projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
ceph: fix request max size
[linux-2.6-block.git] / net / ceph / osd_client.c
CommitLineData
a4ce40a9 1
3d14c5d2 2#include <linux/ceph/ceph_debug.h>
f24e9980 3
3d14c5d2 4#include <linux/module.h>
f24e9980
SW		 5#include <linux/err.h>
6#include <linux/highmem.h>
7#include <linux/mm.h>
8#include <linux/pagemap.h>
9#include <linux/slab.h>
10#include <linux/uaccess.h>
68b4476b
YS		 11#ifdef CONFIG_BLOCK
12#include <linux/bio.h>
13#endif
f24e9980 14
3d14c5d2
YS		 15#include <linux/ceph/libceph.h>
16#include <linux/ceph/osd_client.h>
17#include <linux/ceph/messenger.h>
18#include <linux/ceph/decode.h>
19#include <linux/ceph/auth.h>
20#include <linux/ceph/pagelist.h>
f24e9980 21
c16e7869
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;
ccca4e37
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);
96e4dac6 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 }
96e4dac6 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;
96e4dac6 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
f24e9980
SW		 1227/*
1228 * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
1229 * (as needed), and set the request r_osd appropriately. If there is
25985edc 1230 * no up osd, set r_osd to NULL. Move the request to the appropriate list
6f6c7006 1231 * (unsent, homeless) or leave on in-flight lru.
f24e9980
SW		 1232 *
1233 * Return 0 if unchanged, 1 if changed, or negative on error.
1234 *
1235 * Caller should hold map_sem for read and request_mutex.
1236 */
6f6c7006 1237static int __map_request(struct ceph_osd_client *osdc,
38d6453c 1238 struct ceph_osd_request *req, int force_resend)
f24e9980 1239{
5b191d99 1240 struct ceph_pg pgid;
d85b7056
SW		 1241 int acting[CEPH_PG_MAX_SIZE];
1242 int o = -1, num = 0;
f24e9980 1243 int err;
f24e9980 1244
6f6c7006 1245 dout("map_request %p tid %lld\n", req, req->r_tid);
41766f87
AE		 1246 err = ceph_calc_ceph_pg(&pgid, req->r_oid, osdc->osdmap,
1247 ceph_file_layout_pg_pool(req->r_file_layout));
6f6c7006
SW		 1248 if (err) {
1249 list_move(&req->r_req_lru_item, &osdc->req_notarget);
f24e9980 1250 return err;
6f6c7006 1251 }
7740a42f
SW		 1252 req->r_pgid = pgid;
1253
d85b7056
SW		 1254 err = ceph_calc_pg_acting(osdc->osdmap, pgid, acting);
1255 if (err > 0) {
1256 o = acting[0];
1257 num = err;
1258 }
f24e9980 1259
38d6453c
SW		 1260 if ((!force_resend &&
1261 req->r_osd && req->r_osd->o_osd == o &&
d85b7056
SW		 1262 req->r_sent >= req->r_osd->o_incarnation &&
1263 req->r_num_pg_osds == num &&
1264 memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
f24e9980
SW		 1265 (req->r_osd == NULL && o == -1))
1266 return 0; /* no change */
1267
5b191d99
SW		 1268 dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n",
1269 req->r_tid, pgid.pool, pgid.seed, o,
f24e9980
SW		 1270 req->r_osd ? req->r_osd->o_osd : -1);
1271
d85b7056
SW		 1272 /* record full pg acting set */
1273 memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
1274 req->r_num_pg_osds = num;
1275
f24e9980
SW		 1276 if (req->r_osd) {
1277 __cancel_request(req);
1278 list_del_init(&req->r_osd_item);
f24e9980
SW		 1279 req->r_osd = NULL;
1280 }
1281
1282 req->r_osd = __lookup_osd(osdc, o);
1283 if (!req->r_osd && o >= 0) {
c99eb1c7 1284 err = -ENOMEM;
e10006f8 1285 req->r_osd = create_osd(osdc, o);
6f6c7006
SW		 1286 if (!req->r_osd) {
1287 list_move(&req->r_req_lru_item, &osdc->req_notarget);
c99eb1c7 1288 goto out;
6f6c7006 1289 }
f24e9980 1290
6f6c7006 1291 dout("map_request osd %p is osd%d\n", req->r_osd, o);
f24e9980
SW		 1292 __insert_osd(osdc, req->r_osd);
1293
b7a9e5dd
SW		 1294 ceph_con_open(&req->r_osd->o_con,
1295 CEPH_ENTITY_TYPE_OSD, o,
1296 &osdc->osdmap->osd_addr[o]);
f24e9980
SW		 1297 }
1298
f5a2041b
YS		 1299 if (req->r_osd) {
1300 __remove_osd_from_lru(req->r_osd);
ad885927
AE		 1301 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
1302 list_move_tail(&req->r_req_lru_item, &osdc->req_unsent);
6f6c7006 1303 } else {
ad885927 1304 list_move_tail(&req->r_req_lru_item, &osdc->req_notarget);
f5a2041b 1305 }
d85b7056 1306 err = 1; /* osd or pg changed */
f24e9980
SW		 1307
1308out:
f24e9980
SW		 1309 return err;
1310}
1311
1312/*
1313 * caller should hold map_sem (for read) and request_mutex
1314 */
56e925b6
SW		 1315static void __send_request(struct ceph_osd_client *osdc,
1316 struct ceph_osd_request *req)
f24e9980 1317{
1b83bef2 1318 void *p;
f24e9980 1319
1b83bef2
SW		 1320 dout("send_request %p tid %llu to osd%d flags %d pg %lld.%x\n",
1321 req, req->r_tid, req->r_osd->o_osd, req->r_flags,
1322 (unsigned long long)req->r_pgid.pool, req->r_pgid.seed);
1323
1324 /* fill in message content that changes each time we send it */
1325 put_unaligned_le32(osdc->osdmap->epoch, req->r_request_osdmap_epoch);
1326 put_unaligned_le32(req->r_flags, req->r_request_flags);
1327 put_unaligned_le64(req->r_pgid.pool, req->r_request_pool);
1328 p = req->r_request_pgid;
1329 ceph_encode_64(&p, req->r_pgid.pool);
1330 ceph_encode_32(&p, req->r_pgid.seed);
1331 put_unaligned_le64(1, req->r_request_attempts); /* FIXME */
1332 memcpy(req->r_request_reassert_version, &req->r_reassert_version,
1333 sizeof(req->r_reassert_version));
2169aea6 1334
3dd72fc0 1335 req->r_stamp = jiffies;
07a27e22 1336 list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
f24e9980
SW		 1337
1338 ceph_msg_get(req->r_request); /* send consumes a ref */
26be8808 1339
f24e9980 1340 req->r_sent = req->r_osd->o_incarnation;
26be8808
AE		 1341
1342 ceph_con_send(&req->r_osd->o_con, req->r_request);
f24e9980
SW		 1343}
1344
6f6c7006
SW		 1345/*
1346 * Send any requests in the queue (req_unsent).
1347 */
f9d25199 1348static void __send_queued(struct ceph_osd_client *osdc)
6f6c7006
SW		 1349{
1350 struct ceph_osd_request *req, *tmp;
1351
f9d25199
AE		 1352 dout("__send_queued\n");
1353 list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item)
6f6c7006 1354 __send_request(osdc, req);
6f6c7006
SW		 1355}
1356
f24e9980
SW		 1357/*
1358 * Timeout callback, called every N seconds when 1 or more osd
1359 * requests has been active for more than N seconds. When this
1360 * happens, we ping all OSDs with requests who have timed out to
1361 * ensure any communications channel reset is detected. Reset the
1362 * request timeouts another N seconds in the future as we go.
1363 * Reschedule the timeout event another N seconds in future (unless
1364 * there are no open requests).
1365 */
1366static void handle_timeout(struct work_struct *work)
1367{
1368 struct ceph_osd_client *osdc =
1369 container_of(work, struct ceph_osd_client, timeout_work.work);
83aff95e 1370 struct ceph_osd_request *req;
f24e9980 1371 struct ceph_osd *osd;
422d2cb8 1372 unsigned long keepalive =
3d14c5d2 1373 osdc->client->options->osd_keepalive_timeout * HZ;
422d2cb8 1374 struct list_head slow_osds;
f24e9980
SW		 1375 dout("timeout\n");
1376 down_read(&osdc->map_sem);
1377
1378 ceph_monc_request_next_osdmap(&osdc->client->monc);
1379
1380 mutex_lock(&osdc->request_mutex);
f24e9980 1381
422d2cb8
YS		 1382 /*
1383 * ping osds that are a bit slow. this ensures that if there
1384 * is a break in the TCP connection we will notice, and reopen
1385 * a connection with that osd (from the fault callback).
1386 */
1387 INIT_LIST_HEAD(&slow_osds);
1388 list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
3dd72fc0 1389 if (time_before(jiffies, req->r_stamp + keepalive))
422d2cb8
YS		 1390 break;
1391
1392 osd = req->r_osd;
1393 BUG_ON(!osd);
1394 dout(" tid %llu is slow, will send keepalive on osd%d\n",
f24e9980 1395 req->r_tid, osd->o_osd);
422d2cb8
YS		 1396 list_move_tail(&osd->o_keepalive_item, &slow_osds);
1397 }
1398 while (!list_empty(&slow_osds)) {
1399 osd = list_entry(slow_osds.next, struct ceph_osd,
1400 o_keepalive_item);
1401 list_del_init(&osd->o_keepalive_item);
f24e9980
SW		 1402 ceph_con_keepalive(&osd->o_con);
1403 }
1404
422d2cb8 1405 __schedule_osd_timeout(osdc);
f9d25199 1406 __send_queued(osdc);
f24e9980 1407 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 1408 up_read(&osdc->map_sem);
1409}
1410
f5a2041b
YS		 1411static void handle_osds_timeout(struct work_struct *work)
1412{
1413 struct ceph_osd_client *osdc =
1414 container_of(work, struct ceph_osd_client,
1415 osds_timeout_work.work);
1416 unsigned long delay =
3d14c5d2 1417 osdc->client->options->osd_idle_ttl * HZ >> 2;
f5a2041b
YS		 1418
1419 dout("osds timeout\n");
1420 down_read(&osdc->map_sem);
aca420bc 1421 remove_old_osds(osdc);
f5a2041b
YS		 1422 up_read(&osdc->map_sem);
1423
1424 schedule_delayed_work(&osdc->osds_timeout_work,
1425 round_jiffies_relative(delay));
1426}
1427
25845472
SW		 1428static void complete_request(struct ceph_osd_request *req)
1429{
25845472
SW		 1430 complete_all(&req->r_safe_completion); /* fsync waiter */
1431}
1432
f24e9980
SW		 1433/*
1434 * handle osd op reply. either call the callback if it is specified,
1435 * or do the completion to wake up the waiting thread.
1436 */
350b1c32
SW		 1437static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg,
1438 struct ceph_connection *con)
f24e9980 1439{
1b83bef2 1440 void *p, *end;
f24e9980
SW		 1441 struct ceph_osd_request *req;
1442 u64 tid;
1b83bef2 1443 int object_len;
79528734
AE		 1444 unsigned int numops;
1445 int payload_len, flags;
0ceed5db 1446 s32 result;
1b83bef2
SW		 1447 s32 retry_attempt;
1448 struct ceph_pg pg;
1449 int err;
1450 u32 reassert_epoch;
1451 u64 reassert_version;
1452 u32 osdmap_epoch;
0d5af164 1453 int already_completed;
9fc6e064 1454 u32 bytes;
79528734 1455 unsigned int i;
f24e9980 1456
6df058c0 1457 tid = le64_to_cpu(msg->hdr.tid);
1b83bef2
SW		 1458 dout("handle_reply %p tid %llu\n", msg, tid);
1459
1460 p = msg->front.iov_base;
1461 end = p + msg->front.iov_len;
1462
1463 ceph_decode_need(&p, end, 4, bad);
1464 object_len = ceph_decode_32(&p);
1465 ceph_decode_need(&p, end, object_len, bad);
1466 p += object_len;
1467
ef4859d6 1468 err = ceph_decode_pgid(&p, end, &pg);
1b83bef2 1469 if (err)
f24e9980 1470 goto bad;
1b83bef2
SW		 1471
1472 ceph_decode_need(&p, end, 8 + 4 + 4 + 8 + 4, bad);
1473 flags = ceph_decode_64(&p);
1474 result = ceph_decode_32(&p);
1475 reassert_epoch = ceph_decode_32(&p);
1476 reassert_version = ceph_decode_64(&p);
1477 osdmap_epoch = ceph_decode_32(&p);
1478
f24e9980
SW		 1479 /* lookup */
1480 mutex_lock(&osdc->request_mutex);
1481 req = __lookup_request(osdc, tid);
1482 if (req == NULL) {
1483 dout("handle_reply tid %llu dne\n", tid);
8058fd45 1484 goto bad_mutex;
f24e9980
SW		 1485 }
1486 ceph_osdc_get_request(req);
1b83bef2
SW		 1487
1488 dout("handle_reply %p tid %llu req %p result %d\n", msg, tid,
1489 req, result);
1490
1491 ceph_decode_need(&p, end, 4, bad);
1492 numops = ceph_decode_32(&p);
1493 if (numops > CEPH_OSD_MAX_OP)
1494 goto bad_put;
1495 if (numops != req->r_num_ops)
1496 goto bad_put;
1497 payload_len = 0;
1498 ceph_decode_need(&p, end, numops * sizeof(struct ceph_osd_op), bad);
1499 for (i = 0; i < numops; i++) {
1500 struct ceph_osd_op *op = p;
1501 int len;
1502
1503 len = le32_to_cpu(op->payload_len);
1504 req->r_reply_op_len[i] = len;
1505 dout(" op %d has %d bytes\n", i, len);
1506 payload_len += len;
1507 p += sizeof(*op);
1508 }
9fc6e064
AE		 1509 bytes = le32_to_cpu(msg->hdr.data_len);
1510 if (payload_len != bytes) {
1b83bef2 1511 pr_warning("sum of op payload lens %d != data_len %d",
9fc6e064 1512 payload_len, bytes);
1b83bef2
SW		 1513 goto bad_put;
1514 }
1515
1516 ceph_decode_need(&p, end, 4 + numops * 4, bad);
1517 retry_attempt = ceph_decode_32(&p);
1518 for (i = 0; i < numops; i++)
1519 req->r_reply_op_result[i] = ceph_decode_32(&p);
f24e9980 1520
eb845ff1
YZ		 1521 already_completed = req->r_got_reply;
1522
f24e9980 1523 if (!req->r_got_reply) {
f24e9980 1524
1b83bef2 1525 req->r_result = result;
f24e9980
SW		 1526 dout("handle_reply result %d bytes %d\n", req->r_result,
1527 bytes);
1528 if (req->r_result == 0)
1529 req->r_result = bytes;
1530
1531 /* in case this is a write and we need to replay, */
1b83bef2
SW		 1532 req->r_reassert_version.epoch = cpu_to_le32(reassert_epoch);
1533 req->r_reassert_version.version = cpu_to_le64(reassert_version);
f24e9980
SW		 1534
1535 req->r_got_reply = 1;
1536 } else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) {
1537 dout("handle_reply tid %llu dup ack\n", tid);
34b43a56 1538 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 1539 goto done;
1540 }
1541
1542 dout("handle_reply tid %llu flags %d\n", tid, flags);
1543
a40c4f10
YS		 1544 if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK))
1545 __register_linger_request(osdc, req);
1546
f24e9980 1547 /* either this is a read, or we got the safe response */
0ceed5db
SW		 1548 if (result < 0 ||
1549 (flags & CEPH_OSD_FLAG_ONDISK) ||
f24e9980
SW		 1550 ((flags & CEPH_OSD_FLAG_WRITE) == 0))
1551 __unregister_request(osdc, req);
1552
1553 mutex_unlock(&osdc->request_mutex);
1554
eb845ff1 1555 if (!already_completed) {
61c5d6bf
YZ		 1556 if (req->r_unsafe_callback &&
1557 result >= 0 && !(flags & CEPH_OSD_FLAG_ONDISK))
1558 req->r_unsafe_callback(req, true);
eb845ff1
YZ		 1559 if (req->r_callback)
1560 req->r_callback(req, msg);
1561 else
1562 complete_all(&req->r_completion);
1563 }
f24e9980 1564
61c5d6bf
YZ		 1565 if (flags & CEPH_OSD_FLAG_ONDISK) {
1566 if (req->r_unsafe_callback && already_completed)
1567 req->r_unsafe_callback(req, false);
25845472 1568 complete_request(req);
61c5d6bf 1569 }
f24e9980
SW		 1570
1571done:
a40c4f10 1572 dout("req=%p req->r_linger=%d\n", req, req->r_linger);
f24e9980
SW		 1573 ceph_osdc_put_request(req);
1574 return;
1575
1b83bef2
SW		 1576bad_put:
1577 ceph_osdc_put_request(req);
8058fd45
AE		 1578bad_mutex:
1579 mutex_unlock(&osdc->request_mutex);
f24e9980 1580bad:
1b83bef2
SW		 1581 pr_err("corrupt osd_op_reply got %d %d\n",
1582 (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len));
9ec7cab1 1583 ceph_msg_dump(msg);
f24e9980
SW		 1584}
1585
6f6c7006 1586static void reset_changed_osds(struct ceph_osd_client *osdc)
f24e9980 1587{
f24e9980 1588 struct rb_node *p, *n;
f24e9980 1589
6f6c7006
SW		 1590 for (p = rb_first(&osdc->osds); p; p = n) {
1591 struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
f24e9980 1592
6f6c7006
SW		 1593 n = rb_next(p);
1594 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
1595 memcmp(&osd->o_con.peer_addr,
1596 ceph_osd_addr(osdc->osdmap,
1597 osd->o_osd),
1598 sizeof(struct ceph_entity_addr)) != 0)
1599 __reset_osd(osdc, osd);
f24e9980 1600 }
422d2cb8
YS		 1601}
1602
1603/*
6f6c7006
SW		 1604 * Requeue requests whose mapping to an OSD has changed. If requests map to
1605 * no osd, request a new map.
422d2cb8 1606 *
e6d50f67 1607 * Caller should hold map_sem for read.
422d2cb8 1608 */
38d6453c 1609static void kick_requests(struct ceph_osd_client *osdc, int force_resend)
422d2cb8 1610{
a40c4f10 1611 struct ceph_osd_request *req, *nreq;
6f6c7006
SW		 1612 struct rb_node *p;
1613 int needmap = 0;
1614 int err;
422d2cb8 1615
38d6453c 1616 dout("kick_requests %s\n", force_resend ? " (force resend)" : "");
422d2cb8 1617 mutex_lock(&osdc->request_mutex);
6194ea89 1618 for (p = rb_first(&osdc->requests); p; ) {
6f6c7006 1619 req = rb_entry(p, struct ceph_osd_request, r_node);
6194ea89 1620 p = rb_next(p);
ab60b16d
AE		 1621
1622 /*
1623 * For linger requests that have not yet been
1624 * registered, move them to the linger list; they'll
1625 * be sent to the osd in the loop below. Unregister
1626 * the request before re-registering it as a linger
1627 * request to ensure the __map_request() below
1628 * will decide it needs to be sent.
1629 */
1630 if (req->r_linger && list_empty(&req->r_linger_item)) {
1631 dout("%p tid %llu restart on osd%d\n",
1632 req, req->r_tid,
1633 req->r_osd ? req->r_osd->o_osd : -1);
96e4dac6 1634 ceph_osdc_get_request(req);
ab60b16d
AE		 1635 __unregister_request(osdc, req);
1636 __register_linger_request(osdc, req);
96e4dac6 1637 ceph_osdc_put_request(req);
ab60b16d
AE		 1638 continue;
1639 }
1640
38d6453c 1641 err = __map_request(osdc, req, force_resend);
6f6c7006
SW		 1642 if (err < 0)
1643 continue; /* error */
1644 if (req->r_osd == NULL) {
1645 dout("%p tid %llu maps to no osd\n", req, req->r_tid);
1646 needmap++; /* request a newer map */
1647 } else if (err > 0) {
6194ea89
SW		 1648 if (!req->r_linger) {
1649 dout("%p tid %llu requeued on osd%d\n", req,
1650 req->r_tid,
1651 req->r_osd ? req->r_osd->o_osd : -1);
a40c4f10 1652 req->r_flags |= CEPH_OSD_FLAG_RETRY;
6194ea89
SW		 1653 }
1654 }
a40c4f10
YS		 1655 }
1656
1657 list_for_each_entry_safe(req, nreq, &osdc->req_linger,
1658 r_linger_item) {
1659 dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
1660
38d6453c 1661 err = __map_request(osdc, req, force_resend);
ab60b16d 1662 dout("__map_request returned %d\n", err);
a40c4f10
YS		 1663 if (err == 0)
1664 continue; /* no change and no osd was specified */
1665 if (err < 0)
1666 continue; /* hrm! */
1667 if (req->r_osd == NULL) {
1668 dout("tid %llu maps to no valid osd\n", req->r_tid);
1669 needmap++; /* request a newer map */
1670 continue;
6f6c7006 1671 }
a40c4f10
YS		 1672
1673 dout("kicking lingering %p tid %llu osd%d\n", req, req->r_tid,
1674 req->r_osd ? req->r_osd->o_osd : -1);
a40c4f10 1675 __register_request(osdc, req);
c89ce05e 1676 __unregister_linger_request(osdc, req);
6f6c7006 1677 }
14d2f38d 1678 reset_changed_osds(osdc);
f24e9980
SW		 1679 mutex_unlock(&osdc->request_mutex);
1680
1681 if (needmap) {
1682 dout("%d requests for down osds, need new map\n", needmap);
1683 ceph_monc_request_next_osdmap(&osdc->client->monc);
1684 }
422d2cb8 1685}
6f6c7006
SW		 1686
1687
f24e9980
SW		 1688/*
1689 * Process updated osd map.
1690 *
1691 * The message contains any number of incremental and full maps, normally
1692 * indicating some sort of topology change in the cluster. Kick requests
1693 * off to different OSDs as needed.
1694 */
1695void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1696{
1697 void *p, *end, *next;
1698 u32 nr_maps, maplen;
1699 u32 epoch;
1700 struct ceph_osdmap *newmap = NULL, *oldmap;
1701 int err;
1702 struct ceph_fsid fsid;
1703
1704 dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
1705 p = msg->front.iov_base;
1706 end = p + msg->front.iov_len;
1707
1708 /* verify fsid */
1709 ceph_decode_need(&p, end, sizeof(fsid), bad);
1710 ceph_decode_copy(&p, &fsid, sizeof(fsid));
0743304d
SW		 1711 if (ceph_check_fsid(osdc->client, &fsid) < 0)
1712 return;
f24e9980
SW		 1713
1714 down_write(&osdc->map_sem);
1715
1716 /* incremental maps */
1717 ceph_decode_32_safe(&p, end, nr_maps, bad);
1718 dout(" %d inc maps\n", nr_maps);
1719 while (nr_maps > 0) {
1720 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 1721 epoch = ceph_decode_32(&p);
1722 maplen = ceph_decode_32(&p);
f24e9980
SW		 1723 ceph_decode_need(&p, end, maplen, bad);
1724 next = p + maplen;
1725 if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
1726 dout("applying incremental map %u len %d\n",
1727 epoch, maplen);
1728 newmap = osdmap_apply_incremental(&p, next,
1729 osdc->osdmap,
15d9882c 1730 &osdc->client->msgr);
f24e9980
SW		 1731 if (IS_ERR(newmap)) {
1732 err = PTR_ERR(newmap);
1733 goto bad;
1734 }
30dc6381 1735 BUG_ON(!newmap);
f24e9980
SW		 1736 if (newmap != osdc->osdmap) {
1737 ceph_osdmap_destroy(osdc->osdmap);
1738 osdc->osdmap = newmap;
1739 }
38d6453c 1740 kick_requests(osdc, 0);
f24e9980
SW		 1741 } else {
1742 dout("ignoring incremental map %u len %d\n",
1743 epoch, maplen);
1744 }
1745 p = next;
1746 nr_maps--;
1747 }
1748 if (newmap)
1749 goto done;
1750
1751 /* full maps */
1752 ceph_decode_32_safe(&p, end, nr_maps, bad);
1753 dout(" %d full maps\n", nr_maps);
1754 while (nr_maps) {
1755 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 1756 epoch = ceph_decode_32(&p);
1757 maplen = ceph_decode_32(&p);
f24e9980
SW		 1758 ceph_decode_need(&p, end, maplen, bad);
1759 if (nr_maps > 1) {
1760 dout("skipping non-latest full map %u len %d\n",
1761 epoch, maplen);
1762 } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
1763 dout("skipping full map %u len %d, "
1764 "older than our %u\n", epoch, maplen,
1765 osdc->osdmap->epoch);
1766 } else {
38d6453c
SW		 1767 int skipped_map = 0;
1768
f24e9980
SW		 1769 dout("taking full map %u len %d\n", epoch, maplen);
1770 newmap = osdmap_decode(&p, p+maplen);
1771 if (IS_ERR(newmap)) {
1772 err = PTR_ERR(newmap);
1773 goto bad;
1774 }
30dc6381 1775 BUG_ON(!newmap);
f24e9980
SW		 1776 oldmap = osdc->osdmap;
1777 osdc->osdmap = newmap;
38d6453c
SW		 1778 if (oldmap) {
1779 if (oldmap->epoch + 1 < newmap->epoch)
1780 skipped_map = 1;
f24e9980 1781 ceph_osdmap_destroy(oldmap);
38d6453c
SW		 1782 }
1783 kick_requests(osdc, skipped_map);
f24e9980
SW		 1784 }
1785 p += maplen;
1786 nr_maps--;
1787 }
1788
1789done:
1790 downgrade_write(&osdc->map_sem);
1791 ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch);
cd634fb6
SW		 1792
1793 /*
1794 * subscribe to subsequent osdmap updates if full to ensure
1795 * we find out when we are no longer full and stop returning
1796 * ENOSPC.
1797 */
1798 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL))
1799 ceph_monc_request_next_osdmap(&osdc->client->monc);
1800
f9d25199
AE		 1801 mutex_lock(&osdc->request_mutex);
1802 __send_queued(osdc);
1803 mutex_unlock(&osdc->request_mutex);
f24e9980 1804 up_read(&osdc->map_sem);
03066f23 1805 wake_up_all(&osdc->client->auth_wq);
f24e9980
SW		 1806 return;
1807
1808bad:
1809 pr_err("osdc handle_map corrupt msg\n");
9ec7cab1 1810 ceph_msg_dump(msg);
f24e9980
SW		 1811 up_write(&osdc->map_sem);
1812 return;
1813}
1814
a40c4f10
YS		 1815/*
1816 * watch/notify callback event infrastructure
1817 *
1818 * These callbacks are used both for watch and notify operations.
1819 */
1820static void __release_event(struct kref *kref)
1821{
1822 struct ceph_osd_event *event =
1823 container_of(kref, struct ceph_osd_event, kref);
1824
1825 dout("__release_event %p\n", event);
1826 kfree(event);
1827}
1828
1829static void get_event(struct ceph_osd_event *event)
1830{
1831 kref_get(&event->kref);
1832}
1833
1834void ceph_osdc_put_event(struct ceph_osd_event *event)
1835{
1836 kref_put(&event->kref, __release_event);
1837}
1838EXPORT_SYMBOL(ceph_osdc_put_event);
1839
1840static void __insert_event(struct ceph_osd_client *osdc,
1841 struct ceph_osd_event *new)
1842{
1843 struct rb_node **p = &osdc->event_tree.rb_node;
1844 struct rb_node *parent = NULL;
1845 struct ceph_osd_event *event = NULL;
1846
1847 while (*p) {
1848 parent = *p;
1849 event = rb_entry(parent, struct ceph_osd_event, node);
1850 if (new->cookie < event->cookie)
1851 p = &(*p)->rb_left;
1852 else if (new->cookie > event->cookie)
1853 p = &(*p)->rb_right;
1854 else
1855 BUG();
1856 }
1857
1858 rb_link_node(&new->node, parent, p);
1859 rb_insert_color(&new->node, &osdc->event_tree);
1860}
1861
1862static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
1863 u64 cookie)
1864{
1865 struct rb_node **p = &osdc->event_tree.rb_node;
1866 struct rb_node *parent = NULL;
1867 struct ceph_osd_event *event = NULL;
1868
1869 while (*p) {
1870 parent = *p;
1871 event = rb_entry(parent, struct ceph_osd_event, node);
1872 if (cookie < event->cookie)
1873 p = &(*p)->rb_left;
1874 else if (cookie > event->cookie)
1875 p = &(*p)->rb_right;
1876 else
1877 return event;
1878 }
1879 return NULL;
1880}
1881
1882static void __remove_event(struct ceph_osd_event *event)
1883{
1884 struct ceph_osd_client *osdc = event->osdc;
1885
1886 if (!RB_EMPTY_NODE(&event->node)) {
1887 dout("__remove_event removed %p\n", event);
1888 rb_erase(&event->node, &osdc->event_tree);
1889 ceph_osdc_put_event(event);
1890 } else {
1891 dout("__remove_event didn't remove %p\n", event);
1892 }
1893}
1894
1895int ceph_osdc_create_event(struct ceph_osd_client *osdc,
1896 void (*event_cb)(u64, u64, u8, void *),
3c663bbd 1897 void *data, struct ceph_osd_event **pevent)
a40c4f10
YS		 1898{
1899 struct ceph_osd_event *event;
1900
1901 event = kmalloc(sizeof(*event), GFP_NOIO);
1902 if (!event)
1903 return -ENOMEM;
1904
1905 dout("create_event %p\n", event);
1906 event->cb = event_cb;
3c663bbd 1907 event->one_shot = 0;
a40c4f10
YS		 1908 event->data = data;
1909 event->osdc = osdc;
1910 INIT_LIST_HEAD(&event->osd_node);
3ee5234d 1911 RB_CLEAR_NODE(&event->node);
a40c4f10
YS		 1912 kref_init(&event->kref); /* one ref for us */
1913 kref_get(&event->kref); /* one ref for the caller */
a40c4f10
YS		 1914
1915 spin_lock(&osdc->event_lock);
1916 event->cookie = ++osdc->event_count;
1917 __insert_event(osdc, event);
1918 spin_unlock(&osdc->event_lock);
1919
1920 *pevent = event;
1921 return 0;
1922}
1923EXPORT_SYMBOL(ceph_osdc_create_event);
1924
1925void ceph_osdc_cancel_event(struct ceph_osd_event *event)
1926{
1927 struct ceph_osd_client *osdc = event->osdc;
1928
1929 dout("cancel_event %p\n", event);
1930 spin_lock(&osdc->event_lock);
1931 __remove_event(event);
1932 spin_unlock(&osdc->event_lock);
1933 ceph_osdc_put_event(event); /* caller's */
1934}
1935EXPORT_SYMBOL(ceph_osdc_cancel_event);
1936
1937
1938static void do_event_work(struct work_struct *work)
1939{
1940 struct ceph_osd_event_work *event_work =
1941 container_of(work, struct ceph_osd_event_work, work);
1942 struct ceph_osd_event *event = event_work->event;
1943 u64 ver = event_work->ver;
1944 u64 notify_id = event_work->notify_id;
1945 u8 opcode = event_work->opcode;
1946
1947 dout("do_event_work completing %p\n", event);
1948 event->cb(ver, notify_id, opcode, event->data);
a40c4f10
YS		 1949 dout("do_event_work completed %p\n", event);
1950 ceph_osdc_put_event(event);
1951 kfree(event_work);
1952}
1953
1954
1955/*
1956 * Process osd watch notifications
1957 */
3c663bbd
AE		 1958static void handle_watch_notify(struct ceph_osd_client *osdc,
1959 struct ceph_msg *msg)
a40c4f10
YS		 1960{
1961 void *p, *end;
1962 u8 proto_ver;
1963 u64 cookie, ver, notify_id;
1964 u8 opcode;
1965 struct ceph_osd_event *event;
1966 struct ceph_osd_event_work *event_work;
1967
1968 p = msg->front.iov_base;
1969 end = p + msg->front.iov_len;
1970
1971 ceph_decode_8_safe(&p, end, proto_ver, bad);
1972 ceph_decode_8_safe(&p, end, opcode, bad);
1973 ceph_decode_64_safe(&p, end, cookie, bad);
1974 ceph_decode_64_safe(&p, end, ver, bad);
1975 ceph_decode_64_safe(&p, end, notify_id, bad);
1976
1977 spin_lock(&osdc->event_lock);
1978 event = __find_event(osdc, cookie);
1979 if (event) {
3c663bbd 1980 BUG_ON(event->one_shot);
a40c4f10 1981 get_event(event);
a40c4f10
YS		 1982 }
1983 spin_unlock(&osdc->event_lock);
1984 dout("handle_watch_notify cookie %lld ver %lld event %p\n",
1985 cookie, ver, event);
1986 if (event) {
1987 event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
a40c4f10
YS		 1988 if (!event_work) {
1989 dout("ERROR: could not allocate event_work\n");
1990 goto done_err;
1991 }
6b0ae409 1992 INIT_WORK(&event_work->work, do_event_work);
a40c4f10
YS		 1993 event_work->event = event;
1994 event_work->ver = ver;
1995 event_work->notify_id = notify_id;
1996 event_work->opcode = opcode;
1997 if (!queue_work(osdc->notify_wq, &event_work->work)) {
1998 dout("WARNING: failed to queue notify event work\n");
1999 goto done_err;
2000 }
2001 }
2002
2003 return;
2004
2005done_err:
a40c4f10
YS		 2006 ceph_osdc_put_event(event);
2007 return;
2008
2009bad:
2010 pr_err("osdc handle_watch_notify corrupt msg\n");
2011 return;
2012}
2013
e65550fd
AE		 2014/*
2015 * build new request AND message
2016 *
2017 */
2018void ceph_osdc_build_request(struct ceph_osd_request *req, u64 off,
2019 struct ceph_snap_context *snapc, u64 snap_id,
2020 struct timespec *mtime)
2021{
2022 struct ceph_msg *msg = req->r_request;
2023 void *p;
2024 size_t msg_size;
2025 int flags = req->r_flags;
2026 u64 data_len;
2027 unsigned int i;
2028
2029 req->r_snapid = snap_id;
2030 req->r_snapc = ceph_get_snap_context(snapc);
2031
2032 /* encode request */
2033 msg->hdr.version = cpu_to_le16(4);
2034
2035 p = msg->front.iov_base;
2036 ceph_encode_32(&p, 1); /* client_inc is always 1 */
2037 req->r_request_osdmap_epoch = p;
2038 p += 4;
2039 req->r_request_flags = p;
2040 p += 4;
2041 if (req->r_flags & CEPH_OSD_FLAG_WRITE)
2042 ceph_encode_timespec(p, mtime);
2043 p += sizeof(struct ceph_timespec);
2044 req->r_request_reassert_version = p;
2045 p += sizeof(struct ceph_eversion); /* will get filled in */
2046
2047 /* oloc */
2048 ceph_encode_8(&p, 4);
2049 ceph_encode_8(&p, 4);
2050 ceph_encode_32(&p, 8 + 4 + 4);
2051 req->r_request_pool = p;
2052 p += 8;
2053 ceph_encode_32(&p, -1); /* preferred */
2054 ceph_encode_32(&p, 0); /* key len */
2055
2056 ceph_encode_8(&p, 1);
2057 req->r_request_pgid = p;
2058 p += 8 + 4;
2059 ceph_encode_32(&p, -1); /* preferred */
2060
2061 /* oid */
2062 ceph_encode_32(&p, req->r_oid_len);
2063 memcpy(p, req->r_oid, req->r_oid_len);
2064 dout("oid '%.*s' len %d\n", req->r_oid_len, req->r_oid, req->r_oid_len);
2065 p += req->r_oid_len;
2066
2067 /* ops--can imply data */
2068 ceph_encode_16(&p, (u16)req->r_num_ops);
2069 data_len = 0;
2070 for (i = 0; i < req->r_num_ops; i++) {
2071 data_len += osd_req_encode_op(req, p, i);
2072 p += sizeof(struct ceph_osd_op);
2073 }
2074
2075 /* snaps */
2076 ceph_encode_64(&p, req->r_snapid);
2077 ceph_encode_64(&p, req->r_snapc ? req->r_snapc->seq : 0);
2078 ceph_encode_32(&p, req->r_snapc ? req->r_snapc->num_snaps : 0);
2079 if (req->r_snapc) {
2080 for (i = 0; i < snapc->num_snaps; i++) {
2081 ceph_encode_64(&p, req->r_snapc->snaps[i]);
2082 }
2083 }
2084
2085 req->r_request_attempts = p;
2086 p += 4;
2087
2088 /* data */
2089 if (flags & CEPH_OSD_FLAG_WRITE) {
2090 u16 data_off;
2091
2092 /*
2093 * The header "data_off" is a hint to the receiver
2094 * allowing it to align received data into its
2095 * buffers such that there's no need to re-copy
2096 * it before writing it to disk (direct I/O).
2097 */
2098 data_off = (u16) (off & 0xffff);
2099 req->r_request->hdr.data_off = cpu_to_le16(data_off);
2100 }
2101 req->r_request->hdr.data_len = cpu_to_le32(data_len);
2102
2103 BUG_ON(p > msg->front.iov_base + msg->front.iov_len);
2104 msg_size = p - msg->front.iov_base;
2105 msg->front.iov_len = msg_size;
2106 msg->hdr.front_len = cpu_to_le32(msg_size);
2107
2108 dout("build_request msg_size was %d\n", (int)msg_size);
2109}
2110EXPORT_SYMBOL(ceph_osdc_build_request);
2111
70636773
AE		 2112/*
2113 * Register request, send initial attempt.
2114 */
2115int ceph_osdc_start_request(struct ceph_osd_client *osdc,
2116 struct ceph_osd_request *req,
2117 bool nofail)
2118{
2119 int rc = 0;
2120
f24e9980
SW		 2121 down_read(&osdc->map_sem);
2122 mutex_lock(&osdc->request_mutex);
dc4b870c 2123 __register_request(osdc, req);
c10ebbf5
AE		 2124 req->r_sent = 0;
2125 req->r_got_reply = 0;
92451b49
AE		 2126 rc = __map_request(osdc, req, 0);
2127 if (rc < 0) {
2128 if (nofail) {
2129 dout("osdc_start_request failed map, "
2130 " will retry %lld\n", req->r_tid);
2131 rc = 0;
73d9f7ee 2132 } else {
2133 __unregister_request(osdc, req);
f24e9980 2134 }
92451b49 2135 goto out_unlock;
f24e9980 2136 }
92451b49
AE		 2137 if (req->r_osd == NULL) {
2138 dout("send_request %p no up osds in pg\n", req);
2139 ceph_monc_request_next_osdmap(&osdc->client->monc);
2140 } else {
7e2766a1 2141 __send_queued(osdc);
92451b49
AE		 2142 }
2143 rc = 0;
234af26f 2144out_unlock:
f24e9980
SW		 2145 mutex_unlock(&osdc->request_mutex);
2146 up_read(&osdc->map_sem);
2147 return rc;
2148}
3d14c5d2 2149EXPORT_SYMBOL(ceph_osdc_start_request);
f24e9980
SW		 2150
2151/*
2152 * wait for a request to complete
2153 */
2154int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
2155 struct ceph_osd_request *req)
2156{
2157 int rc;
2158
2159 rc = wait_for_completion_interruptible(&req->r_completion);
2160 if (rc < 0) {
2161 mutex_lock(&osdc->request_mutex);
2162 __cancel_request(req);
529cfcc4 2163 __unregister_request(osdc, req);
f24e9980 2164 mutex_unlock(&osdc->request_mutex);
25845472 2165 complete_request(req);
529cfcc4 2166 dout("wait_request tid %llu canceled/timed out\n", req->r_tid);
f24e9980
SW		 2167 return rc;
2168 }
2169
2170 dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result);
2171 return req->r_result;
2172}
3d14c5d2 2173EXPORT_SYMBOL(ceph_osdc_wait_request);
f24e9980
SW		 2174
2175/*
2176 * sync - wait for all in-flight requests to flush. avoid starvation.
2177 */
2178void ceph_osdc_sync(struct ceph_osd_client *osdc)
2179{
2180 struct ceph_osd_request *req;
2181 u64 last_tid, next_tid = 0;
2182
2183 mutex_lock(&osdc->request_mutex);
2184 last_tid = osdc->last_tid;
2185 while (1) {
2186 req = __lookup_request_ge(osdc, next_tid);
2187 if (!req)
2188 break;
2189 if (req->r_tid > last_tid)
2190 break;
2191
2192 next_tid = req->r_tid + 1;
2193 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
2194 continue;
2195
2196 ceph_osdc_get_request(req);
2197 mutex_unlock(&osdc->request_mutex);
2198 dout("sync waiting on tid %llu (last is %llu)\n",
2199 req->r_tid, last_tid);
2200 wait_for_completion(&req->r_safe_completion);
2201 mutex_lock(&osdc->request_mutex);
2202 ceph_osdc_put_request(req);
2203 }
2204 mutex_unlock(&osdc->request_mutex);
2205 dout("sync done (thru tid %llu)\n", last_tid);
2206}
3d14c5d2 2207EXPORT_SYMBOL(ceph_osdc_sync);
f24e9980
SW		 2208
2209/*
2210 * init, shutdown
2211 */
2212int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
2213{
2214 int err;
2215
2216 dout("init\n");
2217 osdc->client = client;
2218 osdc->osdmap = NULL;
2219 init_rwsem(&osdc->map_sem);
2220 init_completion(&osdc->map_waiters);
2221 osdc->last_requested_map = 0;
2222 mutex_init(&osdc->request_mutex);
f24e9980
SW		 2223 osdc->last_tid = 0;
2224 osdc->osds = RB_ROOT;
f5a2041b 2225 INIT_LIST_HEAD(&osdc->osd_lru);
f24e9980 2226 osdc->requests = RB_ROOT;
422d2cb8 2227 INIT_LIST_HEAD(&osdc->req_lru);
6f6c7006
SW		 2228 INIT_LIST_HEAD(&osdc->req_unsent);
2229 INIT_LIST_HEAD(&osdc->req_notarget);
a40c4f10 2230 INIT_LIST_HEAD(&osdc->req_linger);
f24e9980
SW		 2231 osdc->num_requests = 0;
2232 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
f5a2041b 2233 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
a40c4f10
YS		 2234 spin_lock_init(&osdc->event_lock);
2235 osdc->event_tree = RB_ROOT;
2236 osdc->event_count = 0;
f5a2041b
YS		 2237
2238 schedule_delayed_work(&osdc->osds_timeout_work,
3d14c5d2 2239 round_jiffies_relative(osdc->client->options->osd_idle_ttl * HZ));
f24e9980 2240
5f44f142 2241 err = -ENOMEM;
f24e9980
SW		 2242 osdc->req_mempool = mempool_create_kmalloc_pool(10,
2243 sizeof(struct ceph_osd_request));
2244 if (!osdc->req_mempool)
5f44f142 2245 goto out;
f24e9980 2246
d50b409f
SW		 2247 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
2248 OSD_OP_FRONT_LEN, 10, true,
4f48280e 2249 "osd_op");
f24e9980 2250 if (err < 0)
5f44f142 2251 goto out_mempool;
d50b409f 2252 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
4f48280e
SW		 2253 OSD_OPREPLY_FRONT_LEN, 10, true,
2254 "osd_op_reply");
c16e7869
SW		 2255 if (err < 0)
2256 goto out_msgpool;
a40c4f10
YS		 2257
2258 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
2259 if (IS_ERR(osdc->notify_wq)) {
2260 err = PTR_ERR(osdc->notify_wq);
2261 osdc->notify_wq = NULL;
2262 goto out_msgpool;
2263 }
f24e9980 2264 return 0;
5f44f142 2265
c16e7869
SW		 2266out_msgpool:
2267 ceph_msgpool_destroy(&osdc->msgpool_op);
5f44f142
SW		 2268out_mempool:
2269 mempool_destroy(osdc->req_mempool);
2270out:
2271 return err;
f24e9980
SW		 2272}
2273
2274void ceph_osdc_stop(struct ceph_osd_client *osdc)
2275{
a40c4f10
YS		 2276 flush_workqueue(osdc->notify_wq);
2277 destroy_workqueue(osdc->notify_wq);
f24e9980 2278 cancel_delayed_work_sync(&osdc->timeout_work);
f5a2041b 2279 cancel_delayed_work_sync(&osdc->osds_timeout_work);
f24e9980
SW		 2280 if (osdc->osdmap) {
2281 ceph_osdmap_destroy(osdc->osdmap);
2282 osdc->osdmap = NULL;
2283 }
aca420bc 2284 remove_all_osds(osdc);
f24e9980
SW		 2285 mempool_destroy(osdc->req_mempool);
2286 ceph_msgpool_destroy(&osdc->msgpool_op);
c16e7869 2287 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
f24e9980
SW		 2288}
2289
2290/*
2291 * Read some contiguous pages. If we cross a stripe boundary, shorten
2292 * *plen. Return number of bytes read, or error.
2293 */
2294int ceph_osdc_readpages(struct ceph_osd_client *osdc,
2295 struct ceph_vino vino, struct ceph_file_layout *layout,
2296 u64 off, u64 *plen,
2297 u32 truncate_seq, u64 truncate_size,
b7495fc2 2298 struct page **pages, int num_pages, int page_align)
f24e9980
SW		 2299{
2300 struct ceph_osd_request *req;
2301 int rc = 0;
2302
2303 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
2304 vino.snap, off, *plen);
79528734 2305 req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 1,
f24e9980 2306 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
acead002 2307 NULL, truncate_seq, truncate_size,
153e5167 2308 false);
6816282d
SW		 2309 if (IS_ERR(req))
2310 return PTR_ERR(req);
f24e9980
SW		 2311
2312 /* it may be a short read due to an object boundary */
0fff87ec 2313
406e2c9f 2314 osd_req_op_extent_osd_data_pages(req, 0,
a4ce40a9 2315 pages, *plen, page_align, false, false);
f24e9980 2316
e0c59487 2317 dout("readpages final extent is %llu~%llu (%llu bytes align %d)\n",
43bfe5de 2318 off, *plen, *plen, page_align);
f24e9980 2319
79528734 2320 ceph_osdc_build_request(req, off, NULL, vino.snap, NULL);
02ee07d3 2321
f24e9980
SW		 2322 rc = ceph_osdc_start_request(osdc, req, false);
2323 if (!rc)
2324 rc = ceph_osdc_wait_request(osdc, req);
2325
2326 ceph_osdc_put_request(req);
2327 dout("readpages result %d\n", rc);
2328 return rc;
2329}
3d14c5d2 2330EXPORT_SYMBOL(ceph_osdc_readpages);
f24e9980
SW		 2331
2332/*
2333 * do a synchronous write on N pages
2334 */
2335int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
2336 struct ceph_file_layout *layout,
2337 struct ceph_snap_context *snapc,
2338 u64 off, u64 len,
2339 u32 truncate_seq, u64 truncate_size,
2340 struct timespec *mtime,
24808826 2341 struct page **pages, int num_pages)
f24e9980
SW		 2342{
2343 struct ceph_osd_request *req;
2344 int rc = 0;
b7495fc2 2345 int page_align = off & ~PAGE_MASK;
f24e9980 2346
acead002 2347 BUG_ON(vino.snap != CEPH_NOSNAP); /* snapshots aren't writeable */
79528734 2348 req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 1,
f24e9980 2349 CEPH_OSD_OP_WRITE,
24808826 2350 CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
acead002 2351 snapc, truncate_seq, truncate_size,
153e5167 2352 true);
6816282d
SW		 2353 if (IS_ERR(req))
2354 return PTR_ERR(req);
f24e9980
SW		 2355
2356 /* it may be a short write due to an object boundary */
406e2c9f 2357 osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align,
43bfe5de
AE		 2358 false, false);
2359 dout("writepages %llu~%llu (%llu bytes)\n", off, len, len);
f24e9980 2360
79528734 2361 ceph_osdc_build_request(req, off, snapc, CEPH_NOSNAP, mtime);
02ee07d3 2362
87f979d3 2363 rc = ceph_osdc_start_request(osdc, req, true);
f24e9980
SW		 2364 if (!rc)
2365 rc = ceph_osdc_wait_request(osdc, req);
2366
2367 ceph_osdc_put_request(req);
2368 if (rc == 0)
2369 rc = len;
2370 dout("writepages result %d\n", rc);
2371 return rc;
2372}
3d14c5d2 2373EXPORT_SYMBOL(ceph_osdc_writepages);
f24e9980 2374
5522ae0b
AE		 2375int ceph_osdc_setup(void)
2376{
2377 BUG_ON(ceph_osd_request_cache);
2378 ceph_osd_request_cache = kmem_cache_create("ceph_osd_request",
2379 sizeof (struct ceph_osd_request),
2380 __alignof__(struct ceph_osd_request),
2381 0, NULL);
2382
2383 return ceph_osd_request_cache ? 0 : -ENOMEM;
2384}
2385EXPORT_SYMBOL(ceph_osdc_setup);
2386
2387void ceph_osdc_cleanup(void)
2388{
2389 BUG_ON(!ceph_osd_request_cache);
2390 kmem_cache_destroy(ceph_osd_request_cache);
2391 ceph_osd_request_cache = NULL;
2392}
2393EXPORT_SYMBOL(ceph_osdc_cleanup);
2394
f24e9980
SW		 2395/*
2396 * handle incoming message
2397 */
2398static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
2399{
2400 struct ceph_osd *osd = con->private;
32c895e7 2401 struct ceph_osd_client *osdc;
f24e9980
SW		 2402 int type = le16_to_cpu(msg->hdr.type);
2403
2404 if (!osd)
4a32f93d 2405 goto out;
32c895e7 2406 osdc = osd->o_osdc;
f24e9980
SW		 2407
2408 switch (type) {
2409 case CEPH_MSG_OSD_MAP:
2410 ceph_osdc_handle_map(osdc, msg);
2411 break;
2412 case CEPH_MSG_OSD_OPREPLY:
350b1c32 2413 handle_reply(osdc, msg, con);
f24e9980 2414 break;
a40c4f10
YS		 2415 case CEPH_MSG_WATCH_NOTIFY:
2416 handle_watch_notify(osdc, msg);
2417 break;
f24e9980
SW		 2418
2419 default:
2420 pr_err("received unknown message type %d %s\n", type,
2421 ceph_msg_type_name(type));
2422 }
4a32f93d 2423out:
f24e9980
SW		 2424 ceph_msg_put(msg);
2425}
2426
5b3a4db3 2427/*
21b667f6
SW		 2428 * lookup and return message for incoming reply. set up reply message
2429 * pages.
5b3a4db3
SW		 2430 */
2431static struct ceph_msg *get_reply(struct ceph_connection *con,
2450418c
YS		 2432 struct ceph_msg_header *hdr,
2433 int *skip)
f24e9980
SW		 2434{
2435 struct ceph_osd *osd = con->private;
2436 struct ceph_osd_client *osdc = osd->o_osdc;
2450418c 2437 struct ceph_msg *m;
0547a9b3 2438 struct ceph_osd_request *req;
5b3a4db3
SW		 2439 int front = le32_to_cpu(hdr->front_len);
2440 int data_len = le32_to_cpu(hdr->data_len);
0547a9b3 2441 u64 tid;
f24e9980 2442
0547a9b3
YS		 2443 tid = le64_to_cpu(hdr->tid);
2444 mutex_lock(&osdc->request_mutex);
2445 req = __lookup_request(osdc, tid);
2446 if (!req) {
2447 *skip = 1;
2448 m = NULL;
756a16a5
SW		 2449 dout("get_reply unknown tid %llu from osd%d\n", tid,
2450 osd->o_osd);
0547a9b3
YS		 2451 goto out;
2452 }
c16e7869 2453
ace6d3a9 2454 if (req->r_reply->con)
8921d114 2455 dout("%s revoking msg %p from old con %p\n", __func__,
ace6d3a9
AE		 2456 req->r_reply, req->r_reply->con);
2457 ceph_msg_revoke_incoming(req->r_reply);
0547a9b3 2458
c16e7869 2459 if (front > req->r_reply->front.iov_len) {
4974341e
AE		 2460 pr_warning("get_reply front %d > preallocated %d (%u#%llu)\n",
2461 front, (int)req->r_reply->front.iov_len,
2462 (unsigned int)con->peer_name.type,
2463 le64_to_cpu(con->peer_name.num));
b61c2763 2464 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front, GFP_NOFS, false);
a79832f2 2465 if (!m)
c16e7869
SW		 2466 goto out;
2467 ceph_msg_put(req->r_reply);
2468 req->r_reply = m;
2469 }
2470 m = ceph_msg_get(req->r_reply);
2471
0547a9b3 2472 if (data_len > 0) {
a4ce40a9 2473 struct ceph_osd_data *osd_data;
0fff87ec 2474
a4ce40a9
AE		 2475 /*
2476 * XXX This is assuming there is only one op containing
2477 * XXX page data. Probably OK for reads, but this
2478 * XXX ought to be done more generally.
2479 */
406e2c9f 2480 osd_data = osd_req_op_extent_osd_data(req, 0);
0fff87ec 2481 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
0fff87ec 2482 if (osd_data->pages &&
e0c59487 2483 unlikely(osd_data->length < data_len)) {
2ac2b7a6 2484
e0c59487
AE		 2485 pr_warning("tid %lld reply has %d bytes "
2486 "we had only %llu bytes ready\n",
2487 tid, data_len, osd_data->length);
2ac2b7a6
AE		 2488 *skip = 1;
2489 ceph_msg_put(m);
2490 m = NULL;
2491 goto out;
2492 }
2ac2b7a6 2493 }
0547a9b3 2494 }
5b3a4db3 2495 *skip = 0;
c16e7869 2496 dout("get_reply tid %lld %p\n", tid, m);
0547a9b3
YS		 2497
2498out:
2499 mutex_unlock(&osdc->request_mutex);
2450418c 2500 return m;
5b3a4db3
SW		 2501
2502}
2503
2504static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2505 struct ceph_msg_header *hdr,
2506 int *skip)
2507{
2508 struct ceph_osd *osd = con->private;
2509 int type = le16_to_cpu(hdr->type);
2510 int front = le32_to_cpu(hdr->front_len);
2511
1c20f2d2 2512 *skip = 0;
5b3a4db3
SW		 2513 switch (type) {
2514 case CEPH_MSG_OSD_MAP:
a40c4f10 2515 case CEPH_MSG_WATCH_NOTIFY:
b61c2763 2516 return ceph_msg_new(type, front, GFP_NOFS, false);
5b3a4db3
SW		 2517 case CEPH_MSG_OSD_OPREPLY:
2518 return get_reply(con, hdr, skip);
2519 default:
2520 pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
2521 osd->o_osd);
2522 *skip = 1;
2523 return NULL;
2524 }
f24e9980
SW		 2525}
2526
2527/*
2528 * Wrappers to refcount containing ceph_osd struct
2529 */
2530static struct ceph_connection *get_osd_con(struct ceph_connection *con)
2531{
2532 struct ceph_osd *osd = con->private;
2533 if (get_osd(osd))
2534 return con;
2535 return NULL;
2536}
2537
2538static void put_osd_con(struct ceph_connection *con)
2539{
2540 struct ceph_osd *osd = con->private;
2541 put_osd(osd);
2542}
2543
4e7a5dcd
SW		 2544/*
2545 * authentication
2546 */
a3530df3
AE		 2547/*
2548 * Note: returned pointer is the address of a structure that's
2549 * managed separately. Caller must *not* attempt to free it.
2550 */
2551static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
8f43fb53 2552 int *proto, int force_new)
4e7a5dcd
SW		 2553{
2554 struct ceph_osd *o = con->private;
2555 struct ceph_osd_client *osdc = o->o_osdc;
2556 struct ceph_auth_client *ac = osdc->client->monc.auth;
74f1869f 2557 struct ceph_auth_handshake *auth = &o->o_auth;
4e7a5dcd 2558
74f1869f 2559 if (force_new && auth->authorizer) {
27859f97 2560 ceph_auth_destroy_authorizer(ac, auth->authorizer);
74f1869f
AE		 2561 auth->authorizer = NULL;
2562 }
27859f97
SW		 2563 if (!auth->authorizer) {
2564 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2565 auth);
4e7a5dcd 2566 if (ret)
a3530df3 2567 return ERR_PTR(ret);
27859f97
SW		 2568 } else {
2569 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
0bed9b5c
SW		 2570 auth);
2571 if (ret)
2572 return ERR_PTR(ret);
4e7a5dcd 2573 }
4e7a5dcd 2574 *proto = ac->protocol;
74f1869f 2575
a3530df3 2576 return auth;
4e7a5dcd
SW		 2577}
2578
2579
2580static int verify_authorizer_reply(struct ceph_connection *con, int len)
2581{
2582 struct ceph_osd *o = con->private;
2583 struct ceph_osd_client *osdc = o->o_osdc;
2584 struct ceph_auth_client *ac = osdc->client->monc.auth;
2585
27859f97 2586 return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer, len);
4e7a5dcd
SW		 2587}
2588
9bd2e6f8
SW		 2589static int invalidate_authorizer(struct ceph_connection *con)
2590{
2591 struct ceph_osd *o = con->private;
2592 struct ceph_osd_client *osdc = o->o_osdc;
2593 struct ceph_auth_client *ac = osdc->client->monc.auth;
2594
27859f97 2595 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
9bd2e6f8
SW		 2596 return ceph_monc_validate_auth(&osdc->client->monc);
2597}
4e7a5dcd 2598
9e32789f 2599static const struct ceph_connection_operations osd_con_ops = {
f24e9980
SW		 2600 .get = get_osd_con,
2601 .put = put_osd_con,
2602 .dispatch = dispatch,
4e7a5dcd
SW		 2603 .get_authorizer = get_authorizer,
2604 .verify_authorizer_reply = verify_authorizer_reply,
9bd2e6f8 2605 .invalidate_authorizer = invalidate_authorizer,
f24e9980 2606 .alloc_msg = alloc_msg,
81b024e7 2607 .fault = osd_reset,
f24e9980 2608};
Linux 6.x block layer and io_uring tree(s)