Commit | Line | Data |
---|---|---|
f24e9980 SW |
1 | #include "ceph_debug.h" |
2 | ||
3 | #include <linux/err.h> | |
4 | #include <linux/highmem.h> | |
5 | #include <linux/mm.h> | |
6 | #include <linux/pagemap.h> | |
7 | #include <linux/slab.h> | |
8 | #include <linux/uaccess.h> | |
9 | ||
10 | #include "super.h" | |
11 | #include "osd_client.h" | |
12 | #include "messenger.h" | |
13 | #include "decode.h" | |
4e7a5dcd | 14 | #include "auth.h" |
f24e9980 SW |
15 | |
16 | const static struct ceph_connection_operations osd_con_ops; | |
17 | ||
18 | static void kick_requests(struct ceph_osd_client *osdc, struct ceph_osd *osd); | |
19 | ||
20 | /* | |
21 | * Implement client access to distributed object storage cluster. | |
22 | * | |
23 | * All data objects are stored within a cluster/cloud of OSDs, or | |
24 | * "object storage devices." (Note that Ceph OSDs have _nothing_ to | |
25 | * do with the T10 OSD extensions to SCSI.) Ceph OSDs are simply | |
26 | * remote daemons serving up and coordinating consistent and safe | |
27 | * access to storage. | |
28 | * | |
29 | * Cluster membership and the mapping of data objects onto storage devices | |
30 | * are described by the osd map. | |
31 | * | |
32 | * We keep track of pending OSD requests (read, write), resubmit | |
33 | * requests to different OSDs when the cluster topology/data layout | |
34 | * change, or retry the affected requests when the communications | |
35 | * channel with an OSD is reset. | |
36 | */ | |
37 | ||
38 | /* | |
39 | * calculate the mapping of a file extent onto an object, and fill out the | |
40 | * request accordingly. shorten extent as necessary if it crosses an | |
41 | * object boundary. | |
42 | * | |
43 | * fill osd op in request message. | |
44 | */ | |
45 | static void calc_layout(struct ceph_osd_client *osdc, | |
46 | struct ceph_vino vino, struct ceph_file_layout *layout, | |
47 | u64 off, u64 *plen, | |
48 | struct ceph_osd_request *req) | |
49 | { | |
50 | struct ceph_osd_request_head *reqhead = req->r_request->front.iov_base; | |
51 | struct ceph_osd_op *op = (void *)(reqhead + 1); | |
52 | u64 orig_len = *plen; | |
53 | u64 objoff, objlen; /* extent in object */ | |
54 | u64 bno; | |
55 | ||
56 | reqhead->snapid = cpu_to_le64(vino.snap); | |
57 | ||
58 | /* object extent? */ | |
59 | ceph_calc_file_object_mapping(layout, off, plen, &bno, | |
60 | &objoff, &objlen); | |
61 | if (*plen < orig_len) | |
62 | dout(" skipping last %llu, final file extent %llu~%llu\n", | |
63 | orig_len - *plen, off, *plen); | |
64 | ||
65 | sprintf(req->r_oid, "%llx.%08llx", vino.ino, bno); | |
66 | req->r_oid_len = strlen(req->r_oid); | |
67 | ||
68 | op->extent.offset = cpu_to_le64(objoff); | |
69 | op->extent.length = cpu_to_le64(objlen); | |
70 | req->r_num_pages = calc_pages_for(off, *plen); | |
71 | ||
72 | dout("calc_layout %s (%d) %llu~%llu (%d pages)\n", | |
73 | req->r_oid, req->r_oid_len, objoff, objlen, req->r_num_pages); | |
74 | } | |
75 | ||
76 | ||
77 | /* | |
78 | * requests | |
79 | */ | |
80 | void ceph_osdc_put_request(struct ceph_osd_request *req) | |
81 | { | |
82 | dout("osdc put_request %p %d -> %d\n", req, atomic_read(&req->r_ref), | |
83 | atomic_read(&req->r_ref)-1); | |
84 | BUG_ON(atomic_read(&req->r_ref) <= 0); | |
85 | if (atomic_dec_and_test(&req->r_ref)) { | |
86 | if (req->r_request) | |
87 | ceph_msg_put(req->r_request); | |
88 | if (req->r_reply) | |
89 | ceph_msg_put(req->r_reply); | |
90 | if (req->r_own_pages) | |
91 | ceph_release_page_vector(req->r_pages, | |
92 | req->r_num_pages); | |
93 | ceph_put_snap_context(req->r_snapc); | |
94 | if (req->r_mempool) | |
95 | mempool_free(req, req->r_osdc->req_mempool); | |
96 | else | |
97 | kfree(req); | |
98 | } | |
99 | } | |
100 | ||
101 | /* | |
102 | * build new request AND message, calculate layout, and adjust file | |
103 | * extent as needed. | |
104 | * | |
105 | * if the file was recently truncated, we include information about its | |
106 | * old and new size so that the object can be updated appropriately. (we | |
107 | * avoid synchronously deleting truncated objects because it's slow.) | |
108 | * | |
109 | * if @do_sync, include a 'startsync' command so that the osd will flush | |
110 | * data quickly. | |
111 | */ | |
112 | struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc, | |
113 | struct ceph_file_layout *layout, | |
114 | struct ceph_vino vino, | |
115 | u64 off, u64 *plen, | |
116 | int opcode, int flags, | |
117 | struct ceph_snap_context *snapc, | |
118 | int do_sync, | |
119 | u32 truncate_seq, | |
120 | u64 truncate_size, | |
121 | struct timespec *mtime, | |
122 | bool use_mempool, int num_reply) | |
123 | { | |
124 | struct ceph_osd_request *req; | |
125 | struct ceph_msg *msg; | |
126 | struct ceph_osd_request_head *head; | |
127 | struct ceph_osd_op *op; | |
128 | void *p; | |
129 | int do_trunc = truncate_seq && (off + *plen > truncate_size); | |
130 | int num_op = 1 + do_sync + do_trunc; | |
131 | size_t msg_size = sizeof(*head) + num_op*sizeof(*op); | |
132 | int err, i; | |
133 | u64 prevofs; | |
134 | ||
135 | if (use_mempool) { | |
136 | req = mempool_alloc(osdc->req_mempool, GFP_NOFS); | |
137 | memset(req, 0, sizeof(*req)); | |
138 | } else { | |
139 | req = kzalloc(sizeof(*req), GFP_NOFS); | |
140 | } | |
141 | if (req == NULL) | |
142 | return ERR_PTR(-ENOMEM); | |
143 | ||
144 | err = ceph_msgpool_resv(&osdc->msgpool_op_reply, num_reply); | |
145 | if (err) { | |
146 | ceph_osdc_put_request(req); | |
147 | return ERR_PTR(-ENOMEM); | |
148 | } | |
149 | ||
150 | req->r_osdc = osdc; | |
151 | req->r_mempool = use_mempool; | |
152 | atomic_set(&req->r_ref, 1); | |
153 | init_completion(&req->r_completion); | |
154 | init_completion(&req->r_safe_completion); | |
155 | INIT_LIST_HEAD(&req->r_unsafe_item); | |
156 | req->r_flags = flags; | |
157 | ||
158 | WARN_ON((flags & (CEPH_OSD_FLAG_READ|CEPH_OSD_FLAG_WRITE)) == 0); | |
159 | ||
160 | /* create message; allow space for oid */ | |
161 | msg_size += 40; | |
162 | if (snapc) | |
163 | msg_size += sizeof(u64) * snapc->num_snaps; | |
164 | if (use_mempool) | |
8f3bc053 | 165 | msg = ceph_msgpool_get(&osdc->msgpool_op, 0); |
f24e9980 SW |
166 | else |
167 | msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, 0, 0, NULL); | |
168 | if (IS_ERR(msg)) { | |
169 | ceph_msgpool_resv(&osdc->msgpool_op_reply, num_reply); | |
170 | ceph_osdc_put_request(req); | |
171 | return ERR_PTR(PTR_ERR(msg)); | |
172 | } | |
173 | msg->hdr.type = cpu_to_le16(CEPH_MSG_OSD_OP); | |
174 | memset(msg->front.iov_base, 0, msg->front.iov_len); | |
175 | head = msg->front.iov_base; | |
176 | op = (void *)(head + 1); | |
177 | p = (void *)(op + num_op); | |
178 | ||
179 | req->r_request = msg; | |
180 | req->r_snapc = ceph_get_snap_context(snapc); | |
181 | ||
182 | head->client_inc = cpu_to_le32(1); /* always, for now. */ | |
183 | head->flags = cpu_to_le32(flags); | |
184 | if (flags & CEPH_OSD_FLAG_WRITE) | |
185 | ceph_encode_timespec(&head->mtime, mtime); | |
186 | head->num_ops = cpu_to_le16(num_op); | |
187 | op->op = cpu_to_le16(opcode); | |
188 | ||
189 | /* calculate max write size */ | |
190 | calc_layout(osdc, vino, layout, off, plen, req); | |
191 | req->r_file_layout = *layout; /* keep a copy */ | |
192 | ||
193 | if (flags & CEPH_OSD_FLAG_WRITE) { | |
194 | req->r_request->hdr.data_off = cpu_to_le16(off); | |
195 | req->r_request->hdr.data_len = cpu_to_le32(*plen); | |
196 | op->payload_len = cpu_to_le32(*plen); | |
197 | } | |
198 | ||
199 | /* fill in oid */ | |
200 | head->object_len = cpu_to_le32(req->r_oid_len); | |
201 | memcpy(p, req->r_oid, req->r_oid_len); | |
202 | p += req->r_oid_len; | |
203 | ||
204 | /* additional ops */ | |
205 | if (do_trunc) { | |
206 | op++; | |
207 | op->op = cpu_to_le16(opcode == CEPH_OSD_OP_READ ? | |
208 | CEPH_OSD_OP_MASKTRUNC : CEPH_OSD_OP_SETTRUNC); | |
209 | op->trunc.truncate_seq = cpu_to_le32(truncate_seq); | |
210 | prevofs = le64_to_cpu((op-1)->extent.offset); | |
211 | op->trunc.truncate_size = cpu_to_le64(truncate_size - | |
212 | (off-prevofs)); | |
213 | } | |
214 | if (do_sync) { | |
215 | op++; | |
216 | op->op = cpu_to_le16(CEPH_OSD_OP_STARTSYNC); | |
217 | } | |
218 | if (snapc) { | |
219 | head->snap_seq = cpu_to_le64(snapc->seq); | |
220 | head->num_snaps = cpu_to_le32(snapc->num_snaps); | |
221 | for (i = 0; i < snapc->num_snaps; i++) { | |
222 | put_unaligned_le64(snapc->snaps[i], p); | |
223 | p += sizeof(u64); | |
224 | } | |
225 | } | |
226 | ||
227 | BUG_ON(p > msg->front.iov_base + msg->front.iov_len); | |
228 | return req; | |
229 | } | |
230 | ||
231 | /* | |
232 | * We keep osd requests in an rbtree, sorted by ->r_tid. | |
233 | */ | |
234 | static void __insert_request(struct ceph_osd_client *osdc, | |
235 | struct ceph_osd_request *new) | |
236 | { | |
237 | struct rb_node **p = &osdc->requests.rb_node; | |
238 | struct rb_node *parent = NULL; | |
239 | struct ceph_osd_request *req = NULL; | |
240 | ||
241 | while (*p) { | |
242 | parent = *p; | |
243 | req = rb_entry(parent, struct ceph_osd_request, r_node); | |
244 | if (new->r_tid < req->r_tid) | |
245 | p = &(*p)->rb_left; | |
246 | else if (new->r_tid > req->r_tid) | |
247 | p = &(*p)->rb_right; | |
248 | else | |
249 | BUG(); | |
250 | } | |
251 | ||
252 | rb_link_node(&new->r_node, parent, p); | |
253 | rb_insert_color(&new->r_node, &osdc->requests); | |
254 | } | |
255 | ||
256 | static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc, | |
257 | u64 tid) | |
258 | { | |
259 | struct ceph_osd_request *req; | |
260 | struct rb_node *n = osdc->requests.rb_node; | |
261 | ||
262 | while (n) { | |
263 | req = rb_entry(n, struct ceph_osd_request, r_node); | |
264 | if (tid < req->r_tid) | |
265 | n = n->rb_left; | |
266 | else if (tid > req->r_tid) | |
267 | n = n->rb_right; | |
268 | else | |
269 | return req; | |
270 | } | |
271 | return NULL; | |
272 | } | |
273 | ||
274 | static struct ceph_osd_request * | |
275 | __lookup_request_ge(struct ceph_osd_client *osdc, | |
276 | u64 tid) | |
277 | { | |
278 | struct ceph_osd_request *req; | |
279 | struct rb_node *n = osdc->requests.rb_node; | |
280 | ||
281 | while (n) { | |
282 | req = rb_entry(n, struct ceph_osd_request, r_node); | |
283 | if (tid < req->r_tid) { | |
284 | if (!n->rb_left) | |
285 | return req; | |
286 | n = n->rb_left; | |
287 | } else if (tid > req->r_tid) { | |
288 | n = n->rb_right; | |
289 | } else { | |
290 | return req; | |
291 | } | |
292 | } | |
293 | return NULL; | |
294 | } | |
295 | ||
296 | ||
297 | /* | |
81b024e7 | 298 | * If the osd connection drops, we need to resubmit all requests. |
f24e9980 SW |
299 | */ |
300 | static void osd_reset(struct ceph_connection *con) | |
301 | { | |
302 | struct ceph_osd *osd = con->private; | |
303 | struct ceph_osd_client *osdc; | |
304 | ||
305 | if (!osd) | |
306 | return; | |
307 | dout("osd_reset osd%d\n", osd->o_osd); | |
308 | osdc = osd->o_osdc; | |
309 | osd->o_incarnation++; | |
310 | down_read(&osdc->map_sem); | |
311 | kick_requests(osdc, osd); | |
312 | up_read(&osdc->map_sem); | |
313 | } | |
314 | ||
315 | /* | |
316 | * Track open sessions with osds. | |
317 | */ | |
318 | static struct ceph_osd *create_osd(struct ceph_osd_client *osdc) | |
319 | { | |
320 | struct ceph_osd *osd; | |
321 | ||
322 | osd = kzalloc(sizeof(*osd), GFP_NOFS); | |
323 | if (!osd) | |
324 | return NULL; | |
325 | ||
326 | atomic_set(&osd->o_ref, 1); | |
327 | osd->o_osdc = osdc; | |
328 | INIT_LIST_HEAD(&osd->o_requests); | |
329 | osd->o_incarnation = 1; | |
330 | ||
331 | ceph_con_init(osdc->client->msgr, &osd->o_con); | |
332 | osd->o_con.private = osd; | |
333 | osd->o_con.ops = &osd_con_ops; | |
334 | osd->o_con.peer_name.type = CEPH_ENTITY_TYPE_OSD; | |
4e7a5dcd | 335 | |
f24e9980 SW |
336 | return osd; |
337 | } | |
338 | ||
339 | static struct ceph_osd *get_osd(struct ceph_osd *osd) | |
340 | { | |
341 | if (atomic_inc_not_zero(&osd->o_ref)) { | |
342 | dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1, | |
343 | atomic_read(&osd->o_ref)); | |
344 | return osd; | |
345 | } else { | |
346 | dout("get_osd %p FAIL\n", osd); | |
347 | return NULL; | |
348 | } | |
349 | } | |
350 | ||
351 | static void put_osd(struct ceph_osd *osd) | |
352 | { | |
353 | dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref), | |
354 | atomic_read(&osd->o_ref) - 1); | |
42ce56e5 | 355 | if (atomic_dec_and_test(&osd->o_ref)) |
f24e9980 | 356 | kfree(osd); |
f24e9980 SW |
357 | } |
358 | ||
359 | /* | |
360 | * remove an osd from our map | |
361 | */ | |
362 | static void remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd) | |
363 | { | |
364 | dout("remove_osd %p\n", osd); | |
365 | BUG_ON(!list_empty(&osd->o_requests)); | |
366 | rb_erase(&osd->o_node, &osdc->osds); | |
367 | ceph_con_close(&osd->o_con); | |
368 | put_osd(osd); | |
369 | } | |
370 | ||
371 | /* | |
372 | * reset osd connect | |
373 | */ | |
374 | static int reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd) | |
375 | { | |
376 | int ret = 0; | |
377 | ||
378 | dout("reset_osd %p osd%d\n", osd, osd->o_osd); | |
379 | if (list_empty(&osd->o_requests)) { | |
380 | remove_osd(osdc, osd); | |
381 | } else { | |
382 | ceph_con_close(&osd->o_con); | |
383 | ceph_con_open(&osd->o_con, &osdc->osdmap->osd_addr[osd->o_osd]); | |
384 | osd->o_incarnation++; | |
385 | } | |
386 | return ret; | |
387 | } | |
388 | ||
389 | static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new) | |
390 | { | |
391 | struct rb_node **p = &osdc->osds.rb_node; | |
392 | struct rb_node *parent = NULL; | |
393 | struct ceph_osd *osd = NULL; | |
394 | ||
395 | while (*p) { | |
396 | parent = *p; | |
397 | osd = rb_entry(parent, struct ceph_osd, o_node); | |
398 | if (new->o_osd < osd->o_osd) | |
399 | p = &(*p)->rb_left; | |
400 | else if (new->o_osd > osd->o_osd) | |
401 | p = &(*p)->rb_right; | |
402 | else | |
403 | BUG(); | |
404 | } | |
405 | ||
406 | rb_link_node(&new->o_node, parent, p); | |
407 | rb_insert_color(&new->o_node, &osdc->osds); | |
408 | } | |
409 | ||
410 | static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o) | |
411 | { | |
412 | struct ceph_osd *osd; | |
413 | struct rb_node *n = osdc->osds.rb_node; | |
414 | ||
415 | while (n) { | |
416 | osd = rb_entry(n, struct ceph_osd, o_node); | |
417 | if (o < osd->o_osd) | |
418 | n = n->rb_left; | |
419 | else if (o > osd->o_osd) | |
420 | n = n->rb_right; | |
421 | else | |
422 | return osd; | |
423 | } | |
424 | return NULL; | |
425 | } | |
426 | ||
427 | ||
428 | /* | |
429 | * Register request, assign tid. If this is the first request, set up | |
430 | * the timeout event. | |
431 | */ | |
432 | static void register_request(struct ceph_osd_client *osdc, | |
433 | struct ceph_osd_request *req) | |
434 | { | |
435 | struct ceph_osd_request_head *head = req->r_request->front.iov_base; | |
436 | ||
437 | mutex_lock(&osdc->request_mutex); | |
438 | req->r_tid = ++osdc->last_tid; | |
439 | head->tid = cpu_to_le64(req->r_tid); | |
440 | ||
441 | dout("register_request %p tid %lld\n", req, req->r_tid); | |
442 | __insert_request(osdc, req); | |
443 | ceph_osdc_get_request(req); | |
444 | osdc->num_requests++; | |
445 | ||
446 | req->r_timeout_stamp = | |
6b805185 | 447 | jiffies + osdc->client->mount_args->osd_timeout*HZ; |
f24e9980 SW |
448 | |
449 | if (osdc->num_requests == 1) { | |
450 | osdc->timeout_tid = req->r_tid; | |
451 | dout(" timeout on tid %llu at %lu\n", req->r_tid, | |
452 | req->r_timeout_stamp); | |
453 | schedule_delayed_work(&osdc->timeout_work, | |
454 | round_jiffies_relative(req->r_timeout_stamp - jiffies)); | |
455 | } | |
456 | mutex_unlock(&osdc->request_mutex); | |
457 | } | |
458 | ||
459 | /* | |
460 | * called under osdc->request_mutex | |
461 | */ | |
462 | static void __unregister_request(struct ceph_osd_client *osdc, | |
463 | struct ceph_osd_request *req) | |
464 | { | |
465 | dout("__unregister_request %p tid %lld\n", req, req->r_tid); | |
466 | rb_erase(&req->r_node, &osdc->requests); | |
467 | osdc->num_requests--; | |
468 | ||
0ba6478d SW |
469 | if (req->r_osd) { |
470 | /* make sure the original request isn't in flight. */ | |
471 | ceph_con_revoke(&req->r_osd->o_con, req->r_request); | |
472 | ||
473 | list_del_init(&req->r_osd_item); | |
474 | if (list_empty(&req->r_osd->o_requests)) | |
475 | remove_osd(osdc, req->r_osd); | |
476 | req->r_osd = NULL; | |
477 | } | |
f24e9980 SW |
478 | |
479 | ceph_osdc_put_request(req); | |
480 | ||
481 | if (req->r_tid == osdc->timeout_tid) { | |
482 | if (osdc->num_requests == 0) { | |
483 | dout("no requests, canceling timeout\n"); | |
484 | osdc->timeout_tid = 0; | |
485 | cancel_delayed_work(&osdc->timeout_work); | |
486 | } else { | |
487 | req = rb_entry(rb_first(&osdc->requests), | |
488 | struct ceph_osd_request, r_node); | |
489 | osdc->timeout_tid = req->r_tid; | |
490 | dout("rescheduled timeout on tid %llu at %lu\n", | |
491 | req->r_tid, req->r_timeout_stamp); | |
492 | schedule_delayed_work(&osdc->timeout_work, | |
493 | round_jiffies_relative(req->r_timeout_stamp - | |
494 | jiffies)); | |
495 | } | |
496 | } | |
497 | } | |
498 | ||
499 | /* | |
500 | * Cancel a previously queued request message | |
501 | */ | |
502 | static void __cancel_request(struct ceph_osd_request *req) | |
503 | { | |
504 | if (req->r_sent) { | |
505 | ceph_con_revoke(&req->r_osd->o_con, req->r_request); | |
506 | req->r_sent = 0; | |
507 | } | |
508 | } | |
509 | ||
510 | /* | |
511 | * Pick an osd (the first 'up' osd in the pg), allocate the osd struct | |
512 | * (as needed), and set the request r_osd appropriately. If there is | |
513 | * no up osd, set r_osd to NULL. | |
514 | * | |
515 | * Return 0 if unchanged, 1 if changed, or negative on error. | |
516 | * | |
517 | * Caller should hold map_sem for read and request_mutex. | |
518 | */ | |
519 | static int __map_osds(struct ceph_osd_client *osdc, | |
520 | struct ceph_osd_request *req) | |
521 | { | |
522 | struct ceph_osd_request_head *reqhead = req->r_request->front.iov_base; | |
51042122 | 523 | struct ceph_pg pgid; |
f24e9980 SW |
524 | int o = -1; |
525 | int err; | |
526 | struct ceph_osd *newosd = NULL; | |
527 | ||
528 | dout("map_osds %p tid %lld\n", req, req->r_tid); | |
529 | err = ceph_calc_object_layout(&reqhead->layout, req->r_oid, | |
530 | &req->r_file_layout, osdc->osdmap); | |
531 | if (err) | |
532 | return err; | |
51042122 | 533 | pgid = reqhead->layout.ol_pgid; |
f24e9980 SW |
534 | o = ceph_calc_pg_primary(osdc->osdmap, pgid); |
535 | ||
536 | if ((req->r_osd && req->r_osd->o_osd == o && | |
537 | req->r_sent >= req->r_osd->o_incarnation) || | |
538 | (req->r_osd == NULL && o == -1)) | |
539 | return 0; /* no change */ | |
540 | ||
51042122 SW |
541 | dout("map_osds tid %llu pgid %d.%x osd%d (was osd%d)\n", |
542 | req->r_tid, le32_to_cpu(pgid.pool), le16_to_cpu(pgid.ps), o, | |
f24e9980 SW |
543 | req->r_osd ? req->r_osd->o_osd : -1); |
544 | ||
545 | if (req->r_osd) { | |
546 | __cancel_request(req); | |
547 | list_del_init(&req->r_osd_item); | |
548 | if (list_empty(&req->r_osd->o_requests)) { | |
549 | /* try to re-use r_osd if possible */ | |
550 | newosd = get_osd(req->r_osd); | |
551 | remove_osd(osdc, newosd); | |
552 | } | |
553 | req->r_osd = NULL; | |
554 | } | |
555 | ||
556 | req->r_osd = __lookup_osd(osdc, o); | |
557 | if (!req->r_osd && o >= 0) { | |
558 | if (newosd) { | |
559 | req->r_osd = newosd; | |
560 | newosd = NULL; | |
561 | } else { | |
562 | err = -ENOMEM; | |
563 | req->r_osd = create_osd(osdc); | |
564 | if (!req->r_osd) | |
565 | goto out; | |
566 | } | |
567 | ||
568 | dout("map_osds osd %p is osd%d\n", req->r_osd, o); | |
569 | req->r_osd->o_osd = o; | |
570 | req->r_osd->o_con.peer_name.num = cpu_to_le64(o); | |
571 | __insert_osd(osdc, req->r_osd); | |
572 | ||
573 | ceph_con_open(&req->r_osd->o_con, &osdc->osdmap->osd_addr[o]); | |
574 | } | |
575 | ||
576 | if (req->r_osd) | |
577 | list_add(&req->r_osd_item, &req->r_osd->o_requests); | |
578 | err = 1; /* osd changed */ | |
579 | ||
580 | out: | |
581 | if (newosd) | |
582 | put_osd(newosd); | |
583 | return err; | |
584 | } | |
585 | ||
586 | /* | |
587 | * caller should hold map_sem (for read) and request_mutex | |
588 | */ | |
589 | static int __send_request(struct ceph_osd_client *osdc, | |
590 | struct ceph_osd_request *req) | |
591 | { | |
592 | struct ceph_osd_request_head *reqhead; | |
593 | int err; | |
594 | ||
595 | err = __map_osds(osdc, req); | |
596 | if (err < 0) | |
597 | return err; | |
598 | if (req->r_osd == NULL) { | |
599 | dout("send_request %p no up osds in pg\n", req); | |
600 | ceph_monc_request_next_osdmap(&osdc->client->monc); | |
601 | return 0; | |
602 | } | |
603 | ||
604 | dout("send_request %p tid %llu to osd%d flags %d\n", | |
605 | req, req->r_tid, req->r_osd->o_osd, req->r_flags); | |
606 | ||
607 | reqhead = req->r_request->front.iov_base; | |
608 | reqhead->osdmap_epoch = cpu_to_le32(osdc->osdmap->epoch); | |
609 | reqhead->flags |= cpu_to_le32(req->r_flags); /* e.g., RETRY */ | |
610 | reqhead->reassert_version = req->r_reassert_version; | |
611 | ||
6b805185 | 612 | req->r_timeout_stamp = jiffies+osdc->client->mount_args->osd_timeout*HZ; |
f24e9980 SW |
613 | |
614 | ceph_msg_get(req->r_request); /* send consumes a ref */ | |
615 | ceph_con_send(&req->r_osd->o_con, req->r_request); | |
616 | req->r_sent = req->r_osd->o_incarnation; | |
617 | return 0; | |
618 | } | |
619 | ||
620 | /* | |
621 | * Timeout callback, called every N seconds when 1 or more osd | |
622 | * requests has been active for more than N seconds. When this | |
623 | * happens, we ping all OSDs with requests who have timed out to | |
624 | * ensure any communications channel reset is detected. Reset the | |
625 | * request timeouts another N seconds in the future as we go. | |
626 | * Reschedule the timeout event another N seconds in future (unless | |
627 | * there are no open requests). | |
628 | */ | |
629 | static void handle_timeout(struct work_struct *work) | |
630 | { | |
631 | struct ceph_osd_client *osdc = | |
632 | container_of(work, struct ceph_osd_client, timeout_work.work); | |
633 | struct ceph_osd_request *req; | |
634 | struct ceph_osd *osd; | |
6b805185 | 635 | unsigned long timeout = osdc->client->mount_args->osd_timeout * HZ; |
f24e9980 SW |
636 | unsigned long next_timeout = timeout + jiffies; |
637 | struct rb_node *p; | |
638 | ||
639 | dout("timeout\n"); | |
640 | down_read(&osdc->map_sem); | |
641 | ||
642 | ceph_monc_request_next_osdmap(&osdc->client->monc); | |
643 | ||
644 | mutex_lock(&osdc->request_mutex); | |
645 | for (p = rb_first(&osdc->requests); p; p = rb_next(p)) { | |
646 | req = rb_entry(p, struct ceph_osd_request, r_node); | |
647 | ||
648 | if (req->r_resend) { | |
649 | int err; | |
650 | ||
651 | dout("osdc resending prev failed %lld\n", req->r_tid); | |
652 | err = __send_request(osdc, req); | |
653 | if (err) | |
654 | dout("osdc failed again on %lld\n", req->r_tid); | |
655 | else | |
656 | req->r_resend = false; | |
657 | continue; | |
658 | } | |
659 | } | |
660 | for (p = rb_first(&osdc->osds); p; p = rb_next(p)) { | |
661 | osd = rb_entry(p, struct ceph_osd, o_node); | |
662 | if (list_empty(&osd->o_requests)) | |
663 | continue; | |
664 | req = list_first_entry(&osd->o_requests, | |
665 | struct ceph_osd_request, r_osd_item); | |
666 | if (time_before(jiffies, req->r_timeout_stamp)) | |
667 | continue; | |
668 | ||
669 | dout(" tid %llu (at least) timed out on osd%d\n", | |
670 | req->r_tid, osd->o_osd); | |
671 | req->r_timeout_stamp = next_timeout; | |
672 | ceph_con_keepalive(&osd->o_con); | |
673 | } | |
674 | ||
675 | if (osdc->timeout_tid) | |
676 | schedule_delayed_work(&osdc->timeout_work, | |
677 | round_jiffies_relative(timeout)); | |
678 | ||
679 | mutex_unlock(&osdc->request_mutex); | |
680 | ||
681 | up_read(&osdc->map_sem); | |
682 | } | |
683 | ||
684 | /* | |
685 | * handle osd op reply. either call the callback if it is specified, | |
686 | * or do the completion to wake up the waiting thread. | |
687 | */ | |
688 | static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg) | |
689 | { | |
690 | struct ceph_osd_reply_head *rhead = msg->front.iov_base; | |
691 | struct ceph_osd_request *req; | |
692 | u64 tid; | |
693 | int numops, object_len, flags; | |
694 | ||
695 | if (msg->front.iov_len < sizeof(*rhead)) | |
696 | goto bad; | |
697 | tid = le64_to_cpu(rhead->tid); | |
698 | numops = le32_to_cpu(rhead->num_ops); | |
699 | object_len = le32_to_cpu(rhead->object_len); | |
700 | if (msg->front.iov_len != sizeof(*rhead) + object_len + | |
701 | numops * sizeof(struct ceph_osd_op)) | |
702 | goto bad; | |
703 | dout("handle_reply %p tid %llu\n", msg, tid); | |
704 | ||
705 | /* lookup */ | |
706 | mutex_lock(&osdc->request_mutex); | |
707 | req = __lookup_request(osdc, tid); | |
708 | if (req == NULL) { | |
709 | dout("handle_reply tid %llu dne\n", tid); | |
710 | mutex_unlock(&osdc->request_mutex); | |
711 | return; | |
712 | } | |
713 | ceph_osdc_get_request(req); | |
714 | flags = le32_to_cpu(rhead->flags); | |
715 | ||
716 | if (req->r_reply) { | |
717 | /* | |
718 | * once we see the message has been received, we don't | |
719 | * need a ref (which is only needed for revoking | |
720 | * pages) | |
721 | */ | |
722 | ceph_msg_put(req->r_reply); | |
723 | req->r_reply = NULL; | |
724 | } | |
725 | ||
726 | if (!req->r_got_reply) { | |
727 | unsigned bytes; | |
728 | ||
729 | req->r_result = le32_to_cpu(rhead->result); | |
730 | bytes = le32_to_cpu(msg->hdr.data_len); | |
731 | dout("handle_reply result %d bytes %d\n", req->r_result, | |
732 | bytes); | |
733 | if (req->r_result == 0) | |
734 | req->r_result = bytes; | |
735 | ||
736 | /* in case this is a write and we need to replay, */ | |
737 | req->r_reassert_version = rhead->reassert_version; | |
738 | ||
739 | req->r_got_reply = 1; | |
740 | } else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) { | |
741 | dout("handle_reply tid %llu dup ack\n", tid); | |
742 | goto done; | |
743 | } | |
744 | ||
745 | dout("handle_reply tid %llu flags %d\n", tid, flags); | |
746 | ||
747 | /* either this is a read, or we got the safe response */ | |
748 | if ((flags & CEPH_OSD_FLAG_ONDISK) || | |
749 | ((flags & CEPH_OSD_FLAG_WRITE) == 0)) | |
750 | __unregister_request(osdc, req); | |
751 | ||
752 | mutex_unlock(&osdc->request_mutex); | |
753 | ||
754 | if (req->r_callback) | |
755 | req->r_callback(req, msg); | |
756 | else | |
757 | complete(&req->r_completion); | |
758 | ||
759 | if (flags & CEPH_OSD_FLAG_ONDISK) { | |
760 | if (req->r_safe_callback) | |
761 | req->r_safe_callback(req, msg); | |
762 | complete(&req->r_safe_completion); /* fsync waiter */ | |
763 | } | |
764 | ||
765 | done: | |
766 | ceph_osdc_put_request(req); | |
767 | return; | |
768 | ||
769 | bad: | |
770 | pr_err("corrupt osd_op_reply got %d %d expected %d\n", | |
771 | (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len), | |
772 | (int)sizeof(*rhead)); | |
773 | } | |
774 | ||
775 | ||
776 | /* | |
777 | * Resubmit osd requests whose osd or osd address has changed. Request | |
778 | * a new osd map if osds are down, or we are otherwise unable to determine | |
779 | * how to direct a request. | |
780 | * | |
781 | * Close connections to down osds. | |
782 | * | |
783 | * If @who is specified, resubmit requests for that specific osd. | |
784 | * | |
785 | * Caller should hold map_sem for read and request_mutex. | |
786 | */ | |
787 | static void kick_requests(struct ceph_osd_client *osdc, | |
788 | struct ceph_osd *kickosd) | |
789 | { | |
790 | struct ceph_osd_request *req; | |
791 | struct rb_node *p, *n; | |
792 | int needmap = 0; | |
793 | int err; | |
794 | ||
795 | dout("kick_requests osd%d\n", kickosd ? kickosd->o_osd : -1); | |
796 | mutex_lock(&osdc->request_mutex); | |
797 | if (!kickosd) { | |
798 | for (p = rb_first(&osdc->osds); p; p = n) { | |
799 | struct ceph_osd *osd = | |
800 | rb_entry(p, struct ceph_osd, o_node); | |
801 | ||
802 | n = rb_next(p); | |
803 | if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) || | |
804 | !ceph_entity_addr_equal(&osd->o_con.peer_addr, | |
805 | ceph_osd_addr(osdc->osdmap, | |
806 | osd->o_osd))) | |
807 | reset_osd(osdc, osd); | |
808 | } | |
809 | } | |
810 | ||
811 | for (p = rb_first(&osdc->requests); p; p = rb_next(p)) { | |
812 | req = rb_entry(p, struct ceph_osd_request, r_node); | |
813 | ||
814 | if (req->r_resend) { | |
815 | dout(" r_resend set on tid %llu\n", req->r_tid); | |
266673db | 816 | __cancel_request(req); |
f24e9980 SW |
817 | goto kick; |
818 | } | |
266673db SW |
819 | if (req->r_osd && kickosd == req->r_osd) { |
820 | __cancel_request(req); | |
f24e9980 | 821 | goto kick; |
266673db | 822 | } |
f24e9980 SW |
823 | |
824 | err = __map_osds(osdc, req); | |
825 | if (err == 0) | |
826 | continue; /* no change */ | |
827 | if (err < 0) { | |
828 | /* | |
829 | * FIXME: really, we should set the request | |
830 | * error and fail if this isn't a 'nofail' | |
831 | * request, but that's a fair bit more | |
832 | * complicated to do. So retry! | |
833 | */ | |
834 | dout(" setting r_resend on %llu\n", req->r_tid); | |
835 | req->r_resend = true; | |
836 | continue; | |
837 | } | |
838 | if (req->r_osd == NULL) { | |
839 | dout("tid %llu maps to no valid osd\n", req->r_tid); | |
840 | needmap++; /* request a newer map */ | |
841 | continue; | |
842 | } | |
843 | ||
844 | kick: | |
c1ea8823 SW |
845 | dout("kicking %p tid %llu osd%d\n", req, req->r_tid, |
846 | req->r_osd->o_osd); | |
f24e9980 SW |
847 | req->r_flags |= CEPH_OSD_FLAG_RETRY; |
848 | err = __send_request(osdc, req); | |
849 | if (err) { | |
850 | dout(" setting r_resend on %llu\n", req->r_tid); | |
851 | req->r_resend = true; | |
852 | } | |
853 | } | |
854 | mutex_unlock(&osdc->request_mutex); | |
855 | ||
856 | if (needmap) { | |
857 | dout("%d requests for down osds, need new map\n", needmap); | |
858 | ceph_monc_request_next_osdmap(&osdc->client->monc); | |
859 | } | |
860 | } | |
861 | ||
862 | /* | |
863 | * Process updated osd map. | |
864 | * | |
865 | * The message contains any number of incremental and full maps, normally | |
866 | * indicating some sort of topology change in the cluster. Kick requests | |
867 | * off to different OSDs as needed. | |
868 | */ | |
869 | void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg) | |
870 | { | |
871 | void *p, *end, *next; | |
872 | u32 nr_maps, maplen; | |
873 | u32 epoch; | |
874 | struct ceph_osdmap *newmap = NULL, *oldmap; | |
875 | int err; | |
876 | struct ceph_fsid fsid; | |
877 | ||
878 | dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0); | |
879 | p = msg->front.iov_base; | |
880 | end = p + msg->front.iov_len; | |
881 | ||
882 | /* verify fsid */ | |
883 | ceph_decode_need(&p, end, sizeof(fsid), bad); | |
884 | ceph_decode_copy(&p, &fsid, sizeof(fsid)); | |
0743304d SW |
885 | if (ceph_check_fsid(osdc->client, &fsid) < 0) |
886 | return; | |
f24e9980 SW |
887 | |
888 | down_write(&osdc->map_sem); | |
889 | ||
890 | /* incremental maps */ | |
891 | ceph_decode_32_safe(&p, end, nr_maps, bad); | |
892 | dout(" %d inc maps\n", nr_maps); | |
893 | while (nr_maps > 0) { | |
894 | ceph_decode_need(&p, end, 2*sizeof(u32), bad); | |
c89136ea SW |
895 | epoch = ceph_decode_32(&p); |
896 | maplen = ceph_decode_32(&p); | |
f24e9980 SW |
897 | ceph_decode_need(&p, end, maplen, bad); |
898 | next = p + maplen; | |
899 | if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) { | |
900 | dout("applying incremental map %u len %d\n", | |
901 | epoch, maplen); | |
902 | newmap = osdmap_apply_incremental(&p, next, | |
903 | osdc->osdmap, | |
904 | osdc->client->msgr); | |
905 | if (IS_ERR(newmap)) { | |
906 | err = PTR_ERR(newmap); | |
907 | goto bad; | |
908 | } | |
909 | if (newmap != osdc->osdmap) { | |
910 | ceph_osdmap_destroy(osdc->osdmap); | |
911 | osdc->osdmap = newmap; | |
912 | } | |
913 | } else { | |
914 | dout("ignoring incremental map %u len %d\n", | |
915 | epoch, maplen); | |
916 | } | |
917 | p = next; | |
918 | nr_maps--; | |
919 | } | |
920 | if (newmap) | |
921 | goto done; | |
922 | ||
923 | /* full maps */ | |
924 | ceph_decode_32_safe(&p, end, nr_maps, bad); | |
925 | dout(" %d full maps\n", nr_maps); | |
926 | while (nr_maps) { | |
927 | ceph_decode_need(&p, end, 2*sizeof(u32), bad); | |
c89136ea SW |
928 | epoch = ceph_decode_32(&p); |
929 | maplen = ceph_decode_32(&p); | |
f24e9980 SW |
930 | ceph_decode_need(&p, end, maplen, bad); |
931 | if (nr_maps > 1) { | |
932 | dout("skipping non-latest full map %u len %d\n", | |
933 | epoch, maplen); | |
934 | } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) { | |
935 | dout("skipping full map %u len %d, " | |
936 | "older than our %u\n", epoch, maplen, | |
937 | osdc->osdmap->epoch); | |
938 | } else { | |
939 | dout("taking full map %u len %d\n", epoch, maplen); | |
940 | newmap = osdmap_decode(&p, p+maplen); | |
941 | if (IS_ERR(newmap)) { | |
942 | err = PTR_ERR(newmap); | |
943 | goto bad; | |
944 | } | |
945 | oldmap = osdc->osdmap; | |
946 | osdc->osdmap = newmap; | |
947 | if (oldmap) | |
948 | ceph_osdmap_destroy(oldmap); | |
949 | } | |
950 | p += maplen; | |
951 | nr_maps--; | |
952 | } | |
953 | ||
954 | done: | |
955 | downgrade_write(&osdc->map_sem); | |
956 | ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch); | |
957 | if (newmap) | |
958 | kick_requests(osdc, NULL); | |
959 | up_read(&osdc->map_sem); | |
960 | return; | |
961 | ||
962 | bad: | |
963 | pr_err("osdc handle_map corrupt msg\n"); | |
964 | up_write(&osdc->map_sem); | |
965 | return; | |
966 | } | |
967 | ||
968 | ||
969 | /* | |
970 | * A read request prepares specific pages that data is to be read into. | |
971 | * When a message is being read off the wire, we call prepare_pages to | |
972 | * find those pages. | |
973 | * 0 = success, -1 failure. | |
974 | */ | |
975 | static int prepare_pages(struct ceph_connection *con, struct ceph_msg *m, | |
976 | int want) | |
977 | { | |
978 | struct ceph_osd *osd = con->private; | |
979 | struct ceph_osd_client *osdc; | |
980 | struct ceph_osd_reply_head *rhead = m->front.iov_base; | |
981 | struct ceph_osd_request *req; | |
982 | u64 tid; | |
983 | int ret = -1; | |
984 | int type = le16_to_cpu(m->hdr.type); | |
985 | ||
986 | if (!osd) | |
987 | return -1; | |
988 | osdc = osd->o_osdc; | |
989 | ||
990 | dout("prepare_pages on msg %p want %d\n", m, want); | |
991 | if (unlikely(type != CEPH_MSG_OSD_OPREPLY)) | |
992 | return -1; /* hmm! */ | |
993 | ||
994 | tid = le64_to_cpu(rhead->tid); | |
995 | mutex_lock(&osdc->request_mutex); | |
996 | req = __lookup_request(osdc, tid); | |
997 | if (!req) { | |
998 | dout("prepare_pages unknown tid %llu\n", tid); | |
999 | goto out; | |
1000 | } | |
1001 | dout("prepare_pages tid %llu has %d pages, want %d\n", | |
1002 | tid, req->r_num_pages, want); | |
1003 | if (likely(req->r_num_pages >= want && !req->r_prepared_pages)) { | |
1004 | m->pages = req->r_pages; | |
1005 | m->nr_pages = req->r_num_pages; | |
1006 | req->r_reply = m; /* only for duration of read over socket */ | |
1007 | ceph_msg_get(m); | |
1008 | req->r_prepared_pages = 1; | |
1009 | ret = 0; /* success */ | |
1010 | } | |
1011 | out: | |
1012 | mutex_unlock(&osdc->request_mutex); | |
1013 | return ret; | |
1014 | } | |
1015 | ||
1016 | /* | |
1017 | * Register request, send initial attempt. | |
1018 | */ | |
1019 | int ceph_osdc_start_request(struct ceph_osd_client *osdc, | |
1020 | struct ceph_osd_request *req, | |
1021 | bool nofail) | |
1022 | { | |
c1ea8823 | 1023 | int rc = 0; |
f24e9980 SW |
1024 | |
1025 | req->r_request->pages = req->r_pages; | |
1026 | req->r_request->nr_pages = req->r_num_pages; | |
1027 | ||
1028 | register_request(osdc, req); | |
1029 | ||
1030 | down_read(&osdc->map_sem); | |
1031 | mutex_lock(&osdc->request_mutex); | |
c1ea8823 SW |
1032 | /* |
1033 | * a racing kick_requests() may have sent the message for us | |
1034 | * while we dropped request_mutex above, so only send now if | |
1035 | * the request still han't been touched yet. | |
1036 | */ | |
1037 | if (req->r_sent == 0) { | |
1038 | rc = __send_request(osdc, req); | |
1039 | if (rc) { | |
1040 | if (nofail) { | |
1041 | dout("osdc_start_request failed send, " | |
1042 | " marking %lld\n", req->r_tid); | |
1043 | req->r_resend = true; | |
1044 | rc = 0; | |
1045 | } else { | |
1046 | __unregister_request(osdc, req); | |
1047 | } | |
f24e9980 SW |
1048 | } |
1049 | } | |
1050 | mutex_unlock(&osdc->request_mutex); | |
1051 | up_read(&osdc->map_sem); | |
1052 | return rc; | |
1053 | } | |
1054 | ||
1055 | /* | |
1056 | * wait for a request to complete | |
1057 | */ | |
1058 | int ceph_osdc_wait_request(struct ceph_osd_client *osdc, | |
1059 | struct ceph_osd_request *req) | |
1060 | { | |
1061 | int rc; | |
1062 | ||
1063 | rc = wait_for_completion_interruptible(&req->r_completion); | |
1064 | if (rc < 0) { | |
1065 | mutex_lock(&osdc->request_mutex); | |
1066 | __cancel_request(req); | |
1067 | mutex_unlock(&osdc->request_mutex); | |
1068 | dout("wait_request tid %llu timed out\n", req->r_tid); | |
1069 | return rc; | |
1070 | } | |
1071 | ||
1072 | dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result); | |
1073 | return req->r_result; | |
1074 | } | |
1075 | ||
1076 | /* | |
1077 | * sync - wait for all in-flight requests to flush. avoid starvation. | |
1078 | */ | |
1079 | void ceph_osdc_sync(struct ceph_osd_client *osdc) | |
1080 | { | |
1081 | struct ceph_osd_request *req; | |
1082 | u64 last_tid, next_tid = 0; | |
1083 | ||
1084 | mutex_lock(&osdc->request_mutex); | |
1085 | last_tid = osdc->last_tid; | |
1086 | while (1) { | |
1087 | req = __lookup_request_ge(osdc, next_tid); | |
1088 | if (!req) | |
1089 | break; | |
1090 | if (req->r_tid > last_tid) | |
1091 | break; | |
1092 | ||
1093 | next_tid = req->r_tid + 1; | |
1094 | if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0) | |
1095 | continue; | |
1096 | ||
1097 | ceph_osdc_get_request(req); | |
1098 | mutex_unlock(&osdc->request_mutex); | |
1099 | dout("sync waiting on tid %llu (last is %llu)\n", | |
1100 | req->r_tid, last_tid); | |
1101 | wait_for_completion(&req->r_safe_completion); | |
1102 | mutex_lock(&osdc->request_mutex); | |
1103 | ceph_osdc_put_request(req); | |
1104 | } | |
1105 | mutex_unlock(&osdc->request_mutex); | |
1106 | dout("sync done (thru tid %llu)\n", last_tid); | |
1107 | } | |
1108 | ||
1109 | /* | |
1110 | * init, shutdown | |
1111 | */ | |
1112 | int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client) | |
1113 | { | |
1114 | int err; | |
1115 | ||
1116 | dout("init\n"); | |
1117 | osdc->client = client; | |
1118 | osdc->osdmap = NULL; | |
1119 | init_rwsem(&osdc->map_sem); | |
1120 | init_completion(&osdc->map_waiters); | |
1121 | osdc->last_requested_map = 0; | |
1122 | mutex_init(&osdc->request_mutex); | |
1123 | osdc->timeout_tid = 0; | |
1124 | osdc->last_tid = 0; | |
1125 | osdc->osds = RB_ROOT; | |
1126 | osdc->requests = RB_ROOT; | |
1127 | osdc->num_requests = 0; | |
1128 | INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout); | |
1129 | ||
5f44f142 | 1130 | err = -ENOMEM; |
f24e9980 SW |
1131 | osdc->req_mempool = mempool_create_kmalloc_pool(10, |
1132 | sizeof(struct ceph_osd_request)); | |
1133 | if (!osdc->req_mempool) | |
5f44f142 | 1134 | goto out; |
f24e9980 SW |
1135 | |
1136 | err = ceph_msgpool_init(&osdc->msgpool_op, 4096, 10, true); | |
1137 | if (err < 0) | |
5f44f142 | 1138 | goto out_mempool; |
f24e9980 SW |
1139 | err = ceph_msgpool_init(&osdc->msgpool_op_reply, 512, 0, false); |
1140 | if (err < 0) | |
5f44f142 | 1141 | goto out_msgpool; |
f24e9980 | 1142 | return 0; |
5f44f142 SW |
1143 | |
1144 | out_msgpool: | |
1145 | ceph_msgpool_destroy(&osdc->msgpool_op); | |
1146 | out_mempool: | |
1147 | mempool_destroy(osdc->req_mempool); | |
1148 | out: | |
1149 | return err; | |
f24e9980 SW |
1150 | } |
1151 | ||
1152 | void ceph_osdc_stop(struct ceph_osd_client *osdc) | |
1153 | { | |
1154 | cancel_delayed_work_sync(&osdc->timeout_work); | |
1155 | if (osdc->osdmap) { | |
1156 | ceph_osdmap_destroy(osdc->osdmap); | |
1157 | osdc->osdmap = NULL; | |
1158 | } | |
1159 | mempool_destroy(osdc->req_mempool); | |
1160 | ceph_msgpool_destroy(&osdc->msgpool_op); | |
1161 | ceph_msgpool_destroy(&osdc->msgpool_op_reply); | |
1162 | } | |
1163 | ||
1164 | /* | |
1165 | * Read some contiguous pages. If we cross a stripe boundary, shorten | |
1166 | * *plen. Return number of bytes read, or error. | |
1167 | */ | |
1168 | int ceph_osdc_readpages(struct ceph_osd_client *osdc, | |
1169 | struct ceph_vino vino, struct ceph_file_layout *layout, | |
1170 | u64 off, u64 *plen, | |
1171 | u32 truncate_seq, u64 truncate_size, | |
1172 | struct page **pages, int num_pages) | |
1173 | { | |
1174 | struct ceph_osd_request *req; | |
1175 | int rc = 0; | |
1176 | ||
1177 | dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino, | |
1178 | vino.snap, off, *plen); | |
1179 | req = ceph_osdc_new_request(osdc, layout, vino, off, plen, | |
1180 | CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ, | |
1181 | NULL, 0, truncate_seq, truncate_size, NULL, | |
1182 | false, 1); | |
1183 | if (IS_ERR(req)) | |
1184 | return PTR_ERR(req); | |
1185 | ||
1186 | /* it may be a short read due to an object boundary */ | |
1187 | req->r_pages = pages; | |
1188 | num_pages = calc_pages_for(off, *plen); | |
1189 | req->r_num_pages = num_pages; | |
1190 | ||
1191 | dout("readpages final extent is %llu~%llu (%d pages)\n", | |
1192 | off, *plen, req->r_num_pages); | |
1193 | ||
1194 | rc = ceph_osdc_start_request(osdc, req, false); | |
1195 | if (!rc) | |
1196 | rc = ceph_osdc_wait_request(osdc, req); | |
1197 | ||
1198 | ceph_osdc_put_request(req); | |
1199 | dout("readpages result %d\n", rc); | |
1200 | return rc; | |
1201 | } | |
1202 | ||
1203 | /* | |
1204 | * do a synchronous write on N pages | |
1205 | */ | |
1206 | int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino, | |
1207 | struct ceph_file_layout *layout, | |
1208 | struct ceph_snap_context *snapc, | |
1209 | u64 off, u64 len, | |
1210 | u32 truncate_seq, u64 truncate_size, | |
1211 | struct timespec *mtime, | |
1212 | struct page **pages, int num_pages, | |
1213 | int flags, int do_sync, bool nofail) | |
1214 | { | |
1215 | struct ceph_osd_request *req; | |
1216 | int rc = 0; | |
1217 | ||
1218 | BUG_ON(vino.snap != CEPH_NOSNAP); | |
1219 | req = ceph_osdc_new_request(osdc, layout, vino, off, &len, | |
1220 | CEPH_OSD_OP_WRITE, | |
1221 | flags | CEPH_OSD_FLAG_ONDISK | | |
1222 | CEPH_OSD_FLAG_WRITE, | |
1223 | snapc, do_sync, | |
1224 | truncate_seq, truncate_size, mtime, | |
1225 | nofail, 1); | |
1226 | if (IS_ERR(req)) | |
1227 | return PTR_ERR(req); | |
1228 | ||
1229 | /* it may be a short write due to an object boundary */ | |
1230 | req->r_pages = pages; | |
1231 | req->r_num_pages = calc_pages_for(off, len); | |
1232 | dout("writepages %llu~%llu (%d pages)\n", off, len, | |
1233 | req->r_num_pages); | |
1234 | ||
1235 | rc = ceph_osdc_start_request(osdc, req, nofail); | |
1236 | if (!rc) | |
1237 | rc = ceph_osdc_wait_request(osdc, req); | |
1238 | ||
1239 | ceph_osdc_put_request(req); | |
1240 | if (rc == 0) | |
1241 | rc = len; | |
1242 | dout("writepages result %d\n", rc); | |
1243 | return rc; | |
1244 | } | |
1245 | ||
1246 | /* | |
1247 | * handle incoming message | |
1248 | */ | |
1249 | static void dispatch(struct ceph_connection *con, struct ceph_msg *msg) | |
1250 | { | |
1251 | struct ceph_osd *osd = con->private; | |
1252 | struct ceph_osd_client *osdc = osd->o_osdc; | |
1253 | int type = le16_to_cpu(msg->hdr.type); | |
1254 | ||
1255 | if (!osd) | |
1256 | return; | |
1257 | ||
1258 | switch (type) { | |
1259 | case CEPH_MSG_OSD_MAP: | |
1260 | ceph_osdc_handle_map(osdc, msg); | |
1261 | break; | |
1262 | case CEPH_MSG_OSD_OPREPLY: | |
1263 | handle_reply(osdc, msg); | |
1264 | break; | |
1265 | ||
1266 | default: | |
1267 | pr_err("received unknown message type %d %s\n", type, | |
1268 | ceph_msg_type_name(type)); | |
1269 | } | |
1270 | ceph_msg_put(msg); | |
1271 | } | |
1272 | ||
1273 | static struct ceph_msg *alloc_msg(struct ceph_connection *con, | |
1274 | struct ceph_msg_header *hdr) | |
1275 | { | |
1276 | struct ceph_osd *osd = con->private; | |
1277 | struct ceph_osd_client *osdc = osd->o_osdc; | |
1278 | int type = le16_to_cpu(hdr->type); | |
8f3bc053 | 1279 | int front = le32_to_cpu(hdr->front_len); |
f24e9980 SW |
1280 | |
1281 | switch (type) { | |
1282 | case CEPH_MSG_OSD_OPREPLY: | |
8f3bc053 | 1283 | return ceph_msgpool_get(&osdc->msgpool_op_reply, front); |
f24e9980 SW |
1284 | } |
1285 | return ceph_alloc_msg(con, hdr); | |
1286 | } | |
1287 | ||
1288 | /* | |
1289 | * Wrappers to refcount containing ceph_osd struct | |
1290 | */ | |
1291 | static struct ceph_connection *get_osd_con(struct ceph_connection *con) | |
1292 | { | |
1293 | struct ceph_osd *osd = con->private; | |
1294 | if (get_osd(osd)) | |
1295 | return con; | |
1296 | return NULL; | |
1297 | } | |
1298 | ||
1299 | static void put_osd_con(struct ceph_connection *con) | |
1300 | { | |
1301 | struct ceph_osd *osd = con->private; | |
1302 | put_osd(osd); | |
1303 | } | |
1304 | ||
4e7a5dcd SW |
1305 | /* |
1306 | * authentication | |
1307 | */ | |
1308 | static int get_authorizer(struct ceph_connection *con, | |
1309 | void **buf, int *len, int *proto, | |
1310 | void **reply_buf, int *reply_len, int force_new) | |
1311 | { | |
1312 | struct ceph_osd *o = con->private; | |
1313 | struct ceph_osd_client *osdc = o->o_osdc; | |
1314 | struct ceph_auth_client *ac = osdc->client->monc.auth; | |
1315 | int ret = 0; | |
1316 | ||
1317 | if (force_new && o->o_authorizer) { | |
1318 | ac->ops->destroy_authorizer(ac, o->o_authorizer); | |
1319 | o->o_authorizer = NULL; | |
1320 | } | |
1321 | if (o->o_authorizer == NULL) { | |
1322 | ret = ac->ops->create_authorizer( | |
1323 | ac, CEPH_ENTITY_TYPE_OSD, | |
1324 | &o->o_authorizer, | |
1325 | &o->o_authorizer_buf, | |
1326 | &o->o_authorizer_buf_len, | |
1327 | &o->o_authorizer_reply_buf, | |
1328 | &o->o_authorizer_reply_buf_len); | |
1329 | if (ret) | |
1330 | return ret; | |
1331 | } | |
1332 | ||
1333 | *proto = ac->protocol; | |
1334 | *buf = o->o_authorizer_buf; | |
1335 | *len = o->o_authorizer_buf_len; | |
1336 | *reply_buf = o->o_authorizer_reply_buf; | |
1337 | *reply_len = o->o_authorizer_reply_buf_len; | |
1338 | return 0; | |
1339 | } | |
1340 | ||
1341 | ||
1342 | static int verify_authorizer_reply(struct ceph_connection *con, int len) | |
1343 | { | |
1344 | struct ceph_osd *o = con->private; | |
1345 | struct ceph_osd_client *osdc = o->o_osdc; | |
1346 | struct ceph_auth_client *ac = osdc->client->monc.auth; | |
1347 | ||
1348 | return ac->ops->verify_authorizer_reply(ac, o->o_authorizer, len); | |
1349 | } | |
1350 | ||
1351 | ||
f24e9980 SW |
1352 | const static struct ceph_connection_operations osd_con_ops = { |
1353 | .get = get_osd_con, | |
1354 | .put = put_osd_con, | |
1355 | .dispatch = dispatch, | |
4e7a5dcd SW |
1356 | .get_authorizer = get_authorizer, |
1357 | .verify_authorizer_reply = verify_authorizer_reply, | |
f24e9980 | 1358 | .alloc_msg = alloc_msg, |
81b024e7 | 1359 | .fault = osd_reset, |
f24e9980 SW |
1360 | .alloc_middle = ceph_alloc_middle, |
1361 | .prepare_pages = prepare_pages, | |
1362 | }; |