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 | 34 | |
68b4476b YS |
35 | static int op_has_extent(int op) |
36 | { | |
37 | return (op == CEPH_OSD_OP_READ || | |
38 | op == CEPH_OSD_OP_WRITE); | |
39 | } | |
40 | ||
f24e9980 SW |
41 | /* |
42 | * Implement client access to distributed object storage cluster. | |
43 | * | |
44 | * All data objects are stored within a cluster/cloud of OSDs, or | |
45 | * "object storage devices." (Note that Ceph OSDs have _nothing_ to | |
46 | * do with the T10 OSD extensions to SCSI.) Ceph OSDs are simply | |
47 | * remote daemons serving up and coordinating consistent and safe | |
48 | * access to storage. | |
49 | * | |
50 | * Cluster membership and the mapping of data objects onto storage devices | |
51 | * are described by the osd map. | |
52 | * | |
53 | * We keep track of pending OSD requests (read, write), resubmit | |
54 | * requests to different OSDs when the cluster topology/data layout | |
55 | * change, or retry the affected requests when the communications | |
56 | * channel with an OSD is reset. | |
57 | */ | |
58 | ||
59 | /* | |
60 | * calculate the mapping of a file extent onto an object, and fill out the | |
61 | * request accordingly. shorten extent as necessary if it crosses an | |
62 | * object boundary. | |
63 | * | |
64 | * fill osd op in request message. | |
65 | */ | |
dbe0fc41 | 66 | static int calc_layout(struct ceph_file_layout *layout, u64 off, u64 *plen, |
47a05811 | 67 | struct ceph_osd_req_op *op, u64 *bno) |
f24e9980 | 68 | { |
60e56f13 | 69 | u64 orig_len = *plen; |
60e56f13 AE |
70 | u64 objoff = 0; |
71 | u64 objlen = 0; | |
d63b77f4 | 72 | int r; |
f24e9980 | 73 | |
60e56f13 | 74 | /* object extent? */ |
47a05811 | 75 | r = ceph_calc_file_object_mapping(layout, off, orig_len, bno, |
60e56f13 | 76 | &objoff, &objlen); |
d63b77f4 SW |
77 | if (r < 0) |
78 | return r; | |
60e56f13 AE |
79 | if (objlen < orig_len) { |
80 | *plen = objlen; | |
81 | dout(" skipping last %llu, final file extent %llu~%llu\n", | |
82 | orig_len - *plen, off, *plen); | |
83 | } | |
84 | ||
85 | if (op_has_extent(op->op)) { | |
86 | u32 osize = le32_to_cpu(layout->fl_object_size); | |
87 | op->extent.offset = objoff; | |
88 | op->extent.length = objlen; | |
89 | if (op->extent.truncate_size <= off - objoff) { | |
90 | op->extent.truncate_size = 0; | |
91 | } else { | |
92 | op->extent.truncate_size -= off - objoff; | |
93 | if (op->extent.truncate_size > osize) | |
94 | op->extent.truncate_size = osize; | |
95 | } | |
96 | } | |
60e56f13 AE |
97 | if (op->op == CEPH_OSD_OP_WRITE) |
98 | op->payload_len = *plen; | |
99 | ||
60cf5992 | 100 | dout("calc_layout bno=%llx %llu~%llu\n", *bno, objoff, objlen); |
f24e9980 | 101 | |
3ff5f385 | 102 | return 0; |
f24e9980 SW |
103 | } |
104 | ||
f24e9980 SW |
105 | /* |
106 | * requests | |
107 | */ | |
415e49a9 | 108 | void ceph_osdc_release_request(struct kref *kref) |
f24e9980 | 109 | { |
e0c59487 | 110 | int num_pages; |
415e49a9 SW |
111 | struct ceph_osd_request *req = container_of(kref, |
112 | struct ceph_osd_request, | |
113 | r_kref); | |
114 | ||
115 | if (req->r_request) | |
116 | ceph_msg_put(req->r_request); | |
0d59ab81 | 117 | if (req->r_con_filling_msg) { |
9cbb1d72 AE |
118 | dout("%s revoking msg %p from con %p\n", __func__, |
119 | req->r_reply, req->r_con_filling_msg); | |
8921d114 | 120 | ceph_msg_revoke_incoming(req->r_reply); |
0d47766f | 121 | req->r_con_filling_msg->ops->put(req->r_con_filling_msg); |
9cbb1d72 | 122 | req->r_con_filling_msg = NULL; |
350b1c32 | 123 | } |
ab8cb34a AE |
124 | if (req->r_reply) |
125 | ceph_msg_put(req->r_reply); | |
0fff87ec AE |
126 | |
127 | if (req->r_data_in.type == CEPH_OSD_DATA_TYPE_PAGES && | |
e0c59487 AE |
128 | req->r_data_in.own_pages) { |
129 | num_pages = calc_pages_for((u64)req->r_data_in.alignment, | |
130 | (u64)req->r_data_in.length); | |
131 | ceph_release_page_vector(req->r_data_in.pages, num_pages); | |
132 | } | |
0fff87ec | 133 | if (req->r_data_out.type == CEPH_OSD_DATA_TYPE_PAGES && |
e0c59487 AE |
134 | req->r_data_out.own_pages) { |
135 | num_pages = calc_pages_for((u64)req->r_data_out.alignment, | |
136 | (u64)req->r_data_out.length); | |
137 | ceph_release_page_vector(req->r_data_out.pages, num_pages); | |
138 | } | |
0fff87ec | 139 | |
415e49a9 SW |
140 | ceph_put_snap_context(req->r_snapc); |
141 | if (req->r_mempool) | |
142 | mempool_free(req, req->r_osdc->req_mempool); | |
143 | else | |
144 | kfree(req); | |
f24e9980 | 145 | } |
3d14c5d2 | 146 | EXPORT_SYMBOL(ceph_osdc_release_request); |
68b4476b | 147 | |
3499e8a5 | 148 | struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc, |
f24e9980 | 149 | struct ceph_snap_context *snapc, |
1b83bef2 | 150 | unsigned int num_ops, |
3499e8a5 | 151 | bool use_mempool, |
54a54007 | 152 | gfp_t gfp_flags) |
f24e9980 SW |
153 | { |
154 | struct ceph_osd_request *req; | |
155 | struct ceph_msg *msg; | |
1b83bef2 SW |
156 | size_t msg_size; |
157 | ||
158 | msg_size = 4 + 4 + 8 + 8 + 4+8; | |
159 | msg_size += 2 + 4 + 8 + 4 + 4; /* oloc */ | |
160 | msg_size += 1 + 8 + 4 + 4; /* pg_t */ | |
161 | msg_size += 4 + MAX_OBJ_NAME_SIZE; | |
162 | msg_size += 2 + num_ops*sizeof(struct ceph_osd_op); | |
163 | msg_size += 8; /* snapid */ | |
164 | msg_size += 8; /* snap_seq */ | |
165 | msg_size += 8 * (snapc ? snapc->num_snaps : 0); /* snaps */ | |
166 | msg_size += 4; | |
f24e9980 SW |
167 | |
168 | if (use_mempool) { | |
3499e8a5 | 169 | req = mempool_alloc(osdc->req_mempool, gfp_flags); |
f24e9980 SW |
170 | memset(req, 0, sizeof(*req)); |
171 | } else { | |
3499e8a5 | 172 | req = kzalloc(sizeof(*req), gfp_flags); |
f24e9980 SW |
173 | } |
174 | if (req == NULL) | |
a79832f2 | 175 | return NULL; |
f24e9980 | 176 | |
f24e9980 SW |
177 | req->r_osdc = osdc; |
178 | req->r_mempool = use_mempool; | |
68b4476b | 179 | |
415e49a9 | 180 | kref_init(&req->r_kref); |
f24e9980 SW |
181 | init_completion(&req->r_completion); |
182 | init_completion(&req->r_safe_completion); | |
a978fa20 | 183 | RB_CLEAR_NODE(&req->r_node); |
f24e9980 | 184 | INIT_LIST_HEAD(&req->r_unsafe_item); |
a40c4f10 YS |
185 | INIT_LIST_HEAD(&req->r_linger_item); |
186 | INIT_LIST_HEAD(&req->r_linger_osd); | |
935b639a | 187 | INIT_LIST_HEAD(&req->r_req_lru_item); |
cd43045c SW |
188 | INIT_LIST_HEAD(&req->r_osd_item); |
189 | ||
c16e7869 SW |
190 | /* create reply message */ |
191 | if (use_mempool) | |
192 | msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0); | |
193 | else | |
194 | msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, | |
b61c2763 | 195 | OSD_OPREPLY_FRONT_LEN, gfp_flags, true); |
a79832f2 | 196 | if (!msg) { |
c16e7869 | 197 | ceph_osdc_put_request(req); |
a79832f2 | 198 | return NULL; |
c16e7869 SW |
199 | } |
200 | req->r_reply = msg; | |
201 | ||
0fff87ec AE |
202 | req->r_data_in.type = CEPH_OSD_DATA_TYPE_NONE; |
203 | req->r_data_out.type = CEPH_OSD_DATA_TYPE_NONE; | |
d50b409f | 204 | |
c16e7869 | 205 | /* create request message; allow space for oid */ |
f24e9980 | 206 | if (use_mempool) |
8f3bc053 | 207 | msg = ceph_msgpool_get(&osdc->msgpool_op, 0); |
f24e9980 | 208 | else |
b61c2763 | 209 | msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp_flags, true); |
a79832f2 | 210 | if (!msg) { |
f24e9980 | 211 | ceph_osdc_put_request(req); |
a79832f2 | 212 | return NULL; |
f24e9980 | 213 | } |
68b4476b | 214 | |
f24e9980 | 215 | memset(msg->front.iov_base, 0, msg->front.iov_len); |
3499e8a5 YS |
216 | |
217 | req->r_request = msg; | |
3499e8a5 YS |
218 | |
219 | return req; | |
220 | } | |
3d14c5d2 | 221 | EXPORT_SYMBOL(ceph_osdc_alloc_request); |
3499e8a5 | 222 | |
175face2 | 223 | static u64 osd_req_encode_op(struct ceph_osd_request *req, |
68b4476b YS |
224 | struct ceph_osd_op *dst, |
225 | struct ceph_osd_req_op *src) | |
226 | { | |
175face2 | 227 | u64 out_data_len = 0; |
95e072eb | 228 | struct ceph_pagelist *pagelist; |
175face2 | 229 | |
68b4476b YS |
230 | dst->op = cpu_to_le16(src->op); |
231 | ||
065a68f9 | 232 | switch (src->op) { |
fbfab539 AE |
233 | case CEPH_OSD_OP_STAT: |
234 | break; | |
68b4476b YS |
235 | case CEPH_OSD_OP_READ: |
236 | case CEPH_OSD_OP_WRITE: | |
175face2 AE |
237 | if (src->op == CEPH_OSD_OP_WRITE) |
238 | out_data_len = src->extent.length; | |
239 | dst->extent.offset = cpu_to_le64(src->extent.offset); | |
240 | dst->extent.length = cpu_to_le64(src->extent.length); | |
68b4476b YS |
241 | dst->extent.truncate_size = |
242 | cpu_to_le64(src->extent.truncate_size); | |
243 | dst->extent.truncate_seq = | |
244 | cpu_to_le32(src->extent.truncate_seq); | |
245 | break; | |
ae1533b6 | 246 | case CEPH_OSD_OP_CALL: |
95e072eb AE |
247 | pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS); |
248 | BUG_ON(!pagelist); | |
249 | ceph_pagelist_init(pagelist); | |
250 | ||
ae1533b6 YS |
251 | dst->cls.class_len = src->cls.class_len; |
252 | dst->cls.method_len = src->cls.method_len; | |
253 | dst->cls.indata_len = cpu_to_le32(src->cls.indata_len); | |
95e072eb | 254 | ceph_pagelist_append(pagelist, src->cls.class_name, |
ae1533b6 | 255 | src->cls.class_len); |
95e072eb | 256 | ceph_pagelist_append(pagelist, src->cls.method_name, |
ae1533b6 | 257 | src->cls.method_len); |
95e072eb | 258 | ceph_pagelist_append(pagelist, src->cls.indata, |
ae1533b6 | 259 | src->cls.indata_len); |
95e072eb AE |
260 | |
261 | req->r_data_out.type = CEPH_OSD_DATA_TYPE_PAGELIST; | |
262 | req->r_data_out.pagelist = pagelist; | |
263 | out_data_len = pagelist->length; | |
ae1533b6 | 264 | break; |
68b4476b YS |
265 | case CEPH_OSD_OP_STARTSYNC: |
266 | break; | |
a40c4f10 YS |
267 | case CEPH_OSD_OP_NOTIFY_ACK: |
268 | case CEPH_OSD_OP_WATCH: | |
269 | dst->watch.cookie = cpu_to_le64(src->watch.cookie); | |
270 | dst->watch.ver = cpu_to_le64(src->watch.ver); | |
271 | dst->watch.flag = src->watch.flag; | |
272 | break; | |
68b4476b | 273 | default: |
8f63ca2d | 274 | pr_err("unrecognized osd opcode %d\n", src->op); |
68b4476b YS |
275 | WARN_ON(1); |
276 | break; | |
4c46459c AE |
277 | case CEPH_OSD_OP_MAPEXT: |
278 | case CEPH_OSD_OP_MASKTRUNC: | |
279 | case CEPH_OSD_OP_SPARSE_READ: | |
a9f36c3e | 280 | case CEPH_OSD_OP_NOTIFY: |
4c46459c AE |
281 | case CEPH_OSD_OP_ASSERT_VER: |
282 | case CEPH_OSD_OP_WRITEFULL: | |
283 | case CEPH_OSD_OP_TRUNCATE: | |
284 | case CEPH_OSD_OP_ZERO: | |
285 | case CEPH_OSD_OP_DELETE: | |
286 | case CEPH_OSD_OP_APPEND: | |
287 | case CEPH_OSD_OP_SETTRUNC: | |
288 | case CEPH_OSD_OP_TRIMTRUNC: | |
289 | case CEPH_OSD_OP_TMAPUP: | |
290 | case CEPH_OSD_OP_TMAPPUT: | |
291 | case CEPH_OSD_OP_TMAPGET: | |
292 | case CEPH_OSD_OP_CREATE: | |
a9f36c3e | 293 | case CEPH_OSD_OP_ROLLBACK: |
4c46459c AE |
294 | case CEPH_OSD_OP_OMAPGETKEYS: |
295 | case CEPH_OSD_OP_OMAPGETVALS: | |
296 | case CEPH_OSD_OP_OMAPGETHEADER: | |
297 | case CEPH_OSD_OP_OMAPGETVALSBYKEYS: | |
298 | case CEPH_OSD_OP_MODE_RD: | |
299 | case CEPH_OSD_OP_OMAPSETVALS: | |
300 | case CEPH_OSD_OP_OMAPSETHEADER: | |
301 | case CEPH_OSD_OP_OMAPCLEAR: | |
302 | case CEPH_OSD_OP_OMAPRMKEYS: | |
303 | case CEPH_OSD_OP_OMAP_CMP: | |
304 | case CEPH_OSD_OP_CLONERANGE: | |
305 | case CEPH_OSD_OP_ASSERT_SRC_VERSION: | |
306 | case CEPH_OSD_OP_SRC_CMPXATTR: | |
a9f36c3e | 307 | case CEPH_OSD_OP_GETXATTR: |
4c46459c | 308 | case CEPH_OSD_OP_GETXATTRS: |
a9f36c3e AE |
309 | case CEPH_OSD_OP_CMPXATTR: |
310 | case CEPH_OSD_OP_SETXATTR: | |
4c46459c AE |
311 | case CEPH_OSD_OP_SETXATTRS: |
312 | case CEPH_OSD_OP_RESETXATTRS: | |
313 | case CEPH_OSD_OP_RMXATTR: | |
314 | case CEPH_OSD_OP_PULL: | |
315 | case CEPH_OSD_OP_PUSH: | |
316 | case CEPH_OSD_OP_BALANCEREADS: | |
317 | case CEPH_OSD_OP_UNBALANCEREADS: | |
318 | case CEPH_OSD_OP_SCRUB: | |
319 | case CEPH_OSD_OP_SCRUB_RESERVE: | |
320 | case CEPH_OSD_OP_SCRUB_UNRESERVE: | |
321 | case CEPH_OSD_OP_SCRUB_STOP: | |
322 | case CEPH_OSD_OP_SCRUB_MAP: | |
323 | case CEPH_OSD_OP_WRLOCK: | |
324 | case CEPH_OSD_OP_WRUNLOCK: | |
325 | case CEPH_OSD_OP_RDLOCK: | |
326 | case CEPH_OSD_OP_RDUNLOCK: | |
327 | case CEPH_OSD_OP_UPLOCK: | |
328 | case CEPH_OSD_OP_DNLOCK: | |
329 | case CEPH_OSD_OP_PGLS: | |
330 | case CEPH_OSD_OP_PGLS_FILTER: | |
331 | pr_err("unsupported osd opcode %s\n", | |
8f63ca2d | 332 | ceph_osd_op_name(src->op)); |
4c46459c AE |
333 | WARN_ON(1); |
334 | break; | |
68b4476b YS |
335 | } |
336 | dst->payload_len = cpu_to_le32(src->payload_len); | |
175face2 AE |
337 | |
338 | return out_data_len; | |
68b4476b YS |
339 | } |
340 | ||
3499e8a5 YS |
341 | /* |
342 | * build new request AND message | |
343 | * | |
344 | */ | |
345 | void ceph_osdc_build_request(struct ceph_osd_request *req, | |
175face2 | 346 | u64 off, unsigned int num_ops, |
68b4476b | 347 | struct ceph_osd_req_op *src_ops, |
4d6b250b | 348 | struct ceph_snap_context *snapc, u64 snap_id, |
af77f26c | 349 | struct timespec *mtime) |
3499e8a5 YS |
350 | { |
351 | struct ceph_msg *msg = req->r_request; | |
68b4476b | 352 | struct ceph_osd_req_op *src_op; |
3499e8a5 | 353 | void *p; |
1b83bef2 | 354 | size_t msg_size; |
3499e8a5 | 355 | int flags = req->r_flags; |
f44246e3 | 356 | u64 data_len; |
68b4476b | 357 | int i; |
3499e8a5 | 358 | |
1b83bef2 SW |
359 | req->r_num_ops = num_ops; |
360 | req->r_snapid = snap_id; | |
f24e9980 SW |
361 | req->r_snapc = ceph_get_snap_context(snapc); |
362 | ||
1b83bef2 SW |
363 | /* encode request */ |
364 | msg->hdr.version = cpu_to_le16(4); | |
365 | ||
366 | p = msg->front.iov_base; | |
367 | ceph_encode_32(&p, 1); /* client_inc is always 1 */ | |
368 | req->r_request_osdmap_epoch = p; | |
369 | p += 4; | |
370 | req->r_request_flags = p; | |
371 | p += 4; | |
372 | if (req->r_flags & CEPH_OSD_FLAG_WRITE) | |
373 | ceph_encode_timespec(p, mtime); | |
374 | p += sizeof(struct ceph_timespec); | |
375 | req->r_request_reassert_version = p; | |
376 | p += sizeof(struct ceph_eversion); /* will get filled in */ | |
377 | ||
378 | /* oloc */ | |
379 | ceph_encode_8(&p, 4); | |
380 | ceph_encode_8(&p, 4); | |
381 | ceph_encode_32(&p, 8 + 4 + 4); | |
382 | req->r_request_pool = p; | |
383 | p += 8; | |
384 | ceph_encode_32(&p, -1); /* preferred */ | |
385 | ceph_encode_32(&p, 0); /* key len */ | |
386 | ||
387 | ceph_encode_8(&p, 1); | |
388 | req->r_request_pgid = p; | |
389 | p += 8 + 4; | |
390 | ceph_encode_32(&p, -1); /* preferred */ | |
391 | ||
392 | /* oid */ | |
393 | ceph_encode_32(&p, req->r_oid_len); | |
af77f26c | 394 | memcpy(p, req->r_oid, req->r_oid_len); |
1b83bef2 | 395 | dout("oid '%.*s' len %d\n", req->r_oid_len, req->r_oid, req->r_oid_len); |
af77f26c | 396 | p += req->r_oid_len; |
f24e9980 | 397 | |
175face2 | 398 | /* ops--can imply data */ |
1b83bef2 | 399 | ceph_encode_16(&p, num_ops); |
68b4476b | 400 | src_op = src_ops; |
1b83bef2 | 401 | req->r_request_ops = p; |
175face2 | 402 | data_len = 0; |
1b83bef2 | 403 | for (i = 0; i < num_ops; i++, src_op++) { |
175face2 | 404 | data_len += osd_req_encode_op(req, p, src_op); |
1b83bef2 SW |
405 | p += sizeof(struct ceph_osd_op); |
406 | } | |
68b4476b | 407 | |
1b83bef2 SW |
408 | /* snaps */ |
409 | ceph_encode_64(&p, req->r_snapid); | |
410 | ceph_encode_64(&p, req->r_snapc ? req->r_snapc->seq : 0); | |
411 | ceph_encode_32(&p, req->r_snapc ? req->r_snapc->num_snaps : 0); | |
412 | if (req->r_snapc) { | |
f24e9980 | 413 | for (i = 0; i < snapc->num_snaps; i++) { |
1b83bef2 | 414 | ceph_encode_64(&p, req->r_snapc->snaps[i]); |
f24e9980 SW |
415 | } |
416 | } | |
417 | ||
1b83bef2 SW |
418 | req->r_request_attempts = p; |
419 | p += 4; | |
420 | ||
175face2 AE |
421 | /* data */ |
422 | if (flags & CEPH_OSD_FLAG_WRITE) | |
68b4476b | 423 | req->r_request->hdr.data_off = cpu_to_le16(off); |
f44246e3 | 424 | req->r_request->hdr.data_len = cpu_to_le32(data_len); |
c5c6b19d | 425 | |
f24e9980 | 426 | BUG_ON(p > msg->front.iov_base + msg->front.iov_len); |
6f863e71 SW |
427 | msg_size = p - msg->front.iov_base; |
428 | msg->front.iov_len = msg_size; | |
429 | msg->hdr.front_len = cpu_to_le32(msg_size); | |
1b83bef2 SW |
430 | |
431 | dout("build_request msg_size was %d num_ops %d\n", (int)msg_size, | |
432 | num_ops); | |
3499e8a5 YS |
433 | return; |
434 | } | |
3d14c5d2 | 435 | EXPORT_SYMBOL(ceph_osdc_build_request); |
3499e8a5 YS |
436 | |
437 | /* | |
438 | * build new request AND message, calculate layout, and adjust file | |
439 | * extent as needed. | |
440 | * | |
441 | * if the file was recently truncated, we include information about its | |
442 | * old and new size so that the object can be updated appropriately. (we | |
443 | * avoid synchronously deleting truncated objects because it's slow.) | |
444 | * | |
445 | * if @do_sync, include a 'startsync' command so that the osd will flush | |
446 | * data quickly. | |
447 | */ | |
448 | struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc, | |
449 | struct ceph_file_layout *layout, | |
450 | struct ceph_vino vino, | |
451 | u64 off, u64 *plen, | |
452 | int opcode, int flags, | |
453 | struct ceph_snap_context *snapc, | |
454 | int do_sync, | |
455 | u32 truncate_seq, | |
456 | u64 truncate_size, | |
457 | struct timespec *mtime, | |
153e5167 | 458 | bool use_mempool) |
3499e8a5 | 459 | { |
ae7ca4a3 | 460 | struct ceph_osd_req_op ops[2]; |
68b4476b | 461 | struct ceph_osd_request *req; |
ae7ca4a3 | 462 | unsigned int num_op = 1; |
47a05811 | 463 | u64 bno = 0; |
6816282d | 464 | int r; |
68b4476b | 465 | |
ae7ca4a3 AE |
466 | memset(&ops, 0, sizeof ops); |
467 | ||
68b4476b YS |
468 | ops[0].op = opcode; |
469 | ops[0].extent.truncate_seq = truncate_seq; | |
470 | ops[0].extent.truncate_size = truncate_size; | |
68b4476b YS |
471 | |
472 | if (do_sync) { | |
473 | ops[1].op = CEPH_OSD_OP_STARTSYNC; | |
ae7ca4a3 AE |
474 | num_op++; |
475 | } | |
68b4476b | 476 | |
ae7ca4a3 AE |
477 | req = ceph_osdc_alloc_request(osdc, snapc, num_op, use_mempool, |
478 | GFP_NOFS); | |
4ad12621 | 479 | if (!req) |
6816282d | 480 | return ERR_PTR(-ENOMEM); |
d178a9e7 | 481 | req->r_flags = flags; |
3499e8a5 YS |
482 | |
483 | /* calculate max write size */ | |
60cf5992 | 484 | r = calc_layout(layout, off, plen, ops, &bno); |
3ff5f385 AE |
485 | if (r < 0) { |
486 | ceph_osdc_put_request(req); | |
6816282d | 487 | return ERR_PTR(r); |
3ff5f385 | 488 | } |
3499e8a5 YS |
489 | req->r_file_layout = *layout; /* keep a copy */ |
490 | ||
dbe0fc41 AE |
491 | snprintf(req->r_oid, sizeof(req->r_oid), "%llx.%08llx", vino.ino, bno); |
492 | req->r_oid_len = strlen(req->r_oid); | |
493 | ||
175face2 | 494 | ceph_osdc_build_request(req, off, num_op, ops, |
ae7ca4a3 | 495 | snapc, vino.snap, mtime); |
3499e8a5 | 496 | |
f24e9980 SW |
497 | return req; |
498 | } | |
3d14c5d2 | 499 | EXPORT_SYMBOL(ceph_osdc_new_request); |
f24e9980 SW |
500 | |
501 | /* | |
502 | * We keep osd requests in an rbtree, sorted by ->r_tid. | |
503 | */ | |
504 | static void __insert_request(struct ceph_osd_client *osdc, | |
505 | struct ceph_osd_request *new) | |
506 | { | |
507 | struct rb_node **p = &osdc->requests.rb_node; | |
508 | struct rb_node *parent = NULL; | |
509 | struct ceph_osd_request *req = NULL; | |
510 | ||
511 | while (*p) { | |
512 | parent = *p; | |
513 | req = rb_entry(parent, struct ceph_osd_request, r_node); | |
514 | if (new->r_tid < req->r_tid) | |
515 | p = &(*p)->rb_left; | |
516 | else if (new->r_tid > req->r_tid) | |
517 | p = &(*p)->rb_right; | |
518 | else | |
519 | BUG(); | |
520 | } | |
521 | ||
522 | rb_link_node(&new->r_node, parent, p); | |
523 | rb_insert_color(&new->r_node, &osdc->requests); | |
524 | } | |
525 | ||
526 | static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc, | |
527 | u64 tid) | |
528 | { | |
529 | struct ceph_osd_request *req; | |
530 | struct rb_node *n = osdc->requests.rb_node; | |
531 | ||
532 | while (n) { | |
533 | req = rb_entry(n, struct ceph_osd_request, r_node); | |
534 | if (tid < req->r_tid) | |
535 | n = n->rb_left; | |
536 | else if (tid > req->r_tid) | |
537 | n = n->rb_right; | |
538 | else | |
539 | return req; | |
540 | } | |
541 | return NULL; | |
542 | } | |
543 | ||
544 | static struct ceph_osd_request * | |
545 | __lookup_request_ge(struct ceph_osd_client *osdc, | |
546 | u64 tid) | |
547 | { | |
548 | struct ceph_osd_request *req; | |
549 | struct rb_node *n = osdc->requests.rb_node; | |
550 | ||
551 | while (n) { | |
552 | req = rb_entry(n, struct ceph_osd_request, r_node); | |
553 | if (tid < req->r_tid) { | |
554 | if (!n->rb_left) | |
555 | return req; | |
556 | n = n->rb_left; | |
557 | } else if (tid > req->r_tid) { | |
558 | n = n->rb_right; | |
559 | } else { | |
560 | return req; | |
561 | } | |
562 | } | |
563 | return NULL; | |
564 | } | |
565 | ||
6f6c7006 SW |
566 | /* |
567 | * Resubmit requests pending on the given osd. | |
568 | */ | |
569 | static void __kick_osd_requests(struct ceph_osd_client *osdc, | |
570 | struct ceph_osd *osd) | |
571 | { | |
a40c4f10 | 572 | struct ceph_osd_request *req, *nreq; |
6f6c7006 SW |
573 | int err; |
574 | ||
575 | dout("__kick_osd_requests osd%d\n", osd->o_osd); | |
576 | err = __reset_osd(osdc, osd); | |
685a7555 | 577 | if (err) |
6f6c7006 SW |
578 | return; |
579 | ||
580 | list_for_each_entry(req, &osd->o_requests, r_osd_item) { | |
581 | list_move(&req->r_req_lru_item, &osdc->req_unsent); | |
582 | dout("requeued %p tid %llu osd%d\n", req, req->r_tid, | |
583 | osd->o_osd); | |
a40c4f10 YS |
584 | if (!req->r_linger) |
585 | req->r_flags |= CEPH_OSD_FLAG_RETRY; | |
586 | } | |
587 | ||
588 | list_for_each_entry_safe(req, nreq, &osd->o_linger_requests, | |
589 | r_linger_osd) { | |
77f38e0e SW |
590 | /* |
591 | * reregister request prior to unregistering linger so | |
592 | * that r_osd is preserved. | |
593 | */ | |
594 | BUG_ON(!list_empty(&req->r_req_lru_item)); | |
a40c4f10 | 595 | __register_request(osdc, req); |
77f38e0e SW |
596 | list_add(&req->r_req_lru_item, &osdc->req_unsent); |
597 | list_add(&req->r_osd_item, &req->r_osd->o_requests); | |
598 | __unregister_linger_request(osdc, req); | |
a40c4f10 YS |
599 | dout("requeued lingering %p tid %llu osd%d\n", req, req->r_tid, |
600 | osd->o_osd); | |
6f6c7006 SW |
601 | } |
602 | } | |
603 | ||
f24e9980 | 604 | /* |
81b024e7 | 605 | * If the osd connection drops, we need to resubmit all requests. |
f24e9980 SW |
606 | */ |
607 | static void osd_reset(struct ceph_connection *con) | |
608 | { | |
609 | struct ceph_osd *osd = con->private; | |
610 | struct ceph_osd_client *osdc; | |
611 | ||
612 | if (!osd) | |
613 | return; | |
614 | dout("osd_reset osd%d\n", osd->o_osd); | |
615 | osdc = osd->o_osdc; | |
f24e9980 | 616 | down_read(&osdc->map_sem); |
83aff95e SW |
617 | mutex_lock(&osdc->request_mutex); |
618 | __kick_osd_requests(osdc, osd); | |
f9d25199 | 619 | __send_queued(osdc); |
83aff95e | 620 | mutex_unlock(&osdc->request_mutex); |
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; |
f407731d | 638 | RB_CLEAR_NODE(&osd->o_node); |
f24e9980 | 639 | INIT_LIST_HEAD(&osd->o_requests); |
a40c4f10 | 640 | INIT_LIST_HEAD(&osd->o_linger_requests); |
f5a2041b | 641 | INIT_LIST_HEAD(&osd->o_osd_lru); |
f24e9980 SW |
642 | osd->o_incarnation = 1; |
643 | ||
b7a9e5dd | 644 | ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr); |
4e7a5dcd | 645 | |
422d2cb8 | 646 | INIT_LIST_HEAD(&osd->o_keepalive_item); |
f24e9980 SW |
647 | return osd; |
648 | } | |
649 | ||
650 | static struct ceph_osd *get_osd(struct ceph_osd *osd) | |
651 | { | |
652 | if (atomic_inc_not_zero(&osd->o_ref)) { | |
653 | dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1, | |
654 | atomic_read(&osd->o_ref)); | |
655 | return osd; | |
656 | } else { | |
657 | dout("get_osd %p FAIL\n", osd); | |
658 | return NULL; | |
659 | } | |
660 | } | |
661 | ||
662 | static void put_osd(struct ceph_osd *osd) | |
663 | { | |
664 | dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref), | |
665 | atomic_read(&osd->o_ref) - 1); | |
a255651d | 666 | if (atomic_dec_and_test(&osd->o_ref) && osd->o_auth.authorizer) { |
79494d1b SW |
667 | struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth; |
668 | ||
27859f97 | 669 | ceph_auth_destroy_authorizer(ac, osd->o_auth.authorizer); |
f24e9980 | 670 | kfree(osd); |
79494d1b | 671 | } |
f24e9980 SW |
672 | } |
673 | ||
674 | /* | |
675 | * remove an osd from our map | |
676 | */ | |
f5a2041b | 677 | static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd) |
f24e9980 | 678 | { |
f5a2041b | 679 | dout("__remove_osd %p\n", osd); |
f24e9980 SW |
680 | BUG_ON(!list_empty(&osd->o_requests)); |
681 | rb_erase(&osd->o_node, &osdc->osds); | |
f5a2041b | 682 | list_del_init(&osd->o_osd_lru); |
f24e9980 SW |
683 | ceph_con_close(&osd->o_con); |
684 | put_osd(osd); | |
685 | } | |
686 | ||
aca420bc SW |
687 | static void remove_all_osds(struct ceph_osd_client *osdc) |
688 | { | |
048a9d2d | 689 | dout("%s %p\n", __func__, osdc); |
aca420bc SW |
690 | mutex_lock(&osdc->request_mutex); |
691 | while (!RB_EMPTY_ROOT(&osdc->osds)) { | |
692 | struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds), | |
693 | struct ceph_osd, o_node); | |
694 | __remove_osd(osdc, osd); | |
695 | } | |
696 | mutex_unlock(&osdc->request_mutex); | |
697 | } | |
698 | ||
f5a2041b YS |
699 | static void __move_osd_to_lru(struct ceph_osd_client *osdc, |
700 | struct ceph_osd *osd) | |
701 | { | |
702 | dout("__move_osd_to_lru %p\n", osd); | |
703 | BUG_ON(!list_empty(&osd->o_osd_lru)); | |
704 | list_add_tail(&osd->o_osd_lru, &osdc->osd_lru); | |
3d14c5d2 | 705 | osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl * HZ; |
f5a2041b YS |
706 | } |
707 | ||
708 | static void __remove_osd_from_lru(struct ceph_osd *osd) | |
709 | { | |
710 | dout("__remove_osd_from_lru %p\n", osd); | |
711 | if (!list_empty(&osd->o_osd_lru)) | |
712 | list_del_init(&osd->o_osd_lru); | |
713 | } | |
714 | ||
aca420bc | 715 | static void remove_old_osds(struct ceph_osd_client *osdc) |
f5a2041b YS |
716 | { |
717 | struct ceph_osd *osd, *nosd; | |
718 | ||
719 | dout("__remove_old_osds %p\n", osdc); | |
720 | mutex_lock(&osdc->request_mutex); | |
721 | list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) { | |
aca420bc | 722 | if (time_before(jiffies, osd->lru_ttl)) |
f5a2041b YS |
723 | break; |
724 | __remove_osd(osdc, osd); | |
725 | } | |
726 | mutex_unlock(&osdc->request_mutex); | |
727 | } | |
728 | ||
f24e9980 SW |
729 | /* |
730 | * reset osd connect | |
731 | */ | |
f5a2041b | 732 | static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd) |
f24e9980 | 733 | { |
c3acb181 | 734 | struct ceph_entity_addr *peer_addr; |
f24e9980 | 735 | |
f5a2041b | 736 | dout("__reset_osd %p osd%d\n", osd, osd->o_osd); |
a40c4f10 YS |
737 | if (list_empty(&osd->o_requests) && |
738 | list_empty(&osd->o_linger_requests)) { | |
f5a2041b | 739 | __remove_osd(osdc, osd); |
c3acb181 AE |
740 | |
741 | return -ENODEV; | |
742 | } | |
743 | ||
744 | peer_addr = &osdc->osdmap->osd_addr[osd->o_osd]; | |
745 | if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) && | |
746 | !ceph_con_opened(&osd->o_con)) { | |
747 | struct ceph_osd_request *req; | |
748 | ||
87b315a5 SW |
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; | |
c3acb181 AE |
754 | |
755 | return -EAGAIN; | |
f24e9980 | 756 | } |
c3acb181 AE |
757 | |
758 | ceph_con_close(&osd->o_con); | |
759 | ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr); | |
760 | osd->o_incarnation++; | |
761 | ||
762 | return 0; | |
f24e9980 SW |
763 | } |
764 | ||
765 | static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new) | |
766 | { | |
767 | struct rb_node **p = &osdc->osds.rb_node; | |
768 | struct rb_node *parent = NULL; | |
769 | struct ceph_osd *osd = NULL; | |
770 | ||
aca420bc | 771 | dout("__insert_osd %p osd%d\n", new, new->o_osd); |
f24e9980 SW |
772 | while (*p) { |
773 | parent = *p; | |
774 | osd = rb_entry(parent, struct ceph_osd, o_node); | |
775 | if (new->o_osd < osd->o_osd) | |
776 | p = &(*p)->rb_left; | |
777 | else if (new->o_osd > osd->o_osd) | |
778 | p = &(*p)->rb_right; | |
779 | else | |
780 | BUG(); | |
781 | } | |
782 | ||
783 | rb_link_node(&new->o_node, parent, p); | |
784 | rb_insert_color(&new->o_node, &osdc->osds); | |
785 | } | |
786 | ||
787 | static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o) | |
788 | { | |
789 | struct ceph_osd *osd; | |
790 | struct rb_node *n = osdc->osds.rb_node; | |
791 | ||
792 | while (n) { | |
793 | osd = rb_entry(n, struct ceph_osd, o_node); | |
794 | if (o < osd->o_osd) | |
795 | n = n->rb_left; | |
796 | else if (o > osd->o_osd) | |
797 | n = n->rb_right; | |
798 | else | |
799 | return osd; | |
800 | } | |
801 | return NULL; | |
802 | } | |
803 | ||
422d2cb8 YS |
804 | static void __schedule_osd_timeout(struct ceph_osd_client *osdc) |
805 | { | |
806 | schedule_delayed_work(&osdc->timeout_work, | |
3d14c5d2 | 807 | osdc->client->options->osd_keepalive_timeout * HZ); |
422d2cb8 YS |
808 | } |
809 | ||
810 | static void __cancel_osd_timeout(struct ceph_osd_client *osdc) | |
811 | { | |
812 | cancel_delayed_work(&osdc->timeout_work); | |
813 | } | |
f24e9980 SW |
814 | |
815 | /* | |
816 | * Register request, assign tid. If this is the first request, set up | |
817 | * the timeout event. | |
818 | */ | |
a40c4f10 YS |
819 | static void __register_request(struct ceph_osd_client *osdc, |
820 | struct ceph_osd_request *req) | |
f24e9980 | 821 | { |
f24e9980 | 822 | req->r_tid = ++osdc->last_tid; |
6df058c0 | 823 | req->r_request->hdr.tid = cpu_to_le64(req->r_tid); |
77f38e0e | 824 | dout("__register_request %p tid %lld\n", req, req->r_tid); |
f24e9980 SW |
825 | __insert_request(osdc, req); |
826 | ceph_osdc_get_request(req); | |
827 | osdc->num_requests++; | |
f24e9980 | 828 | if (osdc->num_requests == 1) { |
422d2cb8 YS |
829 | dout(" first request, scheduling timeout\n"); |
830 | __schedule_osd_timeout(osdc); | |
f24e9980 | 831 | } |
a40c4f10 YS |
832 | } |
833 | ||
f24e9980 SW |
834 | /* |
835 | * called under osdc->request_mutex | |
836 | */ | |
837 | static void __unregister_request(struct ceph_osd_client *osdc, | |
838 | struct ceph_osd_request *req) | |
839 | { | |
35f9f8a0 SW |
840 | if (RB_EMPTY_NODE(&req->r_node)) { |
841 | dout("__unregister_request %p tid %lld not registered\n", | |
842 | req, req->r_tid); | |
843 | return; | |
844 | } | |
845 | ||
f24e9980 SW |
846 | dout("__unregister_request %p tid %lld\n", req, req->r_tid); |
847 | rb_erase(&req->r_node, &osdc->requests); | |
848 | osdc->num_requests--; | |
849 | ||
0ba6478d SW |
850 | if (req->r_osd) { |
851 | /* make sure the original request isn't in flight. */ | |
6740a845 | 852 | ceph_msg_revoke(req->r_request); |
0ba6478d SW |
853 | |
854 | list_del_init(&req->r_osd_item); | |
a40c4f10 YS |
855 | if (list_empty(&req->r_osd->o_requests) && |
856 | list_empty(&req->r_osd->o_linger_requests)) { | |
857 | dout("moving osd to %p lru\n", req->r_osd); | |
f5a2041b | 858 | __move_osd_to_lru(osdc, req->r_osd); |
a40c4f10 | 859 | } |
fbdb9190 | 860 | if (list_empty(&req->r_linger_item)) |
a40c4f10 | 861 | req->r_osd = NULL; |
0ba6478d | 862 | } |
f24e9980 | 863 | |
7d5f2481 | 864 | list_del_init(&req->r_req_lru_item); |
f24e9980 SW |
865 | ceph_osdc_put_request(req); |
866 | ||
422d2cb8 YS |
867 | if (osdc->num_requests == 0) { |
868 | dout(" no requests, canceling timeout\n"); | |
869 | __cancel_osd_timeout(osdc); | |
f24e9980 SW |
870 | } |
871 | } | |
872 | ||
873 | /* | |
874 | * Cancel a previously queued request message | |
875 | */ | |
876 | static void __cancel_request(struct ceph_osd_request *req) | |
877 | { | |
6bc18876 | 878 | if (req->r_sent && req->r_osd) { |
6740a845 | 879 | ceph_msg_revoke(req->r_request); |
f24e9980 SW |
880 | req->r_sent = 0; |
881 | } | |
882 | } | |
883 | ||
a40c4f10 YS |
884 | static void __register_linger_request(struct ceph_osd_client *osdc, |
885 | struct ceph_osd_request *req) | |
886 | { | |
887 | dout("__register_linger_request %p\n", req); | |
888 | list_add_tail(&req->r_linger_item, &osdc->req_linger); | |
6194ea89 SW |
889 | if (req->r_osd) |
890 | list_add_tail(&req->r_linger_osd, | |
891 | &req->r_osd->o_linger_requests); | |
a40c4f10 YS |
892 | } |
893 | ||
894 | static void __unregister_linger_request(struct ceph_osd_client *osdc, | |
895 | struct ceph_osd_request *req) | |
896 | { | |
897 | dout("__unregister_linger_request %p\n", req); | |
61c74035 | 898 | list_del_init(&req->r_linger_item); |
a40c4f10 | 899 | if (req->r_osd) { |
a40c4f10 YS |
900 | list_del_init(&req->r_linger_osd); |
901 | ||
902 | if (list_empty(&req->r_osd->o_requests) && | |
903 | list_empty(&req->r_osd->o_linger_requests)) { | |
904 | dout("moving osd to %p lru\n", req->r_osd); | |
905 | __move_osd_to_lru(osdc, req->r_osd); | |
906 | } | |
fbdb9190 SW |
907 | if (list_empty(&req->r_osd_item)) |
908 | req->r_osd = NULL; | |
a40c4f10 YS |
909 | } |
910 | } | |
911 | ||
912 | void ceph_osdc_unregister_linger_request(struct ceph_osd_client *osdc, | |
913 | struct ceph_osd_request *req) | |
914 | { | |
915 | mutex_lock(&osdc->request_mutex); | |
916 | if (req->r_linger) { | |
917 | __unregister_linger_request(osdc, req); | |
918 | ceph_osdc_put_request(req); | |
919 | } | |
920 | mutex_unlock(&osdc->request_mutex); | |
921 | } | |
922 | EXPORT_SYMBOL(ceph_osdc_unregister_linger_request); | |
923 | ||
924 | void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc, | |
925 | struct ceph_osd_request *req) | |
926 | { | |
927 | if (!req->r_linger) { | |
928 | dout("set_request_linger %p\n", req); | |
929 | req->r_linger = 1; | |
930 | /* | |
931 | * caller is now responsible for calling | |
932 | * unregister_linger_request | |
933 | */ | |
934 | ceph_osdc_get_request(req); | |
935 | } | |
936 | } | |
937 | EXPORT_SYMBOL(ceph_osdc_set_request_linger); | |
938 | ||
f24e9980 SW |
939 | /* |
940 | * Pick an osd (the first 'up' osd in the pg), allocate the osd struct | |
941 | * (as needed), and set the request r_osd appropriately. If there is | |
25985edc | 942 | * no up osd, set r_osd to NULL. Move the request to the appropriate list |
6f6c7006 | 943 | * (unsent, homeless) or leave on in-flight lru. |
f24e9980 SW |
944 | * |
945 | * Return 0 if unchanged, 1 if changed, or negative on error. | |
946 | * | |
947 | * Caller should hold map_sem for read and request_mutex. | |
948 | */ | |
6f6c7006 | 949 | static int __map_request(struct ceph_osd_client *osdc, |
38d6453c | 950 | struct ceph_osd_request *req, int force_resend) |
f24e9980 | 951 | { |
5b191d99 | 952 | struct ceph_pg pgid; |
d85b7056 SW |
953 | int acting[CEPH_PG_MAX_SIZE]; |
954 | int o = -1, num = 0; | |
f24e9980 | 955 | int err; |
f24e9980 | 956 | |
6f6c7006 | 957 | dout("map_request %p tid %lld\n", req, req->r_tid); |
41766f87 AE |
958 | err = ceph_calc_ceph_pg(&pgid, req->r_oid, osdc->osdmap, |
959 | ceph_file_layout_pg_pool(req->r_file_layout)); | |
6f6c7006 SW |
960 | if (err) { |
961 | list_move(&req->r_req_lru_item, &osdc->req_notarget); | |
f24e9980 | 962 | return err; |
6f6c7006 | 963 | } |
7740a42f SW |
964 | req->r_pgid = pgid; |
965 | ||
d85b7056 SW |
966 | err = ceph_calc_pg_acting(osdc->osdmap, pgid, acting); |
967 | if (err > 0) { | |
968 | o = acting[0]; | |
969 | num = err; | |
970 | } | |
f24e9980 | 971 | |
38d6453c SW |
972 | if ((!force_resend && |
973 | req->r_osd && req->r_osd->o_osd == o && | |
d85b7056 SW |
974 | req->r_sent >= req->r_osd->o_incarnation && |
975 | req->r_num_pg_osds == num && | |
976 | memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) || | |
f24e9980 SW |
977 | (req->r_osd == NULL && o == -1)) |
978 | return 0; /* no change */ | |
979 | ||
5b191d99 SW |
980 | dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n", |
981 | req->r_tid, pgid.pool, pgid.seed, o, | |
f24e9980 SW |
982 | req->r_osd ? req->r_osd->o_osd : -1); |
983 | ||
d85b7056 SW |
984 | /* record full pg acting set */ |
985 | memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num); | |
986 | req->r_num_pg_osds = num; | |
987 | ||
f24e9980 SW |
988 | if (req->r_osd) { |
989 | __cancel_request(req); | |
990 | list_del_init(&req->r_osd_item); | |
f24e9980 SW |
991 | req->r_osd = NULL; |
992 | } | |
993 | ||
994 | req->r_osd = __lookup_osd(osdc, o); | |
995 | if (!req->r_osd && o >= 0) { | |
c99eb1c7 | 996 | err = -ENOMEM; |
e10006f8 | 997 | req->r_osd = create_osd(osdc, o); |
6f6c7006 SW |
998 | if (!req->r_osd) { |
999 | list_move(&req->r_req_lru_item, &osdc->req_notarget); | |
c99eb1c7 | 1000 | goto out; |
6f6c7006 | 1001 | } |
f24e9980 | 1002 | |
6f6c7006 | 1003 | dout("map_request osd %p is osd%d\n", req->r_osd, o); |
f24e9980 SW |
1004 | __insert_osd(osdc, req->r_osd); |
1005 | ||
b7a9e5dd SW |
1006 | ceph_con_open(&req->r_osd->o_con, |
1007 | CEPH_ENTITY_TYPE_OSD, o, | |
1008 | &osdc->osdmap->osd_addr[o]); | |
f24e9980 SW |
1009 | } |
1010 | ||
f5a2041b YS |
1011 | if (req->r_osd) { |
1012 | __remove_osd_from_lru(req->r_osd); | |
f24e9980 | 1013 | list_add(&req->r_osd_item, &req->r_osd->o_requests); |
6f6c7006 SW |
1014 | list_move(&req->r_req_lru_item, &osdc->req_unsent); |
1015 | } else { | |
1016 | list_move(&req->r_req_lru_item, &osdc->req_notarget); | |
f5a2041b | 1017 | } |
d85b7056 | 1018 | err = 1; /* osd or pg changed */ |
f24e9980 SW |
1019 | |
1020 | out: | |
f24e9980 SW |
1021 | return err; |
1022 | } | |
1023 | ||
1024 | /* | |
1025 | * caller should hold map_sem (for read) and request_mutex | |
1026 | */ | |
56e925b6 SW |
1027 | static void __send_request(struct ceph_osd_client *osdc, |
1028 | struct ceph_osd_request *req) | |
f24e9980 | 1029 | { |
1b83bef2 | 1030 | void *p; |
f24e9980 | 1031 | |
1b83bef2 SW |
1032 | dout("send_request %p tid %llu to osd%d flags %d pg %lld.%x\n", |
1033 | req, req->r_tid, req->r_osd->o_osd, req->r_flags, | |
1034 | (unsigned long long)req->r_pgid.pool, req->r_pgid.seed); | |
1035 | ||
1036 | /* fill in message content that changes each time we send it */ | |
1037 | put_unaligned_le32(osdc->osdmap->epoch, req->r_request_osdmap_epoch); | |
1038 | put_unaligned_le32(req->r_flags, req->r_request_flags); | |
1039 | put_unaligned_le64(req->r_pgid.pool, req->r_request_pool); | |
1040 | p = req->r_request_pgid; | |
1041 | ceph_encode_64(&p, req->r_pgid.pool); | |
1042 | ceph_encode_32(&p, req->r_pgid.seed); | |
1043 | put_unaligned_le64(1, req->r_request_attempts); /* FIXME */ | |
1044 | memcpy(req->r_request_reassert_version, &req->r_reassert_version, | |
1045 | sizeof(req->r_reassert_version)); | |
2169aea6 | 1046 | |
3dd72fc0 | 1047 | req->r_stamp = jiffies; |
07a27e22 | 1048 | list_move_tail(&req->r_req_lru_item, &osdc->req_lru); |
f24e9980 SW |
1049 | |
1050 | ceph_msg_get(req->r_request); /* send consumes a ref */ | |
1051 | ceph_con_send(&req->r_osd->o_con, req->r_request); | |
1052 | req->r_sent = req->r_osd->o_incarnation; | |
f24e9980 SW |
1053 | } |
1054 | ||
6f6c7006 SW |
1055 | /* |
1056 | * Send any requests in the queue (req_unsent). | |
1057 | */ | |
f9d25199 | 1058 | static void __send_queued(struct ceph_osd_client *osdc) |
6f6c7006 SW |
1059 | { |
1060 | struct ceph_osd_request *req, *tmp; | |
1061 | ||
f9d25199 AE |
1062 | dout("__send_queued\n"); |
1063 | list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item) | |
6f6c7006 | 1064 | __send_request(osdc, req); |
6f6c7006 SW |
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); | |
83aff95e | 1080 | struct ceph_osd_request *req; |
f24e9980 | 1081 | struct ceph_osd *osd; |
422d2cb8 | 1082 | unsigned long keepalive = |
3d14c5d2 | 1083 | osdc->client->options->osd_keepalive_timeout * HZ; |
422d2cb8 | 1084 | struct list_head slow_osds; |
f24e9980 SW |
1085 | dout("timeout\n"); |
1086 | down_read(&osdc->map_sem); | |
1087 | ||
1088 | ceph_monc_request_next_osdmap(&osdc->client->monc); | |
1089 | ||
1090 | mutex_lock(&osdc->request_mutex); | |
f24e9980 | 1091 | |
422d2cb8 YS |
1092 | /* |
1093 | * ping osds that are a bit slow. this ensures that if there | |
1094 | * is a break in the TCP connection we will notice, and reopen | |
1095 | * a connection with that osd (from the fault callback). | |
1096 | */ | |
1097 | INIT_LIST_HEAD(&slow_osds); | |
1098 | list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) { | |
3dd72fc0 | 1099 | if (time_before(jiffies, req->r_stamp + keepalive)) |
422d2cb8 YS |
1100 | break; |
1101 | ||
1102 | osd = req->r_osd; | |
1103 | BUG_ON(!osd); | |
1104 | dout(" tid %llu is slow, will send keepalive on osd%d\n", | |
f24e9980 | 1105 | req->r_tid, osd->o_osd); |
422d2cb8 YS |
1106 | list_move_tail(&osd->o_keepalive_item, &slow_osds); |
1107 | } | |
1108 | while (!list_empty(&slow_osds)) { | |
1109 | osd = list_entry(slow_osds.next, struct ceph_osd, | |
1110 | o_keepalive_item); | |
1111 | list_del_init(&osd->o_keepalive_item); | |
f24e9980 SW |
1112 | ceph_con_keepalive(&osd->o_con); |
1113 | } | |
1114 | ||
422d2cb8 | 1115 | __schedule_osd_timeout(osdc); |
f9d25199 | 1116 | __send_queued(osdc); |
f24e9980 | 1117 | mutex_unlock(&osdc->request_mutex); |
f24e9980 SW |
1118 | up_read(&osdc->map_sem); |
1119 | } | |
1120 | ||
f5a2041b YS |
1121 | static void handle_osds_timeout(struct work_struct *work) |
1122 | { | |
1123 | struct ceph_osd_client *osdc = | |
1124 | container_of(work, struct ceph_osd_client, | |
1125 | osds_timeout_work.work); | |
1126 | unsigned long delay = | |
3d14c5d2 | 1127 | osdc->client->options->osd_idle_ttl * HZ >> 2; |
f5a2041b YS |
1128 | |
1129 | dout("osds timeout\n"); | |
1130 | down_read(&osdc->map_sem); | |
aca420bc | 1131 | remove_old_osds(osdc); |
f5a2041b YS |
1132 | up_read(&osdc->map_sem); |
1133 | ||
1134 | schedule_delayed_work(&osdc->osds_timeout_work, | |
1135 | round_jiffies_relative(delay)); | |
1136 | } | |
1137 | ||
25845472 SW |
1138 | static void complete_request(struct ceph_osd_request *req) |
1139 | { | |
1140 | if (req->r_safe_callback) | |
1141 | req->r_safe_callback(req, NULL); | |
1142 | complete_all(&req->r_safe_completion); /* fsync waiter */ | |
1143 | } | |
1144 | ||
1b83bef2 SW |
1145 | static int __decode_pgid(void **p, void *end, struct ceph_pg *pgid) |
1146 | { | |
1147 | __u8 v; | |
1148 | ||
1149 | ceph_decode_need(p, end, 1 + 8 + 4 + 4, bad); | |
1150 | v = ceph_decode_8(p); | |
1151 | if (v > 1) { | |
1152 | pr_warning("do not understand pg encoding %d > 1", v); | |
1153 | return -EINVAL; | |
1154 | } | |
1155 | pgid->pool = ceph_decode_64(p); | |
1156 | pgid->seed = ceph_decode_32(p); | |
1157 | *p += 4; | |
1158 | return 0; | |
1159 | ||
1160 | bad: | |
1161 | pr_warning("incomplete pg encoding"); | |
1162 | return -EINVAL; | |
1163 | } | |
1164 | ||
f24e9980 SW |
1165 | /* |
1166 | * handle osd op reply. either call the callback if it is specified, | |
1167 | * or do the completion to wake up the waiting thread. | |
1168 | */ | |
350b1c32 SW |
1169 | static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg, |
1170 | struct ceph_connection *con) | |
f24e9980 | 1171 | { |
1b83bef2 | 1172 | void *p, *end; |
f24e9980 SW |
1173 | struct ceph_osd_request *req; |
1174 | u64 tid; | |
1b83bef2 SW |
1175 | int object_len; |
1176 | int numops, payload_len, flags; | |
0ceed5db | 1177 | s32 result; |
1b83bef2 SW |
1178 | s32 retry_attempt; |
1179 | struct ceph_pg pg; | |
1180 | int err; | |
1181 | u32 reassert_epoch; | |
1182 | u64 reassert_version; | |
1183 | u32 osdmap_epoch; | |
0d5af164 | 1184 | int already_completed; |
1b83bef2 | 1185 | int i; |
f24e9980 | 1186 | |
6df058c0 | 1187 | tid = le64_to_cpu(msg->hdr.tid); |
1b83bef2 SW |
1188 | dout("handle_reply %p tid %llu\n", msg, tid); |
1189 | ||
1190 | p = msg->front.iov_base; | |
1191 | end = p + msg->front.iov_len; | |
1192 | ||
1193 | ceph_decode_need(&p, end, 4, bad); | |
1194 | object_len = ceph_decode_32(&p); | |
1195 | ceph_decode_need(&p, end, object_len, bad); | |
1196 | p += object_len; | |
1197 | ||
1198 | err = __decode_pgid(&p, end, &pg); | |
1199 | if (err) | |
f24e9980 | 1200 | goto bad; |
1b83bef2 SW |
1201 | |
1202 | ceph_decode_need(&p, end, 8 + 4 + 4 + 8 + 4, bad); | |
1203 | flags = ceph_decode_64(&p); | |
1204 | result = ceph_decode_32(&p); | |
1205 | reassert_epoch = ceph_decode_32(&p); | |
1206 | reassert_version = ceph_decode_64(&p); | |
1207 | osdmap_epoch = ceph_decode_32(&p); | |
1208 | ||
f24e9980 SW |
1209 | /* lookup */ |
1210 | mutex_lock(&osdc->request_mutex); | |
1211 | req = __lookup_request(osdc, tid); | |
1212 | if (req == NULL) { | |
1213 | dout("handle_reply tid %llu dne\n", tid); | |
1214 | mutex_unlock(&osdc->request_mutex); | |
1215 | return; | |
1216 | } | |
1217 | ceph_osdc_get_request(req); | |
1b83bef2 SW |
1218 | |
1219 | dout("handle_reply %p tid %llu req %p result %d\n", msg, tid, | |
1220 | req, result); | |
1221 | ||
1222 | ceph_decode_need(&p, end, 4, bad); | |
1223 | numops = ceph_decode_32(&p); | |
1224 | if (numops > CEPH_OSD_MAX_OP) | |
1225 | goto bad_put; | |
1226 | if (numops != req->r_num_ops) | |
1227 | goto bad_put; | |
1228 | payload_len = 0; | |
1229 | ceph_decode_need(&p, end, numops * sizeof(struct ceph_osd_op), bad); | |
1230 | for (i = 0; i < numops; i++) { | |
1231 | struct ceph_osd_op *op = p; | |
1232 | int len; | |
1233 | ||
1234 | len = le32_to_cpu(op->payload_len); | |
1235 | req->r_reply_op_len[i] = len; | |
1236 | dout(" op %d has %d bytes\n", i, len); | |
1237 | payload_len += len; | |
1238 | p += sizeof(*op); | |
1239 | } | |
1240 | if (payload_len != le32_to_cpu(msg->hdr.data_len)) { | |
1241 | pr_warning("sum of op payload lens %d != data_len %d", | |
1242 | payload_len, le32_to_cpu(msg->hdr.data_len)); | |
1243 | goto bad_put; | |
1244 | } | |
1245 | ||
1246 | ceph_decode_need(&p, end, 4 + numops * 4, bad); | |
1247 | retry_attempt = ceph_decode_32(&p); | |
1248 | for (i = 0; i < numops; i++) | |
1249 | req->r_reply_op_result[i] = ceph_decode_32(&p); | |
f24e9980 | 1250 | |
350b1c32 | 1251 | /* |
0d59ab81 | 1252 | * if this connection filled our message, drop our reference now, to |
350b1c32 SW |
1253 | * avoid a (safe but slower) revoke later. |
1254 | */ | |
0d59ab81 | 1255 | if (req->r_con_filling_msg == con && req->r_reply == msg) { |
c16e7869 | 1256 | dout(" dropping con_filling_msg ref %p\n", con); |
0d59ab81 | 1257 | req->r_con_filling_msg = NULL; |
0d47766f | 1258 | con->ops->put(con); |
350b1c32 SW |
1259 | } |
1260 | ||
f24e9980 | 1261 | if (!req->r_got_reply) { |
95c96174 | 1262 | unsigned int bytes; |
f24e9980 | 1263 | |
1b83bef2 | 1264 | req->r_result = result; |
f24e9980 SW |
1265 | bytes = le32_to_cpu(msg->hdr.data_len); |
1266 | dout("handle_reply result %d bytes %d\n", req->r_result, | |
1267 | bytes); | |
1268 | if (req->r_result == 0) | |
1269 | req->r_result = bytes; | |
1270 | ||
1271 | /* in case this is a write and we need to replay, */ | |
1b83bef2 SW |
1272 | req->r_reassert_version.epoch = cpu_to_le32(reassert_epoch); |
1273 | req->r_reassert_version.version = cpu_to_le64(reassert_version); | |
f24e9980 SW |
1274 | |
1275 | req->r_got_reply = 1; | |
1276 | } else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) { | |
1277 | dout("handle_reply tid %llu dup ack\n", tid); | |
34b43a56 | 1278 | mutex_unlock(&osdc->request_mutex); |
f24e9980 SW |
1279 | goto done; |
1280 | } | |
1281 | ||
1282 | dout("handle_reply tid %llu flags %d\n", tid, flags); | |
1283 | ||
a40c4f10 YS |
1284 | if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK)) |
1285 | __register_linger_request(osdc, req); | |
1286 | ||
f24e9980 | 1287 | /* either this is a read, or we got the safe response */ |
0ceed5db SW |
1288 | if (result < 0 || |
1289 | (flags & CEPH_OSD_FLAG_ONDISK) || | |
f24e9980 SW |
1290 | ((flags & CEPH_OSD_FLAG_WRITE) == 0)) |
1291 | __unregister_request(osdc, req); | |
1292 | ||
0d5af164 AE |
1293 | already_completed = req->r_completed; |
1294 | req->r_completed = 1; | |
f24e9980 | 1295 | mutex_unlock(&osdc->request_mutex); |
0d5af164 AE |
1296 | if (already_completed) |
1297 | goto done; | |
f24e9980 SW |
1298 | |
1299 | if (req->r_callback) | |
1300 | req->r_callback(req, msg); | |
1301 | else | |
03066f23 | 1302 | complete_all(&req->r_completion); |
f24e9980 | 1303 | |
25845472 SW |
1304 | if (flags & CEPH_OSD_FLAG_ONDISK) |
1305 | complete_request(req); | |
f24e9980 SW |
1306 | |
1307 | done: | |
a40c4f10 | 1308 | dout("req=%p req->r_linger=%d\n", req, req->r_linger); |
f24e9980 SW |
1309 | ceph_osdc_put_request(req); |
1310 | return; | |
1311 | ||
1b83bef2 SW |
1312 | bad_put: |
1313 | ceph_osdc_put_request(req); | |
f24e9980 | 1314 | bad: |
1b83bef2 SW |
1315 | pr_err("corrupt osd_op_reply got %d %d\n", |
1316 | (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len)); | |
9ec7cab1 | 1317 | ceph_msg_dump(msg); |
f24e9980 SW |
1318 | } |
1319 | ||
6f6c7006 | 1320 | static void reset_changed_osds(struct ceph_osd_client *osdc) |
f24e9980 | 1321 | { |
f24e9980 | 1322 | struct rb_node *p, *n; |
f24e9980 | 1323 | |
6f6c7006 SW |
1324 | for (p = rb_first(&osdc->osds); p; p = n) { |
1325 | struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node); | |
f24e9980 | 1326 | |
6f6c7006 SW |
1327 | n = rb_next(p); |
1328 | if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) || | |
1329 | memcmp(&osd->o_con.peer_addr, | |
1330 | ceph_osd_addr(osdc->osdmap, | |
1331 | osd->o_osd), | |
1332 | sizeof(struct ceph_entity_addr)) != 0) | |
1333 | __reset_osd(osdc, osd); | |
f24e9980 | 1334 | } |
422d2cb8 YS |
1335 | } |
1336 | ||
1337 | /* | |
6f6c7006 SW |
1338 | * Requeue requests whose mapping to an OSD has changed. If requests map to |
1339 | * no osd, request a new map. | |
422d2cb8 | 1340 | * |
e6d50f67 | 1341 | * Caller should hold map_sem for read. |
422d2cb8 | 1342 | */ |
38d6453c | 1343 | static void kick_requests(struct ceph_osd_client *osdc, int force_resend) |
422d2cb8 | 1344 | { |
a40c4f10 | 1345 | struct ceph_osd_request *req, *nreq; |
6f6c7006 SW |
1346 | struct rb_node *p; |
1347 | int needmap = 0; | |
1348 | int err; | |
422d2cb8 | 1349 | |
38d6453c | 1350 | dout("kick_requests %s\n", force_resend ? " (force resend)" : ""); |
422d2cb8 | 1351 | mutex_lock(&osdc->request_mutex); |
6194ea89 | 1352 | for (p = rb_first(&osdc->requests); p; ) { |
6f6c7006 | 1353 | req = rb_entry(p, struct ceph_osd_request, r_node); |
6194ea89 | 1354 | p = rb_next(p); |
ab60b16d AE |
1355 | |
1356 | /* | |
1357 | * For linger requests that have not yet been | |
1358 | * registered, move them to the linger list; they'll | |
1359 | * be sent to the osd in the loop below. Unregister | |
1360 | * the request before re-registering it as a linger | |
1361 | * request to ensure the __map_request() below | |
1362 | * will decide it needs to be sent. | |
1363 | */ | |
1364 | if (req->r_linger && list_empty(&req->r_linger_item)) { | |
1365 | dout("%p tid %llu restart on osd%d\n", | |
1366 | req, req->r_tid, | |
1367 | req->r_osd ? req->r_osd->o_osd : -1); | |
1368 | __unregister_request(osdc, req); | |
1369 | __register_linger_request(osdc, req); | |
1370 | continue; | |
1371 | } | |
1372 | ||
38d6453c | 1373 | err = __map_request(osdc, req, force_resend); |
6f6c7006 SW |
1374 | if (err < 0) |
1375 | continue; /* error */ | |
1376 | if (req->r_osd == NULL) { | |
1377 | dout("%p tid %llu maps to no osd\n", req, req->r_tid); | |
1378 | needmap++; /* request a newer map */ | |
1379 | } else if (err > 0) { | |
6194ea89 SW |
1380 | if (!req->r_linger) { |
1381 | dout("%p tid %llu requeued on osd%d\n", req, | |
1382 | req->r_tid, | |
1383 | req->r_osd ? req->r_osd->o_osd : -1); | |
a40c4f10 | 1384 | req->r_flags |= CEPH_OSD_FLAG_RETRY; |
6194ea89 SW |
1385 | } |
1386 | } | |
a40c4f10 YS |
1387 | } |
1388 | ||
1389 | list_for_each_entry_safe(req, nreq, &osdc->req_linger, | |
1390 | r_linger_item) { | |
1391 | dout("linger req=%p req->r_osd=%p\n", req, req->r_osd); | |
1392 | ||
38d6453c | 1393 | err = __map_request(osdc, req, force_resend); |
ab60b16d | 1394 | dout("__map_request returned %d\n", err); |
a40c4f10 YS |
1395 | if (err == 0) |
1396 | continue; /* no change and no osd was specified */ | |
1397 | if (err < 0) | |
1398 | continue; /* hrm! */ | |
1399 | if (req->r_osd == NULL) { | |
1400 | dout("tid %llu maps to no valid osd\n", req->r_tid); | |
1401 | needmap++; /* request a newer map */ | |
1402 | continue; | |
6f6c7006 | 1403 | } |
a40c4f10 YS |
1404 | |
1405 | dout("kicking lingering %p tid %llu osd%d\n", req, req->r_tid, | |
1406 | req->r_osd ? req->r_osd->o_osd : -1); | |
a40c4f10 | 1407 | __register_request(osdc, req); |
c89ce05e | 1408 | __unregister_linger_request(osdc, req); |
6f6c7006 | 1409 | } |
f24e9980 SW |
1410 | mutex_unlock(&osdc->request_mutex); |
1411 | ||
1412 | if (needmap) { | |
1413 | dout("%d requests for down osds, need new map\n", needmap); | |
1414 | ceph_monc_request_next_osdmap(&osdc->client->monc); | |
1415 | } | |
e6d50f67 | 1416 | reset_changed_osds(osdc); |
422d2cb8 | 1417 | } |
6f6c7006 SW |
1418 | |
1419 | ||
f24e9980 SW |
1420 | /* |
1421 | * Process updated osd map. | |
1422 | * | |
1423 | * The message contains any number of incremental and full maps, normally | |
1424 | * indicating some sort of topology change in the cluster. Kick requests | |
1425 | * off to different OSDs as needed. | |
1426 | */ | |
1427 | void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg) | |
1428 | { | |
1429 | void *p, *end, *next; | |
1430 | u32 nr_maps, maplen; | |
1431 | u32 epoch; | |
1432 | struct ceph_osdmap *newmap = NULL, *oldmap; | |
1433 | int err; | |
1434 | struct ceph_fsid fsid; | |
1435 | ||
1436 | dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0); | |
1437 | p = msg->front.iov_base; | |
1438 | end = p + msg->front.iov_len; | |
1439 | ||
1440 | /* verify fsid */ | |
1441 | ceph_decode_need(&p, end, sizeof(fsid), bad); | |
1442 | ceph_decode_copy(&p, &fsid, sizeof(fsid)); | |
0743304d SW |
1443 | if (ceph_check_fsid(osdc->client, &fsid) < 0) |
1444 | return; | |
f24e9980 SW |
1445 | |
1446 | down_write(&osdc->map_sem); | |
1447 | ||
1448 | /* incremental maps */ | |
1449 | ceph_decode_32_safe(&p, end, nr_maps, bad); | |
1450 | dout(" %d inc maps\n", nr_maps); | |
1451 | while (nr_maps > 0) { | |
1452 | ceph_decode_need(&p, end, 2*sizeof(u32), bad); | |
c89136ea SW |
1453 | epoch = ceph_decode_32(&p); |
1454 | maplen = ceph_decode_32(&p); | |
f24e9980 SW |
1455 | ceph_decode_need(&p, end, maplen, bad); |
1456 | next = p + maplen; | |
1457 | if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) { | |
1458 | dout("applying incremental map %u len %d\n", | |
1459 | epoch, maplen); | |
1460 | newmap = osdmap_apply_incremental(&p, next, | |
1461 | osdc->osdmap, | |
15d9882c | 1462 | &osdc->client->msgr); |
f24e9980 SW |
1463 | if (IS_ERR(newmap)) { |
1464 | err = PTR_ERR(newmap); | |
1465 | goto bad; | |
1466 | } | |
30dc6381 | 1467 | BUG_ON(!newmap); |
f24e9980 SW |
1468 | if (newmap != osdc->osdmap) { |
1469 | ceph_osdmap_destroy(osdc->osdmap); | |
1470 | osdc->osdmap = newmap; | |
1471 | } | |
38d6453c | 1472 | kick_requests(osdc, 0); |
f24e9980 SW |
1473 | } else { |
1474 | dout("ignoring incremental map %u len %d\n", | |
1475 | epoch, maplen); | |
1476 | } | |
1477 | p = next; | |
1478 | nr_maps--; | |
1479 | } | |
1480 | if (newmap) | |
1481 | goto done; | |
1482 | ||
1483 | /* full maps */ | |
1484 | ceph_decode_32_safe(&p, end, nr_maps, bad); | |
1485 | dout(" %d full maps\n", nr_maps); | |
1486 | while (nr_maps) { | |
1487 | ceph_decode_need(&p, end, 2*sizeof(u32), bad); | |
c89136ea SW |
1488 | epoch = ceph_decode_32(&p); |
1489 | maplen = ceph_decode_32(&p); | |
f24e9980 SW |
1490 | ceph_decode_need(&p, end, maplen, bad); |
1491 | if (nr_maps > 1) { | |
1492 | dout("skipping non-latest full map %u len %d\n", | |
1493 | epoch, maplen); | |
1494 | } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) { | |
1495 | dout("skipping full map %u len %d, " | |
1496 | "older than our %u\n", epoch, maplen, | |
1497 | osdc->osdmap->epoch); | |
1498 | } else { | |
38d6453c SW |
1499 | int skipped_map = 0; |
1500 | ||
f24e9980 SW |
1501 | dout("taking full map %u len %d\n", epoch, maplen); |
1502 | newmap = osdmap_decode(&p, p+maplen); | |
1503 | if (IS_ERR(newmap)) { | |
1504 | err = PTR_ERR(newmap); | |
1505 | goto bad; | |
1506 | } | |
30dc6381 | 1507 | BUG_ON(!newmap); |
f24e9980 SW |
1508 | oldmap = osdc->osdmap; |
1509 | osdc->osdmap = newmap; | |
38d6453c SW |
1510 | if (oldmap) { |
1511 | if (oldmap->epoch + 1 < newmap->epoch) | |
1512 | skipped_map = 1; | |
f24e9980 | 1513 | ceph_osdmap_destroy(oldmap); |
38d6453c SW |
1514 | } |
1515 | kick_requests(osdc, skipped_map); | |
f24e9980 SW |
1516 | } |
1517 | p += maplen; | |
1518 | nr_maps--; | |
1519 | } | |
1520 | ||
1521 | done: | |
1522 | downgrade_write(&osdc->map_sem); | |
1523 | ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch); | |
cd634fb6 SW |
1524 | |
1525 | /* | |
1526 | * subscribe to subsequent osdmap updates if full to ensure | |
1527 | * we find out when we are no longer full and stop returning | |
1528 | * ENOSPC. | |
1529 | */ | |
1530 | if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL)) | |
1531 | ceph_monc_request_next_osdmap(&osdc->client->monc); | |
1532 | ||
f9d25199 AE |
1533 | mutex_lock(&osdc->request_mutex); |
1534 | __send_queued(osdc); | |
1535 | mutex_unlock(&osdc->request_mutex); | |
f24e9980 | 1536 | up_read(&osdc->map_sem); |
03066f23 | 1537 | wake_up_all(&osdc->client->auth_wq); |
f24e9980 SW |
1538 | return; |
1539 | ||
1540 | bad: | |
1541 | pr_err("osdc handle_map corrupt msg\n"); | |
9ec7cab1 | 1542 | ceph_msg_dump(msg); |
f24e9980 SW |
1543 | up_write(&osdc->map_sem); |
1544 | return; | |
1545 | } | |
1546 | ||
a40c4f10 YS |
1547 | /* |
1548 | * watch/notify callback event infrastructure | |
1549 | * | |
1550 | * These callbacks are used both for watch and notify operations. | |
1551 | */ | |
1552 | static void __release_event(struct kref *kref) | |
1553 | { | |
1554 | struct ceph_osd_event *event = | |
1555 | container_of(kref, struct ceph_osd_event, kref); | |
1556 | ||
1557 | dout("__release_event %p\n", event); | |
1558 | kfree(event); | |
1559 | } | |
1560 | ||
1561 | static void get_event(struct ceph_osd_event *event) | |
1562 | { | |
1563 | kref_get(&event->kref); | |
1564 | } | |
1565 | ||
1566 | void ceph_osdc_put_event(struct ceph_osd_event *event) | |
1567 | { | |
1568 | kref_put(&event->kref, __release_event); | |
1569 | } | |
1570 | EXPORT_SYMBOL(ceph_osdc_put_event); | |
1571 | ||
1572 | static void __insert_event(struct ceph_osd_client *osdc, | |
1573 | struct ceph_osd_event *new) | |
1574 | { | |
1575 | struct rb_node **p = &osdc->event_tree.rb_node; | |
1576 | struct rb_node *parent = NULL; | |
1577 | struct ceph_osd_event *event = NULL; | |
1578 | ||
1579 | while (*p) { | |
1580 | parent = *p; | |
1581 | event = rb_entry(parent, struct ceph_osd_event, node); | |
1582 | if (new->cookie < event->cookie) | |
1583 | p = &(*p)->rb_left; | |
1584 | else if (new->cookie > event->cookie) | |
1585 | p = &(*p)->rb_right; | |
1586 | else | |
1587 | BUG(); | |
1588 | } | |
1589 | ||
1590 | rb_link_node(&new->node, parent, p); | |
1591 | rb_insert_color(&new->node, &osdc->event_tree); | |
1592 | } | |
1593 | ||
1594 | static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc, | |
1595 | u64 cookie) | |
1596 | { | |
1597 | struct rb_node **p = &osdc->event_tree.rb_node; | |
1598 | struct rb_node *parent = NULL; | |
1599 | struct ceph_osd_event *event = NULL; | |
1600 | ||
1601 | while (*p) { | |
1602 | parent = *p; | |
1603 | event = rb_entry(parent, struct ceph_osd_event, node); | |
1604 | if (cookie < event->cookie) | |
1605 | p = &(*p)->rb_left; | |
1606 | else if (cookie > event->cookie) | |
1607 | p = &(*p)->rb_right; | |
1608 | else | |
1609 | return event; | |
1610 | } | |
1611 | return NULL; | |
1612 | } | |
1613 | ||
1614 | static void __remove_event(struct ceph_osd_event *event) | |
1615 | { | |
1616 | struct ceph_osd_client *osdc = event->osdc; | |
1617 | ||
1618 | if (!RB_EMPTY_NODE(&event->node)) { | |
1619 | dout("__remove_event removed %p\n", event); | |
1620 | rb_erase(&event->node, &osdc->event_tree); | |
1621 | ceph_osdc_put_event(event); | |
1622 | } else { | |
1623 | dout("__remove_event didn't remove %p\n", event); | |
1624 | } | |
1625 | } | |
1626 | ||
1627 | int ceph_osdc_create_event(struct ceph_osd_client *osdc, | |
1628 | void (*event_cb)(u64, u64, u8, void *), | |
3c663bbd | 1629 | void *data, struct ceph_osd_event **pevent) |
a40c4f10 YS |
1630 | { |
1631 | struct ceph_osd_event *event; | |
1632 | ||
1633 | event = kmalloc(sizeof(*event), GFP_NOIO); | |
1634 | if (!event) | |
1635 | return -ENOMEM; | |
1636 | ||
1637 | dout("create_event %p\n", event); | |
1638 | event->cb = event_cb; | |
3c663bbd | 1639 | event->one_shot = 0; |
a40c4f10 YS |
1640 | event->data = data; |
1641 | event->osdc = osdc; | |
1642 | INIT_LIST_HEAD(&event->osd_node); | |
3ee5234d | 1643 | RB_CLEAR_NODE(&event->node); |
a40c4f10 YS |
1644 | kref_init(&event->kref); /* one ref for us */ |
1645 | kref_get(&event->kref); /* one ref for the caller */ | |
a40c4f10 YS |
1646 | |
1647 | spin_lock(&osdc->event_lock); | |
1648 | event->cookie = ++osdc->event_count; | |
1649 | __insert_event(osdc, event); | |
1650 | spin_unlock(&osdc->event_lock); | |
1651 | ||
1652 | *pevent = event; | |
1653 | return 0; | |
1654 | } | |
1655 | EXPORT_SYMBOL(ceph_osdc_create_event); | |
1656 | ||
1657 | void ceph_osdc_cancel_event(struct ceph_osd_event *event) | |
1658 | { | |
1659 | struct ceph_osd_client *osdc = event->osdc; | |
1660 | ||
1661 | dout("cancel_event %p\n", event); | |
1662 | spin_lock(&osdc->event_lock); | |
1663 | __remove_event(event); | |
1664 | spin_unlock(&osdc->event_lock); | |
1665 | ceph_osdc_put_event(event); /* caller's */ | |
1666 | } | |
1667 | EXPORT_SYMBOL(ceph_osdc_cancel_event); | |
1668 | ||
1669 | ||
1670 | static void do_event_work(struct work_struct *work) | |
1671 | { | |
1672 | struct ceph_osd_event_work *event_work = | |
1673 | container_of(work, struct ceph_osd_event_work, work); | |
1674 | struct ceph_osd_event *event = event_work->event; | |
1675 | u64 ver = event_work->ver; | |
1676 | u64 notify_id = event_work->notify_id; | |
1677 | u8 opcode = event_work->opcode; | |
1678 | ||
1679 | dout("do_event_work completing %p\n", event); | |
1680 | event->cb(ver, notify_id, opcode, event->data); | |
a40c4f10 YS |
1681 | dout("do_event_work completed %p\n", event); |
1682 | ceph_osdc_put_event(event); | |
1683 | kfree(event_work); | |
1684 | } | |
1685 | ||
1686 | ||
1687 | /* | |
1688 | * Process osd watch notifications | |
1689 | */ | |
3c663bbd AE |
1690 | static void handle_watch_notify(struct ceph_osd_client *osdc, |
1691 | struct ceph_msg *msg) | |
a40c4f10 YS |
1692 | { |
1693 | void *p, *end; | |
1694 | u8 proto_ver; | |
1695 | u64 cookie, ver, notify_id; | |
1696 | u8 opcode; | |
1697 | struct ceph_osd_event *event; | |
1698 | struct ceph_osd_event_work *event_work; | |
1699 | ||
1700 | p = msg->front.iov_base; | |
1701 | end = p + msg->front.iov_len; | |
1702 | ||
1703 | ceph_decode_8_safe(&p, end, proto_ver, bad); | |
1704 | ceph_decode_8_safe(&p, end, opcode, bad); | |
1705 | ceph_decode_64_safe(&p, end, cookie, bad); | |
1706 | ceph_decode_64_safe(&p, end, ver, bad); | |
1707 | ceph_decode_64_safe(&p, end, notify_id, bad); | |
1708 | ||
1709 | spin_lock(&osdc->event_lock); | |
1710 | event = __find_event(osdc, cookie); | |
1711 | if (event) { | |
3c663bbd | 1712 | BUG_ON(event->one_shot); |
a40c4f10 | 1713 | get_event(event); |
a40c4f10 YS |
1714 | } |
1715 | spin_unlock(&osdc->event_lock); | |
1716 | dout("handle_watch_notify cookie %lld ver %lld event %p\n", | |
1717 | cookie, ver, event); | |
1718 | if (event) { | |
1719 | event_work = kmalloc(sizeof(*event_work), GFP_NOIO); | |
a40c4f10 YS |
1720 | if (!event_work) { |
1721 | dout("ERROR: could not allocate event_work\n"); | |
1722 | goto done_err; | |
1723 | } | |
6b0ae409 | 1724 | INIT_WORK(&event_work->work, do_event_work); |
a40c4f10 YS |
1725 | event_work->event = event; |
1726 | event_work->ver = ver; | |
1727 | event_work->notify_id = notify_id; | |
1728 | event_work->opcode = opcode; | |
1729 | if (!queue_work(osdc->notify_wq, &event_work->work)) { | |
1730 | dout("WARNING: failed to queue notify event work\n"); | |
1731 | goto done_err; | |
1732 | } | |
1733 | } | |
1734 | ||
1735 | return; | |
1736 | ||
1737 | done_err: | |
a40c4f10 YS |
1738 | ceph_osdc_put_event(event); |
1739 | return; | |
1740 | ||
1741 | bad: | |
1742 | pr_err("osdc handle_watch_notify corrupt msg\n"); | |
1743 | return; | |
1744 | } | |
1745 | ||
70636773 AE |
1746 | static void ceph_osdc_msg_data_set(struct ceph_msg *msg, |
1747 | struct ceph_osd_data *osd_data) | |
f24e9980 | 1748 | { |
0fff87ec | 1749 | if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) { |
f1baeb2b | 1750 | BUG_ON(osd_data->length > (u64) SIZE_MAX); |
ebf18f47 | 1751 | if (osd_data->length) |
70636773 AE |
1752 | ceph_msg_data_set_pages(msg, osd_data->pages, |
1753 | osd_data->length, osd_data->alignment); | |
9a5e6d09 AE |
1754 | } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) { |
1755 | BUG_ON(!osd_data->pagelist->length); | |
1756 | ceph_msg_data_set_pagelist(msg, osd_data->pagelist); | |
68b4476b | 1757 | #ifdef CONFIG_BLOCK |
0fff87ec | 1758 | } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) { |
70636773 | 1759 | ceph_msg_data_set_bio(msg, osd_data->bio); |
68b4476b | 1760 | #endif |
2ac2b7a6 | 1761 | } else { |
0fff87ec | 1762 | BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE); |
2ac2b7a6 | 1763 | } |
70636773 AE |
1764 | } |
1765 | ||
1766 | /* | |
1767 | * Register request, send initial attempt. | |
1768 | */ | |
1769 | int ceph_osdc_start_request(struct ceph_osd_client *osdc, | |
1770 | struct ceph_osd_request *req, | |
1771 | bool nofail) | |
1772 | { | |
1773 | int rc = 0; | |
1774 | ||
1775 | /* Set up response incoming data and request outgoing data fields */ | |
1776 | ||
1777 | ceph_osdc_msg_data_set(req->r_reply, &req->r_data_in); | |
1778 | ceph_osdc_msg_data_set(req->r_request, &req->r_data_out); | |
f24e9980 | 1779 | |
f24e9980 SW |
1780 | down_read(&osdc->map_sem); |
1781 | mutex_lock(&osdc->request_mutex); | |
dc4b870c | 1782 | __register_request(osdc, req); |
92451b49 AE |
1783 | WARN_ON(req->r_sent); |
1784 | rc = __map_request(osdc, req, 0); | |
1785 | if (rc < 0) { | |
1786 | if (nofail) { | |
1787 | dout("osdc_start_request failed map, " | |
1788 | " will retry %lld\n", req->r_tid); | |
1789 | rc = 0; | |
f24e9980 | 1790 | } |
92451b49 | 1791 | goto out_unlock; |
f24e9980 | 1792 | } |
92451b49 AE |
1793 | if (req->r_osd == NULL) { |
1794 | dout("send_request %p no up osds in pg\n", req); | |
1795 | ceph_monc_request_next_osdmap(&osdc->client->monc); | |
1796 | } else { | |
1797 | __send_request(osdc, req); | |
1798 | } | |
1799 | rc = 0; | |
234af26f | 1800 | out_unlock: |
f24e9980 SW |
1801 | mutex_unlock(&osdc->request_mutex); |
1802 | up_read(&osdc->map_sem); | |
1803 | return rc; | |
1804 | } | |
3d14c5d2 | 1805 | EXPORT_SYMBOL(ceph_osdc_start_request); |
f24e9980 SW |
1806 | |
1807 | /* | |
1808 | * wait for a request to complete | |
1809 | */ | |
1810 | int ceph_osdc_wait_request(struct ceph_osd_client *osdc, | |
1811 | struct ceph_osd_request *req) | |
1812 | { | |
1813 | int rc; | |
1814 | ||
1815 | rc = wait_for_completion_interruptible(&req->r_completion); | |
1816 | if (rc < 0) { | |
1817 | mutex_lock(&osdc->request_mutex); | |
1818 | __cancel_request(req); | |
529cfcc4 | 1819 | __unregister_request(osdc, req); |
f24e9980 | 1820 | mutex_unlock(&osdc->request_mutex); |
25845472 | 1821 | complete_request(req); |
529cfcc4 | 1822 | dout("wait_request tid %llu canceled/timed out\n", req->r_tid); |
f24e9980 SW |
1823 | return rc; |
1824 | } | |
1825 | ||
1826 | dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result); | |
1827 | return req->r_result; | |
1828 | } | |
3d14c5d2 | 1829 | EXPORT_SYMBOL(ceph_osdc_wait_request); |
f24e9980 SW |
1830 | |
1831 | /* | |
1832 | * sync - wait for all in-flight requests to flush. avoid starvation. | |
1833 | */ | |
1834 | void ceph_osdc_sync(struct ceph_osd_client *osdc) | |
1835 | { | |
1836 | struct ceph_osd_request *req; | |
1837 | u64 last_tid, next_tid = 0; | |
1838 | ||
1839 | mutex_lock(&osdc->request_mutex); | |
1840 | last_tid = osdc->last_tid; | |
1841 | while (1) { | |
1842 | req = __lookup_request_ge(osdc, next_tid); | |
1843 | if (!req) | |
1844 | break; | |
1845 | if (req->r_tid > last_tid) | |
1846 | break; | |
1847 | ||
1848 | next_tid = req->r_tid + 1; | |
1849 | if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0) | |
1850 | continue; | |
1851 | ||
1852 | ceph_osdc_get_request(req); | |
1853 | mutex_unlock(&osdc->request_mutex); | |
1854 | dout("sync waiting on tid %llu (last is %llu)\n", | |
1855 | req->r_tid, last_tid); | |
1856 | wait_for_completion(&req->r_safe_completion); | |
1857 | mutex_lock(&osdc->request_mutex); | |
1858 | ceph_osdc_put_request(req); | |
1859 | } | |
1860 | mutex_unlock(&osdc->request_mutex); | |
1861 | dout("sync done (thru tid %llu)\n", last_tid); | |
1862 | } | |
3d14c5d2 | 1863 | EXPORT_SYMBOL(ceph_osdc_sync); |
f24e9980 SW |
1864 | |
1865 | /* | |
1866 | * init, shutdown | |
1867 | */ | |
1868 | int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client) | |
1869 | { | |
1870 | int err; | |
1871 | ||
1872 | dout("init\n"); | |
1873 | osdc->client = client; | |
1874 | osdc->osdmap = NULL; | |
1875 | init_rwsem(&osdc->map_sem); | |
1876 | init_completion(&osdc->map_waiters); | |
1877 | osdc->last_requested_map = 0; | |
1878 | mutex_init(&osdc->request_mutex); | |
f24e9980 SW |
1879 | osdc->last_tid = 0; |
1880 | osdc->osds = RB_ROOT; | |
f5a2041b | 1881 | INIT_LIST_HEAD(&osdc->osd_lru); |
f24e9980 | 1882 | osdc->requests = RB_ROOT; |
422d2cb8 | 1883 | INIT_LIST_HEAD(&osdc->req_lru); |
6f6c7006 SW |
1884 | INIT_LIST_HEAD(&osdc->req_unsent); |
1885 | INIT_LIST_HEAD(&osdc->req_notarget); | |
a40c4f10 | 1886 | INIT_LIST_HEAD(&osdc->req_linger); |
f24e9980 SW |
1887 | osdc->num_requests = 0; |
1888 | INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout); | |
f5a2041b | 1889 | INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout); |
a40c4f10 YS |
1890 | spin_lock_init(&osdc->event_lock); |
1891 | osdc->event_tree = RB_ROOT; | |
1892 | osdc->event_count = 0; | |
f5a2041b YS |
1893 | |
1894 | schedule_delayed_work(&osdc->osds_timeout_work, | |
3d14c5d2 | 1895 | round_jiffies_relative(osdc->client->options->osd_idle_ttl * HZ)); |
f24e9980 | 1896 | |
5f44f142 | 1897 | err = -ENOMEM; |
f24e9980 SW |
1898 | osdc->req_mempool = mempool_create_kmalloc_pool(10, |
1899 | sizeof(struct ceph_osd_request)); | |
1900 | if (!osdc->req_mempool) | |
5f44f142 | 1901 | goto out; |
f24e9980 | 1902 | |
d50b409f SW |
1903 | err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP, |
1904 | OSD_OP_FRONT_LEN, 10, true, | |
4f48280e | 1905 | "osd_op"); |
f24e9980 | 1906 | if (err < 0) |
5f44f142 | 1907 | goto out_mempool; |
d50b409f | 1908 | err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY, |
4f48280e SW |
1909 | OSD_OPREPLY_FRONT_LEN, 10, true, |
1910 | "osd_op_reply"); | |
c16e7869 SW |
1911 | if (err < 0) |
1912 | goto out_msgpool; | |
a40c4f10 YS |
1913 | |
1914 | osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify"); | |
1915 | if (IS_ERR(osdc->notify_wq)) { | |
1916 | err = PTR_ERR(osdc->notify_wq); | |
1917 | osdc->notify_wq = NULL; | |
1918 | goto out_msgpool; | |
1919 | } | |
f24e9980 | 1920 | return 0; |
5f44f142 | 1921 | |
c16e7869 SW |
1922 | out_msgpool: |
1923 | ceph_msgpool_destroy(&osdc->msgpool_op); | |
5f44f142 SW |
1924 | out_mempool: |
1925 | mempool_destroy(osdc->req_mempool); | |
1926 | out: | |
1927 | return err; | |
f24e9980 SW |
1928 | } |
1929 | ||
1930 | void ceph_osdc_stop(struct ceph_osd_client *osdc) | |
1931 | { | |
a40c4f10 YS |
1932 | flush_workqueue(osdc->notify_wq); |
1933 | destroy_workqueue(osdc->notify_wq); | |
f24e9980 | 1934 | cancel_delayed_work_sync(&osdc->timeout_work); |
f5a2041b | 1935 | cancel_delayed_work_sync(&osdc->osds_timeout_work); |
f24e9980 SW |
1936 | if (osdc->osdmap) { |
1937 | ceph_osdmap_destroy(osdc->osdmap); | |
1938 | osdc->osdmap = NULL; | |
1939 | } | |
aca420bc | 1940 | remove_all_osds(osdc); |
f24e9980 SW |
1941 | mempool_destroy(osdc->req_mempool); |
1942 | ceph_msgpool_destroy(&osdc->msgpool_op); | |
c16e7869 | 1943 | ceph_msgpool_destroy(&osdc->msgpool_op_reply); |
f24e9980 SW |
1944 | } |
1945 | ||
1946 | /* | |
1947 | * Read some contiguous pages. If we cross a stripe boundary, shorten | |
1948 | * *plen. Return number of bytes read, or error. | |
1949 | */ | |
1950 | int ceph_osdc_readpages(struct ceph_osd_client *osdc, | |
1951 | struct ceph_vino vino, struct ceph_file_layout *layout, | |
1952 | u64 off, u64 *plen, | |
1953 | u32 truncate_seq, u64 truncate_size, | |
b7495fc2 | 1954 | struct page **pages, int num_pages, int page_align) |
f24e9980 SW |
1955 | { |
1956 | struct ceph_osd_request *req; | |
0fff87ec | 1957 | struct ceph_osd_data *osd_data; |
f24e9980 SW |
1958 | int rc = 0; |
1959 | ||
1960 | dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino, | |
1961 | vino.snap, off, *plen); | |
1962 | req = ceph_osdc_new_request(osdc, layout, vino, off, plen, | |
1963 | CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ, | |
1964 | NULL, 0, truncate_seq, truncate_size, NULL, | |
153e5167 | 1965 | false); |
6816282d SW |
1966 | if (IS_ERR(req)) |
1967 | return PTR_ERR(req); | |
f24e9980 SW |
1968 | |
1969 | /* it may be a short read due to an object boundary */ | |
0fff87ec AE |
1970 | |
1971 | osd_data = &req->r_data_in; | |
1972 | osd_data->type = CEPH_OSD_DATA_TYPE_PAGES; | |
1973 | osd_data->pages = pages; | |
e0c59487 | 1974 | osd_data->length = *plen; |
0fff87ec | 1975 | osd_data->alignment = page_align; |
f24e9980 | 1976 | |
e0c59487 AE |
1977 | dout("readpages final extent is %llu~%llu (%llu bytes align %d)\n", |
1978 | off, *plen, osd_data->length, page_align); | |
f24e9980 SW |
1979 | |
1980 | rc = ceph_osdc_start_request(osdc, req, false); | |
1981 | if (!rc) | |
1982 | rc = ceph_osdc_wait_request(osdc, req); | |
1983 | ||
1984 | ceph_osdc_put_request(req); | |
1985 | dout("readpages result %d\n", rc); | |
1986 | return rc; | |
1987 | } | |
3d14c5d2 | 1988 | EXPORT_SYMBOL(ceph_osdc_readpages); |
f24e9980 SW |
1989 | |
1990 | /* | |
1991 | * do a synchronous write on N pages | |
1992 | */ | |
1993 | int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino, | |
1994 | struct ceph_file_layout *layout, | |
1995 | struct ceph_snap_context *snapc, | |
1996 | u64 off, u64 len, | |
1997 | u32 truncate_seq, u64 truncate_size, | |
1998 | struct timespec *mtime, | |
24808826 | 1999 | struct page **pages, int num_pages) |
f24e9980 SW |
2000 | { |
2001 | struct ceph_osd_request *req; | |
0fff87ec | 2002 | struct ceph_osd_data *osd_data; |
f24e9980 | 2003 | int rc = 0; |
b7495fc2 | 2004 | int page_align = off & ~PAGE_MASK; |
f24e9980 SW |
2005 | |
2006 | BUG_ON(vino.snap != CEPH_NOSNAP); | |
2007 | req = ceph_osdc_new_request(osdc, layout, vino, off, &len, | |
2008 | CEPH_OSD_OP_WRITE, | |
24808826 | 2009 | CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE, |
fbf8685f | 2010 | snapc, 0, |
f24e9980 | 2011 | truncate_seq, truncate_size, mtime, |
153e5167 | 2012 | true); |
6816282d SW |
2013 | if (IS_ERR(req)) |
2014 | return PTR_ERR(req); | |
f24e9980 SW |
2015 | |
2016 | /* it may be a short write due to an object boundary */ | |
0fff87ec AE |
2017 | osd_data = &req->r_data_out; |
2018 | osd_data->type = CEPH_OSD_DATA_TYPE_PAGES; | |
2019 | osd_data->pages = pages; | |
e0c59487 | 2020 | osd_data->length = len; |
0fff87ec | 2021 | osd_data->alignment = page_align; |
e0c59487 | 2022 | dout("writepages %llu~%llu (%llu bytes)\n", off, len, osd_data->length); |
f24e9980 | 2023 | |
87f979d3 | 2024 | rc = ceph_osdc_start_request(osdc, req, true); |
f24e9980 SW |
2025 | if (!rc) |
2026 | rc = ceph_osdc_wait_request(osdc, req); | |
2027 | ||
2028 | ceph_osdc_put_request(req); | |
2029 | if (rc == 0) | |
2030 | rc = len; | |
2031 | dout("writepages result %d\n", rc); | |
2032 | return rc; | |
2033 | } | |
3d14c5d2 | 2034 | EXPORT_SYMBOL(ceph_osdc_writepages); |
f24e9980 SW |
2035 | |
2036 | /* | |
2037 | * handle incoming message | |
2038 | */ | |
2039 | static void dispatch(struct ceph_connection *con, struct ceph_msg *msg) | |
2040 | { | |
2041 | struct ceph_osd *osd = con->private; | |
32c895e7 | 2042 | struct ceph_osd_client *osdc; |
f24e9980 SW |
2043 | int type = le16_to_cpu(msg->hdr.type); |
2044 | ||
2045 | if (!osd) | |
4a32f93d | 2046 | goto out; |
32c895e7 | 2047 | osdc = osd->o_osdc; |
f24e9980 SW |
2048 | |
2049 | switch (type) { | |
2050 | case CEPH_MSG_OSD_MAP: | |
2051 | ceph_osdc_handle_map(osdc, msg); | |
2052 | break; | |
2053 | case CEPH_MSG_OSD_OPREPLY: | |
350b1c32 | 2054 | handle_reply(osdc, msg, con); |
f24e9980 | 2055 | break; |
a40c4f10 YS |
2056 | case CEPH_MSG_WATCH_NOTIFY: |
2057 | handle_watch_notify(osdc, msg); | |
2058 | break; | |
f24e9980 SW |
2059 | |
2060 | default: | |
2061 | pr_err("received unknown message type %d %s\n", type, | |
2062 | ceph_msg_type_name(type)); | |
2063 | } | |
4a32f93d | 2064 | out: |
f24e9980 SW |
2065 | ceph_msg_put(msg); |
2066 | } | |
2067 | ||
5b3a4db3 | 2068 | /* |
21b667f6 SW |
2069 | * lookup and return message for incoming reply. set up reply message |
2070 | * pages. | |
5b3a4db3 SW |
2071 | */ |
2072 | static struct ceph_msg *get_reply(struct ceph_connection *con, | |
2450418c YS |
2073 | struct ceph_msg_header *hdr, |
2074 | int *skip) | |
f24e9980 SW |
2075 | { |
2076 | struct ceph_osd *osd = con->private; | |
2077 | struct ceph_osd_client *osdc = osd->o_osdc; | |
2450418c | 2078 | struct ceph_msg *m; |
0547a9b3 | 2079 | struct ceph_osd_request *req; |
5b3a4db3 SW |
2080 | int front = le32_to_cpu(hdr->front_len); |
2081 | int data_len = le32_to_cpu(hdr->data_len); | |
0547a9b3 | 2082 | u64 tid; |
f24e9980 | 2083 | |
0547a9b3 YS |
2084 | tid = le64_to_cpu(hdr->tid); |
2085 | mutex_lock(&osdc->request_mutex); | |
2086 | req = __lookup_request(osdc, tid); | |
2087 | if (!req) { | |
2088 | *skip = 1; | |
2089 | m = NULL; | |
756a16a5 SW |
2090 | dout("get_reply unknown tid %llu from osd%d\n", tid, |
2091 | osd->o_osd); | |
0547a9b3 YS |
2092 | goto out; |
2093 | } | |
c16e7869 SW |
2094 | |
2095 | if (req->r_con_filling_msg) { | |
8921d114 | 2096 | dout("%s revoking msg %p from old con %p\n", __func__, |
c16e7869 | 2097 | req->r_reply, req->r_con_filling_msg); |
8921d114 | 2098 | ceph_msg_revoke_incoming(req->r_reply); |
0d47766f | 2099 | req->r_con_filling_msg->ops->put(req->r_con_filling_msg); |
6f46cb29 | 2100 | req->r_con_filling_msg = NULL; |
0547a9b3 YS |
2101 | } |
2102 | ||
c16e7869 SW |
2103 | if (front > req->r_reply->front.iov_len) { |
2104 | pr_warning("get_reply front %d > preallocated %d\n", | |
2105 | front, (int)req->r_reply->front.iov_len); | |
b61c2763 | 2106 | m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front, GFP_NOFS, false); |
a79832f2 | 2107 | if (!m) |
c16e7869 SW |
2108 | goto out; |
2109 | ceph_msg_put(req->r_reply); | |
2110 | req->r_reply = m; | |
2111 | } | |
2112 | m = ceph_msg_get(req->r_reply); | |
2113 | ||
0547a9b3 | 2114 | if (data_len > 0) { |
0fff87ec AE |
2115 | struct ceph_osd_data *osd_data = &req->r_data_in; |
2116 | ||
2117 | if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) { | |
0fff87ec | 2118 | if (osd_data->pages && |
e0c59487 | 2119 | unlikely(osd_data->length < data_len)) { |
2ac2b7a6 | 2120 | |
e0c59487 AE |
2121 | pr_warning("tid %lld reply has %d bytes " |
2122 | "we had only %llu bytes ready\n", | |
2123 | tid, data_len, osd_data->length); | |
2ac2b7a6 AE |
2124 | *skip = 1; |
2125 | ceph_msg_put(m); | |
2126 | m = NULL; | |
2127 | goto out; | |
2128 | } | |
2ac2b7a6 | 2129 | } |
0547a9b3 | 2130 | } |
5b3a4db3 | 2131 | *skip = 0; |
0d47766f | 2132 | req->r_con_filling_msg = con->ops->get(con); |
c16e7869 | 2133 | dout("get_reply tid %lld %p\n", tid, m); |
0547a9b3 YS |
2134 | |
2135 | out: | |
2136 | mutex_unlock(&osdc->request_mutex); | |
2450418c | 2137 | return m; |
5b3a4db3 SW |
2138 | |
2139 | } | |
2140 | ||
2141 | static struct ceph_msg *alloc_msg(struct ceph_connection *con, | |
2142 | struct ceph_msg_header *hdr, | |
2143 | int *skip) | |
2144 | { | |
2145 | struct ceph_osd *osd = con->private; | |
2146 | int type = le16_to_cpu(hdr->type); | |
2147 | int front = le32_to_cpu(hdr->front_len); | |
2148 | ||
1c20f2d2 | 2149 | *skip = 0; |
5b3a4db3 SW |
2150 | switch (type) { |
2151 | case CEPH_MSG_OSD_MAP: | |
a40c4f10 | 2152 | case CEPH_MSG_WATCH_NOTIFY: |
b61c2763 | 2153 | return ceph_msg_new(type, front, GFP_NOFS, false); |
5b3a4db3 SW |
2154 | case CEPH_MSG_OSD_OPREPLY: |
2155 | return get_reply(con, hdr, skip); | |
2156 | default: | |
2157 | pr_info("alloc_msg unexpected msg type %d from osd%d\n", type, | |
2158 | osd->o_osd); | |
2159 | *skip = 1; | |
2160 | return NULL; | |
2161 | } | |
f24e9980 SW |
2162 | } |
2163 | ||
2164 | /* | |
2165 | * Wrappers to refcount containing ceph_osd struct | |
2166 | */ | |
2167 | static struct ceph_connection *get_osd_con(struct ceph_connection *con) | |
2168 | { | |
2169 | struct ceph_osd *osd = con->private; | |
2170 | if (get_osd(osd)) | |
2171 | return con; | |
2172 | return NULL; | |
2173 | } | |
2174 | ||
2175 | static void put_osd_con(struct ceph_connection *con) | |
2176 | { | |
2177 | struct ceph_osd *osd = con->private; | |
2178 | put_osd(osd); | |
2179 | } | |
2180 | ||
4e7a5dcd SW |
2181 | /* |
2182 | * authentication | |
2183 | */ | |
a3530df3 AE |
2184 | /* |
2185 | * Note: returned pointer is the address of a structure that's | |
2186 | * managed separately. Caller must *not* attempt to free it. | |
2187 | */ | |
2188 | static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con, | |
8f43fb53 | 2189 | int *proto, int force_new) |
4e7a5dcd SW |
2190 | { |
2191 | struct ceph_osd *o = con->private; | |
2192 | struct ceph_osd_client *osdc = o->o_osdc; | |
2193 | struct ceph_auth_client *ac = osdc->client->monc.auth; | |
74f1869f | 2194 | struct ceph_auth_handshake *auth = &o->o_auth; |
4e7a5dcd | 2195 | |
74f1869f | 2196 | if (force_new && auth->authorizer) { |
27859f97 | 2197 | ceph_auth_destroy_authorizer(ac, auth->authorizer); |
74f1869f AE |
2198 | auth->authorizer = NULL; |
2199 | } | |
27859f97 SW |
2200 | if (!auth->authorizer) { |
2201 | int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD, | |
2202 | auth); | |
4e7a5dcd | 2203 | if (ret) |
a3530df3 | 2204 | return ERR_PTR(ret); |
27859f97 SW |
2205 | } else { |
2206 | int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD, | |
0bed9b5c SW |
2207 | auth); |
2208 | if (ret) | |
2209 | return ERR_PTR(ret); | |
4e7a5dcd | 2210 | } |
4e7a5dcd | 2211 | *proto = ac->protocol; |
74f1869f | 2212 | |
a3530df3 | 2213 | return auth; |
4e7a5dcd SW |
2214 | } |
2215 | ||
2216 | ||
2217 | static int verify_authorizer_reply(struct ceph_connection *con, int len) | |
2218 | { | |
2219 | struct ceph_osd *o = con->private; | |
2220 | struct ceph_osd_client *osdc = o->o_osdc; | |
2221 | struct ceph_auth_client *ac = osdc->client->monc.auth; | |
2222 | ||
27859f97 | 2223 | return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer, len); |
4e7a5dcd SW |
2224 | } |
2225 | ||
9bd2e6f8 SW |
2226 | static int invalidate_authorizer(struct ceph_connection *con) |
2227 | { | |
2228 | struct ceph_osd *o = con->private; | |
2229 | struct ceph_osd_client *osdc = o->o_osdc; | |
2230 | struct ceph_auth_client *ac = osdc->client->monc.auth; | |
2231 | ||
27859f97 | 2232 | ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD); |
9bd2e6f8 SW |
2233 | return ceph_monc_validate_auth(&osdc->client->monc); |
2234 | } | |
4e7a5dcd | 2235 | |
9e32789f | 2236 | static const struct ceph_connection_operations osd_con_ops = { |
f24e9980 SW |
2237 | .get = get_osd_con, |
2238 | .put = put_osd_con, | |
2239 | .dispatch = dispatch, | |
4e7a5dcd SW |
2240 | .get_authorizer = get_authorizer, |
2241 | .verify_authorizer_reply = verify_authorizer_reply, | |
9bd2e6f8 | 2242 | .invalidate_authorizer = invalidate_authorizer, |
f24e9980 | 2243 | .alloc_msg = alloc_msg, |
81b024e7 | 2244 | .fault = osd_reset, |
f24e9980 | 2245 | }; |