projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
libceph: rename ceph_osd_request::r_{oloc,oid} to r_base_{oloc,oid}
[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 */
2d0ebc5d 341 msg_size += 4 + CEPH_MAX_OID_NAME_LEN; /* oid */
1b83bef2
SW		 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
3c972c95 371 req->r_base_oloc.pool = -1;
22116525 372
c16e7869
SW		 373 /* create reply message */
374 if (use_mempool)
375 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0);
376 else
377 msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY,
b61c2763 378 OSD_OPREPLY_FRONT_LEN, gfp_flags, true);
a79832f2 379 if (!msg) {
c16e7869 380 ceph_osdc_put_request(req);
a79832f2 381 return NULL;
c16e7869
SW		 382 }
383 req->r_reply = msg;
384
385 /* create request message; allow space for oid */
f24e9980 386 if (use_mempool)
8f3bc053 387 msg = ceph_msgpool_get(&osdc->msgpool_op, 0);
f24e9980 388 else
b61c2763 389 msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp_flags, true);
a79832f2 390 if (!msg) {
f24e9980 391 ceph_osdc_put_request(req);
a79832f2 392 return NULL;
f24e9980 393 }
68b4476b 394
f24e9980 395 memset(msg->front.iov_base, 0, msg->front.iov_len);
3499e8a5
YS		 396
397 req->r_request = msg;
3499e8a5
YS		 398
399 return req;
400}
3d14c5d2 401EXPORT_SYMBOL(ceph_osdc_alloc_request);
3499e8a5 402
a8dd0a37 403static bool osd_req_opcode_valid(u16 opcode)
68b4476b 404{
a8dd0a37 405 switch (opcode) {
68b4476b 406 case CEPH_OSD_OP_READ:
a8dd0a37 407 case CEPH_OSD_OP_STAT:
4c46459c
AE		 408 case CEPH_OSD_OP_MAPEXT:
409 case CEPH_OSD_OP_MASKTRUNC:
410 case CEPH_OSD_OP_SPARSE_READ:
a9f36c3e 411 case CEPH_OSD_OP_NOTIFY:
a8dd0a37 412 case CEPH_OSD_OP_NOTIFY_ACK:
4c46459c 413 case CEPH_OSD_OP_ASSERT_VER:
a8dd0a37 414 case CEPH_OSD_OP_WRITE:
4c46459c
AE		 415 case CEPH_OSD_OP_WRITEFULL:
416 case CEPH_OSD_OP_TRUNCATE:
417 case CEPH_OSD_OP_ZERO:
418 case CEPH_OSD_OP_DELETE:
419 case CEPH_OSD_OP_APPEND:
a8dd0a37 420 case CEPH_OSD_OP_STARTSYNC:
4c46459c
AE		 421 case CEPH_OSD_OP_SETTRUNC:
422 case CEPH_OSD_OP_TRIMTRUNC:
423 case CEPH_OSD_OP_TMAPUP:
424 case CEPH_OSD_OP_TMAPPUT:
425 case CEPH_OSD_OP_TMAPGET:
426 case CEPH_OSD_OP_CREATE:
a9f36c3e 427 case CEPH_OSD_OP_ROLLBACK:
a8dd0a37 428 case CEPH_OSD_OP_WATCH:
4c46459c
AE		 429 case CEPH_OSD_OP_OMAPGETKEYS:
430 case CEPH_OSD_OP_OMAPGETVALS:
431 case CEPH_OSD_OP_OMAPGETHEADER:
432 case CEPH_OSD_OP_OMAPGETVALSBYKEYS:
4c46459c
AE		 433 case CEPH_OSD_OP_OMAPSETVALS:
434 case CEPH_OSD_OP_OMAPSETHEADER:
435 case CEPH_OSD_OP_OMAPCLEAR:
436 case CEPH_OSD_OP_OMAPRMKEYS:
437 case CEPH_OSD_OP_OMAP_CMP:
438 case CEPH_OSD_OP_CLONERANGE:
439 case CEPH_OSD_OP_ASSERT_SRC_VERSION:
440 case CEPH_OSD_OP_SRC_CMPXATTR:
a9f36c3e 441 case CEPH_OSD_OP_GETXATTR:
4c46459c 442 case CEPH_OSD_OP_GETXATTRS:
a9f36c3e
AE		 443 case CEPH_OSD_OP_CMPXATTR:
444 case CEPH_OSD_OP_SETXATTR:
4c46459c
AE		 445 case CEPH_OSD_OP_SETXATTRS:
446 case CEPH_OSD_OP_RESETXATTRS:
447 case CEPH_OSD_OP_RMXATTR:
448 case CEPH_OSD_OP_PULL:
449 case CEPH_OSD_OP_PUSH:
450 case CEPH_OSD_OP_BALANCEREADS:
451 case CEPH_OSD_OP_UNBALANCEREADS:
452 case CEPH_OSD_OP_SCRUB:
453 case CEPH_OSD_OP_SCRUB_RESERVE:
454 case CEPH_OSD_OP_SCRUB_UNRESERVE:
455 case CEPH_OSD_OP_SCRUB_STOP:
456 case CEPH_OSD_OP_SCRUB_MAP:
457 case CEPH_OSD_OP_WRLOCK:
458 case CEPH_OSD_OP_WRUNLOCK:
459 case CEPH_OSD_OP_RDLOCK:
460 case CEPH_OSD_OP_RDUNLOCK:
461 case CEPH_OSD_OP_UPLOCK:
462 case CEPH_OSD_OP_DNLOCK:
a8dd0a37 463 case CEPH_OSD_OP_CALL:
4c46459c
AE		 464 case CEPH_OSD_OP_PGLS:
465 case CEPH_OSD_OP_PGLS_FILTER:
a8dd0a37
AE		 466 return true;
467 default:
468 return false;
469 }
470}
471
33803f33
AE		 472/*
473 * This is an osd op init function for opcodes that have no data or
474 * other information associated with them. It also serves as a
475 * common init routine for all the other init functions, below.
476 */
c99d2d4a 477static struct ceph_osd_req_op *
49719778 478_osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which,
c99d2d4a 479 u16 opcode)
33803f33 480{
c99d2d4a
AE		 481 struct ceph_osd_req_op *op;
482
483 BUG_ON(which >= osd_req->r_num_ops);
33803f33
AE		 484 BUG_ON(!osd_req_opcode_valid(opcode));
485
c99d2d4a 486 op = &osd_req->r_ops[which];
33803f33 487 memset(op, 0, sizeof (*op));
33803f33 488 op->op = opcode;
c99d2d4a
AE		 489
490 return op;
33803f33
AE		 491}
492
49719778
AE		 493void osd_req_op_init(struct ceph_osd_request *osd_req,
494 unsigned int which, u16 opcode)
495{
496 (void)_osd_req_op_init(osd_req, which, opcode);
497}
498EXPORT_SYMBOL(osd_req_op_init);
499
c99d2d4a
AE		 500void osd_req_op_extent_init(struct ceph_osd_request *osd_req,
501 unsigned int which, u16 opcode,
33803f33
AE		 502 u64 offset, u64 length,
503 u64 truncate_size, u32 truncate_seq)
504{
49719778 505 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
33803f33
AE		 506 size_t payload_len = 0;
507
ad7a60de
LW		 508 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
509 opcode != CEPH_OSD_OP_DELETE && opcode != CEPH_OSD_OP_ZERO &&
510 opcode != CEPH_OSD_OP_TRUNCATE);
33803f33 511
33803f33
AE		 512 op->extent.offset = offset;
513 op->extent.length = length;
514 op->extent.truncate_size = truncate_size;
515 op->extent.truncate_seq = truncate_seq;
516 if (opcode == CEPH_OSD_OP_WRITE)
517 payload_len += length;
518
519 op->payload_len = payload_len;
520}
521EXPORT_SYMBOL(osd_req_op_extent_init);
522
c99d2d4a
AE		 523void osd_req_op_extent_update(struct ceph_osd_request *osd_req,
524 unsigned int which, u64 length)
e5975c7c 525{
c99d2d4a
AE		 526 struct ceph_osd_req_op *op;
527 u64 previous;
528
529 BUG_ON(which >= osd_req->r_num_ops);
530 op = &osd_req->r_ops[which];
531 previous = op->extent.length;
e5975c7c
AE		 532
533 if (length == previous)
534 return; /* Nothing to do */
535 BUG_ON(length > previous);
536
537 op->extent.length = length;
538 op->payload_len -= previous - length;
539}
540EXPORT_SYMBOL(osd_req_op_extent_update);
541
c99d2d4a 542void osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
04017e29 543 u16 opcode, const char *class, const char *method)
33803f33 544{
49719778 545 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
5f562df5 546 struct ceph_pagelist *pagelist;
33803f33
AE		 547 size_t payload_len = 0;
548 size_t size;
549
550 BUG_ON(opcode != CEPH_OSD_OP_CALL);
551
5f562df5
AE		 552 pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
553 BUG_ON(!pagelist);
554 ceph_pagelist_init(pagelist);
555
33803f33
AE		 556 op->cls.class_name = class;
557 size = strlen(class);
558 BUG_ON(size > (size_t) U8_MAX);
559 op->cls.class_len = size;
5f562df5 560 ceph_pagelist_append(pagelist, class, size);
33803f33
AE		 561 payload_len += size;
562
563 op->cls.method_name = method;
564 size = strlen(method);
565 BUG_ON(size > (size_t) U8_MAX);
566 op->cls.method_len = size;
5f562df5 567 ceph_pagelist_append(pagelist, method, size);
33803f33
AE		 568 payload_len += size;
569
a4ce40a9 570 osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist);
5f562df5 571
33803f33
AE		 572 op->cls.argc = 0; /* currently unused */
573
574 op->payload_len = payload_len;
575}
576EXPORT_SYMBOL(osd_req_op_cls_init);
8c042b0d 577
c99d2d4a
AE		 578void osd_req_op_watch_init(struct ceph_osd_request *osd_req,
579 unsigned int which, u16 opcode,
33803f33
AE		 580 u64 cookie, u64 version, int flag)
581{
49719778 582 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
33803f33 583
c99d2d4a 584 BUG_ON(opcode != CEPH_OSD_OP_NOTIFY_ACK && opcode != CEPH_OSD_OP_WATCH);
33803f33
AE		 585
586 op->watch.cookie = cookie;
9ef1ee5a 587 op->watch.ver = version;
33803f33 588 if (opcode == CEPH_OSD_OP_WATCH && flag)
c99d2d4a 589 op->watch.flag = (u8)1;
33803f33
AE		 590}
591EXPORT_SYMBOL(osd_req_op_watch_init);
592
90af3602 593static void ceph_osdc_msg_data_add(struct ceph_msg *msg,
ec9123c5
AE		 594 struct ceph_osd_data *osd_data)
595{
596 u64 length = ceph_osd_data_length(osd_data);
597
598 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
599 BUG_ON(length > (u64) SIZE_MAX);
600 if (length)
90af3602 601 ceph_msg_data_add_pages(msg, osd_data->pages,
ec9123c5
AE		 602 length, osd_data->alignment);
603 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
604 BUG_ON(!length);
90af3602 605 ceph_msg_data_add_pagelist(msg, osd_data->pagelist);
ec9123c5
AE		 606#ifdef CONFIG_BLOCK
607 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
90af3602 608 ceph_msg_data_add_bio(msg, osd_data->bio, length);
ec9123c5
AE		 609#endif
610 } else {
611 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
612 }
613}
614
a8dd0a37 615static u64 osd_req_encode_op(struct ceph_osd_request *req,
79528734 616 struct ceph_osd_op *dst, unsigned int which)
a8dd0a37 617{
79528734 618 struct ceph_osd_req_op *src;
04017e29 619 struct ceph_osd_data *osd_data;
54d50649 620 u64 request_data_len = 0;
04017e29 621 u64 data_length;
a8dd0a37 622
79528734
AE		 623 BUG_ON(which >= req->r_num_ops);
624 src = &req->r_ops[which];
a8dd0a37
AE		 625 if (WARN_ON(!osd_req_opcode_valid(src->op))) {
626 pr_err("unrecognized osd opcode %d\n", src->op);
627
628 return 0;
629 }
630
631 switch (src->op) {
632 case CEPH_OSD_OP_STAT:
49719778
AE		 633 osd_data = &src->raw_data_in;
634 ceph_osdc_msg_data_add(req->r_reply, osd_data);
a8dd0a37
AE		 635 break;
636 case CEPH_OSD_OP_READ:
637 case CEPH_OSD_OP_WRITE:
ad7a60de
LW		 638 case CEPH_OSD_OP_ZERO:
639 case CEPH_OSD_OP_DELETE:
640 case CEPH_OSD_OP_TRUNCATE:
a8dd0a37 641 if (src->op == CEPH_OSD_OP_WRITE)
54d50649 642 request_data_len = src->extent.length;
a8dd0a37
AE		 643 dst->extent.offset = cpu_to_le64(src->extent.offset);
644 dst->extent.length = cpu_to_le64(src->extent.length);
645 dst->extent.truncate_size =
646 cpu_to_le64(src->extent.truncate_size);
647 dst->extent.truncate_seq =
648 cpu_to_le32(src->extent.truncate_seq);
04017e29 649 osd_data = &src->extent.osd_data;
5476492f 650 if (src->op == CEPH_OSD_OP_WRITE)
04017e29 651 ceph_osdc_msg_data_add(req->r_request, osd_data);
5476492f 652 else
04017e29 653 ceph_osdc_msg_data_add(req->r_reply, osd_data);
a8dd0a37
AE		 654 break;
655 case CEPH_OSD_OP_CALL:
a8dd0a37
AE		 656 dst->cls.class_len = src->cls.class_len;
657 dst->cls.method_len = src->cls.method_len;
04017e29
AE		 658 osd_data = &src->cls.request_info;
659 ceph_osdc_msg_data_add(req->r_request, osd_data);
660 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGELIST);
661 request_data_len = osd_data->pagelist->length;
662
663 osd_data = &src->cls.request_data;
664 data_length = ceph_osd_data_length(osd_data);
665 if (data_length) {
666 BUG_ON(osd_data->type == CEPH_OSD_DATA_TYPE_NONE);
667 dst->cls.indata_len = cpu_to_le32(data_length);
668 ceph_osdc_msg_data_add(req->r_request, osd_data);
669 src->payload_len += data_length;
670 request_data_len += data_length;
671 }
672 osd_data = &src->cls.response_data;
673 ceph_osdc_msg_data_add(req->r_reply, osd_data);
a8dd0a37
AE		 674 break;
675 case CEPH_OSD_OP_STARTSYNC:
676 break;
677 case CEPH_OSD_OP_NOTIFY_ACK:
678 case CEPH_OSD_OP_WATCH:
679 dst->watch.cookie = cpu_to_le64(src->watch.cookie);
680 dst->watch.ver = cpu_to_le64(src->watch.ver);
681 dst->watch.flag = src->watch.flag;
682 break;
683 default:
4c46459c 684 pr_err("unsupported osd opcode %s\n",
8f63ca2d 685 ceph_osd_op_name(src->op));
4c46459c 686 WARN_ON(1);
a8dd0a37
AE		 687
688 return 0;
68b4476b 689 }
a8dd0a37 690 dst->op = cpu_to_le16(src->op);
68b4476b 691 dst->payload_len = cpu_to_le32(src->payload_len);
175face2 692
54d50649 693 return request_data_len;
68b4476b
YS		 694}
695
3499e8a5
YS		 696/*
697 * build new request AND message, calculate layout, and adjust file
698 * extent as needed.
699 *
700 * if the file was recently truncated, we include information about its
701 * old and new size so that the object can be updated appropriately. (we
702 * avoid synchronously deleting truncated objects because it's slow.)
703 *
704 * if @do_sync, include a 'startsync' command so that the osd will flush
705 * data quickly.
706 */
707struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
708 struct ceph_file_layout *layout,
709 struct ceph_vino vino,
acead002 710 u64 off, u64 *plen, int num_ops,
3499e8a5
YS		 711 int opcode, int flags,
712 struct ceph_snap_context *snapc,
3499e8a5
YS		 713 u32 truncate_seq,
714 u64 truncate_size,
153e5167 715 bool use_mempool)
3499e8a5 716{
68b4476b 717 struct ceph_osd_request *req;
75d1c941
AE		 718 u64 objnum = 0;
719 u64 objoff = 0;
720 u64 objlen = 0;
d18d1e28
AE		 721 u32 object_size;
722 u64 object_base;
6816282d 723 int r;
68b4476b 724
ad7a60de
LW		 725 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
726 opcode != CEPH_OSD_OP_DELETE && opcode != CEPH_OSD_OP_ZERO &&
727 opcode != CEPH_OSD_OP_TRUNCATE);
68b4476b 728
acead002 729 req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool,
ae7ca4a3 730 GFP_NOFS);
4ad12621 731 if (!req)
6816282d 732 return ERR_PTR(-ENOMEM);
79528734 733
d178a9e7 734 req->r_flags = flags;
3499e8a5
YS		 735
736 /* calculate max write size */
a19dadfb 737 r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
3ff5f385
AE		 738 if (r < 0) {
739 ceph_osdc_put_request(req);
6816282d 740 return ERR_PTR(r);
3ff5f385 741 }
a19dadfb 742
d18d1e28
AE		 743 object_size = le32_to_cpu(layout->fl_object_size);
744 object_base = off - objoff;
ccca4e37
YZ		 745 if (!(truncate_seq == 1 && truncate_size == -1ULL)) {
746 if (truncate_size <= object_base) {
747 truncate_size = 0;
748 } else {
749 truncate_size -= object_base;
750 if (truncate_size > object_size)
751 truncate_size = object_size;
752 }
a19dadfb 753 }
d18d1e28 754
c99d2d4a 755 osd_req_op_extent_init(req, 0, opcode, objoff, objlen,
b0270324 756 truncate_size, truncate_seq);
8c042b0d 757
acead002
AE		 758 /*
759 * A second op in the ops array means the caller wants to
760 * also issue a include a 'startsync' command so that the
761 * osd will flush data quickly.
762 */
763 if (num_ops > 1)
c99d2d4a 764 osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC);
d18d1e28 765
3c972c95 766 req->r_base_oloc.pool = ceph_file_layout_pg_pool(*layout);
3499e8a5 767
3c972c95 768 snprintf(req->r_base_oid.name, sizeof(req->r_base_oid.name),
4295f221 769 "%llx.%08llx", vino.ino, objnum);
3c972c95 770 req->r_base_oid.name_len = strlen(req->r_base_oid.name);
dbe0fc41 771
f24e9980
SW		 772 return req;
773}
3d14c5d2 774EXPORT_SYMBOL(ceph_osdc_new_request);
f24e9980
SW		 775
776/*
777 * We keep osd requests in an rbtree, sorted by ->r_tid.
778 */
779static void __insert_request(struct ceph_osd_client *osdc,
780 struct ceph_osd_request *new)
781{
782 struct rb_node **p = &osdc->requests.rb_node;
783 struct rb_node *parent = NULL;
784 struct ceph_osd_request *req = NULL;
785
786 while (*p) {
787 parent = *p;
788 req = rb_entry(parent, struct ceph_osd_request, r_node);
789 if (new->r_tid < req->r_tid)
790 p = &(*p)->rb_left;
791 else if (new->r_tid > req->r_tid)
792 p = &(*p)->rb_right;
793 else
794 BUG();
795 }
796
797 rb_link_node(&new->r_node, parent, p);
798 rb_insert_color(&new->r_node, &osdc->requests);
799}
800
801static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc,
802 u64 tid)
803{
804 struct ceph_osd_request *req;
805 struct rb_node *n = osdc->requests.rb_node;
806
807 while (n) {
808 req = rb_entry(n, struct ceph_osd_request, r_node);
809 if (tid < req->r_tid)
810 n = n->rb_left;
811 else if (tid > req->r_tid)
812 n = n->rb_right;
813 else
814 return req;
815 }
816 return NULL;
817}
818
819static struct ceph_osd_request *
820__lookup_request_ge(struct ceph_osd_client *osdc,
821 u64 tid)
822{
823 struct ceph_osd_request *req;
824 struct rb_node *n = osdc->requests.rb_node;
825
826 while (n) {
827 req = rb_entry(n, struct ceph_osd_request, r_node);
828 if (tid < req->r_tid) {
829 if (!n->rb_left)
830 return req;
831 n = n->rb_left;
832 } else if (tid > req->r_tid) {
833 n = n->rb_right;
834 } else {
835 return req;
836 }
837 }
838 return NULL;
839}
840
6f6c7006
SW		 841/*
842 * Resubmit requests pending on the given osd.
843 */
844static void __kick_osd_requests(struct ceph_osd_client *osdc,
845 struct ceph_osd *osd)
846{
a40c4f10 847 struct ceph_osd_request *req, *nreq;
e02493c0 848 LIST_HEAD(resend);
6f6c7006
SW		 849 int err;
850
851 dout("__kick_osd_requests osd%d\n", osd->o_osd);
852 err = __reset_osd(osdc, osd);
685a7555 853 if (err)
6f6c7006 854 return;
e02493c0
AE		 855 /*
856 * Build up a list of requests to resend by traversing the
857 * osd's list of requests. Requests for a given object are
858 * sent in tid order, and that is also the order they're
859 * kept on this list. Therefore all requests that are in
860 * flight will be found first, followed by all requests that
861 * have not yet been sent. And to resend requests while
862 * preserving this order we will want to put any sent
863 * requests back on the front of the osd client's unsent
864 * list.
865 *
866 * So we build a separate ordered list of already-sent
867 * requests for the affected osd and splice it onto the
868 * front of the osd client's unsent list. Once we've seen a
869 * request that has not yet been sent we're done. Those
870 * requests are already sitting right where they belong.
871 */
6f6c7006 872 list_for_each_entry(req, &osd->o_requests, r_osd_item) {
e02493c0
AE		 873 if (!req->r_sent)
874 break;
875 list_move_tail(&req->r_req_lru_item, &resend);
876 dout("requeueing %p tid %llu osd%d\n", req, req->r_tid,
6f6c7006 877 osd->o_osd);
a40c4f10
YS		 878 if (!req->r_linger)
879 req->r_flags |= CEPH_OSD_FLAG_RETRY;
880 }
e02493c0 881 list_splice(&resend, &osdc->req_unsent);
a40c4f10 882
e02493c0
AE		 883 /*
884 * Linger requests are re-registered before sending, which
885 * sets up a new tid for each. We add them to the unsent
886 * list at the end to keep things in tid order.
887 */
a40c4f10
YS		 888 list_for_each_entry_safe(req, nreq, &osd->o_linger_requests,
889 r_linger_osd) {
77f38e0e
SW		 890 /*
891 * reregister request prior to unregistering linger so
892 * that r_osd is preserved.
893 */
894 BUG_ON(!list_empty(&req->r_req_lru_item));
a40c4f10 895 __register_request(osdc, req);
e02493c0 896 list_add_tail(&req->r_req_lru_item, &osdc->req_unsent);
ad885927 897 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
77f38e0e 898 __unregister_linger_request(osdc, req);
a40c4f10
YS		 899 dout("requeued lingering %p tid %llu osd%d\n", req, req->r_tid,
900 osd->o_osd);
6f6c7006
SW		 901 }
902}
903
f24e9980 904/*
81b024e7 905 * If the osd connection drops, we need to resubmit all requests.
f24e9980
SW		 906 */
907static void osd_reset(struct ceph_connection *con)
908{
909 struct ceph_osd *osd = con->private;
910 struct ceph_osd_client *osdc;
911
912 if (!osd)
913 return;
914 dout("osd_reset osd%d\n", osd->o_osd);
915 osdc = osd->o_osdc;
f24e9980 916 down_read(&osdc->map_sem);
83aff95e
SW		 917 mutex_lock(&osdc->request_mutex);
918 __kick_osd_requests(osdc, osd);
f9d25199 919 __send_queued(osdc);
83aff95e 920 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 921 up_read(&osdc->map_sem);
922}
923
924/*
925 * Track open sessions with osds.
926 */
e10006f8 927static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
f24e9980
SW		 928{
929 struct ceph_osd *osd;
930
931 osd = kzalloc(sizeof(*osd), GFP_NOFS);
932 if (!osd)
933 return NULL;
934
935 atomic_set(&osd->o_ref, 1);
936 osd->o_osdc = osdc;
e10006f8 937 osd->o_osd = onum;
f407731d 938 RB_CLEAR_NODE(&osd->o_node);
f24e9980 939 INIT_LIST_HEAD(&osd->o_requests);
a40c4f10 940 INIT_LIST_HEAD(&osd->o_linger_requests);
f5a2041b 941 INIT_LIST_HEAD(&osd->o_osd_lru);
f24e9980
SW		 942 osd->o_incarnation = 1;
943
b7a9e5dd 944 ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
4e7a5dcd 945
422d2cb8 946 INIT_LIST_HEAD(&osd->o_keepalive_item);
f24e9980
SW		 947 return osd;
948}
949
950static struct ceph_osd *get_osd(struct ceph_osd *osd)
951{
952 if (atomic_inc_not_zero(&osd->o_ref)) {
953 dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
954 atomic_read(&osd->o_ref));
955 return osd;
956 } else {
957 dout("get_osd %p FAIL\n", osd);
958 return NULL;
959 }
960}
961
962static void put_osd(struct ceph_osd *osd)
963{
964 dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
965 atomic_read(&osd->o_ref) - 1);
a255651d 966 if (atomic_dec_and_test(&osd->o_ref) && osd->o_auth.authorizer) {
79494d1b
SW		 967 struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth;
968
27859f97 969 ceph_auth_destroy_authorizer(ac, osd->o_auth.authorizer);
f24e9980 970 kfree(osd);
79494d1b 971 }
f24e9980
SW		 972}
973
974/*
975 * remove an osd from our map
976 */
f5a2041b 977static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
f24e9980 978{
f5a2041b 979 dout("__remove_osd %p\n", osd);
f24e9980
SW		 980 BUG_ON(!list_empty(&osd->o_requests));
981 rb_erase(&osd->o_node, &osdc->osds);
f5a2041b 982 list_del_init(&osd->o_osd_lru);
f24e9980
SW		 983 ceph_con_close(&osd->o_con);
984 put_osd(osd);
985}
986
aca420bc
SW		 987static void remove_all_osds(struct ceph_osd_client *osdc)
988{
048a9d2d 989 dout("%s %p\n", __func__, osdc);
aca420bc
SW		 990 mutex_lock(&osdc->request_mutex);
991 while (!RB_EMPTY_ROOT(&osdc->osds)) {
992 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
993 struct ceph_osd, o_node);
994 __remove_osd(osdc, osd);
995 }
996 mutex_unlock(&osdc->request_mutex);
997}
998
f5a2041b
YS		 999static void __move_osd_to_lru(struct ceph_osd_client *osdc,
1000 struct ceph_osd *osd)
1001{
1002 dout("__move_osd_to_lru %p\n", osd);
1003 BUG_ON(!list_empty(&osd->o_osd_lru));
1004 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
3d14c5d2 1005 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl * HZ;
f5a2041b
YS		 1006}
1007
1008static void __remove_osd_from_lru(struct ceph_osd *osd)
1009{
1010 dout("__remove_osd_from_lru %p\n", osd);
1011 if (!list_empty(&osd->o_osd_lru))
1012 list_del_init(&osd->o_osd_lru);
1013}
1014
aca420bc 1015static void remove_old_osds(struct ceph_osd_client *osdc)
f5a2041b
YS		 1016{
1017 struct ceph_osd *osd, *nosd;
1018
1019 dout("__remove_old_osds %p\n", osdc);
1020 mutex_lock(&osdc->request_mutex);
1021 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
aca420bc 1022 if (time_before(jiffies, osd->lru_ttl))
f5a2041b
YS		 1023 break;
1024 __remove_osd(osdc, osd);
1025 }
1026 mutex_unlock(&osdc->request_mutex);
1027}
1028
f24e9980
SW		 1029/*
1030 * reset osd connect
1031 */
f5a2041b 1032static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
f24e9980 1033{
c3acb181 1034 struct ceph_entity_addr *peer_addr;
f24e9980 1035
f5a2041b 1036 dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
a40c4f10
YS		 1037 if (list_empty(&osd->o_requests) &&
1038 list_empty(&osd->o_linger_requests)) {
f5a2041b 1039 __remove_osd(osdc, osd);
c3acb181
AE		 1040
1041 return -ENODEV;
1042 }
1043
1044 peer_addr = &osdc->osdmap->osd_addr[osd->o_osd];
1045 if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1046 !ceph_con_opened(&osd->o_con)) {
1047 struct ceph_osd_request *req;
1048
0b4af2e8
ID		 1049 dout("osd addr hasn't changed and connection never opened, "
1050 "letting msgr retry\n");
87b315a5
SW		 1051 /* touch each r_stamp for handle_timeout()'s benfit */
1052 list_for_each_entry(req, &osd->o_requests, r_osd_item)
1053 req->r_stamp = jiffies;
c3acb181
AE		 1054
1055 return -EAGAIN;
f24e9980 1056 }
c3acb181
AE		 1057
1058 ceph_con_close(&osd->o_con);
1059 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1060 osd->o_incarnation++;
1061
1062 return 0;
f24e9980
SW		 1063}
1064
1065static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new)
1066{
1067 struct rb_node **p = &osdc->osds.rb_node;
1068 struct rb_node *parent = NULL;
1069 struct ceph_osd *osd = NULL;
1070
aca420bc 1071 dout("__insert_osd %p osd%d\n", new, new->o_osd);
f24e9980
SW		 1072 while (*p) {
1073 parent = *p;
1074 osd = rb_entry(parent, struct ceph_osd, o_node);
1075 if (new->o_osd < osd->o_osd)
1076 p = &(*p)->rb_left;
1077 else if (new->o_osd > osd->o_osd)
1078 p = &(*p)->rb_right;
1079 else
1080 BUG();
1081 }
1082
1083 rb_link_node(&new->o_node, parent, p);
1084 rb_insert_color(&new->o_node, &osdc->osds);
1085}
1086
1087static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o)
1088{
1089 struct ceph_osd *osd;
1090 struct rb_node *n = osdc->osds.rb_node;
1091
1092 while (n) {
1093 osd = rb_entry(n, struct ceph_osd, o_node);
1094 if (o < osd->o_osd)
1095 n = n->rb_left;
1096 else if (o > osd->o_osd)
1097 n = n->rb_right;
1098 else
1099 return osd;
1100 }
1101 return NULL;
1102}
1103
422d2cb8
YS		 1104static void __schedule_osd_timeout(struct ceph_osd_client *osdc)
1105{
1106 schedule_delayed_work(&osdc->timeout_work,
3d14c5d2 1107 osdc->client->options->osd_keepalive_timeout * HZ);
422d2cb8
YS		 1108}
1109
1110static void __cancel_osd_timeout(struct ceph_osd_client *osdc)
1111{
1112 cancel_delayed_work(&osdc->timeout_work);
1113}
f24e9980
SW		 1114
1115/*
1116 * Register request, assign tid. If this is the first request, set up
1117 * the timeout event.
1118 */
a40c4f10
YS		 1119static void __register_request(struct ceph_osd_client *osdc,
1120 struct ceph_osd_request *req)
f24e9980 1121{
f24e9980 1122 req->r_tid = ++osdc->last_tid;
6df058c0 1123 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
77f38e0e 1124 dout("__register_request %p tid %lld\n", req, req->r_tid);
f24e9980
SW		 1125 __insert_request(osdc, req);
1126 ceph_osdc_get_request(req);
1127 osdc->num_requests++;
f24e9980 1128 if (osdc->num_requests == 1) {
422d2cb8
YS		 1129 dout(" first request, scheduling timeout\n");
1130 __schedule_osd_timeout(osdc);
f24e9980 1131 }
a40c4f10
YS		 1132}
1133
f24e9980
SW		 1134/*
1135 * called under osdc->request_mutex
1136 */
1137static void __unregister_request(struct ceph_osd_client *osdc,
1138 struct ceph_osd_request *req)
1139{
35f9f8a0
SW		 1140 if (RB_EMPTY_NODE(&req->r_node)) {
1141 dout("__unregister_request %p tid %lld not registered\n",
1142 req, req->r_tid);
1143 return;
1144 }
1145
f24e9980
SW		 1146 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
1147 rb_erase(&req->r_node, &osdc->requests);
1148 osdc->num_requests--;
1149
0ba6478d
SW		 1150 if (req->r_osd) {
1151 /* make sure the original request isn't in flight. */
6740a845 1152 ceph_msg_revoke(req->r_request);
0ba6478d
SW		 1153
1154 list_del_init(&req->r_osd_item);
a40c4f10
YS		 1155 if (list_empty(&req->r_osd->o_requests) &&
1156 list_empty(&req->r_osd->o_linger_requests)) {
1157 dout("moving osd to %p lru\n", req->r_osd);
f5a2041b 1158 __move_osd_to_lru(osdc, req->r_osd);
a40c4f10 1159 }
fbdb9190 1160 if (list_empty(&req->r_linger_item))
a40c4f10 1161 req->r_osd = NULL;
0ba6478d 1162 }
f24e9980 1163
7d5f2481 1164 list_del_init(&req->r_req_lru_item);
f24e9980
SW		 1165 ceph_osdc_put_request(req);
1166
422d2cb8
YS		 1167 if (osdc->num_requests == 0) {
1168 dout(" no requests, canceling timeout\n");
1169 __cancel_osd_timeout(osdc);
f24e9980
SW		 1170 }
1171}
1172
1173/*
1174 * Cancel a previously queued request message
1175 */
1176static void __cancel_request(struct ceph_osd_request *req)
1177{
6bc18876 1178 if (req->r_sent && req->r_osd) {
6740a845 1179 ceph_msg_revoke(req->r_request);
f24e9980
SW		 1180 req->r_sent = 0;
1181 }
1182}
1183
a40c4f10
YS		 1184static void __register_linger_request(struct ceph_osd_client *osdc,
1185 struct ceph_osd_request *req)
1186{
1187 dout("__register_linger_request %p\n", req);
96e4dac6 1188 ceph_osdc_get_request(req);
a40c4f10 1189 list_add_tail(&req->r_linger_item, &osdc->req_linger);
6194ea89
SW		 1190 if (req->r_osd)
1191 list_add_tail(&req->r_linger_osd,
1192 &req->r_osd->o_linger_requests);
a40c4f10
YS		 1193}
1194
1195static void __unregister_linger_request(struct ceph_osd_client *osdc,
1196 struct ceph_osd_request *req)
1197{
1198 dout("__unregister_linger_request %p\n", req);
61c74035 1199 list_del_init(&req->r_linger_item);
a40c4f10 1200 if (req->r_osd) {
a40c4f10
YS		 1201 list_del_init(&req->r_linger_osd);
1202
1203 if (list_empty(&req->r_osd->o_requests) &&
1204 list_empty(&req->r_osd->o_linger_requests)) {
1205 dout("moving osd to %p lru\n", req->r_osd);
1206 __move_osd_to_lru(osdc, req->r_osd);
1207 }
fbdb9190
SW		 1208 if (list_empty(&req->r_osd_item))
1209 req->r_osd = NULL;
a40c4f10 1210 }
96e4dac6 1211 ceph_osdc_put_request(req);
a40c4f10
YS		 1212}
1213
1214void ceph_osdc_unregister_linger_request(struct ceph_osd_client *osdc,
1215 struct ceph_osd_request *req)
1216{
1217 mutex_lock(&osdc->request_mutex);
1218 if (req->r_linger) {
c10ebbf5 1219 req->r_linger = 0;
96e4dac6 1220 __unregister_linger_request(osdc, req);
a40c4f10
YS		 1221 }
1222 mutex_unlock(&osdc->request_mutex);
1223}
1224EXPORT_SYMBOL(ceph_osdc_unregister_linger_request);
1225
1226void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
1227 struct ceph_osd_request *req)
1228{
1229 if (!req->r_linger) {
1230 dout("set_request_linger %p\n", req);
1231 req->r_linger = 1;
a40c4f10
YS		 1232 }
1233}
1234EXPORT_SYMBOL(ceph_osdc_set_request_linger);
1235
d29adb34
JD		 1236/*
1237 * Returns whether a request should be blocked from being sent
1238 * based on the current osdmap and osd_client settings.
1239 *
1240 * Caller should hold map_sem for read.
1241 */
1242static bool __req_should_be_paused(struct ceph_osd_client *osdc,
1243 struct ceph_osd_request *req)
1244{
1245 bool pauserd = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD);
1246 bool pausewr = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR) ||
1247 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
1248 return (req->r_flags & CEPH_OSD_FLAG_READ && pauserd) ||
1249 (req->r_flags & CEPH_OSD_FLAG_WRITE && pausewr);
1250}
1251
17a13e40
ID		 1252/*
1253 * Calculate mapping of a request to a PG. Takes tiering into account.
1254 */
1255static int __calc_request_pg(struct ceph_osdmap *osdmap,
1256 struct ceph_osd_request *req,
1257 struct ceph_pg *pg_out)
1258{
1259 if ((req->r_flags & CEPH_OSD_FLAG_IGNORE_OVERLAY) == 0) {
1260 struct ceph_pg_pool_info *pi;
1261
3c972c95 1262 pi = ceph_pg_pool_by_id(osdmap, req->r_base_oloc.pool);
17a13e40
ID		 1263 if (pi) {
1264 if ((req->r_flags & CEPH_OSD_FLAG_READ) &&
1265 pi->read_tier >= 0)
3c972c95 1266 req->r_base_oloc.pool = pi->read_tier;
17a13e40
ID		 1267 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
1268 pi->write_tier >= 0)
3c972c95 1269 req->r_base_oloc.pool = pi->write_tier;
17a13e40
ID		 1270 }
1271 /* !pi is caught in ceph_oloc_oid_to_pg() */
1272 }
1273
3c972c95
ID		 1274 return ceph_oloc_oid_to_pg(osdmap, &req->r_base_oloc,
1275 &req->r_base_oid, pg_out);
17a13e40
ID		 1276}
1277
f24e9980
SW		 1278/*
1279 * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
1280 * (as needed), and set the request r_osd appropriately. If there is
25985edc 1281 * no up osd, set r_osd to NULL. Move the request to the appropriate list
6f6c7006 1282 * (unsent, homeless) or leave on in-flight lru.
f24e9980
SW		 1283 *
1284 * Return 0 if unchanged, 1 if changed, or negative on error.
1285 *
1286 * Caller should hold map_sem for read and request_mutex.
1287 */
6f6c7006 1288static int __map_request(struct ceph_osd_client *osdc,
38d6453c 1289 struct ceph_osd_request *req, int force_resend)
f24e9980 1290{
5b191d99 1291 struct ceph_pg pgid;
d85b7056
SW		 1292 int acting[CEPH_PG_MAX_SIZE];
1293 int o = -1, num = 0;
f24e9980 1294 int err;
d29adb34 1295 bool was_paused;
f24e9980 1296
6f6c7006 1297 dout("map_request %p tid %lld\n", req, req->r_tid);
17a13e40
ID		 1298
1299 err = __calc_request_pg(osdc->osdmap, req, &pgid);
6f6c7006
SW		 1300 if (err) {
1301 list_move(&req->r_req_lru_item, &osdc->req_notarget);
f24e9980 1302 return err;
6f6c7006 1303 }
7740a42f
SW		 1304 req->r_pgid = pgid;
1305
d85b7056
SW		 1306 err = ceph_calc_pg_acting(osdc->osdmap, pgid, acting);
1307 if (err > 0) {
1308 o = acting[0];
1309 num = err;
1310 }
f24e9980 1311
d29adb34
JD		 1312 was_paused = req->r_paused;
1313 req->r_paused = __req_should_be_paused(osdc, req);
1314 if (was_paused && !req->r_paused)
1315 force_resend = 1;
1316
38d6453c
SW		 1317 if ((!force_resend &&
1318 req->r_osd && req->r_osd->o_osd == o &&
d85b7056
SW		 1319 req->r_sent >= req->r_osd->o_incarnation &&
1320 req->r_num_pg_osds == num &&
1321 memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
d29adb34
JD		 1322 (req->r_osd == NULL && o == -1) ||
1323 req->r_paused)
f24e9980
SW		 1324 return 0; /* no change */
1325
5b191d99
SW		 1326 dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n",
1327 req->r_tid, pgid.pool, pgid.seed, o,
f24e9980
SW		 1328 req->r_osd ? req->r_osd->o_osd : -1);
1329
d85b7056
SW		 1330 /* record full pg acting set */
1331 memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
1332 req->r_num_pg_osds = num;
1333
f24e9980
SW		 1334 if (req->r_osd) {
1335 __cancel_request(req);
1336 list_del_init(&req->r_osd_item);
f24e9980
SW		 1337 req->r_osd = NULL;
1338 }
1339
1340 req->r_osd = __lookup_osd(osdc, o);
1341 if (!req->r_osd && o >= 0) {
c99eb1c7 1342 err = -ENOMEM;
e10006f8 1343 req->r_osd = create_osd(osdc, o);
6f6c7006
SW		 1344 if (!req->r_osd) {
1345 list_move(&req->r_req_lru_item, &osdc->req_notarget);
c99eb1c7 1346 goto out;
6f6c7006 1347 }
f24e9980 1348
6f6c7006 1349 dout("map_request osd %p is osd%d\n", req->r_osd, o);
f24e9980
SW		 1350 __insert_osd(osdc, req->r_osd);
1351
b7a9e5dd
SW		 1352 ceph_con_open(&req->r_osd->o_con,
1353 CEPH_ENTITY_TYPE_OSD, o,
1354 &osdc->osdmap->osd_addr[o]);
f24e9980
SW		 1355 }
1356
f5a2041b
YS		 1357 if (req->r_osd) {
1358 __remove_osd_from_lru(req->r_osd);
ad885927
AE		 1359 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
1360 list_move_tail(&req->r_req_lru_item, &osdc->req_unsent);
6f6c7006 1361 } else {
ad885927 1362 list_move_tail(&req->r_req_lru_item, &osdc->req_notarget);
f5a2041b 1363 }
d85b7056 1364 err = 1; /* osd or pg changed */
f24e9980
SW		 1365
1366out:
f24e9980
SW		 1367 return err;
1368}
1369
1370/*
1371 * caller should hold map_sem (for read) and request_mutex
1372 */
56e925b6
SW		 1373static void __send_request(struct ceph_osd_client *osdc,
1374 struct ceph_osd_request *req)
f24e9980 1375{
1b83bef2 1376 void *p;
f24e9980 1377
1b83bef2
SW		 1378 dout("send_request %p tid %llu to osd%d flags %d pg %lld.%x\n",
1379 req, req->r_tid, req->r_osd->o_osd, req->r_flags,
1380 (unsigned long long)req->r_pgid.pool, req->r_pgid.seed);
1381
1382 /* fill in message content that changes each time we send it */
1383 put_unaligned_le32(osdc->osdmap->epoch, req->r_request_osdmap_epoch);
1384 put_unaligned_le32(req->r_flags, req->r_request_flags);
3c972c95 1385 put_unaligned_le64(req->r_base_oloc.pool, req->r_request_pool);
1b83bef2
SW		 1386 p = req->r_request_pgid;
1387 ceph_encode_64(&p, req->r_pgid.pool);
1388 ceph_encode_32(&p, req->r_pgid.seed);
1389 put_unaligned_le64(1, req->r_request_attempts); /* FIXME */
1390 memcpy(req->r_request_reassert_version, &req->r_reassert_version,
1391 sizeof(req->r_reassert_version));
2169aea6 1392
3dd72fc0 1393 req->r_stamp = jiffies;
07a27e22 1394 list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
f24e9980
SW		 1395
1396 ceph_msg_get(req->r_request); /* send consumes a ref */
26be8808 1397
f24e9980 1398 req->r_sent = req->r_osd->o_incarnation;
26be8808
AE		 1399
1400 ceph_con_send(&req->r_osd->o_con, req->r_request);
f24e9980
SW		 1401}
1402
6f6c7006
SW		 1403/*
1404 * Send any requests in the queue (req_unsent).
1405 */
f9d25199 1406static void __send_queued(struct ceph_osd_client *osdc)
6f6c7006
SW		 1407{
1408 struct ceph_osd_request *req, *tmp;
1409
f9d25199
AE		 1410 dout("__send_queued\n");
1411 list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item)
6f6c7006 1412 __send_request(osdc, req);
6f6c7006
SW		 1413}
1414
f24e9980
SW		 1415/*
1416 * Timeout callback, called every N seconds when 1 or more osd
1417 * requests has been active for more than N seconds. When this
1418 * happens, we ping all OSDs with requests who have timed out to
1419 * ensure any communications channel reset is detected. Reset the
1420 * request timeouts another N seconds in the future as we go.
1421 * Reschedule the timeout event another N seconds in future (unless
1422 * there are no open requests).
1423 */
1424static void handle_timeout(struct work_struct *work)
1425{
1426 struct ceph_osd_client *osdc =
1427 container_of(work, struct ceph_osd_client, timeout_work.work);
83aff95e 1428 struct ceph_osd_request *req;
f24e9980 1429 struct ceph_osd *osd;
422d2cb8 1430 unsigned long keepalive =
3d14c5d2 1431 osdc->client->options->osd_keepalive_timeout * HZ;
422d2cb8 1432 struct list_head slow_osds;
f24e9980
SW		 1433 dout("timeout\n");
1434 down_read(&osdc->map_sem);
1435
1436 ceph_monc_request_next_osdmap(&osdc->client->monc);
1437
1438 mutex_lock(&osdc->request_mutex);
f24e9980 1439
422d2cb8
YS		 1440 /*
1441 * ping osds that are a bit slow. this ensures that if there
1442 * is a break in the TCP connection we will notice, and reopen
1443 * a connection with that osd (from the fault callback).
1444 */
1445 INIT_LIST_HEAD(&slow_osds);
1446 list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
3dd72fc0 1447 if (time_before(jiffies, req->r_stamp + keepalive))
422d2cb8
YS		 1448 break;
1449
1450 osd = req->r_osd;
1451 BUG_ON(!osd);
1452 dout(" tid %llu is slow, will send keepalive on osd%d\n",
f24e9980 1453 req->r_tid, osd->o_osd);
422d2cb8
YS		 1454 list_move_tail(&osd->o_keepalive_item, &slow_osds);
1455 }
1456 while (!list_empty(&slow_osds)) {
1457 osd = list_entry(slow_osds.next, struct ceph_osd,
1458 o_keepalive_item);
1459 list_del_init(&osd->o_keepalive_item);
f24e9980
SW		 1460 ceph_con_keepalive(&osd->o_con);
1461 }
1462
422d2cb8 1463 __schedule_osd_timeout(osdc);
f9d25199 1464 __send_queued(osdc);
f24e9980 1465 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 1466 up_read(&osdc->map_sem);
1467}
1468
f5a2041b
YS		 1469static void handle_osds_timeout(struct work_struct *work)
1470{
1471 struct ceph_osd_client *osdc =
1472 container_of(work, struct ceph_osd_client,
1473 osds_timeout_work.work);
1474 unsigned long delay =
3d14c5d2 1475 osdc->client->options->osd_idle_ttl * HZ >> 2;
f5a2041b
YS		 1476
1477 dout("osds timeout\n");
1478 down_read(&osdc->map_sem);
aca420bc 1479 remove_old_osds(osdc);
f5a2041b
YS		 1480 up_read(&osdc->map_sem);
1481
1482 schedule_delayed_work(&osdc->osds_timeout_work,
1483 round_jiffies_relative(delay));
1484}
1485
25845472
SW		 1486static void complete_request(struct ceph_osd_request *req)
1487{
25845472
SW		 1488 complete_all(&req->r_safe_completion); /* fsync waiter */
1489}
1490
f24e9980
SW		 1491/*
1492 * handle osd op reply. either call the callback if it is specified,
1493 * or do the completion to wake up the waiting thread.
1494 */
350b1c32
SW		 1495static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg,
1496 struct ceph_connection *con)
f24e9980 1497{
1b83bef2 1498 void *p, *end;
f24e9980
SW		 1499 struct ceph_osd_request *req;
1500 u64 tid;
1b83bef2 1501 int object_len;
79528734
AE		 1502 unsigned int numops;
1503 int payload_len, flags;
0ceed5db 1504 s32 result;
1b83bef2
SW		 1505 s32 retry_attempt;
1506 struct ceph_pg pg;
1507 int err;
1508 u32 reassert_epoch;
1509 u64 reassert_version;
1510 u32 osdmap_epoch;
0d5af164 1511 int already_completed;
9fc6e064 1512 u32 bytes;
79528734 1513 unsigned int i;
f24e9980 1514
6df058c0 1515 tid = le64_to_cpu(msg->hdr.tid);
1b83bef2
SW		 1516 dout("handle_reply %p tid %llu\n", msg, tid);
1517
1518 p = msg->front.iov_base;
1519 end = p + msg->front.iov_len;
1520
1521 ceph_decode_need(&p, end, 4, bad);
1522 object_len = ceph_decode_32(&p);
1523 ceph_decode_need(&p, end, object_len, bad);
1524 p += object_len;
1525
ef4859d6 1526 err = ceph_decode_pgid(&p, end, &pg);
1b83bef2 1527 if (err)
f24e9980 1528 goto bad;
1b83bef2
SW		 1529
1530 ceph_decode_need(&p, end, 8 + 4 + 4 + 8 + 4, bad);
1531 flags = ceph_decode_64(&p);
1532 result = ceph_decode_32(&p);
1533 reassert_epoch = ceph_decode_32(&p);
1534 reassert_version = ceph_decode_64(&p);
1535 osdmap_epoch = ceph_decode_32(&p);
1536
f24e9980
SW		 1537 /* lookup */
1538 mutex_lock(&osdc->request_mutex);
1539 req = __lookup_request(osdc, tid);
1540 if (req == NULL) {
1541 dout("handle_reply tid %llu dne\n", tid);
8058fd45 1542 goto bad_mutex;
f24e9980
SW		 1543 }
1544 ceph_osdc_get_request(req);
1b83bef2
SW		 1545
1546 dout("handle_reply %p tid %llu req %p result %d\n", msg, tid,
1547 req, result);
1548
18741196 1549 ceph_decode_need(&p, end, 4, bad_put);
1b83bef2
SW		 1550 numops = ceph_decode_32(&p);
1551 if (numops > CEPH_OSD_MAX_OP)
1552 goto bad_put;
1553 if (numops != req->r_num_ops)
1554 goto bad_put;
1555 payload_len = 0;
18741196 1556 ceph_decode_need(&p, end, numops * sizeof(struct ceph_osd_op), bad_put);
1b83bef2
SW		 1557 for (i = 0; i < numops; i++) {
1558 struct ceph_osd_op *op = p;
1559 int len;
1560
1561 len = le32_to_cpu(op->payload_len);
1562 req->r_reply_op_len[i] = len;
1563 dout(" op %d has %d bytes\n", i, len);
1564 payload_len += len;
1565 p += sizeof(*op);
1566 }
9fc6e064
AE		 1567 bytes = le32_to_cpu(msg->hdr.data_len);
1568 if (payload_len != bytes) {
1b83bef2 1569 pr_warning("sum of op payload lens %d != data_len %d",
9fc6e064 1570 payload_len, bytes);
1b83bef2
SW		 1571 goto bad_put;
1572 }
1573
18741196 1574 ceph_decode_need(&p, end, 4 + numops * 4, bad_put);
1b83bef2
SW		 1575 retry_attempt = ceph_decode_32(&p);
1576 for (i = 0; i < numops; i++)
1577 req->r_reply_op_result[i] = ceph_decode_32(&p);
f24e9980 1578
eb845ff1
YZ		 1579 already_completed = req->r_got_reply;
1580
f24e9980 1581 if (!req->r_got_reply) {
f24e9980 1582
1b83bef2 1583 req->r_result = result;
f24e9980
SW		 1584 dout("handle_reply result %d bytes %d\n", req->r_result,
1585 bytes);
1586 if (req->r_result == 0)
1587 req->r_result = bytes;
1588
1589 /* in case this is a write and we need to replay, */
1b83bef2
SW		 1590 req->r_reassert_version.epoch = cpu_to_le32(reassert_epoch);
1591 req->r_reassert_version.version = cpu_to_le64(reassert_version);
f24e9980
SW		 1592
1593 req->r_got_reply = 1;
1594 } else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) {
1595 dout("handle_reply tid %llu dup ack\n", tid);
34b43a56 1596 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 1597 goto done;
1598 }
1599
1600 dout("handle_reply tid %llu flags %d\n", tid, flags);
1601
a40c4f10
YS		 1602 if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK))
1603 __register_linger_request(osdc, req);
1604
f24e9980 1605 /* either this is a read, or we got the safe response */
0ceed5db
SW		 1606 if (result < 0 ||
1607 (flags & CEPH_OSD_FLAG_ONDISK) ||
f24e9980
SW		 1608 ((flags & CEPH_OSD_FLAG_WRITE) == 0))
1609 __unregister_request(osdc, req);
1610
1611 mutex_unlock(&osdc->request_mutex);
1612
eb845ff1 1613 if (!already_completed) {
61c5d6bf
YZ		 1614 if (req->r_unsafe_callback &&
1615 result >= 0 && !(flags & CEPH_OSD_FLAG_ONDISK))
1616 req->r_unsafe_callback(req, true);
eb845ff1
YZ		 1617 if (req->r_callback)
1618 req->r_callback(req, msg);
1619 else
1620 complete_all(&req->r_completion);
1621 }
f24e9980 1622
61c5d6bf
YZ		 1623 if (flags & CEPH_OSD_FLAG_ONDISK) {
1624 if (req->r_unsafe_callback && already_completed)
1625 req->r_unsafe_callback(req, false);
25845472 1626 complete_request(req);
61c5d6bf 1627 }
f24e9980
SW		 1628
1629done:
a40c4f10 1630 dout("req=%p req->r_linger=%d\n", req, req->r_linger);
f24e9980
SW		 1631 ceph_osdc_put_request(req);
1632 return;
1633
1b83bef2 1634bad_put:
37c89bde
LW		 1635 req->r_result = -EIO;
1636 __unregister_request(osdc, req);
1637 if (req->r_callback)
1638 req->r_callback(req, msg);
1639 else
1640 complete_all(&req->r_completion);
1641 complete_request(req);
1b83bef2 1642 ceph_osdc_put_request(req);
8058fd45
AE		 1643bad_mutex:
1644 mutex_unlock(&osdc->request_mutex);
f24e9980 1645bad:
1b83bef2
SW		 1646 pr_err("corrupt osd_op_reply got %d %d\n",
1647 (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len));
9ec7cab1 1648 ceph_msg_dump(msg);
f24e9980
SW		 1649}
1650
6f6c7006 1651static void reset_changed_osds(struct ceph_osd_client *osdc)
f24e9980 1652{
f24e9980 1653 struct rb_node *p, *n;
f24e9980 1654
6f6c7006
SW		 1655 for (p = rb_first(&osdc->osds); p; p = n) {
1656 struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
f24e9980 1657
6f6c7006
SW		 1658 n = rb_next(p);
1659 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
1660 memcmp(&osd->o_con.peer_addr,
1661 ceph_osd_addr(osdc->osdmap,
1662 osd->o_osd),
1663 sizeof(struct ceph_entity_addr)) != 0)
1664 __reset_osd(osdc, osd);
f24e9980 1665 }
422d2cb8
YS		 1666}
1667
1668/*
6f6c7006
SW		 1669 * Requeue requests whose mapping to an OSD has changed. If requests map to
1670 * no osd, request a new map.
422d2cb8 1671 *
e6d50f67 1672 * Caller should hold map_sem for read.
422d2cb8 1673 */
9a1ea2db
JD		 1674static void kick_requests(struct ceph_osd_client *osdc, bool force_resend,
1675 bool force_resend_writes)
422d2cb8 1676{
a40c4f10 1677 struct ceph_osd_request *req, *nreq;
6f6c7006
SW		 1678 struct rb_node *p;
1679 int needmap = 0;
1680 int err;
9a1ea2db 1681 bool force_resend_req;
422d2cb8 1682
9a1ea2db
JD		 1683 dout("kick_requests %s %s\n", force_resend ? " (force resend)" : "",
1684 force_resend_writes ? " (force resend writes)" : "");
422d2cb8 1685 mutex_lock(&osdc->request_mutex);
6194ea89 1686 for (p = rb_first(&osdc->requests); p; ) {
6f6c7006 1687 req = rb_entry(p, struct ceph_osd_request, r_node);
6194ea89 1688 p = rb_next(p);
ab60b16d
AE		 1689
1690 /*
1691 * For linger requests that have not yet been
1692 * registered, move them to the linger list; they'll
1693 * be sent to the osd in the loop below. Unregister
1694 * the request before re-registering it as a linger
1695 * request to ensure the __map_request() below
1696 * will decide it needs to be sent.
1697 */
1698 if (req->r_linger && list_empty(&req->r_linger_item)) {
1699 dout("%p tid %llu restart on osd%d\n",
1700 req, req->r_tid,
1701 req->r_osd ? req->r_osd->o_osd : -1);
96e4dac6 1702 ceph_osdc_get_request(req);
ab60b16d
AE		 1703 __unregister_request(osdc, req);
1704 __register_linger_request(osdc, req);
96e4dac6 1705 ceph_osdc_put_request(req);
ab60b16d
AE		 1706 continue;
1707 }
1708
9a1ea2db
JD		 1709 force_resend_req = force_resend ||
1710 (force_resend_writes &&
1711 req->r_flags & CEPH_OSD_FLAG_WRITE);
1712 err = __map_request(osdc, req, force_resend_req);
6f6c7006
SW		 1713 if (err < 0)
1714 continue; /* error */
1715 if (req->r_osd == NULL) {
1716 dout("%p tid %llu maps to no osd\n", req, req->r_tid);
1717 needmap++; /* request a newer map */
1718 } else if (err > 0) {
6194ea89
SW		 1719 if (!req->r_linger) {
1720 dout("%p tid %llu requeued on osd%d\n", req,
1721 req->r_tid,
1722 req->r_osd ? req->r_osd->o_osd : -1);
a40c4f10 1723 req->r_flags |= CEPH_OSD_FLAG_RETRY;
6194ea89
SW		 1724 }
1725 }
a40c4f10
YS		 1726 }
1727
1728 list_for_each_entry_safe(req, nreq, &osdc->req_linger,
1729 r_linger_item) {
1730 dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
1731
9a1ea2db
JD		 1732 err = __map_request(osdc, req,
1733 force_resend || force_resend_writes);
ab60b16d 1734 dout("__map_request returned %d\n", err);
a40c4f10
YS		 1735 if (err == 0)
1736 continue; /* no change and no osd was specified */
1737 if (err < 0)
1738 continue; /* hrm! */
1739 if (req->r_osd == NULL) {
1740 dout("tid %llu maps to no valid osd\n", req->r_tid);
1741 needmap++; /* request a newer map */
1742 continue;
6f6c7006 1743 }
a40c4f10
YS		 1744
1745 dout("kicking lingering %p tid %llu osd%d\n", req, req->r_tid,
1746 req->r_osd ? req->r_osd->o_osd : -1);
a40c4f10 1747 __register_request(osdc, req);
c89ce05e 1748 __unregister_linger_request(osdc, req);
6f6c7006 1749 }
14d2f38d 1750 reset_changed_osds(osdc);
f24e9980
SW		 1751 mutex_unlock(&osdc->request_mutex);
1752
1753 if (needmap) {
1754 dout("%d requests for down osds, need new map\n", needmap);
1755 ceph_monc_request_next_osdmap(&osdc->client->monc);
1756 }
422d2cb8 1757}
6f6c7006
SW		 1758
1759
f24e9980
SW		 1760/*
1761 * Process updated osd map.
1762 *
1763 * The message contains any number of incremental and full maps, normally
1764 * indicating some sort of topology change in the cluster. Kick requests
1765 * off to different OSDs as needed.
1766 */
1767void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1768{
1769 void *p, *end, *next;
1770 u32 nr_maps, maplen;
1771 u32 epoch;
1772 struct ceph_osdmap *newmap = NULL, *oldmap;
1773 int err;
1774 struct ceph_fsid fsid;
9a1ea2db 1775 bool was_full;
f24e9980
SW		 1776
1777 dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
1778 p = msg->front.iov_base;
1779 end = p + msg->front.iov_len;
1780
1781 /* verify fsid */
1782 ceph_decode_need(&p, end, sizeof(fsid), bad);
1783 ceph_decode_copy(&p, &fsid, sizeof(fsid));
0743304d
SW		 1784 if (ceph_check_fsid(osdc->client, &fsid) < 0)
1785 return;
f24e9980
SW		 1786
1787 down_write(&osdc->map_sem);
1788
9a1ea2db
JD		 1789 was_full = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
1790
f24e9980
SW		 1791 /* incremental maps */
1792 ceph_decode_32_safe(&p, end, nr_maps, bad);
1793 dout(" %d inc maps\n", nr_maps);
1794 while (nr_maps > 0) {
1795 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 1796 epoch = ceph_decode_32(&p);
1797 maplen = ceph_decode_32(&p);
f24e9980
SW		 1798 ceph_decode_need(&p, end, maplen, bad);
1799 next = p + maplen;
1800 if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
1801 dout("applying incremental map %u len %d\n",
1802 epoch, maplen);
1803 newmap = osdmap_apply_incremental(&p, next,
1804 osdc->osdmap,
15d9882c 1805 &osdc->client->msgr);
f24e9980
SW		 1806 if (IS_ERR(newmap)) {
1807 err = PTR_ERR(newmap);
1808 goto bad;
1809 }
30dc6381 1810 BUG_ON(!newmap);
f24e9980
SW		 1811 if (newmap != osdc->osdmap) {
1812 ceph_osdmap_destroy(osdc->osdmap);
1813 osdc->osdmap = newmap;
1814 }
9a1ea2db
JD		 1815 was_full = was_full ||
1816 ceph_osdmap_flag(osdc->osdmap,
1817 CEPH_OSDMAP_FULL);
1818 kick_requests(osdc, 0, was_full);
f24e9980
SW		 1819 } else {
1820 dout("ignoring incremental map %u len %d\n",
1821 epoch, maplen);
1822 }
1823 p = next;
1824 nr_maps--;
1825 }
1826 if (newmap)
1827 goto done;
1828
1829 /* full maps */
1830 ceph_decode_32_safe(&p, end, nr_maps, bad);
1831 dout(" %d full maps\n", nr_maps);
1832 while (nr_maps) {
1833 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 1834 epoch = ceph_decode_32(&p);
1835 maplen = ceph_decode_32(&p);
f24e9980
SW		 1836 ceph_decode_need(&p, end, maplen, bad);
1837 if (nr_maps > 1) {
1838 dout("skipping non-latest full map %u len %d\n",
1839 epoch, maplen);
1840 } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
1841 dout("skipping full map %u len %d, "
1842 "older than our %u\n", epoch, maplen,
1843 osdc->osdmap->epoch);
1844 } else {
38d6453c
SW		 1845 int skipped_map = 0;
1846
f24e9980
SW		 1847 dout("taking full map %u len %d\n", epoch, maplen);
1848 newmap = osdmap_decode(&p, p+maplen);
1849 if (IS_ERR(newmap)) {
1850 err = PTR_ERR(newmap);
1851 goto bad;
1852 }
30dc6381 1853 BUG_ON(!newmap);
f24e9980
SW		 1854 oldmap = osdc->osdmap;
1855 osdc->osdmap = newmap;
38d6453c
SW		 1856 if (oldmap) {
1857 if (oldmap->epoch + 1 < newmap->epoch)
1858 skipped_map = 1;
f24e9980 1859 ceph_osdmap_destroy(oldmap);
38d6453c 1860 }
9a1ea2db
JD		 1861 was_full = was_full ||
1862 ceph_osdmap_flag(osdc->osdmap,
1863 CEPH_OSDMAP_FULL);
1864 kick_requests(osdc, skipped_map, was_full);
f24e9980
SW		 1865 }
1866 p += maplen;
1867 nr_maps--;
1868 }
1869
b72e19b9
DC		 1870 if (!osdc->osdmap)
1871 goto bad;
f24e9980
SW		 1872done:
1873 downgrade_write(&osdc->map_sem);
1874 ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch);
cd634fb6
SW		 1875
1876 /*
1877 * subscribe to subsequent osdmap updates if full to ensure
1878 * we find out when we are no longer full and stop returning
1879 * ENOSPC.
1880 */
d29adb34
JD		 1881 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) ||
1882 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD) ||
1883 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR))
cd634fb6
SW		 1884 ceph_monc_request_next_osdmap(&osdc->client->monc);
1885
f9d25199
AE		 1886 mutex_lock(&osdc->request_mutex);
1887 __send_queued(osdc);
1888 mutex_unlock(&osdc->request_mutex);
f24e9980 1889 up_read(&osdc->map_sem);
03066f23 1890 wake_up_all(&osdc->client->auth_wq);
f24e9980
SW		 1891 return;
1892
1893bad:
1894 pr_err("osdc handle_map corrupt msg\n");
9ec7cab1 1895 ceph_msg_dump(msg);
f24e9980
SW		 1896 up_write(&osdc->map_sem);
1897 return;
1898}
1899
a40c4f10
YS		 1900/*
1901 * watch/notify callback event infrastructure
1902 *
1903 * These callbacks are used both for watch and notify operations.
1904 */
1905static void __release_event(struct kref *kref)
1906{
1907 struct ceph_osd_event *event =
1908 container_of(kref, struct ceph_osd_event, kref);
1909
1910 dout("__release_event %p\n", event);
1911 kfree(event);
1912}
1913
1914static void get_event(struct ceph_osd_event *event)
1915{
1916 kref_get(&event->kref);
1917}
1918
1919void ceph_osdc_put_event(struct ceph_osd_event *event)
1920{
1921 kref_put(&event->kref, __release_event);
1922}
1923EXPORT_SYMBOL(ceph_osdc_put_event);
1924
1925static void __insert_event(struct ceph_osd_client *osdc,
1926 struct ceph_osd_event *new)
1927{
1928 struct rb_node **p = &osdc->event_tree.rb_node;
1929 struct rb_node *parent = NULL;
1930 struct ceph_osd_event *event = NULL;
1931
1932 while (*p) {
1933 parent = *p;
1934 event = rb_entry(parent, struct ceph_osd_event, node);
1935 if (new->cookie < event->cookie)
1936 p = &(*p)->rb_left;
1937 else if (new->cookie > event->cookie)
1938 p = &(*p)->rb_right;
1939 else
1940 BUG();
1941 }
1942
1943 rb_link_node(&new->node, parent, p);
1944 rb_insert_color(&new->node, &osdc->event_tree);
1945}
1946
1947static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
1948 u64 cookie)
1949{
1950 struct rb_node **p = &osdc->event_tree.rb_node;
1951 struct rb_node *parent = NULL;
1952 struct ceph_osd_event *event = NULL;
1953
1954 while (*p) {
1955 parent = *p;
1956 event = rb_entry(parent, struct ceph_osd_event, node);
1957 if (cookie < event->cookie)
1958 p = &(*p)->rb_left;
1959 else if (cookie > event->cookie)
1960 p = &(*p)->rb_right;
1961 else
1962 return event;
1963 }
1964 return NULL;
1965}
1966
1967static void __remove_event(struct ceph_osd_event *event)
1968{
1969 struct ceph_osd_client *osdc = event->osdc;
1970
1971 if (!RB_EMPTY_NODE(&event->node)) {
1972 dout("__remove_event removed %p\n", event);
1973 rb_erase(&event->node, &osdc->event_tree);
1974 ceph_osdc_put_event(event);
1975 } else {
1976 dout("__remove_event didn't remove %p\n", event);
1977 }
1978}
1979
1980int ceph_osdc_create_event(struct ceph_osd_client *osdc,
1981 void (*event_cb)(u64, u64, u8, void *),
3c663bbd 1982 void *data, struct ceph_osd_event **pevent)
a40c4f10
YS		 1983{
1984 struct ceph_osd_event *event;
1985
1986 event = kmalloc(sizeof(*event), GFP_NOIO);
1987 if (!event)
1988 return -ENOMEM;
1989
1990 dout("create_event %p\n", event);
1991 event->cb = event_cb;
3c663bbd 1992 event->one_shot = 0;
a40c4f10
YS		 1993 event->data = data;
1994 event->osdc = osdc;
1995 INIT_LIST_HEAD(&event->osd_node);
3ee5234d 1996 RB_CLEAR_NODE(&event->node);
a40c4f10
YS		 1997 kref_init(&event->kref); /* one ref for us */
1998 kref_get(&event->kref); /* one ref for the caller */
a40c4f10
YS		 1999
2000 spin_lock(&osdc->event_lock);
2001 event->cookie = ++osdc->event_count;
2002 __insert_event(osdc, event);
2003 spin_unlock(&osdc->event_lock);
2004
2005 *pevent = event;
2006 return 0;
2007}
2008EXPORT_SYMBOL(ceph_osdc_create_event);
2009
2010void ceph_osdc_cancel_event(struct ceph_osd_event *event)
2011{
2012 struct ceph_osd_client *osdc = event->osdc;
2013
2014 dout("cancel_event %p\n", event);
2015 spin_lock(&osdc->event_lock);
2016 __remove_event(event);
2017 spin_unlock(&osdc->event_lock);
2018 ceph_osdc_put_event(event); /* caller's */
2019}
2020EXPORT_SYMBOL(ceph_osdc_cancel_event);
2021
2022
2023static void do_event_work(struct work_struct *work)
2024{
2025 struct ceph_osd_event_work *event_work =
2026 container_of(work, struct ceph_osd_event_work, work);
2027 struct ceph_osd_event *event = event_work->event;
2028 u64 ver = event_work->ver;
2029 u64 notify_id = event_work->notify_id;
2030 u8 opcode = event_work->opcode;
2031
2032 dout("do_event_work completing %p\n", event);
2033 event->cb(ver, notify_id, opcode, event->data);
a40c4f10
YS		 2034 dout("do_event_work completed %p\n", event);
2035 ceph_osdc_put_event(event);
2036 kfree(event_work);
2037}
2038
2039
2040/*
2041 * Process osd watch notifications
2042 */
3c663bbd
AE		 2043static void handle_watch_notify(struct ceph_osd_client *osdc,
2044 struct ceph_msg *msg)
a40c4f10
YS		 2045{
2046 void *p, *end;
2047 u8 proto_ver;
2048 u64 cookie, ver, notify_id;
2049 u8 opcode;
2050 struct ceph_osd_event *event;
2051 struct ceph_osd_event_work *event_work;
2052
2053 p = msg->front.iov_base;
2054 end = p + msg->front.iov_len;
2055
2056 ceph_decode_8_safe(&p, end, proto_ver, bad);
2057 ceph_decode_8_safe(&p, end, opcode, bad);
2058 ceph_decode_64_safe(&p, end, cookie, bad);
2059 ceph_decode_64_safe(&p, end, ver, bad);
2060 ceph_decode_64_safe(&p, end, notify_id, bad);
2061
2062 spin_lock(&osdc->event_lock);
2063 event = __find_event(osdc, cookie);
2064 if (event) {
3c663bbd 2065 BUG_ON(event->one_shot);
a40c4f10 2066 get_event(event);
a40c4f10
YS		 2067 }
2068 spin_unlock(&osdc->event_lock);
2069 dout("handle_watch_notify cookie %lld ver %lld event %p\n",
2070 cookie, ver, event);
2071 if (event) {
2072 event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
a40c4f10
YS		 2073 if (!event_work) {
2074 dout("ERROR: could not allocate event_work\n");
2075 goto done_err;
2076 }
6b0ae409 2077 INIT_WORK(&event_work->work, do_event_work);
a40c4f10
YS		 2078 event_work->event = event;
2079 event_work->ver = ver;
2080 event_work->notify_id = notify_id;
2081 event_work->opcode = opcode;
2082 if (!queue_work(osdc->notify_wq, &event_work->work)) {
2083 dout("WARNING: failed to queue notify event work\n");
2084 goto done_err;
2085 }
2086 }
2087
2088 return;
2089
2090done_err:
a40c4f10
YS		 2091 ceph_osdc_put_event(event);
2092 return;
2093
2094bad:
2095 pr_err("osdc handle_watch_notify corrupt msg\n");
2096 return;
2097}
2098
e65550fd
AE		 2099/*
2100 * build new request AND message
2101 *
2102 */
2103void ceph_osdc_build_request(struct ceph_osd_request *req, u64 off,
2104 struct ceph_snap_context *snapc, u64 snap_id,
2105 struct timespec *mtime)
2106{
2107 struct ceph_msg *msg = req->r_request;
2108 void *p;
2109 size_t msg_size;
2110 int flags = req->r_flags;
2111 u64 data_len;
2112 unsigned int i;
2113
2114 req->r_snapid = snap_id;
2115 req->r_snapc = ceph_get_snap_context(snapc);
2116
2117 /* encode request */
2118 msg->hdr.version = cpu_to_le16(4);
2119
2120 p = msg->front.iov_base;
2121 ceph_encode_32(&p, 1); /* client_inc is always 1 */
2122 req->r_request_osdmap_epoch = p;
2123 p += 4;
2124 req->r_request_flags = p;
2125 p += 4;
2126 if (req->r_flags & CEPH_OSD_FLAG_WRITE)
2127 ceph_encode_timespec(p, mtime);
2128 p += sizeof(struct ceph_timespec);
2129 req->r_request_reassert_version = p;
2130 p += sizeof(struct ceph_eversion); /* will get filled in */
2131
2132 /* oloc */
2133 ceph_encode_8(&p, 4);
2134 ceph_encode_8(&p, 4);
2135 ceph_encode_32(&p, 8 + 4 + 4);
2136 req->r_request_pool = p;
2137 p += 8;
2138 ceph_encode_32(&p, -1); /* preferred */
2139 ceph_encode_32(&p, 0); /* key len */
2140
2141 ceph_encode_8(&p, 1);
2142 req->r_request_pgid = p;
2143 p += 8 + 4;
2144 ceph_encode_32(&p, -1); /* preferred */
2145
2146 /* oid */
3c972c95
ID		 2147 ceph_encode_32(&p, req->r_base_oid.name_len);
2148 memcpy(p, req->r_base_oid.name, req->r_base_oid.name_len);
2149 dout("oid '%.*s' len %d\n", req->r_base_oid.name_len,
2150 req->r_base_oid.name, req->r_base_oid.name_len);
2151 p += req->r_base_oid.name_len;
e65550fd
AE		 2152
2153 /* ops--can imply data */
2154 ceph_encode_16(&p, (u16)req->r_num_ops);
2155 data_len = 0;
2156 for (i = 0; i < req->r_num_ops; i++) {
2157 data_len += osd_req_encode_op(req, p, i);
2158 p += sizeof(struct ceph_osd_op);
2159 }
2160
2161 /* snaps */
2162 ceph_encode_64(&p, req->r_snapid);
2163 ceph_encode_64(&p, req->r_snapc ? req->r_snapc->seq : 0);
2164 ceph_encode_32(&p, req->r_snapc ? req->r_snapc->num_snaps : 0);
2165 if (req->r_snapc) {
2166 for (i = 0; i < snapc->num_snaps; i++) {
2167 ceph_encode_64(&p, req->r_snapc->snaps[i]);
2168 }
2169 }
2170
2171 req->r_request_attempts = p;
2172 p += 4;
2173
2174 /* data */
2175 if (flags & CEPH_OSD_FLAG_WRITE) {
2176 u16 data_off;
2177
2178 /*
2179 * The header "data_off" is a hint to the receiver
2180 * allowing it to align received data into its
2181 * buffers such that there's no need to re-copy
2182 * it before writing it to disk (direct I/O).
2183 */
2184 data_off = (u16) (off & 0xffff);
2185 req->r_request->hdr.data_off = cpu_to_le16(data_off);
2186 }
2187 req->r_request->hdr.data_len = cpu_to_le32(data_len);
2188
2189 BUG_ON(p > msg->front.iov_base + msg->front.iov_len);
2190 msg_size = p - msg->front.iov_base;
2191 msg->front.iov_len = msg_size;
2192 msg->hdr.front_len = cpu_to_le32(msg_size);
2193
2194 dout("build_request msg_size was %d\n", (int)msg_size);
2195}
2196EXPORT_SYMBOL(ceph_osdc_build_request);
2197
70636773
AE		 2198/*
2199 * Register request, send initial attempt.
2200 */
2201int ceph_osdc_start_request(struct ceph_osd_client *osdc,
2202 struct ceph_osd_request *req,
2203 bool nofail)
2204{
2205 int rc = 0;
2206
f24e9980
SW		 2207 down_read(&osdc->map_sem);
2208 mutex_lock(&osdc->request_mutex);
dc4b870c 2209 __register_request(osdc, req);
c10ebbf5
AE		 2210 req->r_sent = 0;
2211 req->r_got_reply = 0;
92451b49
AE		 2212 rc = __map_request(osdc, req, 0);
2213 if (rc < 0) {
2214 if (nofail) {
2215 dout("osdc_start_request failed map, "
2216 " will retry %lld\n", req->r_tid);
2217 rc = 0;
73d9f7ee 2218 } else {
2219 __unregister_request(osdc, req);
f24e9980 2220 }
92451b49 2221 goto out_unlock;
f24e9980 2222 }
92451b49
AE		 2223 if (req->r_osd == NULL) {
2224 dout("send_request %p no up osds in pg\n", req);
2225 ceph_monc_request_next_osdmap(&osdc->client->monc);
2226 } else {
7e2766a1 2227 __send_queued(osdc);
92451b49
AE		 2228 }
2229 rc = 0;
234af26f 2230out_unlock:
f24e9980
SW		 2231 mutex_unlock(&osdc->request_mutex);
2232 up_read(&osdc->map_sem);
2233 return rc;
2234}
3d14c5d2 2235EXPORT_SYMBOL(ceph_osdc_start_request);
f24e9980
SW		 2236
2237/*
2238 * wait for a request to complete
2239 */
2240int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
2241 struct ceph_osd_request *req)
2242{
2243 int rc;
2244
2245 rc = wait_for_completion_interruptible(&req->r_completion);
2246 if (rc < 0) {
2247 mutex_lock(&osdc->request_mutex);
2248 __cancel_request(req);
529cfcc4 2249 __unregister_request(osdc, req);
f24e9980 2250 mutex_unlock(&osdc->request_mutex);
25845472 2251 complete_request(req);
529cfcc4 2252 dout("wait_request tid %llu canceled/timed out\n", req->r_tid);
f24e9980
SW		 2253 return rc;
2254 }
2255
2256 dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result);
2257 return req->r_result;
2258}
3d14c5d2 2259EXPORT_SYMBOL(ceph_osdc_wait_request);
f24e9980
SW		 2260
2261/*
2262 * sync - wait for all in-flight requests to flush. avoid starvation.
2263 */
2264void ceph_osdc_sync(struct ceph_osd_client *osdc)
2265{
2266 struct ceph_osd_request *req;
2267 u64 last_tid, next_tid = 0;
2268
2269 mutex_lock(&osdc->request_mutex);
2270 last_tid = osdc->last_tid;
2271 while (1) {
2272 req = __lookup_request_ge(osdc, next_tid);
2273 if (!req)
2274 break;
2275 if (req->r_tid > last_tid)
2276 break;
2277
2278 next_tid = req->r_tid + 1;
2279 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
2280 continue;
2281
2282 ceph_osdc_get_request(req);
2283 mutex_unlock(&osdc->request_mutex);
2284 dout("sync waiting on tid %llu (last is %llu)\n",
2285 req->r_tid, last_tid);
2286 wait_for_completion(&req->r_safe_completion);
2287 mutex_lock(&osdc->request_mutex);
2288 ceph_osdc_put_request(req);
2289 }
2290 mutex_unlock(&osdc->request_mutex);
2291 dout("sync done (thru tid %llu)\n", last_tid);
2292}
3d14c5d2 2293EXPORT_SYMBOL(ceph_osdc_sync);
f24e9980 2294
dd935f44
JD		 2295/*
2296 * Call all pending notify callbacks - for use after a watch is
2297 * unregistered, to make sure no more callbacks for it will be invoked
2298 */
2299extern void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc)
2300{
2301 flush_workqueue(osdc->notify_wq);
2302}
2303EXPORT_SYMBOL(ceph_osdc_flush_notifies);
2304
2305
f24e9980
SW		 2306/*
2307 * init, shutdown
2308 */
2309int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
2310{
2311 int err;
2312
2313 dout("init\n");
2314 osdc->client = client;
2315 osdc->osdmap = NULL;
2316 init_rwsem(&osdc->map_sem);
2317 init_completion(&osdc->map_waiters);
2318 osdc->last_requested_map = 0;
2319 mutex_init(&osdc->request_mutex);
f24e9980
SW		 2320 osdc->last_tid = 0;
2321 osdc->osds = RB_ROOT;
f5a2041b 2322 INIT_LIST_HEAD(&osdc->osd_lru);
f24e9980 2323 osdc->requests = RB_ROOT;
422d2cb8 2324 INIT_LIST_HEAD(&osdc->req_lru);
6f6c7006
SW		 2325 INIT_LIST_HEAD(&osdc->req_unsent);
2326 INIT_LIST_HEAD(&osdc->req_notarget);
a40c4f10 2327 INIT_LIST_HEAD(&osdc->req_linger);
f24e9980
SW		 2328 osdc->num_requests = 0;
2329 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
f5a2041b 2330 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
a40c4f10
YS		 2331 spin_lock_init(&osdc->event_lock);
2332 osdc->event_tree = RB_ROOT;
2333 osdc->event_count = 0;
f5a2041b
YS		 2334
2335 schedule_delayed_work(&osdc->osds_timeout_work,
3d14c5d2 2336 round_jiffies_relative(osdc->client->options->osd_idle_ttl * HZ));
f24e9980 2337
5f44f142 2338 err = -ENOMEM;
f24e9980
SW		 2339 osdc->req_mempool = mempool_create_kmalloc_pool(10,
2340 sizeof(struct ceph_osd_request));
2341 if (!osdc->req_mempool)
5f44f142 2342 goto out;
f24e9980 2343
d50b409f
SW		 2344 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
2345 OSD_OP_FRONT_LEN, 10, true,
4f48280e 2346 "osd_op");
f24e9980 2347 if (err < 0)
5f44f142 2348 goto out_mempool;
d50b409f 2349 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
4f48280e
SW		 2350 OSD_OPREPLY_FRONT_LEN, 10, true,
2351 "osd_op_reply");
c16e7869
SW		 2352 if (err < 0)
2353 goto out_msgpool;
a40c4f10 2354
dbcae088 2355 err = -ENOMEM;
a40c4f10 2356 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
dbcae088 2357 if (!osdc->notify_wq)
a40c4f10 2358 goto out_msgpool;
f24e9980 2359 return 0;
5f44f142 2360
c16e7869
SW		 2361out_msgpool:
2362 ceph_msgpool_destroy(&osdc->msgpool_op);
5f44f142
SW		 2363out_mempool:
2364 mempool_destroy(osdc->req_mempool);
2365out:
2366 return err;
f24e9980
SW		 2367}
2368
2369void ceph_osdc_stop(struct ceph_osd_client *osdc)
2370{
a40c4f10
YS		 2371 flush_workqueue(osdc->notify_wq);
2372 destroy_workqueue(osdc->notify_wq);
f24e9980 2373 cancel_delayed_work_sync(&osdc->timeout_work);
f5a2041b 2374 cancel_delayed_work_sync(&osdc->osds_timeout_work);
f24e9980
SW		 2375 if (osdc->osdmap) {
2376 ceph_osdmap_destroy(osdc->osdmap);
2377 osdc->osdmap = NULL;
2378 }
aca420bc 2379 remove_all_osds(osdc);
f24e9980
SW		 2380 mempool_destroy(osdc->req_mempool);
2381 ceph_msgpool_destroy(&osdc->msgpool_op);
c16e7869 2382 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
f24e9980
SW		 2383}
2384
2385/*
2386 * Read some contiguous pages. If we cross a stripe boundary, shorten
2387 * *plen. Return number of bytes read, or error.
2388 */
2389int ceph_osdc_readpages(struct ceph_osd_client *osdc,
2390 struct ceph_vino vino, struct ceph_file_layout *layout,
2391 u64 off, u64 *plen,
2392 u32 truncate_seq, u64 truncate_size,
b7495fc2 2393 struct page **pages, int num_pages, int page_align)
f24e9980
SW		 2394{
2395 struct ceph_osd_request *req;
2396 int rc = 0;
2397
2398 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
2399 vino.snap, off, *plen);
79528734 2400 req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 1,
f24e9980 2401 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
acead002 2402 NULL, truncate_seq, truncate_size,
153e5167 2403 false);
6816282d
SW		 2404 if (IS_ERR(req))
2405 return PTR_ERR(req);
f24e9980
SW		 2406
2407 /* it may be a short read due to an object boundary */
0fff87ec 2408
406e2c9f 2409 osd_req_op_extent_osd_data_pages(req, 0,
a4ce40a9 2410 pages, *plen, page_align, false, false);
f24e9980 2411
e0c59487 2412 dout("readpages final extent is %llu~%llu (%llu bytes align %d)\n",
43bfe5de 2413 off, *plen, *plen, page_align);
f24e9980 2414
79528734 2415 ceph_osdc_build_request(req, off, NULL, vino.snap, NULL);
02ee07d3 2416
f24e9980
SW		 2417 rc = ceph_osdc_start_request(osdc, req, false);
2418 if (!rc)
2419 rc = ceph_osdc_wait_request(osdc, req);
2420
2421 ceph_osdc_put_request(req);
2422 dout("readpages result %d\n", rc);
2423 return rc;
2424}
3d14c5d2 2425EXPORT_SYMBOL(ceph_osdc_readpages);
f24e9980
SW		 2426
2427/*
2428 * do a synchronous write on N pages
2429 */
2430int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
2431 struct ceph_file_layout *layout,
2432 struct ceph_snap_context *snapc,
2433 u64 off, u64 len,
2434 u32 truncate_seq, u64 truncate_size,
2435 struct timespec *mtime,
24808826 2436 struct page **pages, int num_pages)
f24e9980
SW		 2437{
2438 struct ceph_osd_request *req;
2439 int rc = 0;
b7495fc2 2440 int page_align = off & ~PAGE_MASK;
f24e9980 2441
acead002 2442 BUG_ON(vino.snap != CEPH_NOSNAP); /* snapshots aren't writeable */
79528734 2443 req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 1,
f24e9980 2444 CEPH_OSD_OP_WRITE,
24808826 2445 CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
acead002 2446 snapc, truncate_seq, truncate_size,
153e5167 2447 true);
6816282d
SW		 2448 if (IS_ERR(req))
2449 return PTR_ERR(req);
f24e9980
SW		 2450
2451 /* it may be a short write due to an object boundary */
406e2c9f 2452 osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align,
43bfe5de
AE		 2453 false, false);
2454 dout("writepages %llu~%llu (%llu bytes)\n", off, len, len);
f24e9980 2455
79528734 2456 ceph_osdc_build_request(req, off, snapc, CEPH_NOSNAP, mtime);
02ee07d3 2457
87f979d3 2458 rc = ceph_osdc_start_request(osdc, req, true);
f24e9980
SW		 2459 if (!rc)
2460 rc = ceph_osdc_wait_request(osdc, req);
2461
2462 ceph_osdc_put_request(req);
2463 if (rc == 0)
2464 rc = len;
2465 dout("writepages result %d\n", rc);
2466 return rc;
2467}
3d14c5d2 2468EXPORT_SYMBOL(ceph_osdc_writepages);
f24e9980 2469
5522ae0b
AE		 2470int ceph_osdc_setup(void)
2471{
2472 BUG_ON(ceph_osd_request_cache);
2473 ceph_osd_request_cache = kmem_cache_create("ceph_osd_request",
2474 sizeof (struct ceph_osd_request),
2475 __alignof__(struct ceph_osd_request),
2476 0, NULL);
2477
2478 return ceph_osd_request_cache ? 0 : -ENOMEM;
2479}
2480EXPORT_SYMBOL(ceph_osdc_setup);
2481
2482void ceph_osdc_cleanup(void)
2483{
2484 BUG_ON(!ceph_osd_request_cache);
2485 kmem_cache_destroy(ceph_osd_request_cache);
2486 ceph_osd_request_cache = NULL;
2487}
2488EXPORT_SYMBOL(ceph_osdc_cleanup);
2489
f24e9980
SW		 2490/*
2491 * handle incoming message
2492 */
2493static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
2494{
2495 struct ceph_osd *osd = con->private;
32c895e7 2496 struct ceph_osd_client *osdc;
f24e9980
SW		 2497 int type = le16_to_cpu(msg->hdr.type);
2498
2499 if (!osd)
4a32f93d 2500 goto out;
32c895e7 2501 osdc = osd->o_osdc;
f24e9980
SW		 2502
2503 switch (type) {
2504 case CEPH_MSG_OSD_MAP:
2505 ceph_osdc_handle_map(osdc, msg);
2506 break;
2507 case CEPH_MSG_OSD_OPREPLY:
350b1c32 2508 handle_reply(osdc, msg, con);
f24e9980 2509 break;
a40c4f10
YS		 2510 case CEPH_MSG_WATCH_NOTIFY:
2511 handle_watch_notify(osdc, msg);
2512 break;
f24e9980
SW		 2513
2514 default:
2515 pr_err("received unknown message type %d %s\n", type,
2516 ceph_msg_type_name(type));
2517 }
4a32f93d 2518out:
f24e9980
SW		 2519 ceph_msg_put(msg);
2520}
2521
5b3a4db3 2522/*
21b667f6
SW		 2523 * lookup and return message for incoming reply. set up reply message
2524 * pages.
5b3a4db3
SW		 2525 */
2526static struct ceph_msg *get_reply(struct ceph_connection *con,
2450418c
YS		 2527 struct ceph_msg_header *hdr,
2528 int *skip)
f24e9980
SW		 2529{
2530 struct ceph_osd *osd = con->private;
2531 struct ceph_osd_client *osdc = osd->o_osdc;
2450418c 2532 struct ceph_msg *m;
0547a9b3 2533 struct ceph_osd_request *req;
3f0a4ac5 2534 int front_len = le32_to_cpu(hdr->front_len);
5b3a4db3 2535 int data_len = le32_to_cpu(hdr->data_len);
0547a9b3 2536 u64 tid;
f24e9980 2537
0547a9b3
YS		 2538 tid = le64_to_cpu(hdr->tid);
2539 mutex_lock(&osdc->request_mutex);
2540 req = __lookup_request(osdc, tid);
2541 if (!req) {
2542 *skip = 1;
2543 m = NULL;
756a16a5
SW		 2544 dout("get_reply unknown tid %llu from osd%d\n", tid,
2545 osd->o_osd);
0547a9b3
YS		 2546 goto out;
2547 }
c16e7869 2548
ace6d3a9 2549 if (req->r_reply->con)
8921d114 2550 dout("%s revoking msg %p from old con %p\n", __func__,
ace6d3a9
AE		 2551 req->r_reply, req->r_reply->con);
2552 ceph_msg_revoke_incoming(req->r_reply);
0547a9b3 2553
f2be82b0 2554 if (front_len > req->r_reply->front_alloc_len) {
4974341e 2555 pr_warning("get_reply front %d > preallocated %d (%u#%llu)\n",
f2be82b0 2556 front_len, req->r_reply->front_alloc_len,
4974341e
AE		 2557 (unsigned int)con->peer_name.type,
2558 le64_to_cpu(con->peer_name.num));
3f0a4ac5
ID		 2559 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS,
2560 false);
a79832f2 2561 if (!m)
c16e7869
SW		 2562 goto out;
2563 ceph_msg_put(req->r_reply);
2564 req->r_reply = m;
2565 }
2566 m = ceph_msg_get(req->r_reply);
2567
0547a9b3 2568 if (data_len > 0) {
a4ce40a9 2569 struct ceph_osd_data *osd_data;
0fff87ec 2570
a4ce40a9
AE		 2571 /*
2572 * XXX This is assuming there is only one op containing
2573 * XXX page data. Probably OK for reads, but this
2574 * XXX ought to be done more generally.
2575 */
406e2c9f 2576 osd_data = osd_req_op_extent_osd_data(req, 0);
0fff87ec 2577 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
0fff87ec 2578 if (osd_data->pages &&
e0c59487 2579 unlikely(osd_data->length < data_len)) {
2ac2b7a6 2580
e0c59487
AE		 2581 pr_warning("tid %lld reply has %d bytes "
2582 "we had only %llu bytes ready\n",
2583 tid, data_len, osd_data->length);
2ac2b7a6
AE		 2584 *skip = 1;
2585 ceph_msg_put(m);
2586 m = NULL;
2587 goto out;
2588 }
2ac2b7a6 2589 }
0547a9b3 2590 }
5b3a4db3 2591 *skip = 0;
c16e7869 2592 dout("get_reply tid %lld %p\n", tid, m);
0547a9b3
YS		 2593
2594out:
2595 mutex_unlock(&osdc->request_mutex);
2450418c 2596 return m;
5b3a4db3
SW		 2597
2598}
2599
2600static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2601 struct ceph_msg_header *hdr,
2602 int *skip)
2603{
2604 struct ceph_osd *osd = con->private;
2605 int type = le16_to_cpu(hdr->type);
2606 int front = le32_to_cpu(hdr->front_len);
2607
1c20f2d2 2608 *skip = 0;
5b3a4db3
SW		 2609 switch (type) {
2610 case CEPH_MSG_OSD_MAP:
a40c4f10 2611 case CEPH_MSG_WATCH_NOTIFY:
b61c2763 2612 return ceph_msg_new(type, front, GFP_NOFS, false);
5b3a4db3
SW		 2613 case CEPH_MSG_OSD_OPREPLY:
2614 return get_reply(con, hdr, skip);
2615 default:
2616 pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
2617 osd->o_osd);
2618 *skip = 1;
2619 return NULL;
2620 }
f24e9980
SW		 2621}
2622
2623/*
2624 * Wrappers to refcount containing ceph_osd struct
2625 */
2626static struct ceph_connection *get_osd_con(struct ceph_connection *con)
2627{
2628 struct ceph_osd *osd = con->private;
2629 if (get_osd(osd))
2630 return con;
2631 return NULL;
2632}
2633
2634static void put_osd_con(struct ceph_connection *con)
2635{
2636 struct ceph_osd *osd = con->private;
2637 put_osd(osd);
2638}
2639
4e7a5dcd
SW		 2640/*
2641 * authentication
2642 */
a3530df3
AE		 2643/*
2644 * Note: returned pointer is the address of a structure that's
2645 * managed separately. Caller must *not* attempt to free it.
2646 */
2647static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
8f43fb53 2648 int *proto, int force_new)
4e7a5dcd
SW		 2649{
2650 struct ceph_osd *o = con->private;
2651 struct ceph_osd_client *osdc = o->o_osdc;
2652 struct ceph_auth_client *ac = osdc->client->monc.auth;
74f1869f 2653 struct ceph_auth_handshake *auth = &o->o_auth;
4e7a5dcd 2654
74f1869f 2655 if (force_new && auth->authorizer) {
27859f97 2656 ceph_auth_destroy_authorizer(ac, auth->authorizer);
74f1869f
AE		 2657 auth->authorizer = NULL;
2658 }
27859f97
SW		 2659 if (!auth->authorizer) {
2660 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2661 auth);
4e7a5dcd 2662 if (ret)
a3530df3 2663 return ERR_PTR(ret);
27859f97
SW		 2664 } else {
2665 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
0bed9b5c
SW		 2666 auth);
2667 if (ret)
2668 return ERR_PTR(ret);
4e7a5dcd 2669 }
4e7a5dcd 2670 *proto = ac->protocol;
74f1869f 2671
a3530df3 2672 return auth;
4e7a5dcd
SW		 2673}
2674
2675
2676static int verify_authorizer_reply(struct ceph_connection *con, int len)
2677{
2678 struct ceph_osd *o = con->private;
2679 struct ceph_osd_client *osdc = o->o_osdc;
2680 struct ceph_auth_client *ac = osdc->client->monc.auth;
2681
27859f97 2682 return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer, len);
4e7a5dcd
SW		 2683}
2684
9bd2e6f8
SW		 2685static int invalidate_authorizer(struct ceph_connection *con)
2686{
2687 struct ceph_osd *o = con->private;
2688 struct ceph_osd_client *osdc = o->o_osdc;
2689 struct ceph_auth_client *ac = osdc->client->monc.auth;
2690
27859f97 2691 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
9bd2e6f8
SW		 2692 return ceph_monc_validate_auth(&osdc->client->monc);
2693}
4e7a5dcd 2694
9e32789f 2695static const struct ceph_connection_operations osd_con_ops = {
f24e9980
SW		 2696 .get = get_osd_con,
2697 .put = put_osd_con,
2698 .dispatch = dispatch,
4e7a5dcd
SW		 2699 .get_authorizer = get_authorizer,
2700 .verify_authorizer_reply = verify_authorizer_reply,
9bd2e6f8 2701 .invalidate_authorizer = invalidate_authorizer,
f24e9980 2702 .alloc_msg = alloc_msg,
81b024e7 2703 .fault = osd_reset,
f24e9980 2704};
Linux 6.x block layer and io_uring tree(s)