projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
libceph: report connection fault with warning
[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);
cd43045c 224 rb_init_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;
f24e9980 645 INIT_LIST_HEAD(&osd->o_requests);
a40c4f10 646 INIT_LIST_HEAD(&osd->o_linger_requests);
f5a2041b 647 INIT_LIST_HEAD(&osd->o_osd_lru);
f24e9980
SW		 648 osd->o_incarnation = 1;
649
b7a9e5dd 650 ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
4e7a5dcd 651
422d2cb8 652 INIT_LIST_HEAD(&osd->o_keepalive_item);
f24e9980
SW		 653 return osd;
654}
655
656static struct ceph_osd *get_osd(struct ceph_osd *osd)
657{
658 if (atomic_inc_not_zero(&osd->o_ref)) {
659 dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
660 atomic_read(&osd->o_ref));
661 return osd;
662 } else {
663 dout("get_osd %p FAIL\n", osd);
664 return NULL;
665 }
666}
667
668static void put_osd(struct ceph_osd *osd)
669{
670 dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
671 atomic_read(&osd->o_ref) - 1);
a255651d 672 if (atomic_dec_and_test(&osd->o_ref) && osd->o_auth.authorizer) {
79494d1b
SW		 673 struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth;
674
a255651d 675 if (ac->ops && ac->ops->destroy_authorizer)
6c4a1915 676 ac->ops->destroy_authorizer(ac, osd->o_auth.authorizer);
f24e9980 677 kfree(osd);
79494d1b 678 }
f24e9980
SW		 679}
680
681/*
682 * remove an osd from our map
683 */
f5a2041b 684static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
f24e9980 685{
f5a2041b 686 dout("__remove_osd %p\n", osd);
f24e9980
SW		 687 BUG_ON(!list_empty(&osd->o_requests));
688 rb_erase(&osd->o_node, &osdc->osds);
f5a2041b 689 list_del_init(&osd->o_osd_lru);
f24e9980
SW		 690 ceph_con_close(&osd->o_con);
691 put_osd(osd);
692}
693
aca420bc
SW		 694static void remove_all_osds(struct ceph_osd_client *osdc)
695{
048a9d2d 696 dout("%s %p\n", __func__, osdc);
aca420bc
SW		 697 mutex_lock(&osdc->request_mutex);
698 while (!RB_EMPTY_ROOT(&osdc->osds)) {
699 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
700 struct ceph_osd, o_node);
701 __remove_osd(osdc, osd);
702 }
703 mutex_unlock(&osdc->request_mutex);
704}
705
f5a2041b
YS		 706static void __move_osd_to_lru(struct ceph_osd_client *osdc,
707 struct ceph_osd *osd)
708{
709 dout("__move_osd_to_lru %p\n", osd);
710 BUG_ON(!list_empty(&osd->o_osd_lru));
711 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
3d14c5d2 712 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl * HZ;
f5a2041b
YS		 713}
714
715static void __remove_osd_from_lru(struct ceph_osd *osd)
716{
717 dout("__remove_osd_from_lru %p\n", osd);
718 if (!list_empty(&osd->o_osd_lru))
719 list_del_init(&osd->o_osd_lru);
720}
721
aca420bc 722static void remove_old_osds(struct ceph_osd_client *osdc)
f5a2041b
YS		 723{
724 struct ceph_osd *osd, *nosd;
725
726 dout("__remove_old_osds %p\n", osdc);
727 mutex_lock(&osdc->request_mutex);
728 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
aca420bc 729 if (time_before(jiffies, osd->lru_ttl))
f5a2041b
YS		 730 break;
731 __remove_osd(osdc, osd);
732 }
733 mutex_unlock(&osdc->request_mutex);
734}
735
f24e9980
SW		 736/*
737 * reset osd connect
738 */
f5a2041b 739static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
f24e9980 740{
87b315a5 741 struct ceph_osd_request *req;
f24e9980
SW		 742 int ret = 0;
743
f5a2041b 744 dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
a40c4f10
YS		 745 if (list_empty(&osd->o_requests) &&
746 list_empty(&osd->o_linger_requests)) {
f5a2041b 747 __remove_osd(osdc, osd);
685a7555 748 ret = -ENODEV;
87b315a5
SW		 749 } else if (memcmp(&osdc->osdmap->osd_addr[osd->o_osd],
750 &osd->o_con.peer_addr,
751 sizeof(osd->o_con.peer_addr)) == 0 &&
752 !ceph_con_opened(&osd->o_con)) {
753 dout(" osd addr hasn't changed and connection never opened,"
754 " letting msgr retry");
755 /* touch each r_stamp for handle_timeout()'s benfit */
756 list_for_each_entry(req, &osd->o_requests, r_osd_item)
757 req->r_stamp = jiffies;
758 ret = -EAGAIN;
f24e9980
SW		 759 } else {
760 ceph_con_close(&osd->o_con);
b7a9e5dd
SW		 761 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd,
762 &osdc->osdmap->osd_addr[osd->o_osd]);
f24e9980
SW		 763 osd->o_incarnation++;
764 }
765 return ret;
766}
767
768static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new)
769{
770 struct rb_node **p = &osdc->osds.rb_node;
771 struct rb_node *parent = NULL;
772 struct ceph_osd *osd = NULL;
773
aca420bc 774 dout("__insert_osd %p osd%d\n", new, new->o_osd);
f24e9980
SW		 775 while (*p) {
776 parent = *p;
777 osd = rb_entry(parent, struct ceph_osd, o_node);
778 if (new->o_osd < osd->o_osd)
779 p = &(*p)->rb_left;
780 else if (new->o_osd > osd->o_osd)
781 p = &(*p)->rb_right;
782 else
783 BUG();
784 }
785
786 rb_link_node(&new->o_node, parent, p);
787 rb_insert_color(&new->o_node, &osdc->osds);
788}
789
790static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o)
791{
792 struct ceph_osd *osd;
793 struct rb_node *n = osdc->osds.rb_node;
794
795 while (n) {
796 osd = rb_entry(n, struct ceph_osd, o_node);
797 if (o < osd->o_osd)
798 n = n->rb_left;
799 else if (o > osd->o_osd)
800 n = n->rb_right;
801 else
802 return osd;
803 }
804 return NULL;
805}
806
422d2cb8
YS		 807static void __schedule_osd_timeout(struct ceph_osd_client *osdc)
808{
809 schedule_delayed_work(&osdc->timeout_work,
3d14c5d2 810 osdc->client->options->osd_keepalive_timeout * HZ);
422d2cb8
YS		 811}
812
813static void __cancel_osd_timeout(struct ceph_osd_client *osdc)
814{
815 cancel_delayed_work(&osdc->timeout_work);
816}
f24e9980
SW		 817
818/*
819 * Register request, assign tid. If this is the first request, set up
820 * the timeout event.
821 */
a40c4f10
YS		 822static void __register_request(struct ceph_osd_client *osdc,
823 struct ceph_osd_request *req)
f24e9980 824{
f24e9980 825 req->r_tid = ++osdc->last_tid;
6df058c0 826 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
77f38e0e 827 dout("__register_request %p tid %lld\n", req, req->r_tid);
f24e9980
SW		 828 __insert_request(osdc, req);
829 ceph_osdc_get_request(req);
830 osdc->num_requests++;
f24e9980 831 if (osdc->num_requests == 1) {
422d2cb8
YS		 832 dout(" first request, scheduling timeout\n");
833 __schedule_osd_timeout(osdc);
f24e9980 834 }
a40c4f10
YS		 835}
836
837static void register_request(struct ceph_osd_client *osdc,
838 struct ceph_osd_request *req)
839{
840 mutex_lock(&osdc->request_mutex);
841 __register_request(osdc, req);
f24e9980
SW		 842 mutex_unlock(&osdc->request_mutex);
843}
844
845/*
846 * called under osdc->request_mutex
847 */
848static void __unregister_request(struct ceph_osd_client *osdc,
849 struct ceph_osd_request *req)
850{
35f9f8a0
SW		 851 if (RB_EMPTY_NODE(&req->r_node)) {
852 dout("__unregister_request %p tid %lld not registered\n",
853 req, req->r_tid);
854 return;
855 }
856
f24e9980
SW		 857 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
858 rb_erase(&req->r_node, &osdc->requests);
859 osdc->num_requests--;
860
0ba6478d
SW		 861 if (req->r_osd) {
862 /* make sure the original request isn't in flight. */
6740a845 863 ceph_msg_revoke(req->r_request);
0ba6478d
SW		 864
865 list_del_init(&req->r_osd_item);
a40c4f10
YS		 866 if (list_empty(&req->r_osd->o_requests) &&
867 list_empty(&req->r_osd->o_linger_requests)) {
868 dout("moving osd to %p lru\n", req->r_osd);
f5a2041b 869 __move_osd_to_lru(osdc, req->r_osd);
a40c4f10 870 }
fbdb9190 871 if (list_empty(&req->r_linger_item))
a40c4f10 872 req->r_osd = NULL;
0ba6478d 873 }
f24e9980 874
7d5f2481 875 list_del_init(&req->r_req_lru_item);
f24e9980
SW		 876 ceph_osdc_put_request(req);
877
422d2cb8
YS		 878 if (osdc->num_requests == 0) {
879 dout(" no requests, canceling timeout\n");
880 __cancel_osd_timeout(osdc);
f24e9980
SW		 881 }
882}
883
884/*
885 * Cancel a previously queued request message
886 */
887static void __cancel_request(struct ceph_osd_request *req)
888{
6bc18876 889 if (req->r_sent && req->r_osd) {
6740a845 890 ceph_msg_revoke(req->r_request);
f24e9980
SW		 891 req->r_sent = 0;
892 }
893}
894
a40c4f10
YS		 895static void __register_linger_request(struct ceph_osd_client *osdc,
896 struct ceph_osd_request *req)
897{
898 dout("__register_linger_request %p\n", req);
899 list_add_tail(&req->r_linger_item, &osdc->req_linger);
6194ea89
SW		 900 if (req->r_osd)
901 list_add_tail(&req->r_linger_osd,
902 &req->r_osd->o_linger_requests);
a40c4f10
YS		 903}
904
905static void __unregister_linger_request(struct ceph_osd_client *osdc,
906 struct ceph_osd_request *req)
907{
908 dout("__unregister_linger_request %p\n", req);
61c74035 909 list_del_init(&req->r_linger_item);
a40c4f10 910 if (req->r_osd) {
a40c4f10
YS		 911 list_del_init(&req->r_linger_osd);
912
913 if (list_empty(&req->r_osd->o_requests) &&
914 list_empty(&req->r_osd->o_linger_requests)) {
915 dout("moving osd to %p lru\n", req->r_osd);
916 __move_osd_to_lru(osdc, req->r_osd);
917 }
fbdb9190
SW		 918 if (list_empty(&req->r_osd_item))
919 req->r_osd = NULL;
a40c4f10
YS		 920 }
921}
922
923void ceph_osdc_unregister_linger_request(struct ceph_osd_client *osdc,
924 struct ceph_osd_request *req)
925{
926 mutex_lock(&osdc->request_mutex);
927 if (req->r_linger) {
928 __unregister_linger_request(osdc, req);
929 ceph_osdc_put_request(req);
930 }
931 mutex_unlock(&osdc->request_mutex);
932}
933EXPORT_SYMBOL(ceph_osdc_unregister_linger_request);
934
935void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
936 struct ceph_osd_request *req)
937{
938 if (!req->r_linger) {
939 dout("set_request_linger %p\n", req);
940 req->r_linger = 1;
941 /*
942 * caller is now responsible for calling
943 * unregister_linger_request
944 */
945 ceph_osdc_get_request(req);
946 }
947}
948EXPORT_SYMBOL(ceph_osdc_set_request_linger);
949
f24e9980
SW		 950/*
951 * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
952 * (as needed), and set the request r_osd appropriately. If there is
25985edc 953 * no up osd, set r_osd to NULL. Move the request to the appropriate list
6f6c7006 954 * (unsent, homeless) or leave on in-flight lru.
f24e9980
SW		 955 *
956 * Return 0 if unchanged, 1 if changed, or negative on error.
957 *
958 * Caller should hold map_sem for read and request_mutex.
959 */
6f6c7006 960static int __map_request(struct ceph_osd_client *osdc,
38d6453c 961 struct ceph_osd_request *req, int force_resend)
f24e9980
SW		 962{
963 struct ceph_osd_request_head *reqhead = req->r_request->front.iov_base;
51042122 964 struct ceph_pg pgid;
d85b7056
SW		 965 int acting[CEPH_PG_MAX_SIZE];
966 int o = -1, num = 0;
f24e9980 967 int err;
f24e9980 968
6f6c7006 969 dout("map_request %p tid %lld\n", req, req->r_tid);
f24e9980
SW		 970 err = ceph_calc_object_layout(&reqhead->layout, req->r_oid,
971 &req->r_file_layout, osdc->osdmap);
6f6c7006
SW		 972 if (err) {
973 list_move(&req->r_req_lru_item, &osdc->req_notarget);
f24e9980 974 return err;
6f6c7006 975 }
51042122 976 pgid = reqhead->layout.ol_pgid;
7740a42f
SW		 977 req->r_pgid = pgid;
978
d85b7056
SW		 979 err = ceph_calc_pg_acting(osdc->osdmap, pgid, acting);
980 if (err > 0) {
981 o = acting[0];
982 num = err;
983 }
f24e9980 984
38d6453c
SW		 985 if ((!force_resend &&
986 req->r_osd && req->r_osd->o_osd == o &&
d85b7056
SW		 987 req->r_sent >= req->r_osd->o_incarnation &&
988 req->r_num_pg_osds == num &&
989 memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
f24e9980
SW		 990 (req->r_osd == NULL && o == -1))
991 return 0; /* no change */
992
6f6c7006 993 dout("map_request tid %llu pgid %d.%x osd%d (was osd%d)\n",
51042122 994 req->r_tid, le32_to_cpu(pgid.pool), le16_to_cpu(pgid.ps), o,
f24e9980
SW		 995 req->r_osd ? req->r_osd->o_osd : -1);
996
d85b7056
SW		 997 /* record full pg acting set */
998 memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
999 req->r_num_pg_osds = num;
1000
f24e9980
SW		 1001 if (req->r_osd) {
1002 __cancel_request(req);
1003 list_del_init(&req->r_osd_item);
f24e9980
SW		 1004 req->r_osd = NULL;
1005 }
1006
1007 req->r_osd = __lookup_osd(osdc, o);
1008 if (!req->r_osd && o >= 0) {
c99eb1c7 1009 err = -ENOMEM;
e10006f8 1010 req->r_osd = create_osd(osdc, o);
6f6c7006
SW		 1011 if (!req->r_osd) {
1012 list_move(&req->r_req_lru_item, &osdc->req_notarget);
c99eb1c7 1013 goto out;
6f6c7006 1014 }
f24e9980 1015
6f6c7006 1016 dout("map_request osd %p is osd%d\n", req->r_osd, o);
f24e9980
SW		 1017 __insert_osd(osdc, req->r_osd);
1018
b7a9e5dd
SW		 1019 ceph_con_open(&req->r_osd->o_con,
1020 CEPH_ENTITY_TYPE_OSD, o,
1021 &osdc->osdmap->osd_addr[o]);
f24e9980
SW		 1022 }
1023
f5a2041b
YS		 1024 if (req->r_osd) {
1025 __remove_osd_from_lru(req->r_osd);
f24e9980 1026 list_add(&req->r_osd_item, &req->r_osd->o_requests);
6f6c7006
SW		 1027 list_move(&req->r_req_lru_item, &osdc->req_unsent);
1028 } else {
1029 list_move(&req->r_req_lru_item, &osdc->req_notarget);
f5a2041b 1030 }
d85b7056 1031 err = 1; /* osd or pg changed */
f24e9980
SW		 1032
1033out:
f24e9980
SW		 1034 return err;
1035}
1036
1037/*
1038 * caller should hold map_sem (for read) and request_mutex
1039 */
56e925b6
SW		 1040static void __send_request(struct ceph_osd_client *osdc,
1041 struct ceph_osd_request *req)
f24e9980
SW		 1042{
1043 struct ceph_osd_request_head *reqhead;
f24e9980
SW		 1044
1045 dout("send_request %p tid %llu to osd%d flags %d\n",
1046 req, req->r_tid, req->r_osd->o_osd, req->r_flags);
1047
1048 reqhead = req->r_request->front.iov_base;
1049 reqhead->osdmap_epoch = cpu_to_le32(osdc->osdmap->epoch);
1050 reqhead->flags |= cpu_to_le32(req->r_flags); /* e.g., RETRY */
1051 reqhead->reassert_version = req->r_reassert_version;
1052
3dd72fc0 1053 req->r_stamp = jiffies;
07a27e22 1054 list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
f24e9980
SW		 1055
1056 ceph_msg_get(req->r_request); /* send consumes a ref */
1057 ceph_con_send(&req->r_osd->o_con, req->r_request);
1058 req->r_sent = req->r_osd->o_incarnation;
f24e9980
SW		 1059}
1060
6f6c7006
SW		 1061/*
1062 * Send any requests in the queue (req_unsent).
1063 */
1064static void send_queued(struct ceph_osd_client *osdc)
1065{
1066 struct ceph_osd_request *req, *tmp;
1067
1068 dout("send_queued\n");
1069 mutex_lock(&osdc->request_mutex);
1070 list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item) {
1071 __send_request(osdc, req);
1072 }
1073 mutex_unlock(&osdc->request_mutex);
1074}
1075
f24e9980
SW		 1076/*
1077 * Timeout callback, called every N seconds when 1 or more osd
1078 * requests has been active for more than N seconds. When this
1079 * happens, we ping all OSDs with requests who have timed out to
1080 * ensure any communications channel reset is detected. Reset the
1081 * request timeouts another N seconds in the future as we go.
1082 * Reschedule the timeout event another N seconds in future (unless
1083 * there are no open requests).
1084 */
1085static void handle_timeout(struct work_struct *work)
1086{
1087 struct ceph_osd_client *osdc =
1088 container_of(work, struct ceph_osd_client, timeout_work.work);
83aff95e 1089 struct ceph_osd_request *req;
f24e9980 1090 struct ceph_osd *osd;
422d2cb8 1091 unsigned long keepalive =
3d14c5d2 1092 osdc->client->options->osd_keepalive_timeout * HZ;
422d2cb8 1093 struct list_head slow_osds;
f24e9980
SW		 1094 dout("timeout\n");
1095 down_read(&osdc->map_sem);
1096
1097 ceph_monc_request_next_osdmap(&osdc->client->monc);
1098
1099 mutex_lock(&osdc->request_mutex);
f24e9980 1100
422d2cb8
YS		 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);
aca420bc 1140 remove_old_osds(osdc);
f5a2041b
YS		 1141 up_read(&osdc->map_sem);
1142
1143 schedule_delayed_work(&osdc->osds_timeout_work,
1144 round_jiffies_relative(delay));
1145}
1146
25845472
SW		 1147static void complete_request(struct ceph_osd_request *req)
1148{
1149 if (req->r_safe_callback)
1150 req->r_safe_callback(req, NULL);
1151 complete_all(&req->r_safe_completion); /* fsync waiter */
1152}
1153
f24e9980
SW		 1154/*
1155 * handle osd op reply. either call the callback if it is specified,
1156 * or do the completion to wake up the waiting thread.
1157 */
350b1c32
SW		 1158static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg,
1159 struct ceph_connection *con)
f24e9980
SW		 1160{
1161 struct ceph_osd_reply_head *rhead = msg->front.iov_base;
1162 struct ceph_osd_request *req;
1163 u64 tid;
1164 int numops, object_len, flags;
0ceed5db 1165 s32 result;
f24e9980 1166
6df058c0 1167 tid = le64_to_cpu(msg->hdr.tid);
f24e9980
SW		 1168 if (msg->front.iov_len < sizeof(*rhead))
1169 goto bad;
f24e9980
SW		 1170 numops = le32_to_cpu(rhead->num_ops);
1171 object_len = le32_to_cpu(rhead->object_len);
0ceed5db 1172 result = le32_to_cpu(rhead->result);
f24e9980
SW		 1173 if (msg->front.iov_len != sizeof(*rhead) + object_len +
1174 numops * sizeof(struct ceph_osd_op))
1175 goto bad;
0ceed5db 1176 dout("handle_reply %p tid %llu result %d\n", msg, tid, (int)result);
f24e9980
SW		 1177 /* lookup */
1178 mutex_lock(&osdc->request_mutex);
1179 req = __lookup_request(osdc, tid);
1180 if (req == NULL) {
1181 dout("handle_reply tid %llu dne\n", tid);
1182 mutex_unlock(&osdc->request_mutex);
1183 return;
1184 }
1185 ceph_osdc_get_request(req);
1186 flags = le32_to_cpu(rhead->flags);
1187
350b1c32 1188 /*
0d59ab81 1189 * if this connection filled our message, drop our reference now, to
350b1c32
SW		 1190 * avoid a (safe but slower) revoke later.
1191 */
0d59ab81 1192 if (req->r_con_filling_msg == con && req->r_reply == msg) {
c16e7869 1193 dout(" dropping con_filling_msg ref %p\n", con);
0d59ab81 1194 req->r_con_filling_msg = NULL;
0d47766f 1195 con->ops->put(con);
350b1c32
SW		 1196 }
1197
f24e9980 1198 if (!req->r_got_reply) {
95c96174 1199 unsigned int bytes;
f24e9980
SW		 1200
1201 req->r_result = le32_to_cpu(rhead->result);
1202 bytes = le32_to_cpu(msg->hdr.data_len);
1203 dout("handle_reply result %d bytes %d\n", req->r_result,
1204 bytes);
1205 if (req->r_result == 0)
1206 req->r_result = bytes;
1207
1208 /* in case this is a write and we need to replay, */
1209 req->r_reassert_version = rhead->reassert_version;
1210
1211 req->r_got_reply = 1;
1212 } else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) {
1213 dout("handle_reply tid %llu dup ack\n", tid);
34b43a56 1214 mutex_unlock(&osdc->request_mutex);
f24e9980
SW		 1215 goto done;
1216 }
1217
1218 dout("handle_reply tid %llu flags %d\n", tid, flags);
1219
a40c4f10
YS		 1220 if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK))
1221 __register_linger_request(osdc, req);
1222
f24e9980 1223 /* either this is a read, or we got the safe response */
0ceed5db
SW		 1224 if (result < 0 ||
1225 (flags & CEPH_OSD_FLAG_ONDISK) ||
f24e9980
SW		 1226 ((flags & CEPH_OSD_FLAG_WRITE) == 0))
1227 __unregister_request(osdc, req);
1228
1229 mutex_unlock(&osdc->request_mutex);
1230
1231 if (req->r_callback)
1232 req->r_callback(req, msg);
1233 else
03066f23 1234 complete_all(&req->r_completion);
f24e9980 1235
25845472
SW		 1236 if (flags & CEPH_OSD_FLAG_ONDISK)
1237 complete_request(req);
f24e9980
SW		 1238
1239done:
a40c4f10 1240 dout("req=%p req->r_linger=%d\n", req, req->r_linger);
f24e9980
SW		 1241 ceph_osdc_put_request(req);
1242 return;
1243
1244bad:
1245 pr_err("corrupt osd_op_reply got %d %d expected %d\n",
1246 (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len),
1247 (int)sizeof(*rhead));
9ec7cab1 1248 ceph_msg_dump(msg);
f24e9980
SW		 1249}
1250
6f6c7006 1251static void reset_changed_osds(struct ceph_osd_client *osdc)
f24e9980 1252{
f24e9980 1253 struct rb_node *p, *n;
f24e9980 1254
6f6c7006
SW		 1255 for (p = rb_first(&osdc->osds); p; p = n) {
1256 struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
f24e9980 1257
6f6c7006
SW		 1258 n = rb_next(p);
1259 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
1260 memcmp(&osd->o_con.peer_addr,
1261 ceph_osd_addr(osdc->osdmap,
1262 osd->o_osd),
1263 sizeof(struct ceph_entity_addr)) != 0)
1264 __reset_osd(osdc, osd);
f24e9980 1265 }
422d2cb8
YS		 1266}
1267
1268/*
6f6c7006
SW		 1269 * Requeue requests whose mapping to an OSD has changed. If requests map to
1270 * no osd, request a new map.
422d2cb8
YS		 1271 *
1272 * Caller should hold map_sem for read and request_mutex.
1273 */
38d6453c 1274static void kick_requests(struct ceph_osd_client *osdc, int force_resend)
422d2cb8 1275{
a40c4f10 1276 struct ceph_osd_request *req, *nreq;
6f6c7006
SW		 1277 struct rb_node *p;
1278 int needmap = 0;
1279 int err;
422d2cb8 1280
38d6453c 1281 dout("kick_requests %s\n", force_resend ? " (force resend)" : "");
422d2cb8 1282 mutex_lock(&osdc->request_mutex);
6194ea89 1283 for (p = rb_first(&osdc->requests); p; ) {
6f6c7006 1284 req = rb_entry(p, struct ceph_osd_request, r_node);
6194ea89 1285 p = rb_next(p);
38d6453c 1286 err = __map_request(osdc, req, force_resend);
6f6c7006
SW		 1287 if (err < 0)
1288 continue; /* error */
1289 if (req->r_osd == NULL) {
1290 dout("%p tid %llu maps to no osd\n", req, req->r_tid);
1291 needmap++; /* request a newer map */
1292 } else if (err > 0) {
6194ea89
SW		 1293 if (!req->r_linger) {
1294 dout("%p tid %llu requeued on osd%d\n", req,
1295 req->r_tid,
1296 req->r_osd ? req->r_osd->o_osd : -1);
a40c4f10 1297 req->r_flags |= CEPH_OSD_FLAG_RETRY;
6194ea89
SW		 1298 }
1299 }
1300 if (req->r_linger && list_empty(&req->r_linger_item)) {
1301 /*
1302 * register as a linger so that we will
1303 * re-submit below and get a new tid
1304 */
1305 dout("%p tid %llu restart on osd%d\n",
1306 req, req->r_tid,
1307 req->r_osd ? req->r_osd->o_osd : -1);
1308 __register_linger_request(osdc, req);
1309 __unregister_request(osdc, req);
a40c4f10
YS		 1310 }
1311 }
1312
1313 list_for_each_entry_safe(req, nreq, &osdc->req_linger,
1314 r_linger_item) {
1315 dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
1316
38d6453c 1317 err = __map_request(osdc, req, force_resend);
a40c4f10
YS		 1318 if (err == 0)
1319 continue; /* no change and no osd was specified */
1320 if (err < 0)
1321 continue; /* hrm! */
1322 if (req->r_osd == NULL) {
1323 dout("tid %llu maps to no valid osd\n", req->r_tid);
1324 needmap++; /* request a newer map */
1325 continue;
6f6c7006 1326 }
a40c4f10
YS		 1327
1328 dout("kicking lingering %p tid %llu osd%d\n", req, req->r_tid,
1329 req->r_osd ? req->r_osd->o_osd : -1);
1330 __unregister_linger_request(osdc, req);
1331 __register_request(osdc, req);
6f6c7006 1332 }
f24e9980
SW		 1333 mutex_unlock(&osdc->request_mutex);
1334
1335 if (needmap) {
1336 dout("%d requests for down osds, need new map\n", needmap);
1337 ceph_monc_request_next_osdmap(&osdc->client->monc);
1338 }
422d2cb8 1339}
6f6c7006
SW		 1340
1341
f24e9980
SW		 1342/*
1343 * Process updated osd map.
1344 *
1345 * The message contains any number of incremental and full maps, normally
1346 * indicating some sort of topology change in the cluster. Kick requests
1347 * off to different OSDs as needed.
1348 */
1349void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1350{
1351 void *p, *end, *next;
1352 u32 nr_maps, maplen;
1353 u32 epoch;
1354 struct ceph_osdmap *newmap = NULL, *oldmap;
1355 int err;
1356 struct ceph_fsid fsid;
1357
1358 dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
1359 p = msg->front.iov_base;
1360 end = p + msg->front.iov_len;
1361
1362 /* verify fsid */
1363 ceph_decode_need(&p, end, sizeof(fsid), bad);
1364 ceph_decode_copy(&p, &fsid, sizeof(fsid));
0743304d
SW		 1365 if (ceph_check_fsid(osdc->client, &fsid) < 0)
1366 return;
f24e9980
SW		 1367
1368 down_write(&osdc->map_sem);
1369
1370 /* incremental maps */
1371 ceph_decode_32_safe(&p, end, nr_maps, bad);
1372 dout(" %d inc maps\n", nr_maps);
1373 while (nr_maps > 0) {
1374 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 1375 epoch = ceph_decode_32(&p);
1376 maplen = ceph_decode_32(&p);
f24e9980
SW		 1377 ceph_decode_need(&p, end, maplen, bad);
1378 next = p + maplen;
1379 if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
1380 dout("applying incremental map %u len %d\n",
1381 epoch, maplen);
1382 newmap = osdmap_apply_incremental(&p, next,
1383 osdc->osdmap,
15d9882c 1384 &osdc->client->msgr);
f24e9980
SW		 1385 if (IS_ERR(newmap)) {
1386 err = PTR_ERR(newmap);
1387 goto bad;
1388 }
30dc6381 1389 BUG_ON(!newmap);
f24e9980
SW		 1390 if (newmap != osdc->osdmap) {
1391 ceph_osdmap_destroy(osdc->osdmap);
1392 osdc->osdmap = newmap;
1393 }
38d6453c 1394 kick_requests(osdc, 0);
6f6c7006 1395 reset_changed_osds(osdc);
f24e9980
SW		 1396 } else {
1397 dout("ignoring incremental map %u len %d\n",
1398 epoch, maplen);
1399 }
1400 p = next;
1401 nr_maps--;
1402 }
1403 if (newmap)
1404 goto done;
1405
1406 /* full maps */
1407 ceph_decode_32_safe(&p, end, nr_maps, bad);
1408 dout(" %d full maps\n", nr_maps);
1409 while (nr_maps) {
1410 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 1411 epoch = ceph_decode_32(&p);
1412 maplen = ceph_decode_32(&p);
f24e9980
SW		 1413 ceph_decode_need(&p, end, maplen, bad);
1414 if (nr_maps > 1) {
1415 dout("skipping non-latest full map %u len %d\n",
1416 epoch, maplen);
1417 } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
1418 dout("skipping full map %u len %d, "
1419 "older than our %u\n", epoch, maplen,
1420 osdc->osdmap->epoch);
1421 } else {
38d6453c
SW		 1422 int skipped_map = 0;
1423
f24e9980
SW		 1424 dout("taking full map %u len %d\n", epoch, maplen);
1425 newmap = osdmap_decode(&p, p+maplen);
1426 if (IS_ERR(newmap)) {
1427 err = PTR_ERR(newmap);
1428 goto bad;
1429 }
30dc6381 1430 BUG_ON(!newmap);
f24e9980
SW		 1431 oldmap = osdc->osdmap;
1432 osdc->osdmap = newmap;
38d6453c
SW		 1433 if (oldmap) {
1434 if (oldmap->epoch + 1 < newmap->epoch)
1435 skipped_map = 1;
f24e9980 1436 ceph_osdmap_destroy(oldmap);
38d6453c
SW		 1437 }
1438 kick_requests(osdc, skipped_map);
f24e9980
SW		 1439 }
1440 p += maplen;
1441 nr_maps--;
1442 }
1443
1444done:
1445 downgrade_write(&osdc->map_sem);
1446 ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch);
cd634fb6
SW		 1447
1448 /*
1449 * subscribe to subsequent osdmap updates if full to ensure
1450 * we find out when we are no longer full and stop returning
1451 * ENOSPC.
1452 */
1453 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL))
1454 ceph_monc_request_next_osdmap(&osdc->client->monc);
1455
6f6c7006 1456 send_queued(osdc);
f24e9980 1457 up_read(&osdc->map_sem);
03066f23 1458 wake_up_all(&osdc->client->auth_wq);
f24e9980
SW		 1459 return;
1460
1461bad:
1462 pr_err("osdc handle_map corrupt msg\n");
9ec7cab1 1463 ceph_msg_dump(msg);
f24e9980
SW		 1464 up_write(&osdc->map_sem);
1465 return;
1466}
1467
a40c4f10
YS		 1468/*
1469 * watch/notify callback event infrastructure
1470 *
1471 * These callbacks are used both for watch and notify operations.
1472 */
1473static void __release_event(struct kref *kref)
1474{
1475 struct ceph_osd_event *event =
1476 container_of(kref, struct ceph_osd_event, kref);
1477
1478 dout("__release_event %p\n", event);
1479 kfree(event);
1480}
1481
1482static void get_event(struct ceph_osd_event *event)
1483{
1484 kref_get(&event->kref);
1485}
1486
1487void ceph_osdc_put_event(struct ceph_osd_event *event)
1488{
1489 kref_put(&event->kref, __release_event);
1490}
1491EXPORT_SYMBOL(ceph_osdc_put_event);
1492
1493static void __insert_event(struct ceph_osd_client *osdc,
1494 struct ceph_osd_event *new)
1495{
1496 struct rb_node **p = &osdc->event_tree.rb_node;
1497 struct rb_node *parent = NULL;
1498 struct ceph_osd_event *event = NULL;
1499
1500 while (*p) {
1501 parent = *p;
1502 event = rb_entry(parent, struct ceph_osd_event, node);
1503 if (new->cookie < event->cookie)
1504 p = &(*p)->rb_left;
1505 else if (new->cookie > event->cookie)
1506 p = &(*p)->rb_right;
1507 else
1508 BUG();
1509 }
1510
1511 rb_link_node(&new->node, parent, p);
1512 rb_insert_color(&new->node, &osdc->event_tree);
1513}
1514
1515static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
1516 u64 cookie)
1517{
1518 struct rb_node **p = &osdc->event_tree.rb_node;
1519 struct rb_node *parent = NULL;
1520 struct ceph_osd_event *event = NULL;
1521
1522 while (*p) {
1523 parent = *p;
1524 event = rb_entry(parent, struct ceph_osd_event, node);
1525 if (cookie < event->cookie)
1526 p = &(*p)->rb_left;
1527 else if (cookie > event->cookie)
1528 p = &(*p)->rb_right;
1529 else
1530 return event;
1531 }
1532 return NULL;
1533}
1534
1535static void __remove_event(struct ceph_osd_event *event)
1536{
1537 struct ceph_osd_client *osdc = event->osdc;
1538
1539 if (!RB_EMPTY_NODE(&event->node)) {
1540 dout("__remove_event removed %p\n", event);
1541 rb_erase(&event->node, &osdc->event_tree);
1542 ceph_osdc_put_event(event);
1543 } else {
1544 dout("__remove_event didn't remove %p\n", event);
1545 }
1546}
1547
1548int ceph_osdc_create_event(struct ceph_osd_client *osdc,
1549 void (*event_cb)(u64, u64, u8, void *),
1550 int one_shot, void *data,
1551 struct ceph_osd_event **pevent)
1552{
1553 struct ceph_osd_event *event;
1554
1555 event = kmalloc(sizeof(*event), GFP_NOIO);
1556 if (!event)
1557 return -ENOMEM;
1558
1559 dout("create_event %p\n", event);
1560 event->cb = event_cb;
1561 event->one_shot = one_shot;
1562 event->data = data;
1563 event->osdc = osdc;
1564 INIT_LIST_HEAD(&event->osd_node);
1565 kref_init(&event->kref); /* one ref for us */
1566 kref_get(&event->kref); /* one ref for the caller */
1567 init_completion(&event->completion);
1568
1569 spin_lock(&osdc->event_lock);
1570 event->cookie = ++osdc->event_count;
1571 __insert_event(osdc, event);
1572 spin_unlock(&osdc->event_lock);
1573
1574 *pevent = event;
1575 return 0;
1576}
1577EXPORT_SYMBOL(ceph_osdc_create_event);
1578
1579void ceph_osdc_cancel_event(struct ceph_osd_event *event)
1580{
1581 struct ceph_osd_client *osdc = event->osdc;
1582
1583 dout("cancel_event %p\n", event);
1584 spin_lock(&osdc->event_lock);
1585 __remove_event(event);
1586 spin_unlock(&osdc->event_lock);
1587 ceph_osdc_put_event(event); /* caller's */
1588}
1589EXPORT_SYMBOL(ceph_osdc_cancel_event);
1590
1591
1592static void do_event_work(struct work_struct *work)
1593{
1594 struct ceph_osd_event_work *event_work =
1595 container_of(work, struct ceph_osd_event_work, work);
1596 struct ceph_osd_event *event = event_work->event;
1597 u64 ver = event_work->ver;
1598 u64 notify_id = event_work->notify_id;
1599 u8 opcode = event_work->opcode;
1600
1601 dout("do_event_work completing %p\n", event);
1602 event->cb(ver, notify_id, opcode, event->data);
1603 complete(&event->completion);
1604 dout("do_event_work completed %p\n", event);
1605 ceph_osdc_put_event(event);
1606 kfree(event_work);
1607}
1608
1609
1610/*
1611 * Process osd watch notifications
1612 */
1613void handle_watch_notify(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1614{
1615 void *p, *end;
1616 u8 proto_ver;
1617 u64 cookie, ver, notify_id;
1618 u8 opcode;
1619 struct ceph_osd_event *event;
1620 struct ceph_osd_event_work *event_work;
1621
1622 p = msg->front.iov_base;
1623 end = p + msg->front.iov_len;
1624
1625 ceph_decode_8_safe(&p, end, proto_ver, bad);
1626 ceph_decode_8_safe(&p, end, opcode, bad);
1627 ceph_decode_64_safe(&p, end, cookie, bad);
1628 ceph_decode_64_safe(&p, end, ver, bad);
1629 ceph_decode_64_safe(&p, end, notify_id, bad);
1630
1631 spin_lock(&osdc->event_lock);
1632 event = __find_event(osdc, cookie);
1633 if (event) {
1634 get_event(event);
1635 if (event->one_shot)
1636 __remove_event(event);
1637 }
1638 spin_unlock(&osdc->event_lock);
1639 dout("handle_watch_notify cookie %lld ver %lld event %p\n",
1640 cookie, ver, event);
1641 if (event) {
1642 event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
a40c4f10
YS		 1643 if (!event_work) {
1644 dout("ERROR: could not allocate event_work\n");
1645 goto done_err;
1646 }
6b0ae409 1647 INIT_WORK(&event_work->work, do_event_work);
a40c4f10
YS		 1648 event_work->event = event;
1649 event_work->ver = ver;
1650 event_work->notify_id = notify_id;
1651 event_work->opcode = opcode;
1652 if (!queue_work(osdc->notify_wq, &event_work->work)) {
1653 dout("WARNING: failed to queue notify event work\n");
1654 goto done_err;
1655 }
1656 }
1657
1658 return;
1659
1660done_err:
1661 complete(&event->completion);
1662 ceph_osdc_put_event(event);
1663 return;
1664
1665bad:
1666 pr_err("osdc handle_watch_notify corrupt msg\n");
1667 return;
1668}
1669
1670int ceph_osdc_wait_event(struct ceph_osd_event *event, unsigned long timeout)
1671{
1672 int err;
1673
1674 dout("wait_event %p\n", event);
1675 err = wait_for_completion_interruptible_timeout(&event->completion,
1676 timeout * HZ);
1677 ceph_osdc_put_event(event);
1678 if (err > 0)
1679 err = 0;
1680 dout("wait_event %p returns %d\n", event, err);
1681 return err;
1682}
1683EXPORT_SYMBOL(ceph_osdc_wait_event);
1684
f24e9980
SW		 1685/*
1686 * Register request, send initial attempt.
1687 */
1688int ceph_osdc_start_request(struct ceph_osd_client *osdc,
1689 struct ceph_osd_request *req,
1690 bool nofail)
1691{
c1ea8823 1692 int rc = 0;
f24e9980
SW		 1693
1694 req->r_request->pages = req->r_pages;
1695 req->r_request->nr_pages = req->r_num_pages;
68b4476b
YS		 1696#ifdef CONFIG_BLOCK
1697 req->r_request->bio = req->r_bio;
1698#endif
1699 req->r_request->trail = req->r_trail;
f24e9980
SW		 1700
1701 register_request(osdc, req);
1702
1703 down_read(&osdc->map_sem);
1704 mutex_lock(&osdc->request_mutex);
c1ea8823
SW		 1705 /*
1706 * a racing kick_requests() may have sent the message for us
1707 * while we dropped request_mutex above, so only send now if
1708 * the request still han't been touched yet.
1709 */
1710 if (req->r_sent == 0) {
38d6453c 1711 rc = __map_request(osdc, req, 0);
9d6fcb08
SW		 1712 if (rc < 0) {
1713 if (nofail) {
1714 dout("osdc_start_request failed map, "
1715 " will retry %lld\n", req->r_tid);
1716 rc = 0;
1717 }
234af26f 1718 goto out_unlock;
9d6fcb08 1719 }
6f6c7006
SW		 1720 if (req->r_osd == NULL) {
1721 dout("send_request %p no up osds in pg\n", req);
1722 ceph_monc_request_next_osdmap(&osdc->client->monc);
1723 } else {
56e925b6 1724 __send_request(osdc, req);
f24e9980 1725 }
56e925b6 1726 rc = 0;
f24e9980 1727 }
234af26f
DC		 1728
1729out_unlock:
f24e9980
SW		 1730 mutex_unlock(&osdc->request_mutex);
1731 up_read(&osdc->map_sem);
1732 return rc;
1733}
3d14c5d2 1734EXPORT_SYMBOL(ceph_osdc_start_request);
f24e9980
SW		 1735
1736/*
1737 * wait for a request to complete
1738 */
1739int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
1740 struct ceph_osd_request *req)
1741{
1742 int rc;
1743
1744 rc = wait_for_completion_interruptible(&req->r_completion);
1745 if (rc < 0) {
1746 mutex_lock(&osdc->request_mutex);
1747 __cancel_request(req);
529cfcc4 1748 __unregister_request(osdc, req);
f24e9980 1749 mutex_unlock(&osdc->request_mutex);
25845472 1750 complete_request(req);
529cfcc4 1751 dout("wait_request tid %llu canceled/timed out\n", req->r_tid);
f24e9980
SW		 1752 return rc;
1753 }
1754
1755 dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result);
1756 return req->r_result;
1757}
3d14c5d2 1758EXPORT_SYMBOL(ceph_osdc_wait_request);
f24e9980
SW		 1759
1760/*
1761 * sync - wait for all in-flight requests to flush. avoid starvation.
1762 */
1763void ceph_osdc_sync(struct ceph_osd_client *osdc)
1764{
1765 struct ceph_osd_request *req;
1766 u64 last_tid, next_tid = 0;
1767
1768 mutex_lock(&osdc->request_mutex);
1769 last_tid = osdc->last_tid;
1770 while (1) {
1771 req = __lookup_request_ge(osdc, next_tid);
1772 if (!req)
1773 break;
1774 if (req->r_tid > last_tid)
1775 break;
1776
1777 next_tid = req->r_tid + 1;
1778 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
1779 continue;
1780
1781 ceph_osdc_get_request(req);
1782 mutex_unlock(&osdc->request_mutex);
1783 dout("sync waiting on tid %llu (last is %llu)\n",
1784 req->r_tid, last_tid);
1785 wait_for_completion(&req->r_safe_completion);
1786 mutex_lock(&osdc->request_mutex);
1787 ceph_osdc_put_request(req);
1788 }
1789 mutex_unlock(&osdc->request_mutex);
1790 dout("sync done (thru tid %llu)\n", last_tid);
1791}
3d14c5d2 1792EXPORT_SYMBOL(ceph_osdc_sync);
f24e9980
SW		 1793
1794/*
1795 * init, shutdown
1796 */
1797int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
1798{
1799 int err;
1800
1801 dout("init\n");
1802 osdc->client = client;
1803 osdc->osdmap = NULL;
1804 init_rwsem(&osdc->map_sem);
1805 init_completion(&osdc->map_waiters);
1806 osdc->last_requested_map = 0;
1807 mutex_init(&osdc->request_mutex);
f24e9980
SW		 1808 osdc->last_tid = 0;
1809 osdc->osds = RB_ROOT;
f5a2041b 1810 INIT_LIST_HEAD(&osdc->osd_lru);
f24e9980 1811 osdc->requests = RB_ROOT;
422d2cb8 1812 INIT_LIST_HEAD(&osdc->req_lru);
6f6c7006
SW		 1813 INIT_LIST_HEAD(&osdc->req_unsent);
1814 INIT_LIST_HEAD(&osdc->req_notarget);
a40c4f10 1815 INIT_LIST_HEAD(&osdc->req_linger);
f24e9980
SW		 1816 osdc->num_requests = 0;
1817 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
f5a2041b 1818 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
a40c4f10
YS		 1819 spin_lock_init(&osdc->event_lock);
1820 osdc->event_tree = RB_ROOT;
1821 osdc->event_count = 0;
f5a2041b
YS		 1822
1823 schedule_delayed_work(&osdc->osds_timeout_work,
3d14c5d2 1824 round_jiffies_relative(osdc->client->options->osd_idle_ttl * HZ));
f24e9980 1825
5f44f142 1826 err = -ENOMEM;
f24e9980
SW		 1827 osdc->req_mempool = mempool_create_kmalloc_pool(10,
1828 sizeof(struct ceph_osd_request));
1829 if (!osdc->req_mempool)
5f44f142 1830 goto out;
f24e9980 1831
d50b409f
SW		 1832 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
1833 OSD_OP_FRONT_LEN, 10, true,
4f48280e 1834 "osd_op");
f24e9980 1835 if (err < 0)
5f44f142 1836 goto out_mempool;
d50b409f 1837 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
4f48280e
SW		 1838 OSD_OPREPLY_FRONT_LEN, 10, true,
1839 "osd_op_reply");
c16e7869
SW		 1840 if (err < 0)
1841 goto out_msgpool;
a40c4f10
YS		 1842
1843 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
1844 if (IS_ERR(osdc->notify_wq)) {
1845 err = PTR_ERR(osdc->notify_wq);
1846 osdc->notify_wq = NULL;
1847 goto out_msgpool;
1848 }
f24e9980 1849 return 0;
5f44f142 1850
c16e7869
SW		 1851out_msgpool:
1852 ceph_msgpool_destroy(&osdc->msgpool_op);
5f44f142
SW		 1853out_mempool:
1854 mempool_destroy(osdc->req_mempool);
1855out:
1856 return err;
f24e9980 1857}
3d14c5d2 1858EXPORT_SYMBOL(ceph_osdc_init);
f24e9980
SW		 1859
1860void ceph_osdc_stop(struct ceph_osd_client *osdc)
1861{
a40c4f10
YS		 1862 flush_workqueue(osdc->notify_wq);
1863 destroy_workqueue(osdc->notify_wq);
f24e9980 1864 cancel_delayed_work_sync(&osdc->timeout_work);
f5a2041b 1865 cancel_delayed_work_sync(&osdc->osds_timeout_work);
f24e9980
SW		 1866 if (osdc->osdmap) {
1867 ceph_osdmap_destroy(osdc->osdmap);
1868 osdc->osdmap = NULL;
1869 }
aca420bc 1870 remove_all_osds(osdc);
f24e9980
SW		 1871 mempool_destroy(osdc->req_mempool);
1872 ceph_msgpool_destroy(&osdc->msgpool_op);
c16e7869 1873 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
f24e9980 1874}
3d14c5d2 1875EXPORT_SYMBOL(ceph_osdc_stop);
f24e9980
SW		 1876
1877/*
1878 * Read some contiguous pages. If we cross a stripe boundary, shorten
1879 * *plen. Return number of bytes read, or error.
1880 */
1881int ceph_osdc_readpages(struct ceph_osd_client *osdc,
1882 struct ceph_vino vino, struct ceph_file_layout *layout,
1883 u64 off, u64 *plen,
1884 u32 truncate_seq, u64 truncate_size,
b7495fc2 1885 struct page **pages, int num_pages, int page_align)
f24e9980
SW		 1886{
1887 struct ceph_osd_request *req;
1888 int rc = 0;
1889
1890 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
1891 vino.snap, off, *plen);
1892 req = ceph_osdc_new_request(osdc, layout, vino, off, plen,
1893 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
1894 NULL, 0, truncate_seq, truncate_size, NULL,
b7495fc2 1895 false, 1, page_align);
6816282d
SW		 1896 if (IS_ERR(req))
1897 return PTR_ERR(req);
f24e9980
SW		 1898
1899 /* it may be a short read due to an object boundary */
1900 req->r_pages = pages;
f24e9980 1901
b7495fc2
SW		 1902 dout("readpages final extent is %llu~%llu (%d pages align %d)\n",
1903 off, *plen, req->r_num_pages, page_align);
f24e9980
SW		 1904
1905 rc = ceph_osdc_start_request(osdc, req, false);
1906 if (!rc)
1907 rc = ceph_osdc_wait_request(osdc, req);
1908
1909 ceph_osdc_put_request(req);
1910 dout("readpages result %d\n", rc);
1911 return rc;
1912}
3d14c5d2 1913EXPORT_SYMBOL(ceph_osdc_readpages);
f24e9980
SW		 1914
1915/*
1916 * do a synchronous write on N pages
1917 */
1918int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
1919 struct ceph_file_layout *layout,
1920 struct ceph_snap_context *snapc,
1921 u64 off, u64 len,
1922 u32 truncate_seq, u64 truncate_size,
1923 struct timespec *mtime,
1924 struct page **pages, int num_pages,
1925 int flags, int do_sync, bool nofail)
1926{
1927 struct ceph_osd_request *req;
1928 int rc = 0;
b7495fc2 1929 int page_align = off & ~PAGE_MASK;
f24e9980
SW		 1930
1931 BUG_ON(vino.snap != CEPH_NOSNAP);
1932 req = ceph_osdc_new_request(osdc, layout, vino, off, &len,
1933 CEPH_OSD_OP_WRITE,
1934 flags | CEPH_OSD_FLAG_ONDISK |
1935 CEPH_OSD_FLAG_WRITE,
1936 snapc, do_sync,
1937 truncate_seq, truncate_size, mtime,
b7495fc2 1938 nofail, 1, page_align);
6816282d
SW		 1939 if (IS_ERR(req))
1940 return PTR_ERR(req);
f24e9980
SW		 1941
1942 /* it may be a short write due to an object boundary */
1943 req->r_pages = pages;
f24e9980
SW		 1944 dout("writepages %llu~%llu (%d pages)\n", off, len,
1945 req->r_num_pages);
1946
1947 rc = ceph_osdc_start_request(osdc, req, nofail);
1948 if (!rc)
1949 rc = ceph_osdc_wait_request(osdc, req);
1950
1951 ceph_osdc_put_request(req);
1952 if (rc == 0)
1953 rc = len;
1954 dout("writepages result %d\n", rc);
1955 return rc;
1956}
3d14c5d2 1957EXPORT_SYMBOL(ceph_osdc_writepages);
f24e9980
SW		 1958
1959/*
1960 * handle incoming message
1961 */
1962static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
1963{
1964 struct ceph_osd *osd = con->private;
32c895e7 1965 struct ceph_osd_client *osdc;
f24e9980
SW		 1966 int type = le16_to_cpu(msg->hdr.type);
1967
1968 if (!osd)
4a32f93d 1969 goto out;
32c895e7 1970 osdc = osd->o_osdc;
f24e9980
SW		 1971
1972 switch (type) {
1973 case CEPH_MSG_OSD_MAP:
1974 ceph_osdc_handle_map(osdc, msg);
1975 break;
1976 case CEPH_MSG_OSD_OPREPLY:
350b1c32 1977 handle_reply(osdc, msg, con);
f24e9980 1978 break;
a40c4f10
YS		 1979 case CEPH_MSG_WATCH_NOTIFY:
1980 handle_watch_notify(osdc, msg);
1981 break;
f24e9980
SW		 1982
1983 default:
1984 pr_err("received unknown message type %d %s\n", type,
1985 ceph_msg_type_name(type));
1986 }
4a32f93d 1987out:
f24e9980
SW		 1988 ceph_msg_put(msg);
1989}
1990
5b3a4db3 1991/*
21b667f6
SW		 1992 * lookup and return message for incoming reply. set up reply message
1993 * pages.
5b3a4db3
SW		 1994 */
1995static struct ceph_msg *get_reply(struct ceph_connection *con,
2450418c
YS		 1996 struct ceph_msg_header *hdr,
1997 int *skip)
f24e9980
SW		 1998{
1999 struct ceph_osd *osd = con->private;
2000 struct ceph_osd_client *osdc = osd->o_osdc;
2450418c 2001 struct ceph_msg *m;
0547a9b3 2002 struct ceph_osd_request *req;
5b3a4db3
SW		 2003 int front = le32_to_cpu(hdr->front_len);
2004 int data_len = le32_to_cpu(hdr->data_len);
0547a9b3 2005 u64 tid;
f24e9980 2006
0547a9b3
YS		 2007 tid = le64_to_cpu(hdr->tid);
2008 mutex_lock(&osdc->request_mutex);
2009 req = __lookup_request(osdc, tid);
2010 if (!req) {
2011 *skip = 1;
2012 m = NULL;
756a16a5
SW		 2013 dout("get_reply unknown tid %llu from osd%d\n", tid,
2014 osd->o_osd);
0547a9b3
YS		 2015 goto out;
2016 }
c16e7869
SW		 2017
2018 if (req->r_con_filling_msg) {
8921d114 2019 dout("%s revoking msg %p from old con %p\n", __func__,
c16e7869 2020 req->r_reply, req->r_con_filling_msg);
8921d114 2021 ceph_msg_revoke_incoming(req->r_reply);
0d47766f 2022 req->r_con_filling_msg->ops->put(req->r_con_filling_msg);
6f46cb29 2023 req->r_con_filling_msg = NULL;
0547a9b3
YS		 2024 }
2025
c16e7869
SW		 2026 if (front > req->r_reply->front.iov_len) {
2027 pr_warning("get_reply front %d > preallocated %d\n",
2028 front, (int)req->r_reply->front.iov_len);
b61c2763 2029 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front, GFP_NOFS, false);
a79832f2 2030 if (!m)
c16e7869
SW		 2031 goto out;
2032 ceph_msg_put(req->r_reply);
2033 req->r_reply = m;
2034 }
2035 m = ceph_msg_get(req->r_reply);
2036
0547a9b3 2037 if (data_len > 0) {
b7495fc2 2038 int want = calc_pages_for(req->r_page_alignment, data_len);
21b667f6
SW		 2039
2040 if (unlikely(req->r_num_pages < want)) {
9bb0ce2b
SW		 2041 pr_warning("tid %lld reply has %d bytes %d pages, we"
2042 " had only %d pages ready\n", tid, data_len,
2043 want, req->r_num_pages);
0547a9b3
YS		 2044 *skip = 1;
2045 ceph_msg_put(m);
a79832f2 2046 m = NULL;
21b667f6 2047 goto out;
0547a9b3 2048 }
21b667f6
SW		 2049 m->pages = req->r_pages;
2050 m->nr_pages = req->r_num_pages;
c5c6b19d 2051 m->page_alignment = req->r_page_alignment;
68b4476b
YS		 2052#ifdef CONFIG_BLOCK
2053 m->bio = req->r_bio;
2054#endif
0547a9b3 2055 }
5b3a4db3 2056 *skip = 0;
0d47766f 2057 req->r_con_filling_msg = con->ops->get(con);
c16e7869 2058 dout("get_reply tid %lld %p\n", tid, m);
0547a9b3
YS		 2059
2060out:
2061 mutex_unlock(&osdc->request_mutex);
2450418c 2062 return m;
5b3a4db3
SW		 2063
2064}
2065
2066static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2067 struct ceph_msg_header *hdr,
2068 int *skip)
2069{
2070 struct ceph_osd *osd = con->private;
2071 int type = le16_to_cpu(hdr->type);
2072 int front = le32_to_cpu(hdr->front_len);
2073
1c20f2d2 2074 *skip = 0;
5b3a4db3
SW		 2075 switch (type) {
2076 case CEPH_MSG_OSD_MAP:
a40c4f10 2077 case CEPH_MSG_WATCH_NOTIFY:
b61c2763 2078 return ceph_msg_new(type, front, GFP_NOFS, false);
5b3a4db3
SW		 2079 case CEPH_MSG_OSD_OPREPLY:
2080 return get_reply(con, hdr, skip);
2081 default:
2082 pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
2083 osd->o_osd);
2084 *skip = 1;
2085 return NULL;
2086 }
f24e9980
SW		 2087}
2088
2089/*
2090 * Wrappers to refcount containing ceph_osd struct
2091 */
2092static struct ceph_connection *get_osd_con(struct ceph_connection *con)
2093{
2094 struct ceph_osd *osd = con->private;
2095 if (get_osd(osd))
2096 return con;
2097 return NULL;
2098}
2099
2100static void put_osd_con(struct ceph_connection *con)
2101{
2102 struct ceph_osd *osd = con->private;
2103 put_osd(osd);
2104}
2105
4e7a5dcd
SW		 2106/*
2107 * authentication
2108 */
a3530df3
AE		 2109/*
2110 * Note: returned pointer is the address of a structure that's
2111 * managed separately. Caller must *not* attempt to free it.
2112 */
2113static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
8f43fb53 2114 int *proto, int force_new)
4e7a5dcd
SW		 2115{
2116 struct ceph_osd *o = con->private;
2117 struct ceph_osd_client *osdc = o->o_osdc;
2118 struct ceph_auth_client *ac = osdc->client->monc.auth;
74f1869f 2119 struct ceph_auth_handshake *auth = &o->o_auth;
4e7a5dcd 2120
74f1869f 2121 if (force_new && auth->authorizer) {
a255651d
AE		 2122 if (ac->ops && ac->ops->destroy_authorizer)
2123 ac->ops->destroy_authorizer(ac, auth->authorizer);
74f1869f
AE		 2124 auth->authorizer = NULL;
2125 }
a255651d 2126 if (!auth->authorizer && ac->ops && ac->ops->create_authorizer) {
a3530df3
AE		 2127 int ret = ac->ops->create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2128 auth);
4e7a5dcd 2129 if (ret)
a3530df3 2130 return ERR_PTR(ret);
4e7a5dcd 2131 }
4e7a5dcd 2132 *proto = ac->protocol;
74f1869f 2133
a3530df3 2134 return auth;
4e7a5dcd
SW		 2135}
2136
2137
2138static int verify_authorizer_reply(struct ceph_connection *con, int len)
2139{
2140 struct ceph_osd *o = con->private;
2141 struct ceph_osd_client *osdc = o->o_osdc;
2142 struct ceph_auth_client *ac = osdc->client->monc.auth;
2143
a255651d
AE		 2144 /*
2145 * XXX If ac->ops or ac->ops->verify_authorizer_reply is null,
2146 * XXX which do we do: succeed or fail?
2147 */
6c4a1915 2148 return ac->ops->verify_authorizer_reply(ac, o->o_auth.authorizer, len);
4e7a5dcd
SW		 2149}
2150
9bd2e6f8
SW		 2151static int invalidate_authorizer(struct ceph_connection *con)
2152{
2153 struct ceph_osd *o = con->private;
2154 struct ceph_osd_client *osdc = o->o_osdc;
2155 struct ceph_auth_client *ac = osdc->client->monc.auth;
2156
a255651d 2157 if (ac->ops && ac->ops->invalidate_authorizer)
9bd2e6f8
SW		 2158 ac->ops->invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
2159
2160 return ceph_monc_validate_auth(&osdc->client->monc);
2161}
4e7a5dcd 2162
9e32789f 2163static const struct ceph_connection_operations osd_con_ops = {
f24e9980
SW		 2164 .get = get_osd_con,
2165 .put = put_osd_con,
2166 .dispatch = dispatch,
4e7a5dcd
SW		 2167 .get_authorizer = get_authorizer,
2168 .verify_authorizer_reply = verify_authorizer_reply,
9bd2e6f8 2169 .invalidate_authorizer = invalidate_authorizer,
f24e9980 2170 .alloc_msg = alloc_msg,
81b024e7 2171 .fault = osd_reset,
f24e9980 2172};
Linux 6.x block layer and io_uring tree(s)