projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
disable some mediatekl custom warnings
[GitHub/mt8127/android_kernel_alcatel_ttab.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
SW		 22#define OSD_OP_FRONT_LEN 4096
23#define OSD_OPREPLY_FRONT_LEN 512
0d59ab81 24
5522ae0b
AE		 25static struct kmem_cache *ceph_osd_request_cache;
26
9e32789f 27static const struct ceph_connection_operations osd_con_ops;
f24e9980 28
f9d25199 29static void __send_queued(struct ceph_osd_client *osdc);
6f6c7006 30static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd);
a40c4f10
YS		 31static void __register_request(struct ceph_osd_client *osdc,
32 struct ceph_osd_request *req);
33static void __unregister_linger_request(struct ceph_osd_client *osdc,
34 struct ceph_osd_request *req);
56e925b6
SW		 35static void __send_request(struct ceph_osd_client *osdc,
36 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
863c7eb5
AE		 120#define osd_req_op_data(oreq, whch, typ, fld) \
121 ({ \
122 BUG_ON(whch >= (oreq)->r_num_ops); \
123 &(oreq)->r_ops[whch].typ.fld; \
124 })
125
49719778
AE		 126static struct ceph_osd_data *
127osd_req_op_raw_data_in(struct ceph_osd_request *osd_req, unsigned int which)
128{
129 BUG_ON(which >= osd_req->r_num_ops);
130
131 return &osd_req->r_ops[which].raw_data_in;
132}
133
a4ce40a9
AE		 134struct ceph_osd_data *
135osd_req_op_extent_osd_data(struct ceph_osd_request *osd_req,
406e2c9f 136 unsigned int which)
a4ce40a9 137{
863c7eb5 138 return osd_req_op_data(osd_req, which, extent, osd_data);
a4ce40a9
AE		 139}
140EXPORT_SYMBOL(osd_req_op_extent_osd_data);
141
a4ce40a9
AE		 142struct ceph_osd_data *
143osd_req_op_cls_response_data(struct ceph_osd_request *osd_req,
144 unsigned int which)
145{
863c7eb5 146 return osd_req_op_data(osd_req, which, cls, response_data);
a4ce40a9
AE		 147}
148EXPORT_SYMBOL(osd_req_op_cls_response_data); /* ??? */
149
49719778
AE		 150void osd_req_op_raw_data_in_pages(struct ceph_osd_request *osd_req,
151 unsigned int which, struct page **pages,
152 u64 length, u32 alignment,
153 bool pages_from_pool, bool own_pages)
154{
155 struct ceph_osd_data *osd_data;
156
157 osd_data = osd_req_op_raw_data_in(osd_req, which);
158 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
159 pages_from_pool, own_pages);
160}
161EXPORT_SYMBOL(osd_req_op_raw_data_in_pages);
162
a4ce40a9 163void osd_req_op_extent_osd_data_pages(struct ceph_osd_request *osd_req,
406e2c9f
AE		 164 unsigned int which, struct page **pages,
165 u64 length, u32 alignment,
a4ce40a9
AE		 166 bool pages_from_pool, bool own_pages)
167{
168 struct ceph_osd_data *osd_data;
169
863c7eb5 170 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
a4ce40a9
AE		 171 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
172 pages_from_pool, own_pages);
a4ce40a9
AE		 173}
174EXPORT_SYMBOL(osd_req_op_extent_osd_data_pages);
175
176void osd_req_op_extent_osd_data_pagelist(struct ceph_osd_request *osd_req,
406e2c9f 177 unsigned int which, struct ceph_pagelist *pagelist)
a4ce40a9
AE		 178{
179 struct ceph_osd_data *osd_data;
180
863c7eb5 181 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
a4ce40a9 182 ceph_osd_data_pagelist_init(osd_data, pagelist);
a4ce40a9
AE		 183}
184EXPORT_SYMBOL(osd_req_op_extent_osd_data_pagelist);
185
186#ifdef CONFIG_BLOCK
187void osd_req_op_extent_osd_data_bio(struct ceph_osd_request *osd_req,
406e2c9f 188 unsigned int which, struct bio *bio, size_t bio_length)
a4ce40a9
AE		 189{
190 struct ceph_osd_data *osd_data;
863c7eb5
AE		 191
192 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
a4ce40a9 193 ceph_osd_data_bio_init(osd_data, bio, bio_length);
a4ce40a9
AE		 194}
195EXPORT_SYMBOL(osd_req_op_extent_osd_data_bio);
196#endif /* CONFIG_BLOCK */
197
198static void osd_req_op_cls_request_info_pagelist(
199 struct ceph_osd_request *osd_req,
200 unsigned int which, struct ceph_pagelist *pagelist)
201{
202 struct ceph_osd_data *osd_data;
203
863c7eb5 204 osd_data = osd_req_op_data(osd_req, which, cls, request_info);
a4ce40a9 205 ceph_osd_data_pagelist_init(osd_data, pagelist);
a4ce40a9
AE		 206}
207
04017e29
AE		 208void osd_req_op_cls_request_data_pagelist(
209 struct ceph_osd_request *osd_req,
210 unsigned int which, struct ceph_pagelist *pagelist)
211{
212 struct ceph_osd_data *osd_data;
213
863c7eb5 214 osd_data = osd_req_op_data(osd_req, which, cls, request_data);
04017e29
AE		 215 ceph_osd_data_pagelist_init(osd_data, pagelist);
216}
217EXPORT_SYMBOL(osd_req_op_cls_request_data_pagelist);
218
6c57b554
AE		 219void osd_req_op_cls_request_data_pages(struct ceph_osd_request *osd_req,
220 unsigned int which, struct page **pages, u64 length,
221 u32 alignment, bool pages_from_pool, bool own_pages)
222{
223 struct ceph_osd_data *osd_data;
224
225 osd_data = osd_req_op_data(osd_req, which, cls, request_data);
226 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
227 pages_from_pool, own_pages);
228}
229EXPORT_SYMBOL(osd_req_op_cls_request_data_pages);
230
a4ce40a9
AE		 231void osd_req_op_cls_response_data_pages(struct ceph_osd_request *osd_req,
232 unsigned int which, struct page **pages, u64 length,
233 u32 alignment, bool pages_from_pool, bool own_pages)
234{
235 struct ceph_osd_data *osd_data;
236
863c7eb5 237 osd_data = osd_req_op_data(osd_req, which, cls, response_data);
a4ce40a9
AE		 238 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
239 pages_from_pool, own_pages);
a4ce40a9
AE		 240}
241EXPORT_SYMBOL(osd_req_op_cls_response_data_pages);
242
23c08a9c
AE		 243static u64 ceph_osd_data_length(struct ceph_osd_data *osd_data)
244{
245 switch (osd_data->type) {
246 case CEPH_OSD_DATA_TYPE_NONE:
247 return 0;
248 case CEPH_OSD_DATA_TYPE_PAGES:
249 return osd_data->length;
250 case CEPH_OSD_DATA_TYPE_PAGELIST:
251 return (u64)osd_data->pagelist->length;
252#ifdef CONFIG_BLOCK
253 case CEPH_OSD_DATA_TYPE_BIO:
254 return (u64)osd_data->bio_length;
255#endif /* CONFIG_BLOCK */
256 default:
257 WARN(true, "unrecognized data type %d\n", (int)osd_data->type);
258 return 0;
259 }
260}
261
c54d47bf
AE		 262static void ceph_osd_data_release(struct ceph_osd_data *osd_data)
263{
5476492f 264 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES && osd_data->own_pages) {
c54d47bf
AE		 265 int num_pages;
266
267 num_pages = calc_pages_for((u64)osd_data->alignment,
268 (u64)osd_data->length);
269 ceph_release_page_vector(osd_data->pages, num_pages);
270 }
5476492f
AE		 271 ceph_osd_data_init(osd_data);
272}
273
274static void osd_req_op_data_release(struct ceph_osd_request *osd_req,
275 unsigned int which)
276{
277 struct ceph_osd_req_op *op;
278
279 BUG_ON(which >= osd_req->r_num_ops);
280 op = &osd_req->r_ops[which];
281
282 switch (op->op) {
283 case CEPH_OSD_OP_READ:
284 case CEPH_OSD_OP_WRITE:
285 ceph_osd_data_release(&op->extent.osd_data);
286 break;
287 case CEPH_OSD_OP_CALL:
288 ceph_osd_data_release(&op->cls.request_info);
04017e29 289 ceph_osd_data_release(&op->cls.request_data);
5476492f
AE		 290 ceph_osd_data_release(&op->cls.response_data);
291 break;
292 default:
293 break;
294 }
c54d47bf
AE		 295}
296
f24e9980
SW		 297/*
298 * requests
299 */
415e49a9 300void ceph_osdc_release_request(struct kref *kref)
f24e9980 301{
c54d47bf 302 struct ceph_osd_request *req;
5476492f 303 unsigned int which;
415e49a9 304
c54d47bf 305 req = container_of(kref, struct ceph_osd_request, r_kref);
415e49a9
SW		 306 if (req->r_request)
307 ceph_msg_put(req->r_request);
ace6d3a9 308 if (req->r_reply) {
8921d114 309 ceph_msg_revoke_incoming(req->r_reply);
ab8cb34a 310 ceph_msg_put(req->r_reply);
ace6d3a9 311 }
0fff87ec 312
5476492f
AE		 313 for (which = 0; which < req->r_num_ops; which++)
314 osd_req_op_data_release(req, which);
0fff87ec 315
415e49a9
SW		 316 ceph_put_snap_context(req->r_snapc);
317 if (req->r_mempool)
318 mempool_free(req, req->r_osdc->req_mempool);
319 else
5522ae0b
AE		 320 kmem_cache_free(ceph_osd_request_cache, req);
321
f24e9980 322}
3d14c5d2 323EXPORT_SYMBOL(ceph_osdc_release_request);
68b4476b 324
3499e8a5 325struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
f24e9980 326 struct ceph_snap_context *snapc,
1b83bef2 327 unsigned int num_ops,
3499e8a5 328 bool use_mempool,
54a54007 329 gfp_t gfp_flags)
f24e9980
SW		 330{
331 struct ceph_osd_request *req;
332 struct ceph_msg *msg;
1b83bef2
SW		 333 size_t msg_size;
334
79528734
AE		 335 BUILD_BUG_ON(CEPH_OSD_MAX_OP > U16_MAX);
336 BUG_ON(num_ops > CEPH_OSD_MAX_OP);
337
1b83bef2
SW		 338 msg_size = 4 + 4 + 8 + 8 + 4+8;
339 msg_size += 2 + 4 + 8 + 4 + 4; /* oloc */
340 msg_size += 1 + 8 + 4 + 4; /* pg_t */
341 msg_size += 4 + MAX_OBJ_NAME_SIZE;
342 msg_size += 2 + num_ops*sizeof(struct ceph_osd_op);
343 msg_size += 8; /* snapid */
344 msg_size += 8; /* snap_seq */
345 msg_size += 8 * (snapc ? snapc->num_snaps : 0); /* snaps */
346 msg_size += 4;
f24e9980
SW		 347
348 if (use_mempool) {
3499e8a5 349 req = mempool_alloc(osdc->req_mempool, gfp_flags);
f24e9980
SW		 350 memset(req, 0, sizeof(*req));
351 } else {
5522ae0b 352 req = kmem_cache_zalloc(ceph_osd_request_cache, gfp_flags);
f24e9980
SW		 353 }
354 if (req == NULL)
a79832f2 355 return NULL;
f24e9980 356
f24e9980
SW		 357 req->r_osdc = osdc;
358 req->r_mempool = use_mempool;
79528734 359 req->r_num_ops = num_ops;
68b4476b 360
415e49a9 361 kref_init(&req->r_kref);
f24e9980
SW		 362 init_completion(&req->r_completion);
363 init_completion(&req->r_safe_completion);
a978fa20 364 RB_CLEAR_NODE(&req->r_node);
f24e9980 365 INIT_LIST_HEAD(&req->r_unsafe_item);
a40c4f10
YS		 366 INIT_LIST_HEAD(&req->r_linger_item);
367 INIT_LIST_HEAD(&req->r_linger_osd);
935b639a 368 INIT_LIST_HEAD(&req->r_req_lru_item);
cd43045c
SW		 369 INIT_LIST_HEAD(&req->r_osd_item);
370
c16e7869
SW		 371 /* create reply message */
372 if (use_mempool)
373 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0);
374 else
375 msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY,
b61c2763 376 OSD_OPREPLY_FRONT_LEN, gfp_flags, true);
a79832f2 377 if (!msg) {
c16e7869 378 ceph_osdc_put_request(req);
a79832f2 379 return NULL;
c16e7869
SW		 380 }
381 req->r_reply = msg;
382
383 /* create request message; allow space for oid */
f24e9980 384 if (use_mempool)
8f3bc053 385 msg = ceph_msgpool_get(&osdc->msgpool_op, 0);
f24e9980 386 else
b61c2763 387 msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp_flags, true);
a79832f2 388 if (!msg) {
f24e9980 389 ceph_osdc_put_request(req);
a79832f2 390 return NULL;
f24e9980 391 }
68b4476b 392
f24e9980 393 memset(msg->front.iov_base, 0, msg->front.iov_len);
3499e8a5
YS		 394
395 req->r_request = msg;
3499e8a5
YS		 396
397 return req;
398}
3d14c5d2 399EXPORT_SYMBOL(ceph_osdc_alloc_request);
3499e8a5 400
a8dd0a37 401static bool osd_req_opcode_valid(u16 opcode)
68b4476b 402{
a8dd0a37 403 switch (opcode) {
68b4476b 404 case CEPH_OSD_OP_READ:
a8dd0a37 405 case CEPH_OSD_OP_STAT:
4c46459c
AE		 406 case CEPH_OSD_OP_MAPEXT:
407 case CEPH_OSD_OP_MASKTRUNC:
408 case CEPH_OSD_OP_SPARSE_READ:
a9f36c3e 409 case CEPH_OSD_OP_NOTIFY:
a8dd0a37 410 case CEPH_OSD_OP_NOTIFY_ACK:
4c46459c 411 case CEPH_OSD_OP_ASSERT_VER:
a8dd0a37 412 case CEPH_OSD_OP_WRITE:
4c46459c
AE		 413 case CEPH_OSD_OP_WRITEFULL:
414 case CEPH_OSD_OP_TRUNCATE:
415 case CEPH_OSD_OP_ZERO:
416 case CEPH_OSD_OP_DELETE:
417 case CEPH_OSD_OP_APPEND:
a8dd0a37 418 case CEPH_OSD_OP_STARTSYNC:
4c46459c
AE		 419 case CEPH_OSD_OP_SETTRUNC:
420 case CEPH_OSD_OP_TRIMTRUNC:
421 case CEPH_OSD_OP_TMAPUP:
422 case CEPH_OSD_OP_TMAPPUT:
423 case CEPH_OSD_OP_TMAPGET:
424 case CEPH_OSD_OP_CREATE:
a9f36c3e 425 case CEPH_OSD_OP_ROLLBACK:
a8dd0a37 426 case CEPH_OSD_OP_WATCH:
4c46459c
AE		 427 case CEPH_OSD_OP_OMAPGETKEYS:
428 case CEPH_OSD_OP_OMAPGETVALS:
429 case CEPH_OSD_OP_OMAPGETHEADER:
430 case CEPH_OSD_OP_OMAPGETVALSBYKEYS:
4c46459c
AE		 431 case CEPH_OSD_OP_OMAPSETVALS:
432 case CEPH_OSD_OP_OMAPSETHEADER:
433 case CEPH_OSD_OP_OMAPCLEAR:
434 case CEPH_OSD_OP_OMAPRMKEYS:
435 case CEPH_OSD_OP_OMAP_CMP:
436 case CEPH_OSD_OP_CLONERANGE:
437 case CEPH_OSD_OP_ASSERT_SRC_VERSION:
438 case CEPH_OSD_OP_SRC_CMPXATTR:
a9f36c3e 439 case CEPH_OSD_OP_GETXATTR:
4c46459c 440 case CEPH_OSD_OP_GETXATTRS:
a9f36c3e
AE		 441 case CEPH_OSD_OP_CMPXATTR:
442 case CEPH_OSD_OP_SETXATTR:
4c46459c
AE		 443 case CEPH_OSD_OP_SETXATTRS:
444 case CEPH_OSD_OP_RESETXATTRS:
445 case CEPH_OSD_OP_RMXATTR:
446 case CEPH_OSD_OP_PULL:
447 case CEPH_OSD_OP_PUSH:
448 case CEPH_OSD_OP_BALANCEREADS:
449 case CEPH_OSD_OP_UNBALANCEREADS:
450 case CEPH_OSD_OP_SCRUB:
451 case CEPH_OSD_OP_SCRUB_RESERVE:
452 case CEPH_OSD_OP_SCRUB_UNRESERVE:
453 case CEPH_OSD_OP_SCRUB_STOP:
454 case CEPH_OSD_OP_SCRUB_MAP:
455 case CEPH_OSD_OP_WRLOCK:
456 case CEPH_OSD_OP_WRUNLOCK:
457 case CEPH_OSD_OP_RDLOCK:
458 case CEPH_OSD_OP_RDUNLOCK:
459 case CEPH_OSD_OP_UPLOCK:
460 case CEPH_OSD_OP_DNLOCK:
a8dd0a37 461 case CEPH_OSD_OP_CALL:
4c46459c
AE		 462 case CEPH_OSD_OP_PGLS:
463 case CEPH_OSD_OP_PGLS_FILTER:
a8dd0a37
AE		 464 return true;
465 default:
466 return false;
467 }
468}
469
33803f33
AE		 470/*
471 * This is an osd op init function for opcodes that have no data or
472 * other information associated with them. It also serves as a
473 * common init routine for all the other init functions, below.
474 */
c99d2d4a 475static struct ceph_osd_req_op *
49719778 476_osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which,
c99d2d4a 477 u16 opcode)
33803f33 478{
c99d2d4a
AE		 479 struct ceph_osd_req_op *op;
480
481 BUG_ON(which >= osd_req->r_num_ops);
33803f33
AE		 482 BUG_ON(!osd_req_opcode_valid(opcode));
483
c99d2d4a 484 op = &osd_req->r_ops[which];
33803f33 485 memset(op, 0, sizeof (*op));
33803f33 486 op->op = opcode;
c99d2d4a
AE		 487
488 return op;
33803f33
AE		 489}
490
49719778
AE		 491void osd_req_op_init(struct ceph_osd_request *osd_req,
492 unsigned int which, u16 opcode)
493{
494 (void)_osd_req_op_init(osd_req, which, opcode);
495}
496EXPORT_SYMBOL(osd_req_op_init);
497
c99d2d4a
AE		 498void osd_req_op_extent_init(struct ceph_osd_request *osd_req,
499 unsigned int which, u16 opcode,
33803f33
AE		 500 u64 offset, u64 length,
501 u64 truncate_size, u32 truncate_seq)
502{
49719778 503 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
33803f33
AE		 504 size_t payload_len = 0;
505
506 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE);
507
33803f33
AE		 508 op->extent.offset = offset;
509 op->extent.length = length;
510 op->extent.truncate_size = truncate_size;
511 op->extent.truncate_seq = truncate_seq;
512 if (opcode == CEPH_OSD_OP_WRITE)
513 payload_len += length;
514
515 op->payload_len = payload_len;
516}
517EXPORT_SYMBOL(osd_req_op_extent_init);
518
c99d2d4a
AE		 519void osd_req_op_extent_update(struct ceph_osd_request *osd_req,
520 unsigned int which, u64 length)
e5975c7c 521{
c99d2d4a
AE		 522 struct ceph_osd_req_op *op;
523 u64 previous;
524
525 BUG_ON(which >= osd_req->r_num_ops);
526 op = &osd_req->r_ops[which];
527 previous = op->extent.length;
e5975c7c
AE		 528
529 if (length == previous)
530 return; /* Nothing to do */
531 BUG_ON(length > previous);
532
533 op->extent.length = length;
534 op->payload_len -= previous - length;
535}
536EXPORT_SYMBOL(osd_req_op_extent_update);
537
c99d2d4a 538void osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
04017e29 539 u16 opcode, const char *class, const char *method)
33803f33 540{
49719778 541 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
5f562df5 542 struct ceph_pagelist *pagelist;
33803f33
AE		 543 size_t payload_len = 0;
544 size_t size;
545
546 BUG_ON(opcode != CEPH_OSD_OP_CALL);
547
5f562df5
AE		 548 pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
549 BUG_ON(!pagelist);
550 ceph_pagelist_init(pagelist);
551
33803f33
AE		 552 op->cls.class_name = class;
553 size = strlen(class);
554 BUG_ON(size > (size_t) U8_MAX);
555 op->cls.class_len = size;
5f562df5 556 ceph_pagelist_append(pagelist, class, size);
33803f33
AE		 557 payload_len += size;
558
559 op->cls.method_name = method;
560 size = strlen(method);
561 BUG_ON(size > (size_t) U8_MAX);
562 op->cls.method_len = size;
5f562df5 563 ceph_pagelist_append(pagelist, method, size);
33803f33
AE		 564 payload_len += size;
565
a4ce40a9 566 osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist);
5f562df5 567
33803f33
AE		 568 op->cls.argc = 0; /* currently unused */
569
570 op->payload_len = payload_len;
571}
572EXPORT_SYMBOL(osd_req_op_cls_init);
8c042b0d 573
c99d2d4a
AE		 574void osd_req_op_watch_init(struct ceph_osd_request *osd_req,
575 unsigned int which, u16 opcode,
33803f33
AE		 576 u64 cookie, u64 version, int flag)
577{
49719778 578 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
33803f33 579
c99d2d4a 580 BUG_ON(opcode != CEPH_OSD_OP_NOTIFY_ACK && opcode != CEPH_OSD_OP_WATCH);
33803f33
AE		 581
582 op->watch.cookie = cookie;
9ef1ee5a 583 op->watch.ver = version;
33803f33 584 if (opcode == CEPH_OSD_OP_WATCH && flag)
c99d2d4a 585 op->watch.flag = (u8)1;
33803f33
AE		 586}
587EXPORT_SYMBOL(osd_req_op_watch_init);
588
90af3602 589static void ceph_osdc_msg_data_add(struct ceph_msg *msg,
ec9123c5
AE		 590 struct ceph_osd_data *osd_data)
591{
592 u64 length = ceph_osd_data_length(osd_data);
593
594 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
595 BUG_ON(length > (u64) SIZE_MAX);
596 if (length)
90af3602 597 ceph_msg_data_add_pages(msg, osd_data->pages,
ec9123c5
AE		 598 length, osd_data->alignment);
599 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
600 BUG_ON(!length);
90af3602 601 ceph_msg_data_add_pagelist(msg, osd_data->pagelist);
ec9123c5
AE		 602#ifdef CONFIG_BLOCK
603 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
90af3602 604 ceph_msg_data_add_bio(msg, osd_data->bio, length);
ec9123c5
AE		 605#endif
606 } else {
607 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
608 }
609}
610
a8dd0a37 611static u64 osd_req_encode_op(struct ceph_osd_request *req,
79528734 612 struct ceph_osd_op *dst, unsigned int which)
a8dd0a37 613{
79528734 614 struct ceph_osd_req_op *src;
04017e29 615 struct ceph_osd_data *osd_data;
54d50649 616 u64 request_data_len = 0;
04017e29 617 u64 data_length;
a8dd0a37 618
79528734
AE		 619 BUG_ON(which >= req->r_num_ops);
620 src = &req->r_ops[which];
a8dd0a37
AE		 621 if (WARN_ON(!osd_req_opcode_valid(src->op))) {
622 pr_err("unrecognized osd opcode %d\n", src->op);
623
624 return 0;
625 }
626
627 switch (src->op) {
628 case CEPH_OSD_OP_STAT:
49719778
AE		 629 osd_data = &src->raw_data_in;
630 ceph_osdc_msg_data_add(req->r_reply, osd_data);
a8dd0a37
AE		 631 break;
632 case CEPH_OSD_OP_READ:
633 case CEPH_OSD_OP_WRITE:
634 if (src->op == CEPH_OSD_OP_WRITE)
54d50649 635 request_data_len = src->extent.length;
a8dd0a37
AE		 636 dst->extent.offset = cpu_to_le64(src->extent.offset);
637 dst->extent.length = cpu_to_le64(src->extent.length);
638 dst->extent.truncate_size =
639 cpu_to_le64(src->extent.truncate_size);
640 dst->extent.truncate_seq =
641 cpu_to_le32(src->extent.truncate_seq);
04017e29 642 osd_data = &src->extent.osd_data;
5476492f 643 if (src->op == CEPH_OSD_OP_WRITE)
04017e29 644 ceph_osdc_msg_data_add(req->r_request, osd_data);
5476492f 645 else
04017e29 646 ceph_osdc_msg_data_add(req->r_reply, osd_data);
a8dd0a37
AE		 647 break;
648 case CEPH_OSD_OP_CALL:
a8dd0a37
AE		 649 dst->cls.class_len = src->cls.class_len;
650 dst->cls.method_len = src->cls.method_len;
04017e29
AE		 651 osd_data = &src->cls.request_info;
652 ceph_osdc_msg_data_add(req->r_request, osd_data);
653 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGELIST);
654 request_data_len = osd_data->pagelist->length;
655
656 osd_data = &src->cls.request_data;
657 data_length = ceph_osd_data_length(osd_data);
658 if (data_length) {
659 BUG_ON(osd_data->type == CEPH_OSD_DATA_TYPE_NONE);
660 dst->cls.indata_len = cpu_to_le32(data_length);
661 ceph_osdc_msg_data_add(req->r_request, osd_data);
662 src->payload_len += data_length;
663 request_data_len += data_length;
664 }
665 osd_data = &src->cls.response_data;
666 ceph_osdc_msg_data_add(req->r_reply, osd_data);
a8dd0a37
AE		 667 break;
668 case CEPH_OSD_OP_STARTSYNC:
669 break;
670 case CEPH_OSD_OP_NOTIFY_ACK:
671 case CEPH_OSD_OP_WATCH:
672 dst->watch.cookie = cpu_to_le64(src->watch.cookie);
673 dst->watch.ver = cpu_to_le64(src->watch.ver);
674 dst->watch.flag = src->watch.flag;
675 break;
676 default:
4c46459c 677 pr_err("unsupported osd opcode %s\n",
8f63ca2d 678 ceph_osd_op_name(src->op));
4c46459c 679 WARN_ON(1);
a8dd0a37
AE		 680
681 return 0;
68b4476b 682 }
a8dd0a37 683 dst->op = cpu_to_le16(src->op);
68b4476b 684 dst->payload_len = cpu_to_le32(src->payload_len);
175face2 685
54d50649 686 return request_data_len;
68b4476b
YS		 687}
688
3499e8a5
YS		 689/*
690 * build new request AND message, calculate layout, and adjust file
691 * extent as needed.
692 *
693 * if the file was recently truncated, we include information about its
694 * old and new size so that the object can be updated appropriately. (we
695 * avoid synchronously deleting truncated objects because it's slow.)
696 *
697 * if @do_sync, include a 'startsync' command so that the osd will flush
698 * data quickly.
699 */
700struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
701 struct ceph_file_layout *layout,
702 struct ceph_vino vino,
acead002 703 u64 off, u64 *plen, int num_ops,
3499e8a5
YS		 704 int opcode, int flags,
705 struct ceph_snap_context *snapc,
3499e8a5
YS		 706 u32 truncate_seq,
707 u64 truncate_size,
153e5167 708 bool use_mempool)
3499e8a5 709{
68b4476b 710 struct ceph_osd_request *req;
75d1c941
AE		 711 u64 objnum = 0;
712 u64 objoff = 0;
713 u64 objlen = 0;
d18d1e28
AE		 714 u32 object_size;
715 u64 object_base;
6816282d 716 int r;
68b4476b 717
d18d1e28 718 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE);
68b4476b 719
acead002 720 req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool,
ae7ca4a3 721 GFP_NOFS);
4ad12621 722 if (!req)
6816282d 723 return ERR_PTR(-ENOMEM);
79528734 724
d178a9e7 725 req->r_flags = flags;
3499e8a5
YS		 726
727 /* calculate max write size */
a19dadfb 728 r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
3ff5f385
AE		 729 if (r < 0) {
730 ceph_osdc_put_request(req);
6816282d 731 return ERR_PTR(r);
3ff5f385 732 }
a19dadfb 733
d18d1e28
AE		 734 object_size = le32_to_cpu(layout->fl_object_size);
735 object_base = off - objoff;
1f69fb06
YZ		 736 if (!(truncate_seq == 1 && truncate_size == -1ULL)) {
737 if (truncate_size <= object_base) {
738 truncate_size = 0;
739 } else {
740 truncate_size -= object_base;
741 if (truncate_size > object_size)
742 truncate_size = object_size;
743 }
a19dadfb 744 }
d18d1e28 745
c99d2d4a 746 osd_req_op_extent_init(req, 0, opcode, objoff, objlen,
b0270324 747 truncate_size, truncate_seq);
8c042b0d 748
acead002
AE		 749 /*
750 * A second op in the ops array means the caller wants to
751 * also issue a include a 'startsync' command so that the
752 * osd will flush data quickly.
753 */
754 if (num_ops > 1)
c99d2d4a 755 osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC);
d18d1e28 756
3499e8a5
YS		 757 req->r_file_layout = *layout; /* keep a copy */
758
75d1c941
AE		 759 snprintf(req->r_oid, sizeof(req->r_oid), "%llx.%08llx",
760 vino.ino, objnum);
dbe0fc41
AE		 761 req->r_oid_len = strlen(req->r_oid);
762
f24e9980
SW		 763 return req;
764}
3d14c5d2 765EXPORT_SYMBOL(ceph_osdc_new_request);
f24e9980
SW		 766
767/*
768 * We keep osd requests in an rbtree, sorted by ->r_tid.
769 */
770static void __insert_request(struct ceph_osd_client *osdc,
771 struct ceph_osd_request *new)
772{
773 struct rb_node **p = &osdc->requests.rb_node;
774 struct rb_node *parent = NULL;
775 struct ceph_osd_request *req = NULL;
776
777 while (*p) {
778 parent = *p;
779 req = rb_entry(parent, struct ceph_osd_request, r_node);
780 if (new->r_tid < req->r_tid)
781 p = &(*p)->rb_left;
782 else if (new->r_tid > req->r_tid)
783 p = &(*p)->rb_right;
784 else
785 BUG();
786 }
787
788 rb_link_node(&new->r_node, parent, p);
789 rb_insert_color(&new->r_node, &osdc->requests);
790}
791
792static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc,
793 u64 tid)
794{
795 struct ceph_osd_request *req;
796 struct rb_node *n = osdc->requests.rb_node;
797
798 while (n) {
799 req = rb_entry(n, struct ceph_osd_request, r_node);
800 if (tid < req->r_tid)
801 n = n->rb_left;
802 else if (tid > req->r_tid)
803 n = n->rb_right;
804 else
805 return req;
806 }
807 return NULL;
808}
809
810static struct ceph_osd_request *
811__lookup_request_ge(struct ceph_osd_client *osdc,
812 u64 tid)
813{
814 struct ceph_osd_request *req;
815 struct rb_node *n = osdc->requests.rb_node;
816
817 while (n) {
818 req = rb_entry(n, struct ceph_osd_request, r_node);
819 if (tid < req->r_tid) {
820 if (!n->rb_left)
821 return req;
822 n = n->rb_left;
823 } else if (tid > req->r_tid) {
824 n = n->rb_right;
825 } else {
826 return req;
827 }
828 }
829 return NULL;
830}
831
6f6c7006
SW		 832/*
833 * Resubmit requests pending on the given osd.
834 */
835static void __kick_osd_requests(struct ceph_osd_client *osdc,
836 struct ceph_osd *osd)
837{
a40c4f10 838 struct ceph_osd_request *req, *nreq;
e02493c0 839 LIST_HEAD(resend);
6f6c7006
SW		 840 int err;
841
842 dout("__kick_osd_requests osd%d\n", osd->o_osd);
843 err = __reset_osd(osdc, osd);
685a7555 844 if (err)
6f6c7006 845 return;
e02493c0
AE		 846 /*
847 * Build up a list of requests to resend by traversing the
848 * osd's list of requests. Requests for a given object are
849 * sent in tid order, and that is also the order they're
850 * kept on this list. Therefore all requests that are in
851 * flight will be found first, followed by all requests that
852 * have not yet been sent. And to resend requests while
853 * preserving this order we will want to put any sent
854 * requests back on the front of the osd client's unsent
855 * list.
856 *
857 * So we build a separate ordered list of already-sent
858 * requests for the affected osd and splice it onto the
859 * front of the osd client's unsent list. Once we've seen a
860 * request that has not yet been sent we're done. Those
861 * requests are already sitting right where they belong.
862 */
6f6c7006 863 list_for_each_entry(req, &osd->o_requests, r_osd_item) {
e02493c0
AE		 864 if (!req->r_sent)
865 break;
866 list_move_tail(&req->r_req_lru_item, &resend);
867 dout("requeueing %p tid %llu osd%d\n", req, req->r_tid,
6f6c7006 868 osd->o_osd);
a40c4f10
YS		 869 if (!req->r_linger)
870 req->r_flags |= CEPH_OSD_FLAG_RETRY;
871 }
e02493c0 872 list_splice(&resend, &osdc->req_unsent);
a40c4f10 873
e02493c0
AE		 874 /*
875 * Linger requests are re-registered before sending, which
876 * sets up a new tid for each. We add them to the unsent
877 * list at the end to keep things in tid order.
878 */
a40c4f10
YS		 879 list_for_each_entry_safe(req, nreq, &osd->o_linger_requests,
880 r_linger_osd) {
77f38e0e
SW		 881 /*
882 * reregister request prior to unregistering linger so
883 * that r_osd is preserved.
884 */
885 BUG_ON(!list_empty(&req->r_req_lru_item));
a40c4f10 886 __register_request(osdc, req);
e02493c0 887 list_add_tail(&req->r_req_lru_item, &osdc->req_unsent);
ad885927 888 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
77f38e0e 889 __unregister_linger_request(osdc, req);
a40c4f10
YS		 890 dout("requeued lingering %p tid %llu osd%d\n", req, req->r_tid,
891 osd->o_osd);
6f6c7006
SW		 892 }
893}
894
f24e9980 895/*
81b024e7 896 * If the osd connection drops, we need to resubmit all requests.
f24e9980
SW		 897 */
898static void osd_reset(struct ceph_connection *con)
899{
900 struct ceph_osd *osd = con->private;
901 struct ceph_osd_client *osdc;
902
903 if (!osd)
904 return;
905 dout("osd_reset osd%d\n", osd->o_osd);
906 osdc = osd->o_osdc;
f24e9980 907 down_read(&osdc->map_sem);
83aff95e
SW		 908 mutex_lock(&osdc->request_mutex);
909 __kick_osd_requests(osdc, osd);
f9d25199 910 __send_queued(osdc);
83aff95e 911 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 912 up_read(&osdc->map_sem);
913}
914
915/*
916 * Track open sessions with osds.
917 */
e10006f8 918static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
f24e9980
SW		 919{
920 struct ceph_osd *osd;
921
922 osd = kzalloc(sizeof(*osd), GFP_NOFS);
923 if (!osd)
924 return NULL;
925
926 atomic_set(&osd->o_ref, 1);
927 osd->o_osdc = osdc;
e10006f8 928 osd->o_osd = onum;
f407731d 929 RB_CLEAR_NODE(&osd->o_node);
f24e9980 930 INIT_LIST_HEAD(&osd->o_requests);
a40c4f10 931 INIT_LIST_HEAD(&osd->o_linger_requests);
f5a2041b 932 INIT_LIST_HEAD(&osd->o_osd_lru);
f24e9980
SW		 933 osd->o_incarnation = 1;
934
b7a9e5dd 935 ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
4e7a5dcd 936
422d2cb8 937 INIT_LIST_HEAD(&osd->o_keepalive_item);
f24e9980
SW		 938 return osd;
939}
940
941static struct ceph_osd *get_osd(struct ceph_osd *osd)
942{
943 if (atomic_inc_not_zero(&osd->o_ref)) {
944 dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
945 atomic_read(&osd->o_ref));
946 return osd;
947 } else {
948 dout("get_osd %p FAIL\n", osd);
949 return NULL;
950 }
951}
952
953static void put_osd(struct ceph_osd *osd)
954{
955 dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
956 atomic_read(&osd->o_ref) - 1);
a255651d 957 if (atomic_dec_and_test(&osd->o_ref) && osd->o_auth.authorizer) {
79494d1b
SW		 958 struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth;
959
27859f97 960 ceph_auth_destroy_authorizer(ac, osd->o_auth.authorizer);
f24e9980 961 kfree(osd);
79494d1b 962 }
f24e9980
SW		 963}
964
965/*
966 * remove an osd from our map
967 */
f5a2041b 968static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
f24e9980 969{
f5a2041b 970 dout("__remove_osd %p\n", osd);
f24e9980
SW		 971 BUG_ON(!list_empty(&osd->o_requests));
972 rb_erase(&osd->o_node, &osdc->osds);
f5a2041b 973 list_del_init(&osd->o_osd_lru);
f24e9980
SW		 974 ceph_con_close(&osd->o_con);
975 put_osd(osd);
976}
977
aca420bc
SW		 978static void remove_all_osds(struct ceph_osd_client *osdc)
979{
048a9d2d 980 dout("%s %p\n", __func__, osdc);
aca420bc
SW		 981 mutex_lock(&osdc->request_mutex);
982 while (!RB_EMPTY_ROOT(&osdc->osds)) {
983 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
984 struct ceph_osd, o_node);
985 __remove_osd(osdc, osd);
986 }
987 mutex_unlock(&osdc->request_mutex);
988}
989
f5a2041b
YS		 990static void __move_osd_to_lru(struct ceph_osd_client *osdc,
991 struct ceph_osd *osd)
992{
993 dout("__move_osd_to_lru %p\n", osd);
994 BUG_ON(!list_empty(&osd->o_osd_lru));
995 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
3d14c5d2 996 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl * HZ;
f5a2041b
YS		 997}
998
999static void __remove_osd_from_lru(struct ceph_osd *osd)
1000{
1001 dout("__remove_osd_from_lru %p\n", osd);
1002 if (!list_empty(&osd->o_osd_lru))
1003 list_del_init(&osd->o_osd_lru);
1004}
1005
aca420bc 1006static void remove_old_osds(struct ceph_osd_client *osdc)
f5a2041b
YS		 1007{
1008 struct ceph_osd *osd, *nosd;
1009
1010 dout("__remove_old_osds %p\n", osdc);
1011 mutex_lock(&osdc->request_mutex);
1012 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
aca420bc 1013 if (time_before(jiffies, osd->lru_ttl))
f5a2041b
YS		 1014 break;
1015 __remove_osd(osdc, osd);
1016 }
1017 mutex_unlock(&osdc->request_mutex);
1018}
1019
f24e9980
SW		 1020/*
1021 * reset osd connect
1022 */
f5a2041b 1023static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
f24e9980 1024{
c3acb181 1025 struct ceph_entity_addr *peer_addr;
f24e9980 1026
f5a2041b 1027 dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
a40c4f10
YS		 1028 if (list_empty(&osd->o_requests) &&
1029 list_empty(&osd->o_linger_requests)) {
f5a2041b 1030 __remove_osd(osdc, osd);
c3acb181
AE		 1031
1032 return -ENODEV;
1033 }
1034
1035 peer_addr = &osdc->osdmap->osd_addr[osd->o_osd];
1036 if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1037 !ceph_con_opened(&osd->o_con)) {
1038 struct ceph_osd_request *req;
1039
87b315a5
SW		 1040 dout(" osd addr hasn't changed and connection never opened,"
1041 " letting msgr retry");
1042 /* touch each r_stamp for handle_timeout()'s benfit */
1043 list_for_each_entry(req, &osd->o_requests, r_osd_item)
1044 req->r_stamp = jiffies;
c3acb181
AE		 1045
1046 return -EAGAIN;
f24e9980 1047 }
c3acb181
AE		 1048
1049 ceph_con_close(&osd->o_con);
1050 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1051 osd->o_incarnation++;
1052
1053 return 0;
f24e9980
SW		 1054}
1055
1056static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new)
1057{
1058 struct rb_node **p = &osdc->osds.rb_node;
1059 struct rb_node *parent = NULL;
1060 struct ceph_osd *osd = NULL;
1061
aca420bc 1062 dout("__insert_osd %p osd%d\n", new, new->o_osd);
f24e9980
SW		 1063 while (*p) {
1064 parent = *p;
1065 osd = rb_entry(parent, struct ceph_osd, o_node);
1066 if (new->o_osd < osd->o_osd)
1067 p = &(*p)->rb_left;
1068 else if (new->o_osd > osd->o_osd)
1069 p = &(*p)->rb_right;
1070 else
1071 BUG();
1072 }
1073
1074 rb_link_node(&new->o_node, parent, p);
1075 rb_insert_color(&new->o_node, &osdc->osds);
1076}
1077
1078static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o)
1079{
1080 struct ceph_osd *osd;
1081 struct rb_node *n = osdc->osds.rb_node;
1082
1083 while (n) {
1084 osd = rb_entry(n, struct ceph_osd, o_node);
1085 if (o < osd->o_osd)
1086 n = n->rb_left;
1087 else if (o > osd->o_osd)
1088 n = n->rb_right;
1089 else
1090 return osd;
1091 }
1092 return NULL;
1093}
1094
422d2cb8
YS		 1095static void __schedule_osd_timeout(struct ceph_osd_client *osdc)
1096{
1097 schedule_delayed_work(&osdc->timeout_work,
3d14c5d2 1098 osdc->client->options->osd_keepalive_timeout * HZ);
422d2cb8
YS		 1099}
1100
1101static void __cancel_osd_timeout(struct ceph_osd_client *osdc)
1102{
1103 cancel_delayed_work(&osdc->timeout_work);
1104}
f24e9980
SW		 1105
1106/*
1107 * Register request, assign tid. If this is the first request, set up
1108 * the timeout event.
1109 */
a40c4f10
YS		 1110static void __register_request(struct ceph_osd_client *osdc,
1111 struct ceph_osd_request *req)
f24e9980 1112{
f24e9980 1113 req->r_tid = ++osdc->last_tid;
6df058c0 1114 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
77f38e0e 1115 dout("__register_request %p tid %lld\n", req, req->r_tid);
f24e9980
SW		 1116 __insert_request(osdc, req);
1117 ceph_osdc_get_request(req);
1118 osdc->num_requests++;
f24e9980 1119 if (osdc->num_requests == 1) {
422d2cb8
YS		 1120 dout(" first request, scheduling timeout\n");
1121 __schedule_osd_timeout(osdc);
f24e9980 1122 }
a40c4f10
YS		 1123}
1124
f24e9980
SW		 1125/*
1126 * called under osdc->request_mutex
1127 */
1128static void __unregister_request(struct ceph_osd_client *osdc,
1129 struct ceph_osd_request *req)
1130{
35f9f8a0
SW		 1131 if (RB_EMPTY_NODE(&req->r_node)) {
1132 dout("__unregister_request %p tid %lld not registered\n",
1133 req, req->r_tid);
1134 return;
1135 }
1136
f24e9980
SW		 1137 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
1138 rb_erase(&req->r_node, &osdc->requests);
1139 osdc->num_requests--;
1140
0ba6478d
SW		 1141 if (req->r_osd) {
1142 /* make sure the original request isn't in flight. */
6740a845 1143 ceph_msg_revoke(req->r_request);
0ba6478d
SW		 1144
1145 list_del_init(&req->r_osd_item);
a40c4f10
YS		 1146 if (list_empty(&req->r_osd->o_requests) &&
1147 list_empty(&req->r_osd->o_linger_requests)) {
1148 dout("moving osd to %p lru\n", req->r_osd);
f5a2041b 1149 __move_osd_to_lru(osdc, req->r_osd);
a40c4f10 1150 }
fbdb9190 1151 if (list_empty(&req->r_linger_item))
a40c4f10 1152 req->r_osd = NULL;
0ba6478d 1153 }
f24e9980 1154
7d5f2481 1155 list_del_init(&req->r_req_lru_item);
f24e9980
SW		 1156 ceph_osdc_put_request(req);
1157
422d2cb8
YS		 1158 if (osdc->num_requests == 0) {
1159 dout(" no requests, canceling timeout\n");
1160 __cancel_osd_timeout(osdc);
f24e9980
SW		 1161 }
1162}
1163
1164/*
1165 * Cancel a previously queued request message
1166 */
1167static void __cancel_request(struct ceph_osd_request *req)
1168{
6bc18876 1169 if (req->r_sent && req->r_osd) {
6740a845 1170 ceph_msg_revoke(req->r_request);
f24e9980
SW		 1171 req->r_sent = 0;
1172 }
1173}
1174
a40c4f10
YS		 1175static void __register_linger_request(struct ceph_osd_client *osdc,
1176 struct ceph_osd_request *req)
1177{
1178 dout("__register_linger_request %p\n", req);
1d02ec7f 1179 ceph_osdc_get_request(req);
a40c4f10 1180 list_add_tail(&req->r_linger_item, &osdc->req_linger);
6194ea89
SW		 1181 if (req->r_osd)
1182 list_add_tail(&req->r_linger_osd,
1183 &req->r_osd->o_linger_requests);
a40c4f10
YS		 1184}
1185
1186static void __unregister_linger_request(struct ceph_osd_client *osdc,
1187 struct ceph_osd_request *req)
1188{
1189 dout("__unregister_linger_request %p\n", req);
61c74035 1190 list_del_init(&req->r_linger_item);
a40c4f10 1191 if (req->r_osd) {
a40c4f10
YS		 1192 list_del_init(&req->r_linger_osd);
1193
1194 if (list_empty(&req->r_osd->o_requests) &&
1195 list_empty(&req->r_osd->o_linger_requests)) {
1196 dout("moving osd to %p lru\n", req->r_osd);
1197 __move_osd_to_lru(osdc, req->r_osd);
1198 }
fbdb9190
SW		 1199 if (list_empty(&req->r_osd_item))
1200 req->r_osd = NULL;
a40c4f10 1201 }
1d02ec7f 1202 ceph_osdc_put_request(req);
a40c4f10
YS		 1203}
1204
1205void ceph_osdc_unregister_linger_request(struct ceph_osd_client *osdc,
1206 struct ceph_osd_request *req)
1207{
1208 mutex_lock(&osdc->request_mutex);
1209 if (req->r_linger) {
c10ebbf5 1210 req->r_linger = 0;
1d02ec7f 1211 __unregister_linger_request(osdc, req);
a40c4f10
YS		 1212 }
1213 mutex_unlock(&osdc->request_mutex);
1214}
1215EXPORT_SYMBOL(ceph_osdc_unregister_linger_request);
1216
1217void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
1218 struct ceph_osd_request *req)
1219{
1220 if (!req->r_linger) {
1221 dout("set_request_linger %p\n", req);
1222 req->r_linger = 1;
a40c4f10
YS		 1223 }
1224}
1225EXPORT_SYMBOL(ceph_osdc_set_request_linger);
1226
4892ed8d
JD		 1227/*
1228 * Returns whether a request should be blocked from being sent
1229 * based on the current osdmap and osd_client settings.
1230 *
1231 * Caller should hold map_sem for read.
1232 */
1233static bool __req_should_be_paused(struct ceph_osd_client *osdc,
1234 struct ceph_osd_request *req)
1235{
1236 bool pauserd = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD);
1237 bool pausewr = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR) ||
1238 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
1239 return (req->r_flags & CEPH_OSD_FLAG_READ && pauserd) ||
1240 (req->r_flags & CEPH_OSD_FLAG_WRITE && pausewr);
1241}
1242
f24e9980
SW		 1243/*
1244 * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
1245 * (as needed), and set the request r_osd appropriately. If there is
25985edc 1246 * no up osd, set r_osd to NULL. Move the request to the appropriate list
6f6c7006 1247 * (unsent, homeless) or leave on in-flight lru.
f24e9980
SW		 1248 *
1249 * Return 0 if unchanged, 1 if changed, or negative on error.
1250 *
1251 * Caller should hold map_sem for read and request_mutex.
1252 */
6f6c7006 1253static int __map_request(struct ceph_osd_client *osdc,
38d6453c 1254 struct ceph_osd_request *req, int force_resend)
f24e9980 1255{
5b191d99 1256 struct ceph_pg pgid;
d85b7056
SW		 1257 int acting[CEPH_PG_MAX_SIZE];
1258 int o = -1, num = 0;
f24e9980 1259 int err;
4892ed8d 1260 bool was_paused;
f24e9980 1261
6f6c7006 1262 dout("map_request %p tid %lld\n", req, req->r_tid);
41766f87
AE		 1263 err = ceph_calc_ceph_pg(&pgid, req->r_oid, osdc->osdmap,
1264 ceph_file_layout_pg_pool(req->r_file_layout));
6f6c7006
SW		 1265 if (err) {
1266 list_move(&req->r_req_lru_item, &osdc->req_notarget);
f24e9980 1267 return err;
6f6c7006 1268 }
7740a42f
SW		 1269 req->r_pgid = pgid;
1270
d85b7056
SW		 1271 err = ceph_calc_pg_acting(osdc->osdmap, pgid, acting);
1272 if (err > 0) {
1273 o = acting[0];
1274 num = err;
1275 }
f24e9980 1276
4892ed8d
JD		 1277 was_paused = req->r_paused;
1278 req->r_paused = __req_should_be_paused(osdc, req);
1279 if (was_paused && !req->r_paused)
1280 force_resend = 1;
1281
38d6453c
SW		 1282 if ((!force_resend &&
1283 req->r_osd && req->r_osd->o_osd == o &&
d85b7056
SW		 1284 req->r_sent >= req->r_osd->o_incarnation &&
1285 req->r_num_pg_osds == num &&
1286 memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
4892ed8d
JD		 1287 (req->r_osd == NULL && o == -1) ||
1288 req->r_paused)
f24e9980
SW		 1289 return 0; /* no change */
1290
5b191d99
SW		 1291 dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n",
1292 req->r_tid, pgid.pool, pgid.seed, o,
f24e9980
SW		 1293 req->r_osd ? req->r_osd->o_osd : -1);
1294
d85b7056
SW		 1295 /* record full pg acting set */
1296 memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
1297 req->r_num_pg_osds = num;
1298
f24e9980
SW		 1299 if (req->r_osd) {
1300 __cancel_request(req);
1301 list_del_init(&req->r_osd_item);
f24e9980
SW		 1302 req->r_osd = NULL;
1303 }
1304
1305 req->r_osd = __lookup_osd(osdc, o);
1306 if (!req->r_osd && o >= 0) {
c99eb1c7 1307 err = -ENOMEM;
e10006f8 1308 req->r_osd = create_osd(osdc, o);
6f6c7006
SW		 1309 if (!req->r_osd) {
1310 list_move(&req->r_req_lru_item, &osdc->req_notarget);
c99eb1c7 1311 goto out;
6f6c7006 1312 }
f24e9980 1313
6f6c7006 1314 dout("map_request osd %p is osd%d\n", req->r_osd, o);
f24e9980
SW		 1315 __insert_osd(osdc, req->r_osd);
1316
b7a9e5dd
SW		 1317 ceph_con_open(&req->r_osd->o_con,
1318 CEPH_ENTITY_TYPE_OSD, o,
1319 &osdc->osdmap->osd_addr[o]);
f24e9980
SW		 1320 }
1321
f5a2041b
YS		 1322 if (req->r_osd) {
1323 __remove_osd_from_lru(req->r_osd);
ad885927
AE		 1324 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
1325 list_move_tail(&req->r_req_lru_item, &osdc->req_unsent);
6f6c7006 1326 } else {
ad885927 1327 list_move_tail(&req->r_req_lru_item, &osdc->req_notarget);
f5a2041b 1328 }
d85b7056 1329 err = 1; /* osd or pg changed */
f24e9980
SW		 1330
1331out:
f24e9980
SW		 1332 return err;
1333}
1334
1335/*
1336 * caller should hold map_sem (for read) and request_mutex
1337 */
56e925b6
SW		 1338static void __send_request(struct ceph_osd_client *osdc,
1339 struct ceph_osd_request *req)
f24e9980 1340{
1b83bef2 1341 void *p;
f24e9980 1342
1b83bef2
SW		 1343 dout("send_request %p tid %llu to osd%d flags %d pg %lld.%x\n",
1344 req, req->r_tid, req->r_osd->o_osd, req->r_flags,
1345 (unsigned long long)req->r_pgid.pool, req->r_pgid.seed);
1346
1347 /* fill in message content that changes each time we send it */
1348 put_unaligned_le32(osdc->osdmap->epoch, req->r_request_osdmap_epoch);
1349 put_unaligned_le32(req->r_flags, req->r_request_flags);
1350 put_unaligned_le64(req->r_pgid.pool, req->r_request_pool);
1351 p = req->r_request_pgid;
1352 ceph_encode_64(&p, req->r_pgid.pool);
1353 ceph_encode_32(&p, req->r_pgid.seed);
1354 put_unaligned_le64(1, req->r_request_attempts); /* FIXME */
1355 memcpy(req->r_request_reassert_version, &req->r_reassert_version,
1356 sizeof(req->r_reassert_version));
2169aea6 1357
3dd72fc0 1358 req->r_stamp = jiffies;
07a27e22 1359 list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
f24e9980
SW		 1360
1361 ceph_msg_get(req->r_request); /* send consumes a ref */
26be8808 1362
f24e9980 1363 req->r_sent = req->r_osd->o_incarnation;
26be8808
AE		 1364
1365 ceph_con_send(&req->r_osd->o_con, req->r_request);
f24e9980
SW		 1366}
1367
6f6c7006
SW		 1368/*
1369 * Send any requests in the queue (req_unsent).
1370 */
f9d25199 1371static void __send_queued(struct ceph_osd_client *osdc)
6f6c7006
SW		 1372{
1373 struct ceph_osd_request *req, *tmp;
1374
f9d25199
AE		 1375 dout("__send_queued\n");
1376 list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item)
6f6c7006 1377 __send_request(osdc, req);
6f6c7006
SW		 1378}
1379
f24e9980
SW		 1380/*
1381 * Timeout callback, called every N seconds when 1 or more osd
1382 * requests has been active for more than N seconds. When this
1383 * happens, we ping all OSDs with requests who have timed out to
1384 * ensure any communications channel reset is detected. Reset the
1385 * request timeouts another N seconds in the future as we go.
1386 * Reschedule the timeout event another N seconds in future (unless
1387 * there are no open requests).
1388 */
1389static void handle_timeout(struct work_struct *work)
1390{
1391 struct ceph_osd_client *osdc =
1392 container_of(work, struct ceph_osd_client, timeout_work.work);
83aff95e 1393 struct ceph_osd_request *req;
f24e9980 1394 struct ceph_osd *osd;
422d2cb8 1395 unsigned long keepalive =
3d14c5d2 1396 osdc->client->options->osd_keepalive_timeout * HZ;
422d2cb8 1397 struct list_head slow_osds;
f24e9980
SW		 1398 dout("timeout\n");
1399 down_read(&osdc->map_sem);
1400
1401 ceph_monc_request_next_osdmap(&osdc->client->monc);
1402
1403 mutex_lock(&osdc->request_mutex);
f24e9980 1404
422d2cb8
YS		 1405 /*
1406 * ping osds that are a bit slow. this ensures that if there
1407 * is a break in the TCP connection we will notice, and reopen
1408 * a connection with that osd (from the fault callback).
1409 */
1410 INIT_LIST_HEAD(&slow_osds);
1411 list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
3dd72fc0 1412 if (time_before(jiffies, req->r_stamp + keepalive))
422d2cb8
YS		 1413 break;
1414
1415 osd = req->r_osd;
1416 BUG_ON(!osd);
1417 dout(" tid %llu is slow, will send keepalive on osd%d\n",
f24e9980 1418 req->r_tid, osd->o_osd);
422d2cb8
YS		 1419 list_move_tail(&osd->o_keepalive_item, &slow_osds);
1420 }
1421 while (!list_empty(&slow_osds)) {
1422 osd = list_entry(slow_osds.next, struct ceph_osd,
1423 o_keepalive_item);
1424 list_del_init(&osd->o_keepalive_item);
f24e9980
SW		 1425 ceph_con_keepalive(&osd->o_con);
1426 }
1427
422d2cb8 1428 __schedule_osd_timeout(osdc);
f9d25199 1429 __send_queued(osdc);
f24e9980 1430 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 1431 up_read(&osdc->map_sem);
1432}
1433
f5a2041b
YS		 1434static void handle_osds_timeout(struct work_struct *work)
1435{
1436 struct ceph_osd_client *osdc =
1437 container_of(work, struct ceph_osd_client,
1438 osds_timeout_work.work);
1439 unsigned long delay =
3d14c5d2 1440 osdc->client->options->osd_idle_ttl * HZ >> 2;
f5a2041b
YS		 1441
1442 dout("osds timeout\n");
1443 down_read(&osdc->map_sem);
aca420bc 1444 remove_old_osds(osdc);
f5a2041b
YS		 1445 up_read(&osdc->map_sem);
1446
1447 schedule_delayed_work(&osdc->osds_timeout_work,
1448 round_jiffies_relative(delay));
1449}
1450
25845472
SW		 1451static void complete_request(struct ceph_osd_request *req)
1452{
25845472
SW		 1453 complete_all(&req->r_safe_completion); /* fsync waiter */
1454}
1455
f24e9980
SW		 1456/*
1457 * handle osd op reply. either call the callback if it is specified,
1458 * or do the completion to wake up the waiting thread.
1459 */
350b1c32
SW		 1460static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg,
1461 struct ceph_connection *con)
f24e9980 1462{
1b83bef2 1463 void *p, *end;
f24e9980
SW		 1464 struct ceph_osd_request *req;
1465 u64 tid;
1b83bef2 1466 int object_len;
79528734
AE		 1467 unsigned int numops;
1468 int payload_len, flags;
0ceed5db 1469 s32 result;
1b83bef2
SW		 1470 s32 retry_attempt;
1471 struct ceph_pg pg;
1472 int err;
1473 u32 reassert_epoch;
1474 u64 reassert_version;
1475 u32 osdmap_epoch;
0d5af164 1476 int already_completed;
9fc6e064 1477 u32 bytes;
79528734 1478 unsigned int i;
f24e9980 1479
6df058c0 1480 tid = le64_to_cpu(msg->hdr.tid);
1b83bef2
SW		 1481 dout("handle_reply %p tid %llu\n", msg, tid);
1482
1483 p = msg->front.iov_base;
1484 end = p + msg->front.iov_len;
1485
1486 ceph_decode_need(&p, end, 4, bad);
1487 object_len = ceph_decode_32(&p);
1488 ceph_decode_need(&p, end, object_len, bad);
1489 p += object_len;
1490
ef4859d6 1491 err = ceph_decode_pgid(&p, end, &pg);
1b83bef2 1492 if (err)
f24e9980 1493 goto bad;
1b83bef2
SW		 1494
1495 ceph_decode_need(&p, end, 8 + 4 + 4 + 8 + 4, bad);
1496 flags = ceph_decode_64(&p);
1497 result = ceph_decode_32(&p);
1498 reassert_epoch = ceph_decode_32(&p);
1499 reassert_version = ceph_decode_64(&p);
1500 osdmap_epoch = ceph_decode_32(&p);
1501
f24e9980
SW		 1502 /* lookup */
1503 mutex_lock(&osdc->request_mutex);
1504 req = __lookup_request(osdc, tid);
1505 if (req == NULL) {
1506 dout("handle_reply tid %llu dne\n", tid);
8058fd45 1507 goto bad_mutex;
f24e9980
SW		 1508 }
1509 ceph_osdc_get_request(req);
1b83bef2
SW		 1510
1511 dout("handle_reply %p tid %llu req %p result %d\n", msg, tid,
1512 req, result);
1513
e9e4b13a 1514 ceph_decode_need(&p, end, 4, bad_put);
1b83bef2
SW		 1515 numops = ceph_decode_32(&p);
1516 if (numops > CEPH_OSD_MAX_OP)
1517 goto bad_put;
1518 if (numops != req->r_num_ops)
1519 goto bad_put;
1520 payload_len = 0;
e9e4b13a 1521 ceph_decode_need(&p, end, numops * sizeof(struct ceph_osd_op), bad_put);
1b83bef2
SW		 1522 for (i = 0; i < numops; i++) {
1523 struct ceph_osd_op *op = p;
1524 int len;
1525
1526 len = le32_to_cpu(op->payload_len);
1527 req->r_reply_op_len[i] = len;
1528 dout(" op %d has %d bytes\n", i, len);
1529 payload_len += len;
1530 p += sizeof(*op);
1531 }
9fc6e064
AE		 1532 bytes = le32_to_cpu(msg->hdr.data_len);
1533 if (payload_len != bytes) {
1b83bef2 1534 pr_warning("sum of op payload lens %d != data_len %d",
9fc6e064 1535 payload_len, bytes);
1b83bef2
SW		 1536 goto bad_put;
1537 }
1538
e9e4b13a 1539 ceph_decode_need(&p, end, 4 + numops * 4, bad_put);
1b83bef2
SW		 1540 retry_attempt = ceph_decode_32(&p);
1541 for (i = 0; i < numops; i++)
1542 req->r_reply_op_result[i] = ceph_decode_32(&p);
f24e9980 1543
aede2cb5
YZ		 1544 already_completed = req->r_got_reply;
1545
f24e9980 1546 if (!req->r_got_reply) {
f24e9980 1547
1b83bef2 1548 req->r_result = result;
f24e9980
SW		 1549 dout("handle_reply result %d bytes %d\n", req->r_result,
1550 bytes);
1551 if (req->r_result == 0)
1552 req->r_result = bytes;
1553
1554 /* in case this is a write and we need to replay, */
1b83bef2
SW		 1555 req->r_reassert_version.epoch = cpu_to_le32(reassert_epoch);
1556 req->r_reassert_version.version = cpu_to_le64(reassert_version);
f24e9980
SW		 1557
1558 req->r_got_reply = 1;
1559 } else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) {
1560 dout("handle_reply tid %llu dup ack\n", tid);
34b43a56 1561 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 1562 goto done;
1563 }
1564
1565 dout("handle_reply tid %llu flags %d\n", tid, flags);
1566
a40c4f10
YS		 1567 if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK))
1568 __register_linger_request(osdc, req);
1569
f24e9980 1570 /* either this is a read, or we got the safe response */
0ceed5db
SW		 1571 if (result < 0 ||
1572 (flags & CEPH_OSD_FLAG_ONDISK) ||
f24e9980
SW		 1573 ((flags & CEPH_OSD_FLAG_WRITE) == 0))
1574 __unregister_request(osdc, req);
1575
1576 mutex_unlock(&osdc->request_mutex);
1577
aede2cb5 1578 if (!already_completed) {
ef0b67e4
YZ		 1579 if (req->r_unsafe_callback &&
1580 result >= 0 && !(flags & CEPH_OSD_FLAG_ONDISK))
1581 req->r_unsafe_callback(req, true);
aede2cb5
YZ		 1582 if (req->r_callback)
1583 req->r_callback(req, msg);
1584 else
1585 complete_all(&req->r_completion);
1586 }
f24e9980 1587
ef0b67e4
YZ		 1588 if (flags & CEPH_OSD_FLAG_ONDISK) {
1589 if (req->r_unsafe_callback && already_completed)
1590 req->r_unsafe_callback(req, false);
25845472 1591 complete_request(req);
ef0b67e4 1592 }
f24e9980
SW		 1593
1594done:
a40c4f10 1595 dout("req=%p req->r_linger=%d\n", req, req->r_linger);
f24e9980
SW		 1596 ceph_osdc_put_request(req);
1597 return;
1598
1b83bef2
SW		 1599bad_put:
1600 ceph_osdc_put_request(req);
8058fd45
AE		 1601bad_mutex:
1602 mutex_unlock(&osdc->request_mutex);
f24e9980 1603bad:
1b83bef2
SW		 1604 pr_err("corrupt osd_op_reply got %d %d\n",
1605 (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len));
9ec7cab1 1606 ceph_msg_dump(msg);
f24e9980
SW		 1607}
1608
6f6c7006 1609static void reset_changed_osds(struct ceph_osd_client *osdc)
f24e9980 1610{
f24e9980 1611 struct rb_node *p, *n;
f24e9980 1612
6f6c7006
SW		 1613 for (p = rb_first(&osdc->osds); p; p = n) {
1614 struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
f24e9980 1615
6f6c7006
SW		 1616 n = rb_next(p);
1617 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
1618 memcmp(&osd->o_con.peer_addr,
1619 ceph_osd_addr(osdc->osdmap,
1620 osd->o_osd),
1621 sizeof(struct ceph_entity_addr)) != 0)
1622 __reset_osd(osdc, osd);
f24e9980 1623 }
422d2cb8
YS		 1624}
1625
1626/*
6f6c7006
SW		 1627 * Requeue requests whose mapping to an OSD has changed. If requests map to
1628 * no osd, request a new map.
422d2cb8 1629 *
e6d50f67 1630 * Caller should hold map_sem for read.
422d2cb8 1631 */
27ff9f13
JD		 1632static void kick_requests(struct ceph_osd_client *osdc, bool force_resend,
1633 bool force_resend_writes)
422d2cb8 1634{
a40c4f10 1635 struct ceph_osd_request *req, *nreq;
6f6c7006
SW		 1636 struct rb_node *p;
1637 int needmap = 0;
1638 int err;
27ff9f13 1639 bool force_resend_req;
422d2cb8 1640
27ff9f13
JD		 1641 dout("kick_requests %s %s\n", force_resend ? " (force resend)" : "",
1642 force_resend_writes ? " (force resend writes)" : "");
422d2cb8 1643 mutex_lock(&osdc->request_mutex);
6194ea89 1644 for (p = rb_first(&osdc->requests); p; ) {
6f6c7006 1645 req = rb_entry(p, struct ceph_osd_request, r_node);
6194ea89 1646 p = rb_next(p);
ab60b16d
AE		 1647
1648 /*
1649 * For linger requests that have not yet been
1650 * registered, move them to the linger list; they'll
1651 * be sent to the osd in the loop below. Unregister
1652 * the request before re-registering it as a linger
1653 * request to ensure the __map_request() below
1654 * will decide it needs to be sent.
1655 */
1656 if (req->r_linger && list_empty(&req->r_linger_item)) {
1657 dout("%p tid %llu restart on osd%d\n",
1658 req, req->r_tid,
1659 req->r_osd ? req->r_osd->o_osd : -1);
1d02ec7f 1660 ceph_osdc_get_request(req);
ab60b16d
AE		 1661 __unregister_request(osdc, req);
1662 __register_linger_request(osdc, req);
1d02ec7f 1663 ceph_osdc_put_request(req);
ab60b16d
AE		 1664 continue;
1665 }
1666
27ff9f13
JD		 1667 force_resend_req = force_resend ||
1668 (force_resend_writes &&
1669 req->r_flags & CEPH_OSD_FLAG_WRITE);
1670 err = __map_request(osdc, req, force_resend_req);
6f6c7006
SW		 1671 if (err < 0)
1672 continue; /* error */
1673 if (req->r_osd == NULL) {
1674 dout("%p tid %llu maps to no osd\n", req, req->r_tid);
1675 needmap++; /* request a newer map */
1676 } else if (err > 0) {
6194ea89
SW		 1677 if (!req->r_linger) {
1678 dout("%p tid %llu requeued on osd%d\n", req,
1679 req->r_tid,
1680 req->r_osd ? req->r_osd->o_osd : -1);
a40c4f10 1681 req->r_flags |= CEPH_OSD_FLAG_RETRY;
6194ea89
SW		 1682 }
1683 }
a40c4f10
YS		 1684 }
1685
1686 list_for_each_entry_safe(req, nreq, &osdc->req_linger,
1687 r_linger_item) {
1688 dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
1689
27ff9f13
JD		 1690 err = __map_request(osdc, req,
1691 force_resend || force_resend_writes);
ab60b16d 1692 dout("__map_request returned %d\n", err);
a40c4f10
YS		 1693 if (err == 0)
1694 continue; /* no change and no osd was specified */
1695 if (err < 0)
1696 continue; /* hrm! */
1697 if (req->r_osd == NULL) {
1698 dout("tid %llu maps to no valid osd\n", req->r_tid);
1699 needmap++; /* request a newer map */
1700 continue;
6f6c7006 1701 }
a40c4f10
YS		 1702
1703 dout("kicking lingering %p tid %llu osd%d\n", req, req->r_tid,
1704 req->r_osd ? req->r_osd->o_osd : -1);
a40c4f10 1705 __register_request(osdc, req);
c89ce05e 1706 __unregister_linger_request(osdc, req);
6f6c7006 1707 }
14d2f38d 1708 reset_changed_osds(osdc);
f24e9980
SW		 1709 mutex_unlock(&osdc->request_mutex);
1710
1711 if (needmap) {
1712 dout("%d requests for down osds, need new map\n", needmap);
1713 ceph_monc_request_next_osdmap(&osdc->client->monc);
1714 }
422d2cb8 1715}
6f6c7006
SW		 1716
1717
f24e9980
SW		 1718/*
1719 * Process updated osd map.
1720 *
1721 * The message contains any number of incremental and full maps, normally
1722 * indicating some sort of topology change in the cluster. Kick requests
1723 * off to different OSDs as needed.
1724 */
1725void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1726{
1727 void *p, *end, *next;
1728 u32 nr_maps, maplen;
1729 u32 epoch;
1730 struct ceph_osdmap *newmap = NULL, *oldmap;
1731 int err;
1732 struct ceph_fsid fsid;
27ff9f13 1733 bool was_full;
f24e9980
SW		 1734
1735 dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
1736 p = msg->front.iov_base;
1737 end = p + msg->front.iov_len;
1738
1739 /* verify fsid */
1740 ceph_decode_need(&p, end, sizeof(fsid), bad);
1741 ceph_decode_copy(&p, &fsid, sizeof(fsid));
0743304d
SW		 1742 if (ceph_check_fsid(osdc->client, &fsid) < 0)
1743 return;
f24e9980
SW		 1744
1745 down_write(&osdc->map_sem);
1746
27ff9f13
JD		 1747 was_full = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
1748
f24e9980
SW		 1749 /* incremental maps */
1750 ceph_decode_32_safe(&p, end, nr_maps, bad);
1751 dout(" %d inc maps\n", nr_maps);
1752 while (nr_maps > 0) {
1753 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 1754 epoch = ceph_decode_32(&p);
1755 maplen = ceph_decode_32(&p);
f24e9980
SW		 1756 ceph_decode_need(&p, end, maplen, bad);
1757 next = p + maplen;
1758 if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
1759 dout("applying incremental map %u len %d\n",
1760 epoch, maplen);
1761 newmap = osdmap_apply_incremental(&p, next,
1762 osdc->osdmap,
15d9882c 1763 &osdc->client->msgr);
f24e9980
SW		 1764 if (IS_ERR(newmap)) {
1765 err = PTR_ERR(newmap);
1766 goto bad;
1767 }
30dc6381 1768 BUG_ON(!newmap);
f24e9980
SW		 1769 if (newmap != osdc->osdmap) {
1770 ceph_osdmap_destroy(osdc->osdmap);
1771 osdc->osdmap = newmap;
1772 }
27ff9f13
JD		 1773 was_full = was_full ||
1774 ceph_osdmap_flag(osdc->osdmap,
1775 CEPH_OSDMAP_FULL);
1776 kick_requests(osdc, 0, was_full);
f24e9980
SW		 1777 } else {
1778 dout("ignoring incremental map %u len %d\n",
1779 epoch, maplen);
1780 }
1781 p = next;
1782 nr_maps--;
1783 }
1784 if (newmap)
1785 goto done;
1786
1787 /* full maps */
1788 ceph_decode_32_safe(&p, end, nr_maps, bad);
1789 dout(" %d full maps\n", nr_maps);
1790 while (nr_maps) {
1791 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 1792 epoch = ceph_decode_32(&p);
1793 maplen = ceph_decode_32(&p);
f24e9980
SW		 1794 ceph_decode_need(&p, end, maplen, bad);
1795 if (nr_maps > 1) {
1796 dout("skipping non-latest full map %u len %d\n",
1797 epoch, maplen);
1798 } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
1799 dout("skipping full map %u len %d, "
1800 "older than our %u\n", epoch, maplen,
1801 osdc->osdmap->epoch);
1802 } else {
38d6453c
SW		 1803 int skipped_map = 0;
1804
f24e9980
SW		 1805 dout("taking full map %u len %d\n", epoch, maplen);
1806 newmap = osdmap_decode(&p, p+maplen);
1807 if (IS_ERR(newmap)) {
1808 err = PTR_ERR(newmap);
1809 goto bad;
1810 }
30dc6381 1811 BUG_ON(!newmap);
f24e9980
SW		 1812 oldmap = osdc->osdmap;
1813 osdc->osdmap = newmap;
38d6453c
SW		 1814 if (oldmap) {
1815 if (oldmap->epoch + 1 < newmap->epoch)
1816 skipped_map = 1;
f24e9980 1817 ceph_osdmap_destroy(oldmap);
38d6453c 1818 }
27ff9f13
JD		 1819 was_full = was_full ||
1820 ceph_osdmap_flag(osdc->osdmap,
1821 CEPH_OSDMAP_FULL);
1822 kick_requests(osdc, skipped_map, was_full);
f24e9980
SW		 1823 }
1824 p += maplen;
1825 nr_maps--;
1826 }
1827
53341d7d
DC		 1828 if (!osdc->osdmap)
1829 goto bad;
f24e9980
SW		 1830done:
1831 downgrade_write(&osdc->map_sem);
1832 ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch);
cd634fb6
SW		 1833
1834 /*
1835 * subscribe to subsequent osdmap updates if full to ensure
1836 * we find out when we are no longer full and stop returning
1837 * ENOSPC.
1838 */
4892ed8d
JD		 1839 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) ||
1840 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD) ||
1841 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR))
cd634fb6
SW		 1842 ceph_monc_request_next_osdmap(&osdc->client->monc);
1843
f9d25199
AE		 1844 mutex_lock(&osdc->request_mutex);
1845 __send_queued(osdc);
1846 mutex_unlock(&osdc->request_mutex);
f24e9980 1847 up_read(&osdc->map_sem);
03066f23 1848 wake_up_all(&osdc->client->auth_wq);
f24e9980
SW		 1849 return;
1850
1851bad:
1852 pr_err("osdc handle_map corrupt msg\n");
9ec7cab1 1853 ceph_msg_dump(msg);
f24e9980
SW		 1854 up_write(&osdc->map_sem);
1855 return;
1856}
1857
a40c4f10
YS		 1858/*
1859 * watch/notify callback event infrastructure
1860 *
1861 * These callbacks are used both for watch and notify operations.
1862 */
1863static void __release_event(struct kref *kref)
1864{
1865 struct ceph_osd_event *event =
1866 container_of(kref, struct ceph_osd_event, kref);
1867
1868 dout("__release_event %p\n", event);
1869 kfree(event);
1870}
1871
1872static void get_event(struct ceph_osd_event *event)
1873{
1874 kref_get(&event->kref);
1875}
1876
1877void ceph_osdc_put_event(struct ceph_osd_event *event)
1878{
1879 kref_put(&event->kref, __release_event);
1880}
1881EXPORT_SYMBOL(ceph_osdc_put_event);
1882
1883static void __insert_event(struct ceph_osd_client *osdc,
1884 struct ceph_osd_event *new)
1885{
1886 struct rb_node **p = &osdc->event_tree.rb_node;
1887 struct rb_node *parent = NULL;
1888 struct ceph_osd_event *event = NULL;
1889
1890 while (*p) {
1891 parent = *p;
1892 event = rb_entry(parent, struct ceph_osd_event, node);
1893 if (new->cookie < event->cookie)
1894 p = &(*p)->rb_left;
1895 else if (new->cookie > event->cookie)
1896 p = &(*p)->rb_right;
1897 else
1898 BUG();
1899 }
1900
1901 rb_link_node(&new->node, parent, p);
1902 rb_insert_color(&new->node, &osdc->event_tree);
1903}
1904
1905static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
1906 u64 cookie)
1907{
1908 struct rb_node **p = &osdc->event_tree.rb_node;
1909 struct rb_node *parent = NULL;
1910 struct ceph_osd_event *event = NULL;
1911
1912 while (*p) {
1913 parent = *p;
1914 event = rb_entry(parent, struct ceph_osd_event, node);
1915 if (cookie < event->cookie)
1916 p = &(*p)->rb_left;
1917 else if (cookie > event->cookie)
1918 p = &(*p)->rb_right;
1919 else
1920 return event;
1921 }
1922 return NULL;
1923}
1924
1925static void __remove_event(struct ceph_osd_event *event)
1926{
1927 struct ceph_osd_client *osdc = event->osdc;
1928
1929 if (!RB_EMPTY_NODE(&event->node)) {
1930 dout("__remove_event removed %p\n", event);
1931 rb_erase(&event->node, &osdc->event_tree);
1932 ceph_osdc_put_event(event);
1933 } else {
1934 dout("__remove_event didn't remove %p\n", event);
1935 }
1936}
1937
1938int ceph_osdc_create_event(struct ceph_osd_client *osdc,
1939 void (*event_cb)(u64, u64, u8, void *),
3c663bbd 1940 void *data, struct ceph_osd_event **pevent)
a40c4f10
YS		 1941{
1942 struct ceph_osd_event *event;
1943
1944 event = kmalloc(sizeof(*event), GFP_NOIO);
1945 if (!event)
1946 return -ENOMEM;
1947
1948 dout("create_event %p\n", event);
1949 event->cb = event_cb;
3c663bbd 1950 event->one_shot = 0;
a40c4f10
YS		 1951 event->data = data;
1952 event->osdc = osdc;
1953 INIT_LIST_HEAD(&event->osd_node);
3ee5234d 1954 RB_CLEAR_NODE(&event->node);
a40c4f10
YS		 1955 kref_init(&event->kref); /* one ref for us */
1956 kref_get(&event->kref); /* one ref for the caller */
a40c4f10
YS		 1957
1958 spin_lock(&osdc->event_lock);
1959 event->cookie = ++osdc->event_count;
1960 __insert_event(osdc, event);
1961 spin_unlock(&osdc->event_lock);
1962
1963 *pevent = event;
1964 return 0;
1965}
1966EXPORT_SYMBOL(ceph_osdc_create_event);
1967
1968void ceph_osdc_cancel_event(struct ceph_osd_event *event)
1969{
1970 struct ceph_osd_client *osdc = event->osdc;
1971
1972 dout("cancel_event %p\n", event);
1973 spin_lock(&osdc->event_lock);
1974 __remove_event(event);
1975 spin_unlock(&osdc->event_lock);
1976 ceph_osdc_put_event(event); /* caller's */
1977}
1978EXPORT_SYMBOL(ceph_osdc_cancel_event);
1979
1980
1981static void do_event_work(struct work_struct *work)
1982{
1983 struct ceph_osd_event_work *event_work =
1984 container_of(work, struct ceph_osd_event_work, work);
1985 struct ceph_osd_event *event = event_work->event;
1986 u64 ver = event_work->ver;
1987 u64 notify_id = event_work->notify_id;
1988 u8 opcode = event_work->opcode;
1989
1990 dout("do_event_work completing %p\n", event);
1991 event->cb(ver, notify_id, opcode, event->data);
a40c4f10
YS		 1992 dout("do_event_work completed %p\n", event);
1993 ceph_osdc_put_event(event);
1994 kfree(event_work);
1995}
1996
1997
1998/*
1999 * Process osd watch notifications
2000 */
3c663bbd
AE		 2001static void handle_watch_notify(struct ceph_osd_client *osdc,
2002 struct ceph_msg *msg)
a40c4f10
YS		 2003{
2004 void *p, *end;
2005 u8 proto_ver;
2006 u64 cookie, ver, notify_id;
2007 u8 opcode;
2008 struct ceph_osd_event *event;
2009 struct ceph_osd_event_work *event_work;
2010
2011 p = msg->front.iov_base;
2012 end = p + msg->front.iov_len;
2013
2014 ceph_decode_8_safe(&p, end, proto_ver, bad);
2015 ceph_decode_8_safe(&p, end, opcode, bad);
2016 ceph_decode_64_safe(&p, end, cookie, bad);
2017 ceph_decode_64_safe(&p, end, ver, bad);
2018 ceph_decode_64_safe(&p, end, notify_id, bad);
2019
2020 spin_lock(&osdc->event_lock);
2021 event = __find_event(osdc, cookie);
2022 if (event) {
3c663bbd 2023 BUG_ON(event->one_shot);
a40c4f10 2024 get_event(event);
a40c4f10
YS		 2025 }
2026 spin_unlock(&osdc->event_lock);
2027 dout("handle_watch_notify cookie %lld ver %lld event %p\n",
2028 cookie, ver, event);
2029 if (event) {
2030 event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
a40c4f10
YS		 2031 if (!event_work) {
2032 dout("ERROR: could not allocate event_work\n");
2033 goto done_err;
2034 }
6b0ae409 2035 INIT_WORK(&event_work->work, do_event_work);
a40c4f10
YS		 2036 event_work->event = event;
2037 event_work->ver = ver;
2038 event_work->notify_id = notify_id;
2039 event_work->opcode = opcode;
2040 if (!queue_work(osdc->notify_wq, &event_work->work)) {
2041 dout("WARNING: failed to queue notify event work\n");
2042 goto done_err;
2043 }
2044 }
2045
2046 return;
2047
2048done_err:
a40c4f10
YS		 2049 ceph_osdc_put_event(event);
2050 return;
2051
2052bad:
2053 pr_err("osdc handle_watch_notify corrupt msg\n");
2054 return;
2055}
2056
e65550fd
AE		 2057/*
2058 * build new request AND message
2059 *
2060 */
2061void ceph_osdc_build_request(struct ceph_osd_request *req, u64 off,
2062 struct ceph_snap_context *snapc, u64 snap_id,
2063 struct timespec *mtime)
2064{
2065 struct ceph_msg *msg = req->r_request;
2066 void *p;
2067 size_t msg_size;
2068 int flags = req->r_flags;
2069 u64 data_len;
2070 unsigned int i;
2071
2072 req->r_snapid = snap_id;
2073 req->r_snapc = ceph_get_snap_context(snapc);
2074
2075 /* encode request */
2076 msg->hdr.version = cpu_to_le16(4);
2077
2078 p = msg->front.iov_base;
2079 ceph_encode_32(&p, 1); /* client_inc is always 1 */
2080 req->r_request_osdmap_epoch = p;
2081 p += 4;
2082 req->r_request_flags = p;
2083 p += 4;
2084 if (req->r_flags & CEPH_OSD_FLAG_WRITE)
2085 ceph_encode_timespec(p, mtime);
2086 p += sizeof(struct ceph_timespec);
2087 req->r_request_reassert_version = p;
2088 p += sizeof(struct ceph_eversion); /* will get filled in */
2089
2090 /* oloc */
2091 ceph_encode_8(&p, 4);
2092 ceph_encode_8(&p, 4);
2093 ceph_encode_32(&p, 8 + 4 + 4);
2094 req->r_request_pool = p;
2095 p += 8;
2096 ceph_encode_32(&p, -1); /* preferred */
2097 ceph_encode_32(&p, 0); /* key len */
2098
2099 ceph_encode_8(&p, 1);
2100 req->r_request_pgid = p;
2101 p += 8 + 4;
2102 ceph_encode_32(&p, -1); /* preferred */
2103
2104 /* oid */
2105 ceph_encode_32(&p, req->r_oid_len);
2106 memcpy(p, req->r_oid, req->r_oid_len);
2107 dout("oid '%.*s' len %d\n", req->r_oid_len, req->r_oid, req->r_oid_len);
2108 p += req->r_oid_len;
2109
2110 /* ops--can imply data */
2111 ceph_encode_16(&p, (u16)req->r_num_ops);
2112 data_len = 0;
2113 for (i = 0; i < req->r_num_ops; i++) {
2114 data_len += osd_req_encode_op(req, p, i);
2115 p += sizeof(struct ceph_osd_op);
2116 }
2117
2118 /* snaps */
2119 ceph_encode_64(&p, req->r_snapid);
2120 ceph_encode_64(&p, req->r_snapc ? req->r_snapc->seq : 0);
2121 ceph_encode_32(&p, req->r_snapc ? req->r_snapc->num_snaps : 0);
2122 if (req->r_snapc) {
2123 for (i = 0; i < snapc->num_snaps; i++) {
2124 ceph_encode_64(&p, req->r_snapc->snaps[i]);
2125 }
2126 }
2127
2128 req->r_request_attempts = p;
2129 p += 4;
2130
2131 /* data */
2132 if (flags & CEPH_OSD_FLAG_WRITE) {
2133 u16 data_off;
2134
2135 /*
2136 * The header "data_off" is a hint to the receiver
2137 * allowing it to align received data into its
2138 * buffers such that there's no need to re-copy
2139 * it before writing it to disk (direct I/O).
2140 */
2141 data_off = (u16) (off & 0xffff);
2142 req->r_request->hdr.data_off = cpu_to_le16(data_off);
2143 }
2144 req->r_request->hdr.data_len = cpu_to_le32(data_len);
2145
2146 BUG_ON(p > msg->front.iov_base + msg->front.iov_len);
2147 msg_size = p - msg->front.iov_base;
2148 msg->front.iov_len = msg_size;
2149 msg->hdr.front_len = cpu_to_le32(msg_size);
2150
2151 dout("build_request msg_size was %d\n", (int)msg_size);
2152}
2153EXPORT_SYMBOL(ceph_osdc_build_request);
2154
70636773
AE		 2155/*
2156 * Register request, send initial attempt.
2157 */
2158int ceph_osdc_start_request(struct ceph_osd_client *osdc,
2159 struct ceph_osd_request *req,
2160 bool nofail)
2161{
2162 int rc = 0;
2163
f24e9980
SW		 2164 down_read(&osdc->map_sem);
2165 mutex_lock(&osdc->request_mutex);
dc4b870c 2166 __register_request(osdc, req);
c10ebbf5
AE		 2167 req->r_sent = 0;
2168 req->r_got_reply = 0;
92451b49
AE		 2169 rc = __map_request(osdc, req, 0);
2170 if (rc < 0) {
2171 if (nofail) {
2172 dout("osdc_start_request failed map, "
2173 " will retry %lld\n", req->r_tid);
2174 rc = 0;
2ab0ad6a 2175 } else {
2176 __unregister_request(osdc, req);
f24e9980 2177 }
92451b49 2178 goto out_unlock;
f24e9980 2179 }
92451b49
AE		 2180 if (req->r_osd == NULL) {
2181 dout("send_request %p no up osds in pg\n", req);
2182 ceph_monc_request_next_osdmap(&osdc->client->monc);
2183 } else {
7e2766a1 2184 __send_queued(osdc);
92451b49
AE		 2185 }
2186 rc = 0;
234af26f 2187out_unlock:
f24e9980
SW		 2188 mutex_unlock(&osdc->request_mutex);
2189 up_read(&osdc->map_sem);
2190 return rc;
2191}
3d14c5d2 2192EXPORT_SYMBOL(ceph_osdc_start_request);
f24e9980
SW		 2193
2194/*
2195 * wait for a request to complete
2196 */
2197int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
2198 struct ceph_osd_request *req)
2199{
2200 int rc;
2201
2202 rc = wait_for_completion_interruptible(&req->r_completion);
2203 if (rc < 0) {
2204 mutex_lock(&osdc->request_mutex);
2205 __cancel_request(req);
529cfcc4 2206 __unregister_request(osdc, req);
f24e9980 2207 mutex_unlock(&osdc->request_mutex);
25845472 2208 complete_request(req);
529cfcc4 2209 dout("wait_request tid %llu canceled/timed out\n", req->r_tid);
f24e9980
SW		 2210 return rc;
2211 }
2212
2213 dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result);
2214 return req->r_result;
2215}
3d14c5d2 2216EXPORT_SYMBOL(ceph_osdc_wait_request);
f24e9980
SW		 2217
2218/*
2219 * sync - wait for all in-flight requests to flush. avoid starvation.
2220 */
2221void ceph_osdc_sync(struct ceph_osd_client *osdc)
2222{
2223 struct ceph_osd_request *req;
2224 u64 last_tid, next_tid = 0;
2225
2226 mutex_lock(&osdc->request_mutex);
2227 last_tid = osdc->last_tid;
2228 while (1) {
2229 req = __lookup_request_ge(osdc, next_tid);
2230 if (!req)
2231 break;
2232 if (req->r_tid > last_tid)
2233 break;
2234
2235 next_tid = req->r_tid + 1;
2236 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
2237 continue;
2238
2239 ceph_osdc_get_request(req);
2240 mutex_unlock(&osdc->request_mutex);
2241 dout("sync waiting on tid %llu (last is %llu)\n",
2242 req->r_tid, last_tid);
2243 wait_for_completion(&req->r_safe_completion);
2244 mutex_lock(&osdc->request_mutex);
2245 ceph_osdc_put_request(req);
2246 }
2247 mutex_unlock(&osdc->request_mutex);
2248 dout("sync done (thru tid %llu)\n", last_tid);
2249}
3d14c5d2 2250EXPORT_SYMBOL(ceph_osdc_sync);
f24e9980 2251
a2e5951b
JD		 2252/*
2253 * Call all pending notify callbacks - for use after a watch is
2254 * unregistered, to make sure no more callbacks for it will be invoked
2255 */
2256extern void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc)
2257{
2258 flush_workqueue(osdc->notify_wq);
2259}
2260EXPORT_SYMBOL(ceph_osdc_flush_notifies);
2261
2262
f24e9980
SW		 2263/*
2264 * init, shutdown
2265 */
2266int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
2267{
2268 int err;
2269
2270 dout("init\n");
2271 osdc->client = client;
2272 osdc->osdmap = NULL;
2273 init_rwsem(&osdc->map_sem);
2274 init_completion(&osdc->map_waiters);
2275 osdc->last_requested_map = 0;
2276 mutex_init(&osdc->request_mutex);
f24e9980
SW		 2277 osdc->last_tid = 0;
2278 osdc->osds = RB_ROOT;
f5a2041b 2279 INIT_LIST_HEAD(&osdc->osd_lru);
f24e9980 2280 osdc->requests = RB_ROOT;
422d2cb8 2281 INIT_LIST_HEAD(&osdc->req_lru);
6f6c7006
SW		 2282 INIT_LIST_HEAD(&osdc->req_unsent);
2283 INIT_LIST_HEAD(&osdc->req_notarget);
a40c4f10 2284 INIT_LIST_HEAD(&osdc->req_linger);
f24e9980
SW		 2285 osdc->num_requests = 0;
2286 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
f5a2041b 2287 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
a40c4f10
YS		 2288 spin_lock_init(&osdc->event_lock);
2289 osdc->event_tree = RB_ROOT;
2290 osdc->event_count = 0;
f5a2041b
YS		 2291
2292 schedule_delayed_work(&osdc->osds_timeout_work,
3d14c5d2 2293 round_jiffies_relative(osdc->client->options->osd_idle_ttl * HZ));
f24e9980 2294
5f44f142 2295 err = -ENOMEM;
f24e9980
SW		 2296 osdc->req_mempool = mempool_create_kmalloc_pool(10,
2297 sizeof(struct ceph_osd_request));
2298 if (!osdc->req_mempool)
5f44f142 2299 goto out;
f24e9980 2300
d50b409f
SW		 2301 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
2302 OSD_OP_FRONT_LEN, 10, true,
4f48280e 2303 "osd_op");
f24e9980 2304 if (err < 0)
5f44f142 2305 goto out_mempool;
d50b409f 2306 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
4f48280e
SW		 2307 OSD_OPREPLY_FRONT_LEN, 10, true,
2308 "osd_op_reply");
c16e7869
SW		 2309 if (err < 0)
2310 goto out_msgpool;
a40c4f10 2311
3834fb30 2312 err = -ENOMEM;
a40c4f10 2313 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
3834fb30 2314 if (!osdc->notify_wq)
a40c4f10 2315 goto out_msgpool;
f24e9980 2316 return 0;
5f44f142 2317
c16e7869
SW		 2318out_msgpool:
2319 ceph_msgpool_destroy(&osdc->msgpool_op);
5f44f142
SW		 2320out_mempool:
2321 mempool_destroy(osdc->req_mempool);
2322out:
2323 return err;
f24e9980
SW		 2324}
2325
2326void ceph_osdc_stop(struct ceph_osd_client *osdc)
2327{
a40c4f10
YS		 2328 flush_workqueue(osdc->notify_wq);
2329 destroy_workqueue(osdc->notify_wq);
f24e9980 2330 cancel_delayed_work_sync(&osdc->timeout_work);
f5a2041b 2331 cancel_delayed_work_sync(&osdc->osds_timeout_work);
f24e9980
SW		 2332 if (osdc->osdmap) {
2333 ceph_osdmap_destroy(osdc->osdmap);
2334 osdc->osdmap = NULL;
2335 }
aca420bc 2336 remove_all_osds(osdc);
f24e9980
SW		 2337 mempool_destroy(osdc->req_mempool);
2338 ceph_msgpool_destroy(&osdc->msgpool_op);
c16e7869 2339 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
f24e9980
SW		 2340}
2341
2342/*
2343 * Read some contiguous pages. If we cross a stripe boundary, shorten
2344 * *plen. Return number of bytes read, or error.
2345 */
2346int ceph_osdc_readpages(struct ceph_osd_client *osdc,
2347 struct ceph_vino vino, struct ceph_file_layout *layout,
2348 u64 off, u64 *plen,
2349 u32 truncate_seq, u64 truncate_size,
b7495fc2 2350 struct page **pages, int num_pages, int page_align)
f24e9980
SW		 2351{
2352 struct ceph_osd_request *req;
2353 int rc = 0;
2354
2355 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
2356 vino.snap, off, *plen);
79528734 2357 req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 1,
f24e9980 2358 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
acead002 2359 NULL, truncate_seq, truncate_size,
153e5167 2360 false);
6816282d
SW		 2361 if (IS_ERR(req))
2362 return PTR_ERR(req);
f24e9980
SW		 2363
2364 /* it may be a short read due to an object boundary */
0fff87ec 2365
406e2c9f 2366 osd_req_op_extent_osd_data_pages(req, 0,
a4ce40a9 2367 pages, *plen, page_align, false, false);
f24e9980 2368
e0c59487 2369 dout("readpages final extent is %llu~%llu (%llu bytes align %d)\n",
43bfe5de 2370 off, *plen, *plen, page_align);
f24e9980 2371
79528734 2372 ceph_osdc_build_request(req, off, NULL, vino.snap, NULL);
02ee07d3 2373
f24e9980
SW		 2374 rc = ceph_osdc_start_request(osdc, req, false);
2375 if (!rc)
2376 rc = ceph_osdc_wait_request(osdc, req);
2377
2378 ceph_osdc_put_request(req);
2379 dout("readpages result %d\n", rc);
2380 return rc;
2381}
3d14c5d2 2382EXPORT_SYMBOL(ceph_osdc_readpages);
f24e9980
SW		 2383
2384/*
2385 * do a synchronous write on N pages
2386 */
2387int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
2388 struct ceph_file_layout *layout,
2389 struct ceph_snap_context *snapc,
2390 u64 off, u64 len,
2391 u32 truncate_seq, u64 truncate_size,
2392 struct timespec *mtime,
24808826 2393 struct page **pages, int num_pages)
f24e9980
SW		 2394{
2395 struct ceph_osd_request *req;
2396 int rc = 0;
b7495fc2 2397 int page_align = off & ~PAGE_MASK;
f24e9980 2398
acead002 2399 BUG_ON(vino.snap != CEPH_NOSNAP); /* snapshots aren't writeable */
79528734 2400 req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 1,
f24e9980 2401 CEPH_OSD_OP_WRITE,
24808826 2402 CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
acead002 2403 snapc, truncate_seq, truncate_size,
153e5167 2404 true);
6816282d
SW		 2405 if (IS_ERR(req))
2406 return PTR_ERR(req);
f24e9980
SW		 2407
2408 /* it may be a short write due to an object boundary */
406e2c9f 2409 osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align,
43bfe5de
AE		 2410 false, false);
2411 dout("writepages %llu~%llu (%llu bytes)\n", off, len, len);
f24e9980 2412
79528734 2413 ceph_osdc_build_request(req, off, snapc, CEPH_NOSNAP, mtime);
02ee07d3 2414
87f979d3 2415 rc = ceph_osdc_start_request(osdc, req, true);
f24e9980
SW		 2416 if (!rc)
2417 rc = ceph_osdc_wait_request(osdc, req);
2418
2419 ceph_osdc_put_request(req);
2420 if (rc == 0)
2421 rc = len;
2422 dout("writepages result %d\n", rc);
2423 return rc;
2424}
3d14c5d2 2425EXPORT_SYMBOL(ceph_osdc_writepages);
f24e9980 2426
5522ae0b
AE		 2427int ceph_osdc_setup(void)
2428{
2429 BUG_ON(ceph_osd_request_cache);
2430 ceph_osd_request_cache = kmem_cache_create("ceph_osd_request",
2431 sizeof (struct ceph_osd_request),
2432 __alignof__(struct ceph_osd_request),
2433 0, NULL);
2434
2435 return ceph_osd_request_cache ? 0 : -ENOMEM;
2436}
2437EXPORT_SYMBOL(ceph_osdc_setup);
2438
2439void ceph_osdc_cleanup(void)
2440{
2441 BUG_ON(!ceph_osd_request_cache);
2442 kmem_cache_destroy(ceph_osd_request_cache);
2443 ceph_osd_request_cache = NULL;
2444}
2445EXPORT_SYMBOL(ceph_osdc_cleanup);
2446
f24e9980
SW		 2447/*
2448 * handle incoming message
2449 */
2450static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
2451{
2452 struct ceph_osd *osd = con->private;
32c895e7 2453 struct ceph_osd_client *osdc;
f24e9980
SW		 2454 int type = le16_to_cpu(msg->hdr.type);
2455
2456 if (!osd)
4a32f93d 2457 goto out;
32c895e7 2458 osdc = osd->o_osdc;
f24e9980
SW		 2459
2460 switch (type) {
2461 case CEPH_MSG_OSD_MAP:
2462 ceph_osdc_handle_map(osdc, msg);
2463 break;
2464 case CEPH_MSG_OSD_OPREPLY:
350b1c32 2465 handle_reply(osdc, msg, con);
f24e9980 2466 break;
a40c4f10
YS		 2467 case CEPH_MSG_WATCH_NOTIFY:
2468 handle_watch_notify(osdc, msg);
2469 break;
f24e9980
SW		 2470
2471 default:
2472 pr_err("received unknown message type %d %s\n", type,
2473 ceph_msg_type_name(type));
2474 }
4a32f93d 2475out:
f24e9980
SW		 2476 ceph_msg_put(msg);
2477}
2478
5b3a4db3 2479/*
21b667f6
SW		 2480 * lookup and return message for incoming reply. set up reply message
2481 * pages.
5b3a4db3
SW		 2482 */
2483static struct ceph_msg *get_reply(struct ceph_connection *con,
2450418c
YS		 2484 struct ceph_msg_header *hdr,
2485 int *skip)
f24e9980
SW		 2486{
2487 struct ceph_osd *osd = con->private;
2488 struct ceph_osd_client *osdc = osd->o_osdc;
2450418c 2489 struct ceph_msg *m;
0547a9b3 2490 struct ceph_osd_request *req;
5b3a4db3
SW		 2491 int front = le32_to_cpu(hdr->front_len);
2492 int data_len = le32_to_cpu(hdr->data_len);
0547a9b3 2493 u64 tid;
f24e9980 2494
0547a9b3
YS		 2495 tid = le64_to_cpu(hdr->tid);
2496 mutex_lock(&osdc->request_mutex);
2497 req = __lookup_request(osdc, tid);
2498 if (!req) {
2499 *skip = 1;
2500 m = NULL;
756a16a5
SW		 2501 dout("get_reply unknown tid %llu from osd%d\n", tid,
2502 osd->o_osd);
0547a9b3
YS		 2503 goto out;
2504 }
c16e7869 2505
ace6d3a9 2506 if (req->r_reply->con)
8921d114 2507 dout("%s revoking msg %p from old con %p\n", __func__,
ace6d3a9
AE		 2508 req->r_reply, req->r_reply->con);
2509 ceph_msg_revoke_incoming(req->r_reply);
0547a9b3 2510
c16e7869
SW		 2511 if (front > req->r_reply->front.iov_len) {
2512 pr_warning("get_reply front %d > preallocated %d\n",
2513 front, (int)req->r_reply->front.iov_len);
b61c2763 2514 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front, GFP_NOFS, false);
a79832f2 2515 if (!m)
c16e7869
SW		 2516 goto out;
2517 ceph_msg_put(req->r_reply);
2518 req->r_reply = m;
2519 }
2520 m = ceph_msg_get(req->r_reply);
2521
0547a9b3 2522 if (data_len > 0) {
a4ce40a9 2523 struct ceph_osd_data *osd_data;
0fff87ec 2524
a4ce40a9
AE		 2525 /*
2526 * XXX This is assuming there is only one op containing
2527 * XXX page data. Probably OK for reads, but this
2528 * XXX ought to be done more generally.
2529 */
406e2c9f 2530 osd_data = osd_req_op_extent_osd_data(req, 0);
0fff87ec 2531 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
0fff87ec 2532 if (osd_data->pages &&
e0c59487 2533 unlikely(osd_data->length < data_len)) {
2ac2b7a6 2534
e0c59487
AE		 2535 pr_warning("tid %lld reply has %d bytes "
2536 "we had only %llu bytes ready\n",
2537 tid, data_len, osd_data->length);
2ac2b7a6
AE		 2538 *skip = 1;
2539 ceph_msg_put(m);
2540 m = NULL;
2541 goto out;
2542 }
2ac2b7a6 2543 }
0547a9b3 2544 }
5b3a4db3 2545 *skip = 0;
c16e7869 2546 dout("get_reply tid %lld %p\n", tid, m);
0547a9b3
YS		 2547
2548out:
2549 mutex_unlock(&osdc->request_mutex);
2450418c 2550 return m;
5b3a4db3
SW		 2551
2552}
2553
2554static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2555 struct ceph_msg_header *hdr,
2556 int *skip)
2557{
2558 struct ceph_osd *osd = con->private;
2559 int type = le16_to_cpu(hdr->type);
2560 int front = le32_to_cpu(hdr->front_len);
2561
1c20f2d2 2562 *skip = 0;
5b3a4db3
SW		 2563 switch (type) {
2564 case CEPH_MSG_OSD_MAP:
a40c4f10 2565 case CEPH_MSG_WATCH_NOTIFY:
b61c2763 2566 return ceph_msg_new(type, front, GFP_NOFS, false);
5b3a4db3
SW		 2567 case CEPH_MSG_OSD_OPREPLY:
2568 return get_reply(con, hdr, skip);
2569 default:
2570 pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
2571 osd->o_osd);
2572 *skip = 1;
2573 return NULL;
2574 }
f24e9980
SW		 2575}
2576
2577/*
2578 * Wrappers to refcount containing ceph_osd struct
2579 */
2580static struct ceph_connection *get_osd_con(struct ceph_connection *con)
2581{
2582 struct ceph_osd *osd = con->private;
2583 if (get_osd(osd))
2584 return con;
2585 return NULL;
2586}
2587
2588static void put_osd_con(struct ceph_connection *con)
2589{
2590 struct ceph_osd *osd = con->private;
2591 put_osd(osd);
2592}
2593
4e7a5dcd
SW		 2594/*
2595 * authentication
2596 */
a3530df3
AE		 2597/*
2598 * Note: returned pointer is the address of a structure that's
2599 * managed separately. Caller must *not* attempt to free it.
2600 */
2601static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
8f43fb53 2602 int *proto, int force_new)
4e7a5dcd
SW		 2603{
2604 struct ceph_osd *o = con->private;
2605 struct ceph_osd_client *osdc = o->o_osdc;
2606 struct ceph_auth_client *ac = osdc->client->monc.auth;
74f1869f 2607 struct ceph_auth_handshake *auth = &o->o_auth;
4e7a5dcd 2608
74f1869f 2609 if (force_new && auth->authorizer) {
27859f97 2610 ceph_auth_destroy_authorizer(ac, auth->authorizer);
74f1869f
AE		 2611 auth->authorizer = NULL;
2612 }
27859f97
SW		 2613 if (!auth->authorizer) {
2614 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2615 auth);
4e7a5dcd 2616 if (ret)
a3530df3 2617 return ERR_PTR(ret);
27859f97
SW		 2618 } else {
2619 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
0bed9b5c
SW		 2620 auth);
2621 if (ret)
2622 return ERR_PTR(ret);
4e7a5dcd 2623 }
4e7a5dcd 2624 *proto = ac->protocol;
74f1869f 2625
a3530df3 2626 return auth;
4e7a5dcd
SW		 2627}
2628
2629
2630static int verify_authorizer_reply(struct ceph_connection *con, int len)
2631{
2632 struct ceph_osd *o = con->private;
2633 struct ceph_osd_client *osdc = o->o_osdc;
2634 struct ceph_auth_client *ac = osdc->client->monc.auth;
2635
27859f97 2636 return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer, len);
4e7a5dcd
SW		 2637}
2638
9bd2e6f8
SW		 2639static int invalidate_authorizer(struct ceph_connection *con)
2640{
2641 struct ceph_osd *o = con->private;
2642 struct ceph_osd_client *osdc = o->o_osdc;
2643 struct ceph_auth_client *ac = osdc->client->monc.auth;
2644
27859f97 2645 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
9bd2e6f8
SW		 2646 return ceph_monc_validate_auth(&osdc->client->monc);
2647}
4e7a5dcd 2648
9e32789f 2649static const struct ceph_connection_operations osd_con_ops = {
f24e9980
SW		 2650 .get = get_osd_con,
2651 .put = put_osd_con,
2652 .dispatch = dispatch,
4e7a5dcd
SW		 2653 .get_authorizer = get_authorizer,
2654 .verify_authorizer_reply = verify_authorizer_reply,
9bd2e6f8 2655 .invalidate_authorizer = invalidate_authorizer,
f24e9980 2656 .alloc_msg = alloc_msg,
81b024e7 2657 .fault = osd_reset,
f24e9980 2658};
kernel source for telekom puls (alcatel/mediatek)