projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
libceph: complete lingering requests only once
[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 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 */
dbe0fc41 66static int calc_layout(struct ceph_file_layout *layout, u64 off, u64 *plen,
47a05811 67 struct ceph_osd_req_op *op, u64 *bno)
f24e9980 68{
60e56f13 69 u64 orig_len = *plen;
60e56f13
AE		 70 u64 objoff = 0;
71 u64 objlen = 0;
d63b77f4 72 int r;
f24e9980 73
60e56f13 74 /* object extent? */
47a05811 75 r = ceph_calc_file_object_mapping(layout, off, orig_len, bno,
60e56f13 76 &objoff, &objlen);
d63b77f4
SW		 77 if (r < 0)
78 return r;
60e56f13
AE		 79 if (objlen < orig_len) {
80 *plen = objlen;
81 dout(" skipping last %llu, final file extent %llu~%llu\n",
82 orig_len - *plen, off, *plen);
83 }
84
85 if (op_has_extent(op->op)) {
86 u32 osize = le32_to_cpu(layout->fl_object_size);
87 op->extent.offset = objoff;
88 op->extent.length = objlen;
89 if (op->extent.truncate_size <= off - objoff) {
90 op->extent.truncate_size = 0;
91 } else {
92 op->extent.truncate_size -= off - objoff;
93 if (op->extent.truncate_size > osize)
94 op->extent.truncate_size = osize;
95 }
96 }
60e56f13
AE		 97 if (op->op == CEPH_OSD_OP_WRITE)
98 op->payload_len = *plen;
99
60cf5992 100 dout("calc_layout bno=%llx %llu~%llu\n", *bno, objoff, objlen);
f24e9980 101
3ff5f385 102 return 0;
f24e9980
SW		 103}
104
f24e9980
SW		 105/*
106 * requests
107 */
415e49a9 108void ceph_osdc_release_request(struct kref *kref)
f24e9980 109{
415e49a9
SW		 110 struct ceph_osd_request *req = container_of(kref,
111 struct ceph_osd_request,
112 r_kref);
113
114 if (req->r_request)
115 ceph_msg_put(req->r_request);
0d59ab81 116 if (req->r_con_filling_msg) {
9cbb1d72
AE		 117 dout("%s revoking msg %p from con %p\n", __func__,
118 req->r_reply, req->r_con_filling_msg);
8921d114 119 ceph_msg_revoke_incoming(req->r_reply);
0d47766f 120 req->r_con_filling_msg->ops->put(req->r_con_filling_msg);
9cbb1d72 121 req->r_con_filling_msg = NULL;
350b1c32 122 }
ab8cb34a
AE		 123 if (req->r_reply)
124 ceph_msg_put(req->r_reply);
415e49a9
SW		 125 if (req->r_own_pages)
126 ceph_release_page_vector(req->r_pages,
127 req->r_num_pages);
128 ceph_put_snap_context(req->r_snapc);
c885837f 129 ceph_pagelist_release(&req->r_trail);
415e49a9
SW		 130 if (req->r_mempool)
131 mempool_free(req, req->r_osdc->req_mempool);
132 else
133 kfree(req);
f24e9980 134}
3d14c5d2 135EXPORT_SYMBOL(ceph_osdc_release_request);
68b4476b 136
3499e8a5 137struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
f24e9980 138 struct ceph_snap_context *snapc,
1b83bef2 139 unsigned int num_ops,
3499e8a5 140 bool use_mempool,
54a54007 141 gfp_t gfp_flags)
f24e9980
SW		 142{
143 struct ceph_osd_request *req;
144 struct ceph_msg *msg;
1b83bef2
SW		 145 size_t msg_size;
146
147 msg_size = 4 + 4 + 8 + 8 + 4+8;
148 msg_size += 2 + 4 + 8 + 4 + 4; /* oloc */
149 msg_size += 1 + 8 + 4 + 4; /* pg_t */
150 msg_size += 4 + MAX_OBJ_NAME_SIZE;
151 msg_size += 2 + num_ops*sizeof(struct ceph_osd_op);
152 msg_size += 8; /* snapid */
153 msg_size += 8; /* snap_seq */
154 msg_size += 8 * (snapc ? snapc->num_snaps : 0); /* snaps */
155 msg_size += 4;
f24e9980
SW		 156
157 if (use_mempool) {
3499e8a5 158 req = mempool_alloc(osdc->req_mempool, gfp_flags);
f24e9980
SW		 159 memset(req, 0, sizeof(*req));
160 } else {
3499e8a5 161 req = kzalloc(sizeof(*req), gfp_flags);
f24e9980
SW		 162 }
163 if (req == NULL)
a79832f2 164 return NULL;
f24e9980 165
f24e9980
SW		 166 req->r_osdc = osdc;
167 req->r_mempool = use_mempool;
68b4476b 168
415e49a9 169 kref_init(&req->r_kref);
f24e9980
SW		 170 init_completion(&req->r_completion);
171 init_completion(&req->r_safe_completion);
a978fa20 172 RB_CLEAR_NODE(&req->r_node);
f24e9980 173 INIT_LIST_HEAD(&req->r_unsafe_item);
a40c4f10
YS		 174 INIT_LIST_HEAD(&req->r_linger_item);
175 INIT_LIST_HEAD(&req->r_linger_osd);
935b639a 176 INIT_LIST_HEAD(&req->r_req_lru_item);
cd43045c
SW		 177 INIT_LIST_HEAD(&req->r_osd_item);
178
c16e7869
SW		 179 /* create reply message */
180 if (use_mempool)
181 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0);
182 else
183 msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY,
b61c2763 184 OSD_OPREPLY_FRONT_LEN, gfp_flags, true);
a79832f2 185 if (!msg) {
c16e7869 186 ceph_osdc_put_request(req);
a79832f2 187 return NULL;
c16e7869
SW		 188 }
189 req->r_reply = msg;
190
c885837f 191 ceph_pagelist_init(&req->r_trail);
d50b409f 192
c16e7869 193 /* create request message; allow space for oid */
f24e9980 194 if (use_mempool)
8f3bc053 195 msg = ceph_msgpool_get(&osdc->msgpool_op, 0);
f24e9980 196 else
b61c2763 197 msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp_flags, true);
a79832f2 198 if (!msg) {
f24e9980 199 ceph_osdc_put_request(req);
a79832f2 200 return NULL;
f24e9980 201 }
68b4476b 202
f24e9980 203 memset(msg->front.iov_base, 0, msg->front.iov_len);
3499e8a5
YS		 204
205 req->r_request = msg;
3499e8a5
YS		 206
207 return req;
208}
3d14c5d2 209EXPORT_SYMBOL(ceph_osdc_alloc_request);
3499e8a5 210
68b4476b
YS		 211static void osd_req_encode_op(struct ceph_osd_request *req,
212 struct ceph_osd_op *dst,
213 struct ceph_osd_req_op *src)
214{
215 dst->op = cpu_to_le16(src->op);
216
065a68f9 217 switch (src->op) {
fbfab539
AE		 218 case CEPH_OSD_OP_STAT:
219 break;
68b4476b
YS		 220 case CEPH_OSD_OP_READ:
221 case CEPH_OSD_OP_WRITE:
222 dst->extent.offset =
223 cpu_to_le64(src->extent.offset);
224 dst->extent.length =
225 cpu_to_le64(src->extent.length);
226 dst->extent.truncate_size =
227 cpu_to_le64(src->extent.truncate_size);
228 dst->extent.truncate_seq =
229 cpu_to_le32(src->extent.truncate_seq);
230 break;
ae1533b6 231 case CEPH_OSD_OP_CALL:
ae1533b6
YS		 232 dst->cls.class_len = src->cls.class_len;
233 dst->cls.method_len = src->cls.method_len;
234 dst->cls.indata_len = cpu_to_le32(src->cls.indata_len);
235
c885837f 236 ceph_pagelist_append(&req->r_trail, src->cls.class_name,
ae1533b6 237 src->cls.class_len);
c885837f 238 ceph_pagelist_append(&req->r_trail, src->cls.method_name,
ae1533b6 239 src->cls.method_len);
c885837f 240 ceph_pagelist_append(&req->r_trail, src->cls.indata,
ae1533b6
YS		 241 src->cls.indata_len);
242 break;
68b4476b
YS		 243 case CEPH_OSD_OP_STARTSYNC:
244 break;
a40c4f10
YS		 245 case CEPH_OSD_OP_NOTIFY_ACK:
246 case CEPH_OSD_OP_WATCH:
247 dst->watch.cookie = cpu_to_le64(src->watch.cookie);
248 dst->watch.ver = cpu_to_le64(src->watch.ver);
249 dst->watch.flag = src->watch.flag;
250 break;
68b4476b
YS		 251 default:
252 pr_err("unrecognized osd opcode %d\n", dst->op);
253 WARN_ON(1);
254 break;
4c46459c
AE		 255 case CEPH_OSD_OP_MAPEXT:
256 case CEPH_OSD_OP_MASKTRUNC:
257 case CEPH_OSD_OP_SPARSE_READ:
a9f36c3e 258 case CEPH_OSD_OP_NOTIFY:
4c46459c
AE		 259 case CEPH_OSD_OP_ASSERT_VER:
260 case CEPH_OSD_OP_WRITEFULL:
261 case CEPH_OSD_OP_TRUNCATE:
262 case CEPH_OSD_OP_ZERO:
263 case CEPH_OSD_OP_DELETE:
264 case CEPH_OSD_OP_APPEND:
265 case CEPH_OSD_OP_SETTRUNC:
266 case CEPH_OSD_OP_TRIMTRUNC:
267 case CEPH_OSD_OP_TMAPUP:
268 case CEPH_OSD_OP_TMAPPUT:
269 case CEPH_OSD_OP_TMAPGET:
270 case CEPH_OSD_OP_CREATE:
a9f36c3e 271 case CEPH_OSD_OP_ROLLBACK:
4c46459c
AE		 272 case CEPH_OSD_OP_OMAPGETKEYS:
273 case CEPH_OSD_OP_OMAPGETVALS:
274 case CEPH_OSD_OP_OMAPGETHEADER:
275 case CEPH_OSD_OP_OMAPGETVALSBYKEYS:
276 case CEPH_OSD_OP_MODE_RD:
277 case CEPH_OSD_OP_OMAPSETVALS:
278 case CEPH_OSD_OP_OMAPSETHEADER:
279 case CEPH_OSD_OP_OMAPCLEAR:
280 case CEPH_OSD_OP_OMAPRMKEYS:
281 case CEPH_OSD_OP_OMAP_CMP:
282 case CEPH_OSD_OP_CLONERANGE:
283 case CEPH_OSD_OP_ASSERT_SRC_VERSION:
284 case CEPH_OSD_OP_SRC_CMPXATTR:
a9f36c3e 285 case CEPH_OSD_OP_GETXATTR:
4c46459c 286 case CEPH_OSD_OP_GETXATTRS:
a9f36c3e
AE		 287 case CEPH_OSD_OP_CMPXATTR:
288 case CEPH_OSD_OP_SETXATTR:
4c46459c
AE		 289 case CEPH_OSD_OP_SETXATTRS:
290 case CEPH_OSD_OP_RESETXATTRS:
291 case CEPH_OSD_OP_RMXATTR:
292 case CEPH_OSD_OP_PULL:
293 case CEPH_OSD_OP_PUSH:
294 case CEPH_OSD_OP_BALANCEREADS:
295 case CEPH_OSD_OP_UNBALANCEREADS:
296 case CEPH_OSD_OP_SCRUB:
297 case CEPH_OSD_OP_SCRUB_RESERVE:
298 case CEPH_OSD_OP_SCRUB_UNRESERVE:
299 case CEPH_OSD_OP_SCRUB_STOP:
300 case CEPH_OSD_OP_SCRUB_MAP:
301 case CEPH_OSD_OP_WRLOCK:
302 case CEPH_OSD_OP_WRUNLOCK:
303 case CEPH_OSD_OP_RDLOCK:
304 case CEPH_OSD_OP_RDUNLOCK:
305 case CEPH_OSD_OP_UPLOCK:
306 case CEPH_OSD_OP_DNLOCK:
307 case CEPH_OSD_OP_PGLS:
308 case CEPH_OSD_OP_PGLS_FILTER:
309 pr_err("unsupported osd opcode %s\n",
310 ceph_osd_op_name(dst->op));
311 WARN_ON(1);
312 break;
68b4476b
YS		 313 }
314 dst->payload_len = cpu_to_le32(src->payload_len);
315}
316
3499e8a5
YS		 317/*
318 * build new request AND message
319 *
320 */
321void ceph_osdc_build_request(struct ceph_osd_request *req,
1b83bef2 322 u64 off, u64 len, unsigned int num_ops,
68b4476b 323 struct ceph_osd_req_op *src_ops,
4d6b250b 324 struct ceph_snap_context *snapc, u64 snap_id,
af77f26c 325 struct timespec *mtime)
3499e8a5
YS		 326{
327 struct ceph_msg *msg = req->r_request;
68b4476b 328 struct ceph_osd_req_op *src_op;
3499e8a5 329 void *p;
1b83bef2 330 size_t msg_size;
3499e8a5 331 int flags = req->r_flags;
f44246e3 332 u64 data_len;
68b4476b 333 int i;
3499e8a5 334
1b83bef2
SW		 335 req->r_num_ops = num_ops;
336 req->r_snapid = snap_id;
f24e9980
SW		 337 req->r_snapc = ceph_get_snap_context(snapc);
338
1b83bef2
SW		 339 /* encode request */
340 msg->hdr.version = cpu_to_le16(4);
341
342 p = msg->front.iov_base;
343 ceph_encode_32(&p, 1); /* client_inc is always 1 */
344 req->r_request_osdmap_epoch = p;
345 p += 4;
346 req->r_request_flags = p;
347 p += 4;
348 if (req->r_flags & CEPH_OSD_FLAG_WRITE)
349 ceph_encode_timespec(p, mtime);
350 p += sizeof(struct ceph_timespec);
351 req->r_request_reassert_version = p;
352 p += sizeof(struct ceph_eversion); /* will get filled in */
353
354 /* oloc */
355 ceph_encode_8(&p, 4);
356 ceph_encode_8(&p, 4);
357 ceph_encode_32(&p, 8 + 4 + 4);
358 req->r_request_pool = p;
359 p += 8;
360 ceph_encode_32(&p, -1); /* preferred */
361 ceph_encode_32(&p, 0); /* key len */
362
363 ceph_encode_8(&p, 1);
364 req->r_request_pgid = p;
365 p += 8 + 4;
366 ceph_encode_32(&p, -1); /* preferred */
367
368 /* oid */
369 ceph_encode_32(&p, req->r_oid_len);
af77f26c 370 memcpy(p, req->r_oid, req->r_oid_len);
1b83bef2 371 dout("oid '%.*s' len %d\n", req->r_oid_len, req->r_oid, req->r_oid_len);
af77f26c 372 p += req->r_oid_len;
f24e9980 373
1b83bef2
SW		 374 /* ops */
375 ceph_encode_16(&p, num_ops);
68b4476b 376 src_op = src_ops;
1b83bef2
SW		 377 req->r_request_ops = p;
378 for (i = 0; i < num_ops; i++, src_op++) {
379 osd_req_encode_op(req, p, src_op);
380 p += sizeof(struct ceph_osd_op);
381 }
68b4476b 382
1b83bef2
SW		 383 /* snaps */
384 ceph_encode_64(&p, req->r_snapid);
385 ceph_encode_64(&p, req->r_snapc ? req->r_snapc->seq : 0);
386 ceph_encode_32(&p, req->r_snapc ? req->r_snapc->num_snaps : 0);
387 if (req->r_snapc) {
f24e9980 388 for (i = 0; i < snapc->num_snaps; i++) {
1b83bef2 389 ceph_encode_64(&p, req->r_snapc->snaps[i]);
f24e9980
SW		 390 }
391 }
392
1b83bef2
SW		 393 req->r_request_attempts = p;
394 p += 4;
395
f44246e3 396 data_len = req->r_trail.length;
68b4476b
YS		 397 if (flags & CEPH_OSD_FLAG_WRITE) {
398 req->r_request->hdr.data_off = cpu_to_le16(off);
f44246e3 399 data_len += len;
68b4476b 400 }
f44246e3 401 req->r_request->hdr.data_len = cpu_to_le32(data_len);
c5c6b19d 402
f24e9980 403 BUG_ON(p > msg->front.iov_base + msg->front.iov_len);
6f863e71
SW		 404 msg_size = p - msg->front.iov_base;
405 msg->front.iov_len = msg_size;
406 msg->hdr.front_len = cpu_to_le32(msg_size);
1b83bef2
SW		 407
408 dout("build_request msg_size was %d num_ops %d\n", (int)msg_size,
409 num_ops);
3499e8a5
YS		 410 return;
411}
3d14c5d2 412EXPORT_SYMBOL(ceph_osdc_build_request);
3499e8a5
YS		 413
414/*
415 * build new request AND message, calculate layout, and adjust file
416 * extent as needed.
417 *
418 * if the file was recently truncated, we include information about its
419 * old and new size so that the object can be updated appropriately. (we
420 * avoid synchronously deleting truncated objects because it's slow.)
421 *
422 * if @do_sync, include a 'startsync' command so that the osd will flush
423 * data quickly.
424 */
425struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
426 struct ceph_file_layout *layout,
427 struct ceph_vino vino,
428 u64 off, u64 *plen,
429 int opcode, int flags,
430 struct ceph_snap_context *snapc,
431 int do_sync,
432 u32 truncate_seq,
433 u64 truncate_size,
434 struct timespec *mtime,
a3bea47e 435 bool use_mempool,
b7495fc2 436 int page_align)
3499e8a5 437{
ae7ca4a3 438 struct ceph_osd_req_op ops[2];
68b4476b 439 struct ceph_osd_request *req;
ae7ca4a3 440 unsigned int num_op = 1;
47a05811 441 u64 bno = 0;
6816282d 442 int r;
68b4476b 443
ae7ca4a3
AE		 444 memset(&ops, 0, sizeof ops);
445
68b4476b
YS		 446 ops[0].op = opcode;
447 ops[0].extent.truncate_seq = truncate_seq;
448 ops[0].extent.truncate_size = truncate_size;
68b4476b
YS		 449
450 if (do_sync) {
451 ops[1].op = CEPH_OSD_OP_STARTSYNC;
ae7ca4a3
AE		 452 num_op++;
453 }
68b4476b 454
ae7ca4a3
AE		 455 req = ceph_osdc_alloc_request(osdc, snapc, num_op, use_mempool,
456 GFP_NOFS);
4ad12621 457 if (!req)
6816282d 458 return ERR_PTR(-ENOMEM);
d178a9e7 459 req->r_flags = flags;
3499e8a5
YS		 460
461 /* calculate max write size */
60cf5992 462 r = calc_layout(layout, off, plen, ops, &bno);
3ff5f385
AE		 463 if (r < 0) {
464 ceph_osdc_put_request(req);
6816282d 465 return ERR_PTR(r);
3ff5f385 466 }
47a05811 467
3499e8a5
YS		 468 req->r_file_layout = *layout; /* keep a copy */
469
dbe0fc41
AE		 470 snprintf(req->r_oid, sizeof(req->r_oid), "%llx.%08llx", vino.ino, bno);
471 req->r_oid_len = strlen(req->r_oid);
472
60cf5992
AE		 473 /* The alignment may differ from the natural (file) alignment */
474
9bb0ce2b 475 req->r_num_pages = calc_pages_for(page_align, *plen);
b7495fc2
SW		 476 req->r_page_alignment = page_align;
477
ae7ca4a3
AE		 478 ceph_osdc_build_request(req, off, *plen, num_op, ops,
479 snapc, vino.snap, mtime);
3499e8a5 480
f24e9980
SW		 481 return req;
482}
3d14c5d2 483EXPORT_SYMBOL(ceph_osdc_new_request);
f24e9980
SW		 484
485/*
486 * We keep osd requests in an rbtree, sorted by ->r_tid.
487 */
488static void __insert_request(struct ceph_osd_client *osdc,
489 struct ceph_osd_request *new)
490{
491 struct rb_node **p = &osdc->requests.rb_node;
492 struct rb_node *parent = NULL;
493 struct ceph_osd_request *req = NULL;
494
495 while (*p) {
496 parent = *p;
497 req = rb_entry(parent, struct ceph_osd_request, r_node);
498 if (new->r_tid < req->r_tid)
499 p = &(*p)->rb_left;
500 else if (new->r_tid > req->r_tid)
501 p = &(*p)->rb_right;
502 else
503 BUG();
504 }
505
506 rb_link_node(&new->r_node, parent, p);
507 rb_insert_color(&new->r_node, &osdc->requests);
508}
509
510static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc,
511 u64 tid)
512{
513 struct ceph_osd_request *req;
514 struct rb_node *n = osdc->requests.rb_node;
515
516 while (n) {
517 req = rb_entry(n, struct ceph_osd_request, r_node);
518 if (tid < req->r_tid)
519 n = n->rb_left;
520 else if (tid > req->r_tid)
521 n = n->rb_right;
522 else
523 return req;
524 }
525 return NULL;
526}
527
528static struct ceph_osd_request *
529__lookup_request_ge(struct ceph_osd_client *osdc,
530 u64 tid)
531{
532 struct ceph_osd_request *req;
533 struct rb_node *n = osdc->requests.rb_node;
534
535 while (n) {
536 req = rb_entry(n, struct ceph_osd_request, r_node);
537 if (tid < req->r_tid) {
538 if (!n->rb_left)
539 return req;
540 n = n->rb_left;
541 } else if (tid > req->r_tid) {
542 n = n->rb_right;
543 } else {
544 return req;
545 }
546 }
547 return NULL;
548}
549
6f6c7006
SW		 550/*
551 * Resubmit requests pending on the given osd.
552 */
553static void __kick_osd_requests(struct ceph_osd_client *osdc,
554 struct ceph_osd *osd)
555{
a40c4f10 556 struct ceph_osd_request *req, *nreq;
6f6c7006
SW		 557 int err;
558
559 dout("__kick_osd_requests osd%d\n", osd->o_osd);
560 err = __reset_osd(osdc, osd);
685a7555 561 if (err)
6f6c7006
SW		 562 return;
563
564 list_for_each_entry(req, &osd->o_requests, r_osd_item) {
565 list_move(&req->r_req_lru_item, &osdc->req_unsent);
566 dout("requeued %p tid %llu osd%d\n", req, req->r_tid,
567 osd->o_osd);
a40c4f10
YS		 568 if (!req->r_linger)
569 req->r_flags |= CEPH_OSD_FLAG_RETRY;
570 }
571
572 list_for_each_entry_safe(req, nreq, &osd->o_linger_requests,
573 r_linger_osd) {
77f38e0e
SW		 574 /*
575 * reregister request prior to unregistering linger so
576 * that r_osd is preserved.
577 */
578 BUG_ON(!list_empty(&req->r_req_lru_item));
a40c4f10 579 __register_request(osdc, req);
77f38e0e
SW		 580 list_add(&req->r_req_lru_item, &osdc->req_unsent);
581 list_add(&req->r_osd_item, &req->r_osd->o_requests);
582 __unregister_linger_request(osdc, req);
a40c4f10
YS		 583 dout("requeued lingering %p tid %llu osd%d\n", req, req->r_tid,
584 osd->o_osd);
6f6c7006
SW		 585 }
586}
587
f24e9980 588/*
81b024e7 589 * If the osd connection drops, we need to resubmit all requests.
f24e9980
SW		 590 */
591static void osd_reset(struct ceph_connection *con)
592{
593 struct ceph_osd *osd = con->private;
594 struct ceph_osd_client *osdc;
595
596 if (!osd)
597 return;
598 dout("osd_reset osd%d\n", osd->o_osd);
599 osdc = osd->o_osdc;
f24e9980 600 down_read(&osdc->map_sem);
83aff95e
SW		 601 mutex_lock(&osdc->request_mutex);
602 __kick_osd_requests(osdc, osd);
f9d25199 603 __send_queued(osdc);
83aff95e 604 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 605 up_read(&osdc->map_sem);
606}
607
608/*
609 * Track open sessions with osds.
610 */
e10006f8 611static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
f24e9980
SW		 612{
613 struct ceph_osd *osd;
614
615 osd = kzalloc(sizeof(*osd), GFP_NOFS);
616 if (!osd)
617 return NULL;
618
619 atomic_set(&osd->o_ref, 1);
620 osd->o_osdc = osdc;
e10006f8 621 osd->o_osd = onum;
f407731d 622 RB_CLEAR_NODE(&osd->o_node);
f24e9980 623 INIT_LIST_HEAD(&osd->o_requests);
a40c4f10 624 INIT_LIST_HEAD(&osd->o_linger_requests);
f5a2041b 625 INIT_LIST_HEAD(&osd->o_osd_lru);
f24e9980
SW		 626 osd->o_incarnation = 1;
627
b7a9e5dd 628 ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
4e7a5dcd 629
422d2cb8 630 INIT_LIST_HEAD(&osd->o_keepalive_item);
f24e9980
SW		 631 return osd;
632}
633
634static struct ceph_osd *get_osd(struct ceph_osd *osd)
635{
636 if (atomic_inc_not_zero(&osd->o_ref)) {
637 dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
638 atomic_read(&osd->o_ref));
639 return osd;
640 } else {
641 dout("get_osd %p FAIL\n", osd);
642 return NULL;
643 }
644}
645
646static void put_osd(struct ceph_osd *osd)
647{
648 dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
649 atomic_read(&osd->o_ref) - 1);
a255651d 650 if (atomic_dec_and_test(&osd->o_ref) && osd->o_auth.authorizer) {
79494d1b
SW		 651 struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth;
652
a255651d 653 if (ac->ops && ac->ops->destroy_authorizer)
6c4a1915 654 ac->ops->destroy_authorizer(ac, osd->o_auth.authorizer);
f24e9980 655 kfree(osd);
79494d1b 656 }
f24e9980
SW		 657}
658
659/*
660 * remove an osd from our map
661 */
f5a2041b 662static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
f24e9980 663{
f5a2041b 664 dout("__remove_osd %p\n", osd);
f24e9980
SW		 665 BUG_ON(!list_empty(&osd->o_requests));
666 rb_erase(&osd->o_node, &osdc->osds);
f5a2041b 667 list_del_init(&osd->o_osd_lru);
f24e9980
SW		 668 ceph_con_close(&osd->o_con);
669 put_osd(osd);
670}
671
aca420bc
SW		 672static void remove_all_osds(struct ceph_osd_client *osdc)
673{
048a9d2d 674 dout("%s %p\n", __func__, osdc);
aca420bc
SW		 675 mutex_lock(&osdc->request_mutex);
676 while (!RB_EMPTY_ROOT(&osdc->osds)) {
677 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
678 struct ceph_osd, o_node);
679 __remove_osd(osdc, osd);
680 }
681 mutex_unlock(&osdc->request_mutex);
682}
683
f5a2041b
YS		 684static void __move_osd_to_lru(struct ceph_osd_client *osdc,
685 struct ceph_osd *osd)
686{
687 dout("__move_osd_to_lru %p\n", osd);
688 BUG_ON(!list_empty(&osd->o_osd_lru));
689 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
3d14c5d2 690 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl * HZ;
f5a2041b
YS		 691}
692
693static void __remove_osd_from_lru(struct ceph_osd *osd)
694{
695 dout("__remove_osd_from_lru %p\n", osd);
696 if (!list_empty(&osd->o_osd_lru))
697 list_del_init(&osd->o_osd_lru);
698}
699
aca420bc 700static void remove_old_osds(struct ceph_osd_client *osdc)
f5a2041b
YS		 701{
702 struct ceph_osd *osd, *nosd;
703
704 dout("__remove_old_osds %p\n", osdc);
705 mutex_lock(&osdc->request_mutex);
706 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
aca420bc 707 if (time_before(jiffies, osd->lru_ttl))
f5a2041b
YS		 708 break;
709 __remove_osd(osdc, osd);
710 }
711 mutex_unlock(&osdc->request_mutex);
712}
713
f24e9980
SW		 714/*
715 * reset osd connect
716 */
f5a2041b 717static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
f24e9980 718{
c3acb181 719 struct ceph_entity_addr *peer_addr;
f24e9980 720
f5a2041b 721 dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
a40c4f10
YS		 722 if (list_empty(&osd->o_requests) &&
723 list_empty(&osd->o_linger_requests)) {
f5a2041b 724 __remove_osd(osdc, osd);
c3acb181
AE		 725
726 return -ENODEV;
727 }
728
729 peer_addr = &osdc->osdmap->osd_addr[osd->o_osd];
730 if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
731 !ceph_con_opened(&osd->o_con)) {
732 struct ceph_osd_request *req;
733
87b315a5
SW		 734 dout(" osd addr hasn't changed and connection never opened,"
735 " letting msgr retry");
736 /* touch each r_stamp for handle_timeout()'s benfit */
737 list_for_each_entry(req, &osd->o_requests, r_osd_item)
738 req->r_stamp = jiffies;
c3acb181
AE		 739
740 return -EAGAIN;
f24e9980 741 }
c3acb181
AE		 742
743 ceph_con_close(&osd->o_con);
744 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
745 osd->o_incarnation++;
746
747 return 0;
f24e9980
SW		 748}
749
750static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new)
751{
752 struct rb_node **p = &osdc->osds.rb_node;
753 struct rb_node *parent = NULL;
754 struct ceph_osd *osd = NULL;
755
aca420bc 756 dout("__insert_osd %p osd%d\n", new, new->o_osd);
f24e9980
SW		 757 while (*p) {
758 parent = *p;
759 osd = rb_entry(parent, struct ceph_osd, o_node);
760 if (new->o_osd < osd->o_osd)
761 p = &(*p)->rb_left;
762 else if (new->o_osd > osd->o_osd)
763 p = &(*p)->rb_right;
764 else
765 BUG();
766 }
767
768 rb_link_node(&new->o_node, parent, p);
769 rb_insert_color(&new->o_node, &osdc->osds);
770}
771
772static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o)
773{
774 struct ceph_osd *osd;
775 struct rb_node *n = osdc->osds.rb_node;
776
777 while (n) {
778 osd = rb_entry(n, struct ceph_osd, o_node);
779 if (o < osd->o_osd)
780 n = n->rb_left;
781 else if (o > osd->o_osd)
782 n = n->rb_right;
783 else
784 return osd;
785 }
786 return NULL;
787}
788
422d2cb8
YS		 789static void __schedule_osd_timeout(struct ceph_osd_client *osdc)
790{
791 schedule_delayed_work(&osdc->timeout_work,
3d14c5d2 792 osdc->client->options->osd_keepalive_timeout * HZ);
422d2cb8
YS		 793}
794
795static void __cancel_osd_timeout(struct ceph_osd_client *osdc)
796{
797 cancel_delayed_work(&osdc->timeout_work);
798}
f24e9980
SW		 799
800/*
801 * Register request, assign tid. If this is the first request, set up
802 * the timeout event.
803 */
a40c4f10
YS		 804static void __register_request(struct ceph_osd_client *osdc,
805 struct ceph_osd_request *req)
f24e9980 806{
f24e9980 807 req->r_tid = ++osdc->last_tid;
6df058c0 808 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
77f38e0e 809 dout("__register_request %p tid %lld\n", req, req->r_tid);
f24e9980
SW		 810 __insert_request(osdc, req);
811 ceph_osdc_get_request(req);
812 osdc->num_requests++;
f24e9980 813 if (osdc->num_requests == 1) {
422d2cb8
YS		 814 dout(" first request, scheduling timeout\n");
815 __schedule_osd_timeout(osdc);
f24e9980 816 }
a40c4f10
YS		 817}
818
819static void register_request(struct ceph_osd_client *osdc,
820 struct ceph_osd_request *req)
821{
822 mutex_lock(&osdc->request_mutex);
823 __register_request(osdc, req);
f24e9980
SW		 824 mutex_unlock(&osdc->request_mutex);
825}
826
827/*
828 * called under osdc->request_mutex
829 */
830static void __unregister_request(struct ceph_osd_client *osdc,
831 struct ceph_osd_request *req)
832{
35f9f8a0
SW		 833 if (RB_EMPTY_NODE(&req->r_node)) {
834 dout("__unregister_request %p tid %lld not registered\n",
835 req, req->r_tid);
836 return;
837 }
838
f24e9980
SW		 839 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
840 rb_erase(&req->r_node, &osdc->requests);
841 osdc->num_requests--;
842
0ba6478d
SW		 843 if (req->r_osd) {
844 /* make sure the original request isn't in flight. */
6740a845 845 ceph_msg_revoke(req->r_request);
0ba6478d
SW		 846
847 list_del_init(&req->r_osd_item);
a40c4f10
YS		 848 if (list_empty(&req->r_osd->o_requests) &&
849 list_empty(&req->r_osd->o_linger_requests)) {
850 dout("moving osd to %p lru\n", req->r_osd);
f5a2041b 851 __move_osd_to_lru(osdc, req->r_osd);
a40c4f10 852 }
fbdb9190 853 if (list_empty(&req->r_linger_item))
a40c4f10 854 req->r_osd = NULL;
0ba6478d 855 }
f24e9980 856
7d5f2481 857 list_del_init(&req->r_req_lru_item);
f24e9980
SW		 858 ceph_osdc_put_request(req);
859
422d2cb8
YS		 860 if (osdc->num_requests == 0) {
861 dout(" no requests, canceling timeout\n");
862 __cancel_osd_timeout(osdc);
f24e9980
SW		 863 }
864}
865
866/*
867 * Cancel a previously queued request message
868 */
869static void __cancel_request(struct ceph_osd_request *req)
870{
6bc18876 871 if (req->r_sent && req->r_osd) {
6740a845 872 ceph_msg_revoke(req->r_request);
f24e9980
SW		 873 req->r_sent = 0;
874 }
875}
876
a40c4f10
YS		 877static void __register_linger_request(struct ceph_osd_client *osdc,
878 struct ceph_osd_request *req)
879{
880 dout("__register_linger_request %p\n", req);
881 list_add_tail(&req->r_linger_item, &osdc->req_linger);
6194ea89
SW		 882 if (req->r_osd)
883 list_add_tail(&req->r_linger_osd,
884 &req->r_osd->o_linger_requests);
a40c4f10
YS		 885}
886
887static void __unregister_linger_request(struct ceph_osd_client *osdc,
888 struct ceph_osd_request *req)
889{
890 dout("__unregister_linger_request %p\n", req);
61c74035 891 list_del_init(&req->r_linger_item);
a40c4f10 892 if (req->r_osd) {
a40c4f10
YS		 893 list_del_init(&req->r_linger_osd);
894
895 if (list_empty(&req->r_osd->o_requests) &&
896 list_empty(&req->r_osd->o_linger_requests)) {
897 dout("moving osd to %p lru\n", req->r_osd);
898 __move_osd_to_lru(osdc, req->r_osd);
899 }
fbdb9190
SW		 900 if (list_empty(&req->r_osd_item))
901 req->r_osd = NULL;
a40c4f10
YS		 902 }
903}
904
905void ceph_osdc_unregister_linger_request(struct ceph_osd_client *osdc,
906 struct ceph_osd_request *req)
907{
908 mutex_lock(&osdc->request_mutex);
909 if (req->r_linger) {
910 __unregister_linger_request(osdc, req);
911 ceph_osdc_put_request(req);
912 }
913 mutex_unlock(&osdc->request_mutex);
914}
915EXPORT_SYMBOL(ceph_osdc_unregister_linger_request);
916
917void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
918 struct ceph_osd_request *req)
919{
920 if (!req->r_linger) {
921 dout("set_request_linger %p\n", req);
922 req->r_linger = 1;
923 /*
924 * caller is now responsible for calling
925 * unregister_linger_request
926 */
927 ceph_osdc_get_request(req);
928 }
929}
930EXPORT_SYMBOL(ceph_osdc_set_request_linger);
931
f24e9980
SW		 932/*
933 * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
934 * (as needed), and set the request r_osd appropriately. If there is
25985edc 935 * no up osd, set r_osd to NULL. Move the request to the appropriate list
6f6c7006 936 * (unsent, homeless) or leave on in-flight lru.
f24e9980
SW		 937 *
938 * Return 0 if unchanged, 1 if changed, or negative on error.
939 *
940 * Caller should hold map_sem for read and request_mutex.
941 */
6f6c7006 942static int __map_request(struct ceph_osd_client *osdc,
38d6453c 943 struct ceph_osd_request *req, int force_resend)
f24e9980 944{
5b191d99 945 struct ceph_pg pgid;
d85b7056
SW		 946 int acting[CEPH_PG_MAX_SIZE];
947 int o = -1, num = 0;
f24e9980 948 int err;
f24e9980 949
6f6c7006 950 dout("map_request %p tid %lld\n", req, req->r_tid);
2169aea6 951 err = ceph_calc_object_layout(&pgid, req->r_oid,
f24e9980 952 &req->r_file_layout, osdc->osdmap);
6f6c7006
SW		 953 if (err) {
954 list_move(&req->r_req_lru_item, &osdc->req_notarget);
f24e9980 955 return err;
6f6c7006 956 }
7740a42f
SW		 957 req->r_pgid = pgid;
958
d85b7056
SW		 959 err = ceph_calc_pg_acting(osdc->osdmap, pgid, acting);
960 if (err > 0) {
961 o = acting[0];
962 num = err;
963 }
f24e9980 964
38d6453c
SW		 965 if ((!force_resend &&
966 req->r_osd && req->r_osd->o_osd == o &&
d85b7056
SW		 967 req->r_sent >= req->r_osd->o_incarnation &&
968 req->r_num_pg_osds == num &&
969 memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
f24e9980
SW		 970 (req->r_osd == NULL && o == -1))
971 return 0; /* no change */
972
5b191d99
SW		 973 dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n",
974 req->r_tid, pgid.pool, pgid.seed, o,
f24e9980
SW		 975 req->r_osd ? req->r_osd->o_osd : -1);
976
d85b7056
SW		 977 /* record full pg acting set */
978 memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
979 req->r_num_pg_osds = num;
980
f24e9980
SW		 981 if (req->r_osd) {
982 __cancel_request(req);
983 list_del_init(&req->r_osd_item);
f24e9980
SW		 984 req->r_osd = NULL;
985 }
986
987 req->r_osd = __lookup_osd(osdc, o);
988 if (!req->r_osd && o >= 0) {
c99eb1c7 989 err = -ENOMEM;
e10006f8 990 req->r_osd = create_osd(osdc, o);
6f6c7006
SW		 991 if (!req->r_osd) {
992 list_move(&req->r_req_lru_item, &osdc->req_notarget);
c99eb1c7 993 goto out;
6f6c7006 994 }
f24e9980 995
6f6c7006 996 dout("map_request osd %p is osd%d\n", req->r_osd, o);
f24e9980
SW		 997 __insert_osd(osdc, req->r_osd);
998
b7a9e5dd
SW		 999 ceph_con_open(&req->r_osd->o_con,
1000 CEPH_ENTITY_TYPE_OSD, o,
1001 &osdc->osdmap->osd_addr[o]);
f24e9980
SW		 1002 }
1003
f5a2041b
YS		 1004 if (req->r_osd) {
1005 __remove_osd_from_lru(req->r_osd);
f24e9980 1006 list_add(&req->r_osd_item, &req->r_osd->o_requests);
6f6c7006
SW		 1007 list_move(&req->r_req_lru_item, &osdc->req_unsent);
1008 } else {
1009 list_move(&req->r_req_lru_item, &osdc->req_notarget);
f5a2041b 1010 }
d85b7056 1011 err = 1; /* osd or pg changed */
f24e9980
SW		 1012
1013out:
f24e9980
SW		 1014 return err;
1015}
1016
1017/*
1018 * caller should hold map_sem (for read) and request_mutex
1019 */
56e925b6
SW		 1020static void __send_request(struct ceph_osd_client *osdc,
1021 struct ceph_osd_request *req)
f24e9980 1022{
1b83bef2 1023 void *p;
f24e9980 1024
1b83bef2
SW		 1025 dout("send_request %p tid %llu to osd%d flags %d pg %lld.%x\n",
1026 req, req->r_tid, req->r_osd->o_osd, req->r_flags,
1027 (unsigned long long)req->r_pgid.pool, req->r_pgid.seed);
1028
1029 /* fill in message content that changes each time we send it */
1030 put_unaligned_le32(osdc->osdmap->epoch, req->r_request_osdmap_epoch);
1031 put_unaligned_le32(req->r_flags, req->r_request_flags);
1032 put_unaligned_le64(req->r_pgid.pool, req->r_request_pool);
1033 p = req->r_request_pgid;
1034 ceph_encode_64(&p, req->r_pgid.pool);
1035 ceph_encode_32(&p, req->r_pgid.seed);
1036 put_unaligned_le64(1, req->r_request_attempts); /* FIXME */
1037 memcpy(req->r_request_reassert_version, &req->r_reassert_version,
1038 sizeof(req->r_reassert_version));
2169aea6 1039
3dd72fc0 1040 req->r_stamp = jiffies;
07a27e22 1041 list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
f24e9980
SW		 1042
1043 ceph_msg_get(req->r_request); /* send consumes a ref */
1044 ceph_con_send(&req->r_osd->o_con, req->r_request);
1045 req->r_sent = req->r_osd->o_incarnation;
f24e9980
SW		 1046}
1047
6f6c7006
SW		 1048/*
1049 * Send any requests in the queue (req_unsent).
1050 */
f9d25199 1051static void __send_queued(struct ceph_osd_client *osdc)
6f6c7006
SW		 1052{
1053 struct ceph_osd_request *req, *tmp;
1054
f9d25199
AE		 1055 dout("__send_queued\n");
1056 list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item)
6f6c7006 1057 __send_request(osdc, req);
6f6c7006
SW		 1058}
1059
f24e9980
SW		 1060/*
1061 * Timeout callback, called every N seconds when 1 or more osd
1062 * requests has been active for more than N seconds. When this
1063 * happens, we ping all OSDs with requests who have timed out to
1064 * ensure any communications channel reset is detected. Reset the
1065 * request timeouts another N seconds in the future as we go.
1066 * Reschedule the timeout event another N seconds in future (unless
1067 * there are no open requests).
1068 */
1069static void handle_timeout(struct work_struct *work)
1070{
1071 struct ceph_osd_client *osdc =
1072 container_of(work, struct ceph_osd_client, timeout_work.work);
83aff95e 1073 struct ceph_osd_request *req;
f24e9980 1074 struct ceph_osd *osd;
422d2cb8 1075 unsigned long keepalive =
3d14c5d2 1076 osdc->client->options->osd_keepalive_timeout * HZ;
422d2cb8 1077 struct list_head slow_osds;
f24e9980
SW		 1078 dout("timeout\n");
1079 down_read(&osdc->map_sem);
1080
1081 ceph_monc_request_next_osdmap(&osdc->client->monc);
1082
1083 mutex_lock(&osdc->request_mutex);
f24e9980 1084
422d2cb8
YS		 1085 /*
1086 * ping osds that are a bit slow. this ensures that if there
1087 * is a break in the TCP connection we will notice, and reopen
1088 * a connection with that osd (from the fault callback).
1089 */
1090 INIT_LIST_HEAD(&slow_osds);
1091 list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
3dd72fc0 1092 if (time_before(jiffies, req->r_stamp + keepalive))
422d2cb8
YS		 1093 break;
1094
1095 osd = req->r_osd;
1096 BUG_ON(!osd);
1097 dout(" tid %llu is slow, will send keepalive on osd%d\n",
f24e9980 1098 req->r_tid, osd->o_osd);
422d2cb8
YS		 1099 list_move_tail(&osd->o_keepalive_item, &slow_osds);
1100 }
1101 while (!list_empty(&slow_osds)) {
1102 osd = list_entry(slow_osds.next, struct ceph_osd,
1103 o_keepalive_item);
1104 list_del_init(&osd->o_keepalive_item);
f24e9980
SW		 1105 ceph_con_keepalive(&osd->o_con);
1106 }
1107
422d2cb8 1108 __schedule_osd_timeout(osdc);
f9d25199 1109 __send_queued(osdc);
f24e9980 1110 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 1111 up_read(&osdc->map_sem);
1112}
1113
f5a2041b
YS		 1114static void handle_osds_timeout(struct work_struct *work)
1115{
1116 struct ceph_osd_client *osdc =
1117 container_of(work, struct ceph_osd_client,
1118 osds_timeout_work.work);
1119 unsigned long delay =
3d14c5d2 1120 osdc->client->options->osd_idle_ttl * HZ >> 2;
f5a2041b
YS		 1121
1122 dout("osds timeout\n");
1123 down_read(&osdc->map_sem);
aca420bc 1124 remove_old_osds(osdc);
f5a2041b
YS		 1125 up_read(&osdc->map_sem);
1126
1127 schedule_delayed_work(&osdc->osds_timeout_work,
1128 round_jiffies_relative(delay));
1129}
1130
25845472
SW		 1131static void complete_request(struct ceph_osd_request *req)
1132{
1133 if (req->r_safe_callback)
1134 req->r_safe_callback(req, NULL);
1135 complete_all(&req->r_safe_completion); /* fsync waiter */
1136}
1137
1b83bef2
SW		 1138static int __decode_pgid(void **p, void *end, struct ceph_pg *pgid)
1139{
1140 __u8 v;
1141
1142 ceph_decode_need(p, end, 1 + 8 + 4 + 4, bad);
1143 v = ceph_decode_8(p);
1144 if (v > 1) {
1145 pr_warning("do not understand pg encoding %d > 1", v);
1146 return -EINVAL;
1147 }
1148 pgid->pool = ceph_decode_64(p);
1149 pgid->seed = ceph_decode_32(p);
1150 *p += 4;
1151 return 0;
1152
1153bad:
1154 pr_warning("incomplete pg encoding");
1155 return -EINVAL;
1156}
1157
f24e9980
SW		 1158/*
1159 * handle osd op reply. either call the callback if it is specified,
1160 * or do the completion to wake up the waiting thread.
1161 */
350b1c32
SW		 1162static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg,
1163 struct ceph_connection *con)
f24e9980 1164{
1b83bef2 1165 void *p, *end;
f24e9980
SW		 1166 struct ceph_osd_request *req;
1167 u64 tid;
1b83bef2
SW		 1168 int object_len;
1169 int numops, payload_len, flags;
0ceed5db 1170 s32 result;
1b83bef2
SW		 1171 s32 retry_attempt;
1172 struct ceph_pg pg;
1173 int err;
1174 u32 reassert_epoch;
1175 u64 reassert_version;
1176 u32 osdmap_epoch;
0d5af164 1177 int already_completed;
1b83bef2 1178 int i;
f24e9980 1179
6df058c0 1180 tid = le64_to_cpu(msg->hdr.tid);
1b83bef2
SW		 1181 dout("handle_reply %p tid %llu\n", msg, tid);
1182
1183 p = msg->front.iov_base;
1184 end = p + msg->front.iov_len;
1185
1186 ceph_decode_need(&p, end, 4, bad);
1187 object_len = ceph_decode_32(&p);
1188 ceph_decode_need(&p, end, object_len, bad);
1189 p += object_len;
1190
1191 err = __decode_pgid(&p, end, &pg);
1192 if (err)
f24e9980 1193 goto bad;
1b83bef2
SW		 1194
1195 ceph_decode_need(&p, end, 8 + 4 + 4 + 8 + 4, bad);
1196 flags = ceph_decode_64(&p);
1197 result = ceph_decode_32(&p);
1198 reassert_epoch = ceph_decode_32(&p);
1199 reassert_version = ceph_decode_64(&p);
1200 osdmap_epoch = ceph_decode_32(&p);
1201
f24e9980
SW		 1202 /* lookup */
1203 mutex_lock(&osdc->request_mutex);
1204 req = __lookup_request(osdc, tid);
1205 if (req == NULL) {
1206 dout("handle_reply tid %llu dne\n", tid);
1207 mutex_unlock(&osdc->request_mutex);
1208 return;
1209 }
1210 ceph_osdc_get_request(req);
1b83bef2
SW		 1211
1212 dout("handle_reply %p tid %llu req %p result %d\n", msg, tid,
1213 req, result);
1214
1215 ceph_decode_need(&p, end, 4, bad);
1216 numops = ceph_decode_32(&p);
1217 if (numops > CEPH_OSD_MAX_OP)
1218 goto bad_put;
1219 if (numops != req->r_num_ops)
1220 goto bad_put;
1221 payload_len = 0;
1222 ceph_decode_need(&p, end, numops * sizeof(struct ceph_osd_op), bad);
1223 for (i = 0; i < numops; i++) {
1224 struct ceph_osd_op *op = p;
1225 int len;
1226
1227 len = le32_to_cpu(op->payload_len);
1228 req->r_reply_op_len[i] = len;
1229 dout(" op %d has %d bytes\n", i, len);
1230 payload_len += len;
1231 p += sizeof(*op);
1232 }
1233 if (payload_len != le32_to_cpu(msg->hdr.data_len)) {
1234 pr_warning("sum of op payload lens %d != data_len %d",
1235 payload_len, le32_to_cpu(msg->hdr.data_len));
1236 goto bad_put;
1237 }
1238
1239 ceph_decode_need(&p, end, 4 + numops * 4, bad);
1240 retry_attempt = ceph_decode_32(&p);
1241 for (i = 0; i < numops; i++)
1242 req->r_reply_op_result[i] = ceph_decode_32(&p);
f24e9980 1243
350b1c32 1244 /*
0d59ab81 1245 * if this connection filled our message, drop our reference now, to
350b1c32
SW		 1246 * avoid a (safe but slower) revoke later.
1247 */
0d59ab81 1248 if (req->r_con_filling_msg == con && req->r_reply == msg) {
c16e7869 1249 dout(" dropping con_filling_msg ref %p\n", con);
0d59ab81 1250 req->r_con_filling_msg = NULL;
0d47766f 1251 con->ops->put(con);
350b1c32
SW		 1252 }
1253
f24e9980 1254 if (!req->r_got_reply) {
95c96174 1255 unsigned int bytes;
f24e9980 1256
1b83bef2 1257 req->r_result = result;
f24e9980
SW		 1258 bytes = le32_to_cpu(msg->hdr.data_len);
1259 dout("handle_reply result %d bytes %d\n", req->r_result,
1260 bytes);
1261 if (req->r_result == 0)
1262 req->r_result = bytes;
1263
1264 /* in case this is a write and we need to replay, */
1b83bef2
SW		 1265 req->r_reassert_version.epoch = cpu_to_le32(reassert_epoch);
1266 req->r_reassert_version.version = cpu_to_le64(reassert_version);
f24e9980
SW		 1267
1268 req->r_got_reply = 1;
1269 } else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) {
1270 dout("handle_reply tid %llu dup ack\n", tid);
34b43a56 1271 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 1272 goto done;
1273 }
1274
1275 dout("handle_reply tid %llu flags %d\n", tid, flags);
1276
a40c4f10
YS		 1277 if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK))
1278 __register_linger_request(osdc, req);
1279
f24e9980 1280 /* either this is a read, or we got the safe response */
0ceed5db
SW		 1281 if (result < 0 ||
1282 (flags & CEPH_OSD_FLAG_ONDISK) ||
f24e9980
SW		 1283 ((flags & CEPH_OSD_FLAG_WRITE) == 0))
1284 __unregister_request(osdc, req);
1285
0d5af164
AE		 1286 already_completed = req->r_completed;
1287 req->r_completed = 1;
f24e9980 1288 mutex_unlock(&osdc->request_mutex);
0d5af164
AE		 1289 if (already_completed)
1290 goto done;
f24e9980
SW		 1291
1292 if (req->r_callback)
1293 req->r_callback(req, msg);
1294 else
03066f23 1295 complete_all(&req->r_completion);
f24e9980 1296
25845472
SW		 1297 if (flags & CEPH_OSD_FLAG_ONDISK)
1298 complete_request(req);
f24e9980
SW		 1299
1300done:
a40c4f10 1301 dout("req=%p req->r_linger=%d\n", req, req->r_linger);
f24e9980
SW		 1302 ceph_osdc_put_request(req);
1303 return;
1304
1b83bef2
SW		 1305bad_put:
1306 ceph_osdc_put_request(req);
f24e9980 1307bad:
1b83bef2
SW		 1308 pr_err("corrupt osd_op_reply got %d %d\n",
1309 (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len));
9ec7cab1 1310 ceph_msg_dump(msg);
f24e9980
SW		 1311}
1312
6f6c7006 1313static void reset_changed_osds(struct ceph_osd_client *osdc)
f24e9980 1314{
f24e9980 1315 struct rb_node *p, *n;
f24e9980 1316
6f6c7006
SW		 1317 for (p = rb_first(&osdc->osds); p; p = n) {
1318 struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
f24e9980 1319
6f6c7006
SW		 1320 n = rb_next(p);
1321 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
1322 memcmp(&osd->o_con.peer_addr,
1323 ceph_osd_addr(osdc->osdmap,
1324 osd->o_osd),
1325 sizeof(struct ceph_entity_addr)) != 0)
1326 __reset_osd(osdc, osd);
f24e9980 1327 }
422d2cb8
YS		 1328}
1329
1330/*
6f6c7006
SW		 1331 * Requeue requests whose mapping to an OSD has changed. If requests map to
1332 * no osd, request a new map.
422d2cb8 1333 *
e6d50f67 1334 * Caller should hold map_sem for read.
422d2cb8 1335 */
38d6453c 1336static void kick_requests(struct ceph_osd_client *osdc, int force_resend)
422d2cb8 1337{
a40c4f10 1338 struct ceph_osd_request *req, *nreq;
6f6c7006
SW		 1339 struct rb_node *p;
1340 int needmap = 0;
1341 int err;
422d2cb8 1342
38d6453c 1343 dout("kick_requests %s\n", force_resend ? " (force resend)" : "");
422d2cb8 1344 mutex_lock(&osdc->request_mutex);
6194ea89 1345 for (p = rb_first(&osdc->requests); p; ) {
6f6c7006 1346 req = rb_entry(p, struct ceph_osd_request, r_node);
6194ea89 1347 p = rb_next(p);
ab60b16d
AE		 1348
1349 /*
1350 * For linger requests that have not yet been
1351 * registered, move them to the linger list; they'll
1352 * be sent to the osd in the loop below. Unregister
1353 * the request before re-registering it as a linger
1354 * request to ensure the __map_request() below
1355 * will decide it needs to be sent.
1356 */
1357 if (req->r_linger && list_empty(&req->r_linger_item)) {
1358 dout("%p tid %llu restart on osd%d\n",
1359 req, req->r_tid,
1360 req->r_osd ? req->r_osd->o_osd : -1);
1361 __unregister_request(osdc, req);
1362 __register_linger_request(osdc, req);
1363 continue;
1364 }
1365
38d6453c 1366 err = __map_request(osdc, req, force_resend);
6f6c7006
SW		 1367 if (err < 0)
1368 continue; /* error */
1369 if (req->r_osd == NULL) {
1370 dout("%p tid %llu maps to no osd\n", req, req->r_tid);
1371 needmap++; /* request a newer map */
1372 } else if (err > 0) {
6194ea89
SW		 1373 if (!req->r_linger) {
1374 dout("%p tid %llu requeued on osd%d\n", req,
1375 req->r_tid,
1376 req->r_osd ? req->r_osd->o_osd : -1);
a40c4f10 1377 req->r_flags |= CEPH_OSD_FLAG_RETRY;
6194ea89
SW		 1378 }
1379 }
a40c4f10
YS		 1380 }
1381
1382 list_for_each_entry_safe(req, nreq, &osdc->req_linger,
1383 r_linger_item) {
1384 dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
1385
38d6453c 1386 err = __map_request(osdc, req, force_resend);
ab60b16d 1387 dout("__map_request returned %d\n", err);
a40c4f10
YS		 1388 if (err == 0)
1389 continue; /* no change and no osd was specified */
1390 if (err < 0)
1391 continue; /* hrm! */
1392 if (req->r_osd == NULL) {
1393 dout("tid %llu maps to no valid osd\n", req->r_tid);
1394 needmap++; /* request a newer map */
1395 continue;
6f6c7006 1396 }
a40c4f10
YS		 1397
1398 dout("kicking lingering %p tid %llu osd%d\n", req, req->r_tid,
1399 req->r_osd ? req->r_osd->o_osd : -1);
a40c4f10 1400 __register_request(osdc, req);
c89ce05e 1401 __unregister_linger_request(osdc, req);
6f6c7006 1402 }
f24e9980
SW		 1403 mutex_unlock(&osdc->request_mutex);
1404
1405 if (needmap) {
1406 dout("%d requests for down osds, need new map\n", needmap);
1407 ceph_monc_request_next_osdmap(&osdc->client->monc);
1408 }
e6d50f67 1409 reset_changed_osds(osdc);
422d2cb8 1410}
6f6c7006
SW		 1411
1412
f24e9980
SW		 1413/*
1414 * Process updated osd map.
1415 *
1416 * The message contains any number of incremental and full maps, normally
1417 * indicating some sort of topology change in the cluster. Kick requests
1418 * off to different OSDs as needed.
1419 */
1420void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1421{
1422 void *p, *end, *next;
1423 u32 nr_maps, maplen;
1424 u32 epoch;
1425 struct ceph_osdmap *newmap = NULL, *oldmap;
1426 int err;
1427 struct ceph_fsid fsid;
1428
1429 dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
1430 p = msg->front.iov_base;
1431 end = p + msg->front.iov_len;
1432
1433 /* verify fsid */
1434 ceph_decode_need(&p, end, sizeof(fsid), bad);
1435 ceph_decode_copy(&p, &fsid, sizeof(fsid));
0743304d
SW		 1436 if (ceph_check_fsid(osdc->client, &fsid) < 0)
1437 return;
f24e9980
SW		 1438
1439 down_write(&osdc->map_sem);
1440
1441 /* incremental maps */
1442 ceph_decode_32_safe(&p, end, nr_maps, bad);
1443 dout(" %d inc maps\n", nr_maps);
1444 while (nr_maps > 0) {
1445 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 1446 epoch = ceph_decode_32(&p);
1447 maplen = ceph_decode_32(&p);
f24e9980
SW		 1448 ceph_decode_need(&p, end, maplen, bad);
1449 next = p + maplen;
1450 if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
1451 dout("applying incremental map %u len %d\n",
1452 epoch, maplen);
1453 newmap = osdmap_apply_incremental(&p, next,
1454 osdc->osdmap,
15d9882c 1455 &osdc->client->msgr);
f24e9980
SW		 1456 if (IS_ERR(newmap)) {
1457 err = PTR_ERR(newmap);
1458 goto bad;
1459 }
30dc6381 1460 BUG_ON(!newmap);
f24e9980
SW		 1461 if (newmap != osdc->osdmap) {
1462 ceph_osdmap_destroy(osdc->osdmap);
1463 osdc->osdmap = newmap;
1464 }
38d6453c 1465 kick_requests(osdc, 0);
f24e9980
SW		 1466 } else {
1467 dout("ignoring incremental map %u len %d\n",
1468 epoch, maplen);
1469 }
1470 p = next;
1471 nr_maps--;
1472 }
1473 if (newmap)
1474 goto done;
1475
1476 /* full maps */
1477 ceph_decode_32_safe(&p, end, nr_maps, bad);
1478 dout(" %d full maps\n", nr_maps);
1479 while (nr_maps) {
1480 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 1481 epoch = ceph_decode_32(&p);
1482 maplen = ceph_decode_32(&p);
f24e9980
SW		 1483 ceph_decode_need(&p, end, maplen, bad);
1484 if (nr_maps > 1) {
1485 dout("skipping non-latest full map %u len %d\n",
1486 epoch, maplen);
1487 } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
1488 dout("skipping full map %u len %d, "
1489 "older than our %u\n", epoch, maplen,
1490 osdc->osdmap->epoch);
1491 } else {
38d6453c
SW		 1492 int skipped_map = 0;
1493
f24e9980
SW		 1494 dout("taking full map %u len %d\n", epoch, maplen);
1495 newmap = osdmap_decode(&p, p+maplen);
1496 if (IS_ERR(newmap)) {
1497 err = PTR_ERR(newmap);
1498 goto bad;
1499 }
30dc6381 1500 BUG_ON(!newmap);
f24e9980
SW		 1501 oldmap = osdc->osdmap;
1502 osdc->osdmap = newmap;
38d6453c
SW		 1503 if (oldmap) {
1504 if (oldmap->epoch + 1 < newmap->epoch)
1505 skipped_map = 1;
f24e9980 1506 ceph_osdmap_destroy(oldmap);
38d6453c
SW		 1507 }
1508 kick_requests(osdc, skipped_map);
f24e9980
SW		 1509 }
1510 p += maplen;
1511 nr_maps--;
1512 }
1513
1514done:
1515 downgrade_write(&osdc->map_sem);
1516 ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch);
cd634fb6
SW		 1517
1518 /*
1519 * subscribe to subsequent osdmap updates if full to ensure
1520 * we find out when we are no longer full and stop returning
1521 * ENOSPC.
1522 */
1523 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL))
1524 ceph_monc_request_next_osdmap(&osdc->client->monc);
1525
f9d25199
AE		 1526 mutex_lock(&osdc->request_mutex);
1527 __send_queued(osdc);
1528 mutex_unlock(&osdc->request_mutex);
f24e9980 1529 up_read(&osdc->map_sem);
03066f23 1530 wake_up_all(&osdc->client->auth_wq);
f24e9980
SW		 1531 return;
1532
1533bad:
1534 pr_err("osdc handle_map corrupt msg\n");
9ec7cab1 1535 ceph_msg_dump(msg);
f24e9980
SW		 1536 up_write(&osdc->map_sem);
1537 return;
1538}
1539
a40c4f10
YS		 1540/*
1541 * watch/notify callback event infrastructure
1542 *
1543 * These callbacks are used both for watch and notify operations.
1544 */
1545static void __release_event(struct kref *kref)
1546{
1547 struct ceph_osd_event *event =
1548 container_of(kref, struct ceph_osd_event, kref);
1549
1550 dout("__release_event %p\n", event);
1551 kfree(event);
1552}
1553
1554static void get_event(struct ceph_osd_event *event)
1555{
1556 kref_get(&event->kref);
1557}
1558
1559void ceph_osdc_put_event(struct ceph_osd_event *event)
1560{
1561 kref_put(&event->kref, __release_event);
1562}
1563EXPORT_SYMBOL(ceph_osdc_put_event);
1564
1565static void __insert_event(struct ceph_osd_client *osdc,
1566 struct ceph_osd_event *new)
1567{
1568 struct rb_node **p = &osdc->event_tree.rb_node;
1569 struct rb_node *parent = NULL;
1570 struct ceph_osd_event *event = NULL;
1571
1572 while (*p) {
1573 parent = *p;
1574 event = rb_entry(parent, struct ceph_osd_event, node);
1575 if (new->cookie < event->cookie)
1576 p = &(*p)->rb_left;
1577 else if (new->cookie > event->cookie)
1578 p = &(*p)->rb_right;
1579 else
1580 BUG();
1581 }
1582
1583 rb_link_node(&new->node, parent, p);
1584 rb_insert_color(&new->node, &osdc->event_tree);
1585}
1586
1587static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
1588 u64 cookie)
1589{
1590 struct rb_node **p = &osdc->event_tree.rb_node;
1591 struct rb_node *parent = NULL;
1592 struct ceph_osd_event *event = NULL;
1593
1594 while (*p) {
1595 parent = *p;
1596 event = rb_entry(parent, struct ceph_osd_event, node);
1597 if (cookie < event->cookie)
1598 p = &(*p)->rb_left;
1599 else if (cookie > event->cookie)
1600 p = &(*p)->rb_right;
1601 else
1602 return event;
1603 }
1604 return NULL;
1605}
1606
1607static void __remove_event(struct ceph_osd_event *event)
1608{
1609 struct ceph_osd_client *osdc = event->osdc;
1610
1611 if (!RB_EMPTY_NODE(&event->node)) {
1612 dout("__remove_event removed %p\n", event);
1613 rb_erase(&event->node, &osdc->event_tree);
1614 ceph_osdc_put_event(event);
1615 } else {
1616 dout("__remove_event didn't remove %p\n", event);
1617 }
1618}
1619
1620int ceph_osdc_create_event(struct ceph_osd_client *osdc,
1621 void (*event_cb)(u64, u64, u8, void *),
3c663bbd 1622 void *data, struct ceph_osd_event **pevent)
a40c4f10
YS		 1623{
1624 struct ceph_osd_event *event;
1625
1626 event = kmalloc(sizeof(*event), GFP_NOIO);
1627 if (!event)
1628 return -ENOMEM;
1629
1630 dout("create_event %p\n", event);
1631 event->cb = event_cb;
3c663bbd 1632 event->one_shot = 0;
a40c4f10
YS		 1633 event->data = data;
1634 event->osdc = osdc;
1635 INIT_LIST_HEAD(&event->osd_node);
3ee5234d 1636 RB_CLEAR_NODE(&event->node);
a40c4f10
YS		 1637 kref_init(&event->kref); /* one ref for us */
1638 kref_get(&event->kref); /* one ref for the caller */
a40c4f10
YS		 1639
1640 spin_lock(&osdc->event_lock);
1641 event->cookie = ++osdc->event_count;
1642 __insert_event(osdc, event);
1643 spin_unlock(&osdc->event_lock);
1644
1645 *pevent = event;
1646 return 0;
1647}
1648EXPORT_SYMBOL(ceph_osdc_create_event);
1649
1650void ceph_osdc_cancel_event(struct ceph_osd_event *event)
1651{
1652 struct ceph_osd_client *osdc = event->osdc;
1653
1654 dout("cancel_event %p\n", event);
1655 spin_lock(&osdc->event_lock);
1656 __remove_event(event);
1657 spin_unlock(&osdc->event_lock);
1658 ceph_osdc_put_event(event); /* caller's */
1659}
1660EXPORT_SYMBOL(ceph_osdc_cancel_event);
1661
1662
1663static void do_event_work(struct work_struct *work)
1664{
1665 struct ceph_osd_event_work *event_work =
1666 container_of(work, struct ceph_osd_event_work, work);
1667 struct ceph_osd_event *event = event_work->event;
1668 u64 ver = event_work->ver;
1669 u64 notify_id = event_work->notify_id;
1670 u8 opcode = event_work->opcode;
1671
1672 dout("do_event_work completing %p\n", event);
1673 event->cb(ver, notify_id, opcode, event->data);
a40c4f10
YS		 1674 dout("do_event_work completed %p\n", event);
1675 ceph_osdc_put_event(event);
1676 kfree(event_work);
1677}
1678
1679
1680/*
1681 * Process osd watch notifications
1682 */
3c663bbd
AE		 1683static void handle_watch_notify(struct ceph_osd_client *osdc,
1684 struct ceph_msg *msg)
a40c4f10
YS		 1685{
1686 void *p, *end;
1687 u8 proto_ver;
1688 u64 cookie, ver, notify_id;
1689 u8 opcode;
1690 struct ceph_osd_event *event;
1691 struct ceph_osd_event_work *event_work;
1692
1693 p = msg->front.iov_base;
1694 end = p + msg->front.iov_len;
1695
1696 ceph_decode_8_safe(&p, end, proto_ver, bad);
1697 ceph_decode_8_safe(&p, end, opcode, bad);
1698 ceph_decode_64_safe(&p, end, cookie, bad);
1699 ceph_decode_64_safe(&p, end, ver, bad);
1700 ceph_decode_64_safe(&p, end, notify_id, bad);
1701
1702 spin_lock(&osdc->event_lock);
1703 event = __find_event(osdc, cookie);
1704 if (event) {
3c663bbd 1705 BUG_ON(event->one_shot);
a40c4f10 1706 get_event(event);
a40c4f10
YS		 1707 }
1708 spin_unlock(&osdc->event_lock);
1709 dout("handle_watch_notify cookie %lld ver %lld event %p\n",
1710 cookie, ver, event);
1711 if (event) {
1712 event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
a40c4f10
YS		 1713 if (!event_work) {
1714 dout("ERROR: could not allocate event_work\n");
1715 goto done_err;
1716 }
6b0ae409 1717 INIT_WORK(&event_work->work, do_event_work);
a40c4f10
YS		 1718 event_work->event = event;
1719 event_work->ver = ver;
1720 event_work->notify_id = notify_id;
1721 event_work->opcode = opcode;
1722 if (!queue_work(osdc->notify_wq, &event_work->work)) {
1723 dout("WARNING: failed to queue notify event work\n");
1724 goto done_err;
1725 }
1726 }
1727
1728 return;
1729
1730done_err:
a40c4f10
YS		 1731 ceph_osdc_put_event(event);
1732 return;
1733
1734bad:
1735 pr_err("osdc handle_watch_notify corrupt msg\n");
1736 return;
1737}
1738
f24e9980
SW		 1739/*
1740 * Register request, send initial attempt.
1741 */
1742int ceph_osdc_start_request(struct ceph_osd_client *osdc,
1743 struct ceph_osd_request *req,
1744 bool nofail)
1745{
c1ea8823 1746 int rc = 0;
f24e9980
SW		 1747
1748 req->r_request->pages = req->r_pages;
d4b515fa 1749 req->r_request->page_count = req->r_num_pages;
f51a822c 1750 req->r_request->page_alignment = req->r_page_alignment;
68b4476b
YS		 1751#ifdef CONFIG_BLOCK
1752 req->r_request->bio = req->r_bio;
1753#endif
c885837f 1754 req->r_request->trail = &req->r_trail;
f24e9980
SW		 1755
1756 register_request(osdc, req);
1757
1758 down_read(&osdc->map_sem);
1759 mutex_lock(&osdc->request_mutex);
c1ea8823
SW		 1760 /*
1761 * a racing kick_requests() may have sent the message for us
1762 * while we dropped request_mutex above, so only send now if
1763 * the request still han't been touched yet.
1764 */
1765 if (req->r_sent == 0) {
38d6453c 1766 rc = __map_request(osdc, req, 0);
9d6fcb08
SW		 1767 if (rc < 0) {
1768 if (nofail) {
1769 dout("osdc_start_request failed map, "
1770 " will retry %lld\n", req->r_tid);
1771 rc = 0;
1772 }
234af26f 1773 goto out_unlock;
9d6fcb08 1774 }
6f6c7006
SW		 1775 if (req->r_osd == NULL) {
1776 dout("send_request %p no up osds in pg\n", req);
1777 ceph_monc_request_next_osdmap(&osdc->client->monc);
1778 } else {
56e925b6 1779 __send_request(osdc, req);
f24e9980 1780 }
56e925b6 1781 rc = 0;
f24e9980 1782 }
234af26f
DC		 1783
1784out_unlock:
f24e9980
SW		 1785 mutex_unlock(&osdc->request_mutex);
1786 up_read(&osdc->map_sem);
1787 return rc;
1788}
3d14c5d2 1789EXPORT_SYMBOL(ceph_osdc_start_request);
f24e9980
SW		 1790
1791/*
1792 * wait for a request to complete
1793 */
1794int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
1795 struct ceph_osd_request *req)
1796{
1797 int rc;
1798
1799 rc = wait_for_completion_interruptible(&req->r_completion);
1800 if (rc < 0) {
1801 mutex_lock(&osdc->request_mutex);
1802 __cancel_request(req);
529cfcc4 1803 __unregister_request(osdc, req);
f24e9980 1804 mutex_unlock(&osdc->request_mutex);
25845472 1805 complete_request(req);
529cfcc4 1806 dout("wait_request tid %llu canceled/timed out\n", req->r_tid);
f24e9980
SW		 1807 return rc;
1808 }
1809
1810 dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result);
1811 return req->r_result;
1812}
3d14c5d2 1813EXPORT_SYMBOL(ceph_osdc_wait_request);
f24e9980
SW		 1814
1815/*
1816 * sync - wait for all in-flight requests to flush. avoid starvation.
1817 */
1818void ceph_osdc_sync(struct ceph_osd_client *osdc)
1819{
1820 struct ceph_osd_request *req;
1821 u64 last_tid, next_tid = 0;
1822
1823 mutex_lock(&osdc->request_mutex);
1824 last_tid = osdc->last_tid;
1825 while (1) {
1826 req = __lookup_request_ge(osdc, next_tid);
1827 if (!req)
1828 break;
1829 if (req->r_tid > last_tid)
1830 break;
1831
1832 next_tid = req->r_tid + 1;
1833 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
1834 continue;
1835
1836 ceph_osdc_get_request(req);
1837 mutex_unlock(&osdc->request_mutex);
1838 dout("sync waiting on tid %llu (last is %llu)\n",
1839 req->r_tid, last_tid);
1840 wait_for_completion(&req->r_safe_completion);
1841 mutex_lock(&osdc->request_mutex);
1842 ceph_osdc_put_request(req);
1843 }
1844 mutex_unlock(&osdc->request_mutex);
1845 dout("sync done (thru tid %llu)\n", last_tid);
1846}
3d14c5d2 1847EXPORT_SYMBOL(ceph_osdc_sync);
f24e9980
SW		 1848
1849/*
1850 * init, shutdown
1851 */
1852int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
1853{
1854 int err;
1855
1856 dout("init\n");
1857 osdc->client = client;
1858 osdc->osdmap = NULL;
1859 init_rwsem(&osdc->map_sem);
1860 init_completion(&osdc->map_waiters);
1861 osdc->last_requested_map = 0;
1862 mutex_init(&osdc->request_mutex);
f24e9980
SW		 1863 osdc->last_tid = 0;
1864 osdc->osds = RB_ROOT;
f5a2041b 1865 INIT_LIST_HEAD(&osdc->osd_lru);
f24e9980 1866 osdc->requests = RB_ROOT;
422d2cb8 1867 INIT_LIST_HEAD(&osdc->req_lru);
6f6c7006
SW		 1868 INIT_LIST_HEAD(&osdc->req_unsent);
1869 INIT_LIST_HEAD(&osdc->req_notarget);
a40c4f10 1870 INIT_LIST_HEAD(&osdc->req_linger);
f24e9980
SW		 1871 osdc->num_requests = 0;
1872 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
f5a2041b 1873 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
a40c4f10
YS		 1874 spin_lock_init(&osdc->event_lock);
1875 osdc->event_tree = RB_ROOT;
1876 osdc->event_count = 0;
f5a2041b
YS		 1877
1878 schedule_delayed_work(&osdc->osds_timeout_work,
3d14c5d2 1879 round_jiffies_relative(osdc->client->options->osd_idle_ttl * HZ));
f24e9980 1880
5f44f142 1881 err = -ENOMEM;
f24e9980
SW		 1882 osdc->req_mempool = mempool_create_kmalloc_pool(10,
1883 sizeof(struct ceph_osd_request));
1884 if (!osdc->req_mempool)
5f44f142 1885 goto out;
f24e9980 1886
d50b409f
SW		 1887 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
1888 OSD_OP_FRONT_LEN, 10, true,
4f48280e 1889 "osd_op");
f24e9980 1890 if (err < 0)
5f44f142 1891 goto out_mempool;
d50b409f 1892 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
4f48280e
SW		 1893 OSD_OPREPLY_FRONT_LEN, 10, true,
1894 "osd_op_reply");
c16e7869
SW		 1895 if (err < 0)
1896 goto out_msgpool;
a40c4f10
YS		 1897
1898 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
1899 if (IS_ERR(osdc->notify_wq)) {
1900 err = PTR_ERR(osdc->notify_wq);
1901 osdc->notify_wq = NULL;
1902 goto out_msgpool;
1903 }
f24e9980 1904 return 0;
5f44f142 1905
c16e7869
SW		 1906out_msgpool:
1907 ceph_msgpool_destroy(&osdc->msgpool_op);
5f44f142
SW		 1908out_mempool:
1909 mempool_destroy(osdc->req_mempool);
1910out:
1911 return err;
f24e9980
SW		 1912}
1913
1914void ceph_osdc_stop(struct ceph_osd_client *osdc)
1915{
a40c4f10
YS		 1916 flush_workqueue(osdc->notify_wq);
1917 destroy_workqueue(osdc->notify_wq);
f24e9980 1918 cancel_delayed_work_sync(&osdc->timeout_work);
f5a2041b 1919 cancel_delayed_work_sync(&osdc->osds_timeout_work);
f24e9980
SW		 1920 if (osdc->osdmap) {
1921 ceph_osdmap_destroy(osdc->osdmap);
1922 osdc->osdmap = NULL;
1923 }
aca420bc 1924 remove_all_osds(osdc);
f24e9980
SW		 1925 mempool_destroy(osdc->req_mempool);
1926 ceph_msgpool_destroy(&osdc->msgpool_op);
c16e7869 1927 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
f24e9980
SW		 1928}
1929
1930/*
1931 * Read some contiguous pages. If we cross a stripe boundary, shorten
1932 * *plen. Return number of bytes read, or error.
1933 */
1934int ceph_osdc_readpages(struct ceph_osd_client *osdc,
1935 struct ceph_vino vino, struct ceph_file_layout *layout,
1936 u64 off, u64 *plen,
1937 u32 truncate_seq, u64 truncate_size,
b7495fc2 1938 struct page **pages, int num_pages, int page_align)
f24e9980
SW		 1939{
1940 struct ceph_osd_request *req;
1941 int rc = 0;
1942
1943 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
1944 vino.snap, off, *plen);
1945 req = ceph_osdc_new_request(osdc, layout, vino, off, plen,
1946 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
1947 NULL, 0, truncate_seq, truncate_size, NULL,
a3bea47e 1948 false, page_align);
6816282d
SW		 1949 if (IS_ERR(req))
1950 return PTR_ERR(req);
f24e9980
SW		 1951
1952 /* it may be a short read due to an object boundary */
1953 req->r_pages = pages;
f24e9980 1954
b7495fc2
SW		 1955 dout("readpages final extent is %llu~%llu (%d pages align %d)\n",
1956 off, *plen, req->r_num_pages, page_align);
f24e9980
SW		 1957
1958 rc = ceph_osdc_start_request(osdc, req, false);
1959 if (!rc)
1960 rc = ceph_osdc_wait_request(osdc, req);
1961
1962 ceph_osdc_put_request(req);
1963 dout("readpages result %d\n", rc);
1964 return rc;
1965}
3d14c5d2 1966EXPORT_SYMBOL(ceph_osdc_readpages);
f24e9980
SW		 1967
1968/*
1969 * do a synchronous write on N pages
1970 */
1971int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
1972 struct ceph_file_layout *layout,
1973 struct ceph_snap_context *snapc,
1974 u64 off, u64 len,
1975 u32 truncate_seq, u64 truncate_size,
1976 struct timespec *mtime,
24808826 1977 struct page **pages, int num_pages)
f24e9980
SW		 1978{
1979 struct ceph_osd_request *req;
1980 int rc = 0;
b7495fc2 1981 int page_align = off & ~PAGE_MASK;
f24e9980
SW		 1982
1983 BUG_ON(vino.snap != CEPH_NOSNAP);
1984 req = ceph_osdc_new_request(osdc, layout, vino, off, &len,
1985 CEPH_OSD_OP_WRITE,
24808826 1986 CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
fbf8685f 1987 snapc, 0,
f24e9980 1988 truncate_seq, truncate_size, mtime,
a3bea47e 1989 true, page_align);
6816282d
SW		 1990 if (IS_ERR(req))
1991 return PTR_ERR(req);
f24e9980
SW		 1992
1993 /* it may be a short write due to an object boundary */
1994 req->r_pages = pages;
f24e9980
SW		 1995 dout("writepages %llu~%llu (%d pages)\n", off, len,
1996 req->r_num_pages);
1997
87f979d3 1998 rc = ceph_osdc_start_request(osdc, req, true);
f24e9980
SW		 1999 if (!rc)
2000 rc = ceph_osdc_wait_request(osdc, req);
2001
2002 ceph_osdc_put_request(req);
2003 if (rc == 0)
2004 rc = len;
2005 dout("writepages result %d\n", rc);
2006 return rc;
2007}
3d14c5d2 2008EXPORT_SYMBOL(ceph_osdc_writepages);
f24e9980
SW		 2009
2010/*
2011 * handle incoming message
2012 */
2013static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
2014{
2015 struct ceph_osd *osd = con->private;
32c895e7 2016 struct ceph_osd_client *osdc;
f24e9980
SW		 2017 int type = le16_to_cpu(msg->hdr.type);
2018
2019 if (!osd)
4a32f93d 2020 goto out;
32c895e7 2021 osdc = osd->o_osdc;
f24e9980
SW		 2022
2023 switch (type) {
2024 case CEPH_MSG_OSD_MAP:
2025 ceph_osdc_handle_map(osdc, msg);
2026 break;
2027 case CEPH_MSG_OSD_OPREPLY:
350b1c32 2028 handle_reply(osdc, msg, con);
f24e9980 2029 break;
a40c4f10
YS		 2030 case CEPH_MSG_WATCH_NOTIFY:
2031 handle_watch_notify(osdc, msg);
2032 break;
f24e9980
SW		 2033
2034 default:
2035 pr_err("received unknown message type %d %s\n", type,
2036 ceph_msg_type_name(type));
2037 }
4a32f93d 2038out:
f24e9980
SW		 2039 ceph_msg_put(msg);
2040}
2041
5b3a4db3 2042/*
21b667f6
SW		 2043 * lookup and return message for incoming reply. set up reply message
2044 * pages.
5b3a4db3
SW		 2045 */
2046static struct ceph_msg *get_reply(struct ceph_connection *con,
2450418c
YS		 2047 struct ceph_msg_header *hdr,
2048 int *skip)
f24e9980
SW		 2049{
2050 struct ceph_osd *osd = con->private;
2051 struct ceph_osd_client *osdc = osd->o_osdc;
2450418c 2052 struct ceph_msg *m;
0547a9b3 2053 struct ceph_osd_request *req;
5b3a4db3
SW		 2054 int front = le32_to_cpu(hdr->front_len);
2055 int data_len = le32_to_cpu(hdr->data_len);
0547a9b3 2056 u64 tid;
f24e9980 2057
0547a9b3
YS		 2058 tid = le64_to_cpu(hdr->tid);
2059 mutex_lock(&osdc->request_mutex);
2060 req = __lookup_request(osdc, tid);
2061 if (!req) {
2062 *skip = 1;
2063 m = NULL;
756a16a5
SW		 2064 dout("get_reply unknown tid %llu from osd%d\n", tid,
2065 osd->o_osd);
0547a9b3
YS		 2066 goto out;
2067 }
c16e7869
SW		 2068
2069 if (req->r_con_filling_msg) {
8921d114 2070 dout("%s revoking msg %p from old con %p\n", __func__,
c16e7869 2071 req->r_reply, req->r_con_filling_msg);
8921d114 2072 ceph_msg_revoke_incoming(req->r_reply);
0d47766f 2073 req->r_con_filling_msg->ops->put(req->r_con_filling_msg);
6f46cb29 2074 req->r_con_filling_msg = NULL;
0547a9b3
YS		 2075 }
2076
c16e7869
SW		 2077 if (front > req->r_reply->front.iov_len) {
2078 pr_warning("get_reply front %d > preallocated %d\n",
2079 front, (int)req->r_reply->front.iov_len);
b61c2763 2080 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front, GFP_NOFS, false);
a79832f2 2081 if (!m)
c16e7869
SW		 2082 goto out;
2083 ceph_msg_put(req->r_reply);
2084 req->r_reply = m;
2085 }
2086 m = ceph_msg_get(req->r_reply);
2087
0547a9b3 2088 if (data_len > 0) {
b7495fc2 2089 int want = calc_pages_for(req->r_page_alignment, data_len);
21b667f6 2090
9cbb1d72 2091 if (req->r_pages && unlikely(req->r_num_pages < want)) {
9bb0ce2b
SW		 2092 pr_warning("tid %lld reply has %d bytes %d pages, we"
2093 " had only %d pages ready\n", tid, data_len,
2094 want, req->r_num_pages);
0547a9b3
YS		 2095 *skip = 1;
2096 ceph_msg_put(m);
a79832f2 2097 m = NULL;
21b667f6 2098 goto out;
0547a9b3 2099 }
21b667f6 2100 m->pages = req->r_pages;
d4b515fa 2101 m->page_count = req->r_num_pages;
c5c6b19d 2102 m->page_alignment = req->r_page_alignment;
68b4476b
YS		 2103#ifdef CONFIG_BLOCK
2104 m->bio = req->r_bio;
2105#endif
0547a9b3 2106 }
5b3a4db3 2107 *skip = 0;
0d47766f 2108 req->r_con_filling_msg = con->ops->get(con);
c16e7869 2109 dout("get_reply tid %lld %p\n", tid, m);
0547a9b3
YS		 2110
2111out:
2112 mutex_unlock(&osdc->request_mutex);
2450418c 2113 return m;
5b3a4db3
SW		 2114
2115}
2116
2117static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2118 struct ceph_msg_header *hdr,
2119 int *skip)
2120{
2121 struct ceph_osd *osd = con->private;
2122 int type = le16_to_cpu(hdr->type);
2123 int front = le32_to_cpu(hdr->front_len);
2124
1c20f2d2 2125 *skip = 0;
5b3a4db3
SW		 2126 switch (type) {
2127 case CEPH_MSG_OSD_MAP:
a40c4f10 2128 case CEPH_MSG_WATCH_NOTIFY:
b61c2763 2129 return ceph_msg_new(type, front, GFP_NOFS, false);
5b3a4db3
SW		 2130 case CEPH_MSG_OSD_OPREPLY:
2131 return get_reply(con, hdr, skip);
2132 default:
2133 pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
2134 osd->o_osd);
2135 *skip = 1;
2136 return NULL;
2137 }
f24e9980
SW		 2138}
2139
2140/*
2141 * Wrappers to refcount containing ceph_osd struct
2142 */
2143static struct ceph_connection *get_osd_con(struct ceph_connection *con)
2144{
2145 struct ceph_osd *osd = con->private;
2146 if (get_osd(osd))
2147 return con;
2148 return NULL;
2149}
2150
2151static void put_osd_con(struct ceph_connection *con)
2152{
2153 struct ceph_osd *osd = con->private;
2154 put_osd(osd);
2155}
2156
4e7a5dcd
SW		 2157/*
2158 * authentication
2159 */
a3530df3
AE		 2160/*
2161 * Note: returned pointer is the address of a structure that's
2162 * managed separately. Caller must *not* attempt to free it.
2163 */
2164static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
8f43fb53 2165 int *proto, int force_new)
4e7a5dcd
SW		 2166{
2167 struct ceph_osd *o = con->private;
2168 struct ceph_osd_client *osdc = o->o_osdc;
2169 struct ceph_auth_client *ac = osdc->client->monc.auth;
74f1869f 2170 struct ceph_auth_handshake *auth = &o->o_auth;
4e7a5dcd 2171
74f1869f 2172 if (force_new && auth->authorizer) {
a255651d
AE		 2173 if (ac->ops && ac->ops->destroy_authorizer)
2174 ac->ops->destroy_authorizer(ac, auth->authorizer);
74f1869f
AE		 2175 auth->authorizer = NULL;
2176 }
a255651d 2177 if (!auth->authorizer && ac->ops && ac->ops->create_authorizer) {
a3530df3
AE		 2178 int ret = ac->ops->create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2179 auth);
4e7a5dcd 2180 if (ret)
a3530df3 2181 return ERR_PTR(ret);
4e7a5dcd 2182 }
4e7a5dcd 2183 *proto = ac->protocol;
74f1869f 2184
a3530df3 2185 return auth;
4e7a5dcd
SW		 2186}
2187
2188
2189static int verify_authorizer_reply(struct ceph_connection *con, int len)
2190{
2191 struct ceph_osd *o = con->private;
2192 struct ceph_osd_client *osdc = o->o_osdc;
2193 struct ceph_auth_client *ac = osdc->client->monc.auth;
2194
a255651d
AE		 2195 /*
2196 * XXX If ac->ops or ac->ops->verify_authorizer_reply is null,
2197 * XXX which do we do: succeed or fail?
2198 */
6c4a1915 2199 return ac->ops->verify_authorizer_reply(ac, o->o_auth.authorizer, len);
4e7a5dcd
SW		 2200}
2201
9bd2e6f8
SW		 2202static int invalidate_authorizer(struct ceph_connection *con)
2203{
2204 struct ceph_osd *o = con->private;
2205 struct ceph_osd_client *osdc = o->o_osdc;
2206 struct ceph_auth_client *ac = osdc->client->monc.auth;
2207
a255651d 2208 if (ac->ops && ac->ops->invalidate_authorizer)
9bd2e6f8
SW		 2209 ac->ops->invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
2210
2211 return ceph_monc_validate_auth(&osdc->client->monc);
2212}
4e7a5dcd 2213
9e32789f 2214static const struct ceph_connection_operations osd_con_ops = {
f24e9980
SW		 2215 .get = get_osd_con,
2216 .put = put_osd_con,
2217 .dispatch = dispatch,
4e7a5dcd
SW		 2218 .get_authorizer = get_authorizer,
2219 .verify_authorizer_reply = verify_authorizer_reply,
9bd2e6f8 2220 .invalidate_authorizer = invalidate_authorizer,
f24e9980 2221 .alloc_msg = alloc_msg,
81b024e7 2222 .fault = osd_reset,
f24e9980 2223};
Linux 6.x block layer and io_uring tree(s)