projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
ceph: use page_offset() in ceph_writepages_start()
[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
f24e9980
SW		 82/*
83 * requests
84 */
415e49a9 85void ceph_osdc_release_request(struct kref *kref)
f24e9980 86{
e0c59487 87 int num_pages;
415e49a9
SW		 88 struct ceph_osd_request *req = container_of(kref,
89 struct ceph_osd_request,
90 r_kref);
91
92 if (req->r_request)
93 ceph_msg_put(req->r_request);
0d59ab81 94 if (req->r_con_filling_msg) {
9cbb1d72
AE		 95 dout("%s revoking msg %p from con %p\n", __func__,
96 req->r_reply, req->r_con_filling_msg);
8921d114 97 ceph_msg_revoke_incoming(req->r_reply);
0d47766f 98 req->r_con_filling_msg->ops->put(req->r_con_filling_msg);
9cbb1d72 99 req->r_con_filling_msg = NULL;
350b1c32 100 }
ab8cb34a
AE		 101 if (req->r_reply)
102 ceph_msg_put(req->r_reply);
0fff87ec
AE		 103
104 if (req->r_data_in.type == CEPH_OSD_DATA_TYPE_PAGES &&
e0c59487
AE		 105 req->r_data_in.own_pages) {
106 num_pages = calc_pages_for((u64)req->r_data_in.alignment,
107 (u64)req->r_data_in.length);
108 ceph_release_page_vector(req->r_data_in.pages, num_pages);
109 }
0fff87ec 110 if (req->r_data_out.type == CEPH_OSD_DATA_TYPE_PAGES &&
e0c59487
AE		 111 req->r_data_out.own_pages) {
112 num_pages = calc_pages_for((u64)req->r_data_out.alignment,
113 (u64)req->r_data_out.length);
114 ceph_release_page_vector(req->r_data_out.pages, num_pages);
115 }
0fff87ec 116
415e49a9
SW		 117 ceph_put_snap_context(req->r_snapc);
118 if (req->r_mempool)
119 mempool_free(req, req->r_osdc->req_mempool);
120 else
121 kfree(req);
f24e9980 122}
3d14c5d2 123EXPORT_SYMBOL(ceph_osdc_release_request);
68b4476b 124
3499e8a5 125struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
f24e9980 126 struct ceph_snap_context *snapc,
1b83bef2 127 unsigned int num_ops,
3499e8a5 128 bool use_mempool,
54a54007 129 gfp_t gfp_flags)
f24e9980
SW		 130{
131 struct ceph_osd_request *req;
132 struct ceph_msg *msg;
1b83bef2
SW		 133 size_t msg_size;
134
135 msg_size = 4 + 4 + 8 + 8 + 4+8;
136 msg_size += 2 + 4 + 8 + 4 + 4; /* oloc */
137 msg_size += 1 + 8 + 4 + 4; /* pg_t */
138 msg_size += 4 + MAX_OBJ_NAME_SIZE;
139 msg_size += 2 + num_ops*sizeof(struct ceph_osd_op);
140 msg_size += 8; /* snapid */
141 msg_size += 8; /* snap_seq */
142 msg_size += 8 * (snapc ? snapc->num_snaps : 0); /* snaps */
143 msg_size += 4;
f24e9980
SW		 144
145 if (use_mempool) {
3499e8a5 146 req = mempool_alloc(osdc->req_mempool, gfp_flags);
f24e9980
SW		 147 memset(req, 0, sizeof(*req));
148 } else {
3499e8a5 149 req = kzalloc(sizeof(*req), gfp_flags);
f24e9980
SW		 150 }
151 if (req == NULL)
a79832f2 152 return NULL;
f24e9980 153
f24e9980
SW		 154 req->r_osdc = osdc;
155 req->r_mempool = use_mempool;
68b4476b 156
415e49a9 157 kref_init(&req->r_kref);
f24e9980
SW		 158 init_completion(&req->r_completion);
159 init_completion(&req->r_safe_completion);
a978fa20 160 RB_CLEAR_NODE(&req->r_node);
f24e9980 161 INIT_LIST_HEAD(&req->r_unsafe_item);
a40c4f10
YS		 162 INIT_LIST_HEAD(&req->r_linger_item);
163 INIT_LIST_HEAD(&req->r_linger_osd);
935b639a 164 INIT_LIST_HEAD(&req->r_req_lru_item);
cd43045c
SW		 165 INIT_LIST_HEAD(&req->r_osd_item);
166
c16e7869
SW		 167 /* create reply message */
168 if (use_mempool)
169 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0);
170 else
171 msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY,
b61c2763 172 OSD_OPREPLY_FRONT_LEN, gfp_flags, true);
a79832f2 173 if (!msg) {
c16e7869 174 ceph_osdc_put_request(req);
a79832f2 175 return NULL;
c16e7869
SW		 176 }
177 req->r_reply = msg;
178
0fff87ec
AE		 179 req->r_data_in.type = CEPH_OSD_DATA_TYPE_NONE;
180 req->r_data_out.type = CEPH_OSD_DATA_TYPE_NONE;
d50b409f 181
c16e7869 182 /* create request message; allow space for oid */
f24e9980 183 if (use_mempool)
8f3bc053 184 msg = ceph_msgpool_get(&osdc->msgpool_op, 0);
f24e9980 185 else
b61c2763 186 msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp_flags, true);
a79832f2 187 if (!msg) {
f24e9980 188 ceph_osdc_put_request(req);
a79832f2 189 return NULL;
f24e9980 190 }
68b4476b 191
f24e9980 192 memset(msg->front.iov_base, 0, msg->front.iov_len);
3499e8a5
YS		 193
194 req->r_request = msg;
3499e8a5
YS		 195
196 return req;
197}
3d14c5d2 198EXPORT_SYMBOL(ceph_osdc_alloc_request);
3499e8a5 199
a8dd0a37 200static bool osd_req_opcode_valid(u16 opcode)
68b4476b 201{
a8dd0a37 202 switch (opcode) {
68b4476b 203 case CEPH_OSD_OP_READ:
a8dd0a37 204 case CEPH_OSD_OP_STAT:
4c46459c
AE		 205 case CEPH_OSD_OP_MAPEXT:
206 case CEPH_OSD_OP_MASKTRUNC:
207 case CEPH_OSD_OP_SPARSE_READ:
a9f36c3e 208 case CEPH_OSD_OP_NOTIFY:
a8dd0a37 209 case CEPH_OSD_OP_NOTIFY_ACK:
4c46459c 210 case CEPH_OSD_OP_ASSERT_VER:
a8dd0a37 211 case CEPH_OSD_OP_WRITE:
4c46459c
AE		 212 case CEPH_OSD_OP_WRITEFULL:
213 case CEPH_OSD_OP_TRUNCATE:
214 case CEPH_OSD_OP_ZERO:
215 case CEPH_OSD_OP_DELETE:
216 case CEPH_OSD_OP_APPEND:
a8dd0a37 217 case CEPH_OSD_OP_STARTSYNC:
4c46459c
AE		 218 case CEPH_OSD_OP_SETTRUNC:
219 case CEPH_OSD_OP_TRIMTRUNC:
220 case CEPH_OSD_OP_TMAPUP:
221 case CEPH_OSD_OP_TMAPPUT:
222 case CEPH_OSD_OP_TMAPGET:
223 case CEPH_OSD_OP_CREATE:
a9f36c3e 224 case CEPH_OSD_OP_ROLLBACK:
a8dd0a37 225 case CEPH_OSD_OP_WATCH:
4c46459c
AE		 226 case CEPH_OSD_OP_OMAPGETKEYS:
227 case CEPH_OSD_OP_OMAPGETVALS:
228 case CEPH_OSD_OP_OMAPGETHEADER:
229 case CEPH_OSD_OP_OMAPGETVALSBYKEYS:
4c46459c
AE		 230 case CEPH_OSD_OP_OMAPSETVALS:
231 case CEPH_OSD_OP_OMAPSETHEADER:
232 case CEPH_OSD_OP_OMAPCLEAR:
233 case CEPH_OSD_OP_OMAPRMKEYS:
234 case CEPH_OSD_OP_OMAP_CMP:
235 case CEPH_OSD_OP_CLONERANGE:
236 case CEPH_OSD_OP_ASSERT_SRC_VERSION:
237 case CEPH_OSD_OP_SRC_CMPXATTR:
a9f36c3e 238 case CEPH_OSD_OP_GETXATTR:
4c46459c 239 case CEPH_OSD_OP_GETXATTRS:
a9f36c3e
AE		 240 case CEPH_OSD_OP_CMPXATTR:
241 case CEPH_OSD_OP_SETXATTR:
4c46459c
AE		 242 case CEPH_OSD_OP_SETXATTRS:
243 case CEPH_OSD_OP_RESETXATTRS:
244 case CEPH_OSD_OP_RMXATTR:
245 case CEPH_OSD_OP_PULL:
246 case CEPH_OSD_OP_PUSH:
247 case CEPH_OSD_OP_BALANCEREADS:
248 case CEPH_OSD_OP_UNBALANCEREADS:
249 case CEPH_OSD_OP_SCRUB:
250 case CEPH_OSD_OP_SCRUB_RESERVE:
251 case CEPH_OSD_OP_SCRUB_UNRESERVE:
252 case CEPH_OSD_OP_SCRUB_STOP:
253 case CEPH_OSD_OP_SCRUB_MAP:
254 case CEPH_OSD_OP_WRLOCK:
255 case CEPH_OSD_OP_WRUNLOCK:
256 case CEPH_OSD_OP_RDLOCK:
257 case CEPH_OSD_OP_RDUNLOCK:
258 case CEPH_OSD_OP_UPLOCK:
259 case CEPH_OSD_OP_DNLOCK:
a8dd0a37 260 case CEPH_OSD_OP_CALL:
4c46459c
AE		 261 case CEPH_OSD_OP_PGLS:
262 case CEPH_OSD_OP_PGLS_FILTER:
a8dd0a37
AE		 263 return true;
264 default:
265 return false;
266 }
267}
268
33803f33
AE		 269/*
270 * This is an osd op init function for opcodes that have no data or
271 * other information associated with them. It also serves as a
272 * common init routine for all the other init functions, below.
273 */
274void osd_req_op_init(struct ceph_osd_req_op *op, u16 opcode)
275{
276 BUG_ON(!osd_req_opcode_valid(opcode));
277
278 memset(op, 0, sizeof (*op));
279
280 op->op = opcode;
281}
282
283void osd_req_op_extent_init(struct ceph_osd_req_op *op, u16 opcode,
284 u64 offset, u64 length,
285 u64 truncate_size, u32 truncate_seq)
286{
287 size_t payload_len = 0;
288
289 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE);
290
291 osd_req_op_init(op, opcode);
292
293 op->extent.offset = offset;
294 op->extent.length = length;
295 op->extent.truncate_size = truncate_size;
296 op->extent.truncate_seq = truncate_seq;
297 if (opcode == CEPH_OSD_OP_WRITE)
298 payload_len += length;
299
300 op->payload_len = payload_len;
301}
302EXPORT_SYMBOL(osd_req_op_extent_init);
303
304void osd_req_op_cls_init(struct ceph_osd_req_op *op, u16 opcode,
305 const char *class, const char *method,
306 const void *request_data, size_t request_data_size)
307{
308 size_t payload_len = 0;
309 size_t size;
310
311 BUG_ON(opcode != CEPH_OSD_OP_CALL);
312
313 osd_req_op_init(op, opcode);
314
315 op->cls.class_name = class;
316 size = strlen(class);
317 BUG_ON(size > (size_t) U8_MAX);
318 op->cls.class_len = size;
319 payload_len += size;
320
321 op->cls.method_name = method;
322 size = strlen(method);
323 BUG_ON(size > (size_t) U8_MAX);
324 op->cls.method_len = size;
325 payload_len += size;
326
327 op->cls.indata = request_data;
328 BUG_ON(request_data_size > (size_t) U32_MAX);
329 op->cls.indata_len = (u32) request_data_size;
330 payload_len += request_data_size;
331
332 op->cls.argc = 0; /* currently unused */
333
334 op->payload_len = payload_len;
335}
336EXPORT_SYMBOL(osd_req_op_cls_init);
337
338void osd_req_op_watch_init(struct ceph_osd_req_op *op, u16 opcode,
339 u64 cookie, u64 version, int flag)
340{
341 BUG_ON(opcode != CEPH_OSD_OP_NOTIFY_ACK && opcode != CEPH_OSD_OP_WATCH);
342
343 osd_req_op_init(op, opcode);
344
345 op->watch.cookie = cookie;
346 /* op->watch.ver = version; */ /* XXX 3847 */
347 op->watch.ver = cpu_to_le64(version);
348 if (opcode == CEPH_OSD_OP_WATCH && flag)
349 op->watch.flag = (u8) 1;
350}
351EXPORT_SYMBOL(osd_req_op_watch_init);
352
a8dd0a37
AE		 353static u64 osd_req_encode_op(struct ceph_osd_request *req,
354 struct ceph_osd_op *dst,
355 struct ceph_osd_req_op *src)
356{
357 u64 out_data_len = 0;
358 struct ceph_pagelist *pagelist;
359
360 if (WARN_ON(!osd_req_opcode_valid(src->op))) {
361 pr_err("unrecognized osd opcode %d\n", src->op);
362
363 return 0;
364 }
365
366 switch (src->op) {
367 case CEPH_OSD_OP_STAT:
368 break;
369 case CEPH_OSD_OP_READ:
370 case CEPH_OSD_OP_WRITE:
371 if (src->op == CEPH_OSD_OP_WRITE)
372 out_data_len = src->extent.length;
373 dst->extent.offset = cpu_to_le64(src->extent.offset);
374 dst->extent.length = cpu_to_le64(src->extent.length);
375 dst->extent.truncate_size =
376 cpu_to_le64(src->extent.truncate_size);
377 dst->extent.truncate_seq =
378 cpu_to_le32(src->extent.truncate_seq);
379 break;
380 case CEPH_OSD_OP_CALL:
381 pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
382 BUG_ON(!pagelist);
383 ceph_pagelist_init(pagelist);
384
385 dst->cls.class_len = src->cls.class_len;
386 dst->cls.method_len = src->cls.method_len;
387 dst->cls.indata_len = cpu_to_le32(src->cls.indata_len);
388 ceph_pagelist_append(pagelist, src->cls.class_name,
389 src->cls.class_len);
390 ceph_pagelist_append(pagelist, src->cls.method_name,
391 src->cls.method_len);
392 ceph_pagelist_append(pagelist, src->cls.indata,
393 src->cls.indata_len);
394
395 req->r_data_out.type = CEPH_OSD_DATA_TYPE_PAGELIST;
396 req->r_data_out.pagelist = pagelist;
397 out_data_len = pagelist->length;
398 break;
399 case CEPH_OSD_OP_STARTSYNC:
400 break;
401 case CEPH_OSD_OP_NOTIFY_ACK:
402 case CEPH_OSD_OP_WATCH:
403 dst->watch.cookie = cpu_to_le64(src->watch.cookie);
404 dst->watch.ver = cpu_to_le64(src->watch.ver);
405 dst->watch.flag = src->watch.flag;
406 break;
407 default:
4c46459c 408 pr_err("unsupported osd opcode %s\n",
8f63ca2d 409 ceph_osd_op_name(src->op));
4c46459c 410 WARN_ON(1);
a8dd0a37
AE		 411
412 return 0;
68b4476b 413 }
a8dd0a37 414 dst->op = cpu_to_le16(src->op);
68b4476b 415 dst->payload_len = cpu_to_le32(src->payload_len);
175face2
AE		 416
417 return out_data_len;
68b4476b
YS		 418}
419
3499e8a5
YS		 420/*
421 * build new request AND message
422 *
423 */
424void ceph_osdc_build_request(struct ceph_osd_request *req,
175face2 425 u64 off, unsigned int num_ops,
68b4476b 426 struct ceph_osd_req_op *src_ops,
4d6b250b 427 struct ceph_snap_context *snapc, u64 snap_id,
af77f26c 428 struct timespec *mtime)
3499e8a5
YS		 429{
430 struct ceph_msg *msg = req->r_request;
68b4476b 431 struct ceph_osd_req_op *src_op;
3499e8a5 432 void *p;
1b83bef2 433 size_t msg_size;
3499e8a5 434 int flags = req->r_flags;
f44246e3 435 u64 data_len;
68b4476b 436 int i;
3499e8a5 437
1b83bef2
SW		 438 req->r_num_ops = num_ops;
439 req->r_snapid = snap_id;
f24e9980
SW		 440 req->r_snapc = ceph_get_snap_context(snapc);
441
1b83bef2
SW		 442 /* encode request */
443 msg->hdr.version = cpu_to_le16(4);
444
445 p = msg->front.iov_base;
446 ceph_encode_32(&p, 1); /* client_inc is always 1 */
447 req->r_request_osdmap_epoch = p;
448 p += 4;
449 req->r_request_flags = p;
450 p += 4;
451 if (req->r_flags & CEPH_OSD_FLAG_WRITE)
452 ceph_encode_timespec(p, mtime);
453 p += sizeof(struct ceph_timespec);
454 req->r_request_reassert_version = p;
455 p += sizeof(struct ceph_eversion); /* will get filled in */
456
457 /* oloc */
458 ceph_encode_8(&p, 4);
459 ceph_encode_8(&p, 4);
460 ceph_encode_32(&p, 8 + 4 + 4);
461 req->r_request_pool = p;
462 p += 8;
463 ceph_encode_32(&p, -1); /* preferred */
464 ceph_encode_32(&p, 0); /* key len */
465
466 ceph_encode_8(&p, 1);
467 req->r_request_pgid = p;
468 p += 8 + 4;
469 ceph_encode_32(&p, -1); /* preferred */
470
471 /* oid */
472 ceph_encode_32(&p, req->r_oid_len);
af77f26c 473 memcpy(p, req->r_oid, req->r_oid_len);
1b83bef2 474 dout("oid '%.*s' len %d\n", req->r_oid_len, req->r_oid, req->r_oid_len);
af77f26c 475 p += req->r_oid_len;
f24e9980 476
175face2 477 /* ops--can imply data */
1b83bef2 478 ceph_encode_16(&p, num_ops);
68b4476b 479 src_op = src_ops;
1b83bef2 480 req->r_request_ops = p;
175face2 481 data_len = 0;
1b83bef2 482 for (i = 0; i < num_ops; i++, src_op++) {
175face2 483 data_len += osd_req_encode_op(req, p, src_op);
1b83bef2
SW		 484 p += sizeof(struct ceph_osd_op);
485 }
68b4476b 486
1b83bef2
SW		 487 /* snaps */
488 ceph_encode_64(&p, req->r_snapid);
489 ceph_encode_64(&p, req->r_snapc ? req->r_snapc->seq : 0);
490 ceph_encode_32(&p, req->r_snapc ? req->r_snapc->num_snaps : 0);
491 if (req->r_snapc) {
f24e9980 492 for (i = 0; i < snapc->num_snaps; i++) {
1b83bef2 493 ceph_encode_64(&p, req->r_snapc->snaps[i]);
f24e9980
SW		 494 }
495 }
496
1b83bef2
SW		 497 req->r_request_attempts = p;
498 p += 4;
499
175face2 500 /* data */
0baa1bd9
AE		 501 if (flags & CEPH_OSD_FLAG_WRITE) {
502 u16 data_off;
503
504 /*
505 * The header "data_off" is a hint to the receiver
506 * allowing it to align received data into its
507 * buffers such that there's no need to re-copy
508 * it before writing it to disk (direct I/O).
509 */
510 data_off = (u16) (off & 0xffff);
511 req->r_request->hdr.data_off = cpu_to_le16(data_off);
512 }
f44246e3 513 req->r_request->hdr.data_len = cpu_to_le32(data_len);
c5c6b19d 514
f24e9980 515 BUG_ON(p > msg->front.iov_base + msg->front.iov_len);
6f863e71
SW		 516 msg_size = p - msg->front.iov_base;
517 msg->front.iov_len = msg_size;
518 msg->hdr.front_len = cpu_to_le32(msg_size);
1b83bef2
SW		 519
520 dout("build_request msg_size was %d num_ops %d\n", (int)msg_size,
521 num_ops);
3499e8a5
YS		 522 return;
523}
3d14c5d2 524EXPORT_SYMBOL(ceph_osdc_build_request);
3499e8a5
YS		 525
526/*
527 * build new request AND message, calculate layout, and adjust file
528 * extent as needed.
529 *
530 * if the file was recently truncated, we include information about its
531 * old and new size so that the object can be updated appropriately. (we
532 * avoid synchronously deleting truncated objects because it's slow.)
533 *
534 * if @do_sync, include a 'startsync' command so that the osd will flush
535 * data quickly.
536 */
537struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
538 struct ceph_file_layout *layout,
539 struct ceph_vino vino,
540 u64 off, u64 *plen,
541 int opcode, int flags,
542 struct ceph_snap_context *snapc,
543 int do_sync,
544 u32 truncate_seq,
545 u64 truncate_size,
546 struct timespec *mtime,
153e5167 547 bool use_mempool)
3499e8a5 548{
ae7ca4a3 549 struct ceph_osd_req_op ops[2];
68b4476b 550 struct ceph_osd_request *req;
d18d1e28 551 unsigned int num_op = do_sync ? 2 : 1;
75d1c941
AE		 552 u64 objnum = 0;
553 u64 objoff = 0;
554 u64 objlen = 0;
d18d1e28
AE		 555 u32 object_size;
556 u64 object_base;
6816282d 557 int r;
68b4476b 558
d18d1e28 559 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE);
68b4476b 560
ae7ca4a3
AE		 561 req = ceph_osdc_alloc_request(osdc, snapc, num_op, use_mempool,
562 GFP_NOFS);
4ad12621 563 if (!req)
6816282d 564 return ERR_PTR(-ENOMEM);
d178a9e7 565 req->r_flags = flags;
3499e8a5
YS		 566
567 /* calculate max write size */
a19dadfb 568 r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
3ff5f385
AE		 569 if (r < 0) {
570 ceph_osdc_put_request(req);
6816282d 571 return ERR_PTR(r);
3ff5f385 572 }
a19dadfb 573
d18d1e28
AE		 574 object_size = le32_to_cpu(layout->fl_object_size);
575 object_base = off - objoff;
576 if (truncate_size <= object_base) {
577 truncate_size = 0;
578 } else {
579 truncate_size -= object_base;
580 if (truncate_size > object_size)
581 truncate_size = object_size;
a19dadfb 582 }
d18d1e28 583
b0270324
AE		 584 osd_req_op_extent_init(&ops[0], opcode, objoff, objlen,
585 truncate_size, truncate_seq);
d18d1e28 586 if (do_sync)
b0270324 587 osd_req_op_init(&ops[1], CEPH_OSD_OP_STARTSYNC);
d18d1e28 588
3499e8a5
YS		 589 req->r_file_layout = *layout; /* keep a copy */
590
75d1c941
AE		 591 snprintf(req->r_oid, sizeof(req->r_oid), "%llx.%08llx",
592 vino.ino, objnum);
dbe0fc41
AE		 593 req->r_oid_len = strlen(req->r_oid);
594
175face2 595 ceph_osdc_build_request(req, off, num_op, ops,
ae7ca4a3 596 snapc, vino.snap, mtime);
3499e8a5 597
f24e9980
SW		 598 return req;
599}
3d14c5d2 600EXPORT_SYMBOL(ceph_osdc_new_request);
f24e9980
SW		 601
602/*
603 * We keep osd requests in an rbtree, sorted by ->r_tid.
604 */
605static void __insert_request(struct ceph_osd_client *osdc,
606 struct ceph_osd_request *new)
607{
608 struct rb_node **p = &osdc->requests.rb_node;
609 struct rb_node *parent = NULL;
610 struct ceph_osd_request *req = NULL;
611
612 while (*p) {
613 parent = *p;
614 req = rb_entry(parent, struct ceph_osd_request, r_node);
615 if (new->r_tid < req->r_tid)
616 p = &(*p)->rb_left;
617 else if (new->r_tid > req->r_tid)
618 p = &(*p)->rb_right;
619 else
620 BUG();
621 }
622
623 rb_link_node(&new->r_node, parent, p);
624 rb_insert_color(&new->r_node, &osdc->requests);
625}
626
627static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc,
628 u64 tid)
629{
630 struct ceph_osd_request *req;
631 struct rb_node *n = osdc->requests.rb_node;
632
633 while (n) {
634 req = rb_entry(n, struct ceph_osd_request, r_node);
635 if (tid < req->r_tid)
636 n = n->rb_left;
637 else if (tid > req->r_tid)
638 n = n->rb_right;
639 else
640 return req;
641 }
642 return NULL;
643}
644
645static struct ceph_osd_request *
646__lookup_request_ge(struct ceph_osd_client *osdc,
647 u64 tid)
648{
649 struct ceph_osd_request *req;
650 struct rb_node *n = osdc->requests.rb_node;
651
652 while (n) {
653 req = rb_entry(n, struct ceph_osd_request, r_node);
654 if (tid < req->r_tid) {
655 if (!n->rb_left)
656 return req;
657 n = n->rb_left;
658 } else if (tid > req->r_tid) {
659 n = n->rb_right;
660 } else {
661 return req;
662 }
663 }
664 return NULL;
665}
666
6f6c7006
SW		 667/*
668 * Resubmit requests pending on the given osd.
669 */
670static void __kick_osd_requests(struct ceph_osd_client *osdc,
671 struct ceph_osd *osd)
672{
a40c4f10 673 struct ceph_osd_request *req, *nreq;
e02493c0 674 LIST_HEAD(resend);
6f6c7006
SW		 675 int err;
676
677 dout("__kick_osd_requests osd%d\n", osd->o_osd);
678 err = __reset_osd(osdc, osd);
685a7555 679 if (err)
6f6c7006 680 return;
e02493c0
AE		 681 /*
682 * Build up a list of requests to resend by traversing the
683 * osd's list of requests. Requests for a given object are
684 * sent in tid order, and that is also the order they're
685 * kept on this list. Therefore all requests that are in
686 * flight will be found first, followed by all requests that
687 * have not yet been sent. And to resend requests while
688 * preserving this order we will want to put any sent
689 * requests back on the front of the osd client's unsent
690 * list.
691 *
692 * So we build a separate ordered list of already-sent
693 * requests for the affected osd and splice it onto the
694 * front of the osd client's unsent list. Once we've seen a
695 * request that has not yet been sent we're done. Those
696 * requests are already sitting right where they belong.
697 */
6f6c7006 698 list_for_each_entry(req, &osd->o_requests, r_osd_item) {
e02493c0
AE		 699 if (!req->r_sent)
700 break;
701 list_move_tail(&req->r_req_lru_item, &resend);
702 dout("requeueing %p tid %llu osd%d\n", req, req->r_tid,
6f6c7006 703 osd->o_osd);
a40c4f10
YS		 704 if (!req->r_linger)
705 req->r_flags |= CEPH_OSD_FLAG_RETRY;
706 }
e02493c0 707 list_splice(&resend, &osdc->req_unsent);
a40c4f10 708
e02493c0
AE		 709 /*
710 * Linger requests are re-registered before sending, which
711 * sets up a new tid for each. We add them to the unsent
712 * list at the end to keep things in tid order.
713 */
a40c4f10
YS		 714 list_for_each_entry_safe(req, nreq, &osd->o_linger_requests,
715 r_linger_osd) {
77f38e0e
SW		 716 /*
717 * reregister request prior to unregistering linger so
718 * that r_osd is preserved.
719 */
720 BUG_ON(!list_empty(&req->r_req_lru_item));
a40c4f10 721 __register_request(osdc, req);
e02493c0 722 list_add_tail(&req->r_req_lru_item, &osdc->req_unsent);
ad885927 723 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
77f38e0e 724 __unregister_linger_request(osdc, req);
a40c4f10
YS		 725 dout("requeued lingering %p tid %llu osd%d\n", req, req->r_tid,
726 osd->o_osd);
6f6c7006
SW		 727 }
728}
729
f24e9980 730/*
81b024e7 731 * If the osd connection drops, we need to resubmit all requests.
f24e9980
SW		 732 */
733static void osd_reset(struct ceph_connection *con)
734{
735 struct ceph_osd *osd = con->private;
736 struct ceph_osd_client *osdc;
737
738 if (!osd)
739 return;
740 dout("osd_reset osd%d\n", osd->o_osd);
741 osdc = osd->o_osdc;
f24e9980 742 down_read(&osdc->map_sem);
83aff95e
SW		 743 mutex_lock(&osdc->request_mutex);
744 __kick_osd_requests(osdc, osd);
f9d25199 745 __send_queued(osdc);
83aff95e 746 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 747 up_read(&osdc->map_sem);
748}
749
750/*
751 * Track open sessions with osds.
752 */
e10006f8 753static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
f24e9980
SW		 754{
755 struct ceph_osd *osd;
756
757 osd = kzalloc(sizeof(*osd), GFP_NOFS);
758 if (!osd)
759 return NULL;
760
761 atomic_set(&osd->o_ref, 1);
762 osd->o_osdc = osdc;
e10006f8 763 osd->o_osd = onum;
f407731d 764 RB_CLEAR_NODE(&osd->o_node);
f24e9980 765 INIT_LIST_HEAD(&osd->o_requests);
a40c4f10 766 INIT_LIST_HEAD(&osd->o_linger_requests);
f5a2041b 767 INIT_LIST_HEAD(&osd->o_osd_lru);
f24e9980
SW		 768 osd->o_incarnation = 1;
769
b7a9e5dd 770 ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
4e7a5dcd 771
422d2cb8 772 INIT_LIST_HEAD(&osd->o_keepalive_item);
f24e9980
SW		 773 return osd;
774}
775
776static struct ceph_osd *get_osd(struct ceph_osd *osd)
777{
778 if (atomic_inc_not_zero(&osd->o_ref)) {
779 dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
780 atomic_read(&osd->o_ref));
781 return osd;
782 } else {
783 dout("get_osd %p FAIL\n", osd);
784 return NULL;
785 }
786}
787
788static void put_osd(struct ceph_osd *osd)
789{
790 dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
791 atomic_read(&osd->o_ref) - 1);
a255651d 792 if (atomic_dec_and_test(&osd->o_ref) && osd->o_auth.authorizer) {
79494d1b
SW		 793 struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth;
794
27859f97 795 ceph_auth_destroy_authorizer(ac, osd->o_auth.authorizer);
f24e9980 796 kfree(osd);
79494d1b 797 }
f24e9980
SW		 798}
799
800/*
801 * remove an osd from our map
802 */
f5a2041b 803static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
f24e9980 804{
f5a2041b 805 dout("__remove_osd %p\n", osd);
f24e9980
SW		 806 BUG_ON(!list_empty(&osd->o_requests));
807 rb_erase(&osd->o_node, &osdc->osds);
f5a2041b 808 list_del_init(&osd->o_osd_lru);
f24e9980
SW		 809 ceph_con_close(&osd->o_con);
810 put_osd(osd);
811}
812
aca420bc
SW		 813static void remove_all_osds(struct ceph_osd_client *osdc)
814{
048a9d2d 815 dout("%s %p\n", __func__, osdc);
aca420bc
SW		 816 mutex_lock(&osdc->request_mutex);
817 while (!RB_EMPTY_ROOT(&osdc->osds)) {
818 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
819 struct ceph_osd, o_node);
820 __remove_osd(osdc, osd);
821 }
822 mutex_unlock(&osdc->request_mutex);
823}
824
f5a2041b
YS		 825static void __move_osd_to_lru(struct ceph_osd_client *osdc,
826 struct ceph_osd *osd)
827{
828 dout("__move_osd_to_lru %p\n", osd);
829 BUG_ON(!list_empty(&osd->o_osd_lru));
830 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
3d14c5d2 831 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl * HZ;
f5a2041b
YS		 832}
833
834static void __remove_osd_from_lru(struct ceph_osd *osd)
835{
836 dout("__remove_osd_from_lru %p\n", osd);
837 if (!list_empty(&osd->o_osd_lru))
838 list_del_init(&osd->o_osd_lru);
839}
840
aca420bc 841static void remove_old_osds(struct ceph_osd_client *osdc)
f5a2041b
YS		 842{
843 struct ceph_osd *osd, *nosd;
844
845 dout("__remove_old_osds %p\n", osdc);
846 mutex_lock(&osdc->request_mutex);
847 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
aca420bc 848 if (time_before(jiffies, osd->lru_ttl))
f5a2041b
YS		 849 break;
850 __remove_osd(osdc, osd);
851 }
852 mutex_unlock(&osdc->request_mutex);
853}
854
f24e9980
SW		 855/*
856 * reset osd connect
857 */
f5a2041b 858static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
f24e9980 859{
c3acb181 860 struct ceph_entity_addr *peer_addr;
f24e9980 861
f5a2041b 862 dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
a40c4f10
YS		 863 if (list_empty(&osd->o_requests) &&
864 list_empty(&osd->o_linger_requests)) {
f5a2041b 865 __remove_osd(osdc, osd);
c3acb181
AE		 866
867 return -ENODEV;
868 }
869
870 peer_addr = &osdc->osdmap->osd_addr[osd->o_osd];
871 if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
872 !ceph_con_opened(&osd->o_con)) {
873 struct ceph_osd_request *req;
874
87b315a5
SW		 875 dout(" osd addr hasn't changed and connection never opened,"
876 " letting msgr retry");
877 /* touch each r_stamp for handle_timeout()'s benfit */
878 list_for_each_entry(req, &osd->o_requests, r_osd_item)
879 req->r_stamp = jiffies;
c3acb181
AE		 880
881 return -EAGAIN;
f24e9980 882 }
c3acb181
AE		 883
884 ceph_con_close(&osd->o_con);
885 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
886 osd->o_incarnation++;
887
888 return 0;
f24e9980
SW		 889}
890
891static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new)
892{
893 struct rb_node **p = &osdc->osds.rb_node;
894 struct rb_node *parent = NULL;
895 struct ceph_osd *osd = NULL;
896
aca420bc 897 dout("__insert_osd %p osd%d\n", new, new->o_osd);
f24e9980
SW		 898 while (*p) {
899 parent = *p;
900 osd = rb_entry(parent, struct ceph_osd, o_node);
901 if (new->o_osd < osd->o_osd)
902 p = &(*p)->rb_left;
903 else if (new->o_osd > osd->o_osd)
904 p = &(*p)->rb_right;
905 else
906 BUG();
907 }
908
909 rb_link_node(&new->o_node, parent, p);
910 rb_insert_color(&new->o_node, &osdc->osds);
911}
912
913static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o)
914{
915 struct ceph_osd *osd;
916 struct rb_node *n = osdc->osds.rb_node;
917
918 while (n) {
919 osd = rb_entry(n, struct ceph_osd, o_node);
920 if (o < osd->o_osd)
921 n = n->rb_left;
922 else if (o > osd->o_osd)
923 n = n->rb_right;
924 else
925 return osd;
926 }
927 return NULL;
928}
929
422d2cb8
YS		 930static void __schedule_osd_timeout(struct ceph_osd_client *osdc)
931{
932 schedule_delayed_work(&osdc->timeout_work,
3d14c5d2 933 osdc->client->options->osd_keepalive_timeout * HZ);
422d2cb8
YS		 934}
935
936static void __cancel_osd_timeout(struct ceph_osd_client *osdc)
937{
938 cancel_delayed_work(&osdc->timeout_work);
939}
f24e9980
SW		 940
941/*
942 * Register request, assign tid. If this is the first request, set up
943 * the timeout event.
944 */
a40c4f10
YS		 945static void __register_request(struct ceph_osd_client *osdc,
946 struct ceph_osd_request *req)
f24e9980 947{
f24e9980 948 req->r_tid = ++osdc->last_tid;
6df058c0 949 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
77f38e0e 950 dout("__register_request %p tid %lld\n", req, req->r_tid);
f24e9980
SW		 951 __insert_request(osdc, req);
952 ceph_osdc_get_request(req);
953 osdc->num_requests++;
f24e9980 954 if (osdc->num_requests == 1) {
422d2cb8
YS		 955 dout(" first request, scheduling timeout\n");
956 __schedule_osd_timeout(osdc);
f24e9980 957 }
a40c4f10
YS		 958}
959
f24e9980
SW		 960/*
961 * called under osdc->request_mutex
962 */
963static void __unregister_request(struct ceph_osd_client *osdc,
964 struct ceph_osd_request *req)
965{
35f9f8a0
SW		 966 if (RB_EMPTY_NODE(&req->r_node)) {
967 dout("__unregister_request %p tid %lld not registered\n",
968 req, req->r_tid);
969 return;
970 }
971
f24e9980
SW		 972 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
973 rb_erase(&req->r_node, &osdc->requests);
974 osdc->num_requests--;
975
0ba6478d
SW		 976 if (req->r_osd) {
977 /* make sure the original request isn't in flight. */
6740a845 978 ceph_msg_revoke(req->r_request);
0ba6478d
SW		 979
980 list_del_init(&req->r_osd_item);
a40c4f10
YS		 981 if (list_empty(&req->r_osd->o_requests) &&
982 list_empty(&req->r_osd->o_linger_requests)) {
983 dout("moving osd to %p lru\n", req->r_osd);
f5a2041b 984 __move_osd_to_lru(osdc, req->r_osd);
a40c4f10 985 }
fbdb9190 986 if (list_empty(&req->r_linger_item))
a40c4f10 987 req->r_osd = NULL;
0ba6478d 988 }
f24e9980 989
7d5f2481 990 list_del_init(&req->r_req_lru_item);
f24e9980
SW		 991 ceph_osdc_put_request(req);
992
422d2cb8
YS		 993 if (osdc->num_requests == 0) {
994 dout(" no requests, canceling timeout\n");
995 __cancel_osd_timeout(osdc);
f24e9980
SW		 996 }
997}
998
999/*
1000 * Cancel a previously queued request message
1001 */
1002static void __cancel_request(struct ceph_osd_request *req)
1003{
6bc18876 1004 if (req->r_sent && req->r_osd) {
6740a845 1005 ceph_msg_revoke(req->r_request);
f24e9980
SW		 1006 req->r_sent = 0;
1007 }
1008}
1009
a40c4f10
YS		 1010static void __register_linger_request(struct ceph_osd_client *osdc,
1011 struct ceph_osd_request *req)
1012{
1013 dout("__register_linger_request %p\n", req);
1014 list_add_tail(&req->r_linger_item, &osdc->req_linger);
6194ea89
SW		 1015 if (req->r_osd)
1016 list_add_tail(&req->r_linger_osd,
1017 &req->r_osd->o_linger_requests);
a40c4f10
YS		 1018}
1019
1020static void __unregister_linger_request(struct ceph_osd_client *osdc,
1021 struct ceph_osd_request *req)
1022{
1023 dout("__unregister_linger_request %p\n", req);
61c74035 1024 list_del_init(&req->r_linger_item);
a40c4f10 1025 if (req->r_osd) {
a40c4f10
YS		 1026 list_del_init(&req->r_linger_osd);
1027
1028 if (list_empty(&req->r_osd->o_requests) &&
1029 list_empty(&req->r_osd->o_linger_requests)) {
1030 dout("moving osd to %p lru\n", req->r_osd);
1031 __move_osd_to_lru(osdc, req->r_osd);
1032 }
fbdb9190
SW		 1033 if (list_empty(&req->r_osd_item))
1034 req->r_osd = NULL;
a40c4f10
YS		 1035 }
1036}
1037
1038void ceph_osdc_unregister_linger_request(struct ceph_osd_client *osdc,
1039 struct ceph_osd_request *req)
1040{
1041 mutex_lock(&osdc->request_mutex);
1042 if (req->r_linger) {
1043 __unregister_linger_request(osdc, req);
1044 ceph_osdc_put_request(req);
1045 }
1046 mutex_unlock(&osdc->request_mutex);
1047}
1048EXPORT_SYMBOL(ceph_osdc_unregister_linger_request);
1049
1050void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
1051 struct ceph_osd_request *req)
1052{
1053 if (!req->r_linger) {
1054 dout("set_request_linger %p\n", req);
1055 req->r_linger = 1;
1056 /*
1057 * caller is now responsible for calling
1058 * unregister_linger_request
1059 */
1060 ceph_osdc_get_request(req);
1061 }
1062}
1063EXPORT_SYMBOL(ceph_osdc_set_request_linger);
1064
f24e9980
SW		 1065/*
1066 * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
1067 * (as needed), and set the request r_osd appropriately. If there is
25985edc 1068 * no up osd, set r_osd to NULL. Move the request to the appropriate list
6f6c7006 1069 * (unsent, homeless) or leave on in-flight lru.
f24e9980
SW		 1070 *
1071 * Return 0 if unchanged, 1 if changed, or negative on error.
1072 *
1073 * Caller should hold map_sem for read and request_mutex.
1074 */
6f6c7006 1075static int __map_request(struct ceph_osd_client *osdc,
38d6453c 1076 struct ceph_osd_request *req, int force_resend)
f24e9980 1077{
5b191d99 1078 struct ceph_pg pgid;
d85b7056
SW		 1079 int acting[CEPH_PG_MAX_SIZE];
1080 int o = -1, num = 0;
f24e9980 1081 int err;
f24e9980 1082
6f6c7006 1083 dout("map_request %p tid %lld\n", req, req->r_tid);
41766f87
AE		 1084 err = ceph_calc_ceph_pg(&pgid, req->r_oid, osdc->osdmap,
1085 ceph_file_layout_pg_pool(req->r_file_layout));
6f6c7006
SW		 1086 if (err) {
1087 list_move(&req->r_req_lru_item, &osdc->req_notarget);
f24e9980 1088 return err;
6f6c7006 1089 }
7740a42f
SW		 1090 req->r_pgid = pgid;
1091
d85b7056
SW		 1092 err = ceph_calc_pg_acting(osdc->osdmap, pgid, acting);
1093 if (err > 0) {
1094 o = acting[0];
1095 num = err;
1096 }
f24e9980 1097
38d6453c
SW		 1098 if ((!force_resend &&
1099 req->r_osd && req->r_osd->o_osd == o &&
d85b7056
SW		 1100 req->r_sent >= req->r_osd->o_incarnation &&
1101 req->r_num_pg_osds == num &&
1102 memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
f24e9980
SW		 1103 (req->r_osd == NULL && o == -1))
1104 return 0; /* no change */
1105
5b191d99
SW		 1106 dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n",
1107 req->r_tid, pgid.pool, pgid.seed, o,
f24e9980
SW		 1108 req->r_osd ? req->r_osd->o_osd : -1);
1109
d85b7056
SW		 1110 /* record full pg acting set */
1111 memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
1112 req->r_num_pg_osds = num;
1113
f24e9980
SW		 1114 if (req->r_osd) {
1115 __cancel_request(req);
1116 list_del_init(&req->r_osd_item);
f24e9980
SW		 1117 req->r_osd = NULL;
1118 }
1119
1120 req->r_osd = __lookup_osd(osdc, o);
1121 if (!req->r_osd && o >= 0) {
c99eb1c7 1122 err = -ENOMEM;
e10006f8 1123 req->r_osd = create_osd(osdc, o);
6f6c7006
SW		 1124 if (!req->r_osd) {
1125 list_move(&req->r_req_lru_item, &osdc->req_notarget);
c99eb1c7 1126 goto out;
6f6c7006 1127 }
f24e9980 1128
6f6c7006 1129 dout("map_request osd %p is osd%d\n", req->r_osd, o);
f24e9980
SW		 1130 __insert_osd(osdc, req->r_osd);
1131
b7a9e5dd
SW		 1132 ceph_con_open(&req->r_osd->o_con,
1133 CEPH_ENTITY_TYPE_OSD, o,
1134 &osdc->osdmap->osd_addr[o]);
f24e9980
SW		 1135 }
1136
f5a2041b
YS		 1137 if (req->r_osd) {
1138 __remove_osd_from_lru(req->r_osd);
ad885927
AE		 1139 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
1140 list_move_tail(&req->r_req_lru_item, &osdc->req_unsent);
6f6c7006 1141 } else {
ad885927 1142 list_move_tail(&req->r_req_lru_item, &osdc->req_notarget);
f5a2041b 1143 }
d85b7056 1144 err = 1; /* osd or pg changed */
f24e9980
SW		 1145
1146out:
f24e9980
SW		 1147 return err;
1148}
1149
1150/*
1151 * caller should hold map_sem (for read) and request_mutex
1152 */
56e925b6
SW		 1153static void __send_request(struct ceph_osd_client *osdc,
1154 struct ceph_osd_request *req)
f24e9980 1155{
1b83bef2 1156 void *p;
f24e9980 1157
1b83bef2
SW		 1158 dout("send_request %p tid %llu to osd%d flags %d pg %lld.%x\n",
1159 req, req->r_tid, req->r_osd->o_osd, req->r_flags,
1160 (unsigned long long)req->r_pgid.pool, req->r_pgid.seed);
1161
1162 /* fill in message content that changes each time we send it */
1163 put_unaligned_le32(osdc->osdmap->epoch, req->r_request_osdmap_epoch);
1164 put_unaligned_le32(req->r_flags, req->r_request_flags);
1165 put_unaligned_le64(req->r_pgid.pool, req->r_request_pool);
1166 p = req->r_request_pgid;
1167 ceph_encode_64(&p, req->r_pgid.pool);
1168 ceph_encode_32(&p, req->r_pgid.seed);
1169 put_unaligned_le64(1, req->r_request_attempts); /* FIXME */
1170 memcpy(req->r_request_reassert_version, &req->r_reassert_version,
1171 sizeof(req->r_reassert_version));
2169aea6 1172
3dd72fc0 1173 req->r_stamp = jiffies;
07a27e22 1174 list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
f24e9980
SW		 1175
1176 ceph_msg_get(req->r_request); /* send consumes a ref */
1177 ceph_con_send(&req->r_osd->o_con, req->r_request);
1178 req->r_sent = req->r_osd->o_incarnation;
f24e9980
SW		 1179}
1180
6f6c7006
SW		 1181/*
1182 * Send any requests in the queue (req_unsent).
1183 */
f9d25199 1184static void __send_queued(struct ceph_osd_client *osdc)
6f6c7006
SW		 1185{
1186 struct ceph_osd_request *req, *tmp;
1187
f9d25199
AE		 1188 dout("__send_queued\n");
1189 list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item)
6f6c7006 1190 __send_request(osdc, req);
6f6c7006
SW		 1191}
1192
f24e9980
SW		 1193/*
1194 * Timeout callback, called every N seconds when 1 or more osd
1195 * requests has been active for more than N seconds. When this
1196 * happens, we ping all OSDs with requests who have timed out to
1197 * ensure any communications channel reset is detected. Reset the
1198 * request timeouts another N seconds in the future as we go.
1199 * Reschedule the timeout event another N seconds in future (unless
1200 * there are no open requests).
1201 */
1202static void handle_timeout(struct work_struct *work)
1203{
1204 struct ceph_osd_client *osdc =
1205 container_of(work, struct ceph_osd_client, timeout_work.work);
83aff95e 1206 struct ceph_osd_request *req;
f24e9980 1207 struct ceph_osd *osd;
422d2cb8 1208 unsigned long keepalive =
3d14c5d2 1209 osdc->client->options->osd_keepalive_timeout * HZ;
422d2cb8 1210 struct list_head slow_osds;
f24e9980
SW		 1211 dout("timeout\n");
1212 down_read(&osdc->map_sem);
1213
1214 ceph_monc_request_next_osdmap(&osdc->client->monc);
1215
1216 mutex_lock(&osdc->request_mutex);
f24e9980 1217
422d2cb8
YS		 1218 /*
1219 * ping osds that are a bit slow. this ensures that if there
1220 * is a break in the TCP connection we will notice, and reopen
1221 * a connection with that osd (from the fault callback).
1222 */
1223 INIT_LIST_HEAD(&slow_osds);
1224 list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
3dd72fc0 1225 if (time_before(jiffies, req->r_stamp + keepalive))
422d2cb8
YS		 1226 break;
1227
1228 osd = req->r_osd;
1229 BUG_ON(!osd);
1230 dout(" tid %llu is slow, will send keepalive on osd%d\n",
f24e9980 1231 req->r_tid, osd->o_osd);
422d2cb8
YS		 1232 list_move_tail(&osd->o_keepalive_item, &slow_osds);
1233 }
1234 while (!list_empty(&slow_osds)) {
1235 osd = list_entry(slow_osds.next, struct ceph_osd,
1236 o_keepalive_item);
1237 list_del_init(&osd->o_keepalive_item);
f24e9980
SW		 1238 ceph_con_keepalive(&osd->o_con);
1239 }
1240
422d2cb8 1241 __schedule_osd_timeout(osdc);
f9d25199 1242 __send_queued(osdc);
f24e9980 1243 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 1244 up_read(&osdc->map_sem);
1245}
1246
f5a2041b
YS		 1247static void handle_osds_timeout(struct work_struct *work)
1248{
1249 struct ceph_osd_client *osdc =
1250 container_of(work, struct ceph_osd_client,
1251 osds_timeout_work.work);
1252 unsigned long delay =
3d14c5d2 1253 osdc->client->options->osd_idle_ttl * HZ >> 2;
f5a2041b
YS		 1254
1255 dout("osds timeout\n");
1256 down_read(&osdc->map_sem);
aca420bc 1257 remove_old_osds(osdc);
f5a2041b
YS		 1258 up_read(&osdc->map_sem);
1259
1260 schedule_delayed_work(&osdc->osds_timeout_work,
1261 round_jiffies_relative(delay));
1262}
1263
25845472
SW		 1264static void complete_request(struct ceph_osd_request *req)
1265{
1266 if (req->r_safe_callback)
1267 req->r_safe_callback(req, NULL);
1268 complete_all(&req->r_safe_completion); /* fsync waiter */
1269}
1270
f24e9980
SW		 1271/*
1272 * handle osd op reply. either call the callback if it is specified,
1273 * or do the completion to wake up the waiting thread.
1274 */
350b1c32
SW		 1275static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg,
1276 struct ceph_connection *con)
f24e9980 1277{
1b83bef2 1278 void *p, *end;
f24e9980
SW		 1279 struct ceph_osd_request *req;
1280 u64 tid;
1b83bef2
SW		 1281 int object_len;
1282 int numops, payload_len, flags;
0ceed5db 1283 s32 result;
1b83bef2
SW		 1284 s32 retry_attempt;
1285 struct ceph_pg pg;
1286 int err;
1287 u32 reassert_epoch;
1288 u64 reassert_version;
1289 u32 osdmap_epoch;
0d5af164 1290 int already_completed;
1b83bef2 1291 int i;
f24e9980 1292
6df058c0 1293 tid = le64_to_cpu(msg->hdr.tid);
1b83bef2
SW		 1294 dout("handle_reply %p tid %llu\n", msg, tid);
1295
1296 p = msg->front.iov_base;
1297 end = p + msg->front.iov_len;
1298
1299 ceph_decode_need(&p, end, 4, bad);
1300 object_len = ceph_decode_32(&p);
1301 ceph_decode_need(&p, end, object_len, bad);
1302 p += object_len;
1303
ef4859d6 1304 err = ceph_decode_pgid(&p, end, &pg);
1b83bef2 1305 if (err)
f24e9980 1306 goto bad;
1b83bef2
SW		 1307
1308 ceph_decode_need(&p, end, 8 + 4 + 4 + 8 + 4, bad);
1309 flags = ceph_decode_64(&p);
1310 result = ceph_decode_32(&p);
1311 reassert_epoch = ceph_decode_32(&p);
1312 reassert_version = ceph_decode_64(&p);
1313 osdmap_epoch = ceph_decode_32(&p);
1314
f24e9980
SW		 1315 /* lookup */
1316 mutex_lock(&osdc->request_mutex);
1317 req = __lookup_request(osdc, tid);
1318 if (req == NULL) {
1319 dout("handle_reply tid %llu dne\n", tid);
8058fd45 1320 goto bad_mutex;
f24e9980
SW		 1321 }
1322 ceph_osdc_get_request(req);
1b83bef2
SW		 1323
1324 dout("handle_reply %p tid %llu req %p result %d\n", msg, tid,
1325 req, result);
1326
1327 ceph_decode_need(&p, end, 4, bad);
1328 numops = ceph_decode_32(&p);
1329 if (numops > CEPH_OSD_MAX_OP)
1330 goto bad_put;
1331 if (numops != req->r_num_ops)
1332 goto bad_put;
1333 payload_len = 0;
1334 ceph_decode_need(&p, end, numops * sizeof(struct ceph_osd_op), bad);
1335 for (i = 0; i < numops; i++) {
1336 struct ceph_osd_op *op = p;
1337 int len;
1338
1339 len = le32_to_cpu(op->payload_len);
1340 req->r_reply_op_len[i] = len;
1341 dout(" op %d has %d bytes\n", i, len);
1342 payload_len += len;
1343 p += sizeof(*op);
1344 }
1345 if (payload_len != le32_to_cpu(msg->hdr.data_len)) {
1346 pr_warning("sum of op payload lens %d != data_len %d",
1347 payload_len, le32_to_cpu(msg->hdr.data_len));
1348 goto bad_put;
1349 }
1350
1351 ceph_decode_need(&p, end, 4 + numops * 4, bad);
1352 retry_attempt = ceph_decode_32(&p);
1353 for (i = 0; i < numops; i++)
1354 req->r_reply_op_result[i] = ceph_decode_32(&p);
f24e9980 1355
350b1c32 1356 /*
0d59ab81 1357 * if this connection filled our message, drop our reference now, to
350b1c32
SW		 1358 * avoid a (safe but slower) revoke later.
1359 */
0d59ab81 1360 if (req->r_con_filling_msg == con && req->r_reply == msg) {
c16e7869 1361 dout(" dropping con_filling_msg ref %p\n", con);
0d59ab81 1362 req->r_con_filling_msg = NULL;
0d47766f 1363 con->ops->put(con);
350b1c32
SW		 1364 }
1365
f24e9980 1366 if (!req->r_got_reply) {
95c96174 1367 unsigned int bytes;
f24e9980 1368
1b83bef2 1369 req->r_result = result;
f24e9980
SW		 1370 bytes = le32_to_cpu(msg->hdr.data_len);
1371 dout("handle_reply result %d bytes %d\n", req->r_result,
1372 bytes);
1373 if (req->r_result == 0)
1374 req->r_result = bytes;
1375
1376 /* in case this is a write and we need to replay, */
1b83bef2
SW		 1377 req->r_reassert_version.epoch = cpu_to_le32(reassert_epoch);
1378 req->r_reassert_version.version = cpu_to_le64(reassert_version);
f24e9980
SW		 1379
1380 req->r_got_reply = 1;
1381 } else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) {
1382 dout("handle_reply tid %llu dup ack\n", tid);
34b43a56 1383 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 1384 goto done;
1385 }
1386
1387 dout("handle_reply tid %llu flags %d\n", tid, flags);
1388
a40c4f10
YS		 1389 if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK))
1390 __register_linger_request(osdc, req);
1391
f24e9980 1392 /* either this is a read, or we got the safe response */
0ceed5db
SW		 1393 if (result < 0 ||
1394 (flags & CEPH_OSD_FLAG_ONDISK) ||
f24e9980
SW		 1395 ((flags & CEPH_OSD_FLAG_WRITE) == 0))
1396 __unregister_request(osdc, req);
1397
0d5af164
AE		 1398 already_completed = req->r_completed;
1399 req->r_completed = 1;
f24e9980 1400 mutex_unlock(&osdc->request_mutex);
0d5af164
AE		 1401 if (already_completed)
1402 goto done;
f24e9980
SW		 1403
1404 if (req->r_callback)
1405 req->r_callback(req, msg);
1406 else
03066f23 1407 complete_all(&req->r_completion);
f24e9980 1408
25845472
SW		 1409 if (flags & CEPH_OSD_FLAG_ONDISK)
1410 complete_request(req);
f24e9980
SW		 1411
1412done:
a40c4f10 1413 dout("req=%p req->r_linger=%d\n", req, req->r_linger);
f24e9980
SW		 1414 ceph_osdc_put_request(req);
1415 return;
1416
1b83bef2
SW		 1417bad_put:
1418 ceph_osdc_put_request(req);
8058fd45
AE		 1419bad_mutex:
1420 mutex_unlock(&osdc->request_mutex);
f24e9980 1421bad:
1b83bef2
SW		 1422 pr_err("corrupt osd_op_reply got %d %d\n",
1423 (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len));
9ec7cab1 1424 ceph_msg_dump(msg);
f24e9980
SW		 1425}
1426
6f6c7006 1427static void reset_changed_osds(struct ceph_osd_client *osdc)
f24e9980 1428{
f24e9980 1429 struct rb_node *p, *n;
f24e9980 1430
6f6c7006
SW		 1431 for (p = rb_first(&osdc->osds); p; p = n) {
1432 struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
f24e9980 1433
6f6c7006
SW		 1434 n = rb_next(p);
1435 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
1436 memcmp(&osd->o_con.peer_addr,
1437 ceph_osd_addr(osdc->osdmap,
1438 osd->o_osd),
1439 sizeof(struct ceph_entity_addr)) != 0)
1440 __reset_osd(osdc, osd);
f24e9980 1441 }
422d2cb8
YS		 1442}
1443
1444/*
6f6c7006
SW		 1445 * Requeue requests whose mapping to an OSD has changed. If requests map to
1446 * no osd, request a new map.
422d2cb8 1447 *
e6d50f67 1448 * Caller should hold map_sem for read.
422d2cb8 1449 */
38d6453c 1450static void kick_requests(struct ceph_osd_client *osdc, int force_resend)
422d2cb8 1451{
a40c4f10 1452 struct ceph_osd_request *req, *nreq;
6f6c7006
SW		 1453 struct rb_node *p;
1454 int needmap = 0;
1455 int err;
422d2cb8 1456
38d6453c 1457 dout("kick_requests %s\n", force_resend ? " (force resend)" : "");
422d2cb8 1458 mutex_lock(&osdc->request_mutex);
6194ea89 1459 for (p = rb_first(&osdc->requests); p; ) {
6f6c7006 1460 req = rb_entry(p, struct ceph_osd_request, r_node);
6194ea89 1461 p = rb_next(p);
ab60b16d
AE		 1462
1463 /*
1464 * For linger requests that have not yet been
1465 * registered, move them to the linger list; they'll
1466 * be sent to the osd in the loop below. Unregister
1467 * the request before re-registering it as a linger
1468 * request to ensure the __map_request() below
1469 * will decide it needs to be sent.
1470 */
1471 if (req->r_linger && list_empty(&req->r_linger_item)) {
1472 dout("%p tid %llu restart on osd%d\n",
1473 req, req->r_tid,
1474 req->r_osd ? req->r_osd->o_osd : -1);
1475 __unregister_request(osdc, req);
1476 __register_linger_request(osdc, req);
1477 continue;
1478 }
1479
38d6453c 1480 err = __map_request(osdc, req, force_resend);
6f6c7006
SW		 1481 if (err < 0)
1482 continue; /* error */
1483 if (req->r_osd == NULL) {
1484 dout("%p tid %llu maps to no osd\n", req, req->r_tid);
1485 needmap++; /* request a newer map */
1486 } else if (err > 0) {
6194ea89
SW		 1487 if (!req->r_linger) {
1488 dout("%p tid %llu requeued on osd%d\n", req,
1489 req->r_tid,
1490 req->r_osd ? req->r_osd->o_osd : -1);
a40c4f10 1491 req->r_flags |= CEPH_OSD_FLAG_RETRY;
6194ea89
SW		 1492 }
1493 }
a40c4f10
YS		 1494 }
1495
1496 list_for_each_entry_safe(req, nreq, &osdc->req_linger,
1497 r_linger_item) {
1498 dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
1499
38d6453c 1500 err = __map_request(osdc, req, force_resend);
ab60b16d 1501 dout("__map_request returned %d\n", err);
a40c4f10
YS		 1502 if (err == 0)
1503 continue; /* no change and no osd was specified */
1504 if (err < 0)
1505 continue; /* hrm! */
1506 if (req->r_osd == NULL) {
1507 dout("tid %llu maps to no valid osd\n", req->r_tid);
1508 needmap++; /* request a newer map */
1509 continue;
6f6c7006 1510 }
a40c4f10
YS		 1511
1512 dout("kicking lingering %p tid %llu osd%d\n", req, req->r_tid,
1513 req->r_osd ? req->r_osd->o_osd : -1);
a40c4f10 1514 __register_request(osdc, req);
c89ce05e 1515 __unregister_linger_request(osdc, req);
6f6c7006 1516 }
f24e9980
SW		 1517 mutex_unlock(&osdc->request_mutex);
1518
1519 if (needmap) {
1520 dout("%d requests for down osds, need new map\n", needmap);
1521 ceph_monc_request_next_osdmap(&osdc->client->monc);
1522 }
e6d50f67 1523 reset_changed_osds(osdc);
422d2cb8 1524}
6f6c7006
SW		 1525
1526
f24e9980
SW		 1527/*
1528 * Process updated osd map.
1529 *
1530 * The message contains any number of incremental and full maps, normally
1531 * indicating some sort of topology change in the cluster. Kick requests
1532 * off to different OSDs as needed.
1533 */
1534void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1535{
1536 void *p, *end, *next;
1537 u32 nr_maps, maplen;
1538 u32 epoch;
1539 struct ceph_osdmap *newmap = NULL, *oldmap;
1540 int err;
1541 struct ceph_fsid fsid;
1542
1543 dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
1544 p = msg->front.iov_base;
1545 end = p + msg->front.iov_len;
1546
1547 /* verify fsid */
1548 ceph_decode_need(&p, end, sizeof(fsid), bad);
1549 ceph_decode_copy(&p, &fsid, sizeof(fsid));
0743304d
SW		 1550 if (ceph_check_fsid(osdc->client, &fsid) < 0)
1551 return;
f24e9980
SW		 1552
1553 down_write(&osdc->map_sem);
1554
1555 /* incremental maps */
1556 ceph_decode_32_safe(&p, end, nr_maps, bad);
1557 dout(" %d inc maps\n", nr_maps);
1558 while (nr_maps > 0) {
1559 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 1560 epoch = ceph_decode_32(&p);
1561 maplen = ceph_decode_32(&p);
f24e9980
SW		 1562 ceph_decode_need(&p, end, maplen, bad);
1563 next = p + maplen;
1564 if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
1565 dout("applying incremental map %u len %d\n",
1566 epoch, maplen);
1567 newmap = osdmap_apply_incremental(&p, next,
1568 osdc->osdmap,
15d9882c 1569 &osdc->client->msgr);
f24e9980
SW		 1570 if (IS_ERR(newmap)) {
1571 err = PTR_ERR(newmap);
1572 goto bad;
1573 }
30dc6381 1574 BUG_ON(!newmap);
f24e9980
SW		 1575 if (newmap != osdc->osdmap) {
1576 ceph_osdmap_destroy(osdc->osdmap);
1577 osdc->osdmap = newmap;
1578 }
38d6453c 1579 kick_requests(osdc, 0);
f24e9980
SW		 1580 } else {
1581 dout("ignoring incremental map %u len %d\n",
1582 epoch, maplen);
1583 }
1584 p = next;
1585 nr_maps--;
1586 }
1587 if (newmap)
1588 goto done;
1589
1590 /* full maps */
1591 ceph_decode_32_safe(&p, end, nr_maps, bad);
1592 dout(" %d full maps\n", nr_maps);
1593 while (nr_maps) {
1594 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 1595 epoch = ceph_decode_32(&p);
1596 maplen = ceph_decode_32(&p);
f24e9980
SW		 1597 ceph_decode_need(&p, end, maplen, bad);
1598 if (nr_maps > 1) {
1599 dout("skipping non-latest full map %u len %d\n",
1600 epoch, maplen);
1601 } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
1602 dout("skipping full map %u len %d, "
1603 "older than our %u\n", epoch, maplen,
1604 osdc->osdmap->epoch);
1605 } else {
38d6453c
SW		 1606 int skipped_map = 0;
1607
f24e9980
SW		 1608 dout("taking full map %u len %d\n", epoch, maplen);
1609 newmap = osdmap_decode(&p, p+maplen);
1610 if (IS_ERR(newmap)) {
1611 err = PTR_ERR(newmap);
1612 goto bad;
1613 }
30dc6381 1614 BUG_ON(!newmap);
f24e9980
SW		 1615 oldmap = osdc->osdmap;
1616 osdc->osdmap = newmap;
38d6453c
SW		 1617 if (oldmap) {
1618 if (oldmap->epoch + 1 < newmap->epoch)
1619 skipped_map = 1;
f24e9980 1620 ceph_osdmap_destroy(oldmap);
38d6453c
SW		 1621 }
1622 kick_requests(osdc, skipped_map);
f24e9980
SW		 1623 }
1624 p += maplen;
1625 nr_maps--;
1626 }
1627
1628done:
1629 downgrade_write(&osdc->map_sem);
1630 ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch);
cd634fb6
SW		 1631
1632 /*
1633 * subscribe to subsequent osdmap updates if full to ensure
1634 * we find out when we are no longer full and stop returning
1635 * ENOSPC.
1636 */
1637 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL))
1638 ceph_monc_request_next_osdmap(&osdc->client->monc);
1639
f9d25199
AE		 1640 mutex_lock(&osdc->request_mutex);
1641 __send_queued(osdc);
1642 mutex_unlock(&osdc->request_mutex);
f24e9980 1643 up_read(&osdc->map_sem);
03066f23 1644 wake_up_all(&osdc->client->auth_wq);
f24e9980
SW		 1645 return;
1646
1647bad:
1648 pr_err("osdc handle_map corrupt msg\n");
9ec7cab1 1649 ceph_msg_dump(msg);
f24e9980
SW		 1650 up_write(&osdc->map_sem);
1651 return;
1652}
1653
a40c4f10
YS		 1654/*
1655 * watch/notify callback event infrastructure
1656 *
1657 * These callbacks are used both for watch and notify operations.
1658 */
1659static void __release_event(struct kref *kref)
1660{
1661 struct ceph_osd_event *event =
1662 container_of(kref, struct ceph_osd_event, kref);
1663
1664 dout("__release_event %p\n", event);
1665 kfree(event);
1666}
1667
1668static void get_event(struct ceph_osd_event *event)
1669{
1670 kref_get(&event->kref);
1671}
1672
1673void ceph_osdc_put_event(struct ceph_osd_event *event)
1674{
1675 kref_put(&event->kref, __release_event);
1676}
1677EXPORT_SYMBOL(ceph_osdc_put_event);
1678
1679static void __insert_event(struct ceph_osd_client *osdc,
1680 struct ceph_osd_event *new)
1681{
1682 struct rb_node **p = &osdc->event_tree.rb_node;
1683 struct rb_node *parent = NULL;
1684 struct ceph_osd_event *event = NULL;
1685
1686 while (*p) {
1687 parent = *p;
1688 event = rb_entry(parent, struct ceph_osd_event, node);
1689 if (new->cookie < event->cookie)
1690 p = &(*p)->rb_left;
1691 else if (new->cookie > event->cookie)
1692 p = &(*p)->rb_right;
1693 else
1694 BUG();
1695 }
1696
1697 rb_link_node(&new->node, parent, p);
1698 rb_insert_color(&new->node, &osdc->event_tree);
1699}
1700
1701static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
1702 u64 cookie)
1703{
1704 struct rb_node **p = &osdc->event_tree.rb_node;
1705 struct rb_node *parent = NULL;
1706 struct ceph_osd_event *event = NULL;
1707
1708 while (*p) {
1709 parent = *p;
1710 event = rb_entry(parent, struct ceph_osd_event, node);
1711 if (cookie < event->cookie)
1712 p = &(*p)->rb_left;
1713 else if (cookie > event->cookie)
1714 p = &(*p)->rb_right;
1715 else
1716 return event;
1717 }
1718 return NULL;
1719}
1720
1721static void __remove_event(struct ceph_osd_event *event)
1722{
1723 struct ceph_osd_client *osdc = event->osdc;
1724
1725 if (!RB_EMPTY_NODE(&event->node)) {
1726 dout("__remove_event removed %p\n", event);
1727 rb_erase(&event->node, &osdc->event_tree);
1728 ceph_osdc_put_event(event);
1729 } else {
1730 dout("__remove_event didn't remove %p\n", event);
1731 }
1732}
1733
1734int ceph_osdc_create_event(struct ceph_osd_client *osdc,
1735 void (*event_cb)(u64, u64, u8, void *),
3c663bbd 1736 void *data, struct ceph_osd_event **pevent)
a40c4f10
YS		 1737{
1738 struct ceph_osd_event *event;
1739
1740 event = kmalloc(sizeof(*event), GFP_NOIO);
1741 if (!event)
1742 return -ENOMEM;
1743
1744 dout("create_event %p\n", event);
1745 event->cb = event_cb;
3c663bbd 1746 event->one_shot = 0;
a40c4f10
YS		 1747 event->data = data;
1748 event->osdc = osdc;
1749 INIT_LIST_HEAD(&event->osd_node);
3ee5234d 1750 RB_CLEAR_NODE(&event->node);
a40c4f10
YS		 1751 kref_init(&event->kref); /* one ref for us */
1752 kref_get(&event->kref); /* one ref for the caller */
a40c4f10
YS		 1753
1754 spin_lock(&osdc->event_lock);
1755 event->cookie = ++osdc->event_count;
1756 __insert_event(osdc, event);
1757 spin_unlock(&osdc->event_lock);
1758
1759 *pevent = event;
1760 return 0;
1761}
1762EXPORT_SYMBOL(ceph_osdc_create_event);
1763
1764void ceph_osdc_cancel_event(struct ceph_osd_event *event)
1765{
1766 struct ceph_osd_client *osdc = event->osdc;
1767
1768 dout("cancel_event %p\n", event);
1769 spin_lock(&osdc->event_lock);
1770 __remove_event(event);
1771 spin_unlock(&osdc->event_lock);
1772 ceph_osdc_put_event(event); /* caller's */
1773}
1774EXPORT_SYMBOL(ceph_osdc_cancel_event);
1775
1776
1777static void do_event_work(struct work_struct *work)
1778{
1779 struct ceph_osd_event_work *event_work =
1780 container_of(work, struct ceph_osd_event_work, work);
1781 struct ceph_osd_event *event = event_work->event;
1782 u64 ver = event_work->ver;
1783 u64 notify_id = event_work->notify_id;
1784 u8 opcode = event_work->opcode;
1785
1786 dout("do_event_work completing %p\n", event);
1787 event->cb(ver, notify_id, opcode, event->data);
a40c4f10
YS		 1788 dout("do_event_work completed %p\n", event);
1789 ceph_osdc_put_event(event);
1790 kfree(event_work);
1791}
1792
1793
1794/*
1795 * Process osd watch notifications
1796 */
3c663bbd
AE		 1797static void handle_watch_notify(struct ceph_osd_client *osdc,
1798 struct ceph_msg *msg)
a40c4f10
YS		 1799{
1800 void *p, *end;
1801 u8 proto_ver;
1802 u64 cookie, ver, notify_id;
1803 u8 opcode;
1804 struct ceph_osd_event *event;
1805 struct ceph_osd_event_work *event_work;
1806
1807 p = msg->front.iov_base;
1808 end = p + msg->front.iov_len;
1809
1810 ceph_decode_8_safe(&p, end, proto_ver, bad);
1811 ceph_decode_8_safe(&p, end, opcode, bad);
1812 ceph_decode_64_safe(&p, end, cookie, bad);
1813 ceph_decode_64_safe(&p, end, ver, bad);
1814 ceph_decode_64_safe(&p, end, notify_id, bad);
1815
1816 spin_lock(&osdc->event_lock);
1817 event = __find_event(osdc, cookie);
1818 if (event) {
3c663bbd 1819 BUG_ON(event->one_shot);
a40c4f10 1820 get_event(event);
a40c4f10
YS		 1821 }
1822 spin_unlock(&osdc->event_lock);
1823 dout("handle_watch_notify cookie %lld ver %lld event %p\n",
1824 cookie, ver, event);
1825 if (event) {
1826 event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
a40c4f10
YS		 1827 if (!event_work) {
1828 dout("ERROR: could not allocate event_work\n");
1829 goto done_err;
1830 }
6b0ae409 1831 INIT_WORK(&event_work->work, do_event_work);
a40c4f10
YS		 1832 event_work->event = event;
1833 event_work->ver = ver;
1834 event_work->notify_id = notify_id;
1835 event_work->opcode = opcode;
1836 if (!queue_work(osdc->notify_wq, &event_work->work)) {
1837 dout("WARNING: failed to queue notify event work\n");
1838 goto done_err;
1839 }
1840 }
1841
1842 return;
1843
1844done_err:
a40c4f10
YS		 1845 ceph_osdc_put_event(event);
1846 return;
1847
1848bad:
1849 pr_err("osdc handle_watch_notify corrupt msg\n");
1850 return;
1851}
1852
70636773
AE		 1853static void ceph_osdc_msg_data_set(struct ceph_msg *msg,
1854 struct ceph_osd_data *osd_data)
f24e9980 1855{
0fff87ec 1856 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
f1baeb2b 1857 BUG_ON(osd_data->length > (u64) SIZE_MAX);
ebf18f47 1858 if (osd_data->length)
70636773
AE		 1859 ceph_msg_data_set_pages(msg, osd_data->pages,
1860 osd_data->length, osd_data->alignment);
9a5e6d09
AE		 1861 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
1862 BUG_ON(!osd_data->pagelist->length);
1863 ceph_msg_data_set_pagelist(msg, osd_data->pagelist);
68b4476b 1864#ifdef CONFIG_BLOCK
0fff87ec 1865 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
70636773 1866 ceph_msg_data_set_bio(msg, osd_data->bio);
68b4476b 1867#endif
2ac2b7a6 1868 } else {
0fff87ec 1869 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
2ac2b7a6 1870 }
70636773
AE		 1871}
1872
1873/*
1874 * Register request, send initial attempt.
1875 */
1876int ceph_osdc_start_request(struct ceph_osd_client *osdc,
1877 struct ceph_osd_request *req,
1878 bool nofail)
1879{
1880 int rc = 0;
1881
1882 /* Set up response incoming data and request outgoing data fields */
1883
1884 ceph_osdc_msg_data_set(req->r_reply, &req->r_data_in);
1885 ceph_osdc_msg_data_set(req->r_request, &req->r_data_out);
f24e9980 1886
f24e9980
SW		 1887 down_read(&osdc->map_sem);
1888 mutex_lock(&osdc->request_mutex);
dc4b870c 1889 __register_request(osdc, req);
92451b49
AE		 1890 WARN_ON(req->r_sent);
1891 rc = __map_request(osdc, req, 0);
1892 if (rc < 0) {
1893 if (nofail) {
1894 dout("osdc_start_request failed map, "
1895 " will retry %lld\n", req->r_tid);
1896 rc = 0;
f24e9980 1897 }
92451b49 1898 goto out_unlock;
f24e9980 1899 }
92451b49
AE		 1900 if (req->r_osd == NULL) {
1901 dout("send_request %p no up osds in pg\n", req);
1902 ceph_monc_request_next_osdmap(&osdc->client->monc);
1903 } else {
7e2766a1 1904 __send_queued(osdc);
92451b49
AE		 1905 }
1906 rc = 0;
234af26f 1907out_unlock:
f24e9980
SW		 1908 mutex_unlock(&osdc->request_mutex);
1909 up_read(&osdc->map_sem);
1910 return rc;
1911}
3d14c5d2 1912EXPORT_SYMBOL(ceph_osdc_start_request);
f24e9980
SW		 1913
1914/*
1915 * wait for a request to complete
1916 */
1917int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
1918 struct ceph_osd_request *req)
1919{
1920 int rc;
1921
1922 rc = wait_for_completion_interruptible(&req->r_completion);
1923 if (rc < 0) {
1924 mutex_lock(&osdc->request_mutex);
1925 __cancel_request(req);
529cfcc4 1926 __unregister_request(osdc, req);
f24e9980 1927 mutex_unlock(&osdc->request_mutex);
25845472 1928 complete_request(req);
529cfcc4 1929 dout("wait_request tid %llu canceled/timed out\n", req->r_tid);
f24e9980
SW		 1930 return rc;
1931 }
1932
1933 dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result);
1934 return req->r_result;
1935}
3d14c5d2 1936EXPORT_SYMBOL(ceph_osdc_wait_request);
f24e9980
SW		 1937
1938/*
1939 * sync - wait for all in-flight requests to flush. avoid starvation.
1940 */
1941void ceph_osdc_sync(struct ceph_osd_client *osdc)
1942{
1943 struct ceph_osd_request *req;
1944 u64 last_tid, next_tid = 0;
1945
1946 mutex_lock(&osdc->request_mutex);
1947 last_tid = osdc->last_tid;
1948 while (1) {
1949 req = __lookup_request_ge(osdc, next_tid);
1950 if (!req)
1951 break;
1952 if (req->r_tid > last_tid)
1953 break;
1954
1955 next_tid = req->r_tid + 1;
1956 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
1957 continue;
1958
1959 ceph_osdc_get_request(req);
1960 mutex_unlock(&osdc->request_mutex);
1961 dout("sync waiting on tid %llu (last is %llu)\n",
1962 req->r_tid, last_tid);
1963 wait_for_completion(&req->r_safe_completion);
1964 mutex_lock(&osdc->request_mutex);
1965 ceph_osdc_put_request(req);
1966 }
1967 mutex_unlock(&osdc->request_mutex);
1968 dout("sync done (thru tid %llu)\n", last_tid);
1969}
3d14c5d2 1970EXPORT_SYMBOL(ceph_osdc_sync);
f24e9980
SW		 1971
1972/*
1973 * init, shutdown
1974 */
1975int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
1976{
1977 int err;
1978
1979 dout("init\n");
1980 osdc->client = client;
1981 osdc->osdmap = NULL;
1982 init_rwsem(&osdc->map_sem);
1983 init_completion(&osdc->map_waiters);
1984 osdc->last_requested_map = 0;
1985 mutex_init(&osdc->request_mutex);
f24e9980
SW		 1986 osdc->last_tid = 0;
1987 osdc->osds = RB_ROOT;
f5a2041b 1988 INIT_LIST_HEAD(&osdc->osd_lru);
f24e9980 1989 osdc->requests = RB_ROOT;
422d2cb8 1990 INIT_LIST_HEAD(&osdc->req_lru);
6f6c7006
SW		 1991 INIT_LIST_HEAD(&osdc->req_unsent);
1992 INIT_LIST_HEAD(&osdc->req_notarget);
a40c4f10 1993 INIT_LIST_HEAD(&osdc->req_linger);
f24e9980
SW		 1994 osdc->num_requests = 0;
1995 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
f5a2041b 1996 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
a40c4f10
YS		 1997 spin_lock_init(&osdc->event_lock);
1998 osdc->event_tree = RB_ROOT;
1999 osdc->event_count = 0;
f5a2041b
YS		 2000
2001 schedule_delayed_work(&osdc->osds_timeout_work,
3d14c5d2 2002 round_jiffies_relative(osdc->client->options->osd_idle_ttl * HZ));
f24e9980 2003
5f44f142 2004 err = -ENOMEM;
f24e9980
SW		 2005 osdc->req_mempool = mempool_create_kmalloc_pool(10,
2006 sizeof(struct ceph_osd_request));
2007 if (!osdc->req_mempool)
5f44f142 2008 goto out;
f24e9980 2009
d50b409f
SW		 2010 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
2011 OSD_OP_FRONT_LEN, 10, true,
4f48280e 2012 "osd_op");
f24e9980 2013 if (err < 0)
5f44f142 2014 goto out_mempool;
d50b409f 2015 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
4f48280e
SW		 2016 OSD_OPREPLY_FRONT_LEN, 10, true,
2017 "osd_op_reply");
c16e7869
SW		 2018 if (err < 0)
2019 goto out_msgpool;
a40c4f10
YS		 2020
2021 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
2022 if (IS_ERR(osdc->notify_wq)) {
2023 err = PTR_ERR(osdc->notify_wq);
2024 osdc->notify_wq = NULL;
2025 goto out_msgpool;
2026 }
f24e9980 2027 return 0;
5f44f142 2028
c16e7869
SW		 2029out_msgpool:
2030 ceph_msgpool_destroy(&osdc->msgpool_op);
5f44f142
SW		 2031out_mempool:
2032 mempool_destroy(osdc->req_mempool);
2033out:
2034 return err;
f24e9980
SW		 2035}
2036
2037void ceph_osdc_stop(struct ceph_osd_client *osdc)
2038{
a40c4f10
YS		 2039 flush_workqueue(osdc->notify_wq);
2040 destroy_workqueue(osdc->notify_wq);
f24e9980 2041 cancel_delayed_work_sync(&osdc->timeout_work);
f5a2041b 2042 cancel_delayed_work_sync(&osdc->osds_timeout_work);
f24e9980
SW		 2043 if (osdc->osdmap) {
2044 ceph_osdmap_destroy(osdc->osdmap);
2045 osdc->osdmap = NULL;
2046 }
aca420bc 2047 remove_all_osds(osdc);
f24e9980
SW		 2048 mempool_destroy(osdc->req_mempool);
2049 ceph_msgpool_destroy(&osdc->msgpool_op);
c16e7869 2050 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
f24e9980
SW		 2051}
2052
2053/*
2054 * Read some contiguous pages. If we cross a stripe boundary, shorten
2055 * *plen. Return number of bytes read, or error.
2056 */
2057int ceph_osdc_readpages(struct ceph_osd_client *osdc,
2058 struct ceph_vino vino, struct ceph_file_layout *layout,
2059 u64 off, u64 *plen,
2060 u32 truncate_seq, u64 truncate_size,
b7495fc2 2061 struct page **pages, int num_pages, int page_align)
f24e9980
SW		 2062{
2063 struct ceph_osd_request *req;
0fff87ec 2064 struct ceph_osd_data *osd_data;
f24e9980
SW		 2065 int rc = 0;
2066
2067 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
2068 vino.snap, off, *plen);
2069 req = ceph_osdc_new_request(osdc, layout, vino, off, plen,
2070 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
2071 NULL, 0, truncate_seq, truncate_size, NULL,
153e5167 2072 false);
6816282d
SW		 2073 if (IS_ERR(req))
2074 return PTR_ERR(req);
f24e9980
SW		 2075
2076 /* it may be a short read due to an object boundary */
0fff87ec
AE		 2077
2078 osd_data = &req->r_data_in;
2079 osd_data->type = CEPH_OSD_DATA_TYPE_PAGES;
2080 osd_data->pages = pages;
e0c59487 2081 osd_data->length = *plen;
0fff87ec 2082 osd_data->alignment = page_align;
f24e9980 2083
e0c59487
AE		 2084 dout("readpages final extent is %llu~%llu (%llu bytes align %d)\n",
2085 off, *plen, osd_data->length, page_align);
f24e9980
SW		 2086
2087 rc = ceph_osdc_start_request(osdc, req, false);
2088 if (!rc)
2089 rc = ceph_osdc_wait_request(osdc, req);
2090
2091 ceph_osdc_put_request(req);
2092 dout("readpages result %d\n", rc);
2093 return rc;
2094}
3d14c5d2 2095EXPORT_SYMBOL(ceph_osdc_readpages);
f24e9980
SW		 2096
2097/*
2098 * do a synchronous write on N pages
2099 */
2100int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
2101 struct ceph_file_layout *layout,
2102 struct ceph_snap_context *snapc,
2103 u64 off, u64 len,
2104 u32 truncate_seq, u64 truncate_size,
2105 struct timespec *mtime,
24808826 2106 struct page **pages, int num_pages)
f24e9980
SW		 2107{
2108 struct ceph_osd_request *req;
0fff87ec 2109 struct ceph_osd_data *osd_data;
f24e9980 2110 int rc = 0;
b7495fc2 2111 int page_align = off & ~PAGE_MASK;
f24e9980
SW		 2112
2113 BUG_ON(vino.snap != CEPH_NOSNAP);
2114 req = ceph_osdc_new_request(osdc, layout, vino, off, &len,
2115 CEPH_OSD_OP_WRITE,
24808826 2116 CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
fbf8685f 2117 snapc, 0,
f24e9980 2118 truncate_seq, truncate_size, mtime,
153e5167 2119 true);
6816282d
SW		 2120 if (IS_ERR(req))
2121 return PTR_ERR(req);
f24e9980
SW		 2122
2123 /* it may be a short write due to an object boundary */
0fff87ec
AE		 2124 osd_data = &req->r_data_out;
2125 osd_data->type = CEPH_OSD_DATA_TYPE_PAGES;
2126 osd_data->pages = pages;
e0c59487 2127 osd_data->length = len;
0fff87ec 2128 osd_data->alignment = page_align;
e0c59487 2129 dout("writepages %llu~%llu (%llu bytes)\n", off, len, osd_data->length);
f24e9980 2130
87f979d3 2131 rc = ceph_osdc_start_request(osdc, req, true);
f24e9980
SW		 2132 if (!rc)
2133 rc = ceph_osdc_wait_request(osdc, req);
2134
2135 ceph_osdc_put_request(req);
2136 if (rc == 0)
2137 rc = len;
2138 dout("writepages result %d\n", rc);
2139 return rc;
2140}
3d14c5d2 2141EXPORT_SYMBOL(ceph_osdc_writepages);
f24e9980
SW		 2142
2143/*
2144 * handle incoming message
2145 */
2146static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
2147{
2148 struct ceph_osd *osd = con->private;
32c895e7 2149 struct ceph_osd_client *osdc;
f24e9980
SW		 2150 int type = le16_to_cpu(msg->hdr.type);
2151
2152 if (!osd)
4a32f93d 2153 goto out;
32c895e7 2154 osdc = osd->o_osdc;
f24e9980
SW		 2155
2156 switch (type) {
2157 case CEPH_MSG_OSD_MAP:
2158 ceph_osdc_handle_map(osdc, msg);
2159 break;
2160 case CEPH_MSG_OSD_OPREPLY:
350b1c32 2161 handle_reply(osdc, msg, con);
f24e9980 2162 break;
a40c4f10
YS		 2163 case CEPH_MSG_WATCH_NOTIFY:
2164 handle_watch_notify(osdc, msg);
2165 break;
f24e9980
SW		 2166
2167 default:
2168 pr_err("received unknown message type %d %s\n", type,
2169 ceph_msg_type_name(type));
2170 }
4a32f93d 2171out:
f24e9980
SW		 2172 ceph_msg_put(msg);
2173}
2174
5b3a4db3 2175/*
21b667f6
SW		 2176 * lookup and return message for incoming reply. set up reply message
2177 * pages.
5b3a4db3
SW		 2178 */
2179static struct ceph_msg *get_reply(struct ceph_connection *con,
2450418c
YS		 2180 struct ceph_msg_header *hdr,
2181 int *skip)
f24e9980
SW		 2182{
2183 struct ceph_osd *osd = con->private;
2184 struct ceph_osd_client *osdc = osd->o_osdc;
2450418c 2185 struct ceph_msg *m;
0547a9b3 2186 struct ceph_osd_request *req;
5b3a4db3
SW		 2187 int front = le32_to_cpu(hdr->front_len);
2188 int data_len = le32_to_cpu(hdr->data_len);
0547a9b3 2189 u64 tid;
f24e9980 2190
0547a9b3
YS		 2191 tid = le64_to_cpu(hdr->tid);
2192 mutex_lock(&osdc->request_mutex);
2193 req = __lookup_request(osdc, tid);
2194 if (!req) {
2195 *skip = 1;
2196 m = NULL;
756a16a5
SW		 2197 dout("get_reply unknown tid %llu from osd%d\n", tid,
2198 osd->o_osd);
0547a9b3
YS		 2199 goto out;
2200 }
c16e7869
SW		 2201
2202 if (req->r_con_filling_msg) {
8921d114 2203 dout("%s revoking msg %p from old con %p\n", __func__,
c16e7869 2204 req->r_reply, req->r_con_filling_msg);
8921d114 2205 ceph_msg_revoke_incoming(req->r_reply);
0d47766f 2206 req->r_con_filling_msg->ops->put(req->r_con_filling_msg);
6f46cb29 2207 req->r_con_filling_msg = NULL;
0547a9b3
YS		 2208 }
2209
c16e7869
SW		 2210 if (front > req->r_reply->front.iov_len) {
2211 pr_warning("get_reply front %d > preallocated %d\n",
2212 front, (int)req->r_reply->front.iov_len);
b61c2763 2213 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front, GFP_NOFS, false);
a79832f2 2214 if (!m)
c16e7869
SW		 2215 goto out;
2216 ceph_msg_put(req->r_reply);
2217 req->r_reply = m;
2218 }
2219 m = ceph_msg_get(req->r_reply);
2220
0547a9b3 2221 if (data_len > 0) {
0fff87ec
AE		 2222 struct ceph_osd_data *osd_data = &req->r_data_in;
2223
2224 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
0fff87ec 2225 if (osd_data->pages &&
e0c59487 2226 unlikely(osd_data->length < data_len)) {
2ac2b7a6 2227
e0c59487
AE		 2228 pr_warning("tid %lld reply has %d bytes "
2229 "we had only %llu bytes ready\n",
2230 tid, data_len, osd_data->length);
2ac2b7a6
AE		 2231 *skip = 1;
2232 ceph_msg_put(m);
2233 m = NULL;
2234 goto out;
2235 }
2ac2b7a6 2236 }
0547a9b3 2237 }
5b3a4db3 2238 *skip = 0;
0d47766f 2239 req->r_con_filling_msg = con->ops->get(con);
c16e7869 2240 dout("get_reply tid %lld %p\n", tid, m);
0547a9b3
YS		 2241
2242out:
2243 mutex_unlock(&osdc->request_mutex);
2450418c 2244 return m;
5b3a4db3
SW		 2245
2246}
2247
2248static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2249 struct ceph_msg_header *hdr,
2250 int *skip)
2251{
2252 struct ceph_osd *osd = con->private;
2253 int type = le16_to_cpu(hdr->type);
2254 int front = le32_to_cpu(hdr->front_len);
2255
1c20f2d2 2256 *skip = 0;
5b3a4db3
SW		 2257 switch (type) {
2258 case CEPH_MSG_OSD_MAP:
a40c4f10 2259 case CEPH_MSG_WATCH_NOTIFY:
b61c2763 2260 return ceph_msg_new(type, front, GFP_NOFS, false);
5b3a4db3
SW		 2261 case CEPH_MSG_OSD_OPREPLY:
2262 return get_reply(con, hdr, skip);
2263 default:
2264 pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
2265 osd->o_osd);
2266 *skip = 1;
2267 return NULL;
2268 }
f24e9980
SW		 2269}
2270
2271/*
2272 * Wrappers to refcount containing ceph_osd struct
2273 */
2274static struct ceph_connection *get_osd_con(struct ceph_connection *con)
2275{
2276 struct ceph_osd *osd = con->private;
2277 if (get_osd(osd))
2278 return con;
2279 return NULL;
2280}
2281
2282static void put_osd_con(struct ceph_connection *con)
2283{
2284 struct ceph_osd *osd = con->private;
2285 put_osd(osd);
2286}
2287
4e7a5dcd
SW		 2288/*
2289 * authentication
2290 */
a3530df3
AE		 2291/*
2292 * Note: returned pointer is the address of a structure that's
2293 * managed separately. Caller must *not* attempt to free it.
2294 */
2295static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
8f43fb53 2296 int *proto, int force_new)
4e7a5dcd
SW		 2297{
2298 struct ceph_osd *o = con->private;
2299 struct ceph_osd_client *osdc = o->o_osdc;
2300 struct ceph_auth_client *ac = osdc->client->monc.auth;
74f1869f 2301 struct ceph_auth_handshake *auth = &o->o_auth;
4e7a5dcd 2302
74f1869f 2303 if (force_new && auth->authorizer) {
27859f97 2304 ceph_auth_destroy_authorizer(ac, auth->authorizer);
74f1869f
AE		 2305 auth->authorizer = NULL;
2306 }
27859f97
SW		 2307 if (!auth->authorizer) {
2308 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2309 auth);
4e7a5dcd 2310 if (ret)
a3530df3 2311 return ERR_PTR(ret);
27859f97
SW		 2312 } else {
2313 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
0bed9b5c
SW		 2314 auth);
2315 if (ret)
2316 return ERR_PTR(ret);
4e7a5dcd 2317 }
4e7a5dcd 2318 *proto = ac->protocol;
74f1869f 2319
a3530df3 2320 return auth;
4e7a5dcd
SW		 2321}
2322
2323
2324static int verify_authorizer_reply(struct ceph_connection *con, int len)
2325{
2326 struct ceph_osd *o = con->private;
2327 struct ceph_osd_client *osdc = o->o_osdc;
2328 struct ceph_auth_client *ac = osdc->client->monc.auth;
2329
27859f97 2330 return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer, len);
4e7a5dcd
SW		 2331}
2332
9bd2e6f8
SW		 2333static int invalidate_authorizer(struct ceph_connection *con)
2334{
2335 struct ceph_osd *o = con->private;
2336 struct ceph_osd_client *osdc = o->o_osdc;
2337 struct ceph_auth_client *ac = osdc->client->monc.auth;
2338
27859f97 2339 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
9bd2e6f8
SW		 2340 return ceph_monc_validate_auth(&osdc->client->monc);
2341}
4e7a5dcd 2342
9e32789f 2343static const struct ceph_connection_operations osd_con_ops = {
f24e9980
SW		 2344 .get = get_osd_con,
2345 .put = put_osd_con,
2346 .dispatch = dispatch,
4e7a5dcd
SW		 2347 .get_authorizer = get_authorizer,
2348 .verify_authorizer_reply = verify_authorizer_reply,
9bd2e6f8 2349 .invalidate_authorizer = invalidate_authorizer,
f24e9980 2350 .alloc_msg = alloc_msg,
81b024e7 2351 .fault = osd_reset,
f24e9980 2352};
Linux 6.x block layer and io_uring tree(s)