]>
Commit | Line | Data |
---|---|---|
3d14c5d2 | 1 | #include <linux/ceph/ceph_debug.h> |
f24e9980 | 2 | |
3d14c5d2 | 3 | #include <linux/module.h> |
f24e9980 SW |
4 | #include <linux/err.h> |
5 | #include <linux/highmem.h> | |
6 | #include <linux/mm.h> | |
7 | #include <linux/pagemap.h> | |
8 | #include <linux/slab.h> | |
9 | #include <linux/uaccess.h> | |
68b4476b YS |
10 | #ifdef CONFIG_BLOCK |
11 | #include <linux/bio.h> | |
12 | #endif | |
f24e9980 | 13 | |
3d14c5d2 YS |
14 | #include <linux/ceph/libceph.h> |
15 | #include <linux/ceph/osd_client.h> | |
16 | #include <linux/ceph/messenger.h> | |
17 | #include <linux/ceph/decode.h> | |
18 | #include <linux/ceph/auth.h> | |
19 | #include <linux/ceph/pagelist.h> | |
f24e9980 | 20 | |
c16e7869 SW |
21 | #define OSD_OP_FRONT_LEN 4096 |
22 | #define OSD_OPREPLY_FRONT_LEN 512 | |
0d59ab81 | 23 | |
9e32789f | 24 | static const struct ceph_connection_operations osd_con_ops; |
f24e9980 | 25 | |
6f6c7006 SW |
26 | static void send_queued(struct ceph_osd_client *osdc); |
27 | static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd); | |
a40c4f10 YS |
28 | static void __register_request(struct ceph_osd_client *osdc, |
29 | struct ceph_osd_request *req); | |
30 | static void __unregister_linger_request(struct ceph_osd_client *osdc, | |
31 | struct ceph_osd_request *req); | |
56e925b6 SW |
32 | static void __send_request(struct ceph_osd_client *osdc, |
33 | struct ceph_osd_request *req); | |
f24e9980 | 34 | |
68b4476b YS |
35 | static int op_needs_trail(int op) |
36 | { | |
37 | switch (op) { | |
38 | case CEPH_OSD_OP_GETXATTR: | |
39 | case CEPH_OSD_OP_SETXATTR: | |
40 | case CEPH_OSD_OP_CMPXATTR: | |
41 | case CEPH_OSD_OP_CALL: | |
a40c4f10 | 42 | case CEPH_OSD_OP_NOTIFY: |
68b4476b YS |
43 | return 1; |
44 | default: | |
45 | return 0; | |
46 | } | |
47 | } | |
48 | ||
49 | static int op_has_extent(int op) | |
50 | { | |
51 | return (op == CEPH_OSD_OP_READ || | |
52 | op == CEPH_OSD_OP_WRITE); | |
53 | } | |
54 | ||
3499e8a5 YS |
55 | void ceph_calc_raw_layout(struct ceph_osd_client *osdc, |
56 | struct ceph_file_layout *layout, | |
57 | u64 snapid, | |
68b4476b YS |
58 | u64 off, u64 *plen, u64 *bno, |
59 | struct ceph_osd_request *req, | |
60 | struct ceph_osd_req_op *op) | |
3499e8a5 YS |
61 | { |
62 | struct ceph_osd_request_head *reqhead = req->r_request->front.iov_base; | |
68b4476b | 63 | u64 orig_len = *plen; |
3499e8a5 YS |
64 | u64 objoff, objlen; /* extent in object */ |
65 | ||
66 | reqhead->snapid = cpu_to_le64(snapid); | |
67 | ||
68 | /* object extent? */ | |
68b4476b | 69 | ceph_calc_file_object_mapping(layout, off, plen, bno, |
3499e8a5 | 70 | &objoff, &objlen); |
68b4476b | 71 | if (*plen < orig_len) |
3499e8a5 | 72 | dout(" skipping last %llu, final file extent %llu~%llu\n", |
68b4476b | 73 | orig_len - *plen, off, *plen); |
3499e8a5 | 74 | |
68b4476b YS |
75 | if (op_has_extent(op->op)) { |
76 | op->extent.offset = objoff; | |
77 | op->extent.length = objlen; | |
78 | } | |
79 | req->r_num_pages = calc_pages_for(off, *plen); | |
b7495fc2 | 80 | req->r_page_alignment = off & ~PAGE_MASK; |
3d14c5d2 YS |
81 | if (op->op == CEPH_OSD_OP_WRITE) |
82 | op->payload_len = *plen; | |
3499e8a5 YS |
83 | |
84 | dout("calc_layout bno=%llx %llu~%llu (%d pages)\n", | |
85 | *bno, objoff, objlen, req->r_num_pages); | |
86 | ||
87 | } | |
3d14c5d2 | 88 | EXPORT_SYMBOL(ceph_calc_raw_layout); |
3499e8a5 | 89 | |
f24e9980 SW |
90 | /* |
91 | * Implement client access to distributed object storage cluster. | |
92 | * | |
93 | * All data objects are stored within a cluster/cloud of OSDs, or | |
94 | * "object storage devices." (Note that Ceph OSDs have _nothing_ to | |
95 | * do with the T10 OSD extensions to SCSI.) Ceph OSDs are simply | |
96 | * remote daemons serving up and coordinating consistent and safe | |
97 | * access to storage. | |
98 | * | |
99 | * Cluster membership and the mapping of data objects onto storage devices | |
100 | * are described by the osd map. | |
101 | * | |
102 | * We keep track of pending OSD requests (read, write), resubmit | |
103 | * requests to different OSDs when the cluster topology/data layout | |
104 | * change, or retry the affected requests when the communications | |
105 | * channel with an OSD is reset. | |
106 | */ | |
107 | ||
108 | /* | |
109 | * calculate the mapping of a file extent onto an object, and fill out the | |
110 | * request accordingly. shorten extent as necessary if it crosses an | |
111 | * object boundary. | |
112 | * | |
113 | * fill osd op in request message. | |
114 | */ | |
115 | static void calc_layout(struct ceph_osd_client *osdc, | |
3499e8a5 YS |
116 | struct ceph_vino vino, |
117 | struct ceph_file_layout *layout, | |
f24e9980 | 118 | u64 off, u64 *plen, |
68b4476b YS |
119 | struct ceph_osd_request *req, |
120 | struct ceph_osd_req_op *op) | |
f24e9980 | 121 | { |
f24e9980 SW |
122 | u64 bno; |
123 | ||
68b4476b YS |
124 | ceph_calc_raw_layout(osdc, layout, vino.snap, off, |
125 | plen, &bno, req, op); | |
f24e9980 | 126 | |
2dab036b | 127 | snprintf(req->r_oid, sizeof(req->r_oid), "%llx.%08llx", vino.ino, bno); |
f24e9980 | 128 | req->r_oid_len = strlen(req->r_oid); |
f24e9980 SW |
129 | } |
130 | ||
f24e9980 SW |
131 | /* |
132 | * requests | |
133 | */ | |
415e49a9 | 134 | void ceph_osdc_release_request(struct kref *kref) |
f24e9980 | 135 | { |
415e49a9 SW |
136 | struct ceph_osd_request *req = container_of(kref, |
137 | struct ceph_osd_request, | |
138 | r_kref); | |
139 | ||
140 | if (req->r_request) | |
141 | ceph_msg_put(req->r_request); | |
142 | if (req->r_reply) | |
143 | ceph_msg_put(req->r_reply); | |
0d59ab81 | 144 | if (req->r_con_filling_msg) { |
350b1c32 | 145 | dout("release_request revoking pages %p from con %p\n", |
0d59ab81 YS |
146 | req->r_pages, req->r_con_filling_msg); |
147 | ceph_con_revoke_message(req->r_con_filling_msg, | |
148 | req->r_reply); | |
149 | ceph_con_put(req->r_con_filling_msg); | |
350b1c32 | 150 | } |
415e49a9 SW |
151 | if (req->r_own_pages) |
152 | ceph_release_page_vector(req->r_pages, | |
153 | req->r_num_pages); | |
68b4476b YS |
154 | #ifdef CONFIG_BLOCK |
155 | if (req->r_bio) | |
156 | bio_put(req->r_bio); | |
157 | #endif | |
415e49a9 | 158 | ceph_put_snap_context(req->r_snapc); |
68b4476b YS |
159 | if (req->r_trail) { |
160 | ceph_pagelist_release(req->r_trail); | |
161 | kfree(req->r_trail); | |
162 | } | |
415e49a9 SW |
163 | if (req->r_mempool) |
164 | mempool_free(req, req->r_osdc->req_mempool); | |
165 | else | |
166 | kfree(req); | |
f24e9980 | 167 | } |
3d14c5d2 | 168 | EXPORT_SYMBOL(ceph_osdc_release_request); |
68b4476b YS |
169 | |
170 | static int get_num_ops(struct ceph_osd_req_op *ops, int *needs_trail) | |
171 | { | |
172 | int i = 0; | |
173 | ||
174 | if (needs_trail) | |
175 | *needs_trail = 0; | |
176 | while (ops[i].op) { | |
177 | if (needs_trail && op_needs_trail(ops[i].op)) | |
178 | *needs_trail = 1; | |
179 | i++; | |
180 | } | |
181 | ||
182 | return i; | |
183 | } | |
184 | ||
3499e8a5 YS |
185 | struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc, |
186 | int flags, | |
f24e9980 | 187 | struct ceph_snap_context *snapc, |
68b4476b | 188 | struct ceph_osd_req_op *ops, |
3499e8a5 YS |
189 | bool use_mempool, |
190 | gfp_t gfp_flags, | |
68b4476b YS |
191 | struct page **pages, |
192 | struct bio *bio) | |
f24e9980 SW |
193 | { |
194 | struct ceph_osd_request *req; | |
195 | struct ceph_msg *msg; | |
68b4476b YS |
196 | int needs_trail; |
197 | int num_op = get_num_ops(ops, &needs_trail); | |
198 | size_t msg_size = sizeof(struct ceph_osd_request_head); | |
3499e8a5 | 199 | |
68b4476b | 200 | msg_size += num_op*sizeof(struct ceph_osd_op); |
f24e9980 SW |
201 | |
202 | if (use_mempool) { | |
3499e8a5 | 203 | req = mempool_alloc(osdc->req_mempool, gfp_flags); |
f24e9980 SW |
204 | memset(req, 0, sizeof(*req)); |
205 | } else { | |
3499e8a5 | 206 | req = kzalloc(sizeof(*req), gfp_flags); |
f24e9980 SW |
207 | } |
208 | if (req == NULL) | |
a79832f2 | 209 | return NULL; |
f24e9980 | 210 | |
f24e9980 SW |
211 | req->r_osdc = osdc; |
212 | req->r_mempool = use_mempool; | |
68b4476b | 213 | |
415e49a9 | 214 | kref_init(&req->r_kref); |
f24e9980 SW |
215 | init_completion(&req->r_completion); |
216 | init_completion(&req->r_safe_completion); | |
217 | INIT_LIST_HEAD(&req->r_unsafe_item); | |
a40c4f10 YS |
218 | INIT_LIST_HEAD(&req->r_linger_item); |
219 | INIT_LIST_HEAD(&req->r_linger_osd); | |
935b639a | 220 | INIT_LIST_HEAD(&req->r_req_lru_item); |
f24e9980 SW |
221 | req->r_flags = flags; |
222 | ||
223 | WARN_ON((flags & (CEPH_OSD_FLAG_READ|CEPH_OSD_FLAG_WRITE)) == 0); | |
224 | ||
c16e7869 SW |
225 | /* create reply message */ |
226 | if (use_mempool) | |
227 | msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0); | |
228 | else | |
229 | msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, | |
b61c2763 | 230 | OSD_OPREPLY_FRONT_LEN, gfp_flags, true); |
a79832f2 | 231 | if (!msg) { |
c16e7869 | 232 | ceph_osdc_put_request(req); |
a79832f2 | 233 | return NULL; |
c16e7869 SW |
234 | } |
235 | req->r_reply = msg; | |
236 | ||
68b4476b YS |
237 | /* allocate space for the trailing data */ |
238 | if (needs_trail) { | |
239 | req->r_trail = kmalloc(sizeof(struct ceph_pagelist), gfp_flags); | |
240 | if (!req->r_trail) { | |
241 | ceph_osdc_put_request(req); | |
242 | return NULL; | |
243 | } | |
244 | ceph_pagelist_init(req->r_trail); | |
245 | } | |
c16e7869 | 246 | /* create request message; allow space for oid */ |
224736d9 | 247 | msg_size += MAX_OBJ_NAME_SIZE; |
f24e9980 SW |
248 | if (snapc) |
249 | msg_size += sizeof(u64) * snapc->num_snaps; | |
250 | if (use_mempool) | |
8f3bc053 | 251 | msg = ceph_msgpool_get(&osdc->msgpool_op, 0); |
f24e9980 | 252 | else |
b61c2763 | 253 | msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp_flags, true); |
a79832f2 | 254 | if (!msg) { |
f24e9980 | 255 | ceph_osdc_put_request(req); |
a79832f2 | 256 | return NULL; |
f24e9980 | 257 | } |
68b4476b | 258 | |
f24e9980 SW |
259 | msg->hdr.type = cpu_to_le16(CEPH_MSG_OSD_OP); |
260 | memset(msg->front.iov_base, 0, msg->front.iov_len); | |
3499e8a5 YS |
261 | |
262 | req->r_request = msg; | |
263 | req->r_pages = pages; | |
68b4476b YS |
264 | #ifdef CONFIG_BLOCK |
265 | if (bio) { | |
266 | req->r_bio = bio; | |
267 | bio_get(req->r_bio); | |
268 | } | |
269 | #endif | |
3499e8a5 YS |
270 | |
271 | return req; | |
272 | } | |
3d14c5d2 | 273 | EXPORT_SYMBOL(ceph_osdc_alloc_request); |
3499e8a5 | 274 | |
68b4476b YS |
275 | static void osd_req_encode_op(struct ceph_osd_request *req, |
276 | struct ceph_osd_op *dst, | |
277 | struct ceph_osd_req_op *src) | |
278 | { | |
279 | dst->op = cpu_to_le16(src->op); | |
280 | ||
065a68f9 | 281 | switch (src->op) { |
68b4476b YS |
282 | case CEPH_OSD_OP_READ: |
283 | case CEPH_OSD_OP_WRITE: | |
284 | dst->extent.offset = | |
285 | cpu_to_le64(src->extent.offset); | |
286 | dst->extent.length = | |
287 | cpu_to_le64(src->extent.length); | |
288 | dst->extent.truncate_size = | |
289 | cpu_to_le64(src->extent.truncate_size); | |
290 | dst->extent.truncate_seq = | |
291 | cpu_to_le32(src->extent.truncate_seq); | |
292 | break; | |
293 | ||
294 | case CEPH_OSD_OP_GETXATTR: | |
295 | case CEPH_OSD_OP_SETXATTR: | |
296 | case CEPH_OSD_OP_CMPXATTR: | |
297 | BUG_ON(!req->r_trail); | |
298 | ||
299 | dst->xattr.name_len = cpu_to_le32(src->xattr.name_len); | |
300 | dst->xattr.value_len = cpu_to_le32(src->xattr.value_len); | |
301 | dst->xattr.cmp_op = src->xattr.cmp_op; | |
302 | dst->xattr.cmp_mode = src->xattr.cmp_mode; | |
303 | ceph_pagelist_append(req->r_trail, src->xattr.name, | |
304 | src->xattr.name_len); | |
305 | ceph_pagelist_append(req->r_trail, src->xattr.val, | |
306 | src->xattr.value_len); | |
307 | break; | |
ae1533b6 YS |
308 | case CEPH_OSD_OP_CALL: |
309 | BUG_ON(!req->r_trail); | |
310 | ||
311 | dst->cls.class_len = src->cls.class_len; | |
312 | dst->cls.method_len = src->cls.method_len; | |
313 | dst->cls.indata_len = cpu_to_le32(src->cls.indata_len); | |
314 | ||
315 | ceph_pagelist_append(req->r_trail, src->cls.class_name, | |
316 | src->cls.class_len); | |
317 | ceph_pagelist_append(req->r_trail, src->cls.method_name, | |
318 | src->cls.method_len); | |
319 | ceph_pagelist_append(req->r_trail, src->cls.indata, | |
320 | src->cls.indata_len); | |
321 | break; | |
322 | case CEPH_OSD_OP_ROLLBACK: | |
323 | dst->snap.snapid = cpu_to_le64(src->snap.snapid); | |
324 | break; | |
68b4476b YS |
325 | case CEPH_OSD_OP_STARTSYNC: |
326 | break; | |
a40c4f10 YS |
327 | case CEPH_OSD_OP_NOTIFY: |
328 | { | |
329 | __le32 prot_ver = cpu_to_le32(src->watch.prot_ver); | |
330 | __le32 timeout = cpu_to_le32(src->watch.timeout); | |
331 | ||
332 | BUG_ON(!req->r_trail); | |
333 | ||
334 | ceph_pagelist_append(req->r_trail, | |
335 | &prot_ver, sizeof(prot_ver)); | |
336 | ceph_pagelist_append(req->r_trail, | |
337 | &timeout, sizeof(timeout)); | |
338 | } | |
339 | case CEPH_OSD_OP_NOTIFY_ACK: | |
340 | case CEPH_OSD_OP_WATCH: | |
341 | dst->watch.cookie = cpu_to_le64(src->watch.cookie); | |
342 | dst->watch.ver = cpu_to_le64(src->watch.ver); | |
343 | dst->watch.flag = src->watch.flag; | |
344 | break; | |
68b4476b YS |
345 | default: |
346 | pr_err("unrecognized osd opcode %d\n", dst->op); | |
347 | WARN_ON(1); | |
348 | break; | |
349 | } | |
350 | dst->payload_len = cpu_to_le32(src->payload_len); | |
351 | } | |
352 | ||
3499e8a5 YS |
353 | /* |
354 | * build new request AND message | |
355 | * | |
356 | */ | |
357 | void ceph_osdc_build_request(struct ceph_osd_request *req, | |
68b4476b YS |
358 | u64 off, u64 *plen, |
359 | struct ceph_osd_req_op *src_ops, | |
360 | struct ceph_snap_context *snapc, | |
361 | struct timespec *mtime, | |
362 | const char *oid, | |
363 | int oid_len) | |
3499e8a5 YS |
364 | { |
365 | struct ceph_msg *msg = req->r_request; | |
366 | struct ceph_osd_request_head *head; | |
68b4476b | 367 | struct ceph_osd_req_op *src_op; |
3499e8a5 YS |
368 | struct ceph_osd_op *op; |
369 | void *p; | |
68b4476b | 370 | int num_op = get_num_ops(src_ops, NULL); |
3499e8a5 | 371 | size_t msg_size = sizeof(*head) + num_op*sizeof(*op); |
3499e8a5 | 372 | int flags = req->r_flags; |
68b4476b YS |
373 | u64 data_len = 0; |
374 | int i; | |
3499e8a5 | 375 | |
f24e9980 SW |
376 | head = msg->front.iov_base; |
377 | op = (void *)(head + 1); | |
378 | p = (void *)(op + num_op); | |
379 | ||
f24e9980 SW |
380 | req->r_snapc = ceph_get_snap_context(snapc); |
381 | ||
382 | head->client_inc = cpu_to_le32(1); /* always, for now. */ | |
383 | head->flags = cpu_to_le32(flags); | |
384 | if (flags & CEPH_OSD_FLAG_WRITE) | |
385 | ceph_encode_timespec(&head->mtime, mtime); | |
386 | head->num_ops = cpu_to_le16(num_op); | |
f24e9980 | 387 | |
f24e9980 SW |
388 | |
389 | /* fill in oid */ | |
3499e8a5 YS |
390 | head->object_len = cpu_to_le32(oid_len); |
391 | memcpy(p, oid, oid_len); | |
392 | p += oid_len; | |
f24e9980 | 393 | |
68b4476b YS |
394 | src_op = src_ops; |
395 | while (src_op->op) { | |
396 | osd_req_encode_op(req, op, src_op); | |
397 | src_op++; | |
f24e9980 | 398 | op++; |
f24e9980 | 399 | } |
68b4476b YS |
400 | |
401 | if (req->r_trail) | |
402 | data_len += req->r_trail->length; | |
403 | ||
f24e9980 SW |
404 | if (snapc) { |
405 | head->snap_seq = cpu_to_le64(snapc->seq); | |
406 | head->num_snaps = cpu_to_le32(snapc->num_snaps); | |
407 | for (i = 0; i < snapc->num_snaps; i++) { | |
408 | put_unaligned_le64(snapc->snaps[i], p); | |
409 | p += sizeof(u64); | |
410 | } | |
411 | } | |
412 | ||
68b4476b YS |
413 | if (flags & CEPH_OSD_FLAG_WRITE) { |
414 | req->r_request->hdr.data_off = cpu_to_le16(off); | |
415 | req->r_request->hdr.data_len = cpu_to_le32(*plen + data_len); | |
416 | } else if (data_len) { | |
417 | req->r_request->hdr.data_off = 0; | |
418 | req->r_request->hdr.data_len = cpu_to_le32(data_len); | |
419 | } | |
420 | ||
c5c6b19d SW |
421 | req->r_request->page_alignment = req->r_page_alignment; |
422 | ||
f24e9980 | 423 | BUG_ON(p > msg->front.iov_base + msg->front.iov_len); |
6f863e71 SW |
424 | msg_size = p - msg->front.iov_base; |
425 | msg->front.iov_len = msg_size; | |
426 | msg->hdr.front_len = cpu_to_le32(msg_size); | |
3499e8a5 YS |
427 | return; |
428 | } | |
3d14c5d2 | 429 | EXPORT_SYMBOL(ceph_osdc_build_request); |
3499e8a5 YS |
430 | |
431 | /* | |
432 | * build new request AND message, calculate layout, and adjust file | |
433 | * extent as needed. | |
434 | * | |
435 | * if the file was recently truncated, we include information about its | |
436 | * old and new size so that the object can be updated appropriately. (we | |
437 | * avoid synchronously deleting truncated objects because it's slow.) | |
438 | * | |
439 | * if @do_sync, include a 'startsync' command so that the osd will flush | |
440 | * data quickly. | |
441 | */ | |
442 | struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc, | |
443 | struct ceph_file_layout *layout, | |
444 | struct ceph_vino vino, | |
445 | u64 off, u64 *plen, | |
446 | int opcode, int flags, | |
447 | struct ceph_snap_context *snapc, | |
448 | int do_sync, | |
449 | u32 truncate_seq, | |
450 | u64 truncate_size, | |
451 | struct timespec *mtime, | |
b7495fc2 SW |
452 | bool use_mempool, int num_reply, |
453 | int page_align) | |
3499e8a5 | 454 | { |
68b4476b YS |
455 | struct ceph_osd_req_op ops[3]; |
456 | struct ceph_osd_request *req; | |
457 | ||
458 | ops[0].op = opcode; | |
459 | ops[0].extent.truncate_seq = truncate_seq; | |
460 | ops[0].extent.truncate_size = truncate_size; | |
461 | ops[0].payload_len = 0; | |
462 | ||
463 | if (do_sync) { | |
464 | ops[1].op = CEPH_OSD_OP_STARTSYNC; | |
465 | ops[1].payload_len = 0; | |
466 | ops[2].op = 0; | |
467 | } else | |
468 | ops[1].op = 0; | |
469 | ||
470 | req = ceph_osdc_alloc_request(osdc, flags, | |
471 | snapc, ops, | |
3499e8a5 | 472 | use_mempool, |
68b4476b | 473 | GFP_NOFS, NULL, NULL); |
4ad12621 SW |
474 | if (!req) |
475 | return NULL; | |
3499e8a5 YS |
476 | |
477 | /* calculate max write size */ | |
68b4476b | 478 | calc_layout(osdc, vino, layout, off, plen, req, ops); |
3499e8a5 YS |
479 | req->r_file_layout = *layout; /* keep a copy */ |
480 | ||
9bb0ce2b SW |
481 | /* in case it differs from natural (file) alignment that |
482 | calc_layout filled in for us */ | |
483 | req->r_num_pages = calc_pages_for(page_align, *plen); | |
b7495fc2 SW |
484 | req->r_page_alignment = page_align; |
485 | ||
68b4476b YS |
486 | ceph_osdc_build_request(req, off, plen, ops, |
487 | snapc, | |
3499e8a5 YS |
488 | mtime, |
489 | req->r_oid, req->r_oid_len); | |
490 | ||
f24e9980 SW |
491 | return req; |
492 | } | |
3d14c5d2 | 493 | EXPORT_SYMBOL(ceph_osdc_new_request); |
f24e9980 SW |
494 | |
495 | /* | |
496 | * We keep osd requests in an rbtree, sorted by ->r_tid. | |
497 | */ | |
498 | static void __insert_request(struct ceph_osd_client *osdc, | |
499 | struct ceph_osd_request *new) | |
500 | { | |
501 | struct rb_node **p = &osdc->requests.rb_node; | |
502 | struct rb_node *parent = NULL; | |
503 | struct ceph_osd_request *req = NULL; | |
504 | ||
505 | while (*p) { | |
506 | parent = *p; | |
507 | req = rb_entry(parent, struct ceph_osd_request, r_node); | |
508 | if (new->r_tid < req->r_tid) | |
509 | p = &(*p)->rb_left; | |
510 | else if (new->r_tid > req->r_tid) | |
511 | p = &(*p)->rb_right; | |
512 | else | |
513 | BUG(); | |
514 | } | |
515 | ||
516 | rb_link_node(&new->r_node, parent, p); | |
517 | rb_insert_color(&new->r_node, &osdc->requests); | |
518 | } | |
519 | ||
520 | static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc, | |
521 | u64 tid) | |
522 | { | |
523 | struct ceph_osd_request *req; | |
524 | struct rb_node *n = osdc->requests.rb_node; | |
525 | ||
526 | while (n) { | |
527 | req = rb_entry(n, struct ceph_osd_request, r_node); | |
528 | if (tid < req->r_tid) | |
529 | n = n->rb_left; | |
530 | else if (tid > req->r_tid) | |
531 | n = n->rb_right; | |
532 | else | |
533 | return req; | |
534 | } | |
535 | return NULL; | |
536 | } | |
537 | ||
538 | static struct ceph_osd_request * | |
539 | __lookup_request_ge(struct ceph_osd_client *osdc, | |
540 | u64 tid) | |
541 | { | |
542 | struct ceph_osd_request *req; | |
543 | struct rb_node *n = osdc->requests.rb_node; | |
544 | ||
545 | while (n) { | |
546 | req = rb_entry(n, struct ceph_osd_request, r_node); | |
547 | if (tid < req->r_tid) { | |
548 | if (!n->rb_left) | |
549 | return req; | |
550 | n = n->rb_left; | |
551 | } else if (tid > req->r_tid) { | |
552 | n = n->rb_right; | |
553 | } else { | |
554 | return req; | |
555 | } | |
556 | } | |
557 | return NULL; | |
558 | } | |
559 | ||
6f6c7006 SW |
560 | /* |
561 | * Resubmit requests pending on the given osd. | |
562 | */ | |
563 | static void __kick_osd_requests(struct ceph_osd_client *osdc, | |
564 | struct ceph_osd *osd) | |
565 | { | |
a40c4f10 | 566 | struct ceph_osd_request *req, *nreq; |
6f6c7006 SW |
567 | int err; |
568 | ||
569 | dout("__kick_osd_requests osd%d\n", osd->o_osd); | |
570 | err = __reset_osd(osdc, osd); | |
571 | if (err == -EAGAIN) | |
572 | return; | |
573 | ||
574 | list_for_each_entry(req, &osd->o_requests, r_osd_item) { | |
575 | list_move(&req->r_req_lru_item, &osdc->req_unsent); | |
576 | dout("requeued %p tid %llu osd%d\n", req, req->r_tid, | |
577 | osd->o_osd); | |
a40c4f10 YS |
578 | if (!req->r_linger) |
579 | req->r_flags |= CEPH_OSD_FLAG_RETRY; | |
580 | } | |
581 | ||
582 | list_for_each_entry_safe(req, nreq, &osd->o_linger_requests, | |
583 | r_linger_osd) { | |
77f38e0e SW |
584 | /* |
585 | * reregister request prior to unregistering linger so | |
586 | * that r_osd is preserved. | |
587 | */ | |
588 | BUG_ON(!list_empty(&req->r_req_lru_item)); | |
a40c4f10 | 589 | __register_request(osdc, req); |
77f38e0e SW |
590 | list_add(&req->r_req_lru_item, &osdc->req_unsent); |
591 | list_add(&req->r_osd_item, &req->r_osd->o_requests); | |
592 | __unregister_linger_request(osdc, req); | |
a40c4f10 YS |
593 | dout("requeued lingering %p tid %llu osd%d\n", req, req->r_tid, |
594 | osd->o_osd); | |
6f6c7006 SW |
595 | } |
596 | } | |
597 | ||
598 | static void kick_osd_requests(struct ceph_osd_client *osdc, | |
599 | struct ceph_osd *kickosd) | |
600 | { | |
601 | mutex_lock(&osdc->request_mutex); | |
602 | __kick_osd_requests(osdc, kickosd); | |
603 | mutex_unlock(&osdc->request_mutex); | |
604 | } | |
f24e9980 SW |
605 | |
606 | /* | |
81b024e7 | 607 | * If the osd connection drops, we need to resubmit all requests. |
f24e9980 SW |
608 | */ |
609 | static void osd_reset(struct ceph_connection *con) | |
610 | { | |
611 | struct ceph_osd *osd = con->private; | |
612 | struct ceph_osd_client *osdc; | |
613 | ||
614 | if (!osd) | |
615 | return; | |
616 | dout("osd_reset osd%d\n", osd->o_osd); | |
617 | osdc = osd->o_osdc; | |
f24e9980 | 618 | down_read(&osdc->map_sem); |
6f6c7006 SW |
619 | kick_osd_requests(osdc, osd); |
620 | send_queued(osdc); | |
f24e9980 SW |
621 | up_read(&osdc->map_sem); |
622 | } | |
623 | ||
624 | /* | |
625 | * Track open sessions with osds. | |
626 | */ | |
e10006f8 | 627 | static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum) |
f24e9980 SW |
628 | { |
629 | struct ceph_osd *osd; | |
630 | ||
631 | osd = kzalloc(sizeof(*osd), GFP_NOFS); | |
632 | if (!osd) | |
633 | return NULL; | |
634 | ||
635 | atomic_set(&osd->o_ref, 1); | |
636 | osd->o_osdc = osdc; | |
e10006f8 | 637 | osd->o_osd = onum; |
f24e9980 | 638 | INIT_LIST_HEAD(&osd->o_requests); |
a40c4f10 | 639 | INIT_LIST_HEAD(&osd->o_linger_requests); |
f5a2041b | 640 | INIT_LIST_HEAD(&osd->o_osd_lru); |
f24e9980 SW |
641 | osd->o_incarnation = 1; |
642 | ||
15d9882c | 643 | ceph_con_init(&osdc->client->msgr, &osd->o_con); |
f24e9980 SW |
644 | osd->o_con.private = osd; |
645 | osd->o_con.ops = &osd_con_ops; | |
646 | osd->o_con.peer_name.type = CEPH_ENTITY_TYPE_OSD; | |
e10006f8 | 647 | osd->o_con.peer_name.num = cpu_to_le64(onum); |
4e7a5dcd | 648 | |
422d2cb8 | 649 | INIT_LIST_HEAD(&osd->o_keepalive_item); |
f24e9980 SW |
650 | return osd; |
651 | } | |
652 | ||
653 | static struct ceph_osd *get_osd(struct ceph_osd *osd) | |
654 | { | |
655 | if (atomic_inc_not_zero(&osd->o_ref)) { | |
656 | dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1, | |
657 | atomic_read(&osd->o_ref)); | |
658 | return osd; | |
659 | } else { | |
660 | dout("get_osd %p FAIL\n", osd); | |
661 | return NULL; | |
662 | } | |
663 | } | |
664 | ||
665 | static void put_osd(struct ceph_osd *osd) | |
666 | { | |
667 | dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref), | |
668 | atomic_read(&osd->o_ref) - 1); | |
a255651d | 669 | if (atomic_dec_and_test(&osd->o_ref) && osd->o_auth.authorizer) { |
79494d1b SW |
670 | struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth; |
671 | ||
a255651d | 672 | if (ac->ops && ac->ops->destroy_authorizer) |
6c4a1915 | 673 | ac->ops->destroy_authorizer(ac, osd->o_auth.authorizer); |
f24e9980 | 674 | kfree(osd); |
79494d1b | 675 | } |
f24e9980 SW |
676 | } |
677 | ||
678 | /* | |
679 | * remove an osd from our map | |
680 | */ | |
f5a2041b | 681 | static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd) |
f24e9980 | 682 | { |
f5a2041b | 683 | dout("__remove_osd %p\n", osd); |
f24e9980 SW |
684 | BUG_ON(!list_empty(&osd->o_requests)); |
685 | rb_erase(&osd->o_node, &osdc->osds); | |
f5a2041b | 686 | list_del_init(&osd->o_osd_lru); |
f24e9980 SW |
687 | ceph_con_close(&osd->o_con); |
688 | put_osd(osd); | |
689 | } | |
690 | ||
aca420bc SW |
691 | static void remove_all_osds(struct ceph_osd_client *osdc) |
692 | { | |
693 | dout("__remove_old_osds %p\n", osdc); | |
694 | mutex_lock(&osdc->request_mutex); | |
695 | while (!RB_EMPTY_ROOT(&osdc->osds)) { | |
696 | struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds), | |
697 | struct ceph_osd, o_node); | |
698 | __remove_osd(osdc, osd); | |
699 | } | |
700 | mutex_unlock(&osdc->request_mutex); | |
701 | } | |
702 | ||
f5a2041b YS |
703 | static void __move_osd_to_lru(struct ceph_osd_client *osdc, |
704 | struct ceph_osd *osd) | |
705 | { | |
706 | dout("__move_osd_to_lru %p\n", osd); | |
707 | BUG_ON(!list_empty(&osd->o_osd_lru)); | |
708 | list_add_tail(&osd->o_osd_lru, &osdc->osd_lru); | |
3d14c5d2 | 709 | osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl * HZ; |
f5a2041b YS |
710 | } |
711 | ||
712 | static void __remove_osd_from_lru(struct ceph_osd *osd) | |
713 | { | |
714 | dout("__remove_osd_from_lru %p\n", osd); | |
715 | if (!list_empty(&osd->o_osd_lru)) | |
716 | list_del_init(&osd->o_osd_lru); | |
717 | } | |
718 | ||
aca420bc | 719 | static void remove_old_osds(struct ceph_osd_client *osdc) |
f5a2041b YS |
720 | { |
721 | struct ceph_osd *osd, *nosd; | |
722 | ||
723 | dout("__remove_old_osds %p\n", osdc); | |
724 | mutex_lock(&osdc->request_mutex); | |
725 | list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) { | |
aca420bc | 726 | if (time_before(jiffies, osd->lru_ttl)) |
f5a2041b YS |
727 | break; |
728 | __remove_osd(osdc, osd); | |
729 | } | |
730 | mutex_unlock(&osdc->request_mutex); | |
731 | } | |
732 | ||
f24e9980 SW |
733 | /* |
734 | * reset osd connect | |
735 | */ | |
f5a2041b | 736 | static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd) |
f24e9980 | 737 | { |
87b315a5 | 738 | struct ceph_osd_request *req; |
f24e9980 SW |
739 | int ret = 0; |
740 | ||
f5a2041b | 741 | dout("__reset_osd %p osd%d\n", osd, osd->o_osd); |
a40c4f10 YS |
742 | if (list_empty(&osd->o_requests) && |
743 | list_empty(&osd->o_linger_requests)) { | |
f5a2041b | 744 | __remove_osd(osdc, osd); |
87b315a5 SW |
745 | } else if (memcmp(&osdc->osdmap->osd_addr[osd->o_osd], |
746 | &osd->o_con.peer_addr, | |
747 | sizeof(osd->o_con.peer_addr)) == 0 && | |
748 | !ceph_con_opened(&osd->o_con)) { | |
749 | dout(" osd addr hasn't changed and connection never opened," | |
750 | " letting msgr retry"); | |
751 | /* touch each r_stamp for handle_timeout()'s benfit */ | |
752 | list_for_each_entry(req, &osd->o_requests, r_osd_item) | |
753 | req->r_stamp = jiffies; | |
754 | ret = -EAGAIN; | |
f24e9980 SW |
755 | } else { |
756 | ceph_con_close(&osd->o_con); | |
757 | ceph_con_open(&osd->o_con, &osdc->osdmap->osd_addr[osd->o_osd]); | |
758 | osd->o_incarnation++; | |
759 | } | |
760 | return ret; | |
761 | } | |
762 | ||
763 | static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new) | |
764 | { | |
765 | struct rb_node **p = &osdc->osds.rb_node; | |
766 | struct rb_node *parent = NULL; | |
767 | struct ceph_osd *osd = NULL; | |
768 | ||
aca420bc | 769 | dout("__insert_osd %p osd%d\n", new, new->o_osd); |
f24e9980 SW |
770 | while (*p) { |
771 | parent = *p; | |
772 | osd = rb_entry(parent, struct ceph_osd, o_node); | |
773 | if (new->o_osd < osd->o_osd) | |
774 | p = &(*p)->rb_left; | |
775 | else if (new->o_osd > osd->o_osd) | |
776 | p = &(*p)->rb_right; | |
777 | else | |
778 | BUG(); | |
779 | } | |
780 | ||
781 | rb_link_node(&new->o_node, parent, p); | |
782 | rb_insert_color(&new->o_node, &osdc->osds); | |
783 | } | |
784 | ||
785 | static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o) | |
786 | { | |
787 | struct ceph_osd *osd; | |
788 | struct rb_node *n = osdc->osds.rb_node; | |
789 | ||
790 | while (n) { | |
791 | osd = rb_entry(n, struct ceph_osd, o_node); | |
792 | if (o < osd->o_osd) | |
793 | n = n->rb_left; | |
794 | else if (o > osd->o_osd) | |
795 | n = n->rb_right; | |
796 | else | |
797 | return osd; | |
798 | } | |
799 | return NULL; | |
800 | } | |
801 | ||
422d2cb8 YS |
802 | static void __schedule_osd_timeout(struct ceph_osd_client *osdc) |
803 | { | |
804 | schedule_delayed_work(&osdc->timeout_work, | |
3d14c5d2 | 805 | osdc->client->options->osd_keepalive_timeout * HZ); |
422d2cb8 YS |
806 | } |
807 | ||
808 | static void __cancel_osd_timeout(struct ceph_osd_client *osdc) | |
809 | { | |
810 | cancel_delayed_work(&osdc->timeout_work); | |
811 | } | |
f24e9980 SW |
812 | |
813 | /* | |
814 | * Register request, assign tid. If this is the first request, set up | |
815 | * the timeout event. | |
816 | */ | |
a40c4f10 YS |
817 | static void __register_request(struct ceph_osd_client *osdc, |
818 | struct ceph_osd_request *req) | |
f24e9980 | 819 | { |
f24e9980 | 820 | req->r_tid = ++osdc->last_tid; |
6df058c0 | 821 | req->r_request->hdr.tid = cpu_to_le64(req->r_tid); |
77f38e0e | 822 | dout("__register_request %p tid %lld\n", req, req->r_tid); |
f24e9980 SW |
823 | __insert_request(osdc, req); |
824 | ceph_osdc_get_request(req); | |
825 | osdc->num_requests++; | |
f24e9980 | 826 | if (osdc->num_requests == 1) { |
422d2cb8 YS |
827 | dout(" first request, scheduling timeout\n"); |
828 | __schedule_osd_timeout(osdc); | |
f24e9980 | 829 | } |
a40c4f10 YS |
830 | } |
831 | ||
832 | static void register_request(struct ceph_osd_client *osdc, | |
833 | struct ceph_osd_request *req) | |
834 | { | |
835 | mutex_lock(&osdc->request_mutex); | |
836 | __register_request(osdc, req); | |
f24e9980 SW |
837 | mutex_unlock(&osdc->request_mutex); |
838 | } | |
839 | ||
840 | /* | |
841 | * called under osdc->request_mutex | |
842 | */ | |
843 | static void __unregister_request(struct ceph_osd_client *osdc, | |
844 | struct ceph_osd_request *req) | |
845 | { | |
35f9f8a0 SW |
846 | if (RB_EMPTY_NODE(&req->r_node)) { |
847 | dout("__unregister_request %p tid %lld not registered\n", | |
848 | req, req->r_tid); | |
849 | return; | |
850 | } | |
851 | ||
f24e9980 SW |
852 | dout("__unregister_request %p tid %lld\n", req, req->r_tid); |
853 | rb_erase(&req->r_node, &osdc->requests); | |
854 | osdc->num_requests--; | |
855 | ||
0ba6478d SW |
856 | if (req->r_osd) { |
857 | /* make sure the original request isn't in flight. */ | |
858 | ceph_con_revoke(&req->r_osd->o_con, req->r_request); | |
859 | ||
860 | list_del_init(&req->r_osd_item); | |
a40c4f10 YS |
861 | if (list_empty(&req->r_osd->o_requests) && |
862 | list_empty(&req->r_osd->o_linger_requests)) { | |
863 | dout("moving osd to %p lru\n", req->r_osd); | |
f5a2041b | 864 | __move_osd_to_lru(osdc, req->r_osd); |
a40c4f10 | 865 | } |
fbdb9190 | 866 | if (list_empty(&req->r_linger_item)) |
a40c4f10 | 867 | req->r_osd = NULL; |
0ba6478d | 868 | } |
f24e9980 SW |
869 | |
870 | ceph_osdc_put_request(req); | |
871 | ||
422d2cb8 YS |
872 | list_del_init(&req->r_req_lru_item); |
873 | if (osdc->num_requests == 0) { | |
874 | dout(" no requests, canceling timeout\n"); | |
875 | __cancel_osd_timeout(osdc); | |
f24e9980 SW |
876 | } |
877 | } | |
878 | ||
879 | /* | |
880 | * Cancel a previously queued request message | |
881 | */ | |
882 | static void __cancel_request(struct ceph_osd_request *req) | |
883 | { | |
6bc18876 | 884 | if (req->r_sent && req->r_osd) { |
f24e9980 SW |
885 | ceph_con_revoke(&req->r_osd->o_con, req->r_request); |
886 | req->r_sent = 0; | |
887 | } | |
888 | } | |
889 | ||
a40c4f10 YS |
890 | static void __register_linger_request(struct ceph_osd_client *osdc, |
891 | struct ceph_osd_request *req) | |
892 | { | |
893 | dout("__register_linger_request %p\n", req); | |
894 | list_add_tail(&req->r_linger_item, &osdc->req_linger); | |
895 | list_add_tail(&req->r_linger_osd, &req->r_osd->o_linger_requests); | |
896 | } | |
897 | ||
898 | static void __unregister_linger_request(struct ceph_osd_client *osdc, | |
899 | struct ceph_osd_request *req) | |
900 | { | |
901 | dout("__unregister_linger_request %p\n", req); | |
902 | if (req->r_osd) { | |
903 | list_del_init(&req->r_linger_item); | |
904 | list_del_init(&req->r_linger_osd); | |
905 | ||
906 | if (list_empty(&req->r_osd->o_requests) && | |
907 | list_empty(&req->r_osd->o_linger_requests)) { | |
908 | dout("moving osd to %p lru\n", req->r_osd); | |
909 | __move_osd_to_lru(osdc, req->r_osd); | |
910 | } | |
fbdb9190 SW |
911 | if (list_empty(&req->r_osd_item)) |
912 | req->r_osd = NULL; | |
a40c4f10 YS |
913 | } |
914 | } | |
915 | ||
916 | void ceph_osdc_unregister_linger_request(struct ceph_osd_client *osdc, | |
917 | struct ceph_osd_request *req) | |
918 | { | |
919 | mutex_lock(&osdc->request_mutex); | |
920 | if (req->r_linger) { | |
921 | __unregister_linger_request(osdc, req); | |
922 | ceph_osdc_put_request(req); | |
923 | } | |
924 | mutex_unlock(&osdc->request_mutex); | |
925 | } | |
926 | EXPORT_SYMBOL(ceph_osdc_unregister_linger_request); | |
927 | ||
928 | void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc, | |
929 | struct ceph_osd_request *req) | |
930 | { | |
931 | if (!req->r_linger) { | |
932 | dout("set_request_linger %p\n", req); | |
933 | req->r_linger = 1; | |
934 | /* | |
935 | * caller is now responsible for calling | |
936 | * unregister_linger_request | |
937 | */ | |
938 | ceph_osdc_get_request(req); | |
939 | } | |
940 | } | |
941 | EXPORT_SYMBOL(ceph_osdc_set_request_linger); | |
942 | ||
f24e9980 SW |
943 | /* |
944 | * Pick an osd (the first 'up' osd in the pg), allocate the osd struct | |
945 | * (as needed), and set the request r_osd appropriately. If there is | |
25985edc | 946 | * no up osd, set r_osd to NULL. Move the request to the appropriate list |
6f6c7006 | 947 | * (unsent, homeless) or leave on in-flight lru. |
f24e9980 SW |
948 | * |
949 | * Return 0 if unchanged, 1 if changed, or negative on error. | |
950 | * | |
951 | * Caller should hold map_sem for read and request_mutex. | |
952 | */ | |
6f6c7006 | 953 | static int __map_request(struct ceph_osd_client *osdc, |
38d6453c | 954 | struct ceph_osd_request *req, int force_resend) |
f24e9980 SW |
955 | { |
956 | struct ceph_osd_request_head *reqhead = req->r_request->front.iov_base; | |
51042122 | 957 | struct ceph_pg pgid; |
d85b7056 SW |
958 | int acting[CEPH_PG_MAX_SIZE]; |
959 | int o = -1, num = 0; | |
f24e9980 | 960 | int err; |
f24e9980 | 961 | |
6f6c7006 | 962 | dout("map_request %p tid %lld\n", req, req->r_tid); |
f24e9980 SW |
963 | err = ceph_calc_object_layout(&reqhead->layout, req->r_oid, |
964 | &req->r_file_layout, osdc->osdmap); | |
6f6c7006 SW |
965 | if (err) { |
966 | list_move(&req->r_req_lru_item, &osdc->req_notarget); | |
f24e9980 | 967 | return err; |
6f6c7006 | 968 | } |
51042122 | 969 | pgid = reqhead->layout.ol_pgid; |
7740a42f SW |
970 | req->r_pgid = pgid; |
971 | ||
d85b7056 SW |
972 | err = ceph_calc_pg_acting(osdc->osdmap, pgid, acting); |
973 | if (err > 0) { | |
974 | o = acting[0]; | |
975 | num = err; | |
976 | } | |
f24e9980 | 977 | |
38d6453c SW |
978 | if ((!force_resend && |
979 | req->r_osd && req->r_osd->o_osd == o && | |
d85b7056 SW |
980 | req->r_sent >= req->r_osd->o_incarnation && |
981 | req->r_num_pg_osds == num && | |
982 | memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) || | |
f24e9980 SW |
983 | (req->r_osd == NULL && o == -1)) |
984 | return 0; /* no change */ | |
985 | ||
6f6c7006 | 986 | dout("map_request tid %llu pgid %d.%x osd%d (was osd%d)\n", |
51042122 | 987 | req->r_tid, le32_to_cpu(pgid.pool), le16_to_cpu(pgid.ps), o, |
f24e9980 SW |
988 | req->r_osd ? req->r_osd->o_osd : -1); |
989 | ||
d85b7056 SW |
990 | /* record full pg acting set */ |
991 | memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num); | |
992 | req->r_num_pg_osds = num; | |
993 | ||
f24e9980 SW |
994 | if (req->r_osd) { |
995 | __cancel_request(req); | |
996 | list_del_init(&req->r_osd_item); | |
f24e9980 SW |
997 | req->r_osd = NULL; |
998 | } | |
999 | ||
1000 | req->r_osd = __lookup_osd(osdc, o); | |
1001 | if (!req->r_osd && o >= 0) { | |
c99eb1c7 | 1002 | err = -ENOMEM; |
e10006f8 | 1003 | req->r_osd = create_osd(osdc, o); |
6f6c7006 SW |
1004 | if (!req->r_osd) { |
1005 | list_move(&req->r_req_lru_item, &osdc->req_notarget); | |
c99eb1c7 | 1006 | goto out; |
6f6c7006 | 1007 | } |
f24e9980 | 1008 | |
6f6c7006 | 1009 | dout("map_request osd %p is osd%d\n", req->r_osd, o); |
f24e9980 SW |
1010 | __insert_osd(osdc, req->r_osd); |
1011 | ||
1012 | ceph_con_open(&req->r_osd->o_con, &osdc->osdmap->osd_addr[o]); | |
1013 | } | |
1014 | ||
f5a2041b YS |
1015 | if (req->r_osd) { |
1016 | __remove_osd_from_lru(req->r_osd); | |
f24e9980 | 1017 | list_add(&req->r_osd_item, &req->r_osd->o_requests); |
6f6c7006 SW |
1018 | list_move(&req->r_req_lru_item, &osdc->req_unsent); |
1019 | } else { | |
1020 | list_move(&req->r_req_lru_item, &osdc->req_notarget); | |
f5a2041b | 1021 | } |
d85b7056 | 1022 | err = 1; /* osd or pg changed */ |
f24e9980 SW |
1023 | |
1024 | out: | |
f24e9980 SW |
1025 | return err; |
1026 | } | |
1027 | ||
1028 | /* | |
1029 | * caller should hold map_sem (for read) and request_mutex | |
1030 | */ | |
56e925b6 SW |
1031 | static void __send_request(struct ceph_osd_client *osdc, |
1032 | struct ceph_osd_request *req) | |
f24e9980 SW |
1033 | { |
1034 | struct ceph_osd_request_head *reqhead; | |
f24e9980 SW |
1035 | |
1036 | dout("send_request %p tid %llu to osd%d flags %d\n", | |
1037 | req, req->r_tid, req->r_osd->o_osd, req->r_flags); | |
1038 | ||
1039 | reqhead = req->r_request->front.iov_base; | |
1040 | reqhead->osdmap_epoch = cpu_to_le32(osdc->osdmap->epoch); | |
1041 | reqhead->flags |= cpu_to_le32(req->r_flags); /* e.g., RETRY */ | |
1042 | reqhead->reassert_version = req->r_reassert_version; | |
1043 | ||
3dd72fc0 | 1044 | req->r_stamp = jiffies; |
07a27e22 | 1045 | list_move_tail(&req->r_req_lru_item, &osdc->req_lru); |
f24e9980 SW |
1046 | |
1047 | ceph_msg_get(req->r_request); /* send consumes a ref */ | |
1048 | ceph_con_send(&req->r_osd->o_con, req->r_request); | |
1049 | req->r_sent = req->r_osd->o_incarnation; | |
f24e9980 SW |
1050 | } |
1051 | ||
6f6c7006 SW |
1052 | /* |
1053 | * Send any requests in the queue (req_unsent). | |
1054 | */ | |
1055 | static void send_queued(struct ceph_osd_client *osdc) | |
1056 | { | |
1057 | struct ceph_osd_request *req, *tmp; | |
1058 | ||
1059 | dout("send_queued\n"); | |
1060 | mutex_lock(&osdc->request_mutex); | |
1061 | list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item) { | |
1062 | __send_request(osdc, req); | |
1063 | } | |
1064 | mutex_unlock(&osdc->request_mutex); | |
1065 | } | |
1066 | ||
f24e9980 SW |
1067 | /* |
1068 | * Timeout callback, called every N seconds when 1 or more osd | |
1069 | * requests has been active for more than N seconds. When this | |
1070 | * happens, we ping all OSDs with requests who have timed out to | |
1071 | * ensure any communications channel reset is detected. Reset the | |
1072 | * request timeouts another N seconds in the future as we go. | |
1073 | * Reschedule the timeout event another N seconds in future (unless | |
1074 | * there are no open requests). | |
1075 | */ | |
1076 | static void handle_timeout(struct work_struct *work) | |
1077 | { | |
1078 | struct ceph_osd_client *osdc = | |
1079 | container_of(work, struct ceph_osd_client, timeout_work.work); | |
422d2cb8 | 1080 | struct ceph_osd_request *req, *last_req = NULL; |
f24e9980 | 1081 | struct ceph_osd *osd; |
3d14c5d2 | 1082 | unsigned long timeout = osdc->client->options->osd_timeout * HZ; |
422d2cb8 | 1083 | unsigned long keepalive = |
3d14c5d2 | 1084 | osdc->client->options->osd_keepalive_timeout * HZ; |
3dd72fc0 | 1085 | unsigned long last_stamp = 0; |
422d2cb8 | 1086 | struct list_head slow_osds; |
f24e9980 SW |
1087 | dout("timeout\n"); |
1088 | down_read(&osdc->map_sem); | |
1089 | ||
1090 | ceph_monc_request_next_osdmap(&osdc->client->monc); | |
1091 | ||
1092 | mutex_lock(&osdc->request_mutex); | |
f24e9980 | 1093 | |
422d2cb8 YS |
1094 | /* |
1095 | * reset osds that appear to be _really_ unresponsive. this | |
1096 | * is a failsafe measure.. we really shouldn't be getting to | |
1097 | * this point if the system is working properly. the monitors | |
1098 | * should mark the osd as failed and we should find out about | |
1099 | * it from an updated osd map. | |
1100 | */ | |
f26e681d | 1101 | while (timeout && !list_empty(&osdc->req_lru)) { |
422d2cb8 YS |
1102 | req = list_entry(osdc->req_lru.next, struct ceph_osd_request, |
1103 | r_req_lru_item); | |
1104 | ||
4cf9d544 | 1105 | /* hasn't been long enough since we sent it? */ |
3dd72fc0 | 1106 | if (time_before(jiffies, req->r_stamp + timeout)) |
422d2cb8 | 1107 | break; |
4cf9d544 SW |
1108 | |
1109 | /* hasn't been long enough since it was acked? */ | |
1110 | if (req->r_request->ack_stamp == 0 || | |
1111 | time_before(jiffies, req->r_request->ack_stamp + timeout)) | |
1112 | break; | |
422d2cb8 | 1113 | |
3dd72fc0 | 1114 | BUG_ON(req == last_req && req->r_stamp == last_stamp); |
422d2cb8 | 1115 | last_req = req; |
3dd72fc0 | 1116 | last_stamp = req->r_stamp; |
422d2cb8 YS |
1117 | |
1118 | osd = req->r_osd; | |
1119 | BUG_ON(!osd); | |
1120 | pr_warning(" tid %llu timed out on osd%d, will reset osd\n", | |
1121 | req->r_tid, osd->o_osd); | |
6f6c7006 | 1122 | __kick_osd_requests(osdc, osd); |
422d2cb8 YS |
1123 | } |
1124 | ||
1125 | /* | |
1126 | * ping osds that are a bit slow. this ensures that if there | |
1127 | * is a break in the TCP connection we will notice, and reopen | |
1128 | * a connection with that osd (from the fault callback). | |
1129 | */ | |
1130 | INIT_LIST_HEAD(&slow_osds); | |
1131 | list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) { | |
3dd72fc0 | 1132 | if (time_before(jiffies, req->r_stamp + keepalive)) |
422d2cb8 YS |
1133 | break; |
1134 | ||
1135 | osd = req->r_osd; | |
1136 | BUG_ON(!osd); | |
1137 | dout(" tid %llu is slow, will send keepalive on osd%d\n", | |
f24e9980 | 1138 | req->r_tid, osd->o_osd); |
422d2cb8 YS |
1139 | list_move_tail(&osd->o_keepalive_item, &slow_osds); |
1140 | } | |
1141 | while (!list_empty(&slow_osds)) { | |
1142 | osd = list_entry(slow_osds.next, struct ceph_osd, | |
1143 | o_keepalive_item); | |
1144 | list_del_init(&osd->o_keepalive_item); | |
f24e9980 SW |
1145 | ceph_con_keepalive(&osd->o_con); |
1146 | } | |
1147 | ||
422d2cb8 | 1148 | __schedule_osd_timeout(osdc); |
f24e9980 | 1149 | mutex_unlock(&osdc->request_mutex); |
6f6c7006 | 1150 | send_queued(osdc); |
f24e9980 SW |
1151 | up_read(&osdc->map_sem); |
1152 | } | |
1153 | ||
f5a2041b YS |
1154 | static void handle_osds_timeout(struct work_struct *work) |
1155 | { | |
1156 | struct ceph_osd_client *osdc = | |
1157 | container_of(work, struct ceph_osd_client, | |
1158 | osds_timeout_work.work); | |
1159 | unsigned long delay = | |
3d14c5d2 | 1160 | osdc->client->options->osd_idle_ttl * HZ >> 2; |
f5a2041b YS |
1161 | |
1162 | dout("osds timeout\n"); | |
1163 | down_read(&osdc->map_sem); | |
aca420bc | 1164 | remove_old_osds(osdc); |
f5a2041b YS |
1165 | up_read(&osdc->map_sem); |
1166 | ||
1167 | schedule_delayed_work(&osdc->osds_timeout_work, | |
1168 | round_jiffies_relative(delay)); | |
1169 | } | |
1170 | ||
25845472 SW |
1171 | static void complete_request(struct ceph_osd_request *req) |
1172 | { | |
1173 | if (req->r_safe_callback) | |
1174 | req->r_safe_callback(req, NULL); | |
1175 | complete_all(&req->r_safe_completion); /* fsync waiter */ | |
1176 | } | |
1177 | ||
f24e9980 SW |
1178 | /* |
1179 | * handle osd op reply. either call the callback if it is specified, | |
1180 | * or do the completion to wake up the waiting thread. | |
1181 | */ | |
350b1c32 SW |
1182 | static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg, |
1183 | struct ceph_connection *con) | |
f24e9980 SW |
1184 | { |
1185 | struct ceph_osd_reply_head *rhead = msg->front.iov_base; | |
1186 | struct ceph_osd_request *req; | |
1187 | u64 tid; | |
1188 | int numops, object_len, flags; | |
0ceed5db | 1189 | s32 result; |
f24e9980 | 1190 | |
6df058c0 | 1191 | tid = le64_to_cpu(msg->hdr.tid); |
f24e9980 SW |
1192 | if (msg->front.iov_len < sizeof(*rhead)) |
1193 | goto bad; | |
f24e9980 SW |
1194 | numops = le32_to_cpu(rhead->num_ops); |
1195 | object_len = le32_to_cpu(rhead->object_len); | |
0ceed5db | 1196 | result = le32_to_cpu(rhead->result); |
f24e9980 SW |
1197 | if (msg->front.iov_len != sizeof(*rhead) + object_len + |
1198 | numops * sizeof(struct ceph_osd_op)) | |
1199 | goto bad; | |
0ceed5db | 1200 | dout("handle_reply %p tid %llu result %d\n", msg, tid, (int)result); |
f24e9980 SW |
1201 | /* lookup */ |
1202 | mutex_lock(&osdc->request_mutex); | |
1203 | req = __lookup_request(osdc, tid); | |
1204 | if (req == NULL) { | |
1205 | dout("handle_reply tid %llu dne\n", tid); | |
1206 | mutex_unlock(&osdc->request_mutex); | |
1207 | return; | |
1208 | } | |
1209 | ceph_osdc_get_request(req); | |
1210 | flags = le32_to_cpu(rhead->flags); | |
1211 | ||
350b1c32 | 1212 | /* |
0d59ab81 | 1213 | * if this connection filled our message, drop our reference now, to |
350b1c32 SW |
1214 | * avoid a (safe but slower) revoke later. |
1215 | */ | |
0d59ab81 | 1216 | if (req->r_con_filling_msg == con && req->r_reply == msg) { |
c16e7869 | 1217 | dout(" dropping con_filling_msg ref %p\n", con); |
0d59ab81 | 1218 | req->r_con_filling_msg = NULL; |
350b1c32 SW |
1219 | ceph_con_put(con); |
1220 | } | |
1221 | ||
f24e9980 SW |
1222 | if (!req->r_got_reply) { |
1223 | unsigned bytes; | |
1224 | ||
1225 | req->r_result = le32_to_cpu(rhead->result); | |
1226 | bytes = le32_to_cpu(msg->hdr.data_len); | |
1227 | dout("handle_reply result %d bytes %d\n", req->r_result, | |
1228 | bytes); | |
1229 | if (req->r_result == 0) | |
1230 | req->r_result = bytes; | |
1231 | ||
1232 | /* in case this is a write and we need to replay, */ | |
1233 | req->r_reassert_version = rhead->reassert_version; | |
1234 | ||
1235 | req->r_got_reply = 1; | |
1236 | } else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) { | |
1237 | dout("handle_reply tid %llu dup ack\n", tid); | |
34b43a56 | 1238 | mutex_unlock(&osdc->request_mutex); |
f24e9980 SW |
1239 | goto done; |
1240 | } | |
1241 | ||
1242 | dout("handle_reply tid %llu flags %d\n", tid, flags); | |
1243 | ||
a40c4f10 YS |
1244 | if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK)) |
1245 | __register_linger_request(osdc, req); | |
1246 | ||
f24e9980 | 1247 | /* either this is a read, or we got the safe response */ |
0ceed5db SW |
1248 | if (result < 0 || |
1249 | (flags & CEPH_OSD_FLAG_ONDISK) || | |
f24e9980 SW |
1250 | ((flags & CEPH_OSD_FLAG_WRITE) == 0)) |
1251 | __unregister_request(osdc, req); | |
1252 | ||
1253 | mutex_unlock(&osdc->request_mutex); | |
1254 | ||
1255 | if (req->r_callback) | |
1256 | req->r_callback(req, msg); | |
1257 | else | |
03066f23 | 1258 | complete_all(&req->r_completion); |
f24e9980 | 1259 | |
25845472 SW |
1260 | if (flags & CEPH_OSD_FLAG_ONDISK) |
1261 | complete_request(req); | |
f24e9980 SW |
1262 | |
1263 | done: | |
a40c4f10 | 1264 | dout("req=%p req->r_linger=%d\n", req, req->r_linger); |
f24e9980 SW |
1265 | ceph_osdc_put_request(req); |
1266 | return; | |
1267 | ||
1268 | bad: | |
1269 | pr_err("corrupt osd_op_reply got %d %d expected %d\n", | |
1270 | (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len), | |
1271 | (int)sizeof(*rhead)); | |
9ec7cab1 | 1272 | ceph_msg_dump(msg); |
f24e9980 SW |
1273 | } |
1274 | ||
6f6c7006 | 1275 | static void reset_changed_osds(struct ceph_osd_client *osdc) |
f24e9980 | 1276 | { |
f24e9980 | 1277 | struct rb_node *p, *n; |
f24e9980 | 1278 | |
6f6c7006 SW |
1279 | for (p = rb_first(&osdc->osds); p; p = n) { |
1280 | struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node); | |
f24e9980 | 1281 | |
6f6c7006 SW |
1282 | n = rb_next(p); |
1283 | if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) || | |
1284 | memcmp(&osd->o_con.peer_addr, | |
1285 | ceph_osd_addr(osdc->osdmap, | |
1286 | osd->o_osd), | |
1287 | sizeof(struct ceph_entity_addr)) != 0) | |
1288 | __reset_osd(osdc, osd); | |
f24e9980 | 1289 | } |
422d2cb8 YS |
1290 | } |
1291 | ||
1292 | /* | |
6f6c7006 SW |
1293 | * Requeue requests whose mapping to an OSD has changed. If requests map to |
1294 | * no osd, request a new map. | |
422d2cb8 YS |
1295 | * |
1296 | * Caller should hold map_sem for read and request_mutex. | |
1297 | */ | |
38d6453c | 1298 | static void kick_requests(struct ceph_osd_client *osdc, int force_resend) |
422d2cb8 | 1299 | { |
a40c4f10 | 1300 | struct ceph_osd_request *req, *nreq; |
6f6c7006 SW |
1301 | struct rb_node *p; |
1302 | int needmap = 0; | |
1303 | int err; | |
422d2cb8 | 1304 | |
38d6453c | 1305 | dout("kick_requests %s\n", force_resend ? " (force resend)" : ""); |
422d2cb8 | 1306 | mutex_lock(&osdc->request_mutex); |
6f6c7006 SW |
1307 | for (p = rb_first(&osdc->requests); p; p = rb_next(p)) { |
1308 | req = rb_entry(p, struct ceph_osd_request, r_node); | |
38d6453c | 1309 | err = __map_request(osdc, req, force_resend); |
6f6c7006 SW |
1310 | if (err < 0) |
1311 | continue; /* error */ | |
1312 | if (req->r_osd == NULL) { | |
1313 | dout("%p tid %llu maps to no osd\n", req, req->r_tid); | |
1314 | needmap++; /* request a newer map */ | |
1315 | } else if (err > 0) { | |
1316 | dout("%p tid %llu requeued on osd%d\n", req, req->r_tid, | |
1317 | req->r_osd ? req->r_osd->o_osd : -1); | |
a40c4f10 YS |
1318 | if (!req->r_linger) |
1319 | req->r_flags |= CEPH_OSD_FLAG_RETRY; | |
1320 | } | |
1321 | } | |
1322 | ||
1323 | list_for_each_entry_safe(req, nreq, &osdc->req_linger, | |
1324 | r_linger_item) { | |
1325 | dout("linger req=%p req->r_osd=%p\n", req, req->r_osd); | |
1326 | ||
38d6453c | 1327 | err = __map_request(osdc, req, force_resend); |
a40c4f10 YS |
1328 | if (err == 0) |
1329 | continue; /* no change and no osd was specified */ | |
1330 | if (err < 0) | |
1331 | continue; /* hrm! */ | |
1332 | if (req->r_osd == NULL) { | |
1333 | dout("tid %llu maps to no valid osd\n", req->r_tid); | |
1334 | needmap++; /* request a newer map */ | |
1335 | continue; | |
6f6c7006 | 1336 | } |
a40c4f10 YS |
1337 | |
1338 | dout("kicking lingering %p tid %llu osd%d\n", req, req->r_tid, | |
1339 | req->r_osd ? req->r_osd->o_osd : -1); | |
1340 | __unregister_linger_request(osdc, req); | |
1341 | __register_request(osdc, req); | |
6f6c7006 | 1342 | } |
f24e9980 SW |
1343 | mutex_unlock(&osdc->request_mutex); |
1344 | ||
1345 | if (needmap) { | |
1346 | dout("%d requests for down osds, need new map\n", needmap); | |
1347 | ceph_monc_request_next_osdmap(&osdc->client->monc); | |
1348 | } | |
422d2cb8 | 1349 | } |
6f6c7006 SW |
1350 | |
1351 | ||
f24e9980 SW |
1352 | /* |
1353 | * Process updated osd map. | |
1354 | * | |
1355 | * The message contains any number of incremental and full maps, normally | |
1356 | * indicating some sort of topology change in the cluster. Kick requests | |
1357 | * off to different OSDs as needed. | |
1358 | */ | |
1359 | void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg) | |
1360 | { | |
1361 | void *p, *end, *next; | |
1362 | u32 nr_maps, maplen; | |
1363 | u32 epoch; | |
1364 | struct ceph_osdmap *newmap = NULL, *oldmap; | |
1365 | int err; | |
1366 | struct ceph_fsid fsid; | |
1367 | ||
1368 | dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0); | |
1369 | p = msg->front.iov_base; | |
1370 | end = p + msg->front.iov_len; | |
1371 | ||
1372 | /* verify fsid */ | |
1373 | ceph_decode_need(&p, end, sizeof(fsid), bad); | |
1374 | ceph_decode_copy(&p, &fsid, sizeof(fsid)); | |
0743304d SW |
1375 | if (ceph_check_fsid(osdc->client, &fsid) < 0) |
1376 | return; | |
f24e9980 SW |
1377 | |
1378 | down_write(&osdc->map_sem); | |
1379 | ||
1380 | /* incremental maps */ | |
1381 | ceph_decode_32_safe(&p, end, nr_maps, bad); | |
1382 | dout(" %d inc maps\n", nr_maps); | |
1383 | while (nr_maps > 0) { | |
1384 | ceph_decode_need(&p, end, 2*sizeof(u32), bad); | |
c89136ea SW |
1385 | epoch = ceph_decode_32(&p); |
1386 | maplen = ceph_decode_32(&p); | |
f24e9980 SW |
1387 | ceph_decode_need(&p, end, maplen, bad); |
1388 | next = p + maplen; | |
1389 | if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) { | |
1390 | dout("applying incremental map %u len %d\n", | |
1391 | epoch, maplen); | |
1392 | newmap = osdmap_apply_incremental(&p, next, | |
1393 | osdc->osdmap, | |
15d9882c | 1394 | &osdc->client->msgr); |
f24e9980 SW |
1395 | if (IS_ERR(newmap)) { |
1396 | err = PTR_ERR(newmap); | |
1397 | goto bad; | |
1398 | } | |
30dc6381 | 1399 | BUG_ON(!newmap); |
f24e9980 SW |
1400 | if (newmap != osdc->osdmap) { |
1401 | ceph_osdmap_destroy(osdc->osdmap); | |
1402 | osdc->osdmap = newmap; | |
1403 | } | |
38d6453c | 1404 | kick_requests(osdc, 0); |
6f6c7006 | 1405 | reset_changed_osds(osdc); |
f24e9980 SW |
1406 | } else { |
1407 | dout("ignoring incremental map %u len %d\n", | |
1408 | epoch, maplen); | |
1409 | } | |
1410 | p = next; | |
1411 | nr_maps--; | |
1412 | } | |
1413 | if (newmap) | |
1414 | goto done; | |
1415 | ||
1416 | /* full maps */ | |
1417 | ceph_decode_32_safe(&p, end, nr_maps, bad); | |
1418 | dout(" %d full maps\n", nr_maps); | |
1419 | while (nr_maps) { | |
1420 | ceph_decode_need(&p, end, 2*sizeof(u32), bad); | |
c89136ea SW |
1421 | epoch = ceph_decode_32(&p); |
1422 | maplen = ceph_decode_32(&p); | |
f24e9980 SW |
1423 | ceph_decode_need(&p, end, maplen, bad); |
1424 | if (nr_maps > 1) { | |
1425 | dout("skipping non-latest full map %u len %d\n", | |
1426 | epoch, maplen); | |
1427 | } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) { | |
1428 | dout("skipping full map %u len %d, " | |
1429 | "older than our %u\n", epoch, maplen, | |
1430 | osdc->osdmap->epoch); | |
1431 | } else { | |
38d6453c SW |
1432 | int skipped_map = 0; |
1433 | ||
f24e9980 SW |
1434 | dout("taking full map %u len %d\n", epoch, maplen); |
1435 | newmap = osdmap_decode(&p, p+maplen); | |
1436 | if (IS_ERR(newmap)) { | |
1437 | err = PTR_ERR(newmap); | |
1438 | goto bad; | |
1439 | } | |
30dc6381 | 1440 | BUG_ON(!newmap); |
f24e9980 SW |
1441 | oldmap = osdc->osdmap; |
1442 | osdc->osdmap = newmap; | |
38d6453c SW |
1443 | if (oldmap) { |
1444 | if (oldmap->epoch + 1 < newmap->epoch) | |
1445 | skipped_map = 1; | |
f24e9980 | 1446 | ceph_osdmap_destroy(oldmap); |
38d6453c SW |
1447 | } |
1448 | kick_requests(osdc, skipped_map); | |
f24e9980 SW |
1449 | } |
1450 | p += maplen; | |
1451 | nr_maps--; | |
1452 | } | |
1453 | ||
1454 | done: | |
1455 | downgrade_write(&osdc->map_sem); | |
1456 | ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch); | |
cd634fb6 SW |
1457 | |
1458 | /* | |
1459 | * subscribe to subsequent osdmap updates if full to ensure | |
1460 | * we find out when we are no longer full and stop returning | |
1461 | * ENOSPC. | |
1462 | */ | |
1463 | if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL)) | |
1464 | ceph_monc_request_next_osdmap(&osdc->client->monc); | |
1465 | ||
6f6c7006 | 1466 | send_queued(osdc); |
f24e9980 | 1467 | up_read(&osdc->map_sem); |
03066f23 | 1468 | wake_up_all(&osdc->client->auth_wq); |
f24e9980 SW |
1469 | return; |
1470 | ||
1471 | bad: | |
1472 | pr_err("osdc handle_map corrupt msg\n"); | |
9ec7cab1 | 1473 | ceph_msg_dump(msg); |
f24e9980 SW |
1474 | up_write(&osdc->map_sem); |
1475 | return; | |
1476 | } | |
1477 | ||
a40c4f10 YS |
1478 | /* |
1479 | * watch/notify callback event infrastructure | |
1480 | * | |
1481 | * These callbacks are used both for watch and notify operations. | |
1482 | */ | |
1483 | static void __release_event(struct kref *kref) | |
1484 | { | |
1485 | struct ceph_osd_event *event = | |
1486 | container_of(kref, struct ceph_osd_event, kref); | |
1487 | ||
1488 | dout("__release_event %p\n", event); | |
1489 | kfree(event); | |
1490 | } | |
1491 | ||
1492 | static void get_event(struct ceph_osd_event *event) | |
1493 | { | |
1494 | kref_get(&event->kref); | |
1495 | } | |
1496 | ||
1497 | void ceph_osdc_put_event(struct ceph_osd_event *event) | |
1498 | { | |
1499 | kref_put(&event->kref, __release_event); | |
1500 | } | |
1501 | EXPORT_SYMBOL(ceph_osdc_put_event); | |
1502 | ||
1503 | static void __insert_event(struct ceph_osd_client *osdc, | |
1504 | struct ceph_osd_event *new) | |
1505 | { | |
1506 | struct rb_node **p = &osdc->event_tree.rb_node; | |
1507 | struct rb_node *parent = NULL; | |
1508 | struct ceph_osd_event *event = NULL; | |
1509 | ||
1510 | while (*p) { | |
1511 | parent = *p; | |
1512 | event = rb_entry(parent, struct ceph_osd_event, node); | |
1513 | if (new->cookie < event->cookie) | |
1514 | p = &(*p)->rb_left; | |
1515 | else if (new->cookie > event->cookie) | |
1516 | p = &(*p)->rb_right; | |
1517 | else | |
1518 | BUG(); | |
1519 | } | |
1520 | ||
1521 | rb_link_node(&new->node, parent, p); | |
1522 | rb_insert_color(&new->node, &osdc->event_tree); | |
1523 | } | |
1524 | ||
1525 | static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc, | |
1526 | u64 cookie) | |
1527 | { | |
1528 | struct rb_node **p = &osdc->event_tree.rb_node; | |
1529 | struct rb_node *parent = NULL; | |
1530 | struct ceph_osd_event *event = NULL; | |
1531 | ||
1532 | while (*p) { | |
1533 | parent = *p; | |
1534 | event = rb_entry(parent, struct ceph_osd_event, node); | |
1535 | if (cookie < event->cookie) | |
1536 | p = &(*p)->rb_left; | |
1537 | else if (cookie > event->cookie) | |
1538 | p = &(*p)->rb_right; | |
1539 | else | |
1540 | return event; | |
1541 | } | |
1542 | return NULL; | |
1543 | } | |
1544 | ||
1545 | static void __remove_event(struct ceph_osd_event *event) | |
1546 | { | |
1547 | struct ceph_osd_client *osdc = event->osdc; | |
1548 | ||
1549 | if (!RB_EMPTY_NODE(&event->node)) { | |
1550 | dout("__remove_event removed %p\n", event); | |
1551 | rb_erase(&event->node, &osdc->event_tree); | |
1552 | ceph_osdc_put_event(event); | |
1553 | } else { | |
1554 | dout("__remove_event didn't remove %p\n", event); | |
1555 | } | |
1556 | } | |
1557 | ||
1558 | int ceph_osdc_create_event(struct ceph_osd_client *osdc, | |
1559 | void (*event_cb)(u64, u64, u8, void *), | |
1560 | int one_shot, void *data, | |
1561 | struct ceph_osd_event **pevent) | |
1562 | { | |
1563 | struct ceph_osd_event *event; | |
1564 | ||
1565 | event = kmalloc(sizeof(*event), GFP_NOIO); | |
1566 | if (!event) | |
1567 | return -ENOMEM; | |
1568 | ||
1569 | dout("create_event %p\n", event); | |
1570 | event->cb = event_cb; | |
1571 | event->one_shot = one_shot; | |
1572 | event->data = data; | |
1573 | event->osdc = osdc; | |
1574 | INIT_LIST_HEAD(&event->osd_node); | |
1575 | kref_init(&event->kref); /* one ref for us */ | |
1576 | kref_get(&event->kref); /* one ref for the caller */ | |
1577 | init_completion(&event->completion); | |
1578 | ||
1579 | spin_lock(&osdc->event_lock); | |
1580 | event->cookie = ++osdc->event_count; | |
1581 | __insert_event(osdc, event); | |
1582 | spin_unlock(&osdc->event_lock); | |
1583 | ||
1584 | *pevent = event; | |
1585 | return 0; | |
1586 | } | |
1587 | EXPORT_SYMBOL(ceph_osdc_create_event); | |
1588 | ||
1589 | void ceph_osdc_cancel_event(struct ceph_osd_event *event) | |
1590 | { | |
1591 | struct ceph_osd_client *osdc = event->osdc; | |
1592 | ||
1593 | dout("cancel_event %p\n", event); | |
1594 | spin_lock(&osdc->event_lock); | |
1595 | __remove_event(event); | |
1596 | spin_unlock(&osdc->event_lock); | |
1597 | ceph_osdc_put_event(event); /* caller's */ | |
1598 | } | |
1599 | EXPORT_SYMBOL(ceph_osdc_cancel_event); | |
1600 | ||
1601 | ||
1602 | static void do_event_work(struct work_struct *work) | |
1603 | { | |
1604 | struct ceph_osd_event_work *event_work = | |
1605 | container_of(work, struct ceph_osd_event_work, work); | |
1606 | struct ceph_osd_event *event = event_work->event; | |
1607 | u64 ver = event_work->ver; | |
1608 | u64 notify_id = event_work->notify_id; | |
1609 | u8 opcode = event_work->opcode; | |
1610 | ||
1611 | dout("do_event_work completing %p\n", event); | |
1612 | event->cb(ver, notify_id, opcode, event->data); | |
1613 | complete(&event->completion); | |
1614 | dout("do_event_work completed %p\n", event); | |
1615 | ceph_osdc_put_event(event); | |
1616 | kfree(event_work); | |
1617 | } | |
1618 | ||
1619 | ||
1620 | /* | |
1621 | * Process osd watch notifications | |
1622 | */ | |
1623 | void handle_watch_notify(struct ceph_osd_client *osdc, struct ceph_msg *msg) | |
1624 | { | |
1625 | void *p, *end; | |
1626 | u8 proto_ver; | |
1627 | u64 cookie, ver, notify_id; | |
1628 | u8 opcode; | |
1629 | struct ceph_osd_event *event; | |
1630 | struct ceph_osd_event_work *event_work; | |
1631 | ||
1632 | p = msg->front.iov_base; | |
1633 | end = p + msg->front.iov_len; | |
1634 | ||
1635 | ceph_decode_8_safe(&p, end, proto_ver, bad); | |
1636 | ceph_decode_8_safe(&p, end, opcode, bad); | |
1637 | ceph_decode_64_safe(&p, end, cookie, bad); | |
1638 | ceph_decode_64_safe(&p, end, ver, bad); | |
1639 | ceph_decode_64_safe(&p, end, notify_id, bad); | |
1640 | ||
1641 | spin_lock(&osdc->event_lock); | |
1642 | event = __find_event(osdc, cookie); | |
1643 | if (event) { | |
1644 | get_event(event); | |
1645 | if (event->one_shot) | |
1646 | __remove_event(event); | |
1647 | } | |
1648 | spin_unlock(&osdc->event_lock); | |
1649 | dout("handle_watch_notify cookie %lld ver %lld event %p\n", | |
1650 | cookie, ver, event); | |
1651 | if (event) { | |
1652 | event_work = kmalloc(sizeof(*event_work), GFP_NOIO); | |
a40c4f10 YS |
1653 | if (!event_work) { |
1654 | dout("ERROR: could not allocate event_work\n"); | |
1655 | goto done_err; | |
1656 | } | |
6b0ae409 | 1657 | INIT_WORK(&event_work->work, do_event_work); |
a40c4f10 YS |
1658 | event_work->event = event; |
1659 | event_work->ver = ver; | |
1660 | event_work->notify_id = notify_id; | |
1661 | event_work->opcode = opcode; | |
1662 | if (!queue_work(osdc->notify_wq, &event_work->work)) { | |
1663 | dout("WARNING: failed to queue notify event work\n"); | |
1664 | goto done_err; | |
1665 | } | |
1666 | } | |
1667 | ||
1668 | return; | |
1669 | ||
1670 | done_err: | |
1671 | complete(&event->completion); | |
1672 | ceph_osdc_put_event(event); | |
1673 | return; | |
1674 | ||
1675 | bad: | |
1676 | pr_err("osdc handle_watch_notify corrupt msg\n"); | |
1677 | return; | |
1678 | } | |
1679 | ||
1680 | int ceph_osdc_wait_event(struct ceph_osd_event *event, unsigned long timeout) | |
1681 | { | |
1682 | int err; | |
1683 | ||
1684 | dout("wait_event %p\n", event); | |
1685 | err = wait_for_completion_interruptible_timeout(&event->completion, | |
1686 | timeout * HZ); | |
1687 | ceph_osdc_put_event(event); | |
1688 | if (err > 0) | |
1689 | err = 0; | |
1690 | dout("wait_event %p returns %d\n", event, err); | |
1691 | return err; | |
1692 | } | |
1693 | EXPORT_SYMBOL(ceph_osdc_wait_event); | |
1694 | ||
f24e9980 SW |
1695 | /* |
1696 | * Register request, send initial attempt. | |
1697 | */ | |
1698 | int ceph_osdc_start_request(struct ceph_osd_client *osdc, | |
1699 | struct ceph_osd_request *req, | |
1700 | bool nofail) | |
1701 | { | |
c1ea8823 | 1702 | int rc = 0; |
f24e9980 SW |
1703 | |
1704 | req->r_request->pages = req->r_pages; | |
1705 | req->r_request->nr_pages = req->r_num_pages; | |
68b4476b YS |
1706 | #ifdef CONFIG_BLOCK |
1707 | req->r_request->bio = req->r_bio; | |
1708 | #endif | |
1709 | req->r_request->trail = req->r_trail; | |
f24e9980 SW |
1710 | |
1711 | register_request(osdc, req); | |
1712 | ||
1713 | down_read(&osdc->map_sem); | |
1714 | mutex_lock(&osdc->request_mutex); | |
c1ea8823 SW |
1715 | /* |
1716 | * a racing kick_requests() may have sent the message for us | |
1717 | * while we dropped request_mutex above, so only send now if | |
1718 | * the request still han't been touched yet. | |
1719 | */ | |
1720 | if (req->r_sent == 0) { | |
38d6453c | 1721 | rc = __map_request(osdc, req, 0); |
9d6fcb08 SW |
1722 | if (rc < 0) { |
1723 | if (nofail) { | |
1724 | dout("osdc_start_request failed map, " | |
1725 | " will retry %lld\n", req->r_tid); | |
1726 | rc = 0; | |
1727 | } | |
234af26f | 1728 | goto out_unlock; |
9d6fcb08 | 1729 | } |
6f6c7006 SW |
1730 | if (req->r_osd == NULL) { |
1731 | dout("send_request %p no up osds in pg\n", req); | |
1732 | ceph_monc_request_next_osdmap(&osdc->client->monc); | |
1733 | } else { | |
56e925b6 | 1734 | __send_request(osdc, req); |
f24e9980 | 1735 | } |
56e925b6 | 1736 | rc = 0; |
f24e9980 | 1737 | } |
234af26f DC |
1738 | |
1739 | out_unlock: | |
f24e9980 SW |
1740 | mutex_unlock(&osdc->request_mutex); |
1741 | up_read(&osdc->map_sem); | |
1742 | return rc; | |
1743 | } | |
3d14c5d2 | 1744 | EXPORT_SYMBOL(ceph_osdc_start_request); |
f24e9980 SW |
1745 | |
1746 | /* | |
1747 | * wait for a request to complete | |
1748 | */ | |
1749 | int ceph_osdc_wait_request(struct ceph_osd_client *osdc, | |
1750 | struct ceph_osd_request *req) | |
1751 | { | |
1752 | int rc; | |
1753 | ||
1754 | rc = wait_for_completion_interruptible(&req->r_completion); | |
1755 | if (rc < 0) { | |
1756 | mutex_lock(&osdc->request_mutex); | |
1757 | __cancel_request(req); | |
529cfcc4 | 1758 | __unregister_request(osdc, req); |
f24e9980 | 1759 | mutex_unlock(&osdc->request_mutex); |
25845472 | 1760 | complete_request(req); |
529cfcc4 | 1761 | dout("wait_request tid %llu canceled/timed out\n", req->r_tid); |
f24e9980 SW |
1762 | return rc; |
1763 | } | |
1764 | ||
1765 | dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result); | |
1766 | return req->r_result; | |
1767 | } | |
3d14c5d2 | 1768 | EXPORT_SYMBOL(ceph_osdc_wait_request); |
f24e9980 SW |
1769 | |
1770 | /* | |
1771 | * sync - wait for all in-flight requests to flush. avoid starvation. | |
1772 | */ | |
1773 | void ceph_osdc_sync(struct ceph_osd_client *osdc) | |
1774 | { | |
1775 | struct ceph_osd_request *req; | |
1776 | u64 last_tid, next_tid = 0; | |
1777 | ||
1778 | mutex_lock(&osdc->request_mutex); | |
1779 | last_tid = osdc->last_tid; | |
1780 | while (1) { | |
1781 | req = __lookup_request_ge(osdc, next_tid); | |
1782 | if (!req) | |
1783 | break; | |
1784 | if (req->r_tid > last_tid) | |
1785 | break; | |
1786 | ||
1787 | next_tid = req->r_tid + 1; | |
1788 | if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0) | |
1789 | continue; | |
1790 | ||
1791 | ceph_osdc_get_request(req); | |
1792 | mutex_unlock(&osdc->request_mutex); | |
1793 | dout("sync waiting on tid %llu (last is %llu)\n", | |
1794 | req->r_tid, last_tid); | |
1795 | wait_for_completion(&req->r_safe_completion); | |
1796 | mutex_lock(&osdc->request_mutex); | |
1797 | ceph_osdc_put_request(req); | |
1798 | } | |
1799 | mutex_unlock(&osdc->request_mutex); | |
1800 | dout("sync done (thru tid %llu)\n", last_tid); | |
1801 | } | |
3d14c5d2 | 1802 | EXPORT_SYMBOL(ceph_osdc_sync); |
f24e9980 SW |
1803 | |
1804 | /* | |
1805 | * init, shutdown | |
1806 | */ | |
1807 | int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client) | |
1808 | { | |
1809 | int err; | |
1810 | ||
1811 | dout("init\n"); | |
1812 | osdc->client = client; | |
1813 | osdc->osdmap = NULL; | |
1814 | init_rwsem(&osdc->map_sem); | |
1815 | init_completion(&osdc->map_waiters); | |
1816 | osdc->last_requested_map = 0; | |
1817 | mutex_init(&osdc->request_mutex); | |
f24e9980 SW |
1818 | osdc->last_tid = 0; |
1819 | osdc->osds = RB_ROOT; | |
f5a2041b | 1820 | INIT_LIST_HEAD(&osdc->osd_lru); |
f24e9980 | 1821 | osdc->requests = RB_ROOT; |
422d2cb8 | 1822 | INIT_LIST_HEAD(&osdc->req_lru); |
6f6c7006 SW |
1823 | INIT_LIST_HEAD(&osdc->req_unsent); |
1824 | INIT_LIST_HEAD(&osdc->req_notarget); | |
a40c4f10 | 1825 | INIT_LIST_HEAD(&osdc->req_linger); |
f24e9980 SW |
1826 | osdc->num_requests = 0; |
1827 | INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout); | |
f5a2041b | 1828 | INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout); |
a40c4f10 YS |
1829 | spin_lock_init(&osdc->event_lock); |
1830 | osdc->event_tree = RB_ROOT; | |
1831 | osdc->event_count = 0; | |
f5a2041b YS |
1832 | |
1833 | schedule_delayed_work(&osdc->osds_timeout_work, | |
3d14c5d2 | 1834 | round_jiffies_relative(osdc->client->options->osd_idle_ttl * HZ)); |
f24e9980 | 1835 | |
5f44f142 | 1836 | err = -ENOMEM; |
f24e9980 SW |
1837 | osdc->req_mempool = mempool_create_kmalloc_pool(10, |
1838 | sizeof(struct ceph_osd_request)); | |
1839 | if (!osdc->req_mempool) | |
5f44f142 | 1840 | goto out; |
f24e9980 | 1841 | |
4f48280e SW |
1842 | err = ceph_msgpool_init(&osdc->msgpool_op, OSD_OP_FRONT_LEN, 10, true, |
1843 | "osd_op"); | |
f24e9980 | 1844 | if (err < 0) |
5f44f142 | 1845 | goto out_mempool; |
c16e7869 | 1846 | err = ceph_msgpool_init(&osdc->msgpool_op_reply, |
4f48280e SW |
1847 | OSD_OPREPLY_FRONT_LEN, 10, true, |
1848 | "osd_op_reply"); | |
c16e7869 SW |
1849 | if (err < 0) |
1850 | goto out_msgpool; | |
a40c4f10 YS |
1851 | |
1852 | osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify"); | |
1853 | if (IS_ERR(osdc->notify_wq)) { | |
1854 | err = PTR_ERR(osdc->notify_wq); | |
1855 | osdc->notify_wq = NULL; | |
1856 | goto out_msgpool; | |
1857 | } | |
f24e9980 | 1858 | return 0; |
5f44f142 | 1859 | |
c16e7869 SW |
1860 | out_msgpool: |
1861 | ceph_msgpool_destroy(&osdc->msgpool_op); | |
5f44f142 SW |
1862 | out_mempool: |
1863 | mempool_destroy(osdc->req_mempool); | |
1864 | out: | |
1865 | return err; | |
f24e9980 | 1866 | } |
3d14c5d2 | 1867 | EXPORT_SYMBOL(ceph_osdc_init); |
f24e9980 SW |
1868 | |
1869 | void ceph_osdc_stop(struct ceph_osd_client *osdc) | |
1870 | { | |
a40c4f10 YS |
1871 | flush_workqueue(osdc->notify_wq); |
1872 | destroy_workqueue(osdc->notify_wq); | |
f24e9980 | 1873 | cancel_delayed_work_sync(&osdc->timeout_work); |
f5a2041b | 1874 | cancel_delayed_work_sync(&osdc->osds_timeout_work); |
f24e9980 SW |
1875 | if (osdc->osdmap) { |
1876 | ceph_osdmap_destroy(osdc->osdmap); | |
1877 | osdc->osdmap = NULL; | |
1878 | } | |
aca420bc | 1879 | remove_all_osds(osdc); |
f24e9980 SW |
1880 | mempool_destroy(osdc->req_mempool); |
1881 | ceph_msgpool_destroy(&osdc->msgpool_op); | |
c16e7869 | 1882 | ceph_msgpool_destroy(&osdc->msgpool_op_reply); |
f24e9980 | 1883 | } |
3d14c5d2 | 1884 | EXPORT_SYMBOL(ceph_osdc_stop); |
f24e9980 SW |
1885 | |
1886 | /* | |
1887 | * Read some contiguous pages. If we cross a stripe boundary, shorten | |
1888 | * *plen. Return number of bytes read, or error. | |
1889 | */ | |
1890 | int ceph_osdc_readpages(struct ceph_osd_client *osdc, | |
1891 | struct ceph_vino vino, struct ceph_file_layout *layout, | |
1892 | u64 off, u64 *plen, | |
1893 | u32 truncate_seq, u64 truncate_size, | |
b7495fc2 | 1894 | struct page **pages, int num_pages, int page_align) |
f24e9980 SW |
1895 | { |
1896 | struct ceph_osd_request *req; | |
1897 | int rc = 0; | |
1898 | ||
1899 | dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino, | |
1900 | vino.snap, off, *plen); | |
1901 | req = ceph_osdc_new_request(osdc, layout, vino, off, plen, | |
1902 | CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ, | |
1903 | NULL, 0, truncate_seq, truncate_size, NULL, | |
b7495fc2 | 1904 | false, 1, page_align); |
a79832f2 SW |
1905 | if (!req) |
1906 | return -ENOMEM; | |
f24e9980 SW |
1907 | |
1908 | /* it may be a short read due to an object boundary */ | |
1909 | req->r_pages = pages; | |
f24e9980 | 1910 | |
b7495fc2 SW |
1911 | dout("readpages final extent is %llu~%llu (%d pages align %d)\n", |
1912 | off, *plen, req->r_num_pages, page_align); | |
f24e9980 SW |
1913 | |
1914 | rc = ceph_osdc_start_request(osdc, req, false); | |
1915 | if (!rc) | |
1916 | rc = ceph_osdc_wait_request(osdc, req); | |
1917 | ||
1918 | ceph_osdc_put_request(req); | |
1919 | dout("readpages result %d\n", rc); | |
1920 | return rc; | |
1921 | } | |
3d14c5d2 | 1922 | EXPORT_SYMBOL(ceph_osdc_readpages); |
f24e9980 SW |
1923 | |
1924 | /* | |
1925 | * do a synchronous write on N pages | |
1926 | */ | |
1927 | int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino, | |
1928 | struct ceph_file_layout *layout, | |
1929 | struct ceph_snap_context *snapc, | |
1930 | u64 off, u64 len, | |
1931 | u32 truncate_seq, u64 truncate_size, | |
1932 | struct timespec *mtime, | |
1933 | struct page **pages, int num_pages, | |
1934 | int flags, int do_sync, bool nofail) | |
1935 | { | |
1936 | struct ceph_osd_request *req; | |
1937 | int rc = 0; | |
b7495fc2 | 1938 | int page_align = off & ~PAGE_MASK; |
f24e9980 SW |
1939 | |
1940 | BUG_ON(vino.snap != CEPH_NOSNAP); | |
1941 | req = ceph_osdc_new_request(osdc, layout, vino, off, &len, | |
1942 | CEPH_OSD_OP_WRITE, | |
1943 | flags | CEPH_OSD_FLAG_ONDISK | | |
1944 | CEPH_OSD_FLAG_WRITE, | |
1945 | snapc, do_sync, | |
1946 | truncate_seq, truncate_size, mtime, | |
b7495fc2 | 1947 | nofail, 1, page_align); |
a79832f2 SW |
1948 | if (!req) |
1949 | return -ENOMEM; | |
f24e9980 SW |
1950 | |
1951 | /* it may be a short write due to an object boundary */ | |
1952 | req->r_pages = pages; | |
f24e9980 SW |
1953 | dout("writepages %llu~%llu (%d pages)\n", off, len, |
1954 | req->r_num_pages); | |
1955 | ||
1956 | rc = ceph_osdc_start_request(osdc, req, nofail); | |
1957 | if (!rc) | |
1958 | rc = ceph_osdc_wait_request(osdc, req); | |
1959 | ||
1960 | ceph_osdc_put_request(req); | |
1961 | if (rc == 0) | |
1962 | rc = len; | |
1963 | dout("writepages result %d\n", rc); | |
1964 | return rc; | |
1965 | } | |
3d14c5d2 | 1966 | EXPORT_SYMBOL(ceph_osdc_writepages); |
f24e9980 SW |
1967 | |
1968 | /* | |
1969 | * handle incoming message | |
1970 | */ | |
1971 | static void dispatch(struct ceph_connection *con, struct ceph_msg *msg) | |
1972 | { | |
1973 | struct ceph_osd *osd = con->private; | |
32c895e7 | 1974 | struct ceph_osd_client *osdc; |
f24e9980 SW |
1975 | int type = le16_to_cpu(msg->hdr.type); |
1976 | ||
1977 | if (!osd) | |
4a32f93d | 1978 | goto out; |
32c895e7 | 1979 | osdc = osd->o_osdc; |
f24e9980 SW |
1980 | |
1981 | switch (type) { | |
1982 | case CEPH_MSG_OSD_MAP: | |
1983 | ceph_osdc_handle_map(osdc, msg); | |
1984 | break; | |
1985 | case CEPH_MSG_OSD_OPREPLY: | |
350b1c32 | 1986 | handle_reply(osdc, msg, con); |
f24e9980 | 1987 | break; |
a40c4f10 YS |
1988 | case CEPH_MSG_WATCH_NOTIFY: |
1989 | handle_watch_notify(osdc, msg); | |
1990 | break; | |
f24e9980 SW |
1991 | |
1992 | default: | |
1993 | pr_err("received unknown message type %d %s\n", type, | |
1994 | ceph_msg_type_name(type)); | |
1995 | } | |
4a32f93d | 1996 | out: |
f24e9980 SW |
1997 | ceph_msg_put(msg); |
1998 | } | |
1999 | ||
5b3a4db3 | 2000 | /* |
21b667f6 SW |
2001 | * lookup and return message for incoming reply. set up reply message |
2002 | * pages. | |
5b3a4db3 SW |
2003 | */ |
2004 | static struct ceph_msg *get_reply(struct ceph_connection *con, | |
2450418c YS |
2005 | struct ceph_msg_header *hdr, |
2006 | int *skip) | |
f24e9980 SW |
2007 | { |
2008 | struct ceph_osd *osd = con->private; | |
2009 | struct ceph_osd_client *osdc = osd->o_osdc; | |
2450418c | 2010 | struct ceph_msg *m; |
0547a9b3 | 2011 | struct ceph_osd_request *req; |
5b3a4db3 SW |
2012 | int front = le32_to_cpu(hdr->front_len); |
2013 | int data_len = le32_to_cpu(hdr->data_len); | |
0547a9b3 | 2014 | u64 tid; |
f24e9980 | 2015 | |
0547a9b3 YS |
2016 | tid = le64_to_cpu(hdr->tid); |
2017 | mutex_lock(&osdc->request_mutex); | |
2018 | req = __lookup_request(osdc, tid); | |
2019 | if (!req) { | |
2020 | *skip = 1; | |
2021 | m = NULL; | |
c16e7869 | 2022 | pr_info("get_reply unknown tid %llu from osd%d\n", tid, |
5b3a4db3 | 2023 | osd->o_osd); |
0547a9b3 YS |
2024 | goto out; |
2025 | } | |
c16e7869 SW |
2026 | |
2027 | if (req->r_con_filling_msg) { | |
2028 | dout("get_reply revoking msg %p from old con %p\n", | |
2029 | req->r_reply, req->r_con_filling_msg); | |
2030 | ceph_con_revoke_message(req->r_con_filling_msg, req->r_reply); | |
2031 | ceph_con_put(req->r_con_filling_msg); | |
6f46cb29 | 2032 | req->r_con_filling_msg = NULL; |
0547a9b3 YS |
2033 | } |
2034 | ||
c16e7869 SW |
2035 | if (front > req->r_reply->front.iov_len) { |
2036 | pr_warning("get_reply front %d > preallocated %d\n", | |
2037 | front, (int)req->r_reply->front.iov_len); | |
b61c2763 | 2038 | m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front, GFP_NOFS, false); |
a79832f2 | 2039 | if (!m) |
c16e7869 SW |
2040 | goto out; |
2041 | ceph_msg_put(req->r_reply); | |
2042 | req->r_reply = m; | |
2043 | } | |
2044 | m = ceph_msg_get(req->r_reply); | |
2045 | ||
0547a9b3 | 2046 | if (data_len > 0) { |
b7495fc2 | 2047 | int want = calc_pages_for(req->r_page_alignment, data_len); |
21b667f6 SW |
2048 | |
2049 | if (unlikely(req->r_num_pages < want)) { | |
9bb0ce2b SW |
2050 | pr_warning("tid %lld reply has %d bytes %d pages, we" |
2051 | " had only %d pages ready\n", tid, data_len, | |
2052 | want, req->r_num_pages); | |
0547a9b3 YS |
2053 | *skip = 1; |
2054 | ceph_msg_put(m); | |
a79832f2 | 2055 | m = NULL; |
21b667f6 | 2056 | goto out; |
0547a9b3 | 2057 | } |
21b667f6 SW |
2058 | m->pages = req->r_pages; |
2059 | m->nr_pages = req->r_num_pages; | |
c5c6b19d | 2060 | m->page_alignment = req->r_page_alignment; |
68b4476b YS |
2061 | #ifdef CONFIG_BLOCK |
2062 | m->bio = req->r_bio; | |
2063 | #endif | |
0547a9b3 | 2064 | } |
5b3a4db3 | 2065 | *skip = 0; |
c16e7869 SW |
2066 | req->r_con_filling_msg = ceph_con_get(con); |
2067 | dout("get_reply tid %lld %p\n", tid, m); | |
0547a9b3 YS |
2068 | |
2069 | out: | |
2070 | mutex_unlock(&osdc->request_mutex); | |
2450418c | 2071 | return m; |
5b3a4db3 SW |
2072 | |
2073 | } | |
2074 | ||
2075 | static struct ceph_msg *alloc_msg(struct ceph_connection *con, | |
2076 | struct ceph_msg_header *hdr, | |
2077 | int *skip) | |
2078 | { | |
2079 | struct ceph_osd *osd = con->private; | |
2080 | int type = le16_to_cpu(hdr->type); | |
2081 | int front = le32_to_cpu(hdr->front_len); | |
2082 | ||
2083 | switch (type) { | |
2084 | case CEPH_MSG_OSD_MAP: | |
a40c4f10 | 2085 | case CEPH_MSG_WATCH_NOTIFY: |
b61c2763 | 2086 | return ceph_msg_new(type, front, GFP_NOFS, false); |
5b3a4db3 SW |
2087 | case CEPH_MSG_OSD_OPREPLY: |
2088 | return get_reply(con, hdr, skip); | |
2089 | default: | |
2090 | pr_info("alloc_msg unexpected msg type %d from osd%d\n", type, | |
2091 | osd->o_osd); | |
2092 | *skip = 1; | |
2093 | return NULL; | |
2094 | } | |
f24e9980 SW |
2095 | } |
2096 | ||
2097 | /* | |
2098 | * Wrappers to refcount containing ceph_osd struct | |
2099 | */ | |
2100 | static struct ceph_connection *get_osd_con(struct ceph_connection *con) | |
2101 | { | |
2102 | struct ceph_osd *osd = con->private; | |
2103 | if (get_osd(osd)) | |
2104 | return con; | |
2105 | return NULL; | |
2106 | } | |
2107 | ||
2108 | static void put_osd_con(struct ceph_connection *con) | |
2109 | { | |
2110 | struct ceph_osd *osd = con->private; | |
2111 | put_osd(osd); | |
2112 | } | |
2113 | ||
4e7a5dcd SW |
2114 | /* |
2115 | * authentication | |
2116 | */ | |
a3530df3 AE |
2117 | /* |
2118 | * Note: returned pointer is the address of a structure that's | |
2119 | * managed separately. Caller must *not* attempt to free it. | |
2120 | */ | |
2121 | static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con, | |
8f43fb53 | 2122 | int *proto, int force_new) |
4e7a5dcd SW |
2123 | { |
2124 | struct ceph_osd *o = con->private; | |
2125 | struct ceph_osd_client *osdc = o->o_osdc; | |
2126 | struct ceph_auth_client *ac = osdc->client->monc.auth; | |
74f1869f | 2127 | struct ceph_auth_handshake *auth = &o->o_auth; |
4e7a5dcd | 2128 | |
74f1869f | 2129 | if (force_new && auth->authorizer) { |
a255651d AE |
2130 | if (ac->ops && ac->ops->destroy_authorizer) |
2131 | ac->ops->destroy_authorizer(ac, auth->authorizer); | |
74f1869f AE |
2132 | auth->authorizer = NULL; |
2133 | } | |
a255651d | 2134 | if (!auth->authorizer && ac->ops && ac->ops->create_authorizer) { |
a3530df3 AE |
2135 | int ret = ac->ops->create_authorizer(ac, CEPH_ENTITY_TYPE_OSD, |
2136 | auth); | |
4e7a5dcd | 2137 | if (ret) |
a3530df3 | 2138 | return ERR_PTR(ret); |
4e7a5dcd | 2139 | } |
4e7a5dcd | 2140 | *proto = ac->protocol; |
74f1869f | 2141 | |
a3530df3 | 2142 | return auth; |
4e7a5dcd SW |
2143 | } |
2144 | ||
2145 | ||
2146 | static int verify_authorizer_reply(struct ceph_connection *con, int len) | |
2147 | { | |
2148 | struct ceph_osd *o = con->private; | |
2149 | struct ceph_osd_client *osdc = o->o_osdc; | |
2150 | struct ceph_auth_client *ac = osdc->client->monc.auth; | |
2151 | ||
a255651d AE |
2152 | /* |
2153 | * XXX If ac->ops or ac->ops->verify_authorizer_reply is null, | |
2154 | * XXX which do we do: succeed or fail? | |
2155 | */ | |
6c4a1915 | 2156 | return ac->ops->verify_authorizer_reply(ac, o->o_auth.authorizer, len); |
4e7a5dcd SW |
2157 | } |
2158 | ||
9bd2e6f8 SW |
2159 | static int invalidate_authorizer(struct ceph_connection *con) |
2160 | { | |
2161 | struct ceph_osd *o = con->private; | |
2162 | struct ceph_osd_client *osdc = o->o_osdc; | |
2163 | struct ceph_auth_client *ac = osdc->client->monc.auth; | |
2164 | ||
a255651d | 2165 | if (ac->ops && ac->ops->invalidate_authorizer) |
9bd2e6f8 SW |
2166 | ac->ops->invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD); |
2167 | ||
2168 | return ceph_monc_validate_auth(&osdc->client->monc); | |
2169 | } | |
4e7a5dcd | 2170 | |
9e32789f | 2171 | static const struct ceph_connection_operations osd_con_ops = { |
f24e9980 SW |
2172 | .get = get_osd_con, |
2173 | .put = put_osd_con, | |
2174 | .dispatch = dispatch, | |
4e7a5dcd SW |
2175 | .get_authorizer = get_authorizer, |
2176 | .verify_authorizer_reply = verify_authorizer_reply, | |
9bd2e6f8 | 2177 | .invalidate_authorizer = invalidate_authorizer, |
f24e9980 | 2178 | .alloc_msg = alloc_msg, |
81b024e7 | 2179 | .fault = osd_reset, |
f24e9980 | 2180 | }; |