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 | 15 | |
0d59ab81 YS |
16 | #define OSD_REPLY_RESERVE_FRONT_LEN 512 |
17 | ||
f24e9980 SW |
18 | const static struct ceph_connection_operations osd_con_ops; |
19 | ||
20 | static void kick_requests(struct ceph_osd_client *osdc, struct ceph_osd *osd); | |
21 | ||
22 | /* | |
23 | * Implement client access to distributed object storage cluster. | |
24 | * | |
25 | * All data objects are stored within a cluster/cloud of OSDs, or | |
26 | * "object storage devices." (Note that Ceph OSDs have _nothing_ to | |
27 | * do with the T10 OSD extensions to SCSI.) Ceph OSDs are simply | |
28 | * remote daemons serving up and coordinating consistent and safe | |
29 | * access to storage. | |
30 | * | |
31 | * Cluster membership and the mapping of data objects onto storage devices | |
32 | * are described by the osd map. | |
33 | * | |
34 | * We keep track of pending OSD requests (read, write), resubmit | |
35 | * requests to different OSDs when the cluster topology/data layout | |
36 | * change, or retry the affected requests when the communications | |
37 | * channel with an OSD is reset. | |
38 | */ | |
39 | ||
40 | /* | |
41 | * calculate the mapping of a file extent onto an object, and fill out the | |
42 | * request accordingly. shorten extent as necessary if it crosses an | |
43 | * object boundary. | |
44 | * | |
45 | * fill osd op in request message. | |
46 | */ | |
47 | static void calc_layout(struct ceph_osd_client *osdc, | |
48 | struct ceph_vino vino, struct ceph_file_layout *layout, | |
49 | u64 off, u64 *plen, | |
50 | struct ceph_osd_request *req) | |
51 | { | |
52 | struct ceph_osd_request_head *reqhead = req->r_request->front.iov_base; | |
53 | struct ceph_osd_op *op = (void *)(reqhead + 1); | |
54 | u64 orig_len = *plen; | |
55 | u64 objoff, objlen; /* extent in object */ | |
56 | u64 bno; | |
57 | ||
58 | reqhead->snapid = cpu_to_le64(vino.snap); | |
59 | ||
60 | /* object extent? */ | |
61 | ceph_calc_file_object_mapping(layout, off, plen, &bno, | |
62 | &objoff, &objlen); | |
63 | if (*plen < orig_len) | |
64 | dout(" skipping last %llu, final file extent %llu~%llu\n", | |
65 | orig_len - *plen, off, *plen); | |
66 | ||
67 | sprintf(req->r_oid, "%llx.%08llx", vino.ino, bno); | |
68 | req->r_oid_len = strlen(req->r_oid); | |
69 | ||
70 | op->extent.offset = cpu_to_le64(objoff); | |
71 | op->extent.length = cpu_to_le64(objlen); | |
72 | req->r_num_pages = calc_pages_for(off, *plen); | |
73 | ||
74 | dout("calc_layout %s (%d) %llu~%llu (%d pages)\n", | |
75 | req->r_oid, req->r_oid_len, objoff, objlen, req->r_num_pages); | |
76 | } | |
77 | ||
0d59ab81 YS |
78 | static void remove_replies(struct ceph_osd_request *req) |
79 | { | |
80 | int i; | |
81 | int max = ARRAY_SIZE(req->replies); | |
82 | ||
83 | for (i=0; i<max; i++) { | |
84 | if (req->replies[i]) | |
85 | ceph_msg_put(req->replies[i]); | |
86 | } | |
87 | } | |
f24e9980 SW |
88 | |
89 | /* | |
90 | * requests | |
91 | */ | |
415e49a9 | 92 | void ceph_osdc_release_request(struct kref *kref) |
f24e9980 | 93 | { |
415e49a9 SW |
94 | struct ceph_osd_request *req = container_of(kref, |
95 | struct ceph_osd_request, | |
96 | r_kref); | |
97 | ||
98 | if (req->r_request) | |
99 | ceph_msg_put(req->r_request); | |
100 | if (req->r_reply) | |
101 | ceph_msg_put(req->r_reply); | |
0d59ab81 YS |
102 | remove_replies(req); |
103 | if (req->r_con_filling_msg) { | |
350b1c32 | 104 | dout("release_request revoking pages %p from con %p\n", |
0d59ab81 YS |
105 | req->r_pages, req->r_con_filling_msg); |
106 | ceph_con_revoke_message(req->r_con_filling_msg, | |
107 | req->r_reply); | |
108 | ceph_con_put(req->r_con_filling_msg); | |
350b1c32 | 109 | } |
415e49a9 SW |
110 | if (req->r_own_pages) |
111 | ceph_release_page_vector(req->r_pages, | |
112 | req->r_num_pages); | |
113 | ceph_put_snap_context(req->r_snapc); | |
114 | if (req->r_mempool) | |
115 | mempool_free(req, req->r_osdc->req_mempool); | |
116 | else | |
117 | kfree(req); | |
f24e9980 SW |
118 | } |
119 | ||
0d59ab81 YS |
120 | static int alloc_replies(struct ceph_osd_request *req, int num_reply) |
121 | { | |
122 | int i; | |
123 | int max = ARRAY_SIZE(req->replies); | |
124 | ||
125 | BUG_ON(num_reply > max); | |
126 | ||
127 | for (i=0; i<num_reply; i++) { | |
128 | req->replies[i] = ceph_msg_new(0, OSD_REPLY_RESERVE_FRONT_LEN, 0, 0, NULL); | |
129 | if (IS_ERR(req->replies[i])) { | |
130 | int j; | |
131 | int err = PTR_ERR(req->replies[i]); | |
132 | for (j = 0; j<=i; j++) { | |
133 | ceph_msg_put(req->replies[j]); | |
134 | } | |
135 | return err; | |
136 | } | |
137 | } | |
138 | ||
139 | for (; i<max; i++) { | |
140 | req->replies[i] = NULL; | |
141 | } | |
142 | ||
143 | req->cur_reply = 0; | |
144 | ||
145 | return 0; | |
146 | } | |
147 | ||
148 | static struct ceph_msg *__get_next_reply(struct ceph_connection *con, | |
149 | struct ceph_osd_request *req, | |
150 | int front_len) | |
151 | { | |
152 | struct ceph_msg *reply; | |
153 | if (req->r_con_filling_msg) { | |
154 | dout("revoking reply msg %p from old con %p\n", req->r_reply, | |
155 | req->r_con_filling_msg); | |
156 | ceph_con_revoke_message(req->r_con_filling_msg, req->r_reply); | |
157 | ceph_con_put(req->r_con_filling_msg); | |
158 | req->cur_reply = 0; | |
159 | } | |
160 | reply = req->replies[req->cur_reply]; | |
161 | if (!reply || front_len > OSD_REPLY_RESERVE_FRONT_LEN) { | |
162 | /* maybe we can allocate it now? */ | |
163 | reply = ceph_msg_new(0, front_len, 0, 0, NULL); | |
164 | if (!reply || IS_ERR(reply)) { | |
165 | pr_err(" reply alloc failed, front_len=%d\n", front_len); | |
166 | return ERR_PTR(-ENOMEM); | |
167 | } | |
168 | } | |
169 | req->r_con_filling_msg = ceph_con_get(con); | |
170 | req->r_reply = ceph_msg_get(reply); /* for duration of read over socket */ | |
171 | return ceph_msg_get(reply); | |
172 | } | |
173 | ||
f24e9980 SW |
174 | /* |
175 | * build new request AND message, calculate layout, and adjust file | |
176 | * extent as needed. | |
177 | * | |
178 | * if the file was recently truncated, we include information about its | |
179 | * old and new size so that the object can be updated appropriately. (we | |
180 | * avoid synchronously deleting truncated objects because it's slow.) | |
181 | * | |
182 | * if @do_sync, include a 'startsync' command so that the osd will flush | |
183 | * data quickly. | |
184 | */ | |
185 | struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc, | |
186 | struct ceph_file_layout *layout, | |
187 | struct ceph_vino vino, | |
188 | u64 off, u64 *plen, | |
189 | int opcode, int flags, | |
190 | struct ceph_snap_context *snapc, | |
191 | int do_sync, | |
192 | u32 truncate_seq, | |
193 | u64 truncate_size, | |
194 | struct timespec *mtime, | |
195 | bool use_mempool, int num_reply) | |
196 | { | |
197 | struct ceph_osd_request *req; | |
198 | struct ceph_msg *msg; | |
199 | struct ceph_osd_request_head *head; | |
200 | struct ceph_osd_op *op; | |
201 | void *p; | |
0c948992 | 202 | int num_op = 1 + do_sync; |
f24e9980 SW |
203 | size_t msg_size = sizeof(*head) + num_op*sizeof(*op); |
204 | int err, i; | |
f24e9980 SW |
205 | |
206 | if (use_mempool) { | |
207 | req = mempool_alloc(osdc->req_mempool, GFP_NOFS); | |
208 | memset(req, 0, sizeof(*req)); | |
209 | } else { | |
210 | req = kzalloc(sizeof(*req), GFP_NOFS); | |
211 | } | |
212 | if (req == NULL) | |
213 | return ERR_PTR(-ENOMEM); | |
214 | ||
0d59ab81 | 215 | err = alloc_replies(req, num_reply); |
f24e9980 SW |
216 | if (err) { |
217 | ceph_osdc_put_request(req); | |
218 | return ERR_PTR(-ENOMEM); | |
219 | } | |
93c20d98 | 220 | req->r_num_prealloc_reply = num_reply; |
f24e9980 SW |
221 | |
222 | req->r_osdc = osdc; | |
223 | req->r_mempool = use_mempool; | |
415e49a9 | 224 | kref_init(&req->r_kref); |
f24e9980 SW |
225 | init_completion(&req->r_completion); |
226 | init_completion(&req->r_safe_completion); | |
227 | INIT_LIST_HEAD(&req->r_unsafe_item); | |
228 | req->r_flags = flags; | |
229 | ||
230 | WARN_ON((flags & (CEPH_OSD_FLAG_READ|CEPH_OSD_FLAG_WRITE)) == 0); | |
231 | ||
232 | /* create message; allow space for oid */ | |
233 | msg_size += 40; | |
234 | if (snapc) | |
235 | msg_size += sizeof(u64) * snapc->num_snaps; | |
236 | if (use_mempool) | |
8f3bc053 | 237 | msg = ceph_msgpool_get(&osdc->msgpool_op, 0); |
f24e9980 SW |
238 | else |
239 | msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, 0, 0, NULL); | |
240 | if (IS_ERR(msg)) { | |
f24e9980 SW |
241 | ceph_osdc_put_request(req); |
242 | return ERR_PTR(PTR_ERR(msg)); | |
243 | } | |
244 | msg->hdr.type = cpu_to_le16(CEPH_MSG_OSD_OP); | |
245 | memset(msg->front.iov_base, 0, msg->front.iov_len); | |
246 | head = msg->front.iov_base; | |
247 | op = (void *)(head + 1); | |
248 | p = (void *)(op + num_op); | |
249 | ||
250 | req->r_request = msg; | |
251 | req->r_snapc = ceph_get_snap_context(snapc); | |
252 | ||
253 | head->client_inc = cpu_to_le32(1); /* always, for now. */ | |
254 | head->flags = cpu_to_le32(flags); | |
255 | if (flags & CEPH_OSD_FLAG_WRITE) | |
256 | ceph_encode_timespec(&head->mtime, mtime); | |
257 | head->num_ops = cpu_to_le16(num_op); | |
258 | op->op = cpu_to_le16(opcode); | |
259 | ||
260 | /* calculate max write size */ | |
261 | calc_layout(osdc, vino, layout, off, plen, req); | |
262 | req->r_file_layout = *layout; /* keep a copy */ | |
263 | ||
264 | if (flags & CEPH_OSD_FLAG_WRITE) { | |
265 | req->r_request->hdr.data_off = cpu_to_le16(off); | |
266 | req->r_request->hdr.data_len = cpu_to_le32(*plen); | |
267 | op->payload_len = cpu_to_le32(*plen); | |
268 | } | |
0c948992 YS |
269 | op->extent.truncate_size = cpu_to_le64(truncate_size); |
270 | op->extent.truncate_seq = cpu_to_le32(truncate_seq); | |
f24e9980 SW |
271 | |
272 | /* fill in oid */ | |
273 | head->object_len = cpu_to_le32(req->r_oid_len); | |
274 | memcpy(p, req->r_oid, req->r_oid_len); | |
275 | p += req->r_oid_len; | |
276 | ||
f24e9980 SW |
277 | if (do_sync) { |
278 | op++; | |
279 | op->op = cpu_to_le16(CEPH_OSD_OP_STARTSYNC); | |
280 | } | |
281 | if (snapc) { | |
282 | head->snap_seq = cpu_to_le64(snapc->seq); | |
283 | head->num_snaps = cpu_to_le32(snapc->num_snaps); | |
284 | for (i = 0; i < snapc->num_snaps; i++) { | |
285 | put_unaligned_le64(snapc->snaps[i], p); | |
286 | p += sizeof(u64); | |
287 | } | |
288 | } | |
289 | ||
290 | BUG_ON(p > msg->front.iov_base + msg->front.iov_len); | |
291 | return req; | |
292 | } | |
293 | ||
294 | /* | |
295 | * We keep osd requests in an rbtree, sorted by ->r_tid. | |
296 | */ | |
297 | static void __insert_request(struct ceph_osd_client *osdc, | |
298 | struct ceph_osd_request *new) | |
299 | { | |
300 | struct rb_node **p = &osdc->requests.rb_node; | |
301 | struct rb_node *parent = NULL; | |
302 | struct ceph_osd_request *req = NULL; | |
303 | ||
304 | while (*p) { | |
305 | parent = *p; | |
306 | req = rb_entry(parent, struct ceph_osd_request, r_node); | |
307 | if (new->r_tid < req->r_tid) | |
308 | p = &(*p)->rb_left; | |
309 | else if (new->r_tid > req->r_tid) | |
310 | p = &(*p)->rb_right; | |
311 | else | |
312 | BUG(); | |
313 | } | |
314 | ||
315 | rb_link_node(&new->r_node, parent, p); | |
316 | rb_insert_color(&new->r_node, &osdc->requests); | |
317 | } | |
318 | ||
319 | static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc, | |
320 | u64 tid) | |
321 | { | |
322 | struct ceph_osd_request *req; | |
323 | struct rb_node *n = osdc->requests.rb_node; | |
324 | ||
325 | while (n) { | |
326 | req = rb_entry(n, struct ceph_osd_request, r_node); | |
327 | if (tid < req->r_tid) | |
328 | n = n->rb_left; | |
329 | else if (tid > req->r_tid) | |
330 | n = n->rb_right; | |
331 | else | |
332 | return req; | |
333 | } | |
334 | return NULL; | |
335 | } | |
336 | ||
337 | static struct ceph_osd_request * | |
338 | __lookup_request_ge(struct ceph_osd_client *osdc, | |
339 | u64 tid) | |
340 | { | |
341 | struct ceph_osd_request *req; | |
342 | struct rb_node *n = osdc->requests.rb_node; | |
343 | ||
344 | while (n) { | |
345 | req = rb_entry(n, struct ceph_osd_request, r_node); | |
346 | if (tid < req->r_tid) { | |
347 | if (!n->rb_left) | |
348 | return req; | |
349 | n = n->rb_left; | |
350 | } else if (tid > req->r_tid) { | |
351 | n = n->rb_right; | |
352 | } else { | |
353 | return req; | |
354 | } | |
355 | } | |
356 | return NULL; | |
357 | } | |
358 | ||
359 | ||
360 | /* | |
81b024e7 | 361 | * If the osd connection drops, we need to resubmit all requests. |
f24e9980 SW |
362 | */ |
363 | static void osd_reset(struct ceph_connection *con) | |
364 | { | |
365 | struct ceph_osd *osd = con->private; | |
366 | struct ceph_osd_client *osdc; | |
367 | ||
368 | if (!osd) | |
369 | return; | |
370 | dout("osd_reset osd%d\n", osd->o_osd); | |
371 | osdc = osd->o_osdc; | |
372 | osd->o_incarnation++; | |
373 | down_read(&osdc->map_sem); | |
374 | kick_requests(osdc, osd); | |
375 | up_read(&osdc->map_sem); | |
376 | } | |
377 | ||
378 | /* | |
379 | * Track open sessions with osds. | |
380 | */ | |
381 | static struct ceph_osd *create_osd(struct ceph_osd_client *osdc) | |
382 | { | |
383 | struct ceph_osd *osd; | |
384 | ||
385 | osd = kzalloc(sizeof(*osd), GFP_NOFS); | |
386 | if (!osd) | |
387 | return NULL; | |
388 | ||
389 | atomic_set(&osd->o_ref, 1); | |
390 | osd->o_osdc = osdc; | |
391 | INIT_LIST_HEAD(&osd->o_requests); | |
392 | osd->o_incarnation = 1; | |
393 | ||
394 | ceph_con_init(osdc->client->msgr, &osd->o_con); | |
395 | osd->o_con.private = osd; | |
396 | osd->o_con.ops = &osd_con_ops; | |
397 | osd->o_con.peer_name.type = CEPH_ENTITY_TYPE_OSD; | |
4e7a5dcd | 398 | |
f24e9980 SW |
399 | return osd; |
400 | } | |
401 | ||
402 | static struct ceph_osd *get_osd(struct ceph_osd *osd) | |
403 | { | |
404 | if (atomic_inc_not_zero(&osd->o_ref)) { | |
405 | dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1, | |
406 | atomic_read(&osd->o_ref)); | |
407 | return osd; | |
408 | } else { | |
409 | dout("get_osd %p FAIL\n", osd); | |
410 | return NULL; | |
411 | } | |
412 | } | |
413 | ||
414 | static void put_osd(struct ceph_osd *osd) | |
415 | { | |
416 | dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref), | |
417 | atomic_read(&osd->o_ref) - 1); | |
42ce56e5 | 418 | if (atomic_dec_and_test(&osd->o_ref)) |
f24e9980 | 419 | kfree(osd); |
f24e9980 SW |
420 | } |
421 | ||
422 | /* | |
423 | * remove an osd from our map | |
424 | */ | |
425 | static void remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd) | |
426 | { | |
427 | dout("remove_osd %p\n", osd); | |
428 | BUG_ON(!list_empty(&osd->o_requests)); | |
429 | rb_erase(&osd->o_node, &osdc->osds); | |
430 | ceph_con_close(&osd->o_con); | |
431 | put_osd(osd); | |
432 | } | |
433 | ||
434 | /* | |
435 | * reset osd connect | |
436 | */ | |
437 | static int reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd) | |
438 | { | |
439 | int ret = 0; | |
440 | ||
441 | dout("reset_osd %p osd%d\n", osd, osd->o_osd); | |
442 | if (list_empty(&osd->o_requests)) { | |
443 | remove_osd(osdc, osd); | |
444 | } else { | |
445 | ceph_con_close(&osd->o_con); | |
446 | ceph_con_open(&osd->o_con, &osdc->osdmap->osd_addr[osd->o_osd]); | |
447 | osd->o_incarnation++; | |
448 | } | |
449 | return ret; | |
450 | } | |
451 | ||
452 | static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new) | |
453 | { | |
454 | struct rb_node **p = &osdc->osds.rb_node; | |
455 | struct rb_node *parent = NULL; | |
456 | struct ceph_osd *osd = NULL; | |
457 | ||
458 | while (*p) { | |
459 | parent = *p; | |
460 | osd = rb_entry(parent, struct ceph_osd, o_node); | |
461 | if (new->o_osd < osd->o_osd) | |
462 | p = &(*p)->rb_left; | |
463 | else if (new->o_osd > osd->o_osd) | |
464 | p = &(*p)->rb_right; | |
465 | else | |
466 | BUG(); | |
467 | } | |
468 | ||
469 | rb_link_node(&new->o_node, parent, p); | |
470 | rb_insert_color(&new->o_node, &osdc->osds); | |
471 | } | |
472 | ||
473 | static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o) | |
474 | { | |
475 | struct ceph_osd *osd; | |
476 | struct rb_node *n = osdc->osds.rb_node; | |
477 | ||
478 | while (n) { | |
479 | osd = rb_entry(n, struct ceph_osd, o_node); | |
480 | if (o < osd->o_osd) | |
481 | n = n->rb_left; | |
482 | else if (o > osd->o_osd) | |
483 | n = n->rb_right; | |
484 | else | |
485 | return osd; | |
486 | } | |
487 | return NULL; | |
488 | } | |
489 | ||
490 | ||
491 | /* | |
492 | * Register request, assign tid. If this is the first request, set up | |
493 | * the timeout event. | |
494 | */ | |
495 | static void register_request(struct ceph_osd_client *osdc, | |
496 | struct ceph_osd_request *req) | |
497 | { | |
f24e9980 SW |
498 | mutex_lock(&osdc->request_mutex); |
499 | req->r_tid = ++osdc->last_tid; | |
6df058c0 | 500 | req->r_request->hdr.tid = cpu_to_le64(req->r_tid); |
f24e9980 SW |
501 | |
502 | dout("register_request %p tid %lld\n", req, req->r_tid); | |
503 | __insert_request(osdc, req); | |
504 | ceph_osdc_get_request(req); | |
505 | osdc->num_requests++; | |
506 | ||
507 | req->r_timeout_stamp = | |
6b805185 | 508 | jiffies + osdc->client->mount_args->osd_timeout*HZ; |
f24e9980 SW |
509 | |
510 | if (osdc->num_requests == 1) { | |
511 | osdc->timeout_tid = req->r_tid; | |
512 | dout(" timeout on tid %llu at %lu\n", req->r_tid, | |
513 | req->r_timeout_stamp); | |
514 | schedule_delayed_work(&osdc->timeout_work, | |
515 | round_jiffies_relative(req->r_timeout_stamp - jiffies)); | |
516 | } | |
517 | mutex_unlock(&osdc->request_mutex); | |
518 | } | |
519 | ||
520 | /* | |
521 | * called under osdc->request_mutex | |
522 | */ | |
523 | static void __unregister_request(struct ceph_osd_client *osdc, | |
524 | struct ceph_osd_request *req) | |
525 | { | |
526 | dout("__unregister_request %p tid %lld\n", req, req->r_tid); | |
527 | rb_erase(&req->r_node, &osdc->requests); | |
528 | osdc->num_requests--; | |
529 | ||
0ba6478d SW |
530 | if (req->r_osd) { |
531 | /* make sure the original request isn't in flight. */ | |
532 | ceph_con_revoke(&req->r_osd->o_con, req->r_request); | |
533 | ||
534 | list_del_init(&req->r_osd_item); | |
535 | if (list_empty(&req->r_osd->o_requests)) | |
536 | remove_osd(osdc, req->r_osd); | |
537 | req->r_osd = NULL; | |
538 | } | |
f24e9980 SW |
539 | |
540 | ceph_osdc_put_request(req); | |
541 | ||
542 | if (req->r_tid == osdc->timeout_tid) { | |
543 | if (osdc->num_requests == 0) { | |
544 | dout("no requests, canceling timeout\n"); | |
545 | osdc->timeout_tid = 0; | |
546 | cancel_delayed_work(&osdc->timeout_work); | |
547 | } else { | |
548 | req = rb_entry(rb_first(&osdc->requests), | |
549 | struct ceph_osd_request, r_node); | |
550 | osdc->timeout_tid = req->r_tid; | |
551 | dout("rescheduled timeout on tid %llu at %lu\n", | |
552 | req->r_tid, req->r_timeout_stamp); | |
553 | schedule_delayed_work(&osdc->timeout_work, | |
554 | round_jiffies_relative(req->r_timeout_stamp - | |
555 | jiffies)); | |
556 | } | |
557 | } | |
558 | } | |
559 | ||
560 | /* | |
561 | * Cancel a previously queued request message | |
562 | */ | |
563 | static void __cancel_request(struct ceph_osd_request *req) | |
564 | { | |
565 | if (req->r_sent) { | |
566 | ceph_con_revoke(&req->r_osd->o_con, req->r_request); | |
567 | req->r_sent = 0; | |
568 | } | |
569 | } | |
570 | ||
571 | /* | |
572 | * Pick an osd (the first 'up' osd in the pg), allocate the osd struct | |
573 | * (as needed), and set the request r_osd appropriately. If there is | |
574 | * no up osd, set r_osd to NULL. | |
575 | * | |
576 | * Return 0 if unchanged, 1 if changed, or negative on error. | |
577 | * | |
578 | * Caller should hold map_sem for read and request_mutex. | |
579 | */ | |
580 | static int __map_osds(struct ceph_osd_client *osdc, | |
581 | struct ceph_osd_request *req) | |
582 | { | |
583 | struct ceph_osd_request_head *reqhead = req->r_request->front.iov_base; | |
51042122 | 584 | struct ceph_pg pgid; |
f24e9980 SW |
585 | int o = -1; |
586 | int err; | |
587 | struct ceph_osd *newosd = NULL; | |
588 | ||
589 | dout("map_osds %p tid %lld\n", req, req->r_tid); | |
590 | err = ceph_calc_object_layout(&reqhead->layout, req->r_oid, | |
591 | &req->r_file_layout, osdc->osdmap); | |
592 | if (err) | |
593 | return err; | |
51042122 | 594 | pgid = reqhead->layout.ol_pgid; |
7740a42f SW |
595 | req->r_pgid = pgid; |
596 | ||
f24e9980 SW |
597 | o = ceph_calc_pg_primary(osdc->osdmap, pgid); |
598 | ||
599 | if ((req->r_osd && req->r_osd->o_osd == o && | |
600 | req->r_sent >= req->r_osd->o_incarnation) || | |
601 | (req->r_osd == NULL && o == -1)) | |
602 | return 0; /* no change */ | |
603 | ||
51042122 SW |
604 | dout("map_osds tid %llu pgid %d.%x osd%d (was osd%d)\n", |
605 | req->r_tid, le32_to_cpu(pgid.pool), le16_to_cpu(pgid.ps), o, | |
f24e9980 SW |
606 | req->r_osd ? req->r_osd->o_osd : -1); |
607 | ||
608 | if (req->r_osd) { | |
609 | __cancel_request(req); | |
610 | list_del_init(&req->r_osd_item); | |
611 | if (list_empty(&req->r_osd->o_requests)) { | |
612 | /* try to re-use r_osd if possible */ | |
613 | newosd = get_osd(req->r_osd); | |
614 | remove_osd(osdc, newosd); | |
615 | } | |
616 | req->r_osd = NULL; | |
617 | } | |
618 | ||
619 | req->r_osd = __lookup_osd(osdc, o); | |
620 | if (!req->r_osd && o >= 0) { | |
621 | if (newosd) { | |
622 | req->r_osd = newosd; | |
623 | newosd = NULL; | |
624 | } else { | |
625 | err = -ENOMEM; | |
626 | req->r_osd = create_osd(osdc); | |
627 | if (!req->r_osd) | |
628 | goto out; | |
629 | } | |
630 | ||
631 | dout("map_osds osd %p is osd%d\n", req->r_osd, o); | |
632 | req->r_osd->o_osd = o; | |
633 | req->r_osd->o_con.peer_name.num = cpu_to_le64(o); | |
634 | __insert_osd(osdc, req->r_osd); | |
635 | ||
636 | ceph_con_open(&req->r_osd->o_con, &osdc->osdmap->osd_addr[o]); | |
637 | } | |
638 | ||
639 | if (req->r_osd) | |
640 | list_add(&req->r_osd_item, &req->r_osd->o_requests); | |
641 | err = 1; /* osd changed */ | |
642 | ||
643 | out: | |
644 | if (newosd) | |
645 | put_osd(newosd); | |
646 | return err; | |
647 | } | |
648 | ||
649 | /* | |
650 | * caller should hold map_sem (for read) and request_mutex | |
651 | */ | |
652 | static int __send_request(struct ceph_osd_client *osdc, | |
653 | struct ceph_osd_request *req) | |
654 | { | |
655 | struct ceph_osd_request_head *reqhead; | |
656 | int err; | |
657 | ||
658 | err = __map_osds(osdc, req); | |
659 | if (err < 0) | |
660 | return err; | |
661 | if (req->r_osd == NULL) { | |
662 | dout("send_request %p no up osds in pg\n", req); | |
663 | ceph_monc_request_next_osdmap(&osdc->client->monc); | |
664 | return 0; | |
665 | } | |
666 | ||
667 | dout("send_request %p tid %llu to osd%d flags %d\n", | |
668 | req, req->r_tid, req->r_osd->o_osd, req->r_flags); | |
669 | ||
670 | reqhead = req->r_request->front.iov_base; | |
671 | reqhead->osdmap_epoch = cpu_to_le32(osdc->osdmap->epoch); | |
672 | reqhead->flags |= cpu_to_le32(req->r_flags); /* e.g., RETRY */ | |
673 | reqhead->reassert_version = req->r_reassert_version; | |
674 | ||
6b805185 | 675 | req->r_timeout_stamp = jiffies+osdc->client->mount_args->osd_timeout*HZ; |
f24e9980 SW |
676 | |
677 | ceph_msg_get(req->r_request); /* send consumes a ref */ | |
678 | ceph_con_send(&req->r_osd->o_con, req->r_request); | |
679 | req->r_sent = req->r_osd->o_incarnation; | |
680 | return 0; | |
681 | } | |
682 | ||
683 | /* | |
684 | * Timeout callback, called every N seconds when 1 or more osd | |
685 | * requests has been active for more than N seconds. When this | |
686 | * happens, we ping all OSDs with requests who have timed out to | |
687 | * ensure any communications channel reset is detected. Reset the | |
688 | * request timeouts another N seconds in the future as we go. | |
689 | * Reschedule the timeout event another N seconds in future (unless | |
690 | * there are no open requests). | |
691 | */ | |
692 | static void handle_timeout(struct work_struct *work) | |
693 | { | |
694 | struct ceph_osd_client *osdc = | |
695 | container_of(work, struct ceph_osd_client, timeout_work.work); | |
696 | struct ceph_osd_request *req; | |
697 | struct ceph_osd *osd; | |
6b805185 | 698 | unsigned long timeout = osdc->client->mount_args->osd_timeout * HZ; |
f24e9980 SW |
699 | unsigned long next_timeout = timeout + jiffies; |
700 | struct rb_node *p; | |
701 | ||
702 | dout("timeout\n"); | |
703 | down_read(&osdc->map_sem); | |
704 | ||
705 | ceph_monc_request_next_osdmap(&osdc->client->monc); | |
706 | ||
707 | mutex_lock(&osdc->request_mutex); | |
708 | for (p = rb_first(&osdc->requests); p; p = rb_next(p)) { | |
709 | req = rb_entry(p, struct ceph_osd_request, r_node); | |
710 | ||
711 | if (req->r_resend) { | |
712 | int err; | |
713 | ||
714 | dout("osdc resending prev failed %lld\n", req->r_tid); | |
715 | err = __send_request(osdc, req); | |
716 | if (err) | |
717 | dout("osdc failed again on %lld\n", req->r_tid); | |
718 | else | |
719 | req->r_resend = false; | |
720 | continue; | |
721 | } | |
722 | } | |
723 | for (p = rb_first(&osdc->osds); p; p = rb_next(p)) { | |
724 | osd = rb_entry(p, struct ceph_osd, o_node); | |
725 | if (list_empty(&osd->o_requests)) | |
726 | continue; | |
727 | req = list_first_entry(&osd->o_requests, | |
728 | struct ceph_osd_request, r_osd_item); | |
729 | if (time_before(jiffies, req->r_timeout_stamp)) | |
730 | continue; | |
731 | ||
732 | dout(" tid %llu (at least) timed out on osd%d\n", | |
733 | req->r_tid, osd->o_osd); | |
734 | req->r_timeout_stamp = next_timeout; | |
735 | ceph_con_keepalive(&osd->o_con); | |
736 | } | |
737 | ||
738 | if (osdc->timeout_tid) | |
739 | schedule_delayed_work(&osdc->timeout_work, | |
740 | round_jiffies_relative(timeout)); | |
741 | ||
742 | mutex_unlock(&osdc->request_mutex); | |
743 | ||
744 | up_read(&osdc->map_sem); | |
745 | } | |
746 | ||
747 | /* | |
748 | * handle osd op reply. either call the callback if it is specified, | |
749 | * or do the completion to wake up the waiting thread. | |
750 | */ | |
350b1c32 SW |
751 | static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg, |
752 | struct ceph_connection *con) | |
f24e9980 SW |
753 | { |
754 | struct ceph_osd_reply_head *rhead = msg->front.iov_base; | |
755 | struct ceph_osd_request *req; | |
756 | u64 tid; | |
757 | int numops, object_len, flags; | |
758 | ||
6df058c0 | 759 | tid = le64_to_cpu(msg->hdr.tid); |
f24e9980 SW |
760 | if (msg->front.iov_len < sizeof(*rhead)) |
761 | goto bad; | |
f24e9980 SW |
762 | numops = le32_to_cpu(rhead->num_ops); |
763 | object_len = le32_to_cpu(rhead->object_len); | |
764 | if (msg->front.iov_len != sizeof(*rhead) + object_len + | |
765 | numops * sizeof(struct ceph_osd_op)) | |
766 | goto bad; | |
767 | dout("handle_reply %p tid %llu\n", msg, tid); | |
768 | ||
769 | /* lookup */ | |
770 | mutex_lock(&osdc->request_mutex); | |
771 | req = __lookup_request(osdc, tid); | |
772 | if (req == NULL) { | |
773 | dout("handle_reply tid %llu dne\n", tid); | |
774 | mutex_unlock(&osdc->request_mutex); | |
775 | return; | |
776 | } | |
777 | ceph_osdc_get_request(req); | |
778 | flags = le32_to_cpu(rhead->flags); | |
779 | ||
350b1c32 | 780 | /* |
0d59ab81 | 781 | * if this connection filled our message, drop our reference now, to |
350b1c32 SW |
782 | * avoid a (safe but slower) revoke later. |
783 | */ | |
0d59ab81 YS |
784 | if (req->r_con_filling_msg == con && req->r_reply == msg) { |
785 | dout(" got pages, dropping con_filling_msg ref %p\n", con); | |
786 | req->r_con_filling_msg = NULL; | |
350b1c32 SW |
787 | ceph_con_put(con); |
788 | } | |
789 | ||
f24e9980 SW |
790 | if (req->r_reply) { |
791 | /* | |
792 | * once we see the message has been received, we don't | |
793 | * need a ref (which is only needed for revoking | |
794 | * pages) | |
795 | */ | |
796 | ceph_msg_put(req->r_reply); | |
797 | req->r_reply = NULL; | |
798 | } | |
799 | ||
800 | if (!req->r_got_reply) { | |
801 | unsigned bytes; | |
802 | ||
803 | req->r_result = le32_to_cpu(rhead->result); | |
804 | bytes = le32_to_cpu(msg->hdr.data_len); | |
805 | dout("handle_reply result %d bytes %d\n", req->r_result, | |
806 | bytes); | |
807 | if (req->r_result == 0) | |
808 | req->r_result = bytes; | |
809 | ||
810 | /* in case this is a write and we need to replay, */ | |
811 | req->r_reassert_version = rhead->reassert_version; | |
812 | ||
813 | req->r_got_reply = 1; | |
814 | } else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) { | |
815 | dout("handle_reply tid %llu dup ack\n", tid); | |
34b43a56 | 816 | mutex_unlock(&osdc->request_mutex); |
f24e9980 SW |
817 | goto done; |
818 | } | |
819 | ||
820 | dout("handle_reply tid %llu flags %d\n", tid, flags); | |
821 | ||
822 | /* either this is a read, or we got the safe response */ | |
823 | if ((flags & CEPH_OSD_FLAG_ONDISK) || | |
824 | ((flags & CEPH_OSD_FLAG_WRITE) == 0)) | |
825 | __unregister_request(osdc, req); | |
826 | ||
827 | mutex_unlock(&osdc->request_mutex); | |
828 | ||
829 | if (req->r_callback) | |
830 | req->r_callback(req, msg); | |
831 | else | |
832 | complete(&req->r_completion); | |
833 | ||
834 | if (flags & CEPH_OSD_FLAG_ONDISK) { | |
835 | if (req->r_safe_callback) | |
836 | req->r_safe_callback(req, msg); | |
837 | complete(&req->r_safe_completion); /* fsync waiter */ | |
838 | } | |
839 | ||
840 | done: | |
841 | ceph_osdc_put_request(req); | |
842 | return; | |
843 | ||
844 | bad: | |
845 | pr_err("corrupt osd_op_reply got %d %d expected %d\n", | |
846 | (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len), | |
847 | (int)sizeof(*rhead)); | |
9ec7cab1 | 848 | ceph_msg_dump(msg); |
f24e9980 SW |
849 | } |
850 | ||
851 | ||
852 | /* | |
853 | * Resubmit osd requests whose osd or osd address has changed. Request | |
854 | * a new osd map if osds are down, or we are otherwise unable to determine | |
855 | * how to direct a request. | |
856 | * | |
857 | * Close connections to down osds. | |
858 | * | |
859 | * If @who is specified, resubmit requests for that specific osd. | |
860 | * | |
861 | * Caller should hold map_sem for read and request_mutex. | |
862 | */ | |
863 | static void kick_requests(struct ceph_osd_client *osdc, | |
864 | struct ceph_osd *kickosd) | |
865 | { | |
866 | struct ceph_osd_request *req; | |
867 | struct rb_node *p, *n; | |
868 | int needmap = 0; | |
869 | int err; | |
870 | ||
871 | dout("kick_requests osd%d\n", kickosd ? kickosd->o_osd : -1); | |
872 | mutex_lock(&osdc->request_mutex); | |
873 | if (!kickosd) { | |
874 | for (p = rb_first(&osdc->osds); p; p = n) { | |
875 | struct ceph_osd *osd = | |
876 | rb_entry(p, struct ceph_osd, o_node); | |
877 | ||
878 | n = rb_next(p); | |
879 | if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) || | |
103e2d3a SW |
880 | memcmp(&osd->o_con.peer_addr, |
881 | ceph_osd_addr(osdc->osdmap, | |
882 | osd->o_osd), | |
883 | sizeof(struct ceph_entity_addr)) != 0) | |
f24e9980 SW |
884 | reset_osd(osdc, osd); |
885 | } | |
886 | } | |
887 | ||
888 | for (p = rb_first(&osdc->requests); p; p = rb_next(p)) { | |
889 | req = rb_entry(p, struct ceph_osd_request, r_node); | |
890 | ||
891 | if (req->r_resend) { | |
892 | dout(" r_resend set on tid %llu\n", req->r_tid); | |
266673db | 893 | __cancel_request(req); |
f24e9980 SW |
894 | goto kick; |
895 | } | |
266673db SW |
896 | if (req->r_osd && kickosd == req->r_osd) { |
897 | __cancel_request(req); | |
f24e9980 | 898 | goto kick; |
266673db | 899 | } |
f24e9980 SW |
900 | |
901 | err = __map_osds(osdc, req); | |
902 | if (err == 0) | |
903 | continue; /* no change */ | |
904 | if (err < 0) { | |
905 | /* | |
906 | * FIXME: really, we should set the request | |
907 | * error and fail if this isn't a 'nofail' | |
908 | * request, but that's a fair bit more | |
909 | * complicated to do. So retry! | |
910 | */ | |
911 | dout(" setting r_resend on %llu\n", req->r_tid); | |
912 | req->r_resend = true; | |
913 | continue; | |
914 | } | |
915 | if (req->r_osd == NULL) { | |
916 | dout("tid %llu maps to no valid osd\n", req->r_tid); | |
917 | needmap++; /* request a newer map */ | |
918 | continue; | |
919 | } | |
920 | ||
921 | kick: | |
c1ea8823 SW |
922 | dout("kicking %p tid %llu osd%d\n", req, req->r_tid, |
923 | req->r_osd->o_osd); | |
f24e9980 SW |
924 | req->r_flags |= CEPH_OSD_FLAG_RETRY; |
925 | err = __send_request(osdc, req); | |
926 | if (err) { | |
927 | dout(" setting r_resend on %llu\n", req->r_tid); | |
928 | req->r_resend = true; | |
929 | } | |
930 | } | |
931 | mutex_unlock(&osdc->request_mutex); | |
932 | ||
933 | if (needmap) { | |
934 | dout("%d requests for down osds, need new map\n", needmap); | |
935 | ceph_monc_request_next_osdmap(&osdc->client->monc); | |
936 | } | |
937 | } | |
938 | ||
939 | /* | |
940 | * Process updated osd map. | |
941 | * | |
942 | * The message contains any number of incremental and full maps, normally | |
943 | * indicating some sort of topology change in the cluster. Kick requests | |
944 | * off to different OSDs as needed. | |
945 | */ | |
946 | void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg) | |
947 | { | |
948 | void *p, *end, *next; | |
949 | u32 nr_maps, maplen; | |
950 | u32 epoch; | |
951 | struct ceph_osdmap *newmap = NULL, *oldmap; | |
952 | int err; | |
953 | struct ceph_fsid fsid; | |
954 | ||
955 | dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0); | |
956 | p = msg->front.iov_base; | |
957 | end = p + msg->front.iov_len; | |
958 | ||
959 | /* verify fsid */ | |
960 | ceph_decode_need(&p, end, sizeof(fsid), bad); | |
961 | ceph_decode_copy(&p, &fsid, sizeof(fsid)); | |
0743304d SW |
962 | if (ceph_check_fsid(osdc->client, &fsid) < 0) |
963 | return; | |
f24e9980 SW |
964 | |
965 | down_write(&osdc->map_sem); | |
966 | ||
967 | /* incremental maps */ | |
968 | ceph_decode_32_safe(&p, end, nr_maps, bad); | |
969 | dout(" %d inc maps\n", nr_maps); | |
970 | while (nr_maps > 0) { | |
971 | ceph_decode_need(&p, end, 2*sizeof(u32), bad); | |
c89136ea SW |
972 | epoch = ceph_decode_32(&p); |
973 | maplen = ceph_decode_32(&p); | |
f24e9980 SW |
974 | ceph_decode_need(&p, end, maplen, bad); |
975 | next = p + maplen; | |
976 | if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) { | |
977 | dout("applying incremental map %u len %d\n", | |
978 | epoch, maplen); | |
979 | newmap = osdmap_apply_incremental(&p, next, | |
980 | osdc->osdmap, | |
981 | osdc->client->msgr); | |
982 | if (IS_ERR(newmap)) { | |
983 | err = PTR_ERR(newmap); | |
984 | goto bad; | |
985 | } | |
30dc6381 | 986 | BUG_ON(!newmap); |
f24e9980 SW |
987 | if (newmap != osdc->osdmap) { |
988 | ceph_osdmap_destroy(osdc->osdmap); | |
989 | osdc->osdmap = newmap; | |
990 | } | |
991 | } else { | |
992 | dout("ignoring incremental map %u len %d\n", | |
993 | epoch, maplen); | |
994 | } | |
995 | p = next; | |
996 | nr_maps--; | |
997 | } | |
998 | if (newmap) | |
999 | goto done; | |
1000 | ||
1001 | /* full maps */ | |
1002 | ceph_decode_32_safe(&p, end, nr_maps, bad); | |
1003 | dout(" %d full maps\n", nr_maps); | |
1004 | while (nr_maps) { | |
1005 | ceph_decode_need(&p, end, 2*sizeof(u32), bad); | |
c89136ea SW |
1006 | epoch = ceph_decode_32(&p); |
1007 | maplen = ceph_decode_32(&p); | |
f24e9980 SW |
1008 | ceph_decode_need(&p, end, maplen, bad); |
1009 | if (nr_maps > 1) { | |
1010 | dout("skipping non-latest full map %u len %d\n", | |
1011 | epoch, maplen); | |
1012 | } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) { | |
1013 | dout("skipping full map %u len %d, " | |
1014 | "older than our %u\n", epoch, maplen, | |
1015 | osdc->osdmap->epoch); | |
1016 | } else { | |
1017 | dout("taking full map %u len %d\n", epoch, maplen); | |
1018 | newmap = osdmap_decode(&p, p+maplen); | |
1019 | if (IS_ERR(newmap)) { | |
1020 | err = PTR_ERR(newmap); | |
1021 | goto bad; | |
1022 | } | |
30dc6381 | 1023 | BUG_ON(!newmap); |
f24e9980 SW |
1024 | oldmap = osdc->osdmap; |
1025 | osdc->osdmap = newmap; | |
1026 | if (oldmap) | |
1027 | ceph_osdmap_destroy(oldmap); | |
1028 | } | |
1029 | p += maplen; | |
1030 | nr_maps--; | |
1031 | } | |
1032 | ||
1033 | done: | |
1034 | downgrade_write(&osdc->map_sem); | |
1035 | ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch); | |
1036 | if (newmap) | |
1037 | kick_requests(osdc, NULL); | |
1038 | up_read(&osdc->map_sem); | |
1039 | return; | |
1040 | ||
1041 | bad: | |
1042 | pr_err("osdc handle_map corrupt msg\n"); | |
9ec7cab1 | 1043 | ceph_msg_dump(msg); |
f24e9980 SW |
1044 | up_write(&osdc->map_sem); |
1045 | return; | |
1046 | } | |
1047 | ||
1048 | ||
1049 | /* | |
1050 | * A read request prepares specific pages that data is to be read into. | |
1051 | * When a message is being read off the wire, we call prepare_pages to | |
1052 | * find those pages. | |
1053 | * 0 = success, -1 failure. | |
1054 | */ | |
0d59ab81 | 1055 | static int __prepare_pages(struct ceph_connection *con, |
0547a9b3 YS |
1056 | struct ceph_msg_header *hdr, |
1057 | struct ceph_osd_request *req, | |
1058 | u64 tid, | |
1059 | struct ceph_msg *m) | |
f24e9980 SW |
1060 | { |
1061 | struct ceph_osd *osd = con->private; | |
1062 | struct ceph_osd_client *osdc; | |
f24e9980 | 1063 | int ret = -1; |
0547a9b3 YS |
1064 | int data_len = le32_to_cpu(hdr->data_len); |
1065 | unsigned data_off = le16_to_cpu(hdr->data_off); | |
1066 | ||
1067 | int want = calc_pages_for(data_off & ~PAGE_MASK, data_len); | |
f24e9980 SW |
1068 | |
1069 | if (!osd) | |
1070 | return -1; | |
0547a9b3 | 1071 | |
f24e9980 SW |
1072 | osdc = osd->o_osdc; |
1073 | ||
0d59ab81 | 1074 | dout("__prepare_pages on msg %p tid %llu, has %d pages, want %d\n", m, |
f24e9980 | 1075 | tid, req->r_num_pages, want); |
350b1c32 SW |
1076 | if (unlikely(req->r_num_pages < want)) |
1077 | goto out; | |
350b1c32 SW |
1078 | m->pages = req->r_pages; |
1079 | m->nr_pages = req->r_num_pages; | |
1080 | ret = 0; /* success */ | |
f24e9980 | 1081 | out: |
0547a9b3 YS |
1082 | BUG_ON(ret < 0 || m->nr_pages < want); |
1083 | ||
f24e9980 SW |
1084 | return ret; |
1085 | } | |
1086 | ||
1087 | /* | |
1088 | * Register request, send initial attempt. | |
1089 | */ | |
1090 | int ceph_osdc_start_request(struct ceph_osd_client *osdc, | |
1091 | struct ceph_osd_request *req, | |
1092 | bool nofail) | |
1093 | { | |
c1ea8823 | 1094 | int rc = 0; |
f24e9980 SW |
1095 | |
1096 | req->r_request->pages = req->r_pages; | |
1097 | req->r_request->nr_pages = req->r_num_pages; | |
1098 | ||
1099 | register_request(osdc, req); | |
1100 | ||
1101 | down_read(&osdc->map_sem); | |
1102 | mutex_lock(&osdc->request_mutex); | |
c1ea8823 SW |
1103 | /* |
1104 | * a racing kick_requests() may have sent the message for us | |
1105 | * while we dropped request_mutex above, so only send now if | |
1106 | * the request still han't been touched yet. | |
1107 | */ | |
1108 | if (req->r_sent == 0) { | |
1109 | rc = __send_request(osdc, req); | |
1110 | if (rc) { | |
1111 | if (nofail) { | |
1112 | dout("osdc_start_request failed send, " | |
1113 | " marking %lld\n", req->r_tid); | |
1114 | req->r_resend = true; | |
1115 | rc = 0; | |
1116 | } else { | |
1117 | __unregister_request(osdc, req); | |
1118 | } | |
f24e9980 SW |
1119 | } |
1120 | } | |
1121 | mutex_unlock(&osdc->request_mutex); | |
1122 | up_read(&osdc->map_sem); | |
1123 | return rc; | |
1124 | } | |
1125 | ||
1126 | /* | |
1127 | * wait for a request to complete | |
1128 | */ | |
1129 | int ceph_osdc_wait_request(struct ceph_osd_client *osdc, | |
1130 | struct ceph_osd_request *req) | |
1131 | { | |
1132 | int rc; | |
1133 | ||
1134 | rc = wait_for_completion_interruptible(&req->r_completion); | |
1135 | if (rc < 0) { | |
1136 | mutex_lock(&osdc->request_mutex); | |
1137 | __cancel_request(req); | |
529cfcc4 | 1138 | __unregister_request(osdc, req); |
f24e9980 | 1139 | mutex_unlock(&osdc->request_mutex); |
529cfcc4 | 1140 | dout("wait_request tid %llu canceled/timed out\n", req->r_tid); |
f24e9980 SW |
1141 | return rc; |
1142 | } | |
1143 | ||
1144 | dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result); | |
1145 | return req->r_result; | |
1146 | } | |
1147 | ||
1148 | /* | |
1149 | * sync - wait for all in-flight requests to flush. avoid starvation. | |
1150 | */ | |
1151 | void ceph_osdc_sync(struct ceph_osd_client *osdc) | |
1152 | { | |
1153 | struct ceph_osd_request *req; | |
1154 | u64 last_tid, next_tid = 0; | |
1155 | ||
1156 | mutex_lock(&osdc->request_mutex); | |
1157 | last_tid = osdc->last_tid; | |
1158 | while (1) { | |
1159 | req = __lookup_request_ge(osdc, next_tid); | |
1160 | if (!req) | |
1161 | break; | |
1162 | if (req->r_tid > last_tid) | |
1163 | break; | |
1164 | ||
1165 | next_tid = req->r_tid + 1; | |
1166 | if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0) | |
1167 | continue; | |
1168 | ||
1169 | ceph_osdc_get_request(req); | |
1170 | mutex_unlock(&osdc->request_mutex); | |
1171 | dout("sync waiting on tid %llu (last is %llu)\n", | |
1172 | req->r_tid, last_tid); | |
1173 | wait_for_completion(&req->r_safe_completion); | |
1174 | mutex_lock(&osdc->request_mutex); | |
1175 | ceph_osdc_put_request(req); | |
1176 | } | |
1177 | mutex_unlock(&osdc->request_mutex); | |
1178 | dout("sync done (thru tid %llu)\n", last_tid); | |
1179 | } | |
1180 | ||
1181 | /* | |
1182 | * init, shutdown | |
1183 | */ | |
1184 | int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client) | |
1185 | { | |
1186 | int err; | |
1187 | ||
1188 | dout("init\n"); | |
1189 | osdc->client = client; | |
1190 | osdc->osdmap = NULL; | |
1191 | init_rwsem(&osdc->map_sem); | |
1192 | init_completion(&osdc->map_waiters); | |
1193 | osdc->last_requested_map = 0; | |
1194 | mutex_init(&osdc->request_mutex); | |
1195 | osdc->timeout_tid = 0; | |
1196 | osdc->last_tid = 0; | |
1197 | osdc->osds = RB_ROOT; | |
1198 | osdc->requests = RB_ROOT; | |
1199 | osdc->num_requests = 0; | |
1200 | INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout); | |
1201 | ||
5f44f142 | 1202 | err = -ENOMEM; |
f24e9980 SW |
1203 | osdc->req_mempool = mempool_create_kmalloc_pool(10, |
1204 | sizeof(struct ceph_osd_request)); | |
1205 | if (!osdc->req_mempool) | |
5f44f142 | 1206 | goto out; |
f24e9980 SW |
1207 | |
1208 | err = ceph_msgpool_init(&osdc->msgpool_op, 4096, 10, true); | |
1209 | if (err < 0) | |
5f44f142 | 1210 | goto out_mempool; |
f24e9980 | 1211 | return 0; |
5f44f142 | 1212 | |
5f44f142 SW |
1213 | out_mempool: |
1214 | mempool_destroy(osdc->req_mempool); | |
1215 | out: | |
1216 | return err; | |
f24e9980 SW |
1217 | } |
1218 | ||
1219 | void ceph_osdc_stop(struct ceph_osd_client *osdc) | |
1220 | { | |
1221 | cancel_delayed_work_sync(&osdc->timeout_work); | |
1222 | if (osdc->osdmap) { | |
1223 | ceph_osdmap_destroy(osdc->osdmap); | |
1224 | osdc->osdmap = NULL; | |
1225 | } | |
1226 | mempool_destroy(osdc->req_mempool); | |
1227 | ceph_msgpool_destroy(&osdc->msgpool_op); | |
f24e9980 SW |
1228 | } |
1229 | ||
1230 | /* | |
1231 | * Read some contiguous pages. If we cross a stripe boundary, shorten | |
1232 | * *plen. Return number of bytes read, or error. | |
1233 | */ | |
1234 | int ceph_osdc_readpages(struct ceph_osd_client *osdc, | |
1235 | struct ceph_vino vino, struct ceph_file_layout *layout, | |
1236 | u64 off, u64 *plen, | |
1237 | u32 truncate_seq, u64 truncate_size, | |
1238 | struct page **pages, int num_pages) | |
1239 | { | |
1240 | struct ceph_osd_request *req; | |
1241 | int rc = 0; | |
1242 | ||
1243 | dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino, | |
1244 | vino.snap, off, *plen); | |
1245 | req = ceph_osdc_new_request(osdc, layout, vino, off, plen, | |
1246 | CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ, | |
1247 | NULL, 0, truncate_seq, truncate_size, NULL, | |
1248 | false, 1); | |
1249 | if (IS_ERR(req)) | |
1250 | return PTR_ERR(req); | |
1251 | ||
1252 | /* it may be a short read due to an object boundary */ | |
1253 | req->r_pages = pages; | |
1254 | num_pages = calc_pages_for(off, *plen); | |
1255 | req->r_num_pages = num_pages; | |
1256 | ||
1257 | dout("readpages final extent is %llu~%llu (%d pages)\n", | |
1258 | off, *plen, req->r_num_pages); | |
1259 | ||
1260 | rc = ceph_osdc_start_request(osdc, req, false); | |
1261 | if (!rc) | |
1262 | rc = ceph_osdc_wait_request(osdc, req); | |
1263 | ||
1264 | ceph_osdc_put_request(req); | |
1265 | dout("readpages result %d\n", rc); | |
1266 | return rc; | |
1267 | } | |
1268 | ||
1269 | /* | |
1270 | * do a synchronous write on N pages | |
1271 | */ | |
1272 | int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino, | |
1273 | struct ceph_file_layout *layout, | |
1274 | struct ceph_snap_context *snapc, | |
1275 | u64 off, u64 len, | |
1276 | u32 truncate_seq, u64 truncate_size, | |
1277 | struct timespec *mtime, | |
1278 | struct page **pages, int num_pages, | |
1279 | int flags, int do_sync, bool nofail) | |
1280 | { | |
1281 | struct ceph_osd_request *req; | |
1282 | int rc = 0; | |
1283 | ||
1284 | BUG_ON(vino.snap != CEPH_NOSNAP); | |
1285 | req = ceph_osdc_new_request(osdc, layout, vino, off, &len, | |
1286 | CEPH_OSD_OP_WRITE, | |
1287 | flags | CEPH_OSD_FLAG_ONDISK | | |
1288 | CEPH_OSD_FLAG_WRITE, | |
1289 | snapc, do_sync, | |
1290 | truncate_seq, truncate_size, mtime, | |
1291 | nofail, 1); | |
1292 | if (IS_ERR(req)) | |
1293 | return PTR_ERR(req); | |
1294 | ||
1295 | /* it may be a short write due to an object boundary */ | |
1296 | req->r_pages = pages; | |
1297 | req->r_num_pages = calc_pages_for(off, len); | |
1298 | dout("writepages %llu~%llu (%d pages)\n", off, len, | |
1299 | req->r_num_pages); | |
1300 | ||
1301 | rc = ceph_osdc_start_request(osdc, req, nofail); | |
1302 | if (!rc) | |
1303 | rc = ceph_osdc_wait_request(osdc, req); | |
1304 | ||
1305 | ceph_osdc_put_request(req); | |
1306 | if (rc == 0) | |
1307 | rc = len; | |
1308 | dout("writepages result %d\n", rc); | |
1309 | return rc; | |
1310 | } | |
1311 | ||
1312 | /* | |
1313 | * handle incoming message | |
1314 | */ | |
1315 | static void dispatch(struct ceph_connection *con, struct ceph_msg *msg) | |
1316 | { | |
1317 | struct ceph_osd *osd = con->private; | |
32c895e7 | 1318 | struct ceph_osd_client *osdc; |
f24e9980 SW |
1319 | int type = le16_to_cpu(msg->hdr.type); |
1320 | ||
1321 | if (!osd) | |
1322 | return; | |
32c895e7 | 1323 | osdc = osd->o_osdc; |
f24e9980 SW |
1324 | |
1325 | switch (type) { | |
1326 | case CEPH_MSG_OSD_MAP: | |
1327 | ceph_osdc_handle_map(osdc, msg); | |
1328 | break; | |
1329 | case CEPH_MSG_OSD_OPREPLY: | |
350b1c32 | 1330 | handle_reply(osdc, msg, con); |
f24e9980 SW |
1331 | break; |
1332 | ||
1333 | default: | |
1334 | pr_err("received unknown message type %d %s\n", type, | |
1335 | ceph_msg_type_name(type)); | |
1336 | } | |
1337 | ceph_msg_put(msg); | |
1338 | } | |
1339 | ||
1340 | static struct ceph_msg *alloc_msg(struct ceph_connection *con, | |
2450418c YS |
1341 | struct ceph_msg_header *hdr, |
1342 | int *skip) | |
f24e9980 SW |
1343 | { |
1344 | struct ceph_osd *osd = con->private; | |
1345 | struct ceph_osd_client *osdc = osd->o_osdc; | |
1346 | int type = le16_to_cpu(hdr->type); | |
8f3bc053 | 1347 | int front = le32_to_cpu(hdr->front_len); |
0547a9b3 | 1348 | int data_len = le32_to_cpu(hdr->data_len); |
2450418c | 1349 | struct ceph_msg *m; |
0547a9b3 YS |
1350 | struct ceph_osd_request *req; |
1351 | u64 tid; | |
1352 | int err; | |
f24e9980 | 1353 | |
2450418c | 1354 | *skip = 0; |
0547a9b3 | 1355 | if (type != CEPH_MSG_OSD_OPREPLY) |
2450418c | 1356 | return NULL; |
2450418c | 1357 | |
0547a9b3 YS |
1358 | tid = le64_to_cpu(hdr->tid); |
1359 | mutex_lock(&osdc->request_mutex); | |
1360 | req = __lookup_request(osdc, tid); | |
1361 | if (!req) { | |
1362 | *skip = 1; | |
1363 | m = NULL; | |
0d59ab81 | 1364 | dout("alloc_msg unknown tid %llu\n", tid); |
0547a9b3 YS |
1365 | goto out; |
1366 | } | |
0d59ab81 YS |
1367 | m = __get_next_reply(con, req, front); |
1368 | if (!m || IS_ERR(m)) { | |
2450418c | 1369 | *skip = 1; |
0547a9b3 YS |
1370 | goto out; |
1371 | } | |
1372 | ||
1373 | if (data_len > 0) { | |
0d59ab81 | 1374 | err = __prepare_pages(con, hdr, req, tid, m); |
0547a9b3 YS |
1375 | if (err < 0) { |
1376 | *skip = 1; | |
1377 | ceph_msg_put(m); | |
1378 | m = ERR_PTR(err); | |
1379 | } | |
1380 | } | |
1381 | ||
1382 | out: | |
1383 | mutex_unlock(&osdc->request_mutex); | |
2450418c YS |
1384 | |
1385 | return m; | |
f24e9980 SW |
1386 | } |
1387 | ||
1388 | /* | |
1389 | * Wrappers to refcount containing ceph_osd struct | |
1390 | */ | |
1391 | static struct ceph_connection *get_osd_con(struct ceph_connection *con) | |
1392 | { | |
1393 | struct ceph_osd *osd = con->private; | |
1394 | if (get_osd(osd)) | |
1395 | return con; | |
1396 | return NULL; | |
1397 | } | |
1398 | ||
1399 | static void put_osd_con(struct ceph_connection *con) | |
1400 | { | |
1401 | struct ceph_osd *osd = con->private; | |
1402 | put_osd(osd); | |
1403 | } | |
1404 | ||
4e7a5dcd SW |
1405 | /* |
1406 | * authentication | |
1407 | */ | |
1408 | static int get_authorizer(struct ceph_connection *con, | |
50b885b9 SW |
1409 | void **buf, int *len, int *proto, |
1410 | void **reply_buf, int *reply_len, int force_new) | |
4e7a5dcd SW |
1411 | { |
1412 | struct ceph_osd *o = con->private; | |
1413 | struct ceph_osd_client *osdc = o->o_osdc; | |
1414 | struct ceph_auth_client *ac = osdc->client->monc.auth; | |
1415 | int ret = 0; | |
1416 | ||
1417 | if (force_new && o->o_authorizer) { | |
1418 | ac->ops->destroy_authorizer(ac, o->o_authorizer); | |
1419 | o->o_authorizer = NULL; | |
1420 | } | |
1421 | if (o->o_authorizer == NULL) { | |
1422 | ret = ac->ops->create_authorizer( | |
1423 | ac, CEPH_ENTITY_TYPE_OSD, | |
1424 | &o->o_authorizer, | |
1425 | &o->o_authorizer_buf, | |
1426 | &o->o_authorizer_buf_len, | |
1427 | &o->o_authorizer_reply_buf, | |
1428 | &o->o_authorizer_reply_buf_len); | |
1429 | if (ret) | |
1430 | return ret; | |
1431 | } | |
1432 | ||
1433 | *proto = ac->protocol; | |
1434 | *buf = o->o_authorizer_buf; | |
1435 | *len = o->o_authorizer_buf_len; | |
1436 | *reply_buf = o->o_authorizer_reply_buf; | |
1437 | *reply_len = o->o_authorizer_reply_buf_len; | |
1438 | return 0; | |
1439 | } | |
1440 | ||
1441 | ||
1442 | static int verify_authorizer_reply(struct ceph_connection *con, int len) | |
1443 | { | |
1444 | struct ceph_osd *o = con->private; | |
1445 | struct ceph_osd_client *osdc = o->o_osdc; | |
1446 | struct ceph_auth_client *ac = osdc->client->monc.auth; | |
1447 | ||
1448 | return ac->ops->verify_authorizer_reply(ac, o->o_authorizer, len); | |
1449 | } | |
1450 | ||
1451 | ||
f24e9980 SW |
1452 | const static struct ceph_connection_operations osd_con_ops = { |
1453 | .get = get_osd_con, | |
1454 | .put = put_osd_con, | |
1455 | .dispatch = dispatch, | |
4e7a5dcd SW |
1456 | .get_authorizer = get_authorizer, |
1457 | .verify_authorizer_reply = verify_authorizer_reply, | |
f24e9980 | 1458 | .alloc_msg = alloc_msg, |
81b024e7 | 1459 | .fault = osd_reset, |
f24e9980 | 1460 | }; |