projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
libceph: block I/O when PAUSE or FULL osd map flags are set
[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
ad7a60de
LW		 506 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
507 opcode != CEPH_OSD_OP_DELETE && opcode != CEPH_OSD_OP_ZERO &&
508 opcode != CEPH_OSD_OP_TRUNCATE);
33803f33 509
33803f33
AE		 510 op->extent.offset = offset;
511 op->extent.length = length;
512 op->extent.truncate_size = truncate_size;
513 op->extent.truncate_seq = truncate_seq;
514 if (opcode == CEPH_OSD_OP_WRITE)
515 payload_len += length;
516
517 op->payload_len = payload_len;
518}
519EXPORT_SYMBOL(osd_req_op_extent_init);
520
c99d2d4a
AE		 521void osd_req_op_extent_update(struct ceph_osd_request *osd_req,
522 unsigned int which, u64 length)
e5975c7c 523{
c99d2d4a
AE		 524 struct ceph_osd_req_op *op;
525 u64 previous;
526
527 BUG_ON(which >= osd_req->r_num_ops);
528 op = &osd_req->r_ops[which];
529 previous = op->extent.length;
e5975c7c
AE		 530
531 if (length == previous)
532 return; /* Nothing to do */
533 BUG_ON(length > previous);
534
535 op->extent.length = length;
536 op->payload_len -= previous - length;
537}
538EXPORT_SYMBOL(osd_req_op_extent_update);
539
c99d2d4a 540void osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
04017e29 541 u16 opcode, const char *class, const char *method)
33803f33 542{
49719778 543 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
5f562df5 544 struct ceph_pagelist *pagelist;
33803f33
AE		 545 size_t payload_len = 0;
546 size_t size;
547
548 BUG_ON(opcode != CEPH_OSD_OP_CALL);
549
5f562df5
AE		 550 pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
551 BUG_ON(!pagelist);
552 ceph_pagelist_init(pagelist);
553
33803f33
AE		 554 op->cls.class_name = class;
555 size = strlen(class);
556 BUG_ON(size > (size_t) U8_MAX);
557 op->cls.class_len = size;
5f562df5 558 ceph_pagelist_append(pagelist, class, size);
33803f33
AE		 559 payload_len += size;
560
561 op->cls.method_name = method;
562 size = strlen(method);
563 BUG_ON(size > (size_t) U8_MAX);
564 op->cls.method_len = size;
5f562df5 565 ceph_pagelist_append(pagelist, method, size);
33803f33
AE		 566 payload_len += size;
567
a4ce40a9 568 osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist);
5f562df5 569
33803f33
AE		 570 op->cls.argc = 0; /* currently unused */
571
572 op->payload_len = payload_len;
573}
574EXPORT_SYMBOL(osd_req_op_cls_init);
8c042b0d 575
c99d2d4a
AE		 576void osd_req_op_watch_init(struct ceph_osd_request *osd_req,
577 unsigned int which, u16 opcode,
33803f33
AE		 578 u64 cookie, u64 version, int flag)
579{
49719778 580 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
33803f33 581
c99d2d4a 582 BUG_ON(opcode != CEPH_OSD_OP_NOTIFY_ACK && opcode != CEPH_OSD_OP_WATCH);
33803f33
AE		 583
584 op->watch.cookie = cookie;
9ef1ee5a 585 op->watch.ver = version;
33803f33 586 if (opcode == CEPH_OSD_OP_WATCH && flag)
c99d2d4a 587 op->watch.flag = (u8)1;
33803f33
AE		 588}
589EXPORT_SYMBOL(osd_req_op_watch_init);
590
90af3602 591static void ceph_osdc_msg_data_add(struct ceph_msg *msg,
ec9123c5
AE		 592 struct ceph_osd_data *osd_data)
593{
594 u64 length = ceph_osd_data_length(osd_data);
595
596 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
597 BUG_ON(length > (u64) SIZE_MAX);
598 if (length)
90af3602 599 ceph_msg_data_add_pages(msg, osd_data->pages,
ec9123c5
AE		 600 length, osd_data->alignment);
601 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
602 BUG_ON(!length);
90af3602 603 ceph_msg_data_add_pagelist(msg, osd_data->pagelist);
ec9123c5
AE		 604#ifdef CONFIG_BLOCK
605 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
90af3602 606 ceph_msg_data_add_bio(msg, osd_data->bio, length);
ec9123c5
AE		 607#endif
608 } else {
609 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
610 }
611}
612
a8dd0a37 613static u64 osd_req_encode_op(struct ceph_osd_request *req,
79528734 614 struct ceph_osd_op *dst, unsigned int which)
a8dd0a37 615{
79528734 616 struct ceph_osd_req_op *src;
04017e29 617 struct ceph_osd_data *osd_data;
54d50649 618 u64 request_data_len = 0;
04017e29 619 u64 data_length;
a8dd0a37 620
79528734
AE		 621 BUG_ON(which >= req->r_num_ops);
622 src = &req->r_ops[which];
a8dd0a37
AE		 623 if (WARN_ON(!osd_req_opcode_valid(src->op))) {
624 pr_err("unrecognized osd opcode %d\n", src->op);
625
626 return 0;
627 }
628
629 switch (src->op) {
630 case CEPH_OSD_OP_STAT:
49719778
AE		 631 osd_data = &src->raw_data_in;
632 ceph_osdc_msg_data_add(req->r_reply, osd_data);
a8dd0a37
AE		 633 break;
634 case CEPH_OSD_OP_READ:
635 case CEPH_OSD_OP_WRITE:
ad7a60de
LW		 636 case CEPH_OSD_OP_ZERO:
637 case CEPH_OSD_OP_DELETE:
638 case CEPH_OSD_OP_TRUNCATE:
a8dd0a37 639 if (src->op == CEPH_OSD_OP_WRITE)
54d50649 640 request_data_len = src->extent.length;
a8dd0a37
AE		 641 dst->extent.offset = cpu_to_le64(src->extent.offset);
642 dst->extent.length = cpu_to_le64(src->extent.length);
643 dst->extent.truncate_size =
644 cpu_to_le64(src->extent.truncate_size);
645 dst->extent.truncate_seq =
646 cpu_to_le32(src->extent.truncate_seq);
04017e29 647 osd_data = &src->extent.osd_data;
5476492f 648 if (src->op == CEPH_OSD_OP_WRITE)
04017e29 649 ceph_osdc_msg_data_add(req->r_request, osd_data);
5476492f 650 else
04017e29 651 ceph_osdc_msg_data_add(req->r_reply, osd_data);
a8dd0a37
AE		 652 break;
653 case CEPH_OSD_OP_CALL:
a8dd0a37
AE		 654 dst->cls.class_len = src->cls.class_len;
655 dst->cls.method_len = src->cls.method_len;
04017e29
AE		 656 osd_data = &src->cls.request_info;
657 ceph_osdc_msg_data_add(req->r_request, osd_data);
658 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGELIST);
659 request_data_len = osd_data->pagelist->length;
660
661 osd_data = &src->cls.request_data;
662 data_length = ceph_osd_data_length(osd_data);
663 if (data_length) {
664 BUG_ON(osd_data->type == CEPH_OSD_DATA_TYPE_NONE);
665 dst->cls.indata_len = cpu_to_le32(data_length);
666 ceph_osdc_msg_data_add(req->r_request, osd_data);
667 src->payload_len += data_length;
668 request_data_len += data_length;
669 }
670 osd_data = &src->cls.response_data;
671 ceph_osdc_msg_data_add(req->r_reply, osd_data);
a8dd0a37
AE		 672 break;
673 case CEPH_OSD_OP_STARTSYNC:
674 break;
675 case CEPH_OSD_OP_NOTIFY_ACK:
676 case CEPH_OSD_OP_WATCH:
677 dst->watch.cookie = cpu_to_le64(src->watch.cookie);
678 dst->watch.ver = cpu_to_le64(src->watch.ver);
679 dst->watch.flag = src->watch.flag;
680 break;
681 default:
4c46459c 682 pr_err("unsupported osd opcode %s\n",
8f63ca2d 683 ceph_osd_op_name(src->op));
4c46459c 684 WARN_ON(1);
a8dd0a37
AE		 685
686 return 0;
68b4476b 687 }
a8dd0a37 688 dst->op = cpu_to_le16(src->op);
68b4476b 689 dst->payload_len = cpu_to_le32(src->payload_len);
175face2 690
54d50649 691 return request_data_len;
68b4476b
YS		 692}
693
3499e8a5
YS		 694/*
695 * build new request AND message, calculate layout, and adjust file
696 * extent as needed.
697 *
698 * if the file was recently truncated, we include information about its
699 * old and new size so that the object can be updated appropriately. (we
700 * avoid synchronously deleting truncated objects because it's slow.)
701 *
702 * if @do_sync, include a 'startsync' command so that the osd will flush
703 * data quickly.
704 */
705struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
706 struct ceph_file_layout *layout,
707 struct ceph_vino vino,
acead002 708 u64 off, u64 *plen, int num_ops,
3499e8a5
YS		 709 int opcode, int flags,
710 struct ceph_snap_context *snapc,
3499e8a5
YS		 711 u32 truncate_seq,
712 u64 truncate_size,
153e5167 713 bool use_mempool)
3499e8a5 714{
68b4476b 715 struct ceph_osd_request *req;
75d1c941
AE		 716 u64 objnum = 0;
717 u64 objoff = 0;
718 u64 objlen = 0;
d18d1e28
AE		 719 u32 object_size;
720 u64 object_base;
6816282d 721 int r;
68b4476b 722
ad7a60de
LW		 723 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
724 opcode != CEPH_OSD_OP_DELETE && opcode != CEPH_OSD_OP_ZERO &&
725 opcode != CEPH_OSD_OP_TRUNCATE);
68b4476b 726
acead002 727 req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool,
ae7ca4a3 728 GFP_NOFS);
4ad12621 729 if (!req)
6816282d 730 return ERR_PTR(-ENOMEM);
79528734 731
d178a9e7 732 req->r_flags = flags;
3499e8a5
YS		 733
734 /* calculate max write size */
a19dadfb 735 r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
3ff5f385
AE		 736 if (r < 0) {
737 ceph_osdc_put_request(req);
6816282d 738 return ERR_PTR(r);
3ff5f385 739 }
a19dadfb 740
d18d1e28
AE		 741 object_size = le32_to_cpu(layout->fl_object_size);
742 object_base = off - objoff;
ccca4e37
YZ		 743 if (!(truncate_seq == 1 && truncate_size == -1ULL)) {
744 if (truncate_size <= object_base) {
745 truncate_size = 0;
746 } else {
747 truncate_size -= object_base;
748 if (truncate_size > object_size)
749 truncate_size = object_size;
750 }
a19dadfb 751 }
d18d1e28 752
c99d2d4a 753 osd_req_op_extent_init(req, 0, opcode, objoff, objlen,
b0270324 754 truncate_size, truncate_seq);
8c042b0d 755
acead002
AE		 756 /*
757 * A second op in the ops array means the caller wants to
758 * also issue a include a 'startsync' command so that the
759 * osd will flush data quickly.
760 */
761 if (num_ops > 1)
c99d2d4a 762 osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC);
d18d1e28 763
3499e8a5
YS		 764 req->r_file_layout = *layout; /* keep a copy */
765
75d1c941
AE		 766 snprintf(req->r_oid, sizeof(req->r_oid), "%llx.%08llx",
767 vino.ino, objnum);
dbe0fc41
AE		 768 req->r_oid_len = strlen(req->r_oid);
769
f24e9980
SW		 770 return req;
771}
3d14c5d2 772EXPORT_SYMBOL(ceph_osdc_new_request);
f24e9980
SW		 773
774/*
775 * We keep osd requests in an rbtree, sorted by ->r_tid.
776 */
777static void __insert_request(struct ceph_osd_client *osdc,
778 struct ceph_osd_request *new)
779{
780 struct rb_node **p = &osdc->requests.rb_node;
781 struct rb_node *parent = NULL;
782 struct ceph_osd_request *req = NULL;
783
784 while (*p) {
785 parent = *p;
786 req = rb_entry(parent, struct ceph_osd_request, r_node);
787 if (new->r_tid < req->r_tid)
788 p = &(*p)->rb_left;
789 else if (new->r_tid > req->r_tid)
790 p = &(*p)->rb_right;
791 else
792 BUG();
793 }
794
795 rb_link_node(&new->r_node, parent, p);
796 rb_insert_color(&new->r_node, &osdc->requests);
797}
798
799static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc,
800 u64 tid)
801{
802 struct ceph_osd_request *req;
803 struct rb_node *n = osdc->requests.rb_node;
804
805 while (n) {
806 req = rb_entry(n, struct ceph_osd_request, r_node);
807 if (tid < req->r_tid)
808 n = n->rb_left;
809 else if (tid > req->r_tid)
810 n = n->rb_right;
811 else
812 return req;
813 }
814 return NULL;
815}
816
817static struct ceph_osd_request *
818__lookup_request_ge(struct ceph_osd_client *osdc,
819 u64 tid)
820{
821 struct ceph_osd_request *req;
822 struct rb_node *n = osdc->requests.rb_node;
823
824 while (n) {
825 req = rb_entry(n, struct ceph_osd_request, r_node);
826 if (tid < req->r_tid) {
827 if (!n->rb_left)
828 return req;
829 n = n->rb_left;
830 } else if (tid > req->r_tid) {
831 n = n->rb_right;
832 } else {
833 return req;
834 }
835 }
836 return NULL;
837}
838
6f6c7006
SW		 839/*
840 * Resubmit requests pending on the given osd.
841 */
842static void __kick_osd_requests(struct ceph_osd_client *osdc,
843 struct ceph_osd *osd)
844{
a40c4f10 845 struct ceph_osd_request *req, *nreq;
e02493c0 846 LIST_HEAD(resend);
6f6c7006
SW		 847 int err;
848
849 dout("__kick_osd_requests osd%d\n", osd->o_osd);
850 err = __reset_osd(osdc, osd);
685a7555 851 if (err)
6f6c7006 852 return;
e02493c0
AE		 853 /*
854 * Build up a list of requests to resend by traversing the
855 * osd's list of requests. Requests for a given object are
856 * sent in tid order, and that is also the order they're
857 * kept on this list. Therefore all requests that are in
858 * flight will be found first, followed by all requests that
859 * have not yet been sent. And to resend requests while
860 * preserving this order we will want to put any sent
861 * requests back on the front of the osd client's unsent
862 * list.
863 *
864 * So we build a separate ordered list of already-sent
865 * requests for the affected osd and splice it onto the
866 * front of the osd client's unsent list. Once we've seen a
867 * request that has not yet been sent we're done. Those
868 * requests are already sitting right where they belong.
869 */
6f6c7006 870 list_for_each_entry(req, &osd->o_requests, r_osd_item) {
e02493c0
AE		 871 if (!req->r_sent)
872 break;
873 list_move_tail(&req->r_req_lru_item, &resend);
874 dout("requeueing %p tid %llu osd%d\n", req, req->r_tid,
6f6c7006 875 osd->o_osd);
a40c4f10
YS		 876 if (!req->r_linger)
877 req->r_flags |= CEPH_OSD_FLAG_RETRY;
878 }
e02493c0 879 list_splice(&resend, &osdc->req_unsent);
a40c4f10 880
e02493c0
AE		 881 /*
882 * Linger requests are re-registered before sending, which
883 * sets up a new tid for each. We add them to the unsent
884 * list at the end to keep things in tid order.
885 */
a40c4f10
YS		 886 list_for_each_entry_safe(req, nreq, &osd->o_linger_requests,
887 r_linger_osd) {
77f38e0e
SW		 888 /*
889 * reregister request prior to unregistering linger so
890 * that r_osd is preserved.
891 */
892 BUG_ON(!list_empty(&req->r_req_lru_item));
a40c4f10 893 __register_request(osdc, req);
e02493c0 894 list_add_tail(&req->r_req_lru_item, &osdc->req_unsent);
ad885927 895 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
77f38e0e 896 __unregister_linger_request(osdc, req);
a40c4f10
YS		 897 dout("requeued lingering %p tid %llu osd%d\n", req, req->r_tid,
898 osd->o_osd);
6f6c7006
SW		 899 }
900}
901
f24e9980 902/*
81b024e7 903 * If the osd connection drops, we need to resubmit all requests.
f24e9980
SW		 904 */
905static void osd_reset(struct ceph_connection *con)
906{
907 struct ceph_osd *osd = con->private;
908 struct ceph_osd_client *osdc;
909
910 if (!osd)
911 return;
912 dout("osd_reset osd%d\n", osd->o_osd);
913 osdc = osd->o_osdc;
f24e9980 914 down_read(&osdc->map_sem);
83aff95e
SW		 915 mutex_lock(&osdc->request_mutex);
916 __kick_osd_requests(osdc, osd);
f9d25199 917 __send_queued(osdc);
83aff95e 918 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 919 up_read(&osdc->map_sem);
920}
921
922/*
923 * Track open sessions with osds.
924 */
e10006f8 925static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
f24e9980
SW		 926{
927 struct ceph_osd *osd;
928
929 osd = kzalloc(sizeof(*osd), GFP_NOFS);
930 if (!osd)
931 return NULL;
932
933 atomic_set(&osd->o_ref, 1);
934 osd->o_osdc = osdc;
e10006f8 935 osd->o_osd = onum;
f407731d 936 RB_CLEAR_NODE(&osd->o_node);
f24e9980 937 INIT_LIST_HEAD(&osd->o_requests);
a40c4f10 938 INIT_LIST_HEAD(&osd->o_linger_requests);
f5a2041b 939 INIT_LIST_HEAD(&osd->o_osd_lru);
f24e9980
SW		 940 osd->o_incarnation = 1;
941
b7a9e5dd 942 ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
4e7a5dcd 943
422d2cb8 944 INIT_LIST_HEAD(&osd->o_keepalive_item);
f24e9980
SW		 945 return osd;
946}
947
948static struct ceph_osd *get_osd(struct ceph_osd *osd)
949{
950 if (atomic_inc_not_zero(&osd->o_ref)) {
951 dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
952 atomic_read(&osd->o_ref));
953 return osd;
954 } else {
955 dout("get_osd %p FAIL\n", osd);
956 return NULL;
957 }
958}
959
960static void put_osd(struct ceph_osd *osd)
961{
962 dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
963 atomic_read(&osd->o_ref) - 1);
a255651d 964 if (atomic_dec_and_test(&osd->o_ref) && osd->o_auth.authorizer) {
79494d1b
SW		 965 struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth;
966
27859f97 967 ceph_auth_destroy_authorizer(ac, osd->o_auth.authorizer);
f24e9980 968 kfree(osd);
79494d1b 969 }
f24e9980
SW		 970}
971
972/*
973 * remove an osd from our map
974 */
f5a2041b 975static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
f24e9980 976{
f5a2041b 977 dout("__remove_osd %p\n", osd);
f24e9980
SW		 978 BUG_ON(!list_empty(&osd->o_requests));
979 rb_erase(&osd->o_node, &osdc->osds);
f5a2041b 980 list_del_init(&osd->o_osd_lru);
f24e9980
SW		 981 ceph_con_close(&osd->o_con);
982 put_osd(osd);
983}
984
aca420bc
SW		 985static void remove_all_osds(struct ceph_osd_client *osdc)
986{
048a9d2d 987 dout("%s %p\n", __func__, osdc);
aca420bc
SW		 988 mutex_lock(&osdc->request_mutex);
989 while (!RB_EMPTY_ROOT(&osdc->osds)) {
990 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
991 struct ceph_osd, o_node);
992 __remove_osd(osdc, osd);
993 }
994 mutex_unlock(&osdc->request_mutex);
995}
996
f5a2041b
YS		 997static void __move_osd_to_lru(struct ceph_osd_client *osdc,
998 struct ceph_osd *osd)
999{
1000 dout("__move_osd_to_lru %p\n", osd);
1001 BUG_ON(!list_empty(&osd->o_osd_lru));
1002 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
3d14c5d2 1003 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl * HZ;
f5a2041b
YS		 1004}
1005
1006static void __remove_osd_from_lru(struct ceph_osd *osd)
1007{
1008 dout("__remove_osd_from_lru %p\n", osd);
1009 if (!list_empty(&osd->o_osd_lru))
1010 list_del_init(&osd->o_osd_lru);
1011}
1012
aca420bc 1013static void remove_old_osds(struct ceph_osd_client *osdc)
f5a2041b
YS		 1014{
1015 struct ceph_osd *osd, *nosd;
1016
1017 dout("__remove_old_osds %p\n", osdc);
1018 mutex_lock(&osdc->request_mutex);
1019 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
aca420bc 1020 if (time_before(jiffies, osd->lru_ttl))
f5a2041b
YS		 1021 break;
1022 __remove_osd(osdc, osd);
1023 }
1024 mutex_unlock(&osdc->request_mutex);
1025}
1026
f24e9980
SW		 1027/*
1028 * reset osd connect
1029 */
f5a2041b 1030static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
f24e9980 1031{
c3acb181 1032 struct ceph_entity_addr *peer_addr;
f24e9980 1033
f5a2041b 1034 dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
a40c4f10
YS		 1035 if (list_empty(&osd->o_requests) &&
1036 list_empty(&osd->o_linger_requests)) {
f5a2041b 1037 __remove_osd(osdc, osd);
c3acb181
AE		 1038
1039 return -ENODEV;
1040 }
1041
1042 peer_addr = &osdc->osdmap->osd_addr[osd->o_osd];
1043 if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1044 !ceph_con_opened(&osd->o_con)) {
1045 struct ceph_osd_request *req;
1046
87b315a5
SW		 1047 dout(" osd addr hasn't changed and connection never opened,"
1048 " letting msgr retry");
1049 /* touch each r_stamp for handle_timeout()'s benfit */
1050 list_for_each_entry(req, &osd->o_requests, r_osd_item)
1051 req->r_stamp = jiffies;
c3acb181
AE		 1052
1053 return -EAGAIN;
f24e9980 1054 }
c3acb181
AE		 1055
1056 ceph_con_close(&osd->o_con);
1057 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1058 osd->o_incarnation++;
1059
1060 return 0;
f24e9980
SW		 1061}
1062
1063static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new)
1064{
1065 struct rb_node **p = &osdc->osds.rb_node;
1066 struct rb_node *parent = NULL;
1067 struct ceph_osd *osd = NULL;
1068
aca420bc 1069 dout("__insert_osd %p osd%d\n", new, new->o_osd);
f24e9980
SW		 1070 while (*p) {
1071 parent = *p;
1072 osd = rb_entry(parent, struct ceph_osd, o_node);
1073 if (new->o_osd < osd->o_osd)
1074 p = &(*p)->rb_left;
1075 else if (new->o_osd > osd->o_osd)
1076 p = &(*p)->rb_right;
1077 else
1078 BUG();
1079 }
1080
1081 rb_link_node(&new->o_node, parent, p);
1082 rb_insert_color(&new->o_node, &osdc->osds);
1083}
1084
1085static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o)
1086{
1087 struct ceph_osd *osd;
1088 struct rb_node *n = osdc->osds.rb_node;
1089
1090 while (n) {
1091 osd = rb_entry(n, struct ceph_osd, o_node);
1092 if (o < osd->o_osd)
1093 n = n->rb_left;
1094 else if (o > osd->o_osd)
1095 n = n->rb_right;
1096 else
1097 return osd;
1098 }
1099 return NULL;
1100}
1101
422d2cb8
YS		 1102static void __schedule_osd_timeout(struct ceph_osd_client *osdc)
1103{
1104 schedule_delayed_work(&osdc->timeout_work,
3d14c5d2 1105 osdc->client->options->osd_keepalive_timeout * HZ);
422d2cb8
YS		 1106}
1107
1108static void __cancel_osd_timeout(struct ceph_osd_client *osdc)
1109{
1110 cancel_delayed_work(&osdc->timeout_work);
1111}
f24e9980
SW		 1112
1113/*
1114 * Register request, assign tid. If this is the first request, set up
1115 * the timeout event.
1116 */
a40c4f10
YS		 1117static void __register_request(struct ceph_osd_client *osdc,
1118 struct ceph_osd_request *req)
f24e9980 1119{
f24e9980 1120 req->r_tid = ++osdc->last_tid;
6df058c0 1121 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
77f38e0e 1122 dout("__register_request %p tid %lld\n", req, req->r_tid);
f24e9980
SW		 1123 __insert_request(osdc, req);
1124 ceph_osdc_get_request(req);
1125 osdc->num_requests++;
f24e9980 1126 if (osdc->num_requests == 1) {
422d2cb8
YS		 1127 dout(" first request, scheduling timeout\n");
1128 __schedule_osd_timeout(osdc);
f24e9980 1129 }
a40c4f10
YS		 1130}
1131
f24e9980
SW		 1132/*
1133 * called under osdc->request_mutex
1134 */
1135static void __unregister_request(struct ceph_osd_client *osdc,
1136 struct ceph_osd_request *req)
1137{
35f9f8a0
SW		 1138 if (RB_EMPTY_NODE(&req->r_node)) {
1139 dout("__unregister_request %p tid %lld not registered\n",
1140 req, req->r_tid);
1141 return;
1142 }
1143
f24e9980
SW		 1144 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
1145 rb_erase(&req->r_node, &osdc->requests);
1146 osdc->num_requests--;
1147
0ba6478d
SW		 1148 if (req->r_osd) {
1149 /* make sure the original request isn't in flight. */
6740a845 1150 ceph_msg_revoke(req->r_request);
0ba6478d
SW		 1151
1152 list_del_init(&req->r_osd_item);
a40c4f10
YS		 1153 if (list_empty(&req->r_osd->o_requests) &&
1154 list_empty(&req->r_osd->o_linger_requests)) {
1155 dout("moving osd to %p lru\n", req->r_osd);
f5a2041b 1156 __move_osd_to_lru(osdc, req->r_osd);
a40c4f10 1157 }
fbdb9190 1158 if (list_empty(&req->r_linger_item))
a40c4f10 1159 req->r_osd = NULL;
0ba6478d 1160 }
f24e9980 1161
7d5f2481 1162 list_del_init(&req->r_req_lru_item);
f24e9980
SW		 1163 ceph_osdc_put_request(req);
1164
422d2cb8
YS		 1165 if (osdc->num_requests == 0) {
1166 dout(" no requests, canceling timeout\n");
1167 __cancel_osd_timeout(osdc);
f24e9980
SW		 1168 }
1169}
1170
1171/*
1172 * Cancel a previously queued request message
1173 */
1174static void __cancel_request(struct ceph_osd_request *req)
1175{
6bc18876 1176 if (req->r_sent && req->r_osd) {
6740a845 1177 ceph_msg_revoke(req->r_request);
f24e9980
SW		 1178 req->r_sent = 0;
1179 }
1180}
1181
a40c4f10
YS		 1182static void __register_linger_request(struct ceph_osd_client *osdc,
1183 struct ceph_osd_request *req)
1184{
1185 dout("__register_linger_request %p\n", req);
96e4dac6 1186 ceph_osdc_get_request(req);
a40c4f10 1187 list_add_tail(&req->r_linger_item, &osdc->req_linger);
6194ea89
SW		 1188 if (req->r_osd)
1189 list_add_tail(&req->r_linger_osd,
1190 &req->r_osd->o_linger_requests);
a40c4f10
YS		 1191}
1192
1193static void __unregister_linger_request(struct ceph_osd_client *osdc,
1194 struct ceph_osd_request *req)
1195{
1196 dout("__unregister_linger_request %p\n", req);
61c74035 1197 list_del_init(&req->r_linger_item);
a40c4f10 1198 if (req->r_osd) {
a40c4f10
YS		 1199 list_del_init(&req->r_linger_osd);
1200
1201 if (list_empty(&req->r_osd->o_requests) &&
1202 list_empty(&req->r_osd->o_linger_requests)) {
1203 dout("moving osd to %p lru\n", req->r_osd);
1204 __move_osd_to_lru(osdc, req->r_osd);
1205 }
fbdb9190
SW		 1206 if (list_empty(&req->r_osd_item))
1207 req->r_osd = NULL;
a40c4f10 1208 }
96e4dac6 1209 ceph_osdc_put_request(req);
a40c4f10
YS		 1210}
1211
1212void ceph_osdc_unregister_linger_request(struct ceph_osd_client *osdc,
1213 struct ceph_osd_request *req)
1214{
1215 mutex_lock(&osdc->request_mutex);
1216 if (req->r_linger) {
c10ebbf5 1217 req->r_linger = 0;
96e4dac6 1218 __unregister_linger_request(osdc, req);
a40c4f10
YS		 1219 }
1220 mutex_unlock(&osdc->request_mutex);
1221}
1222EXPORT_SYMBOL(ceph_osdc_unregister_linger_request);
1223
1224void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
1225 struct ceph_osd_request *req)
1226{
1227 if (!req->r_linger) {
1228 dout("set_request_linger %p\n", req);
1229 req->r_linger = 1;
a40c4f10
YS		 1230 }
1231}
1232EXPORT_SYMBOL(ceph_osdc_set_request_linger);
1233
d29adb34
JD		 1234/*
1235 * Returns whether a request should be blocked from being sent
1236 * based on the current osdmap and osd_client settings.
1237 *
1238 * Caller should hold map_sem for read.
1239 */
1240static bool __req_should_be_paused(struct ceph_osd_client *osdc,
1241 struct ceph_osd_request *req)
1242{
1243 bool pauserd = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD);
1244 bool pausewr = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR) ||
1245 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
1246 return (req->r_flags & CEPH_OSD_FLAG_READ && pauserd) ||
1247 (req->r_flags & CEPH_OSD_FLAG_WRITE && pausewr);
1248}
1249
f24e9980
SW		 1250/*
1251 * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
1252 * (as needed), and set the request r_osd appropriately. If there is
25985edc 1253 * no up osd, set r_osd to NULL. Move the request to the appropriate list
6f6c7006 1254 * (unsent, homeless) or leave on in-flight lru.
f24e9980
SW		 1255 *
1256 * Return 0 if unchanged, 1 if changed, or negative on error.
1257 *
1258 * Caller should hold map_sem for read and request_mutex.
1259 */
6f6c7006 1260static int __map_request(struct ceph_osd_client *osdc,
38d6453c 1261 struct ceph_osd_request *req, int force_resend)
f24e9980 1262{
5b191d99 1263 struct ceph_pg pgid;
d85b7056
SW		 1264 int acting[CEPH_PG_MAX_SIZE];
1265 int o = -1, num = 0;
f24e9980 1266 int err;
d29adb34 1267 bool was_paused;
f24e9980 1268
6f6c7006 1269 dout("map_request %p tid %lld\n", req, req->r_tid);
41766f87
AE		 1270 err = ceph_calc_ceph_pg(&pgid, req->r_oid, osdc->osdmap,
1271 ceph_file_layout_pg_pool(req->r_file_layout));
6f6c7006
SW		 1272 if (err) {
1273 list_move(&req->r_req_lru_item, &osdc->req_notarget);
f24e9980 1274 return err;
6f6c7006 1275 }
7740a42f
SW		 1276 req->r_pgid = pgid;
1277
d85b7056
SW		 1278 err = ceph_calc_pg_acting(osdc->osdmap, pgid, acting);
1279 if (err > 0) {
1280 o = acting[0];
1281 num = err;
1282 }
f24e9980 1283
d29adb34
JD		 1284 was_paused = req->r_paused;
1285 req->r_paused = __req_should_be_paused(osdc, req);
1286 if (was_paused && !req->r_paused)
1287 force_resend = 1;
1288
38d6453c
SW		 1289 if ((!force_resend &&
1290 req->r_osd && req->r_osd->o_osd == o &&
d85b7056
SW		 1291 req->r_sent >= req->r_osd->o_incarnation &&
1292 req->r_num_pg_osds == num &&
1293 memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
d29adb34
JD		 1294 (req->r_osd == NULL && o == -1) ||
1295 req->r_paused)
f24e9980
SW		 1296 return 0; /* no change */
1297
5b191d99
SW		 1298 dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n",
1299 req->r_tid, pgid.pool, pgid.seed, o,
f24e9980
SW		 1300 req->r_osd ? req->r_osd->o_osd : -1);
1301
d85b7056
SW		 1302 /* record full pg acting set */
1303 memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
1304 req->r_num_pg_osds = num;
1305
f24e9980
SW		 1306 if (req->r_osd) {
1307 __cancel_request(req);
1308 list_del_init(&req->r_osd_item);
f24e9980
SW		 1309 req->r_osd = NULL;
1310 }
1311
1312 req->r_osd = __lookup_osd(osdc, o);
1313 if (!req->r_osd && o >= 0) {
c99eb1c7 1314 err = -ENOMEM;
e10006f8 1315 req->r_osd = create_osd(osdc, o);
6f6c7006
SW		 1316 if (!req->r_osd) {
1317 list_move(&req->r_req_lru_item, &osdc->req_notarget);
c99eb1c7 1318 goto out;
6f6c7006 1319 }
f24e9980 1320
6f6c7006 1321 dout("map_request osd %p is osd%d\n", req->r_osd, o);
f24e9980
SW		 1322 __insert_osd(osdc, req->r_osd);
1323
b7a9e5dd
SW		 1324 ceph_con_open(&req->r_osd->o_con,
1325 CEPH_ENTITY_TYPE_OSD, o,
1326 &osdc->osdmap->osd_addr[o]);
f24e9980
SW		 1327 }
1328
f5a2041b
YS		 1329 if (req->r_osd) {
1330 __remove_osd_from_lru(req->r_osd);
ad885927
AE		 1331 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
1332 list_move_tail(&req->r_req_lru_item, &osdc->req_unsent);
6f6c7006 1333 } else {
ad885927 1334 list_move_tail(&req->r_req_lru_item, &osdc->req_notarget);
f5a2041b 1335 }
d85b7056 1336 err = 1; /* osd or pg changed */
f24e9980
SW		 1337
1338out:
f24e9980
SW		 1339 return err;
1340}
1341
1342/*
1343 * caller should hold map_sem (for read) and request_mutex
1344 */
56e925b6
SW		 1345static void __send_request(struct ceph_osd_client *osdc,
1346 struct ceph_osd_request *req)
f24e9980 1347{
1b83bef2 1348 void *p;
f24e9980 1349
1b83bef2
SW		 1350 dout("send_request %p tid %llu to osd%d flags %d pg %lld.%x\n",
1351 req, req->r_tid, req->r_osd->o_osd, req->r_flags,
1352 (unsigned long long)req->r_pgid.pool, req->r_pgid.seed);
1353
1354 /* fill in message content that changes each time we send it */
1355 put_unaligned_le32(osdc->osdmap->epoch, req->r_request_osdmap_epoch);
1356 put_unaligned_le32(req->r_flags, req->r_request_flags);
1357 put_unaligned_le64(req->r_pgid.pool, req->r_request_pool);
1358 p = req->r_request_pgid;
1359 ceph_encode_64(&p, req->r_pgid.pool);
1360 ceph_encode_32(&p, req->r_pgid.seed);
1361 put_unaligned_le64(1, req->r_request_attempts); /* FIXME */
1362 memcpy(req->r_request_reassert_version, &req->r_reassert_version,
1363 sizeof(req->r_reassert_version));
2169aea6 1364
3dd72fc0 1365 req->r_stamp = jiffies;
07a27e22 1366 list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
f24e9980
SW		 1367
1368 ceph_msg_get(req->r_request); /* send consumes a ref */
26be8808 1369
f24e9980 1370 req->r_sent = req->r_osd->o_incarnation;
26be8808
AE		 1371
1372 ceph_con_send(&req->r_osd->o_con, req->r_request);
f24e9980
SW		 1373}
1374
6f6c7006
SW		 1375/*
1376 * Send any requests in the queue (req_unsent).
1377 */
f9d25199 1378static void __send_queued(struct ceph_osd_client *osdc)
6f6c7006
SW		 1379{
1380 struct ceph_osd_request *req, *tmp;
1381
f9d25199
AE		 1382 dout("__send_queued\n");
1383 list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item)
6f6c7006 1384 __send_request(osdc, req);
6f6c7006
SW		 1385}
1386
f24e9980
SW		 1387/*
1388 * Timeout callback, called every N seconds when 1 or more osd
1389 * requests has been active for more than N seconds. When this
1390 * happens, we ping all OSDs with requests who have timed out to
1391 * ensure any communications channel reset is detected. Reset the
1392 * request timeouts another N seconds in the future as we go.
1393 * Reschedule the timeout event another N seconds in future (unless
1394 * there are no open requests).
1395 */
1396static void handle_timeout(struct work_struct *work)
1397{
1398 struct ceph_osd_client *osdc =
1399 container_of(work, struct ceph_osd_client, timeout_work.work);
83aff95e 1400 struct ceph_osd_request *req;
f24e9980 1401 struct ceph_osd *osd;
422d2cb8 1402 unsigned long keepalive =
3d14c5d2 1403 osdc->client->options->osd_keepalive_timeout * HZ;
422d2cb8 1404 struct list_head slow_osds;
f24e9980
SW		 1405 dout("timeout\n");
1406 down_read(&osdc->map_sem);
1407
1408 ceph_monc_request_next_osdmap(&osdc->client->monc);
1409
1410 mutex_lock(&osdc->request_mutex);
f24e9980 1411
422d2cb8
YS		 1412 /*
1413 * ping osds that are a bit slow. this ensures that if there
1414 * is a break in the TCP connection we will notice, and reopen
1415 * a connection with that osd (from the fault callback).
1416 */
1417 INIT_LIST_HEAD(&slow_osds);
1418 list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
3dd72fc0 1419 if (time_before(jiffies, req->r_stamp + keepalive))
422d2cb8
YS		 1420 break;
1421
1422 osd = req->r_osd;
1423 BUG_ON(!osd);
1424 dout(" tid %llu is slow, will send keepalive on osd%d\n",
f24e9980 1425 req->r_tid, osd->o_osd);
422d2cb8
YS		 1426 list_move_tail(&osd->o_keepalive_item, &slow_osds);
1427 }
1428 while (!list_empty(&slow_osds)) {
1429 osd = list_entry(slow_osds.next, struct ceph_osd,
1430 o_keepalive_item);
1431 list_del_init(&osd->o_keepalive_item);
f24e9980
SW		 1432 ceph_con_keepalive(&osd->o_con);
1433 }
1434
422d2cb8 1435 __schedule_osd_timeout(osdc);
f9d25199 1436 __send_queued(osdc);
f24e9980 1437 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 1438 up_read(&osdc->map_sem);
1439}
1440
f5a2041b
YS		 1441static void handle_osds_timeout(struct work_struct *work)
1442{
1443 struct ceph_osd_client *osdc =
1444 container_of(work, struct ceph_osd_client,
1445 osds_timeout_work.work);
1446 unsigned long delay =
3d14c5d2 1447 osdc->client->options->osd_idle_ttl * HZ >> 2;
f5a2041b
YS		 1448
1449 dout("osds timeout\n");
1450 down_read(&osdc->map_sem);
aca420bc 1451 remove_old_osds(osdc);
f5a2041b
YS		 1452 up_read(&osdc->map_sem);
1453
1454 schedule_delayed_work(&osdc->osds_timeout_work,
1455 round_jiffies_relative(delay));
1456}
1457
25845472
SW		 1458static void complete_request(struct ceph_osd_request *req)
1459{
25845472
SW		 1460 complete_all(&req->r_safe_completion); /* fsync waiter */
1461}
1462
f24e9980
SW		 1463/*
1464 * handle osd op reply. either call the callback if it is specified,
1465 * or do the completion to wake up the waiting thread.
1466 */
350b1c32
SW		 1467static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg,
1468 struct ceph_connection *con)
f24e9980 1469{
1b83bef2 1470 void *p, *end;
f24e9980
SW		 1471 struct ceph_osd_request *req;
1472 u64 tid;
1b83bef2 1473 int object_len;
79528734
AE		 1474 unsigned int numops;
1475 int payload_len, flags;
0ceed5db 1476 s32 result;
1b83bef2
SW		 1477 s32 retry_attempt;
1478 struct ceph_pg pg;
1479 int err;
1480 u32 reassert_epoch;
1481 u64 reassert_version;
1482 u32 osdmap_epoch;
0d5af164 1483 int already_completed;
9fc6e064 1484 u32 bytes;
79528734 1485 unsigned int i;
f24e9980 1486
6df058c0 1487 tid = le64_to_cpu(msg->hdr.tid);
1b83bef2
SW		 1488 dout("handle_reply %p tid %llu\n", msg, tid);
1489
1490 p = msg->front.iov_base;
1491 end = p + msg->front.iov_len;
1492
1493 ceph_decode_need(&p, end, 4, bad);
1494 object_len = ceph_decode_32(&p);
1495 ceph_decode_need(&p, end, object_len, bad);
1496 p += object_len;
1497
ef4859d6 1498 err = ceph_decode_pgid(&p, end, &pg);
1b83bef2 1499 if (err)
f24e9980 1500 goto bad;
1b83bef2
SW		 1501
1502 ceph_decode_need(&p, end, 8 + 4 + 4 + 8 + 4, bad);
1503 flags = ceph_decode_64(&p);
1504 result = ceph_decode_32(&p);
1505 reassert_epoch = ceph_decode_32(&p);
1506 reassert_version = ceph_decode_64(&p);
1507 osdmap_epoch = ceph_decode_32(&p);
1508
f24e9980
SW		 1509 /* lookup */
1510 mutex_lock(&osdc->request_mutex);
1511 req = __lookup_request(osdc, tid);
1512 if (req == NULL) {
1513 dout("handle_reply tid %llu dne\n", tid);
8058fd45 1514 goto bad_mutex;
f24e9980
SW		 1515 }
1516 ceph_osdc_get_request(req);
1b83bef2
SW		 1517
1518 dout("handle_reply %p tid %llu req %p result %d\n", msg, tid,
1519 req, result);
1520
18741196 1521 ceph_decode_need(&p, end, 4, bad_put);
1b83bef2
SW		 1522 numops = ceph_decode_32(&p);
1523 if (numops > CEPH_OSD_MAX_OP)
1524 goto bad_put;
1525 if (numops != req->r_num_ops)
1526 goto bad_put;
1527 payload_len = 0;
18741196 1528 ceph_decode_need(&p, end, numops * sizeof(struct ceph_osd_op), bad_put);
1b83bef2
SW		 1529 for (i = 0; i < numops; i++) {
1530 struct ceph_osd_op *op = p;
1531 int len;
1532
1533 len = le32_to_cpu(op->payload_len);
1534 req->r_reply_op_len[i] = len;
1535 dout(" op %d has %d bytes\n", i, len);
1536 payload_len += len;
1537 p += sizeof(*op);
1538 }
9fc6e064
AE		 1539 bytes = le32_to_cpu(msg->hdr.data_len);
1540 if (payload_len != bytes) {
1b83bef2 1541 pr_warning("sum of op payload lens %d != data_len %d",
9fc6e064 1542 payload_len, bytes);
1b83bef2
SW		 1543 goto bad_put;
1544 }
1545
18741196 1546 ceph_decode_need(&p, end, 4 + numops * 4, bad_put);
1b83bef2
SW		 1547 retry_attempt = ceph_decode_32(&p);
1548 for (i = 0; i < numops; i++)
1549 req->r_reply_op_result[i] = ceph_decode_32(&p);
f24e9980 1550
eb845ff1
YZ		 1551 already_completed = req->r_got_reply;
1552
f24e9980 1553 if (!req->r_got_reply) {
f24e9980 1554
1b83bef2 1555 req->r_result = result;
f24e9980
SW		 1556 dout("handle_reply result %d bytes %d\n", req->r_result,
1557 bytes);
1558 if (req->r_result == 0)
1559 req->r_result = bytes;
1560
1561 /* in case this is a write and we need to replay, */
1b83bef2
SW		 1562 req->r_reassert_version.epoch = cpu_to_le32(reassert_epoch);
1563 req->r_reassert_version.version = cpu_to_le64(reassert_version);
f24e9980
SW		 1564
1565 req->r_got_reply = 1;
1566 } else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) {
1567 dout("handle_reply tid %llu dup ack\n", tid);
34b43a56 1568 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 1569 goto done;
1570 }
1571
1572 dout("handle_reply tid %llu flags %d\n", tid, flags);
1573
a40c4f10
YS		 1574 if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK))
1575 __register_linger_request(osdc, req);
1576
f24e9980 1577 /* either this is a read, or we got the safe response */
0ceed5db
SW		 1578 if (result < 0 ||
1579 (flags & CEPH_OSD_FLAG_ONDISK) ||
f24e9980
SW		 1580 ((flags & CEPH_OSD_FLAG_WRITE) == 0))
1581 __unregister_request(osdc, req);
1582
1583 mutex_unlock(&osdc->request_mutex);
1584
eb845ff1 1585 if (!already_completed) {
61c5d6bf
YZ		 1586 if (req->r_unsafe_callback &&
1587 result >= 0 && !(flags & CEPH_OSD_FLAG_ONDISK))
1588 req->r_unsafe_callback(req, true);
eb845ff1
YZ		 1589 if (req->r_callback)
1590 req->r_callback(req, msg);
1591 else
1592 complete_all(&req->r_completion);
1593 }
f24e9980 1594
61c5d6bf
YZ		 1595 if (flags & CEPH_OSD_FLAG_ONDISK) {
1596 if (req->r_unsafe_callback && already_completed)
1597 req->r_unsafe_callback(req, false);
25845472 1598 complete_request(req);
61c5d6bf 1599 }
f24e9980
SW		 1600
1601done:
a40c4f10 1602 dout("req=%p req->r_linger=%d\n", req, req->r_linger);
f24e9980
SW		 1603 ceph_osdc_put_request(req);
1604 return;
1605
1b83bef2 1606bad_put:
37c89bde
LW		 1607 req->r_result = -EIO;
1608 __unregister_request(osdc, req);
1609 if (req->r_callback)
1610 req->r_callback(req, msg);
1611 else
1612 complete_all(&req->r_completion);
1613 complete_request(req);
1b83bef2 1614 ceph_osdc_put_request(req);
8058fd45
AE		 1615bad_mutex:
1616 mutex_unlock(&osdc->request_mutex);
f24e9980 1617bad:
1b83bef2
SW		 1618 pr_err("corrupt osd_op_reply got %d %d\n",
1619 (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len));
9ec7cab1 1620 ceph_msg_dump(msg);
f24e9980
SW		 1621}
1622
6f6c7006 1623static void reset_changed_osds(struct ceph_osd_client *osdc)
f24e9980 1624{
f24e9980 1625 struct rb_node *p, *n;
f24e9980 1626
6f6c7006
SW		 1627 for (p = rb_first(&osdc->osds); p; p = n) {
1628 struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
f24e9980 1629
6f6c7006
SW		 1630 n = rb_next(p);
1631 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
1632 memcmp(&osd->o_con.peer_addr,
1633 ceph_osd_addr(osdc->osdmap,
1634 osd->o_osd),
1635 sizeof(struct ceph_entity_addr)) != 0)
1636 __reset_osd(osdc, osd);
f24e9980 1637 }
422d2cb8
YS		 1638}
1639
1640/*
6f6c7006
SW		 1641 * Requeue requests whose mapping to an OSD has changed. If requests map to
1642 * no osd, request a new map.
422d2cb8 1643 *
e6d50f67 1644 * Caller should hold map_sem for read.
422d2cb8 1645 */
38d6453c 1646static void kick_requests(struct ceph_osd_client *osdc, int force_resend)
422d2cb8 1647{
a40c4f10 1648 struct ceph_osd_request *req, *nreq;
6f6c7006
SW		 1649 struct rb_node *p;
1650 int needmap = 0;
1651 int err;
422d2cb8 1652
38d6453c 1653 dout("kick_requests %s\n", force_resend ? " (force resend)" : "");
422d2cb8 1654 mutex_lock(&osdc->request_mutex);
6194ea89 1655 for (p = rb_first(&osdc->requests); p; ) {
6f6c7006 1656 req = rb_entry(p, struct ceph_osd_request, r_node);
6194ea89 1657 p = rb_next(p);
ab60b16d
AE		 1658
1659 /*
1660 * For linger requests that have not yet been
1661 * registered, move them to the linger list; they'll
1662 * be sent to the osd in the loop below. Unregister
1663 * the request before re-registering it as a linger
1664 * request to ensure the __map_request() below
1665 * will decide it needs to be sent.
1666 */
1667 if (req->r_linger && list_empty(&req->r_linger_item)) {
1668 dout("%p tid %llu restart on osd%d\n",
1669 req, req->r_tid,
1670 req->r_osd ? req->r_osd->o_osd : -1);
96e4dac6 1671 ceph_osdc_get_request(req);
ab60b16d
AE		 1672 __unregister_request(osdc, req);
1673 __register_linger_request(osdc, req);
96e4dac6 1674 ceph_osdc_put_request(req);
ab60b16d
AE		 1675 continue;
1676 }
1677
38d6453c 1678 err = __map_request(osdc, req, force_resend);
6f6c7006
SW		 1679 if (err < 0)
1680 continue; /* error */
1681 if (req->r_osd == NULL) {
1682 dout("%p tid %llu maps to no osd\n", req, req->r_tid);
1683 needmap++; /* request a newer map */
1684 } else if (err > 0) {
6194ea89
SW		 1685 if (!req->r_linger) {
1686 dout("%p tid %llu requeued on osd%d\n", req,
1687 req->r_tid,
1688 req->r_osd ? req->r_osd->o_osd : -1);
a40c4f10 1689 req->r_flags |= CEPH_OSD_FLAG_RETRY;
6194ea89
SW		 1690 }
1691 }
a40c4f10
YS		 1692 }
1693
1694 list_for_each_entry_safe(req, nreq, &osdc->req_linger,
1695 r_linger_item) {
1696 dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
1697
38d6453c 1698 err = __map_request(osdc, req, force_resend);
ab60b16d 1699 dout("__map_request returned %d\n", err);
a40c4f10
YS		 1700 if (err == 0)
1701 continue; /* no change and no osd was specified */
1702 if (err < 0)
1703 continue; /* hrm! */
1704 if (req->r_osd == NULL) {
1705 dout("tid %llu maps to no valid osd\n", req->r_tid);
1706 needmap++; /* request a newer map */
1707 continue;
6f6c7006 1708 }
a40c4f10
YS		 1709
1710 dout("kicking lingering %p tid %llu osd%d\n", req, req->r_tid,
1711 req->r_osd ? req->r_osd->o_osd : -1);
a40c4f10 1712 __register_request(osdc, req);
c89ce05e 1713 __unregister_linger_request(osdc, req);
6f6c7006 1714 }
14d2f38d 1715 reset_changed_osds(osdc);
f24e9980
SW		 1716 mutex_unlock(&osdc->request_mutex);
1717
1718 if (needmap) {
1719 dout("%d requests for down osds, need new map\n", needmap);
1720 ceph_monc_request_next_osdmap(&osdc->client->monc);
1721 }
422d2cb8 1722}
6f6c7006
SW		 1723
1724
f24e9980
SW		 1725/*
1726 * Process updated osd map.
1727 *
1728 * The message contains any number of incremental and full maps, normally
1729 * indicating some sort of topology change in the cluster. Kick requests
1730 * off to different OSDs as needed.
1731 */
1732void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1733{
1734 void *p, *end, *next;
1735 u32 nr_maps, maplen;
1736 u32 epoch;
1737 struct ceph_osdmap *newmap = NULL, *oldmap;
1738 int err;
1739 struct ceph_fsid fsid;
1740
1741 dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
1742 p = msg->front.iov_base;
1743 end = p + msg->front.iov_len;
1744
1745 /* verify fsid */
1746 ceph_decode_need(&p, end, sizeof(fsid), bad);
1747 ceph_decode_copy(&p, &fsid, sizeof(fsid));
0743304d
SW		 1748 if (ceph_check_fsid(osdc->client, &fsid) < 0)
1749 return;
f24e9980
SW		 1750
1751 down_write(&osdc->map_sem);
1752
1753 /* incremental maps */
1754 ceph_decode_32_safe(&p, end, nr_maps, bad);
1755 dout(" %d inc maps\n", nr_maps);
1756 while (nr_maps > 0) {
1757 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 1758 epoch = ceph_decode_32(&p);
1759 maplen = ceph_decode_32(&p);
f24e9980
SW		 1760 ceph_decode_need(&p, end, maplen, bad);
1761 next = p + maplen;
1762 if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
1763 dout("applying incremental map %u len %d\n",
1764 epoch, maplen);
1765 newmap = osdmap_apply_incremental(&p, next,
1766 osdc->osdmap,
15d9882c 1767 &osdc->client->msgr);
f24e9980
SW		 1768 if (IS_ERR(newmap)) {
1769 err = PTR_ERR(newmap);
1770 goto bad;
1771 }
30dc6381 1772 BUG_ON(!newmap);
f24e9980
SW		 1773 if (newmap != osdc->osdmap) {
1774 ceph_osdmap_destroy(osdc->osdmap);
1775 osdc->osdmap = newmap;
1776 }
38d6453c 1777 kick_requests(osdc, 0);
f24e9980
SW		 1778 } else {
1779 dout("ignoring incremental map %u len %d\n",
1780 epoch, maplen);
1781 }
1782 p = next;
1783 nr_maps--;
1784 }
1785 if (newmap)
1786 goto done;
1787
1788 /* full maps */
1789 ceph_decode_32_safe(&p, end, nr_maps, bad);
1790 dout(" %d full maps\n", nr_maps);
1791 while (nr_maps) {
1792 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 1793 epoch = ceph_decode_32(&p);
1794 maplen = ceph_decode_32(&p);
f24e9980
SW		 1795 ceph_decode_need(&p, end, maplen, bad);
1796 if (nr_maps > 1) {
1797 dout("skipping non-latest full map %u len %d\n",
1798 epoch, maplen);
1799 } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
1800 dout("skipping full map %u len %d, "
1801 "older than our %u\n", epoch, maplen,
1802 osdc->osdmap->epoch);
1803 } else {
38d6453c
SW		 1804 int skipped_map = 0;
1805
f24e9980
SW		 1806 dout("taking full map %u len %d\n", epoch, maplen);
1807 newmap = osdmap_decode(&p, p+maplen);
1808 if (IS_ERR(newmap)) {
1809 err = PTR_ERR(newmap);
1810 goto bad;
1811 }
30dc6381 1812 BUG_ON(!newmap);
f24e9980
SW		 1813 oldmap = osdc->osdmap;
1814 osdc->osdmap = newmap;
38d6453c
SW		 1815 if (oldmap) {
1816 if (oldmap->epoch + 1 < newmap->epoch)
1817 skipped_map = 1;
f24e9980 1818 ceph_osdmap_destroy(oldmap);
38d6453c
SW		 1819 }
1820 kick_requests(osdc, skipped_map);
f24e9980
SW		 1821 }
1822 p += maplen;
1823 nr_maps--;
1824 }
1825
b72e19b9
DC		 1826 if (!osdc->osdmap)
1827 goto bad;
f24e9980
SW		 1828done:
1829 downgrade_write(&osdc->map_sem);
1830 ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch);
cd634fb6
SW		 1831
1832 /*
1833 * subscribe to subsequent osdmap updates if full to ensure
1834 * we find out when we are no longer full and stop returning
1835 * ENOSPC.
1836 */
d29adb34
JD		 1837 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) ||
1838 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD) ||
1839 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR))
cd634fb6
SW		 1840 ceph_monc_request_next_osdmap(&osdc->client->monc);
1841
f9d25199
AE		 1842 mutex_lock(&osdc->request_mutex);
1843 __send_queued(osdc);
1844 mutex_unlock(&osdc->request_mutex);
f24e9980 1845 up_read(&osdc->map_sem);
03066f23 1846 wake_up_all(&osdc->client->auth_wq);
f24e9980
SW		 1847 return;
1848
1849bad:
1850 pr_err("osdc handle_map corrupt msg\n");
9ec7cab1 1851 ceph_msg_dump(msg);
f24e9980
SW		 1852 up_write(&osdc->map_sem);
1853 return;
1854}
1855
a40c4f10
YS		 1856/*
1857 * watch/notify callback event infrastructure
1858 *
1859 * These callbacks are used both for watch and notify operations.
1860 */
1861static void __release_event(struct kref *kref)
1862{
1863 struct ceph_osd_event *event =
1864 container_of(kref, struct ceph_osd_event, kref);
1865
1866 dout("__release_event %p\n", event);
1867 kfree(event);
1868}
1869
1870static void get_event(struct ceph_osd_event *event)
1871{
1872 kref_get(&event->kref);
1873}
1874
1875void ceph_osdc_put_event(struct ceph_osd_event *event)
1876{
1877 kref_put(&event->kref, __release_event);
1878}
1879EXPORT_SYMBOL(ceph_osdc_put_event);
1880
1881static void __insert_event(struct ceph_osd_client *osdc,
1882 struct ceph_osd_event *new)
1883{
1884 struct rb_node **p = &osdc->event_tree.rb_node;
1885 struct rb_node *parent = NULL;
1886 struct ceph_osd_event *event = NULL;
1887
1888 while (*p) {
1889 parent = *p;
1890 event = rb_entry(parent, struct ceph_osd_event, node);
1891 if (new->cookie < event->cookie)
1892 p = &(*p)->rb_left;
1893 else if (new->cookie > event->cookie)
1894 p = &(*p)->rb_right;
1895 else
1896 BUG();
1897 }
1898
1899 rb_link_node(&new->node, parent, p);
1900 rb_insert_color(&new->node, &osdc->event_tree);
1901}
1902
1903static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
1904 u64 cookie)
1905{
1906 struct rb_node **p = &osdc->event_tree.rb_node;
1907 struct rb_node *parent = NULL;
1908 struct ceph_osd_event *event = NULL;
1909
1910 while (*p) {
1911 parent = *p;
1912 event = rb_entry(parent, struct ceph_osd_event, node);
1913 if (cookie < event->cookie)
1914 p = &(*p)->rb_left;
1915 else if (cookie > event->cookie)
1916 p = &(*p)->rb_right;
1917 else
1918 return event;
1919 }
1920 return NULL;
1921}
1922
1923static void __remove_event(struct ceph_osd_event *event)
1924{
1925 struct ceph_osd_client *osdc = event->osdc;
1926
1927 if (!RB_EMPTY_NODE(&event->node)) {
1928 dout("__remove_event removed %p\n", event);
1929 rb_erase(&event->node, &osdc->event_tree);
1930 ceph_osdc_put_event(event);
1931 } else {
1932 dout("__remove_event didn't remove %p\n", event);
1933 }
1934}
1935
1936int ceph_osdc_create_event(struct ceph_osd_client *osdc,
1937 void (*event_cb)(u64, u64, u8, void *),
3c663bbd 1938 void *data, struct ceph_osd_event **pevent)
a40c4f10
YS		 1939{
1940 struct ceph_osd_event *event;
1941
1942 event = kmalloc(sizeof(*event), GFP_NOIO);
1943 if (!event)
1944 return -ENOMEM;
1945
1946 dout("create_event %p\n", event);
1947 event->cb = event_cb;
3c663bbd 1948 event->one_shot = 0;
a40c4f10
YS		 1949 event->data = data;
1950 event->osdc = osdc;
1951 INIT_LIST_HEAD(&event->osd_node);
3ee5234d 1952 RB_CLEAR_NODE(&event->node);
a40c4f10
YS		 1953 kref_init(&event->kref); /* one ref for us */
1954 kref_get(&event->kref); /* one ref for the caller */
a40c4f10
YS		 1955
1956 spin_lock(&osdc->event_lock);
1957 event->cookie = ++osdc->event_count;
1958 __insert_event(osdc, event);
1959 spin_unlock(&osdc->event_lock);
1960
1961 *pevent = event;
1962 return 0;
1963}
1964EXPORT_SYMBOL(ceph_osdc_create_event);
1965
1966void ceph_osdc_cancel_event(struct ceph_osd_event *event)
1967{
1968 struct ceph_osd_client *osdc = event->osdc;
1969
1970 dout("cancel_event %p\n", event);
1971 spin_lock(&osdc->event_lock);
1972 __remove_event(event);
1973 spin_unlock(&osdc->event_lock);
1974 ceph_osdc_put_event(event); /* caller's */
1975}
1976EXPORT_SYMBOL(ceph_osdc_cancel_event);
1977
1978
1979static void do_event_work(struct work_struct *work)
1980{
1981 struct ceph_osd_event_work *event_work =
1982 container_of(work, struct ceph_osd_event_work, work);
1983 struct ceph_osd_event *event = event_work->event;
1984 u64 ver = event_work->ver;
1985 u64 notify_id = event_work->notify_id;
1986 u8 opcode = event_work->opcode;
1987
1988 dout("do_event_work completing %p\n", event);
1989 event->cb(ver, notify_id, opcode, event->data);
a40c4f10
YS		 1990 dout("do_event_work completed %p\n", event);
1991 ceph_osdc_put_event(event);
1992 kfree(event_work);
1993}
1994
1995
1996/*
1997 * Process osd watch notifications
1998 */
3c663bbd
AE		 1999static void handle_watch_notify(struct ceph_osd_client *osdc,
2000 struct ceph_msg *msg)
a40c4f10
YS		 2001{
2002 void *p, *end;
2003 u8 proto_ver;
2004 u64 cookie, ver, notify_id;
2005 u8 opcode;
2006 struct ceph_osd_event *event;
2007 struct ceph_osd_event_work *event_work;
2008
2009 p = msg->front.iov_base;
2010 end = p + msg->front.iov_len;
2011
2012 ceph_decode_8_safe(&p, end, proto_ver, bad);
2013 ceph_decode_8_safe(&p, end, opcode, bad);
2014 ceph_decode_64_safe(&p, end, cookie, bad);
2015 ceph_decode_64_safe(&p, end, ver, bad);
2016 ceph_decode_64_safe(&p, end, notify_id, bad);
2017
2018 spin_lock(&osdc->event_lock);
2019 event = __find_event(osdc, cookie);
2020 if (event) {
3c663bbd 2021 BUG_ON(event->one_shot);
a40c4f10 2022 get_event(event);
a40c4f10
YS		 2023 }
2024 spin_unlock(&osdc->event_lock);
2025 dout("handle_watch_notify cookie %lld ver %lld event %p\n",
2026 cookie, ver, event);
2027 if (event) {
2028 event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
a40c4f10
YS		 2029 if (!event_work) {
2030 dout("ERROR: could not allocate event_work\n");
2031 goto done_err;
2032 }
6b0ae409 2033 INIT_WORK(&event_work->work, do_event_work);
a40c4f10
YS		 2034 event_work->event = event;
2035 event_work->ver = ver;
2036 event_work->notify_id = notify_id;
2037 event_work->opcode = opcode;
2038 if (!queue_work(osdc->notify_wq, &event_work->work)) {
2039 dout("WARNING: failed to queue notify event work\n");
2040 goto done_err;
2041 }
2042 }
2043
2044 return;
2045
2046done_err:
a40c4f10
YS		 2047 ceph_osdc_put_event(event);
2048 return;
2049
2050bad:
2051 pr_err("osdc handle_watch_notify corrupt msg\n");
2052 return;
2053}
2054
e65550fd
AE		 2055/*
2056 * build new request AND message
2057 *
2058 */
2059void ceph_osdc_build_request(struct ceph_osd_request *req, u64 off,
2060 struct ceph_snap_context *snapc, u64 snap_id,
2061 struct timespec *mtime)
2062{
2063 struct ceph_msg *msg = req->r_request;
2064 void *p;
2065 size_t msg_size;
2066 int flags = req->r_flags;
2067 u64 data_len;
2068 unsigned int i;
2069
2070 req->r_snapid = snap_id;
2071 req->r_snapc = ceph_get_snap_context(snapc);
2072
2073 /* encode request */
2074 msg->hdr.version = cpu_to_le16(4);
2075
2076 p = msg->front.iov_base;
2077 ceph_encode_32(&p, 1); /* client_inc is always 1 */
2078 req->r_request_osdmap_epoch = p;
2079 p += 4;
2080 req->r_request_flags = p;
2081 p += 4;
2082 if (req->r_flags & CEPH_OSD_FLAG_WRITE)
2083 ceph_encode_timespec(p, mtime);
2084 p += sizeof(struct ceph_timespec);
2085 req->r_request_reassert_version = p;
2086 p += sizeof(struct ceph_eversion); /* will get filled in */
2087
2088 /* oloc */
2089 ceph_encode_8(&p, 4);
2090 ceph_encode_8(&p, 4);
2091 ceph_encode_32(&p, 8 + 4 + 4);
2092 req->r_request_pool = p;
2093 p += 8;
2094 ceph_encode_32(&p, -1); /* preferred */
2095 ceph_encode_32(&p, 0); /* key len */
2096
2097 ceph_encode_8(&p, 1);
2098 req->r_request_pgid = p;
2099 p += 8 + 4;
2100 ceph_encode_32(&p, -1); /* preferred */
2101
2102 /* oid */
2103 ceph_encode_32(&p, req->r_oid_len);
2104 memcpy(p, req->r_oid, req->r_oid_len);
2105 dout("oid '%.*s' len %d\n", req->r_oid_len, req->r_oid, req->r_oid_len);
2106 p += req->r_oid_len;
2107
2108 /* ops--can imply data */
2109 ceph_encode_16(&p, (u16)req->r_num_ops);
2110 data_len = 0;
2111 for (i = 0; i < req->r_num_ops; i++) {
2112 data_len += osd_req_encode_op(req, p, i);
2113 p += sizeof(struct ceph_osd_op);
2114 }
2115
2116 /* snaps */
2117 ceph_encode_64(&p, req->r_snapid);
2118 ceph_encode_64(&p, req->r_snapc ? req->r_snapc->seq : 0);
2119 ceph_encode_32(&p, req->r_snapc ? req->r_snapc->num_snaps : 0);
2120 if (req->r_snapc) {
2121 for (i = 0; i < snapc->num_snaps; i++) {
2122 ceph_encode_64(&p, req->r_snapc->snaps[i]);
2123 }
2124 }
2125
2126 req->r_request_attempts = p;
2127 p += 4;
2128
2129 /* data */
2130 if (flags & CEPH_OSD_FLAG_WRITE) {
2131 u16 data_off;
2132
2133 /*
2134 * The header "data_off" is a hint to the receiver
2135 * allowing it to align received data into its
2136 * buffers such that there's no need to re-copy
2137 * it before writing it to disk (direct I/O).
2138 */
2139 data_off = (u16) (off & 0xffff);
2140 req->r_request->hdr.data_off = cpu_to_le16(data_off);
2141 }
2142 req->r_request->hdr.data_len = cpu_to_le32(data_len);
2143
2144 BUG_ON(p > msg->front.iov_base + msg->front.iov_len);
2145 msg_size = p - msg->front.iov_base;
2146 msg->front.iov_len = msg_size;
2147 msg->hdr.front_len = cpu_to_le32(msg_size);
2148
2149 dout("build_request msg_size was %d\n", (int)msg_size);
2150}
2151EXPORT_SYMBOL(ceph_osdc_build_request);
2152
70636773
AE		 2153/*
2154 * Register request, send initial attempt.
2155 */
2156int ceph_osdc_start_request(struct ceph_osd_client *osdc,
2157 struct ceph_osd_request *req,
2158 bool nofail)
2159{
2160 int rc = 0;
2161
f24e9980
SW		 2162 down_read(&osdc->map_sem);
2163 mutex_lock(&osdc->request_mutex);
dc4b870c 2164 __register_request(osdc, req);
c10ebbf5
AE		 2165 req->r_sent = 0;
2166 req->r_got_reply = 0;
92451b49
AE		 2167 rc = __map_request(osdc, req, 0);
2168 if (rc < 0) {
2169 if (nofail) {
2170 dout("osdc_start_request failed map, "
2171 " will retry %lld\n", req->r_tid);
2172 rc = 0;
73d9f7ee 2173 } else {
2174 __unregister_request(osdc, req);
f24e9980 2175 }
92451b49 2176 goto out_unlock;
f24e9980 2177 }
92451b49
AE		 2178 if (req->r_osd == NULL) {
2179 dout("send_request %p no up osds in pg\n", req);
2180 ceph_monc_request_next_osdmap(&osdc->client->monc);
2181 } else {
7e2766a1 2182 __send_queued(osdc);
92451b49
AE		 2183 }
2184 rc = 0;
234af26f 2185out_unlock:
f24e9980
SW		 2186 mutex_unlock(&osdc->request_mutex);
2187 up_read(&osdc->map_sem);
2188 return rc;
2189}
3d14c5d2 2190EXPORT_SYMBOL(ceph_osdc_start_request);
f24e9980
SW		 2191
2192/*
2193 * wait for a request to complete
2194 */
2195int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
2196 struct ceph_osd_request *req)
2197{
2198 int rc;
2199
2200 rc = wait_for_completion_interruptible(&req->r_completion);
2201 if (rc < 0) {
2202 mutex_lock(&osdc->request_mutex);
2203 __cancel_request(req);
529cfcc4 2204 __unregister_request(osdc, req);
f24e9980 2205 mutex_unlock(&osdc->request_mutex);
25845472 2206 complete_request(req);
529cfcc4 2207 dout("wait_request tid %llu canceled/timed out\n", req->r_tid);
f24e9980
SW		 2208 return rc;
2209 }
2210
2211 dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result);
2212 return req->r_result;
2213}
3d14c5d2 2214EXPORT_SYMBOL(ceph_osdc_wait_request);
f24e9980
SW		 2215
2216/*
2217 * sync - wait for all in-flight requests to flush. avoid starvation.
2218 */
2219void ceph_osdc_sync(struct ceph_osd_client *osdc)
2220{
2221 struct ceph_osd_request *req;
2222 u64 last_tid, next_tid = 0;
2223
2224 mutex_lock(&osdc->request_mutex);
2225 last_tid = osdc->last_tid;
2226 while (1) {
2227 req = __lookup_request_ge(osdc, next_tid);
2228 if (!req)
2229 break;
2230 if (req->r_tid > last_tid)
2231 break;
2232
2233 next_tid = req->r_tid + 1;
2234 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
2235 continue;
2236
2237 ceph_osdc_get_request(req);
2238 mutex_unlock(&osdc->request_mutex);
2239 dout("sync waiting on tid %llu (last is %llu)\n",
2240 req->r_tid, last_tid);
2241 wait_for_completion(&req->r_safe_completion);
2242 mutex_lock(&osdc->request_mutex);
2243 ceph_osdc_put_request(req);
2244 }
2245 mutex_unlock(&osdc->request_mutex);
2246 dout("sync done (thru tid %llu)\n", last_tid);
2247}
3d14c5d2 2248EXPORT_SYMBOL(ceph_osdc_sync);
f24e9980 2249
dd935f44
JD		 2250/*
2251 * Call all pending notify callbacks - for use after a watch is
2252 * unregistered, to make sure no more callbacks for it will be invoked
2253 */
2254extern void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc)
2255{
2256 flush_workqueue(osdc->notify_wq);
2257}
2258EXPORT_SYMBOL(ceph_osdc_flush_notifies);
2259
2260
f24e9980
SW		 2261/*
2262 * init, shutdown
2263 */
2264int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
2265{
2266 int err;
2267
2268 dout("init\n");
2269 osdc->client = client;
2270 osdc->osdmap = NULL;
2271 init_rwsem(&osdc->map_sem);
2272 init_completion(&osdc->map_waiters);
2273 osdc->last_requested_map = 0;
2274 mutex_init(&osdc->request_mutex);
f24e9980
SW		 2275 osdc->last_tid = 0;
2276 osdc->osds = RB_ROOT;
f5a2041b 2277 INIT_LIST_HEAD(&osdc->osd_lru);
f24e9980 2278 osdc->requests = RB_ROOT;
422d2cb8 2279 INIT_LIST_HEAD(&osdc->req_lru);
6f6c7006
SW		 2280 INIT_LIST_HEAD(&osdc->req_unsent);
2281 INIT_LIST_HEAD(&osdc->req_notarget);
a40c4f10 2282 INIT_LIST_HEAD(&osdc->req_linger);
f24e9980
SW		 2283 osdc->num_requests = 0;
2284 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
f5a2041b 2285 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
a40c4f10
YS		 2286 spin_lock_init(&osdc->event_lock);
2287 osdc->event_tree = RB_ROOT;
2288 osdc->event_count = 0;
f5a2041b
YS		 2289
2290 schedule_delayed_work(&osdc->osds_timeout_work,
3d14c5d2 2291 round_jiffies_relative(osdc->client->options->osd_idle_ttl * HZ));
f24e9980 2292
5f44f142 2293 err = -ENOMEM;
f24e9980
SW		 2294 osdc->req_mempool = mempool_create_kmalloc_pool(10,
2295 sizeof(struct ceph_osd_request));
2296 if (!osdc->req_mempool)
5f44f142 2297 goto out;
f24e9980 2298
d50b409f
SW		 2299 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
2300 OSD_OP_FRONT_LEN, 10, true,
4f48280e 2301 "osd_op");
f24e9980 2302 if (err < 0)
5f44f142 2303 goto out_mempool;
d50b409f 2304 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
4f48280e
SW		 2305 OSD_OPREPLY_FRONT_LEN, 10, true,
2306 "osd_op_reply");
c16e7869
SW		 2307 if (err < 0)
2308 goto out_msgpool;
a40c4f10 2309
dbcae088 2310 err = -ENOMEM;
a40c4f10 2311 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
dbcae088 2312 if (!osdc->notify_wq)
a40c4f10 2313 goto out_msgpool;
f24e9980 2314 return 0;
5f44f142 2315
c16e7869
SW		 2316out_msgpool:
2317 ceph_msgpool_destroy(&osdc->msgpool_op);
5f44f142
SW		 2318out_mempool:
2319 mempool_destroy(osdc->req_mempool);
2320out:
2321 return err;
f24e9980
SW		 2322}
2323
2324void ceph_osdc_stop(struct ceph_osd_client *osdc)
2325{
a40c4f10
YS		 2326 flush_workqueue(osdc->notify_wq);
2327 destroy_workqueue(osdc->notify_wq);
f24e9980 2328 cancel_delayed_work_sync(&osdc->timeout_work);
f5a2041b 2329 cancel_delayed_work_sync(&osdc->osds_timeout_work);
f24e9980
SW		 2330 if (osdc->osdmap) {
2331 ceph_osdmap_destroy(osdc->osdmap);
2332 osdc->osdmap = NULL;
2333 }
aca420bc 2334 remove_all_osds(osdc);
f24e9980
SW		 2335 mempool_destroy(osdc->req_mempool);
2336 ceph_msgpool_destroy(&osdc->msgpool_op);
c16e7869 2337 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
f24e9980
SW		 2338}
2339
2340/*
2341 * Read some contiguous pages. If we cross a stripe boundary, shorten
2342 * *plen. Return number of bytes read, or error.
2343 */
2344int ceph_osdc_readpages(struct ceph_osd_client *osdc,
2345 struct ceph_vino vino, struct ceph_file_layout *layout,
2346 u64 off, u64 *plen,
2347 u32 truncate_seq, u64 truncate_size,
b7495fc2 2348 struct page **pages, int num_pages, int page_align)
f24e9980
SW		 2349{
2350 struct ceph_osd_request *req;
2351 int rc = 0;
2352
2353 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
2354 vino.snap, off, *plen);
79528734 2355 req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 1,
f24e9980 2356 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
acead002 2357 NULL, truncate_seq, truncate_size,
153e5167 2358 false);
6816282d
SW		 2359 if (IS_ERR(req))
2360 return PTR_ERR(req);
f24e9980
SW		 2361
2362 /* it may be a short read due to an object boundary */
0fff87ec 2363
406e2c9f 2364 osd_req_op_extent_osd_data_pages(req, 0,
a4ce40a9 2365 pages, *plen, page_align, false, false);
f24e9980 2366
e0c59487 2367 dout("readpages final extent is %llu~%llu (%llu bytes align %d)\n",
43bfe5de 2368 off, *plen, *plen, page_align);
f24e9980 2369
79528734 2370 ceph_osdc_build_request(req, off, NULL, vino.snap, NULL);
02ee07d3 2371
f24e9980
SW		 2372 rc = ceph_osdc_start_request(osdc, req, false);
2373 if (!rc)
2374 rc = ceph_osdc_wait_request(osdc, req);
2375
2376 ceph_osdc_put_request(req);
2377 dout("readpages result %d\n", rc);
2378 return rc;
2379}
3d14c5d2 2380EXPORT_SYMBOL(ceph_osdc_readpages);
f24e9980
SW		 2381
2382/*
2383 * do a synchronous write on N pages
2384 */
2385int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
2386 struct ceph_file_layout *layout,
2387 struct ceph_snap_context *snapc,
2388 u64 off, u64 len,
2389 u32 truncate_seq, u64 truncate_size,
2390 struct timespec *mtime,
24808826 2391 struct page **pages, int num_pages)
f24e9980
SW		 2392{
2393 struct ceph_osd_request *req;
2394 int rc = 0;
b7495fc2 2395 int page_align = off & ~PAGE_MASK;
f24e9980 2396
acead002 2397 BUG_ON(vino.snap != CEPH_NOSNAP); /* snapshots aren't writeable */
79528734 2398 req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 1,
f24e9980 2399 CEPH_OSD_OP_WRITE,
24808826 2400 CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
acead002 2401 snapc, truncate_seq, truncate_size,
153e5167 2402 true);
6816282d
SW		 2403 if (IS_ERR(req))
2404 return PTR_ERR(req);
f24e9980
SW		 2405
2406 /* it may be a short write due to an object boundary */
406e2c9f 2407 osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align,
43bfe5de
AE		 2408 false, false);
2409 dout("writepages %llu~%llu (%llu bytes)\n", off, len, len);
f24e9980 2410
79528734 2411 ceph_osdc_build_request(req, off, snapc, CEPH_NOSNAP, mtime);
02ee07d3 2412
87f979d3 2413 rc = ceph_osdc_start_request(osdc, req, true);
f24e9980
SW		 2414 if (!rc)
2415 rc = ceph_osdc_wait_request(osdc, req);
2416
2417 ceph_osdc_put_request(req);
2418 if (rc == 0)
2419 rc = len;
2420 dout("writepages result %d\n", rc);
2421 return rc;
2422}
3d14c5d2 2423EXPORT_SYMBOL(ceph_osdc_writepages);
f24e9980 2424
5522ae0b
AE		 2425int ceph_osdc_setup(void)
2426{
2427 BUG_ON(ceph_osd_request_cache);
2428 ceph_osd_request_cache = kmem_cache_create("ceph_osd_request",
2429 sizeof (struct ceph_osd_request),
2430 __alignof__(struct ceph_osd_request),
2431 0, NULL);
2432
2433 return ceph_osd_request_cache ? 0 : -ENOMEM;
2434}
2435EXPORT_SYMBOL(ceph_osdc_setup);
2436
2437void ceph_osdc_cleanup(void)
2438{
2439 BUG_ON(!ceph_osd_request_cache);
2440 kmem_cache_destroy(ceph_osd_request_cache);
2441 ceph_osd_request_cache = NULL;
2442}
2443EXPORT_SYMBOL(ceph_osdc_cleanup);
2444
f24e9980
SW		 2445/*
2446 * handle incoming message
2447 */
2448static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
2449{
2450 struct ceph_osd *osd = con->private;
32c895e7 2451 struct ceph_osd_client *osdc;
f24e9980
SW		 2452 int type = le16_to_cpu(msg->hdr.type);
2453
2454 if (!osd)
4a32f93d 2455 goto out;
32c895e7 2456 osdc = osd->o_osdc;
f24e9980
SW		 2457
2458 switch (type) {
2459 case CEPH_MSG_OSD_MAP:
2460 ceph_osdc_handle_map(osdc, msg);
2461 break;
2462 case CEPH_MSG_OSD_OPREPLY:
350b1c32 2463 handle_reply(osdc, msg, con);
f24e9980 2464 break;
a40c4f10
YS		 2465 case CEPH_MSG_WATCH_NOTIFY:
2466 handle_watch_notify(osdc, msg);
2467 break;
f24e9980
SW		 2468
2469 default:
2470 pr_err("received unknown message type %d %s\n", type,
2471 ceph_msg_type_name(type));
2472 }
4a32f93d 2473out:
f24e9980
SW		 2474 ceph_msg_put(msg);
2475}
2476
5b3a4db3 2477/*
21b667f6
SW		 2478 * lookup and return message for incoming reply. set up reply message
2479 * pages.
5b3a4db3
SW		 2480 */
2481static struct ceph_msg *get_reply(struct ceph_connection *con,
2450418c
YS		 2482 struct ceph_msg_header *hdr,
2483 int *skip)
f24e9980
SW		 2484{
2485 struct ceph_osd *osd = con->private;
2486 struct ceph_osd_client *osdc = osd->o_osdc;
2450418c 2487 struct ceph_msg *m;
0547a9b3 2488 struct ceph_osd_request *req;
5b3a4db3
SW		 2489 int front = le32_to_cpu(hdr->front_len);
2490 int data_len = le32_to_cpu(hdr->data_len);
0547a9b3 2491 u64 tid;
f24e9980 2492
0547a9b3
YS		 2493 tid = le64_to_cpu(hdr->tid);
2494 mutex_lock(&osdc->request_mutex);
2495 req = __lookup_request(osdc, tid);
2496 if (!req) {
2497 *skip = 1;
2498 m = NULL;
756a16a5
SW		 2499 dout("get_reply unknown tid %llu from osd%d\n", tid,
2500 osd->o_osd);
0547a9b3
YS		 2501 goto out;
2502 }
c16e7869 2503
ace6d3a9 2504 if (req->r_reply->con)
8921d114 2505 dout("%s revoking msg %p from old con %p\n", __func__,
ace6d3a9
AE		 2506 req->r_reply, req->r_reply->con);
2507 ceph_msg_revoke_incoming(req->r_reply);
0547a9b3 2508
c16e7869 2509 if (front > req->r_reply->front.iov_len) {
4974341e
AE		 2510 pr_warning("get_reply front %d > preallocated %d (%u#%llu)\n",
2511 front, (int)req->r_reply->front.iov_len,
2512 (unsigned int)con->peer_name.type,
2513 le64_to_cpu(con->peer_name.num));
b61c2763 2514 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front, GFP_NOFS, false);
a79832f2 2515 if (!m)
c16e7869
SW		 2516 goto out;
2517 ceph_msg_put(req->r_reply);
2518 req->r_reply = m;
2519 }
2520 m = ceph_msg_get(req->r_reply);
2521
0547a9b3 2522 if (data_len > 0) {
a4ce40a9 2523 struct ceph_osd_data *osd_data;
0fff87ec 2524
a4ce40a9
AE		 2525 /*
2526 * XXX This is assuming there is only one op containing
2527 * XXX page data. Probably OK for reads, but this
2528 * XXX ought to be done more generally.
2529 */
406e2c9f 2530 osd_data = osd_req_op_extent_osd_data(req, 0);
0fff87ec 2531 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
0fff87ec 2532 if (osd_data->pages &&
e0c59487 2533 unlikely(osd_data->length < data_len)) {
2ac2b7a6 2534
e0c59487
AE		 2535 pr_warning("tid %lld reply has %d bytes "
2536 "we had only %llu bytes ready\n",
2537 tid, data_len, osd_data->length);
2ac2b7a6
AE		 2538 *skip = 1;
2539 ceph_msg_put(m);
2540 m = NULL;
2541 goto out;
2542 }
2ac2b7a6 2543 }
0547a9b3 2544 }
5b3a4db3 2545 *skip = 0;
c16e7869 2546 dout("get_reply tid %lld %p\n", tid, m);
0547a9b3
YS		 2547
2548out:
2549 mutex_unlock(&osdc->request_mutex);
2450418c 2550 return m;
5b3a4db3
SW		 2551
2552}
2553
2554static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2555 struct ceph_msg_header *hdr,
2556 int *skip)
2557{
2558 struct ceph_osd *osd = con->private;
2559 int type = le16_to_cpu(hdr->type);
2560 int front = le32_to_cpu(hdr->front_len);
2561
1c20f2d2 2562 *skip = 0;
5b3a4db3
SW		 2563 switch (type) {
2564 case CEPH_MSG_OSD_MAP:
a40c4f10 2565 case CEPH_MSG_WATCH_NOTIFY:
b61c2763 2566 return ceph_msg_new(type, front, GFP_NOFS, false);
5b3a4db3
SW		 2567 case CEPH_MSG_OSD_OPREPLY:
2568 return get_reply(con, hdr, skip);
2569 default:
2570 pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
2571 osd->o_osd);
2572 *skip = 1;
2573 return NULL;
2574 }
f24e9980
SW		 2575}
2576
2577/*
2578 * Wrappers to refcount containing ceph_osd struct
2579 */
2580static struct ceph_connection *get_osd_con(struct ceph_connection *con)
2581{
2582 struct ceph_osd *osd = con->private;
2583 if (get_osd(osd))
2584 return con;
2585 return NULL;
2586}
2587
2588static void put_osd_con(struct ceph_connection *con)
2589{
2590 struct ceph_osd *osd = con->private;
2591 put_osd(osd);
2592}
2593
4e7a5dcd
SW		 2594/*
2595 * authentication
2596 */
a3530df3
AE		 2597/*
2598 * Note: returned pointer is the address of a structure that's
2599 * managed separately. Caller must *not* attempt to free it.
2600 */
2601static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
8f43fb53 2602 int *proto, int force_new)
4e7a5dcd
SW		 2603{
2604 struct ceph_osd *o = con->private;
2605 struct ceph_osd_client *osdc = o->o_osdc;
2606 struct ceph_auth_client *ac = osdc->client->monc.auth;
74f1869f 2607 struct ceph_auth_handshake *auth = &o->o_auth;
4e7a5dcd 2608
74f1869f 2609 if (force_new && auth->authorizer) {
27859f97 2610 ceph_auth_destroy_authorizer(ac, auth->authorizer);
74f1869f
AE		 2611 auth->authorizer = NULL;
2612 }
27859f97
SW		 2613 if (!auth->authorizer) {
2614 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2615 auth);
4e7a5dcd 2616 if (ret)
a3530df3 2617 return ERR_PTR(ret);
27859f97
SW		 2618 } else {
2619 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
0bed9b5c
SW		 2620 auth);
2621 if (ret)
2622 return ERR_PTR(ret);
4e7a5dcd 2623 }
4e7a5dcd 2624 *proto = ac->protocol;
74f1869f 2625
a3530df3 2626 return auth;
4e7a5dcd
SW		 2627}
2628
2629
2630static int verify_authorizer_reply(struct ceph_connection *con, int len)
2631{
2632 struct ceph_osd *o = con->private;
2633 struct ceph_osd_client *osdc = o->o_osdc;
2634 struct ceph_auth_client *ac = osdc->client->monc.auth;
2635
27859f97 2636 return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer, len);
4e7a5dcd
SW		 2637}
2638
9bd2e6f8
SW		 2639static int invalidate_authorizer(struct ceph_connection *con)
2640{
2641 struct ceph_osd *o = con->private;
2642 struct ceph_osd_client *osdc = o->o_osdc;
2643 struct ceph_auth_client *ac = osdc->client->monc.auth;
2644
27859f97 2645 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
9bd2e6f8
SW		 2646 return ceph_monc_validate_auth(&osdc->client->monc);
2647}
4e7a5dcd 2648
9e32789f 2649static const struct ceph_connection_operations osd_con_ops = {
f24e9980
SW		 2650 .get = get_osd_con,
2651 .put = put_osd_con,
2652 .dispatch = dispatch,
4e7a5dcd
SW		 2653 .get_authorizer = get_authorizer,
2654 .verify_authorizer_reply = verify_authorizer_reply,
9bd2e6f8 2655 .invalidate_authorizer = invalidate_authorizer,
f24e9980 2656 .alloc_msg = alloc_msg,
81b024e7 2657 .fault = osd_reset,
f24e9980 2658};
Linux 6.x block layer and io_uring tree(s)