projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
libceph: always preallocate mon connection
[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
6f6c7006
SW		 26static void send_queued(struct ceph_osd_client *osdc);
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);
32static int __send_request(struct ceph_osd_client *osdc,
33 struct ceph_osd_request *req);
f24e9980 34
68b4476b
YS		 35static int op_needs_trail(int op)
36{
37 switch (op) {
38 case CEPH_OSD_OP_GETXATTR:
39 case CEPH_OSD_OP_SETXATTR:
40 case CEPH_OSD_OP_CMPXATTR:
41 case CEPH_OSD_OP_CALL:
a40c4f10 42 case CEPH_OSD_OP_NOTIFY:
68b4476b
YS		 43 return 1;
44 default:
45 return 0;
46 }
47}
48
49static int op_has_extent(int op)
50{
51 return (op == CEPH_OSD_OP_READ ||
52 op == CEPH_OSD_OP_WRITE);
53}
54
3499e8a5
YS		 55void ceph_calc_raw_layout(struct ceph_osd_client *osdc,
56 struct ceph_file_layout *layout,
57 u64 snapid,
68b4476b
YS		 58 u64 off, u64 *plen, u64 *bno,
59 struct ceph_osd_request *req,
60 struct ceph_osd_req_op *op)
3499e8a5
YS		 61{
62 struct ceph_osd_request_head *reqhead = req->r_request->front.iov_base;
68b4476b 63 u64 orig_len = *plen;
3499e8a5
YS		 64 u64 objoff, objlen; /* extent in object */
65
66 reqhead->snapid = cpu_to_le64(snapid);
67
68 /* object extent? */
68b4476b 69 ceph_calc_file_object_mapping(layout, off, plen, bno,
3499e8a5 70 &objoff, &objlen);
68b4476b 71 if (*plen < orig_len)
3499e8a5 72 dout(" skipping last %llu, final file extent %llu~%llu\n",
68b4476b 73 orig_len - *plen, off, *plen);
3499e8a5 74
68b4476b
YS		 75 if (op_has_extent(op->op)) {
76 op->extent.offset = objoff;
77 op->extent.length = objlen;
78 }
79 req->r_num_pages = calc_pages_for(off, *plen);
b7495fc2 80 req->r_page_alignment = off & ~PAGE_MASK;
3d14c5d2
YS		 81 if (op->op == CEPH_OSD_OP_WRITE)
82 op->payload_len = *plen;
3499e8a5
YS		 83
84 dout("calc_layout bno=%llx %llu~%llu (%d pages)\n",
85 *bno, objoff, objlen, req->r_num_pages);
86
87}
3d14c5d2 88EXPORT_SYMBOL(ceph_calc_raw_layout);
3499e8a5 89
f24e9980
SW		 90/*
91 * Implement client access to distributed object storage cluster.
92 *
93 * All data objects are stored within a cluster/cloud of OSDs, or
94 * "object storage devices." (Note that Ceph OSDs have _nothing_ to
95 * do with the T10 OSD extensions to SCSI.) Ceph OSDs are simply
96 * remote daemons serving up and coordinating consistent and safe
97 * access to storage.
98 *
99 * Cluster membership and the mapping of data objects onto storage devices
100 * are described by the osd map.
101 *
102 * We keep track of pending OSD requests (read, write), resubmit
103 * requests to different OSDs when the cluster topology/data layout
104 * change, or retry the affected requests when the communications
105 * channel with an OSD is reset.
106 */
107
108/*
109 * calculate the mapping of a file extent onto an object, and fill out the
110 * request accordingly. shorten extent as necessary if it crosses an
111 * object boundary.
112 *
113 * fill osd op in request message.
114 */
115static void calc_layout(struct ceph_osd_client *osdc,
3499e8a5
YS		 116 struct ceph_vino vino,
117 struct ceph_file_layout *layout,
f24e9980 118 u64 off, u64 *plen,
68b4476b
YS		 119 struct ceph_osd_request *req,
120 struct ceph_osd_req_op *op)
f24e9980 121{
f24e9980
SW		 122 u64 bno;
123
68b4476b
YS		 124 ceph_calc_raw_layout(osdc, layout, vino.snap, off,
125 plen, &bno, req, op);
f24e9980 126
2dab036b 127 snprintf(req->r_oid, sizeof(req->r_oid), "%llx.%08llx", vino.ino, bno);
f24e9980 128 req->r_oid_len = strlen(req->r_oid);
f24e9980
SW		 129}
130
f24e9980
SW		 131/*
132 * requests
133 */
415e49a9 134void ceph_osdc_release_request(struct kref *kref)
f24e9980 135{
415e49a9
SW		 136 struct ceph_osd_request *req = container_of(kref,
137 struct ceph_osd_request,
138 r_kref);
139
140 if (req->r_request)
141 ceph_msg_put(req->r_request);
142 if (req->r_reply)
143 ceph_msg_put(req->r_reply);
0d59ab81 144 if (req->r_con_filling_msg) {
350b1c32 145 dout("release_request revoking pages %p from con %p\n",
0d59ab81
YS		 146 req->r_pages, req->r_con_filling_msg);
147 ceph_con_revoke_message(req->r_con_filling_msg,
148 req->r_reply);
149 ceph_con_put(req->r_con_filling_msg);
350b1c32 150 }
415e49a9
SW		 151 if (req->r_own_pages)
152 ceph_release_page_vector(req->r_pages,
153 req->r_num_pages);
68b4476b
YS		 154#ifdef CONFIG_BLOCK
155 if (req->r_bio)
156 bio_put(req->r_bio);
157#endif
415e49a9 158 ceph_put_snap_context(req->r_snapc);
68b4476b
YS		 159 if (req->r_trail) {
160 ceph_pagelist_release(req->r_trail);
161 kfree(req->r_trail);
162 }
415e49a9
SW		 163 if (req->r_mempool)
164 mempool_free(req, req->r_osdc->req_mempool);
165 else
166 kfree(req);
f24e9980 167}
3d14c5d2 168EXPORT_SYMBOL(ceph_osdc_release_request);
68b4476b
YS		 169
170static int get_num_ops(struct ceph_osd_req_op *ops, int *needs_trail)
171{
172 int i = 0;
173
174 if (needs_trail)
175 *needs_trail = 0;
176 while (ops[i].op) {
177 if (needs_trail && op_needs_trail(ops[i].op))
178 *needs_trail = 1;
179 i++;
180 }
181
182 return i;
183}
184
3499e8a5
YS		 185struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
186 int flags,
f24e9980 187 struct ceph_snap_context *snapc,
68b4476b 188 struct ceph_osd_req_op *ops,
3499e8a5
YS		 189 bool use_mempool,
190 gfp_t gfp_flags,
68b4476b
YS		 191 struct page **pages,
192 struct bio *bio)
f24e9980
SW		 193{
194 struct ceph_osd_request *req;
195 struct ceph_msg *msg;
68b4476b
YS		 196 int needs_trail;
197 int num_op = get_num_ops(ops, &needs_trail);
198 size_t msg_size = sizeof(struct ceph_osd_request_head);
3499e8a5 199
68b4476b 200 msg_size += num_op*sizeof(struct ceph_osd_op);
f24e9980
SW		 201
202 if (use_mempool) {
3499e8a5 203 req = mempool_alloc(osdc->req_mempool, gfp_flags);
f24e9980
SW		 204 memset(req, 0, sizeof(*req));
205 } else {
3499e8a5 206 req = kzalloc(sizeof(*req), gfp_flags);
f24e9980
SW		 207 }
208 if (req == NULL)
a79832f2 209 return NULL;
f24e9980 210
f24e9980
SW		 211 req->r_osdc = osdc;
212 req->r_mempool = use_mempool;
68b4476b 213
415e49a9 214 kref_init(&req->r_kref);
f24e9980
SW		 215 init_completion(&req->r_completion);
216 init_completion(&req->r_safe_completion);
217 INIT_LIST_HEAD(&req->r_unsafe_item);
a40c4f10
YS		 218 INIT_LIST_HEAD(&req->r_linger_item);
219 INIT_LIST_HEAD(&req->r_linger_osd);
935b639a 220 INIT_LIST_HEAD(&req->r_req_lru_item);
f24e9980
SW		 221 req->r_flags = flags;
222
223 WARN_ON((flags & (CEPH_OSD_FLAG_READ|CEPH_OSD_FLAG_WRITE)) == 0);
224
c16e7869
SW		 225 /* create reply message */
226 if (use_mempool)
227 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0);
228 else
229 msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY,
3499e8a5 230 OSD_OPREPLY_FRONT_LEN, gfp_flags);
a79832f2 231 if (!msg) {
c16e7869 232 ceph_osdc_put_request(req);
a79832f2 233 return NULL;
c16e7869
SW		 234 }
235 req->r_reply = msg;
236
68b4476b
YS		 237 /* allocate space for the trailing data */
238 if (needs_trail) {
239 req->r_trail = kmalloc(sizeof(struct ceph_pagelist), gfp_flags);
240 if (!req->r_trail) {
241 ceph_osdc_put_request(req);
242 return NULL;
243 }
244 ceph_pagelist_init(req->r_trail);
245 }
c16e7869 246 /* create request message; allow space for oid */
f24e9980
SW		 247 msg_size += 40;
248 if (snapc)
249 msg_size += sizeof(u64) * snapc->num_snaps;
250 if (use_mempool)
8f3bc053 251 msg = ceph_msgpool_get(&osdc->msgpool_op, 0);
f24e9980 252 else
3499e8a5 253 msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp_flags);
a79832f2 254 if (!msg) {
f24e9980 255 ceph_osdc_put_request(req);
a79832f2 256 return NULL;
f24e9980 257 }
68b4476b 258
f24e9980
SW		 259 msg->hdr.type = cpu_to_le16(CEPH_MSG_OSD_OP);
260 memset(msg->front.iov_base, 0, msg->front.iov_len);
3499e8a5
YS		 261
262 req->r_request = msg;
263 req->r_pages = pages;
68b4476b
YS		 264#ifdef CONFIG_BLOCK
265 if (bio) {
266 req->r_bio = bio;
267 bio_get(req->r_bio);
268 }
269#endif
3499e8a5
YS		 270
271 return req;
272}
3d14c5d2 273EXPORT_SYMBOL(ceph_osdc_alloc_request);
3499e8a5 274
68b4476b
YS		 275static void osd_req_encode_op(struct ceph_osd_request *req,
276 struct ceph_osd_op *dst,
277 struct ceph_osd_req_op *src)
278{
279 dst->op = cpu_to_le16(src->op);
280
281 switch (dst->op) {
282 case CEPH_OSD_OP_READ:
283 case CEPH_OSD_OP_WRITE:
284 dst->extent.offset =
285 cpu_to_le64(src->extent.offset);
286 dst->extent.length =
287 cpu_to_le64(src->extent.length);
288 dst->extent.truncate_size =
289 cpu_to_le64(src->extent.truncate_size);
290 dst->extent.truncate_seq =
291 cpu_to_le32(src->extent.truncate_seq);
292 break;
293
294 case CEPH_OSD_OP_GETXATTR:
295 case CEPH_OSD_OP_SETXATTR:
296 case CEPH_OSD_OP_CMPXATTR:
297 BUG_ON(!req->r_trail);
298
299 dst->xattr.name_len = cpu_to_le32(src->xattr.name_len);
300 dst->xattr.value_len = cpu_to_le32(src->xattr.value_len);
301 dst->xattr.cmp_op = src->xattr.cmp_op;
302 dst->xattr.cmp_mode = src->xattr.cmp_mode;
303 ceph_pagelist_append(req->r_trail, src->xattr.name,
304 src->xattr.name_len);
305 ceph_pagelist_append(req->r_trail, src->xattr.val,
306 src->xattr.value_len);
307 break;
ae1533b6
YS		 308 case CEPH_OSD_OP_CALL:
309 BUG_ON(!req->r_trail);
310
311 dst->cls.class_len = src->cls.class_len;
312 dst->cls.method_len = src->cls.method_len;
313 dst->cls.indata_len = cpu_to_le32(src->cls.indata_len);
314
315 ceph_pagelist_append(req->r_trail, src->cls.class_name,
316 src->cls.class_len);
317 ceph_pagelist_append(req->r_trail, src->cls.method_name,
318 src->cls.method_len);
319 ceph_pagelist_append(req->r_trail, src->cls.indata,
320 src->cls.indata_len);
321 break;
322 case CEPH_OSD_OP_ROLLBACK:
323 dst->snap.snapid = cpu_to_le64(src->snap.snapid);
324 break;
68b4476b
YS		 325 case CEPH_OSD_OP_STARTSYNC:
326 break;
a40c4f10
YS		 327 case CEPH_OSD_OP_NOTIFY:
328 {
329 __le32 prot_ver = cpu_to_le32(src->watch.prot_ver);
330 __le32 timeout = cpu_to_le32(src->watch.timeout);
331
332 BUG_ON(!req->r_trail);
333
334 ceph_pagelist_append(req->r_trail,
335 &prot_ver, sizeof(prot_ver));
336 ceph_pagelist_append(req->r_trail,
337 &timeout, sizeof(timeout));
338 }
339 case CEPH_OSD_OP_NOTIFY_ACK:
340 case CEPH_OSD_OP_WATCH:
341 dst->watch.cookie = cpu_to_le64(src->watch.cookie);
342 dst->watch.ver = cpu_to_le64(src->watch.ver);
343 dst->watch.flag = src->watch.flag;
344 break;
68b4476b
YS		 345 default:
346 pr_err("unrecognized osd opcode %d\n", dst->op);
347 WARN_ON(1);
348 break;
349 }
350 dst->payload_len = cpu_to_le32(src->payload_len);
351}
352
3499e8a5
YS		 353/*
354 * build new request AND message
355 *
356 */
357void ceph_osdc_build_request(struct ceph_osd_request *req,
68b4476b
YS		 358 u64 off, u64 *plen,
359 struct ceph_osd_req_op *src_ops,
360 struct ceph_snap_context *snapc,
361 struct timespec *mtime,
362 const char *oid,
363 int oid_len)
3499e8a5
YS		 364{
365 struct ceph_msg *msg = req->r_request;
366 struct ceph_osd_request_head *head;
68b4476b 367 struct ceph_osd_req_op *src_op;
3499e8a5
YS		 368 struct ceph_osd_op *op;
369 void *p;
68b4476b 370 int num_op = get_num_ops(src_ops, NULL);
3499e8a5 371 size_t msg_size = sizeof(*head) + num_op*sizeof(*op);
3499e8a5 372 int flags = req->r_flags;
68b4476b
YS		 373 u64 data_len = 0;
374 int i;
3499e8a5 375
f24e9980
SW		 376 head = msg->front.iov_base;
377 op = (void *)(head + 1);
378 p = (void *)(op + num_op);
379
f24e9980
SW		 380 req->r_snapc = ceph_get_snap_context(snapc);
381
382 head->client_inc = cpu_to_le32(1); /* always, for now. */
383 head->flags = cpu_to_le32(flags);
384 if (flags & CEPH_OSD_FLAG_WRITE)
385 ceph_encode_timespec(&head->mtime, mtime);
386 head->num_ops = cpu_to_le16(num_op);
f24e9980 387
f24e9980
SW		 388
389 /* fill in oid */
3499e8a5
YS		 390 head->object_len = cpu_to_le32(oid_len);
391 memcpy(p, oid, oid_len);
392 p += oid_len;
f24e9980 393
68b4476b
YS		 394 src_op = src_ops;
395 while (src_op->op) {
396 osd_req_encode_op(req, op, src_op);
397 src_op++;
f24e9980 398 op++;
f24e9980 399 }
68b4476b
YS		 400
401 if (req->r_trail)
402 data_len += req->r_trail->length;
403
f24e9980
SW		 404 if (snapc) {
405 head->snap_seq = cpu_to_le64(snapc->seq);
406 head->num_snaps = cpu_to_le32(snapc->num_snaps);
407 for (i = 0; i < snapc->num_snaps; i++) {
408 put_unaligned_le64(snapc->snaps[i], p);
409 p += sizeof(u64);
410 }
411 }
412
68b4476b
YS		 413 if (flags & CEPH_OSD_FLAG_WRITE) {
414 req->r_request->hdr.data_off = cpu_to_le16(off);
415 req->r_request->hdr.data_len = cpu_to_le32(*plen + data_len);
416 } else if (data_len) {
417 req->r_request->hdr.data_off = 0;
418 req->r_request->hdr.data_len = cpu_to_le32(data_len);
419 }
420
c5c6b19d
SW		 421 req->r_request->page_alignment = req->r_page_alignment;
422
f24e9980 423 BUG_ON(p > msg->front.iov_base + msg->front.iov_len);
6f863e71
SW		 424 msg_size = p - msg->front.iov_base;
425 msg->front.iov_len = msg_size;
426 msg->hdr.front_len = cpu_to_le32(msg_size);
3499e8a5
YS		 427 return;
428}
3d14c5d2 429EXPORT_SYMBOL(ceph_osdc_build_request);
3499e8a5
YS		 430
431/*
432 * build new request AND message, calculate layout, and adjust file
433 * extent as needed.
434 *
435 * if the file was recently truncated, we include information about its
436 * old and new size so that the object can be updated appropriately. (we
437 * avoid synchronously deleting truncated objects because it's slow.)
438 *
439 * if @do_sync, include a 'startsync' command so that the osd will flush
440 * data quickly.
441 */
442struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
443 struct ceph_file_layout *layout,
444 struct ceph_vino vino,
445 u64 off, u64 *plen,
446 int opcode, int flags,
447 struct ceph_snap_context *snapc,
448 int do_sync,
449 u32 truncate_seq,
450 u64 truncate_size,
451 struct timespec *mtime,
b7495fc2
SW		 452 bool use_mempool, int num_reply,
453 int page_align)
3499e8a5 454{
68b4476b
YS		 455 struct ceph_osd_req_op ops[3];
456 struct ceph_osd_request *req;
457
458 ops[0].op = opcode;
459 ops[0].extent.truncate_seq = truncate_seq;
460 ops[0].extent.truncate_size = truncate_size;
461 ops[0].payload_len = 0;
462
463 if (do_sync) {
464 ops[1].op = CEPH_OSD_OP_STARTSYNC;
465 ops[1].payload_len = 0;
466 ops[2].op = 0;
467 } else
468 ops[1].op = 0;
469
470 req = ceph_osdc_alloc_request(osdc, flags,
471 snapc, ops,
3499e8a5 472 use_mempool,
68b4476b 473 GFP_NOFS, NULL, NULL);
4ad12621
SW		 474 if (!req)
475 return NULL;
3499e8a5
YS		 476
477 /* calculate max write size */
68b4476b 478 calc_layout(osdc, vino, layout, off, plen, req, ops);
3499e8a5
YS		 479 req->r_file_layout = *layout; /* keep a copy */
480
9bb0ce2b
SW		 481 /* in case it differs from natural (file) alignment that
482 calc_layout filled in for us */
483 req->r_num_pages = calc_pages_for(page_align, *plen);
b7495fc2
SW		 484 req->r_page_alignment = page_align;
485
68b4476b
YS		 486 ceph_osdc_build_request(req, off, plen, ops,
487 snapc,
3499e8a5
YS		 488 mtime,
489 req->r_oid, req->r_oid_len);
490
f24e9980
SW		 491 return req;
492}
3d14c5d2 493EXPORT_SYMBOL(ceph_osdc_new_request);
f24e9980
SW		 494
495/*
496 * We keep osd requests in an rbtree, sorted by ->r_tid.
497 */
498static void __insert_request(struct ceph_osd_client *osdc,
499 struct ceph_osd_request *new)
500{
501 struct rb_node **p = &osdc->requests.rb_node;
502 struct rb_node *parent = NULL;
503 struct ceph_osd_request *req = NULL;
504
505 while (*p) {
506 parent = *p;
507 req = rb_entry(parent, struct ceph_osd_request, r_node);
508 if (new->r_tid < req->r_tid)
509 p = &(*p)->rb_left;
510 else if (new->r_tid > req->r_tid)
511 p = &(*p)->rb_right;
512 else
513 BUG();
514 }
515
516 rb_link_node(&new->r_node, parent, p);
517 rb_insert_color(&new->r_node, &osdc->requests);
518}
519
520static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc,
521 u64 tid)
522{
523 struct ceph_osd_request *req;
524 struct rb_node *n = osdc->requests.rb_node;
525
526 while (n) {
527 req = rb_entry(n, struct ceph_osd_request, r_node);
528 if (tid < req->r_tid)
529 n = n->rb_left;
530 else if (tid > req->r_tid)
531 n = n->rb_right;
532 else
533 return req;
534 }
535 return NULL;
536}
537
538static struct ceph_osd_request *
539__lookup_request_ge(struct ceph_osd_client *osdc,
540 u64 tid)
541{
542 struct ceph_osd_request *req;
543 struct rb_node *n = osdc->requests.rb_node;
544
545 while (n) {
546 req = rb_entry(n, struct ceph_osd_request, r_node);
547 if (tid < req->r_tid) {
548 if (!n->rb_left)
549 return req;
550 n = n->rb_left;
551 } else if (tid > req->r_tid) {
552 n = n->rb_right;
553 } else {
554 return req;
555 }
556 }
557 return NULL;
558}
559
6f6c7006
SW		 560/*
561 * Resubmit requests pending on the given osd.
562 */
563static void __kick_osd_requests(struct ceph_osd_client *osdc,
564 struct ceph_osd *osd)
565{
a40c4f10 566 struct ceph_osd_request *req, *nreq;
6f6c7006
SW		 567 int err;
568
569 dout("__kick_osd_requests osd%d\n", osd->o_osd);
570 err = __reset_osd(osdc, osd);
571 if (err == -EAGAIN)
572 return;
573
574 list_for_each_entry(req, &osd->o_requests, r_osd_item) {
575 list_move(&req->r_req_lru_item, &osdc->req_unsent);
576 dout("requeued %p tid %llu osd%d\n", req, req->r_tid,
577 osd->o_osd);
a40c4f10
YS		 578 if (!req->r_linger)
579 req->r_flags |= CEPH_OSD_FLAG_RETRY;
580 }
581
582 list_for_each_entry_safe(req, nreq, &osd->o_linger_requests,
583 r_linger_osd) {
77f38e0e
SW		 584 /*
585 * reregister request prior to unregistering linger so
586 * that r_osd is preserved.
587 */
588 BUG_ON(!list_empty(&req->r_req_lru_item));
a40c4f10 589 __register_request(osdc, req);
77f38e0e
SW		 590 list_add(&req->r_req_lru_item, &osdc->req_unsent);
591 list_add(&req->r_osd_item, &req->r_osd->o_requests);
592 __unregister_linger_request(osdc, req);
a40c4f10
YS		 593 dout("requeued lingering %p tid %llu osd%d\n", req, req->r_tid,
594 osd->o_osd);
6f6c7006
SW		 595 }
596}
597
598static void kick_osd_requests(struct ceph_osd_client *osdc,
599 struct ceph_osd *kickosd)
600{
601 mutex_lock(&osdc->request_mutex);
602 __kick_osd_requests(osdc, kickosd);
603 mutex_unlock(&osdc->request_mutex);
604}
f24e9980
SW		 605
606/*
81b024e7 607 * If the osd connection drops, we need to resubmit all requests.
f24e9980
SW		 608 */
609static void osd_reset(struct ceph_connection *con)
610{
611 struct ceph_osd *osd = con->private;
612 struct ceph_osd_client *osdc;
613
614 if (!osd)
615 return;
616 dout("osd_reset osd%d\n", osd->o_osd);
617 osdc = osd->o_osdc;
f24e9980 618 down_read(&osdc->map_sem);
6f6c7006
SW		 619 kick_osd_requests(osdc, osd);
620 send_queued(osdc);
f24e9980
SW		 621 up_read(&osdc->map_sem);
622}
623
624/*
625 * Track open sessions with osds.
626 */
627static struct ceph_osd *create_osd(struct ceph_osd_client *osdc)
628{
629 struct ceph_osd *osd;
630
631 osd = kzalloc(sizeof(*osd), GFP_NOFS);
632 if (!osd)
633 return NULL;
634
635 atomic_set(&osd->o_ref, 1);
636 osd->o_osdc = osdc;
637 INIT_LIST_HEAD(&osd->o_requests);
a40c4f10 638 INIT_LIST_HEAD(&osd->o_linger_requests);
f5a2041b 639 INIT_LIST_HEAD(&osd->o_osd_lru);
f24e9980
SW		 640 osd->o_incarnation = 1;
641
642 ceph_con_init(osdc->client->msgr, &osd->o_con);
643 osd->o_con.private = osd;
644 osd->o_con.ops = &osd_con_ops;
645 osd->o_con.peer_name.type = CEPH_ENTITY_TYPE_OSD;
4e7a5dcd 646
422d2cb8 647 INIT_LIST_HEAD(&osd->o_keepalive_item);
f24e9980
SW		 648 return osd;
649}
650
651static struct ceph_osd *get_osd(struct ceph_osd *osd)
652{
653 if (atomic_inc_not_zero(&osd->o_ref)) {
654 dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
655 atomic_read(&osd->o_ref));
656 return osd;
657 } else {
658 dout("get_osd %p FAIL\n", osd);
659 return NULL;
660 }
661}
662
663static void put_osd(struct ceph_osd *osd)
664{
665 dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
666 atomic_read(&osd->o_ref) - 1);
79494d1b
SW		 667 if (atomic_dec_and_test(&osd->o_ref)) {
668 struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth;
669
670 if (osd->o_authorizer)
671 ac->ops->destroy_authorizer(ac, osd->o_authorizer);
f24e9980 672 kfree(osd);
79494d1b 673 }
f24e9980
SW		 674}
675
676/*
677 * remove an osd from our map
678 */
f5a2041b 679static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
f24e9980 680{
f5a2041b 681 dout("__remove_osd %p\n", osd);
f24e9980
SW		 682 BUG_ON(!list_empty(&osd->o_requests));
683 rb_erase(&osd->o_node, &osdc->osds);
f5a2041b 684 list_del_init(&osd->o_osd_lru);
f24e9980
SW		 685 ceph_con_close(&osd->o_con);
686 put_osd(osd);
687}
688
aca420bc
SW		 689static void remove_all_osds(struct ceph_osd_client *osdc)
690{
691 dout("__remove_old_osds %p\n", osdc);
692 mutex_lock(&osdc->request_mutex);
693 while (!RB_EMPTY_ROOT(&osdc->osds)) {
694 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
695 struct ceph_osd, o_node);
696 __remove_osd(osdc, osd);
697 }
698 mutex_unlock(&osdc->request_mutex);
699}
700
f5a2041b
YS		 701static void __move_osd_to_lru(struct ceph_osd_client *osdc,
702 struct ceph_osd *osd)
703{
704 dout("__move_osd_to_lru %p\n", osd);
705 BUG_ON(!list_empty(&osd->o_osd_lru));
706 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
3d14c5d2 707 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl * HZ;
f5a2041b
YS		 708}
709
710static void __remove_osd_from_lru(struct ceph_osd *osd)
711{
712 dout("__remove_osd_from_lru %p\n", osd);
713 if (!list_empty(&osd->o_osd_lru))
714 list_del_init(&osd->o_osd_lru);
715}
716
aca420bc 717static void remove_old_osds(struct ceph_osd_client *osdc)
f5a2041b
YS		 718{
719 struct ceph_osd *osd, *nosd;
720
721 dout("__remove_old_osds %p\n", osdc);
722 mutex_lock(&osdc->request_mutex);
723 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
aca420bc 724 if (time_before(jiffies, osd->lru_ttl))
f5a2041b
YS		 725 break;
726 __remove_osd(osdc, osd);
727 }
728 mutex_unlock(&osdc->request_mutex);
729}
730
f24e9980
SW		 731/*
732 * reset osd connect
733 */
f5a2041b 734static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
f24e9980 735{
87b315a5 736 struct ceph_osd_request *req;
f24e9980
SW		 737 int ret = 0;
738
f5a2041b 739 dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
a40c4f10
YS		 740 if (list_empty(&osd->o_requests) &&
741 list_empty(&osd->o_linger_requests)) {
f5a2041b 742 __remove_osd(osdc, osd);
87b315a5
SW		 743 } else if (memcmp(&osdc->osdmap->osd_addr[osd->o_osd],
744 &osd->o_con.peer_addr,
745 sizeof(osd->o_con.peer_addr)) == 0 &&
746 !ceph_con_opened(&osd->o_con)) {
747 dout(" osd addr hasn't changed and connection never opened,"
748 " letting msgr retry");
749 /* touch each r_stamp for handle_timeout()'s benfit */
750 list_for_each_entry(req, &osd->o_requests, r_osd_item)
751 req->r_stamp = jiffies;
752 ret = -EAGAIN;
f24e9980
SW		 753 } else {
754 ceph_con_close(&osd->o_con);
755 ceph_con_open(&osd->o_con, &osdc->osdmap->osd_addr[osd->o_osd]);
756 osd->o_incarnation++;
757 }
758 return ret;
759}
760
761static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new)
762{
763 struct rb_node **p = &osdc->osds.rb_node;
764 struct rb_node *parent = NULL;
765 struct ceph_osd *osd = NULL;
766
aca420bc 767 dout("__insert_osd %p osd%d\n", new, new->o_osd);
f24e9980
SW		 768 while (*p) {
769 parent = *p;
770 osd = rb_entry(parent, struct ceph_osd, o_node);
771 if (new->o_osd < osd->o_osd)
772 p = &(*p)->rb_left;
773 else if (new->o_osd > osd->o_osd)
774 p = &(*p)->rb_right;
775 else
776 BUG();
777 }
778
779 rb_link_node(&new->o_node, parent, p);
780 rb_insert_color(&new->o_node, &osdc->osds);
781}
782
783static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o)
784{
785 struct ceph_osd *osd;
786 struct rb_node *n = osdc->osds.rb_node;
787
788 while (n) {
789 osd = rb_entry(n, struct ceph_osd, o_node);
790 if (o < osd->o_osd)
791 n = n->rb_left;
792 else if (o > osd->o_osd)
793 n = n->rb_right;
794 else
795 return osd;
796 }
797 return NULL;
798}
799
422d2cb8
YS		 800static void __schedule_osd_timeout(struct ceph_osd_client *osdc)
801{
802 schedule_delayed_work(&osdc->timeout_work,
3d14c5d2 803 osdc->client->options->osd_keepalive_timeout * HZ);
422d2cb8
YS		 804}
805
806static void __cancel_osd_timeout(struct ceph_osd_client *osdc)
807{
808 cancel_delayed_work(&osdc->timeout_work);
809}
f24e9980
SW		 810
811/*
812 * Register request, assign tid. If this is the first request, set up
813 * the timeout event.
814 */
a40c4f10
YS		 815static void __register_request(struct ceph_osd_client *osdc,
816 struct ceph_osd_request *req)
f24e9980 817{
f24e9980 818 req->r_tid = ++osdc->last_tid;
6df058c0 819 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
77f38e0e 820 dout("__register_request %p tid %lld\n", req, req->r_tid);
f24e9980
SW		 821 __insert_request(osdc, req);
822 ceph_osdc_get_request(req);
823 osdc->num_requests++;
f24e9980 824 if (osdc->num_requests == 1) {
422d2cb8
YS		 825 dout(" first request, scheduling timeout\n");
826 __schedule_osd_timeout(osdc);
f24e9980 827 }
a40c4f10
YS		 828}
829
830static void register_request(struct ceph_osd_client *osdc,
831 struct ceph_osd_request *req)
832{
833 mutex_lock(&osdc->request_mutex);
834 __register_request(osdc, req);
f24e9980
SW		 835 mutex_unlock(&osdc->request_mutex);
836}
837
838/*
839 * called under osdc->request_mutex
840 */
841static void __unregister_request(struct ceph_osd_client *osdc,
842 struct ceph_osd_request *req)
843{
844 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
845 rb_erase(&req->r_node, &osdc->requests);
846 osdc->num_requests--;
847
0ba6478d
SW		 848 if (req->r_osd) {
849 /* make sure the original request isn't in flight. */
850 ceph_con_revoke(&req->r_osd->o_con, req->r_request);
851
852 list_del_init(&req->r_osd_item);
a40c4f10
YS		 853 if (list_empty(&req->r_osd->o_requests) &&
854 list_empty(&req->r_osd->o_linger_requests)) {
855 dout("moving osd to %p lru\n", req->r_osd);
f5a2041b 856 __move_osd_to_lru(osdc, req->r_osd);
a40c4f10 857 }
fbdb9190 858 if (list_empty(&req->r_linger_item))
a40c4f10 859 req->r_osd = NULL;
0ba6478d 860 }
f24e9980
SW		 861
862 ceph_osdc_put_request(req);
863
422d2cb8
YS		 864 list_del_init(&req->r_req_lru_item);
865 if (osdc->num_requests == 0) {
866 dout(" no requests, canceling timeout\n");
867 __cancel_osd_timeout(osdc);
f24e9980
SW		 868 }
869}
870
871/*
872 * Cancel a previously queued request message
873 */
874static void __cancel_request(struct ceph_osd_request *req)
875{
6bc18876 876 if (req->r_sent && req->r_osd) {
f24e9980
SW		 877 ceph_con_revoke(&req->r_osd->o_con, req->r_request);
878 req->r_sent = 0;
879 }
880}
881
a40c4f10
YS		 882static void __register_linger_request(struct ceph_osd_client *osdc,
883 struct ceph_osd_request *req)
884{
885 dout("__register_linger_request %p\n", req);
886 list_add_tail(&req->r_linger_item, &osdc->req_linger);
887 list_add_tail(&req->r_linger_osd, &req->r_osd->o_linger_requests);
888}
889
890static void __unregister_linger_request(struct ceph_osd_client *osdc,
891 struct ceph_osd_request *req)
892{
893 dout("__unregister_linger_request %p\n", req);
894 if (req->r_osd) {
895 list_del_init(&req->r_linger_item);
896 list_del_init(&req->r_linger_osd);
897
898 if (list_empty(&req->r_osd->o_requests) &&
899 list_empty(&req->r_osd->o_linger_requests)) {
900 dout("moving osd to %p lru\n", req->r_osd);
901 __move_osd_to_lru(osdc, req->r_osd);
902 }
fbdb9190
SW		 903 if (list_empty(&req->r_osd_item))
904 req->r_osd = NULL;
a40c4f10
YS		 905 }
906}
907
908void ceph_osdc_unregister_linger_request(struct ceph_osd_client *osdc,
909 struct ceph_osd_request *req)
910{
911 mutex_lock(&osdc->request_mutex);
912 if (req->r_linger) {
913 __unregister_linger_request(osdc, req);
914 ceph_osdc_put_request(req);
915 }
916 mutex_unlock(&osdc->request_mutex);
917}
918EXPORT_SYMBOL(ceph_osdc_unregister_linger_request);
919
920void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
921 struct ceph_osd_request *req)
922{
923 if (!req->r_linger) {
924 dout("set_request_linger %p\n", req);
925 req->r_linger = 1;
926 /*
927 * caller is now responsible for calling
928 * unregister_linger_request
929 */
930 ceph_osdc_get_request(req);
931 }
932}
933EXPORT_SYMBOL(ceph_osdc_set_request_linger);
934
f24e9980
SW		 935/*
936 * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
937 * (as needed), and set the request r_osd appropriately. If there is
25985edc 938 * no up osd, set r_osd to NULL. Move the request to the appropriate list
6f6c7006 939 * (unsent, homeless) or leave on in-flight lru.
f24e9980
SW		 940 *
941 * Return 0 if unchanged, 1 if changed, or negative on error.
942 *
943 * Caller should hold map_sem for read and request_mutex.
944 */
6f6c7006
SW		 945static int __map_request(struct ceph_osd_client *osdc,
946 struct ceph_osd_request *req)
f24e9980
SW		 947{
948 struct ceph_osd_request_head *reqhead = req->r_request->front.iov_base;
51042122 949 struct ceph_pg pgid;
d85b7056
SW		 950 int acting[CEPH_PG_MAX_SIZE];
951 int o = -1, num = 0;
f24e9980 952 int err;
f24e9980 953
6f6c7006 954 dout("map_request %p tid %lld\n", req, req->r_tid);
f24e9980
SW		 955 err = ceph_calc_object_layout(&reqhead->layout, req->r_oid,
956 &req->r_file_layout, osdc->osdmap);
6f6c7006
SW		 957 if (err) {
958 list_move(&req->r_req_lru_item, &osdc->req_notarget);
f24e9980 959 return err;
6f6c7006 960 }
51042122 961 pgid = reqhead->layout.ol_pgid;
7740a42f
SW		 962 req->r_pgid = pgid;
963
d85b7056
SW		 964 err = ceph_calc_pg_acting(osdc->osdmap, pgid, acting);
965 if (err > 0) {
966 o = acting[0];
967 num = err;
968 }
f24e9980
SW		 969
970 if ((req->r_osd && req->r_osd->o_osd == o &&
d85b7056
SW		 971 req->r_sent >= req->r_osd->o_incarnation &&
972 req->r_num_pg_osds == num &&
973 memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
f24e9980
SW		 974 (req->r_osd == NULL && o == -1))
975 return 0; /* no change */
976
6f6c7006 977 dout("map_request tid %llu pgid %d.%x osd%d (was osd%d)\n",
51042122 978 req->r_tid, le32_to_cpu(pgid.pool), le16_to_cpu(pgid.ps), o,
f24e9980
SW		 979 req->r_osd ? req->r_osd->o_osd : -1);
980
d85b7056
SW		 981 /* record full pg acting set */
982 memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
983 req->r_num_pg_osds = num;
984
f24e9980
SW		 985 if (req->r_osd) {
986 __cancel_request(req);
987 list_del_init(&req->r_osd_item);
f24e9980
SW		 988 req->r_osd = NULL;
989 }
990
991 req->r_osd = __lookup_osd(osdc, o);
992 if (!req->r_osd && o >= 0) {
c99eb1c7
SW		 993 err = -ENOMEM;
994 req->r_osd = create_osd(osdc);
6f6c7006
SW		 995 if (!req->r_osd) {
996 list_move(&req->r_req_lru_item, &osdc->req_notarget);
c99eb1c7 997 goto out;
6f6c7006 998 }
f24e9980 999
6f6c7006 1000 dout("map_request osd %p is osd%d\n", req->r_osd, o);
f24e9980
SW		 1001 req->r_osd->o_osd = o;
1002 req->r_osd->o_con.peer_name.num = cpu_to_le64(o);
1003 __insert_osd(osdc, req->r_osd);
1004
1005 ceph_con_open(&req->r_osd->o_con, &osdc->osdmap->osd_addr[o]);
1006 }
1007
f5a2041b
YS		 1008 if (req->r_osd) {
1009 __remove_osd_from_lru(req->r_osd);
f24e9980 1010 list_add(&req->r_osd_item, &req->r_osd->o_requests);
6f6c7006
SW		 1011 list_move(&req->r_req_lru_item, &osdc->req_unsent);
1012 } else {
1013 list_move(&req->r_req_lru_item, &osdc->req_notarget);
f5a2041b 1014 }
d85b7056 1015 err = 1; /* osd or pg changed */
f24e9980
SW		 1016
1017out:
f24e9980
SW		 1018 return err;
1019}
1020
1021/*
1022 * caller should hold map_sem (for read) and request_mutex
1023 */
1024static int __send_request(struct ceph_osd_client *osdc,
1025 struct ceph_osd_request *req)
1026{
1027 struct ceph_osd_request_head *reqhead;
f24e9980
SW		 1028
1029 dout("send_request %p tid %llu to osd%d flags %d\n",
1030 req, req->r_tid, req->r_osd->o_osd, req->r_flags);
1031
1032 reqhead = req->r_request->front.iov_base;
1033 reqhead->osdmap_epoch = cpu_to_le32(osdc->osdmap->epoch);
1034 reqhead->flags |= cpu_to_le32(req->r_flags); /* e.g., RETRY */
1035 reqhead->reassert_version = req->r_reassert_version;
1036
3dd72fc0 1037 req->r_stamp = jiffies;
07a27e22 1038 list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
f24e9980
SW		 1039
1040 ceph_msg_get(req->r_request); /* send consumes a ref */
1041 ceph_con_send(&req->r_osd->o_con, req->r_request);
1042 req->r_sent = req->r_osd->o_incarnation;
1043 return 0;
1044}
1045
6f6c7006
SW		 1046/*
1047 * Send any requests in the queue (req_unsent).
1048 */
1049static void send_queued(struct ceph_osd_client *osdc)
1050{
1051 struct ceph_osd_request *req, *tmp;
1052
1053 dout("send_queued\n");
1054 mutex_lock(&osdc->request_mutex);
1055 list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item) {
1056 __send_request(osdc, req);
1057 }
1058 mutex_unlock(&osdc->request_mutex);
1059}
1060
f24e9980
SW		 1061/*
1062 * Timeout callback, called every N seconds when 1 or more osd
1063 * requests has been active for more than N seconds. When this
1064 * happens, we ping all OSDs with requests who have timed out to
1065 * ensure any communications channel reset is detected. Reset the
1066 * request timeouts another N seconds in the future as we go.
1067 * Reschedule the timeout event another N seconds in future (unless
1068 * there are no open requests).
1069 */
1070static void handle_timeout(struct work_struct *work)
1071{
1072 struct ceph_osd_client *osdc =
1073 container_of(work, struct ceph_osd_client, timeout_work.work);
422d2cb8 1074 struct ceph_osd_request *req, *last_req = NULL;
f24e9980 1075 struct ceph_osd *osd;
3d14c5d2 1076 unsigned long timeout = osdc->client->options->osd_timeout * HZ;
422d2cb8 1077 unsigned long keepalive =
3d14c5d2 1078 osdc->client->options->osd_keepalive_timeout * HZ;
3dd72fc0 1079 unsigned long last_stamp = 0;
422d2cb8 1080 struct list_head slow_osds;
f24e9980
SW		 1081 dout("timeout\n");
1082 down_read(&osdc->map_sem);
1083
1084 ceph_monc_request_next_osdmap(&osdc->client->monc);
1085
1086 mutex_lock(&osdc->request_mutex);
f24e9980 1087
422d2cb8
YS		 1088 /*
1089 * reset osds that appear to be _really_ unresponsive. this
1090 * is a failsafe measure.. we really shouldn't be getting to
1091 * this point if the system is working properly. the monitors
1092 * should mark the osd as failed and we should find out about
1093 * it from an updated osd map.
1094 */
f26e681d 1095 while (timeout && !list_empty(&osdc->req_lru)) {
422d2cb8
YS		 1096 req = list_entry(osdc->req_lru.next, struct ceph_osd_request,
1097 r_req_lru_item);
1098
4cf9d544 1099 /* hasn't been long enough since we sent it? */
3dd72fc0 1100 if (time_before(jiffies, req->r_stamp + timeout))
422d2cb8 1101 break;
4cf9d544
SW		 1102
1103 /* hasn't been long enough since it was acked? */
1104 if (req->r_request->ack_stamp == 0 ||
1105 time_before(jiffies, req->r_request->ack_stamp + timeout))
1106 break;
422d2cb8 1107
3dd72fc0 1108 BUG_ON(req == last_req && req->r_stamp == last_stamp);
422d2cb8 1109 last_req = req;
3dd72fc0 1110 last_stamp = req->r_stamp;
422d2cb8
YS		 1111
1112 osd = req->r_osd;
1113 BUG_ON(!osd);
1114 pr_warning(" tid %llu timed out on osd%d, will reset osd\n",
1115 req->r_tid, osd->o_osd);
6f6c7006 1116 __kick_osd_requests(osdc, osd);
422d2cb8
YS		 1117 }
1118
1119 /*
1120 * ping osds that are a bit slow. this ensures that if there
1121 * is a break in the TCP connection we will notice, and reopen
1122 * a connection with that osd (from the fault callback).
1123 */
1124 INIT_LIST_HEAD(&slow_osds);
1125 list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
3dd72fc0 1126 if (time_before(jiffies, req->r_stamp + keepalive))
422d2cb8
YS		 1127 break;
1128
1129 osd = req->r_osd;
1130 BUG_ON(!osd);
1131 dout(" tid %llu is slow, will send keepalive on osd%d\n",
f24e9980 1132 req->r_tid, osd->o_osd);
422d2cb8
YS		 1133 list_move_tail(&osd->o_keepalive_item, &slow_osds);
1134 }
1135 while (!list_empty(&slow_osds)) {
1136 osd = list_entry(slow_osds.next, struct ceph_osd,
1137 o_keepalive_item);
1138 list_del_init(&osd->o_keepalive_item);
f24e9980
SW		 1139 ceph_con_keepalive(&osd->o_con);
1140 }
1141
422d2cb8 1142 __schedule_osd_timeout(osdc);
f24e9980 1143 mutex_unlock(&osdc->request_mutex);
6f6c7006 1144 send_queued(osdc);
f24e9980
SW		 1145 up_read(&osdc->map_sem);
1146}
1147
f5a2041b
YS		 1148static void handle_osds_timeout(struct work_struct *work)
1149{
1150 struct ceph_osd_client *osdc =
1151 container_of(work, struct ceph_osd_client,
1152 osds_timeout_work.work);
1153 unsigned long delay =
3d14c5d2 1154 osdc->client->options->osd_idle_ttl * HZ >> 2;
f5a2041b
YS		 1155
1156 dout("osds timeout\n");
1157 down_read(&osdc->map_sem);
aca420bc 1158 remove_old_osds(osdc);
f5a2041b
YS		 1159 up_read(&osdc->map_sem);
1160
1161 schedule_delayed_work(&osdc->osds_timeout_work,
1162 round_jiffies_relative(delay));
1163}
1164
25845472
SW		 1165static void complete_request(struct ceph_osd_request *req)
1166{
1167 if (req->r_safe_callback)
1168 req->r_safe_callback(req, NULL);
1169 complete_all(&req->r_safe_completion); /* fsync waiter */
1170}
1171
f24e9980
SW		 1172/*
1173 * handle osd op reply. either call the callback if it is specified,
1174 * or do the completion to wake up the waiting thread.
1175 */
350b1c32
SW		 1176static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg,
1177 struct ceph_connection *con)
f24e9980
SW		 1178{
1179 struct ceph_osd_reply_head *rhead = msg->front.iov_base;
1180 struct ceph_osd_request *req;
1181 u64 tid;
1182 int numops, object_len, flags;
0ceed5db 1183 s32 result;
f24e9980 1184
6df058c0 1185 tid = le64_to_cpu(msg->hdr.tid);
f24e9980
SW		 1186 if (msg->front.iov_len < sizeof(*rhead))
1187 goto bad;
f24e9980
SW		 1188 numops = le32_to_cpu(rhead->num_ops);
1189 object_len = le32_to_cpu(rhead->object_len);
0ceed5db 1190 result = le32_to_cpu(rhead->result);
f24e9980
SW		 1191 if (msg->front.iov_len != sizeof(*rhead) + object_len +
1192 numops * sizeof(struct ceph_osd_op))
1193 goto bad;
0ceed5db 1194 dout("handle_reply %p tid %llu result %d\n", msg, tid, (int)result);
f24e9980
SW		 1195 /* lookup */
1196 mutex_lock(&osdc->request_mutex);
1197 req = __lookup_request(osdc, tid);
1198 if (req == NULL) {
1199 dout("handle_reply tid %llu dne\n", tid);
1200 mutex_unlock(&osdc->request_mutex);
1201 return;
1202 }
1203 ceph_osdc_get_request(req);
1204 flags = le32_to_cpu(rhead->flags);
1205
350b1c32 1206 /*
0d59ab81 1207 * if this connection filled our message, drop our reference now, to
350b1c32
SW		 1208 * avoid a (safe but slower) revoke later.
1209 */
0d59ab81 1210 if (req->r_con_filling_msg == con && req->r_reply == msg) {
c16e7869 1211 dout(" dropping con_filling_msg ref %p\n", con);
0d59ab81 1212 req->r_con_filling_msg = NULL;
350b1c32
SW		 1213 ceph_con_put(con);
1214 }
1215
f24e9980
SW		 1216 if (!req->r_got_reply) {
1217 unsigned bytes;
1218
1219 req->r_result = le32_to_cpu(rhead->result);
1220 bytes = le32_to_cpu(msg->hdr.data_len);
1221 dout("handle_reply result %d bytes %d\n", req->r_result,
1222 bytes);
1223 if (req->r_result == 0)
1224 req->r_result = bytes;
1225
1226 /* in case this is a write and we need to replay, */
1227 req->r_reassert_version = rhead->reassert_version;
1228
1229 req->r_got_reply = 1;
1230 } else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) {
1231 dout("handle_reply tid %llu dup ack\n", tid);
34b43a56 1232 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 1233 goto done;
1234 }
1235
1236 dout("handle_reply tid %llu flags %d\n", tid, flags);
1237
a40c4f10
YS		 1238 if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK))
1239 __register_linger_request(osdc, req);
1240
f24e9980 1241 /* either this is a read, or we got the safe response */
0ceed5db
SW		 1242 if (result < 0 ||
1243 (flags & CEPH_OSD_FLAG_ONDISK) ||
f24e9980
SW		 1244 ((flags & CEPH_OSD_FLAG_WRITE) == 0))
1245 __unregister_request(osdc, req);
1246
1247 mutex_unlock(&osdc->request_mutex);
1248
1249 if (req->r_callback)
1250 req->r_callback(req, msg);
1251 else
03066f23 1252 complete_all(&req->r_completion);
f24e9980 1253
25845472
SW		 1254 if (flags & CEPH_OSD_FLAG_ONDISK)
1255 complete_request(req);
f24e9980
SW		 1256
1257done:
a40c4f10 1258 dout("req=%p req->r_linger=%d\n", req, req->r_linger);
f24e9980
SW		 1259 ceph_osdc_put_request(req);
1260 return;
1261
1262bad:
1263 pr_err("corrupt osd_op_reply got %d %d expected %d\n",
1264 (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len),
1265 (int)sizeof(*rhead));
9ec7cab1 1266 ceph_msg_dump(msg);
f24e9980
SW		 1267}
1268
6f6c7006 1269static void reset_changed_osds(struct ceph_osd_client *osdc)
f24e9980 1270{
f24e9980 1271 struct rb_node *p, *n;
f24e9980 1272
6f6c7006
SW		 1273 for (p = rb_first(&osdc->osds); p; p = n) {
1274 struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
f24e9980 1275
6f6c7006
SW		 1276 n = rb_next(p);
1277 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
1278 memcmp(&osd->o_con.peer_addr,
1279 ceph_osd_addr(osdc->osdmap,
1280 osd->o_osd),
1281 sizeof(struct ceph_entity_addr)) != 0)
1282 __reset_osd(osdc, osd);
f24e9980 1283 }
422d2cb8
YS		 1284}
1285
1286/*
6f6c7006
SW		 1287 * Requeue requests whose mapping to an OSD has changed. If requests map to
1288 * no osd, request a new map.
422d2cb8
YS		 1289 *
1290 * Caller should hold map_sem for read and request_mutex.
1291 */
6f6c7006 1292static void kick_requests(struct ceph_osd_client *osdc)
422d2cb8 1293{
a40c4f10 1294 struct ceph_osd_request *req, *nreq;
6f6c7006
SW		 1295 struct rb_node *p;
1296 int needmap = 0;
1297 int err;
422d2cb8 1298
6f6c7006 1299 dout("kick_requests\n");
422d2cb8 1300 mutex_lock(&osdc->request_mutex);
6f6c7006
SW		 1301 for (p = rb_first(&osdc->requests); p; p = rb_next(p)) {
1302 req = rb_entry(p, struct ceph_osd_request, r_node);
1303 err = __map_request(osdc, req);
1304 if (err < 0)
1305 continue; /* error */
1306 if (req->r_osd == NULL) {
1307 dout("%p tid %llu maps to no osd\n", req, req->r_tid);
1308 needmap++; /* request a newer map */
1309 } else if (err > 0) {
1310 dout("%p tid %llu requeued on osd%d\n", req, req->r_tid,
1311 req->r_osd ? req->r_osd->o_osd : -1);
a40c4f10
YS		 1312 if (!req->r_linger)
1313 req->r_flags |= CEPH_OSD_FLAG_RETRY;
1314 }
1315 }
1316
1317 list_for_each_entry_safe(req, nreq, &osdc->req_linger,
1318 r_linger_item) {
1319 dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
1320
1321 err = __map_request(osdc, req);
1322 if (err == 0)
1323 continue; /* no change and no osd was specified */
1324 if (err < 0)
1325 continue; /* hrm! */
1326 if (req->r_osd == NULL) {
1327 dout("tid %llu maps to no valid osd\n", req->r_tid);
1328 needmap++; /* request a newer map */
1329 continue;
6f6c7006 1330 }
a40c4f10
YS		 1331
1332 dout("kicking lingering %p tid %llu osd%d\n", req, req->r_tid,
1333 req->r_osd ? req->r_osd->o_osd : -1);
1334 __unregister_linger_request(osdc, req);
1335 __register_request(osdc, req);
6f6c7006 1336 }
f24e9980
SW		 1337 mutex_unlock(&osdc->request_mutex);
1338
1339 if (needmap) {
1340 dout("%d requests for down osds, need new map\n", needmap);
1341 ceph_monc_request_next_osdmap(&osdc->client->monc);
1342 }
422d2cb8 1343}
6f6c7006
SW		 1344
1345
f24e9980
SW		 1346/*
1347 * Process updated osd map.
1348 *
1349 * The message contains any number of incremental and full maps, normally
1350 * indicating some sort of topology change in the cluster. Kick requests
1351 * off to different OSDs as needed.
1352 */
1353void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1354{
1355 void *p, *end, *next;
1356 u32 nr_maps, maplen;
1357 u32 epoch;
1358 struct ceph_osdmap *newmap = NULL, *oldmap;
1359 int err;
1360 struct ceph_fsid fsid;
1361
1362 dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
1363 p = msg->front.iov_base;
1364 end = p + msg->front.iov_len;
1365
1366 /* verify fsid */
1367 ceph_decode_need(&p, end, sizeof(fsid), bad);
1368 ceph_decode_copy(&p, &fsid, sizeof(fsid));
0743304d
SW		 1369 if (ceph_check_fsid(osdc->client, &fsid) < 0)
1370 return;
f24e9980
SW		 1371
1372 down_write(&osdc->map_sem);
1373
1374 /* incremental maps */
1375 ceph_decode_32_safe(&p, end, nr_maps, bad);
1376 dout(" %d inc maps\n", nr_maps);
1377 while (nr_maps > 0) {
1378 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 1379 epoch = ceph_decode_32(&p);
1380 maplen = ceph_decode_32(&p);
f24e9980
SW		 1381 ceph_decode_need(&p, end, maplen, bad);
1382 next = p + maplen;
1383 if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
1384 dout("applying incremental map %u len %d\n",
1385 epoch, maplen);
1386 newmap = osdmap_apply_incremental(&p, next,
1387 osdc->osdmap,
1388 osdc->client->msgr);
1389 if (IS_ERR(newmap)) {
1390 err = PTR_ERR(newmap);
1391 goto bad;
1392 }
30dc6381 1393 BUG_ON(!newmap);
f24e9980
SW		 1394 if (newmap != osdc->osdmap) {
1395 ceph_osdmap_destroy(osdc->osdmap);
1396 osdc->osdmap = newmap;
1397 }
6f6c7006
SW		 1398 kick_requests(osdc);
1399 reset_changed_osds(osdc);
f24e9980
SW		 1400 } else {
1401 dout("ignoring incremental map %u len %d\n",
1402 epoch, maplen);
1403 }
1404 p = next;
1405 nr_maps--;
1406 }
1407 if (newmap)
1408 goto done;
1409
1410 /* full maps */
1411 ceph_decode_32_safe(&p, end, nr_maps, bad);
1412 dout(" %d full maps\n", nr_maps);
1413 while (nr_maps) {
1414 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 1415 epoch = ceph_decode_32(&p);
1416 maplen = ceph_decode_32(&p);
f24e9980
SW		 1417 ceph_decode_need(&p, end, maplen, bad);
1418 if (nr_maps > 1) {
1419 dout("skipping non-latest full map %u len %d\n",
1420 epoch, maplen);
1421 } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
1422 dout("skipping full map %u len %d, "
1423 "older than our %u\n", epoch, maplen,
1424 osdc->osdmap->epoch);
1425 } else {
1426 dout("taking full map %u len %d\n", epoch, maplen);
1427 newmap = osdmap_decode(&p, p+maplen);
1428 if (IS_ERR(newmap)) {
1429 err = PTR_ERR(newmap);
1430 goto bad;
1431 }
30dc6381 1432 BUG_ON(!newmap);
f24e9980
SW		 1433 oldmap = osdc->osdmap;
1434 osdc->osdmap = newmap;
1435 if (oldmap)
1436 ceph_osdmap_destroy(oldmap);
6f6c7006 1437 kick_requests(osdc);
f24e9980
SW		 1438 }
1439 p += maplen;
1440 nr_maps--;
1441 }
1442
1443done:
1444 downgrade_write(&osdc->map_sem);
1445 ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch);
cd634fb6
SW		 1446
1447 /*
1448 * subscribe to subsequent osdmap updates if full to ensure
1449 * we find out when we are no longer full and stop returning
1450 * ENOSPC.
1451 */
1452 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL))
1453 ceph_monc_request_next_osdmap(&osdc->client->monc);
1454
6f6c7006 1455 send_queued(osdc);
f24e9980 1456 up_read(&osdc->map_sem);
03066f23 1457 wake_up_all(&osdc->client->auth_wq);
f24e9980
SW		 1458 return;
1459
1460bad:
1461 pr_err("osdc handle_map corrupt msg\n");
9ec7cab1 1462 ceph_msg_dump(msg);
f24e9980
SW		 1463 up_write(&osdc->map_sem);
1464 return;
1465}
1466
a40c4f10
YS		 1467/*
1468 * watch/notify callback event infrastructure
1469 *
1470 * These callbacks are used both for watch and notify operations.
1471 */
1472static void __release_event(struct kref *kref)
1473{
1474 struct ceph_osd_event *event =
1475 container_of(kref, struct ceph_osd_event, kref);
1476
1477 dout("__release_event %p\n", event);
1478 kfree(event);
1479}
1480
1481static void get_event(struct ceph_osd_event *event)
1482{
1483 kref_get(&event->kref);
1484}
1485
1486void ceph_osdc_put_event(struct ceph_osd_event *event)
1487{
1488 kref_put(&event->kref, __release_event);
1489}
1490EXPORT_SYMBOL(ceph_osdc_put_event);
1491
1492static void __insert_event(struct ceph_osd_client *osdc,
1493 struct ceph_osd_event *new)
1494{
1495 struct rb_node **p = &osdc->event_tree.rb_node;
1496 struct rb_node *parent = NULL;
1497 struct ceph_osd_event *event = NULL;
1498
1499 while (*p) {
1500 parent = *p;
1501 event = rb_entry(parent, struct ceph_osd_event, node);
1502 if (new->cookie < event->cookie)
1503 p = &(*p)->rb_left;
1504 else if (new->cookie > event->cookie)
1505 p = &(*p)->rb_right;
1506 else
1507 BUG();
1508 }
1509
1510 rb_link_node(&new->node, parent, p);
1511 rb_insert_color(&new->node, &osdc->event_tree);
1512}
1513
1514static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
1515 u64 cookie)
1516{
1517 struct rb_node **p = &osdc->event_tree.rb_node;
1518 struct rb_node *parent = NULL;
1519 struct ceph_osd_event *event = NULL;
1520
1521 while (*p) {
1522 parent = *p;
1523 event = rb_entry(parent, struct ceph_osd_event, node);
1524 if (cookie < event->cookie)
1525 p = &(*p)->rb_left;
1526 else if (cookie > event->cookie)
1527 p = &(*p)->rb_right;
1528 else
1529 return event;
1530 }
1531 return NULL;
1532}
1533
1534static void __remove_event(struct ceph_osd_event *event)
1535{
1536 struct ceph_osd_client *osdc = event->osdc;
1537
1538 if (!RB_EMPTY_NODE(&event->node)) {
1539 dout("__remove_event removed %p\n", event);
1540 rb_erase(&event->node, &osdc->event_tree);
1541 ceph_osdc_put_event(event);
1542 } else {
1543 dout("__remove_event didn't remove %p\n", event);
1544 }
1545}
1546
1547int ceph_osdc_create_event(struct ceph_osd_client *osdc,
1548 void (*event_cb)(u64, u64, u8, void *),
1549 int one_shot, void *data,
1550 struct ceph_osd_event **pevent)
1551{
1552 struct ceph_osd_event *event;
1553
1554 event = kmalloc(sizeof(*event), GFP_NOIO);
1555 if (!event)
1556 return -ENOMEM;
1557
1558 dout("create_event %p\n", event);
1559 event->cb = event_cb;
1560 event->one_shot = one_shot;
1561 event->data = data;
1562 event->osdc = osdc;
1563 INIT_LIST_HEAD(&event->osd_node);
1564 kref_init(&event->kref); /* one ref for us */
1565 kref_get(&event->kref); /* one ref for the caller */
1566 init_completion(&event->completion);
1567
1568 spin_lock(&osdc->event_lock);
1569 event->cookie = ++osdc->event_count;
1570 __insert_event(osdc, event);
1571 spin_unlock(&osdc->event_lock);
1572
1573 *pevent = event;
1574 return 0;
1575}
1576EXPORT_SYMBOL(ceph_osdc_create_event);
1577
1578void ceph_osdc_cancel_event(struct ceph_osd_event *event)
1579{
1580 struct ceph_osd_client *osdc = event->osdc;
1581
1582 dout("cancel_event %p\n", event);
1583 spin_lock(&osdc->event_lock);
1584 __remove_event(event);
1585 spin_unlock(&osdc->event_lock);
1586 ceph_osdc_put_event(event); /* caller's */
1587}
1588EXPORT_SYMBOL(ceph_osdc_cancel_event);
1589
1590
1591static void do_event_work(struct work_struct *work)
1592{
1593 struct ceph_osd_event_work *event_work =
1594 container_of(work, struct ceph_osd_event_work, work);
1595 struct ceph_osd_event *event = event_work->event;
1596 u64 ver = event_work->ver;
1597 u64 notify_id = event_work->notify_id;
1598 u8 opcode = event_work->opcode;
1599
1600 dout("do_event_work completing %p\n", event);
1601 event->cb(ver, notify_id, opcode, event->data);
1602 complete(&event->completion);
1603 dout("do_event_work completed %p\n", event);
1604 ceph_osdc_put_event(event);
1605 kfree(event_work);
1606}
1607
1608
1609/*
1610 * Process osd watch notifications
1611 */
1612void handle_watch_notify(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1613{
1614 void *p, *end;
1615 u8 proto_ver;
1616 u64 cookie, ver, notify_id;
1617 u8 opcode;
1618 struct ceph_osd_event *event;
1619 struct ceph_osd_event_work *event_work;
1620
1621 p = msg->front.iov_base;
1622 end = p + msg->front.iov_len;
1623
1624 ceph_decode_8_safe(&p, end, proto_ver, bad);
1625 ceph_decode_8_safe(&p, end, opcode, bad);
1626 ceph_decode_64_safe(&p, end, cookie, bad);
1627 ceph_decode_64_safe(&p, end, ver, bad);
1628 ceph_decode_64_safe(&p, end, notify_id, bad);
1629
1630 spin_lock(&osdc->event_lock);
1631 event = __find_event(osdc, cookie);
1632 if (event) {
1633 get_event(event);
1634 if (event->one_shot)
1635 __remove_event(event);
1636 }
1637 spin_unlock(&osdc->event_lock);
1638 dout("handle_watch_notify cookie %lld ver %lld event %p\n",
1639 cookie, ver, event);
1640 if (event) {
1641 event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
a40c4f10
YS		 1642 if (!event_work) {
1643 dout("ERROR: could not allocate event_work\n");
1644 goto done_err;
1645 }
6b0ae409 1646 INIT_WORK(&event_work->work, do_event_work);
a40c4f10
YS		 1647 event_work->event = event;
1648 event_work->ver = ver;
1649 event_work->notify_id = notify_id;
1650 event_work->opcode = opcode;
1651 if (!queue_work(osdc->notify_wq, &event_work->work)) {
1652 dout("WARNING: failed to queue notify event work\n");
1653 goto done_err;
1654 }
1655 }
1656
1657 return;
1658
1659done_err:
1660 complete(&event->completion);
1661 ceph_osdc_put_event(event);
1662 return;
1663
1664bad:
1665 pr_err("osdc handle_watch_notify corrupt msg\n");
1666 return;
1667}
1668
1669int ceph_osdc_wait_event(struct ceph_osd_event *event, unsigned long timeout)
1670{
1671 int err;
1672
1673 dout("wait_event %p\n", event);
1674 err = wait_for_completion_interruptible_timeout(&event->completion,
1675 timeout * HZ);
1676 ceph_osdc_put_event(event);
1677 if (err > 0)
1678 err = 0;
1679 dout("wait_event %p returns %d\n", event, err);
1680 return err;
1681}
1682EXPORT_SYMBOL(ceph_osdc_wait_event);
1683
f24e9980
SW		 1684/*
1685 * Register request, send initial attempt.
1686 */
1687int ceph_osdc_start_request(struct ceph_osd_client *osdc,
1688 struct ceph_osd_request *req,
1689 bool nofail)
1690{
c1ea8823 1691 int rc = 0;
f24e9980
SW		 1692
1693 req->r_request->pages = req->r_pages;
1694 req->r_request->nr_pages = req->r_num_pages;
68b4476b
YS		 1695#ifdef CONFIG_BLOCK
1696 req->r_request->bio = req->r_bio;
1697#endif
1698 req->r_request->trail = req->r_trail;
f24e9980
SW		 1699
1700 register_request(osdc, req);
1701
1702 down_read(&osdc->map_sem);
1703 mutex_lock(&osdc->request_mutex);
c1ea8823
SW		 1704 /*
1705 * a racing kick_requests() may have sent the message for us
1706 * while we dropped request_mutex above, so only send now if
1707 * the request still han't been touched yet.
1708 */
1709 if (req->r_sent == 0) {
6f6c7006 1710 rc = __map_request(osdc, req);
9d6fcb08
SW		 1711 if (rc < 0) {
1712 if (nofail) {
1713 dout("osdc_start_request failed map, "
1714 " will retry %lld\n", req->r_tid);
1715 rc = 0;
1716 }
234af26f 1717 goto out_unlock;
9d6fcb08 1718 }
6f6c7006
SW		 1719 if (req->r_osd == NULL) {
1720 dout("send_request %p no up osds in pg\n", req);
1721 ceph_monc_request_next_osdmap(&osdc->client->monc);
1722 } else {
1723 rc = __send_request(osdc, req);
1724 if (rc) {
1725 if (nofail) {
1726 dout("osdc_start_request failed send, "
1727 " will retry %lld\n", req->r_tid);
1728 rc = 0;
1729 } else {
1730 __unregister_request(osdc, req);
1731 }
c1ea8823 1732 }
f24e9980
SW		 1733 }
1734 }
234af26f
DC		 1735
1736out_unlock:
f24e9980
SW		 1737 mutex_unlock(&osdc->request_mutex);
1738 up_read(&osdc->map_sem);
1739 return rc;
1740}
3d14c5d2 1741EXPORT_SYMBOL(ceph_osdc_start_request);
f24e9980
SW		 1742
1743/*
1744 * wait for a request to complete
1745 */
1746int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
1747 struct ceph_osd_request *req)
1748{
1749 int rc;
1750
1751 rc = wait_for_completion_interruptible(&req->r_completion);
1752 if (rc < 0) {
1753 mutex_lock(&osdc->request_mutex);
1754 __cancel_request(req);
529cfcc4 1755 __unregister_request(osdc, req);
f24e9980 1756 mutex_unlock(&osdc->request_mutex);
25845472 1757 complete_request(req);
529cfcc4 1758 dout("wait_request tid %llu canceled/timed out\n", req->r_tid);
f24e9980
SW		 1759 return rc;
1760 }
1761
1762 dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result);
1763 return req->r_result;
1764}
3d14c5d2 1765EXPORT_SYMBOL(ceph_osdc_wait_request);
f24e9980
SW		 1766
1767/*
1768 * sync - wait for all in-flight requests to flush. avoid starvation.
1769 */
1770void ceph_osdc_sync(struct ceph_osd_client *osdc)
1771{
1772 struct ceph_osd_request *req;
1773 u64 last_tid, next_tid = 0;
1774
1775 mutex_lock(&osdc->request_mutex);
1776 last_tid = osdc->last_tid;
1777 while (1) {
1778 req = __lookup_request_ge(osdc, next_tid);
1779 if (!req)
1780 break;
1781 if (req->r_tid > last_tid)
1782 break;
1783
1784 next_tid = req->r_tid + 1;
1785 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
1786 continue;
1787
1788 ceph_osdc_get_request(req);
1789 mutex_unlock(&osdc->request_mutex);
1790 dout("sync waiting on tid %llu (last is %llu)\n",
1791 req->r_tid, last_tid);
1792 wait_for_completion(&req->r_safe_completion);
1793 mutex_lock(&osdc->request_mutex);
1794 ceph_osdc_put_request(req);
1795 }
1796 mutex_unlock(&osdc->request_mutex);
1797 dout("sync done (thru tid %llu)\n", last_tid);
1798}
3d14c5d2 1799EXPORT_SYMBOL(ceph_osdc_sync);
f24e9980
SW		 1800
1801/*
1802 * init, shutdown
1803 */
1804int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
1805{
1806 int err;
1807
1808 dout("init\n");
1809 osdc->client = client;
1810 osdc->osdmap = NULL;
1811 init_rwsem(&osdc->map_sem);
1812 init_completion(&osdc->map_waiters);
1813 osdc->last_requested_map = 0;
1814 mutex_init(&osdc->request_mutex);
f24e9980
SW		 1815 osdc->last_tid = 0;
1816 osdc->osds = RB_ROOT;
f5a2041b 1817 INIT_LIST_HEAD(&osdc->osd_lru);
f24e9980 1818 osdc->requests = RB_ROOT;
422d2cb8 1819 INIT_LIST_HEAD(&osdc->req_lru);
6f6c7006
SW		 1820 INIT_LIST_HEAD(&osdc->req_unsent);
1821 INIT_LIST_HEAD(&osdc->req_notarget);
a40c4f10 1822 INIT_LIST_HEAD(&osdc->req_linger);
f24e9980
SW		 1823 osdc->num_requests = 0;
1824 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
f5a2041b 1825 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
a40c4f10
YS		 1826 spin_lock_init(&osdc->event_lock);
1827 osdc->event_tree = RB_ROOT;
1828 osdc->event_count = 0;
f5a2041b
YS		 1829
1830 schedule_delayed_work(&osdc->osds_timeout_work,
3d14c5d2 1831 round_jiffies_relative(osdc->client->options->osd_idle_ttl * HZ));
f24e9980 1832
5f44f142 1833 err = -ENOMEM;
f24e9980
SW		 1834 osdc->req_mempool = mempool_create_kmalloc_pool(10,
1835 sizeof(struct ceph_osd_request));
1836 if (!osdc->req_mempool)
5f44f142 1837 goto out;
f24e9980 1838
4f48280e
SW		 1839 err = ceph_msgpool_init(&osdc->msgpool_op, OSD_OP_FRONT_LEN, 10, true,
1840 "osd_op");
f24e9980 1841 if (err < 0)
5f44f142 1842 goto out_mempool;
c16e7869 1843 err = ceph_msgpool_init(&osdc->msgpool_op_reply,
4f48280e
SW		 1844 OSD_OPREPLY_FRONT_LEN, 10, true,
1845 "osd_op_reply");
c16e7869
SW		 1846 if (err < 0)
1847 goto out_msgpool;
a40c4f10
YS		 1848
1849 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
1850 if (IS_ERR(osdc->notify_wq)) {
1851 err = PTR_ERR(osdc->notify_wq);
1852 osdc->notify_wq = NULL;
1853 goto out_msgpool;
1854 }
f24e9980 1855 return 0;
5f44f142 1856
c16e7869
SW		 1857out_msgpool:
1858 ceph_msgpool_destroy(&osdc->msgpool_op);
5f44f142
SW		 1859out_mempool:
1860 mempool_destroy(osdc->req_mempool);
1861out:
1862 return err;
f24e9980 1863}
3d14c5d2 1864EXPORT_SYMBOL(ceph_osdc_init);
f24e9980
SW		 1865
1866void ceph_osdc_stop(struct ceph_osd_client *osdc)
1867{
a40c4f10
YS		 1868 flush_workqueue(osdc->notify_wq);
1869 destroy_workqueue(osdc->notify_wq);
f24e9980 1870 cancel_delayed_work_sync(&osdc->timeout_work);
f5a2041b 1871 cancel_delayed_work_sync(&osdc->osds_timeout_work);
f24e9980
SW		 1872 if (osdc->osdmap) {
1873 ceph_osdmap_destroy(osdc->osdmap);
1874 osdc->osdmap = NULL;
1875 }
aca420bc 1876 remove_all_osds(osdc);
f24e9980
SW		 1877 mempool_destroy(osdc->req_mempool);
1878 ceph_msgpool_destroy(&osdc->msgpool_op);
c16e7869 1879 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
f24e9980 1880}
3d14c5d2 1881EXPORT_SYMBOL(ceph_osdc_stop);
f24e9980
SW		 1882
1883/*
1884 * Read some contiguous pages. If we cross a stripe boundary, shorten
1885 * *plen. Return number of bytes read, or error.
1886 */
1887int ceph_osdc_readpages(struct ceph_osd_client *osdc,
1888 struct ceph_vino vino, struct ceph_file_layout *layout,
1889 u64 off, u64 *plen,
1890 u32 truncate_seq, u64 truncate_size,
b7495fc2 1891 struct page **pages, int num_pages, int page_align)
f24e9980
SW		 1892{
1893 struct ceph_osd_request *req;
1894 int rc = 0;
1895
1896 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
1897 vino.snap, off, *plen);
1898 req = ceph_osdc_new_request(osdc, layout, vino, off, plen,
1899 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
1900 NULL, 0, truncate_seq, truncate_size, NULL,
b7495fc2 1901 false, 1, page_align);
a79832f2
SW		 1902 if (!req)
1903 return -ENOMEM;
f24e9980
SW		 1904
1905 /* it may be a short read due to an object boundary */
1906 req->r_pages = pages;
f24e9980 1907
b7495fc2
SW		 1908 dout("readpages final extent is %llu~%llu (%d pages align %d)\n",
1909 off, *plen, req->r_num_pages, page_align);
f24e9980
SW		 1910
1911 rc = ceph_osdc_start_request(osdc, req, false);
1912 if (!rc)
1913 rc = ceph_osdc_wait_request(osdc, req);
1914
1915 ceph_osdc_put_request(req);
1916 dout("readpages result %d\n", rc);
1917 return rc;
1918}
3d14c5d2 1919EXPORT_SYMBOL(ceph_osdc_readpages);
f24e9980
SW		 1920
1921/*
1922 * do a synchronous write on N pages
1923 */
1924int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
1925 struct ceph_file_layout *layout,
1926 struct ceph_snap_context *snapc,
1927 u64 off, u64 len,
1928 u32 truncate_seq, u64 truncate_size,
1929 struct timespec *mtime,
1930 struct page **pages, int num_pages,
1931 int flags, int do_sync, bool nofail)
1932{
1933 struct ceph_osd_request *req;
1934 int rc = 0;
b7495fc2 1935 int page_align = off & ~PAGE_MASK;
f24e9980
SW		 1936
1937 BUG_ON(vino.snap != CEPH_NOSNAP);
1938 req = ceph_osdc_new_request(osdc, layout, vino, off, &len,
1939 CEPH_OSD_OP_WRITE,
1940 flags | CEPH_OSD_FLAG_ONDISK |
1941 CEPH_OSD_FLAG_WRITE,
1942 snapc, do_sync,
1943 truncate_seq, truncate_size, mtime,
b7495fc2 1944 nofail, 1, page_align);
a79832f2
SW		 1945 if (!req)
1946 return -ENOMEM;
f24e9980
SW		 1947
1948 /* it may be a short write due to an object boundary */
1949 req->r_pages = pages;
f24e9980
SW		 1950 dout("writepages %llu~%llu (%d pages)\n", off, len,
1951 req->r_num_pages);
1952
1953 rc = ceph_osdc_start_request(osdc, req, nofail);
1954 if (!rc)
1955 rc = ceph_osdc_wait_request(osdc, req);
1956
1957 ceph_osdc_put_request(req);
1958 if (rc == 0)
1959 rc = len;
1960 dout("writepages result %d\n", rc);
1961 return rc;
1962}
3d14c5d2 1963EXPORT_SYMBOL(ceph_osdc_writepages);
f24e9980
SW		 1964
1965/*
1966 * handle incoming message
1967 */
1968static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
1969{
1970 struct ceph_osd *osd = con->private;
32c895e7 1971 struct ceph_osd_client *osdc;
f24e9980
SW		 1972 int type = le16_to_cpu(msg->hdr.type);
1973
1974 if (!osd)
4a32f93d 1975 goto out;
32c895e7 1976 osdc = osd->o_osdc;
f24e9980
SW		 1977
1978 switch (type) {
1979 case CEPH_MSG_OSD_MAP:
1980 ceph_osdc_handle_map(osdc, msg);
1981 break;
1982 case CEPH_MSG_OSD_OPREPLY:
350b1c32 1983 handle_reply(osdc, msg, con);
f24e9980 1984 break;
a40c4f10
YS		 1985 case CEPH_MSG_WATCH_NOTIFY:
1986 handle_watch_notify(osdc, msg);
1987 break;
f24e9980
SW		 1988
1989 default:
1990 pr_err("received unknown message type %d %s\n", type,
1991 ceph_msg_type_name(type));
1992 }
4a32f93d 1993out:
f24e9980
SW		 1994 ceph_msg_put(msg);
1995}
1996
5b3a4db3 1997/*
21b667f6
SW		 1998 * lookup and return message for incoming reply. set up reply message
1999 * pages.
5b3a4db3
SW		 2000 */
2001static struct ceph_msg *get_reply(struct ceph_connection *con,
2450418c
YS		 2002 struct ceph_msg_header *hdr,
2003 int *skip)
f24e9980
SW		 2004{
2005 struct ceph_osd *osd = con->private;
2006 struct ceph_osd_client *osdc = osd->o_osdc;
2450418c 2007 struct ceph_msg *m;
0547a9b3 2008 struct ceph_osd_request *req;
5b3a4db3
SW		 2009 int front = le32_to_cpu(hdr->front_len);
2010 int data_len = le32_to_cpu(hdr->data_len);
0547a9b3 2011 u64 tid;
f24e9980 2012
0547a9b3
YS		 2013 tid = le64_to_cpu(hdr->tid);
2014 mutex_lock(&osdc->request_mutex);
2015 req = __lookup_request(osdc, tid);
2016 if (!req) {
2017 *skip = 1;
2018 m = NULL;
c16e7869 2019 pr_info("get_reply unknown tid %llu from osd%d\n", tid,
5b3a4db3 2020 osd->o_osd);
0547a9b3
YS		 2021 goto out;
2022 }
c16e7869
SW		 2023
2024 if (req->r_con_filling_msg) {
2025 dout("get_reply revoking msg %p from old con %p\n",
2026 req->r_reply, req->r_con_filling_msg);
2027 ceph_con_revoke_message(req->r_con_filling_msg, req->r_reply);
2028 ceph_con_put(req->r_con_filling_msg);
6f46cb29 2029 req->r_con_filling_msg = NULL;
0547a9b3
YS		 2030 }
2031
c16e7869
SW		 2032 if (front > req->r_reply->front.iov_len) {
2033 pr_warning("get_reply front %d > preallocated %d\n",
2034 front, (int)req->r_reply->front.iov_len);
34d23762 2035 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front, GFP_NOFS);
a79832f2 2036 if (!m)
c16e7869
SW		 2037 goto out;
2038 ceph_msg_put(req->r_reply);
2039 req->r_reply = m;
2040 }
2041 m = ceph_msg_get(req->r_reply);
2042
0547a9b3 2043 if (data_len > 0) {
b7495fc2 2044 int want = calc_pages_for(req->r_page_alignment, data_len);
21b667f6
SW		 2045
2046 if (unlikely(req->r_num_pages < want)) {
9bb0ce2b
SW		 2047 pr_warning("tid %lld reply has %d bytes %d pages, we"
2048 " had only %d pages ready\n", tid, data_len,
2049 want, req->r_num_pages);
0547a9b3
YS		 2050 *skip = 1;
2051 ceph_msg_put(m);
a79832f2 2052 m = NULL;
21b667f6 2053 goto out;
0547a9b3 2054 }
21b667f6
SW		 2055 m->pages = req->r_pages;
2056 m->nr_pages = req->r_num_pages;
c5c6b19d 2057 m->page_alignment = req->r_page_alignment;
68b4476b
YS		 2058#ifdef CONFIG_BLOCK
2059 m->bio = req->r_bio;
2060#endif
0547a9b3 2061 }
5b3a4db3 2062 *skip = 0;
c16e7869
SW		 2063 req->r_con_filling_msg = ceph_con_get(con);
2064 dout("get_reply tid %lld %p\n", tid, m);
0547a9b3
YS		 2065
2066out:
2067 mutex_unlock(&osdc->request_mutex);
2450418c 2068 return m;
5b3a4db3
SW		 2069
2070}
2071
2072static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2073 struct ceph_msg_header *hdr,
2074 int *skip)
2075{
2076 struct ceph_osd *osd = con->private;
2077 int type = le16_to_cpu(hdr->type);
2078 int front = le32_to_cpu(hdr->front_len);
2079
2080 switch (type) {
2081 case CEPH_MSG_OSD_MAP:
a40c4f10 2082 case CEPH_MSG_WATCH_NOTIFY:
34d23762 2083 return ceph_msg_new(type, front, GFP_NOFS);
5b3a4db3
SW		 2084 case CEPH_MSG_OSD_OPREPLY:
2085 return get_reply(con, hdr, skip);
2086 default:
2087 pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
2088 osd->o_osd);
2089 *skip = 1;
2090 return NULL;
2091 }
f24e9980
SW		 2092}
2093
2094/*
2095 * Wrappers to refcount containing ceph_osd struct
2096 */
2097static struct ceph_connection *get_osd_con(struct ceph_connection *con)
2098{
2099 struct ceph_osd *osd = con->private;
2100 if (get_osd(osd))
2101 return con;
2102 return NULL;
2103}
2104
2105static void put_osd_con(struct ceph_connection *con)
2106{
2107 struct ceph_osd *osd = con->private;
2108 put_osd(osd);
2109}
2110
4e7a5dcd
SW		 2111/*
2112 * authentication
2113 */
2114static int get_authorizer(struct ceph_connection *con,
213c99ee
SW		 2115 void **buf, int *len, int *proto,
2116 void **reply_buf, int *reply_len, int force_new)
4e7a5dcd
SW		 2117{
2118 struct ceph_osd *o = con->private;
2119 struct ceph_osd_client *osdc = o->o_osdc;
2120 struct ceph_auth_client *ac = osdc->client->monc.auth;
2121 int ret = 0;
2122
2123 if (force_new && o->o_authorizer) {
2124 ac->ops->destroy_authorizer(ac, o->o_authorizer);
2125 o->o_authorizer = NULL;
2126 }
2127 if (o->o_authorizer == NULL) {
2128 ret = ac->ops->create_authorizer(
2129 ac, CEPH_ENTITY_TYPE_OSD,
2130 &o->o_authorizer,
2131 &o->o_authorizer_buf,
2132 &o->o_authorizer_buf_len,
2133 &o->o_authorizer_reply_buf,
2134 &o->o_authorizer_reply_buf_len);
2135 if (ret)
213c99ee 2136 return ret;
4e7a5dcd
SW		 2137 }
2138
2139 *proto = ac->protocol;
2140 *buf = o->o_authorizer_buf;
2141 *len = o->o_authorizer_buf_len;
2142 *reply_buf = o->o_authorizer_reply_buf;
2143 *reply_len = o->o_authorizer_reply_buf_len;
2144 return 0;
2145}
2146
2147
2148static int verify_authorizer_reply(struct ceph_connection *con, int len)
2149{
2150 struct ceph_osd *o = con->private;
2151 struct ceph_osd_client *osdc = o->o_osdc;
2152 struct ceph_auth_client *ac = osdc->client->monc.auth;
2153
2154 return ac->ops->verify_authorizer_reply(ac, o->o_authorizer, len);
2155}
2156
9bd2e6f8
SW		 2157static int invalidate_authorizer(struct ceph_connection *con)
2158{
2159 struct ceph_osd *o = con->private;
2160 struct ceph_osd_client *osdc = o->o_osdc;
2161 struct ceph_auth_client *ac = osdc->client->monc.auth;
2162
2163 if (ac->ops->invalidate_authorizer)
2164 ac->ops->invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
2165
2166 return ceph_monc_validate_auth(&osdc->client->monc);
2167}
4e7a5dcd 2168
9e32789f 2169static const struct ceph_connection_operations osd_con_ops = {
f24e9980
SW		 2170 .get = get_osd_con,
2171 .put = put_osd_con,
2172 .dispatch = dispatch,
4e7a5dcd
SW		 2173 .get_authorizer = get_authorizer,
2174 .verify_authorizer_reply = verify_authorizer_reply,
9bd2e6f8 2175 .invalidate_authorizer = invalidate_authorizer,
f24e9980 2176 .alloc_msg = alloc_msg,
81b024e7 2177 .fault = osd_reset,
f24e9980 2178};
Linux 6.x block layer and io_uring tree(s)