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