projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
drivers: power: report battery voltage in AOSP compatible format
[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{
6af167fb 970 dout("%s %p osd%d\n", __func__, osd, osd->o_osd);
54ff4c89
ID		 971 WARN_ON(!list_empty(&osd->o_requests));
972 WARN_ON(!list_empty(&osd->o_linger_requests));
5d3c6d27 973
f5a2041b 974 list_del_init(&osd->o_osd_lru);
6af167fb
ID		 975 rb_erase(&osd->o_node, &osdc->osds);
976 RB_CLEAR_NODE(&osd->o_node);
977}
978
979static void remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
980{
981 dout("%s %p osd%d\n", __func__, osd, osd->o_osd);
982
983 if (!RB_EMPTY_NODE(&osd->o_node)) {
984 ceph_con_close(&osd->o_con);
985 __remove_osd(osdc, osd);
986 put_osd(osd);
987 }
f24e9980
SW		 988}
989
aca420bc
SW		 990static void remove_all_osds(struct ceph_osd_client *osdc)
991{
048a9d2d 992 dout("%s %p\n", __func__, osdc);
aca420bc
SW		 993 mutex_lock(&osdc->request_mutex);
994 while (!RB_EMPTY_ROOT(&osdc->osds)) {
995 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
996 struct ceph_osd, o_node);
6af167fb 997 remove_osd(osdc, osd);
aca420bc
SW		 998 }
999 mutex_unlock(&osdc->request_mutex);
1000}
1001
f5a2041b
YS		 1002static void __move_osd_to_lru(struct ceph_osd_client *osdc,
1003 struct ceph_osd *osd)
1004{
1005 dout("__move_osd_to_lru %p\n", osd);
1006 BUG_ON(!list_empty(&osd->o_osd_lru));
1007 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
3d14c5d2 1008 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl * HZ;
f5a2041b
YS		 1009}
1010
1011static void __remove_osd_from_lru(struct ceph_osd *osd)
1012{
1013 dout("__remove_osd_from_lru %p\n", osd);
1014 if (!list_empty(&osd->o_osd_lru))
1015 list_del_init(&osd->o_osd_lru);
1016}
1017
aca420bc 1018static void remove_old_osds(struct ceph_osd_client *osdc)
f5a2041b
YS		 1019{
1020 struct ceph_osd *osd, *nosd;
1021
1022 dout("__remove_old_osds %p\n", osdc);
1023 mutex_lock(&osdc->request_mutex);
1024 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
aca420bc 1025 if (time_before(jiffies, osd->lru_ttl))
f5a2041b 1026 break;
6af167fb 1027 remove_osd(osdc, osd);
f5a2041b
YS		 1028 }
1029 mutex_unlock(&osdc->request_mutex);
1030}
1031
f24e9980
SW		 1032/*
1033 * reset osd connect
1034 */
f5a2041b 1035static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
f24e9980 1036{
c3acb181 1037 struct ceph_entity_addr *peer_addr;
f24e9980 1038
f5a2041b 1039 dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
a40c4f10
YS		 1040 if (list_empty(&osd->o_requests) &&
1041 list_empty(&osd->o_linger_requests)) {
6af167fb 1042 remove_osd(osdc, osd);
c3acb181
AE		 1043 return -ENODEV;
1044 }
1045
1046 peer_addr = &osdc->osdmap->osd_addr[osd->o_osd];
1047 if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1048 !ceph_con_opened(&osd->o_con)) {
1049 struct ceph_osd_request *req;
1050
87b315a5
SW		 1051 dout(" osd addr hasn't changed and connection never opened,"
1052 " letting msgr retry");
1053 /* touch each r_stamp for handle_timeout()'s benfit */
1054 list_for_each_entry(req, &osd->o_requests, r_osd_item)
1055 req->r_stamp = jiffies;
c3acb181
AE		 1056
1057 return -EAGAIN;
f24e9980 1058 }
c3acb181
AE		 1059
1060 ceph_con_close(&osd->o_con);
1061 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1062 osd->o_incarnation++;
1063
1064 return 0;
f24e9980
SW		 1065}
1066
1067static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new)
1068{
1069 struct rb_node **p = &osdc->osds.rb_node;
1070 struct rb_node *parent = NULL;
1071 struct ceph_osd *osd = NULL;
1072
aca420bc 1073 dout("__insert_osd %p osd%d\n", new, new->o_osd);
f24e9980
SW		 1074 while (*p) {
1075 parent = *p;
1076 osd = rb_entry(parent, struct ceph_osd, o_node);
1077 if (new->o_osd < osd->o_osd)
1078 p = &(*p)->rb_left;
1079 else if (new->o_osd > osd->o_osd)
1080 p = &(*p)->rb_right;
1081 else
1082 BUG();
1083 }
1084
1085 rb_link_node(&new->o_node, parent, p);
1086 rb_insert_color(&new->o_node, &osdc->osds);
1087}
1088
1089static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o)
1090{
1091 struct ceph_osd *osd;
1092 struct rb_node *n = osdc->osds.rb_node;
1093
1094 while (n) {
1095 osd = rb_entry(n, struct ceph_osd, o_node);
1096 if (o < osd->o_osd)
1097 n = n->rb_left;
1098 else if (o > osd->o_osd)
1099 n = n->rb_right;
1100 else
1101 return osd;
1102 }
1103 return NULL;
1104}
1105
422d2cb8
YS		 1106static void __schedule_osd_timeout(struct ceph_osd_client *osdc)
1107{
1108 schedule_delayed_work(&osdc->timeout_work,
3d14c5d2 1109 osdc->client->options->osd_keepalive_timeout * HZ);
422d2cb8
YS		 1110}
1111
1112static void __cancel_osd_timeout(struct ceph_osd_client *osdc)
1113{
1114 cancel_delayed_work(&osdc->timeout_work);
1115}
f24e9980
SW		 1116
1117/*
1118 * Register request, assign tid. If this is the first request, set up
1119 * the timeout event.
1120 */
a40c4f10
YS		 1121static void __register_request(struct ceph_osd_client *osdc,
1122 struct ceph_osd_request *req)
f24e9980 1123{
f24e9980 1124 req->r_tid = ++osdc->last_tid;
6df058c0 1125 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
77f38e0e 1126 dout("__register_request %p tid %lld\n", req, req->r_tid);
f24e9980
SW		 1127 __insert_request(osdc, req);
1128 ceph_osdc_get_request(req);
1129 osdc->num_requests++;
f24e9980 1130 if (osdc->num_requests == 1) {
422d2cb8
YS		 1131 dout(" first request, scheduling timeout\n");
1132 __schedule_osd_timeout(osdc);
f24e9980 1133 }
a40c4f10
YS		 1134}
1135
f24e9980
SW		 1136/*
1137 * called under osdc->request_mutex
1138 */
1139static void __unregister_request(struct ceph_osd_client *osdc,
1140 struct ceph_osd_request *req)
1141{
35f9f8a0
SW		 1142 if (RB_EMPTY_NODE(&req->r_node)) {
1143 dout("__unregister_request %p tid %lld not registered\n",
1144 req, req->r_tid);
1145 return;
1146 }
1147
f24e9980
SW		 1148 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
1149 rb_erase(&req->r_node, &osdc->requests);
1150 osdc->num_requests--;
1151
0ba6478d
SW		 1152 if (req->r_osd) {
1153 /* make sure the original request isn't in flight. */
6740a845 1154 ceph_msg_revoke(req->r_request);
0ba6478d
SW		 1155
1156 list_del_init(&req->r_osd_item);
a40c4f10
YS		 1157 if (list_empty(&req->r_osd->o_requests) &&
1158 list_empty(&req->r_osd->o_linger_requests)) {
1159 dout("moving osd to %p lru\n", req->r_osd);
f5a2041b 1160 __move_osd_to_lru(osdc, req->r_osd);
a40c4f10 1161 }
fbdb9190 1162 if (list_empty(&req->r_linger_item))
a40c4f10 1163 req->r_osd = NULL;
0ba6478d 1164 }
f24e9980 1165
7d5f2481 1166 list_del_init(&req->r_req_lru_item);
f24e9980
SW		 1167 ceph_osdc_put_request(req);
1168
422d2cb8
YS		 1169 if (osdc->num_requests == 0) {
1170 dout(" no requests, canceling timeout\n");
1171 __cancel_osd_timeout(osdc);
f24e9980
SW		 1172 }
1173}
1174
1175/*
1176 * Cancel a previously queued request message
1177 */
1178static void __cancel_request(struct ceph_osd_request *req)
1179{
6bc18876 1180 if (req->r_sent && req->r_osd) {
6740a845 1181 ceph_msg_revoke(req->r_request);
f24e9980
SW		 1182 req->r_sent = 0;
1183 }
1184}
1185
a40c4f10
YS		 1186static void __register_linger_request(struct ceph_osd_client *osdc,
1187 struct ceph_osd_request *req)
1188{
1189 dout("__register_linger_request %p\n", req);
1d02ec7f 1190 ceph_osdc_get_request(req);
a40c4f10 1191 list_add_tail(&req->r_linger_item, &osdc->req_linger);
6194ea89
SW		 1192 if (req->r_osd)
1193 list_add_tail(&req->r_linger_osd,
1194 &req->r_osd->o_linger_requests);
a40c4f10
YS		 1195}
1196
1197static void __unregister_linger_request(struct ceph_osd_client *osdc,
1198 struct ceph_osd_request *req)
1199{
1200 dout("__unregister_linger_request %p\n", req);
61c74035 1201 list_del_init(&req->r_linger_item);
a40c4f10 1202 if (req->r_osd) {
a40c4f10
YS		 1203 list_del_init(&req->r_linger_osd);
1204
1205 if (list_empty(&req->r_osd->o_requests) &&
1206 list_empty(&req->r_osd->o_linger_requests)) {
1207 dout("moving osd to %p lru\n", req->r_osd);
1208 __move_osd_to_lru(osdc, req->r_osd);
1209 }
fbdb9190
SW		 1210 if (list_empty(&req->r_osd_item))
1211 req->r_osd = NULL;
a40c4f10 1212 }
1d02ec7f 1213 ceph_osdc_put_request(req);
a40c4f10
YS		 1214}
1215
1216void ceph_osdc_unregister_linger_request(struct ceph_osd_client *osdc,
1217 struct ceph_osd_request *req)
1218{
1219 mutex_lock(&osdc->request_mutex);
1220 if (req->r_linger) {
c10ebbf5 1221 req->r_linger = 0;
1d02ec7f 1222 __unregister_linger_request(osdc, req);
a40c4f10
YS		 1223 }
1224 mutex_unlock(&osdc->request_mutex);
1225}
1226EXPORT_SYMBOL(ceph_osdc_unregister_linger_request);
1227
1228void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
1229 struct ceph_osd_request *req)
1230{
1231 if (!req->r_linger) {
1232 dout("set_request_linger %p\n", req);
1233 req->r_linger = 1;
a40c4f10
YS		 1234 }
1235}
1236EXPORT_SYMBOL(ceph_osdc_set_request_linger);
1237
4892ed8d
JD		 1238/*
1239 * Returns whether a request should be blocked from being sent
1240 * based on the current osdmap and osd_client settings.
1241 *
1242 * Caller should hold map_sem for read.
1243 */
1244static bool __req_should_be_paused(struct ceph_osd_client *osdc,
1245 struct ceph_osd_request *req)
1246{
1247 bool pauserd = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD);
1248 bool pausewr = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR) ||
1249 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
1250 return (req->r_flags & CEPH_OSD_FLAG_READ && pauserd) ||
1251 (req->r_flags & CEPH_OSD_FLAG_WRITE && pausewr);
1252}
1253
f24e9980
SW		 1254/*
1255 * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
1256 * (as needed), and set the request r_osd appropriately. If there is
25985edc 1257 * no up osd, set r_osd to NULL. Move the request to the appropriate list
6f6c7006 1258 * (unsent, homeless) or leave on in-flight lru.
f24e9980
SW		 1259 *
1260 * Return 0 if unchanged, 1 if changed, or negative on error.
1261 *
1262 * Caller should hold map_sem for read and request_mutex.
1263 */
6f6c7006 1264static int __map_request(struct ceph_osd_client *osdc,
38d6453c 1265 struct ceph_osd_request *req, int force_resend)
f24e9980 1266{
5b191d99 1267 struct ceph_pg pgid;
d85b7056
SW		 1268 int acting[CEPH_PG_MAX_SIZE];
1269 int o = -1, num = 0;
f24e9980 1270 int err;
4892ed8d 1271 bool was_paused;
f24e9980 1272
6f6c7006 1273 dout("map_request %p tid %lld\n", req, req->r_tid);
41766f87
AE		 1274 err = ceph_calc_ceph_pg(&pgid, req->r_oid, osdc->osdmap,
1275 ceph_file_layout_pg_pool(req->r_file_layout));
6f6c7006
SW		 1276 if (err) {
1277 list_move(&req->r_req_lru_item, &osdc->req_notarget);
f24e9980 1278 return err;
6f6c7006 1279 }
7740a42f
SW		 1280 req->r_pgid = pgid;
1281
d85b7056
SW		 1282 err = ceph_calc_pg_acting(osdc->osdmap, pgid, acting);
1283 if (err > 0) {
1284 o = acting[0];
1285 num = err;
1286 }
f24e9980 1287
4892ed8d
JD		 1288 was_paused = req->r_paused;
1289 req->r_paused = __req_should_be_paused(osdc, req);
1290 if (was_paused && !req->r_paused)
1291 force_resend = 1;
1292
38d6453c
SW		 1293 if ((!force_resend &&
1294 req->r_osd && req->r_osd->o_osd == o &&
d85b7056
SW		 1295 req->r_sent >= req->r_osd->o_incarnation &&
1296 req->r_num_pg_osds == num &&
1297 memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
4892ed8d
JD		 1298 (req->r_osd == NULL && o == -1) ||
1299 req->r_paused)
f24e9980
SW		 1300 return 0; /* no change */
1301
5b191d99
SW		 1302 dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n",
1303 req->r_tid, pgid.pool, pgid.seed, o,
f24e9980
SW		 1304 req->r_osd ? req->r_osd->o_osd : -1);
1305
d85b7056
SW		 1306 /* record full pg acting set */
1307 memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
1308 req->r_num_pg_osds = num;
1309
f24e9980
SW		 1310 if (req->r_osd) {
1311 __cancel_request(req);
1312 list_del_init(&req->r_osd_item);
f24e9980
SW		 1313 req->r_osd = NULL;
1314 }
1315
1316 req->r_osd = __lookup_osd(osdc, o);
1317 if (!req->r_osd && o >= 0) {
c99eb1c7 1318 err = -ENOMEM;
e10006f8 1319 req->r_osd = create_osd(osdc, o);
6f6c7006
SW		 1320 if (!req->r_osd) {
1321 list_move(&req->r_req_lru_item, &osdc->req_notarget);
c99eb1c7 1322 goto out;
6f6c7006 1323 }
f24e9980 1324
6f6c7006 1325 dout("map_request osd %p is osd%d\n", req->r_osd, o);
f24e9980
SW		 1326 __insert_osd(osdc, req->r_osd);
1327
b7a9e5dd
SW		 1328 ceph_con_open(&req->r_osd->o_con,
1329 CEPH_ENTITY_TYPE_OSD, o,
1330 &osdc->osdmap->osd_addr[o]);
f24e9980
SW		 1331 }
1332
f5a2041b
YS		 1333 if (req->r_osd) {
1334 __remove_osd_from_lru(req->r_osd);
ad885927
AE		 1335 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
1336 list_move_tail(&req->r_req_lru_item, &osdc->req_unsent);
6f6c7006 1337 } else {
ad885927 1338 list_move_tail(&req->r_req_lru_item, &osdc->req_notarget);
f5a2041b 1339 }
d85b7056 1340 err = 1; /* osd or pg changed */
f24e9980
SW		 1341
1342out:
f24e9980
SW		 1343 return err;
1344}
1345
1346/*
1347 * caller should hold map_sem (for read) and request_mutex
1348 */
56e925b6
SW		 1349static void __send_request(struct ceph_osd_client *osdc,
1350 struct ceph_osd_request *req)
f24e9980 1351{
1b83bef2 1352 void *p;
f24e9980 1353
1b83bef2
SW		 1354 dout("send_request %p tid %llu to osd%d flags %d pg %lld.%x\n",
1355 req, req->r_tid, req->r_osd->o_osd, req->r_flags,
1356 (unsigned long long)req->r_pgid.pool, req->r_pgid.seed);
1357
1358 /* fill in message content that changes each time we send it */
1359 put_unaligned_le32(osdc->osdmap->epoch, req->r_request_osdmap_epoch);
1360 put_unaligned_le32(req->r_flags, req->r_request_flags);
1361 put_unaligned_le64(req->r_pgid.pool, req->r_request_pool);
1362 p = req->r_request_pgid;
1363 ceph_encode_64(&p, req->r_pgid.pool);
1364 ceph_encode_32(&p, req->r_pgid.seed);
1365 put_unaligned_le64(1, req->r_request_attempts); /* FIXME */
1366 memcpy(req->r_request_reassert_version, &req->r_reassert_version,
1367 sizeof(req->r_reassert_version));
2169aea6 1368
3dd72fc0 1369 req->r_stamp = jiffies;
07a27e22 1370 list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
f24e9980
SW		 1371
1372 ceph_msg_get(req->r_request); /* send consumes a ref */
26be8808 1373
f24e9980 1374 req->r_sent = req->r_osd->o_incarnation;
26be8808
AE		 1375
1376 ceph_con_send(&req->r_osd->o_con, req->r_request);
f24e9980
SW		 1377}
1378
6f6c7006
SW		 1379/*
1380 * Send any requests in the queue (req_unsent).
1381 */
f9d25199 1382static void __send_queued(struct ceph_osd_client *osdc)
6f6c7006
SW		 1383{
1384 struct ceph_osd_request *req, *tmp;
1385
f9d25199
AE		 1386 dout("__send_queued\n");
1387 list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item)
6f6c7006 1388 __send_request(osdc, req);
6f6c7006
SW		 1389}
1390
f24e9980
SW		 1391/*
1392 * Timeout callback, called every N seconds when 1 or more osd
1393 * requests has been active for more than N seconds. When this
1394 * happens, we ping all OSDs with requests who have timed out to
1395 * ensure any communications channel reset is detected. Reset the
1396 * request timeouts another N seconds in the future as we go.
1397 * Reschedule the timeout event another N seconds in future (unless
1398 * there are no open requests).
1399 */
1400static void handle_timeout(struct work_struct *work)
1401{
1402 struct ceph_osd_client *osdc =
1403 container_of(work, struct ceph_osd_client, timeout_work.work);
83aff95e 1404 struct ceph_osd_request *req;
f24e9980 1405 struct ceph_osd *osd;
422d2cb8 1406 unsigned long keepalive =
3d14c5d2 1407 osdc->client->options->osd_keepalive_timeout * HZ;
422d2cb8 1408 struct list_head slow_osds;
f24e9980
SW		 1409 dout("timeout\n");
1410 down_read(&osdc->map_sem);
1411
1412 ceph_monc_request_next_osdmap(&osdc->client->monc);
1413
1414 mutex_lock(&osdc->request_mutex);
f24e9980 1415
422d2cb8
YS		 1416 /*
1417 * ping osds that are a bit slow. this ensures that if there
1418 * is a break in the TCP connection we will notice, and reopen
1419 * a connection with that osd (from the fault callback).
1420 */
1421 INIT_LIST_HEAD(&slow_osds);
1422 list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
3dd72fc0 1423 if (time_before(jiffies, req->r_stamp + keepalive))
422d2cb8
YS		 1424 break;
1425
1426 osd = req->r_osd;
1427 BUG_ON(!osd);
1428 dout(" tid %llu is slow, will send keepalive on osd%d\n",
f24e9980 1429 req->r_tid, osd->o_osd);
422d2cb8
YS		 1430 list_move_tail(&osd->o_keepalive_item, &slow_osds);
1431 }
1432 while (!list_empty(&slow_osds)) {
1433 osd = list_entry(slow_osds.next, struct ceph_osd,
1434 o_keepalive_item);
1435 list_del_init(&osd->o_keepalive_item);
f24e9980
SW		 1436 ceph_con_keepalive(&osd->o_con);
1437 }
1438
422d2cb8 1439 __schedule_osd_timeout(osdc);
f9d25199 1440 __send_queued(osdc);
f24e9980 1441 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 1442 up_read(&osdc->map_sem);
1443}
1444
f5a2041b
YS		 1445static void handle_osds_timeout(struct work_struct *work)
1446{
1447 struct ceph_osd_client *osdc =
1448 container_of(work, struct ceph_osd_client,
1449 osds_timeout_work.work);
1450 unsigned long delay =
3d14c5d2 1451 osdc->client->options->osd_idle_ttl * HZ >> 2;
f5a2041b
YS		 1452
1453 dout("osds timeout\n");
1454 down_read(&osdc->map_sem);
aca420bc 1455 remove_old_osds(osdc);
f5a2041b
YS		 1456 up_read(&osdc->map_sem);
1457
1458 schedule_delayed_work(&osdc->osds_timeout_work,
1459 round_jiffies_relative(delay));
1460}
1461
25845472
SW		 1462static void complete_request(struct ceph_osd_request *req)
1463{
25845472
SW		 1464 complete_all(&req->r_safe_completion); /* fsync waiter */
1465}
1466
f24e9980
SW		 1467/*
1468 * handle osd op reply. either call the callback if it is specified,
1469 * or do the completion to wake up the waiting thread.
1470 */
350b1c32
SW		 1471static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg,
1472 struct ceph_connection *con)
f24e9980 1473{
1b83bef2 1474 void *p, *end;
f24e9980
SW		 1475 struct ceph_osd_request *req;
1476 u64 tid;
1b83bef2 1477 int object_len;
79528734
AE		 1478 unsigned int numops;
1479 int payload_len, flags;
0ceed5db 1480 s32 result;
1b83bef2
SW		 1481 s32 retry_attempt;
1482 struct ceph_pg pg;
1483 int err;
1484 u32 reassert_epoch;
1485 u64 reassert_version;
1486 u32 osdmap_epoch;
0d5af164 1487 int already_completed;
9fc6e064 1488 u32 bytes;
79528734 1489 unsigned int i;
f24e9980 1490
6df058c0 1491 tid = le64_to_cpu(msg->hdr.tid);
1b83bef2
SW		 1492 dout("handle_reply %p tid %llu\n", msg, tid);
1493
1494 p = msg->front.iov_base;
1495 end = p + msg->front.iov_len;
1496
1497 ceph_decode_need(&p, end, 4, bad);
1498 object_len = ceph_decode_32(&p);
1499 ceph_decode_need(&p, end, object_len, bad);
1500 p += object_len;
1501
ef4859d6 1502 err = ceph_decode_pgid(&p, end, &pg);
1b83bef2 1503 if (err)
f24e9980 1504 goto bad;
1b83bef2
SW		 1505
1506 ceph_decode_need(&p, end, 8 + 4 + 4 + 8 + 4, bad);
1507 flags = ceph_decode_64(&p);
1508 result = ceph_decode_32(&p);
1509 reassert_epoch = ceph_decode_32(&p);
1510 reassert_version = ceph_decode_64(&p);
1511 osdmap_epoch = ceph_decode_32(&p);
1512
f24e9980
SW		 1513 /* lookup */
1514 mutex_lock(&osdc->request_mutex);
1515 req = __lookup_request(osdc, tid);
1516 if (req == NULL) {
1517 dout("handle_reply tid %llu dne\n", tid);
8058fd45 1518 goto bad_mutex;
f24e9980
SW		 1519 }
1520 ceph_osdc_get_request(req);
1b83bef2
SW		 1521
1522 dout("handle_reply %p tid %llu req %p result %d\n", msg, tid,
1523 req, result);
1524
e9e4b13a 1525 ceph_decode_need(&p, end, 4, bad_put);
1b83bef2
SW		 1526 numops = ceph_decode_32(&p);
1527 if (numops > CEPH_OSD_MAX_OP)
1528 goto bad_put;
1529 if (numops != req->r_num_ops)
1530 goto bad_put;
1531 payload_len = 0;
e9e4b13a 1532 ceph_decode_need(&p, end, numops * sizeof(struct ceph_osd_op), bad_put);
1b83bef2
SW		 1533 for (i = 0; i < numops; i++) {
1534 struct ceph_osd_op *op = p;
1535 int len;
1536
1537 len = le32_to_cpu(op->payload_len);
1538 req->r_reply_op_len[i] = len;
1539 dout(" op %d has %d bytes\n", i, len);
1540 payload_len += len;
1541 p += sizeof(*op);
1542 }
9fc6e064
AE		 1543 bytes = le32_to_cpu(msg->hdr.data_len);
1544 if (payload_len != bytes) {
1b83bef2 1545 pr_warning("sum of op payload lens %d != data_len %d",
9fc6e064 1546 payload_len, bytes);
1b83bef2
SW		 1547 goto bad_put;
1548 }
1549
e9e4b13a 1550 ceph_decode_need(&p, end, 4 + numops * 4, bad_put);
1b83bef2
SW		 1551 retry_attempt = ceph_decode_32(&p);
1552 for (i = 0; i < numops; i++)
1553 req->r_reply_op_result[i] = ceph_decode_32(&p);
f24e9980 1554
aede2cb5
YZ		 1555 already_completed = req->r_got_reply;
1556
f24e9980 1557 if (!req->r_got_reply) {
f24e9980 1558
1b83bef2 1559 req->r_result = result;
f24e9980
SW		 1560 dout("handle_reply result %d bytes %d\n", req->r_result,
1561 bytes);
1562 if (req->r_result == 0)
1563 req->r_result = bytes;
1564
1565 /* in case this is a write and we need to replay, */
1b83bef2
SW		 1566 req->r_reassert_version.epoch = cpu_to_le32(reassert_epoch);
1567 req->r_reassert_version.version = cpu_to_le64(reassert_version);
f24e9980
SW		 1568
1569 req->r_got_reply = 1;
1570 } else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) {
1571 dout("handle_reply tid %llu dup ack\n", tid);
34b43a56 1572 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 1573 goto done;
1574 }
1575
1576 dout("handle_reply tid %llu flags %d\n", tid, flags);
1577
a40c4f10
YS		 1578 if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK))
1579 __register_linger_request(osdc, req);
1580
f24e9980 1581 /* either this is a read, or we got the safe response */
0ceed5db
SW		 1582 if (result < 0 ||
1583 (flags & CEPH_OSD_FLAG_ONDISK) ||
f24e9980
SW		 1584 ((flags & CEPH_OSD_FLAG_WRITE) == 0))
1585 __unregister_request(osdc, req);
1586
1587 mutex_unlock(&osdc->request_mutex);
1588
aede2cb5 1589 if (!already_completed) {
ef0b67e4
YZ		 1590 if (req->r_unsafe_callback &&
1591 result >= 0 && !(flags & CEPH_OSD_FLAG_ONDISK))
1592 req->r_unsafe_callback(req, true);
aede2cb5
YZ		 1593 if (req->r_callback)
1594 req->r_callback(req, msg);
1595 else
1596 complete_all(&req->r_completion);
1597 }
f24e9980 1598
ef0b67e4
YZ		 1599 if (flags & CEPH_OSD_FLAG_ONDISK) {
1600 if (req->r_unsafe_callback && already_completed)
1601 req->r_unsafe_callback(req, false);
25845472 1602 complete_request(req);
ef0b67e4 1603 }
f24e9980
SW		 1604
1605done:
a40c4f10 1606 dout("req=%p req->r_linger=%d\n", req, req->r_linger);
f24e9980
SW		 1607 ceph_osdc_put_request(req);
1608 return;
1609
1b83bef2
SW		 1610bad_put:
1611 ceph_osdc_put_request(req);
8058fd45
AE		 1612bad_mutex:
1613 mutex_unlock(&osdc->request_mutex);
f24e9980 1614bad:
1b83bef2
SW		 1615 pr_err("corrupt osd_op_reply got %d %d\n",
1616 (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len));
9ec7cab1 1617 ceph_msg_dump(msg);
f24e9980
SW		 1618}
1619
6f6c7006 1620static void reset_changed_osds(struct ceph_osd_client *osdc)
f24e9980 1621{
f24e9980 1622 struct rb_node *p, *n;
f24e9980 1623
6af167fb 1624 dout("%s %p\n", __func__, osdc);
6f6c7006
SW		 1625 for (p = rb_first(&osdc->osds); p; p = n) {
1626 struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
f24e9980 1627
6f6c7006
SW		 1628 n = rb_next(p);
1629 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
1630 memcmp(&osd->o_con.peer_addr,
1631 ceph_osd_addr(osdc->osdmap,
1632 osd->o_osd),
1633 sizeof(struct ceph_entity_addr)) != 0)
1634 __reset_osd(osdc, osd);
f24e9980 1635 }
422d2cb8
YS		 1636}
1637
1638/*
6f6c7006
SW		 1639 * Requeue requests whose mapping to an OSD has changed. If requests map to
1640 * no osd, request a new map.
422d2cb8 1641 *
e6d50f67 1642 * Caller should hold map_sem for read.
422d2cb8 1643 */
27ff9f13
JD		 1644static void kick_requests(struct ceph_osd_client *osdc, bool force_resend,
1645 bool force_resend_writes)
422d2cb8 1646{
a40c4f10 1647 struct ceph_osd_request *req, *nreq;
6f6c7006
SW		 1648 struct rb_node *p;
1649 int needmap = 0;
1650 int err;
27ff9f13 1651 bool force_resend_req;
422d2cb8 1652
27ff9f13
JD		 1653 dout("kick_requests %s %s\n", force_resend ? " (force resend)" : "",
1654 force_resend_writes ? " (force resend writes)" : "");
422d2cb8 1655 mutex_lock(&osdc->request_mutex);
6194ea89 1656 for (p = rb_first(&osdc->requests); p; ) {
6f6c7006 1657 req = rb_entry(p, struct ceph_osd_request, r_node);
6194ea89 1658 p = rb_next(p);
ab60b16d
AE		 1659
1660 /*
1661 * For linger requests that have not yet been
1662 * registered, move them to the linger list; they'll
1663 * be sent to the osd in the loop below. Unregister
1664 * the request before re-registering it as a linger
1665 * request to ensure the __map_request() below
1666 * will decide it needs to be sent.
1667 */
1668 if (req->r_linger && list_empty(&req->r_linger_item)) {
1669 dout("%p tid %llu restart on osd%d\n",
1670 req, req->r_tid,
1671 req->r_osd ? req->r_osd->o_osd : -1);
1d02ec7f 1672 ceph_osdc_get_request(req);
ab60b16d
AE		 1673 __unregister_request(osdc, req);
1674 __register_linger_request(osdc, req);
1d02ec7f 1675 ceph_osdc_put_request(req);
ab60b16d
AE		 1676 continue;
1677 }
1678
27ff9f13
JD		 1679 force_resend_req = force_resend ||
1680 (force_resend_writes &&
1681 req->r_flags & CEPH_OSD_FLAG_WRITE);
1682 err = __map_request(osdc, req, force_resend_req);
6f6c7006
SW		 1683 if (err < 0)
1684 continue; /* error */
1685 if (req->r_osd == NULL) {
1686 dout("%p tid %llu maps to no osd\n", req, req->r_tid);
1687 needmap++; /* request a newer map */
1688 } else if (err > 0) {
6194ea89
SW		 1689 if (!req->r_linger) {
1690 dout("%p tid %llu requeued on osd%d\n", req,
1691 req->r_tid,
1692 req->r_osd ? req->r_osd->o_osd : -1);
a40c4f10 1693 req->r_flags |= CEPH_OSD_FLAG_RETRY;
6194ea89
SW		 1694 }
1695 }
a40c4f10
YS		 1696 }
1697
1698 list_for_each_entry_safe(req, nreq, &osdc->req_linger,
1699 r_linger_item) {
1700 dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
1701
27ff9f13
JD		 1702 err = __map_request(osdc, req,
1703 force_resend || force_resend_writes);
ab60b16d 1704 dout("__map_request returned %d\n", err);
a40c4f10
YS		 1705 if (err < 0)
1706 continue; /* hrm! */
d2e3dbf9
ID		 1707 if (req->r_osd == NULL || err > 0) {
1708 if (req->r_osd == NULL) {
1709 dout("lingering %p tid %llu maps to no osd\n",
1710 req, req->r_tid);
1711 /*
1712 * A homeless lingering request makes
1713 * no sense, as it's job is to keep
1714 * a particular OSD connection open.
1715 * Request a newer map and kick the
1716 * request, knowing that it won't be
1717 * resent until we actually get a map
1718 * that can tell us where to send it.
1719 */
1720 needmap++;
1721 }
a40c4f10 1722
d2e3dbf9
ID		 1723 dout("kicking lingering %p tid %llu osd%d\n", req,
1724 req->r_tid, req->r_osd ? req->r_osd->o_osd : -1);
1725 __register_request(osdc, req);
1726 __unregister_linger_request(osdc, req);
1727 }
6f6c7006 1728 }
14d2f38d 1729 reset_changed_osds(osdc);
f24e9980
SW		 1730 mutex_unlock(&osdc->request_mutex);
1731
1732 if (needmap) {
1733 dout("%d requests for down osds, need new map\n", needmap);
1734 ceph_monc_request_next_osdmap(&osdc->client->monc);
1735 }
422d2cb8 1736}
6f6c7006
SW		 1737
1738
f24e9980
SW		 1739/*
1740 * Process updated osd map.
1741 *
1742 * The message contains any number of incremental and full maps, normally
1743 * indicating some sort of topology change in the cluster. Kick requests
1744 * off to different OSDs as needed.
1745 */
1746void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1747{
1748 void *p, *end, *next;
1749 u32 nr_maps, maplen;
1750 u32 epoch;
1751 struct ceph_osdmap *newmap = NULL, *oldmap;
1752 int err;
1753 struct ceph_fsid fsid;
27ff9f13 1754 bool was_full;
f24e9980
SW		 1755
1756 dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
1757 p = msg->front.iov_base;
1758 end = p + msg->front.iov_len;
1759
1760 /* verify fsid */
1761 ceph_decode_need(&p, end, sizeof(fsid), bad);
1762 ceph_decode_copy(&p, &fsid, sizeof(fsid));
0743304d
SW		 1763 if (ceph_check_fsid(osdc->client, &fsid) < 0)
1764 return;
f24e9980
SW		 1765
1766 down_write(&osdc->map_sem);
1767
27ff9f13
JD		 1768 was_full = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
1769
f24e9980
SW		 1770 /* incremental maps */
1771 ceph_decode_32_safe(&p, end, nr_maps, bad);
1772 dout(" %d inc maps\n", nr_maps);
1773 while (nr_maps > 0) {
1774 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 1775 epoch = ceph_decode_32(&p);
1776 maplen = ceph_decode_32(&p);
f24e9980
SW		 1777 ceph_decode_need(&p, end, maplen, bad);
1778 next = p + maplen;
1779 if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
1780 dout("applying incremental map %u len %d\n",
1781 epoch, maplen);
1782 newmap = osdmap_apply_incremental(&p, next,
1783 osdc->osdmap,
15d9882c 1784 &osdc->client->msgr);
f24e9980
SW		 1785 if (IS_ERR(newmap)) {
1786 err = PTR_ERR(newmap);
1787 goto bad;
1788 }
30dc6381 1789 BUG_ON(!newmap);
f24e9980
SW		 1790 if (newmap != osdc->osdmap) {
1791 ceph_osdmap_destroy(osdc->osdmap);
1792 osdc->osdmap = newmap;
1793 }
27ff9f13
JD		 1794 was_full = was_full ||
1795 ceph_osdmap_flag(osdc->osdmap,
1796 CEPH_OSDMAP_FULL);
1797 kick_requests(osdc, 0, was_full);
f24e9980
SW		 1798 } else {
1799 dout("ignoring incremental map %u len %d\n",
1800 epoch, maplen);
1801 }
1802 p = next;
1803 nr_maps--;
1804 }
1805 if (newmap)
1806 goto done;
1807
1808 /* full maps */
1809 ceph_decode_32_safe(&p, end, nr_maps, bad);
1810 dout(" %d full maps\n", nr_maps);
1811 while (nr_maps) {
1812 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 1813 epoch = ceph_decode_32(&p);
1814 maplen = ceph_decode_32(&p);
f24e9980
SW		 1815 ceph_decode_need(&p, end, maplen, bad);
1816 if (nr_maps > 1) {
1817 dout("skipping non-latest full map %u len %d\n",
1818 epoch, maplen);
1819 } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
1820 dout("skipping full map %u len %d, "
1821 "older than our %u\n", epoch, maplen,
1822 osdc->osdmap->epoch);
1823 } else {
38d6453c
SW		 1824 int skipped_map = 0;
1825
f24e9980
SW		 1826 dout("taking full map %u len %d\n", epoch, maplen);
1827 newmap = osdmap_decode(&p, p+maplen);
1828 if (IS_ERR(newmap)) {
1829 err = PTR_ERR(newmap);
1830 goto bad;
1831 }
30dc6381 1832 BUG_ON(!newmap);
f24e9980
SW		 1833 oldmap = osdc->osdmap;
1834 osdc->osdmap = newmap;
38d6453c
SW		 1835 if (oldmap) {
1836 if (oldmap->epoch + 1 < newmap->epoch)
1837 skipped_map = 1;
f24e9980 1838 ceph_osdmap_destroy(oldmap);
38d6453c 1839 }
27ff9f13
JD		 1840 was_full = was_full ||
1841 ceph_osdmap_flag(osdc->osdmap,
1842 CEPH_OSDMAP_FULL);
1843 kick_requests(osdc, skipped_map, was_full);
f24e9980
SW		 1844 }
1845 p += maplen;
1846 nr_maps--;
1847 }
1848
53341d7d
DC		 1849 if (!osdc->osdmap)
1850 goto bad;
f24e9980
SW		 1851done:
1852 downgrade_write(&osdc->map_sem);
1853 ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch);
cd634fb6
SW		 1854
1855 /*
1856 * subscribe to subsequent osdmap updates if full to ensure
1857 * we find out when we are no longer full and stop returning
1858 * ENOSPC.
1859 */
4892ed8d
JD		 1860 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) ||
1861 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD) ||
1862 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR))
cd634fb6
SW		 1863 ceph_monc_request_next_osdmap(&osdc->client->monc);
1864
f9d25199
AE		 1865 mutex_lock(&osdc->request_mutex);
1866 __send_queued(osdc);
1867 mutex_unlock(&osdc->request_mutex);
f24e9980 1868 up_read(&osdc->map_sem);
03066f23 1869 wake_up_all(&osdc->client->auth_wq);
f24e9980
SW		 1870 return;
1871
1872bad:
1873 pr_err("osdc handle_map corrupt msg\n");
9ec7cab1 1874 ceph_msg_dump(msg);
f24e9980
SW		 1875 up_write(&osdc->map_sem);
1876 return;
1877}
1878
a40c4f10
YS		 1879/*
1880 * watch/notify callback event infrastructure
1881 *
1882 * These callbacks are used both for watch and notify operations.
1883 */
1884static void __release_event(struct kref *kref)
1885{
1886 struct ceph_osd_event *event =
1887 container_of(kref, struct ceph_osd_event, kref);
1888
1889 dout("__release_event %p\n", event);
1890 kfree(event);
1891}
1892
1893static void get_event(struct ceph_osd_event *event)
1894{
1895 kref_get(&event->kref);
1896}
1897
1898void ceph_osdc_put_event(struct ceph_osd_event *event)
1899{
1900 kref_put(&event->kref, __release_event);
1901}
1902EXPORT_SYMBOL(ceph_osdc_put_event);
1903
1904static void __insert_event(struct ceph_osd_client *osdc,
1905 struct ceph_osd_event *new)
1906{
1907 struct rb_node **p = &osdc->event_tree.rb_node;
1908 struct rb_node *parent = NULL;
1909 struct ceph_osd_event *event = NULL;
1910
1911 while (*p) {
1912 parent = *p;
1913 event = rb_entry(parent, struct ceph_osd_event, node);
1914 if (new->cookie < event->cookie)
1915 p = &(*p)->rb_left;
1916 else if (new->cookie > event->cookie)
1917 p = &(*p)->rb_right;
1918 else
1919 BUG();
1920 }
1921
1922 rb_link_node(&new->node, parent, p);
1923 rb_insert_color(&new->node, &osdc->event_tree);
1924}
1925
1926static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
1927 u64 cookie)
1928{
1929 struct rb_node **p = &osdc->event_tree.rb_node;
1930 struct rb_node *parent = NULL;
1931 struct ceph_osd_event *event = NULL;
1932
1933 while (*p) {
1934 parent = *p;
1935 event = rb_entry(parent, struct ceph_osd_event, node);
1936 if (cookie < event->cookie)
1937 p = &(*p)->rb_left;
1938 else if (cookie > event->cookie)
1939 p = &(*p)->rb_right;
1940 else
1941 return event;
1942 }
1943 return NULL;
1944}
1945
1946static void __remove_event(struct ceph_osd_event *event)
1947{
1948 struct ceph_osd_client *osdc = event->osdc;
1949
1950 if (!RB_EMPTY_NODE(&event->node)) {
1951 dout("__remove_event removed %p\n", event);
1952 rb_erase(&event->node, &osdc->event_tree);
1953 ceph_osdc_put_event(event);
1954 } else {
1955 dout("__remove_event didn't remove %p\n", event);
1956 }
1957}
1958
1959int ceph_osdc_create_event(struct ceph_osd_client *osdc,
1960 void (*event_cb)(u64, u64, u8, void *),
3c663bbd 1961 void *data, struct ceph_osd_event **pevent)
a40c4f10
YS		 1962{
1963 struct ceph_osd_event *event;
1964
1965 event = kmalloc(sizeof(*event), GFP_NOIO);
1966 if (!event)
1967 return -ENOMEM;
1968
1969 dout("create_event %p\n", event);
1970 event->cb = event_cb;
3c663bbd 1971 event->one_shot = 0;
a40c4f10
YS		 1972 event->data = data;
1973 event->osdc = osdc;
1974 INIT_LIST_HEAD(&event->osd_node);
3ee5234d 1975 RB_CLEAR_NODE(&event->node);
a40c4f10
YS		 1976 kref_init(&event->kref); /* one ref for us */
1977 kref_get(&event->kref); /* one ref for the caller */
a40c4f10
YS		 1978
1979 spin_lock(&osdc->event_lock);
1980 event->cookie = ++osdc->event_count;
1981 __insert_event(osdc, event);
1982 spin_unlock(&osdc->event_lock);
1983
1984 *pevent = event;
1985 return 0;
1986}
1987EXPORT_SYMBOL(ceph_osdc_create_event);
1988
1989void ceph_osdc_cancel_event(struct ceph_osd_event *event)
1990{
1991 struct ceph_osd_client *osdc = event->osdc;
1992
1993 dout("cancel_event %p\n", event);
1994 spin_lock(&osdc->event_lock);
1995 __remove_event(event);
1996 spin_unlock(&osdc->event_lock);
1997 ceph_osdc_put_event(event); /* caller's */
1998}
1999EXPORT_SYMBOL(ceph_osdc_cancel_event);
2000
2001
2002static void do_event_work(struct work_struct *work)
2003{
2004 struct ceph_osd_event_work *event_work =
2005 container_of(work, struct ceph_osd_event_work, work);
2006 struct ceph_osd_event *event = event_work->event;
2007 u64 ver = event_work->ver;
2008 u64 notify_id = event_work->notify_id;
2009 u8 opcode = event_work->opcode;
2010
2011 dout("do_event_work completing %p\n", event);
2012 event->cb(ver, notify_id, opcode, event->data);
a40c4f10
YS		 2013 dout("do_event_work completed %p\n", event);
2014 ceph_osdc_put_event(event);
2015 kfree(event_work);
2016}
2017
2018
2019/*
2020 * Process osd watch notifications
2021 */
3c663bbd
AE		 2022static void handle_watch_notify(struct ceph_osd_client *osdc,
2023 struct ceph_msg *msg)
a40c4f10
YS		 2024{
2025 void *p, *end;
2026 u8 proto_ver;
2027 u64 cookie, ver, notify_id;
2028 u8 opcode;
2029 struct ceph_osd_event *event;
2030 struct ceph_osd_event_work *event_work;
2031
2032 p = msg->front.iov_base;
2033 end = p + msg->front.iov_len;
2034
2035 ceph_decode_8_safe(&p, end, proto_ver, bad);
2036 ceph_decode_8_safe(&p, end, opcode, bad);
2037 ceph_decode_64_safe(&p, end, cookie, bad);
2038 ceph_decode_64_safe(&p, end, ver, bad);
2039 ceph_decode_64_safe(&p, end, notify_id, bad);
2040
2041 spin_lock(&osdc->event_lock);
2042 event = __find_event(osdc, cookie);
2043 if (event) {
3c663bbd 2044 BUG_ON(event->one_shot);
a40c4f10 2045 get_event(event);
a40c4f10
YS		 2046 }
2047 spin_unlock(&osdc->event_lock);
2048 dout("handle_watch_notify cookie %lld ver %lld event %p\n",
2049 cookie, ver, event);
2050 if (event) {
2051 event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
a40c4f10
YS		 2052 if (!event_work) {
2053 dout("ERROR: could not allocate event_work\n");
2054 goto done_err;
2055 }
6b0ae409 2056 INIT_WORK(&event_work->work, do_event_work);
a40c4f10
YS		 2057 event_work->event = event;
2058 event_work->ver = ver;
2059 event_work->notify_id = notify_id;
2060 event_work->opcode = opcode;
2061 if (!queue_work(osdc->notify_wq, &event_work->work)) {
2062 dout("WARNING: failed to queue notify event work\n");
2063 goto done_err;
2064 }
2065 }
2066
2067 return;
2068
2069done_err:
a40c4f10
YS		 2070 ceph_osdc_put_event(event);
2071 return;
2072
2073bad:
2074 pr_err("osdc handle_watch_notify corrupt msg\n");
2075 return;
2076}
2077
e65550fd
AE		 2078/*
2079 * build new request AND message
2080 *
2081 */
2082void ceph_osdc_build_request(struct ceph_osd_request *req, u64 off,
2083 struct ceph_snap_context *snapc, u64 snap_id,
2084 struct timespec *mtime)
2085{
2086 struct ceph_msg *msg = req->r_request;
2087 void *p;
2088 size_t msg_size;
2089 int flags = req->r_flags;
2090 u64 data_len;
2091 unsigned int i;
2092
2093 req->r_snapid = snap_id;
2094 req->r_snapc = ceph_get_snap_context(snapc);
2095
2096 /* encode request */
2097 msg->hdr.version = cpu_to_le16(4);
2098
2099 p = msg->front.iov_base;
2100 ceph_encode_32(&p, 1); /* client_inc is always 1 */
2101 req->r_request_osdmap_epoch = p;
2102 p += 4;
2103 req->r_request_flags = p;
2104 p += 4;
2105 if (req->r_flags & CEPH_OSD_FLAG_WRITE)
2106 ceph_encode_timespec(p, mtime);
2107 p += sizeof(struct ceph_timespec);
2108 req->r_request_reassert_version = p;
2109 p += sizeof(struct ceph_eversion); /* will get filled in */
2110
2111 /* oloc */
2112 ceph_encode_8(&p, 4);
2113 ceph_encode_8(&p, 4);
2114 ceph_encode_32(&p, 8 + 4 + 4);
2115 req->r_request_pool = p;
2116 p += 8;
2117 ceph_encode_32(&p, -1); /* preferred */
2118 ceph_encode_32(&p, 0); /* key len */
2119
2120 ceph_encode_8(&p, 1);
2121 req->r_request_pgid = p;
2122 p += 8 + 4;
2123 ceph_encode_32(&p, -1); /* preferred */
2124
2125 /* oid */
2126 ceph_encode_32(&p, req->r_oid_len);
2127 memcpy(p, req->r_oid, req->r_oid_len);
2128 dout("oid '%.*s' len %d\n", req->r_oid_len, req->r_oid, req->r_oid_len);
2129 p += req->r_oid_len;
2130
2131 /* ops--can imply data */
2132 ceph_encode_16(&p, (u16)req->r_num_ops);
2133 data_len = 0;
2134 for (i = 0; i < req->r_num_ops; i++) {
2135 data_len += osd_req_encode_op(req, p, i);
2136 p += sizeof(struct ceph_osd_op);
2137 }
2138
2139 /* snaps */
2140 ceph_encode_64(&p, req->r_snapid);
2141 ceph_encode_64(&p, req->r_snapc ? req->r_snapc->seq : 0);
2142 ceph_encode_32(&p, req->r_snapc ? req->r_snapc->num_snaps : 0);
2143 if (req->r_snapc) {
2144 for (i = 0; i < snapc->num_snaps; i++) {
2145 ceph_encode_64(&p, req->r_snapc->snaps[i]);
2146 }
2147 }
2148
2149 req->r_request_attempts = p;
2150 p += 4;
2151
2152 /* data */
2153 if (flags & CEPH_OSD_FLAG_WRITE) {
2154 u16 data_off;
2155
2156 /*
2157 * The header "data_off" is a hint to the receiver
2158 * allowing it to align received data into its
2159 * buffers such that there's no need to re-copy
2160 * it before writing it to disk (direct I/O).
2161 */
2162 data_off = (u16) (off & 0xffff);
2163 req->r_request->hdr.data_off = cpu_to_le16(data_off);
2164 }
2165 req->r_request->hdr.data_len = cpu_to_le32(data_len);
2166
2167 BUG_ON(p > msg->front.iov_base + msg->front.iov_len);
2168 msg_size = p - msg->front.iov_base;
2169 msg->front.iov_len = msg_size;
2170 msg->hdr.front_len = cpu_to_le32(msg_size);
2171
2172 dout("build_request msg_size was %d\n", (int)msg_size);
2173}
2174EXPORT_SYMBOL(ceph_osdc_build_request);
2175
70636773
AE		 2176/*
2177 * Register request, send initial attempt.
2178 */
2179int ceph_osdc_start_request(struct ceph_osd_client *osdc,
2180 struct ceph_osd_request *req,
2181 bool nofail)
2182{
2183 int rc = 0;
2184
f24e9980
SW		 2185 down_read(&osdc->map_sem);
2186 mutex_lock(&osdc->request_mutex);
dc4b870c 2187 __register_request(osdc, req);
c10ebbf5
AE		 2188 req->r_sent = 0;
2189 req->r_got_reply = 0;
92451b49
AE		 2190 rc = __map_request(osdc, req, 0);
2191 if (rc < 0) {
2192 if (nofail) {
2193 dout("osdc_start_request failed map, "
2194 " will retry %lld\n", req->r_tid);
2195 rc = 0;
2ab0ad6a 2196 } else {
2197 __unregister_request(osdc, req);
f24e9980 2198 }
92451b49 2199 goto out_unlock;
f24e9980 2200 }
92451b49
AE		 2201 if (req->r_osd == NULL) {
2202 dout("send_request %p no up osds in pg\n", req);
2203 ceph_monc_request_next_osdmap(&osdc->client->monc);
2204 } else {
7e2766a1 2205 __send_queued(osdc);
92451b49
AE		 2206 }
2207 rc = 0;
234af26f 2208out_unlock:
f24e9980
SW		 2209 mutex_unlock(&osdc->request_mutex);
2210 up_read(&osdc->map_sem);
2211 return rc;
2212}
3d14c5d2 2213EXPORT_SYMBOL(ceph_osdc_start_request);
f24e9980
SW		 2214
2215/*
2216 * wait for a request to complete
2217 */
2218int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
2219 struct ceph_osd_request *req)
2220{
2221 int rc;
2222
2223 rc = wait_for_completion_interruptible(&req->r_completion);
2224 if (rc < 0) {
2225 mutex_lock(&osdc->request_mutex);
2226 __cancel_request(req);
529cfcc4 2227 __unregister_request(osdc, req);
f24e9980 2228 mutex_unlock(&osdc->request_mutex);
25845472 2229 complete_request(req);
529cfcc4 2230 dout("wait_request tid %llu canceled/timed out\n", req->r_tid);
f24e9980
SW		 2231 return rc;
2232 }
2233
2234 dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result);
2235 return req->r_result;
2236}
3d14c5d2 2237EXPORT_SYMBOL(ceph_osdc_wait_request);
f24e9980
SW		 2238
2239/*
2240 * sync - wait for all in-flight requests to flush. avoid starvation.
2241 */
2242void ceph_osdc_sync(struct ceph_osd_client *osdc)
2243{
2244 struct ceph_osd_request *req;
2245 u64 last_tid, next_tid = 0;
2246
2247 mutex_lock(&osdc->request_mutex);
2248 last_tid = osdc->last_tid;
2249 while (1) {
2250 req = __lookup_request_ge(osdc, next_tid);
2251 if (!req)
2252 break;
2253 if (req->r_tid > last_tid)
2254 break;
2255
2256 next_tid = req->r_tid + 1;
2257 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
2258 continue;
2259
2260 ceph_osdc_get_request(req);
2261 mutex_unlock(&osdc->request_mutex);
2262 dout("sync waiting on tid %llu (last is %llu)\n",
2263 req->r_tid, last_tid);
2264 wait_for_completion(&req->r_safe_completion);
2265 mutex_lock(&osdc->request_mutex);
2266 ceph_osdc_put_request(req);
2267 }
2268 mutex_unlock(&osdc->request_mutex);
2269 dout("sync done (thru tid %llu)\n", last_tid);
2270}
3d14c5d2 2271EXPORT_SYMBOL(ceph_osdc_sync);
f24e9980 2272
a2e5951b
JD		 2273/*
2274 * Call all pending notify callbacks - for use after a watch is
2275 * unregistered, to make sure no more callbacks for it will be invoked
2276 */
2277extern void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc)
2278{
2279 flush_workqueue(osdc->notify_wq);
2280}
2281EXPORT_SYMBOL(ceph_osdc_flush_notifies);
2282
2283
f24e9980
SW		 2284/*
2285 * init, shutdown
2286 */
2287int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
2288{
2289 int err;
2290
2291 dout("init\n");
2292 osdc->client = client;
2293 osdc->osdmap = NULL;
2294 init_rwsem(&osdc->map_sem);
2295 init_completion(&osdc->map_waiters);
2296 osdc->last_requested_map = 0;
2297 mutex_init(&osdc->request_mutex);
f24e9980
SW		 2298 osdc->last_tid = 0;
2299 osdc->osds = RB_ROOT;
f5a2041b 2300 INIT_LIST_HEAD(&osdc->osd_lru);
f24e9980 2301 osdc->requests = RB_ROOT;
422d2cb8 2302 INIT_LIST_HEAD(&osdc->req_lru);
6f6c7006
SW		 2303 INIT_LIST_HEAD(&osdc->req_unsent);
2304 INIT_LIST_HEAD(&osdc->req_notarget);
a40c4f10 2305 INIT_LIST_HEAD(&osdc->req_linger);
f24e9980
SW		 2306 osdc->num_requests = 0;
2307 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
f5a2041b 2308 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
a40c4f10
YS		 2309 spin_lock_init(&osdc->event_lock);
2310 osdc->event_tree = RB_ROOT;
2311 osdc->event_count = 0;
f5a2041b
YS		 2312
2313 schedule_delayed_work(&osdc->osds_timeout_work,
3d14c5d2 2314 round_jiffies_relative(osdc->client->options->osd_idle_ttl * HZ));
f24e9980 2315
5f44f142 2316 err = -ENOMEM;
f24e9980
SW		 2317 osdc->req_mempool = mempool_create_kmalloc_pool(10,
2318 sizeof(struct ceph_osd_request));
2319 if (!osdc->req_mempool)
5f44f142 2320 goto out;
f24e9980 2321
d50b409f
SW		 2322 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
2323 OSD_OP_FRONT_LEN, 10, true,
4f48280e 2324 "osd_op");
f24e9980 2325 if (err < 0)
5f44f142 2326 goto out_mempool;
d50b409f 2327 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
4f48280e
SW		 2328 OSD_OPREPLY_FRONT_LEN, 10, true,
2329 "osd_op_reply");
c16e7869
SW		 2330 if (err < 0)
2331 goto out_msgpool;
a40c4f10 2332
3834fb30 2333 err = -ENOMEM;
a40c4f10 2334 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
3834fb30 2335 if (!osdc->notify_wq)
a40c4f10 2336 goto out_msgpool;
f24e9980 2337 return 0;
5f44f142 2338
c16e7869
SW		 2339out_msgpool:
2340 ceph_msgpool_destroy(&osdc->msgpool_op);
5f44f142
SW		 2341out_mempool:
2342 mempool_destroy(osdc->req_mempool);
2343out:
2344 return err;
f24e9980
SW		 2345}
2346
2347void ceph_osdc_stop(struct ceph_osd_client *osdc)
2348{
a40c4f10
YS		 2349 flush_workqueue(osdc->notify_wq);
2350 destroy_workqueue(osdc->notify_wq);
f24e9980 2351 cancel_delayed_work_sync(&osdc->timeout_work);
f5a2041b 2352 cancel_delayed_work_sync(&osdc->osds_timeout_work);
f24e9980
SW		 2353 if (osdc->osdmap) {
2354 ceph_osdmap_destroy(osdc->osdmap);
2355 osdc->osdmap = NULL;
2356 }
aca420bc 2357 remove_all_osds(osdc);
f24e9980
SW		 2358 mempool_destroy(osdc->req_mempool);
2359 ceph_msgpool_destroy(&osdc->msgpool_op);
c16e7869 2360 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
f24e9980
SW		 2361}
2362
2363/*
2364 * Read some contiguous pages. If we cross a stripe boundary, shorten
2365 * *plen. Return number of bytes read, or error.
2366 */
2367int ceph_osdc_readpages(struct ceph_osd_client *osdc,
2368 struct ceph_vino vino, struct ceph_file_layout *layout,
2369 u64 off, u64 *plen,
2370 u32 truncate_seq, u64 truncate_size,
b7495fc2 2371 struct page **pages, int num_pages, int page_align)
f24e9980
SW		 2372{
2373 struct ceph_osd_request *req;
2374 int rc = 0;
2375
2376 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
2377 vino.snap, off, *plen);
79528734 2378 req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 1,
f24e9980 2379 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
acead002 2380 NULL, truncate_seq, truncate_size,
153e5167 2381 false);
6816282d
SW		 2382 if (IS_ERR(req))
2383 return PTR_ERR(req);
f24e9980
SW		 2384
2385 /* it may be a short read due to an object boundary */
0fff87ec 2386
406e2c9f 2387 osd_req_op_extent_osd_data_pages(req, 0,
a4ce40a9 2388 pages, *plen, page_align, false, false);
f24e9980 2389
e0c59487 2390 dout("readpages final extent is %llu~%llu (%llu bytes align %d)\n",
43bfe5de 2391 off, *plen, *plen, page_align);
f24e9980 2392
79528734 2393 ceph_osdc_build_request(req, off, NULL, vino.snap, NULL);
02ee07d3 2394
f24e9980
SW		 2395 rc = ceph_osdc_start_request(osdc, req, false);
2396 if (!rc)
2397 rc = ceph_osdc_wait_request(osdc, req);
2398
2399 ceph_osdc_put_request(req);
2400 dout("readpages result %d\n", rc);
2401 return rc;
2402}
3d14c5d2 2403EXPORT_SYMBOL(ceph_osdc_readpages);
f24e9980
SW		 2404
2405/*
2406 * do a synchronous write on N pages
2407 */
2408int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
2409 struct ceph_file_layout *layout,
2410 struct ceph_snap_context *snapc,
2411 u64 off, u64 len,
2412 u32 truncate_seq, u64 truncate_size,
2413 struct timespec *mtime,
24808826 2414 struct page **pages, int num_pages)
f24e9980
SW		 2415{
2416 struct ceph_osd_request *req;
2417 int rc = 0;
b7495fc2 2418 int page_align = off & ~PAGE_MASK;
f24e9980 2419
acead002 2420 BUG_ON(vino.snap != CEPH_NOSNAP); /* snapshots aren't writeable */
79528734 2421 req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 1,
f24e9980 2422 CEPH_OSD_OP_WRITE,
24808826 2423 CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
acead002 2424 snapc, truncate_seq, truncate_size,
153e5167 2425 true);
6816282d
SW		 2426 if (IS_ERR(req))
2427 return PTR_ERR(req);
f24e9980
SW		 2428
2429 /* it may be a short write due to an object boundary */
406e2c9f 2430 osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align,
43bfe5de
AE		 2431 false, false);
2432 dout("writepages %llu~%llu (%llu bytes)\n", off, len, len);
f24e9980 2433
79528734 2434 ceph_osdc_build_request(req, off, snapc, CEPH_NOSNAP, mtime);
02ee07d3 2435
87f979d3 2436 rc = ceph_osdc_start_request(osdc, req, true);
f24e9980
SW		 2437 if (!rc)
2438 rc = ceph_osdc_wait_request(osdc, req);
2439
2440 ceph_osdc_put_request(req);
2441 if (rc == 0)
2442 rc = len;
2443 dout("writepages result %d\n", rc);
2444 return rc;
2445}
3d14c5d2 2446EXPORT_SYMBOL(ceph_osdc_writepages);
f24e9980 2447
5522ae0b
AE		 2448int ceph_osdc_setup(void)
2449{
2450 BUG_ON(ceph_osd_request_cache);
2451 ceph_osd_request_cache = kmem_cache_create("ceph_osd_request",
2452 sizeof (struct ceph_osd_request),
2453 __alignof__(struct ceph_osd_request),
2454 0, NULL);
2455
2456 return ceph_osd_request_cache ? 0 : -ENOMEM;
2457}
2458EXPORT_SYMBOL(ceph_osdc_setup);
2459
2460void ceph_osdc_cleanup(void)
2461{
2462 BUG_ON(!ceph_osd_request_cache);
2463 kmem_cache_destroy(ceph_osd_request_cache);
2464 ceph_osd_request_cache = NULL;
2465}
2466EXPORT_SYMBOL(ceph_osdc_cleanup);
2467
f24e9980
SW		 2468/*
2469 * handle incoming message
2470 */
2471static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
2472{
2473 struct ceph_osd *osd = con->private;
32c895e7 2474 struct ceph_osd_client *osdc;
f24e9980
SW		 2475 int type = le16_to_cpu(msg->hdr.type);
2476
2477 if (!osd)
4a32f93d 2478 goto out;
32c895e7 2479 osdc = osd->o_osdc;
f24e9980
SW		 2480
2481 switch (type) {
2482 case CEPH_MSG_OSD_MAP:
2483 ceph_osdc_handle_map(osdc, msg);
2484 break;
2485 case CEPH_MSG_OSD_OPREPLY:
350b1c32 2486 handle_reply(osdc, msg, con);
f24e9980 2487 break;
a40c4f10
YS		 2488 case CEPH_MSG_WATCH_NOTIFY:
2489 handle_watch_notify(osdc, msg);
2490 break;
f24e9980
SW		 2491
2492 default:
2493 pr_err("received unknown message type %d %s\n", type,
2494 ceph_msg_type_name(type));
2495 }
4a32f93d 2496out:
f24e9980
SW		 2497 ceph_msg_put(msg);
2498}
2499
5b3a4db3 2500/*
21b667f6
SW		 2501 * lookup and return message for incoming reply. set up reply message
2502 * pages.
5b3a4db3
SW		 2503 */
2504static struct ceph_msg *get_reply(struct ceph_connection *con,
2450418c
YS		 2505 struct ceph_msg_header *hdr,
2506 int *skip)
f24e9980
SW		 2507{
2508 struct ceph_osd *osd = con->private;
2509 struct ceph_osd_client *osdc = osd->o_osdc;
2450418c 2510 struct ceph_msg *m;
0547a9b3 2511 struct ceph_osd_request *req;
5b3a4db3
SW		 2512 int front = le32_to_cpu(hdr->front_len);
2513 int data_len = le32_to_cpu(hdr->data_len);
0547a9b3 2514 u64 tid;
f24e9980 2515
0547a9b3
YS		 2516 tid = le64_to_cpu(hdr->tid);
2517 mutex_lock(&osdc->request_mutex);
2518 req = __lookup_request(osdc, tid);
2519 if (!req) {
2520 *skip = 1;
2521 m = NULL;
756a16a5
SW		 2522 dout("get_reply unknown tid %llu from osd%d\n", tid,
2523 osd->o_osd);
0547a9b3
YS		 2524 goto out;
2525 }
c16e7869 2526
ace6d3a9 2527 if (req->r_reply->con)
8921d114 2528 dout("%s revoking msg %p from old con %p\n", __func__,
ace6d3a9
AE		 2529 req->r_reply, req->r_reply->con);
2530 ceph_msg_revoke_incoming(req->r_reply);
0547a9b3 2531
c16e7869
SW		 2532 if (front > req->r_reply->front.iov_len) {
2533 pr_warning("get_reply front %d > preallocated %d\n",
2534 front, (int)req->r_reply->front.iov_len);
b61c2763 2535 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front, GFP_NOFS, false);
a79832f2 2536 if (!m)
c16e7869
SW		 2537 goto out;
2538 ceph_msg_put(req->r_reply);
2539 req->r_reply = m;
2540 }
2541 m = ceph_msg_get(req->r_reply);
2542
0547a9b3 2543 if (data_len > 0) {
a4ce40a9 2544 struct ceph_osd_data *osd_data;
0fff87ec 2545
a4ce40a9
AE		 2546 /*
2547 * XXX This is assuming there is only one op containing
2548 * XXX page data. Probably OK for reads, but this
2549 * XXX ought to be done more generally.
2550 */
406e2c9f 2551 osd_data = osd_req_op_extent_osd_data(req, 0);
0fff87ec 2552 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
0fff87ec 2553 if (osd_data->pages &&
e0c59487 2554 unlikely(osd_data->length < data_len)) {
2ac2b7a6 2555
e0c59487
AE		 2556 pr_warning("tid %lld reply has %d bytes "
2557 "we had only %llu bytes ready\n",
2558 tid, data_len, osd_data->length);
2ac2b7a6
AE		 2559 *skip = 1;
2560 ceph_msg_put(m);
2561 m = NULL;
2562 goto out;
2563 }
2ac2b7a6 2564 }
0547a9b3 2565 }
5b3a4db3 2566 *skip = 0;
c16e7869 2567 dout("get_reply tid %lld %p\n", tid, m);
0547a9b3
YS		 2568
2569out:
2570 mutex_unlock(&osdc->request_mutex);
2450418c 2571 return m;
5b3a4db3
SW		 2572
2573}
2574
2575static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2576 struct ceph_msg_header *hdr,
2577 int *skip)
2578{
2579 struct ceph_osd *osd = con->private;
2580 int type = le16_to_cpu(hdr->type);
2581 int front = le32_to_cpu(hdr->front_len);
2582
1c20f2d2 2583 *skip = 0;
5b3a4db3
SW		 2584 switch (type) {
2585 case CEPH_MSG_OSD_MAP:
a40c4f10 2586 case CEPH_MSG_WATCH_NOTIFY:
b61c2763 2587 return ceph_msg_new(type, front, GFP_NOFS, false);
5b3a4db3
SW		 2588 case CEPH_MSG_OSD_OPREPLY:
2589 return get_reply(con, hdr, skip);
2590 default:
2591 pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
2592 osd->o_osd);
2593 *skip = 1;
2594 return NULL;
2595 }
f24e9980
SW		 2596}
2597
2598/*
2599 * Wrappers to refcount containing ceph_osd struct
2600 */
2601static struct ceph_connection *get_osd_con(struct ceph_connection *con)
2602{
2603 struct ceph_osd *osd = con->private;
2604 if (get_osd(osd))
2605 return con;
2606 return NULL;
2607}
2608
2609static void put_osd_con(struct ceph_connection *con)
2610{
2611 struct ceph_osd *osd = con->private;
2612 put_osd(osd);
2613}
2614
4e7a5dcd
SW		 2615/*
2616 * authentication
2617 */
a3530df3
AE		 2618/*
2619 * Note: returned pointer is the address of a structure that's
2620 * managed separately. Caller must *not* attempt to free it.
2621 */
2622static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
8f43fb53 2623 int *proto, int force_new)
4e7a5dcd
SW		 2624{
2625 struct ceph_osd *o = con->private;
2626 struct ceph_osd_client *osdc = o->o_osdc;
2627 struct ceph_auth_client *ac = osdc->client->monc.auth;
74f1869f 2628 struct ceph_auth_handshake *auth = &o->o_auth;
4e7a5dcd 2629
74f1869f 2630 if (force_new && auth->authorizer) {
27859f97 2631 ceph_auth_destroy_authorizer(ac, auth->authorizer);
74f1869f
AE		 2632 auth->authorizer = NULL;
2633 }
27859f97
SW		 2634 if (!auth->authorizer) {
2635 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2636 auth);
4e7a5dcd 2637 if (ret)
a3530df3 2638 return ERR_PTR(ret);
27859f97
SW		 2639 } else {
2640 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
0bed9b5c
SW		 2641 auth);
2642 if (ret)
2643 return ERR_PTR(ret);
4e7a5dcd 2644 }
4e7a5dcd 2645 *proto = ac->protocol;
74f1869f 2646
a3530df3 2647 return auth;
4e7a5dcd
SW		 2648}
2649
2650
2651static int verify_authorizer_reply(struct ceph_connection *con, int len)
2652{
2653 struct ceph_osd *o = con->private;
2654 struct ceph_osd_client *osdc = o->o_osdc;
2655 struct ceph_auth_client *ac = osdc->client->monc.auth;
2656
27859f97 2657 return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer, len);
4e7a5dcd
SW		 2658}
2659
9bd2e6f8
SW		 2660static int invalidate_authorizer(struct ceph_connection *con)
2661{
2662 struct ceph_osd *o = con->private;
2663 struct ceph_osd_client *osdc = o->o_osdc;
2664 struct ceph_auth_client *ac = osdc->client->monc.auth;
2665
27859f97 2666 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
9bd2e6f8
SW		 2667 return ceph_monc_validate_auth(&osdc->client->monc);
2668}
4e7a5dcd 2669
9e32789f 2670static const struct ceph_connection_operations osd_con_ops = {
f24e9980
SW		 2671 .get = get_osd_con,
2672 .put = put_osd_con,
2673 .dispatch = dispatch,
4e7a5dcd
SW		 2674 .get_authorizer = get_authorizer,
2675 .verify_authorizer_reply = verify_authorizer_reply,
9bd2e6f8 2676 .invalidate_authorizer = invalidate_authorizer,
f24e9980 2677 .alloc_msg = alloc_msg,
81b024e7 2678 .fault = osd_reset,
f24e9980 2679};
kernel source for telekom puls (alcatel/mediatek)