projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
libceph: update osd request/reply encoding
[deliverable/linux.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 34
68b4476b
YS		 35static int op_has_extent(int op)
36{
37 return (op == CEPH_OSD_OP_READ ||
38 op == CEPH_OSD_OP_WRITE);
39}
40
f24e9980
SW		 41/*
42 * Implement client access to distributed object storage cluster.
43 *
44 * All data objects are stored within a cluster/cloud of OSDs, or
45 * "object storage devices." (Note that Ceph OSDs have _nothing_ to
46 * do with the T10 OSD extensions to SCSI.) Ceph OSDs are simply
47 * remote daemons serving up and coordinating consistent and safe
48 * access to storage.
49 *
50 * Cluster membership and the mapping of data objects onto storage devices
51 * are described by the osd map.
52 *
53 * We keep track of pending OSD requests (read, write), resubmit
54 * requests to different OSDs when the cluster topology/data layout
55 * change, or retry the affected requests when the communications
56 * channel with an OSD is reset.
57 */
58
59/*
60 * calculate the mapping of a file extent onto an object, and fill out the
61 * request accordingly. shorten extent as necessary if it crosses an
62 * object boundary.
63 *
64 * fill osd op in request message.
65 */
e75b45cf 66static int calc_layout(struct ceph_vino vino,
d63b77f4
SW		 67 struct ceph_file_layout *layout,
68 u64 off, u64 *plen,
69 struct ceph_osd_request *req,
70 struct ceph_osd_req_op *op)
f24e9980 71{
60e56f13
AE		 72 u64 orig_len = *plen;
73 u64 bno = 0;
74 u64 objoff = 0;
75 u64 objlen = 0;
d63b77f4 76 int r;
f24e9980 77
60e56f13
AE		 78 /* object extent? */
79 r = ceph_calc_file_object_mapping(layout, off, orig_len, &bno,
80 &objoff, &objlen);
d63b77f4
SW		 81 if (r < 0)
82 return r;
60e56f13
AE		 83 if (objlen < orig_len) {
84 *plen = objlen;
85 dout(" skipping last %llu, final file extent %llu~%llu\n",
86 orig_len - *plen, off, *plen);
87 }
88
89 if (op_has_extent(op->op)) {
90 u32 osize = le32_to_cpu(layout->fl_object_size);
91 op->extent.offset = objoff;
92 op->extent.length = objlen;
93 if (op->extent.truncate_size <= off - objoff) {
94 op->extent.truncate_size = 0;
95 } else {
96 op->extent.truncate_size -= off - objoff;
97 if (op->extent.truncate_size > osize)
98 op->extent.truncate_size = osize;
99 }
100 }
101 req->r_num_pages = calc_pages_for(off, *plen);
102 req->r_page_alignment = off & ~PAGE_MASK;
103 if (op->op == CEPH_OSD_OP_WRITE)
104 op->payload_len = *plen;
105
106 dout("calc_layout bno=%llx %llu~%llu (%d pages)\n",
107 bno, objoff, objlen, req->r_num_pages);
f24e9980 108
2dab036b 109 snprintf(req->r_oid, sizeof(req->r_oid), "%llx.%08llx", vino.ino, bno);
f24e9980 110 req->r_oid_len = strlen(req->r_oid);
d63b77f4
SW		 111
112 return r;
f24e9980
SW		 113}
114
f24e9980
SW		 115/*
116 * requests
117 */
415e49a9 118void ceph_osdc_release_request(struct kref *kref)
f24e9980 119{
415e49a9
SW		 120 struct ceph_osd_request *req = container_of(kref,
121 struct ceph_osd_request,
122 r_kref);
123
124 if (req->r_request)
125 ceph_msg_put(req->r_request);
0d59ab81 126 if (req->r_con_filling_msg) {
9cbb1d72
AE		 127 dout("%s revoking msg %p from con %p\n", __func__,
128 req->r_reply, req->r_con_filling_msg);
8921d114 129 ceph_msg_revoke_incoming(req->r_reply);
0d47766f 130 req->r_con_filling_msg->ops->put(req->r_con_filling_msg);
9cbb1d72 131 req->r_con_filling_msg = NULL;
350b1c32 132 }
ab8cb34a
AE		 133 if (req->r_reply)
134 ceph_msg_put(req->r_reply);
415e49a9
SW		 135 if (req->r_own_pages)
136 ceph_release_page_vector(req->r_pages,
137 req->r_num_pages);
138 ceph_put_snap_context(req->r_snapc);
c885837f 139 ceph_pagelist_release(&req->r_trail);
415e49a9
SW		 140 if (req->r_mempool)
141 mempool_free(req, req->r_osdc->req_mempool);
142 else
143 kfree(req);
f24e9980 144}
3d14c5d2 145EXPORT_SYMBOL(ceph_osdc_release_request);
68b4476b 146
3499e8a5 147struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
f24e9980 148 struct ceph_snap_context *snapc,
1b83bef2 149 unsigned int num_ops,
3499e8a5 150 bool use_mempool,
54a54007 151 gfp_t gfp_flags)
f24e9980
SW		 152{
153 struct ceph_osd_request *req;
154 struct ceph_msg *msg;
1b83bef2
SW		 155 size_t msg_size;
156
157 msg_size = 4 + 4 + 8 + 8 + 4+8;
158 msg_size += 2 + 4 + 8 + 4 + 4; /* oloc */
159 msg_size += 1 + 8 + 4 + 4; /* pg_t */
160 msg_size += 4 + MAX_OBJ_NAME_SIZE;
161 msg_size += 2 + num_ops*sizeof(struct ceph_osd_op);
162 msg_size += 8; /* snapid */
163 msg_size += 8; /* snap_seq */
164 msg_size += 8 * (snapc ? snapc->num_snaps : 0); /* snaps */
165 msg_size += 4;
f24e9980
SW		 166
167 if (use_mempool) {
3499e8a5 168 req = mempool_alloc(osdc->req_mempool, gfp_flags);
f24e9980
SW		 169 memset(req, 0, sizeof(*req));
170 } else {
3499e8a5 171 req = kzalloc(sizeof(*req), gfp_flags);
f24e9980
SW		 172 }
173 if (req == NULL)
a79832f2 174 return NULL;
f24e9980 175
f24e9980
SW		 176 req->r_osdc = osdc;
177 req->r_mempool = use_mempool;
68b4476b 178
415e49a9 179 kref_init(&req->r_kref);
f24e9980
SW		 180 init_completion(&req->r_completion);
181 init_completion(&req->r_safe_completion);
a978fa20 182 RB_CLEAR_NODE(&req->r_node);
f24e9980 183 INIT_LIST_HEAD(&req->r_unsafe_item);
a40c4f10
YS		 184 INIT_LIST_HEAD(&req->r_linger_item);
185 INIT_LIST_HEAD(&req->r_linger_osd);
935b639a 186 INIT_LIST_HEAD(&req->r_req_lru_item);
cd43045c
SW		 187 INIT_LIST_HEAD(&req->r_osd_item);
188
c16e7869
SW		 189 /* create reply message */
190 if (use_mempool)
191 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0);
192 else
193 msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY,
b61c2763 194 OSD_OPREPLY_FRONT_LEN, gfp_flags, true);
a79832f2 195 if (!msg) {
c16e7869 196 ceph_osdc_put_request(req);
a79832f2 197 return NULL;
c16e7869
SW		 198 }
199 req->r_reply = msg;
200
c885837f 201 ceph_pagelist_init(&req->r_trail);
d50b409f 202
c16e7869 203 /* create request message; allow space for oid */
f24e9980 204 if (use_mempool)
8f3bc053 205 msg = ceph_msgpool_get(&osdc->msgpool_op, 0);
f24e9980 206 else
b61c2763 207 msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp_flags, true);
a79832f2 208 if (!msg) {
f24e9980 209 ceph_osdc_put_request(req);
a79832f2 210 return NULL;
f24e9980 211 }
68b4476b 212
f24e9980 213 memset(msg->front.iov_base, 0, msg->front.iov_len);
3499e8a5
YS		 214
215 req->r_request = msg;
3499e8a5
YS		 216
217 return req;
218}
3d14c5d2 219EXPORT_SYMBOL(ceph_osdc_alloc_request);
3499e8a5 220
68b4476b
YS		 221static void osd_req_encode_op(struct ceph_osd_request *req,
222 struct ceph_osd_op *dst,
223 struct ceph_osd_req_op *src)
224{
225 dst->op = cpu_to_le16(src->op);
226
065a68f9 227 switch (src->op) {
fbfab539
AE		 228 case CEPH_OSD_OP_STAT:
229 break;
68b4476b
YS		 230 case CEPH_OSD_OP_READ:
231 case CEPH_OSD_OP_WRITE:
232 dst->extent.offset =
233 cpu_to_le64(src->extent.offset);
234 dst->extent.length =
235 cpu_to_le64(src->extent.length);
236 dst->extent.truncate_size =
237 cpu_to_le64(src->extent.truncate_size);
238 dst->extent.truncate_seq =
239 cpu_to_le32(src->extent.truncate_seq);
240 break;
ae1533b6 241 case CEPH_OSD_OP_CALL:
ae1533b6
YS		 242 dst->cls.class_len = src->cls.class_len;
243 dst->cls.method_len = src->cls.method_len;
244 dst->cls.indata_len = cpu_to_le32(src->cls.indata_len);
245
c885837f 246 ceph_pagelist_append(&req->r_trail, src->cls.class_name,
ae1533b6 247 src->cls.class_len);
c885837f 248 ceph_pagelist_append(&req->r_trail, src->cls.method_name,
ae1533b6 249 src->cls.method_len);
c885837f 250 ceph_pagelist_append(&req->r_trail, src->cls.indata,
ae1533b6
YS		 251 src->cls.indata_len);
252 break;
68b4476b
YS		 253 case CEPH_OSD_OP_STARTSYNC:
254 break;
a40c4f10
YS		 255 case CEPH_OSD_OP_NOTIFY_ACK:
256 case CEPH_OSD_OP_WATCH:
257 dst->watch.cookie = cpu_to_le64(src->watch.cookie);
258 dst->watch.ver = cpu_to_le64(src->watch.ver);
259 dst->watch.flag = src->watch.flag;
260 break;
68b4476b
YS		 261 default:
262 pr_err("unrecognized osd opcode %d\n", dst->op);
263 WARN_ON(1);
264 break;
4c46459c
AE		 265 case CEPH_OSD_OP_MAPEXT:
266 case CEPH_OSD_OP_MASKTRUNC:
267 case CEPH_OSD_OP_SPARSE_READ:
a9f36c3e 268 case CEPH_OSD_OP_NOTIFY:
4c46459c
AE		 269 case CEPH_OSD_OP_ASSERT_VER:
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:
275 case CEPH_OSD_OP_SETTRUNC:
276 case CEPH_OSD_OP_TRIMTRUNC:
277 case CEPH_OSD_OP_TMAPUP:
278 case CEPH_OSD_OP_TMAPPUT:
279 case CEPH_OSD_OP_TMAPGET:
280 case CEPH_OSD_OP_CREATE:
a9f36c3e 281 case CEPH_OSD_OP_ROLLBACK:
4c46459c
AE		 282 case CEPH_OSD_OP_OMAPGETKEYS:
283 case CEPH_OSD_OP_OMAPGETVALS:
284 case CEPH_OSD_OP_OMAPGETHEADER:
285 case CEPH_OSD_OP_OMAPGETVALSBYKEYS:
286 case CEPH_OSD_OP_MODE_RD:
287 case CEPH_OSD_OP_OMAPSETVALS:
288 case CEPH_OSD_OP_OMAPSETHEADER:
289 case CEPH_OSD_OP_OMAPCLEAR:
290 case CEPH_OSD_OP_OMAPRMKEYS:
291 case CEPH_OSD_OP_OMAP_CMP:
292 case CEPH_OSD_OP_CLONERANGE:
293 case CEPH_OSD_OP_ASSERT_SRC_VERSION:
294 case CEPH_OSD_OP_SRC_CMPXATTR:
a9f36c3e 295 case CEPH_OSD_OP_GETXATTR:
4c46459c 296 case CEPH_OSD_OP_GETXATTRS:
a9f36c3e
AE		 297 case CEPH_OSD_OP_CMPXATTR:
298 case CEPH_OSD_OP_SETXATTR:
4c46459c
AE		 299 case CEPH_OSD_OP_SETXATTRS:
300 case CEPH_OSD_OP_RESETXATTRS:
301 case CEPH_OSD_OP_RMXATTR:
302 case CEPH_OSD_OP_PULL:
303 case CEPH_OSD_OP_PUSH:
304 case CEPH_OSD_OP_BALANCEREADS:
305 case CEPH_OSD_OP_UNBALANCEREADS:
306 case CEPH_OSD_OP_SCRUB:
307 case CEPH_OSD_OP_SCRUB_RESERVE:
308 case CEPH_OSD_OP_SCRUB_UNRESERVE:
309 case CEPH_OSD_OP_SCRUB_STOP:
310 case CEPH_OSD_OP_SCRUB_MAP:
311 case CEPH_OSD_OP_WRLOCK:
312 case CEPH_OSD_OP_WRUNLOCK:
313 case CEPH_OSD_OP_RDLOCK:
314 case CEPH_OSD_OP_RDUNLOCK:
315 case CEPH_OSD_OP_UPLOCK:
316 case CEPH_OSD_OP_DNLOCK:
317 case CEPH_OSD_OP_PGLS:
318 case CEPH_OSD_OP_PGLS_FILTER:
319 pr_err("unsupported osd opcode %s\n",
320 ceph_osd_op_name(dst->op));
321 WARN_ON(1);
322 break;
68b4476b
YS		 323 }
324 dst->payload_len = cpu_to_le32(src->payload_len);
325}
326
3499e8a5
YS		 327/*
328 * build new request AND message
329 *
330 */
331void ceph_osdc_build_request(struct ceph_osd_request *req,
1b83bef2 332 u64 off, u64 len, unsigned int num_ops,
68b4476b 333 struct ceph_osd_req_op *src_ops,
4d6b250b 334 struct ceph_snap_context *snapc, u64 snap_id,
af77f26c 335 struct timespec *mtime)
3499e8a5
YS		 336{
337 struct ceph_msg *msg = req->r_request;
68b4476b 338 struct ceph_osd_req_op *src_op;
3499e8a5 339 void *p;
1b83bef2 340 size_t msg_size;
3499e8a5 341 int flags = req->r_flags;
f44246e3 342 u64 data_len;
68b4476b 343 int i;
3499e8a5 344
1b83bef2
SW		 345 req->r_num_ops = num_ops;
346 req->r_snapid = snap_id;
f24e9980
SW		 347 req->r_snapc = ceph_get_snap_context(snapc);
348
1b83bef2
SW		 349 /* encode request */
350 msg->hdr.version = cpu_to_le16(4);
351
352 p = msg->front.iov_base;
353 ceph_encode_32(&p, 1); /* client_inc is always 1 */
354 req->r_request_osdmap_epoch = p;
355 p += 4;
356 req->r_request_flags = p;
357 p += 4;
358 if (req->r_flags & CEPH_OSD_FLAG_WRITE)
359 ceph_encode_timespec(p, mtime);
360 p += sizeof(struct ceph_timespec);
361 req->r_request_reassert_version = p;
362 p += sizeof(struct ceph_eversion); /* will get filled in */
363
364 /* oloc */
365 ceph_encode_8(&p, 4);
366 ceph_encode_8(&p, 4);
367 ceph_encode_32(&p, 8 + 4 + 4);
368 req->r_request_pool = p;
369 p += 8;
370 ceph_encode_32(&p, -1); /* preferred */
371 ceph_encode_32(&p, 0); /* key len */
372
373 ceph_encode_8(&p, 1);
374 req->r_request_pgid = p;
375 p += 8 + 4;
376 ceph_encode_32(&p, -1); /* preferred */
377
378 /* oid */
379 ceph_encode_32(&p, req->r_oid_len);
af77f26c 380 memcpy(p, req->r_oid, req->r_oid_len);
1b83bef2 381 dout("oid '%.*s' len %d\n", req->r_oid_len, req->r_oid, req->r_oid_len);
af77f26c 382 p += req->r_oid_len;
f24e9980 383
1b83bef2
SW		 384 /* ops */
385 ceph_encode_16(&p, num_ops);
68b4476b 386 src_op = src_ops;
1b83bef2
SW		 387 req->r_request_ops = p;
388 for (i = 0; i < num_ops; i++, src_op++) {
389 osd_req_encode_op(req, p, src_op);
390 p += sizeof(struct ceph_osd_op);
391 }
68b4476b 392
1b83bef2
SW		 393 /* snaps */
394 ceph_encode_64(&p, req->r_snapid);
395 ceph_encode_64(&p, req->r_snapc ? req->r_snapc->seq : 0);
396 ceph_encode_32(&p, req->r_snapc ? req->r_snapc->num_snaps : 0);
397 if (req->r_snapc) {
f24e9980 398 for (i = 0; i < snapc->num_snaps; i++) {
1b83bef2 399 ceph_encode_64(&p, req->r_snapc->snaps[i]);
f24e9980
SW		 400 }
401 }
402
1b83bef2
SW		 403 req->r_request_attempts = p;
404 p += 4;
405
f44246e3 406 data_len = req->r_trail.length;
68b4476b
YS		 407 if (flags & CEPH_OSD_FLAG_WRITE) {
408 req->r_request->hdr.data_off = cpu_to_le16(off);
f44246e3 409 data_len += len;
68b4476b 410 }
f44246e3 411 req->r_request->hdr.data_len = cpu_to_le32(data_len);
c5c6b19d
SW		 412 req->r_request->page_alignment = req->r_page_alignment;
413
f24e9980 414 BUG_ON(p > msg->front.iov_base + msg->front.iov_len);
6f863e71
SW		 415 msg_size = p - msg->front.iov_base;
416 msg->front.iov_len = msg_size;
417 msg->hdr.front_len = cpu_to_le32(msg_size);
1b83bef2
SW		 418
419 dout("build_request msg_size was %d num_ops %d\n", (int)msg_size,
420 num_ops);
3499e8a5
YS		 421 return;
422}
3d14c5d2 423EXPORT_SYMBOL(ceph_osdc_build_request);
3499e8a5
YS		 424
425/*
426 * build new request AND message, calculate layout, and adjust file
427 * extent as needed.
428 *
429 * if the file was recently truncated, we include information about its
430 * old and new size so that the object can be updated appropriately. (we
431 * avoid synchronously deleting truncated objects because it's slow.)
432 *
433 * if @do_sync, include a 'startsync' command so that the osd will flush
434 * data quickly.
435 */
436struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
437 struct ceph_file_layout *layout,
438 struct ceph_vino vino,
439 u64 off, u64 *plen,
440 int opcode, int flags,
441 struct ceph_snap_context *snapc,
442 int do_sync,
443 u32 truncate_seq,
444 u64 truncate_size,
445 struct timespec *mtime,
a3bea47e 446 bool use_mempool,
b7495fc2 447 int page_align)
3499e8a5 448{
ae7ca4a3 449 struct ceph_osd_req_op ops[2];
68b4476b 450 struct ceph_osd_request *req;
ae7ca4a3 451 unsigned int num_op = 1;
6816282d 452 int r;
68b4476b 453
ae7ca4a3
AE		 454 memset(&ops, 0, sizeof ops);
455
68b4476b
YS		 456 ops[0].op = opcode;
457 ops[0].extent.truncate_seq = truncate_seq;
458 ops[0].extent.truncate_size = truncate_size;
68b4476b
YS		 459
460 if (do_sync) {
461 ops[1].op = CEPH_OSD_OP_STARTSYNC;
ae7ca4a3
AE		 462 num_op++;
463 }
68b4476b 464
ae7ca4a3
AE		 465 req = ceph_osdc_alloc_request(osdc, snapc, num_op, use_mempool,
466 GFP_NOFS);
4ad12621 467 if (!req)
6816282d 468 return ERR_PTR(-ENOMEM);
d178a9e7 469 req->r_flags = flags;
3499e8a5
YS		 470
471 /* calculate max write size */
e75b45cf 472 r = calc_layout(vino, layout, off, plen, req, ops);
6816282d
SW		 473 if (r < 0)
474 return ERR_PTR(r);
3499e8a5
YS		 475 req->r_file_layout = *layout; /* keep a copy */
476
9bb0ce2b
SW		 477 /* in case it differs from natural (file) alignment that
478 calc_layout filled in for us */
479 req->r_num_pages = calc_pages_for(page_align, *plen);
b7495fc2
SW		 480 req->r_page_alignment = page_align;
481
ae7ca4a3
AE		 482 ceph_osdc_build_request(req, off, *plen, num_op, ops,
483 snapc, vino.snap, mtime);
3499e8a5 484
f24e9980
SW		 485 return req;
486}
3d14c5d2 487EXPORT_SYMBOL(ceph_osdc_new_request);
f24e9980
SW		 488
489/*
490 * We keep osd requests in an rbtree, sorted by ->r_tid.
491 */
492static void __insert_request(struct ceph_osd_client *osdc,
493 struct ceph_osd_request *new)
494{
495 struct rb_node **p = &osdc->requests.rb_node;
496 struct rb_node *parent = NULL;
497 struct ceph_osd_request *req = NULL;
498
499 while (*p) {
500 parent = *p;
501 req = rb_entry(parent, struct ceph_osd_request, r_node);
502 if (new->r_tid < req->r_tid)
503 p = &(*p)->rb_left;
504 else if (new->r_tid > req->r_tid)
505 p = &(*p)->rb_right;
506 else
507 BUG();
508 }
509
510 rb_link_node(&new->r_node, parent, p);
511 rb_insert_color(&new->r_node, &osdc->requests);
512}
513
514static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc,
515 u64 tid)
516{
517 struct ceph_osd_request *req;
518 struct rb_node *n = osdc->requests.rb_node;
519
520 while (n) {
521 req = rb_entry(n, struct ceph_osd_request, r_node);
522 if (tid < req->r_tid)
523 n = n->rb_left;
524 else if (tid > req->r_tid)
525 n = n->rb_right;
526 else
527 return req;
528 }
529 return NULL;
530}
531
532static struct ceph_osd_request *
533__lookup_request_ge(struct ceph_osd_client *osdc,
534 u64 tid)
535{
536 struct ceph_osd_request *req;
537 struct rb_node *n = osdc->requests.rb_node;
538
539 while (n) {
540 req = rb_entry(n, struct ceph_osd_request, r_node);
541 if (tid < req->r_tid) {
542 if (!n->rb_left)
543 return req;
544 n = n->rb_left;
545 } else if (tid > req->r_tid) {
546 n = n->rb_right;
547 } else {
548 return req;
549 }
550 }
551 return NULL;
552}
553
6f6c7006
SW		 554/*
555 * Resubmit requests pending on the given osd.
556 */
557static void __kick_osd_requests(struct ceph_osd_client *osdc,
558 struct ceph_osd *osd)
559{
a40c4f10 560 struct ceph_osd_request *req, *nreq;
6f6c7006
SW		 561 int err;
562
563 dout("__kick_osd_requests osd%d\n", osd->o_osd);
564 err = __reset_osd(osdc, osd);
685a7555 565 if (err)
6f6c7006
SW		 566 return;
567
568 list_for_each_entry(req, &osd->o_requests, r_osd_item) {
569 list_move(&req->r_req_lru_item, &osdc->req_unsent);
570 dout("requeued %p tid %llu osd%d\n", req, req->r_tid,
571 osd->o_osd);
a40c4f10
YS		 572 if (!req->r_linger)
573 req->r_flags |= CEPH_OSD_FLAG_RETRY;
574 }
575
576 list_for_each_entry_safe(req, nreq, &osd->o_linger_requests,
577 r_linger_osd) {
77f38e0e
SW		 578 /*
579 * reregister request prior to unregistering linger so
580 * that r_osd is preserved.
581 */
582 BUG_ON(!list_empty(&req->r_req_lru_item));
a40c4f10 583 __register_request(osdc, req);
77f38e0e
SW		 584 list_add(&req->r_req_lru_item, &osdc->req_unsent);
585 list_add(&req->r_osd_item, &req->r_osd->o_requests);
586 __unregister_linger_request(osdc, req);
a40c4f10
YS		 587 dout("requeued lingering %p tid %llu osd%d\n", req, req->r_tid,
588 osd->o_osd);
6f6c7006
SW		 589 }
590}
591
f24e9980 592/*
81b024e7 593 * If the osd connection drops, we need to resubmit all requests.
f24e9980
SW		 594 */
595static void osd_reset(struct ceph_connection *con)
596{
597 struct ceph_osd *osd = con->private;
598 struct ceph_osd_client *osdc;
599
600 if (!osd)
601 return;
602 dout("osd_reset osd%d\n", osd->o_osd);
603 osdc = osd->o_osdc;
f24e9980 604 down_read(&osdc->map_sem);
83aff95e
SW		 605 mutex_lock(&osdc->request_mutex);
606 __kick_osd_requests(osdc, osd);
f9d25199 607 __send_queued(osdc);
83aff95e 608 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 609 up_read(&osdc->map_sem);
610}
611
612/*
613 * Track open sessions with osds.
614 */
e10006f8 615static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
f24e9980
SW		 616{
617 struct ceph_osd *osd;
618
619 osd = kzalloc(sizeof(*osd), GFP_NOFS);
620 if (!osd)
621 return NULL;
622
623 atomic_set(&osd->o_ref, 1);
624 osd->o_osdc = osdc;
e10006f8 625 osd->o_osd = onum;
f407731d 626 RB_CLEAR_NODE(&osd->o_node);
f24e9980 627 INIT_LIST_HEAD(&osd->o_requests);
a40c4f10 628 INIT_LIST_HEAD(&osd->o_linger_requests);
f5a2041b 629 INIT_LIST_HEAD(&osd->o_osd_lru);
f24e9980
SW		 630 osd->o_incarnation = 1;
631
b7a9e5dd 632 ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
4e7a5dcd 633
422d2cb8 634 INIT_LIST_HEAD(&osd->o_keepalive_item);
f24e9980
SW		 635 return osd;
636}
637
638static struct ceph_osd *get_osd(struct ceph_osd *osd)
639{
640 if (atomic_inc_not_zero(&osd->o_ref)) {
641 dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
642 atomic_read(&osd->o_ref));
643 return osd;
644 } else {
645 dout("get_osd %p FAIL\n", osd);
646 return NULL;
647 }
648}
649
650static void put_osd(struct ceph_osd *osd)
651{
652 dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
653 atomic_read(&osd->o_ref) - 1);
a255651d 654 if (atomic_dec_and_test(&osd->o_ref) && osd->o_auth.authorizer) {
79494d1b
SW		 655 struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth;
656
a255651d 657 if (ac->ops && ac->ops->destroy_authorizer)
6c4a1915 658 ac->ops->destroy_authorizer(ac, osd->o_auth.authorizer);
f24e9980 659 kfree(osd);
79494d1b 660 }
f24e9980
SW		 661}
662
663/*
664 * remove an osd from our map
665 */
f5a2041b 666static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
f24e9980 667{
f5a2041b 668 dout("__remove_osd %p\n", osd);
f24e9980
SW		 669 BUG_ON(!list_empty(&osd->o_requests));
670 rb_erase(&osd->o_node, &osdc->osds);
f5a2041b 671 list_del_init(&osd->o_osd_lru);
f24e9980
SW		 672 ceph_con_close(&osd->o_con);
673 put_osd(osd);
674}
675
aca420bc
SW		 676static void remove_all_osds(struct ceph_osd_client *osdc)
677{
048a9d2d 678 dout("%s %p\n", __func__, osdc);
aca420bc
SW		 679 mutex_lock(&osdc->request_mutex);
680 while (!RB_EMPTY_ROOT(&osdc->osds)) {
681 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
682 struct ceph_osd, o_node);
683 __remove_osd(osdc, osd);
684 }
685 mutex_unlock(&osdc->request_mutex);
686}
687
f5a2041b
YS		 688static void __move_osd_to_lru(struct ceph_osd_client *osdc,
689 struct ceph_osd *osd)
690{
691 dout("__move_osd_to_lru %p\n", osd);
692 BUG_ON(!list_empty(&osd->o_osd_lru));
693 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
3d14c5d2 694 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl * HZ;
f5a2041b
YS		 695}
696
697static void __remove_osd_from_lru(struct ceph_osd *osd)
698{
699 dout("__remove_osd_from_lru %p\n", osd);
700 if (!list_empty(&osd->o_osd_lru))
701 list_del_init(&osd->o_osd_lru);
702}
703
aca420bc 704static void remove_old_osds(struct ceph_osd_client *osdc)
f5a2041b
YS		 705{
706 struct ceph_osd *osd, *nosd;
707
708 dout("__remove_old_osds %p\n", osdc);
709 mutex_lock(&osdc->request_mutex);
710 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
aca420bc 711 if (time_before(jiffies, osd->lru_ttl))
f5a2041b
YS		 712 break;
713 __remove_osd(osdc, osd);
714 }
715 mutex_unlock(&osdc->request_mutex);
716}
717
f24e9980
SW		 718/*
719 * reset osd connect
720 */
f5a2041b 721static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
f24e9980 722{
c3acb181 723 struct ceph_entity_addr *peer_addr;
f24e9980 724
f5a2041b 725 dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
a40c4f10
YS		 726 if (list_empty(&osd->o_requests) &&
727 list_empty(&osd->o_linger_requests)) {
f5a2041b 728 __remove_osd(osdc, osd);
c3acb181
AE		 729
730 return -ENODEV;
731 }
732
733 peer_addr = &osdc->osdmap->osd_addr[osd->o_osd];
734 if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
735 !ceph_con_opened(&osd->o_con)) {
736 struct ceph_osd_request *req;
737
87b315a5
SW		 738 dout(" osd addr hasn't changed and connection never opened,"
739 " letting msgr retry");
740 /* touch each r_stamp for handle_timeout()'s benfit */
741 list_for_each_entry(req, &osd->o_requests, r_osd_item)
742 req->r_stamp = jiffies;
c3acb181
AE		 743
744 return -EAGAIN;
f24e9980 745 }
c3acb181
AE		 746
747 ceph_con_close(&osd->o_con);
748 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
749 osd->o_incarnation++;
750
751 return 0;
f24e9980
SW		 752}
753
754static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new)
755{
756 struct rb_node **p = &osdc->osds.rb_node;
757 struct rb_node *parent = NULL;
758 struct ceph_osd *osd = NULL;
759
aca420bc 760 dout("__insert_osd %p osd%d\n", new, new->o_osd);
f24e9980
SW		 761 while (*p) {
762 parent = *p;
763 osd = rb_entry(parent, struct ceph_osd, o_node);
764 if (new->o_osd < osd->o_osd)
765 p = &(*p)->rb_left;
766 else if (new->o_osd > osd->o_osd)
767 p = &(*p)->rb_right;
768 else
769 BUG();
770 }
771
772 rb_link_node(&new->o_node, parent, p);
773 rb_insert_color(&new->o_node, &osdc->osds);
774}
775
776static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o)
777{
778 struct ceph_osd *osd;
779 struct rb_node *n = osdc->osds.rb_node;
780
781 while (n) {
782 osd = rb_entry(n, struct ceph_osd, o_node);
783 if (o < osd->o_osd)
784 n = n->rb_left;
785 else if (o > osd->o_osd)
786 n = n->rb_right;
787 else
788 return osd;
789 }
790 return NULL;
791}
792
422d2cb8
YS		 793static void __schedule_osd_timeout(struct ceph_osd_client *osdc)
794{
795 schedule_delayed_work(&osdc->timeout_work,
3d14c5d2 796 osdc->client->options->osd_keepalive_timeout * HZ);
422d2cb8
YS		 797}
798
799static void __cancel_osd_timeout(struct ceph_osd_client *osdc)
800{
801 cancel_delayed_work(&osdc->timeout_work);
802}
f24e9980
SW		 803
804/*
805 * Register request, assign tid. If this is the first request, set up
806 * the timeout event.
807 */
a40c4f10
YS		 808static void __register_request(struct ceph_osd_client *osdc,
809 struct ceph_osd_request *req)
f24e9980 810{
f24e9980 811 req->r_tid = ++osdc->last_tid;
6df058c0 812 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
77f38e0e 813 dout("__register_request %p tid %lld\n", req, req->r_tid);
f24e9980
SW		 814 __insert_request(osdc, req);
815 ceph_osdc_get_request(req);
816 osdc->num_requests++;
f24e9980 817 if (osdc->num_requests == 1) {
422d2cb8
YS		 818 dout(" first request, scheduling timeout\n");
819 __schedule_osd_timeout(osdc);
f24e9980 820 }
a40c4f10
YS		 821}
822
823static void register_request(struct ceph_osd_client *osdc,
824 struct ceph_osd_request *req)
825{
826 mutex_lock(&osdc->request_mutex);
827 __register_request(osdc, req);
f24e9980
SW		 828 mutex_unlock(&osdc->request_mutex);
829}
830
831/*
832 * called under osdc->request_mutex
833 */
834static void __unregister_request(struct ceph_osd_client *osdc,
835 struct ceph_osd_request *req)
836{
35f9f8a0
SW		 837 if (RB_EMPTY_NODE(&req->r_node)) {
838 dout("__unregister_request %p tid %lld not registered\n",
839 req, req->r_tid);
840 return;
841 }
842
f24e9980
SW		 843 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
844 rb_erase(&req->r_node, &osdc->requests);
845 osdc->num_requests--;
846
0ba6478d
SW		 847 if (req->r_osd) {
848 /* make sure the original request isn't in flight. */
6740a845 849 ceph_msg_revoke(req->r_request);
0ba6478d
SW		 850
851 list_del_init(&req->r_osd_item);
a40c4f10
YS		 852 if (list_empty(&req->r_osd->o_requests) &&
853 list_empty(&req->r_osd->o_linger_requests)) {
854 dout("moving osd to %p lru\n", req->r_osd);
f5a2041b 855 __move_osd_to_lru(osdc, req->r_osd);
a40c4f10 856 }
fbdb9190 857 if (list_empty(&req->r_linger_item))
a40c4f10 858 req->r_osd = NULL;
0ba6478d 859 }
f24e9980 860
7d5f2481 861 list_del_init(&req->r_req_lru_item);
f24e9980
SW		 862 ceph_osdc_put_request(req);
863
422d2cb8
YS		 864 if (osdc->num_requests == 0) {
865 dout(" no requests, canceling timeout\n");
866 __cancel_osd_timeout(osdc);
f24e9980
SW		 867 }
868}
869
870/*
871 * Cancel a previously queued request message
872 */
873static void __cancel_request(struct ceph_osd_request *req)
874{
6bc18876 875 if (req->r_sent && req->r_osd) {
6740a845 876 ceph_msg_revoke(req->r_request);
f24e9980
SW		 877 req->r_sent = 0;
878 }
879}
880
a40c4f10
YS		 881static void __register_linger_request(struct ceph_osd_client *osdc,
882 struct ceph_osd_request *req)
883{
884 dout("__register_linger_request %p\n", req);
885 list_add_tail(&req->r_linger_item, &osdc->req_linger);
6194ea89
SW		 886 if (req->r_osd)
887 list_add_tail(&req->r_linger_osd,
888 &req->r_osd->o_linger_requests);
a40c4f10
YS		 889}
890
891static void __unregister_linger_request(struct ceph_osd_client *osdc,
892 struct ceph_osd_request *req)
893{
894 dout("__unregister_linger_request %p\n", req);
61c74035 895 list_del_init(&req->r_linger_item);
a40c4f10 896 if (req->r_osd) {
a40c4f10
YS		 897 list_del_init(&req->r_linger_osd);
898
899 if (list_empty(&req->r_osd->o_requests) &&
900 list_empty(&req->r_osd->o_linger_requests)) {
901 dout("moving osd to %p lru\n", req->r_osd);
902 __move_osd_to_lru(osdc, req->r_osd);
903 }
fbdb9190
SW		 904 if (list_empty(&req->r_osd_item))
905 req->r_osd = NULL;
a40c4f10
YS		 906 }
907}
908
909void ceph_osdc_unregister_linger_request(struct ceph_osd_client *osdc,
910 struct ceph_osd_request *req)
911{
912 mutex_lock(&osdc->request_mutex);
913 if (req->r_linger) {
914 __unregister_linger_request(osdc, req);
915 ceph_osdc_put_request(req);
916 }
917 mutex_unlock(&osdc->request_mutex);
918}
919EXPORT_SYMBOL(ceph_osdc_unregister_linger_request);
920
921void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
922 struct ceph_osd_request *req)
923{
924 if (!req->r_linger) {
925 dout("set_request_linger %p\n", req);
926 req->r_linger = 1;
927 /*
928 * caller is now responsible for calling
929 * unregister_linger_request
930 */
931 ceph_osdc_get_request(req);
932 }
933}
934EXPORT_SYMBOL(ceph_osdc_set_request_linger);
935
f24e9980
SW		 936/*
937 * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
938 * (as needed), and set the request r_osd appropriately. If there is
25985edc 939 * no up osd, set r_osd to NULL. Move the request to the appropriate list
6f6c7006 940 * (unsent, homeless) or leave on in-flight lru.
f24e9980
SW		 941 *
942 * Return 0 if unchanged, 1 if changed, or negative on error.
943 *
944 * Caller should hold map_sem for read and request_mutex.
945 */
6f6c7006 946static int __map_request(struct ceph_osd_client *osdc,
38d6453c 947 struct ceph_osd_request *req, int force_resend)
f24e9980 948{
5b191d99 949 struct ceph_pg pgid;
d85b7056
SW		 950 int acting[CEPH_PG_MAX_SIZE];
951 int o = -1, num = 0;
f24e9980 952 int err;
f24e9980 953
6f6c7006 954 dout("map_request %p tid %lld\n", req, req->r_tid);
2169aea6 955 err = ceph_calc_object_layout(&pgid, req->r_oid,
f24e9980 956 &req->r_file_layout, osdc->osdmap);
6f6c7006
SW		 957 if (err) {
958 list_move(&req->r_req_lru_item, &osdc->req_notarget);
f24e9980 959 return err;
6f6c7006 960 }
7740a42f
SW		 961 req->r_pgid = pgid;
962
d85b7056
SW		 963 err = ceph_calc_pg_acting(osdc->osdmap, pgid, acting);
964 if (err > 0) {
965 o = acting[0];
966 num = err;
967 }
f24e9980 968
38d6453c
SW		 969 if ((!force_resend &&
970 req->r_osd && req->r_osd->o_osd == o &&
d85b7056
SW		 971 req->r_sent >= req->r_osd->o_incarnation &&
972 req->r_num_pg_osds == num &&
973 memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
f24e9980
SW		 974 (req->r_osd == NULL && o == -1))
975 return 0; /* no change */
976
5b191d99
SW		 977 dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n",
978 req->r_tid, pgid.pool, pgid.seed, o,
f24e9980
SW		 979 req->r_osd ? req->r_osd->o_osd : -1);
980
d85b7056
SW		 981 /* record full pg acting set */
982 memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
983 req->r_num_pg_osds = num;
984
f24e9980
SW		 985 if (req->r_osd) {
986 __cancel_request(req);
987 list_del_init(&req->r_osd_item);
f24e9980
SW		 988 req->r_osd = NULL;
989 }
990
991 req->r_osd = __lookup_osd(osdc, o);
992 if (!req->r_osd && o >= 0) {
c99eb1c7 993 err = -ENOMEM;
e10006f8 994 req->r_osd = create_osd(osdc, o);
6f6c7006
SW		 995 if (!req->r_osd) {
996 list_move(&req->r_req_lru_item, &osdc->req_notarget);
c99eb1c7 997 goto out;
6f6c7006 998 }
f24e9980 999
6f6c7006 1000 dout("map_request osd %p is osd%d\n", req->r_osd, o);
f24e9980
SW		 1001 __insert_osd(osdc, req->r_osd);
1002
b7a9e5dd
SW		 1003 ceph_con_open(&req->r_osd->o_con,
1004 CEPH_ENTITY_TYPE_OSD, o,
1005 &osdc->osdmap->osd_addr[o]);
f24e9980
SW		 1006 }
1007
f5a2041b
YS		 1008 if (req->r_osd) {
1009 __remove_osd_from_lru(req->r_osd);
f24e9980 1010 list_add(&req->r_osd_item, &req->r_osd->o_requests);
6f6c7006
SW		 1011 list_move(&req->r_req_lru_item, &osdc->req_unsent);
1012 } else {
1013 list_move(&req->r_req_lru_item, &osdc->req_notarget);
f5a2041b 1014 }
d85b7056 1015 err = 1; /* osd or pg changed */
f24e9980
SW		 1016
1017out:
f24e9980
SW		 1018 return err;
1019}
1020
1021/*
1022 * caller should hold map_sem (for read) and request_mutex
1023 */
56e925b6
SW		 1024static void __send_request(struct ceph_osd_client *osdc,
1025 struct ceph_osd_request *req)
f24e9980 1026{
1b83bef2 1027 void *p;
f24e9980 1028
1b83bef2
SW		 1029 dout("send_request %p tid %llu to osd%d flags %d pg %lld.%x\n",
1030 req, req->r_tid, req->r_osd->o_osd, req->r_flags,
1031 (unsigned long long)req->r_pgid.pool, req->r_pgid.seed);
1032
1033 /* fill in message content that changes each time we send it */
1034 put_unaligned_le32(osdc->osdmap->epoch, req->r_request_osdmap_epoch);
1035 put_unaligned_le32(req->r_flags, req->r_request_flags);
1036 put_unaligned_le64(req->r_pgid.pool, req->r_request_pool);
1037 p = req->r_request_pgid;
1038 ceph_encode_64(&p, req->r_pgid.pool);
1039 ceph_encode_32(&p, req->r_pgid.seed);
1040 put_unaligned_le64(1, req->r_request_attempts); /* FIXME */
1041 memcpy(req->r_request_reassert_version, &req->r_reassert_version,
1042 sizeof(req->r_reassert_version));
2169aea6 1043
3dd72fc0 1044 req->r_stamp = jiffies;
07a27e22 1045 list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
f24e9980
SW		 1046
1047 ceph_msg_get(req->r_request); /* send consumes a ref */
1048 ceph_con_send(&req->r_osd->o_con, req->r_request);
1049 req->r_sent = req->r_osd->o_incarnation;
f24e9980
SW		 1050}
1051
6f6c7006
SW		 1052/*
1053 * Send any requests in the queue (req_unsent).
1054 */
f9d25199 1055static void __send_queued(struct ceph_osd_client *osdc)
6f6c7006
SW		 1056{
1057 struct ceph_osd_request *req, *tmp;
1058
f9d25199
AE		 1059 dout("__send_queued\n");
1060 list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item)
6f6c7006 1061 __send_request(osdc, req);
6f6c7006
SW		 1062}
1063
f24e9980
SW		 1064/*
1065 * Timeout callback, called every N seconds when 1 or more osd
1066 * requests has been active for more than N seconds. When this
1067 * happens, we ping all OSDs with requests who have timed out to
1068 * ensure any communications channel reset is detected. Reset the
1069 * request timeouts another N seconds in the future as we go.
1070 * Reschedule the timeout event another N seconds in future (unless
1071 * there are no open requests).
1072 */
1073static void handle_timeout(struct work_struct *work)
1074{
1075 struct ceph_osd_client *osdc =
1076 container_of(work, struct ceph_osd_client, timeout_work.work);
83aff95e 1077 struct ceph_osd_request *req;
f24e9980 1078 struct ceph_osd *osd;
422d2cb8 1079 unsigned long keepalive =
3d14c5d2 1080 osdc->client->options->osd_keepalive_timeout * HZ;
422d2cb8 1081 struct list_head slow_osds;
f24e9980
SW		 1082 dout("timeout\n");
1083 down_read(&osdc->map_sem);
1084
1085 ceph_monc_request_next_osdmap(&osdc->client->monc);
1086
1087 mutex_lock(&osdc->request_mutex);
f24e9980 1088
422d2cb8
YS		 1089 /*
1090 * ping osds that are a bit slow. this ensures that if there
1091 * is a break in the TCP connection we will notice, and reopen
1092 * a connection with that osd (from the fault callback).
1093 */
1094 INIT_LIST_HEAD(&slow_osds);
1095 list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
3dd72fc0 1096 if (time_before(jiffies, req->r_stamp + keepalive))
422d2cb8
YS		 1097 break;
1098
1099 osd = req->r_osd;
1100 BUG_ON(!osd);
1101 dout(" tid %llu is slow, will send keepalive on osd%d\n",
f24e9980 1102 req->r_tid, osd->o_osd);
422d2cb8
YS		 1103 list_move_tail(&osd->o_keepalive_item, &slow_osds);
1104 }
1105 while (!list_empty(&slow_osds)) {
1106 osd = list_entry(slow_osds.next, struct ceph_osd,
1107 o_keepalive_item);
1108 list_del_init(&osd->o_keepalive_item);
f24e9980
SW		 1109 ceph_con_keepalive(&osd->o_con);
1110 }
1111
422d2cb8 1112 __schedule_osd_timeout(osdc);
f9d25199 1113 __send_queued(osdc);
f24e9980 1114 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 1115 up_read(&osdc->map_sem);
1116}
1117
f5a2041b
YS		 1118static void handle_osds_timeout(struct work_struct *work)
1119{
1120 struct ceph_osd_client *osdc =
1121 container_of(work, struct ceph_osd_client,
1122 osds_timeout_work.work);
1123 unsigned long delay =
3d14c5d2 1124 osdc->client->options->osd_idle_ttl * HZ >> 2;
f5a2041b
YS		 1125
1126 dout("osds timeout\n");
1127 down_read(&osdc->map_sem);
aca420bc 1128 remove_old_osds(osdc);
f5a2041b
YS		 1129 up_read(&osdc->map_sem);
1130
1131 schedule_delayed_work(&osdc->osds_timeout_work,
1132 round_jiffies_relative(delay));
1133}
1134
25845472
SW		 1135static void complete_request(struct ceph_osd_request *req)
1136{
1137 if (req->r_safe_callback)
1138 req->r_safe_callback(req, NULL);
1139 complete_all(&req->r_safe_completion); /* fsync waiter */
1140}
1141
1b83bef2
SW		 1142static int __decode_pgid(void **p, void *end, struct ceph_pg *pgid)
1143{
1144 __u8 v;
1145
1146 ceph_decode_need(p, end, 1 + 8 + 4 + 4, bad);
1147 v = ceph_decode_8(p);
1148 if (v > 1) {
1149 pr_warning("do not understand pg encoding %d > 1", v);
1150 return -EINVAL;
1151 }
1152 pgid->pool = ceph_decode_64(p);
1153 pgid->seed = ceph_decode_32(p);
1154 *p += 4;
1155 return 0;
1156
1157bad:
1158 pr_warning("incomplete pg encoding");
1159 return -EINVAL;
1160}
1161
f24e9980
SW		 1162/*
1163 * handle osd op reply. either call the callback if it is specified,
1164 * or do the completion to wake up the waiting thread.
1165 */
350b1c32
SW		 1166static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg,
1167 struct ceph_connection *con)
f24e9980 1168{
1b83bef2 1169 void *p, *end;
f24e9980
SW		 1170 struct ceph_osd_request *req;
1171 u64 tid;
1b83bef2
SW		 1172 int object_len;
1173 int numops, payload_len, flags;
0ceed5db 1174 s32 result;
1b83bef2
SW		 1175 s32 retry_attempt;
1176 struct ceph_pg pg;
1177 int err;
1178 u32 reassert_epoch;
1179 u64 reassert_version;
1180 u32 osdmap_epoch;
1181 int i;
f24e9980 1182
6df058c0 1183 tid = le64_to_cpu(msg->hdr.tid);
1b83bef2
SW		 1184 dout("handle_reply %p tid %llu\n", msg, tid);
1185
1186 p = msg->front.iov_base;
1187 end = p + msg->front.iov_len;
1188
1189 ceph_decode_need(&p, end, 4, bad);
1190 object_len = ceph_decode_32(&p);
1191 ceph_decode_need(&p, end, object_len, bad);
1192 p += object_len;
1193
1194 err = __decode_pgid(&p, end, &pg);
1195 if (err)
f24e9980 1196 goto bad;
1b83bef2
SW		 1197
1198 ceph_decode_need(&p, end, 8 + 4 + 4 + 8 + 4, bad);
1199 flags = ceph_decode_64(&p);
1200 result = ceph_decode_32(&p);
1201 reassert_epoch = ceph_decode_32(&p);
1202 reassert_version = ceph_decode_64(&p);
1203 osdmap_epoch = ceph_decode_32(&p);
1204
f24e9980
SW		 1205 /* lookup */
1206 mutex_lock(&osdc->request_mutex);
1207 req = __lookup_request(osdc, tid);
1208 if (req == NULL) {
1209 dout("handle_reply tid %llu dne\n", tid);
1210 mutex_unlock(&osdc->request_mutex);
1211 return;
1212 }
1213 ceph_osdc_get_request(req);
1b83bef2
SW		 1214
1215 dout("handle_reply %p tid %llu req %p result %d\n", msg, tid,
1216 req, result);
1217
1218 ceph_decode_need(&p, end, 4, bad);
1219 numops = ceph_decode_32(&p);
1220 if (numops > CEPH_OSD_MAX_OP)
1221 goto bad_put;
1222 if (numops != req->r_num_ops)
1223 goto bad_put;
1224 payload_len = 0;
1225 ceph_decode_need(&p, end, numops * sizeof(struct ceph_osd_op), bad);
1226 for (i = 0; i < numops; i++) {
1227 struct ceph_osd_op *op = p;
1228 int len;
1229
1230 len = le32_to_cpu(op->payload_len);
1231 req->r_reply_op_len[i] = len;
1232 dout(" op %d has %d bytes\n", i, len);
1233 payload_len += len;
1234 p += sizeof(*op);
1235 }
1236 if (payload_len != le32_to_cpu(msg->hdr.data_len)) {
1237 pr_warning("sum of op payload lens %d != data_len %d",
1238 payload_len, le32_to_cpu(msg->hdr.data_len));
1239 goto bad_put;
1240 }
1241
1242 ceph_decode_need(&p, end, 4 + numops * 4, bad);
1243 retry_attempt = ceph_decode_32(&p);
1244 for (i = 0; i < numops; i++)
1245 req->r_reply_op_result[i] = ceph_decode_32(&p);
f24e9980 1246
350b1c32 1247 /*
0d59ab81 1248 * if this connection filled our message, drop our reference now, to
350b1c32
SW		 1249 * avoid a (safe but slower) revoke later.
1250 */
0d59ab81 1251 if (req->r_con_filling_msg == con && req->r_reply == msg) {
c16e7869 1252 dout(" dropping con_filling_msg ref %p\n", con);
0d59ab81 1253 req->r_con_filling_msg = NULL;
0d47766f 1254 con->ops->put(con);
350b1c32
SW		 1255 }
1256
f24e9980 1257 if (!req->r_got_reply) {
95c96174 1258 unsigned int bytes;
f24e9980 1259
1b83bef2 1260 req->r_result = result;
f24e9980
SW		 1261 bytes = le32_to_cpu(msg->hdr.data_len);
1262 dout("handle_reply result %d bytes %d\n", req->r_result,
1263 bytes);
1264 if (req->r_result == 0)
1265 req->r_result = bytes;
1266
1267 /* in case this is a write and we need to replay, */
1b83bef2
SW		 1268 req->r_reassert_version.epoch = cpu_to_le32(reassert_epoch);
1269 req->r_reassert_version.version = cpu_to_le64(reassert_version);
f24e9980
SW		 1270
1271 req->r_got_reply = 1;
1272 } else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) {
1273 dout("handle_reply tid %llu dup ack\n", tid);
34b43a56 1274 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 1275 goto done;
1276 }
1277
1278 dout("handle_reply tid %llu flags %d\n", tid, flags);
1279
a40c4f10
YS		 1280 if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK))
1281 __register_linger_request(osdc, req);
1282
f24e9980 1283 /* either this is a read, or we got the safe response */
0ceed5db
SW		 1284 if (result < 0 ||
1285 (flags & CEPH_OSD_FLAG_ONDISK) ||
f24e9980
SW		 1286 ((flags & CEPH_OSD_FLAG_WRITE) == 0))
1287 __unregister_request(osdc, req);
1288
1289 mutex_unlock(&osdc->request_mutex);
1290
1291 if (req->r_callback)
1292 req->r_callback(req, msg);
1293 else
03066f23 1294 complete_all(&req->r_completion);
f24e9980 1295
25845472
SW		 1296 if (flags & CEPH_OSD_FLAG_ONDISK)
1297 complete_request(req);
f24e9980
SW		 1298
1299done:
a40c4f10 1300 dout("req=%p req->r_linger=%d\n", req, req->r_linger);
f24e9980
SW		 1301 ceph_osdc_put_request(req);
1302 return;
1303
1b83bef2
SW		 1304bad_put:
1305 ceph_osdc_put_request(req);
f24e9980 1306bad:
1b83bef2
SW		 1307 pr_err("corrupt osd_op_reply got %d %d\n",
1308 (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len));
9ec7cab1 1309 ceph_msg_dump(msg);
f24e9980
SW		 1310}
1311
6f6c7006 1312static void reset_changed_osds(struct ceph_osd_client *osdc)
f24e9980 1313{
f24e9980 1314 struct rb_node *p, *n;
f24e9980 1315
6f6c7006
SW		 1316 for (p = rb_first(&osdc->osds); p; p = n) {
1317 struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
f24e9980 1318
6f6c7006
SW		 1319 n = rb_next(p);
1320 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
1321 memcmp(&osd->o_con.peer_addr,
1322 ceph_osd_addr(osdc->osdmap,
1323 osd->o_osd),
1324 sizeof(struct ceph_entity_addr)) != 0)
1325 __reset_osd(osdc, osd);
f24e9980 1326 }
422d2cb8
YS		 1327}
1328
1329/*
6f6c7006
SW		 1330 * Requeue requests whose mapping to an OSD has changed. If requests map to
1331 * no osd, request a new map.
422d2cb8 1332 *
e6d50f67 1333 * Caller should hold map_sem for read.
422d2cb8 1334 */
38d6453c 1335static void kick_requests(struct ceph_osd_client *osdc, int force_resend)
422d2cb8 1336{
a40c4f10 1337 struct ceph_osd_request *req, *nreq;
6f6c7006
SW		 1338 struct rb_node *p;
1339 int needmap = 0;
1340 int err;
422d2cb8 1341
38d6453c 1342 dout("kick_requests %s\n", force_resend ? " (force resend)" : "");
422d2cb8 1343 mutex_lock(&osdc->request_mutex);
6194ea89 1344 for (p = rb_first(&osdc->requests); p; ) {
6f6c7006 1345 req = rb_entry(p, struct ceph_osd_request, r_node);
6194ea89 1346 p = rb_next(p);
ab60b16d
AE		 1347
1348 /*
1349 * For linger requests that have not yet been
1350 * registered, move them to the linger list; they'll
1351 * be sent to the osd in the loop below. Unregister
1352 * the request before re-registering it as a linger
1353 * request to ensure the __map_request() below
1354 * will decide it needs to be sent.
1355 */
1356 if (req->r_linger && list_empty(&req->r_linger_item)) {
1357 dout("%p tid %llu restart on osd%d\n",
1358 req, req->r_tid,
1359 req->r_osd ? req->r_osd->o_osd : -1);
1360 __unregister_request(osdc, req);
1361 __register_linger_request(osdc, req);
1362 continue;
1363 }
1364
38d6453c 1365 err = __map_request(osdc, req, force_resend);
6f6c7006
SW		 1366 if (err < 0)
1367 continue; /* error */
1368 if (req->r_osd == NULL) {
1369 dout("%p tid %llu maps to no osd\n", req, req->r_tid);
1370 needmap++; /* request a newer map */
1371 } else if (err > 0) {
6194ea89
SW		 1372 if (!req->r_linger) {
1373 dout("%p tid %llu requeued on osd%d\n", req,
1374 req->r_tid,
1375 req->r_osd ? req->r_osd->o_osd : -1);
a40c4f10 1376 req->r_flags |= CEPH_OSD_FLAG_RETRY;
6194ea89
SW		 1377 }
1378 }
a40c4f10
YS		 1379 }
1380
1381 list_for_each_entry_safe(req, nreq, &osdc->req_linger,
1382 r_linger_item) {
1383 dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
1384
38d6453c 1385 err = __map_request(osdc, req, force_resend);
ab60b16d 1386 dout("__map_request returned %d\n", err);
a40c4f10
YS		 1387 if (err == 0)
1388 continue; /* no change and no osd was specified */
1389 if (err < 0)
1390 continue; /* hrm! */
1391 if (req->r_osd == NULL) {
1392 dout("tid %llu maps to no valid osd\n", req->r_tid);
1393 needmap++; /* request a newer map */
1394 continue;
6f6c7006 1395 }
a40c4f10
YS		 1396
1397 dout("kicking lingering %p tid %llu osd%d\n", req, req->r_tid,
1398 req->r_osd ? req->r_osd->o_osd : -1);
a40c4f10 1399 __register_request(osdc, req);
c89ce05e 1400 __unregister_linger_request(osdc, req);
6f6c7006 1401 }
f24e9980
SW		 1402 mutex_unlock(&osdc->request_mutex);
1403
1404 if (needmap) {
1405 dout("%d requests for down osds, need new map\n", needmap);
1406 ceph_monc_request_next_osdmap(&osdc->client->monc);
1407 }
e6d50f67 1408 reset_changed_osds(osdc);
422d2cb8 1409}
6f6c7006
SW		 1410
1411
f24e9980
SW		 1412/*
1413 * Process updated osd map.
1414 *
1415 * The message contains any number of incremental and full maps, normally
1416 * indicating some sort of topology change in the cluster. Kick requests
1417 * off to different OSDs as needed.
1418 */
1419void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1420{
1421 void *p, *end, *next;
1422 u32 nr_maps, maplen;
1423 u32 epoch;
1424 struct ceph_osdmap *newmap = NULL, *oldmap;
1425 int err;
1426 struct ceph_fsid fsid;
1427
1428 dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
1429 p = msg->front.iov_base;
1430 end = p + msg->front.iov_len;
1431
1432 /* verify fsid */
1433 ceph_decode_need(&p, end, sizeof(fsid), bad);
1434 ceph_decode_copy(&p, &fsid, sizeof(fsid));
0743304d
SW		 1435 if (ceph_check_fsid(osdc->client, &fsid) < 0)
1436 return;
f24e9980
SW		 1437
1438 down_write(&osdc->map_sem);
1439
1440 /* incremental maps */
1441 ceph_decode_32_safe(&p, end, nr_maps, bad);
1442 dout(" %d inc maps\n", nr_maps);
1443 while (nr_maps > 0) {
1444 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 1445 epoch = ceph_decode_32(&p);
1446 maplen = ceph_decode_32(&p);
f24e9980
SW		 1447 ceph_decode_need(&p, end, maplen, bad);
1448 next = p + maplen;
1449 if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
1450 dout("applying incremental map %u len %d\n",
1451 epoch, maplen);
1452 newmap = osdmap_apply_incremental(&p, next,
1453 osdc->osdmap,
15d9882c 1454 &osdc->client->msgr);
f24e9980
SW		 1455 if (IS_ERR(newmap)) {
1456 err = PTR_ERR(newmap);
1457 goto bad;
1458 }
30dc6381 1459 BUG_ON(!newmap);
f24e9980
SW		 1460 if (newmap != osdc->osdmap) {
1461 ceph_osdmap_destroy(osdc->osdmap);
1462 osdc->osdmap = newmap;
1463 }
38d6453c 1464 kick_requests(osdc, 0);
f24e9980
SW		 1465 } else {
1466 dout("ignoring incremental map %u len %d\n",
1467 epoch, maplen);
1468 }
1469 p = next;
1470 nr_maps--;
1471 }
1472 if (newmap)
1473 goto done;
1474
1475 /* full maps */
1476 ceph_decode_32_safe(&p, end, nr_maps, bad);
1477 dout(" %d full maps\n", nr_maps);
1478 while (nr_maps) {
1479 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 1480 epoch = ceph_decode_32(&p);
1481 maplen = ceph_decode_32(&p);
f24e9980
SW		 1482 ceph_decode_need(&p, end, maplen, bad);
1483 if (nr_maps > 1) {
1484 dout("skipping non-latest full map %u len %d\n",
1485 epoch, maplen);
1486 } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
1487 dout("skipping full map %u len %d, "
1488 "older than our %u\n", epoch, maplen,
1489 osdc->osdmap->epoch);
1490 } else {
38d6453c
SW		 1491 int skipped_map = 0;
1492
f24e9980
SW		 1493 dout("taking full map %u len %d\n", epoch, maplen);
1494 newmap = osdmap_decode(&p, p+maplen);
1495 if (IS_ERR(newmap)) {
1496 err = PTR_ERR(newmap);
1497 goto bad;
1498 }
30dc6381 1499 BUG_ON(!newmap);
f24e9980
SW		 1500 oldmap = osdc->osdmap;
1501 osdc->osdmap = newmap;
38d6453c
SW		 1502 if (oldmap) {
1503 if (oldmap->epoch + 1 < newmap->epoch)
1504 skipped_map = 1;
f24e9980 1505 ceph_osdmap_destroy(oldmap);
38d6453c
SW		 1506 }
1507 kick_requests(osdc, skipped_map);
f24e9980
SW		 1508 }
1509 p += maplen;
1510 nr_maps--;
1511 }
1512
1513done:
1514 downgrade_write(&osdc->map_sem);
1515 ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch);
cd634fb6
SW		 1516
1517 /*
1518 * subscribe to subsequent osdmap updates if full to ensure
1519 * we find out when we are no longer full and stop returning
1520 * ENOSPC.
1521 */
1522 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL))
1523 ceph_monc_request_next_osdmap(&osdc->client->monc);
1524
f9d25199
AE		 1525 mutex_lock(&osdc->request_mutex);
1526 __send_queued(osdc);
1527 mutex_unlock(&osdc->request_mutex);
f24e9980 1528 up_read(&osdc->map_sem);
03066f23 1529 wake_up_all(&osdc->client->auth_wq);
f24e9980
SW		 1530 return;
1531
1532bad:
1533 pr_err("osdc handle_map corrupt msg\n");
9ec7cab1 1534 ceph_msg_dump(msg);
f24e9980
SW		 1535 up_write(&osdc->map_sem);
1536 return;
1537}
1538
a40c4f10
YS		 1539/*
1540 * watch/notify callback event infrastructure
1541 *
1542 * These callbacks are used both for watch and notify operations.
1543 */
1544static void __release_event(struct kref *kref)
1545{
1546 struct ceph_osd_event *event =
1547 container_of(kref, struct ceph_osd_event, kref);
1548
1549 dout("__release_event %p\n", event);
1550 kfree(event);
1551}
1552
1553static void get_event(struct ceph_osd_event *event)
1554{
1555 kref_get(&event->kref);
1556}
1557
1558void ceph_osdc_put_event(struct ceph_osd_event *event)
1559{
1560 kref_put(&event->kref, __release_event);
1561}
1562EXPORT_SYMBOL(ceph_osdc_put_event);
1563
1564static void __insert_event(struct ceph_osd_client *osdc,
1565 struct ceph_osd_event *new)
1566{
1567 struct rb_node **p = &osdc->event_tree.rb_node;
1568 struct rb_node *parent = NULL;
1569 struct ceph_osd_event *event = NULL;
1570
1571 while (*p) {
1572 parent = *p;
1573 event = rb_entry(parent, struct ceph_osd_event, node);
1574 if (new->cookie < event->cookie)
1575 p = &(*p)->rb_left;
1576 else if (new->cookie > event->cookie)
1577 p = &(*p)->rb_right;
1578 else
1579 BUG();
1580 }
1581
1582 rb_link_node(&new->node, parent, p);
1583 rb_insert_color(&new->node, &osdc->event_tree);
1584}
1585
1586static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
1587 u64 cookie)
1588{
1589 struct rb_node **p = &osdc->event_tree.rb_node;
1590 struct rb_node *parent = NULL;
1591 struct ceph_osd_event *event = NULL;
1592
1593 while (*p) {
1594 parent = *p;
1595 event = rb_entry(parent, struct ceph_osd_event, node);
1596 if (cookie < event->cookie)
1597 p = &(*p)->rb_left;
1598 else if (cookie > event->cookie)
1599 p = &(*p)->rb_right;
1600 else
1601 return event;
1602 }
1603 return NULL;
1604}
1605
1606static void __remove_event(struct ceph_osd_event *event)
1607{
1608 struct ceph_osd_client *osdc = event->osdc;
1609
1610 if (!RB_EMPTY_NODE(&event->node)) {
1611 dout("__remove_event removed %p\n", event);
1612 rb_erase(&event->node, &osdc->event_tree);
1613 ceph_osdc_put_event(event);
1614 } else {
1615 dout("__remove_event didn't remove %p\n", event);
1616 }
1617}
1618
1619int ceph_osdc_create_event(struct ceph_osd_client *osdc,
1620 void (*event_cb)(u64, u64, u8, void *),
3c663bbd 1621 void *data, struct ceph_osd_event **pevent)
a40c4f10
YS		 1622{
1623 struct ceph_osd_event *event;
1624
1625 event = kmalloc(sizeof(*event), GFP_NOIO);
1626 if (!event)
1627 return -ENOMEM;
1628
1629 dout("create_event %p\n", event);
1630 event->cb = event_cb;
3c663bbd 1631 event->one_shot = 0;
a40c4f10
YS		 1632 event->data = data;
1633 event->osdc = osdc;
1634 INIT_LIST_HEAD(&event->osd_node);
3ee5234d 1635 RB_CLEAR_NODE(&event->node);
a40c4f10
YS		 1636 kref_init(&event->kref); /* one ref for us */
1637 kref_get(&event->kref); /* one ref for the caller */
a40c4f10
YS		 1638
1639 spin_lock(&osdc->event_lock);
1640 event->cookie = ++osdc->event_count;
1641 __insert_event(osdc, event);
1642 spin_unlock(&osdc->event_lock);
1643
1644 *pevent = event;
1645 return 0;
1646}
1647EXPORT_SYMBOL(ceph_osdc_create_event);
1648
1649void ceph_osdc_cancel_event(struct ceph_osd_event *event)
1650{
1651 struct ceph_osd_client *osdc = event->osdc;
1652
1653 dout("cancel_event %p\n", event);
1654 spin_lock(&osdc->event_lock);
1655 __remove_event(event);
1656 spin_unlock(&osdc->event_lock);
1657 ceph_osdc_put_event(event); /* caller's */
1658}
1659EXPORT_SYMBOL(ceph_osdc_cancel_event);
1660
1661
1662static void do_event_work(struct work_struct *work)
1663{
1664 struct ceph_osd_event_work *event_work =
1665 container_of(work, struct ceph_osd_event_work, work);
1666 struct ceph_osd_event *event = event_work->event;
1667 u64 ver = event_work->ver;
1668 u64 notify_id = event_work->notify_id;
1669 u8 opcode = event_work->opcode;
1670
1671 dout("do_event_work completing %p\n", event);
1672 event->cb(ver, notify_id, opcode, event->data);
a40c4f10
YS		 1673 dout("do_event_work completed %p\n", event);
1674 ceph_osdc_put_event(event);
1675 kfree(event_work);
1676}
1677
1678
1679/*
1680 * Process osd watch notifications
1681 */
3c663bbd
AE		 1682static void handle_watch_notify(struct ceph_osd_client *osdc,
1683 struct ceph_msg *msg)
a40c4f10
YS		 1684{
1685 void *p, *end;
1686 u8 proto_ver;
1687 u64 cookie, ver, notify_id;
1688 u8 opcode;
1689 struct ceph_osd_event *event;
1690 struct ceph_osd_event_work *event_work;
1691
1692 p = msg->front.iov_base;
1693 end = p + msg->front.iov_len;
1694
1695 ceph_decode_8_safe(&p, end, proto_ver, bad);
1696 ceph_decode_8_safe(&p, end, opcode, bad);
1697 ceph_decode_64_safe(&p, end, cookie, bad);
1698 ceph_decode_64_safe(&p, end, ver, bad);
1699 ceph_decode_64_safe(&p, end, notify_id, bad);
1700
1701 spin_lock(&osdc->event_lock);
1702 event = __find_event(osdc, cookie);
1703 if (event) {
3c663bbd 1704 BUG_ON(event->one_shot);
a40c4f10 1705 get_event(event);
a40c4f10
YS		 1706 }
1707 spin_unlock(&osdc->event_lock);
1708 dout("handle_watch_notify cookie %lld ver %lld event %p\n",
1709 cookie, ver, event);
1710 if (event) {
1711 event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
a40c4f10
YS		 1712 if (!event_work) {
1713 dout("ERROR: could not allocate event_work\n");
1714 goto done_err;
1715 }
6b0ae409 1716 INIT_WORK(&event_work->work, do_event_work);
a40c4f10
YS		 1717 event_work->event = event;
1718 event_work->ver = ver;
1719 event_work->notify_id = notify_id;
1720 event_work->opcode = opcode;
1721 if (!queue_work(osdc->notify_wq, &event_work->work)) {
1722 dout("WARNING: failed to queue notify event work\n");
1723 goto done_err;
1724 }
1725 }
1726
1727 return;
1728
1729done_err:
a40c4f10
YS		 1730 ceph_osdc_put_event(event);
1731 return;
1732
1733bad:
1734 pr_err("osdc handle_watch_notify corrupt msg\n");
1735 return;
1736}
1737
f24e9980
SW		 1738/*
1739 * Register request, send initial attempt.
1740 */
1741int ceph_osdc_start_request(struct ceph_osd_client *osdc,
1742 struct ceph_osd_request *req,
1743 bool nofail)
1744{
c1ea8823 1745 int rc = 0;
f24e9980
SW		 1746
1747 req->r_request->pages = req->r_pages;
1748 req->r_request->nr_pages = req->r_num_pages;
68b4476b
YS		 1749#ifdef CONFIG_BLOCK
1750 req->r_request->bio = req->r_bio;
1751#endif
c885837f 1752 req->r_request->trail = &req->r_trail;
f24e9980
SW		 1753
1754 register_request(osdc, req);
1755
1756 down_read(&osdc->map_sem);
1757 mutex_lock(&osdc->request_mutex);
c1ea8823
SW		 1758 /*
1759 * a racing kick_requests() may have sent the message for us
1760 * while we dropped request_mutex above, so only send now if
1761 * the request still han't been touched yet.
1762 */
1763 if (req->r_sent == 0) {
38d6453c 1764 rc = __map_request(osdc, req, 0);
9d6fcb08
SW		 1765 if (rc < 0) {
1766 if (nofail) {
1767 dout("osdc_start_request failed map, "
1768 " will retry %lld\n", req->r_tid);
1769 rc = 0;
1770 }
234af26f 1771 goto out_unlock;
9d6fcb08 1772 }
6f6c7006
SW		 1773 if (req->r_osd == NULL) {
1774 dout("send_request %p no up osds in pg\n", req);
1775 ceph_monc_request_next_osdmap(&osdc->client->monc);
1776 } else {
56e925b6 1777 __send_request(osdc, req);
f24e9980 1778 }
56e925b6 1779 rc = 0;
f24e9980 1780 }
234af26f
DC		 1781
1782out_unlock:
f24e9980
SW		 1783 mutex_unlock(&osdc->request_mutex);
1784 up_read(&osdc->map_sem);
1785 return rc;
1786}
3d14c5d2 1787EXPORT_SYMBOL(ceph_osdc_start_request);
f24e9980
SW		 1788
1789/*
1790 * wait for a request to complete
1791 */
1792int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
1793 struct ceph_osd_request *req)
1794{
1795 int rc;
1796
1797 rc = wait_for_completion_interruptible(&req->r_completion);
1798 if (rc < 0) {
1799 mutex_lock(&osdc->request_mutex);
1800 __cancel_request(req);
529cfcc4 1801 __unregister_request(osdc, req);
f24e9980 1802 mutex_unlock(&osdc->request_mutex);
25845472 1803 complete_request(req);
529cfcc4 1804 dout("wait_request tid %llu canceled/timed out\n", req->r_tid);
f24e9980
SW		 1805 return rc;
1806 }
1807
1808 dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result);
1809 return req->r_result;
1810}
3d14c5d2 1811EXPORT_SYMBOL(ceph_osdc_wait_request);
f24e9980
SW		 1812
1813/*
1814 * sync - wait for all in-flight requests to flush. avoid starvation.
1815 */
1816void ceph_osdc_sync(struct ceph_osd_client *osdc)
1817{
1818 struct ceph_osd_request *req;
1819 u64 last_tid, next_tid = 0;
1820
1821 mutex_lock(&osdc->request_mutex);
1822 last_tid = osdc->last_tid;
1823 while (1) {
1824 req = __lookup_request_ge(osdc, next_tid);
1825 if (!req)
1826 break;
1827 if (req->r_tid > last_tid)
1828 break;
1829
1830 next_tid = req->r_tid + 1;
1831 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
1832 continue;
1833
1834 ceph_osdc_get_request(req);
1835 mutex_unlock(&osdc->request_mutex);
1836 dout("sync waiting on tid %llu (last is %llu)\n",
1837 req->r_tid, last_tid);
1838 wait_for_completion(&req->r_safe_completion);
1839 mutex_lock(&osdc->request_mutex);
1840 ceph_osdc_put_request(req);
1841 }
1842 mutex_unlock(&osdc->request_mutex);
1843 dout("sync done (thru tid %llu)\n", last_tid);
1844}
3d14c5d2 1845EXPORT_SYMBOL(ceph_osdc_sync);
f24e9980
SW		 1846
1847/*
1848 * init, shutdown
1849 */
1850int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
1851{
1852 int err;
1853
1854 dout("init\n");
1855 osdc->client = client;
1856 osdc->osdmap = NULL;
1857 init_rwsem(&osdc->map_sem);
1858 init_completion(&osdc->map_waiters);
1859 osdc->last_requested_map = 0;
1860 mutex_init(&osdc->request_mutex);
f24e9980
SW		 1861 osdc->last_tid = 0;
1862 osdc->osds = RB_ROOT;
f5a2041b 1863 INIT_LIST_HEAD(&osdc->osd_lru);
f24e9980 1864 osdc->requests = RB_ROOT;
422d2cb8 1865 INIT_LIST_HEAD(&osdc->req_lru);
6f6c7006
SW		 1866 INIT_LIST_HEAD(&osdc->req_unsent);
1867 INIT_LIST_HEAD(&osdc->req_notarget);
a40c4f10 1868 INIT_LIST_HEAD(&osdc->req_linger);
f24e9980
SW		 1869 osdc->num_requests = 0;
1870 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
f5a2041b 1871 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
a40c4f10
YS		 1872 spin_lock_init(&osdc->event_lock);
1873 osdc->event_tree = RB_ROOT;
1874 osdc->event_count = 0;
f5a2041b
YS		 1875
1876 schedule_delayed_work(&osdc->osds_timeout_work,
3d14c5d2 1877 round_jiffies_relative(osdc->client->options->osd_idle_ttl * HZ));
f24e9980 1878
5f44f142 1879 err = -ENOMEM;
f24e9980
SW		 1880 osdc->req_mempool = mempool_create_kmalloc_pool(10,
1881 sizeof(struct ceph_osd_request));
1882 if (!osdc->req_mempool)
5f44f142 1883 goto out;
f24e9980 1884
d50b409f
SW		 1885 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
1886 OSD_OP_FRONT_LEN, 10, true,
4f48280e 1887 "osd_op");
f24e9980 1888 if (err < 0)
5f44f142 1889 goto out_mempool;
d50b409f 1890 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
4f48280e
SW		 1891 OSD_OPREPLY_FRONT_LEN, 10, true,
1892 "osd_op_reply");
c16e7869
SW		 1893 if (err < 0)
1894 goto out_msgpool;
a40c4f10
YS		 1895
1896 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
1897 if (IS_ERR(osdc->notify_wq)) {
1898 err = PTR_ERR(osdc->notify_wq);
1899 osdc->notify_wq = NULL;
1900 goto out_msgpool;
1901 }
f24e9980 1902 return 0;
5f44f142 1903
c16e7869
SW		 1904out_msgpool:
1905 ceph_msgpool_destroy(&osdc->msgpool_op);
5f44f142
SW		 1906out_mempool:
1907 mempool_destroy(osdc->req_mempool);
1908out:
1909 return err;
f24e9980
SW		 1910}
1911
1912void ceph_osdc_stop(struct ceph_osd_client *osdc)
1913{
a40c4f10
YS		 1914 flush_workqueue(osdc->notify_wq);
1915 destroy_workqueue(osdc->notify_wq);
f24e9980 1916 cancel_delayed_work_sync(&osdc->timeout_work);
f5a2041b 1917 cancel_delayed_work_sync(&osdc->osds_timeout_work);
f24e9980
SW		 1918 if (osdc->osdmap) {
1919 ceph_osdmap_destroy(osdc->osdmap);
1920 osdc->osdmap = NULL;
1921 }
aca420bc 1922 remove_all_osds(osdc);
f24e9980
SW		 1923 mempool_destroy(osdc->req_mempool);
1924 ceph_msgpool_destroy(&osdc->msgpool_op);
c16e7869 1925 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
f24e9980
SW		 1926}
1927
1928/*
1929 * Read some contiguous pages. If we cross a stripe boundary, shorten
1930 * *plen. Return number of bytes read, or error.
1931 */
1932int ceph_osdc_readpages(struct ceph_osd_client *osdc,
1933 struct ceph_vino vino, struct ceph_file_layout *layout,
1934 u64 off, u64 *plen,
1935 u32 truncate_seq, u64 truncate_size,
b7495fc2 1936 struct page **pages, int num_pages, int page_align)
f24e9980
SW		 1937{
1938 struct ceph_osd_request *req;
1939 int rc = 0;
1940
1941 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
1942 vino.snap, off, *plen);
1943 req = ceph_osdc_new_request(osdc, layout, vino, off, plen,
1944 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
1945 NULL, 0, truncate_seq, truncate_size, NULL,
a3bea47e 1946 false, page_align);
6816282d
SW		 1947 if (IS_ERR(req))
1948 return PTR_ERR(req);
f24e9980
SW		 1949
1950 /* it may be a short read due to an object boundary */
1951 req->r_pages = pages;
f24e9980 1952
b7495fc2
SW		 1953 dout("readpages final extent is %llu~%llu (%d pages align %d)\n",
1954 off, *plen, req->r_num_pages, page_align);
f24e9980
SW		 1955
1956 rc = ceph_osdc_start_request(osdc, req, false);
1957 if (!rc)
1958 rc = ceph_osdc_wait_request(osdc, req);
1959
1960 ceph_osdc_put_request(req);
1961 dout("readpages result %d\n", rc);
1962 return rc;
1963}
3d14c5d2 1964EXPORT_SYMBOL(ceph_osdc_readpages);
f24e9980
SW		 1965
1966/*
1967 * do a synchronous write on N pages
1968 */
1969int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
1970 struct ceph_file_layout *layout,
1971 struct ceph_snap_context *snapc,
1972 u64 off, u64 len,
1973 u32 truncate_seq, u64 truncate_size,
1974 struct timespec *mtime,
24808826 1975 struct page **pages, int num_pages)
f24e9980
SW		 1976{
1977 struct ceph_osd_request *req;
1978 int rc = 0;
b7495fc2 1979 int page_align = off & ~PAGE_MASK;
f24e9980
SW		 1980
1981 BUG_ON(vino.snap != CEPH_NOSNAP);
1982 req = ceph_osdc_new_request(osdc, layout, vino, off, &len,
1983 CEPH_OSD_OP_WRITE,
24808826 1984 CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
fbf8685f 1985 snapc, 0,
f24e9980 1986 truncate_seq, truncate_size, mtime,
a3bea47e 1987 true, page_align);
6816282d
SW		 1988 if (IS_ERR(req))
1989 return PTR_ERR(req);
f24e9980
SW		 1990
1991 /* it may be a short write due to an object boundary */
1992 req->r_pages = pages;
f24e9980
SW		 1993 dout("writepages %llu~%llu (%d pages)\n", off, len,
1994 req->r_num_pages);
1995
87f979d3 1996 rc = ceph_osdc_start_request(osdc, req, true);
f24e9980
SW		 1997 if (!rc)
1998 rc = ceph_osdc_wait_request(osdc, req);
1999
2000 ceph_osdc_put_request(req);
2001 if (rc == 0)
2002 rc = len;
2003 dout("writepages result %d\n", rc);
2004 return rc;
2005}
3d14c5d2 2006EXPORT_SYMBOL(ceph_osdc_writepages);
f24e9980
SW		 2007
2008/*
2009 * handle incoming message
2010 */
2011static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
2012{
2013 struct ceph_osd *osd = con->private;
32c895e7 2014 struct ceph_osd_client *osdc;
f24e9980
SW		 2015 int type = le16_to_cpu(msg->hdr.type);
2016
2017 if (!osd)
4a32f93d 2018 goto out;
32c895e7 2019 osdc = osd->o_osdc;
f24e9980
SW		 2020
2021 switch (type) {
2022 case CEPH_MSG_OSD_MAP:
2023 ceph_osdc_handle_map(osdc, msg);
2024 break;
2025 case CEPH_MSG_OSD_OPREPLY:
350b1c32 2026 handle_reply(osdc, msg, con);
f24e9980 2027 break;
a40c4f10
YS		 2028 case CEPH_MSG_WATCH_NOTIFY:
2029 handle_watch_notify(osdc, msg);
2030 break;
f24e9980
SW		 2031
2032 default:
2033 pr_err("received unknown message type %d %s\n", type,
2034 ceph_msg_type_name(type));
2035 }
4a32f93d 2036out:
f24e9980
SW		 2037 ceph_msg_put(msg);
2038}
2039
5b3a4db3 2040/*
21b667f6
SW		 2041 * lookup and return message for incoming reply. set up reply message
2042 * pages.
5b3a4db3
SW		 2043 */
2044static struct ceph_msg *get_reply(struct ceph_connection *con,
2450418c
YS		 2045 struct ceph_msg_header *hdr,
2046 int *skip)
f24e9980
SW		 2047{
2048 struct ceph_osd *osd = con->private;
2049 struct ceph_osd_client *osdc = osd->o_osdc;
2450418c 2050 struct ceph_msg *m;
0547a9b3 2051 struct ceph_osd_request *req;
5b3a4db3
SW		 2052 int front = le32_to_cpu(hdr->front_len);
2053 int data_len = le32_to_cpu(hdr->data_len);
0547a9b3 2054 u64 tid;
f24e9980 2055
0547a9b3
YS		 2056 tid = le64_to_cpu(hdr->tid);
2057 mutex_lock(&osdc->request_mutex);
2058 req = __lookup_request(osdc, tid);
2059 if (!req) {
2060 *skip = 1;
2061 m = NULL;
756a16a5
SW		 2062 dout("get_reply unknown tid %llu from osd%d\n", tid,
2063 osd->o_osd);
0547a9b3
YS		 2064 goto out;
2065 }
c16e7869
SW		 2066
2067 if (req->r_con_filling_msg) {
8921d114 2068 dout("%s revoking msg %p from old con %p\n", __func__,
c16e7869 2069 req->r_reply, req->r_con_filling_msg);
8921d114 2070 ceph_msg_revoke_incoming(req->r_reply);
0d47766f 2071 req->r_con_filling_msg->ops->put(req->r_con_filling_msg);
6f46cb29 2072 req->r_con_filling_msg = NULL;
0547a9b3
YS		 2073 }
2074
c16e7869
SW		 2075 if (front > req->r_reply->front.iov_len) {
2076 pr_warning("get_reply front %d > preallocated %d\n",
2077 front, (int)req->r_reply->front.iov_len);
b61c2763 2078 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front, GFP_NOFS, false);
a79832f2 2079 if (!m)
c16e7869
SW		 2080 goto out;
2081 ceph_msg_put(req->r_reply);
2082 req->r_reply = m;
2083 }
2084 m = ceph_msg_get(req->r_reply);
2085
0547a9b3 2086 if (data_len > 0) {
b7495fc2 2087 int want = calc_pages_for(req->r_page_alignment, data_len);
21b667f6 2088
9cbb1d72 2089 if (req->r_pages && unlikely(req->r_num_pages < want)) {
9bb0ce2b
SW		 2090 pr_warning("tid %lld reply has %d bytes %d pages, we"
2091 " had only %d pages ready\n", tid, data_len,
2092 want, req->r_num_pages);
0547a9b3
YS		 2093 *skip = 1;
2094 ceph_msg_put(m);
a79832f2 2095 m = NULL;
21b667f6 2096 goto out;
0547a9b3 2097 }
21b667f6
SW		 2098 m->pages = req->r_pages;
2099 m->nr_pages = req->r_num_pages;
c5c6b19d 2100 m->page_alignment = req->r_page_alignment;
68b4476b
YS		 2101#ifdef CONFIG_BLOCK
2102 m->bio = req->r_bio;
2103#endif
0547a9b3 2104 }
5b3a4db3 2105 *skip = 0;
0d47766f 2106 req->r_con_filling_msg = con->ops->get(con);
c16e7869 2107 dout("get_reply tid %lld %p\n", tid, m);
0547a9b3
YS		 2108
2109out:
2110 mutex_unlock(&osdc->request_mutex);
2450418c 2111 return m;
5b3a4db3
SW		 2112
2113}
2114
2115static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2116 struct ceph_msg_header *hdr,
2117 int *skip)
2118{
2119 struct ceph_osd *osd = con->private;
2120 int type = le16_to_cpu(hdr->type);
2121 int front = le32_to_cpu(hdr->front_len);
2122
1c20f2d2 2123 *skip = 0;
5b3a4db3
SW		 2124 switch (type) {
2125 case CEPH_MSG_OSD_MAP:
a40c4f10 2126 case CEPH_MSG_WATCH_NOTIFY:
b61c2763 2127 return ceph_msg_new(type, front, GFP_NOFS, false);
5b3a4db3
SW		 2128 case CEPH_MSG_OSD_OPREPLY:
2129 return get_reply(con, hdr, skip);
2130 default:
2131 pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
2132 osd->o_osd);
2133 *skip = 1;
2134 return NULL;
2135 }
f24e9980
SW		 2136}
2137
2138/*
2139 * Wrappers to refcount containing ceph_osd struct
2140 */
2141static struct ceph_connection *get_osd_con(struct ceph_connection *con)
2142{
2143 struct ceph_osd *osd = con->private;
2144 if (get_osd(osd))
2145 return con;
2146 return NULL;
2147}
2148
2149static void put_osd_con(struct ceph_connection *con)
2150{
2151 struct ceph_osd *osd = con->private;
2152 put_osd(osd);
2153}
2154
4e7a5dcd
SW		 2155/*
2156 * authentication
2157 */
a3530df3
AE		 2158/*
2159 * Note: returned pointer is the address of a structure that's
2160 * managed separately. Caller must *not* attempt to free it.
2161 */
2162static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
8f43fb53 2163 int *proto, int force_new)
4e7a5dcd
SW		 2164{
2165 struct ceph_osd *o = con->private;
2166 struct ceph_osd_client *osdc = o->o_osdc;
2167 struct ceph_auth_client *ac = osdc->client->monc.auth;
74f1869f 2168 struct ceph_auth_handshake *auth = &o->o_auth;
4e7a5dcd 2169
74f1869f 2170 if (force_new && auth->authorizer) {
a255651d
AE		 2171 if (ac->ops && ac->ops->destroy_authorizer)
2172 ac->ops->destroy_authorizer(ac, auth->authorizer);
74f1869f
AE		 2173 auth->authorizer = NULL;
2174 }
a255651d 2175 if (!auth->authorizer && ac->ops && ac->ops->create_authorizer) {
a3530df3
AE		 2176 int ret = ac->ops->create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2177 auth);
4e7a5dcd 2178 if (ret)
a3530df3 2179 return ERR_PTR(ret);
4e7a5dcd 2180 }
4e7a5dcd 2181 *proto = ac->protocol;
74f1869f 2182
a3530df3 2183 return auth;
4e7a5dcd
SW		 2184}
2185
2186
2187static int verify_authorizer_reply(struct ceph_connection *con, int len)
2188{
2189 struct ceph_osd *o = con->private;
2190 struct ceph_osd_client *osdc = o->o_osdc;
2191 struct ceph_auth_client *ac = osdc->client->monc.auth;
2192
a255651d
AE		 2193 /*
2194 * XXX If ac->ops or ac->ops->verify_authorizer_reply is null,
2195 * XXX which do we do: succeed or fail?
2196 */
6c4a1915 2197 return ac->ops->verify_authorizer_reply(ac, o->o_auth.authorizer, len);
4e7a5dcd
SW		 2198}
2199
9bd2e6f8
SW		 2200static int invalidate_authorizer(struct ceph_connection *con)
2201{
2202 struct ceph_osd *o = con->private;
2203 struct ceph_osd_client *osdc = o->o_osdc;
2204 struct ceph_auth_client *ac = osdc->client->monc.auth;
2205
a255651d 2206 if (ac->ops && ac->ops->invalidate_authorizer)
9bd2e6f8
SW		 2207 ac->ops->invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
2208
2209 return ceph_monc_validate_auth(&osdc->client->monc);
2210}
4e7a5dcd 2211
9e32789f 2212static const struct ceph_connection_operations osd_con_ops = {
f24e9980
SW		 2213 .get = get_osd_con,
2214 .put = put_osd_con,
2215 .dispatch = dispatch,
4e7a5dcd
SW		 2216 .get_authorizer = get_authorizer,
2217 .verify_authorizer_reply = verify_authorizer_reply,
9bd2e6f8 2218 .invalidate_authorizer = invalidate_authorizer,
f24e9980 2219 .alloc_msg = alloc_msg,
81b024e7 2220 .fault = osd_reset,
f24e9980 2221};
Linux kernel - Deliverables
This page took 0.238668 seconds and 5 git commands to generate.