projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
ceph: update support for PGID64, PGPOOL3, OSDENC protocol features
[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 */
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,
ae7ca4a3 149 unsigned int num_op,
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;
68b4476b 155 size_t msg_size = sizeof(struct ceph_osd_request_head);
3499e8a5 156
68b4476b 157 msg_size += num_op*sizeof(struct ceph_osd_op);
f24e9980
SW		 158
159 if (use_mempool) {
3499e8a5 160 req = mempool_alloc(osdc->req_mempool, gfp_flags);
f24e9980
SW		 161 memset(req, 0, sizeof(*req));
162 } else {
3499e8a5 163 req = kzalloc(sizeof(*req), gfp_flags);
f24e9980
SW		 164 }
165 if (req == NULL)
a79832f2 166 return NULL;
f24e9980 167
f24e9980
SW		 168 req->r_osdc = osdc;
169 req->r_mempool = use_mempool;
68b4476b 170
415e49a9 171 kref_init(&req->r_kref);
f24e9980
SW		 172 init_completion(&req->r_completion);
173 init_completion(&req->r_safe_completion);
a978fa20 174 RB_CLEAR_NODE(&req->r_node);
f24e9980 175 INIT_LIST_HEAD(&req->r_unsafe_item);
a40c4f10
YS		 176 INIT_LIST_HEAD(&req->r_linger_item);
177 INIT_LIST_HEAD(&req->r_linger_osd);
935b639a 178 INIT_LIST_HEAD(&req->r_req_lru_item);
cd43045c
SW		 179 INIT_LIST_HEAD(&req->r_osd_item);
180
c16e7869
SW		 181 /* create reply message */
182 if (use_mempool)
183 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0);
184 else
185 msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY,
b61c2763 186 OSD_OPREPLY_FRONT_LEN, gfp_flags, true);
a79832f2 187 if (!msg) {
c16e7869 188 ceph_osdc_put_request(req);
a79832f2 189 return NULL;
c16e7869
SW		 190 }
191 req->r_reply = msg;
192
c885837f 193 ceph_pagelist_init(&req->r_trail);
d50b409f 194
c16e7869 195 /* create request message; allow space for oid */
224736d9 196 msg_size += MAX_OBJ_NAME_SIZE;
f24e9980
SW		 197 if (snapc)
198 msg_size += sizeof(u64) * snapc->num_snaps;
199 if (use_mempool)
8f3bc053 200 msg = ceph_msgpool_get(&osdc->msgpool_op, 0);
f24e9980 201 else
b61c2763 202 msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp_flags, true);
a79832f2 203 if (!msg) {
f24e9980 204 ceph_osdc_put_request(req);
a79832f2 205 return NULL;
f24e9980 206 }
68b4476b 207
f24e9980 208 memset(msg->front.iov_base, 0, msg->front.iov_len);
3499e8a5
YS		 209
210 req->r_request = msg;
3499e8a5
YS		 211
212 return req;
213}
3d14c5d2 214EXPORT_SYMBOL(ceph_osdc_alloc_request);
3499e8a5 215
68b4476b
YS		 216static void osd_req_encode_op(struct ceph_osd_request *req,
217 struct ceph_osd_op *dst,
218 struct ceph_osd_req_op *src)
219{
220 dst->op = cpu_to_le16(src->op);
221
065a68f9 222 switch (src->op) {
fbfab539
AE		 223 case CEPH_OSD_OP_STAT:
224 break;
68b4476b
YS		 225 case CEPH_OSD_OP_READ:
226 case CEPH_OSD_OP_WRITE:
227 dst->extent.offset =
228 cpu_to_le64(src->extent.offset);
229 dst->extent.length =
230 cpu_to_le64(src->extent.length);
231 dst->extent.truncate_size =
232 cpu_to_le64(src->extent.truncate_size);
233 dst->extent.truncate_seq =
234 cpu_to_le32(src->extent.truncate_seq);
235 break;
ae1533b6 236 case CEPH_OSD_OP_CALL:
ae1533b6
YS		 237 dst->cls.class_len = src->cls.class_len;
238 dst->cls.method_len = src->cls.method_len;
239 dst->cls.indata_len = cpu_to_le32(src->cls.indata_len);
240
c885837f 241 ceph_pagelist_append(&req->r_trail, src->cls.class_name,
ae1533b6 242 src->cls.class_len);
c885837f 243 ceph_pagelist_append(&req->r_trail, src->cls.method_name,
ae1533b6 244 src->cls.method_len);
c885837f 245 ceph_pagelist_append(&req->r_trail, src->cls.indata,
ae1533b6
YS		 246 src->cls.indata_len);
247 break;
68b4476b
YS		 248 case CEPH_OSD_OP_STARTSYNC:
249 break;
a40c4f10
YS		 250 case CEPH_OSD_OP_NOTIFY_ACK:
251 case CEPH_OSD_OP_WATCH:
252 dst->watch.cookie = cpu_to_le64(src->watch.cookie);
253 dst->watch.ver = cpu_to_le64(src->watch.ver);
254 dst->watch.flag = src->watch.flag;
255 break;
68b4476b
YS		 256 default:
257 pr_err("unrecognized osd opcode %d\n", dst->op);
258 WARN_ON(1);
259 break;
4c46459c
AE		 260 case CEPH_OSD_OP_MAPEXT:
261 case CEPH_OSD_OP_MASKTRUNC:
262 case CEPH_OSD_OP_SPARSE_READ:
a9f36c3e 263 case CEPH_OSD_OP_NOTIFY:
4c46459c
AE		 264 case CEPH_OSD_OP_ASSERT_VER:
265 case CEPH_OSD_OP_WRITEFULL:
266 case CEPH_OSD_OP_TRUNCATE:
267 case CEPH_OSD_OP_ZERO:
268 case CEPH_OSD_OP_DELETE:
269 case CEPH_OSD_OP_APPEND:
270 case CEPH_OSD_OP_SETTRUNC:
271 case CEPH_OSD_OP_TRIMTRUNC:
272 case CEPH_OSD_OP_TMAPUP:
273 case CEPH_OSD_OP_TMAPPUT:
274 case CEPH_OSD_OP_TMAPGET:
275 case CEPH_OSD_OP_CREATE:
a9f36c3e 276 case CEPH_OSD_OP_ROLLBACK:
4c46459c
AE		 277 case CEPH_OSD_OP_OMAPGETKEYS:
278 case CEPH_OSD_OP_OMAPGETVALS:
279 case CEPH_OSD_OP_OMAPGETHEADER:
280 case CEPH_OSD_OP_OMAPGETVALSBYKEYS:
281 case CEPH_OSD_OP_MODE_RD:
282 case CEPH_OSD_OP_OMAPSETVALS:
283 case CEPH_OSD_OP_OMAPSETHEADER:
284 case CEPH_OSD_OP_OMAPCLEAR:
285 case CEPH_OSD_OP_OMAPRMKEYS:
286 case CEPH_OSD_OP_OMAP_CMP:
287 case CEPH_OSD_OP_CLONERANGE:
288 case CEPH_OSD_OP_ASSERT_SRC_VERSION:
289 case CEPH_OSD_OP_SRC_CMPXATTR:
a9f36c3e 290 case CEPH_OSD_OP_GETXATTR:
4c46459c 291 case CEPH_OSD_OP_GETXATTRS:
a9f36c3e
AE		 292 case CEPH_OSD_OP_CMPXATTR:
293 case CEPH_OSD_OP_SETXATTR:
4c46459c
AE		 294 case CEPH_OSD_OP_SETXATTRS:
295 case CEPH_OSD_OP_RESETXATTRS:
296 case CEPH_OSD_OP_RMXATTR:
297 case CEPH_OSD_OP_PULL:
298 case CEPH_OSD_OP_PUSH:
299 case CEPH_OSD_OP_BALANCEREADS:
300 case CEPH_OSD_OP_UNBALANCEREADS:
301 case CEPH_OSD_OP_SCRUB:
302 case CEPH_OSD_OP_SCRUB_RESERVE:
303 case CEPH_OSD_OP_SCRUB_UNRESERVE:
304 case CEPH_OSD_OP_SCRUB_STOP:
305 case CEPH_OSD_OP_SCRUB_MAP:
306 case CEPH_OSD_OP_WRLOCK:
307 case CEPH_OSD_OP_WRUNLOCK:
308 case CEPH_OSD_OP_RDLOCK:
309 case CEPH_OSD_OP_RDUNLOCK:
310 case CEPH_OSD_OP_UPLOCK:
311 case CEPH_OSD_OP_DNLOCK:
312 case CEPH_OSD_OP_PGLS:
313 case CEPH_OSD_OP_PGLS_FILTER:
314 pr_err("unsupported osd opcode %s\n",
315 ceph_osd_op_name(dst->op));
316 WARN_ON(1);
317 break;
68b4476b
YS		 318 }
319 dst->payload_len = cpu_to_le32(src->payload_len);
320}
321
3499e8a5
YS		 322/*
323 * build new request AND message
324 *
325 */
326void ceph_osdc_build_request(struct ceph_osd_request *req,
ae7ca4a3 327 u64 off, u64 len, unsigned int num_op,
68b4476b 328 struct ceph_osd_req_op *src_ops,
4d6b250b 329 struct ceph_snap_context *snapc, u64 snap_id,
af77f26c 330 struct timespec *mtime)
3499e8a5
YS		 331{
332 struct ceph_msg *msg = req->r_request;
333 struct ceph_osd_request_head *head;
68b4476b 334 struct ceph_osd_req_op *src_op;
3499e8a5
YS		 335 struct ceph_osd_op *op;
336 void *p;
3499e8a5 337 size_t msg_size = sizeof(*head) + num_op*sizeof(*op);
3499e8a5 338 int flags = req->r_flags;
f44246e3 339 u64 data_len;
68b4476b 340 int i;
3499e8a5 341
d178a9e7
AE		 342 WARN_ON((flags & (CEPH_OSD_FLAG_READ|CEPH_OSD_FLAG_WRITE)) == 0);
343
f24e9980 344 head = msg->front.iov_base;
4d6b250b 345 head->snapid = cpu_to_le64(snap_id);
f24e9980
SW		 346 op = (void *)(head + 1);
347 p = (void *)(op + num_op);
348
f24e9980
SW		 349 req->r_snapc = ceph_get_snap_context(snapc);
350
351 head->client_inc = cpu_to_le32(1); /* always, for now. */
352 head->flags = cpu_to_le32(flags);
353 if (flags & CEPH_OSD_FLAG_WRITE)
354 ceph_encode_timespec(&head->mtime, mtime);
ae7ca4a3 355 BUG_ON(num_op > (unsigned int) ((u16) -1));
f24e9980 356 head->num_ops = cpu_to_le16(num_op);
f24e9980 357
f24e9980 358 /* fill in oid */
af77f26c
AE		 359 head->object_len = cpu_to_le32(req->r_oid_len);
360 memcpy(p, req->r_oid, req->r_oid_len);
361 p += req->r_oid_len;
f24e9980 362
68b4476b 363 src_op = src_ops;
ae7ca4a3
AE		 364 while (num_op--)
365 osd_req_encode_op(req, op++, src_op++);
68b4476b 366
f24e9980
SW		 367 if (snapc) {
368 head->snap_seq = cpu_to_le64(snapc->seq);
369 head->num_snaps = cpu_to_le32(snapc->num_snaps);
370 for (i = 0; i < snapc->num_snaps; i++) {
371 put_unaligned_le64(snapc->snaps[i], p);
372 p += sizeof(u64);
373 }
374 }
375
f44246e3 376 data_len = req->r_trail.length;
68b4476b
YS		 377 if (flags & CEPH_OSD_FLAG_WRITE) {
378 req->r_request->hdr.data_off = cpu_to_le16(off);
f44246e3 379 data_len += len;
68b4476b 380 }
f44246e3 381 req->r_request->hdr.data_len = cpu_to_le32(data_len);
c5c6b19d
SW		 382 req->r_request->page_alignment = req->r_page_alignment;
383
f24e9980 384 BUG_ON(p > msg->front.iov_base + msg->front.iov_len);
6f863e71
SW		 385 msg_size = p - msg->front.iov_base;
386 msg->front.iov_len = msg_size;
387 msg->hdr.front_len = cpu_to_le32(msg_size);
3499e8a5
YS		 388 return;
389}
3d14c5d2 390EXPORT_SYMBOL(ceph_osdc_build_request);
3499e8a5
YS		 391
392/*
393 * build new request AND message, calculate layout, and adjust file
394 * extent as needed.
395 *
396 * if the file was recently truncated, we include information about its
397 * old and new size so that the object can be updated appropriately. (we
398 * avoid synchronously deleting truncated objects because it's slow.)
399 *
400 * if @do_sync, include a 'startsync' command so that the osd will flush
401 * data quickly.
402 */
403struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
404 struct ceph_file_layout *layout,
405 struct ceph_vino vino,
406 u64 off, u64 *plen,
407 int opcode, int flags,
408 struct ceph_snap_context *snapc,
409 int do_sync,
410 u32 truncate_seq,
411 u64 truncate_size,
412 struct timespec *mtime,
a3bea47e 413 bool use_mempool,
b7495fc2 414 int page_align)
3499e8a5 415{
ae7ca4a3 416 struct ceph_osd_req_op ops[2];
68b4476b 417 struct ceph_osd_request *req;
ae7ca4a3 418 unsigned int num_op = 1;
6816282d 419 int r;
68b4476b 420
ae7ca4a3
AE		 421 memset(&ops, 0, sizeof ops);
422
68b4476b
YS		 423 ops[0].op = opcode;
424 ops[0].extent.truncate_seq = truncate_seq;
425 ops[0].extent.truncate_size = truncate_size;
68b4476b
YS		 426
427 if (do_sync) {
428 ops[1].op = CEPH_OSD_OP_STARTSYNC;
ae7ca4a3
AE		 429 num_op++;
430 }
68b4476b 431
ae7ca4a3
AE		 432 req = ceph_osdc_alloc_request(osdc, snapc, num_op, use_mempool,
433 GFP_NOFS);
4ad12621 434 if (!req)
6816282d 435 return ERR_PTR(-ENOMEM);
d178a9e7 436 req->r_flags = flags;
3499e8a5
YS		 437
438 /* calculate max write size */
e75b45cf 439 r = calc_layout(vino, layout, off, plen, req, ops);
6816282d
SW		 440 if (r < 0)
441 return ERR_PTR(r);
3499e8a5
YS		 442 req->r_file_layout = *layout; /* keep a copy */
443
9bb0ce2b
SW		 444 /* in case it differs from natural (file) alignment that
445 calc_layout filled in for us */
446 req->r_num_pages = calc_pages_for(page_align, *plen);
b7495fc2
SW		 447 req->r_page_alignment = page_align;
448
ae7ca4a3
AE		 449 ceph_osdc_build_request(req, off, *plen, num_op, ops,
450 snapc, vino.snap, mtime);
3499e8a5 451
f24e9980
SW		 452 return req;
453}
3d14c5d2 454EXPORT_SYMBOL(ceph_osdc_new_request);
f24e9980
SW		 455
456/*
457 * We keep osd requests in an rbtree, sorted by ->r_tid.
458 */
459static void __insert_request(struct ceph_osd_client *osdc,
460 struct ceph_osd_request *new)
461{
462 struct rb_node **p = &osdc->requests.rb_node;
463 struct rb_node *parent = NULL;
464 struct ceph_osd_request *req = NULL;
465
466 while (*p) {
467 parent = *p;
468 req = rb_entry(parent, struct ceph_osd_request, r_node);
469 if (new->r_tid < req->r_tid)
470 p = &(*p)->rb_left;
471 else if (new->r_tid > req->r_tid)
472 p = &(*p)->rb_right;
473 else
474 BUG();
475 }
476
477 rb_link_node(&new->r_node, parent, p);
478 rb_insert_color(&new->r_node, &osdc->requests);
479}
480
481static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc,
482 u64 tid)
483{
484 struct ceph_osd_request *req;
485 struct rb_node *n = osdc->requests.rb_node;
486
487 while (n) {
488 req = rb_entry(n, struct ceph_osd_request, r_node);
489 if (tid < req->r_tid)
490 n = n->rb_left;
491 else if (tid > req->r_tid)
492 n = n->rb_right;
493 else
494 return req;
495 }
496 return NULL;
497}
498
499static struct ceph_osd_request *
500__lookup_request_ge(struct ceph_osd_client *osdc,
501 u64 tid)
502{
503 struct ceph_osd_request *req;
504 struct rb_node *n = osdc->requests.rb_node;
505
506 while (n) {
507 req = rb_entry(n, struct ceph_osd_request, r_node);
508 if (tid < req->r_tid) {
509 if (!n->rb_left)
510 return req;
511 n = n->rb_left;
512 } else if (tid > req->r_tid) {
513 n = n->rb_right;
514 } else {
515 return req;
516 }
517 }
518 return NULL;
519}
520
6f6c7006
SW		 521/*
522 * Resubmit requests pending on the given osd.
523 */
524static void __kick_osd_requests(struct ceph_osd_client *osdc,
525 struct ceph_osd *osd)
526{
a40c4f10 527 struct ceph_osd_request *req, *nreq;
6f6c7006
SW		 528 int err;
529
530 dout("__kick_osd_requests osd%d\n", osd->o_osd);
531 err = __reset_osd(osdc, osd);
685a7555 532 if (err)
6f6c7006
SW		 533 return;
534
535 list_for_each_entry(req, &osd->o_requests, r_osd_item) {
536 list_move(&req->r_req_lru_item, &osdc->req_unsent);
537 dout("requeued %p tid %llu osd%d\n", req, req->r_tid,
538 osd->o_osd);
a40c4f10
YS		 539 if (!req->r_linger)
540 req->r_flags |= CEPH_OSD_FLAG_RETRY;
541 }
542
543 list_for_each_entry_safe(req, nreq, &osd->o_linger_requests,
544 r_linger_osd) {
77f38e0e
SW		 545 /*
546 * reregister request prior to unregistering linger so
547 * that r_osd is preserved.
548 */
549 BUG_ON(!list_empty(&req->r_req_lru_item));
a40c4f10 550 __register_request(osdc, req);
77f38e0e
SW		 551 list_add(&req->r_req_lru_item, &osdc->req_unsent);
552 list_add(&req->r_osd_item, &req->r_osd->o_requests);
553 __unregister_linger_request(osdc, req);
a40c4f10
YS		 554 dout("requeued lingering %p tid %llu osd%d\n", req, req->r_tid,
555 osd->o_osd);
6f6c7006
SW		 556 }
557}
558
f24e9980 559/*
81b024e7 560 * If the osd connection drops, we need to resubmit all requests.
f24e9980
SW		 561 */
562static void osd_reset(struct ceph_connection *con)
563{
564 struct ceph_osd *osd = con->private;
565 struct ceph_osd_client *osdc;
566
567 if (!osd)
568 return;
569 dout("osd_reset osd%d\n", osd->o_osd);
570 osdc = osd->o_osdc;
f24e9980 571 down_read(&osdc->map_sem);
83aff95e
SW		 572 mutex_lock(&osdc->request_mutex);
573 __kick_osd_requests(osdc, osd);
f9d25199 574 __send_queued(osdc);
83aff95e 575 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 576 up_read(&osdc->map_sem);
577}
578
579/*
580 * Track open sessions with osds.
581 */
e10006f8 582static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
f24e9980
SW		 583{
584 struct ceph_osd *osd;
585
586 osd = kzalloc(sizeof(*osd), GFP_NOFS);
587 if (!osd)
588 return NULL;
589
590 atomic_set(&osd->o_ref, 1);
591 osd->o_osdc = osdc;
e10006f8 592 osd->o_osd = onum;
f407731d 593 RB_CLEAR_NODE(&osd->o_node);
f24e9980 594 INIT_LIST_HEAD(&osd->o_requests);
a40c4f10 595 INIT_LIST_HEAD(&osd->o_linger_requests);
f5a2041b 596 INIT_LIST_HEAD(&osd->o_osd_lru);
f24e9980
SW		 597 osd->o_incarnation = 1;
598
b7a9e5dd 599 ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
4e7a5dcd 600
422d2cb8 601 INIT_LIST_HEAD(&osd->o_keepalive_item);
f24e9980
SW		 602 return osd;
603}
604
605static struct ceph_osd *get_osd(struct ceph_osd *osd)
606{
607 if (atomic_inc_not_zero(&osd->o_ref)) {
608 dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
609 atomic_read(&osd->o_ref));
610 return osd;
611 } else {
612 dout("get_osd %p FAIL\n", osd);
613 return NULL;
614 }
615}
616
617static void put_osd(struct ceph_osd *osd)
618{
619 dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
620 atomic_read(&osd->o_ref) - 1);
a255651d 621 if (atomic_dec_and_test(&osd->o_ref) && osd->o_auth.authorizer) {
79494d1b
SW		 622 struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth;
623
a255651d 624 if (ac->ops && ac->ops->destroy_authorizer)
6c4a1915 625 ac->ops->destroy_authorizer(ac, osd->o_auth.authorizer);
f24e9980 626 kfree(osd);
79494d1b 627 }
f24e9980
SW		 628}
629
630/*
631 * remove an osd from our map
632 */
f5a2041b 633static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
f24e9980 634{
f5a2041b 635 dout("__remove_osd %p\n", osd);
f24e9980
SW		 636 BUG_ON(!list_empty(&osd->o_requests));
637 rb_erase(&osd->o_node, &osdc->osds);
f5a2041b 638 list_del_init(&osd->o_osd_lru);
f24e9980
SW		 639 ceph_con_close(&osd->o_con);
640 put_osd(osd);
641}
642
aca420bc
SW		 643static void remove_all_osds(struct ceph_osd_client *osdc)
644{
048a9d2d 645 dout("%s %p\n", __func__, osdc);
aca420bc
SW		 646 mutex_lock(&osdc->request_mutex);
647 while (!RB_EMPTY_ROOT(&osdc->osds)) {
648 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
649 struct ceph_osd, o_node);
650 __remove_osd(osdc, osd);
651 }
652 mutex_unlock(&osdc->request_mutex);
653}
654
f5a2041b
YS		 655static void __move_osd_to_lru(struct ceph_osd_client *osdc,
656 struct ceph_osd *osd)
657{
658 dout("__move_osd_to_lru %p\n", osd);
659 BUG_ON(!list_empty(&osd->o_osd_lru));
660 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
3d14c5d2 661 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl * HZ;
f5a2041b
YS		 662}
663
664static void __remove_osd_from_lru(struct ceph_osd *osd)
665{
666 dout("__remove_osd_from_lru %p\n", osd);
667 if (!list_empty(&osd->o_osd_lru))
668 list_del_init(&osd->o_osd_lru);
669}
670
aca420bc 671static void remove_old_osds(struct ceph_osd_client *osdc)
f5a2041b
YS		 672{
673 struct ceph_osd *osd, *nosd;
674
675 dout("__remove_old_osds %p\n", osdc);
676 mutex_lock(&osdc->request_mutex);
677 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
aca420bc 678 if (time_before(jiffies, osd->lru_ttl))
f5a2041b
YS		 679 break;
680 __remove_osd(osdc, osd);
681 }
682 mutex_unlock(&osdc->request_mutex);
683}
684
f24e9980
SW		 685/*
686 * reset osd connect
687 */
f5a2041b 688static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
f24e9980 689{
c3acb181 690 struct ceph_entity_addr *peer_addr;
f24e9980 691
f5a2041b 692 dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
a40c4f10
YS		 693 if (list_empty(&osd->o_requests) &&
694 list_empty(&osd->o_linger_requests)) {
f5a2041b 695 __remove_osd(osdc, osd);
c3acb181
AE		 696
697 return -ENODEV;
698 }
699
700 peer_addr = &osdc->osdmap->osd_addr[osd->o_osd];
701 if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
702 !ceph_con_opened(&osd->o_con)) {
703 struct ceph_osd_request *req;
704
87b315a5
SW		 705 dout(" osd addr hasn't changed and connection never opened,"
706 " letting msgr retry");
707 /* touch each r_stamp for handle_timeout()'s benfit */
708 list_for_each_entry(req, &osd->o_requests, r_osd_item)
709 req->r_stamp = jiffies;
c3acb181
AE		 710
711 return -EAGAIN;
f24e9980 712 }
c3acb181
AE		 713
714 ceph_con_close(&osd->o_con);
715 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
716 osd->o_incarnation++;
717
718 return 0;
f24e9980
SW		 719}
720
721static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new)
722{
723 struct rb_node **p = &osdc->osds.rb_node;
724 struct rb_node *parent = NULL;
725 struct ceph_osd *osd = NULL;
726
aca420bc 727 dout("__insert_osd %p osd%d\n", new, new->o_osd);
f24e9980
SW		 728 while (*p) {
729 parent = *p;
730 osd = rb_entry(parent, struct ceph_osd, o_node);
731 if (new->o_osd < osd->o_osd)
732 p = &(*p)->rb_left;
733 else if (new->o_osd > osd->o_osd)
734 p = &(*p)->rb_right;
735 else
736 BUG();
737 }
738
739 rb_link_node(&new->o_node, parent, p);
740 rb_insert_color(&new->o_node, &osdc->osds);
741}
742
743static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o)
744{
745 struct ceph_osd *osd;
746 struct rb_node *n = osdc->osds.rb_node;
747
748 while (n) {
749 osd = rb_entry(n, struct ceph_osd, o_node);
750 if (o < osd->o_osd)
751 n = n->rb_left;
752 else if (o > osd->o_osd)
753 n = n->rb_right;
754 else
755 return osd;
756 }
757 return NULL;
758}
759
422d2cb8
YS		 760static void __schedule_osd_timeout(struct ceph_osd_client *osdc)
761{
762 schedule_delayed_work(&osdc->timeout_work,
3d14c5d2 763 osdc->client->options->osd_keepalive_timeout * HZ);
422d2cb8
YS		 764}
765
766static void __cancel_osd_timeout(struct ceph_osd_client *osdc)
767{
768 cancel_delayed_work(&osdc->timeout_work);
769}
f24e9980
SW		 770
771/*
772 * Register request, assign tid. If this is the first request, set up
773 * the timeout event.
774 */
a40c4f10
YS		 775static void __register_request(struct ceph_osd_client *osdc,
776 struct ceph_osd_request *req)
f24e9980 777{
f24e9980 778 req->r_tid = ++osdc->last_tid;
6df058c0 779 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
77f38e0e 780 dout("__register_request %p tid %lld\n", req, req->r_tid);
f24e9980
SW		 781 __insert_request(osdc, req);
782 ceph_osdc_get_request(req);
783 osdc->num_requests++;
f24e9980 784 if (osdc->num_requests == 1) {
422d2cb8
YS		 785 dout(" first request, scheduling timeout\n");
786 __schedule_osd_timeout(osdc);
f24e9980 787 }
a40c4f10
YS		 788}
789
790static void register_request(struct ceph_osd_client *osdc,
791 struct ceph_osd_request *req)
792{
793 mutex_lock(&osdc->request_mutex);
794 __register_request(osdc, req);
f24e9980
SW		 795 mutex_unlock(&osdc->request_mutex);
796}
797
798/*
799 * called under osdc->request_mutex
800 */
801static void __unregister_request(struct ceph_osd_client *osdc,
802 struct ceph_osd_request *req)
803{
35f9f8a0
SW		 804 if (RB_EMPTY_NODE(&req->r_node)) {
805 dout("__unregister_request %p tid %lld not registered\n",
806 req, req->r_tid);
807 return;
808 }
809
f24e9980
SW		 810 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
811 rb_erase(&req->r_node, &osdc->requests);
812 osdc->num_requests--;
813
0ba6478d
SW		 814 if (req->r_osd) {
815 /* make sure the original request isn't in flight. */
6740a845 816 ceph_msg_revoke(req->r_request);
0ba6478d
SW		 817
818 list_del_init(&req->r_osd_item);
a40c4f10
YS		 819 if (list_empty(&req->r_osd->o_requests) &&
820 list_empty(&req->r_osd->o_linger_requests)) {
821 dout("moving osd to %p lru\n", req->r_osd);
f5a2041b 822 __move_osd_to_lru(osdc, req->r_osd);
a40c4f10 823 }
fbdb9190 824 if (list_empty(&req->r_linger_item))
a40c4f10 825 req->r_osd = NULL;
0ba6478d 826 }
f24e9980 827
7d5f2481 828 list_del_init(&req->r_req_lru_item);
f24e9980
SW		 829 ceph_osdc_put_request(req);
830
422d2cb8
YS		 831 if (osdc->num_requests == 0) {
832 dout(" no requests, canceling timeout\n");
833 __cancel_osd_timeout(osdc);
f24e9980
SW		 834 }
835}
836
837/*
838 * Cancel a previously queued request message
839 */
840static void __cancel_request(struct ceph_osd_request *req)
841{
6bc18876 842 if (req->r_sent && req->r_osd) {
6740a845 843 ceph_msg_revoke(req->r_request);
f24e9980
SW		 844 req->r_sent = 0;
845 }
846}
847
a40c4f10
YS		 848static void __register_linger_request(struct ceph_osd_client *osdc,
849 struct ceph_osd_request *req)
850{
851 dout("__register_linger_request %p\n", req);
852 list_add_tail(&req->r_linger_item, &osdc->req_linger);
6194ea89
SW		 853 if (req->r_osd)
854 list_add_tail(&req->r_linger_osd,
855 &req->r_osd->o_linger_requests);
a40c4f10
YS		 856}
857
858static void __unregister_linger_request(struct ceph_osd_client *osdc,
859 struct ceph_osd_request *req)
860{
861 dout("__unregister_linger_request %p\n", req);
61c74035 862 list_del_init(&req->r_linger_item);
a40c4f10 863 if (req->r_osd) {
a40c4f10
YS		 864 list_del_init(&req->r_linger_osd);
865
866 if (list_empty(&req->r_osd->o_requests) &&
867 list_empty(&req->r_osd->o_linger_requests)) {
868 dout("moving osd to %p lru\n", req->r_osd);
869 __move_osd_to_lru(osdc, req->r_osd);
870 }
fbdb9190
SW		 871 if (list_empty(&req->r_osd_item))
872 req->r_osd = NULL;
a40c4f10
YS		 873 }
874}
875
876void ceph_osdc_unregister_linger_request(struct ceph_osd_client *osdc,
877 struct ceph_osd_request *req)
878{
879 mutex_lock(&osdc->request_mutex);
880 if (req->r_linger) {
881 __unregister_linger_request(osdc, req);
882 ceph_osdc_put_request(req);
883 }
884 mutex_unlock(&osdc->request_mutex);
885}
886EXPORT_SYMBOL(ceph_osdc_unregister_linger_request);
887
888void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
889 struct ceph_osd_request *req)
890{
891 if (!req->r_linger) {
892 dout("set_request_linger %p\n", req);
893 req->r_linger = 1;
894 /*
895 * caller is now responsible for calling
896 * unregister_linger_request
897 */
898 ceph_osdc_get_request(req);
899 }
900}
901EXPORT_SYMBOL(ceph_osdc_set_request_linger);
902
f24e9980
SW		 903/*
904 * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
905 * (as needed), and set the request r_osd appropriately. If there is
25985edc 906 * no up osd, set r_osd to NULL. Move the request to the appropriate list
6f6c7006 907 * (unsent, homeless) or leave on in-flight lru.
f24e9980
SW		 908 *
909 * Return 0 if unchanged, 1 if changed, or negative on error.
910 *
911 * Caller should hold map_sem for read and request_mutex.
912 */
6f6c7006 913static int __map_request(struct ceph_osd_client *osdc,
38d6453c 914 struct ceph_osd_request *req, int force_resend)
f24e9980
SW		 915{
916 struct ceph_osd_request_head *reqhead = req->r_request->front.iov_base;
5b191d99 917 struct ceph_pg pgid;
d85b7056
SW		 918 int acting[CEPH_PG_MAX_SIZE];
919 int o = -1, num = 0;
f24e9980 920 int err;
f24e9980 921
6f6c7006 922 dout("map_request %p tid %lld\n", req, req->r_tid);
f24e9980
SW		 923 err = ceph_calc_object_layout(&reqhead->layout, req->r_oid,
924 &req->r_file_layout, osdc->osdmap);
6f6c7006
SW		 925 if (err) {
926 list_move(&req->r_req_lru_item, &osdc->req_notarget);
f24e9980 927 return err;
6f6c7006 928 }
5b191d99
SW		 929 pgid.pool = le32_to_cpu(reqhead->layout.ol_pgid.pool);
930 pgid.seed = le16_to_cpu(reqhead->layout.ol_pgid.ps);
7740a42f
SW		 931 req->r_pgid = pgid;
932
d85b7056
SW		 933 err = ceph_calc_pg_acting(osdc->osdmap, pgid, acting);
934 if (err > 0) {
935 o = acting[0];
936 num = err;
937 }
f24e9980 938
38d6453c
SW		 939 if ((!force_resend &&
940 req->r_osd && req->r_osd->o_osd == o &&
d85b7056
SW		 941 req->r_sent >= req->r_osd->o_incarnation &&
942 req->r_num_pg_osds == num &&
943 memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
f24e9980
SW		 944 (req->r_osd == NULL && o == -1))
945 return 0; /* no change */
946
5b191d99
SW		 947 dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n",
948 req->r_tid, pgid.pool, pgid.seed, o,
f24e9980
SW		 949 req->r_osd ? req->r_osd->o_osd : -1);
950
d85b7056
SW		 951 /* record full pg acting set */
952 memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
953 req->r_num_pg_osds = num;
954
f24e9980
SW		 955 if (req->r_osd) {
956 __cancel_request(req);
957 list_del_init(&req->r_osd_item);
f24e9980
SW		 958 req->r_osd = NULL;
959 }
960
961 req->r_osd = __lookup_osd(osdc, o);
962 if (!req->r_osd && o >= 0) {
c99eb1c7 963 err = -ENOMEM;
e10006f8 964 req->r_osd = create_osd(osdc, o);
6f6c7006
SW		 965 if (!req->r_osd) {
966 list_move(&req->r_req_lru_item, &osdc->req_notarget);
c99eb1c7 967 goto out;
6f6c7006 968 }
f24e9980 969
6f6c7006 970 dout("map_request osd %p is osd%d\n", req->r_osd, o);
f24e9980
SW		 971 __insert_osd(osdc, req->r_osd);
972
b7a9e5dd
SW		 973 ceph_con_open(&req->r_osd->o_con,
974 CEPH_ENTITY_TYPE_OSD, o,
975 &osdc->osdmap->osd_addr[o]);
f24e9980
SW		 976 }
977
f5a2041b
YS		 978 if (req->r_osd) {
979 __remove_osd_from_lru(req->r_osd);
f24e9980 980 list_add(&req->r_osd_item, &req->r_osd->o_requests);
6f6c7006
SW		 981 list_move(&req->r_req_lru_item, &osdc->req_unsent);
982 } else {
983 list_move(&req->r_req_lru_item, &osdc->req_notarget);
f5a2041b 984 }
d85b7056 985 err = 1; /* osd or pg changed */
f24e9980
SW		 986
987out:
f24e9980
SW		 988 return err;
989}
990
991/*
992 * caller should hold map_sem (for read) and request_mutex
993 */
56e925b6
SW		 994static void __send_request(struct ceph_osd_client *osdc,
995 struct ceph_osd_request *req)
f24e9980
SW		 996{
997 struct ceph_osd_request_head *reqhead;
f24e9980
SW		 998
999 dout("send_request %p tid %llu to osd%d flags %d\n",
1000 req, req->r_tid, req->r_osd->o_osd, req->r_flags);
1001
1002 reqhead = req->r_request->front.iov_base;
1003 reqhead->osdmap_epoch = cpu_to_le32(osdc->osdmap->epoch);
1004 reqhead->flags |= cpu_to_le32(req->r_flags); /* e.g., RETRY */
1005 reqhead->reassert_version = req->r_reassert_version;
1006
3dd72fc0 1007 req->r_stamp = jiffies;
07a27e22 1008 list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
f24e9980
SW		 1009
1010 ceph_msg_get(req->r_request); /* send consumes a ref */
1011 ceph_con_send(&req->r_osd->o_con, req->r_request);
1012 req->r_sent = req->r_osd->o_incarnation;
f24e9980
SW		 1013}
1014
6f6c7006
SW		 1015/*
1016 * Send any requests in the queue (req_unsent).
1017 */
f9d25199 1018static void __send_queued(struct ceph_osd_client *osdc)
6f6c7006
SW		 1019{
1020 struct ceph_osd_request *req, *tmp;
1021
f9d25199
AE		 1022 dout("__send_queued\n");
1023 list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item)
6f6c7006 1024 __send_request(osdc, req);
6f6c7006
SW		 1025}
1026
f24e9980
SW		 1027/*
1028 * Timeout callback, called every N seconds when 1 or more osd
1029 * requests has been active for more than N seconds. When this
1030 * happens, we ping all OSDs with requests who have timed out to
1031 * ensure any communications channel reset is detected. Reset the
1032 * request timeouts another N seconds in the future as we go.
1033 * Reschedule the timeout event another N seconds in future (unless
1034 * there are no open requests).
1035 */
1036static void handle_timeout(struct work_struct *work)
1037{
1038 struct ceph_osd_client *osdc =
1039 container_of(work, struct ceph_osd_client, timeout_work.work);
83aff95e 1040 struct ceph_osd_request *req;
f24e9980 1041 struct ceph_osd *osd;
422d2cb8 1042 unsigned long keepalive =
3d14c5d2 1043 osdc->client->options->osd_keepalive_timeout * HZ;
422d2cb8 1044 struct list_head slow_osds;
f24e9980
SW		 1045 dout("timeout\n");
1046 down_read(&osdc->map_sem);
1047
1048 ceph_monc_request_next_osdmap(&osdc->client->monc);
1049
1050 mutex_lock(&osdc->request_mutex);
f24e9980 1051
422d2cb8
YS		 1052 /*
1053 * ping osds that are a bit slow. this ensures that if there
1054 * is a break in the TCP connection we will notice, and reopen
1055 * a connection with that osd (from the fault callback).
1056 */
1057 INIT_LIST_HEAD(&slow_osds);
1058 list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
3dd72fc0 1059 if (time_before(jiffies, req->r_stamp + keepalive))
422d2cb8
YS		 1060 break;
1061
1062 osd = req->r_osd;
1063 BUG_ON(!osd);
1064 dout(" tid %llu is slow, will send keepalive on osd%d\n",
f24e9980 1065 req->r_tid, osd->o_osd);
422d2cb8
YS		 1066 list_move_tail(&osd->o_keepalive_item, &slow_osds);
1067 }
1068 while (!list_empty(&slow_osds)) {
1069 osd = list_entry(slow_osds.next, struct ceph_osd,
1070 o_keepalive_item);
1071 list_del_init(&osd->o_keepalive_item);
f24e9980
SW		 1072 ceph_con_keepalive(&osd->o_con);
1073 }
1074
422d2cb8 1075 __schedule_osd_timeout(osdc);
f9d25199 1076 __send_queued(osdc);
f24e9980 1077 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 1078 up_read(&osdc->map_sem);
1079}
1080
f5a2041b
YS		 1081static void handle_osds_timeout(struct work_struct *work)
1082{
1083 struct ceph_osd_client *osdc =
1084 container_of(work, struct ceph_osd_client,
1085 osds_timeout_work.work);
1086 unsigned long delay =
3d14c5d2 1087 osdc->client->options->osd_idle_ttl * HZ >> 2;
f5a2041b
YS		 1088
1089 dout("osds timeout\n");
1090 down_read(&osdc->map_sem);
aca420bc 1091 remove_old_osds(osdc);
f5a2041b
YS		 1092 up_read(&osdc->map_sem);
1093
1094 schedule_delayed_work(&osdc->osds_timeout_work,
1095 round_jiffies_relative(delay));
1096}
1097
25845472
SW		 1098static void complete_request(struct ceph_osd_request *req)
1099{
1100 if (req->r_safe_callback)
1101 req->r_safe_callback(req, NULL);
1102 complete_all(&req->r_safe_completion); /* fsync waiter */
1103}
1104
f24e9980
SW		 1105/*
1106 * handle osd op reply. either call the callback if it is specified,
1107 * or do the completion to wake up the waiting thread.
1108 */
350b1c32
SW		 1109static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg,
1110 struct ceph_connection *con)
f24e9980
SW		 1111{
1112 struct ceph_osd_reply_head *rhead = msg->front.iov_base;
1113 struct ceph_osd_request *req;
1114 u64 tid;
1115 int numops, object_len, flags;
0ceed5db 1116 s32 result;
f24e9980 1117
6df058c0 1118 tid = le64_to_cpu(msg->hdr.tid);
f24e9980
SW		 1119 if (msg->front.iov_len < sizeof(*rhead))
1120 goto bad;
f24e9980
SW		 1121 numops = le32_to_cpu(rhead->num_ops);
1122 object_len = le32_to_cpu(rhead->object_len);
0ceed5db 1123 result = le32_to_cpu(rhead->result);
f24e9980
SW		 1124 if (msg->front.iov_len != sizeof(*rhead) + object_len +
1125 numops * sizeof(struct ceph_osd_op))
1126 goto bad;
0ceed5db 1127 dout("handle_reply %p tid %llu result %d\n", msg, tid, (int)result);
f24e9980
SW		 1128 /* lookup */
1129 mutex_lock(&osdc->request_mutex);
1130 req = __lookup_request(osdc, tid);
1131 if (req == NULL) {
1132 dout("handle_reply tid %llu dne\n", tid);
1133 mutex_unlock(&osdc->request_mutex);
1134 return;
1135 }
1136 ceph_osdc_get_request(req);
1137 flags = le32_to_cpu(rhead->flags);
1138
350b1c32 1139 /*
0d59ab81 1140 * if this connection filled our message, drop our reference now, to
350b1c32
SW		 1141 * avoid a (safe but slower) revoke later.
1142 */
0d59ab81 1143 if (req->r_con_filling_msg == con && req->r_reply == msg) {
c16e7869 1144 dout(" dropping con_filling_msg ref %p\n", con);
0d59ab81 1145 req->r_con_filling_msg = NULL;
0d47766f 1146 con->ops->put(con);
350b1c32
SW		 1147 }
1148
f24e9980 1149 if (!req->r_got_reply) {
95c96174 1150 unsigned int bytes;
f24e9980
SW		 1151
1152 req->r_result = le32_to_cpu(rhead->result);
1153 bytes = le32_to_cpu(msg->hdr.data_len);
1154 dout("handle_reply result %d bytes %d\n", req->r_result,
1155 bytes);
1156 if (req->r_result == 0)
1157 req->r_result = bytes;
1158
1159 /* in case this is a write and we need to replay, */
1160 req->r_reassert_version = rhead->reassert_version;
1161
1162 req->r_got_reply = 1;
1163 } else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) {
1164 dout("handle_reply tid %llu dup ack\n", tid);
34b43a56 1165 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 1166 goto done;
1167 }
1168
1169 dout("handle_reply tid %llu flags %d\n", tid, flags);
1170
a40c4f10
YS		 1171 if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK))
1172 __register_linger_request(osdc, req);
1173
f24e9980 1174 /* either this is a read, or we got the safe response */
0ceed5db
SW		 1175 if (result < 0 ||
1176 (flags & CEPH_OSD_FLAG_ONDISK) ||
f24e9980
SW		 1177 ((flags & CEPH_OSD_FLAG_WRITE) == 0))
1178 __unregister_request(osdc, req);
1179
1180 mutex_unlock(&osdc->request_mutex);
1181
1182 if (req->r_callback)
1183 req->r_callback(req, msg);
1184 else
03066f23 1185 complete_all(&req->r_completion);
f24e9980 1186
25845472
SW		 1187 if (flags & CEPH_OSD_FLAG_ONDISK)
1188 complete_request(req);
f24e9980
SW		 1189
1190done:
a40c4f10 1191 dout("req=%p req->r_linger=%d\n", req, req->r_linger);
f24e9980
SW		 1192 ceph_osdc_put_request(req);
1193 return;
1194
1195bad:
1196 pr_err("corrupt osd_op_reply got %d %d expected %d\n",
1197 (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len),
1198 (int)sizeof(*rhead));
9ec7cab1 1199 ceph_msg_dump(msg);
f24e9980
SW		 1200}
1201
6f6c7006 1202static void reset_changed_osds(struct ceph_osd_client *osdc)
f24e9980 1203{
f24e9980 1204 struct rb_node *p, *n;
f24e9980 1205
6f6c7006
SW		 1206 for (p = rb_first(&osdc->osds); p; p = n) {
1207 struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
f24e9980 1208
6f6c7006
SW		 1209 n = rb_next(p);
1210 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
1211 memcmp(&osd->o_con.peer_addr,
1212 ceph_osd_addr(osdc->osdmap,
1213 osd->o_osd),
1214 sizeof(struct ceph_entity_addr)) != 0)
1215 __reset_osd(osdc, osd);
f24e9980 1216 }
422d2cb8
YS		 1217}
1218
1219/*
6f6c7006
SW		 1220 * Requeue requests whose mapping to an OSD has changed. If requests map to
1221 * no osd, request a new map.
422d2cb8 1222 *
e6d50f67 1223 * Caller should hold map_sem for read.
422d2cb8 1224 */
38d6453c 1225static void kick_requests(struct ceph_osd_client *osdc, int force_resend)
422d2cb8 1226{
a40c4f10 1227 struct ceph_osd_request *req, *nreq;
6f6c7006
SW		 1228 struct rb_node *p;
1229 int needmap = 0;
1230 int err;
422d2cb8 1231
38d6453c 1232 dout("kick_requests %s\n", force_resend ? " (force resend)" : "");
422d2cb8 1233 mutex_lock(&osdc->request_mutex);
6194ea89 1234 for (p = rb_first(&osdc->requests); p; ) {
6f6c7006 1235 req = rb_entry(p, struct ceph_osd_request, r_node);
6194ea89 1236 p = rb_next(p);
ab60b16d
AE		 1237
1238 /*
1239 * For linger requests that have not yet been
1240 * registered, move them to the linger list; they'll
1241 * be sent to the osd in the loop below. Unregister
1242 * the request before re-registering it as a linger
1243 * request to ensure the __map_request() below
1244 * will decide it needs to be sent.
1245 */
1246 if (req->r_linger && list_empty(&req->r_linger_item)) {
1247 dout("%p tid %llu restart on osd%d\n",
1248 req, req->r_tid,
1249 req->r_osd ? req->r_osd->o_osd : -1);
1250 __unregister_request(osdc, req);
1251 __register_linger_request(osdc, req);
1252 continue;
1253 }
1254
38d6453c 1255 err = __map_request(osdc, req, force_resend);
6f6c7006
SW		 1256 if (err < 0)
1257 continue; /* error */
1258 if (req->r_osd == NULL) {
1259 dout("%p tid %llu maps to no osd\n", req, req->r_tid);
1260 needmap++; /* request a newer map */
1261 } else if (err > 0) {
6194ea89
SW		 1262 if (!req->r_linger) {
1263 dout("%p tid %llu requeued on osd%d\n", req,
1264 req->r_tid,
1265 req->r_osd ? req->r_osd->o_osd : -1);
a40c4f10 1266 req->r_flags |= CEPH_OSD_FLAG_RETRY;
6194ea89
SW		 1267 }
1268 }
a40c4f10
YS		 1269 }
1270
1271 list_for_each_entry_safe(req, nreq, &osdc->req_linger,
1272 r_linger_item) {
1273 dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
1274
38d6453c 1275 err = __map_request(osdc, req, force_resend);
ab60b16d 1276 dout("__map_request returned %d\n", err);
a40c4f10
YS		 1277 if (err == 0)
1278 continue; /* no change and no osd was specified */
1279 if (err < 0)
1280 continue; /* hrm! */
1281 if (req->r_osd == NULL) {
1282 dout("tid %llu maps to no valid osd\n", req->r_tid);
1283 needmap++; /* request a newer map */
1284 continue;
6f6c7006 1285 }
a40c4f10
YS		 1286
1287 dout("kicking lingering %p tid %llu osd%d\n", req, req->r_tid,
1288 req->r_osd ? req->r_osd->o_osd : -1);
a40c4f10 1289 __register_request(osdc, req);
c89ce05e 1290 __unregister_linger_request(osdc, req);
6f6c7006 1291 }
f24e9980
SW		 1292 mutex_unlock(&osdc->request_mutex);
1293
1294 if (needmap) {
1295 dout("%d requests for down osds, need new map\n", needmap);
1296 ceph_monc_request_next_osdmap(&osdc->client->monc);
1297 }
e6d50f67 1298 reset_changed_osds(osdc);
422d2cb8 1299}
6f6c7006
SW		 1300
1301
f24e9980
SW		 1302/*
1303 * Process updated osd map.
1304 *
1305 * The message contains any number of incremental and full maps, normally
1306 * indicating some sort of topology change in the cluster. Kick requests
1307 * off to different OSDs as needed.
1308 */
1309void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1310{
1311 void *p, *end, *next;
1312 u32 nr_maps, maplen;
1313 u32 epoch;
1314 struct ceph_osdmap *newmap = NULL, *oldmap;
1315 int err;
1316 struct ceph_fsid fsid;
1317
1318 dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
1319 p = msg->front.iov_base;
1320 end = p + msg->front.iov_len;
1321
1322 /* verify fsid */
1323 ceph_decode_need(&p, end, sizeof(fsid), bad);
1324 ceph_decode_copy(&p, &fsid, sizeof(fsid));
0743304d
SW		 1325 if (ceph_check_fsid(osdc->client, &fsid) < 0)
1326 return;
f24e9980
SW		 1327
1328 down_write(&osdc->map_sem);
1329
1330 /* incremental maps */
1331 ceph_decode_32_safe(&p, end, nr_maps, bad);
1332 dout(" %d inc maps\n", nr_maps);
1333 while (nr_maps > 0) {
1334 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 1335 epoch = ceph_decode_32(&p);
1336 maplen = ceph_decode_32(&p);
f24e9980
SW		 1337 ceph_decode_need(&p, end, maplen, bad);
1338 next = p + maplen;
1339 if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
1340 dout("applying incremental map %u len %d\n",
1341 epoch, maplen);
1342 newmap = osdmap_apply_incremental(&p, next,
1343 osdc->osdmap,
15d9882c 1344 &osdc->client->msgr);
f24e9980
SW		 1345 if (IS_ERR(newmap)) {
1346 err = PTR_ERR(newmap);
1347 goto bad;
1348 }
30dc6381 1349 BUG_ON(!newmap);
f24e9980
SW		 1350 if (newmap != osdc->osdmap) {
1351 ceph_osdmap_destroy(osdc->osdmap);
1352 osdc->osdmap = newmap;
1353 }
38d6453c 1354 kick_requests(osdc, 0);
f24e9980
SW		 1355 } else {
1356 dout("ignoring incremental map %u len %d\n",
1357 epoch, maplen);
1358 }
1359 p = next;
1360 nr_maps--;
1361 }
1362 if (newmap)
1363 goto done;
1364
1365 /* full maps */
1366 ceph_decode_32_safe(&p, end, nr_maps, bad);
1367 dout(" %d full maps\n", nr_maps);
1368 while (nr_maps) {
1369 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 1370 epoch = ceph_decode_32(&p);
1371 maplen = ceph_decode_32(&p);
f24e9980
SW		 1372 ceph_decode_need(&p, end, maplen, bad);
1373 if (nr_maps > 1) {
1374 dout("skipping non-latest full map %u len %d\n",
1375 epoch, maplen);
1376 } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
1377 dout("skipping full map %u len %d, "
1378 "older than our %u\n", epoch, maplen,
1379 osdc->osdmap->epoch);
1380 } else {
38d6453c
SW		 1381 int skipped_map = 0;
1382
f24e9980
SW		 1383 dout("taking full map %u len %d\n", epoch, maplen);
1384 newmap = osdmap_decode(&p, p+maplen);
1385 if (IS_ERR(newmap)) {
1386 err = PTR_ERR(newmap);
1387 goto bad;
1388 }
30dc6381 1389 BUG_ON(!newmap);
f24e9980
SW		 1390 oldmap = osdc->osdmap;
1391 osdc->osdmap = newmap;
38d6453c
SW		 1392 if (oldmap) {
1393 if (oldmap->epoch + 1 < newmap->epoch)
1394 skipped_map = 1;
f24e9980 1395 ceph_osdmap_destroy(oldmap);
38d6453c
SW		 1396 }
1397 kick_requests(osdc, skipped_map);
f24e9980
SW		 1398 }
1399 p += maplen;
1400 nr_maps--;
1401 }
1402
1403done:
1404 downgrade_write(&osdc->map_sem);
1405 ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch);
cd634fb6
SW		 1406
1407 /*
1408 * subscribe to subsequent osdmap updates if full to ensure
1409 * we find out when we are no longer full and stop returning
1410 * ENOSPC.
1411 */
1412 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL))
1413 ceph_monc_request_next_osdmap(&osdc->client->monc);
1414
f9d25199
AE		 1415 mutex_lock(&osdc->request_mutex);
1416 __send_queued(osdc);
1417 mutex_unlock(&osdc->request_mutex);
f24e9980 1418 up_read(&osdc->map_sem);
03066f23 1419 wake_up_all(&osdc->client->auth_wq);
f24e9980
SW		 1420 return;
1421
1422bad:
1423 pr_err("osdc handle_map corrupt msg\n");
9ec7cab1 1424 ceph_msg_dump(msg);
f24e9980
SW		 1425 up_write(&osdc->map_sem);
1426 return;
1427}
1428
a40c4f10
YS		 1429/*
1430 * watch/notify callback event infrastructure
1431 *
1432 * These callbacks are used both for watch and notify operations.
1433 */
1434static void __release_event(struct kref *kref)
1435{
1436 struct ceph_osd_event *event =
1437 container_of(kref, struct ceph_osd_event, kref);
1438
1439 dout("__release_event %p\n", event);
1440 kfree(event);
1441}
1442
1443static void get_event(struct ceph_osd_event *event)
1444{
1445 kref_get(&event->kref);
1446}
1447
1448void ceph_osdc_put_event(struct ceph_osd_event *event)
1449{
1450 kref_put(&event->kref, __release_event);
1451}
1452EXPORT_SYMBOL(ceph_osdc_put_event);
1453
1454static void __insert_event(struct ceph_osd_client *osdc,
1455 struct ceph_osd_event *new)
1456{
1457 struct rb_node **p = &osdc->event_tree.rb_node;
1458 struct rb_node *parent = NULL;
1459 struct ceph_osd_event *event = NULL;
1460
1461 while (*p) {
1462 parent = *p;
1463 event = rb_entry(parent, struct ceph_osd_event, node);
1464 if (new->cookie < event->cookie)
1465 p = &(*p)->rb_left;
1466 else if (new->cookie > event->cookie)
1467 p = &(*p)->rb_right;
1468 else
1469 BUG();
1470 }
1471
1472 rb_link_node(&new->node, parent, p);
1473 rb_insert_color(&new->node, &osdc->event_tree);
1474}
1475
1476static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
1477 u64 cookie)
1478{
1479 struct rb_node **p = &osdc->event_tree.rb_node;
1480 struct rb_node *parent = NULL;
1481 struct ceph_osd_event *event = NULL;
1482
1483 while (*p) {
1484 parent = *p;
1485 event = rb_entry(parent, struct ceph_osd_event, node);
1486 if (cookie < event->cookie)
1487 p = &(*p)->rb_left;
1488 else if (cookie > event->cookie)
1489 p = &(*p)->rb_right;
1490 else
1491 return event;
1492 }
1493 return NULL;
1494}
1495
1496static void __remove_event(struct ceph_osd_event *event)
1497{
1498 struct ceph_osd_client *osdc = event->osdc;
1499
1500 if (!RB_EMPTY_NODE(&event->node)) {
1501 dout("__remove_event removed %p\n", event);
1502 rb_erase(&event->node, &osdc->event_tree);
1503 ceph_osdc_put_event(event);
1504 } else {
1505 dout("__remove_event didn't remove %p\n", event);
1506 }
1507}
1508
1509int ceph_osdc_create_event(struct ceph_osd_client *osdc,
1510 void (*event_cb)(u64, u64, u8, void *),
3c663bbd 1511 void *data, struct ceph_osd_event **pevent)
a40c4f10
YS		 1512{
1513 struct ceph_osd_event *event;
1514
1515 event = kmalloc(sizeof(*event), GFP_NOIO);
1516 if (!event)
1517 return -ENOMEM;
1518
1519 dout("create_event %p\n", event);
1520 event->cb = event_cb;
3c663bbd 1521 event->one_shot = 0;
a40c4f10
YS		 1522 event->data = data;
1523 event->osdc = osdc;
1524 INIT_LIST_HEAD(&event->osd_node);
3ee5234d 1525 RB_CLEAR_NODE(&event->node);
a40c4f10
YS		 1526 kref_init(&event->kref); /* one ref for us */
1527 kref_get(&event->kref); /* one ref for the caller */
a40c4f10
YS		 1528
1529 spin_lock(&osdc->event_lock);
1530 event->cookie = ++osdc->event_count;
1531 __insert_event(osdc, event);
1532 spin_unlock(&osdc->event_lock);
1533
1534 *pevent = event;
1535 return 0;
1536}
1537EXPORT_SYMBOL(ceph_osdc_create_event);
1538
1539void ceph_osdc_cancel_event(struct ceph_osd_event *event)
1540{
1541 struct ceph_osd_client *osdc = event->osdc;
1542
1543 dout("cancel_event %p\n", event);
1544 spin_lock(&osdc->event_lock);
1545 __remove_event(event);
1546 spin_unlock(&osdc->event_lock);
1547 ceph_osdc_put_event(event); /* caller's */
1548}
1549EXPORT_SYMBOL(ceph_osdc_cancel_event);
1550
1551
1552static void do_event_work(struct work_struct *work)
1553{
1554 struct ceph_osd_event_work *event_work =
1555 container_of(work, struct ceph_osd_event_work, work);
1556 struct ceph_osd_event *event = event_work->event;
1557 u64 ver = event_work->ver;
1558 u64 notify_id = event_work->notify_id;
1559 u8 opcode = event_work->opcode;
1560
1561 dout("do_event_work completing %p\n", event);
1562 event->cb(ver, notify_id, opcode, event->data);
a40c4f10
YS		 1563 dout("do_event_work completed %p\n", event);
1564 ceph_osdc_put_event(event);
1565 kfree(event_work);
1566}
1567
1568
1569/*
1570 * Process osd watch notifications
1571 */
3c663bbd
AE		 1572static void handle_watch_notify(struct ceph_osd_client *osdc,
1573 struct ceph_msg *msg)
a40c4f10
YS		 1574{
1575 void *p, *end;
1576 u8 proto_ver;
1577 u64 cookie, ver, notify_id;
1578 u8 opcode;
1579 struct ceph_osd_event *event;
1580 struct ceph_osd_event_work *event_work;
1581
1582 p = msg->front.iov_base;
1583 end = p + msg->front.iov_len;
1584
1585 ceph_decode_8_safe(&p, end, proto_ver, bad);
1586 ceph_decode_8_safe(&p, end, opcode, bad);
1587 ceph_decode_64_safe(&p, end, cookie, bad);
1588 ceph_decode_64_safe(&p, end, ver, bad);
1589 ceph_decode_64_safe(&p, end, notify_id, bad);
1590
1591 spin_lock(&osdc->event_lock);
1592 event = __find_event(osdc, cookie);
1593 if (event) {
3c663bbd 1594 BUG_ON(event->one_shot);
a40c4f10 1595 get_event(event);
a40c4f10
YS		 1596 }
1597 spin_unlock(&osdc->event_lock);
1598 dout("handle_watch_notify cookie %lld ver %lld event %p\n",
1599 cookie, ver, event);
1600 if (event) {
1601 event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
a40c4f10
YS		 1602 if (!event_work) {
1603 dout("ERROR: could not allocate event_work\n");
1604 goto done_err;
1605 }
6b0ae409 1606 INIT_WORK(&event_work->work, do_event_work);
a40c4f10
YS		 1607 event_work->event = event;
1608 event_work->ver = ver;
1609 event_work->notify_id = notify_id;
1610 event_work->opcode = opcode;
1611 if (!queue_work(osdc->notify_wq, &event_work->work)) {
1612 dout("WARNING: failed to queue notify event work\n");
1613 goto done_err;
1614 }
1615 }
1616
1617 return;
1618
1619done_err:
a40c4f10
YS		 1620 ceph_osdc_put_event(event);
1621 return;
1622
1623bad:
1624 pr_err("osdc handle_watch_notify corrupt msg\n");
1625 return;
1626}
1627
f24e9980
SW		 1628/*
1629 * Register request, send initial attempt.
1630 */
1631int ceph_osdc_start_request(struct ceph_osd_client *osdc,
1632 struct ceph_osd_request *req,
1633 bool nofail)
1634{
c1ea8823 1635 int rc = 0;
f24e9980
SW		 1636
1637 req->r_request->pages = req->r_pages;
1638 req->r_request->nr_pages = req->r_num_pages;
68b4476b
YS		 1639#ifdef CONFIG_BLOCK
1640 req->r_request->bio = req->r_bio;
1641#endif
c885837f 1642 req->r_request->trail = &req->r_trail;
f24e9980
SW		 1643
1644 register_request(osdc, req);
1645
1646 down_read(&osdc->map_sem);
1647 mutex_lock(&osdc->request_mutex);
c1ea8823
SW		 1648 /*
1649 * a racing kick_requests() may have sent the message for us
1650 * while we dropped request_mutex above, so only send now if
1651 * the request still han't been touched yet.
1652 */
1653 if (req->r_sent == 0) {
38d6453c 1654 rc = __map_request(osdc, req, 0);
9d6fcb08
SW		 1655 if (rc < 0) {
1656 if (nofail) {
1657 dout("osdc_start_request failed map, "
1658 " will retry %lld\n", req->r_tid);
1659 rc = 0;
1660 }
234af26f 1661 goto out_unlock;
9d6fcb08 1662 }
6f6c7006
SW		 1663 if (req->r_osd == NULL) {
1664 dout("send_request %p no up osds in pg\n", req);
1665 ceph_monc_request_next_osdmap(&osdc->client->monc);
1666 } else {
56e925b6 1667 __send_request(osdc, req);
f24e9980 1668 }
56e925b6 1669 rc = 0;
f24e9980 1670 }
234af26f
DC		 1671
1672out_unlock:
f24e9980
SW		 1673 mutex_unlock(&osdc->request_mutex);
1674 up_read(&osdc->map_sem);
1675 return rc;
1676}
3d14c5d2 1677EXPORT_SYMBOL(ceph_osdc_start_request);
f24e9980
SW		 1678
1679/*
1680 * wait for a request to complete
1681 */
1682int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
1683 struct ceph_osd_request *req)
1684{
1685 int rc;
1686
1687 rc = wait_for_completion_interruptible(&req->r_completion);
1688 if (rc < 0) {
1689 mutex_lock(&osdc->request_mutex);
1690 __cancel_request(req);
529cfcc4 1691 __unregister_request(osdc, req);
f24e9980 1692 mutex_unlock(&osdc->request_mutex);
25845472 1693 complete_request(req);
529cfcc4 1694 dout("wait_request tid %llu canceled/timed out\n", req->r_tid);
f24e9980
SW		 1695 return rc;
1696 }
1697
1698 dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result);
1699 return req->r_result;
1700}
3d14c5d2 1701EXPORT_SYMBOL(ceph_osdc_wait_request);
f24e9980
SW		 1702
1703/*
1704 * sync - wait for all in-flight requests to flush. avoid starvation.
1705 */
1706void ceph_osdc_sync(struct ceph_osd_client *osdc)
1707{
1708 struct ceph_osd_request *req;
1709 u64 last_tid, next_tid = 0;
1710
1711 mutex_lock(&osdc->request_mutex);
1712 last_tid = osdc->last_tid;
1713 while (1) {
1714 req = __lookup_request_ge(osdc, next_tid);
1715 if (!req)
1716 break;
1717 if (req->r_tid > last_tid)
1718 break;
1719
1720 next_tid = req->r_tid + 1;
1721 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
1722 continue;
1723
1724 ceph_osdc_get_request(req);
1725 mutex_unlock(&osdc->request_mutex);
1726 dout("sync waiting on tid %llu (last is %llu)\n",
1727 req->r_tid, last_tid);
1728 wait_for_completion(&req->r_safe_completion);
1729 mutex_lock(&osdc->request_mutex);
1730 ceph_osdc_put_request(req);
1731 }
1732 mutex_unlock(&osdc->request_mutex);
1733 dout("sync done (thru tid %llu)\n", last_tid);
1734}
3d14c5d2 1735EXPORT_SYMBOL(ceph_osdc_sync);
f24e9980
SW		 1736
1737/*
1738 * init, shutdown
1739 */
1740int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
1741{
1742 int err;
1743
1744 dout("init\n");
1745 osdc->client = client;
1746 osdc->osdmap = NULL;
1747 init_rwsem(&osdc->map_sem);
1748 init_completion(&osdc->map_waiters);
1749 osdc->last_requested_map = 0;
1750 mutex_init(&osdc->request_mutex);
f24e9980
SW		 1751 osdc->last_tid = 0;
1752 osdc->osds = RB_ROOT;
f5a2041b 1753 INIT_LIST_HEAD(&osdc->osd_lru);
f24e9980 1754 osdc->requests = RB_ROOT;
422d2cb8 1755 INIT_LIST_HEAD(&osdc->req_lru);
6f6c7006
SW		 1756 INIT_LIST_HEAD(&osdc->req_unsent);
1757 INIT_LIST_HEAD(&osdc->req_notarget);
a40c4f10 1758 INIT_LIST_HEAD(&osdc->req_linger);
f24e9980
SW		 1759 osdc->num_requests = 0;
1760 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
f5a2041b 1761 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
a40c4f10
YS		 1762 spin_lock_init(&osdc->event_lock);
1763 osdc->event_tree = RB_ROOT;
1764 osdc->event_count = 0;
f5a2041b
YS		 1765
1766 schedule_delayed_work(&osdc->osds_timeout_work,
3d14c5d2 1767 round_jiffies_relative(osdc->client->options->osd_idle_ttl * HZ));
f24e9980 1768
5f44f142 1769 err = -ENOMEM;
f24e9980
SW		 1770 osdc->req_mempool = mempool_create_kmalloc_pool(10,
1771 sizeof(struct ceph_osd_request));
1772 if (!osdc->req_mempool)
5f44f142 1773 goto out;
f24e9980 1774
d50b409f
SW		 1775 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
1776 OSD_OP_FRONT_LEN, 10, true,
4f48280e 1777 "osd_op");
f24e9980 1778 if (err < 0)
5f44f142 1779 goto out_mempool;
d50b409f 1780 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
4f48280e
SW		 1781 OSD_OPREPLY_FRONT_LEN, 10, true,
1782 "osd_op_reply");
c16e7869
SW		 1783 if (err < 0)
1784 goto out_msgpool;
a40c4f10
YS		 1785
1786 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
1787 if (IS_ERR(osdc->notify_wq)) {
1788 err = PTR_ERR(osdc->notify_wq);
1789 osdc->notify_wq = NULL;
1790 goto out_msgpool;
1791 }
f24e9980 1792 return 0;
5f44f142 1793
c16e7869
SW		 1794out_msgpool:
1795 ceph_msgpool_destroy(&osdc->msgpool_op);
5f44f142
SW		 1796out_mempool:
1797 mempool_destroy(osdc->req_mempool);
1798out:
1799 return err;
f24e9980
SW		 1800}
1801
1802void ceph_osdc_stop(struct ceph_osd_client *osdc)
1803{
a40c4f10
YS		 1804 flush_workqueue(osdc->notify_wq);
1805 destroy_workqueue(osdc->notify_wq);
f24e9980 1806 cancel_delayed_work_sync(&osdc->timeout_work);
f5a2041b 1807 cancel_delayed_work_sync(&osdc->osds_timeout_work);
f24e9980
SW		 1808 if (osdc->osdmap) {
1809 ceph_osdmap_destroy(osdc->osdmap);
1810 osdc->osdmap = NULL;
1811 }
aca420bc 1812 remove_all_osds(osdc);
f24e9980
SW		 1813 mempool_destroy(osdc->req_mempool);
1814 ceph_msgpool_destroy(&osdc->msgpool_op);
c16e7869 1815 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
f24e9980
SW		 1816}
1817
1818/*
1819 * Read some contiguous pages. If we cross a stripe boundary, shorten
1820 * *plen. Return number of bytes read, or error.
1821 */
1822int ceph_osdc_readpages(struct ceph_osd_client *osdc,
1823 struct ceph_vino vino, struct ceph_file_layout *layout,
1824 u64 off, u64 *plen,
1825 u32 truncate_seq, u64 truncate_size,
b7495fc2 1826 struct page **pages, int num_pages, int page_align)
f24e9980
SW		 1827{
1828 struct ceph_osd_request *req;
1829 int rc = 0;
1830
1831 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
1832 vino.snap, off, *plen);
1833 req = ceph_osdc_new_request(osdc, layout, vino, off, plen,
1834 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
1835 NULL, 0, truncate_seq, truncate_size, NULL,
a3bea47e 1836 false, page_align);
6816282d
SW		 1837 if (IS_ERR(req))
1838 return PTR_ERR(req);
f24e9980
SW		 1839
1840 /* it may be a short read due to an object boundary */
1841 req->r_pages = pages;
f24e9980 1842
b7495fc2
SW		 1843 dout("readpages final extent is %llu~%llu (%d pages align %d)\n",
1844 off, *plen, req->r_num_pages, page_align);
f24e9980
SW		 1845
1846 rc = ceph_osdc_start_request(osdc, req, false);
1847 if (!rc)
1848 rc = ceph_osdc_wait_request(osdc, req);
1849
1850 ceph_osdc_put_request(req);
1851 dout("readpages result %d\n", rc);
1852 return rc;
1853}
3d14c5d2 1854EXPORT_SYMBOL(ceph_osdc_readpages);
f24e9980
SW		 1855
1856/*
1857 * do a synchronous write on N pages
1858 */
1859int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
1860 struct ceph_file_layout *layout,
1861 struct ceph_snap_context *snapc,
1862 u64 off, u64 len,
1863 u32 truncate_seq, u64 truncate_size,
1864 struct timespec *mtime,
24808826 1865 struct page **pages, int num_pages)
f24e9980
SW		 1866{
1867 struct ceph_osd_request *req;
1868 int rc = 0;
b7495fc2 1869 int page_align = off & ~PAGE_MASK;
f24e9980
SW		 1870
1871 BUG_ON(vino.snap != CEPH_NOSNAP);
1872 req = ceph_osdc_new_request(osdc, layout, vino, off, &len,
1873 CEPH_OSD_OP_WRITE,
24808826 1874 CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
fbf8685f 1875 snapc, 0,
f24e9980 1876 truncate_seq, truncate_size, mtime,
a3bea47e 1877 true, page_align);
6816282d
SW		 1878 if (IS_ERR(req))
1879 return PTR_ERR(req);
f24e9980
SW		 1880
1881 /* it may be a short write due to an object boundary */
1882 req->r_pages = pages;
f24e9980
SW		 1883 dout("writepages %llu~%llu (%d pages)\n", off, len,
1884 req->r_num_pages);
1885
87f979d3 1886 rc = ceph_osdc_start_request(osdc, req, true);
f24e9980
SW		 1887 if (!rc)
1888 rc = ceph_osdc_wait_request(osdc, req);
1889
1890 ceph_osdc_put_request(req);
1891 if (rc == 0)
1892 rc = len;
1893 dout("writepages result %d\n", rc);
1894 return rc;
1895}
3d14c5d2 1896EXPORT_SYMBOL(ceph_osdc_writepages);
f24e9980
SW		 1897
1898/*
1899 * handle incoming message
1900 */
1901static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
1902{
1903 struct ceph_osd *osd = con->private;
32c895e7 1904 struct ceph_osd_client *osdc;
f24e9980
SW		 1905 int type = le16_to_cpu(msg->hdr.type);
1906
1907 if (!osd)
4a32f93d 1908 goto out;
32c895e7 1909 osdc = osd->o_osdc;
f24e9980
SW		 1910
1911 switch (type) {
1912 case CEPH_MSG_OSD_MAP:
1913 ceph_osdc_handle_map(osdc, msg);
1914 break;
1915 case CEPH_MSG_OSD_OPREPLY:
350b1c32 1916 handle_reply(osdc, msg, con);
f24e9980 1917 break;
a40c4f10
YS		 1918 case CEPH_MSG_WATCH_NOTIFY:
1919 handle_watch_notify(osdc, msg);
1920 break;
f24e9980
SW		 1921
1922 default:
1923 pr_err("received unknown message type %d %s\n", type,
1924 ceph_msg_type_name(type));
1925 }
4a32f93d 1926out:
f24e9980
SW		 1927 ceph_msg_put(msg);
1928}
1929
5b3a4db3 1930/*
21b667f6
SW		 1931 * lookup and return message for incoming reply. set up reply message
1932 * pages.
5b3a4db3
SW		 1933 */
1934static struct ceph_msg *get_reply(struct ceph_connection *con,
2450418c
YS		 1935 struct ceph_msg_header *hdr,
1936 int *skip)
f24e9980
SW		 1937{
1938 struct ceph_osd *osd = con->private;
1939 struct ceph_osd_client *osdc = osd->o_osdc;
2450418c 1940 struct ceph_msg *m;
0547a9b3 1941 struct ceph_osd_request *req;
5b3a4db3
SW		 1942 int front = le32_to_cpu(hdr->front_len);
1943 int data_len = le32_to_cpu(hdr->data_len);
0547a9b3 1944 u64 tid;
f24e9980 1945
0547a9b3
YS		 1946 tid = le64_to_cpu(hdr->tid);
1947 mutex_lock(&osdc->request_mutex);
1948 req = __lookup_request(osdc, tid);
1949 if (!req) {
1950 *skip = 1;
1951 m = NULL;
756a16a5
SW		 1952 dout("get_reply unknown tid %llu from osd%d\n", tid,
1953 osd->o_osd);
0547a9b3
YS		 1954 goto out;
1955 }
c16e7869
SW		 1956
1957 if (req->r_con_filling_msg) {
8921d114 1958 dout("%s revoking msg %p from old con %p\n", __func__,
c16e7869 1959 req->r_reply, req->r_con_filling_msg);
8921d114 1960 ceph_msg_revoke_incoming(req->r_reply);
0d47766f 1961 req->r_con_filling_msg->ops->put(req->r_con_filling_msg);
6f46cb29 1962 req->r_con_filling_msg = NULL;
0547a9b3
YS		 1963 }
1964
c16e7869
SW		 1965 if (front > req->r_reply->front.iov_len) {
1966 pr_warning("get_reply front %d > preallocated %d\n",
1967 front, (int)req->r_reply->front.iov_len);
b61c2763 1968 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front, GFP_NOFS, false);
a79832f2 1969 if (!m)
c16e7869
SW		 1970 goto out;
1971 ceph_msg_put(req->r_reply);
1972 req->r_reply = m;
1973 }
1974 m = ceph_msg_get(req->r_reply);
1975
0547a9b3 1976 if (data_len > 0) {
b7495fc2 1977 int want = calc_pages_for(req->r_page_alignment, data_len);
21b667f6 1978
9cbb1d72 1979 if (req->r_pages && unlikely(req->r_num_pages < want)) {
9bb0ce2b
SW		 1980 pr_warning("tid %lld reply has %d bytes %d pages, we"
1981 " had only %d pages ready\n", tid, data_len,
1982 want, req->r_num_pages);
0547a9b3
YS		 1983 *skip = 1;
1984 ceph_msg_put(m);
a79832f2 1985 m = NULL;
21b667f6 1986 goto out;
0547a9b3 1987 }
21b667f6
SW		 1988 m->pages = req->r_pages;
1989 m->nr_pages = req->r_num_pages;
c5c6b19d 1990 m->page_alignment = req->r_page_alignment;
68b4476b
YS		 1991#ifdef CONFIG_BLOCK
1992 m->bio = req->r_bio;
1993#endif
0547a9b3 1994 }
5b3a4db3 1995 *skip = 0;
0d47766f 1996 req->r_con_filling_msg = con->ops->get(con);
c16e7869 1997 dout("get_reply tid %lld %p\n", tid, m);
0547a9b3
YS		 1998
1999out:
2000 mutex_unlock(&osdc->request_mutex);
2450418c 2001 return m;
5b3a4db3
SW		 2002
2003}
2004
2005static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2006 struct ceph_msg_header *hdr,
2007 int *skip)
2008{
2009 struct ceph_osd *osd = con->private;
2010 int type = le16_to_cpu(hdr->type);
2011 int front = le32_to_cpu(hdr->front_len);
2012
1c20f2d2 2013 *skip = 0;
5b3a4db3
SW		 2014 switch (type) {
2015 case CEPH_MSG_OSD_MAP:
a40c4f10 2016 case CEPH_MSG_WATCH_NOTIFY:
b61c2763 2017 return ceph_msg_new(type, front, GFP_NOFS, false);
5b3a4db3
SW		 2018 case CEPH_MSG_OSD_OPREPLY:
2019 return get_reply(con, hdr, skip);
2020 default:
2021 pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
2022 osd->o_osd);
2023 *skip = 1;
2024 return NULL;
2025 }
f24e9980
SW		 2026}
2027
2028/*
2029 * Wrappers to refcount containing ceph_osd struct
2030 */
2031static struct ceph_connection *get_osd_con(struct ceph_connection *con)
2032{
2033 struct ceph_osd *osd = con->private;
2034 if (get_osd(osd))
2035 return con;
2036 return NULL;
2037}
2038
2039static void put_osd_con(struct ceph_connection *con)
2040{
2041 struct ceph_osd *osd = con->private;
2042 put_osd(osd);
2043}
2044
4e7a5dcd
SW		 2045/*
2046 * authentication
2047 */
a3530df3
AE		 2048/*
2049 * Note: returned pointer is the address of a structure that's
2050 * managed separately. Caller must *not* attempt to free it.
2051 */
2052static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
8f43fb53 2053 int *proto, int force_new)
4e7a5dcd
SW		 2054{
2055 struct ceph_osd *o = con->private;
2056 struct ceph_osd_client *osdc = o->o_osdc;
2057 struct ceph_auth_client *ac = osdc->client->monc.auth;
74f1869f 2058 struct ceph_auth_handshake *auth = &o->o_auth;
4e7a5dcd 2059
74f1869f 2060 if (force_new && auth->authorizer) {
a255651d
AE		 2061 if (ac->ops && ac->ops->destroy_authorizer)
2062 ac->ops->destroy_authorizer(ac, auth->authorizer);
74f1869f
AE		 2063 auth->authorizer = NULL;
2064 }
a255651d 2065 if (!auth->authorizer && ac->ops && ac->ops->create_authorizer) {
a3530df3
AE		 2066 int ret = ac->ops->create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2067 auth);
4e7a5dcd 2068 if (ret)
a3530df3 2069 return ERR_PTR(ret);
4e7a5dcd 2070 }
4e7a5dcd 2071 *proto = ac->protocol;
74f1869f 2072
a3530df3 2073 return auth;
4e7a5dcd
SW		 2074}
2075
2076
2077static int verify_authorizer_reply(struct ceph_connection *con, int len)
2078{
2079 struct ceph_osd *o = con->private;
2080 struct ceph_osd_client *osdc = o->o_osdc;
2081 struct ceph_auth_client *ac = osdc->client->monc.auth;
2082
a255651d
AE		 2083 /*
2084 * XXX If ac->ops or ac->ops->verify_authorizer_reply is null,
2085 * XXX which do we do: succeed or fail?
2086 */
6c4a1915 2087 return ac->ops->verify_authorizer_reply(ac, o->o_auth.authorizer, len);
4e7a5dcd
SW		 2088}
2089
9bd2e6f8
SW		 2090static int invalidate_authorizer(struct ceph_connection *con)
2091{
2092 struct ceph_osd *o = con->private;
2093 struct ceph_osd_client *osdc = o->o_osdc;
2094 struct ceph_auth_client *ac = osdc->client->monc.auth;
2095
a255651d 2096 if (ac->ops && ac->ops->invalidate_authorizer)
9bd2e6f8
SW		 2097 ac->ops->invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
2098
2099 return ceph_monc_validate_auth(&osdc->client->monc);
2100}
4e7a5dcd 2101
9e32789f 2102static const struct ceph_connection_operations osd_con_ops = {
f24e9980
SW		 2103 .get = get_osd_con,
2104 .put = put_osd_con,
2105 .dispatch = dispatch,
4e7a5dcd
SW		 2106 .get_authorizer = get_authorizer,
2107 .verify_authorizer_reply = verify_authorizer_reply,
9bd2e6f8 2108 .invalidate_authorizer = invalidate_authorizer,
f24e9980 2109 .alloc_msg = alloc_msg,
81b024e7 2110 .fault = osd_reset,
f24e9980 2111};
Linux 6.x block layer and io_uring tree(s)