projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
libceph: factor out logic from ceph_osdc_start_request()
[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;
205ee118 372 req->r_target_oloc.pool = -1;
22116525 373
c16e7869
SW		 374 /* create reply message */
375 if (use_mempool)
376 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0);
377 else
378 msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY,
b61c2763 379 OSD_OPREPLY_FRONT_LEN, gfp_flags, true);
a79832f2 380 if (!msg) {
c16e7869 381 ceph_osdc_put_request(req);
a79832f2 382 return NULL;
c16e7869
SW		 383 }
384 req->r_reply = msg;
385
386 /* create request message; allow space for oid */
f24e9980 387 if (use_mempool)
8f3bc053 388 msg = ceph_msgpool_get(&osdc->msgpool_op, 0);
f24e9980 389 else
b61c2763 390 msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp_flags, true);
a79832f2 391 if (!msg) {
f24e9980 392 ceph_osdc_put_request(req);
a79832f2 393 return NULL;
f24e9980 394 }
68b4476b 395
f24e9980 396 memset(msg->front.iov_base, 0, msg->front.iov_len);
3499e8a5
YS		 397
398 req->r_request = msg;
3499e8a5
YS		 399
400 return req;
401}
3d14c5d2 402EXPORT_SYMBOL(ceph_osdc_alloc_request);
3499e8a5 403
a8dd0a37 404static bool osd_req_opcode_valid(u16 opcode)
68b4476b 405{
a8dd0a37 406 switch (opcode) {
68b4476b 407 case CEPH_OSD_OP_READ:
a8dd0a37 408 case CEPH_OSD_OP_STAT:
4c46459c
AE		 409 case CEPH_OSD_OP_MAPEXT:
410 case CEPH_OSD_OP_MASKTRUNC:
411 case CEPH_OSD_OP_SPARSE_READ:
a9f36c3e 412 case CEPH_OSD_OP_NOTIFY:
a8dd0a37 413 case CEPH_OSD_OP_NOTIFY_ACK:
4c46459c 414 case CEPH_OSD_OP_ASSERT_VER:
a8dd0a37 415 case CEPH_OSD_OP_WRITE:
4c46459c
AE		 416 case CEPH_OSD_OP_WRITEFULL:
417 case CEPH_OSD_OP_TRUNCATE:
418 case CEPH_OSD_OP_ZERO:
419 case CEPH_OSD_OP_DELETE:
420 case CEPH_OSD_OP_APPEND:
a8dd0a37 421 case CEPH_OSD_OP_STARTSYNC:
4c46459c
AE		 422 case CEPH_OSD_OP_SETTRUNC:
423 case CEPH_OSD_OP_TRIMTRUNC:
424 case CEPH_OSD_OP_TMAPUP:
425 case CEPH_OSD_OP_TMAPPUT:
426 case CEPH_OSD_OP_TMAPGET:
427 case CEPH_OSD_OP_CREATE:
a9f36c3e 428 case CEPH_OSD_OP_ROLLBACK:
a8dd0a37 429 case CEPH_OSD_OP_WATCH:
4c46459c
AE		 430 case CEPH_OSD_OP_OMAPGETKEYS:
431 case CEPH_OSD_OP_OMAPGETVALS:
432 case CEPH_OSD_OP_OMAPGETHEADER:
433 case CEPH_OSD_OP_OMAPGETVALSBYKEYS:
4c46459c
AE		 434 case CEPH_OSD_OP_OMAPSETVALS:
435 case CEPH_OSD_OP_OMAPSETHEADER:
436 case CEPH_OSD_OP_OMAPCLEAR:
437 case CEPH_OSD_OP_OMAPRMKEYS:
438 case CEPH_OSD_OP_OMAP_CMP:
439 case CEPH_OSD_OP_CLONERANGE:
440 case CEPH_OSD_OP_ASSERT_SRC_VERSION:
441 case CEPH_OSD_OP_SRC_CMPXATTR:
a9f36c3e 442 case CEPH_OSD_OP_GETXATTR:
4c46459c 443 case CEPH_OSD_OP_GETXATTRS:
a9f36c3e
AE		 444 case CEPH_OSD_OP_CMPXATTR:
445 case CEPH_OSD_OP_SETXATTR:
4c46459c
AE		 446 case CEPH_OSD_OP_SETXATTRS:
447 case CEPH_OSD_OP_RESETXATTRS:
448 case CEPH_OSD_OP_RMXATTR:
449 case CEPH_OSD_OP_PULL:
450 case CEPH_OSD_OP_PUSH:
451 case CEPH_OSD_OP_BALANCEREADS:
452 case CEPH_OSD_OP_UNBALANCEREADS:
453 case CEPH_OSD_OP_SCRUB:
454 case CEPH_OSD_OP_SCRUB_RESERVE:
455 case CEPH_OSD_OP_SCRUB_UNRESERVE:
456 case CEPH_OSD_OP_SCRUB_STOP:
457 case CEPH_OSD_OP_SCRUB_MAP:
458 case CEPH_OSD_OP_WRLOCK:
459 case CEPH_OSD_OP_WRUNLOCK:
460 case CEPH_OSD_OP_RDLOCK:
461 case CEPH_OSD_OP_RDUNLOCK:
462 case CEPH_OSD_OP_UPLOCK:
463 case CEPH_OSD_OP_DNLOCK:
a8dd0a37 464 case CEPH_OSD_OP_CALL:
4c46459c
AE		 465 case CEPH_OSD_OP_PGLS:
466 case CEPH_OSD_OP_PGLS_FILTER:
a8dd0a37
AE		 467 return true;
468 default:
469 return false;
470 }
471}
472
33803f33
AE		 473/*
474 * This is an osd op init function for opcodes that have no data or
475 * other information associated with them. It also serves as a
476 * common init routine for all the other init functions, below.
477 */
c99d2d4a 478static struct ceph_osd_req_op *
49719778 479_osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which,
c99d2d4a 480 u16 opcode)
33803f33 481{
c99d2d4a
AE		 482 struct ceph_osd_req_op *op;
483
484 BUG_ON(which >= osd_req->r_num_ops);
33803f33
AE		 485 BUG_ON(!osd_req_opcode_valid(opcode));
486
c99d2d4a 487 op = &osd_req->r_ops[which];
33803f33 488 memset(op, 0, sizeof (*op));
33803f33 489 op->op = opcode;
c99d2d4a
AE		 490
491 return op;
33803f33
AE		 492}
493
49719778
AE		 494void osd_req_op_init(struct ceph_osd_request *osd_req,
495 unsigned int which, u16 opcode)
496{
497 (void)_osd_req_op_init(osd_req, which, opcode);
498}
499EXPORT_SYMBOL(osd_req_op_init);
500
c99d2d4a
AE		 501void osd_req_op_extent_init(struct ceph_osd_request *osd_req,
502 unsigned int which, u16 opcode,
33803f33
AE		 503 u64 offset, u64 length,
504 u64 truncate_size, u32 truncate_seq)
505{
49719778 506 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
33803f33
AE		 507 size_t payload_len = 0;
508
ad7a60de
LW		 509 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
510 opcode != CEPH_OSD_OP_DELETE && opcode != CEPH_OSD_OP_ZERO &&
511 opcode != CEPH_OSD_OP_TRUNCATE);
33803f33 512
33803f33
AE		 513 op->extent.offset = offset;
514 op->extent.length = length;
515 op->extent.truncate_size = truncate_size;
516 op->extent.truncate_seq = truncate_seq;
517 if (opcode == CEPH_OSD_OP_WRITE)
518 payload_len += length;
519
520 op->payload_len = payload_len;
521}
522EXPORT_SYMBOL(osd_req_op_extent_init);
523
c99d2d4a
AE		 524void osd_req_op_extent_update(struct ceph_osd_request *osd_req,
525 unsigned int which, u64 length)
e5975c7c 526{
c99d2d4a
AE		 527 struct ceph_osd_req_op *op;
528 u64 previous;
529
530 BUG_ON(which >= osd_req->r_num_ops);
531 op = &osd_req->r_ops[which];
532 previous = op->extent.length;
e5975c7c
AE		 533
534 if (length == previous)
535 return; /* Nothing to do */
536 BUG_ON(length > previous);
537
538 op->extent.length = length;
539 op->payload_len -= previous - length;
540}
541EXPORT_SYMBOL(osd_req_op_extent_update);
542
c99d2d4a 543void osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
04017e29 544 u16 opcode, const char *class, const char *method)
33803f33 545{
49719778 546 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
5f562df5 547 struct ceph_pagelist *pagelist;
33803f33
AE		 548 size_t payload_len = 0;
549 size_t size;
550
551 BUG_ON(opcode != CEPH_OSD_OP_CALL);
552
5f562df5
AE		 553 pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
554 BUG_ON(!pagelist);
555 ceph_pagelist_init(pagelist);
556
33803f33
AE		 557 op->cls.class_name = class;
558 size = strlen(class);
559 BUG_ON(size > (size_t) U8_MAX);
560 op->cls.class_len = size;
5f562df5 561 ceph_pagelist_append(pagelist, class, size);
33803f33
AE		 562 payload_len += size;
563
564 op->cls.method_name = method;
565 size = strlen(method);
566 BUG_ON(size > (size_t) U8_MAX);
567 op->cls.method_len = size;
5f562df5 568 ceph_pagelist_append(pagelist, method, size);
33803f33
AE		 569 payload_len += size;
570
a4ce40a9 571 osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist);
5f562df5 572
33803f33
AE		 573 op->cls.argc = 0; /* currently unused */
574
575 op->payload_len = payload_len;
576}
577EXPORT_SYMBOL(osd_req_op_cls_init);
8c042b0d 578
c99d2d4a
AE		 579void osd_req_op_watch_init(struct ceph_osd_request *osd_req,
580 unsigned int which, u16 opcode,
33803f33
AE		 581 u64 cookie, u64 version, int flag)
582{
49719778 583 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
33803f33 584
c99d2d4a 585 BUG_ON(opcode != CEPH_OSD_OP_NOTIFY_ACK && opcode != CEPH_OSD_OP_WATCH);
33803f33
AE		 586
587 op->watch.cookie = cookie;
9ef1ee5a 588 op->watch.ver = version;
33803f33 589 if (opcode == CEPH_OSD_OP_WATCH && flag)
c99d2d4a 590 op->watch.flag = (u8)1;
33803f33
AE		 591}
592EXPORT_SYMBOL(osd_req_op_watch_init);
593
90af3602 594static void ceph_osdc_msg_data_add(struct ceph_msg *msg,
ec9123c5
AE		 595 struct ceph_osd_data *osd_data)
596{
597 u64 length = ceph_osd_data_length(osd_data);
598
599 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
600 BUG_ON(length > (u64) SIZE_MAX);
601 if (length)
90af3602 602 ceph_msg_data_add_pages(msg, osd_data->pages,
ec9123c5
AE		 603 length, osd_data->alignment);
604 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
605 BUG_ON(!length);
90af3602 606 ceph_msg_data_add_pagelist(msg, osd_data->pagelist);
ec9123c5
AE		 607#ifdef CONFIG_BLOCK
608 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
90af3602 609 ceph_msg_data_add_bio(msg, osd_data->bio, length);
ec9123c5
AE		 610#endif
611 } else {
612 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
613 }
614}
615
a8dd0a37 616static u64 osd_req_encode_op(struct ceph_osd_request *req,
79528734 617 struct ceph_osd_op *dst, unsigned int which)
a8dd0a37 618{
79528734 619 struct ceph_osd_req_op *src;
04017e29 620 struct ceph_osd_data *osd_data;
54d50649 621 u64 request_data_len = 0;
04017e29 622 u64 data_length;
a8dd0a37 623
79528734
AE		 624 BUG_ON(which >= req->r_num_ops);
625 src = &req->r_ops[which];
a8dd0a37
AE		 626 if (WARN_ON(!osd_req_opcode_valid(src->op))) {
627 pr_err("unrecognized osd opcode %d\n", src->op);
628
629 return 0;
630 }
631
632 switch (src->op) {
633 case CEPH_OSD_OP_STAT:
49719778
AE		 634 osd_data = &src->raw_data_in;
635 ceph_osdc_msg_data_add(req->r_reply, osd_data);
a8dd0a37
AE		 636 break;
637 case CEPH_OSD_OP_READ:
638 case CEPH_OSD_OP_WRITE:
ad7a60de
LW		 639 case CEPH_OSD_OP_ZERO:
640 case CEPH_OSD_OP_DELETE:
641 case CEPH_OSD_OP_TRUNCATE:
a8dd0a37 642 if (src->op == CEPH_OSD_OP_WRITE)
54d50649 643 request_data_len = src->extent.length;
a8dd0a37
AE		 644 dst->extent.offset = cpu_to_le64(src->extent.offset);
645 dst->extent.length = cpu_to_le64(src->extent.length);
646 dst->extent.truncate_size =
647 cpu_to_le64(src->extent.truncate_size);
648 dst->extent.truncate_seq =
649 cpu_to_le32(src->extent.truncate_seq);
04017e29 650 osd_data = &src->extent.osd_data;
5476492f 651 if (src->op == CEPH_OSD_OP_WRITE)
04017e29 652 ceph_osdc_msg_data_add(req->r_request, osd_data);
5476492f 653 else
04017e29 654 ceph_osdc_msg_data_add(req->r_reply, osd_data);
a8dd0a37
AE		 655 break;
656 case CEPH_OSD_OP_CALL:
a8dd0a37
AE		 657 dst->cls.class_len = src->cls.class_len;
658 dst->cls.method_len = src->cls.method_len;
04017e29
AE		 659 osd_data = &src->cls.request_info;
660 ceph_osdc_msg_data_add(req->r_request, osd_data);
661 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGELIST);
662 request_data_len = osd_data->pagelist->length;
663
664 osd_data = &src->cls.request_data;
665 data_length = ceph_osd_data_length(osd_data);
666 if (data_length) {
667 BUG_ON(osd_data->type == CEPH_OSD_DATA_TYPE_NONE);
668 dst->cls.indata_len = cpu_to_le32(data_length);
669 ceph_osdc_msg_data_add(req->r_request, osd_data);
670 src->payload_len += data_length;
671 request_data_len += data_length;
672 }
673 osd_data = &src->cls.response_data;
674 ceph_osdc_msg_data_add(req->r_reply, osd_data);
a8dd0a37
AE		 675 break;
676 case CEPH_OSD_OP_STARTSYNC:
677 break;
678 case CEPH_OSD_OP_NOTIFY_ACK:
679 case CEPH_OSD_OP_WATCH:
680 dst->watch.cookie = cpu_to_le64(src->watch.cookie);
681 dst->watch.ver = cpu_to_le64(src->watch.ver);
682 dst->watch.flag = src->watch.flag;
683 break;
684 default:
4c46459c 685 pr_err("unsupported osd opcode %s\n",
8f63ca2d 686 ceph_osd_op_name(src->op));
4c46459c 687 WARN_ON(1);
a8dd0a37
AE		 688
689 return 0;
68b4476b 690 }
a8dd0a37 691 dst->op = cpu_to_le16(src->op);
68b4476b 692 dst->payload_len = cpu_to_le32(src->payload_len);
175face2 693
54d50649 694 return request_data_len;
68b4476b
YS		 695}
696
3499e8a5
YS		 697/*
698 * build new request AND message, calculate layout, and adjust file
699 * extent as needed.
700 *
701 * if the file was recently truncated, we include information about its
702 * old and new size so that the object can be updated appropriately. (we
703 * avoid synchronously deleting truncated objects because it's slow.)
704 *
705 * if @do_sync, include a 'startsync' command so that the osd will flush
706 * data quickly.
707 */
708struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
709 struct ceph_file_layout *layout,
710 struct ceph_vino vino,
acead002 711 u64 off, u64 *plen, int num_ops,
3499e8a5
YS		 712 int opcode, int flags,
713 struct ceph_snap_context *snapc,
3499e8a5
YS		 714 u32 truncate_seq,
715 u64 truncate_size,
153e5167 716 bool use_mempool)
3499e8a5 717{
68b4476b 718 struct ceph_osd_request *req;
75d1c941
AE		 719 u64 objnum = 0;
720 u64 objoff = 0;
721 u64 objlen = 0;
d18d1e28
AE		 722 u32 object_size;
723 u64 object_base;
6816282d 724 int r;
68b4476b 725
ad7a60de
LW		 726 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
727 opcode != CEPH_OSD_OP_DELETE && opcode != CEPH_OSD_OP_ZERO &&
728 opcode != CEPH_OSD_OP_TRUNCATE);
68b4476b 729
acead002 730 req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool,
ae7ca4a3 731 GFP_NOFS);
4ad12621 732 if (!req)
6816282d 733 return ERR_PTR(-ENOMEM);
79528734 734
d178a9e7 735 req->r_flags = flags;
3499e8a5
YS		 736
737 /* calculate max write size */
a19dadfb 738 r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
3ff5f385
AE		 739 if (r < 0) {
740 ceph_osdc_put_request(req);
6816282d 741 return ERR_PTR(r);
3ff5f385 742 }
a19dadfb 743
d18d1e28
AE		 744 object_size = le32_to_cpu(layout->fl_object_size);
745 object_base = off - objoff;
ccca4e37
YZ		 746 if (!(truncate_seq == 1 && truncate_size == -1ULL)) {
747 if (truncate_size <= object_base) {
748 truncate_size = 0;
749 } else {
750 truncate_size -= object_base;
751 if (truncate_size > object_size)
752 truncate_size = object_size;
753 }
a19dadfb 754 }
d18d1e28 755
c99d2d4a 756 osd_req_op_extent_init(req, 0, opcode, objoff, objlen,
b0270324 757 truncate_size, truncate_seq);
8c042b0d 758
acead002
AE		 759 /*
760 * A second op in the ops array means the caller wants to
761 * also issue a include a 'startsync' command so that the
762 * osd will flush data quickly.
763 */
764 if (num_ops > 1)
c99d2d4a 765 osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC);
d18d1e28 766
3c972c95 767 req->r_base_oloc.pool = ceph_file_layout_pg_pool(*layout);
3499e8a5 768
3c972c95 769 snprintf(req->r_base_oid.name, sizeof(req->r_base_oid.name),
4295f221 770 "%llx.%08llx", vino.ino, objnum);
3c972c95 771 req->r_base_oid.name_len = strlen(req->r_base_oid.name);
dbe0fc41 772
f24e9980
SW		 773 return req;
774}
3d14c5d2 775EXPORT_SYMBOL(ceph_osdc_new_request);
f24e9980
SW		 776
777/*
778 * We keep osd requests in an rbtree, sorted by ->r_tid.
779 */
780static void __insert_request(struct ceph_osd_client *osdc,
781 struct ceph_osd_request *new)
782{
783 struct rb_node **p = &osdc->requests.rb_node;
784 struct rb_node *parent = NULL;
785 struct ceph_osd_request *req = NULL;
786
787 while (*p) {
788 parent = *p;
789 req = rb_entry(parent, struct ceph_osd_request, r_node);
790 if (new->r_tid < req->r_tid)
791 p = &(*p)->rb_left;
792 else if (new->r_tid > req->r_tid)
793 p = &(*p)->rb_right;
794 else
795 BUG();
796 }
797
798 rb_link_node(&new->r_node, parent, p);
799 rb_insert_color(&new->r_node, &osdc->requests);
800}
801
802static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc,
803 u64 tid)
804{
805 struct ceph_osd_request *req;
806 struct rb_node *n = osdc->requests.rb_node;
807
808 while (n) {
809 req = rb_entry(n, struct ceph_osd_request, r_node);
810 if (tid < req->r_tid)
811 n = n->rb_left;
812 else if (tid > req->r_tid)
813 n = n->rb_right;
814 else
815 return req;
816 }
817 return NULL;
818}
819
820static struct ceph_osd_request *
821__lookup_request_ge(struct ceph_osd_client *osdc,
822 u64 tid)
823{
824 struct ceph_osd_request *req;
825 struct rb_node *n = osdc->requests.rb_node;
826
827 while (n) {
828 req = rb_entry(n, struct ceph_osd_request, r_node);
829 if (tid < req->r_tid) {
830 if (!n->rb_left)
831 return req;
832 n = n->rb_left;
833 } else if (tid > req->r_tid) {
834 n = n->rb_right;
835 } else {
836 return req;
837 }
838 }
839 return NULL;
840}
841
6f6c7006
SW		 842/*
843 * Resubmit requests pending on the given osd.
844 */
845static void __kick_osd_requests(struct ceph_osd_client *osdc,
846 struct ceph_osd *osd)
847{
a40c4f10 848 struct ceph_osd_request *req, *nreq;
e02493c0 849 LIST_HEAD(resend);
6f6c7006
SW		 850 int err;
851
852 dout("__kick_osd_requests osd%d\n", osd->o_osd);
853 err = __reset_osd(osdc, osd);
685a7555 854 if (err)
6f6c7006 855 return;
e02493c0
AE		 856 /*
857 * Build up a list of requests to resend by traversing the
858 * osd's list of requests. Requests for a given object are
859 * sent in tid order, and that is also the order they're
860 * kept on this list. Therefore all requests that are in
861 * flight will be found first, followed by all requests that
862 * have not yet been sent. And to resend requests while
863 * preserving this order we will want to put any sent
864 * requests back on the front of the osd client's unsent
865 * list.
866 *
867 * So we build a separate ordered list of already-sent
868 * requests for the affected osd and splice it onto the
869 * front of the osd client's unsent list. Once we've seen a
870 * request that has not yet been sent we're done. Those
871 * requests are already sitting right where they belong.
872 */
6f6c7006 873 list_for_each_entry(req, &osd->o_requests, r_osd_item) {
e02493c0
AE		 874 if (!req->r_sent)
875 break;
876 list_move_tail(&req->r_req_lru_item, &resend);
877 dout("requeueing %p tid %llu osd%d\n", req, req->r_tid,
6f6c7006 878 osd->o_osd);
a40c4f10
YS		 879 if (!req->r_linger)
880 req->r_flags |= CEPH_OSD_FLAG_RETRY;
881 }
e02493c0 882 list_splice(&resend, &osdc->req_unsent);
a40c4f10 883
e02493c0
AE		 884 /*
885 * Linger requests are re-registered before sending, which
886 * sets up a new tid for each. We add them to the unsent
887 * list at the end to keep things in tid order.
888 */
a40c4f10
YS		 889 list_for_each_entry_safe(req, nreq, &osd->o_linger_requests,
890 r_linger_osd) {
77f38e0e
SW		 891 /*
892 * reregister request prior to unregistering linger so
893 * that r_osd is preserved.
894 */
895 BUG_ON(!list_empty(&req->r_req_lru_item));
a40c4f10 896 __register_request(osdc, req);
e02493c0 897 list_add_tail(&req->r_req_lru_item, &osdc->req_unsent);
ad885927 898 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
77f38e0e 899 __unregister_linger_request(osdc, req);
a40c4f10
YS		 900 dout("requeued lingering %p tid %llu osd%d\n", req, req->r_tid,
901 osd->o_osd);
6f6c7006
SW		 902 }
903}
904
f24e9980 905/*
81b024e7 906 * If the osd connection drops, we need to resubmit all requests.
f24e9980
SW		 907 */
908static void osd_reset(struct ceph_connection *con)
909{
910 struct ceph_osd *osd = con->private;
911 struct ceph_osd_client *osdc;
912
913 if (!osd)
914 return;
915 dout("osd_reset osd%d\n", osd->o_osd);
916 osdc = osd->o_osdc;
f24e9980 917 down_read(&osdc->map_sem);
83aff95e
SW		 918 mutex_lock(&osdc->request_mutex);
919 __kick_osd_requests(osdc, osd);
f9d25199 920 __send_queued(osdc);
83aff95e 921 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 922 up_read(&osdc->map_sem);
923}
924
925/*
926 * Track open sessions with osds.
927 */
e10006f8 928static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
f24e9980
SW		 929{
930 struct ceph_osd *osd;
931
932 osd = kzalloc(sizeof(*osd), GFP_NOFS);
933 if (!osd)
934 return NULL;
935
936 atomic_set(&osd->o_ref, 1);
937 osd->o_osdc = osdc;
e10006f8 938 osd->o_osd = onum;
f407731d 939 RB_CLEAR_NODE(&osd->o_node);
f24e9980 940 INIT_LIST_HEAD(&osd->o_requests);
a40c4f10 941 INIT_LIST_HEAD(&osd->o_linger_requests);
f5a2041b 942 INIT_LIST_HEAD(&osd->o_osd_lru);
f24e9980
SW		 943 osd->o_incarnation = 1;
944
b7a9e5dd 945 ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
4e7a5dcd 946
422d2cb8 947 INIT_LIST_HEAD(&osd->o_keepalive_item);
f24e9980
SW		 948 return osd;
949}
950
951static struct ceph_osd *get_osd(struct ceph_osd *osd)
952{
953 if (atomic_inc_not_zero(&osd->o_ref)) {
954 dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
955 atomic_read(&osd->o_ref));
956 return osd;
957 } else {
958 dout("get_osd %p FAIL\n", osd);
959 return NULL;
960 }
961}
962
963static void put_osd(struct ceph_osd *osd)
964{
965 dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
966 atomic_read(&osd->o_ref) - 1);
a255651d 967 if (atomic_dec_and_test(&osd->o_ref) && osd->o_auth.authorizer) {
79494d1b
SW		 968 struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth;
969
27859f97 970 ceph_auth_destroy_authorizer(ac, osd->o_auth.authorizer);
f24e9980 971 kfree(osd);
79494d1b 972 }
f24e9980
SW		 973}
974
975/*
976 * remove an osd from our map
977 */
f5a2041b 978static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
f24e9980 979{
f5a2041b 980 dout("__remove_osd %p\n", osd);
f24e9980
SW		 981 BUG_ON(!list_empty(&osd->o_requests));
982 rb_erase(&osd->o_node, &osdc->osds);
f5a2041b 983 list_del_init(&osd->o_osd_lru);
f24e9980
SW		 984 ceph_con_close(&osd->o_con);
985 put_osd(osd);
986}
987
aca420bc
SW		 988static void remove_all_osds(struct ceph_osd_client *osdc)
989{
048a9d2d 990 dout("%s %p\n", __func__, osdc);
aca420bc
SW		 991 mutex_lock(&osdc->request_mutex);
992 while (!RB_EMPTY_ROOT(&osdc->osds)) {
993 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
994 struct ceph_osd, o_node);
995 __remove_osd(osdc, osd);
996 }
997 mutex_unlock(&osdc->request_mutex);
998}
999
f5a2041b
YS		 1000static void __move_osd_to_lru(struct ceph_osd_client *osdc,
1001 struct ceph_osd *osd)
1002{
1003 dout("__move_osd_to_lru %p\n", osd);
1004 BUG_ON(!list_empty(&osd->o_osd_lru));
1005 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
3d14c5d2 1006 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl * HZ;
f5a2041b
YS		 1007}
1008
1009static void __remove_osd_from_lru(struct ceph_osd *osd)
1010{
1011 dout("__remove_osd_from_lru %p\n", osd);
1012 if (!list_empty(&osd->o_osd_lru))
1013 list_del_init(&osd->o_osd_lru);
1014}
1015
aca420bc 1016static void remove_old_osds(struct ceph_osd_client *osdc)
f5a2041b
YS		 1017{
1018 struct ceph_osd *osd, *nosd;
1019
1020 dout("__remove_old_osds %p\n", osdc);
1021 mutex_lock(&osdc->request_mutex);
1022 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
aca420bc 1023 if (time_before(jiffies, osd->lru_ttl))
f5a2041b
YS		 1024 break;
1025 __remove_osd(osdc, osd);
1026 }
1027 mutex_unlock(&osdc->request_mutex);
1028}
1029
f24e9980
SW		 1030/*
1031 * reset osd connect
1032 */
f5a2041b 1033static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
f24e9980 1034{
c3acb181 1035 struct ceph_entity_addr *peer_addr;
f24e9980 1036
f5a2041b 1037 dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
a40c4f10
YS		 1038 if (list_empty(&osd->o_requests) &&
1039 list_empty(&osd->o_linger_requests)) {
f5a2041b 1040 __remove_osd(osdc, osd);
c3acb181
AE		 1041
1042 return -ENODEV;
1043 }
1044
1045 peer_addr = &osdc->osdmap->osd_addr[osd->o_osd];
1046 if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1047 !ceph_con_opened(&osd->o_con)) {
1048 struct ceph_osd_request *req;
1049
0b4af2e8
ID		 1050 dout("osd addr hasn't changed and connection never opened, "
1051 "letting msgr retry\n");
87b315a5
SW		 1052 /* touch each r_stamp for handle_timeout()'s benfit */
1053 list_for_each_entry(req, &osd->o_requests, r_osd_item)
1054 req->r_stamp = jiffies;
c3acb181
AE		 1055
1056 return -EAGAIN;
f24e9980 1057 }
c3acb181
AE		 1058
1059 ceph_con_close(&osd->o_con);
1060 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1061 osd->o_incarnation++;
1062
1063 return 0;
f24e9980
SW		 1064}
1065
1066static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new)
1067{
1068 struct rb_node **p = &osdc->osds.rb_node;
1069 struct rb_node *parent = NULL;
1070 struct ceph_osd *osd = NULL;
1071
aca420bc 1072 dout("__insert_osd %p osd%d\n", new, new->o_osd);
f24e9980
SW		 1073 while (*p) {
1074 parent = *p;
1075 osd = rb_entry(parent, struct ceph_osd, o_node);
1076 if (new->o_osd < osd->o_osd)
1077 p = &(*p)->rb_left;
1078 else if (new->o_osd > osd->o_osd)
1079 p = &(*p)->rb_right;
1080 else
1081 BUG();
1082 }
1083
1084 rb_link_node(&new->o_node, parent, p);
1085 rb_insert_color(&new->o_node, &osdc->osds);
1086}
1087
1088static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o)
1089{
1090 struct ceph_osd *osd;
1091 struct rb_node *n = osdc->osds.rb_node;
1092
1093 while (n) {
1094 osd = rb_entry(n, struct ceph_osd, o_node);
1095 if (o < osd->o_osd)
1096 n = n->rb_left;
1097 else if (o > osd->o_osd)
1098 n = n->rb_right;
1099 else
1100 return osd;
1101 }
1102 return NULL;
1103}
1104
422d2cb8
YS		 1105static void __schedule_osd_timeout(struct ceph_osd_client *osdc)
1106{
1107 schedule_delayed_work(&osdc->timeout_work,
3d14c5d2 1108 osdc->client->options->osd_keepalive_timeout * HZ);
422d2cb8
YS		 1109}
1110
1111static void __cancel_osd_timeout(struct ceph_osd_client *osdc)
1112{
1113 cancel_delayed_work(&osdc->timeout_work);
1114}
f24e9980
SW		 1115
1116/*
1117 * Register request, assign tid. If this is the first request, set up
1118 * the timeout event.
1119 */
a40c4f10
YS		 1120static void __register_request(struct ceph_osd_client *osdc,
1121 struct ceph_osd_request *req)
f24e9980 1122{
f24e9980 1123 req->r_tid = ++osdc->last_tid;
6df058c0 1124 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
77f38e0e 1125 dout("__register_request %p tid %lld\n", req, req->r_tid);
f24e9980
SW		 1126 __insert_request(osdc, req);
1127 ceph_osdc_get_request(req);
1128 osdc->num_requests++;
f24e9980 1129 if (osdc->num_requests == 1) {
422d2cb8
YS		 1130 dout(" first request, scheduling timeout\n");
1131 __schedule_osd_timeout(osdc);
f24e9980 1132 }
a40c4f10
YS		 1133}
1134
f24e9980
SW		 1135/*
1136 * called under osdc->request_mutex
1137 */
1138static void __unregister_request(struct ceph_osd_client *osdc,
1139 struct ceph_osd_request *req)
1140{
35f9f8a0
SW		 1141 if (RB_EMPTY_NODE(&req->r_node)) {
1142 dout("__unregister_request %p tid %lld not registered\n",
1143 req, req->r_tid);
1144 return;
1145 }
1146
f24e9980
SW		 1147 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
1148 rb_erase(&req->r_node, &osdc->requests);
1149 osdc->num_requests--;
1150
0ba6478d
SW		 1151 if (req->r_osd) {
1152 /* make sure the original request isn't in flight. */
6740a845 1153 ceph_msg_revoke(req->r_request);
0ba6478d
SW		 1154
1155 list_del_init(&req->r_osd_item);
a40c4f10
YS		 1156 if (list_empty(&req->r_osd->o_requests) &&
1157 list_empty(&req->r_osd->o_linger_requests)) {
1158 dout("moving osd to %p lru\n", req->r_osd);
f5a2041b 1159 __move_osd_to_lru(osdc, req->r_osd);
a40c4f10 1160 }
fbdb9190 1161 if (list_empty(&req->r_linger_item))
a40c4f10 1162 req->r_osd = NULL;
0ba6478d 1163 }
f24e9980 1164
7d5f2481 1165 list_del_init(&req->r_req_lru_item);
f24e9980
SW		 1166 ceph_osdc_put_request(req);
1167
422d2cb8
YS		 1168 if (osdc->num_requests == 0) {
1169 dout(" no requests, canceling timeout\n");
1170 __cancel_osd_timeout(osdc);
f24e9980
SW		 1171 }
1172}
1173
1174/*
1175 * Cancel a previously queued request message
1176 */
1177static void __cancel_request(struct ceph_osd_request *req)
1178{
6bc18876 1179 if (req->r_sent && req->r_osd) {
6740a845 1180 ceph_msg_revoke(req->r_request);
f24e9980
SW		 1181 req->r_sent = 0;
1182 }
1183}
1184
a40c4f10
YS		 1185static void __register_linger_request(struct ceph_osd_client *osdc,
1186 struct ceph_osd_request *req)
1187{
1188 dout("__register_linger_request %p\n", req);
96e4dac6 1189 ceph_osdc_get_request(req);
a40c4f10 1190 list_add_tail(&req->r_linger_item, &osdc->req_linger);
6194ea89
SW		 1191 if (req->r_osd)
1192 list_add_tail(&req->r_linger_osd,
1193 &req->r_osd->o_linger_requests);
a40c4f10
YS		 1194}
1195
1196static void __unregister_linger_request(struct ceph_osd_client *osdc,
1197 struct ceph_osd_request *req)
1198{
1199 dout("__unregister_linger_request %p\n", req);
61c74035 1200 list_del_init(&req->r_linger_item);
a40c4f10 1201 if (req->r_osd) {
a40c4f10
YS		 1202 list_del_init(&req->r_linger_osd);
1203
1204 if (list_empty(&req->r_osd->o_requests) &&
1205 list_empty(&req->r_osd->o_linger_requests)) {
1206 dout("moving osd to %p lru\n", req->r_osd);
1207 __move_osd_to_lru(osdc, req->r_osd);
1208 }
fbdb9190
SW		 1209 if (list_empty(&req->r_osd_item))
1210 req->r_osd = NULL;
a40c4f10 1211 }
96e4dac6 1212 ceph_osdc_put_request(req);
a40c4f10
YS		 1213}
1214
1215void ceph_osdc_unregister_linger_request(struct ceph_osd_client *osdc,
1216 struct ceph_osd_request *req)
1217{
1218 mutex_lock(&osdc->request_mutex);
1219 if (req->r_linger) {
c10ebbf5 1220 req->r_linger = 0;
96e4dac6 1221 __unregister_linger_request(osdc, req);
a40c4f10
YS		 1222 }
1223 mutex_unlock(&osdc->request_mutex);
1224}
1225EXPORT_SYMBOL(ceph_osdc_unregister_linger_request);
1226
1227void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
1228 struct ceph_osd_request *req)
1229{
1230 if (!req->r_linger) {
1231 dout("set_request_linger %p\n", req);
1232 req->r_linger = 1;
a40c4f10
YS		 1233 }
1234}
1235EXPORT_SYMBOL(ceph_osdc_set_request_linger);
1236
d29adb34
JD		 1237/*
1238 * Returns whether a request should be blocked from being sent
1239 * based on the current osdmap and osd_client settings.
1240 *
1241 * Caller should hold map_sem for read.
1242 */
1243static bool __req_should_be_paused(struct ceph_osd_client *osdc,
1244 struct ceph_osd_request *req)
1245{
1246 bool pauserd = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD);
1247 bool pausewr = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR) ||
1248 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
1249 return (req->r_flags & CEPH_OSD_FLAG_READ && pauserd) ||
1250 (req->r_flags & CEPH_OSD_FLAG_WRITE && pausewr);
1251}
1252
17a13e40
ID		 1253/*
1254 * Calculate mapping of a request to a PG. Takes tiering into account.
1255 */
1256static int __calc_request_pg(struct ceph_osdmap *osdmap,
1257 struct ceph_osd_request *req,
1258 struct ceph_pg *pg_out)
1259{
205ee118
ID		 1260 bool need_check_tiering;
1261
1262 need_check_tiering = false;
1263 if (req->r_target_oloc.pool == -1) {
1264 req->r_target_oloc = req->r_base_oloc; /* struct */
1265 need_check_tiering = true;
1266 }
1267 if (req->r_target_oid.name_len == 0) {
1268 ceph_oid_copy(&req->r_target_oid, &req->r_base_oid);
1269 need_check_tiering = true;
1270 }
1271
1272 if (need_check_tiering &&
1273 (req->r_flags & CEPH_OSD_FLAG_IGNORE_OVERLAY) == 0) {
17a13e40
ID		 1274 struct ceph_pg_pool_info *pi;
1275
205ee118 1276 pi = ceph_pg_pool_by_id(osdmap, req->r_target_oloc.pool);
17a13e40
ID		 1277 if (pi) {
1278 if ((req->r_flags & CEPH_OSD_FLAG_READ) &&
1279 pi->read_tier >= 0)
205ee118 1280 req->r_target_oloc.pool = pi->read_tier;
17a13e40
ID		 1281 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
1282 pi->write_tier >= 0)
205ee118 1283 req->r_target_oloc.pool = pi->write_tier;
17a13e40
ID		 1284 }
1285 /* !pi is caught in ceph_oloc_oid_to_pg() */
1286 }
1287
205ee118
ID		 1288 return ceph_oloc_oid_to_pg(osdmap, &req->r_target_oloc,
1289 &req->r_target_oid, pg_out);
17a13e40
ID		 1290}
1291
f24e9980
SW		 1292/*
1293 * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
1294 * (as needed), and set the request r_osd appropriately. If there is
25985edc 1295 * no up osd, set r_osd to NULL. Move the request to the appropriate list
6f6c7006 1296 * (unsent, homeless) or leave on in-flight lru.
f24e9980
SW		 1297 *
1298 * Return 0 if unchanged, 1 if changed, or negative on error.
1299 *
1300 * Caller should hold map_sem for read and request_mutex.
1301 */
6f6c7006 1302static int __map_request(struct ceph_osd_client *osdc,
38d6453c 1303 struct ceph_osd_request *req, int force_resend)
f24e9980 1304{
5b191d99 1305 struct ceph_pg pgid;
d85b7056
SW		 1306 int acting[CEPH_PG_MAX_SIZE];
1307 int o = -1, num = 0;
f24e9980 1308 int err;
d29adb34 1309 bool was_paused;
f24e9980 1310
6f6c7006 1311 dout("map_request %p tid %lld\n", req, req->r_tid);
17a13e40
ID		 1312
1313 err = __calc_request_pg(osdc->osdmap, req, &pgid);
6f6c7006
SW		 1314 if (err) {
1315 list_move(&req->r_req_lru_item, &osdc->req_notarget);
f24e9980 1316 return err;
6f6c7006 1317 }
7740a42f
SW		 1318 req->r_pgid = pgid;
1319
d85b7056
SW		 1320 err = ceph_calc_pg_acting(osdc->osdmap, pgid, acting);
1321 if (err > 0) {
1322 o = acting[0];
1323 num = err;
1324 }
f24e9980 1325
d29adb34
JD		 1326 was_paused = req->r_paused;
1327 req->r_paused = __req_should_be_paused(osdc, req);
1328 if (was_paused && !req->r_paused)
1329 force_resend = 1;
1330
38d6453c
SW		 1331 if ((!force_resend &&
1332 req->r_osd && req->r_osd->o_osd == o &&
d85b7056
SW		 1333 req->r_sent >= req->r_osd->o_incarnation &&
1334 req->r_num_pg_osds == num &&
1335 memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
d29adb34
JD		 1336 (req->r_osd == NULL && o == -1) ||
1337 req->r_paused)
f24e9980
SW		 1338 return 0; /* no change */
1339
5b191d99
SW		 1340 dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n",
1341 req->r_tid, pgid.pool, pgid.seed, o,
f24e9980
SW		 1342 req->r_osd ? req->r_osd->o_osd : -1);
1343
d85b7056
SW		 1344 /* record full pg acting set */
1345 memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
1346 req->r_num_pg_osds = num;
1347
f24e9980
SW		 1348 if (req->r_osd) {
1349 __cancel_request(req);
1350 list_del_init(&req->r_osd_item);
f24e9980
SW		 1351 req->r_osd = NULL;
1352 }
1353
1354 req->r_osd = __lookup_osd(osdc, o);
1355 if (!req->r_osd && o >= 0) {
c99eb1c7 1356 err = -ENOMEM;
e10006f8 1357 req->r_osd = create_osd(osdc, o);
6f6c7006
SW		 1358 if (!req->r_osd) {
1359 list_move(&req->r_req_lru_item, &osdc->req_notarget);
c99eb1c7 1360 goto out;
6f6c7006 1361 }
f24e9980 1362
6f6c7006 1363 dout("map_request osd %p is osd%d\n", req->r_osd, o);
f24e9980
SW		 1364 __insert_osd(osdc, req->r_osd);
1365
b7a9e5dd
SW		 1366 ceph_con_open(&req->r_osd->o_con,
1367 CEPH_ENTITY_TYPE_OSD, o,
1368 &osdc->osdmap->osd_addr[o]);
f24e9980
SW		 1369 }
1370
f5a2041b
YS		 1371 if (req->r_osd) {
1372 __remove_osd_from_lru(req->r_osd);
ad885927
AE		 1373 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
1374 list_move_tail(&req->r_req_lru_item, &osdc->req_unsent);
6f6c7006 1375 } else {
ad885927 1376 list_move_tail(&req->r_req_lru_item, &osdc->req_notarget);
f5a2041b 1377 }
d85b7056 1378 err = 1; /* osd or pg changed */
f24e9980
SW		 1379
1380out:
f24e9980
SW		 1381 return err;
1382}
1383
1384/*
1385 * caller should hold map_sem (for read) and request_mutex
1386 */
56e925b6
SW		 1387static void __send_request(struct ceph_osd_client *osdc,
1388 struct ceph_osd_request *req)
f24e9980 1389{
1b83bef2 1390 void *p;
f24e9980 1391
1b83bef2
SW		 1392 dout("send_request %p tid %llu to osd%d flags %d pg %lld.%x\n",
1393 req, req->r_tid, req->r_osd->o_osd, req->r_flags,
1394 (unsigned long long)req->r_pgid.pool, req->r_pgid.seed);
1395
1396 /* fill in message content that changes each time we send it */
1397 put_unaligned_le32(osdc->osdmap->epoch, req->r_request_osdmap_epoch);
1398 put_unaligned_le32(req->r_flags, req->r_request_flags);
205ee118 1399 put_unaligned_le64(req->r_target_oloc.pool, req->r_request_pool);
1b83bef2
SW		 1400 p = req->r_request_pgid;
1401 ceph_encode_64(&p, req->r_pgid.pool);
1402 ceph_encode_32(&p, req->r_pgid.seed);
1403 put_unaligned_le64(1, req->r_request_attempts); /* FIXME */
1404 memcpy(req->r_request_reassert_version, &req->r_reassert_version,
1405 sizeof(req->r_reassert_version));
2169aea6 1406
3dd72fc0 1407 req->r_stamp = jiffies;
07a27e22 1408 list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
f24e9980
SW		 1409
1410 ceph_msg_get(req->r_request); /* send consumes a ref */
26be8808 1411
f24e9980 1412 req->r_sent = req->r_osd->o_incarnation;
26be8808
AE		 1413
1414 ceph_con_send(&req->r_osd->o_con, req->r_request);
f24e9980
SW		 1415}
1416
6f6c7006
SW		 1417/*
1418 * Send any requests in the queue (req_unsent).
1419 */
f9d25199 1420static void __send_queued(struct ceph_osd_client *osdc)
6f6c7006
SW		 1421{
1422 struct ceph_osd_request *req, *tmp;
1423
f9d25199
AE		 1424 dout("__send_queued\n");
1425 list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item)
6f6c7006 1426 __send_request(osdc, req);
6f6c7006
SW		 1427}
1428
0bbfdfe8
ID		 1429/*
1430 * Caller should hold map_sem for read and request_mutex.
1431 */
1432static int __ceph_osdc_start_request(struct ceph_osd_client *osdc,
1433 struct ceph_osd_request *req,
1434 bool nofail)
1435{
1436 int rc;
1437
1438 __register_request(osdc, req);
1439 req->r_sent = 0;
1440 req->r_got_reply = 0;
1441 rc = __map_request(osdc, req, 0);
1442 if (rc < 0) {
1443 if (nofail) {
1444 dout("osdc_start_request failed map, "
1445 " will retry %lld\n", req->r_tid);
1446 rc = 0;
1447 } else {
1448 __unregister_request(osdc, req);
1449 }
1450 return rc;
1451 }
1452
1453 if (req->r_osd == NULL) {
1454 dout("send_request %p no up osds in pg\n", req);
1455 ceph_monc_request_next_osdmap(&osdc->client->monc);
1456 } else {
1457 __send_queued(osdc);
1458 }
1459
1460 return 0;
1461}
1462
f24e9980
SW		 1463/*
1464 * Timeout callback, called every N seconds when 1 or more osd
1465 * requests has been active for more than N seconds. When this
1466 * happens, we ping all OSDs with requests who have timed out to
1467 * ensure any communications channel reset is detected. Reset the
1468 * request timeouts another N seconds in the future as we go.
1469 * Reschedule the timeout event another N seconds in future (unless
1470 * there are no open requests).
1471 */
1472static void handle_timeout(struct work_struct *work)
1473{
1474 struct ceph_osd_client *osdc =
1475 container_of(work, struct ceph_osd_client, timeout_work.work);
83aff95e 1476 struct ceph_osd_request *req;
f24e9980 1477 struct ceph_osd *osd;
422d2cb8 1478 unsigned long keepalive =
3d14c5d2 1479 osdc->client->options->osd_keepalive_timeout * HZ;
422d2cb8 1480 struct list_head slow_osds;
f24e9980
SW		 1481 dout("timeout\n");
1482 down_read(&osdc->map_sem);
1483
1484 ceph_monc_request_next_osdmap(&osdc->client->monc);
1485
1486 mutex_lock(&osdc->request_mutex);
f24e9980 1487
422d2cb8
YS		 1488 /*
1489 * ping osds that are a bit slow. this ensures that if there
1490 * is a break in the TCP connection we will notice, and reopen
1491 * a connection with that osd (from the fault callback).
1492 */
1493 INIT_LIST_HEAD(&slow_osds);
1494 list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
3dd72fc0 1495 if (time_before(jiffies, req->r_stamp + keepalive))
422d2cb8
YS		 1496 break;
1497
1498 osd = req->r_osd;
1499 BUG_ON(!osd);
1500 dout(" tid %llu is slow, will send keepalive on osd%d\n",
f24e9980 1501 req->r_tid, osd->o_osd);
422d2cb8
YS		 1502 list_move_tail(&osd->o_keepalive_item, &slow_osds);
1503 }
1504 while (!list_empty(&slow_osds)) {
1505 osd = list_entry(slow_osds.next, struct ceph_osd,
1506 o_keepalive_item);
1507 list_del_init(&osd->o_keepalive_item);
f24e9980
SW		 1508 ceph_con_keepalive(&osd->o_con);
1509 }
1510
422d2cb8 1511 __schedule_osd_timeout(osdc);
f9d25199 1512 __send_queued(osdc);
f24e9980 1513 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 1514 up_read(&osdc->map_sem);
1515}
1516
f5a2041b
YS		 1517static void handle_osds_timeout(struct work_struct *work)
1518{
1519 struct ceph_osd_client *osdc =
1520 container_of(work, struct ceph_osd_client,
1521 osds_timeout_work.work);
1522 unsigned long delay =
3d14c5d2 1523 osdc->client->options->osd_idle_ttl * HZ >> 2;
f5a2041b
YS		 1524
1525 dout("osds timeout\n");
1526 down_read(&osdc->map_sem);
aca420bc 1527 remove_old_osds(osdc);
f5a2041b
YS		 1528 up_read(&osdc->map_sem);
1529
1530 schedule_delayed_work(&osdc->osds_timeout_work,
1531 round_jiffies_relative(delay));
1532}
1533
205ee118
ID		 1534static int ceph_oloc_decode(void **p, void *end,
1535 struct ceph_object_locator *oloc)
1536{
1537 u8 struct_v, struct_cv;
1538 u32 len;
1539 void *struct_end;
1540 int ret = 0;
1541
1542 ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
1543 struct_v = ceph_decode_8(p);
1544 struct_cv = ceph_decode_8(p);
1545 if (struct_v < 3) {
1546 pr_warn("got v %d < 3 cv %d of ceph_object_locator\n",
1547 struct_v, struct_cv);
1548 goto e_inval;
1549 }
1550 if (struct_cv > 6) {
1551 pr_warn("got v %d cv %d > 6 of ceph_object_locator\n",
1552 struct_v, struct_cv);
1553 goto e_inval;
1554 }
1555 len = ceph_decode_32(p);
1556 ceph_decode_need(p, end, len, e_inval);
1557 struct_end = *p + len;
1558
1559 oloc->pool = ceph_decode_64(p);
1560 *p += 4; /* skip preferred */
1561
1562 len = ceph_decode_32(p);
1563 if (len > 0) {
1564 pr_warn("ceph_object_locator::key is set\n");
1565 goto e_inval;
1566 }
1567
1568 if (struct_v >= 5) {
1569 len = ceph_decode_32(p);
1570 if (len > 0) {
1571 pr_warn("ceph_object_locator::nspace is set\n");
1572 goto e_inval;
1573 }
1574 }
1575
1576 if (struct_v >= 6) {
1577 s64 hash = ceph_decode_64(p);
1578 if (hash != -1) {
1579 pr_warn("ceph_object_locator::hash is set\n");
1580 goto e_inval;
1581 }
1582 }
1583
1584 /* skip the rest */
1585 *p = struct_end;
1586out:
1587 return ret;
1588
1589e_inval:
1590 ret = -EINVAL;
1591 goto out;
1592}
1593
1594static int ceph_redirect_decode(void **p, void *end,
1595 struct ceph_request_redirect *redir)
1596{
1597 u8 struct_v, struct_cv;
1598 u32 len;
1599 void *struct_end;
1600 int ret;
1601
1602 ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
1603 struct_v = ceph_decode_8(p);
1604 struct_cv = ceph_decode_8(p);
1605 if (struct_cv > 1) {
1606 pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n",
1607 struct_v, struct_cv);
1608 goto e_inval;
1609 }
1610 len = ceph_decode_32(p);
1611 ceph_decode_need(p, end, len, e_inval);
1612 struct_end = *p + len;
1613
1614 ret = ceph_oloc_decode(p, end, &redir->oloc);
1615 if (ret)
1616 goto out;
1617
1618 len = ceph_decode_32(p);
1619 if (len > 0) {
1620 pr_warn("ceph_request_redirect::object_name is set\n");
1621 goto e_inval;
1622 }
1623
1624 len = ceph_decode_32(p);
1625 *p += len; /* skip osd_instructions */
1626
1627 /* skip the rest */
1628 *p = struct_end;
1629out:
1630 return ret;
1631
1632e_inval:
1633 ret = -EINVAL;
1634 goto out;
1635}
1636
25845472
SW		 1637static void complete_request(struct ceph_osd_request *req)
1638{
25845472
SW		 1639 complete_all(&req->r_safe_completion); /* fsync waiter */
1640}
1641
f24e9980
SW		 1642/*
1643 * handle osd op reply. either call the callback if it is specified,
1644 * or do the completion to wake up the waiting thread.
1645 */
350b1c32
SW		 1646static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg,
1647 struct ceph_connection *con)
f24e9980 1648{
1b83bef2 1649 void *p, *end;
f24e9980 1650 struct ceph_osd_request *req;
205ee118 1651 struct ceph_request_redirect redir;
f24e9980 1652 u64 tid;
1b83bef2 1653 int object_len;
79528734
AE		 1654 unsigned int numops;
1655 int payload_len, flags;
0ceed5db 1656 s32 result;
1b83bef2
SW		 1657 s32 retry_attempt;
1658 struct ceph_pg pg;
1659 int err;
1660 u32 reassert_epoch;
1661 u64 reassert_version;
1662 u32 osdmap_epoch;
0d5af164 1663 int already_completed;
9fc6e064 1664 u32 bytes;
79528734 1665 unsigned int i;
f24e9980 1666
6df058c0 1667 tid = le64_to_cpu(msg->hdr.tid);
1b83bef2
SW		 1668 dout("handle_reply %p tid %llu\n", msg, tid);
1669
1670 p = msg->front.iov_base;
1671 end = p + msg->front.iov_len;
1672
1673 ceph_decode_need(&p, end, 4, bad);
1674 object_len = ceph_decode_32(&p);
1675 ceph_decode_need(&p, end, object_len, bad);
1676 p += object_len;
1677
ef4859d6 1678 err = ceph_decode_pgid(&p, end, &pg);
1b83bef2 1679 if (err)
f24e9980 1680 goto bad;
1b83bef2
SW		 1681
1682 ceph_decode_need(&p, end, 8 + 4 + 4 + 8 + 4, bad);
1683 flags = ceph_decode_64(&p);
1684 result = ceph_decode_32(&p);
1685 reassert_epoch = ceph_decode_32(&p);
1686 reassert_version = ceph_decode_64(&p);
1687 osdmap_epoch = ceph_decode_32(&p);
1688
f24e9980
SW		 1689 /* lookup */
1690 mutex_lock(&osdc->request_mutex);
1691 req = __lookup_request(osdc, tid);
1692 if (req == NULL) {
1693 dout("handle_reply tid %llu dne\n", tid);
8058fd45 1694 goto bad_mutex;
f24e9980
SW		 1695 }
1696 ceph_osdc_get_request(req);
1b83bef2
SW		 1697
1698 dout("handle_reply %p tid %llu req %p result %d\n", msg, tid,
1699 req, result);
1700
18741196 1701 ceph_decode_need(&p, end, 4, bad_put);
1b83bef2
SW		 1702 numops = ceph_decode_32(&p);
1703 if (numops > CEPH_OSD_MAX_OP)
1704 goto bad_put;
1705 if (numops != req->r_num_ops)
1706 goto bad_put;
1707 payload_len = 0;
18741196 1708 ceph_decode_need(&p, end, numops * sizeof(struct ceph_osd_op), bad_put);
1b83bef2
SW		 1709 for (i = 0; i < numops; i++) {
1710 struct ceph_osd_op *op = p;
1711 int len;
1712
1713 len = le32_to_cpu(op->payload_len);
1714 req->r_reply_op_len[i] = len;
1715 dout(" op %d has %d bytes\n", i, len);
1716 payload_len += len;
1717 p += sizeof(*op);
1718 }
9fc6e064
AE		 1719 bytes = le32_to_cpu(msg->hdr.data_len);
1720 if (payload_len != bytes) {
1b83bef2 1721 pr_warning("sum of op payload lens %d != data_len %d",
9fc6e064 1722 payload_len, bytes);
1b83bef2
SW		 1723 goto bad_put;
1724 }
1725
18741196 1726 ceph_decode_need(&p, end, 4 + numops * 4, bad_put);
1b83bef2
SW		 1727 retry_attempt = ceph_decode_32(&p);
1728 for (i = 0; i < numops; i++)
1729 req->r_reply_op_result[i] = ceph_decode_32(&p);
f24e9980 1730
205ee118
ID		 1731 if (le16_to_cpu(msg->hdr.version) >= 6) {
1732 p += 8 + 4; /* skip replay_version */
1733 p += 8; /* skip user_version */
eb845ff1 1734
205ee118
ID		 1735 err = ceph_redirect_decode(&p, end, &redir);
1736 if (err)
1737 goto bad_put;
1738 } else {
1739 redir.oloc.pool = -1;
1740 }
f24e9980 1741
205ee118
ID		 1742 if (redir.oloc.pool != -1) {
1743 dout("redirect pool %lld\n", redir.oloc.pool);
1744
1745 __unregister_request(osdc, req);
1746 mutex_unlock(&osdc->request_mutex);
1747
1748 req->r_target_oloc = redir.oloc; /* struct */
1749
1750 /*
1751 * Start redirect requests with nofail=true. If
1752 * mapping fails, request will end up on the notarget
1753 * list, waiting for the new osdmap (which can take
1754 * a while), even though the original request mapped
1755 * successfully. In the future we might want to follow
1756 * original request's nofail setting here.
1757 */
1758 err = ceph_osdc_start_request(osdc, req, true);
1759 BUG_ON(err);
1760
1761 goto done;
1762 }
1763
1764 already_completed = req->r_got_reply;
1765 if (!req->r_got_reply) {
1b83bef2 1766 req->r_result = result;
f24e9980
SW		 1767 dout("handle_reply result %d bytes %d\n", req->r_result,
1768 bytes);
1769 if (req->r_result == 0)
1770 req->r_result = bytes;
1771
1772 /* in case this is a write and we need to replay, */
1b83bef2
SW		 1773 req->r_reassert_version.epoch = cpu_to_le32(reassert_epoch);
1774 req->r_reassert_version.version = cpu_to_le64(reassert_version);
f24e9980
SW		 1775
1776 req->r_got_reply = 1;
1777 } else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) {
1778 dout("handle_reply tid %llu dup ack\n", tid);
34b43a56 1779 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 1780 goto done;
1781 }
1782
1783 dout("handle_reply tid %llu flags %d\n", tid, flags);
1784
a40c4f10
YS		 1785 if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK))
1786 __register_linger_request(osdc, req);
1787
f24e9980 1788 /* either this is a read, or we got the safe response */
0ceed5db
SW		 1789 if (result < 0 ||
1790 (flags & CEPH_OSD_FLAG_ONDISK) ||
f24e9980
SW		 1791 ((flags & CEPH_OSD_FLAG_WRITE) == 0))
1792 __unregister_request(osdc, req);
1793
1794 mutex_unlock(&osdc->request_mutex);
1795
eb845ff1 1796 if (!already_completed) {
61c5d6bf
YZ		 1797 if (req->r_unsafe_callback &&
1798 result >= 0 && !(flags & CEPH_OSD_FLAG_ONDISK))
1799 req->r_unsafe_callback(req, true);
eb845ff1
YZ		 1800 if (req->r_callback)
1801 req->r_callback(req, msg);
1802 else
1803 complete_all(&req->r_completion);
1804 }
f24e9980 1805
61c5d6bf
YZ		 1806 if (flags & CEPH_OSD_FLAG_ONDISK) {
1807 if (req->r_unsafe_callback && already_completed)
1808 req->r_unsafe_callback(req, false);
25845472 1809 complete_request(req);
61c5d6bf 1810 }
f24e9980
SW		 1811
1812done:
a40c4f10 1813 dout("req=%p req->r_linger=%d\n", req, req->r_linger);
f24e9980
SW		 1814 ceph_osdc_put_request(req);
1815 return;
1816
1b83bef2 1817bad_put:
37c89bde
LW		 1818 req->r_result = -EIO;
1819 __unregister_request(osdc, req);
1820 if (req->r_callback)
1821 req->r_callback(req, msg);
1822 else
1823 complete_all(&req->r_completion);
1824 complete_request(req);
1b83bef2 1825 ceph_osdc_put_request(req);
8058fd45
AE		 1826bad_mutex:
1827 mutex_unlock(&osdc->request_mutex);
f24e9980 1828bad:
1b83bef2
SW		 1829 pr_err("corrupt osd_op_reply got %d %d\n",
1830 (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len));
9ec7cab1 1831 ceph_msg_dump(msg);
f24e9980
SW		 1832}
1833
6f6c7006 1834static void reset_changed_osds(struct ceph_osd_client *osdc)
f24e9980 1835{
f24e9980 1836 struct rb_node *p, *n;
f24e9980 1837
6f6c7006
SW		 1838 for (p = rb_first(&osdc->osds); p; p = n) {
1839 struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
f24e9980 1840
6f6c7006
SW		 1841 n = rb_next(p);
1842 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
1843 memcmp(&osd->o_con.peer_addr,
1844 ceph_osd_addr(osdc->osdmap,
1845 osd->o_osd),
1846 sizeof(struct ceph_entity_addr)) != 0)
1847 __reset_osd(osdc, osd);
f24e9980 1848 }
422d2cb8
YS		 1849}
1850
1851/*
6f6c7006
SW		 1852 * Requeue requests whose mapping to an OSD has changed. If requests map to
1853 * no osd, request a new map.
422d2cb8 1854 *
e6d50f67 1855 * Caller should hold map_sem for read.
422d2cb8 1856 */
9a1ea2db
JD		 1857static void kick_requests(struct ceph_osd_client *osdc, bool force_resend,
1858 bool force_resend_writes)
422d2cb8 1859{
a40c4f10 1860 struct ceph_osd_request *req, *nreq;
6f6c7006
SW		 1861 struct rb_node *p;
1862 int needmap = 0;
1863 int err;
9a1ea2db 1864 bool force_resend_req;
422d2cb8 1865
9a1ea2db
JD		 1866 dout("kick_requests %s %s\n", force_resend ? " (force resend)" : "",
1867 force_resend_writes ? " (force resend writes)" : "");
422d2cb8 1868 mutex_lock(&osdc->request_mutex);
6194ea89 1869 for (p = rb_first(&osdc->requests); p; ) {
6f6c7006 1870 req = rb_entry(p, struct ceph_osd_request, r_node);
6194ea89 1871 p = rb_next(p);
ab60b16d
AE		 1872
1873 /*
1874 * For linger requests that have not yet been
1875 * registered, move them to the linger list; they'll
1876 * be sent to the osd in the loop below. Unregister
1877 * the request before re-registering it as a linger
1878 * request to ensure the __map_request() below
1879 * will decide it needs to be sent.
1880 */
1881 if (req->r_linger && list_empty(&req->r_linger_item)) {
1882 dout("%p tid %llu restart on osd%d\n",
1883 req, req->r_tid,
1884 req->r_osd ? req->r_osd->o_osd : -1);
96e4dac6 1885 ceph_osdc_get_request(req);
ab60b16d
AE		 1886 __unregister_request(osdc, req);
1887 __register_linger_request(osdc, req);
96e4dac6 1888 ceph_osdc_put_request(req);
ab60b16d
AE		 1889 continue;
1890 }
1891
9a1ea2db
JD		 1892 force_resend_req = force_resend ||
1893 (force_resend_writes &&
1894 req->r_flags & CEPH_OSD_FLAG_WRITE);
1895 err = __map_request(osdc, req, force_resend_req);
6f6c7006
SW		 1896 if (err < 0)
1897 continue; /* error */
1898 if (req->r_osd == NULL) {
1899 dout("%p tid %llu maps to no osd\n", req, req->r_tid);
1900 needmap++; /* request a newer map */
1901 } else if (err > 0) {
6194ea89
SW		 1902 if (!req->r_linger) {
1903 dout("%p tid %llu requeued on osd%d\n", req,
1904 req->r_tid,
1905 req->r_osd ? req->r_osd->o_osd : -1);
a40c4f10 1906 req->r_flags |= CEPH_OSD_FLAG_RETRY;
6194ea89
SW		 1907 }
1908 }
a40c4f10
YS		 1909 }
1910
1911 list_for_each_entry_safe(req, nreq, &osdc->req_linger,
1912 r_linger_item) {
1913 dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
1914
9a1ea2db
JD		 1915 err = __map_request(osdc, req,
1916 force_resend || force_resend_writes);
ab60b16d 1917 dout("__map_request returned %d\n", err);
a40c4f10
YS		 1918 if (err == 0)
1919 continue; /* no change and no osd was specified */
1920 if (err < 0)
1921 continue; /* hrm! */
1922 if (req->r_osd == NULL) {
1923 dout("tid %llu maps to no valid osd\n", req->r_tid);
1924 needmap++; /* request a newer map */
1925 continue;
6f6c7006 1926 }
a40c4f10
YS		 1927
1928 dout("kicking lingering %p tid %llu osd%d\n", req, req->r_tid,
1929 req->r_osd ? req->r_osd->o_osd : -1);
a40c4f10 1930 __register_request(osdc, req);
c89ce05e 1931 __unregister_linger_request(osdc, req);
6f6c7006 1932 }
14d2f38d 1933 reset_changed_osds(osdc);
f24e9980
SW		 1934 mutex_unlock(&osdc->request_mutex);
1935
1936 if (needmap) {
1937 dout("%d requests for down osds, need new map\n", needmap);
1938 ceph_monc_request_next_osdmap(&osdc->client->monc);
1939 }
422d2cb8 1940}
6f6c7006
SW		 1941
1942
f24e9980
SW		 1943/*
1944 * Process updated osd map.
1945 *
1946 * The message contains any number of incremental and full maps, normally
1947 * indicating some sort of topology change in the cluster. Kick requests
1948 * off to different OSDs as needed.
1949 */
1950void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1951{
1952 void *p, *end, *next;
1953 u32 nr_maps, maplen;
1954 u32 epoch;
1955 struct ceph_osdmap *newmap = NULL, *oldmap;
1956 int err;
1957 struct ceph_fsid fsid;
9a1ea2db 1958 bool was_full;
f24e9980
SW		 1959
1960 dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
1961 p = msg->front.iov_base;
1962 end = p + msg->front.iov_len;
1963
1964 /* verify fsid */
1965 ceph_decode_need(&p, end, sizeof(fsid), bad);
1966 ceph_decode_copy(&p, &fsid, sizeof(fsid));
0743304d
SW		 1967 if (ceph_check_fsid(osdc->client, &fsid) < 0)
1968 return;
f24e9980
SW		 1969
1970 down_write(&osdc->map_sem);
1971
9a1ea2db
JD		 1972 was_full = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
1973
f24e9980
SW		 1974 /* incremental maps */
1975 ceph_decode_32_safe(&p, end, nr_maps, bad);
1976 dout(" %d inc maps\n", nr_maps);
1977 while (nr_maps > 0) {
1978 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 1979 epoch = ceph_decode_32(&p);
1980 maplen = ceph_decode_32(&p);
f24e9980
SW		 1981 ceph_decode_need(&p, end, maplen, bad);
1982 next = p + maplen;
1983 if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
1984 dout("applying incremental map %u len %d\n",
1985 epoch, maplen);
1986 newmap = osdmap_apply_incremental(&p, next,
1987 osdc->osdmap,
15d9882c 1988 &osdc->client->msgr);
f24e9980
SW		 1989 if (IS_ERR(newmap)) {
1990 err = PTR_ERR(newmap);
1991 goto bad;
1992 }
30dc6381 1993 BUG_ON(!newmap);
f24e9980
SW		 1994 if (newmap != osdc->osdmap) {
1995 ceph_osdmap_destroy(osdc->osdmap);
1996 osdc->osdmap = newmap;
1997 }
9a1ea2db
JD		 1998 was_full = was_full ||
1999 ceph_osdmap_flag(osdc->osdmap,
2000 CEPH_OSDMAP_FULL);
2001 kick_requests(osdc, 0, was_full);
f24e9980
SW		 2002 } else {
2003 dout("ignoring incremental map %u len %d\n",
2004 epoch, maplen);
2005 }
2006 p = next;
2007 nr_maps--;
2008 }
2009 if (newmap)
2010 goto done;
2011
2012 /* full maps */
2013 ceph_decode_32_safe(&p, end, nr_maps, bad);
2014 dout(" %d full maps\n", nr_maps);
2015 while (nr_maps) {
2016 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 2017 epoch = ceph_decode_32(&p);
2018 maplen = ceph_decode_32(&p);
f24e9980
SW		 2019 ceph_decode_need(&p, end, maplen, bad);
2020 if (nr_maps > 1) {
2021 dout("skipping non-latest full map %u len %d\n",
2022 epoch, maplen);
2023 } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
2024 dout("skipping full map %u len %d, "
2025 "older than our %u\n", epoch, maplen,
2026 osdc->osdmap->epoch);
2027 } else {
38d6453c
SW		 2028 int skipped_map = 0;
2029
f24e9980
SW		 2030 dout("taking full map %u len %d\n", epoch, maplen);
2031 newmap = osdmap_decode(&p, p+maplen);
2032 if (IS_ERR(newmap)) {
2033 err = PTR_ERR(newmap);
2034 goto bad;
2035 }
30dc6381 2036 BUG_ON(!newmap);
f24e9980
SW		 2037 oldmap = osdc->osdmap;
2038 osdc->osdmap = newmap;
38d6453c
SW		 2039 if (oldmap) {
2040 if (oldmap->epoch + 1 < newmap->epoch)
2041 skipped_map = 1;
f24e9980 2042 ceph_osdmap_destroy(oldmap);
38d6453c 2043 }
9a1ea2db
JD		 2044 was_full = was_full ||
2045 ceph_osdmap_flag(osdc->osdmap,
2046 CEPH_OSDMAP_FULL);
2047 kick_requests(osdc, skipped_map, was_full);
f24e9980
SW		 2048 }
2049 p += maplen;
2050 nr_maps--;
2051 }
2052
b72e19b9
DC		 2053 if (!osdc->osdmap)
2054 goto bad;
f24e9980
SW		 2055done:
2056 downgrade_write(&osdc->map_sem);
2057 ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch);
cd634fb6
SW		 2058
2059 /*
2060 * subscribe to subsequent osdmap updates if full to ensure
2061 * we find out when we are no longer full and stop returning
2062 * ENOSPC.
2063 */
d29adb34
JD		 2064 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) ||
2065 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD) ||
2066 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR))
cd634fb6
SW		 2067 ceph_monc_request_next_osdmap(&osdc->client->monc);
2068
f9d25199
AE		 2069 mutex_lock(&osdc->request_mutex);
2070 __send_queued(osdc);
2071 mutex_unlock(&osdc->request_mutex);
f24e9980 2072 up_read(&osdc->map_sem);
03066f23 2073 wake_up_all(&osdc->client->auth_wq);
f24e9980
SW		 2074 return;
2075
2076bad:
2077 pr_err("osdc handle_map corrupt msg\n");
9ec7cab1 2078 ceph_msg_dump(msg);
f24e9980
SW		 2079 up_write(&osdc->map_sem);
2080 return;
2081}
2082
a40c4f10
YS		 2083/*
2084 * watch/notify callback event infrastructure
2085 *
2086 * These callbacks are used both for watch and notify operations.
2087 */
2088static void __release_event(struct kref *kref)
2089{
2090 struct ceph_osd_event *event =
2091 container_of(kref, struct ceph_osd_event, kref);
2092
2093 dout("__release_event %p\n", event);
2094 kfree(event);
2095}
2096
2097static void get_event(struct ceph_osd_event *event)
2098{
2099 kref_get(&event->kref);
2100}
2101
2102void ceph_osdc_put_event(struct ceph_osd_event *event)
2103{
2104 kref_put(&event->kref, __release_event);
2105}
2106EXPORT_SYMBOL(ceph_osdc_put_event);
2107
2108static void __insert_event(struct ceph_osd_client *osdc,
2109 struct ceph_osd_event *new)
2110{
2111 struct rb_node **p = &osdc->event_tree.rb_node;
2112 struct rb_node *parent = NULL;
2113 struct ceph_osd_event *event = NULL;
2114
2115 while (*p) {
2116 parent = *p;
2117 event = rb_entry(parent, struct ceph_osd_event, node);
2118 if (new->cookie < event->cookie)
2119 p = &(*p)->rb_left;
2120 else if (new->cookie > event->cookie)
2121 p = &(*p)->rb_right;
2122 else
2123 BUG();
2124 }
2125
2126 rb_link_node(&new->node, parent, p);
2127 rb_insert_color(&new->node, &osdc->event_tree);
2128}
2129
2130static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
2131 u64 cookie)
2132{
2133 struct rb_node **p = &osdc->event_tree.rb_node;
2134 struct rb_node *parent = NULL;
2135 struct ceph_osd_event *event = NULL;
2136
2137 while (*p) {
2138 parent = *p;
2139 event = rb_entry(parent, struct ceph_osd_event, node);
2140 if (cookie < event->cookie)
2141 p = &(*p)->rb_left;
2142 else if (cookie > event->cookie)
2143 p = &(*p)->rb_right;
2144 else
2145 return event;
2146 }
2147 return NULL;
2148}
2149
2150static void __remove_event(struct ceph_osd_event *event)
2151{
2152 struct ceph_osd_client *osdc = event->osdc;
2153
2154 if (!RB_EMPTY_NODE(&event->node)) {
2155 dout("__remove_event removed %p\n", event);
2156 rb_erase(&event->node, &osdc->event_tree);
2157 ceph_osdc_put_event(event);
2158 } else {
2159 dout("__remove_event didn't remove %p\n", event);
2160 }
2161}
2162
2163int ceph_osdc_create_event(struct ceph_osd_client *osdc,
2164 void (*event_cb)(u64, u64, u8, void *),
3c663bbd 2165 void *data, struct ceph_osd_event **pevent)
a40c4f10
YS		 2166{
2167 struct ceph_osd_event *event;
2168
2169 event = kmalloc(sizeof(*event), GFP_NOIO);
2170 if (!event)
2171 return -ENOMEM;
2172
2173 dout("create_event %p\n", event);
2174 event->cb = event_cb;
3c663bbd 2175 event->one_shot = 0;
a40c4f10
YS		 2176 event->data = data;
2177 event->osdc = osdc;
2178 INIT_LIST_HEAD(&event->osd_node);
3ee5234d 2179 RB_CLEAR_NODE(&event->node);
a40c4f10
YS		 2180 kref_init(&event->kref); /* one ref for us */
2181 kref_get(&event->kref); /* one ref for the caller */
a40c4f10
YS		 2182
2183 spin_lock(&osdc->event_lock);
2184 event->cookie = ++osdc->event_count;
2185 __insert_event(osdc, event);
2186 spin_unlock(&osdc->event_lock);
2187
2188 *pevent = event;
2189 return 0;
2190}
2191EXPORT_SYMBOL(ceph_osdc_create_event);
2192
2193void ceph_osdc_cancel_event(struct ceph_osd_event *event)
2194{
2195 struct ceph_osd_client *osdc = event->osdc;
2196
2197 dout("cancel_event %p\n", event);
2198 spin_lock(&osdc->event_lock);
2199 __remove_event(event);
2200 spin_unlock(&osdc->event_lock);
2201 ceph_osdc_put_event(event); /* caller's */
2202}
2203EXPORT_SYMBOL(ceph_osdc_cancel_event);
2204
2205
2206static void do_event_work(struct work_struct *work)
2207{
2208 struct ceph_osd_event_work *event_work =
2209 container_of(work, struct ceph_osd_event_work, work);
2210 struct ceph_osd_event *event = event_work->event;
2211 u64 ver = event_work->ver;
2212 u64 notify_id = event_work->notify_id;
2213 u8 opcode = event_work->opcode;
2214
2215 dout("do_event_work completing %p\n", event);
2216 event->cb(ver, notify_id, opcode, event->data);
a40c4f10
YS		 2217 dout("do_event_work completed %p\n", event);
2218 ceph_osdc_put_event(event);
2219 kfree(event_work);
2220}
2221
2222
2223/*
2224 * Process osd watch notifications
2225 */
3c663bbd
AE		 2226static void handle_watch_notify(struct ceph_osd_client *osdc,
2227 struct ceph_msg *msg)
a40c4f10
YS		 2228{
2229 void *p, *end;
2230 u8 proto_ver;
2231 u64 cookie, ver, notify_id;
2232 u8 opcode;
2233 struct ceph_osd_event *event;
2234 struct ceph_osd_event_work *event_work;
2235
2236 p = msg->front.iov_base;
2237 end = p + msg->front.iov_len;
2238
2239 ceph_decode_8_safe(&p, end, proto_ver, bad);
2240 ceph_decode_8_safe(&p, end, opcode, bad);
2241 ceph_decode_64_safe(&p, end, cookie, bad);
2242 ceph_decode_64_safe(&p, end, ver, bad);
2243 ceph_decode_64_safe(&p, end, notify_id, bad);
2244
2245 spin_lock(&osdc->event_lock);
2246 event = __find_event(osdc, cookie);
2247 if (event) {
3c663bbd 2248 BUG_ON(event->one_shot);
a40c4f10 2249 get_event(event);
a40c4f10
YS		 2250 }
2251 spin_unlock(&osdc->event_lock);
2252 dout("handle_watch_notify cookie %lld ver %lld event %p\n",
2253 cookie, ver, event);
2254 if (event) {
2255 event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
a40c4f10
YS		 2256 if (!event_work) {
2257 dout("ERROR: could not allocate event_work\n");
2258 goto done_err;
2259 }
6b0ae409 2260 INIT_WORK(&event_work->work, do_event_work);
a40c4f10
YS		 2261 event_work->event = event;
2262 event_work->ver = ver;
2263 event_work->notify_id = notify_id;
2264 event_work->opcode = opcode;
2265 if (!queue_work(osdc->notify_wq, &event_work->work)) {
2266 dout("WARNING: failed to queue notify event work\n");
2267 goto done_err;
2268 }
2269 }
2270
2271 return;
2272
2273done_err:
a40c4f10
YS		 2274 ceph_osdc_put_event(event);
2275 return;
2276
2277bad:
2278 pr_err("osdc handle_watch_notify corrupt msg\n");
2279 return;
2280}
2281
e65550fd
AE		 2282/*
2283 * build new request AND message
2284 *
2285 */
2286void ceph_osdc_build_request(struct ceph_osd_request *req, u64 off,
2287 struct ceph_snap_context *snapc, u64 snap_id,
2288 struct timespec *mtime)
2289{
2290 struct ceph_msg *msg = req->r_request;
2291 void *p;
2292 size_t msg_size;
2293 int flags = req->r_flags;
2294 u64 data_len;
2295 unsigned int i;
2296
2297 req->r_snapid = snap_id;
2298 req->r_snapc = ceph_get_snap_context(snapc);
2299
2300 /* encode request */
2301 msg->hdr.version = cpu_to_le16(4);
2302
2303 p = msg->front.iov_base;
2304 ceph_encode_32(&p, 1); /* client_inc is always 1 */
2305 req->r_request_osdmap_epoch = p;
2306 p += 4;
2307 req->r_request_flags = p;
2308 p += 4;
2309 if (req->r_flags & CEPH_OSD_FLAG_WRITE)
2310 ceph_encode_timespec(p, mtime);
2311 p += sizeof(struct ceph_timespec);
2312 req->r_request_reassert_version = p;
2313 p += sizeof(struct ceph_eversion); /* will get filled in */
2314
2315 /* oloc */
2316 ceph_encode_8(&p, 4);
2317 ceph_encode_8(&p, 4);
2318 ceph_encode_32(&p, 8 + 4 + 4);
2319 req->r_request_pool = p;
2320 p += 8;
2321 ceph_encode_32(&p, -1); /* preferred */
2322 ceph_encode_32(&p, 0); /* key len */
2323
2324 ceph_encode_8(&p, 1);
2325 req->r_request_pgid = p;
2326 p += 8 + 4;
2327 ceph_encode_32(&p, -1); /* preferred */
2328
2329 /* oid */
3c972c95
ID		 2330 ceph_encode_32(&p, req->r_base_oid.name_len);
2331 memcpy(p, req->r_base_oid.name, req->r_base_oid.name_len);
2332 dout("oid '%.*s' len %d\n", req->r_base_oid.name_len,
2333 req->r_base_oid.name, req->r_base_oid.name_len);
2334 p += req->r_base_oid.name_len;
e65550fd
AE		 2335
2336 /* ops--can imply data */
2337 ceph_encode_16(&p, (u16)req->r_num_ops);
2338 data_len = 0;
2339 for (i = 0; i < req->r_num_ops; i++) {
2340 data_len += osd_req_encode_op(req, p, i);
2341 p += sizeof(struct ceph_osd_op);
2342 }
2343
2344 /* snaps */
2345 ceph_encode_64(&p, req->r_snapid);
2346 ceph_encode_64(&p, req->r_snapc ? req->r_snapc->seq : 0);
2347 ceph_encode_32(&p, req->r_snapc ? req->r_snapc->num_snaps : 0);
2348 if (req->r_snapc) {
2349 for (i = 0; i < snapc->num_snaps; i++) {
2350 ceph_encode_64(&p, req->r_snapc->snaps[i]);
2351 }
2352 }
2353
2354 req->r_request_attempts = p;
2355 p += 4;
2356
2357 /* data */
2358 if (flags & CEPH_OSD_FLAG_WRITE) {
2359 u16 data_off;
2360
2361 /*
2362 * The header "data_off" is a hint to the receiver
2363 * allowing it to align received data into its
2364 * buffers such that there's no need to re-copy
2365 * it before writing it to disk (direct I/O).
2366 */
2367 data_off = (u16) (off & 0xffff);
2368 req->r_request->hdr.data_off = cpu_to_le16(data_off);
2369 }
2370 req->r_request->hdr.data_len = cpu_to_le32(data_len);
2371
2372 BUG_ON(p > msg->front.iov_base + msg->front.iov_len);
2373 msg_size = p - msg->front.iov_base;
2374 msg->front.iov_len = msg_size;
2375 msg->hdr.front_len = cpu_to_le32(msg_size);
2376
2377 dout("build_request msg_size was %d\n", (int)msg_size);
2378}
2379EXPORT_SYMBOL(ceph_osdc_build_request);
2380
70636773
AE		 2381/*
2382 * Register request, send initial attempt.
2383 */
2384int ceph_osdc_start_request(struct ceph_osd_client *osdc,
2385 struct ceph_osd_request *req,
2386 bool nofail)
2387{
0bbfdfe8 2388 int rc;
70636773 2389
f24e9980
SW		 2390 down_read(&osdc->map_sem);
2391 mutex_lock(&osdc->request_mutex);
0bbfdfe8
ID		 2392
2393 rc = __ceph_osdc_start_request(osdc, req, nofail);
2394
f24e9980
SW		 2395 mutex_unlock(&osdc->request_mutex);
2396 up_read(&osdc->map_sem);
0bbfdfe8 2397
f24e9980
SW		 2398 return rc;
2399}
3d14c5d2 2400EXPORT_SYMBOL(ceph_osdc_start_request);
f24e9980
SW		 2401
2402/*
2403 * wait for a request to complete
2404 */
2405int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
2406 struct ceph_osd_request *req)
2407{
2408 int rc;
2409
2410 rc = wait_for_completion_interruptible(&req->r_completion);
2411 if (rc < 0) {
2412 mutex_lock(&osdc->request_mutex);
2413 __cancel_request(req);
529cfcc4 2414 __unregister_request(osdc, req);
f24e9980 2415 mutex_unlock(&osdc->request_mutex);
25845472 2416 complete_request(req);
529cfcc4 2417 dout("wait_request tid %llu canceled/timed out\n", req->r_tid);
f24e9980
SW		 2418 return rc;
2419 }
2420
2421 dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result);
2422 return req->r_result;
2423}
3d14c5d2 2424EXPORT_SYMBOL(ceph_osdc_wait_request);
f24e9980
SW		 2425
2426/*
2427 * sync - wait for all in-flight requests to flush. avoid starvation.
2428 */
2429void ceph_osdc_sync(struct ceph_osd_client *osdc)
2430{
2431 struct ceph_osd_request *req;
2432 u64 last_tid, next_tid = 0;
2433
2434 mutex_lock(&osdc->request_mutex);
2435 last_tid = osdc->last_tid;
2436 while (1) {
2437 req = __lookup_request_ge(osdc, next_tid);
2438 if (!req)
2439 break;
2440 if (req->r_tid > last_tid)
2441 break;
2442
2443 next_tid = req->r_tid + 1;
2444 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
2445 continue;
2446
2447 ceph_osdc_get_request(req);
2448 mutex_unlock(&osdc->request_mutex);
2449 dout("sync waiting on tid %llu (last is %llu)\n",
2450 req->r_tid, last_tid);
2451 wait_for_completion(&req->r_safe_completion);
2452 mutex_lock(&osdc->request_mutex);
2453 ceph_osdc_put_request(req);
2454 }
2455 mutex_unlock(&osdc->request_mutex);
2456 dout("sync done (thru tid %llu)\n", last_tid);
2457}
3d14c5d2 2458EXPORT_SYMBOL(ceph_osdc_sync);
f24e9980 2459
dd935f44
JD		 2460/*
2461 * Call all pending notify callbacks - for use after a watch is
2462 * unregistered, to make sure no more callbacks for it will be invoked
2463 */
2464extern void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc)
2465{
2466 flush_workqueue(osdc->notify_wq);
2467}
2468EXPORT_SYMBOL(ceph_osdc_flush_notifies);
2469
2470
f24e9980
SW		 2471/*
2472 * init, shutdown
2473 */
2474int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
2475{
2476 int err;
2477
2478 dout("init\n");
2479 osdc->client = client;
2480 osdc->osdmap = NULL;
2481 init_rwsem(&osdc->map_sem);
2482 init_completion(&osdc->map_waiters);
2483 osdc->last_requested_map = 0;
2484 mutex_init(&osdc->request_mutex);
f24e9980
SW		 2485 osdc->last_tid = 0;
2486 osdc->osds = RB_ROOT;
f5a2041b 2487 INIT_LIST_HEAD(&osdc->osd_lru);
f24e9980 2488 osdc->requests = RB_ROOT;
422d2cb8 2489 INIT_LIST_HEAD(&osdc->req_lru);
6f6c7006
SW		 2490 INIT_LIST_HEAD(&osdc->req_unsent);
2491 INIT_LIST_HEAD(&osdc->req_notarget);
a40c4f10 2492 INIT_LIST_HEAD(&osdc->req_linger);
f24e9980
SW		 2493 osdc->num_requests = 0;
2494 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
f5a2041b 2495 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
a40c4f10
YS		 2496 spin_lock_init(&osdc->event_lock);
2497 osdc->event_tree = RB_ROOT;
2498 osdc->event_count = 0;
f5a2041b
YS		 2499
2500 schedule_delayed_work(&osdc->osds_timeout_work,
3d14c5d2 2501 round_jiffies_relative(osdc->client->options->osd_idle_ttl * HZ));
f24e9980 2502
5f44f142 2503 err = -ENOMEM;
f24e9980
SW		 2504 osdc->req_mempool = mempool_create_kmalloc_pool(10,
2505 sizeof(struct ceph_osd_request));
2506 if (!osdc->req_mempool)
5f44f142 2507 goto out;
f24e9980 2508
d50b409f
SW		 2509 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
2510 OSD_OP_FRONT_LEN, 10, true,
4f48280e 2511 "osd_op");
f24e9980 2512 if (err < 0)
5f44f142 2513 goto out_mempool;
d50b409f 2514 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
4f48280e
SW		 2515 OSD_OPREPLY_FRONT_LEN, 10, true,
2516 "osd_op_reply");
c16e7869
SW		 2517 if (err < 0)
2518 goto out_msgpool;
a40c4f10 2519
dbcae088 2520 err = -ENOMEM;
a40c4f10 2521 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
dbcae088 2522 if (!osdc->notify_wq)
c172ec5c
ID		 2523 goto out_msgpool_reply;
2524
f24e9980 2525 return 0;
5f44f142 2526
c172ec5c
ID		 2527out_msgpool_reply:
2528 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
c16e7869
SW		 2529out_msgpool:
2530 ceph_msgpool_destroy(&osdc->msgpool_op);
5f44f142
SW		 2531out_mempool:
2532 mempool_destroy(osdc->req_mempool);
2533out:
2534 return err;
f24e9980
SW		 2535}
2536
2537void ceph_osdc_stop(struct ceph_osd_client *osdc)
2538{
a40c4f10
YS		 2539 flush_workqueue(osdc->notify_wq);
2540 destroy_workqueue(osdc->notify_wq);
f24e9980 2541 cancel_delayed_work_sync(&osdc->timeout_work);
f5a2041b 2542 cancel_delayed_work_sync(&osdc->osds_timeout_work);
f24e9980
SW		 2543 if (osdc->osdmap) {
2544 ceph_osdmap_destroy(osdc->osdmap);
2545 osdc->osdmap = NULL;
2546 }
aca420bc 2547 remove_all_osds(osdc);
f24e9980
SW		 2548 mempool_destroy(osdc->req_mempool);
2549 ceph_msgpool_destroy(&osdc->msgpool_op);
c16e7869 2550 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
f24e9980
SW		 2551}
2552
2553/*
2554 * Read some contiguous pages. If we cross a stripe boundary, shorten
2555 * *plen. Return number of bytes read, or error.
2556 */
2557int ceph_osdc_readpages(struct ceph_osd_client *osdc,
2558 struct ceph_vino vino, struct ceph_file_layout *layout,
2559 u64 off, u64 *plen,
2560 u32 truncate_seq, u64 truncate_size,
b7495fc2 2561 struct page **pages, int num_pages, int page_align)
f24e9980
SW		 2562{
2563 struct ceph_osd_request *req;
2564 int rc = 0;
2565
2566 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
2567 vino.snap, off, *plen);
79528734 2568 req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 1,
f24e9980 2569 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
acead002 2570 NULL, truncate_seq, truncate_size,
153e5167 2571 false);
6816282d
SW		 2572 if (IS_ERR(req))
2573 return PTR_ERR(req);
f24e9980
SW		 2574
2575 /* it may be a short read due to an object boundary */
0fff87ec 2576
406e2c9f 2577 osd_req_op_extent_osd_data_pages(req, 0,
a4ce40a9 2578 pages, *plen, page_align, false, false);
f24e9980 2579
e0c59487 2580 dout("readpages final extent is %llu~%llu (%llu bytes align %d)\n",
43bfe5de 2581 off, *plen, *plen, page_align);
f24e9980 2582
79528734 2583 ceph_osdc_build_request(req, off, NULL, vino.snap, NULL);
02ee07d3 2584
f24e9980
SW		 2585 rc = ceph_osdc_start_request(osdc, req, false);
2586 if (!rc)
2587 rc = ceph_osdc_wait_request(osdc, req);
2588
2589 ceph_osdc_put_request(req);
2590 dout("readpages result %d\n", rc);
2591 return rc;
2592}
3d14c5d2 2593EXPORT_SYMBOL(ceph_osdc_readpages);
f24e9980
SW		 2594
2595/*
2596 * do a synchronous write on N pages
2597 */
2598int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
2599 struct ceph_file_layout *layout,
2600 struct ceph_snap_context *snapc,
2601 u64 off, u64 len,
2602 u32 truncate_seq, u64 truncate_size,
2603 struct timespec *mtime,
24808826 2604 struct page **pages, int num_pages)
f24e9980
SW		 2605{
2606 struct ceph_osd_request *req;
2607 int rc = 0;
b7495fc2 2608 int page_align = off & ~PAGE_MASK;
f24e9980 2609
acead002 2610 BUG_ON(vino.snap != CEPH_NOSNAP); /* snapshots aren't writeable */
79528734 2611 req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 1,
f24e9980 2612 CEPH_OSD_OP_WRITE,
24808826 2613 CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
acead002 2614 snapc, truncate_seq, truncate_size,
153e5167 2615 true);
6816282d
SW		 2616 if (IS_ERR(req))
2617 return PTR_ERR(req);
f24e9980
SW		 2618
2619 /* it may be a short write due to an object boundary */
406e2c9f 2620 osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align,
43bfe5de
AE		 2621 false, false);
2622 dout("writepages %llu~%llu (%llu bytes)\n", off, len, len);
f24e9980 2623
79528734 2624 ceph_osdc_build_request(req, off, snapc, CEPH_NOSNAP, mtime);
02ee07d3 2625
87f979d3 2626 rc = ceph_osdc_start_request(osdc, req, true);
f24e9980
SW		 2627 if (!rc)
2628 rc = ceph_osdc_wait_request(osdc, req);
2629
2630 ceph_osdc_put_request(req);
2631 if (rc == 0)
2632 rc = len;
2633 dout("writepages result %d\n", rc);
2634 return rc;
2635}
3d14c5d2 2636EXPORT_SYMBOL(ceph_osdc_writepages);
f24e9980 2637
5522ae0b
AE		 2638int ceph_osdc_setup(void)
2639{
2640 BUG_ON(ceph_osd_request_cache);
2641 ceph_osd_request_cache = kmem_cache_create("ceph_osd_request",
2642 sizeof (struct ceph_osd_request),
2643 __alignof__(struct ceph_osd_request),
2644 0, NULL);
2645
2646 return ceph_osd_request_cache ? 0 : -ENOMEM;
2647}
2648EXPORT_SYMBOL(ceph_osdc_setup);
2649
2650void ceph_osdc_cleanup(void)
2651{
2652 BUG_ON(!ceph_osd_request_cache);
2653 kmem_cache_destroy(ceph_osd_request_cache);
2654 ceph_osd_request_cache = NULL;
2655}
2656EXPORT_SYMBOL(ceph_osdc_cleanup);
2657
f24e9980
SW		 2658/*
2659 * handle incoming message
2660 */
2661static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
2662{
2663 struct ceph_osd *osd = con->private;
32c895e7 2664 struct ceph_osd_client *osdc;
f24e9980
SW		 2665 int type = le16_to_cpu(msg->hdr.type);
2666
2667 if (!osd)
4a32f93d 2668 goto out;
32c895e7 2669 osdc = osd->o_osdc;
f24e9980
SW		 2670
2671 switch (type) {
2672 case CEPH_MSG_OSD_MAP:
2673 ceph_osdc_handle_map(osdc, msg);
2674 break;
2675 case CEPH_MSG_OSD_OPREPLY:
350b1c32 2676 handle_reply(osdc, msg, con);
f24e9980 2677 break;
a40c4f10
YS		 2678 case CEPH_MSG_WATCH_NOTIFY:
2679 handle_watch_notify(osdc, msg);
2680 break;
f24e9980
SW		 2681
2682 default:
2683 pr_err("received unknown message type %d %s\n", type,
2684 ceph_msg_type_name(type));
2685 }
4a32f93d 2686out:
f24e9980
SW		 2687 ceph_msg_put(msg);
2688}
2689
5b3a4db3 2690/*
21b667f6
SW		 2691 * lookup and return message for incoming reply. set up reply message
2692 * pages.
5b3a4db3
SW		 2693 */
2694static struct ceph_msg *get_reply(struct ceph_connection *con,
2450418c
YS		 2695 struct ceph_msg_header *hdr,
2696 int *skip)
f24e9980
SW		 2697{
2698 struct ceph_osd *osd = con->private;
2699 struct ceph_osd_client *osdc = osd->o_osdc;
2450418c 2700 struct ceph_msg *m;
0547a9b3 2701 struct ceph_osd_request *req;
3f0a4ac5 2702 int front_len = le32_to_cpu(hdr->front_len);
5b3a4db3 2703 int data_len = le32_to_cpu(hdr->data_len);
0547a9b3 2704 u64 tid;
f24e9980 2705
0547a9b3
YS		 2706 tid = le64_to_cpu(hdr->tid);
2707 mutex_lock(&osdc->request_mutex);
2708 req = __lookup_request(osdc, tid);
2709 if (!req) {
2710 *skip = 1;
2711 m = NULL;
756a16a5
SW		 2712 dout("get_reply unknown tid %llu from osd%d\n", tid,
2713 osd->o_osd);
0547a9b3
YS		 2714 goto out;
2715 }
c16e7869 2716
ace6d3a9 2717 if (req->r_reply->con)
8921d114 2718 dout("%s revoking msg %p from old con %p\n", __func__,
ace6d3a9
AE		 2719 req->r_reply, req->r_reply->con);
2720 ceph_msg_revoke_incoming(req->r_reply);
0547a9b3 2721
f2be82b0 2722 if (front_len > req->r_reply->front_alloc_len) {
4974341e 2723 pr_warning("get_reply front %d > preallocated %d (%u#%llu)\n",
f2be82b0 2724 front_len, req->r_reply->front_alloc_len,
4974341e
AE		 2725 (unsigned int)con->peer_name.type,
2726 le64_to_cpu(con->peer_name.num));
3f0a4ac5
ID		 2727 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS,
2728 false);
a79832f2 2729 if (!m)
c16e7869
SW		 2730 goto out;
2731 ceph_msg_put(req->r_reply);
2732 req->r_reply = m;
2733 }
2734 m = ceph_msg_get(req->r_reply);
2735
0547a9b3 2736 if (data_len > 0) {
a4ce40a9 2737 struct ceph_osd_data *osd_data;
0fff87ec 2738
a4ce40a9
AE		 2739 /*
2740 * XXX This is assuming there is only one op containing
2741 * XXX page data. Probably OK for reads, but this
2742 * XXX ought to be done more generally.
2743 */
406e2c9f 2744 osd_data = osd_req_op_extent_osd_data(req, 0);
0fff87ec 2745 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
0fff87ec 2746 if (osd_data->pages &&
e0c59487 2747 unlikely(osd_data->length < data_len)) {
2ac2b7a6 2748
e0c59487
AE		 2749 pr_warning("tid %lld reply has %d bytes "
2750 "we had only %llu bytes ready\n",
2751 tid, data_len, osd_data->length);
2ac2b7a6
AE		 2752 *skip = 1;
2753 ceph_msg_put(m);
2754 m = NULL;
2755 goto out;
2756 }
2ac2b7a6 2757 }
0547a9b3 2758 }
5b3a4db3 2759 *skip = 0;
c16e7869 2760 dout("get_reply tid %lld %p\n", tid, m);
0547a9b3
YS		 2761
2762out:
2763 mutex_unlock(&osdc->request_mutex);
2450418c 2764 return m;
5b3a4db3
SW		 2765
2766}
2767
2768static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2769 struct ceph_msg_header *hdr,
2770 int *skip)
2771{
2772 struct ceph_osd *osd = con->private;
2773 int type = le16_to_cpu(hdr->type);
2774 int front = le32_to_cpu(hdr->front_len);
2775
1c20f2d2 2776 *skip = 0;
5b3a4db3
SW		 2777 switch (type) {
2778 case CEPH_MSG_OSD_MAP:
a40c4f10 2779 case CEPH_MSG_WATCH_NOTIFY:
b61c2763 2780 return ceph_msg_new(type, front, GFP_NOFS, false);
5b3a4db3
SW		 2781 case CEPH_MSG_OSD_OPREPLY:
2782 return get_reply(con, hdr, skip);
2783 default:
2784 pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
2785 osd->o_osd);
2786 *skip = 1;
2787 return NULL;
2788 }
f24e9980
SW		 2789}
2790
2791/*
2792 * Wrappers to refcount containing ceph_osd struct
2793 */
2794static struct ceph_connection *get_osd_con(struct ceph_connection *con)
2795{
2796 struct ceph_osd *osd = con->private;
2797 if (get_osd(osd))
2798 return con;
2799 return NULL;
2800}
2801
2802static void put_osd_con(struct ceph_connection *con)
2803{
2804 struct ceph_osd *osd = con->private;
2805 put_osd(osd);
2806}
2807
4e7a5dcd
SW		 2808/*
2809 * authentication
2810 */
a3530df3
AE		 2811/*
2812 * Note: returned pointer is the address of a structure that's
2813 * managed separately. Caller must *not* attempt to free it.
2814 */
2815static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
8f43fb53 2816 int *proto, int force_new)
4e7a5dcd
SW		 2817{
2818 struct ceph_osd *o = con->private;
2819 struct ceph_osd_client *osdc = o->o_osdc;
2820 struct ceph_auth_client *ac = osdc->client->monc.auth;
74f1869f 2821 struct ceph_auth_handshake *auth = &o->o_auth;
4e7a5dcd 2822
74f1869f 2823 if (force_new && auth->authorizer) {
27859f97 2824 ceph_auth_destroy_authorizer(ac, auth->authorizer);
74f1869f
AE		 2825 auth->authorizer = NULL;
2826 }
27859f97
SW		 2827 if (!auth->authorizer) {
2828 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2829 auth);
4e7a5dcd 2830 if (ret)
a3530df3 2831 return ERR_PTR(ret);
27859f97
SW		 2832 } else {
2833 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
0bed9b5c
SW		 2834 auth);
2835 if (ret)
2836 return ERR_PTR(ret);
4e7a5dcd 2837 }
4e7a5dcd 2838 *proto = ac->protocol;
74f1869f 2839
a3530df3 2840 return auth;
4e7a5dcd
SW		 2841}
2842
2843
2844static int verify_authorizer_reply(struct ceph_connection *con, int len)
2845{
2846 struct ceph_osd *o = con->private;
2847 struct ceph_osd_client *osdc = o->o_osdc;
2848 struct ceph_auth_client *ac = osdc->client->monc.auth;
2849
27859f97 2850 return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer, len);
4e7a5dcd
SW		 2851}
2852
9bd2e6f8
SW		 2853static int invalidate_authorizer(struct ceph_connection *con)
2854{
2855 struct ceph_osd *o = con->private;
2856 struct ceph_osd_client *osdc = o->o_osdc;
2857 struct ceph_auth_client *ac = osdc->client->monc.auth;
2858
27859f97 2859 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
9bd2e6f8
SW		 2860 return ceph_monc_validate_auth(&osdc->client->monc);
2861}
4e7a5dcd 2862
9e32789f 2863static const struct ceph_connection_operations osd_con_ops = {
f24e9980
SW		 2864 .get = get_osd_con,
2865 .put = put_osd_con,
2866 .dispatch = dispatch,
4e7a5dcd
SW		 2867 .get_authorizer = get_authorizer,
2868 .verify_authorizer_reply = verify_authorizer_reply,
9bd2e6f8 2869 .invalidate_authorizer = invalidate_authorizer,
f24e9980 2870 .alloc_msg = alloc_msg,
81b024e7 2871 .fault = osd_reset,
f24e9980 2872};
Linux 6.x block layer and io_uring tree(s)