytes align %d)\n",
43bfe5de 2901 off, *plen, *plen, page_align);
f24e9980
SW		 2902
2903 rc = ceph_osdc_start_request(osdc, req, false);
2904 if (!rc)
2905 rc = ceph_osdc_wait_request(osdc, req);
2906
2907 ceph_osdc_put_request(req);
2908 dout("readpages result %d\n", rc);
2909 return rc;
2910}
3d14c5d2 2911EXPORT_SYMBOL(ceph_osdc_readpages);
f24e9980
SW		 2912
2913/*
2914 * do a synchronous write on N pages
2915 */
2916int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
2917 struct ceph_file_layout *layout,
2918 struct ceph_snap_context *snapc,
2919 u64 off, u64 len,
2920 u32 truncate_seq, u64 truncate_size,
2921 struct timespec *mtime,
24808826 2922 struct page **pages, int num_pages)
f24e9980
SW		 2923{
2924 struct ceph_osd_request *req;
2925 int rc = 0;
b7495fc2 2926 int page_align = off & ~PAGE_MASK;
f24e9980 2927
715e4cd4 2928 req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 0, 1,
f24e9980 2929 CEPH_OSD_OP_WRITE,
24808826 2930 CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
acead002 2931 snapc, truncate_seq, truncate_size,
153e5167 2932 true);
6816282d
SW		 2933 if (IS_ERR(req))
2934 return PTR_ERR(req);
f24e9980
SW		 2935
2936 /* it may be a short write due to an object boundary */
406e2c9f 2937 osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align,
43bfe5de
AE		 2938 false, false);
2939 dout("writepages %llu~%llu (%llu bytes)\n", off, len, len);
f24e9980 2940
bb873b53 2941 req->r_mtime = *mtime;
87f979d3 2942 rc = ceph_osdc_start_request(osdc, req, true);
f24e9980
SW		 2943 if (!rc)
2944 rc = ceph_osdc_wait_request(osdc, req);
2945
2946 ceph_osdc_put_request(req);
2947 if (rc == 0)
2948 rc = len;
2949 dout("writepages result %d\n", rc);
2950 return rc;
2951}
3d14c5d2 2952EXPORT_SYMBOL(ceph_osdc_writepages);
f24e9980 2953
5522ae0b
AE		 2954int ceph_osdc_setup(void)
2955{
3f1af42a
ID		 2956 size_t size = sizeof(struct ceph_osd_request) +
2957 CEPH_OSD_SLAB_OPS * sizeof(struct ceph_osd_req_op);
2958
5522ae0b 2959 BUG_ON(ceph_osd_request_cache);
3f1af42a
ID		 2960 ceph_osd_request_cache = kmem_cache_create("ceph_osd_request", size,
2961 0, 0, NULL);
5522ae0b
AE		 2962
2963 return ceph_osd_request_cache ? 0 : -ENOMEM;
2964}
2965EXPORT_SYMBOL(ceph_osdc_setup);
2966
2967void ceph_osdc_cleanup(void)
2968{
2969 BUG_ON(!ceph_osd_request_cache);
2970 kmem_cache_destroy(ceph_osd_request_cache);
2971 ceph_osd_request_cache = NULL;
2972}
2973EXPORT_SYMBOL(ceph_osdc_cleanup);
2974
f24e9980
SW		 2975/*
2976 * handle incoming message
2977 */
2978static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
2979{
2980 struct ceph_osd *osd = con->private;
32c895e7 2981 struct ceph_osd_client *osdc;
f24e9980
SW		 2982 int type = le16_to_cpu(msg->hdr.type);
2983
2984 if (!osd)
4a32f93d 2985 goto out;
32c895e7 2986 osdc = osd->o_osdc;
f24e9980
SW		 2987
2988 switch (type) {
2989 case CEPH_MSG_OSD_MAP:
2990 ceph_osdc_handle_map(osdc, msg);
2991 break;
2992 case CEPH_MSG_OSD_OPREPLY:
70cf052d 2993 handle_reply(osdc, msg);
f24e9980 2994 break;
a40c4f10
YS		 2995 case CEPH_MSG_WATCH_NOTIFY:
2996 handle_watch_notify(osdc, msg);
2997 break;
f24e9980
SW		 2998
2999 default:
3000 pr_err("received unknown message type %d %s\n", type,
3001 ceph_msg_type_name(type));
3002 }
4a32f93d 3003out:
f24e9980
SW		 3004 ceph_msg_put(msg);
3005}
3006
5b3a4db3 3007/*
d15f9d69
ID		 3008 * Lookup and return message for incoming reply. Don't try to do
3009 * anything about a larger than preallocated data portion of the
3010 * message at the moment - for now, just skip the message.
5b3a4db3
SW		 3011 */
3012static struct ceph_msg *get_reply(struct ceph_connection *con,
2450418c
YS		 3013 struct ceph_msg_header *hdr,
3014 int *skip)
f24e9980
SW		 3015{
3016 struct ceph_osd *osd = con->private;
3017 struct ceph_osd_client *osdc = osd->o_osdc;
2450418c 3018 struct ceph_msg *m;
0547a9b3 3019 struct ceph_osd_request *req;
3f0a4ac5 3020 int front_len = le32_to_cpu(hdr->front_len);
5b3a4db3 3021 int data_len = le32_to_cpu(hdr->data_len);
0547a9b3 3022 u64 tid;
f24e9980 3023
0547a9b3
YS		 3024 tid = le64_to_cpu(hdr->tid);
3025 mutex_lock(&osdc->request_mutex);
fcd00b68 3026 req = lookup_request(&osdc->requests, tid);
0547a9b3 3027 if (!req) {
cd8140c6
ID		 3028 dout("%s osd%d tid %llu unknown, skipping\n", __func__,
3029 osd->o_osd, tid);
0547a9b3 3030 m = NULL;
d15f9d69 3031 *skip = 1;
0547a9b3
YS		 3032 goto out;
3033 }
c16e7869 3034
ace6d3a9 3035 ceph_msg_revoke_incoming(req->r_reply);
0547a9b3 3036
f2be82b0 3037 if (front_len > req->r_reply->front_alloc_len) {
d15f9d69
ID		 3038 pr_warn("%s osd%d tid %llu front %d > preallocated %d\n",
3039 __func__, osd->o_osd, req->r_tid, front_len,
3040 req->r_reply->front_alloc_len);
3f0a4ac5
ID		 3041 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS,
3042 false);
a79832f2 3043 if (!m)
c16e7869
SW		 3044 goto out;
3045 ceph_msg_put(req->r_reply);
3046 req->r_reply = m;
3047 }
0fff87ec 3048
d15f9d69
ID		 3049 if (data_len > req->r_reply->data_length) {
3050 pr_warn("%s osd%d tid %llu data %d > preallocated %zu, skipping\n",
3051 __func__, osd->o_osd, req->r_tid, data_len,
3052 req->r_reply->data_length);
3053 m = NULL;
3054 *skip = 1;
3055 goto out;
0547a9b3 3056 }
d15f9d69
ID		 3057
3058 m = ceph_msg_get(req->r_reply);
c16e7869 3059 dout("get_reply tid %lld %p\n", tid, m);
0547a9b3
YS		 3060
3061out:
3062 mutex_unlock(&osdc->request_mutex);
2450418c 3063 return m;
5b3a4db3
SW		 3064}
3065
3066static struct ceph_msg *alloc_msg(struct ceph_connection *con,
3067 struct ceph_msg_header *hdr,
3068 int *skip)
3069{
3070 struct ceph_osd *osd = con->private;
3071 int type = le16_to_cpu(hdr->type);
3072 int front = le32_to_cpu(hdr->front_len);
3073
1c20f2d2 3074 *skip = 0;
5b3a4db3
SW		 3075 switch (type) {
3076 case CEPH_MSG_OSD_MAP:
a40c4f10 3077 case CEPH_MSG_WATCH_NOTIFY:
b61c2763 3078 return ceph_msg_new(type, front, GFP_NOFS, false);
5b3a4db3
SW		 3079 case CEPH_MSG_OSD_OPREPLY:
3080 return get_reply(con, hdr, skip);
3081 default:
3082 pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
3083 osd->o_osd);
3084 *skip = 1;
3085 return NULL;
3086 }
f24e9980
SW		 3087}
3088
3089/*
3090 * Wrappers to refcount containing ceph_osd struct
3091 */
3092static struct ceph_connection *get_osd_con(struct ceph_connection *con)
3093{
3094 struct ceph_osd *osd = con->private;
3095 if (get_osd(osd))
3096 return con;
3097 return NULL;
3098}
3099
3100static void put_osd_con(struct ceph_connection *con)
3101{
3102 struct ceph_osd *osd = con->private;
3103 put_osd(osd);
3104}
3105
4e7a5dcd
SW		 3106/*
3107 * authentication
3108 */
a3530df3
AE		 3109/*
3110 * Note: returned pointer is the address of a structure that's
3111 * managed separately. Caller must *not* attempt to free it.
3112 */
3113static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
8f43fb53 3114 int *proto, int force_new)
4e7a5dcd
SW		 3115{
3116 struct ceph_osd *o = con->private;
3117 struct ceph_osd_client *osdc = o->o_osdc;
3118 struct ceph_auth_client *ac = osdc->client->monc.auth;
74f1869f 3119 struct ceph_auth_handshake *auth = &o->o_auth;
4e7a5dcd 3120
74f1869f 3121 if (force_new && auth->authorizer) {
6c1ea260 3122 ceph_auth_destroy_authorizer(auth->authorizer);
74f1869f
AE		 3123 auth->authorizer = NULL;
3124 }
27859f97
SW		 3125 if (!auth->authorizer) {
3126 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
3127 auth);
4e7a5dcd 3128 if (ret)
a3530df3 3129 return ERR_PTR(ret);
27859f97
SW		 3130 } else {
3131 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
0bed9b5c
SW		 3132 auth);
3133 if (ret)
3134 return ERR_PTR(ret);
4e7a5dcd 3135 }
4e7a5dcd 3136 *proto = ac->protocol;
74f1869f 3137
a3530df3 3138 return auth;
4e7a5dcd
SW		 3139}
3140
3141
3142static int verify_authorizer_reply(struct ceph_connection *con, int len)
3143{
3144 struct ceph_osd *o = con->private;
3145 struct ceph_osd_client *osdc = o->o_osdc;
3146 struct ceph_auth_client *ac = osdc->client->monc.auth;
3147
27859f97 3148 return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer, len);
4e7a5dcd
SW		 3149}
3150
9bd2e6f8
SW		 3151static int invalidate_authorizer(struct ceph_connection *con)
3152{
3153 struct ceph_osd *o = con->private;
3154 struct ceph_osd_client *osdc = o->o_osdc;
3155 struct ceph_auth_client *ac = osdc->client->monc.auth;
3156
27859f97 3157 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
9bd2e6f8
SW		 3158 return ceph_monc_validate_auth(&osdc->client->monc);
3159}
4e7a5dcd 3160
79dbd1ba 3161static int osd_sign_message(struct ceph_msg *msg)
33d07337 3162{
79dbd1ba 3163 struct ceph_osd *o = msg->con->private;
33d07337 3164 struct ceph_auth_handshake *auth = &o->o_auth;
79dbd1ba 3165
33d07337
YZ		 3166 return ceph_auth_sign_message(auth, msg);
3167}
3168
79dbd1ba 3169static int osd_check_message_signature(struct ceph_msg *msg)
33d07337 3170{
79dbd1ba 3171 struct ceph_osd *o = msg->con->private;
33d07337 3172 struct ceph_auth_handshake *auth = &o->o_auth;
79dbd1ba 3173
33d07337
YZ		 3174 return ceph_auth_check_message_signature(auth, msg);
3175}
3176
9e32789f 3177static const struct ceph_connection_operations osd_con_ops = {
f24e9980
SW		 3178 .get = get_osd_con,
3179 .put = put_osd_con,
3180 .dispatch = dispatch,
4e7a5dcd
SW		 3181 .get_authorizer = get_authorizer,
3182 .verify_authorizer_reply = verify_authorizer_reply,
9bd2e6f8 3183 .invalidate_authorizer = invalidate_authorizer,
f24e9980 3184 .alloc_msg = alloc_msg,
79dbd1ba
ID		 3185 .sign_message = osd_sign_message,
3186 .check_message_signature = osd_check_message_signature,
81b024e7 3187 .fault = osd_reset,
f24e9980 3188};
Linux 6.x block layer and io_uring tree(s)