projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
libceph: move linger requests sooner in kick_requests()
[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);
56e925b6
SW		 32static void __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
d63b77f4 55int ceph_calc_raw_layout(struct ceph_osd_client *osdc,
3499e8a5
YS		 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 64 u64 objoff, objlen; /* extent in object */
d63b77f4 65 int r;
3499e8a5
YS		 66
67 reqhead->snapid = cpu_to_le64(snapid);
68
69 /* object extent? */
d63b77f4
SW		 70 r = ceph_calc_file_object_mapping(layout, off, plen, bno,
71 &objoff, &objlen);
72 if (r < 0)
73 return r;
68b4476b 74 if (*plen < orig_len)
3499e8a5 75 dout(" skipping last %llu, final file extent %llu~%llu\n",
68b4476b 76 orig_len - *plen, off, *plen);
3499e8a5 77
68b4476b
YS		 78 if (op_has_extent(op->op)) {
79 op->extent.offset = objoff;
80 op->extent.length = objlen;
81 }
82 req->r_num_pages = calc_pages_for(off, *plen);
b7495fc2 83 req->r_page_alignment = off & ~PAGE_MASK;
3d14c5d2
YS		 84 if (op->op == CEPH_OSD_OP_WRITE)
85 op->payload_len = *plen;
3499e8a5
YS		 86
87 dout("calc_layout bno=%llx %llu~%llu (%d pages)\n",
88 *bno, objoff, objlen, req->r_num_pages);
d63b77f4 89 return 0;
3499e8a5 90}
3d14c5d2 91EXPORT_SYMBOL(ceph_calc_raw_layout);
3499e8a5 92
f24e9980
SW		 93/*
94 * Implement client access to distributed object storage cluster.
95 *
96 * All data objects are stored within a cluster/cloud of OSDs, or
97 * "object storage devices." (Note that Ceph OSDs have _nothing_ to
98 * do with the T10 OSD extensions to SCSI.) Ceph OSDs are simply
99 * remote daemons serving up and coordinating consistent and safe
100 * access to storage.
101 *
102 * Cluster membership and the mapping of data objects onto storage devices
103 * are described by the osd map.
104 *
105 * We keep track of pending OSD requests (read, write), resubmit
106 * requests to different OSDs when the cluster topology/data layout
107 * change, or retry the affected requests when the communications
108 * channel with an OSD is reset.
109 */
110
111/*
112 * calculate the mapping of a file extent onto an object, and fill out the
113 * request accordingly. shorten extent as necessary if it crosses an
114 * object boundary.
115 *
116 * fill osd op in request message.
117 */
d63b77f4
SW		 118static int calc_layout(struct ceph_osd_client *osdc,
119 struct ceph_vino vino,
120 struct ceph_file_layout *layout,
121 u64 off, u64 *plen,
122 struct ceph_osd_request *req,
123 struct ceph_osd_req_op *op)
f24e9980 124{
f24e9980 125 u64 bno;
d63b77f4 126 int r;
f24e9980 127
d63b77f4
SW		 128 r = ceph_calc_raw_layout(osdc, layout, vino.snap, off,
129 plen, &bno, req, op);
130 if (r < 0)
131 return r;
f24e9980 132
2dab036b 133 snprintf(req->r_oid, sizeof(req->r_oid), "%llx.%08llx", vino.ino, bno);
f24e9980 134 req->r_oid_len = strlen(req->r_oid);
d63b77f4
SW		 135
136 return r;
f24e9980
SW		 137}
138
f24e9980
SW		 139/*
140 * requests
141 */
415e49a9 142void ceph_osdc_release_request(struct kref *kref)
f24e9980 143{
415e49a9
SW		 144 struct ceph_osd_request *req = container_of(kref,
145 struct ceph_osd_request,
146 r_kref);
147
148 if (req->r_request)
149 ceph_msg_put(req->r_request);
0d59ab81 150 if (req->r_con_filling_msg) {
8921d114 151 dout("%s revoking pages %p from con %p\n", __func__,
0d59ab81 152 req->r_pages, req->r_con_filling_msg);
8921d114 153 ceph_msg_revoke_incoming(req->r_reply);
0d47766f 154 req->r_con_filling_msg->ops->put(req->r_con_filling_msg);
350b1c32 155 }
ab8cb34a
AE		 156 if (req->r_reply)
157 ceph_msg_put(req->r_reply);
415e49a9
SW		 158 if (req->r_own_pages)
159 ceph_release_page_vector(req->r_pages,
160 req->r_num_pages);
68b4476b
YS		 161#ifdef CONFIG_BLOCK
162 if (req->r_bio)
163 bio_put(req->r_bio);
164#endif
415e49a9 165 ceph_put_snap_context(req->r_snapc);
68b4476b
YS		 166 if (req->r_trail) {
167 ceph_pagelist_release(req->r_trail);
168 kfree(req->r_trail);
169 }
415e49a9
SW		 170 if (req->r_mempool)
171 mempool_free(req, req->r_osdc->req_mempool);
172 else
173 kfree(req);
f24e9980 174}
3d14c5d2 175EXPORT_SYMBOL(ceph_osdc_release_request);
68b4476b
YS		 176
177static int get_num_ops(struct ceph_osd_req_op *ops, int *needs_trail)
178{
179 int i = 0;
180
181 if (needs_trail)
182 *needs_trail = 0;
183 while (ops[i].op) {
184 if (needs_trail && op_needs_trail(ops[i].op))
185 *needs_trail = 1;
186 i++;
187 }
188
189 return i;
190}
191
3499e8a5
YS		 192struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
193 int flags,
f24e9980 194 struct ceph_snap_context *snapc,
68b4476b 195 struct ceph_osd_req_op *ops,
3499e8a5
YS		 196 bool use_mempool,
197 gfp_t gfp_flags,
68b4476b
YS		 198 struct page **pages,
199 struct bio *bio)
f24e9980
SW		 200{
201 struct ceph_osd_request *req;
202 struct ceph_msg *msg;
68b4476b
YS		 203 int needs_trail;
204 int num_op = get_num_ops(ops, &needs_trail);
205 size_t msg_size = sizeof(struct ceph_osd_request_head);
3499e8a5 206
68b4476b 207 msg_size += num_op*sizeof(struct ceph_osd_op);
f24e9980
SW		 208
209 if (use_mempool) {
3499e8a5 210 req = mempool_alloc(osdc->req_mempool, gfp_flags);
f24e9980
SW		 211 memset(req, 0, sizeof(*req));
212 } else {
3499e8a5 213 req = kzalloc(sizeof(*req), gfp_flags);
f24e9980
SW		 214 }
215 if (req == NULL)
a79832f2 216 return NULL;
f24e9980 217
f24e9980
SW		 218 req->r_osdc = osdc;
219 req->r_mempool = use_mempool;
68b4476b 220
415e49a9 221 kref_init(&req->r_kref);
f24e9980
SW		 222 init_completion(&req->r_completion);
223 init_completion(&req->r_safe_completion);
a978fa20 224 RB_CLEAR_NODE(&req->r_node);
f24e9980 225 INIT_LIST_HEAD(&req->r_unsafe_item);
a40c4f10
YS		 226 INIT_LIST_HEAD(&req->r_linger_item);
227 INIT_LIST_HEAD(&req->r_linger_osd);
935b639a 228 INIT_LIST_HEAD(&req->r_req_lru_item);
cd43045c
SW		 229 INIT_LIST_HEAD(&req->r_osd_item);
230
f24e9980
SW		 231 req->r_flags = flags;
232
233 WARN_ON((flags & (CEPH_OSD_FLAG_READ|CEPH_OSD_FLAG_WRITE)) == 0);
234
c16e7869
SW		 235 /* create reply message */
236 if (use_mempool)
237 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0);
238 else
239 msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY,
b61c2763 240 OSD_OPREPLY_FRONT_LEN, gfp_flags, true);
a79832f2 241 if (!msg) {
c16e7869 242 ceph_osdc_put_request(req);
a79832f2 243 return NULL;
c16e7869
SW		 244 }
245 req->r_reply = msg;
246
68b4476b
YS		 247 /* allocate space for the trailing data */
248 if (needs_trail) {
249 req->r_trail = kmalloc(sizeof(struct ceph_pagelist), gfp_flags);
250 if (!req->r_trail) {
251 ceph_osdc_put_request(req);
252 return NULL;
253 }
254 ceph_pagelist_init(req->r_trail);
255 }
d50b409f 256
c16e7869 257 /* create request message; allow space for oid */
224736d9 258 msg_size += MAX_OBJ_NAME_SIZE;
f24e9980
SW		 259 if (snapc)
260 msg_size += sizeof(u64) * snapc->num_snaps;
261 if (use_mempool)
8f3bc053 262 msg = ceph_msgpool_get(&osdc->msgpool_op, 0);
f24e9980 263 else
b61c2763 264 msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp_flags, true);
a79832f2 265 if (!msg) {
f24e9980 266 ceph_osdc_put_request(req);
a79832f2 267 return NULL;
f24e9980 268 }
68b4476b 269
f24e9980 270 memset(msg->front.iov_base, 0, msg->front.iov_len);
3499e8a5
YS		 271
272 req->r_request = msg;
273 req->r_pages = pages;
68b4476b
YS		 274#ifdef CONFIG_BLOCK
275 if (bio) {
276 req->r_bio = bio;
277 bio_get(req->r_bio);
278 }
279#endif
3499e8a5
YS		 280
281 return req;
282}
3d14c5d2 283EXPORT_SYMBOL(ceph_osdc_alloc_request);
3499e8a5 284
68b4476b
YS		 285static void osd_req_encode_op(struct ceph_osd_request *req,
286 struct ceph_osd_op *dst,
287 struct ceph_osd_req_op *src)
288{
289 dst->op = cpu_to_le16(src->op);
290
065a68f9 291 switch (src->op) {
68b4476b
YS		 292 case CEPH_OSD_OP_READ:
293 case CEPH_OSD_OP_WRITE:
294 dst->extent.offset =
295 cpu_to_le64(src->extent.offset);
296 dst->extent.length =
297 cpu_to_le64(src->extent.length);
298 dst->extent.truncate_size =
299 cpu_to_le64(src->extent.truncate_size);
300 dst->extent.truncate_seq =
301 cpu_to_le32(src->extent.truncate_seq);
302 break;
303
304 case CEPH_OSD_OP_GETXATTR:
305 case CEPH_OSD_OP_SETXATTR:
306 case CEPH_OSD_OP_CMPXATTR:
307 BUG_ON(!req->r_trail);
308
309 dst->xattr.name_len = cpu_to_le32(src->xattr.name_len);
310 dst->xattr.value_len = cpu_to_le32(src->xattr.value_len);
311 dst->xattr.cmp_op = src->xattr.cmp_op;
312 dst->xattr.cmp_mode = src->xattr.cmp_mode;
313 ceph_pagelist_append(req->r_trail, src->xattr.name,
314 src->xattr.name_len);
315 ceph_pagelist_append(req->r_trail, src->xattr.val,
316 src->xattr.value_len);
317 break;
ae1533b6
YS		 318 case CEPH_OSD_OP_CALL:
319 BUG_ON(!req->r_trail);
320
321 dst->cls.class_len = src->cls.class_len;
322 dst->cls.method_len = src->cls.method_len;
323 dst->cls.indata_len = cpu_to_le32(src->cls.indata_len);
324
325 ceph_pagelist_append(req->r_trail, src->cls.class_name,
326 src->cls.class_len);
327 ceph_pagelist_append(req->r_trail, src->cls.method_name,
328 src->cls.method_len);
329 ceph_pagelist_append(req->r_trail, src->cls.indata,
330 src->cls.indata_len);
331 break;
332 case CEPH_OSD_OP_ROLLBACK:
333 dst->snap.snapid = cpu_to_le64(src->snap.snapid);
334 break;
68b4476b
YS		 335 case CEPH_OSD_OP_STARTSYNC:
336 break;
a40c4f10
YS		 337 case CEPH_OSD_OP_NOTIFY:
338 {
339 __le32 prot_ver = cpu_to_le32(src->watch.prot_ver);
340 __le32 timeout = cpu_to_le32(src->watch.timeout);
341
342 BUG_ON(!req->r_trail);
343
344 ceph_pagelist_append(req->r_trail,
345 &prot_ver, sizeof(prot_ver));
346 ceph_pagelist_append(req->r_trail,
347 &timeout, sizeof(timeout));
348 }
349 case CEPH_OSD_OP_NOTIFY_ACK:
350 case CEPH_OSD_OP_WATCH:
351 dst->watch.cookie = cpu_to_le64(src->watch.cookie);
352 dst->watch.ver = cpu_to_le64(src->watch.ver);
353 dst->watch.flag = src->watch.flag;
354 break;
68b4476b
YS		 355 default:
356 pr_err("unrecognized osd opcode %d\n", dst->op);
357 WARN_ON(1);
358 break;
359 }
360 dst->payload_len = cpu_to_le32(src->payload_len);
361}
362
3499e8a5
YS		 363/*
364 * build new request AND message
365 *
366 */
367void ceph_osdc_build_request(struct ceph_osd_request *req,
68b4476b
YS		 368 u64 off, u64 *plen,
369 struct ceph_osd_req_op *src_ops,
370 struct ceph_snap_context *snapc,
371 struct timespec *mtime,
372 const char *oid,
373 int oid_len)
3499e8a5
YS		 374{
375 struct ceph_msg *msg = req->r_request;
376 struct ceph_osd_request_head *head;
68b4476b 377 struct ceph_osd_req_op *src_op;
3499e8a5
YS		 378 struct ceph_osd_op *op;
379 void *p;
68b4476b 380 int num_op = get_num_ops(src_ops, NULL);
3499e8a5 381 size_t msg_size = sizeof(*head) + num_op*sizeof(*op);
3499e8a5 382 int flags = req->r_flags;
68b4476b
YS		 383 u64 data_len = 0;
384 int i;
3499e8a5 385
f24e9980
SW		 386 head = msg->front.iov_base;
387 op = (void *)(head + 1);
388 p = (void *)(op + num_op);
389
f24e9980
SW		 390 req->r_snapc = ceph_get_snap_context(snapc);
391
392 head->client_inc = cpu_to_le32(1); /* always, for now. */
393 head->flags = cpu_to_le32(flags);
394 if (flags & CEPH_OSD_FLAG_WRITE)
395 ceph_encode_timespec(&head->mtime, mtime);
396 head->num_ops = cpu_to_le16(num_op);
f24e9980 397
f24e9980
SW		 398
399 /* fill in oid */
3499e8a5
YS		 400 head->object_len = cpu_to_le32(oid_len);
401 memcpy(p, oid, oid_len);
402 p += oid_len;
f24e9980 403
68b4476b
YS		 404 src_op = src_ops;
405 while (src_op->op) {
406 osd_req_encode_op(req, op, src_op);
407 src_op++;
f24e9980 408 op++;
f24e9980 409 }
68b4476b
YS		 410
411 if (req->r_trail)
412 data_len += req->r_trail->length;
413
f24e9980
SW		 414 if (snapc) {
415 head->snap_seq = cpu_to_le64(snapc->seq);
416 head->num_snaps = cpu_to_le32(snapc->num_snaps);
417 for (i = 0; i < snapc->num_snaps; i++) {
418 put_unaligned_le64(snapc->snaps[i], p);
419 p += sizeof(u64);
420 }
421 }
422
68b4476b
YS		 423 if (flags & CEPH_OSD_FLAG_WRITE) {
424 req->r_request->hdr.data_off = cpu_to_le16(off);
425 req->r_request->hdr.data_len = cpu_to_le32(*plen + data_len);
426 } else if (data_len) {
427 req->r_request->hdr.data_off = 0;
428 req->r_request->hdr.data_len = cpu_to_le32(data_len);
429 }
430
c5c6b19d
SW		 431 req->r_request->page_alignment = req->r_page_alignment;
432
f24e9980 433 BUG_ON(p > msg->front.iov_base + msg->front.iov_len);
6f863e71
SW		 434 msg_size = p - msg->front.iov_base;
435 msg->front.iov_len = msg_size;
436 msg->hdr.front_len = cpu_to_le32(msg_size);
3499e8a5
YS		 437 return;
438}
3d14c5d2 439EXPORT_SYMBOL(ceph_osdc_build_request);
3499e8a5
YS		 440
441/*
442 * build new request AND message, calculate layout, and adjust file
443 * extent as needed.
444 *
445 * if the file was recently truncated, we include information about its
446 * old and new size so that the object can be updated appropriately. (we
447 * avoid synchronously deleting truncated objects because it's slow.)
448 *
449 * if @do_sync, include a 'startsync' command so that the osd will flush
450 * data quickly.
451 */
452struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
453 struct ceph_file_layout *layout,
454 struct ceph_vino vino,
455 u64 off, u64 *plen,
456 int opcode, int flags,
457 struct ceph_snap_context *snapc,
458 int do_sync,
459 u32 truncate_seq,
460 u64 truncate_size,
461 struct timespec *mtime,
b7495fc2
SW		 462 bool use_mempool, int num_reply,
463 int page_align)
3499e8a5 464{
68b4476b
YS		 465 struct ceph_osd_req_op ops[3];
466 struct ceph_osd_request *req;
6816282d 467 int r;
68b4476b
YS		 468
469 ops[0].op = opcode;
470 ops[0].extent.truncate_seq = truncate_seq;
471 ops[0].extent.truncate_size = truncate_size;
472 ops[0].payload_len = 0;
473
474 if (do_sync) {
475 ops[1].op = CEPH_OSD_OP_STARTSYNC;
476 ops[1].payload_len = 0;
477 ops[2].op = 0;
478 } else
479 ops[1].op = 0;
480
481 req = ceph_osdc_alloc_request(osdc, flags,
482 snapc, ops,
3499e8a5 483 use_mempool,
68b4476b 484 GFP_NOFS, NULL, NULL);
4ad12621 485 if (!req)
6816282d 486 return ERR_PTR(-ENOMEM);
3499e8a5
YS		 487
488 /* calculate max write size */
6816282d
SW		 489 r = calc_layout(osdc, vino, layout, off, plen, req, ops);
490 if (r < 0)
491 return ERR_PTR(r);
3499e8a5
YS		 492 req->r_file_layout = *layout; /* keep a copy */
493
9bb0ce2b
SW		 494 /* in case it differs from natural (file) alignment that
495 calc_layout filled in for us */
496 req->r_num_pages = calc_pages_for(page_align, *plen);
b7495fc2
SW		 497 req->r_page_alignment = page_align;
498
68b4476b
YS		 499 ceph_osdc_build_request(req, off, plen, ops,
500 snapc,
3499e8a5
YS		 501 mtime,
502 req->r_oid, req->r_oid_len);
503
f24e9980
SW		 504 return req;
505}
3d14c5d2 506EXPORT_SYMBOL(ceph_osdc_new_request);
f24e9980
SW		 507
508/*
509 * We keep osd requests in an rbtree, sorted by ->r_tid.
510 */
511static void __insert_request(struct ceph_osd_client *osdc,
512 struct ceph_osd_request *new)
513{
514 struct rb_node **p = &osdc->requests.rb_node;
515 struct rb_node *parent = NULL;
516 struct ceph_osd_request *req = NULL;
517
518 while (*p) {
519 parent = *p;
520 req = rb_entry(parent, struct ceph_osd_request, r_node);
521 if (new->r_tid < req->r_tid)
522 p = &(*p)->rb_left;
523 else if (new->r_tid > req->r_tid)
524 p = &(*p)->rb_right;
525 else
526 BUG();
527 }
528
529 rb_link_node(&new->r_node, parent, p);
530 rb_insert_color(&new->r_node, &osdc->requests);
531}
532
533static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc,
534 u64 tid)
535{
536 struct ceph_osd_request *req;
537 struct rb_node *n = osdc->requests.rb_node;
538
539 while (n) {
540 req = rb_entry(n, struct ceph_osd_request, r_node);
541 if (tid < req->r_tid)
542 n = n->rb_left;
543 else if (tid > req->r_tid)
544 n = n->rb_right;
545 else
546 return req;
547 }
548 return NULL;
549}
550
551static struct ceph_osd_request *
552__lookup_request_ge(struct ceph_osd_client *osdc,
553 u64 tid)
554{
555 struct ceph_osd_request *req;
556 struct rb_node *n = osdc->requests.rb_node;
557
558 while (n) {
559 req = rb_entry(n, struct ceph_osd_request, r_node);
560 if (tid < req->r_tid) {
561 if (!n->rb_left)
562 return req;
563 n = n->rb_left;
564 } else if (tid > req->r_tid) {
565 n = n->rb_right;
566 } else {
567 return req;
568 }
569 }
570 return NULL;
571}
572
6f6c7006
SW		 573/*
574 * Resubmit requests pending on the given osd.
575 */
576static void __kick_osd_requests(struct ceph_osd_client *osdc,
577 struct ceph_osd *osd)
578{
a40c4f10 579 struct ceph_osd_request *req, *nreq;
6f6c7006
SW		 580 int err;
581
582 dout("__kick_osd_requests osd%d\n", osd->o_osd);
583 err = __reset_osd(osdc, osd);
685a7555 584 if (err)
6f6c7006
SW		 585 return;
586
587 list_for_each_entry(req, &osd->o_requests, r_osd_item) {
588 list_move(&req->r_req_lru_item, &osdc->req_unsent);
589 dout("requeued %p tid %llu osd%d\n", req, req->r_tid,
590 osd->o_osd);
a40c4f10
YS		 591 if (!req->r_linger)
592 req->r_flags |= CEPH_OSD_FLAG_RETRY;
593 }
594
595 list_for_each_entry_safe(req, nreq, &osd->o_linger_requests,
596 r_linger_osd) {
77f38e0e
SW		 597 /*
598 * reregister request prior to unregistering linger so
599 * that r_osd is preserved.
600 */
601 BUG_ON(!list_empty(&req->r_req_lru_item));
a40c4f10 602 __register_request(osdc, req);
77f38e0e
SW		 603 list_add(&req->r_req_lru_item, &osdc->req_unsent);
604 list_add(&req->r_osd_item, &req->r_osd->o_requests);
605 __unregister_linger_request(osdc, req);
a40c4f10
YS		 606 dout("requeued lingering %p tid %llu osd%d\n", req, req->r_tid,
607 osd->o_osd);
6f6c7006
SW		 608 }
609}
610
f24e9980 611/*
81b024e7 612 * If the osd connection drops, we need to resubmit all requests.
f24e9980
SW		 613 */
614static void osd_reset(struct ceph_connection *con)
615{
616 struct ceph_osd *osd = con->private;
617 struct ceph_osd_client *osdc;
618
619 if (!osd)
620 return;
621 dout("osd_reset osd%d\n", osd->o_osd);
622 osdc = osd->o_osdc;
f24e9980 623 down_read(&osdc->map_sem);
83aff95e
SW		 624 mutex_lock(&osdc->request_mutex);
625 __kick_osd_requests(osdc, osd);
626 mutex_unlock(&osdc->request_mutex);
6f6c7006 627 send_queued(osdc);
f24e9980
SW		 628 up_read(&osdc->map_sem);
629}
630
631/*
632 * Track open sessions with osds.
633 */
e10006f8 634static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
f24e9980
SW		 635{
636 struct ceph_osd *osd;
637
638 osd = kzalloc(sizeof(*osd), GFP_NOFS);
639 if (!osd)
640 return NULL;
641
642 atomic_set(&osd->o_ref, 1);
643 osd->o_osdc = osdc;
e10006f8 644 osd->o_osd = onum;
f407731d 645 RB_CLEAR_NODE(&osd->o_node);
f24e9980 646 INIT_LIST_HEAD(&osd->o_requests);
a40c4f10 647 INIT_LIST_HEAD(&osd->o_linger_requests);
f5a2041b 648 INIT_LIST_HEAD(&osd->o_osd_lru);
f24e9980
SW		 649 osd->o_incarnation = 1;
650
b7a9e5dd 651 ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
4e7a5dcd 652
422d2cb8 653 INIT_LIST_HEAD(&osd->o_keepalive_item);
f24e9980
SW		 654 return osd;
655}
656
657static struct ceph_osd *get_osd(struct ceph_osd *osd)
658{
659 if (atomic_inc_not_zero(&osd->o_ref)) {
660 dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
661 atomic_read(&osd->o_ref));
662 return osd;
663 } else {
664 dout("get_osd %p FAIL\n", osd);
665 return NULL;
666 }
667}
668
669static void put_osd(struct ceph_osd *osd)
670{
671 dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
672 atomic_read(&osd->o_ref) - 1);
a255651d 673 if (atomic_dec_and_test(&osd->o_ref) && osd->o_auth.authorizer) {
79494d1b
SW		 674 struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth;
675
a255651d 676 if (ac->ops && ac->ops->destroy_authorizer)
6c4a1915 677 ac->ops->destroy_authorizer(ac, osd->o_auth.authorizer);
f24e9980 678 kfree(osd);
79494d1b 679 }
f24e9980
SW		 680}
681
682/*
683 * remove an osd from our map
684 */
f5a2041b 685static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
f24e9980 686{
f5a2041b 687 dout("__remove_osd %p\n", osd);
f24e9980
SW		 688 BUG_ON(!list_empty(&osd->o_requests));
689 rb_erase(&osd->o_node, &osdc->osds);
f5a2041b 690 list_del_init(&osd->o_osd_lru);
f24e9980
SW		 691 ceph_con_close(&osd->o_con);
692 put_osd(osd);
693}
694
aca420bc
SW		 695static void remove_all_osds(struct ceph_osd_client *osdc)
696{
048a9d2d 697 dout("%s %p\n", __func__, osdc);
aca420bc
SW		 698 mutex_lock(&osdc->request_mutex);
699 while (!RB_EMPTY_ROOT(&osdc->osds)) {
700 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
701 struct ceph_osd, o_node);
702 __remove_osd(osdc, osd);
703 }
704 mutex_unlock(&osdc->request_mutex);
705}
706
f5a2041b
YS		 707static void __move_osd_to_lru(struct ceph_osd_client *osdc,
708 struct ceph_osd *osd)
709{
710 dout("__move_osd_to_lru %p\n", osd);
711 BUG_ON(!list_empty(&osd->o_osd_lru));
712 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
3d14c5d2 713 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl * HZ;
f5a2041b
YS		 714}
715
716static void __remove_osd_from_lru(struct ceph_osd *osd)
717{
718 dout("__remove_osd_from_lru %p\n", osd);
719 if (!list_empty(&osd->o_osd_lru))
720 list_del_init(&osd->o_osd_lru);
721}
722
aca420bc 723static void remove_old_osds(struct ceph_osd_client *osdc)
f5a2041b
YS		 724{
725 struct ceph_osd *osd, *nosd;
726
727 dout("__remove_old_osds %p\n", osdc);
728 mutex_lock(&osdc->request_mutex);
729 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
aca420bc 730 if (time_before(jiffies, osd->lru_ttl))
f5a2041b
YS		 731 break;
732 __remove_osd(osdc, osd);
733 }
734 mutex_unlock(&osdc->request_mutex);
735}
736
f24e9980
SW		 737/*
738 * reset osd connect
739 */
f5a2041b 740static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
f24e9980 741{
87b315a5 742 struct ceph_osd_request *req;
f24e9980
SW		 743 int ret = 0;
744
f5a2041b 745 dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
a40c4f10
YS		 746 if (list_empty(&osd->o_requests) &&
747 list_empty(&osd->o_linger_requests)) {
f5a2041b 748 __remove_osd(osdc, osd);
685a7555 749 ret = -ENODEV;
87b315a5
SW		 750 } else if (memcmp(&osdc->osdmap->osd_addr[osd->o_osd],
751 &osd->o_con.peer_addr,
752 sizeof(osd->o_con.peer_addr)) == 0 &&
753 !ceph_con_opened(&osd->o_con)) {
754 dout(" osd addr hasn't changed and connection never opened,"
755 " letting msgr retry");
756 /* touch each r_stamp for handle_timeout()'s benfit */
757 list_for_each_entry(req, &osd->o_requests, r_osd_item)
758 req->r_stamp = jiffies;
759 ret = -EAGAIN;
f24e9980
SW		 760 } else {
761 ceph_con_close(&osd->o_con);
b7a9e5dd
SW		 762 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd,
763 &osdc->osdmap->osd_addr[osd->o_osd]);
f24e9980
SW		 764 osd->o_incarnation++;
765 }
766 return ret;
767}
768
769static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new)
770{
771 struct rb_node **p = &osdc->osds.rb_node;
772 struct rb_node *parent = NULL;
773 struct ceph_osd *osd = NULL;
774
aca420bc 775 dout("__insert_osd %p osd%d\n", new, new->o_osd);
f24e9980
SW		 776 while (*p) {
777 parent = *p;
778 osd = rb_entry(parent, struct ceph_osd, o_node);
779 if (new->o_osd < osd->o_osd)
780 p = &(*p)->rb_left;
781 else if (new->o_osd > osd->o_osd)
782 p = &(*p)->rb_right;
783 else
784 BUG();
785 }
786
787 rb_link_node(&new->o_node, parent, p);
788 rb_insert_color(&new->o_node, &osdc->osds);
789}
790
791static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o)
792{
793 struct ceph_osd *osd;
794 struct rb_node *n = osdc->osds.rb_node;
795
796 while (n) {
797 osd = rb_entry(n, struct ceph_osd, o_node);
798 if (o < osd->o_osd)
799 n = n->rb_left;
800 else if (o > osd->o_osd)
801 n = n->rb_right;
802 else
803 return osd;
804 }
805 return NULL;
806}
807
422d2cb8
YS		 808static void __schedule_osd_timeout(struct ceph_osd_client *osdc)
809{
810 schedule_delayed_work(&osdc->timeout_work,
3d14c5d2 811 osdc->client->options->osd_keepalive_timeout * HZ);
422d2cb8
YS		 812}
813
814static void __cancel_osd_timeout(struct ceph_osd_client *osdc)
815{
816 cancel_delayed_work(&osdc->timeout_work);
817}
f24e9980
SW		 818
819/*
820 * Register request, assign tid. If this is the first request, set up
821 * the timeout event.
822 */
a40c4f10
YS		 823static void __register_request(struct ceph_osd_client *osdc,
824 struct ceph_osd_request *req)
f24e9980 825{
f24e9980 826 req->r_tid = ++osdc->last_tid;
6df058c0 827 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
77f38e0e 828 dout("__register_request %p tid %lld\n", req, req->r_tid);
f24e9980
SW		 829 __insert_request(osdc, req);
830 ceph_osdc_get_request(req);
831 osdc->num_requests++;
f24e9980 832 if (osdc->num_requests == 1) {
422d2cb8
YS		 833 dout(" first request, scheduling timeout\n");
834 __schedule_osd_timeout(osdc);
f24e9980 835 }
a40c4f10
YS		 836}
837
838static void register_request(struct ceph_osd_client *osdc,
839 struct ceph_osd_request *req)
840{
841 mutex_lock(&osdc->request_mutex);
842 __register_request(osdc, req);
f24e9980
SW		 843 mutex_unlock(&osdc->request_mutex);
844}
845
846/*
847 * called under osdc->request_mutex
848 */
849static void __unregister_request(struct ceph_osd_client *osdc,
850 struct ceph_osd_request *req)
851{
35f9f8a0
SW		 852 if (RB_EMPTY_NODE(&req->r_node)) {
853 dout("__unregister_request %p tid %lld not registered\n",
854 req, req->r_tid);
855 return;
856 }
857
f24e9980
SW		 858 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
859 rb_erase(&req->r_node, &osdc->requests);
860 osdc->num_requests--;
861
0ba6478d
SW		 862 if (req->r_osd) {
863 /* make sure the original request isn't in flight. */
6740a845 864 ceph_msg_revoke(req->r_request);
0ba6478d
SW		 865
866 list_del_init(&req->r_osd_item);
a40c4f10
YS		 867 if (list_empty(&req->r_osd->o_requests) &&
868 list_empty(&req->r_osd->o_linger_requests)) {
869 dout("moving osd to %p lru\n", req->r_osd);
f5a2041b 870 __move_osd_to_lru(osdc, req->r_osd);
a40c4f10 871 }
fbdb9190 872 if (list_empty(&req->r_linger_item))
a40c4f10 873 req->r_osd = NULL;
0ba6478d 874 }
f24e9980 875
7d5f2481 876 list_del_init(&req->r_req_lru_item);
f24e9980
SW		 877 ceph_osdc_put_request(req);
878
422d2cb8
YS		 879 if (osdc->num_requests == 0) {
880 dout(" no requests, canceling timeout\n");
881 __cancel_osd_timeout(osdc);
f24e9980
SW		 882 }
883}
884
885/*
886 * Cancel a previously queued request message
887 */
888static void __cancel_request(struct ceph_osd_request *req)
889{
6bc18876 890 if (req->r_sent && req->r_osd) {
6740a845 891 ceph_msg_revoke(req->r_request);
f24e9980
SW		 892 req->r_sent = 0;
893 }
894}
895
a40c4f10
YS		 896static void __register_linger_request(struct ceph_osd_client *osdc,
897 struct ceph_osd_request *req)
898{
899 dout("__register_linger_request %p\n", req);
900 list_add_tail(&req->r_linger_item, &osdc->req_linger);
6194ea89
SW		 901 if (req->r_osd)
902 list_add_tail(&req->r_linger_osd,
903 &req->r_osd->o_linger_requests);
a40c4f10
YS		 904}
905
906static void __unregister_linger_request(struct ceph_osd_client *osdc,
907 struct ceph_osd_request *req)
908{
909 dout("__unregister_linger_request %p\n", req);
61c74035 910 list_del_init(&req->r_linger_item);
a40c4f10 911 if (req->r_osd) {
a40c4f10
YS		 912 list_del_init(&req->r_linger_osd);
913
914 if (list_empty(&req->r_osd->o_requests) &&
915 list_empty(&req->r_osd->o_linger_requests)) {
916 dout("moving osd to %p lru\n", req->r_osd);
917 __move_osd_to_lru(osdc, req->r_osd);
918 }
fbdb9190
SW		 919 if (list_empty(&req->r_osd_item))
920 req->r_osd = NULL;
a40c4f10
YS		 921 }
922}
923
924void ceph_osdc_unregister_linger_request(struct ceph_osd_client *osdc,
925 struct ceph_osd_request *req)
926{
927 mutex_lock(&osdc->request_mutex);
928 if (req->r_linger) {
929 __unregister_linger_request(osdc, req);
930 ceph_osdc_put_request(req);
931 }
932 mutex_unlock(&osdc->request_mutex);
933}
934EXPORT_SYMBOL(ceph_osdc_unregister_linger_request);
935
936void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
937 struct ceph_osd_request *req)
938{
939 if (!req->r_linger) {
940 dout("set_request_linger %p\n", req);
941 req->r_linger = 1;
942 /*
943 * caller is now responsible for calling
944 * unregister_linger_request
945 */
946 ceph_osdc_get_request(req);
947 }
948}
949EXPORT_SYMBOL(ceph_osdc_set_request_linger);
950
f24e9980
SW		 951/*
952 * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
953 * (as needed), and set the request r_osd appropriately. If there is
25985edc 954 * no up osd, set r_osd to NULL. Move the request to the appropriate list
6f6c7006 955 * (unsent, homeless) or leave on in-flight lru.
f24e9980
SW		 956 *
957 * Return 0 if unchanged, 1 if changed, or negative on error.
958 *
959 * Caller should hold map_sem for read and request_mutex.
960 */
6f6c7006 961static int __map_request(struct ceph_osd_client *osdc,
38d6453c 962 struct ceph_osd_request *req, int force_resend)
f24e9980
SW		 963{
964 struct ceph_osd_request_head *reqhead = req->r_request->front.iov_base;
51042122 965 struct ceph_pg pgid;
d85b7056
SW		 966 int acting[CEPH_PG_MAX_SIZE];
967 int o = -1, num = 0;
f24e9980 968 int err;
f24e9980 969
6f6c7006 970 dout("map_request %p tid %lld\n", req, req->r_tid);
f24e9980
SW		 971 err = ceph_calc_object_layout(&reqhead->layout, req->r_oid,
972 &req->r_file_layout, osdc->osdmap);
6f6c7006
SW		 973 if (err) {
974 list_move(&req->r_req_lru_item, &osdc->req_notarget);
f24e9980 975 return err;
6f6c7006 976 }
51042122 977 pgid = reqhead->layout.ol_pgid;
7740a42f
SW		 978 req->r_pgid = pgid;
979
d85b7056
SW		 980 err = ceph_calc_pg_acting(osdc->osdmap, pgid, acting);
981 if (err > 0) {
982 o = acting[0];
983 num = err;
984 }
f24e9980 985
38d6453c
SW		 986 if ((!force_resend &&
987 req->r_osd && req->r_osd->o_osd == o &&
d85b7056
SW		 988 req->r_sent >= req->r_osd->o_incarnation &&
989 req->r_num_pg_osds == num &&
990 memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
f24e9980
SW		 991 (req->r_osd == NULL && o == -1))
992 return 0; /* no change */
993
6f6c7006 994 dout("map_request tid %llu pgid %d.%x osd%d (was osd%d)\n",
51042122 995 req->r_tid, le32_to_cpu(pgid.pool), le16_to_cpu(pgid.ps), o,
f24e9980
SW		 996 req->r_osd ? req->r_osd->o_osd : -1);
997
d85b7056
SW		 998 /* record full pg acting set */
999 memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
1000 req->r_num_pg_osds = num;
1001
f24e9980
SW		 1002 if (req->r_osd) {
1003 __cancel_request(req);
1004 list_del_init(&req->r_osd_item);
f24e9980
SW		 1005 req->r_osd = NULL;
1006 }
1007
1008 req->r_osd = __lookup_osd(osdc, o);
1009 if (!req->r_osd && o >= 0) {
c99eb1c7 1010 err = -ENOMEM;
e10006f8 1011 req->r_osd = create_osd(osdc, o);
6f6c7006
SW		 1012 if (!req->r_osd) {
1013 list_move(&req->r_req_lru_item, &osdc->req_notarget);
c99eb1c7 1014 goto out;
6f6c7006 1015 }
f24e9980 1016
6f6c7006 1017 dout("map_request osd %p is osd%d\n", req->r_osd, o);
f24e9980
SW		 1018 __insert_osd(osdc, req->r_osd);
1019
b7a9e5dd
SW		 1020 ceph_con_open(&req->r_osd->o_con,
1021 CEPH_ENTITY_TYPE_OSD, o,
1022 &osdc->osdmap->osd_addr[o]);
f24e9980
SW		 1023 }
1024
f5a2041b
YS		 1025 if (req->r_osd) {
1026 __remove_osd_from_lru(req->r_osd);
f24e9980 1027 list_add(&req->r_osd_item, &req->r_osd->o_requests);
6f6c7006
SW		 1028 list_move(&req->r_req_lru_item, &osdc->req_unsent);
1029 } else {
1030 list_move(&req->r_req_lru_item, &osdc->req_notarget);
f5a2041b 1031 }
d85b7056 1032 err = 1; /* osd or pg changed */
f24e9980
SW		 1033
1034out:
f24e9980
SW		 1035 return err;
1036}
1037
1038/*
1039 * caller should hold map_sem (for read) and request_mutex
1040 */
56e925b6
SW		 1041static void __send_request(struct ceph_osd_client *osdc,
1042 struct ceph_osd_request *req)
f24e9980
SW		 1043{
1044 struct ceph_osd_request_head *reqhead;
f24e9980
SW		 1045
1046 dout("send_request %p tid %llu to osd%d flags %d\n",
1047 req, req->r_tid, req->r_osd->o_osd, req->r_flags);
1048
1049 reqhead = req->r_request->front.iov_base;
1050 reqhead->osdmap_epoch = cpu_to_le32(osdc->osdmap->epoch);
1051 reqhead->flags |= cpu_to_le32(req->r_flags); /* e.g., RETRY */
1052 reqhead->reassert_version = req->r_reassert_version;
1053
3dd72fc0 1054 req->r_stamp = jiffies;
07a27e22 1055 list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
f24e9980
SW		 1056
1057 ceph_msg_get(req->r_request); /* send consumes a ref */
1058 ceph_con_send(&req->r_osd->o_con, req->r_request);
1059 req->r_sent = req->r_osd->o_incarnation;
f24e9980
SW		 1060}
1061
6f6c7006
SW		 1062/*
1063 * Send any requests in the queue (req_unsent).
1064 */
1065static void send_queued(struct ceph_osd_client *osdc)
1066{
1067 struct ceph_osd_request *req, *tmp;
1068
1069 dout("send_queued\n");
1070 mutex_lock(&osdc->request_mutex);
1071 list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item) {
1072 __send_request(osdc, req);
1073 }
1074 mutex_unlock(&osdc->request_mutex);
1075}
1076
f24e9980
SW		 1077/*
1078 * Timeout callback, called every N seconds when 1 or more osd
1079 * requests has been active for more than N seconds. When this
1080 * happens, we ping all OSDs with requests who have timed out to
1081 * ensure any communications channel reset is detected. Reset the
1082 * request timeouts another N seconds in the future as we go.
1083 * Reschedule the timeout event another N seconds in future (unless
1084 * there are no open requests).
1085 */
1086static void handle_timeout(struct work_struct *work)
1087{
1088 struct ceph_osd_client *osdc =
1089 container_of(work, struct ceph_osd_client, timeout_work.work);
83aff95e 1090 struct ceph_osd_request *req;
f24e9980 1091 struct ceph_osd *osd;
422d2cb8 1092 unsigned long keepalive =
3d14c5d2 1093 osdc->client->options->osd_keepalive_timeout * HZ;
422d2cb8 1094 struct list_head slow_osds;
f24e9980
SW		 1095 dout("timeout\n");
1096 down_read(&osdc->map_sem);
1097
1098 ceph_monc_request_next_osdmap(&osdc->client->monc);
1099
1100 mutex_lock(&osdc->request_mutex);
f24e9980 1101
422d2cb8
YS		 1102 /*
1103 * ping osds that are a bit slow. this ensures that if there
1104 * is a break in the TCP connection we will notice, and reopen
1105 * a connection with that osd (from the fault callback).
1106 */
1107 INIT_LIST_HEAD(&slow_osds);
1108 list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
3dd72fc0 1109 if (time_before(jiffies, req->r_stamp + keepalive))
422d2cb8
YS		 1110 break;
1111
1112 osd = req->r_osd;
1113 BUG_ON(!osd);
1114 dout(" tid %llu is slow, will send keepalive on osd%d\n",
f24e9980 1115 req->r_tid, osd->o_osd);
422d2cb8
YS		 1116 list_move_tail(&osd->o_keepalive_item, &slow_osds);
1117 }
1118 while (!list_empty(&slow_osds)) {
1119 osd = list_entry(slow_osds.next, struct ceph_osd,
1120 o_keepalive_item);
1121 list_del_init(&osd->o_keepalive_item);
f24e9980
SW		 1122 ceph_con_keepalive(&osd->o_con);
1123 }
1124
422d2cb8 1125 __schedule_osd_timeout(osdc);
f24e9980 1126 mutex_unlock(&osdc->request_mutex);
6f6c7006 1127 send_queued(osdc);
f24e9980
SW		 1128 up_read(&osdc->map_sem);
1129}
1130
f5a2041b
YS		 1131static void handle_osds_timeout(struct work_struct *work)
1132{
1133 struct ceph_osd_client *osdc =
1134 container_of(work, struct ceph_osd_client,
1135 osds_timeout_work.work);
1136 unsigned long delay =
3d14c5d2 1137 osdc->client->options->osd_idle_ttl * HZ >> 2;
f5a2041b
YS		 1138
1139 dout("osds timeout\n");
1140 down_read(&osdc->map_sem);
aca420bc 1141 remove_old_osds(osdc);
f5a2041b
YS		 1142 up_read(&osdc->map_sem);
1143
1144 schedule_delayed_work(&osdc->osds_timeout_work,
1145 round_jiffies_relative(delay));
1146}
1147
25845472
SW		 1148static void complete_request(struct ceph_osd_request *req)
1149{
1150 if (req->r_safe_callback)
1151 req->r_safe_callback(req, NULL);
1152 complete_all(&req->r_safe_completion); /* fsync waiter */
1153}
1154
f24e9980
SW		 1155/*
1156 * handle osd op reply. either call the callback if it is specified,
1157 * or do the completion to wake up the waiting thread.
1158 */
350b1c32
SW		 1159static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg,
1160 struct ceph_connection *con)
f24e9980
SW		 1161{
1162 struct ceph_osd_reply_head *rhead = msg->front.iov_base;
1163 struct ceph_osd_request *req;
1164 u64 tid;
1165 int numops, object_len, flags;
0ceed5db 1166 s32 result;
f24e9980 1167
6df058c0 1168 tid = le64_to_cpu(msg->hdr.tid);
f24e9980
SW		 1169 if (msg->front.iov_len < sizeof(*rhead))
1170 goto bad;
f24e9980
SW		 1171 numops = le32_to_cpu(rhead->num_ops);
1172 object_len = le32_to_cpu(rhead->object_len);
0ceed5db 1173 result = le32_to_cpu(rhead->result);
f24e9980
SW		 1174 if (msg->front.iov_len != sizeof(*rhead) + object_len +
1175 numops * sizeof(struct ceph_osd_op))
1176 goto bad;
0ceed5db 1177 dout("handle_reply %p tid %llu result %d\n", msg, tid, (int)result);
f24e9980
SW		 1178 /* lookup */
1179 mutex_lock(&osdc->request_mutex);
1180 req = __lookup_request(osdc, tid);
1181 if (req == NULL) {
1182 dout("handle_reply tid %llu dne\n", tid);
1183 mutex_unlock(&osdc->request_mutex);
1184 return;
1185 }
1186 ceph_osdc_get_request(req);
1187 flags = le32_to_cpu(rhead->flags);
1188
350b1c32 1189 /*
0d59ab81 1190 * if this connection filled our message, drop our reference now, to
350b1c32
SW		 1191 * avoid a (safe but slower) revoke later.
1192 */
0d59ab81 1193 if (req->r_con_filling_msg == con && req->r_reply == msg) {
c16e7869 1194 dout(" dropping con_filling_msg ref %p\n", con);
0d59ab81 1195 req->r_con_filling_msg = NULL;
0d47766f 1196 con->ops->put(con);
350b1c32
SW		 1197 }
1198
f24e9980 1199 if (!req->r_got_reply) {
95c96174 1200 unsigned int bytes;
f24e9980
SW		 1201
1202 req->r_result = le32_to_cpu(rhead->result);
1203 bytes = le32_to_cpu(msg->hdr.data_len);
1204 dout("handle_reply result %d bytes %d\n", req->r_result,
1205 bytes);
1206 if (req->r_result == 0)
1207 req->r_result = bytes;
1208
1209 /* in case this is a write and we need to replay, */
1210 req->r_reassert_version = rhead->reassert_version;
1211
1212 req->r_got_reply = 1;
1213 } else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) {
1214 dout("handle_reply tid %llu dup ack\n", tid);
34b43a56 1215 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 1216 goto done;
1217 }
1218
1219 dout("handle_reply tid %llu flags %d\n", tid, flags);
1220
a40c4f10
YS		 1221 if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK))
1222 __register_linger_request(osdc, req);
1223
f24e9980 1224 /* either this is a read, or we got the safe response */
0ceed5db
SW		 1225 if (result < 0 ||
1226 (flags & CEPH_OSD_FLAG_ONDISK) ||
f24e9980
SW		 1227 ((flags & CEPH_OSD_FLAG_WRITE) == 0))
1228 __unregister_request(osdc, req);
1229
1230 mutex_unlock(&osdc->request_mutex);
1231
1232 if (req->r_callback)
1233 req->r_callback(req, msg);
1234 else
03066f23 1235 complete_all(&req->r_completion);
f24e9980 1236
25845472
SW		 1237 if (flags & CEPH_OSD_FLAG_ONDISK)
1238 complete_request(req);
f24e9980
SW		 1239
1240done:
a40c4f10 1241 dout("req=%p req->r_linger=%d\n", req, req->r_linger);
f24e9980
SW		 1242 ceph_osdc_put_request(req);
1243 return;
1244
1245bad:
1246 pr_err("corrupt osd_op_reply got %d %d expected %d\n",
1247 (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len),
1248 (int)sizeof(*rhead));
9ec7cab1 1249 ceph_msg_dump(msg);
f24e9980
SW		 1250}
1251
6f6c7006 1252static void reset_changed_osds(struct ceph_osd_client *osdc)
f24e9980 1253{
f24e9980 1254 struct rb_node *p, *n;
f24e9980 1255
6f6c7006
SW		 1256 for (p = rb_first(&osdc->osds); p; p = n) {
1257 struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
f24e9980 1258
6f6c7006
SW		 1259 n = rb_next(p);
1260 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
1261 memcmp(&osd->o_con.peer_addr,
1262 ceph_osd_addr(osdc->osdmap,
1263 osd->o_osd),
1264 sizeof(struct ceph_entity_addr)) != 0)
1265 __reset_osd(osdc, osd);
f24e9980 1266 }
422d2cb8
YS		 1267}
1268
1269/*
6f6c7006
SW		 1270 * Requeue requests whose mapping to an OSD has changed. If requests map to
1271 * no osd, request a new map.
422d2cb8
YS		 1272 *
1273 * Caller should hold map_sem for read and request_mutex.
1274 */
38d6453c 1275static void kick_requests(struct ceph_osd_client *osdc, int force_resend)
422d2cb8 1276{
a40c4f10 1277 struct ceph_osd_request *req, *nreq;
6f6c7006
SW		 1278 struct rb_node *p;
1279 int needmap = 0;
1280 int err;
422d2cb8 1281
38d6453c 1282 dout("kick_requests %s\n", force_resend ? " (force resend)" : "");
422d2cb8 1283 mutex_lock(&osdc->request_mutex);
6194ea89 1284 for (p = rb_first(&osdc->requests); p; ) {
6f6c7006 1285 req = rb_entry(p, struct ceph_osd_request, r_node);
6194ea89 1286 p = rb_next(p);
ab60b16d
AE		 1287
1288 /*
1289 * For linger requests that have not yet been
1290 * registered, move them to the linger list; they'll
1291 * be sent to the osd in the loop below. Unregister
1292 * the request before re-registering it as a linger
1293 * request to ensure the __map_request() below
1294 * will decide it needs to be sent.
1295 */
1296 if (req->r_linger && list_empty(&req->r_linger_item)) {
1297 dout("%p tid %llu restart on osd%d\n",
1298 req, req->r_tid,
1299 req->r_osd ? req->r_osd->o_osd : -1);
1300 __unregister_request(osdc, req);
1301 __register_linger_request(osdc, req);
1302 continue;
1303 }
1304
38d6453c 1305 err = __map_request(osdc, req, force_resend);
6f6c7006
SW		 1306 if (err < 0)
1307 continue; /* error */
1308 if (req->r_osd == NULL) {
1309 dout("%p tid %llu maps to no osd\n", req, req->r_tid);
1310 needmap++; /* request a newer map */
1311 } else if (err > 0) {
6194ea89
SW		 1312 if (!req->r_linger) {
1313 dout("%p tid %llu requeued on osd%d\n", req,
1314 req->r_tid,
1315 req->r_osd ? req->r_osd->o_osd : -1);
a40c4f10 1316 req->r_flags |= CEPH_OSD_FLAG_RETRY;
6194ea89
SW		 1317 }
1318 }
a40c4f10
YS		 1319 }
1320
1321 list_for_each_entry_safe(req, nreq, &osdc->req_linger,
1322 r_linger_item) {
1323 dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
1324
38d6453c 1325 err = __map_request(osdc, req, force_resend);
ab60b16d 1326 dout("__map_request returned %d\n", err);
a40c4f10
YS		 1327 if (err == 0)
1328 continue; /* no change and no osd was specified */
1329 if (err < 0)
1330 continue; /* hrm! */
1331 if (req->r_osd == NULL) {
1332 dout("tid %llu maps to no valid osd\n", req->r_tid);
1333 needmap++; /* request a newer map */
1334 continue;
6f6c7006 1335 }
a40c4f10
YS		 1336
1337 dout("kicking lingering %p tid %llu osd%d\n", req, req->r_tid,
1338 req->r_osd ? req->r_osd->o_osd : -1);
a40c4f10 1339 __register_request(osdc, req);
c89ce05e 1340 __unregister_linger_request(osdc, req);
6f6c7006 1341 }
f24e9980
SW		 1342 mutex_unlock(&osdc->request_mutex);
1343
1344 if (needmap) {
1345 dout("%d requests for down osds, need new map\n", needmap);
1346 ceph_monc_request_next_osdmap(&osdc->client->monc);
1347 }
422d2cb8 1348}
6f6c7006
SW		 1349
1350
f24e9980
SW		 1351/*
1352 * Process updated osd map.
1353 *
1354 * The message contains any number of incremental and full maps, normally
1355 * indicating some sort of topology change in the cluster. Kick requests
1356 * off to different OSDs as needed.
1357 */
1358void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1359{
1360 void *p, *end, *next;
1361 u32 nr_maps, maplen;
1362 u32 epoch;
1363 struct ceph_osdmap *newmap = NULL, *oldmap;
1364 int err;
1365 struct ceph_fsid fsid;
1366
1367 dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
1368 p = msg->front.iov_base;
1369 end = p + msg->front.iov_len;
1370
1371 /* verify fsid */
1372 ceph_decode_need(&p, end, sizeof(fsid), bad);
1373 ceph_decode_copy(&p, &fsid, sizeof(fsid));
0743304d
SW		 1374 if (ceph_check_fsid(osdc->client, &fsid) < 0)
1375 return;
f24e9980
SW		 1376
1377 down_write(&osdc->map_sem);
1378
1379 /* incremental maps */
1380 ceph_decode_32_safe(&p, end, nr_maps, bad);
1381 dout(" %d inc maps\n", nr_maps);
1382 while (nr_maps > 0) {
1383 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 1384 epoch = ceph_decode_32(&p);
1385 maplen = ceph_decode_32(&p);
f24e9980
SW		 1386 ceph_decode_need(&p, end, maplen, bad);
1387 next = p + maplen;
1388 if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
1389 dout("applying incremental map %u len %d\n",
1390 epoch, maplen);
1391 newmap = osdmap_apply_incremental(&p, next,
1392 osdc->osdmap,
15d9882c 1393 &osdc->client->msgr);
f24e9980
SW		 1394 if (IS_ERR(newmap)) {
1395 err = PTR_ERR(newmap);
1396 goto bad;
1397 }
30dc6381 1398 BUG_ON(!newmap);
f24e9980
SW		 1399 if (newmap != osdc->osdmap) {
1400 ceph_osdmap_destroy(osdc->osdmap);
1401 osdc->osdmap = newmap;
1402 }
38d6453c 1403 kick_requests(osdc, 0);
6f6c7006 1404 reset_changed_osds(osdc);
f24e9980
SW		 1405 } else {
1406 dout("ignoring incremental map %u len %d\n",
1407 epoch, maplen);
1408 }
1409 p = next;
1410 nr_maps--;
1411 }
1412 if (newmap)
1413 goto done;
1414
1415 /* full maps */
1416 ceph_decode_32_safe(&p, end, nr_maps, bad);
1417 dout(" %d full maps\n", nr_maps);
1418 while (nr_maps) {
1419 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 1420 epoch = ceph_decode_32(&p);
1421 maplen = ceph_decode_32(&p);
f24e9980
SW		 1422 ceph_decode_need(&p, end, maplen, bad);
1423 if (nr_maps > 1) {
1424 dout("skipping non-latest full map %u len %d\n",
1425 epoch, maplen);
1426 } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
1427 dout("skipping full map %u len %d, "
1428 "older than our %u\n", epoch, maplen,
1429 osdc->osdmap->epoch);
1430 } else {
38d6453c
SW		 1431 int skipped_map = 0;
1432
f24e9980
SW		 1433 dout("taking full map %u len %d\n", epoch, maplen);
1434 newmap = osdmap_decode(&p, p+maplen);
1435 if (IS_ERR(newmap)) {
1436 err = PTR_ERR(newmap);
1437 goto bad;
1438 }
30dc6381 1439 BUG_ON(!newmap);
f24e9980
SW		 1440 oldmap = osdc->osdmap;
1441 osdc->osdmap = newmap;
38d6453c
SW		 1442 if (oldmap) {
1443 if (oldmap->epoch + 1 < newmap->epoch)
1444 skipped_map = 1;
f24e9980 1445 ceph_osdmap_destroy(oldmap);
38d6453c
SW		 1446 }
1447 kick_requests(osdc, skipped_map);
f24e9980
SW		 1448 }
1449 p += maplen;
1450 nr_maps--;
1451 }
1452
1453done:
1454 downgrade_write(&osdc->map_sem);
1455 ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch);
cd634fb6
SW		 1456
1457 /*
1458 * subscribe to subsequent osdmap updates if full to ensure
1459 * we find out when we are no longer full and stop returning
1460 * ENOSPC.
1461 */
1462 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL))
1463 ceph_monc_request_next_osdmap(&osdc->client->monc);
1464
6f6c7006 1465 send_queued(osdc);
f24e9980 1466 up_read(&osdc->map_sem);
03066f23 1467 wake_up_all(&osdc->client->auth_wq);
f24e9980
SW		 1468 return;
1469
1470bad:
1471 pr_err("osdc handle_map corrupt msg\n");
9ec7cab1 1472 ceph_msg_dump(msg);
f24e9980
SW		 1473 up_write(&osdc->map_sem);
1474 return;
1475}
1476
a40c4f10
YS		 1477/*
1478 * watch/notify callback event infrastructure
1479 *
1480 * These callbacks are used both for watch and notify operations.
1481 */
1482static void __release_event(struct kref *kref)
1483{
1484 struct ceph_osd_event *event =
1485 container_of(kref, struct ceph_osd_event, kref);
1486
1487 dout("__release_event %p\n", event);
1488 kfree(event);
1489}
1490
1491static void get_event(struct ceph_osd_event *event)
1492{
1493 kref_get(&event->kref);
1494}
1495
1496void ceph_osdc_put_event(struct ceph_osd_event *event)
1497{
1498 kref_put(&event->kref, __release_event);
1499}
1500EXPORT_SYMBOL(ceph_osdc_put_event);
1501
1502static void __insert_event(struct ceph_osd_client *osdc,
1503 struct ceph_osd_event *new)
1504{
1505 struct rb_node **p = &osdc->event_tree.rb_node;
1506 struct rb_node *parent = NULL;
1507 struct ceph_osd_event *event = NULL;
1508
1509 while (*p) {
1510 parent = *p;
1511 event = rb_entry(parent, struct ceph_osd_event, node);
1512 if (new->cookie < event->cookie)
1513 p = &(*p)->rb_left;
1514 else if (new->cookie > event->cookie)
1515 p = &(*p)->rb_right;
1516 else
1517 BUG();
1518 }
1519
1520 rb_link_node(&new->node, parent, p);
1521 rb_insert_color(&new->node, &osdc->event_tree);
1522}
1523
1524static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
1525 u64 cookie)
1526{
1527 struct rb_node **p = &osdc->event_tree.rb_node;
1528 struct rb_node *parent = NULL;
1529 struct ceph_osd_event *event = NULL;
1530
1531 while (*p) {
1532 parent = *p;
1533 event = rb_entry(parent, struct ceph_osd_event, node);
1534 if (cookie < event->cookie)
1535 p = &(*p)->rb_left;
1536 else if (cookie > event->cookie)
1537 p = &(*p)->rb_right;
1538 else
1539 return event;
1540 }
1541 return NULL;
1542}
1543
1544static void __remove_event(struct ceph_osd_event *event)
1545{
1546 struct ceph_osd_client *osdc = event->osdc;
1547
1548 if (!RB_EMPTY_NODE(&event->node)) {
1549 dout("__remove_event removed %p\n", event);
1550 rb_erase(&event->node, &osdc->event_tree);
1551 ceph_osdc_put_event(event);
1552 } else {
1553 dout("__remove_event didn't remove %p\n", event);
1554 }
1555}
1556
1557int ceph_osdc_create_event(struct ceph_osd_client *osdc,
1558 void (*event_cb)(u64, u64, u8, void *),
1559 int one_shot, void *data,
1560 struct ceph_osd_event **pevent)
1561{
1562 struct ceph_osd_event *event;
1563
1564 event = kmalloc(sizeof(*event), GFP_NOIO);
1565 if (!event)
1566 return -ENOMEM;
1567
1568 dout("create_event %p\n", event);
1569 event->cb = event_cb;
1570 event->one_shot = one_shot;
1571 event->data = data;
1572 event->osdc = osdc;
1573 INIT_LIST_HEAD(&event->osd_node);
3ee5234d 1574 RB_CLEAR_NODE(&event->node);
a40c4f10
YS		 1575 kref_init(&event->kref); /* one ref for us */
1576 kref_get(&event->kref); /* one ref for the caller */
1577 init_completion(&event->completion);
1578
1579 spin_lock(&osdc->event_lock);
1580 event->cookie = ++osdc->event_count;
1581 __insert_event(osdc, event);
1582 spin_unlock(&osdc->event_lock);
1583
1584 *pevent = event;
1585 return 0;
1586}
1587EXPORT_SYMBOL(ceph_osdc_create_event);
1588
1589void ceph_osdc_cancel_event(struct ceph_osd_event *event)
1590{
1591 struct ceph_osd_client *osdc = event->osdc;
1592
1593 dout("cancel_event %p\n", event);
1594 spin_lock(&osdc->event_lock);
1595 __remove_event(event);
1596 spin_unlock(&osdc->event_lock);
1597 ceph_osdc_put_event(event); /* caller's */
1598}
1599EXPORT_SYMBOL(ceph_osdc_cancel_event);
1600
1601
1602static void do_event_work(struct work_struct *work)
1603{
1604 struct ceph_osd_event_work *event_work =
1605 container_of(work, struct ceph_osd_event_work, work);
1606 struct ceph_osd_event *event = event_work->event;
1607 u64 ver = event_work->ver;
1608 u64 notify_id = event_work->notify_id;
1609 u8 opcode = event_work->opcode;
1610
1611 dout("do_event_work completing %p\n", event);
1612 event->cb(ver, notify_id, opcode, event->data);
1613 complete(&event->completion);
1614 dout("do_event_work completed %p\n", event);
1615 ceph_osdc_put_event(event);
1616 kfree(event_work);
1617}
1618
1619
1620/*
1621 * Process osd watch notifications
1622 */
1623void handle_watch_notify(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1624{
1625 void *p, *end;
1626 u8 proto_ver;
1627 u64 cookie, ver, notify_id;
1628 u8 opcode;
1629 struct ceph_osd_event *event;
1630 struct ceph_osd_event_work *event_work;
1631
1632 p = msg->front.iov_base;
1633 end = p + msg->front.iov_len;
1634
1635 ceph_decode_8_safe(&p, end, proto_ver, bad);
1636 ceph_decode_8_safe(&p, end, opcode, bad);
1637 ceph_decode_64_safe(&p, end, cookie, bad);
1638 ceph_decode_64_safe(&p, end, ver, bad);
1639 ceph_decode_64_safe(&p, end, notify_id, bad);
1640
1641 spin_lock(&osdc->event_lock);
1642 event = __find_event(osdc, cookie);
1643 if (event) {
1644 get_event(event);
1645 if (event->one_shot)
1646 __remove_event(event);
1647 }
1648 spin_unlock(&osdc->event_lock);
1649 dout("handle_watch_notify cookie %lld ver %lld event %p\n",
1650 cookie, ver, event);
1651 if (event) {
1652 event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
a40c4f10
YS		 1653 if (!event_work) {
1654 dout("ERROR: could not allocate event_work\n");
1655 goto done_err;
1656 }
6b0ae409 1657 INIT_WORK(&event_work->work, do_event_work);
a40c4f10
YS		 1658 event_work->event = event;
1659 event_work->ver = ver;
1660 event_work->notify_id = notify_id;
1661 event_work->opcode = opcode;
1662 if (!queue_work(osdc->notify_wq, &event_work->work)) {
1663 dout("WARNING: failed to queue notify event work\n");
1664 goto done_err;
1665 }
1666 }
1667
1668 return;
1669
1670done_err:
1671 complete(&event->completion);
1672 ceph_osdc_put_event(event);
1673 return;
1674
1675bad:
1676 pr_err("osdc handle_watch_notify corrupt msg\n");
1677 return;
1678}
1679
1680int ceph_osdc_wait_event(struct ceph_osd_event *event, unsigned long timeout)
1681{
1682 int err;
1683
1684 dout("wait_event %p\n", event);
1685 err = wait_for_completion_interruptible_timeout(&event->completion,
1686 timeout * HZ);
1687 ceph_osdc_put_event(event);
1688 if (err > 0)
1689 err = 0;
1690 dout("wait_event %p returns %d\n", event, err);
1691 return err;
1692}
1693EXPORT_SYMBOL(ceph_osdc_wait_event);
1694
f24e9980
SW		 1695/*
1696 * Register request, send initial attempt.
1697 */
1698int ceph_osdc_start_request(struct ceph_osd_client *osdc,
1699 struct ceph_osd_request *req,
1700 bool nofail)
1701{
c1ea8823 1702 int rc = 0;
f24e9980
SW		 1703
1704 req->r_request->pages = req->r_pages;
1705 req->r_request->nr_pages = req->r_num_pages;
68b4476b
YS		 1706#ifdef CONFIG_BLOCK
1707 req->r_request->bio = req->r_bio;
1708#endif
1709 req->r_request->trail = req->r_trail;
f24e9980
SW		 1710
1711 register_request(osdc, req);
1712
1713 down_read(&osdc->map_sem);
1714 mutex_lock(&osdc->request_mutex);
c1ea8823
SW		 1715 /*
1716 * a racing kick_requests() may have sent the message for us
1717 * while we dropped request_mutex above, so only send now if
1718 * the request still han't been touched yet.
1719 */
1720 if (req->r_sent == 0) {
38d6453c 1721 rc = __map_request(osdc, req, 0);
9d6fcb08
SW		 1722 if (rc < 0) {
1723 if (nofail) {
1724 dout("osdc_start_request failed map, "
1725 " will retry %lld\n", req->r_tid);
1726 rc = 0;
1727 }
234af26f 1728 goto out_unlock;
9d6fcb08 1729 }
6f6c7006
SW		 1730 if (req->r_osd == NULL) {
1731 dout("send_request %p no up osds in pg\n", req);
1732 ceph_monc_request_next_osdmap(&osdc->client->monc);
1733 } else {
56e925b6 1734 __send_request(osdc, req);
f24e9980 1735 }
56e925b6 1736 rc = 0;
f24e9980 1737 }
234af26f
DC		 1738
1739out_unlock:
f24e9980
SW		 1740 mutex_unlock(&osdc->request_mutex);
1741 up_read(&osdc->map_sem);
1742 return rc;
1743}
3d14c5d2 1744EXPORT_SYMBOL(ceph_osdc_start_request);
f24e9980
SW		 1745
1746/*
1747 * wait for a request to complete
1748 */
1749int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
1750 struct ceph_osd_request *req)
1751{
1752 int rc;
1753
1754 rc = wait_for_completion_interruptible(&req->r_completion);
1755 if (rc < 0) {
1756 mutex_lock(&osdc->request_mutex);
1757 __cancel_request(req);
529cfcc4 1758 __unregister_request(osdc, req);
f24e9980 1759 mutex_unlock(&osdc->request_mutex);
25845472 1760 complete_request(req);
529cfcc4 1761 dout("wait_request tid %llu canceled/timed out\n", req->r_tid);
f24e9980
SW		 1762 return rc;
1763 }
1764
1765 dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result);
1766 return req->r_result;
1767}
3d14c5d2 1768EXPORT_SYMBOL(ceph_osdc_wait_request);
f24e9980
SW		 1769
1770/*
1771 * sync - wait for all in-flight requests to flush. avoid starvation.
1772 */
1773void ceph_osdc_sync(struct ceph_osd_client *osdc)
1774{
1775 struct ceph_osd_request *req;
1776 u64 last_tid, next_tid = 0;
1777
1778 mutex_lock(&osdc->request_mutex);
1779 last_tid = osdc->last_tid;
1780 while (1) {
1781 req = __lookup_request_ge(osdc, next_tid);
1782 if (!req)
1783 break;
1784 if (req->r_tid > last_tid)
1785 break;
1786
1787 next_tid = req->r_tid + 1;
1788 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
1789 continue;
1790
1791 ceph_osdc_get_request(req);
1792 mutex_unlock(&osdc->request_mutex);
1793 dout("sync waiting on tid %llu (last is %llu)\n",
1794 req->r_tid, last_tid);
1795 wait_for_completion(&req->r_safe_completion);
1796 mutex_lock(&osdc->request_mutex);
1797 ceph_osdc_put_request(req);
1798 }
1799 mutex_unlock(&osdc->request_mutex);
1800 dout("sync done (thru tid %llu)\n", last_tid);
1801}
3d14c5d2 1802EXPORT_SYMBOL(ceph_osdc_sync);
f24e9980
SW		 1803
1804/*
1805 * init, shutdown
1806 */
1807int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
1808{
1809 int err;
1810
1811 dout("init\n");
1812 osdc->client = client;
1813 osdc->osdmap = NULL;
1814 init_rwsem(&osdc->map_sem);
1815 init_completion(&osdc->map_waiters);
1816 osdc->last_requested_map = 0;
1817 mutex_init(&osdc->request_mutex);
f24e9980
SW		 1818 osdc->last_tid = 0;
1819 osdc->osds = RB_ROOT;
f5a2041b 1820 INIT_LIST_HEAD(&osdc->osd_lru);
f24e9980 1821 osdc->requests = RB_ROOT;
422d2cb8 1822 INIT_LIST_HEAD(&osdc->req_lru);
6f6c7006
SW		 1823 INIT_LIST_HEAD(&osdc->req_unsent);
1824 INIT_LIST_HEAD(&osdc->req_notarget);
a40c4f10 1825 INIT_LIST_HEAD(&osdc->req_linger);
f24e9980
SW		 1826 osdc->num_requests = 0;
1827 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
f5a2041b 1828 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
a40c4f10
YS		 1829 spin_lock_init(&osdc->event_lock);
1830 osdc->event_tree = RB_ROOT;
1831 osdc->event_count = 0;
f5a2041b
YS		 1832
1833 schedule_delayed_work(&osdc->osds_timeout_work,
3d14c5d2 1834 round_jiffies_relative(osdc->client->options->osd_idle_ttl * HZ));
f24e9980 1835
5f44f142 1836 err = -ENOMEM;
f24e9980
SW		 1837 osdc->req_mempool = mempool_create_kmalloc_pool(10,
1838 sizeof(struct ceph_osd_request));
1839 if (!osdc->req_mempool)
5f44f142 1840 goto out;
f24e9980 1841
d50b409f
SW		 1842 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
1843 OSD_OP_FRONT_LEN, 10, true,
4f48280e 1844 "osd_op");
f24e9980 1845 if (err < 0)
5f44f142 1846 goto out_mempool;
d50b409f 1847 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
4f48280e
SW		 1848 OSD_OPREPLY_FRONT_LEN, 10, true,
1849 "osd_op_reply");
c16e7869
SW		 1850 if (err < 0)
1851 goto out_msgpool;
a40c4f10
YS		 1852
1853 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
1854 if (IS_ERR(osdc->notify_wq)) {
1855 err = PTR_ERR(osdc->notify_wq);
1856 osdc->notify_wq = NULL;
1857 goto out_msgpool;
1858 }
f24e9980 1859 return 0;
5f44f142 1860
c16e7869
SW		 1861out_msgpool:
1862 ceph_msgpool_destroy(&osdc->msgpool_op);
5f44f142
SW		 1863out_mempool:
1864 mempool_destroy(osdc->req_mempool);
1865out:
1866 return err;
f24e9980 1867}
3d14c5d2 1868EXPORT_SYMBOL(ceph_osdc_init);
f24e9980
SW		 1869
1870void ceph_osdc_stop(struct ceph_osd_client *osdc)
1871{
a40c4f10
YS		 1872 flush_workqueue(osdc->notify_wq);
1873 destroy_workqueue(osdc->notify_wq);
f24e9980 1874 cancel_delayed_work_sync(&osdc->timeout_work);
f5a2041b 1875 cancel_delayed_work_sync(&osdc->osds_timeout_work);
f24e9980
SW		 1876 if (osdc->osdmap) {
1877 ceph_osdmap_destroy(osdc->osdmap);
1878 osdc->osdmap = NULL;
1879 }
aca420bc 1880 remove_all_osds(osdc);
f24e9980
SW		 1881 mempool_destroy(osdc->req_mempool);
1882 ceph_msgpool_destroy(&osdc->msgpool_op);
c16e7869 1883 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
f24e9980 1884}
3d14c5d2 1885EXPORT_SYMBOL(ceph_osdc_stop);
f24e9980
SW		 1886
1887/*
1888 * Read some contiguous pages. If we cross a stripe boundary, shorten
1889 * *plen. Return number of bytes read, or error.
1890 */
1891int ceph_osdc_readpages(struct ceph_osd_client *osdc,
1892 struct ceph_vino vino, struct ceph_file_layout *layout,
1893 u64 off, u64 *plen,
1894 u32 truncate_seq, u64 truncate_size,
b7495fc2 1895 struct page **pages, int num_pages, int page_align)
f24e9980
SW		 1896{
1897 struct ceph_osd_request *req;
1898 int rc = 0;
1899
1900 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
1901 vino.snap, off, *plen);
1902 req = ceph_osdc_new_request(osdc, layout, vino, off, plen,
1903 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
1904 NULL, 0, truncate_seq, truncate_size, NULL,
b7495fc2 1905 false, 1, page_align);
6816282d
SW		 1906 if (IS_ERR(req))
1907 return PTR_ERR(req);
f24e9980
SW		 1908
1909 /* it may be a short read due to an object boundary */
1910 req->r_pages = pages;
f24e9980 1911
b7495fc2
SW		 1912 dout("readpages final extent is %llu~%llu (%d pages align %d)\n",
1913 off, *plen, req->r_num_pages, page_align);
f24e9980
SW		 1914
1915 rc = ceph_osdc_start_request(osdc, req, false);
1916 if (!rc)
1917 rc = ceph_osdc_wait_request(osdc, req);
1918
1919 ceph_osdc_put_request(req);
1920 dout("readpages result %d\n", rc);
1921 return rc;
1922}
3d14c5d2 1923EXPORT_SYMBOL(ceph_osdc_readpages);
f24e9980
SW		 1924
1925/*
1926 * do a synchronous write on N pages
1927 */
1928int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
1929 struct ceph_file_layout *layout,
1930 struct ceph_snap_context *snapc,
1931 u64 off, u64 len,
1932 u32 truncate_seq, u64 truncate_size,
1933 struct timespec *mtime,
1934 struct page **pages, int num_pages,
1935 int flags, int do_sync, bool nofail)
1936{
1937 struct ceph_osd_request *req;
1938 int rc = 0;
b7495fc2 1939 int page_align = off & ~PAGE_MASK;
f24e9980
SW		 1940
1941 BUG_ON(vino.snap != CEPH_NOSNAP);
1942 req = ceph_osdc_new_request(osdc, layout, vino, off, &len,
1943 CEPH_OSD_OP_WRITE,
1944 flags | CEPH_OSD_FLAG_ONDISK |
1945 CEPH_OSD_FLAG_WRITE,
1946 snapc, do_sync,
1947 truncate_seq, truncate_size, mtime,
b7495fc2 1948 nofail, 1, page_align);
6816282d
SW		 1949 if (IS_ERR(req))
1950 return PTR_ERR(req);
f24e9980
SW		 1951
1952 /* it may be a short write due to an object boundary */
1953 req->r_pages = pages;
f24e9980
SW		 1954 dout("writepages %llu~%llu (%d pages)\n", off, len,
1955 req->r_num_pages);
1956
1957 rc = ceph_osdc_start_request(osdc, req, nofail);
1958 if (!rc)
1959 rc = ceph_osdc_wait_request(osdc, req);
1960
1961 ceph_osdc_put_request(req);
1962 if (rc == 0)
1963 rc = len;
1964 dout("writepages result %d\n", rc);
1965 return rc;
1966}
3d14c5d2 1967EXPORT_SYMBOL(ceph_osdc_writepages);
f24e9980
SW		 1968
1969/*
1970 * handle incoming message
1971 */
1972static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
1973{
1974 struct ceph_osd *osd = con->private;
32c895e7 1975 struct ceph_osd_client *osdc;
f24e9980
SW		 1976 int type = le16_to_cpu(msg->hdr.type);
1977
1978 if (!osd)
4a32f93d 1979 goto out;
32c895e7 1980 osdc = osd->o_osdc;
f24e9980
SW		 1981
1982 switch (type) {
1983 case CEPH_MSG_OSD_MAP:
1984 ceph_osdc_handle_map(osdc, msg);
1985 break;
1986 case CEPH_MSG_OSD_OPREPLY:
350b1c32 1987 handle_reply(osdc, msg, con);
f24e9980 1988 break;
a40c4f10
YS		 1989 case CEPH_MSG_WATCH_NOTIFY:
1990 handle_watch_notify(osdc, msg);
1991 break;
f24e9980
SW		 1992
1993 default:
1994 pr_err("received unknown message type %d %s\n", type,
1995 ceph_msg_type_name(type));
1996 }
4a32f93d 1997out:
f24e9980
SW		 1998 ceph_msg_put(msg);
1999}
2000
5b3a4db3 2001/*
21b667f6
SW		 2002 * lookup and return message for incoming reply. set up reply message
2003 * pages.
5b3a4db3
SW		 2004 */
2005static struct ceph_msg *get_reply(struct ceph_connection *con,
2450418c
YS		 2006 struct ceph_msg_header *hdr,
2007 int *skip)
f24e9980
SW		 2008{
2009 struct ceph_osd *osd = con->private;
2010 struct ceph_osd_client *osdc = osd->o_osdc;
2450418c 2011 struct ceph_msg *m;
0547a9b3 2012 struct ceph_osd_request *req;
5b3a4db3
SW		 2013 int front = le32_to_cpu(hdr->front_len);
2014 int data_len = le32_to_cpu(hdr->data_len);
0547a9b3 2015 u64 tid;
f24e9980 2016
0547a9b3
YS		 2017 tid = le64_to_cpu(hdr->tid);
2018 mutex_lock(&osdc->request_mutex);
2019 req = __lookup_request(osdc, tid);
2020 if (!req) {
2021 *skip = 1;
2022 m = NULL;
756a16a5
SW		 2023 dout("get_reply unknown tid %llu from osd%d\n", tid,
2024 osd->o_osd);
0547a9b3
YS		 2025 goto out;
2026 }
c16e7869
SW		 2027
2028 if (req->r_con_filling_msg) {
8921d114 2029 dout("%s revoking msg %p from old con %p\n", __func__,
c16e7869 2030 req->r_reply, req->r_con_filling_msg);
8921d114 2031 ceph_msg_revoke_incoming(req->r_reply);
0d47766f 2032 req->r_con_filling_msg->ops->put(req->r_con_filling_msg);
6f46cb29 2033 req->r_con_filling_msg = NULL;
0547a9b3
YS		 2034 }
2035
c16e7869
SW		 2036 if (front > req->r_reply->front.iov_len) {
2037 pr_warning("get_reply front %d > preallocated %d\n",
2038 front, (int)req->r_reply->front.iov_len);
b61c2763 2039 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front, GFP_NOFS, false);
a79832f2 2040 if (!m)
c16e7869
SW		 2041 goto out;
2042 ceph_msg_put(req->r_reply);
2043 req->r_reply = m;
2044 }
2045 m = ceph_msg_get(req->r_reply);
2046
0547a9b3 2047 if (data_len > 0) {
b7495fc2 2048 int want = calc_pages_for(req->r_page_alignment, data_len);
21b667f6
SW		 2049
2050 if (unlikely(req->r_num_pages < want)) {
9bb0ce2b
SW		 2051 pr_warning("tid %lld reply has %d bytes %d pages, we"
2052 " had only %d pages ready\n", tid, data_len,
2053 want, req->r_num_pages);
0547a9b3
YS		 2054 *skip = 1;
2055 ceph_msg_put(m);
a79832f2 2056 m = NULL;
21b667f6 2057 goto out;
0547a9b3 2058 }
21b667f6
SW		 2059 m->pages = req->r_pages;
2060 m->nr_pages = req->r_num_pages;
c5c6b19d 2061 m->page_alignment = req->r_page_alignment;
68b4476b
YS		 2062#ifdef CONFIG_BLOCK
2063 m->bio = req->r_bio;
2064#endif
0547a9b3 2065 }
5b3a4db3 2066 *skip = 0;
0d47766f 2067 req->r_con_filling_msg = con->ops->get(con);
c16e7869 2068 dout("get_reply tid %lld %p\n", tid, m);
0547a9b3
YS		 2069
2070out:
2071 mutex_unlock(&osdc->request_mutex);
2450418c 2072 return m;
5b3a4db3
SW		 2073
2074}
2075
2076static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2077 struct ceph_msg_header *hdr,
2078 int *skip)
2079{
2080 struct ceph_osd *osd = con->private;
2081 int type = le16_to_cpu(hdr->type);
2082 int front = le32_to_cpu(hdr->front_len);
2083
1c20f2d2 2084 *skip = 0;
5b3a4db3
SW		 2085 switch (type) {
2086 case CEPH_MSG_OSD_MAP:
a40c4f10 2087 case CEPH_MSG_WATCH_NOTIFY:
b61c2763 2088 return ceph_msg_new(type, front, GFP_NOFS, false);
5b3a4db3
SW		 2089 case CEPH_MSG_OSD_OPREPLY:
2090 return get_reply(con, hdr, skip);
2091 default:
2092 pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
2093 osd->o_osd);
2094 *skip = 1;
2095 return NULL;
2096 }
f24e9980
SW		 2097}
2098
2099/*
2100 * Wrappers to refcount containing ceph_osd struct
2101 */
2102static struct ceph_connection *get_osd_con(struct ceph_connection *con)
2103{
2104 struct ceph_osd *osd = con->private;
2105 if (get_osd(osd))
2106 return con;
2107 return NULL;
2108}
2109
2110static void put_osd_con(struct ceph_connection *con)
2111{
2112 struct ceph_osd *osd = con->private;
2113 put_osd(osd);
2114}
2115
4e7a5dcd
SW		 2116/*
2117 * authentication
2118 */
a3530df3
AE		 2119/*
2120 * Note: returned pointer is the address of a structure that's
2121 * managed separately. Caller must *not* attempt to free it.
2122 */
2123static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
8f43fb53 2124 int *proto, int force_new)
4e7a5dcd
SW		 2125{
2126 struct ceph_osd *o = con->private;
2127 struct ceph_osd_client *osdc = o->o_osdc;
2128 struct ceph_auth_client *ac = osdc->client->monc.auth;
74f1869f 2129 struct ceph_auth_handshake *auth = &o->o_auth;
4e7a5dcd 2130
74f1869f 2131 if (force_new && auth->authorizer) {
a255651d
AE		 2132 if (ac->ops && ac->ops->destroy_authorizer)
2133 ac->ops->destroy_authorizer(ac, auth->authorizer);
74f1869f
AE		 2134 auth->authorizer = NULL;
2135 }
a255651d 2136 if (!auth->authorizer && ac->ops && ac->ops->create_authorizer) {
a3530df3
AE		 2137 int ret = ac->ops->create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2138 auth);
4e7a5dcd 2139 if (ret)
a3530df3 2140 return ERR_PTR(ret);
4e7a5dcd 2141 }
4e7a5dcd 2142 *proto = ac->protocol;
74f1869f 2143
a3530df3 2144 return auth;
4e7a5dcd
SW		 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
a255651d
AE		 2154 /*
2155 * XXX If ac->ops or ac->ops->verify_authorizer_reply is null,
2156 * XXX which do we do: succeed or fail?
2157 */
6c4a1915 2158 return ac->ops->verify_authorizer_reply(ac, o->o_auth.authorizer, len);
4e7a5dcd
SW		 2159}
2160
9bd2e6f8
SW		 2161static int invalidate_authorizer(struct ceph_connection *con)
2162{
2163 struct ceph_osd *o = con->private;
2164 struct ceph_osd_client *osdc = o->o_osdc;
2165 struct ceph_auth_client *ac = osdc->client->monc.auth;
2166
a255651d 2167 if (ac->ops && ac->ops->invalidate_authorizer)
9bd2e6f8
SW		 2168 ac->ops->invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
2169
2170 return ceph_monc_validate_auth(&osdc->client->monc);
2171}
4e7a5dcd 2172
9e32789f 2173static const struct ceph_connection_operations osd_con_ops = {
f24e9980
SW		 2174 .get = get_osd_con,
2175 .put = put_osd_con,
2176 .dispatch = dispatch,
4e7a5dcd
SW		 2177 .get_authorizer = get_authorizer,
2178 .verify_authorizer_reply = verify_authorizer_reply,
9bd2e6f8 2179 .invalidate_authorizer = invalidate_authorizer,
f24e9980 2180 .alloc_msg = alloc_msg,
81b024e7 2181 .fault = osd_reset,
f24e9980 2182};
Linux 6.x block layer and io_uring tree(s)