projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
libceph: add data pointers in osd op structures
[linux-2.6-block.git] / net / ceph / osd_client.c
CommitLineData
3d14c5d2 1#include <linux/ceph/ceph_debug.h>
f24e9980 2
3d14c5d2 3#include <linux/module.h>
f24e9980
SW		 4#include <linux/err.h>
5#include <linux/highmem.h>
6#include <linux/mm.h>
7#include <linux/pagemap.h>
8#include <linux/slab.h>
9#include <linux/uaccess.h>
68b4476b
YS		 10#ifdef CONFIG_BLOCK
11#include <linux/bio.h>
12#endif
f24e9980 13
3d14c5d2
YS		 14#include <linux/ceph/libceph.h>
15#include <linux/ceph/osd_client.h>
16#include <linux/ceph/messenger.h>
17#include <linux/ceph/decode.h>
18#include <linux/ceph/auth.h>
19#include <linux/ceph/pagelist.h>
f24e9980 20
c16e7869
SW		 21#define OSD_OP_FRONT_LEN 4096
22#define OSD_OPREPLY_FRONT_LEN 512
0d59ab81 23
9e32789f 24static const struct ceph_connection_operations osd_con_ops;
f24e9980 25
f9d25199 26static void __send_queued(struct ceph_osd_client *osdc);
6f6c7006 27static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd);
a40c4f10
YS		 28static void __register_request(struct ceph_osd_client *osdc,
29 struct ceph_osd_request *req);
30static void __unregister_linger_request(struct ceph_osd_client *osdc,
31 struct ceph_osd_request *req);
56e925b6
SW		 32static void __send_request(struct ceph_osd_client *osdc,
33 struct ceph_osd_request *req);
f24e9980
SW		 34
35/*
36 * Implement client access to distributed object storage cluster.
37 *
38 * All data objects are stored within a cluster/cloud of OSDs, or
39 * "object storage devices." (Note that Ceph OSDs have _nothing_ to
40 * do with the T10 OSD extensions to SCSI.) Ceph OSDs are simply
41 * remote daemons serving up and coordinating consistent and safe
42 * access to storage.
43 *
44 * Cluster membership and the mapping of data objects onto storage devices
45 * are described by the osd map.
46 *
47 * We keep track of pending OSD requests (read, write), resubmit
48 * requests to different OSDs when the cluster topology/data layout
49 * change, or retry the affected requests when the communications
50 * channel with an OSD is reset.
51 */
52
53/*
54 * calculate the mapping of a file extent onto an object, and fill out the
55 * request accordingly. shorten extent as necessary if it crosses an
56 * object boundary.
57 *
58 * fill osd op in request message.
59 */
dbe0fc41 60static int calc_layout(struct ceph_file_layout *layout, u64 off, u64 *plen,
a19dadfb 61 u64 *objnum, u64 *objoff, u64 *objlen)
f24e9980 62{
60e56f13 63 u64 orig_len = *plen;
d63b77f4 64 int r;
f24e9980 65
60e56f13 66 /* object extent? */
75d1c941
AE		 67 r = ceph_calc_file_object_mapping(layout, off, orig_len, objnum,
68 objoff, objlen);
d63b77f4
SW		 69 if (r < 0)
70 return r;
75d1c941
AE		 71 if (*objlen < orig_len) {
72 *plen = *objlen;
60e56f13
AE		 73 dout(" skipping last %llu, final file extent %llu~%llu\n",
74 orig_len - *plen, off, *plen);
75 }
76
75d1c941 77 dout("calc_layout objnum=%llx %llu~%llu\n", *objnum, *objoff, *objlen);
f24e9980 78
3ff5f385 79 return 0;
f24e9980
SW		 80}
81
c54d47bf
AE		 82static void ceph_osd_data_init(struct ceph_osd_data *osd_data)
83{
84 memset(osd_data, 0, sizeof (*osd_data));
85 osd_data->type = CEPH_OSD_DATA_TYPE_NONE;
86}
87
43bfe5de
AE		 88void ceph_osd_data_pages_init(struct ceph_osd_data *osd_data,
89 struct page **pages, u64 length, u32 alignment,
90 bool pages_from_pool, bool own_pages)
91{
92 osd_data->type = CEPH_OSD_DATA_TYPE_PAGES;
93 osd_data->pages = pages;
94 osd_data->length = length;
95 osd_data->alignment = alignment;
96 osd_data->pages_from_pool = pages_from_pool;
97 osd_data->own_pages = own_pages;
98}
99EXPORT_SYMBOL(ceph_osd_data_pages_init);
100
101void ceph_osd_data_pagelist_init(struct ceph_osd_data *osd_data,
102 struct ceph_pagelist *pagelist)
103{
104 osd_data->type = CEPH_OSD_DATA_TYPE_PAGELIST;
105 osd_data->pagelist = pagelist;
106}
107EXPORT_SYMBOL(ceph_osd_data_pagelist_init);
108
109#ifdef CONFIG_BLOCK
110void ceph_osd_data_bio_init(struct ceph_osd_data *osd_data,
111 struct bio *bio, size_t bio_length)
112{
113 osd_data->type = CEPH_OSD_DATA_TYPE_BIO;
114 osd_data->bio = bio;
115 osd_data->bio_length = bio_length;
116}
117EXPORT_SYMBOL(ceph_osd_data_bio_init);
118#endif /* CONFIG_BLOCK */
119
23c08a9c
AE		 120static u64 ceph_osd_data_length(struct ceph_osd_data *osd_data)
121{
122 switch (osd_data->type) {
123 case CEPH_OSD_DATA_TYPE_NONE:
124 return 0;
125 case CEPH_OSD_DATA_TYPE_PAGES:
126 return osd_data->length;
127 case CEPH_OSD_DATA_TYPE_PAGELIST:
128 return (u64)osd_data->pagelist->length;
129#ifdef CONFIG_BLOCK
130 case CEPH_OSD_DATA_TYPE_BIO:
131 return (u64)osd_data->bio_length;
132#endif /* CONFIG_BLOCK */
133 default:
134 WARN(true, "unrecognized data type %d\n", (int)osd_data->type);
135 return 0;
136 }
137}
138
c54d47bf
AE		 139static void ceph_osd_data_release(struct ceph_osd_data *osd_data)
140{
141 if (osd_data->type != CEPH_OSD_DATA_TYPE_PAGES)
142 return;
143
144 if (osd_data->own_pages) {
145 int num_pages;
146
147 num_pages = calc_pages_for((u64)osd_data->alignment,
148 (u64)osd_data->length);
149 ceph_release_page_vector(osd_data->pages, num_pages);
150 }
151}
152
f24e9980
SW		 153/*
154 * requests
155 */
415e49a9 156void ceph_osdc_release_request(struct kref *kref)
f24e9980 157{
c54d47bf 158 struct ceph_osd_request *req;
415e49a9 159
c54d47bf 160 req = container_of(kref, struct ceph_osd_request, r_kref);
415e49a9
SW		 161 if (req->r_request)
162 ceph_msg_put(req->r_request);
ace6d3a9 163 if (req->r_reply) {
8921d114 164 ceph_msg_revoke_incoming(req->r_reply);
ab8cb34a 165 ceph_msg_put(req->r_reply);
ace6d3a9 166 }
0fff87ec 167
c54d47bf
AE		 168 ceph_osd_data_release(&req->r_data_in);
169 ceph_osd_data_release(&req->r_data_out);
0fff87ec 170
415e49a9
SW		 171 ceph_put_snap_context(req->r_snapc);
172 if (req->r_mempool)
173 mempool_free(req, req->r_osdc->req_mempool);
174 else
175 kfree(req);
f24e9980 176}
3d14c5d2 177EXPORT_SYMBOL(ceph_osdc_release_request);
68b4476b 178
3499e8a5 179struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
f24e9980 180 struct ceph_snap_context *snapc,
1b83bef2 181 unsigned int num_ops,
3499e8a5 182 bool use_mempool,
54a54007 183 gfp_t gfp_flags)
f24e9980
SW		 184{
185 struct ceph_osd_request *req;
186 struct ceph_msg *msg;
1b83bef2
SW		 187 size_t msg_size;
188
79528734
AE		 189 BUILD_BUG_ON(CEPH_OSD_MAX_OP > U16_MAX);
190 BUG_ON(num_ops > CEPH_OSD_MAX_OP);
191
1b83bef2
SW		 192 msg_size = 4 + 4 + 8 + 8 + 4+8;
193 msg_size += 2 + 4 + 8 + 4 + 4; /* oloc */
194 msg_size += 1 + 8 + 4 + 4; /* pg_t */
195 msg_size += 4 + MAX_OBJ_NAME_SIZE;
196 msg_size += 2 + num_ops*sizeof(struct ceph_osd_op);
197 msg_size += 8; /* snapid */
198 msg_size += 8; /* snap_seq */
199 msg_size += 8 * (snapc ? snapc->num_snaps : 0); /* snaps */
200 msg_size += 4;
f24e9980
SW		 201
202 if (use_mempool) {
3499e8a5 203 req = mempool_alloc(osdc->req_mempool, gfp_flags);
f24e9980
SW		 204 memset(req, 0, sizeof(*req));
205 } else {
3499e8a5 206 req = kzalloc(sizeof(*req), gfp_flags);
f24e9980
SW		 207 }
208 if (req == NULL)
a79832f2 209 return NULL;
f24e9980 210
f24e9980
SW		 211 req->r_osdc = osdc;
212 req->r_mempool = use_mempool;
79528734 213 req->r_num_ops = num_ops;
68b4476b 214
415e49a9 215 kref_init(&req->r_kref);
f24e9980
SW		 216 init_completion(&req->r_completion);
217 init_completion(&req->r_safe_completion);
a978fa20 218 RB_CLEAR_NODE(&req->r_node);
f24e9980 219 INIT_LIST_HEAD(&req->r_unsafe_item);
a40c4f10
YS		 220 INIT_LIST_HEAD(&req->r_linger_item);
221 INIT_LIST_HEAD(&req->r_linger_osd);
935b639a 222 INIT_LIST_HEAD(&req->r_req_lru_item);
cd43045c
SW		 223 INIT_LIST_HEAD(&req->r_osd_item);
224
c16e7869
SW		 225 /* create reply message */
226 if (use_mempool)
227 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0);
228 else
229 msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY,
b61c2763 230 OSD_OPREPLY_FRONT_LEN, gfp_flags, true);
a79832f2 231 if (!msg) {
c16e7869 232 ceph_osdc_put_request(req);
a79832f2 233 return NULL;
c16e7869
SW		 234 }
235 req->r_reply = msg;
236
c54d47bf
AE		 237 ceph_osd_data_init(&req->r_data_in);
238 ceph_osd_data_init(&req->r_data_out);
d50b409f 239
c16e7869 240 /* create request message; allow space for oid */
f24e9980 241 if (use_mempool)
8f3bc053 242 msg = ceph_msgpool_get(&osdc->msgpool_op, 0);
f24e9980 243 else
b61c2763 244 msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp_flags, true);
a79832f2 245 if (!msg) {
f24e9980 246 ceph_osdc_put_request(req);
a79832f2 247 return NULL;
f24e9980 248 }
68b4476b 249
f24e9980 250 memset(msg->front.iov_base, 0, msg->front.iov_len);
3499e8a5
YS		 251
252 req->r_request = msg;
3499e8a5
YS		 253
254 return req;
255}
3d14c5d2 256EXPORT_SYMBOL(ceph_osdc_alloc_request);
3499e8a5 257
a8dd0a37 258static bool osd_req_opcode_valid(u16 opcode)
68b4476b 259{
a8dd0a37 260 switch (opcode) {
68b4476b 261 case CEPH_OSD_OP_READ:
a8dd0a37 262 case CEPH_OSD_OP_STAT:
4c46459c
AE		 263 case CEPH_OSD_OP_MAPEXT:
264 case CEPH_OSD_OP_MASKTRUNC:
265 case CEPH_OSD_OP_SPARSE_READ:
a9f36c3e 266 case CEPH_OSD_OP_NOTIFY:
a8dd0a37 267 case CEPH_OSD_OP_NOTIFY_ACK:
4c46459c 268 case CEPH_OSD_OP_ASSERT_VER:
a8dd0a37 269 case CEPH_OSD_OP_WRITE:
4c46459c
AE		 270 case CEPH_OSD_OP_WRITEFULL:
271 case CEPH_OSD_OP_TRUNCATE:
272 case CEPH_OSD_OP_ZERO:
273 case CEPH_OSD_OP_DELETE:
274 case CEPH_OSD_OP_APPEND:
a8dd0a37 275 case CEPH_OSD_OP_STARTSYNC:
4c46459c
AE		 276 case CEPH_OSD_OP_SETTRUNC:
277 case CEPH_OSD_OP_TRIMTRUNC:
278 case CEPH_OSD_OP_TMAPUP:
279 case CEPH_OSD_OP_TMAPPUT:
280 case CEPH_OSD_OP_TMAPGET:
281 case CEPH_OSD_OP_CREATE:
a9f36c3e 282 case CEPH_OSD_OP_ROLLBACK:
a8dd0a37 283 case CEPH_OSD_OP_WATCH:
4c46459c
AE		 284 case CEPH_OSD_OP_OMAPGETKEYS:
285 case CEPH_OSD_OP_OMAPGETVALS:
286 case CEPH_OSD_OP_OMAPGETHEADER:
287 case CEPH_OSD_OP_OMAPGETVALSBYKEYS:
4c46459c
AE		 288 case CEPH_OSD_OP_OMAPSETVALS:
289 case CEPH_OSD_OP_OMAPSETHEADER:
290 case CEPH_OSD_OP_OMAPCLEAR:
291 case CEPH_OSD_OP_OMAPRMKEYS:
292 case CEPH_OSD_OP_OMAP_CMP:
293 case CEPH_OSD_OP_CLONERANGE:
294 case CEPH_OSD_OP_ASSERT_SRC_VERSION:
295 case CEPH_OSD_OP_SRC_CMPXATTR:
a9f36c3e 296 case CEPH_OSD_OP_GETXATTR:
4c46459c 297 case CEPH_OSD_OP_GETXATTRS:
a9f36c3e
AE		 298 case CEPH_OSD_OP_CMPXATTR:
299 case CEPH_OSD_OP_SETXATTR:
4c46459c
AE		 300 case CEPH_OSD_OP_SETXATTRS:
301 case CEPH_OSD_OP_RESETXATTRS:
302 case CEPH_OSD_OP_RMXATTR:
303 case CEPH_OSD_OP_PULL:
304 case CEPH_OSD_OP_PUSH:
305 case CEPH_OSD_OP_BALANCEREADS:
306 case CEPH_OSD_OP_UNBALANCEREADS:
307 case CEPH_OSD_OP_SCRUB:
308 case CEPH_OSD_OP_SCRUB_RESERVE:
309 case CEPH_OSD_OP_SCRUB_UNRESERVE:
310 case CEPH_OSD_OP_SCRUB_STOP:
311 case CEPH_OSD_OP_SCRUB_MAP:
312 case CEPH_OSD_OP_WRLOCK:
313 case CEPH_OSD_OP_WRUNLOCK:
314 case CEPH_OSD_OP_RDLOCK:
315 case CEPH_OSD_OP_RDUNLOCK:
316 case CEPH_OSD_OP_UPLOCK:
317 case CEPH_OSD_OP_DNLOCK:
a8dd0a37 318 case CEPH_OSD_OP_CALL:
4c46459c
AE		 319 case CEPH_OSD_OP_PGLS:
320 case CEPH_OSD_OP_PGLS_FILTER:
a8dd0a37
AE		 321 return true;
322 default:
323 return false;
324 }
325}
326
33803f33
AE		 327/*
328 * This is an osd op init function for opcodes that have no data or
329 * other information associated with them. It also serves as a
330 * common init routine for all the other init functions, below.
331 */
332void osd_req_op_init(struct ceph_osd_req_op *op, u16 opcode)
333{
334 BUG_ON(!osd_req_opcode_valid(opcode));
335
336 memset(op, 0, sizeof (*op));
337
338 op->op = opcode;
339}
340
341void osd_req_op_extent_init(struct ceph_osd_req_op *op, u16 opcode,
342 u64 offset, u64 length,
343 u64 truncate_size, u32 truncate_seq)
344{
345 size_t payload_len = 0;
346
347 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE);
348
349 osd_req_op_init(op, opcode);
350
351 op->extent.offset = offset;
352 op->extent.length = length;
353 op->extent.truncate_size = truncate_size;
354 op->extent.truncate_seq = truncate_seq;
355 if (opcode == CEPH_OSD_OP_WRITE)
356 payload_len += length;
357
358 op->payload_len = payload_len;
359}
360EXPORT_SYMBOL(osd_req_op_extent_init);
361
e5975c7c
AE		 362void osd_req_op_extent_update(struct ceph_osd_req_op *op, u64 length)
363{
364 u64 previous = op->extent.length;
365
366 if (length == previous)
367 return; /* Nothing to do */
368 BUG_ON(length > previous);
369
370 op->extent.length = length;
371 op->payload_len -= previous - length;
372}
373EXPORT_SYMBOL(osd_req_op_extent_update);
374
8c042b0d
AE		 375void osd_req_op_extent_osd_data(struct ceph_osd_req_op *op,
376 struct ceph_osd_data *osd_data)
377{
378 op->extent.osd_data = osd_data;
379}
380EXPORT_SYMBOL(osd_req_op_extent_osd_data);
381
33803f33
AE		 382void osd_req_op_cls_init(struct ceph_osd_req_op *op, u16 opcode,
383 const char *class, const char *method,
384 const void *request_data, size_t request_data_size)
385{
386 size_t payload_len = 0;
387 size_t size;
388
389 BUG_ON(opcode != CEPH_OSD_OP_CALL);
390
391 osd_req_op_init(op, opcode);
392
393 op->cls.class_name = class;
394 size = strlen(class);
395 BUG_ON(size > (size_t) U8_MAX);
396 op->cls.class_len = size;
397 payload_len += size;
398
399 op->cls.method_name = method;
400 size = strlen(method);
401 BUG_ON(size > (size_t) U8_MAX);
402 op->cls.method_len = size;
403 payload_len += size;
404
54d50649 405 op->cls.request_data = request_data;
33803f33 406 BUG_ON(request_data_size > (size_t) U32_MAX);
54d50649 407 op->cls.request_data_len = (u32) request_data_size;
33803f33
AE		 408 payload_len += request_data_size;
409
410 op->cls.argc = 0; /* currently unused */
411
412 op->payload_len = payload_len;
413}
414EXPORT_SYMBOL(osd_req_op_cls_init);
415
8c042b0d
AE		 416void osd_req_op_cls_response_data(struct ceph_osd_req_op *op,
417 struct ceph_osd_data *response_data)
418{
419 op->cls.response_data = response_data;
420}
421EXPORT_SYMBOL(osd_req_op_cls_response_data);
422
33803f33
AE		 423void osd_req_op_watch_init(struct ceph_osd_req_op *op, u16 opcode,
424 u64 cookie, u64 version, int flag)
425{
426 BUG_ON(opcode != CEPH_OSD_OP_NOTIFY_ACK && opcode != CEPH_OSD_OP_WATCH);
427
428 osd_req_op_init(op, opcode);
429
430 op->watch.cookie = cookie;
431 /* op->watch.ver = version; */ /* XXX 3847 */
432 op->watch.ver = cpu_to_le64(version);
433 if (opcode == CEPH_OSD_OP_WATCH && flag)
434 op->watch.flag = (u8) 1;
435}
436EXPORT_SYMBOL(osd_req_op_watch_init);
437
a8dd0a37 438static u64 osd_req_encode_op(struct ceph_osd_request *req,
79528734 439 struct ceph_osd_op *dst, unsigned int which)
a8dd0a37 440{
79528734 441 struct ceph_osd_req_op *src;
54d50649 442 u64 request_data_len = 0;
a8dd0a37
AE		 443 struct ceph_pagelist *pagelist;
444
79528734
AE		 445 BUG_ON(which >= req->r_num_ops);
446 src = &req->r_ops[which];
a8dd0a37
AE		 447 if (WARN_ON(!osd_req_opcode_valid(src->op))) {
448 pr_err("unrecognized osd opcode %d\n", src->op);
449
450 return 0;
451 }
452
453 switch (src->op) {
454 case CEPH_OSD_OP_STAT:
455 break;
456 case CEPH_OSD_OP_READ:
457 case CEPH_OSD_OP_WRITE:
458 if (src->op == CEPH_OSD_OP_WRITE)
54d50649 459 request_data_len = src->extent.length;
a8dd0a37
AE		 460 dst->extent.offset = cpu_to_le64(src->extent.offset);
461 dst->extent.length = cpu_to_le64(src->extent.length);
462 dst->extent.truncate_size =
463 cpu_to_le64(src->extent.truncate_size);
464 dst->extent.truncate_seq =
465 cpu_to_le32(src->extent.truncate_seq);
8c042b0d
AE		 466 if (src->op == CEPH_OSD_OP_WRITE)
467 WARN_ON(src->extent.osd_data != &req->r_data_out);
468 else
469 WARN_ON(src->extent.osd_data != &req->r_data_in);
a8dd0a37
AE		 470 break;
471 case CEPH_OSD_OP_CALL:
472 pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
473 BUG_ON(!pagelist);
474 ceph_pagelist_init(pagelist);
475
476 dst->cls.class_len = src->cls.class_len;
477 dst->cls.method_len = src->cls.method_len;
54d50649 478 dst->cls.indata_len = cpu_to_le32(src->cls.request_data_len);
a8dd0a37
AE		 479 ceph_pagelist_append(pagelist, src->cls.class_name,
480 src->cls.class_len);
481 ceph_pagelist_append(pagelist, src->cls.method_name,
482 src->cls.method_len);
54d50649
AE		 483 ceph_pagelist_append(pagelist, src->cls.request_data,
484 src->cls.request_data_len);
43bfe5de 485 ceph_osd_data_pagelist_init(&req->r_data_out, pagelist);
8c042b0d
AE		 486
487 WARN_ON(src->cls.response_data != &req->r_data_in);
54d50649 488 request_data_len = pagelist->length;
a8dd0a37
AE		 489 break;
490 case CEPH_OSD_OP_STARTSYNC:
491 break;
492 case CEPH_OSD_OP_NOTIFY_ACK:
493 case CEPH_OSD_OP_WATCH:
494 dst->watch.cookie = cpu_to_le64(src->watch.cookie);
495 dst->watch.ver = cpu_to_le64(src->watch.ver);
496 dst->watch.flag = src->watch.flag;
497 break;
498 default:
4c46459c 499 pr_err("unsupported osd opcode %s\n",
8f63ca2d 500 ceph_osd_op_name(src->op));
4c46459c 501 WARN_ON(1);
a8dd0a37
AE		 502
503 return 0;
68b4476b 504 }
a8dd0a37 505 dst->op = cpu_to_le16(src->op);
68b4476b 506 dst->payload_len = cpu_to_le32(src->payload_len);
175face2 507
54d50649 508 return request_data_len;
68b4476b
YS		 509}
510
3499e8a5
YS		 511/*
512 * build new request AND message
513 *
514 */
79528734
AE		 515void ceph_osdc_build_request(struct ceph_osd_request *req, u64 off,
516 struct ceph_snap_context *snapc, u64 snap_id,
517 struct timespec *mtime)
3499e8a5
YS		 518{
519 struct ceph_msg *msg = req->r_request;
3499e8a5 520 void *p;
1b83bef2 521 size_t msg_size;
3499e8a5 522 int flags = req->r_flags;
f44246e3 523 u64 data_len;
79528734 524 unsigned int i;
3499e8a5 525
1b83bef2 526 req->r_snapid = snap_id;
f24e9980
SW		 527 req->r_snapc = ceph_get_snap_context(snapc);
528
1b83bef2
SW		 529 /* encode request */
530 msg->hdr.version = cpu_to_le16(4);
531
532 p = msg->front.iov_base;
533 ceph_encode_32(&p, 1); /* client_inc is always 1 */
534 req->r_request_osdmap_epoch = p;
535 p += 4;
536 req->r_request_flags = p;
537 p += 4;
538 if (req->r_flags & CEPH_OSD_FLAG_WRITE)
539 ceph_encode_timespec(p, mtime);
540 p += sizeof(struct ceph_timespec);
541 req->r_request_reassert_version = p;
542 p += sizeof(struct ceph_eversion); /* will get filled in */
543
544 /* oloc */
545 ceph_encode_8(&p, 4);
546 ceph_encode_8(&p, 4);
547 ceph_encode_32(&p, 8 + 4 + 4);
548 req->r_request_pool = p;
549 p += 8;
550 ceph_encode_32(&p, -1); /* preferred */
551 ceph_encode_32(&p, 0); /* key len */
552
553 ceph_encode_8(&p, 1);
554 req->r_request_pgid = p;
555 p += 8 + 4;
556 ceph_encode_32(&p, -1); /* preferred */
557
558 /* oid */
559 ceph_encode_32(&p, req->r_oid_len);
af77f26c 560 memcpy(p, req->r_oid, req->r_oid_len);
1b83bef2 561 dout("oid '%.*s' len %d\n", req->r_oid_len, req->r_oid, req->r_oid_len);
af77f26c 562 p += req->r_oid_len;
f24e9980 563
175face2 564 /* ops--can imply data */
79528734 565 ceph_encode_16(&p, (u16)req->r_num_ops);
175face2 566 data_len = 0;
79528734
AE		 567 for (i = 0; i < req->r_num_ops; i++) {
568 data_len += osd_req_encode_op(req, p, i);
1b83bef2
SW		 569 p += sizeof(struct ceph_osd_op);
570 }
68b4476b 571
1b83bef2
SW		 572 /* snaps */
573 ceph_encode_64(&p, req->r_snapid);
574 ceph_encode_64(&p, req->r_snapc ? req->r_snapc->seq : 0);
575 ceph_encode_32(&p, req->r_snapc ? req->r_snapc->num_snaps : 0);
576 if (req->r_snapc) {
f24e9980 577 for (i = 0; i < snapc->num_snaps; i++) {
1b83bef2 578 ceph_encode_64(&p, req->r_snapc->snaps[i]);
f24e9980
SW		 579 }
580 }
581
1b83bef2
SW		 582 req->r_request_attempts = p;
583 p += 4;
584
175face2 585 /* data */
0baa1bd9
AE		 586 if (flags & CEPH_OSD_FLAG_WRITE) {
587 u16 data_off;
588
589 /*
590 * The header "data_off" is a hint to the receiver
591 * allowing it to align received data into its
592 * buffers such that there's no need to re-copy
593 * it before writing it to disk (direct I/O).
594 */
595 data_off = (u16) (off & 0xffff);
596 req->r_request->hdr.data_off = cpu_to_le16(data_off);
597 }
f44246e3 598 req->r_request->hdr.data_len = cpu_to_le32(data_len);
c5c6b19d 599
f24e9980 600 BUG_ON(p > msg->front.iov_base + msg->front.iov_len);
6f863e71
SW		 601 msg_size = p - msg->front.iov_base;
602 msg->front.iov_len = msg_size;
603 msg->hdr.front_len = cpu_to_le32(msg_size);
1b83bef2 604
acead002 605 dout("build_request msg_size was %d\n", (int)msg_size);
3499e8a5 606}
3d14c5d2 607EXPORT_SYMBOL(ceph_osdc_build_request);
3499e8a5
YS		 608
609/*
610 * build new request AND message, calculate layout, and adjust file
611 * extent as needed.
612 *
613 * if the file was recently truncated, we include information about its
614 * old and new size so that the object can be updated appropriately. (we
615 * avoid synchronously deleting truncated objects because it's slow.)
616 *
617 * if @do_sync, include a 'startsync' command so that the osd will flush
618 * data quickly.
619 */
620struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
621 struct ceph_file_layout *layout,
622 struct ceph_vino vino,
acead002 623 u64 off, u64 *plen, int num_ops,
3499e8a5
YS		 624 int opcode, int flags,
625 struct ceph_snap_context *snapc,
3499e8a5
YS		 626 u32 truncate_seq,
627 u64 truncate_size,
153e5167 628 bool use_mempool)
3499e8a5 629{
68b4476b 630 struct ceph_osd_request *req;
8c042b0d 631 struct ceph_osd_data *osd_data;
79528734 632 struct ceph_osd_req_op *op;
75d1c941
AE		 633 u64 objnum = 0;
634 u64 objoff = 0;
635 u64 objlen = 0;
d18d1e28
AE		 636 u32 object_size;
637 u64 object_base;
6816282d 638 int r;
68b4476b 639
d18d1e28 640 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE);
68b4476b 641
acead002 642 req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool,
ae7ca4a3 643 GFP_NOFS);
4ad12621 644 if (!req)
6816282d 645 return ERR_PTR(-ENOMEM);
8c042b0d
AE		 646 osd_data = opcode == CEPH_OSD_OP_WRITE ? &req->r_data_out
647 : &req->r_data_in;
79528734 648
d178a9e7 649 req->r_flags = flags;
3499e8a5
YS		 650
651 /* calculate max write size */
a19dadfb 652 r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
3ff5f385
AE		 653 if (r < 0) {
654 ceph_osdc_put_request(req);
6816282d 655 return ERR_PTR(r);
3ff5f385 656 }
a19dadfb 657
d18d1e28
AE		 658 object_size = le32_to_cpu(layout->fl_object_size);
659 object_base = off - objoff;
660 if (truncate_size <= object_base) {
661 truncate_size = 0;
662 } else {
663 truncate_size -= object_base;
664 if (truncate_size > object_size)
665 truncate_size = object_size;
a19dadfb 666 }
d18d1e28 667
79528734
AE		 668 op = &req->r_ops[0];
669 osd_req_op_extent_init(op, opcode, objoff, objlen,
b0270324 670 truncate_size, truncate_seq);
8c042b0d
AE		 671 osd_req_op_extent_osd_data(op, osd_data);
672
acead002
AE		 673 /*
674 * A second op in the ops array means the caller wants to
675 * also issue a include a 'startsync' command so that the
676 * osd will flush data quickly.
677 */
678 if (num_ops > 1)
79528734 679 osd_req_op_init(++op, CEPH_OSD_OP_STARTSYNC);
d18d1e28 680
3499e8a5
YS		 681 req->r_file_layout = *layout; /* keep a copy */
682
75d1c941
AE		 683 snprintf(req->r_oid, sizeof(req->r_oid), "%llx.%08llx",
684 vino.ino, objnum);
dbe0fc41
AE		 685 req->r_oid_len = strlen(req->r_oid);
686
f24e9980
SW		 687 return req;
688}
3d14c5d2 689EXPORT_SYMBOL(ceph_osdc_new_request);
f24e9980
SW		 690
691/*
692 * We keep osd requests in an rbtree, sorted by ->r_tid.
693 */
694static void __insert_request(struct ceph_osd_client *osdc,
695 struct ceph_osd_request *new)
696{
697 struct rb_node **p = &osdc->requests.rb_node;
698 struct rb_node *parent = NULL;
699 struct ceph_osd_request *req = NULL;
700
701 while (*p) {
702 parent = *p;
703 req = rb_entry(parent, struct ceph_osd_request, r_node);
704 if (new->r_tid < req->r_tid)
705 p = &(*p)->rb_left;
706 else if (new->r_tid > req->r_tid)
707 p = &(*p)->rb_right;
708 else
709 BUG();
710 }
711
712 rb_link_node(&new->r_node, parent, p);
713 rb_insert_color(&new->r_node, &osdc->requests);
714}
715
716static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc,
717 u64 tid)
718{
719 struct ceph_osd_request *req;
720 struct rb_node *n = osdc->requests.rb_node;
721
722 while (n) {
723 req = rb_entry(n, struct ceph_osd_request, r_node);
724 if (tid < req->r_tid)
725 n = n->rb_left;
726 else if (tid > req->r_tid)
727 n = n->rb_right;
728 else
729 return req;
730 }
731 return NULL;
732}
733
734static struct ceph_osd_request *
735__lookup_request_ge(struct ceph_osd_client *osdc,
736 u64 tid)
737{
738 struct ceph_osd_request *req;
739 struct rb_node *n = osdc->requests.rb_node;
740
741 while (n) {
742 req = rb_entry(n, struct ceph_osd_request, r_node);
743 if (tid < req->r_tid) {
744 if (!n->rb_left)
745 return req;
746 n = n->rb_left;
747 } else if (tid > req->r_tid) {
748 n = n->rb_right;
749 } else {
750 return req;
751 }
752 }
753 return NULL;
754}
755
6f6c7006
SW		 756/*
757 * Resubmit requests pending on the given osd.
758 */
759static void __kick_osd_requests(struct ceph_osd_client *osdc,
760 struct ceph_osd *osd)
761{
a40c4f10 762 struct ceph_osd_request *req, *nreq;
e02493c0 763 LIST_HEAD(resend);
6f6c7006
SW		 764 int err;
765
766 dout("__kick_osd_requests osd%d\n", osd->o_osd);
767 err = __reset_osd(osdc, osd);
685a7555 768 if (err)
6f6c7006 769 return;
e02493c0
AE		 770 /*
771 * Build up a list of requests to resend by traversing the
772 * osd's list of requests. Requests for a given object are
773 * sent in tid order, and that is also the order they're
774 * kept on this list. Therefore all requests that are in
775 * flight will be found first, followed by all requests that
776 * have not yet been sent. And to resend requests while
777 * preserving this order we will want to put any sent
778 * requests back on the front of the osd client's unsent
779 * list.
780 *
781 * So we build a separate ordered list of already-sent
782 * requests for the affected osd and splice it onto the
783 * front of the osd client's unsent list. Once we've seen a
784 * request that has not yet been sent we're done. Those
785 * requests are already sitting right where they belong.
786 */
6f6c7006 787 list_for_each_entry(req, &osd->o_requests, r_osd_item) {
e02493c0
AE		 788 if (!req->r_sent)
789 break;
790 list_move_tail(&req->r_req_lru_item, &resend);
791 dout("requeueing %p tid %llu osd%d\n", req, req->r_tid,
6f6c7006 792 osd->o_osd);
a40c4f10
YS		 793 if (!req->r_linger)
794 req->r_flags |= CEPH_OSD_FLAG_RETRY;
795 }
e02493c0 796 list_splice(&resend, &osdc->req_unsent);
a40c4f10 797
e02493c0
AE		 798 /*
799 * Linger requests are re-registered before sending, which
800 * sets up a new tid for each. We add them to the unsent
801 * list at the end to keep things in tid order.
802 */
a40c4f10
YS		 803 list_for_each_entry_safe(req, nreq, &osd->o_linger_requests,
804 r_linger_osd) {
77f38e0e
SW		 805 /*
806 * reregister request prior to unregistering linger so
807 * that r_osd is preserved.
808 */
809 BUG_ON(!list_empty(&req->r_req_lru_item));
a40c4f10 810 __register_request(osdc, req);
e02493c0 811 list_add_tail(&req->r_req_lru_item, &osdc->req_unsent);
ad885927 812 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
77f38e0e 813 __unregister_linger_request(osdc, req);
a40c4f10
YS		 814 dout("requeued lingering %p tid %llu osd%d\n", req, req->r_tid,
815 osd->o_osd);
6f6c7006
SW		 816 }
817}
818
f24e9980 819/*
81b024e7 820 * If the osd connection drops, we need to resubmit all requests.
f24e9980
SW		 821 */
822static void osd_reset(struct ceph_connection *con)
823{
824 struct ceph_osd *osd = con->private;
825 struct ceph_osd_client *osdc;
826
827 if (!osd)
828 return;
829 dout("osd_reset osd%d\n", osd->o_osd);
830 osdc = osd->o_osdc;
f24e9980 831 down_read(&osdc->map_sem);
83aff95e
SW		 832 mutex_lock(&osdc->request_mutex);
833 __kick_osd_requests(osdc, osd);
f9d25199 834 __send_queued(osdc);
83aff95e 835 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 836 up_read(&osdc->map_sem);
837}
838
839/*
840 * Track open sessions with osds.
841 */
e10006f8 842static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
f24e9980
SW		 843{
844 struct ceph_osd *osd;
845
846 osd = kzalloc(sizeof(*osd), GFP_NOFS);
847 if (!osd)
848 return NULL;
849
850 atomic_set(&osd->o_ref, 1);
851 osd->o_osdc = osdc;
e10006f8 852 osd->o_osd = onum;
f407731d 853 RB_CLEAR_NODE(&osd->o_node);
f24e9980 854 INIT_LIST_HEAD(&osd->o_requests);
a40c4f10 855 INIT_LIST_HEAD(&osd->o_linger_requests);
f5a2041b 856 INIT_LIST_HEAD(&osd->o_osd_lru);
f24e9980
SW		 857 osd->o_incarnation = 1;
858
b7a9e5dd 859 ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
4e7a5dcd 860
422d2cb8 861 INIT_LIST_HEAD(&osd->o_keepalive_item);
f24e9980
SW		 862 return osd;
863}
864
865static struct ceph_osd *get_osd(struct ceph_osd *osd)
866{
867 if (atomic_inc_not_zero(&osd->o_ref)) {
868 dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
869 atomic_read(&osd->o_ref));
870 return osd;
871 } else {
872 dout("get_osd %p FAIL\n", osd);
873 return NULL;
874 }
875}
876
877static void put_osd(struct ceph_osd *osd)
878{
879 dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
880 atomic_read(&osd->o_ref) - 1);
a255651d 881 if (atomic_dec_and_test(&osd->o_ref) && osd->o_auth.authorizer) {
79494d1b
SW		 882 struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth;
883
27859f97 884 ceph_auth_destroy_authorizer(ac, osd->o_auth.authorizer);
f24e9980 885 kfree(osd);
79494d1b 886 }
f24e9980
SW		 887}
888
889/*
890 * remove an osd from our map
891 */
f5a2041b 892static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
f24e9980 893{
f5a2041b 894 dout("__remove_osd %p\n", osd);
f24e9980
SW		 895 BUG_ON(!list_empty(&osd->o_requests));
896 rb_erase(&osd->o_node, &osdc->osds);
f5a2041b 897 list_del_init(&osd->o_osd_lru);
f24e9980
SW		 898 ceph_con_close(&osd->o_con);
899 put_osd(osd);
900}
901
aca420bc
SW		 902static void remove_all_osds(struct ceph_osd_client *osdc)
903{
048a9d2d 904 dout("%s %p\n", __func__, osdc);
aca420bc
SW		 905 mutex_lock(&osdc->request_mutex);
906 while (!RB_EMPTY_ROOT(&osdc->osds)) {
907 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
908 struct ceph_osd, o_node);
909 __remove_osd(osdc, osd);
910 }
911 mutex_unlock(&osdc->request_mutex);
912}
913
f5a2041b
YS		 914static void __move_osd_to_lru(struct ceph_osd_client *osdc,
915 struct ceph_osd *osd)
916{
917 dout("__move_osd_to_lru %p\n", osd);
918 BUG_ON(!list_empty(&osd->o_osd_lru));
919 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
3d14c5d2 920 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl * HZ;
f5a2041b
YS		 921}
922
923static void __remove_osd_from_lru(struct ceph_osd *osd)
924{
925 dout("__remove_osd_from_lru %p\n", osd);
926 if (!list_empty(&osd->o_osd_lru))
927 list_del_init(&osd->o_osd_lru);
928}
929
aca420bc 930static void remove_old_osds(struct ceph_osd_client *osdc)
f5a2041b
YS		 931{
932 struct ceph_osd *osd, *nosd;
933
934 dout("__remove_old_osds %p\n", osdc);
935 mutex_lock(&osdc->request_mutex);
936 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
aca420bc 937 if (time_before(jiffies, osd->lru_ttl))
f5a2041b
YS		 938 break;
939 __remove_osd(osdc, osd);
940 }
941 mutex_unlock(&osdc->request_mutex);
942}
943
f24e9980
SW		 944/*
945 * reset osd connect
946 */
f5a2041b 947static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
f24e9980 948{
c3acb181 949 struct ceph_entity_addr *peer_addr;
f24e9980 950
f5a2041b 951 dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
a40c4f10
YS		 952 if (list_empty(&osd->o_requests) &&
953 list_empty(&osd->o_linger_requests)) {
f5a2041b 954 __remove_osd(osdc, osd);
c3acb181
AE		 955
956 return -ENODEV;
957 }
958
959 peer_addr = &osdc->osdmap->osd_addr[osd->o_osd];
960 if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
961 !ceph_con_opened(&osd->o_con)) {
962 struct ceph_osd_request *req;
963
87b315a5
SW		 964 dout(" osd addr hasn't changed and connection never opened,"
965 " letting msgr retry");
966 /* touch each r_stamp for handle_timeout()'s benfit */
967 list_for_each_entry(req, &osd->o_requests, r_osd_item)
968 req->r_stamp = jiffies;
c3acb181
AE		 969
970 return -EAGAIN;
f24e9980 971 }
c3acb181
AE		 972
973 ceph_con_close(&osd->o_con);
974 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
975 osd->o_incarnation++;
976
977 return 0;
f24e9980
SW		 978}
979
980static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new)
981{
982 struct rb_node **p = &osdc->osds.rb_node;
983 struct rb_node *parent = NULL;
984 struct ceph_osd *osd = NULL;
985
aca420bc 986 dout("__insert_osd %p osd%d\n", new, new->o_osd);
f24e9980
SW		 987 while (*p) {
988 parent = *p;
989 osd = rb_entry(parent, struct ceph_osd, o_node);
990 if (new->o_osd < osd->o_osd)
991 p = &(*p)->rb_left;
992 else if (new->o_osd > osd->o_osd)
993 p = &(*p)->rb_right;
994 else
995 BUG();
996 }
997
998 rb_link_node(&new->o_node, parent, p);
999 rb_insert_color(&new->o_node, &osdc->osds);
1000}
1001
1002static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o)
1003{
1004 struct ceph_osd *osd;
1005 struct rb_node *n = osdc->osds.rb_node;
1006
1007 while (n) {
1008 osd = rb_entry(n, struct ceph_osd, o_node);
1009 if (o < osd->o_osd)
1010 n = n->rb_left;
1011 else if (o > osd->o_osd)
1012 n = n->rb_right;
1013 else
1014 return osd;
1015 }
1016 return NULL;
1017}
1018
422d2cb8
YS		 1019static void __schedule_osd_timeout(struct ceph_osd_client *osdc)
1020{
1021 schedule_delayed_work(&osdc->timeout_work,
3d14c5d2 1022 osdc->client->options->osd_keepalive_timeout * HZ);
422d2cb8
YS		 1023}
1024
1025static void __cancel_osd_timeout(struct ceph_osd_client *osdc)
1026{
1027 cancel_delayed_work(&osdc->timeout_work);
1028}
f24e9980
SW		 1029
1030/*
1031 * Register request, assign tid. If this is the first request, set up
1032 * the timeout event.
1033 */
a40c4f10
YS		 1034static void __register_request(struct ceph_osd_client *osdc,
1035 struct ceph_osd_request *req)
f24e9980 1036{
f24e9980 1037 req->r_tid = ++osdc->last_tid;
6df058c0 1038 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
77f38e0e 1039 dout("__register_request %p tid %lld\n", req, req->r_tid);
f24e9980
SW		 1040 __insert_request(osdc, req);
1041 ceph_osdc_get_request(req);
1042 osdc->num_requests++;
f24e9980 1043 if (osdc->num_requests == 1) {
422d2cb8
YS		 1044 dout(" first request, scheduling timeout\n");
1045 __schedule_osd_timeout(osdc);
f24e9980 1046 }
a40c4f10
YS		 1047}
1048
f24e9980
SW		 1049/*
1050 * called under osdc->request_mutex
1051 */
1052static void __unregister_request(struct ceph_osd_client *osdc,
1053 struct ceph_osd_request *req)
1054{
35f9f8a0
SW		 1055 if (RB_EMPTY_NODE(&req->r_node)) {
1056 dout("__unregister_request %p tid %lld not registered\n",
1057 req, req->r_tid);
1058 return;
1059 }
1060
f24e9980
SW		 1061 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
1062 rb_erase(&req->r_node, &osdc->requests);
1063 osdc->num_requests--;
1064
0ba6478d
SW		 1065 if (req->r_osd) {
1066 /* make sure the original request isn't in flight. */
6740a845 1067 ceph_msg_revoke(req->r_request);
0ba6478d
SW		 1068
1069 list_del_init(&req->r_osd_item);
a40c4f10
YS		 1070 if (list_empty(&req->r_osd->o_requests) &&
1071 list_empty(&req->r_osd->o_linger_requests)) {
1072 dout("moving osd to %p lru\n", req->r_osd);
f5a2041b 1073 __move_osd_to_lru(osdc, req->r_osd);
a40c4f10 1074 }
fbdb9190 1075 if (list_empty(&req->r_linger_item))
a40c4f10 1076 req->r_osd = NULL;
0ba6478d 1077 }
f24e9980 1078
7d5f2481 1079 list_del_init(&req->r_req_lru_item);
f24e9980
SW		 1080 ceph_osdc_put_request(req);
1081
422d2cb8
YS		 1082 if (osdc->num_requests == 0) {
1083 dout(" no requests, canceling timeout\n");
1084 __cancel_osd_timeout(osdc);
f24e9980
SW		 1085 }
1086}
1087
1088/*
1089 * Cancel a previously queued request message
1090 */
1091static void __cancel_request(struct ceph_osd_request *req)
1092{
6bc18876 1093 if (req->r_sent && req->r_osd) {
6740a845 1094 ceph_msg_revoke(req->r_request);
f24e9980
SW		 1095 req->r_sent = 0;
1096 }
1097}
1098
a40c4f10
YS		 1099static void __register_linger_request(struct ceph_osd_client *osdc,
1100 struct ceph_osd_request *req)
1101{
1102 dout("__register_linger_request %p\n", req);
1103 list_add_tail(&req->r_linger_item, &osdc->req_linger);
6194ea89
SW		 1104 if (req->r_osd)
1105 list_add_tail(&req->r_linger_osd,
1106 &req->r_osd->o_linger_requests);
a40c4f10
YS		 1107}
1108
1109static void __unregister_linger_request(struct ceph_osd_client *osdc,
1110 struct ceph_osd_request *req)
1111{
1112 dout("__unregister_linger_request %p\n", req);
61c74035 1113 list_del_init(&req->r_linger_item);
a40c4f10 1114 if (req->r_osd) {
a40c4f10
YS		 1115 list_del_init(&req->r_linger_osd);
1116
1117 if (list_empty(&req->r_osd->o_requests) &&
1118 list_empty(&req->r_osd->o_linger_requests)) {
1119 dout("moving osd to %p lru\n", req->r_osd);
1120 __move_osd_to_lru(osdc, req->r_osd);
1121 }
fbdb9190
SW		 1122 if (list_empty(&req->r_osd_item))
1123 req->r_osd = NULL;
a40c4f10
YS		 1124 }
1125}
1126
1127void ceph_osdc_unregister_linger_request(struct ceph_osd_client *osdc,
1128 struct ceph_osd_request *req)
1129{
1130 mutex_lock(&osdc->request_mutex);
1131 if (req->r_linger) {
1132 __unregister_linger_request(osdc, req);
1133 ceph_osdc_put_request(req);
1134 }
1135 mutex_unlock(&osdc->request_mutex);
1136}
1137EXPORT_SYMBOL(ceph_osdc_unregister_linger_request);
1138
1139void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
1140 struct ceph_osd_request *req)
1141{
1142 if (!req->r_linger) {
1143 dout("set_request_linger %p\n", req);
1144 req->r_linger = 1;
1145 /*
1146 * caller is now responsible for calling
1147 * unregister_linger_request
1148 */
1149 ceph_osdc_get_request(req);
1150 }
1151}
1152EXPORT_SYMBOL(ceph_osdc_set_request_linger);
1153
f24e9980
SW		 1154/*
1155 * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
1156 * (as needed), and set the request r_osd appropriately. If there is
25985edc 1157 * no up osd, set r_osd to NULL. Move the request to the appropriate list
6f6c7006 1158 * (unsent, homeless) or leave on in-flight lru.
f24e9980
SW		 1159 *
1160 * Return 0 if unchanged, 1 if changed, or negative on error.
1161 *
1162 * Caller should hold map_sem for read and request_mutex.
1163 */
6f6c7006 1164static int __map_request(struct ceph_osd_client *osdc,
38d6453c 1165 struct ceph_osd_request *req, int force_resend)
f24e9980 1166{
5b191d99 1167 struct ceph_pg pgid;
d85b7056
SW		 1168 int acting[CEPH_PG_MAX_SIZE];
1169 int o = -1, num = 0;
f24e9980 1170 int err;
f24e9980 1171
6f6c7006 1172 dout("map_request %p tid %lld\n", req, req->r_tid);
41766f87
AE		 1173 err = ceph_calc_ceph_pg(&pgid, req->r_oid, osdc->osdmap,
1174 ceph_file_layout_pg_pool(req->r_file_layout));
6f6c7006
SW		 1175 if (err) {
1176 list_move(&req->r_req_lru_item, &osdc->req_notarget);
f24e9980 1177 return err;
6f6c7006 1178 }
7740a42f
SW		 1179 req->r_pgid = pgid;
1180
d85b7056
SW		 1181 err = ceph_calc_pg_acting(osdc->osdmap, pgid, acting);
1182 if (err > 0) {
1183 o = acting[0];
1184 num = err;
1185 }
f24e9980 1186
38d6453c
SW		 1187 if ((!force_resend &&
1188 req->r_osd && req->r_osd->o_osd == o &&
d85b7056
SW		 1189 req->r_sent >= req->r_osd->o_incarnation &&
1190 req->r_num_pg_osds == num &&
1191 memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
f24e9980
SW		 1192 (req->r_osd == NULL && o == -1))
1193 return 0; /* no change */
1194
5b191d99
SW		 1195 dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n",
1196 req->r_tid, pgid.pool, pgid.seed, o,
f24e9980
SW		 1197 req->r_osd ? req->r_osd->o_osd : -1);
1198
d85b7056
SW		 1199 /* record full pg acting set */
1200 memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
1201 req->r_num_pg_osds = num;
1202
f24e9980
SW		 1203 if (req->r_osd) {
1204 __cancel_request(req);
1205 list_del_init(&req->r_osd_item);
f24e9980
SW		 1206 req->r_osd = NULL;
1207 }
1208
1209 req->r_osd = __lookup_osd(osdc, o);
1210 if (!req->r_osd && o >= 0) {
c99eb1c7 1211 err = -ENOMEM;
e10006f8 1212 req->r_osd = create_osd(osdc, o);
6f6c7006
SW		 1213 if (!req->r_osd) {
1214 list_move(&req->r_req_lru_item, &osdc->req_notarget);
c99eb1c7 1215 goto out;
6f6c7006 1216 }
f24e9980 1217
6f6c7006 1218 dout("map_request osd %p is osd%d\n", req->r_osd, o);
f24e9980
SW		 1219 __insert_osd(osdc, req->r_osd);
1220
b7a9e5dd
SW		 1221 ceph_con_open(&req->r_osd->o_con,
1222 CEPH_ENTITY_TYPE_OSD, o,
1223 &osdc->osdmap->osd_addr[o]);
f24e9980
SW		 1224 }
1225
f5a2041b
YS		 1226 if (req->r_osd) {
1227 __remove_osd_from_lru(req->r_osd);
ad885927
AE		 1228 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
1229 list_move_tail(&req->r_req_lru_item, &osdc->req_unsent);
6f6c7006 1230 } else {
ad885927 1231 list_move_tail(&req->r_req_lru_item, &osdc->req_notarget);
f5a2041b 1232 }
d85b7056 1233 err = 1; /* osd or pg changed */
f24e9980
SW		 1234
1235out:
f24e9980
SW		 1236 return err;
1237}
1238
1239/*
1240 * caller should hold map_sem (for read) and request_mutex
1241 */
56e925b6
SW		 1242static void __send_request(struct ceph_osd_client *osdc,
1243 struct ceph_osd_request *req)
f24e9980 1244{
1b83bef2 1245 void *p;
f24e9980 1246
1b83bef2
SW		 1247 dout("send_request %p tid %llu to osd%d flags %d pg %lld.%x\n",
1248 req, req->r_tid, req->r_osd->o_osd, req->r_flags,
1249 (unsigned long long)req->r_pgid.pool, req->r_pgid.seed);
1250
1251 /* fill in message content that changes each time we send it */
1252 put_unaligned_le32(osdc->osdmap->epoch, req->r_request_osdmap_epoch);
1253 put_unaligned_le32(req->r_flags, req->r_request_flags);
1254 put_unaligned_le64(req->r_pgid.pool, req->r_request_pool);
1255 p = req->r_request_pgid;
1256 ceph_encode_64(&p, req->r_pgid.pool);
1257 ceph_encode_32(&p, req->r_pgid.seed);
1258 put_unaligned_le64(1, req->r_request_attempts); /* FIXME */
1259 memcpy(req->r_request_reassert_version, &req->r_reassert_version,
1260 sizeof(req->r_reassert_version));
2169aea6 1261
3dd72fc0 1262 req->r_stamp = jiffies;
07a27e22 1263 list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
f24e9980
SW		 1264
1265 ceph_msg_get(req->r_request); /* send consumes a ref */
1266 ceph_con_send(&req->r_osd->o_con, req->r_request);
1267 req->r_sent = req->r_osd->o_incarnation;
f24e9980
SW		 1268}
1269
6f6c7006
SW		 1270/*
1271 * Send any requests in the queue (req_unsent).
1272 */
f9d25199 1273static void __send_queued(struct ceph_osd_client *osdc)
6f6c7006
SW		 1274{
1275 struct ceph_osd_request *req, *tmp;
1276
f9d25199
AE		 1277 dout("__send_queued\n");
1278 list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item)
6f6c7006 1279 __send_request(osdc, req);
6f6c7006
SW		 1280}
1281
f24e9980
SW		 1282/*
1283 * Timeout callback, called every N seconds when 1 or more osd
1284 * requests has been active for more than N seconds. When this
1285 * happens, we ping all OSDs with requests who have timed out to
1286 * ensure any communications channel reset is detected. Reset the
1287 * request timeouts another N seconds in the future as we go.
1288 * Reschedule the timeout event another N seconds in future (unless
1289 * there are no open requests).
1290 */
1291static void handle_timeout(struct work_struct *work)
1292{
1293 struct ceph_osd_client *osdc =
1294 container_of(work, struct ceph_osd_client, timeout_work.work);
83aff95e 1295 struct ceph_osd_request *req;
f24e9980 1296 struct ceph_osd *osd;
422d2cb8 1297 unsigned long keepalive =
3d14c5d2 1298 osdc->client->options->osd_keepalive_timeout * HZ;
422d2cb8 1299 struct list_head slow_osds;
f24e9980
SW		 1300 dout("timeout\n");
1301 down_read(&osdc->map_sem);
1302
1303 ceph_monc_request_next_osdmap(&osdc->client->monc);
1304
1305 mutex_lock(&osdc->request_mutex);
f24e9980 1306
422d2cb8
YS		 1307 /*
1308 * ping osds that are a bit slow. this ensures that if there
1309 * is a break in the TCP connection we will notice, and reopen
1310 * a connection with that osd (from the fault callback).
1311 */
1312 INIT_LIST_HEAD(&slow_osds);
1313 list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
3dd72fc0 1314 if (time_before(jiffies, req->r_stamp + keepalive))
422d2cb8
YS		 1315 break;
1316
1317 osd = req->r_osd;
1318 BUG_ON(!osd);
1319 dout(" tid %llu is slow, will send keepalive on osd%d\n",
f24e9980 1320 req->r_tid, osd->o_osd);
422d2cb8
YS		 1321 list_move_tail(&osd->o_keepalive_item, &slow_osds);
1322 }
1323 while (!list_empty(&slow_osds)) {
1324 osd = list_entry(slow_osds.next, struct ceph_osd,
1325 o_keepalive_item);
1326 list_del_init(&osd->o_keepalive_item);
f24e9980
SW		 1327 ceph_con_keepalive(&osd->o_con);
1328 }
1329
422d2cb8 1330 __schedule_osd_timeout(osdc);
f9d25199 1331 __send_queued(osdc);
f24e9980 1332 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 1333 up_read(&osdc->map_sem);
1334}
1335
f5a2041b
YS		 1336static void handle_osds_timeout(struct work_struct *work)
1337{
1338 struct ceph_osd_client *osdc =
1339 container_of(work, struct ceph_osd_client,
1340 osds_timeout_work.work);
1341 unsigned long delay =
3d14c5d2 1342 osdc->client->options->osd_idle_ttl * HZ >> 2;
f5a2041b
YS		 1343
1344 dout("osds timeout\n");
1345 down_read(&osdc->map_sem);
aca420bc 1346 remove_old_osds(osdc);
f5a2041b
YS		 1347 up_read(&osdc->map_sem);
1348
1349 schedule_delayed_work(&osdc->osds_timeout_work,
1350 round_jiffies_relative(delay));
1351}
1352
25845472
SW		 1353static void complete_request(struct ceph_osd_request *req)
1354{
1355 if (req->r_safe_callback)
1356 req->r_safe_callback(req, NULL);
1357 complete_all(&req->r_safe_completion); /* fsync waiter */
1358}
1359
f24e9980
SW		 1360/*
1361 * handle osd op reply. either call the callback if it is specified,
1362 * or do the completion to wake up the waiting thread.
1363 */
350b1c32
SW		 1364static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg,
1365 struct ceph_connection *con)
f24e9980 1366{
1b83bef2 1367 void *p, *end;
f24e9980
SW		 1368 struct ceph_osd_request *req;
1369 u64 tid;
1b83bef2 1370 int object_len;
79528734
AE		 1371 unsigned int numops;
1372 int payload_len, flags;
0ceed5db 1373 s32 result;
1b83bef2
SW		 1374 s32 retry_attempt;
1375 struct ceph_pg pg;
1376 int err;
1377 u32 reassert_epoch;
1378 u64 reassert_version;
1379 u32 osdmap_epoch;
0d5af164 1380 int already_completed;
9fc6e064 1381 u32 bytes;
79528734 1382 unsigned int i;
f24e9980 1383
6df058c0 1384 tid = le64_to_cpu(msg->hdr.tid);
1b83bef2
SW		 1385 dout("handle_reply %p tid %llu\n", msg, tid);
1386
1387 p = msg->front.iov_base;
1388 end = p + msg->front.iov_len;
1389
1390 ceph_decode_need(&p, end, 4, bad);
1391 object_len = ceph_decode_32(&p);
1392 ceph_decode_need(&p, end, object_len, bad);
1393 p += object_len;
1394
ef4859d6 1395 err = ceph_decode_pgid(&p, end, &pg);
1b83bef2 1396 if (err)
f24e9980 1397 goto bad;
1b83bef2
SW		 1398
1399 ceph_decode_need(&p, end, 8 + 4 + 4 + 8 + 4, bad);
1400 flags = ceph_decode_64(&p);
1401 result = ceph_decode_32(&p);
1402 reassert_epoch = ceph_decode_32(&p);
1403 reassert_version = ceph_decode_64(&p);
1404 osdmap_epoch = ceph_decode_32(&p);
1405
f24e9980
SW		 1406 /* lookup */
1407 mutex_lock(&osdc->request_mutex);
1408 req = __lookup_request(osdc, tid);
1409 if (req == NULL) {
1410 dout("handle_reply tid %llu dne\n", tid);
8058fd45 1411 goto bad_mutex;
f24e9980
SW		 1412 }
1413 ceph_osdc_get_request(req);
1b83bef2
SW		 1414
1415 dout("handle_reply %p tid %llu req %p result %d\n", msg, tid,
1416 req, result);
1417
1418 ceph_decode_need(&p, end, 4, bad);
1419 numops = ceph_decode_32(&p);
1420 if (numops > CEPH_OSD_MAX_OP)
1421 goto bad_put;
1422 if (numops != req->r_num_ops)
1423 goto bad_put;
1424 payload_len = 0;
1425 ceph_decode_need(&p, end, numops * sizeof(struct ceph_osd_op), bad);
1426 for (i = 0; i < numops; i++) {
1427 struct ceph_osd_op *op = p;
1428 int len;
1429
1430 len = le32_to_cpu(op->payload_len);
1431 req->r_reply_op_len[i] = len;
1432 dout(" op %d has %d bytes\n", i, len);
1433 payload_len += len;
1434 p += sizeof(*op);
1435 }
9fc6e064
AE		 1436 bytes = le32_to_cpu(msg->hdr.data_len);
1437 if (payload_len != bytes) {
1b83bef2 1438 pr_warning("sum of op payload lens %d != data_len %d",
9fc6e064 1439 payload_len, bytes);
1b83bef2
SW		 1440 goto bad_put;
1441 }
1442
1443 ceph_decode_need(&p, end, 4 + numops * 4, bad);
1444 retry_attempt = ceph_decode_32(&p);
1445 for (i = 0; i < numops; i++)
1446 req->r_reply_op_result[i] = ceph_decode_32(&p);
f24e9980 1447
f24e9980 1448 if (!req->r_got_reply) {
f24e9980 1449
1b83bef2 1450 req->r_result = result;
f24e9980
SW		 1451 dout("handle_reply result %d bytes %d\n", req->r_result,
1452 bytes);
1453 if (req->r_result == 0)
1454 req->r_result = bytes;
1455
1456 /* in case this is a write and we need to replay, */
1b83bef2
SW		 1457 req->r_reassert_version.epoch = cpu_to_le32(reassert_epoch);
1458 req->r_reassert_version.version = cpu_to_le64(reassert_version);
f24e9980
SW		 1459
1460 req->r_got_reply = 1;
1461 } else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) {
1462 dout("handle_reply tid %llu dup ack\n", tid);
34b43a56 1463 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 1464 goto done;
1465 }
1466
1467 dout("handle_reply tid %llu flags %d\n", tid, flags);
1468
a40c4f10
YS		 1469 if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK))
1470 __register_linger_request(osdc, req);
1471
f24e9980 1472 /* either this is a read, or we got the safe response */
0ceed5db
SW		 1473 if (result < 0 ||
1474 (flags & CEPH_OSD_FLAG_ONDISK) ||
f24e9980
SW		 1475 ((flags & CEPH_OSD_FLAG_WRITE) == 0))
1476 __unregister_request(osdc, req);
1477
0d5af164
AE		 1478 already_completed = req->r_completed;
1479 req->r_completed = 1;
f24e9980 1480 mutex_unlock(&osdc->request_mutex);
0d5af164
AE		 1481 if (already_completed)
1482 goto done;
f24e9980
SW		 1483
1484 if (req->r_callback)
1485 req->r_callback(req, msg);
1486 else
03066f23 1487 complete_all(&req->r_completion);
f24e9980 1488
25845472
SW		 1489 if (flags & CEPH_OSD_FLAG_ONDISK)
1490 complete_request(req);
f24e9980
SW		 1491
1492done:
a40c4f10 1493 dout("req=%p req->r_linger=%d\n", req, req->r_linger);
f24e9980
SW		 1494 ceph_osdc_put_request(req);
1495 return;
1496
1b83bef2
SW		 1497bad_put:
1498 ceph_osdc_put_request(req);
8058fd45
AE		 1499bad_mutex:
1500 mutex_unlock(&osdc->request_mutex);
f24e9980 1501bad:
1b83bef2
SW		 1502 pr_err("corrupt osd_op_reply got %d %d\n",
1503 (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len));
9ec7cab1 1504 ceph_msg_dump(msg);
f24e9980
SW		 1505}
1506
6f6c7006 1507static void reset_changed_osds(struct ceph_osd_client *osdc)
f24e9980 1508{
f24e9980 1509 struct rb_node *p, *n;
f24e9980 1510
6f6c7006
SW		 1511 for (p = rb_first(&osdc->osds); p; p = n) {
1512 struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
f24e9980 1513
6f6c7006
SW		 1514 n = rb_next(p);
1515 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
1516 memcmp(&osd->o_con.peer_addr,
1517 ceph_osd_addr(osdc->osdmap,
1518 osd->o_osd),
1519 sizeof(struct ceph_entity_addr)) != 0)
1520 __reset_osd(osdc, osd);
f24e9980 1521 }
422d2cb8
YS		 1522}
1523
1524/*
6f6c7006
SW		 1525 * Requeue requests whose mapping to an OSD has changed. If requests map to
1526 * no osd, request a new map.
422d2cb8 1527 *
e6d50f67 1528 * Caller should hold map_sem for read.
422d2cb8 1529 */
38d6453c 1530static void kick_requests(struct ceph_osd_client *osdc, int force_resend)
422d2cb8 1531{
a40c4f10 1532 struct ceph_osd_request *req, *nreq;
6f6c7006
SW		 1533 struct rb_node *p;
1534 int needmap = 0;
1535 int err;
422d2cb8 1536
38d6453c 1537 dout("kick_requests %s\n", force_resend ? " (force resend)" : "");
422d2cb8 1538 mutex_lock(&osdc->request_mutex);
6194ea89 1539 for (p = rb_first(&osdc->requests); p; ) {
6f6c7006 1540 req = rb_entry(p, struct ceph_osd_request, r_node);
6194ea89 1541 p = rb_next(p);
ab60b16d
AE		 1542
1543 /*
1544 * For linger requests that have not yet been
1545 * registered, move them to the linger list; they'll
1546 * be sent to the osd in the loop below. Unregister
1547 * the request before re-registering it as a linger
1548 * request to ensure the __map_request() below
1549 * will decide it needs to be sent.
1550 */
1551 if (req->r_linger && list_empty(&req->r_linger_item)) {
1552 dout("%p tid %llu restart on osd%d\n",
1553 req, req->r_tid,
1554 req->r_osd ? req->r_osd->o_osd : -1);
1555 __unregister_request(osdc, req);
1556 __register_linger_request(osdc, req);
1557 continue;
1558 }
1559
38d6453c 1560 err = __map_request(osdc, req, force_resend);
6f6c7006
SW		 1561 if (err < 0)
1562 continue; /* error */
1563 if (req->r_osd == NULL) {
1564 dout("%p tid %llu maps to no osd\n", req, req->r_tid);
1565 needmap++; /* request a newer map */
1566 } else if (err > 0) {
6194ea89
SW		 1567 if (!req->r_linger) {
1568 dout("%p tid %llu requeued on osd%d\n", req,
1569 req->r_tid,
1570 req->r_osd ? req->r_osd->o_osd : -1);
a40c4f10 1571 req->r_flags |= CEPH_OSD_FLAG_RETRY;
6194ea89
SW		 1572 }
1573 }
a40c4f10
YS		 1574 }
1575
1576 list_for_each_entry_safe(req, nreq, &osdc->req_linger,
1577 r_linger_item) {
1578 dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
1579
38d6453c 1580 err = __map_request(osdc, req, force_resend);
ab60b16d 1581 dout("__map_request returned %d\n", err);
a40c4f10
YS		 1582 if (err == 0)
1583 continue; /* no change and no osd was specified */
1584 if (err < 0)
1585 continue; /* hrm! */
1586 if (req->r_osd == NULL) {
1587 dout("tid %llu maps to no valid osd\n", req->r_tid);
1588 needmap++; /* request a newer map */
1589 continue;
6f6c7006 1590 }
a40c4f10
YS		 1591
1592 dout("kicking lingering %p tid %llu osd%d\n", req, req->r_tid,
1593 req->r_osd ? req->r_osd->o_osd : -1);
a40c4f10 1594 __register_request(osdc, req);
c89ce05e 1595 __unregister_linger_request(osdc, req);
6f6c7006 1596 }
f24e9980
SW		 1597 mutex_unlock(&osdc->request_mutex);
1598
1599 if (needmap) {
1600 dout("%d requests for down osds, need new map\n", needmap);
1601 ceph_monc_request_next_osdmap(&osdc->client->monc);
1602 }
e6d50f67 1603 reset_changed_osds(osdc);
422d2cb8 1604}
6f6c7006
SW		 1605
1606
f24e9980
SW		 1607/*
1608 * Process updated osd map.
1609 *
1610 * The message contains any number of incremental and full maps, normally
1611 * indicating some sort of topology change in the cluster. Kick requests
1612 * off to different OSDs as needed.
1613 */
1614void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1615{
1616 void *p, *end, *next;
1617 u32 nr_maps, maplen;
1618 u32 epoch;
1619 struct ceph_osdmap *newmap = NULL, *oldmap;
1620 int err;
1621 struct ceph_fsid fsid;
1622
1623 dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
1624 p = msg->front.iov_base;
1625 end = p + msg->front.iov_len;
1626
1627 /* verify fsid */
1628 ceph_decode_need(&p, end, sizeof(fsid), bad);
1629 ceph_decode_copy(&p, &fsid, sizeof(fsid));
0743304d
SW		 1630 if (ceph_check_fsid(osdc->client, &fsid) < 0)
1631 return;
f24e9980
SW		 1632
1633 down_write(&osdc->map_sem);
1634
1635 /* incremental maps */
1636 ceph_decode_32_safe(&p, end, nr_maps, bad);
1637 dout(" %d inc maps\n", nr_maps);
1638 while (nr_maps > 0) {
1639 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 1640 epoch = ceph_decode_32(&p);
1641 maplen = ceph_decode_32(&p);
f24e9980
SW		 1642 ceph_decode_need(&p, end, maplen, bad);
1643 next = p + maplen;
1644 if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
1645 dout("applying incremental map %u len %d\n",
1646 epoch, maplen);
1647 newmap = osdmap_apply_incremental(&p, next,
1648 osdc->osdmap,
15d9882c 1649 &osdc->client->msgr);
f24e9980
SW		 1650 if (IS_ERR(newmap)) {
1651 err = PTR_ERR(newmap);
1652 goto bad;
1653 }
30dc6381 1654 BUG_ON(!newmap);
f24e9980
SW		 1655 if (newmap != osdc->osdmap) {
1656 ceph_osdmap_destroy(osdc->osdmap);
1657 osdc->osdmap = newmap;
1658 }
38d6453c 1659 kick_requests(osdc, 0);
f24e9980
SW		 1660 } else {
1661 dout("ignoring incremental map %u len %d\n",
1662 epoch, maplen);
1663 }
1664 p = next;
1665 nr_maps--;
1666 }
1667 if (newmap)
1668 goto done;
1669
1670 /* full maps */
1671 ceph_decode_32_safe(&p, end, nr_maps, bad);
1672 dout(" %d full maps\n", nr_maps);
1673 while (nr_maps) {
1674 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 1675 epoch = ceph_decode_32(&p);
1676 maplen = ceph_decode_32(&p);
f24e9980
SW		 1677 ceph_decode_need(&p, end, maplen, bad);
1678 if (nr_maps > 1) {
1679 dout("skipping non-latest full map %u len %d\n",
1680 epoch, maplen);
1681 } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
1682 dout("skipping full map %u len %d, "
1683 "older than our %u\n", epoch, maplen,
1684 osdc->osdmap->epoch);
1685 } else {
38d6453c
SW		 1686 int skipped_map = 0;
1687
f24e9980
SW		 1688 dout("taking full map %u len %d\n", epoch, maplen);
1689 newmap = osdmap_decode(&p, p+maplen);
1690 if (IS_ERR(newmap)) {
1691 err = PTR_ERR(newmap);
1692 goto bad;
1693 }
30dc6381 1694 BUG_ON(!newmap);
f24e9980
SW		 1695 oldmap = osdc->osdmap;
1696 osdc->osdmap = newmap;
38d6453c
SW		 1697 if (oldmap) {
1698 if (oldmap->epoch + 1 < newmap->epoch)
1699 skipped_map = 1;
f24e9980 1700 ceph_osdmap_destroy(oldmap);
38d6453c
SW		 1701 }
1702 kick_requests(osdc, skipped_map);
f24e9980
SW		 1703 }
1704 p += maplen;
1705 nr_maps--;
1706 }
1707
1708done:
1709 downgrade_write(&osdc->map_sem);
1710 ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch);
cd634fb6
SW		 1711
1712 /*
1713 * subscribe to subsequent osdmap updates if full to ensure
1714 * we find out when we are no longer full and stop returning
1715 * ENOSPC.
1716 */
1717 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL))
1718 ceph_monc_request_next_osdmap(&osdc->client->monc);
1719
f9d25199
AE		 1720 mutex_lock(&osdc->request_mutex);
1721 __send_queued(osdc);
1722 mutex_unlock(&osdc->request_mutex);
f24e9980 1723 up_read(&osdc->map_sem);
03066f23 1724 wake_up_all(&osdc->client->auth_wq);
f24e9980
SW		 1725 return;
1726
1727bad:
1728 pr_err("osdc handle_map corrupt msg\n");
9ec7cab1 1729 ceph_msg_dump(msg);
f24e9980
SW		 1730 up_write(&osdc->map_sem);
1731 return;
1732}
1733
a40c4f10
YS		 1734/*
1735 * watch/notify callback event infrastructure
1736 *
1737 * These callbacks are used both for watch and notify operations.
1738 */
1739static void __release_event(struct kref *kref)
1740{
1741 struct ceph_osd_event *event =
1742 container_of(kref, struct ceph_osd_event, kref);
1743
1744 dout("__release_event %p\n", event);
1745 kfree(event);
1746}
1747
1748static void get_event(struct ceph_osd_event *event)
1749{
1750 kref_get(&event->kref);
1751}
1752
1753void ceph_osdc_put_event(struct ceph_osd_event *event)
1754{
1755 kref_put(&event->kref, __release_event);
1756}
1757EXPORT_SYMBOL(ceph_osdc_put_event);
1758
1759static void __insert_event(struct ceph_osd_client *osdc,
1760 struct ceph_osd_event *new)
1761{
1762 struct rb_node **p = &osdc->event_tree.rb_node;
1763 struct rb_node *parent = NULL;
1764 struct ceph_osd_event *event = NULL;
1765
1766 while (*p) {
1767 parent = *p;
1768 event = rb_entry(parent, struct ceph_osd_event, node);
1769 if (new->cookie < event->cookie)
1770 p = &(*p)->rb_left;
1771 else if (new->cookie > event->cookie)
1772 p = &(*p)->rb_right;
1773 else
1774 BUG();
1775 }
1776
1777 rb_link_node(&new->node, parent, p);
1778 rb_insert_color(&new->node, &osdc->event_tree);
1779}
1780
1781static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
1782 u64 cookie)
1783{
1784 struct rb_node **p = &osdc->event_tree.rb_node;
1785 struct rb_node *parent = NULL;
1786 struct ceph_osd_event *event = NULL;
1787
1788 while (*p) {
1789 parent = *p;
1790 event = rb_entry(parent, struct ceph_osd_event, node);
1791 if (cookie < event->cookie)
1792 p = &(*p)->rb_left;
1793 else if (cookie > event->cookie)
1794 p = &(*p)->rb_right;
1795 else
1796 return event;
1797 }
1798 return NULL;
1799}
1800
1801static void __remove_event(struct ceph_osd_event *event)
1802{
1803 struct ceph_osd_client *osdc = event->osdc;
1804
1805 if (!RB_EMPTY_NODE(&event->node)) {
1806 dout("__remove_event removed %p\n", event);
1807 rb_erase(&event->node, &osdc->event_tree);
1808 ceph_osdc_put_event(event);
1809 } else {
1810 dout("__remove_event didn't remove %p\n", event);
1811 }
1812}
1813
1814int ceph_osdc_create_event(struct ceph_osd_client *osdc,
1815 void (*event_cb)(u64, u64, u8, void *),
3c663bbd 1816 void *data, struct ceph_osd_event **pevent)
a40c4f10
YS		 1817{
1818 struct ceph_osd_event *event;
1819
1820 event = kmalloc(sizeof(*event), GFP_NOIO);
1821 if (!event)
1822 return -ENOMEM;
1823
1824 dout("create_event %p\n", event);
1825 event->cb = event_cb;
3c663bbd 1826 event->one_shot = 0;
a40c4f10
YS		 1827 event->data = data;
1828 event->osdc = osdc;
1829 INIT_LIST_HEAD(&event->osd_node);
3ee5234d 1830 RB_CLEAR_NODE(&event->node);
a40c4f10
YS		 1831 kref_init(&event->kref); /* one ref for us */
1832 kref_get(&event->kref); /* one ref for the caller */
a40c4f10
YS		 1833
1834 spin_lock(&osdc->event_lock);
1835 event->cookie = ++osdc->event_count;
1836 __insert_event(osdc, event);
1837 spin_unlock(&osdc->event_lock);
1838
1839 *pevent = event;
1840 return 0;
1841}
1842EXPORT_SYMBOL(ceph_osdc_create_event);
1843
1844void ceph_osdc_cancel_event(struct ceph_osd_event *event)
1845{
1846 struct ceph_osd_client *osdc = event->osdc;
1847
1848 dout("cancel_event %p\n", event);
1849 spin_lock(&osdc->event_lock);
1850 __remove_event(event);
1851 spin_unlock(&osdc->event_lock);
1852 ceph_osdc_put_event(event); /* caller's */
1853}
1854EXPORT_SYMBOL(ceph_osdc_cancel_event);
1855
1856
1857static void do_event_work(struct work_struct *work)
1858{
1859 struct ceph_osd_event_work *event_work =
1860 container_of(work, struct ceph_osd_event_work, work);
1861 struct ceph_osd_event *event = event_work->event;
1862 u64 ver = event_work->ver;
1863 u64 notify_id = event_work->notify_id;
1864 u8 opcode = event_work->opcode;
1865
1866 dout("do_event_work completing %p\n", event);
1867 event->cb(ver, notify_id, opcode, event->data);
a40c4f10
YS		 1868 dout("do_event_work completed %p\n", event);
1869 ceph_osdc_put_event(event);
1870 kfree(event_work);
1871}
1872
1873
1874/*
1875 * Process osd watch notifications
1876 */
3c663bbd
AE		 1877static void handle_watch_notify(struct ceph_osd_client *osdc,
1878 struct ceph_msg *msg)
a40c4f10
YS		 1879{
1880 void *p, *end;
1881 u8 proto_ver;
1882 u64 cookie, ver, notify_id;
1883 u8 opcode;
1884 struct ceph_osd_event *event;
1885 struct ceph_osd_event_work *event_work;
1886
1887 p = msg->front.iov_base;
1888 end = p + msg->front.iov_len;
1889
1890 ceph_decode_8_safe(&p, end, proto_ver, bad);
1891 ceph_decode_8_safe(&p, end, opcode, bad);
1892 ceph_decode_64_safe(&p, end, cookie, bad);
1893 ceph_decode_64_safe(&p, end, ver, bad);
1894 ceph_decode_64_safe(&p, end, notify_id, bad);
1895
1896 spin_lock(&osdc->event_lock);
1897 event = __find_event(osdc, cookie);
1898 if (event) {
3c663bbd 1899 BUG_ON(event->one_shot);
a40c4f10 1900 get_event(event);
a40c4f10
YS		 1901 }
1902 spin_unlock(&osdc->event_lock);
1903 dout("handle_watch_notify cookie %lld ver %lld event %p\n",
1904 cookie, ver, event);
1905 if (event) {
1906 event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
a40c4f10
YS		 1907 if (!event_work) {
1908 dout("ERROR: could not allocate event_work\n");
1909 goto done_err;
1910 }
6b0ae409 1911 INIT_WORK(&event_work->work, do_event_work);
a40c4f10
YS		 1912 event_work->event = event;
1913 event_work->ver = ver;
1914 event_work->notify_id = notify_id;
1915 event_work->opcode = opcode;
1916 if (!queue_work(osdc->notify_wq, &event_work->work)) {
1917 dout("WARNING: failed to queue notify event work\n");
1918 goto done_err;
1919 }
1920 }
1921
1922 return;
1923
1924done_err:
a40c4f10
YS		 1925 ceph_osdc_put_event(event);
1926 return;
1927
1928bad:
1929 pr_err("osdc handle_watch_notify corrupt msg\n");
1930 return;
1931}
1932
70636773
AE		 1933static void ceph_osdc_msg_data_set(struct ceph_msg *msg,
1934 struct ceph_osd_data *osd_data)
f24e9980 1935{
23c08a9c
AE		 1936 u64 length = ceph_osd_data_length(osd_data);
1937
0fff87ec 1938 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
23c08a9c
AE		 1939 BUG_ON(length > (u64) SIZE_MAX);
1940 if (length)
70636773 1941 ceph_msg_data_set_pages(msg, osd_data->pages,
23c08a9c 1942 length, osd_data->alignment);
9a5e6d09 1943 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
23c08a9c 1944 BUG_ON(!length);
9a5e6d09 1945 ceph_msg_data_set_pagelist(msg, osd_data->pagelist);
68b4476b 1946#ifdef CONFIG_BLOCK
0fff87ec 1947 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
23c08a9c 1948 ceph_msg_data_set_bio(msg, osd_data->bio, length);
68b4476b 1949#endif
2ac2b7a6 1950 } else {
0fff87ec 1951 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
2ac2b7a6 1952 }
70636773
AE		 1953}
1954
1955/*
1956 * Register request, send initial attempt.
1957 */
1958int ceph_osdc_start_request(struct ceph_osd_client *osdc,
1959 struct ceph_osd_request *req,
1960 bool nofail)
1961{
1962 int rc = 0;
1963
1964 /* Set up response incoming data and request outgoing data fields */
1965
1966 ceph_osdc_msg_data_set(req->r_reply, &req->r_data_in);
1967 ceph_osdc_msg_data_set(req->r_request, &req->r_data_out);
f24e9980 1968
f24e9980
SW		 1969 down_read(&osdc->map_sem);
1970 mutex_lock(&osdc->request_mutex);
dc4b870c 1971 __register_request(osdc, req);
92451b49
AE		 1972 WARN_ON(req->r_sent);
1973 rc = __map_request(osdc, req, 0);
1974 if (rc < 0) {
1975 if (nofail) {
1976 dout("osdc_start_request failed map, "
1977 " will retry %lld\n", req->r_tid);
1978 rc = 0;
f24e9980 1979 }
92451b49 1980 goto out_unlock;
f24e9980 1981 }
92451b49
AE		 1982 if (req->r_osd == NULL) {
1983 dout("send_request %p no up osds in pg\n", req);
1984 ceph_monc_request_next_osdmap(&osdc->client->monc);
1985 } else {
7e2766a1 1986 __send_queued(osdc);
92451b49
AE		 1987 }
1988 rc = 0;
234af26f 1989out_unlock:
f24e9980
SW		 1990 mutex_unlock(&osdc->request_mutex);
1991 up_read(&osdc->map_sem);
1992 return rc;
1993}
3d14c5d2 1994EXPORT_SYMBOL(ceph_osdc_start_request);
f24e9980
SW		 1995
1996/*
1997 * wait for a request to complete
1998 */
1999int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
2000 struct ceph_osd_request *req)
2001{
2002 int rc;
2003
2004 rc = wait_for_completion_interruptible(&req->r_completion);
2005 if (rc < 0) {
2006 mutex_lock(&osdc->request_mutex);
2007 __cancel_request(req);
529cfcc4 2008 __unregister_request(osdc, req);
f24e9980 2009 mutex_unlock(&osdc->request_mutex);
25845472 2010 complete_request(req);
529cfcc4 2011 dout("wait_request tid %llu canceled/timed out\n", req->r_tid);
f24e9980
SW		 2012 return rc;
2013 }
2014
2015 dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result);
2016 return req->r_result;
2017}
3d14c5d2 2018EXPORT_SYMBOL(ceph_osdc_wait_request);
f24e9980
SW		 2019
2020/*
2021 * sync - wait for all in-flight requests to flush. avoid starvation.
2022 */
2023void ceph_osdc_sync(struct ceph_osd_client *osdc)
2024{
2025 struct ceph_osd_request *req;
2026 u64 last_tid, next_tid = 0;
2027
2028 mutex_lock(&osdc->request_mutex);
2029 last_tid = osdc->last_tid;
2030 while (1) {
2031 req = __lookup_request_ge(osdc, next_tid);
2032 if (!req)
2033 break;
2034 if (req->r_tid > last_tid)
2035 break;
2036
2037 next_tid = req->r_tid + 1;
2038 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
2039 continue;
2040
2041 ceph_osdc_get_request(req);
2042 mutex_unlock(&osdc->request_mutex);
2043 dout("sync waiting on tid %llu (last is %llu)\n",
2044 req->r_tid, last_tid);
2045 wait_for_completion(&req->r_safe_completion);
2046 mutex_lock(&osdc->request_mutex);
2047 ceph_osdc_put_request(req);
2048 }
2049 mutex_unlock(&osdc->request_mutex);
2050 dout("sync done (thru tid %llu)\n", last_tid);
2051}
3d14c5d2 2052EXPORT_SYMBOL(ceph_osdc_sync);
f24e9980
SW		 2053
2054/*
2055 * init, shutdown
2056 */
2057int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
2058{
2059 int err;
2060
2061 dout("init\n");
2062 osdc->client = client;
2063 osdc->osdmap = NULL;
2064 init_rwsem(&osdc->map_sem);
2065 init_completion(&osdc->map_waiters);
2066 osdc->last_requested_map = 0;
2067 mutex_init(&osdc->request_mutex);
f24e9980
SW		 2068 osdc->last_tid = 0;
2069 osdc->osds = RB_ROOT;
f5a2041b 2070 INIT_LIST_HEAD(&osdc->osd_lru);
f24e9980 2071 osdc->requests = RB_ROOT;
422d2cb8 2072 INIT_LIST_HEAD(&osdc->req_lru);
6f6c7006
SW		 2073 INIT_LIST_HEAD(&osdc->req_unsent);
2074 INIT_LIST_HEAD(&osdc->req_notarget);
a40c4f10 2075 INIT_LIST_HEAD(&osdc->req_linger);
f24e9980
SW		 2076 osdc->num_requests = 0;
2077 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
f5a2041b 2078 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
a40c4f10
YS		 2079 spin_lock_init(&osdc->event_lock);
2080 osdc->event_tree = RB_ROOT;
2081 osdc->event_count = 0;
f5a2041b
YS		 2082
2083 schedule_delayed_work(&osdc->osds_timeout_work,
3d14c5d2 2084 round_jiffies_relative(osdc->client->options->osd_idle_ttl * HZ));
f24e9980 2085
5f44f142 2086 err = -ENOMEM;
f24e9980
SW		 2087 osdc->req_mempool = mempool_create_kmalloc_pool(10,
2088 sizeof(struct ceph_osd_request));
2089 if (!osdc->req_mempool)
5f44f142 2090 goto out;
f24e9980 2091
d50b409f
SW		 2092 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
2093 OSD_OP_FRONT_LEN, 10, true,
4f48280e 2094 "osd_op");
f24e9980 2095 if (err < 0)
5f44f142 2096 goto out_mempool;
d50b409f 2097 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
4f48280e
SW		 2098 OSD_OPREPLY_FRONT_LEN, 10, true,
2099 "osd_op_reply");
c16e7869
SW		 2100 if (err < 0)
2101 goto out_msgpool;
a40c4f10
YS		 2102
2103 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
2104 if (IS_ERR(osdc->notify_wq)) {
2105 err = PTR_ERR(osdc->notify_wq);
2106 osdc->notify_wq = NULL;
2107 goto out_msgpool;
2108 }
f24e9980 2109 return 0;
5f44f142 2110
c16e7869
SW		 2111out_msgpool:
2112 ceph_msgpool_destroy(&osdc->msgpool_op);
5f44f142
SW		 2113out_mempool:
2114 mempool_destroy(osdc->req_mempool);
2115out:
2116 return err;
f24e9980
SW		 2117}
2118
2119void ceph_osdc_stop(struct ceph_osd_client *osdc)
2120{
a40c4f10
YS		 2121 flush_workqueue(osdc->notify_wq);
2122 destroy_workqueue(osdc->notify_wq);
f24e9980 2123 cancel_delayed_work_sync(&osdc->timeout_work);
f5a2041b 2124 cancel_delayed_work_sync(&osdc->osds_timeout_work);
f24e9980
SW		 2125 if (osdc->osdmap) {
2126 ceph_osdmap_destroy(osdc->osdmap);
2127 osdc->osdmap = NULL;
2128 }
aca420bc 2129 remove_all_osds(osdc);
f24e9980
SW		 2130 mempool_destroy(osdc->req_mempool);
2131 ceph_msgpool_destroy(&osdc->msgpool_op);
c16e7869 2132 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
f24e9980
SW		 2133}
2134
2135/*
2136 * Read some contiguous pages. If we cross a stripe boundary, shorten
2137 * *plen. Return number of bytes read, or error.
2138 */
2139int ceph_osdc_readpages(struct ceph_osd_client *osdc,
2140 struct ceph_vino vino, struct ceph_file_layout *layout,
2141 u64 off, u64 *plen,
2142 u32 truncate_seq, u64 truncate_size,
b7495fc2 2143 struct page **pages, int num_pages, int page_align)
f24e9980
SW		 2144{
2145 struct ceph_osd_request *req;
2146 int rc = 0;
2147
2148 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
2149 vino.snap, off, *plen);
79528734 2150 req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 1,
f24e9980 2151 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
acead002 2152 NULL, truncate_seq, truncate_size,
153e5167 2153 false);
6816282d
SW		 2154 if (IS_ERR(req))
2155 return PTR_ERR(req);
f24e9980
SW		 2156
2157 /* it may be a short read due to an object boundary */
0fff87ec 2158
43bfe5de
AE		 2159 ceph_osd_data_pages_init(&req->r_data_in, pages, *plen, page_align,
2160 false, false);
f24e9980 2161
e0c59487 2162 dout("readpages final extent is %llu~%llu (%llu bytes align %d)\n",
43bfe5de 2163 off, *plen, *plen, page_align);
f24e9980 2164
79528734 2165 ceph_osdc_build_request(req, off, NULL, vino.snap, NULL);
02ee07d3 2166
f24e9980
SW		 2167 rc = ceph_osdc_start_request(osdc, req, false);
2168 if (!rc)
2169 rc = ceph_osdc_wait_request(osdc, req);
2170
2171 ceph_osdc_put_request(req);
2172 dout("readpages result %d\n", rc);
2173 return rc;
2174}
3d14c5d2 2175EXPORT_SYMBOL(ceph_osdc_readpages);
f24e9980
SW		 2176
2177/*
2178 * do a synchronous write on N pages
2179 */
2180int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
2181 struct ceph_file_layout *layout,
2182 struct ceph_snap_context *snapc,
2183 u64 off, u64 len,
2184 u32 truncate_seq, u64 truncate_size,
2185 struct timespec *mtime,
24808826 2186 struct page **pages, int num_pages)
f24e9980
SW		 2187{
2188 struct ceph_osd_request *req;
2189 int rc = 0;
b7495fc2 2190 int page_align = off & ~PAGE_MASK;
f24e9980 2191
acead002 2192 BUG_ON(vino.snap != CEPH_NOSNAP); /* snapshots aren't writeable */
79528734 2193 req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 1,
f24e9980 2194 CEPH_OSD_OP_WRITE,
24808826 2195 CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
acead002 2196 snapc, truncate_seq, truncate_size,
153e5167 2197 true);
6816282d
SW		 2198 if (IS_ERR(req))
2199 return PTR_ERR(req);
f24e9980
SW		 2200
2201 /* it may be a short write due to an object boundary */
43bfe5de
AE		 2202 ceph_osd_data_pages_init(&req->r_data_out, pages, len, page_align,
2203 false, false);
2204 dout("writepages %llu~%llu (%llu bytes)\n", off, len, len);
f24e9980 2205
79528734 2206 ceph_osdc_build_request(req, off, snapc, CEPH_NOSNAP, mtime);
02ee07d3 2207
87f979d3 2208 rc = ceph_osdc_start_request(osdc, req, true);
f24e9980
SW		 2209 if (!rc)
2210 rc = ceph_osdc_wait_request(osdc, req);
2211
2212 ceph_osdc_put_request(req);
2213 if (rc == 0)
2214 rc = len;
2215 dout("writepages result %d\n", rc);
2216 return rc;
2217}
3d14c5d2 2218EXPORT_SYMBOL(ceph_osdc_writepages);
f24e9980
SW		 2219
2220/*
2221 * handle incoming message
2222 */
2223static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
2224{
2225 struct ceph_osd *osd = con->private;
32c895e7 2226 struct ceph_osd_client *osdc;
f24e9980
SW		 2227 int type = le16_to_cpu(msg->hdr.type);
2228
2229 if (!osd)
4a32f93d 2230 goto out;
32c895e7 2231 osdc = osd->o_osdc;
f24e9980
SW		 2232
2233 switch (type) {
2234 case CEPH_MSG_OSD_MAP:
2235 ceph_osdc_handle_map(osdc, msg);
2236 break;
2237 case CEPH_MSG_OSD_OPREPLY:
350b1c32 2238 handle_reply(osdc, msg, con);
f24e9980 2239 break;
a40c4f10
YS		 2240 case CEPH_MSG_WATCH_NOTIFY:
2241 handle_watch_notify(osdc, msg);
2242 break;
f24e9980
SW		 2243
2244 default:
2245 pr_err("received unknown message type %d %s\n", type,
2246 ceph_msg_type_name(type));
2247 }
4a32f93d 2248out:
f24e9980
SW		 2249 ceph_msg_put(msg);
2250}
2251
5b3a4db3 2252/*
21b667f6
SW		 2253 * lookup and return message for incoming reply. set up reply message
2254 * pages.
5b3a4db3
SW		 2255 */
2256static struct ceph_msg *get_reply(struct ceph_connection *con,
2450418c
YS		 2257 struct ceph_msg_header *hdr,
2258 int *skip)
f24e9980
SW		 2259{
2260 struct ceph_osd *osd = con->private;
2261 struct ceph_osd_client *osdc = osd->o_osdc;
2450418c 2262 struct ceph_msg *m;
0547a9b3 2263 struct ceph_osd_request *req;
5b3a4db3
SW		 2264 int front = le32_to_cpu(hdr->front_len);
2265 int data_len = le32_to_cpu(hdr->data_len);
0547a9b3 2266 u64 tid;
f24e9980 2267
0547a9b3
YS		 2268 tid = le64_to_cpu(hdr->tid);
2269 mutex_lock(&osdc->request_mutex);
2270 req = __lookup_request(osdc, tid);
2271 if (!req) {
2272 *skip = 1;
2273 m = NULL;
756a16a5
SW		 2274 dout("get_reply unknown tid %llu from osd%d\n", tid,
2275 osd->o_osd);
0547a9b3
YS		 2276 goto out;
2277 }
c16e7869 2278
ace6d3a9 2279 if (req->r_reply->con)
8921d114 2280 dout("%s revoking msg %p from old con %p\n", __func__,
ace6d3a9
AE		 2281 req->r_reply, req->r_reply->con);
2282 ceph_msg_revoke_incoming(req->r_reply);
0547a9b3 2283
c16e7869
SW		 2284 if (front > req->r_reply->front.iov_len) {
2285 pr_warning("get_reply front %d > preallocated %d\n",
2286 front, (int)req->r_reply->front.iov_len);
b61c2763 2287 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front, GFP_NOFS, false);
a79832f2 2288 if (!m)
c16e7869
SW		 2289 goto out;
2290 ceph_msg_put(req->r_reply);
2291 req->r_reply = m;
2292 }
2293 m = ceph_msg_get(req->r_reply);
2294
0547a9b3 2295 if (data_len > 0) {
0fff87ec
AE		 2296 struct ceph_osd_data *osd_data = &req->r_data_in;
2297
2298 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
0fff87ec 2299 if (osd_data->pages &&
e0c59487 2300 unlikely(osd_data->length < data_len)) {
2ac2b7a6 2301
e0c59487
AE		 2302 pr_warning("tid %lld reply has %d bytes "
2303 "we had only %llu bytes ready\n",
2304 tid, data_len, osd_data->length);
2ac2b7a6
AE		 2305 *skip = 1;
2306 ceph_msg_put(m);
2307 m = NULL;
2308 goto out;
2309 }
2ac2b7a6 2310 }
0547a9b3 2311 }
5b3a4db3 2312 *skip = 0;
c16e7869 2313 dout("get_reply tid %lld %p\n", tid, m);
0547a9b3
YS		 2314
2315out:
2316 mutex_unlock(&osdc->request_mutex);
2450418c 2317 return m;
5b3a4db3
SW		 2318
2319}
2320
2321static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2322 struct ceph_msg_header *hdr,
2323 int *skip)
2324{
2325 struct ceph_osd *osd = con->private;
2326 int type = le16_to_cpu(hdr->type);
2327 int front = le32_to_cpu(hdr->front_len);
2328
1c20f2d2 2329 *skip = 0;
5b3a4db3
SW		 2330 switch (type) {
2331 case CEPH_MSG_OSD_MAP:
a40c4f10 2332 case CEPH_MSG_WATCH_NOTIFY:
b61c2763 2333 return ceph_msg_new(type, front, GFP_NOFS, false);
5b3a4db3
SW		 2334 case CEPH_MSG_OSD_OPREPLY:
2335 return get_reply(con, hdr, skip);
2336 default:
2337 pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
2338 osd->o_osd);
2339 *skip = 1;
2340 return NULL;
2341 }
f24e9980
SW		 2342}
2343
2344/*
2345 * Wrappers to refcount containing ceph_osd struct
2346 */
2347static struct ceph_connection *get_osd_con(struct ceph_connection *con)
2348{
2349 struct ceph_osd *osd = con->private;
2350 if (get_osd(osd))
2351 return con;
2352 return NULL;
2353}
2354
2355static void put_osd_con(struct ceph_connection *con)
2356{
2357 struct ceph_osd *osd = con->private;
2358 put_osd(osd);
2359}
2360
4e7a5dcd
SW		 2361/*
2362 * authentication
2363 */
a3530df3
AE		 2364/*
2365 * Note: returned pointer is the address of a structure that's
2366 * managed separately. Caller must *not* attempt to free it.
2367 */
2368static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
8f43fb53 2369 int *proto, int force_new)
4e7a5dcd
SW		 2370{
2371 struct ceph_osd *o = con->private;
2372 struct ceph_osd_client *osdc = o->o_osdc;
2373 struct ceph_auth_client *ac = osdc->client->monc.auth;
74f1869f 2374 struct ceph_auth_handshake *auth = &o->o_auth;
4e7a5dcd 2375
74f1869f 2376 if (force_new && auth->authorizer) {
27859f97 2377 ceph_auth_destroy_authorizer(ac, auth->authorizer);
74f1869f
AE		 2378 auth->authorizer = NULL;
2379 }
27859f97
SW		 2380 if (!auth->authorizer) {
2381 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2382 auth);
4e7a5dcd 2383 if (ret)
a3530df3 2384 return ERR_PTR(ret);
27859f97
SW		 2385 } else {
2386 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
0bed9b5c
SW		 2387 auth);
2388 if (ret)
2389 return ERR_PTR(ret);
4e7a5dcd 2390 }
4e7a5dcd 2391 *proto = ac->protocol;
74f1869f 2392
a3530df3 2393 return auth;
4e7a5dcd
SW		 2394}
2395
2396
2397static int verify_authorizer_reply(struct ceph_connection *con, int len)
2398{
2399 struct ceph_osd *o = con->private;
2400 struct ceph_osd_client *osdc = o->o_osdc;
2401 struct ceph_auth_client *ac = osdc->client->monc.auth;
2402
27859f97 2403 return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer, len);
4e7a5dcd
SW		 2404}
2405
9bd2e6f8
SW		 2406static int invalidate_authorizer(struct ceph_connection *con)
2407{
2408 struct ceph_osd *o = con->private;
2409 struct ceph_osd_client *osdc = o->o_osdc;
2410 struct ceph_auth_client *ac = osdc->client->monc.auth;
2411
27859f97 2412 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
9bd2e6f8
SW		 2413 return ceph_monc_validate_auth(&osdc->client->monc);
2414}
4e7a5dcd 2415
9e32789f 2416static const struct ceph_connection_operations osd_con_ops = {
f24e9980
SW		 2417 .get = get_osd_con,
2418 .put = put_osd_con,
2419 .dispatch = dispatch,
4e7a5dcd
SW		 2420 .get_authorizer = get_authorizer,
2421 .verify_authorizer_reply = verify_authorizer_reply,
9bd2e6f8 2422 .invalidate_authorizer = invalidate_authorizer,
f24e9980 2423 .alloc_msg = alloc_msg,
81b024e7 2424 .fault = osd_reset,
f24e9980 2425};
Linux 6.x block layer and io_uring tree(s)