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 | |
f9d25199 | 26 | static void __send_queued(struct ceph_osd_client *osdc); |
6f6c7006 | 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 SW |
34 | |
35 | /* | |
36 | * Implement client access to distributed object storage cluster. | |
37 | * | |
38 | * All data objects are stored within a cluster/cloud of OSDs, or | |
39 | * "object storage devices." (Note that Ceph OSDs have _nothing_ to | |
40 | * do with the T10 OSD extensions to SCSI.) Ceph OSDs are simply | |
41 | * remote daemons serving up and coordinating consistent and safe | |
42 | * access to storage. | |
43 | * | |
44 | * Cluster membership and the mapping of data objects onto storage devices | |
45 | * are described by the osd map. | |
46 | * | |
47 | * We keep track of pending OSD requests (read, write), resubmit | |
48 | * requests to different OSDs when the cluster topology/data layout | |
49 | * change, or retry the affected requests when the communications | |
50 | * channel with an OSD is reset. | |
51 | */ | |
52 | ||
53 | /* | |
54 | * calculate the mapping of a file extent onto an object, and fill out the | |
55 | * request accordingly. shorten extent as necessary if it crosses an | |
56 | * object boundary. | |
57 | * | |
58 | * fill osd op in request message. | |
59 | */ | |
dbe0fc41 | 60 | static int calc_layout(struct ceph_file_layout *layout, u64 off, u64 *plen, |
a19dadfb | 61 | u64 *objnum, u64 *objoff, u64 *objlen) |
f24e9980 | 62 | { |
60e56f13 | 63 | u64 orig_len = *plen; |
d63b77f4 | 64 | int r; |
f24e9980 | 65 | |
60e56f13 | 66 | /* object extent? */ |
75d1c941 AE |
67 | r = ceph_calc_file_object_mapping(layout, off, orig_len, objnum, |
68 | objoff, objlen); | |
d63b77f4 SW |
69 | if (r < 0) |
70 | return r; | |
75d1c941 AE |
71 | if (*objlen < orig_len) { |
72 | *plen = *objlen; | |
60e56f13 AE |
73 | dout(" skipping last %llu, final file extent %llu~%llu\n", |
74 | orig_len - *plen, off, *plen); | |
75 | } | |
76 | ||
75d1c941 | 77 | dout("calc_layout objnum=%llx %llu~%llu\n", *objnum, *objoff, *objlen); |
f24e9980 | 78 | |
3ff5f385 | 79 | return 0; |
f24e9980 SW |
80 | } |
81 | ||
f24e9980 SW |
82 | /* |
83 | * requests | |
84 | */ | |
415e49a9 | 85 | void ceph_osdc_release_request(struct kref *kref) |
f24e9980 | 86 | { |
e0c59487 | 87 | int num_pages; |
415e49a9 SW |
88 | struct ceph_osd_request *req = container_of(kref, |
89 | struct ceph_osd_request, | |
90 | r_kref); | |
91 | ||
92 | if (req->r_request) | |
93 | ceph_msg_put(req->r_request); | |
0d59ab81 | 94 | if (req->r_con_filling_msg) { |
9cbb1d72 AE |
95 | dout("%s revoking msg %p from con %p\n", __func__, |
96 | req->r_reply, req->r_con_filling_msg); | |
8921d114 | 97 | ceph_msg_revoke_incoming(req->r_reply); |
0d47766f | 98 | req->r_con_filling_msg->ops->put(req->r_con_filling_msg); |
9cbb1d72 | 99 | req->r_con_filling_msg = NULL; |
350b1c32 | 100 | } |
ab8cb34a AE |
101 | if (req->r_reply) |
102 | ceph_msg_put(req->r_reply); | |
0fff87ec AE |
103 | |
104 | if (req->r_data_in.type == CEPH_OSD_DATA_TYPE_PAGES && | |
e0c59487 AE |
105 | req->r_data_in.own_pages) { |
106 | num_pages = calc_pages_for((u64)req->r_data_in.alignment, | |
107 | (u64)req->r_data_in.length); | |
108 | ceph_release_page_vector(req->r_data_in.pages, num_pages); | |
109 | } | |
0fff87ec | 110 | if (req->r_data_out.type == CEPH_OSD_DATA_TYPE_PAGES && |
e0c59487 AE |
111 | req->r_data_out.own_pages) { |
112 | num_pages = calc_pages_for((u64)req->r_data_out.alignment, | |
113 | (u64)req->r_data_out.length); | |
114 | ceph_release_page_vector(req->r_data_out.pages, num_pages); | |
115 | } | |
0fff87ec | 116 | |
415e49a9 SW |
117 | ceph_put_snap_context(req->r_snapc); |
118 | if (req->r_mempool) | |
119 | mempool_free(req, req->r_osdc->req_mempool); | |
120 | else | |
121 | kfree(req); | |
f24e9980 | 122 | } |
3d14c5d2 | 123 | EXPORT_SYMBOL(ceph_osdc_release_request); |
68b4476b | 124 | |
3499e8a5 | 125 | struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc, |
f24e9980 | 126 | struct ceph_snap_context *snapc, |
1b83bef2 | 127 | unsigned int num_ops, |
3499e8a5 | 128 | bool use_mempool, |
54a54007 | 129 | gfp_t gfp_flags) |
f24e9980 SW |
130 | { |
131 | struct ceph_osd_request *req; | |
132 | struct ceph_msg *msg; | |
1b83bef2 SW |
133 | size_t msg_size; |
134 | ||
135 | msg_size = 4 + 4 + 8 + 8 + 4+8; | |
136 | msg_size += 2 + 4 + 8 + 4 + 4; /* oloc */ | |
137 | msg_size += 1 + 8 + 4 + 4; /* pg_t */ | |
138 | msg_size += 4 + MAX_OBJ_NAME_SIZE; | |
139 | msg_size += 2 + num_ops*sizeof(struct ceph_osd_op); | |
140 | msg_size += 8; /* snapid */ | |
141 | msg_size += 8; /* snap_seq */ | |
142 | msg_size += 8 * (snapc ? snapc->num_snaps : 0); /* snaps */ | |
143 | msg_size += 4; | |
f24e9980 SW |
144 | |
145 | if (use_mempool) { | |
3499e8a5 | 146 | req = mempool_alloc(osdc->req_mempool, gfp_flags); |
f24e9980 SW |
147 | memset(req, 0, sizeof(*req)); |
148 | } else { | |
3499e8a5 | 149 | req = kzalloc(sizeof(*req), gfp_flags); |
f24e9980 SW |
150 | } |
151 | if (req == NULL) | |
a79832f2 | 152 | return NULL; |
f24e9980 | 153 | |
f24e9980 SW |
154 | req->r_osdc = osdc; |
155 | req->r_mempool = use_mempool; | |
68b4476b | 156 | |
415e49a9 | 157 | kref_init(&req->r_kref); |
f24e9980 SW |
158 | init_completion(&req->r_completion); |
159 | init_completion(&req->r_safe_completion); | |
a978fa20 | 160 | RB_CLEAR_NODE(&req->r_node); |
f24e9980 | 161 | INIT_LIST_HEAD(&req->r_unsafe_item); |
a40c4f10 YS |
162 | INIT_LIST_HEAD(&req->r_linger_item); |
163 | INIT_LIST_HEAD(&req->r_linger_osd); | |
935b639a | 164 | INIT_LIST_HEAD(&req->r_req_lru_item); |
cd43045c SW |
165 | INIT_LIST_HEAD(&req->r_osd_item); |
166 | ||
c16e7869 SW |
167 | /* create reply message */ |
168 | if (use_mempool) | |
169 | msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0); | |
170 | else | |
171 | msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, | |
b61c2763 | 172 | OSD_OPREPLY_FRONT_LEN, gfp_flags, true); |
a79832f2 | 173 | if (!msg) { |
c16e7869 | 174 | ceph_osdc_put_request(req); |
a79832f2 | 175 | return NULL; |
c16e7869 SW |
176 | } |
177 | req->r_reply = msg; | |
178 | ||
0fff87ec AE |
179 | req->r_data_in.type = CEPH_OSD_DATA_TYPE_NONE; |
180 | req->r_data_out.type = CEPH_OSD_DATA_TYPE_NONE; | |
d50b409f | 181 | |
c16e7869 | 182 | /* create request message; allow space for oid */ |
f24e9980 | 183 | if (use_mempool) |
8f3bc053 | 184 | msg = ceph_msgpool_get(&osdc->msgpool_op, 0); |
f24e9980 | 185 | else |
b61c2763 | 186 | msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp_flags, true); |
a79832f2 | 187 | if (!msg) { |
f24e9980 | 188 | ceph_osdc_put_request(req); |
a79832f2 | 189 | return NULL; |
f24e9980 | 190 | } |
68b4476b | 191 | |
f24e9980 | 192 | memset(msg->front.iov_base, 0, msg->front.iov_len); |
3499e8a5 YS |
193 | |
194 | req->r_request = msg; | |
3499e8a5 YS |
195 | |
196 | return req; | |
197 | } | |
3d14c5d2 | 198 | EXPORT_SYMBOL(ceph_osdc_alloc_request); |
3499e8a5 | 199 | |
a8dd0a37 | 200 | static bool osd_req_opcode_valid(u16 opcode) |
68b4476b | 201 | { |
a8dd0a37 | 202 | switch (opcode) { |
68b4476b | 203 | case CEPH_OSD_OP_READ: |
a8dd0a37 | 204 | case CEPH_OSD_OP_STAT: |
4c46459c AE |
205 | case CEPH_OSD_OP_MAPEXT: |
206 | case CEPH_OSD_OP_MASKTRUNC: | |
207 | case CEPH_OSD_OP_SPARSE_READ: | |
a9f36c3e | 208 | case CEPH_OSD_OP_NOTIFY: |
a8dd0a37 | 209 | case CEPH_OSD_OP_NOTIFY_ACK: |
4c46459c | 210 | case CEPH_OSD_OP_ASSERT_VER: |
a8dd0a37 | 211 | case CEPH_OSD_OP_WRITE: |
4c46459c AE |
212 | case CEPH_OSD_OP_WRITEFULL: |
213 | case CEPH_OSD_OP_TRUNCATE: | |
214 | case CEPH_OSD_OP_ZERO: | |
215 | case CEPH_OSD_OP_DELETE: | |
216 | case CEPH_OSD_OP_APPEND: | |
a8dd0a37 | 217 | case CEPH_OSD_OP_STARTSYNC: |
4c46459c AE |
218 | case CEPH_OSD_OP_SETTRUNC: |
219 | case CEPH_OSD_OP_TRIMTRUNC: | |
220 | case CEPH_OSD_OP_TMAPUP: | |
221 | case CEPH_OSD_OP_TMAPPUT: | |
222 | case CEPH_OSD_OP_TMAPGET: | |
223 | case CEPH_OSD_OP_CREATE: | |
a9f36c3e | 224 | case CEPH_OSD_OP_ROLLBACK: |
a8dd0a37 | 225 | case CEPH_OSD_OP_WATCH: |
4c46459c AE |
226 | case CEPH_OSD_OP_OMAPGETKEYS: |
227 | case CEPH_OSD_OP_OMAPGETVALS: | |
228 | case CEPH_OSD_OP_OMAPGETHEADER: | |
229 | case CEPH_OSD_OP_OMAPGETVALSBYKEYS: | |
4c46459c AE |
230 | case CEPH_OSD_OP_OMAPSETVALS: |
231 | case CEPH_OSD_OP_OMAPSETHEADER: | |
232 | case CEPH_OSD_OP_OMAPCLEAR: | |
233 | case CEPH_OSD_OP_OMAPRMKEYS: | |
234 | case CEPH_OSD_OP_OMAP_CMP: | |
235 | case CEPH_OSD_OP_CLONERANGE: | |
236 | case CEPH_OSD_OP_ASSERT_SRC_VERSION: | |
237 | case CEPH_OSD_OP_SRC_CMPXATTR: | |
a9f36c3e | 238 | case CEPH_OSD_OP_GETXATTR: |
4c46459c | 239 | case CEPH_OSD_OP_GETXATTRS: |
a9f36c3e AE |
240 | case CEPH_OSD_OP_CMPXATTR: |
241 | case CEPH_OSD_OP_SETXATTR: | |
4c46459c AE |
242 | case CEPH_OSD_OP_SETXATTRS: |
243 | case CEPH_OSD_OP_RESETXATTRS: | |
244 | case CEPH_OSD_OP_RMXATTR: | |
245 | case CEPH_OSD_OP_PULL: | |
246 | case CEPH_OSD_OP_PUSH: | |
247 | case CEPH_OSD_OP_BALANCEREADS: | |
248 | case CEPH_OSD_OP_UNBALANCEREADS: | |
249 | case CEPH_OSD_OP_SCRUB: | |
250 | case CEPH_OSD_OP_SCRUB_RESERVE: | |
251 | case CEPH_OSD_OP_SCRUB_UNRESERVE: | |
252 | case CEPH_OSD_OP_SCRUB_STOP: | |
253 | case CEPH_OSD_OP_SCRUB_MAP: | |
254 | case CEPH_OSD_OP_WRLOCK: | |
255 | case CEPH_OSD_OP_WRUNLOCK: | |
256 | case CEPH_OSD_OP_RDLOCK: | |
257 | case CEPH_OSD_OP_RDUNLOCK: | |
258 | case CEPH_OSD_OP_UPLOCK: | |
259 | case CEPH_OSD_OP_DNLOCK: | |
a8dd0a37 | 260 | case CEPH_OSD_OP_CALL: |
4c46459c AE |
261 | case CEPH_OSD_OP_PGLS: |
262 | case CEPH_OSD_OP_PGLS_FILTER: | |
a8dd0a37 AE |
263 | return true; |
264 | default: | |
265 | return false; | |
266 | } | |
267 | } | |
268 | ||
33803f33 AE |
269 | /* |
270 | * This is an osd op init function for opcodes that have no data or | |
271 | * other information associated with them. It also serves as a | |
272 | * common init routine for all the other init functions, below. | |
273 | */ | |
274 | void osd_req_op_init(struct ceph_osd_req_op *op, u16 opcode) | |
275 | { | |
276 | BUG_ON(!osd_req_opcode_valid(opcode)); | |
277 | ||
278 | memset(op, 0, sizeof (*op)); | |
279 | ||
280 | op->op = opcode; | |
281 | } | |
282 | ||
283 | void osd_req_op_extent_init(struct ceph_osd_req_op *op, u16 opcode, | |
284 | u64 offset, u64 length, | |
285 | u64 truncate_size, u32 truncate_seq) | |
286 | { | |
287 | size_t payload_len = 0; | |
288 | ||
289 | BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE); | |
290 | ||
291 | osd_req_op_init(op, opcode); | |
292 | ||
293 | op->extent.offset = offset; | |
294 | op->extent.length = length; | |
295 | op->extent.truncate_size = truncate_size; | |
296 | op->extent.truncate_seq = truncate_seq; | |
297 | if (opcode == CEPH_OSD_OP_WRITE) | |
298 | payload_len += length; | |
299 | ||
300 | op->payload_len = payload_len; | |
301 | } | |
302 | EXPORT_SYMBOL(osd_req_op_extent_init); | |
303 | ||
304 | void osd_req_op_cls_init(struct ceph_osd_req_op *op, u16 opcode, | |
305 | const char *class, const char *method, | |
306 | const void *request_data, size_t request_data_size) | |
307 | { | |
308 | size_t payload_len = 0; | |
309 | size_t size; | |
310 | ||
311 | BUG_ON(opcode != CEPH_OSD_OP_CALL); | |
312 | ||
313 | osd_req_op_init(op, opcode); | |
314 | ||
315 | op->cls.class_name = class; | |
316 | size = strlen(class); | |
317 | BUG_ON(size > (size_t) U8_MAX); | |
318 | op->cls.class_len = size; | |
319 | payload_len += size; | |
320 | ||
321 | op->cls.method_name = method; | |
322 | size = strlen(method); | |
323 | BUG_ON(size > (size_t) U8_MAX); | |
324 | op->cls.method_len = size; | |
325 | payload_len += size; | |
326 | ||
327 | op->cls.indata = request_data; | |
328 | BUG_ON(request_data_size > (size_t) U32_MAX); | |
329 | op->cls.indata_len = (u32) request_data_size; | |
330 | payload_len += request_data_size; | |
331 | ||
332 | op->cls.argc = 0; /* currently unused */ | |
333 | ||
334 | op->payload_len = payload_len; | |
335 | } | |
336 | EXPORT_SYMBOL(osd_req_op_cls_init); | |
337 | ||
338 | void osd_req_op_watch_init(struct ceph_osd_req_op *op, u16 opcode, | |
339 | u64 cookie, u64 version, int flag) | |
340 | { | |
341 | BUG_ON(opcode != CEPH_OSD_OP_NOTIFY_ACK && opcode != CEPH_OSD_OP_WATCH); | |
342 | ||
343 | osd_req_op_init(op, opcode); | |
344 | ||
345 | op->watch.cookie = cookie; | |
346 | /* op->watch.ver = version; */ /* XXX 3847 */ | |
347 | op->watch.ver = cpu_to_le64(version); | |
348 | if (opcode == CEPH_OSD_OP_WATCH && flag) | |
349 | op->watch.flag = (u8) 1; | |
350 | } | |
351 | EXPORT_SYMBOL(osd_req_op_watch_init); | |
352 | ||
a8dd0a37 AE |
353 | static u64 osd_req_encode_op(struct ceph_osd_request *req, |
354 | struct ceph_osd_op *dst, | |
355 | struct ceph_osd_req_op *src) | |
356 | { | |
357 | u64 out_data_len = 0; | |
358 | struct ceph_pagelist *pagelist; | |
359 | ||
360 | if (WARN_ON(!osd_req_opcode_valid(src->op))) { | |
361 | pr_err("unrecognized osd opcode %d\n", src->op); | |
362 | ||
363 | return 0; | |
364 | } | |
365 | ||
366 | switch (src->op) { | |
367 | case CEPH_OSD_OP_STAT: | |
368 | break; | |
369 | case CEPH_OSD_OP_READ: | |
370 | case CEPH_OSD_OP_WRITE: | |
371 | if (src->op == CEPH_OSD_OP_WRITE) | |
372 | out_data_len = src->extent.length; | |
373 | dst->extent.offset = cpu_to_le64(src->extent.offset); | |
374 | dst->extent.length = cpu_to_le64(src->extent.length); | |
375 | dst->extent.truncate_size = | |
376 | cpu_to_le64(src->extent.truncate_size); | |
377 | dst->extent.truncate_seq = | |
378 | cpu_to_le32(src->extent.truncate_seq); | |
379 | break; | |
380 | case CEPH_OSD_OP_CALL: | |
381 | pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS); | |
382 | BUG_ON(!pagelist); | |
383 | ceph_pagelist_init(pagelist); | |
384 | ||
385 | dst->cls.class_len = src->cls.class_len; | |
386 | dst->cls.method_len = src->cls.method_len; | |
387 | dst->cls.indata_len = cpu_to_le32(src->cls.indata_len); | |
388 | ceph_pagelist_append(pagelist, src->cls.class_name, | |
389 | src->cls.class_len); | |
390 | ceph_pagelist_append(pagelist, src->cls.method_name, | |
391 | src->cls.method_len); | |
392 | ceph_pagelist_append(pagelist, src->cls.indata, | |
393 | src->cls.indata_len); | |
394 | ||
395 | req->r_data_out.type = CEPH_OSD_DATA_TYPE_PAGELIST; | |
396 | req->r_data_out.pagelist = pagelist; | |
397 | out_data_len = pagelist->length; | |
398 | break; | |
399 | case CEPH_OSD_OP_STARTSYNC: | |
400 | break; | |
401 | case CEPH_OSD_OP_NOTIFY_ACK: | |
402 | case CEPH_OSD_OP_WATCH: | |
403 | dst->watch.cookie = cpu_to_le64(src->watch.cookie); | |
404 | dst->watch.ver = cpu_to_le64(src->watch.ver); | |
405 | dst->watch.flag = src->watch.flag; | |
406 | break; | |
407 | default: | |
4c46459c | 408 | pr_err("unsupported osd opcode %s\n", |
8f63ca2d | 409 | ceph_osd_op_name(src->op)); |
4c46459c | 410 | WARN_ON(1); |
a8dd0a37 AE |
411 | |
412 | return 0; | |
68b4476b | 413 | } |
a8dd0a37 | 414 | dst->op = cpu_to_le16(src->op); |
68b4476b | 415 | dst->payload_len = cpu_to_le32(src->payload_len); |
175face2 AE |
416 | |
417 | return out_data_len; | |
68b4476b YS |
418 | } |
419 | ||
3499e8a5 YS |
420 | /* |
421 | * build new request AND message | |
422 | * | |
423 | */ | |
424 | void ceph_osdc_build_request(struct ceph_osd_request *req, | |
175face2 | 425 | u64 off, unsigned int num_ops, |
68b4476b | 426 | struct ceph_osd_req_op *src_ops, |
4d6b250b | 427 | struct ceph_snap_context *snapc, u64 snap_id, |
af77f26c | 428 | struct timespec *mtime) |
3499e8a5 YS |
429 | { |
430 | struct ceph_msg *msg = req->r_request; | |
68b4476b | 431 | struct ceph_osd_req_op *src_op; |
3499e8a5 | 432 | void *p; |
1b83bef2 | 433 | size_t msg_size; |
3499e8a5 | 434 | int flags = req->r_flags; |
f44246e3 | 435 | u64 data_len; |
68b4476b | 436 | int i; |
3499e8a5 | 437 | |
1b83bef2 SW |
438 | req->r_num_ops = num_ops; |
439 | req->r_snapid = snap_id; | |
f24e9980 SW |
440 | req->r_snapc = ceph_get_snap_context(snapc); |
441 | ||
1b83bef2 SW |
442 | /* encode request */ |
443 | msg->hdr.version = cpu_to_le16(4); | |
444 | ||
445 | p = msg->front.iov_base; | |
446 | ceph_encode_32(&p, 1); /* client_inc is always 1 */ | |
447 | req->r_request_osdmap_epoch = p; | |
448 | p += 4; | |
449 | req->r_request_flags = p; | |
450 | p += 4; | |
451 | if (req->r_flags & CEPH_OSD_FLAG_WRITE) | |
452 | ceph_encode_timespec(p, mtime); | |
453 | p += sizeof(struct ceph_timespec); | |
454 | req->r_request_reassert_version = p; | |
455 | p += sizeof(struct ceph_eversion); /* will get filled in */ | |
456 | ||
457 | /* oloc */ | |
458 | ceph_encode_8(&p, 4); | |
459 | ceph_encode_8(&p, 4); | |
460 | ceph_encode_32(&p, 8 + 4 + 4); | |
461 | req->r_request_pool = p; | |
462 | p += 8; | |
463 | ceph_encode_32(&p, -1); /* preferred */ | |
464 | ceph_encode_32(&p, 0); /* key len */ | |
465 | ||
466 | ceph_encode_8(&p, 1); | |
467 | req->r_request_pgid = p; | |
468 | p += 8 + 4; | |
469 | ceph_encode_32(&p, -1); /* preferred */ | |
470 | ||
471 | /* oid */ | |
472 | ceph_encode_32(&p, req->r_oid_len); | |
af77f26c | 473 | memcpy(p, req->r_oid, req->r_oid_len); |
1b83bef2 | 474 | dout("oid '%.*s' len %d\n", req->r_oid_len, req->r_oid, req->r_oid_len); |
af77f26c | 475 | p += req->r_oid_len; |
f24e9980 | 476 | |
175face2 | 477 | /* ops--can imply data */ |
1b83bef2 | 478 | ceph_encode_16(&p, num_ops); |
68b4476b | 479 | src_op = src_ops; |
1b83bef2 | 480 | req->r_request_ops = p; |
175face2 | 481 | data_len = 0; |
1b83bef2 | 482 | for (i = 0; i < num_ops; i++, src_op++) { |
175face2 | 483 | data_len += osd_req_encode_op(req, p, src_op); |
1b83bef2 SW |
484 | p += sizeof(struct ceph_osd_op); |
485 | } | |
68b4476b | 486 | |
1b83bef2 SW |
487 | /* snaps */ |
488 | ceph_encode_64(&p, req->r_snapid); | |
489 | ceph_encode_64(&p, req->r_snapc ? req->r_snapc->seq : 0); | |
490 | ceph_encode_32(&p, req->r_snapc ? req->r_snapc->num_snaps : 0); | |
491 | if (req->r_snapc) { | |
f24e9980 | 492 | for (i = 0; i < snapc->num_snaps; i++) { |
1b83bef2 | 493 | ceph_encode_64(&p, req->r_snapc->snaps[i]); |
f24e9980 SW |
494 | } |
495 | } | |
496 | ||
1b83bef2 SW |
497 | req->r_request_attempts = p; |
498 | p += 4; | |
499 | ||
175face2 | 500 | /* data */ |
0baa1bd9 AE |
501 | if (flags & CEPH_OSD_FLAG_WRITE) { |
502 | u16 data_off; | |
503 | ||
504 | /* | |
505 | * The header "data_off" is a hint to the receiver | |
506 | * allowing it to align received data into its | |
507 | * buffers such that there's no need to re-copy | |
508 | * it before writing it to disk (direct I/O). | |
509 | */ | |
510 | data_off = (u16) (off & 0xffff); | |
511 | req->r_request->hdr.data_off = cpu_to_le16(data_off); | |
512 | } | |
f44246e3 | 513 | req->r_request->hdr.data_len = cpu_to_le32(data_len); |
c5c6b19d | 514 | |
f24e9980 | 515 | BUG_ON(p > msg->front.iov_base + msg->front.iov_len); |
6f863e71 SW |
516 | msg_size = p - msg->front.iov_base; |
517 | msg->front.iov_len = msg_size; | |
518 | msg->hdr.front_len = cpu_to_le32(msg_size); | |
1b83bef2 SW |
519 | |
520 | dout("build_request msg_size was %d num_ops %d\n", (int)msg_size, | |
521 | num_ops); | |
3499e8a5 YS |
522 | return; |
523 | } | |
3d14c5d2 | 524 | EXPORT_SYMBOL(ceph_osdc_build_request); |
3499e8a5 YS |
525 | |
526 | /* | |
527 | * build new request AND message, calculate layout, and adjust file | |
528 | * extent as needed. | |
529 | * | |
530 | * if the file was recently truncated, we include information about its | |
531 | * old and new size so that the object can be updated appropriately. (we | |
532 | * avoid synchronously deleting truncated objects because it's slow.) | |
533 | * | |
534 | * if @do_sync, include a 'startsync' command so that the osd will flush | |
535 | * data quickly. | |
536 | */ | |
537 | struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc, | |
538 | struct ceph_file_layout *layout, | |
539 | struct ceph_vino vino, | |
540 | u64 off, u64 *plen, | |
541 | int opcode, int flags, | |
542 | struct ceph_snap_context *snapc, | |
543 | int do_sync, | |
544 | u32 truncate_seq, | |
545 | u64 truncate_size, | |
546 | struct timespec *mtime, | |
153e5167 | 547 | bool use_mempool) |
3499e8a5 | 548 | { |
ae7ca4a3 | 549 | struct ceph_osd_req_op ops[2]; |
68b4476b | 550 | struct ceph_osd_request *req; |
d18d1e28 | 551 | unsigned int num_op = do_sync ? 2 : 1; |
75d1c941 AE |
552 | u64 objnum = 0; |
553 | u64 objoff = 0; | |
554 | u64 objlen = 0; | |
d18d1e28 AE |
555 | u32 object_size; |
556 | u64 object_base; | |
6816282d | 557 | int r; |
68b4476b | 558 | |
d18d1e28 | 559 | BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE); |
68b4476b | 560 | |
ae7ca4a3 AE |
561 | req = ceph_osdc_alloc_request(osdc, snapc, num_op, use_mempool, |
562 | GFP_NOFS); | |
4ad12621 | 563 | if (!req) |
6816282d | 564 | return ERR_PTR(-ENOMEM); |
d178a9e7 | 565 | req->r_flags = flags; |
3499e8a5 YS |
566 | |
567 | /* calculate max write size */ | |
a19dadfb | 568 | r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen); |
3ff5f385 AE |
569 | if (r < 0) { |
570 | ceph_osdc_put_request(req); | |
6816282d | 571 | return ERR_PTR(r); |
3ff5f385 | 572 | } |
a19dadfb | 573 | |
d18d1e28 AE |
574 | object_size = le32_to_cpu(layout->fl_object_size); |
575 | object_base = off - objoff; | |
576 | if (truncate_size <= object_base) { | |
577 | truncate_size = 0; | |
578 | } else { | |
579 | truncate_size -= object_base; | |
580 | if (truncate_size > object_size) | |
581 | truncate_size = object_size; | |
a19dadfb | 582 | } |
d18d1e28 AE |
583 | |
584 | memset(&ops, 0, sizeof ops); | |
585 | ops[0].op = opcode; | |
586 | ops[0].extent.offset = objoff; | |
587 | ops[0].extent.length = objlen; | |
588 | ops[0].extent.truncate_size = truncate_size; | |
589 | ops[0].extent.truncate_seq = truncate_seq; | |
a19dadfb AE |
590 | if (ops[0].op == CEPH_OSD_OP_WRITE) |
591 | ops[0].payload_len = *plen; | |
592 | ||
d18d1e28 AE |
593 | if (do_sync) |
594 | ops[1].op = CEPH_OSD_OP_STARTSYNC; | |
595 | ||
3499e8a5 YS |
596 | req->r_file_layout = *layout; /* keep a copy */ |
597 | ||
75d1c941 AE |
598 | snprintf(req->r_oid, sizeof(req->r_oid), "%llx.%08llx", |
599 | vino.ino, objnum); | |
dbe0fc41 AE |
600 | req->r_oid_len = strlen(req->r_oid); |
601 | ||
175face2 | 602 | ceph_osdc_build_request(req, off, num_op, ops, |
ae7ca4a3 | 603 | snapc, vino.snap, mtime); |
3499e8a5 | 604 | |
f24e9980 SW |
605 | return req; |
606 | } | |
3d14c5d2 | 607 | EXPORT_SYMBOL(ceph_osdc_new_request); |
f24e9980 SW |
608 | |
609 | /* | |
610 | * We keep osd requests in an rbtree, sorted by ->r_tid. | |
611 | */ | |
612 | static void __insert_request(struct ceph_osd_client *osdc, | |
613 | struct ceph_osd_request *new) | |
614 | { | |
615 | struct rb_node **p = &osdc->requests.rb_node; | |
616 | struct rb_node *parent = NULL; | |
617 | struct ceph_osd_request *req = NULL; | |
618 | ||
619 | while (*p) { | |
620 | parent = *p; | |
621 | req = rb_entry(parent, struct ceph_osd_request, r_node); | |
622 | if (new->r_tid < req->r_tid) | |
623 | p = &(*p)->rb_left; | |
624 | else if (new->r_tid > req->r_tid) | |
625 | p = &(*p)->rb_right; | |
626 | else | |
627 | BUG(); | |
628 | } | |
629 | ||
630 | rb_link_node(&new->r_node, parent, p); | |
631 | rb_insert_color(&new->r_node, &osdc->requests); | |
632 | } | |
633 | ||
634 | static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc, | |
635 | u64 tid) | |
636 | { | |
637 | struct ceph_osd_request *req; | |
638 | struct rb_node *n = osdc->requests.rb_node; | |
639 | ||
640 | while (n) { | |
641 | req = rb_entry(n, struct ceph_osd_request, r_node); | |
642 | if (tid < req->r_tid) | |
643 | n = n->rb_left; | |
644 | else if (tid > req->r_tid) | |
645 | n = n->rb_right; | |
646 | else | |
647 | return req; | |
648 | } | |
649 | return NULL; | |
650 | } | |
651 | ||
652 | static struct ceph_osd_request * | |
653 | __lookup_request_ge(struct ceph_osd_client *osdc, | |
654 | u64 tid) | |
655 | { | |
656 | struct ceph_osd_request *req; | |
657 | struct rb_node *n = osdc->requests.rb_node; | |
658 | ||
659 | while (n) { | |
660 | req = rb_entry(n, struct ceph_osd_request, r_node); | |
661 | if (tid < req->r_tid) { | |
662 | if (!n->rb_left) | |
663 | return req; | |
664 | n = n->rb_left; | |
665 | } else if (tid > req->r_tid) { | |
666 | n = n->rb_right; | |
667 | } else { | |
668 | return req; | |
669 | } | |
670 | } | |
671 | return NULL; | |
672 | } | |
673 | ||
6f6c7006 SW |
674 | /* |
675 | * Resubmit requests pending on the given osd. | |
676 | */ | |
677 | static void __kick_osd_requests(struct ceph_osd_client *osdc, | |
678 | struct ceph_osd *osd) | |
679 | { | |
a40c4f10 | 680 | struct ceph_osd_request *req, *nreq; |
e02493c0 | 681 | LIST_HEAD(resend); |
6f6c7006 SW |
682 | int err; |
683 | ||
684 | dout("__kick_osd_requests osd%d\n", osd->o_osd); | |
685 | err = __reset_osd(osdc, osd); | |
685a7555 | 686 | if (err) |
6f6c7006 | 687 | return; |
e02493c0 AE |
688 | /* |
689 | * Build up a list of requests to resend by traversing the | |
690 | * osd's list of requests. Requests for a given object are | |
691 | * sent in tid order, and that is also the order they're | |
692 | * kept on this list. Therefore all requests that are in | |
693 | * flight will be found first, followed by all requests that | |
694 | * have not yet been sent. And to resend requests while | |
695 | * preserving this order we will want to put any sent | |
696 | * requests back on the front of the osd client's unsent | |
697 | * list. | |
698 | * | |
699 | * So we build a separate ordered list of already-sent | |
700 | * requests for the affected osd and splice it onto the | |
701 | * front of the osd client's unsent list. Once we've seen a | |
702 | * request that has not yet been sent we're done. Those | |
703 | * requests are already sitting right where they belong. | |
704 | */ | |
6f6c7006 | 705 | list_for_each_entry(req, &osd->o_requests, r_osd_item) { |
e02493c0 AE |
706 | if (!req->r_sent) |
707 | break; | |
708 | list_move_tail(&req->r_req_lru_item, &resend); | |
709 | dout("requeueing %p tid %llu osd%d\n", req, req->r_tid, | |
6f6c7006 | 710 | osd->o_osd); |
a40c4f10 YS |
711 | if (!req->r_linger) |
712 | req->r_flags |= CEPH_OSD_FLAG_RETRY; | |
713 | } | |
e02493c0 | 714 | list_splice(&resend, &osdc->req_unsent); |
a40c4f10 | 715 | |
e02493c0 AE |
716 | /* |
717 | * Linger requests are re-registered before sending, which | |
718 | * sets up a new tid for each. We add them to the unsent | |
719 | * list at the end to keep things in tid order. | |
720 | */ | |
a40c4f10 YS |
721 | list_for_each_entry_safe(req, nreq, &osd->o_linger_requests, |
722 | r_linger_osd) { | |
77f38e0e SW |
723 | /* |
724 | * reregister request prior to unregistering linger so | |
725 | * that r_osd is preserved. | |
726 | */ | |
727 | BUG_ON(!list_empty(&req->r_req_lru_item)); | |
a40c4f10 | 728 | __register_request(osdc, req); |
e02493c0 | 729 | list_add_tail(&req->r_req_lru_item, &osdc->req_unsent); |
ad885927 | 730 | list_add_tail(&req->r_osd_item, &req->r_osd->o_requests); |
77f38e0e | 731 | __unregister_linger_request(osdc, req); |
a40c4f10 YS |
732 | dout("requeued lingering %p tid %llu osd%d\n", req, req->r_tid, |
733 | osd->o_osd); | |
6f6c7006 SW |
734 | } |
735 | } | |
736 | ||
f24e9980 | 737 | /* |
81b024e7 | 738 | * If the osd connection drops, we need to resubmit all requests. |
f24e9980 SW |
739 | */ |
740 | static void osd_reset(struct ceph_connection *con) | |
741 | { | |
742 | struct ceph_osd *osd = con->private; | |
743 | struct ceph_osd_client *osdc; | |
744 | ||
745 | if (!osd) | |
746 | return; | |
747 | dout("osd_reset osd%d\n", osd->o_osd); | |
748 | osdc = osd->o_osdc; | |
f24e9980 | 749 | down_read(&osdc->map_sem); |
83aff95e SW |
750 | mutex_lock(&osdc->request_mutex); |
751 | __kick_osd_requests(osdc, osd); | |
f9d25199 | 752 | __send_queued(osdc); |
83aff95e | 753 | mutex_unlock(&osdc->request_mutex); |
f24e9980 SW |
754 | up_read(&osdc->map_sem); |
755 | } | |
756 | ||
757 | /* | |
758 | * Track open sessions with osds. | |
759 | */ | |
e10006f8 | 760 | static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum) |
f24e9980 SW |
761 | { |
762 | struct ceph_osd *osd; | |
763 | ||
764 | osd = kzalloc(sizeof(*osd), GFP_NOFS); | |
765 | if (!osd) | |
766 | return NULL; | |
767 | ||
768 | atomic_set(&osd->o_ref, 1); | |
769 | osd->o_osdc = osdc; | |
e10006f8 | 770 | osd->o_osd = onum; |
f407731d | 771 | RB_CLEAR_NODE(&osd->o_node); |
f24e9980 | 772 | INIT_LIST_HEAD(&osd->o_requests); |
a40c4f10 | 773 | INIT_LIST_HEAD(&osd->o_linger_requests); |
f5a2041b | 774 | INIT_LIST_HEAD(&osd->o_osd_lru); |
f24e9980 SW |
775 | osd->o_incarnation = 1; |
776 | ||
b7a9e5dd | 777 | ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr); |
4e7a5dcd | 778 | |
422d2cb8 | 779 | INIT_LIST_HEAD(&osd->o_keepalive_item); |
f24e9980 SW |
780 | return osd; |
781 | } | |
782 | ||
783 | static struct ceph_osd *get_osd(struct ceph_osd *osd) | |
784 | { | |
785 | if (atomic_inc_not_zero(&osd->o_ref)) { | |
786 | dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1, | |
787 | atomic_read(&osd->o_ref)); | |
788 | return osd; | |
789 | } else { | |
790 | dout("get_osd %p FAIL\n", osd); | |
791 | return NULL; | |
792 | } | |
793 | } | |
794 | ||
795 | static void put_osd(struct ceph_osd *osd) | |
796 | { | |
797 | dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref), | |
798 | atomic_read(&osd->o_ref) - 1); | |
a255651d | 799 | if (atomic_dec_and_test(&osd->o_ref) && osd->o_auth.authorizer) { |
79494d1b SW |
800 | struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth; |
801 | ||
27859f97 | 802 | ceph_auth_destroy_authorizer(ac, osd->o_auth.authorizer); |
f24e9980 | 803 | kfree(osd); |
79494d1b | 804 | } |
f24e9980 SW |
805 | } |
806 | ||
807 | /* | |
808 | * remove an osd from our map | |
809 | */ | |
f5a2041b | 810 | static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd) |
f24e9980 | 811 | { |
f5a2041b | 812 | dout("__remove_osd %p\n", osd); |
f24e9980 SW |
813 | BUG_ON(!list_empty(&osd->o_requests)); |
814 | rb_erase(&osd->o_node, &osdc->osds); | |
f5a2041b | 815 | list_del_init(&osd->o_osd_lru); |
f24e9980 SW |
816 | ceph_con_close(&osd->o_con); |
817 | put_osd(osd); | |
818 | } | |
819 | ||
aca420bc SW |
820 | static void remove_all_osds(struct ceph_osd_client *osdc) |
821 | { | |
048a9d2d | 822 | dout("%s %p\n", __func__, osdc); |
aca420bc SW |
823 | mutex_lock(&osdc->request_mutex); |
824 | while (!RB_EMPTY_ROOT(&osdc->osds)) { | |
825 | struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds), | |
826 | struct ceph_osd, o_node); | |
827 | __remove_osd(osdc, osd); | |
828 | } | |
829 | mutex_unlock(&osdc->request_mutex); | |
830 | } | |
831 | ||
f5a2041b YS |
832 | static void __move_osd_to_lru(struct ceph_osd_client *osdc, |
833 | struct ceph_osd *osd) | |
834 | { | |
835 | dout("__move_osd_to_lru %p\n", osd); | |
836 | BUG_ON(!list_empty(&osd->o_osd_lru)); | |
837 | list_add_tail(&osd->o_osd_lru, &osdc->osd_lru); | |
3d14c5d2 | 838 | osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl * HZ; |
f5a2041b YS |
839 | } |
840 | ||
841 | static void __remove_osd_from_lru(struct ceph_osd *osd) | |
842 | { | |
843 | dout("__remove_osd_from_lru %p\n", osd); | |
844 | if (!list_empty(&osd->o_osd_lru)) | |
845 | list_del_init(&osd->o_osd_lru); | |
846 | } | |
847 | ||
aca420bc | 848 | static void remove_old_osds(struct ceph_osd_client *osdc) |
f5a2041b YS |
849 | { |
850 | struct ceph_osd *osd, *nosd; | |
851 | ||
852 | dout("__remove_old_osds %p\n", osdc); | |
853 | mutex_lock(&osdc->request_mutex); | |
854 | list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) { | |
aca420bc | 855 | if (time_before(jiffies, osd->lru_ttl)) |
f5a2041b YS |
856 | break; |
857 | __remove_osd(osdc, osd); | |
858 | } | |
859 | mutex_unlock(&osdc->request_mutex); | |
860 | } | |
861 | ||
f24e9980 SW |
862 | /* |
863 | * reset osd connect | |
864 | */ | |
f5a2041b | 865 | static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd) |
f24e9980 | 866 | { |
c3acb181 | 867 | struct ceph_entity_addr *peer_addr; |
f24e9980 | 868 | |
f5a2041b | 869 | dout("__reset_osd %p osd%d\n", osd, osd->o_osd); |
a40c4f10 YS |
870 | if (list_empty(&osd->o_requests) && |
871 | list_empty(&osd->o_linger_requests)) { | |
f5a2041b | 872 | __remove_osd(osdc, osd); |
c3acb181 AE |
873 | |
874 | return -ENODEV; | |
875 | } | |
876 | ||
877 | peer_addr = &osdc->osdmap->osd_addr[osd->o_osd]; | |
878 | if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) && | |
879 | !ceph_con_opened(&osd->o_con)) { | |
880 | struct ceph_osd_request *req; | |
881 | ||
87b315a5 SW |
882 | dout(" osd addr hasn't changed and connection never opened," |
883 | " letting msgr retry"); | |
884 | /* touch each r_stamp for handle_timeout()'s benfit */ | |
885 | list_for_each_entry(req, &osd->o_requests, r_osd_item) | |
886 | req->r_stamp = jiffies; | |
c3acb181 AE |
887 | |
888 | return -EAGAIN; | |
f24e9980 | 889 | } |
c3acb181 AE |
890 | |
891 | ceph_con_close(&osd->o_con); | |
892 | ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr); | |
893 | osd->o_incarnation++; | |
894 | ||
895 | return 0; | |
f24e9980 SW |
896 | } |
897 | ||
898 | static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new) | |
899 | { | |
900 | struct rb_node **p = &osdc->osds.rb_node; | |
901 | struct rb_node *parent = NULL; | |
902 | struct ceph_osd *osd = NULL; | |
903 | ||
aca420bc | 904 | dout("__insert_osd %p osd%d\n", new, new->o_osd); |
f24e9980 SW |
905 | while (*p) { |
906 | parent = *p; | |
907 | osd = rb_entry(parent, struct ceph_osd, o_node); | |
908 | if (new->o_osd < osd->o_osd) | |
909 | p = &(*p)->rb_left; | |
910 | else if (new->o_osd > osd->o_osd) | |
911 | p = &(*p)->rb_right; | |
912 | else | |
913 | BUG(); | |
914 | } | |
915 | ||
916 | rb_link_node(&new->o_node, parent, p); | |
917 | rb_insert_color(&new->o_node, &osdc->osds); | |
918 | } | |
919 | ||
920 | static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o) | |
921 | { | |
922 | struct ceph_osd *osd; | |
923 | struct rb_node *n = osdc->osds.rb_node; | |
924 | ||
925 | while (n) { | |
926 | osd = rb_entry(n, struct ceph_osd, o_node); | |
927 | if (o < osd->o_osd) | |
928 | n = n->rb_left; | |
929 | else if (o > osd->o_osd) | |
930 | n = n->rb_right; | |
931 | else | |
932 | return osd; | |
933 | } | |
934 | return NULL; | |
935 | } | |
936 | ||
422d2cb8 YS |
937 | static void __schedule_osd_timeout(struct ceph_osd_client *osdc) |
938 | { | |
939 | schedule_delayed_work(&osdc->timeout_work, | |
3d14c5d2 | 940 | osdc->client->options->osd_keepalive_timeout * HZ); |
422d2cb8 YS |
941 | } |
942 | ||
943 | static void __cancel_osd_timeout(struct ceph_osd_client *osdc) | |
944 | { | |
945 | cancel_delayed_work(&osdc->timeout_work); | |
946 | } | |
f24e9980 SW |
947 | |
948 | /* | |
949 | * Register request, assign tid. If this is the first request, set up | |
950 | * the timeout event. | |
951 | */ | |
a40c4f10 YS |
952 | static void __register_request(struct ceph_osd_client *osdc, |
953 | struct ceph_osd_request *req) | |
f24e9980 | 954 | { |
f24e9980 | 955 | req->r_tid = ++osdc->last_tid; |
6df058c0 | 956 | req->r_request->hdr.tid = cpu_to_le64(req->r_tid); |
77f38e0e | 957 | dout("__register_request %p tid %lld\n", req, req->r_tid); |
f24e9980 SW |
958 | __insert_request(osdc, req); |
959 | ceph_osdc_get_request(req); | |
960 | osdc->num_requests++; | |
f24e9980 | 961 | if (osdc->num_requests == 1) { |
422d2cb8 YS |
962 | dout(" first request, scheduling timeout\n"); |
963 | __schedule_osd_timeout(osdc); | |
f24e9980 | 964 | } |
a40c4f10 YS |
965 | } |
966 | ||
f24e9980 SW |
967 | /* |
968 | * called under osdc->request_mutex | |
969 | */ | |
970 | static void __unregister_request(struct ceph_osd_client *osdc, | |
971 | struct ceph_osd_request *req) | |
972 | { | |
35f9f8a0 SW |
973 | if (RB_EMPTY_NODE(&req->r_node)) { |
974 | dout("__unregister_request %p tid %lld not registered\n", | |
975 | req, req->r_tid); | |
976 | return; | |
977 | } | |
978 | ||
f24e9980 SW |
979 | dout("__unregister_request %p tid %lld\n", req, req->r_tid); |
980 | rb_erase(&req->r_node, &osdc->requests); | |
981 | osdc->num_requests--; | |
982 | ||
0ba6478d SW |
983 | if (req->r_osd) { |
984 | /* make sure the original request isn't in flight. */ | |
6740a845 | 985 | ceph_msg_revoke(req->r_request); |
0ba6478d SW |
986 | |
987 | list_del_init(&req->r_osd_item); | |
a40c4f10 YS |
988 | if (list_empty(&req->r_osd->o_requests) && |
989 | list_empty(&req->r_osd->o_linger_requests)) { | |
990 | dout("moving osd to %p lru\n", req->r_osd); | |
f5a2041b | 991 | __move_osd_to_lru(osdc, req->r_osd); |
a40c4f10 | 992 | } |
fbdb9190 | 993 | if (list_empty(&req->r_linger_item)) |
a40c4f10 | 994 | req->r_osd = NULL; |
0ba6478d | 995 | } |
f24e9980 | 996 | |
7d5f2481 | 997 | list_del_init(&req->r_req_lru_item); |
f24e9980 SW |
998 | ceph_osdc_put_request(req); |
999 | ||
422d2cb8 YS |
1000 | if (osdc->num_requests == 0) { |
1001 | dout(" no requests, canceling timeout\n"); | |
1002 | __cancel_osd_timeout(osdc); | |
f24e9980 SW |
1003 | } |
1004 | } | |
1005 | ||
1006 | /* | |
1007 | * Cancel a previously queued request message | |
1008 | */ | |
1009 | static void __cancel_request(struct ceph_osd_request *req) | |
1010 | { | |
6bc18876 | 1011 | if (req->r_sent && req->r_osd) { |
6740a845 | 1012 | ceph_msg_revoke(req->r_request); |
f24e9980 SW |
1013 | req->r_sent = 0; |
1014 | } | |
1015 | } | |
1016 | ||
a40c4f10 YS |
1017 | static void __register_linger_request(struct ceph_osd_client *osdc, |
1018 | struct ceph_osd_request *req) | |
1019 | { | |
1020 | dout("__register_linger_request %p\n", req); | |
1021 | list_add_tail(&req->r_linger_item, &osdc->req_linger); | |
6194ea89 SW |
1022 | if (req->r_osd) |
1023 | list_add_tail(&req->r_linger_osd, | |
1024 | &req->r_osd->o_linger_requests); | |
a40c4f10 YS |
1025 | } |
1026 | ||
1027 | static void __unregister_linger_request(struct ceph_osd_client *osdc, | |
1028 | struct ceph_osd_request *req) | |
1029 | { | |
1030 | dout("__unregister_linger_request %p\n", req); | |
61c74035 | 1031 | list_del_init(&req->r_linger_item); |
a40c4f10 | 1032 | if (req->r_osd) { |
a40c4f10 YS |
1033 | list_del_init(&req->r_linger_osd); |
1034 | ||
1035 | if (list_empty(&req->r_osd->o_requests) && | |
1036 | list_empty(&req->r_osd->o_linger_requests)) { | |
1037 | dout("moving osd to %p lru\n", req->r_osd); | |
1038 | __move_osd_to_lru(osdc, req->r_osd); | |
1039 | } | |
fbdb9190 SW |
1040 | if (list_empty(&req->r_osd_item)) |
1041 | req->r_osd = NULL; | |
a40c4f10 YS |
1042 | } |
1043 | } | |
1044 | ||
1045 | void ceph_osdc_unregister_linger_request(struct ceph_osd_client *osdc, | |
1046 | struct ceph_osd_request *req) | |
1047 | { | |
1048 | mutex_lock(&osdc->request_mutex); | |
1049 | if (req->r_linger) { | |
1050 | __unregister_linger_request(osdc, req); | |
1051 | ceph_osdc_put_request(req); | |
1052 | } | |
1053 | mutex_unlock(&osdc->request_mutex); | |
1054 | } | |
1055 | EXPORT_SYMBOL(ceph_osdc_unregister_linger_request); | |
1056 | ||
1057 | void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc, | |
1058 | struct ceph_osd_request *req) | |
1059 | { | |
1060 | if (!req->r_linger) { | |
1061 | dout("set_request_linger %p\n", req); | |
1062 | req->r_linger = 1; | |
1063 | /* | |
1064 | * caller is now responsible for calling | |
1065 | * unregister_linger_request | |
1066 | */ | |
1067 | ceph_osdc_get_request(req); | |
1068 | } | |
1069 | } | |
1070 | EXPORT_SYMBOL(ceph_osdc_set_request_linger); | |
1071 | ||
f24e9980 SW |
1072 | /* |
1073 | * Pick an osd (the first 'up' osd in the pg), allocate the osd struct | |
1074 | * (as needed), and set the request r_osd appropriately. If there is | |
25985edc | 1075 | * no up osd, set r_osd to NULL. Move the request to the appropriate list |
6f6c7006 | 1076 | * (unsent, homeless) or leave on in-flight lru. |
f24e9980 SW |
1077 | * |
1078 | * Return 0 if unchanged, 1 if changed, or negative on error. | |
1079 | * | |
1080 | * Caller should hold map_sem for read and request_mutex. | |
1081 | */ | |
6f6c7006 | 1082 | static int __map_request(struct ceph_osd_client *osdc, |
38d6453c | 1083 | struct ceph_osd_request *req, int force_resend) |
f24e9980 | 1084 | { |
5b191d99 | 1085 | struct ceph_pg pgid; |
d85b7056 SW |
1086 | int acting[CEPH_PG_MAX_SIZE]; |
1087 | int o = -1, num = 0; | |
f24e9980 | 1088 | int err; |
f24e9980 | 1089 | |
6f6c7006 | 1090 | dout("map_request %p tid %lld\n", req, req->r_tid); |
41766f87 AE |
1091 | err = ceph_calc_ceph_pg(&pgid, req->r_oid, osdc->osdmap, |
1092 | ceph_file_layout_pg_pool(req->r_file_layout)); | |
6f6c7006 SW |
1093 | if (err) { |
1094 | list_move(&req->r_req_lru_item, &osdc->req_notarget); | |
f24e9980 | 1095 | return err; |
6f6c7006 | 1096 | } |
7740a42f SW |
1097 | req->r_pgid = pgid; |
1098 | ||
d85b7056 SW |
1099 | err = ceph_calc_pg_acting(osdc->osdmap, pgid, acting); |
1100 | if (err > 0) { | |
1101 | o = acting[0]; | |
1102 | num = err; | |
1103 | } | |
f24e9980 | 1104 | |
38d6453c SW |
1105 | if ((!force_resend && |
1106 | req->r_osd && req->r_osd->o_osd == o && | |
d85b7056 SW |
1107 | req->r_sent >= req->r_osd->o_incarnation && |
1108 | req->r_num_pg_osds == num && | |
1109 | memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) || | |
f24e9980 SW |
1110 | (req->r_osd == NULL && o == -1)) |
1111 | return 0; /* no change */ | |
1112 | ||
5b191d99 SW |
1113 | dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n", |
1114 | req->r_tid, pgid.pool, pgid.seed, o, | |
f24e9980 SW |
1115 | req->r_osd ? req->r_osd->o_osd : -1); |
1116 | ||
d85b7056 SW |
1117 | /* record full pg acting set */ |
1118 | memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num); | |
1119 | req->r_num_pg_osds = num; | |
1120 | ||
f24e9980 SW |
1121 | if (req->r_osd) { |
1122 | __cancel_request(req); | |
1123 | list_del_init(&req->r_osd_item); | |
f24e9980 SW |
1124 | req->r_osd = NULL; |
1125 | } | |
1126 | ||
1127 | req->r_osd = __lookup_osd(osdc, o); | |
1128 | if (!req->r_osd && o >= 0) { | |
c99eb1c7 | 1129 | err = -ENOMEM; |
e10006f8 | 1130 | req->r_osd = create_osd(osdc, o); |
6f6c7006 SW |
1131 | if (!req->r_osd) { |
1132 | list_move(&req->r_req_lru_item, &osdc->req_notarget); | |
c99eb1c7 | 1133 | goto out; |
6f6c7006 | 1134 | } |
f24e9980 | 1135 | |
6f6c7006 | 1136 | dout("map_request osd %p is osd%d\n", req->r_osd, o); |
f24e9980 SW |
1137 | __insert_osd(osdc, req->r_osd); |
1138 | ||
b7a9e5dd SW |
1139 | ceph_con_open(&req->r_osd->o_con, |
1140 | CEPH_ENTITY_TYPE_OSD, o, | |
1141 | &osdc->osdmap->osd_addr[o]); | |
f24e9980 SW |
1142 | } |
1143 | ||
f5a2041b YS |
1144 | if (req->r_osd) { |
1145 | __remove_osd_from_lru(req->r_osd); | |
ad885927 AE |
1146 | list_add_tail(&req->r_osd_item, &req->r_osd->o_requests); |
1147 | list_move_tail(&req->r_req_lru_item, &osdc->req_unsent); | |
6f6c7006 | 1148 | } else { |
ad885927 | 1149 | list_move_tail(&req->r_req_lru_item, &osdc->req_notarget); |
f5a2041b | 1150 | } |
d85b7056 | 1151 | err = 1; /* osd or pg changed */ |
f24e9980 SW |
1152 | |
1153 | out: | |
f24e9980 SW |
1154 | return err; |
1155 | } | |
1156 | ||
1157 | /* | |
1158 | * caller should hold map_sem (for read) and request_mutex | |
1159 | */ | |
56e925b6 SW |
1160 | static void __send_request(struct ceph_osd_client *osdc, |
1161 | struct ceph_osd_request *req) | |
f24e9980 | 1162 | { |
1b83bef2 | 1163 | void *p; |
f24e9980 | 1164 | |
1b83bef2 SW |
1165 | dout("send_request %p tid %llu to osd%d flags %d pg %lld.%x\n", |
1166 | req, req->r_tid, req->r_osd->o_osd, req->r_flags, | |
1167 | (unsigned long long)req->r_pgid.pool, req->r_pgid.seed); | |
1168 | ||
1169 | /* fill in message content that changes each time we send it */ | |
1170 | put_unaligned_le32(osdc->osdmap->epoch, req->r_request_osdmap_epoch); | |
1171 | put_unaligned_le32(req->r_flags, req->r_request_flags); | |
1172 | put_unaligned_le64(req->r_pgid.pool, req->r_request_pool); | |
1173 | p = req->r_request_pgid; | |
1174 | ceph_encode_64(&p, req->r_pgid.pool); | |
1175 | ceph_encode_32(&p, req->r_pgid.seed); | |
1176 | put_unaligned_le64(1, req->r_request_attempts); /* FIXME */ | |
1177 | memcpy(req->r_request_reassert_version, &req->r_reassert_version, | |
1178 | sizeof(req->r_reassert_version)); | |
2169aea6 | 1179 | |
3dd72fc0 | 1180 | req->r_stamp = jiffies; |
07a27e22 | 1181 | list_move_tail(&req->r_req_lru_item, &osdc->req_lru); |
f24e9980 SW |
1182 | |
1183 | ceph_msg_get(req->r_request); /* send consumes a ref */ | |
1184 | ceph_con_send(&req->r_osd->o_con, req->r_request); | |
1185 | req->r_sent = req->r_osd->o_incarnation; | |
f24e9980 SW |
1186 | } |
1187 | ||
6f6c7006 SW |
1188 | /* |
1189 | * Send any requests in the queue (req_unsent). | |
1190 | */ | |
f9d25199 | 1191 | static void __send_queued(struct ceph_osd_client *osdc) |
6f6c7006 SW |
1192 | { |
1193 | struct ceph_osd_request *req, *tmp; | |
1194 | ||
f9d25199 AE |
1195 | dout("__send_queued\n"); |
1196 | list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item) | |
6f6c7006 | 1197 | __send_request(osdc, req); |
6f6c7006 SW |
1198 | } |
1199 | ||
f24e9980 SW |
1200 | /* |
1201 | * Timeout callback, called every N seconds when 1 or more osd | |
1202 | * requests has been active for more than N seconds. When this | |
1203 | * happens, we ping all OSDs with requests who have timed out to | |
1204 | * ensure any communications channel reset is detected. Reset the | |
1205 | * request timeouts another N seconds in the future as we go. | |
1206 | * Reschedule the timeout event another N seconds in future (unless | |
1207 | * there are no open requests). | |
1208 | */ | |
1209 | static void handle_timeout(struct work_struct *work) | |
1210 | { | |
1211 | struct ceph_osd_client *osdc = | |
1212 | container_of(work, struct ceph_osd_client, timeout_work.work); | |
83aff95e | 1213 | struct ceph_osd_request *req; |
f24e9980 | 1214 | struct ceph_osd *osd; |
422d2cb8 | 1215 | unsigned long keepalive = |
3d14c5d2 | 1216 | osdc->client->options->osd_keepalive_timeout * HZ; |
422d2cb8 | 1217 | struct list_head slow_osds; |
f24e9980 SW |
1218 | dout("timeout\n"); |
1219 | down_read(&osdc->map_sem); | |
1220 | ||
1221 | ceph_monc_request_next_osdmap(&osdc->client->monc); | |
1222 | ||
1223 | mutex_lock(&osdc->request_mutex); | |
f24e9980 | 1224 | |
422d2cb8 YS |
1225 | /* |
1226 | * ping osds that are a bit slow. this ensures that if there | |
1227 | * is a break in the TCP connection we will notice, and reopen | |
1228 | * a connection with that osd (from the fault callback). | |
1229 | */ | |
1230 | INIT_LIST_HEAD(&slow_osds); | |
1231 | list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) { | |
3dd72fc0 | 1232 | if (time_before(jiffies, req->r_stamp + keepalive)) |
422d2cb8 YS |
1233 | break; |
1234 | ||
1235 | osd = req->r_osd; | |
1236 | BUG_ON(!osd); | |
1237 | dout(" tid %llu is slow, will send keepalive on osd%d\n", | |
f24e9980 | 1238 | req->r_tid, osd->o_osd); |
422d2cb8 YS |
1239 | list_move_tail(&osd->o_keepalive_item, &slow_osds); |
1240 | } | |
1241 | while (!list_empty(&slow_osds)) { | |
1242 | osd = list_entry(slow_osds.next, struct ceph_osd, | |
1243 | o_keepalive_item); | |
1244 | list_del_init(&osd->o_keepalive_item); | |
f24e9980 SW |
1245 | ceph_con_keepalive(&osd->o_con); |
1246 | } | |
1247 | ||
422d2cb8 | 1248 | __schedule_osd_timeout(osdc); |
f9d25199 | 1249 | __send_queued(osdc); |
f24e9980 | 1250 | mutex_unlock(&osdc->request_mutex); |
f24e9980 SW |
1251 | up_read(&osdc->map_sem); |
1252 | } | |
1253 | ||
f5a2041b YS |
1254 | static void handle_osds_timeout(struct work_struct *work) |
1255 | { | |
1256 | struct ceph_osd_client *osdc = | |
1257 | container_of(work, struct ceph_osd_client, | |
1258 | osds_timeout_work.work); | |
1259 | unsigned long delay = | |
3d14c5d2 | 1260 | osdc->client->options->osd_idle_ttl * HZ >> 2; |
f5a2041b YS |
1261 | |
1262 | dout("osds timeout\n"); | |
1263 | down_read(&osdc->map_sem); | |
aca420bc | 1264 | remove_old_osds(osdc); |
f5a2041b YS |
1265 | up_read(&osdc->map_sem); |
1266 | ||
1267 | schedule_delayed_work(&osdc->osds_timeout_work, | |
1268 | round_jiffies_relative(delay)); | |
1269 | } | |
1270 | ||
25845472 SW |
1271 | static void complete_request(struct ceph_osd_request *req) |
1272 | { | |
1273 | if (req->r_safe_callback) | |
1274 | req->r_safe_callback(req, NULL); | |
1275 | complete_all(&req->r_safe_completion); /* fsync waiter */ | |
1276 | } | |
1277 | ||
1b83bef2 SW |
1278 | static int __decode_pgid(void **p, void *end, struct ceph_pg *pgid) |
1279 | { | |
1280 | __u8 v; | |
1281 | ||
1282 | ceph_decode_need(p, end, 1 + 8 + 4 + 4, bad); | |
1283 | v = ceph_decode_8(p); | |
1284 | if (v > 1) { | |
1285 | pr_warning("do not understand pg encoding %d > 1", v); | |
1286 | return -EINVAL; | |
1287 | } | |
1288 | pgid->pool = ceph_decode_64(p); | |
1289 | pgid->seed = ceph_decode_32(p); | |
1290 | *p += 4; | |
1291 | return 0; | |
1292 | ||
1293 | bad: | |
1294 | pr_warning("incomplete pg encoding"); | |
1295 | return -EINVAL; | |
1296 | } | |
1297 | ||
f24e9980 SW |
1298 | /* |
1299 | * handle osd op reply. either call the callback if it is specified, | |
1300 | * or do the completion to wake up the waiting thread. | |
1301 | */ | |
350b1c32 SW |
1302 | static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg, |
1303 | struct ceph_connection *con) | |
f24e9980 | 1304 | { |
1b83bef2 | 1305 | void *p, *end; |
f24e9980 SW |
1306 | struct ceph_osd_request *req; |
1307 | u64 tid; | |
1b83bef2 SW |
1308 | int object_len; |
1309 | int numops, payload_len, flags; | |
0ceed5db | 1310 | s32 result; |
1b83bef2 SW |
1311 | s32 retry_attempt; |
1312 | struct ceph_pg pg; | |
1313 | int err; | |
1314 | u32 reassert_epoch; | |
1315 | u64 reassert_version; | |
1316 | u32 osdmap_epoch; | |
0d5af164 | 1317 | int already_completed; |
1b83bef2 | 1318 | int i; |
f24e9980 | 1319 | |
6df058c0 | 1320 | tid = le64_to_cpu(msg->hdr.tid); |
1b83bef2 SW |
1321 | dout("handle_reply %p tid %llu\n", msg, tid); |
1322 | ||
1323 | p = msg->front.iov_base; | |
1324 | end = p + msg->front.iov_len; | |
1325 | ||
1326 | ceph_decode_need(&p, end, 4, bad); | |
1327 | object_len = ceph_decode_32(&p); | |
1328 | ceph_decode_need(&p, end, object_len, bad); | |
1329 | p += object_len; | |
1330 | ||
1331 | err = __decode_pgid(&p, end, &pg); | |
1332 | if (err) | |
f24e9980 | 1333 | goto bad; |
1b83bef2 SW |
1334 | |
1335 | ceph_decode_need(&p, end, 8 + 4 + 4 + 8 + 4, bad); | |
1336 | flags = ceph_decode_64(&p); | |
1337 | result = ceph_decode_32(&p); | |
1338 | reassert_epoch = ceph_decode_32(&p); | |
1339 | reassert_version = ceph_decode_64(&p); | |
1340 | osdmap_epoch = ceph_decode_32(&p); | |
1341 | ||
f24e9980 SW |
1342 | /* lookup */ |
1343 | mutex_lock(&osdc->request_mutex); | |
1344 | req = __lookup_request(osdc, tid); | |
1345 | if (req == NULL) { | |
1346 | dout("handle_reply tid %llu dne\n", tid); | |
1347 | mutex_unlock(&osdc->request_mutex); | |
1348 | return; | |
1349 | } | |
1350 | ceph_osdc_get_request(req); | |
1b83bef2 SW |
1351 | |
1352 | dout("handle_reply %p tid %llu req %p result %d\n", msg, tid, | |
1353 | req, result); | |
1354 | ||
1355 | ceph_decode_need(&p, end, 4, bad); | |
1356 | numops = ceph_decode_32(&p); | |
1357 | if (numops > CEPH_OSD_MAX_OP) | |
1358 | goto bad_put; | |
1359 | if (numops != req->r_num_ops) | |
1360 | goto bad_put; | |
1361 | payload_len = 0; | |
1362 | ceph_decode_need(&p, end, numops * sizeof(struct ceph_osd_op), bad); | |
1363 | for (i = 0; i < numops; i++) { | |
1364 | struct ceph_osd_op *op = p; | |
1365 | int len; | |
1366 | ||
1367 | len = le32_to_cpu(op->payload_len); | |
1368 | req->r_reply_op_len[i] = len; | |
1369 | dout(" op %d has %d bytes\n", i, len); | |
1370 | payload_len += len; | |
1371 | p += sizeof(*op); | |
1372 | } | |
1373 | if (payload_len != le32_to_cpu(msg->hdr.data_len)) { | |
1374 | pr_warning("sum of op payload lens %d != data_len %d", | |
1375 | payload_len, le32_to_cpu(msg->hdr.data_len)); | |
1376 | goto bad_put; | |
1377 | } | |
1378 | ||
1379 | ceph_decode_need(&p, end, 4 + numops * 4, bad); | |
1380 | retry_attempt = ceph_decode_32(&p); | |
1381 | for (i = 0; i < numops; i++) | |
1382 | req->r_reply_op_result[i] = ceph_decode_32(&p); | |
f24e9980 | 1383 | |
350b1c32 | 1384 | /* |
0d59ab81 | 1385 | * if this connection filled our message, drop our reference now, to |
350b1c32 SW |
1386 | * avoid a (safe but slower) revoke later. |
1387 | */ | |
0d59ab81 | 1388 | if (req->r_con_filling_msg == con && req->r_reply == msg) { |
c16e7869 | 1389 | dout(" dropping con_filling_msg ref %p\n", con); |
0d59ab81 | 1390 | req->r_con_filling_msg = NULL; |
0d47766f | 1391 | con->ops->put(con); |
350b1c32 SW |
1392 | } |
1393 | ||
f24e9980 | 1394 | if (!req->r_got_reply) { |
95c96174 | 1395 | unsigned int bytes; |
f24e9980 | 1396 | |
1b83bef2 | 1397 | req->r_result = result; |
f24e9980 SW |
1398 | bytes = le32_to_cpu(msg->hdr.data_len); |
1399 | dout("handle_reply result %d bytes %d\n", req->r_result, | |
1400 | bytes); | |
1401 | if (req->r_result == 0) | |
1402 | req->r_result = bytes; | |
1403 | ||
1404 | /* in case this is a write and we need to replay, */ | |
1b83bef2 SW |
1405 | req->r_reassert_version.epoch = cpu_to_le32(reassert_epoch); |
1406 | req->r_reassert_version.version = cpu_to_le64(reassert_version); | |
f24e9980 SW |
1407 | |
1408 | req->r_got_reply = 1; | |
1409 | } else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) { | |
1410 | dout("handle_reply tid %llu dup ack\n", tid); | |
34b43a56 | 1411 | mutex_unlock(&osdc->request_mutex); |
f24e9980 SW |
1412 | goto done; |
1413 | } | |
1414 | ||
1415 | dout("handle_reply tid %llu flags %d\n", tid, flags); | |
1416 | ||
a40c4f10 YS |
1417 | if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK)) |
1418 | __register_linger_request(osdc, req); | |
1419 | ||
f24e9980 | 1420 | /* either this is a read, or we got the safe response */ |
0ceed5db SW |
1421 | if (result < 0 || |
1422 | (flags & CEPH_OSD_FLAG_ONDISK) || | |
f24e9980 SW |
1423 | ((flags & CEPH_OSD_FLAG_WRITE) == 0)) |
1424 | __unregister_request(osdc, req); | |
1425 | ||
0d5af164 AE |
1426 | already_completed = req->r_completed; |
1427 | req->r_completed = 1; | |
f24e9980 | 1428 | mutex_unlock(&osdc->request_mutex); |
0d5af164 AE |
1429 | if (already_completed) |
1430 | goto done; | |
f24e9980 SW |
1431 | |
1432 | if (req->r_callback) | |
1433 | req->r_callback(req, msg); | |
1434 | else | |
03066f23 | 1435 | complete_all(&req->r_completion); |
f24e9980 | 1436 | |
25845472 SW |
1437 | if (flags & CEPH_OSD_FLAG_ONDISK) |
1438 | complete_request(req); | |
f24e9980 SW |
1439 | |
1440 | done: | |
a40c4f10 | 1441 | dout("req=%p req->r_linger=%d\n", req, req->r_linger); |
f24e9980 SW |
1442 | ceph_osdc_put_request(req); |
1443 | return; | |
1444 | ||
1b83bef2 SW |
1445 | bad_put: |
1446 | ceph_osdc_put_request(req); | |
f24e9980 | 1447 | bad: |
1b83bef2 SW |
1448 | pr_err("corrupt osd_op_reply got %d %d\n", |
1449 | (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len)); | |
9ec7cab1 | 1450 | ceph_msg_dump(msg); |
f24e9980 SW |
1451 | } |
1452 | ||
6f6c7006 | 1453 | static void reset_changed_osds(struct ceph_osd_client *osdc) |
f24e9980 | 1454 | { |
f24e9980 | 1455 | struct rb_node *p, *n; |
f24e9980 | 1456 | |
6f6c7006 SW |
1457 | for (p = rb_first(&osdc->osds); p; p = n) { |
1458 | struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node); | |
f24e9980 | 1459 | |
6f6c7006 SW |
1460 | n = rb_next(p); |
1461 | if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) || | |
1462 | memcmp(&osd->o_con.peer_addr, | |
1463 | ceph_osd_addr(osdc->osdmap, | |
1464 | osd->o_osd), | |
1465 | sizeof(struct ceph_entity_addr)) != 0) | |
1466 | __reset_osd(osdc, osd); | |
f24e9980 | 1467 | } |
422d2cb8 YS |
1468 | } |
1469 | ||
1470 | /* | |
6f6c7006 SW |
1471 | * Requeue requests whose mapping to an OSD has changed. If requests map to |
1472 | * no osd, request a new map. | |
422d2cb8 | 1473 | * |
e6d50f67 | 1474 | * Caller should hold map_sem for read. |
422d2cb8 | 1475 | */ |
38d6453c | 1476 | static void kick_requests(struct ceph_osd_client *osdc, int force_resend) |
422d2cb8 | 1477 | { |
a40c4f10 | 1478 | struct ceph_osd_request *req, *nreq; |
6f6c7006 SW |
1479 | struct rb_node *p; |
1480 | int needmap = 0; | |
1481 | int err; | |
422d2cb8 | 1482 | |
38d6453c | 1483 | dout("kick_requests %s\n", force_resend ? " (force resend)" : ""); |
422d2cb8 | 1484 | mutex_lock(&osdc->request_mutex); |
6194ea89 | 1485 | for (p = rb_first(&osdc->requests); p; ) { |
6f6c7006 | 1486 | req = rb_entry(p, struct ceph_osd_request, r_node); |
6194ea89 | 1487 | p = rb_next(p); |
ab60b16d AE |
1488 | |
1489 | /* | |
1490 | * For linger requests that have not yet been | |
1491 | * registered, move them to the linger list; they'll | |
1492 | * be sent to the osd in the loop below. Unregister | |
1493 | * the request before re-registering it as a linger | |
1494 | * request to ensure the __map_request() below | |
1495 | * will decide it needs to be sent. | |
1496 | */ | |
1497 | if (req->r_linger && list_empty(&req->r_linger_item)) { | |
1498 | dout("%p tid %llu restart on osd%d\n", | |
1499 | req, req->r_tid, | |
1500 | req->r_osd ? req->r_osd->o_osd : -1); | |
1501 | __unregister_request(osdc, req); | |
1502 | __register_linger_request(osdc, req); | |
1503 | continue; | |
1504 | } | |
1505 | ||
38d6453c | 1506 | err = __map_request(osdc, req, force_resend); |
6f6c7006 SW |
1507 | if (err < 0) |
1508 | continue; /* error */ | |
1509 | if (req->r_osd == NULL) { | |
1510 | dout("%p tid %llu maps to no osd\n", req, req->r_tid); | |
1511 | needmap++; /* request a newer map */ | |
1512 | } else if (err > 0) { | |
6194ea89 SW |
1513 | if (!req->r_linger) { |
1514 | dout("%p tid %llu requeued on osd%d\n", req, | |
1515 | req->r_tid, | |
1516 | req->r_osd ? req->r_osd->o_osd : -1); | |
a40c4f10 | 1517 | req->r_flags |= CEPH_OSD_FLAG_RETRY; |
6194ea89 SW |
1518 | } |
1519 | } | |
a40c4f10 YS |
1520 | } |
1521 | ||
1522 | list_for_each_entry_safe(req, nreq, &osdc->req_linger, | |
1523 | r_linger_item) { | |
1524 | dout("linger req=%p req->r_osd=%p\n", req, req->r_osd); | |
1525 | ||
38d6453c | 1526 | err = __map_request(osdc, req, force_resend); |
ab60b16d | 1527 | dout("__map_request returned %d\n", err); |
a40c4f10 YS |
1528 | if (err == 0) |
1529 | continue; /* no change and no osd was specified */ | |
1530 | if (err < 0) | |
1531 | continue; /* hrm! */ | |
1532 | if (req->r_osd == NULL) { | |
1533 | dout("tid %llu maps to no valid osd\n", req->r_tid); | |
1534 | needmap++; /* request a newer map */ | |
1535 | continue; | |
6f6c7006 | 1536 | } |
a40c4f10 YS |
1537 | |
1538 | dout("kicking lingering %p tid %llu osd%d\n", req, req->r_tid, | |
1539 | req->r_osd ? req->r_osd->o_osd : -1); | |
a40c4f10 | 1540 | __register_request(osdc, req); |
c89ce05e | 1541 | __unregister_linger_request(osdc, req); |
6f6c7006 | 1542 | } |
f24e9980 SW |
1543 | mutex_unlock(&osdc->request_mutex); |
1544 | ||
1545 | if (needmap) { | |
1546 | dout("%d requests for down osds, need new map\n", needmap); | |
1547 | ceph_monc_request_next_osdmap(&osdc->client->monc); | |
1548 | } | |
e6d50f67 | 1549 | reset_changed_osds(osdc); |
422d2cb8 | 1550 | } |
6f6c7006 SW |
1551 | |
1552 | ||
f24e9980 SW |
1553 | /* |
1554 | * Process updated osd map. | |
1555 | * | |
1556 | * The message contains any number of incremental and full maps, normally | |
1557 | * indicating some sort of topology change in the cluster. Kick requests | |
1558 | * off to different OSDs as needed. | |
1559 | */ | |
1560 | void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg) | |
1561 | { | |
1562 | void *p, *end, *next; | |
1563 | u32 nr_maps, maplen; | |
1564 | u32 epoch; | |
1565 | struct ceph_osdmap *newmap = NULL, *oldmap; | |
1566 | int err; | |
1567 | struct ceph_fsid fsid; | |
1568 | ||
1569 | dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0); | |
1570 | p = msg->front.iov_base; | |
1571 | end = p + msg->front.iov_len; | |
1572 | ||
1573 | /* verify fsid */ | |
1574 | ceph_decode_need(&p, end, sizeof(fsid), bad); | |
1575 | ceph_decode_copy(&p, &fsid, sizeof(fsid)); | |
0743304d SW |
1576 | if (ceph_check_fsid(osdc->client, &fsid) < 0) |
1577 | return; | |
f24e9980 SW |
1578 | |
1579 | down_write(&osdc->map_sem); | |
1580 | ||
1581 | /* incremental maps */ | |
1582 | ceph_decode_32_safe(&p, end, nr_maps, bad); | |
1583 | dout(" %d inc maps\n", nr_maps); | |
1584 | while (nr_maps > 0) { | |
1585 | ceph_decode_need(&p, end, 2*sizeof(u32), bad); | |
c89136ea SW |
1586 | epoch = ceph_decode_32(&p); |
1587 | maplen = ceph_decode_32(&p); | |
f24e9980 SW |
1588 | ceph_decode_need(&p, end, maplen, bad); |
1589 | next = p + maplen; | |
1590 | if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) { | |
1591 | dout("applying incremental map %u len %d\n", | |
1592 | epoch, maplen); | |
1593 | newmap = osdmap_apply_incremental(&p, next, | |
1594 | osdc->osdmap, | |
15d9882c | 1595 | &osdc->client->msgr); |
f24e9980 SW |
1596 | if (IS_ERR(newmap)) { |
1597 | err = PTR_ERR(newmap); | |
1598 | goto bad; | |
1599 | } | |
30dc6381 | 1600 | BUG_ON(!newmap); |
f24e9980 SW |
1601 | if (newmap != osdc->osdmap) { |
1602 | ceph_osdmap_destroy(osdc->osdmap); | |
1603 | osdc->osdmap = newmap; | |
1604 | } | |
38d6453c | 1605 | kick_requests(osdc, 0); |
f24e9980 SW |
1606 | } else { |
1607 | dout("ignoring incremental map %u len %d\n", | |
1608 | epoch, maplen); | |
1609 | } | |
1610 | p = next; | |
1611 | nr_maps--; | |
1612 | } | |
1613 | if (newmap) | |
1614 | goto done; | |
1615 | ||
1616 | /* full maps */ | |
1617 | ceph_decode_32_safe(&p, end, nr_maps, bad); | |
1618 | dout(" %d full maps\n", nr_maps); | |
1619 | while (nr_maps) { | |
1620 | ceph_decode_need(&p, end, 2*sizeof(u32), bad); | |
c89136ea SW |
1621 | epoch = ceph_decode_32(&p); |
1622 | maplen = ceph_decode_32(&p); | |
f24e9980 SW |
1623 | ceph_decode_need(&p, end, maplen, bad); |
1624 | if (nr_maps > 1) { | |
1625 | dout("skipping non-latest full map %u len %d\n", | |
1626 | epoch, maplen); | |
1627 | } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) { | |
1628 | dout("skipping full map %u len %d, " | |
1629 | "older than our %u\n", epoch, maplen, | |
1630 | osdc->osdmap->epoch); | |
1631 | } else { | |
38d6453c SW |
1632 | int skipped_map = 0; |
1633 | ||
f24e9980 SW |
1634 | dout("taking full map %u len %d\n", epoch, maplen); |
1635 | newmap = osdmap_decode(&p, p+maplen); | |
1636 | if (IS_ERR(newmap)) { | |
1637 | err = PTR_ERR(newmap); | |
1638 | goto bad; | |
1639 | } | |
30dc6381 | 1640 | BUG_ON(!newmap); |
f24e9980 SW |
1641 | oldmap = osdc->osdmap; |
1642 | osdc->osdmap = newmap; | |
38d6453c SW |
1643 | if (oldmap) { |
1644 | if (oldmap->epoch + 1 < newmap->epoch) | |
1645 | skipped_map = 1; | |
f24e9980 | 1646 | ceph_osdmap_destroy(oldmap); |
38d6453c SW |
1647 | } |
1648 | kick_requests(osdc, skipped_map); | |
f24e9980 SW |
1649 | } |
1650 | p += maplen; | |
1651 | nr_maps--; | |
1652 | } | |
1653 | ||
1654 | done: | |
1655 | downgrade_write(&osdc->map_sem); | |
1656 | ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch); | |
cd634fb6 SW |
1657 | |
1658 | /* | |
1659 | * subscribe to subsequent osdmap updates if full to ensure | |
1660 | * we find out when we are no longer full and stop returning | |
1661 | * ENOSPC. | |
1662 | */ | |
1663 | if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL)) | |
1664 | ceph_monc_request_next_osdmap(&osdc->client->monc); | |
1665 | ||
f9d25199 AE |
1666 | mutex_lock(&osdc->request_mutex); |
1667 | __send_queued(osdc); | |
1668 | mutex_unlock(&osdc->request_mutex); | |
f24e9980 | 1669 | up_read(&osdc->map_sem); |
03066f23 | 1670 | wake_up_all(&osdc->client->auth_wq); |
f24e9980 SW |
1671 | return; |
1672 | ||
1673 | bad: | |
1674 | pr_err("osdc handle_map corrupt msg\n"); | |
9ec7cab1 | 1675 | ceph_msg_dump(msg); |
f24e9980 SW |
1676 | up_write(&osdc->map_sem); |
1677 | return; | |
1678 | } | |
1679 | ||
a40c4f10 YS |
1680 | /* |
1681 | * watch/notify callback event infrastructure | |
1682 | * | |
1683 | * These callbacks are used both for watch and notify operations. | |
1684 | */ | |
1685 | static void __release_event(struct kref *kref) | |
1686 | { | |
1687 | struct ceph_osd_event *event = | |
1688 | container_of(kref, struct ceph_osd_event, kref); | |
1689 | ||
1690 | dout("__release_event %p\n", event); | |
1691 | kfree(event); | |
1692 | } | |
1693 | ||
1694 | static void get_event(struct ceph_osd_event *event) | |
1695 | { | |
1696 | kref_get(&event->kref); | |
1697 | } | |
1698 | ||
1699 | void ceph_osdc_put_event(struct ceph_osd_event *event) | |
1700 | { | |
1701 | kref_put(&event->kref, __release_event); | |
1702 | } | |
1703 | EXPORT_SYMBOL(ceph_osdc_put_event); | |
1704 | ||
1705 | static void __insert_event(struct ceph_osd_client *osdc, | |
1706 | struct ceph_osd_event *new) | |
1707 | { | |
1708 | struct rb_node **p = &osdc->event_tree.rb_node; | |
1709 | struct rb_node *parent = NULL; | |
1710 | struct ceph_osd_event *event = NULL; | |
1711 | ||
1712 | while (*p) { | |
1713 | parent = *p; | |
1714 | event = rb_entry(parent, struct ceph_osd_event, node); | |
1715 | if (new->cookie < event->cookie) | |
1716 | p = &(*p)->rb_left; | |
1717 | else if (new->cookie > event->cookie) | |
1718 | p = &(*p)->rb_right; | |
1719 | else | |
1720 | BUG(); | |
1721 | } | |
1722 | ||
1723 | rb_link_node(&new->node, parent, p); | |
1724 | rb_insert_color(&new->node, &osdc->event_tree); | |
1725 | } | |
1726 | ||
1727 | static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc, | |
1728 | u64 cookie) | |
1729 | { | |
1730 | struct rb_node **p = &osdc->event_tree.rb_node; | |
1731 | struct rb_node *parent = NULL; | |
1732 | struct ceph_osd_event *event = NULL; | |
1733 | ||
1734 | while (*p) { | |
1735 | parent = *p; | |
1736 | event = rb_entry(parent, struct ceph_osd_event, node); | |
1737 | if (cookie < event->cookie) | |
1738 | p = &(*p)->rb_left; | |
1739 | else if (cookie > event->cookie) | |
1740 | p = &(*p)->rb_right; | |
1741 | else | |
1742 | return event; | |
1743 | } | |
1744 | return NULL; | |
1745 | } | |
1746 | ||
1747 | static void __remove_event(struct ceph_osd_event *event) | |
1748 | { | |
1749 | struct ceph_osd_client *osdc = event->osdc; | |
1750 | ||
1751 | if (!RB_EMPTY_NODE(&event->node)) { | |
1752 | dout("__remove_event removed %p\n", event); | |
1753 | rb_erase(&event->node, &osdc->event_tree); | |
1754 | ceph_osdc_put_event(event); | |
1755 | } else { | |
1756 | dout("__remove_event didn't remove %p\n", event); | |
1757 | } | |
1758 | } | |
1759 | ||
1760 | int ceph_osdc_create_event(struct ceph_osd_client *osdc, | |
1761 | void (*event_cb)(u64, u64, u8, void *), | |
3c663bbd | 1762 | void *data, struct ceph_osd_event **pevent) |
a40c4f10 YS |
1763 | { |
1764 | struct ceph_osd_event *event; | |
1765 | ||
1766 | event = kmalloc(sizeof(*event), GFP_NOIO); | |
1767 | if (!event) | |
1768 | return -ENOMEM; | |
1769 | ||
1770 | dout("create_event %p\n", event); | |
1771 | event->cb = event_cb; | |
3c663bbd | 1772 | event->one_shot = 0; |
a40c4f10 YS |
1773 | event->data = data; |
1774 | event->osdc = osdc; | |
1775 | INIT_LIST_HEAD(&event->osd_node); | |
3ee5234d | 1776 | RB_CLEAR_NODE(&event->node); |
a40c4f10 YS |
1777 | kref_init(&event->kref); /* one ref for us */ |
1778 | kref_get(&event->kref); /* one ref for the caller */ | |
a40c4f10 YS |
1779 | |
1780 | spin_lock(&osdc->event_lock); | |
1781 | event->cookie = ++osdc->event_count; | |
1782 | __insert_event(osdc, event); | |
1783 | spin_unlock(&osdc->event_lock); | |
1784 | ||
1785 | *pevent = event; | |
1786 | return 0; | |
1787 | } | |
1788 | EXPORT_SYMBOL(ceph_osdc_create_event); | |
1789 | ||
1790 | void ceph_osdc_cancel_event(struct ceph_osd_event *event) | |
1791 | { | |
1792 | struct ceph_osd_client *osdc = event->osdc; | |
1793 | ||
1794 | dout("cancel_event %p\n", event); | |
1795 | spin_lock(&osdc->event_lock); | |
1796 | __remove_event(event); | |
1797 | spin_unlock(&osdc->event_lock); | |
1798 | ceph_osdc_put_event(event); /* caller's */ | |
1799 | } | |
1800 | EXPORT_SYMBOL(ceph_osdc_cancel_event); | |
1801 | ||
1802 | ||
1803 | static void do_event_work(struct work_struct *work) | |
1804 | { | |
1805 | struct ceph_osd_event_work *event_work = | |
1806 | container_of(work, struct ceph_osd_event_work, work); | |
1807 | struct ceph_osd_event *event = event_work->event; | |
1808 | u64 ver = event_work->ver; | |
1809 | u64 notify_id = event_work->notify_id; | |
1810 | u8 opcode = event_work->opcode; | |
1811 | ||
1812 | dout("do_event_work completing %p\n", event); | |
1813 | event->cb(ver, notify_id, opcode, event->data); | |
a40c4f10 YS |
1814 | dout("do_event_work completed %p\n", event); |
1815 | ceph_osdc_put_event(event); | |
1816 | kfree(event_work); | |
1817 | } | |
1818 | ||
1819 | ||
1820 | /* | |
1821 | * Process osd watch notifications | |
1822 | */ | |
3c663bbd AE |
1823 | static void handle_watch_notify(struct ceph_osd_client *osdc, |
1824 | struct ceph_msg *msg) | |
a40c4f10 YS |
1825 | { |
1826 | void *p, *end; | |
1827 | u8 proto_ver; | |
1828 | u64 cookie, ver, notify_id; | |
1829 | u8 opcode; | |
1830 | struct ceph_osd_event *event; | |
1831 | struct ceph_osd_event_work *event_work; | |
1832 | ||
1833 | p = msg->front.iov_base; | |
1834 | end = p + msg->front.iov_len; | |
1835 | ||
1836 | ceph_decode_8_safe(&p, end, proto_ver, bad); | |
1837 | ceph_decode_8_safe(&p, end, opcode, bad); | |
1838 | ceph_decode_64_safe(&p, end, cookie, bad); | |
1839 | ceph_decode_64_safe(&p, end, ver, bad); | |
1840 | ceph_decode_64_safe(&p, end, notify_id, bad); | |
1841 | ||
1842 | spin_lock(&osdc->event_lock); | |
1843 | event = __find_event(osdc, cookie); | |
1844 | if (event) { | |
3c663bbd | 1845 | BUG_ON(event->one_shot); |
a40c4f10 | 1846 | get_event(event); |
a40c4f10 YS |
1847 | } |
1848 | spin_unlock(&osdc->event_lock); | |
1849 | dout("handle_watch_notify cookie %lld ver %lld event %p\n", | |
1850 | cookie, ver, event); | |
1851 | if (event) { | |
1852 | event_work = kmalloc(sizeof(*event_work), GFP_NOIO); | |
a40c4f10 YS |
1853 | if (!event_work) { |
1854 | dout("ERROR: could not allocate event_work\n"); | |
1855 | goto done_err; | |
1856 | } | |
6b0ae409 | 1857 | INIT_WORK(&event_work->work, do_event_work); |
a40c4f10 YS |
1858 | event_work->event = event; |
1859 | event_work->ver = ver; | |
1860 | event_work->notify_id = notify_id; | |
1861 | event_work->opcode = opcode; | |
1862 | if (!queue_work(osdc->notify_wq, &event_work->work)) { | |
1863 | dout("WARNING: failed to queue notify event work\n"); | |
1864 | goto done_err; | |
1865 | } | |
1866 | } | |
1867 | ||
1868 | return; | |
1869 | ||
1870 | done_err: | |
a40c4f10 YS |
1871 | ceph_osdc_put_event(event); |
1872 | return; | |
1873 | ||
1874 | bad: | |
1875 | pr_err("osdc handle_watch_notify corrupt msg\n"); | |
1876 | return; | |
1877 | } | |
1878 | ||
70636773 AE |
1879 | static void ceph_osdc_msg_data_set(struct ceph_msg *msg, |
1880 | struct ceph_osd_data *osd_data) | |
f24e9980 | 1881 | { |
0fff87ec | 1882 | if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) { |
f1baeb2b | 1883 | BUG_ON(osd_data->length > (u64) SIZE_MAX); |
ebf18f47 | 1884 | if (osd_data->length) |
70636773 AE |
1885 | ceph_msg_data_set_pages(msg, osd_data->pages, |
1886 | osd_data->length, osd_data->alignment); | |
9a5e6d09 AE |
1887 | } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) { |
1888 | BUG_ON(!osd_data->pagelist->length); | |
1889 | ceph_msg_data_set_pagelist(msg, osd_data->pagelist); | |
68b4476b | 1890 | #ifdef CONFIG_BLOCK |
0fff87ec | 1891 | } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) { |
70636773 | 1892 | ceph_msg_data_set_bio(msg, osd_data->bio); |
68b4476b | 1893 | #endif |
2ac2b7a6 | 1894 | } else { |
0fff87ec | 1895 | BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE); |
2ac2b7a6 | 1896 | } |
70636773 AE |
1897 | } |
1898 | ||
1899 | /* | |
1900 | * Register request, send initial attempt. | |
1901 | */ | |
1902 | int ceph_osdc_start_request(struct ceph_osd_client *osdc, | |
1903 | struct ceph_osd_request *req, | |
1904 | bool nofail) | |
1905 | { | |
1906 | int rc = 0; | |
1907 | ||
1908 | /* Set up response incoming data and request outgoing data fields */ | |
1909 | ||
1910 | ceph_osdc_msg_data_set(req->r_reply, &req->r_data_in); | |
1911 | ceph_osdc_msg_data_set(req->r_request, &req->r_data_out); | |
f24e9980 | 1912 | |
f24e9980 SW |
1913 | down_read(&osdc->map_sem); |
1914 | mutex_lock(&osdc->request_mutex); | |
dc4b870c | 1915 | __register_request(osdc, req); |
92451b49 AE |
1916 | WARN_ON(req->r_sent); |
1917 | rc = __map_request(osdc, req, 0); | |
1918 | if (rc < 0) { | |
1919 | if (nofail) { | |
1920 | dout("osdc_start_request failed map, " | |
1921 | " will retry %lld\n", req->r_tid); | |
1922 | rc = 0; | |
f24e9980 | 1923 | } |
92451b49 | 1924 | goto out_unlock; |
f24e9980 | 1925 | } |
92451b49 AE |
1926 | if (req->r_osd == NULL) { |
1927 | dout("send_request %p no up osds in pg\n", req); | |
1928 | ceph_monc_request_next_osdmap(&osdc->client->monc); | |
1929 | } else { | |
7e2766a1 | 1930 | __send_queued(osdc); |
92451b49 AE |
1931 | } |
1932 | rc = 0; | |
234af26f | 1933 | out_unlock: |
f24e9980 SW |
1934 | mutex_unlock(&osdc->request_mutex); |
1935 | up_read(&osdc->map_sem); | |
1936 | return rc; | |
1937 | } | |
3d14c5d2 | 1938 | EXPORT_SYMBOL(ceph_osdc_start_request); |
f24e9980 SW |
1939 | |
1940 | /* | |
1941 | * wait for a request to complete | |
1942 | */ | |
1943 | int ceph_osdc_wait_request(struct ceph_osd_client *osdc, | |
1944 | struct ceph_osd_request *req) | |
1945 | { | |
1946 | int rc; | |
1947 | ||
1948 | rc = wait_for_completion_interruptible(&req->r_completion); | |
1949 | if (rc < 0) { | |
1950 | mutex_lock(&osdc->request_mutex); | |
1951 | __cancel_request(req); | |
529cfcc4 | 1952 | __unregister_request(osdc, req); |
f24e9980 | 1953 | mutex_unlock(&osdc->request_mutex); |
25845472 | 1954 | complete_request(req); |
529cfcc4 | 1955 | dout("wait_request tid %llu canceled/timed out\n", req->r_tid); |
f24e9980 SW |
1956 | return rc; |
1957 | } | |
1958 | ||
1959 | dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result); | |
1960 | return req->r_result; | |
1961 | } | |
3d14c5d2 | 1962 | EXPORT_SYMBOL(ceph_osdc_wait_request); |
f24e9980 SW |
1963 | |
1964 | /* | |
1965 | * sync - wait for all in-flight requests to flush. avoid starvation. | |
1966 | */ | |
1967 | void ceph_osdc_sync(struct ceph_osd_client *osdc) | |
1968 | { | |
1969 | struct ceph_osd_request *req; | |
1970 | u64 last_tid, next_tid = 0; | |
1971 | ||
1972 | mutex_lock(&osdc->request_mutex); | |
1973 | last_tid = osdc->last_tid; | |
1974 | while (1) { | |
1975 | req = __lookup_request_ge(osdc, next_tid); | |
1976 | if (!req) | |
1977 | break; | |
1978 | if (req->r_tid > last_tid) | |
1979 | break; | |
1980 | ||
1981 | next_tid = req->r_tid + 1; | |
1982 | if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0) | |
1983 | continue; | |
1984 | ||
1985 | ceph_osdc_get_request(req); | |
1986 | mutex_unlock(&osdc->request_mutex); | |
1987 | dout("sync waiting on tid %llu (last is %llu)\n", | |
1988 | req->r_tid, last_tid); | |
1989 | wait_for_completion(&req->r_safe_completion); | |
1990 | mutex_lock(&osdc->request_mutex); | |
1991 | ceph_osdc_put_request(req); | |
1992 | } | |
1993 | mutex_unlock(&osdc->request_mutex); | |
1994 | dout("sync done (thru tid %llu)\n", last_tid); | |
1995 | } | |
3d14c5d2 | 1996 | EXPORT_SYMBOL(ceph_osdc_sync); |
f24e9980 SW |
1997 | |
1998 | /* | |
1999 | * init, shutdown | |
2000 | */ | |
2001 | int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client) | |
2002 | { | |
2003 | int err; | |
2004 | ||
2005 | dout("init\n"); | |
2006 | osdc->client = client; | |
2007 | osdc->osdmap = NULL; | |
2008 | init_rwsem(&osdc->map_sem); | |
2009 | init_completion(&osdc->map_waiters); | |
2010 | osdc->last_requested_map = 0; | |
2011 | mutex_init(&osdc->request_mutex); | |
f24e9980 SW |
2012 | osdc->last_tid = 0; |
2013 | osdc->osds = RB_ROOT; | |
f5a2041b | 2014 | INIT_LIST_HEAD(&osdc->osd_lru); |
f24e9980 | 2015 | osdc->requests = RB_ROOT; |
422d2cb8 | 2016 | INIT_LIST_HEAD(&osdc->req_lru); |
6f6c7006 SW |
2017 | INIT_LIST_HEAD(&osdc->req_unsent); |
2018 | INIT_LIST_HEAD(&osdc->req_notarget); | |
a40c4f10 | 2019 | INIT_LIST_HEAD(&osdc->req_linger); |
f24e9980 SW |
2020 | osdc->num_requests = 0; |
2021 | INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout); | |
f5a2041b | 2022 | INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout); |
a40c4f10 YS |
2023 | spin_lock_init(&osdc->event_lock); |
2024 | osdc->event_tree = RB_ROOT; | |
2025 | osdc->event_count = 0; | |
f5a2041b YS |
2026 | |
2027 | schedule_delayed_work(&osdc->osds_timeout_work, | |
3d14c5d2 | 2028 | round_jiffies_relative(osdc->client->options->osd_idle_ttl * HZ)); |
f24e9980 | 2029 | |
5f44f142 | 2030 | err = -ENOMEM; |
f24e9980 SW |
2031 | osdc->req_mempool = mempool_create_kmalloc_pool(10, |
2032 | sizeof(struct ceph_osd_request)); | |
2033 | if (!osdc->req_mempool) | |
5f44f142 | 2034 | goto out; |
f24e9980 | 2035 | |
d50b409f SW |
2036 | err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP, |
2037 | OSD_OP_FRONT_LEN, 10, true, | |
4f48280e | 2038 | "osd_op"); |
f24e9980 | 2039 | if (err < 0) |
5f44f142 | 2040 | goto out_mempool; |
d50b409f | 2041 | err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY, |
4f48280e SW |
2042 | OSD_OPREPLY_FRONT_LEN, 10, true, |
2043 | "osd_op_reply"); | |
c16e7869 SW |
2044 | if (err < 0) |
2045 | goto out_msgpool; | |
a40c4f10 YS |
2046 | |
2047 | osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify"); | |
2048 | if (IS_ERR(osdc->notify_wq)) { | |
2049 | err = PTR_ERR(osdc->notify_wq); | |
2050 | osdc->notify_wq = NULL; | |
2051 | goto out_msgpool; | |
2052 | } | |
f24e9980 | 2053 | return 0; |
5f44f142 | 2054 | |
c16e7869 SW |
2055 | out_msgpool: |
2056 | ceph_msgpool_destroy(&osdc->msgpool_op); | |
5f44f142 SW |
2057 | out_mempool: |
2058 | mempool_destroy(osdc->req_mempool); | |
2059 | out: | |
2060 | return err; | |
f24e9980 SW |
2061 | } |
2062 | ||
2063 | void ceph_osdc_stop(struct ceph_osd_client *osdc) | |
2064 | { | |
a40c4f10 YS |
2065 | flush_workqueue(osdc->notify_wq); |
2066 | destroy_workqueue(osdc->notify_wq); | |
f24e9980 | 2067 | cancel_delayed_work_sync(&osdc->timeout_work); |
f5a2041b | 2068 | cancel_delayed_work_sync(&osdc->osds_timeout_work); |
f24e9980 SW |
2069 | if (osdc->osdmap) { |
2070 | ceph_osdmap_destroy(osdc->osdmap); | |
2071 | osdc->osdmap = NULL; | |
2072 | } | |
aca420bc | 2073 | remove_all_osds(osdc); |
f24e9980 SW |
2074 | mempool_destroy(osdc->req_mempool); |
2075 | ceph_msgpool_destroy(&osdc->msgpool_op); | |
c16e7869 | 2076 | ceph_msgpool_destroy(&osdc->msgpool_op_reply); |
f24e9980 SW |
2077 | } |
2078 | ||
2079 | /* | |
2080 | * Read some contiguous pages. If we cross a stripe boundary, shorten | |
2081 | * *plen. Return number of bytes read, or error. | |
2082 | */ | |
2083 | int ceph_osdc_readpages(struct ceph_osd_client *osdc, | |
2084 | struct ceph_vino vino, struct ceph_file_layout *layout, | |
2085 | u64 off, u64 *plen, | |
2086 | u32 truncate_seq, u64 truncate_size, | |
b7495fc2 | 2087 | struct page **pages, int num_pages, int page_align) |
f24e9980 SW |
2088 | { |
2089 | struct ceph_osd_request *req; | |
0fff87ec | 2090 | struct ceph_osd_data *osd_data; |
f24e9980 SW |
2091 | int rc = 0; |
2092 | ||
2093 | dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino, | |
2094 | vino.snap, off, *plen); | |
2095 | req = ceph_osdc_new_request(osdc, layout, vino, off, plen, | |
2096 | CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ, | |
2097 | NULL, 0, truncate_seq, truncate_size, NULL, | |
153e5167 | 2098 | false); |
6816282d SW |
2099 | if (IS_ERR(req)) |
2100 | return PTR_ERR(req); | |
f24e9980 SW |
2101 | |
2102 | /* it may be a short read due to an object boundary */ | |
0fff87ec AE |
2103 | |
2104 | osd_data = &req->r_data_in; | |
2105 | osd_data->type = CEPH_OSD_DATA_TYPE_PAGES; | |
2106 | osd_data->pages = pages; | |
e0c59487 | 2107 | osd_data->length = *plen; |
0fff87ec | 2108 | osd_data->alignment = page_align; |
f24e9980 | 2109 | |
e0c59487 AE |
2110 | dout("readpages final extent is %llu~%llu (%llu bytes align %d)\n", |
2111 | off, *plen, osd_data->length, page_align); | |
f24e9980 SW |
2112 | |
2113 | rc = ceph_osdc_start_request(osdc, req, false); | |
2114 | if (!rc) | |
2115 | rc = ceph_osdc_wait_request(osdc, req); | |
2116 | ||
2117 | ceph_osdc_put_request(req); | |
2118 | dout("readpages result %d\n", rc); | |
2119 | return rc; | |
2120 | } | |
3d14c5d2 | 2121 | EXPORT_SYMBOL(ceph_osdc_readpages); |
f24e9980 SW |
2122 | |
2123 | /* | |
2124 | * do a synchronous write on N pages | |
2125 | */ | |
2126 | int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino, | |
2127 | struct ceph_file_layout *layout, | |
2128 | struct ceph_snap_context *snapc, | |
2129 | u64 off, u64 len, | |
2130 | u32 truncate_seq, u64 truncate_size, | |
2131 | struct timespec *mtime, | |
24808826 | 2132 | struct page **pages, int num_pages) |
f24e9980 SW |
2133 | { |
2134 | struct ceph_osd_request *req; | |
0fff87ec | 2135 | struct ceph_osd_data *osd_data; |
f24e9980 | 2136 | int rc = 0; |
b7495fc2 | 2137 | int page_align = off & ~PAGE_MASK; |
f24e9980 SW |
2138 | |
2139 | BUG_ON(vino.snap != CEPH_NOSNAP); | |
2140 | req = ceph_osdc_new_request(osdc, layout, vino, off, &len, | |
2141 | CEPH_OSD_OP_WRITE, | |
24808826 | 2142 | CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE, |
fbf8685f | 2143 | snapc, 0, |
f24e9980 | 2144 | truncate_seq, truncate_size, mtime, |
153e5167 | 2145 | true); |
6816282d SW |
2146 | if (IS_ERR(req)) |
2147 | return PTR_ERR(req); | |
f24e9980 SW |
2148 | |
2149 | /* it may be a short write due to an object boundary */ | |
0fff87ec AE |
2150 | osd_data = &req->r_data_out; |
2151 | osd_data->type = CEPH_OSD_DATA_TYPE_PAGES; | |
2152 | osd_data->pages = pages; | |
e0c59487 | 2153 | osd_data->length = len; |
0fff87ec | 2154 | osd_data->alignment = page_align; |
e0c59487 | 2155 | dout("writepages %llu~%llu (%llu bytes)\n", off, len, osd_data->length); |
f24e9980 | 2156 | |
87f979d3 | 2157 | rc = ceph_osdc_start_request(osdc, req, true); |
f24e9980 SW |
2158 | if (!rc) |
2159 | rc = ceph_osdc_wait_request(osdc, req); | |
2160 | ||
2161 | ceph_osdc_put_request(req); | |
2162 | if (rc == 0) | |
2163 | rc = len; | |
2164 | dout("writepages result %d\n", rc); | |
2165 | return rc; | |
2166 | } | |
3d14c5d2 | 2167 | EXPORT_SYMBOL(ceph_osdc_writepages); |
f24e9980 SW |
2168 | |
2169 | /* | |
2170 | * handle incoming message | |
2171 | */ | |
2172 | static void dispatch(struct ceph_connection *con, struct ceph_msg *msg) | |
2173 | { | |
2174 | struct ceph_osd *osd = con->private; | |
32c895e7 | 2175 | struct ceph_osd_client *osdc; |
f24e9980 SW |
2176 | int type = le16_to_cpu(msg->hdr.type); |
2177 | ||
2178 | if (!osd) | |
4a32f93d | 2179 | goto out; |
32c895e7 | 2180 | osdc = osd->o_osdc; |
f24e9980 SW |
2181 | |
2182 | switch (type) { | |
2183 | case CEPH_MSG_OSD_MAP: | |
2184 | ceph_osdc_handle_map(osdc, msg); | |
2185 | break; | |
2186 | case CEPH_MSG_OSD_OPREPLY: | |
350b1c32 | 2187 | handle_reply(osdc, msg, con); |
f24e9980 | 2188 | break; |
a40c4f10 YS |
2189 | case CEPH_MSG_WATCH_NOTIFY: |
2190 | handle_watch_notify(osdc, msg); | |
2191 | break; | |
f24e9980 SW |
2192 | |
2193 | default: | |
2194 | pr_err("received unknown message type %d %s\n", type, | |
2195 | ceph_msg_type_name(type)); | |
2196 | } | |
4a32f93d | 2197 | out: |
f24e9980 SW |
2198 | ceph_msg_put(msg); |
2199 | } | |
2200 | ||
5b3a4db3 | 2201 | /* |
21b667f6 SW |
2202 | * lookup and return message for incoming reply. set up reply message |
2203 | * pages. | |
5b3a4db3 SW |
2204 | */ |
2205 | static struct ceph_msg *get_reply(struct ceph_connection *con, | |
2450418c YS |
2206 | struct ceph_msg_header *hdr, |
2207 | int *skip) | |
f24e9980 SW |
2208 | { |
2209 | struct ceph_osd *osd = con->private; | |
2210 | struct ceph_osd_client *osdc = osd->o_osdc; | |
2450418c | 2211 | struct ceph_msg *m; |
0547a9b3 | 2212 | struct ceph_osd_request *req; |
5b3a4db3 SW |
2213 | int front = le32_to_cpu(hdr->front_len); |
2214 | int data_len = le32_to_cpu(hdr->data_len); | |
0547a9b3 | 2215 | u64 tid; |
f24e9980 | 2216 | |
0547a9b3 YS |
2217 | tid = le64_to_cpu(hdr->tid); |
2218 | mutex_lock(&osdc->request_mutex); | |
2219 | req = __lookup_request(osdc, tid); | |
2220 | if (!req) { | |
2221 | *skip = 1; | |
2222 | m = NULL; | |
756a16a5 SW |
2223 | dout("get_reply unknown tid %llu from osd%d\n", tid, |
2224 | osd->o_osd); | |
0547a9b3 YS |
2225 | goto out; |
2226 | } | |
c16e7869 SW |
2227 | |
2228 | if (req->r_con_filling_msg) { | |
8921d114 | 2229 | dout("%s revoking msg %p from old con %p\n", __func__, |
c16e7869 | 2230 | req->r_reply, req->r_con_filling_msg); |
8921d114 | 2231 | ceph_msg_revoke_incoming(req->r_reply); |
0d47766f | 2232 | req->r_con_filling_msg->ops->put(req->r_con_filling_msg); |
6f46cb29 | 2233 | req->r_con_filling_msg = NULL; |
0547a9b3 YS |
2234 | } |
2235 | ||
c16e7869 SW |
2236 | if (front > req->r_reply->front.iov_len) { |
2237 | pr_warning("get_reply front %d > preallocated %d\n", | |
2238 | front, (int)req->r_reply->front.iov_len); | |
b61c2763 | 2239 | m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front, GFP_NOFS, false); |
a79832f2 | 2240 | if (!m) |
c16e7869 SW |
2241 | goto out; |
2242 | ceph_msg_put(req->r_reply); | |
2243 | req->r_reply = m; | |
2244 | } | |
2245 | m = ceph_msg_get(req->r_reply); | |
2246 | ||
0547a9b3 | 2247 | if (data_len > 0) { |
0fff87ec AE |
2248 | struct ceph_osd_data *osd_data = &req->r_data_in; |
2249 | ||
2250 | if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) { | |
0fff87ec | 2251 | if (osd_data->pages && |
e0c59487 | 2252 | unlikely(osd_data->length < data_len)) { |
2ac2b7a6 | 2253 | |
e0c59487 AE |
2254 | pr_warning("tid %lld reply has %d bytes " |
2255 | "we had only %llu bytes ready\n", | |
2256 | tid, data_len, osd_data->length); | |
2ac2b7a6 AE |
2257 | *skip = 1; |
2258 | ceph_msg_put(m); | |
2259 | m = NULL; | |
2260 | goto out; | |
2261 | } | |
2ac2b7a6 | 2262 | } |
0547a9b3 | 2263 | } |
5b3a4db3 | 2264 | *skip = 0; |
0d47766f | 2265 | req->r_con_filling_msg = con->ops->get(con); |
c16e7869 | 2266 | dout("get_reply tid %lld %p\n", tid, m); |
0547a9b3 YS |
2267 | |
2268 | out: | |
2269 | mutex_unlock(&osdc->request_mutex); | |
2450418c | 2270 | return m; |
5b3a4db3 SW |
2271 | |
2272 | } | |
2273 | ||
2274 | static struct ceph_msg *alloc_msg(struct ceph_connection *con, | |
2275 | struct ceph_msg_header *hdr, | |
2276 | int *skip) | |
2277 | { | |
2278 | struct ceph_osd *osd = con->private; | |
2279 | int type = le16_to_cpu(hdr->type); | |
2280 | int front = le32_to_cpu(hdr->front_len); | |
2281 | ||
1c20f2d2 | 2282 | *skip = 0; |
5b3a4db3 SW |
2283 | switch (type) { |
2284 | case CEPH_MSG_OSD_MAP: | |
a40c4f10 | 2285 | case CEPH_MSG_WATCH_NOTIFY: |
b61c2763 | 2286 | return ceph_msg_new(type, front, GFP_NOFS, false); |
5b3a4db3 SW |
2287 | case CEPH_MSG_OSD_OPREPLY: |
2288 | return get_reply(con, hdr, skip); | |
2289 | default: | |
2290 | pr_info("alloc_msg unexpected msg type %d from osd%d\n", type, | |
2291 | osd->o_osd); | |
2292 | *skip = 1; | |
2293 | return NULL; | |
2294 | } | |
f24e9980 SW |
2295 | } |
2296 | ||
2297 | /* | |
2298 | * Wrappers to refcount containing ceph_osd struct | |
2299 | */ | |
2300 | static struct ceph_connection *get_osd_con(struct ceph_connection *con) | |
2301 | { | |
2302 | struct ceph_osd *osd = con->private; | |
2303 | if (get_osd(osd)) | |
2304 | return con; | |
2305 | return NULL; | |
2306 | } | |
2307 | ||
2308 | static void put_osd_con(struct ceph_connection *con) | |
2309 | { | |
2310 | struct ceph_osd *osd = con->private; | |
2311 | put_osd(osd); | |
2312 | } | |
2313 | ||
4e7a5dcd SW |
2314 | /* |
2315 | * authentication | |
2316 | */ | |
a3530df3 AE |
2317 | /* |
2318 | * Note: returned pointer is the address of a structure that's | |
2319 | * managed separately. Caller must *not* attempt to free it. | |
2320 | */ | |
2321 | static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con, | |
8f43fb53 | 2322 | int *proto, int force_new) |
4e7a5dcd SW |
2323 | { |
2324 | struct ceph_osd *o = con->private; | |
2325 | struct ceph_osd_client *osdc = o->o_osdc; | |
2326 | struct ceph_auth_client *ac = osdc->client->monc.auth; | |
74f1869f | 2327 | struct ceph_auth_handshake *auth = &o->o_auth; |
4e7a5dcd | 2328 | |
74f1869f | 2329 | if (force_new && auth->authorizer) { |
27859f97 | 2330 | ceph_auth_destroy_authorizer(ac, auth->authorizer); |
74f1869f AE |
2331 | auth->authorizer = NULL; |
2332 | } | |
27859f97 SW |
2333 | if (!auth->authorizer) { |
2334 | int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD, | |
2335 | auth); | |
4e7a5dcd | 2336 | if (ret) |
a3530df3 | 2337 | return ERR_PTR(ret); |
27859f97 SW |
2338 | } else { |
2339 | int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD, | |
0bed9b5c SW |
2340 | auth); |
2341 | if (ret) | |
2342 | return ERR_PTR(ret); | |
4e7a5dcd | 2343 | } |
4e7a5dcd | 2344 | *proto = ac->protocol; |
74f1869f | 2345 | |
a3530df3 | 2346 | return auth; |
4e7a5dcd SW |
2347 | } |
2348 | ||
2349 | ||
2350 | static int verify_authorizer_reply(struct ceph_connection *con, int len) | |
2351 | { | |
2352 | struct ceph_osd *o = con->private; | |
2353 | struct ceph_osd_client *osdc = o->o_osdc; | |
2354 | struct ceph_auth_client *ac = osdc->client->monc.auth; | |
2355 | ||
27859f97 | 2356 | return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer, len); |
4e7a5dcd SW |
2357 | } |
2358 | ||
9bd2e6f8 SW |
2359 | static int invalidate_authorizer(struct ceph_connection *con) |
2360 | { | |
2361 | struct ceph_osd *o = con->private; | |
2362 | struct ceph_osd_client *osdc = o->o_osdc; | |
2363 | struct ceph_auth_client *ac = osdc->client->monc.auth; | |
2364 | ||
27859f97 | 2365 | ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD); |
9bd2e6f8 SW |
2366 | return ceph_monc_validate_auth(&osdc->client->monc); |
2367 | } | |
4e7a5dcd | 2368 | |
9e32789f | 2369 | static const struct ceph_connection_operations osd_con_ops = { |
f24e9980 SW |
2370 | .get = get_osd_con, |
2371 | .put = put_osd_con, | |
2372 | .dispatch = dispatch, | |
4e7a5dcd SW |
2373 | .get_authorizer = get_authorizer, |
2374 | .verify_authorizer_reply = verify_authorizer_reply, | |
9bd2e6f8 | 2375 | .invalidate_authorizer = invalidate_authorizer, |
f24e9980 | 2376 | .alloc_msg = alloc_msg, |
81b024e7 | 2377 | .fault = osd_reset, |
f24e9980 | 2378 | }; |