Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1 | #include "ceph_debug.h" |
| 2 | |
| 3 | #include <linux/err.h> |
| 4 | #include <linux/highmem.h> |
| 5 | #include <linux/mm.h> |
| 6 | #include <linux/pagemap.h> |
| 7 | #include <linux/slab.h> |
| 8 | #include <linux/uaccess.h> |
| 9 | |
| 10 | #include "super.h" |
| 11 | #include "osd_client.h" |
| 12 | #include "messenger.h" |
| 13 | #include "decode.h" |
Sage Weil | 4e7a5dc | 2009-11-18 16:19:57 -0800 | [diff] [blame] | 14 | #include "auth.h" |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 15 | |
Sage Weil | c16e786 | 2010-03-01 13:02:00 -0800 | [diff] [blame] | 16 | #define OSD_OP_FRONT_LEN 4096 |
| 17 | #define OSD_OPREPLY_FRONT_LEN 512 |
Yehuda Sadeh | 0d59ab8 | 2010-01-13 17:03:23 -0800 | [diff] [blame] | 18 | |
Tobias Klauser | 9e32789 | 2010-05-20 10:40:19 +0200 | [diff] [blame] | 19 | static const struct ceph_connection_operations osd_con_ops; |
Yehuda Sadeh | 422d2cb | 2010-02-26 15:32:31 -0800 | [diff] [blame] | 20 | static int __kick_requests(struct ceph_osd_client *osdc, |
| 21 | struct ceph_osd *kickosd); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 22 | |
| 23 | static void kick_requests(struct ceph_osd_client *osdc, struct ceph_osd *osd); |
| 24 | |
Yehuda Sadeh | 3499e8a | 2010-04-06 14:51:47 -0700 | [diff] [blame^] | 25 | void ceph_calc_raw_layout(struct ceph_osd_client *osdc, |
| 26 | struct ceph_file_layout *layout, |
| 27 | u64 snapid, |
| 28 | u64 off, u64 len, u64 *bno, |
| 29 | struct ceph_osd_request *req) |
| 30 | { |
| 31 | struct ceph_osd_request_head *reqhead = req->r_request->front.iov_base; |
| 32 | struct ceph_osd_op *op = (void *)(reqhead + 1); |
| 33 | u64 orig_len = len; |
| 34 | u64 objoff, objlen; /* extent in object */ |
| 35 | |
| 36 | reqhead->snapid = cpu_to_le64(snapid); |
| 37 | |
| 38 | /* object extent? */ |
| 39 | ceph_calc_file_object_mapping(layout, off, &len, bno, |
| 40 | &objoff, &objlen); |
| 41 | if (len < orig_len) |
| 42 | dout(" skipping last %llu, final file extent %llu~%llu\n", |
| 43 | orig_len - len, off, len); |
| 44 | |
| 45 | op->extent.offset = cpu_to_le64(objoff); |
| 46 | op->extent.length = cpu_to_le64(objlen); |
| 47 | req->r_num_pages = calc_pages_for(off, len); |
| 48 | |
| 49 | dout("calc_layout bno=%llx %llu~%llu (%d pages)\n", |
| 50 | *bno, objoff, objlen, req->r_num_pages); |
| 51 | |
| 52 | } |
| 53 | |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 54 | /* |
| 55 | * Implement client access to distributed object storage cluster. |
| 56 | * |
| 57 | * All data objects are stored within a cluster/cloud of OSDs, or |
| 58 | * "object storage devices." (Note that Ceph OSDs have _nothing_ to |
| 59 | * do with the T10 OSD extensions to SCSI.) Ceph OSDs are simply |
| 60 | * remote daemons serving up and coordinating consistent and safe |
| 61 | * access to storage. |
| 62 | * |
| 63 | * Cluster membership and the mapping of data objects onto storage devices |
| 64 | * are described by the osd map. |
| 65 | * |
| 66 | * We keep track of pending OSD requests (read, write), resubmit |
| 67 | * requests to different OSDs when the cluster topology/data layout |
| 68 | * change, or retry the affected requests when the communications |
| 69 | * channel with an OSD is reset. |
| 70 | */ |
| 71 | |
| 72 | /* |
| 73 | * calculate the mapping of a file extent onto an object, and fill out the |
| 74 | * request accordingly. shorten extent as necessary if it crosses an |
| 75 | * object boundary. |
| 76 | * |
| 77 | * fill osd op in request message. |
| 78 | */ |
| 79 | static void calc_layout(struct ceph_osd_client *osdc, |
Yehuda Sadeh | 3499e8a | 2010-04-06 14:51:47 -0700 | [diff] [blame^] | 80 | struct ceph_vino vino, |
| 81 | struct ceph_file_layout *layout, |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 82 | u64 off, u64 *plen, |
| 83 | struct ceph_osd_request *req) |
| 84 | { |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 85 | u64 bno; |
| 86 | |
Yehuda Sadeh | 3499e8a | 2010-04-06 14:51:47 -0700 | [diff] [blame^] | 87 | ceph_calc_raw_layout(osdc, layout, vino.snap, off, *plen, &bno, req); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 88 | |
| 89 | sprintf(req->r_oid, "%llx.%08llx", vino.ino, bno); |
| 90 | req->r_oid_len = strlen(req->r_oid); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 91 | } |
| 92 | |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 93 | /* |
| 94 | * requests |
| 95 | */ |
Sage Weil | 415e49a | 2009-12-07 13:37:03 -0800 | [diff] [blame] | 96 | void ceph_osdc_release_request(struct kref *kref) |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 97 | { |
Sage Weil | 415e49a | 2009-12-07 13:37:03 -0800 | [diff] [blame] | 98 | struct ceph_osd_request *req = container_of(kref, |
| 99 | struct ceph_osd_request, |
| 100 | r_kref); |
| 101 | |
| 102 | if (req->r_request) |
| 103 | ceph_msg_put(req->r_request); |
| 104 | if (req->r_reply) |
| 105 | ceph_msg_put(req->r_reply); |
Yehuda Sadeh | 0d59ab8 | 2010-01-13 17:03:23 -0800 | [diff] [blame] | 106 | if (req->r_con_filling_msg) { |
Sage Weil | 350b1c3 | 2009-12-22 10:45:45 -0800 | [diff] [blame] | 107 | dout("release_request revoking pages %p from con %p\n", |
Yehuda Sadeh | 0d59ab8 | 2010-01-13 17:03:23 -0800 | [diff] [blame] | 108 | req->r_pages, req->r_con_filling_msg); |
| 109 | ceph_con_revoke_message(req->r_con_filling_msg, |
| 110 | req->r_reply); |
| 111 | ceph_con_put(req->r_con_filling_msg); |
Sage Weil | 350b1c3 | 2009-12-22 10:45:45 -0800 | [diff] [blame] | 112 | } |
Sage Weil | 415e49a | 2009-12-07 13:37:03 -0800 | [diff] [blame] | 113 | if (req->r_own_pages) |
| 114 | ceph_release_page_vector(req->r_pages, |
| 115 | req->r_num_pages); |
| 116 | ceph_put_snap_context(req->r_snapc); |
| 117 | if (req->r_mempool) |
| 118 | mempool_free(req, req->r_osdc->req_mempool); |
| 119 | else |
| 120 | kfree(req); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 121 | } |
| 122 | |
Yehuda Sadeh | 3499e8a | 2010-04-06 14:51:47 -0700 | [diff] [blame^] | 123 | struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc, |
| 124 | int flags, |
| 125 | struct ceph_snap_context *snapc, |
| 126 | int do_sync, |
| 127 | bool use_mempool, |
| 128 | gfp_t gfp_flags, |
| 129 | struct page **pages) |
| 130 | { |
| 131 | struct ceph_osd_request *req; |
| 132 | struct ceph_msg *msg; |
| 133 | int num_op = 1 + do_sync; |
| 134 | size_t msg_size = sizeof(struct ceph_osd_request_head) + |
| 135 | num_op*sizeof(struct ceph_osd_op); |
| 136 | |
| 137 | if (use_mempool) { |
| 138 | req = mempool_alloc(osdc->req_mempool, gfp_flags); |
| 139 | memset(req, 0, sizeof(*req)); |
| 140 | } else { |
| 141 | req = kzalloc(sizeof(*req), gfp_flags); |
| 142 | } |
| 143 | if (!req) |
| 144 | return NULL; |
| 145 | |
| 146 | if (use_mempool) { |
| 147 | req = mempool_alloc(osdc->req_mempool, gfp_flags); |
| 148 | memset(req, 0, sizeof(*req)); |
| 149 | } else { |
| 150 | req = kzalloc(sizeof(*req), gfp_flags); |
| 151 | } |
| 152 | if (req == NULL) |
| 153 | return NULL; |
| 154 | |
| 155 | req->r_osdc = osdc; |
| 156 | req->r_mempool = use_mempool; |
| 157 | kref_init(&req->r_kref); |
| 158 | init_completion(&req->r_completion); |
| 159 | init_completion(&req->r_safe_completion); |
| 160 | INIT_LIST_HEAD(&req->r_unsafe_item); |
| 161 | req->r_flags = flags; |
| 162 | |
| 163 | WARN_ON((flags & (CEPH_OSD_FLAG_READ|CEPH_OSD_FLAG_WRITE)) == 0); |
| 164 | |
| 165 | /* create reply message */ |
| 166 | if (use_mempool) |
| 167 | msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0); |
| 168 | else |
| 169 | msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, |
| 170 | OSD_OPREPLY_FRONT_LEN, gfp_flags); |
| 171 | if (!msg) { |
| 172 | ceph_osdc_put_request(req); |
| 173 | return NULL; |
| 174 | } |
| 175 | req->r_reply = msg; |
| 176 | |
| 177 | /* create request message; allow space for oid */ |
| 178 | msg_size += 40; |
| 179 | if (snapc) |
| 180 | msg_size += sizeof(u64) * snapc->num_snaps; |
| 181 | if (use_mempool) |
| 182 | msg = ceph_msgpool_get(&osdc->msgpool_op, 0); |
| 183 | else |
| 184 | msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp_flags); |
| 185 | if (!msg) { |
| 186 | ceph_osdc_put_request(req); |
| 187 | return NULL; |
| 188 | } |
| 189 | msg->hdr.type = cpu_to_le16(CEPH_MSG_OSD_OP); |
| 190 | memset(msg->front.iov_base, 0, msg->front.iov_len); |
| 191 | |
| 192 | req->r_request = msg; |
| 193 | req->r_pages = pages; |
| 194 | |
| 195 | return req; |
| 196 | } |
| 197 | |
| 198 | /* |
| 199 | * build new request AND message |
| 200 | * |
| 201 | */ |
| 202 | void ceph_osdc_build_request(struct ceph_osd_request *req, |
| 203 | u64 off, u64 *plen, |
| 204 | int opcode, |
| 205 | struct ceph_snap_context *snapc, |
| 206 | int do_sync, |
| 207 | u32 truncate_seq, |
| 208 | u64 truncate_size, |
| 209 | struct timespec *mtime, |
| 210 | const char *oid, |
| 211 | int oid_len) |
| 212 | { |
| 213 | struct ceph_msg *msg = req->r_request; |
| 214 | struct ceph_osd_request_head *head; |
| 215 | struct ceph_osd_op *op; |
| 216 | void *p; |
| 217 | int num_op = 1 + do_sync; |
| 218 | size_t msg_size = sizeof(*head) + num_op*sizeof(*op); |
| 219 | int i; |
| 220 | int flags = req->r_flags; |
| 221 | |
| 222 | head = msg->front.iov_base; |
| 223 | op = (void *)(head + 1); |
| 224 | p = (void *)(op + num_op); |
| 225 | |
| 226 | req->r_snapc = ceph_get_snap_context(snapc); |
| 227 | |
| 228 | head->client_inc = cpu_to_le32(1); /* always, for now. */ |
| 229 | head->flags = cpu_to_le32(flags); |
| 230 | if (flags & CEPH_OSD_FLAG_WRITE) |
| 231 | ceph_encode_timespec(&head->mtime, mtime); |
| 232 | head->num_ops = cpu_to_le16(num_op); |
| 233 | op->op = cpu_to_le16(opcode); |
| 234 | |
| 235 | if (flags & CEPH_OSD_FLAG_WRITE) { |
| 236 | req->r_request->hdr.data_off = cpu_to_le16(off); |
| 237 | req->r_request->hdr.data_len = cpu_to_le32(*plen); |
| 238 | op->payload_len = cpu_to_le32(*plen); |
| 239 | } |
| 240 | op->extent.truncate_size = cpu_to_le64(truncate_size); |
| 241 | op->extent.truncate_seq = cpu_to_le32(truncate_seq); |
| 242 | |
| 243 | /* fill in oid */ |
| 244 | head->object_len = cpu_to_le32(oid_len); |
| 245 | memcpy(p, oid, oid_len); |
| 246 | p += oid_len; |
| 247 | |
| 248 | if (do_sync) { |
| 249 | op++; |
| 250 | op->op = cpu_to_le16(CEPH_OSD_OP_STARTSYNC); |
| 251 | } |
| 252 | if (snapc) { |
| 253 | head->snap_seq = cpu_to_le64(snapc->seq); |
| 254 | head->num_snaps = cpu_to_le32(snapc->num_snaps); |
| 255 | for (i = 0; i < snapc->num_snaps; i++) { |
| 256 | put_unaligned_le64(snapc->snaps[i], p); |
| 257 | p += sizeof(u64); |
| 258 | } |
| 259 | } |
| 260 | |
| 261 | BUG_ON(p > msg->front.iov_base + msg->front.iov_len); |
| 262 | msg_size = p - msg->front.iov_base; |
| 263 | msg->front.iov_len = msg_size; |
| 264 | msg->hdr.front_len = cpu_to_le32(msg_size); |
| 265 | return; |
| 266 | } |
| 267 | |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 268 | /* |
| 269 | * build new request AND message, calculate layout, and adjust file |
| 270 | * extent as needed. |
| 271 | * |
| 272 | * if the file was recently truncated, we include information about its |
| 273 | * old and new size so that the object can be updated appropriately. (we |
| 274 | * avoid synchronously deleting truncated objects because it's slow.) |
| 275 | * |
| 276 | * if @do_sync, include a 'startsync' command so that the osd will flush |
| 277 | * data quickly. |
| 278 | */ |
| 279 | struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc, |
| 280 | struct ceph_file_layout *layout, |
| 281 | struct ceph_vino vino, |
| 282 | u64 off, u64 *plen, |
| 283 | int opcode, int flags, |
| 284 | struct ceph_snap_context *snapc, |
| 285 | int do_sync, |
| 286 | u32 truncate_seq, |
| 287 | u64 truncate_size, |
| 288 | struct timespec *mtime, |
| 289 | bool use_mempool, int num_reply) |
| 290 | { |
Yehuda Sadeh | 3499e8a | 2010-04-06 14:51:47 -0700 | [diff] [blame^] | 291 | struct ceph_osd_request *req = |
| 292 | ceph_osdc_alloc_request(osdc, flags, |
| 293 | snapc, do_sync, |
| 294 | use_mempool, |
| 295 | GFP_NOFS, NULL); |
| 296 | if (IS_ERR(req)) |
| 297 | return req; |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 298 | |
| 299 | /* calculate max write size */ |
| 300 | calc_layout(osdc, vino, layout, off, plen, req); |
| 301 | req->r_file_layout = *layout; /* keep a copy */ |
| 302 | |
Yehuda Sadeh | 3499e8a | 2010-04-06 14:51:47 -0700 | [diff] [blame^] | 303 | ceph_osdc_build_request(req, off, plen, opcode, |
| 304 | snapc, do_sync, |
| 305 | truncate_seq, truncate_size, |
| 306 | mtime, |
| 307 | req->r_oid, req->r_oid_len); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 308 | |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 309 | return req; |
| 310 | } |
| 311 | |
| 312 | /* |
| 313 | * We keep osd requests in an rbtree, sorted by ->r_tid. |
| 314 | */ |
| 315 | static void __insert_request(struct ceph_osd_client *osdc, |
| 316 | struct ceph_osd_request *new) |
| 317 | { |
| 318 | struct rb_node **p = &osdc->requests.rb_node; |
| 319 | struct rb_node *parent = NULL; |
| 320 | struct ceph_osd_request *req = NULL; |
| 321 | |
| 322 | while (*p) { |
| 323 | parent = *p; |
| 324 | req = rb_entry(parent, struct ceph_osd_request, r_node); |
| 325 | if (new->r_tid < req->r_tid) |
| 326 | p = &(*p)->rb_left; |
| 327 | else if (new->r_tid > req->r_tid) |
| 328 | p = &(*p)->rb_right; |
| 329 | else |
| 330 | BUG(); |
| 331 | } |
| 332 | |
| 333 | rb_link_node(&new->r_node, parent, p); |
| 334 | rb_insert_color(&new->r_node, &osdc->requests); |
| 335 | } |
| 336 | |
| 337 | static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc, |
| 338 | u64 tid) |
| 339 | { |
| 340 | struct ceph_osd_request *req; |
| 341 | struct rb_node *n = osdc->requests.rb_node; |
| 342 | |
| 343 | while (n) { |
| 344 | req = rb_entry(n, struct ceph_osd_request, r_node); |
| 345 | if (tid < req->r_tid) |
| 346 | n = n->rb_left; |
| 347 | else if (tid > req->r_tid) |
| 348 | n = n->rb_right; |
| 349 | else |
| 350 | return req; |
| 351 | } |
| 352 | return NULL; |
| 353 | } |
| 354 | |
| 355 | static struct ceph_osd_request * |
| 356 | __lookup_request_ge(struct ceph_osd_client *osdc, |
| 357 | u64 tid) |
| 358 | { |
| 359 | struct ceph_osd_request *req; |
| 360 | struct rb_node *n = osdc->requests.rb_node; |
| 361 | |
| 362 | while (n) { |
| 363 | req = rb_entry(n, struct ceph_osd_request, r_node); |
| 364 | if (tid < req->r_tid) { |
| 365 | if (!n->rb_left) |
| 366 | return req; |
| 367 | n = n->rb_left; |
| 368 | } else if (tid > req->r_tid) { |
| 369 | n = n->rb_right; |
| 370 | } else { |
| 371 | return req; |
| 372 | } |
| 373 | } |
| 374 | return NULL; |
| 375 | } |
| 376 | |
| 377 | |
| 378 | /* |
Sage Weil | 81b024e | 2009-10-09 10:29:18 -0700 | [diff] [blame] | 379 | * If the osd connection drops, we need to resubmit all requests. |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 380 | */ |
| 381 | static void osd_reset(struct ceph_connection *con) |
| 382 | { |
| 383 | struct ceph_osd *osd = con->private; |
| 384 | struct ceph_osd_client *osdc; |
| 385 | |
| 386 | if (!osd) |
| 387 | return; |
| 388 | dout("osd_reset osd%d\n", osd->o_osd); |
| 389 | osdc = osd->o_osdc; |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 390 | down_read(&osdc->map_sem); |
| 391 | kick_requests(osdc, osd); |
| 392 | up_read(&osdc->map_sem); |
| 393 | } |
| 394 | |
| 395 | /* |
| 396 | * Track open sessions with osds. |
| 397 | */ |
| 398 | static struct ceph_osd *create_osd(struct ceph_osd_client *osdc) |
| 399 | { |
| 400 | struct ceph_osd *osd; |
| 401 | |
| 402 | osd = kzalloc(sizeof(*osd), GFP_NOFS); |
| 403 | if (!osd) |
| 404 | return NULL; |
| 405 | |
| 406 | atomic_set(&osd->o_ref, 1); |
| 407 | osd->o_osdc = osdc; |
| 408 | INIT_LIST_HEAD(&osd->o_requests); |
Yehuda Sadeh | f5a2041 | 2010-02-03 11:00:26 -0800 | [diff] [blame] | 409 | INIT_LIST_HEAD(&osd->o_osd_lru); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 410 | osd->o_incarnation = 1; |
| 411 | |
| 412 | ceph_con_init(osdc->client->msgr, &osd->o_con); |
| 413 | osd->o_con.private = osd; |
| 414 | osd->o_con.ops = &osd_con_ops; |
| 415 | osd->o_con.peer_name.type = CEPH_ENTITY_TYPE_OSD; |
Sage Weil | 4e7a5dc | 2009-11-18 16:19:57 -0800 | [diff] [blame] | 416 | |
Yehuda Sadeh | 422d2cb | 2010-02-26 15:32:31 -0800 | [diff] [blame] | 417 | INIT_LIST_HEAD(&osd->o_keepalive_item); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 418 | return osd; |
| 419 | } |
| 420 | |
| 421 | static struct ceph_osd *get_osd(struct ceph_osd *osd) |
| 422 | { |
| 423 | if (atomic_inc_not_zero(&osd->o_ref)) { |
| 424 | dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1, |
| 425 | atomic_read(&osd->o_ref)); |
| 426 | return osd; |
| 427 | } else { |
| 428 | dout("get_osd %p FAIL\n", osd); |
| 429 | return NULL; |
| 430 | } |
| 431 | } |
| 432 | |
| 433 | static void put_osd(struct ceph_osd *osd) |
| 434 | { |
| 435 | dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref), |
| 436 | atomic_read(&osd->o_ref) - 1); |
Sage Weil | 79494d1 | 2010-05-27 14:15:49 -0700 | [diff] [blame] | 437 | if (atomic_dec_and_test(&osd->o_ref)) { |
| 438 | struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth; |
| 439 | |
| 440 | if (osd->o_authorizer) |
| 441 | ac->ops->destroy_authorizer(ac, osd->o_authorizer); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 442 | kfree(osd); |
Sage Weil | 79494d1 | 2010-05-27 14:15:49 -0700 | [diff] [blame] | 443 | } |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 444 | } |
| 445 | |
| 446 | /* |
| 447 | * remove an osd from our map |
| 448 | */ |
Yehuda Sadeh | f5a2041 | 2010-02-03 11:00:26 -0800 | [diff] [blame] | 449 | static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd) |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 450 | { |
Yehuda Sadeh | f5a2041 | 2010-02-03 11:00:26 -0800 | [diff] [blame] | 451 | dout("__remove_osd %p\n", osd); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 452 | BUG_ON(!list_empty(&osd->o_requests)); |
| 453 | rb_erase(&osd->o_node, &osdc->osds); |
Yehuda Sadeh | f5a2041 | 2010-02-03 11:00:26 -0800 | [diff] [blame] | 454 | list_del_init(&osd->o_osd_lru); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 455 | ceph_con_close(&osd->o_con); |
| 456 | put_osd(osd); |
| 457 | } |
| 458 | |
Yehuda Sadeh | f5a2041 | 2010-02-03 11:00:26 -0800 | [diff] [blame] | 459 | static void __move_osd_to_lru(struct ceph_osd_client *osdc, |
| 460 | struct ceph_osd *osd) |
| 461 | { |
| 462 | dout("__move_osd_to_lru %p\n", osd); |
| 463 | BUG_ON(!list_empty(&osd->o_osd_lru)); |
| 464 | list_add_tail(&osd->o_osd_lru, &osdc->osd_lru); |
| 465 | osd->lru_ttl = jiffies + osdc->client->mount_args->osd_idle_ttl * HZ; |
| 466 | } |
| 467 | |
| 468 | static void __remove_osd_from_lru(struct ceph_osd *osd) |
| 469 | { |
| 470 | dout("__remove_osd_from_lru %p\n", osd); |
| 471 | if (!list_empty(&osd->o_osd_lru)) |
| 472 | list_del_init(&osd->o_osd_lru); |
| 473 | } |
| 474 | |
| 475 | static void remove_old_osds(struct ceph_osd_client *osdc, int remove_all) |
| 476 | { |
| 477 | struct ceph_osd *osd, *nosd; |
| 478 | |
| 479 | dout("__remove_old_osds %p\n", osdc); |
| 480 | mutex_lock(&osdc->request_mutex); |
| 481 | list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) { |
| 482 | if (!remove_all && time_before(jiffies, osd->lru_ttl)) |
| 483 | break; |
| 484 | __remove_osd(osdc, osd); |
| 485 | } |
| 486 | mutex_unlock(&osdc->request_mutex); |
| 487 | } |
| 488 | |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 489 | /* |
| 490 | * reset osd connect |
| 491 | */ |
Yehuda Sadeh | f5a2041 | 2010-02-03 11:00:26 -0800 | [diff] [blame] | 492 | static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd) |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 493 | { |
Sage Weil | 87b315a | 2010-03-22 14:51:18 -0700 | [diff] [blame] | 494 | struct ceph_osd_request *req; |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 495 | int ret = 0; |
| 496 | |
Yehuda Sadeh | f5a2041 | 2010-02-03 11:00:26 -0800 | [diff] [blame] | 497 | dout("__reset_osd %p osd%d\n", osd, osd->o_osd); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 498 | if (list_empty(&osd->o_requests)) { |
Yehuda Sadeh | f5a2041 | 2010-02-03 11:00:26 -0800 | [diff] [blame] | 499 | __remove_osd(osdc, osd); |
Sage Weil | 87b315a | 2010-03-22 14:51:18 -0700 | [diff] [blame] | 500 | } else if (memcmp(&osdc->osdmap->osd_addr[osd->o_osd], |
| 501 | &osd->o_con.peer_addr, |
| 502 | sizeof(osd->o_con.peer_addr)) == 0 && |
| 503 | !ceph_con_opened(&osd->o_con)) { |
| 504 | dout(" osd addr hasn't changed and connection never opened," |
| 505 | " letting msgr retry"); |
| 506 | /* touch each r_stamp for handle_timeout()'s benfit */ |
| 507 | list_for_each_entry(req, &osd->o_requests, r_osd_item) |
| 508 | req->r_stamp = jiffies; |
| 509 | ret = -EAGAIN; |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 510 | } else { |
| 511 | ceph_con_close(&osd->o_con); |
| 512 | ceph_con_open(&osd->o_con, &osdc->osdmap->osd_addr[osd->o_osd]); |
| 513 | osd->o_incarnation++; |
| 514 | } |
| 515 | return ret; |
| 516 | } |
| 517 | |
| 518 | static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new) |
| 519 | { |
| 520 | struct rb_node **p = &osdc->osds.rb_node; |
| 521 | struct rb_node *parent = NULL; |
| 522 | struct ceph_osd *osd = NULL; |
| 523 | |
| 524 | while (*p) { |
| 525 | parent = *p; |
| 526 | osd = rb_entry(parent, struct ceph_osd, o_node); |
| 527 | if (new->o_osd < osd->o_osd) |
| 528 | p = &(*p)->rb_left; |
| 529 | else if (new->o_osd > osd->o_osd) |
| 530 | p = &(*p)->rb_right; |
| 531 | else |
| 532 | BUG(); |
| 533 | } |
| 534 | |
| 535 | rb_link_node(&new->o_node, parent, p); |
| 536 | rb_insert_color(&new->o_node, &osdc->osds); |
| 537 | } |
| 538 | |
| 539 | static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o) |
| 540 | { |
| 541 | struct ceph_osd *osd; |
| 542 | struct rb_node *n = osdc->osds.rb_node; |
| 543 | |
| 544 | while (n) { |
| 545 | osd = rb_entry(n, struct ceph_osd, o_node); |
| 546 | if (o < osd->o_osd) |
| 547 | n = n->rb_left; |
| 548 | else if (o > osd->o_osd) |
| 549 | n = n->rb_right; |
| 550 | else |
| 551 | return osd; |
| 552 | } |
| 553 | return NULL; |
| 554 | } |
| 555 | |
Yehuda Sadeh | 422d2cb | 2010-02-26 15:32:31 -0800 | [diff] [blame] | 556 | static void __schedule_osd_timeout(struct ceph_osd_client *osdc) |
| 557 | { |
| 558 | schedule_delayed_work(&osdc->timeout_work, |
| 559 | osdc->client->mount_args->osd_keepalive_timeout * HZ); |
| 560 | } |
| 561 | |
| 562 | static void __cancel_osd_timeout(struct ceph_osd_client *osdc) |
| 563 | { |
| 564 | cancel_delayed_work(&osdc->timeout_work); |
| 565 | } |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 566 | |
| 567 | /* |
| 568 | * Register request, assign tid. If this is the first request, set up |
| 569 | * the timeout event. |
| 570 | */ |
| 571 | static void register_request(struct ceph_osd_client *osdc, |
| 572 | struct ceph_osd_request *req) |
| 573 | { |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 574 | mutex_lock(&osdc->request_mutex); |
| 575 | req->r_tid = ++osdc->last_tid; |
Sage Weil | 6df058c | 2009-12-22 11:24:33 -0800 | [diff] [blame] | 576 | req->r_request->hdr.tid = cpu_to_le64(req->r_tid); |
Yehuda Sadeh | 422d2cb | 2010-02-26 15:32:31 -0800 | [diff] [blame] | 577 | INIT_LIST_HEAD(&req->r_req_lru_item); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 578 | |
| 579 | dout("register_request %p tid %lld\n", req, req->r_tid); |
| 580 | __insert_request(osdc, req); |
| 581 | ceph_osdc_get_request(req); |
| 582 | osdc->num_requests++; |
| 583 | |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 584 | if (osdc->num_requests == 1) { |
Yehuda Sadeh | 422d2cb | 2010-02-26 15:32:31 -0800 | [diff] [blame] | 585 | dout(" first request, scheduling timeout\n"); |
| 586 | __schedule_osd_timeout(osdc); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 587 | } |
| 588 | mutex_unlock(&osdc->request_mutex); |
| 589 | } |
| 590 | |
| 591 | /* |
| 592 | * called under osdc->request_mutex |
| 593 | */ |
| 594 | static void __unregister_request(struct ceph_osd_client *osdc, |
| 595 | struct ceph_osd_request *req) |
| 596 | { |
| 597 | dout("__unregister_request %p tid %lld\n", req, req->r_tid); |
| 598 | rb_erase(&req->r_node, &osdc->requests); |
| 599 | osdc->num_requests--; |
| 600 | |
Sage Weil | 0ba6478 | 2009-10-08 16:57:16 -0700 | [diff] [blame] | 601 | if (req->r_osd) { |
| 602 | /* make sure the original request isn't in flight. */ |
| 603 | ceph_con_revoke(&req->r_osd->o_con, req->r_request); |
| 604 | |
| 605 | list_del_init(&req->r_osd_item); |
| 606 | if (list_empty(&req->r_osd->o_requests)) |
Yehuda Sadeh | f5a2041 | 2010-02-03 11:00:26 -0800 | [diff] [blame] | 607 | __move_osd_to_lru(osdc, req->r_osd); |
Sage Weil | 0ba6478 | 2009-10-08 16:57:16 -0700 | [diff] [blame] | 608 | req->r_osd = NULL; |
| 609 | } |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 610 | |
| 611 | ceph_osdc_put_request(req); |
| 612 | |
Yehuda Sadeh | 422d2cb | 2010-02-26 15:32:31 -0800 | [diff] [blame] | 613 | list_del_init(&req->r_req_lru_item); |
| 614 | if (osdc->num_requests == 0) { |
| 615 | dout(" no requests, canceling timeout\n"); |
| 616 | __cancel_osd_timeout(osdc); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 617 | } |
| 618 | } |
| 619 | |
| 620 | /* |
| 621 | * Cancel a previously queued request message |
| 622 | */ |
| 623 | static void __cancel_request(struct ceph_osd_request *req) |
| 624 | { |
Sage Weil | 6bc1887 | 2010-09-27 10:18:52 -0700 | [diff] [blame] | 625 | if (req->r_sent && req->r_osd) { |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 626 | ceph_con_revoke(&req->r_osd->o_con, req->r_request); |
| 627 | req->r_sent = 0; |
| 628 | } |
Yehuda Sadeh | 422d2cb | 2010-02-26 15:32:31 -0800 | [diff] [blame] | 629 | list_del_init(&req->r_req_lru_item); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 630 | } |
| 631 | |
| 632 | /* |
| 633 | * Pick an osd (the first 'up' osd in the pg), allocate the osd struct |
| 634 | * (as needed), and set the request r_osd appropriately. If there is |
| 635 | * no up osd, set r_osd to NULL. |
| 636 | * |
| 637 | * Return 0 if unchanged, 1 if changed, or negative on error. |
| 638 | * |
| 639 | * Caller should hold map_sem for read and request_mutex. |
| 640 | */ |
| 641 | static int __map_osds(struct ceph_osd_client *osdc, |
| 642 | struct ceph_osd_request *req) |
| 643 | { |
| 644 | struct ceph_osd_request_head *reqhead = req->r_request->front.iov_base; |
Sage Weil | 5104212 | 2009-11-04 11:39:12 -0800 | [diff] [blame] | 645 | struct ceph_pg pgid; |
Sage Weil | d85b705 | 2010-05-10 10:24:48 -0700 | [diff] [blame] | 646 | int acting[CEPH_PG_MAX_SIZE]; |
| 647 | int o = -1, num = 0; |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 648 | int err; |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 649 | |
| 650 | dout("map_osds %p tid %lld\n", req, req->r_tid); |
| 651 | err = ceph_calc_object_layout(&reqhead->layout, req->r_oid, |
| 652 | &req->r_file_layout, osdc->osdmap); |
| 653 | if (err) |
| 654 | return err; |
Sage Weil | 5104212 | 2009-11-04 11:39:12 -0800 | [diff] [blame] | 655 | pgid = reqhead->layout.ol_pgid; |
Sage Weil | 7740a42 | 2010-01-08 15:58:25 -0800 | [diff] [blame] | 656 | req->r_pgid = pgid; |
| 657 | |
Sage Weil | d85b705 | 2010-05-10 10:24:48 -0700 | [diff] [blame] | 658 | err = ceph_calc_pg_acting(osdc->osdmap, pgid, acting); |
| 659 | if (err > 0) { |
| 660 | o = acting[0]; |
| 661 | num = err; |
| 662 | } |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 663 | |
| 664 | if ((req->r_osd && req->r_osd->o_osd == o && |
Sage Weil | d85b705 | 2010-05-10 10:24:48 -0700 | [diff] [blame] | 665 | req->r_sent >= req->r_osd->o_incarnation && |
| 666 | req->r_num_pg_osds == num && |
| 667 | memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) || |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 668 | (req->r_osd == NULL && o == -1)) |
| 669 | return 0; /* no change */ |
| 670 | |
Sage Weil | 5104212 | 2009-11-04 11:39:12 -0800 | [diff] [blame] | 671 | dout("map_osds tid %llu pgid %d.%x osd%d (was osd%d)\n", |
| 672 | req->r_tid, le32_to_cpu(pgid.pool), le16_to_cpu(pgid.ps), o, |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 673 | req->r_osd ? req->r_osd->o_osd : -1); |
| 674 | |
Sage Weil | d85b705 | 2010-05-10 10:24:48 -0700 | [diff] [blame] | 675 | /* record full pg acting set */ |
| 676 | memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num); |
| 677 | req->r_num_pg_osds = num; |
| 678 | |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 679 | if (req->r_osd) { |
| 680 | __cancel_request(req); |
| 681 | list_del_init(&req->r_osd_item); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 682 | req->r_osd = NULL; |
| 683 | } |
| 684 | |
| 685 | req->r_osd = __lookup_osd(osdc, o); |
| 686 | if (!req->r_osd && o >= 0) { |
Sage Weil | c99eb1c | 2010-02-26 09:37:33 -0800 | [diff] [blame] | 687 | err = -ENOMEM; |
| 688 | req->r_osd = create_osd(osdc); |
| 689 | if (!req->r_osd) |
| 690 | goto out; |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 691 | |
| 692 | dout("map_osds osd %p is osd%d\n", req->r_osd, o); |
| 693 | req->r_osd->o_osd = o; |
| 694 | req->r_osd->o_con.peer_name.num = cpu_to_le64(o); |
| 695 | __insert_osd(osdc, req->r_osd); |
| 696 | |
| 697 | ceph_con_open(&req->r_osd->o_con, &osdc->osdmap->osd_addr[o]); |
| 698 | } |
| 699 | |
Yehuda Sadeh | f5a2041 | 2010-02-03 11:00:26 -0800 | [diff] [blame] | 700 | if (req->r_osd) { |
| 701 | __remove_osd_from_lru(req->r_osd); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 702 | list_add(&req->r_osd_item, &req->r_osd->o_requests); |
Yehuda Sadeh | f5a2041 | 2010-02-03 11:00:26 -0800 | [diff] [blame] | 703 | } |
Sage Weil | d85b705 | 2010-05-10 10:24:48 -0700 | [diff] [blame] | 704 | err = 1; /* osd or pg changed */ |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 705 | |
| 706 | out: |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 707 | return err; |
| 708 | } |
| 709 | |
| 710 | /* |
| 711 | * caller should hold map_sem (for read) and request_mutex |
| 712 | */ |
| 713 | static int __send_request(struct ceph_osd_client *osdc, |
| 714 | struct ceph_osd_request *req) |
| 715 | { |
| 716 | struct ceph_osd_request_head *reqhead; |
| 717 | int err; |
| 718 | |
| 719 | err = __map_osds(osdc, req); |
| 720 | if (err < 0) |
| 721 | return err; |
| 722 | if (req->r_osd == NULL) { |
| 723 | dout("send_request %p no up osds in pg\n", req); |
| 724 | ceph_monc_request_next_osdmap(&osdc->client->monc); |
| 725 | return 0; |
| 726 | } |
| 727 | |
| 728 | dout("send_request %p tid %llu to osd%d flags %d\n", |
| 729 | req, req->r_tid, req->r_osd->o_osd, req->r_flags); |
| 730 | |
| 731 | reqhead = req->r_request->front.iov_base; |
| 732 | reqhead->osdmap_epoch = cpu_to_le32(osdc->osdmap->epoch); |
| 733 | reqhead->flags |= cpu_to_le32(req->r_flags); /* e.g., RETRY */ |
| 734 | reqhead->reassert_version = req->r_reassert_version; |
| 735 | |
Sage Weil | 3dd72fc | 2010-03-22 14:42:30 -0700 | [diff] [blame] | 736 | req->r_stamp = jiffies; |
Henry C Chang | 07a27e2 | 2010-08-22 21:34:27 -0700 | [diff] [blame] | 737 | list_move_tail(&req->r_req_lru_item, &osdc->req_lru); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 738 | |
| 739 | ceph_msg_get(req->r_request); /* send consumes a ref */ |
| 740 | ceph_con_send(&req->r_osd->o_con, req->r_request); |
| 741 | req->r_sent = req->r_osd->o_incarnation; |
| 742 | return 0; |
| 743 | } |
| 744 | |
| 745 | /* |
| 746 | * Timeout callback, called every N seconds when 1 or more osd |
| 747 | * requests has been active for more than N seconds. When this |
| 748 | * happens, we ping all OSDs with requests who have timed out to |
| 749 | * ensure any communications channel reset is detected. Reset the |
| 750 | * request timeouts another N seconds in the future as we go. |
| 751 | * Reschedule the timeout event another N seconds in future (unless |
| 752 | * there are no open requests). |
| 753 | */ |
| 754 | static void handle_timeout(struct work_struct *work) |
| 755 | { |
| 756 | struct ceph_osd_client *osdc = |
| 757 | container_of(work, struct ceph_osd_client, timeout_work.work); |
Yehuda Sadeh | 422d2cb | 2010-02-26 15:32:31 -0800 | [diff] [blame] | 758 | struct ceph_osd_request *req, *last_req = NULL; |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 759 | struct ceph_osd *osd; |
Sage Weil | 6b80518 | 2009-10-27 11:50:50 -0700 | [diff] [blame] | 760 | unsigned long timeout = osdc->client->mount_args->osd_timeout * HZ; |
Yehuda Sadeh | 422d2cb | 2010-02-26 15:32:31 -0800 | [diff] [blame] | 761 | unsigned long keepalive = |
| 762 | osdc->client->mount_args->osd_keepalive_timeout * HZ; |
Sage Weil | 3dd72fc | 2010-03-22 14:42:30 -0700 | [diff] [blame] | 763 | unsigned long last_stamp = 0; |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 764 | struct rb_node *p; |
Yehuda Sadeh | 422d2cb | 2010-02-26 15:32:31 -0800 | [diff] [blame] | 765 | struct list_head slow_osds; |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 766 | |
| 767 | dout("timeout\n"); |
| 768 | down_read(&osdc->map_sem); |
| 769 | |
| 770 | ceph_monc_request_next_osdmap(&osdc->client->monc); |
| 771 | |
| 772 | mutex_lock(&osdc->request_mutex); |
| 773 | for (p = rb_first(&osdc->requests); p; p = rb_next(p)) { |
| 774 | req = rb_entry(p, struct ceph_osd_request, r_node); |
| 775 | |
| 776 | if (req->r_resend) { |
| 777 | int err; |
| 778 | |
| 779 | dout("osdc resending prev failed %lld\n", req->r_tid); |
| 780 | err = __send_request(osdc, req); |
| 781 | if (err) |
| 782 | dout("osdc failed again on %lld\n", req->r_tid); |
| 783 | else |
| 784 | req->r_resend = false; |
| 785 | continue; |
| 786 | } |
| 787 | } |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 788 | |
Yehuda Sadeh | 422d2cb | 2010-02-26 15:32:31 -0800 | [diff] [blame] | 789 | /* |
| 790 | * reset osds that appear to be _really_ unresponsive. this |
| 791 | * is a failsafe measure.. we really shouldn't be getting to |
| 792 | * this point if the system is working properly. the monitors |
| 793 | * should mark the osd as failed and we should find out about |
| 794 | * it from an updated osd map. |
| 795 | */ |
Sage Weil | f26e681 | 2010-04-21 11:09:38 -0700 | [diff] [blame] | 796 | while (timeout && !list_empty(&osdc->req_lru)) { |
Yehuda Sadeh | 422d2cb | 2010-02-26 15:32:31 -0800 | [diff] [blame] | 797 | req = list_entry(osdc->req_lru.next, struct ceph_osd_request, |
| 798 | r_req_lru_item); |
| 799 | |
Sage Weil | 3dd72fc | 2010-03-22 14:42:30 -0700 | [diff] [blame] | 800 | if (time_before(jiffies, req->r_stamp + timeout)) |
Yehuda Sadeh | 422d2cb | 2010-02-26 15:32:31 -0800 | [diff] [blame] | 801 | break; |
| 802 | |
Sage Weil | 3dd72fc | 2010-03-22 14:42:30 -0700 | [diff] [blame] | 803 | BUG_ON(req == last_req && req->r_stamp == last_stamp); |
Yehuda Sadeh | 422d2cb | 2010-02-26 15:32:31 -0800 | [diff] [blame] | 804 | last_req = req; |
Sage Weil | 3dd72fc | 2010-03-22 14:42:30 -0700 | [diff] [blame] | 805 | last_stamp = req->r_stamp; |
Yehuda Sadeh | 422d2cb | 2010-02-26 15:32:31 -0800 | [diff] [blame] | 806 | |
| 807 | osd = req->r_osd; |
| 808 | BUG_ON(!osd); |
| 809 | pr_warning(" tid %llu timed out on osd%d, will reset osd\n", |
| 810 | req->r_tid, osd->o_osd); |
| 811 | __kick_requests(osdc, osd); |
| 812 | } |
| 813 | |
| 814 | /* |
| 815 | * ping osds that are a bit slow. this ensures that if there |
| 816 | * is a break in the TCP connection we will notice, and reopen |
| 817 | * a connection with that osd (from the fault callback). |
| 818 | */ |
| 819 | INIT_LIST_HEAD(&slow_osds); |
| 820 | list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) { |
Sage Weil | 3dd72fc | 2010-03-22 14:42:30 -0700 | [diff] [blame] | 821 | if (time_before(jiffies, req->r_stamp + keepalive)) |
Yehuda Sadeh | 422d2cb | 2010-02-26 15:32:31 -0800 | [diff] [blame] | 822 | break; |
| 823 | |
| 824 | osd = req->r_osd; |
| 825 | BUG_ON(!osd); |
| 826 | dout(" tid %llu is slow, will send keepalive on osd%d\n", |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 827 | req->r_tid, osd->o_osd); |
Yehuda Sadeh | 422d2cb | 2010-02-26 15:32:31 -0800 | [diff] [blame] | 828 | list_move_tail(&osd->o_keepalive_item, &slow_osds); |
| 829 | } |
| 830 | while (!list_empty(&slow_osds)) { |
| 831 | osd = list_entry(slow_osds.next, struct ceph_osd, |
| 832 | o_keepalive_item); |
| 833 | list_del_init(&osd->o_keepalive_item); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 834 | ceph_con_keepalive(&osd->o_con); |
| 835 | } |
| 836 | |
Yehuda Sadeh | 422d2cb | 2010-02-26 15:32:31 -0800 | [diff] [blame] | 837 | __schedule_osd_timeout(osdc); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 838 | mutex_unlock(&osdc->request_mutex); |
| 839 | |
| 840 | up_read(&osdc->map_sem); |
| 841 | } |
| 842 | |
Yehuda Sadeh | f5a2041 | 2010-02-03 11:00:26 -0800 | [diff] [blame] | 843 | static void handle_osds_timeout(struct work_struct *work) |
| 844 | { |
| 845 | struct ceph_osd_client *osdc = |
| 846 | container_of(work, struct ceph_osd_client, |
| 847 | osds_timeout_work.work); |
| 848 | unsigned long delay = |
| 849 | osdc->client->mount_args->osd_idle_ttl * HZ >> 2; |
| 850 | |
| 851 | dout("osds timeout\n"); |
| 852 | down_read(&osdc->map_sem); |
| 853 | remove_old_osds(osdc, 0); |
| 854 | up_read(&osdc->map_sem); |
| 855 | |
| 856 | schedule_delayed_work(&osdc->osds_timeout_work, |
| 857 | round_jiffies_relative(delay)); |
| 858 | } |
| 859 | |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 860 | /* |
| 861 | * handle osd op reply. either call the callback if it is specified, |
| 862 | * or do the completion to wake up the waiting thread. |
| 863 | */ |
Sage Weil | 350b1c3 | 2009-12-22 10:45:45 -0800 | [diff] [blame] | 864 | static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg, |
| 865 | struct ceph_connection *con) |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 866 | { |
| 867 | struct ceph_osd_reply_head *rhead = msg->front.iov_base; |
| 868 | struct ceph_osd_request *req; |
| 869 | u64 tid; |
| 870 | int numops, object_len, flags; |
Sage Weil | 0ceed5d | 2010-05-11 09:53:18 -0700 | [diff] [blame] | 871 | s32 result; |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 872 | |
Sage Weil | 6df058c | 2009-12-22 11:24:33 -0800 | [diff] [blame] | 873 | tid = le64_to_cpu(msg->hdr.tid); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 874 | if (msg->front.iov_len < sizeof(*rhead)) |
| 875 | goto bad; |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 876 | numops = le32_to_cpu(rhead->num_ops); |
| 877 | object_len = le32_to_cpu(rhead->object_len); |
Sage Weil | 0ceed5d | 2010-05-11 09:53:18 -0700 | [diff] [blame] | 878 | result = le32_to_cpu(rhead->result); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 879 | if (msg->front.iov_len != sizeof(*rhead) + object_len + |
| 880 | numops * sizeof(struct ceph_osd_op)) |
| 881 | goto bad; |
Sage Weil | 0ceed5d | 2010-05-11 09:53:18 -0700 | [diff] [blame] | 882 | dout("handle_reply %p tid %llu result %d\n", msg, tid, (int)result); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 883 | |
| 884 | /* lookup */ |
| 885 | mutex_lock(&osdc->request_mutex); |
| 886 | req = __lookup_request(osdc, tid); |
| 887 | if (req == NULL) { |
| 888 | dout("handle_reply tid %llu dne\n", tid); |
| 889 | mutex_unlock(&osdc->request_mutex); |
| 890 | return; |
| 891 | } |
| 892 | ceph_osdc_get_request(req); |
| 893 | flags = le32_to_cpu(rhead->flags); |
| 894 | |
Sage Weil | 350b1c3 | 2009-12-22 10:45:45 -0800 | [diff] [blame] | 895 | /* |
Yehuda Sadeh | 0d59ab8 | 2010-01-13 17:03:23 -0800 | [diff] [blame] | 896 | * if this connection filled our message, drop our reference now, to |
Sage Weil | 350b1c3 | 2009-12-22 10:45:45 -0800 | [diff] [blame] | 897 | * avoid a (safe but slower) revoke later. |
| 898 | */ |
Yehuda Sadeh | 0d59ab8 | 2010-01-13 17:03:23 -0800 | [diff] [blame] | 899 | if (req->r_con_filling_msg == con && req->r_reply == msg) { |
Sage Weil | c16e786 | 2010-03-01 13:02:00 -0800 | [diff] [blame] | 900 | dout(" dropping con_filling_msg ref %p\n", con); |
Yehuda Sadeh | 0d59ab8 | 2010-01-13 17:03:23 -0800 | [diff] [blame] | 901 | req->r_con_filling_msg = NULL; |
Sage Weil | 350b1c3 | 2009-12-22 10:45:45 -0800 | [diff] [blame] | 902 | ceph_con_put(con); |
| 903 | } |
| 904 | |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 905 | if (!req->r_got_reply) { |
| 906 | unsigned bytes; |
| 907 | |
| 908 | req->r_result = le32_to_cpu(rhead->result); |
| 909 | bytes = le32_to_cpu(msg->hdr.data_len); |
| 910 | dout("handle_reply result %d bytes %d\n", req->r_result, |
| 911 | bytes); |
| 912 | if (req->r_result == 0) |
| 913 | req->r_result = bytes; |
| 914 | |
| 915 | /* in case this is a write and we need to replay, */ |
| 916 | req->r_reassert_version = rhead->reassert_version; |
| 917 | |
| 918 | req->r_got_reply = 1; |
| 919 | } else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) { |
| 920 | dout("handle_reply tid %llu dup ack\n", tid); |
Sage Weil | 34b43a5 | 2009-12-01 12:23:54 -0800 | [diff] [blame] | 921 | mutex_unlock(&osdc->request_mutex); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 922 | goto done; |
| 923 | } |
| 924 | |
| 925 | dout("handle_reply tid %llu flags %d\n", tid, flags); |
| 926 | |
| 927 | /* either this is a read, or we got the safe response */ |
Sage Weil | 0ceed5d | 2010-05-11 09:53:18 -0700 | [diff] [blame] | 928 | if (result < 0 || |
| 929 | (flags & CEPH_OSD_FLAG_ONDISK) || |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 930 | ((flags & CEPH_OSD_FLAG_WRITE) == 0)) |
| 931 | __unregister_request(osdc, req); |
| 932 | |
| 933 | mutex_unlock(&osdc->request_mutex); |
| 934 | |
| 935 | if (req->r_callback) |
| 936 | req->r_callback(req, msg); |
| 937 | else |
Yehuda Sadeh | 03066f2 | 2010-07-27 13:11:08 -0700 | [diff] [blame] | 938 | complete_all(&req->r_completion); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 939 | |
| 940 | if (flags & CEPH_OSD_FLAG_ONDISK) { |
| 941 | if (req->r_safe_callback) |
| 942 | req->r_safe_callback(req, msg); |
Yehuda Sadeh | 03066f2 | 2010-07-27 13:11:08 -0700 | [diff] [blame] | 943 | complete_all(&req->r_safe_completion); /* fsync waiter */ |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 944 | } |
| 945 | |
| 946 | done: |
| 947 | ceph_osdc_put_request(req); |
| 948 | return; |
| 949 | |
| 950 | bad: |
| 951 | pr_err("corrupt osd_op_reply got %d %d expected %d\n", |
| 952 | (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len), |
| 953 | (int)sizeof(*rhead)); |
Sage Weil | 9ec7cab | 2009-12-14 15:13:47 -0800 | [diff] [blame] | 954 | ceph_msg_dump(msg); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 955 | } |
| 956 | |
| 957 | |
Yehuda Sadeh | 422d2cb | 2010-02-26 15:32:31 -0800 | [diff] [blame] | 958 | static int __kick_requests(struct ceph_osd_client *osdc, |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 959 | struct ceph_osd *kickosd) |
| 960 | { |
| 961 | struct ceph_osd_request *req; |
| 962 | struct rb_node *p, *n; |
| 963 | int needmap = 0; |
| 964 | int err; |
| 965 | |
| 966 | dout("kick_requests osd%d\n", kickosd ? kickosd->o_osd : -1); |
Sage Weil | 153a008 | 2010-02-15 12:11:51 -0800 | [diff] [blame] | 967 | if (kickosd) { |
Sage Weil | 87b315a | 2010-03-22 14:51:18 -0700 | [diff] [blame] | 968 | err = __reset_osd(osdc, kickosd); |
| 969 | if (err == -EAGAIN) |
| 970 | return 1; |
Sage Weil | 153a008 | 2010-02-15 12:11:51 -0800 | [diff] [blame] | 971 | } else { |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 972 | for (p = rb_first(&osdc->osds); p; p = n) { |
| 973 | struct ceph_osd *osd = |
| 974 | rb_entry(p, struct ceph_osd, o_node); |
| 975 | |
| 976 | n = rb_next(p); |
| 977 | if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) || |
Sage Weil | 103e2d3 | 2010-01-07 16:12:36 -0800 | [diff] [blame] | 978 | memcmp(&osd->o_con.peer_addr, |
| 979 | ceph_osd_addr(osdc->osdmap, |
| 980 | osd->o_osd), |
| 981 | sizeof(struct ceph_entity_addr)) != 0) |
Yehuda Sadeh | f5a2041 | 2010-02-03 11:00:26 -0800 | [diff] [blame] | 982 | __reset_osd(osdc, osd); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 983 | } |
| 984 | } |
| 985 | |
| 986 | for (p = rb_first(&osdc->requests); p; p = rb_next(p)) { |
| 987 | req = rb_entry(p, struct ceph_osd_request, r_node); |
| 988 | |
| 989 | if (req->r_resend) { |
| 990 | dout(" r_resend set on tid %llu\n", req->r_tid); |
Sage Weil | 266673d | 2009-10-09 10:31:32 -0700 | [diff] [blame] | 991 | __cancel_request(req); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 992 | goto kick; |
| 993 | } |
Sage Weil | 266673d | 2009-10-09 10:31:32 -0700 | [diff] [blame] | 994 | if (req->r_osd && kickosd == req->r_osd) { |
| 995 | __cancel_request(req); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 996 | goto kick; |
Sage Weil | 266673d | 2009-10-09 10:31:32 -0700 | [diff] [blame] | 997 | } |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 998 | |
| 999 | err = __map_osds(osdc, req); |
| 1000 | if (err == 0) |
| 1001 | continue; /* no change */ |
| 1002 | if (err < 0) { |
| 1003 | /* |
| 1004 | * FIXME: really, we should set the request |
| 1005 | * error and fail if this isn't a 'nofail' |
| 1006 | * request, but that's a fair bit more |
| 1007 | * complicated to do. So retry! |
| 1008 | */ |
| 1009 | dout(" setting r_resend on %llu\n", req->r_tid); |
| 1010 | req->r_resend = true; |
| 1011 | continue; |
| 1012 | } |
| 1013 | if (req->r_osd == NULL) { |
| 1014 | dout("tid %llu maps to no valid osd\n", req->r_tid); |
| 1015 | needmap++; /* request a newer map */ |
| 1016 | continue; |
| 1017 | } |
| 1018 | |
| 1019 | kick: |
Sage Weil | c1ea882 | 2009-10-08 16:55:47 -0700 | [diff] [blame] | 1020 | dout("kicking %p tid %llu osd%d\n", req, req->r_tid, |
Sage Weil | 12eadc1 | 2010-03-15 22:20:39 -0700 | [diff] [blame] | 1021 | req->r_osd ? req->r_osd->o_osd : -1); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1022 | req->r_flags |= CEPH_OSD_FLAG_RETRY; |
| 1023 | err = __send_request(osdc, req); |
| 1024 | if (err) { |
| 1025 | dout(" setting r_resend on %llu\n", req->r_tid); |
| 1026 | req->r_resend = true; |
| 1027 | } |
| 1028 | } |
Yehuda Sadeh | 422d2cb | 2010-02-26 15:32:31 -0800 | [diff] [blame] | 1029 | |
| 1030 | return needmap; |
| 1031 | } |
| 1032 | |
| 1033 | /* |
| 1034 | * Resubmit osd requests whose osd or osd address has changed. Request |
| 1035 | * a new osd map if osds are down, or we are otherwise unable to determine |
| 1036 | * how to direct a request. |
| 1037 | * |
| 1038 | * Close connections to down osds. |
| 1039 | * |
| 1040 | * If @who is specified, resubmit requests for that specific osd. |
| 1041 | * |
| 1042 | * Caller should hold map_sem for read and request_mutex. |
| 1043 | */ |
| 1044 | static void kick_requests(struct ceph_osd_client *osdc, |
| 1045 | struct ceph_osd *kickosd) |
| 1046 | { |
| 1047 | int needmap; |
| 1048 | |
| 1049 | mutex_lock(&osdc->request_mutex); |
| 1050 | needmap = __kick_requests(osdc, kickosd); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1051 | mutex_unlock(&osdc->request_mutex); |
| 1052 | |
| 1053 | if (needmap) { |
| 1054 | dout("%d requests for down osds, need new map\n", needmap); |
| 1055 | ceph_monc_request_next_osdmap(&osdc->client->monc); |
| 1056 | } |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1057 | |
Yehuda Sadeh | 422d2cb | 2010-02-26 15:32:31 -0800 | [diff] [blame] | 1058 | } |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1059 | /* |
| 1060 | * Process updated osd map. |
| 1061 | * |
| 1062 | * The message contains any number of incremental and full maps, normally |
| 1063 | * indicating some sort of topology change in the cluster. Kick requests |
| 1064 | * off to different OSDs as needed. |
| 1065 | */ |
| 1066 | void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg) |
| 1067 | { |
| 1068 | void *p, *end, *next; |
| 1069 | u32 nr_maps, maplen; |
| 1070 | u32 epoch; |
| 1071 | struct ceph_osdmap *newmap = NULL, *oldmap; |
| 1072 | int err; |
| 1073 | struct ceph_fsid fsid; |
| 1074 | |
| 1075 | dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0); |
| 1076 | p = msg->front.iov_base; |
| 1077 | end = p + msg->front.iov_len; |
| 1078 | |
| 1079 | /* verify fsid */ |
| 1080 | ceph_decode_need(&p, end, sizeof(fsid), bad); |
| 1081 | ceph_decode_copy(&p, &fsid, sizeof(fsid)); |
Sage Weil | 0743304 | 2009-11-18 16:50:41 -0800 | [diff] [blame] | 1082 | if (ceph_check_fsid(osdc->client, &fsid) < 0) |
| 1083 | return; |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1084 | |
| 1085 | down_write(&osdc->map_sem); |
| 1086 | |
| 1087 | /* incremental maps */ |
| 1088 | ceph_decode_32_safe(&p, end, nr_maps, bad); |
| 1089 | dout(" %d inc maps\n", nr_maps); |
| 1090 | while (nr_maps > 0) { |
| 1091 | ceph_decode_need(&p, end, 2*sizeof(u32), bad); |
Sage Weil | c89136e | 2009-10-14 09:59:09 -0700 | [diff] [blame] | 1092 | epoch = ceph_decode_32(&p); |
| 1093 | maplen = ceph_decode_32(&p); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1094 | ceph_decode_need(&p, end, maplen, bad); |
| 1095 | next = p + maplen; |
| 1096 | if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) { |
| 1097 | dout("applying incremental map %u len %d\n", |
| 1098 | epoch, maplen); |
| 1099 | newmap = osdmap_apply_incremental(&p, next, |
| 1100 | osdc->osdmap, |
| 1101 | osdc->client->msgr); |
| 1102 | if (IS_ERR(newmap)) { |
| 1103 | err = PTR_ERR(newmap); |
| 1104 | goto bad; |
| 1105 | } |
Sage Weil | 30dc638 | 2009-12-21 14:49:37 -0800 | [diff] [blame] | 1106 | BUG_ON(!newmap); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1107 | if (newmap != osdc->osdmap) { |
| 1108 | ceph_osdmap_destroy(osdc->osdmap); |
| 1109 | osdc->osdmap = newmap; |
| 1110 | } |
| 1111 | } else { |
| 1112 | dout("ignoring incremental map %u len %d\n", |
| 1113 | epoch, maplen); |
| 1114 | } |
| 1115 | p = next; |
| 1116 | nr_maps--; |
| 1117 | } |
| 1118 | if (newmap) |
| 1119 | goto done; |
| 1120 | |
| 1121 | /* full maps */ |
| 1122 | ceph_decode_32_safe(&p, end, nr_maps, bad); |
| 1123 | dout(" %d full maps\n", nr_maps); |
| 1124 | while (nr_maps) { |
| 1125 | ceph_decode_need(&p, end, 2*sizeof(u32), bad); |
Sage Weil | c89136e | 2009-10-14 09:59:09 -0700 | [diff] [blame] | 1126 | epoch = ceph_decode_32(&p); |
| 1127 | maplen = ceph_decode_32(&p); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1128 | ceph_decode_need(&p, end, maplen, bad); |
| 1129 | if (nr_maps > 1) { |
| 1130 | dout("skipping non-latest full map %u len %d\n", |
| 1131 | epoch, maplen); |
| 1132 | } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) { |
| 1133 | dout("skipping full map %u len %d, " |
| 1134 | "older than our %u\n", epoch, maplen, |
| 1135 | osdc->osdmap->epoch); |
| 1136 | } else { |
| 1137 | dout("taking full map %u len %d\n", epoch, maplen); |
| 1138 | newmap = osdmap_decode(&p, p+maplen); |
| 1139 | if (IS_ERR(newmap)) { |
| 1140 | err = PTR_ERR(newmap); |
| 1141 | goto bad; |
| 1142 | } |
Sage Weil | 30dc638 | 2009-12-21 14:49:37 -0800 | [diff] [blame] | 1143 | BUG_ON(!newmap); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1144 | oldmap = osdc->osdmap; |
| 1145 | osdc->osdmap = newmap; |
| 1146 | if (oldmap) |
| 1147 | ceph_osdmap_destroy(oldmap); |
| 1148 | } |
| 1149 | p += maplen; |
| 1150 | nr_maps--; |
| 1151 | } |
| 1152 | |
| 1153 | done: |
| 1154 | downgrade_write(&osdc->map_sem); |
| 1155 | ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch); |
| 1156 | if (newmap) |
| 1157 | kick_requests(osdc, NULL); |
| 1158 | up_read(&osdc->map_sem); |
Yehuda Sadeh | 03066f2 | 2010-07-27 13:11:08 -0700 | [diff] [blame] | 1159 | wake_up_all(&osdc->client->auth_wq); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1160 | return; |
| 1161 | |
| 1162 | bad: |
| 1163 | pr_err("osdc handle_map corrupt msg\n"); |
Sage Weil | 9ec7cab | 2009-12-14 15:13:47 -0800 | [diff] [blame] | 1164 | ceph_msg_dump(msg); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1165 | up_write(&osdc->map_sem); |
| 1166 | return; |
| 1167 | } |
| 1168 | |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1169 | /* |
| 1170 | * Register request, send initial attempt. |
| 1171 | */ |
| 1172 | int ceph_osdc_start_request(struct ceph_osd_client *osdc, |
| 1173 | struct ceph_osd_request *req, |
| 1174 | bool nofail) |
| 1175 | { |
Sage Weil | c1ea882 | 2009-10-08 16:55:47 -0700 | [diff] [blame] | 1176 | int rc = 0; |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1177 | |
| 1178 | req->r_request->pages = req->r_pages; |
| 1179 | req->r_request->nr_pages = req->r_num_pages; |
| 1180 | |
| 1181 | register_request(osdc, req); |
| 1182 | |
| 1183 | down_read(&osdc->map_sem); |
| 1184 | mutex_lock(&osdc->request_mutex); |
Sage Weil | c1ea882 | 2009-10-08 16:55:47 -0700 | [diff] [blame] | 1185 | /* |
| 1186 | * a racing kick_requests() may have sent the message for us |
| 1187 | * while we dropped request_mutex above, so only send now if |
| 1188 | * the request still han't been touched yet. |
| 1189 | */ |
| 1190 | if (req->r_sent == 0) { |
| 1191 | rc = __send_request(osdc, req); |
| 1192 | if (rc) { |
| 1193 | if (nofail) { |
| 1194 | dout("osdc_start_request failed send, " |
| 1195 | " marking %lld\n", req->r_tid); |
| 1196 | req->r_resend = true; |
| 1197 | rc = 0; |
| 1198 | } else { |
| 1199 | __unregister_request(osdc, req); |
| 1200 | } |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1201 | } |
| 1202 | } |
| 1203 | mutex_unlock(&osdc->request_mutex); |
| 1204 | up_read(&osdc->map_sem); |
| 1205 | return rc; |
| 1206 | } |
| 1207 | |
| 1208 | /* |
| 1209 | * wait for a request to complete |
| 1210 | */ |
| 1211 | int ceph_osdc_wait_request(struct ceph_osd_client *osdc, |
| 1212 | struct ceph_osd_request *req) |
| 1213 | { |
| 1214 | int rc; |
| 1215 | |
| 1216 | rc = wait_for_completion_interruptible(&req->r_completion); |
| 1217 | if (rc < 0) { |
| 1218 | mutex_lock(&osdc->request_mutex); |
| 1219 | __cancel_request(req); |
Sage Weil | 529cfcc | 2009-12-22 10:29:39 -0800 | [diff] [blame] | 1220 | __unregister_request(osdc, req); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1221 | mutex_unlock(&osdc->request_mutex); |
Sage Weil | 529cfcc | 2009-12-22 10:29:39 -0800 | [diff] [blame] | 1222 | dout("wait_request tid %llu canceled/timed out\n", req->r_tid); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1223 | return rc; |
| 1224 | } |
| 1225 | |
| 1226 | dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result); |
| 1227 | return req->r_result; |
| 1228 | } |
| 1229 | |
| 1230 | /* |
| 1231 | * sync - wait for all in-flight requests to flush. avoid starvation. |
| 1232 | */ |
| 1233 | void ceph_osdc_sync(struct ceph_osd_client *osdc) |
| 1234 | { |
| 1235 | struct ceph_osd_request *req; |
| 1236 | u64 last_tid, next_tid = 0; |
| 1237 | |
| 1238 | mutex_lock(&osdc->request_mutex); |
| 1239 | last_tid = osdc->last_tid; |
| 1240 | while (1) { |
| 1241 | req = __lookup_request_ge(osdc, next_tid); |
| 1242 | if (!req) |
| 1243 | break; |
| 1244 | if (req->r_tid > last_tid) |
| 1245 | break; |
| 1246 | |
| 1247 | next_tid = req->r_tid + 1; |
| 1248 | if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0) |
| 1249 | continue; |
| 1250 | |
| 1251 | ceph_osdc_get_request(req); |
| 1252 | mutex_unlock(&osdc->request_mutex); |
| 1253 | dout("sync waiting on tid %llu (last is %llu)\n", |
| 1254 | req->r_tid, last_tid); |
| 1255 | wait_for_completion(&req->r_safe_completion); |
| 1256 | mutex_lock(&osdc->request_mutex); |
| 1257 | ceph_osdc_put_request(req); |
| 1258 | } |
| 1259 | mutex_unlock(&osdc->request_mutex); |
| 1260 | dout("sync done (thru tid %llu)\n", last_tid); |
| 1261 | } |
| 1262 | |
| 1263 | /* |
| 1264 | * init, shutdown |
| 1265 | */ |
| 1266 | int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client) |
| 1267 | { |
| 1268 | int err; |
| 1269 | |
| 1270 | dout("init\n"); |
| 1271 | osdc->client = client; |
| 1272 | osdc->osdmap = NULL; |
| 1273 | init_rwsem(&osdc->map_sem); |
| 1274 | init_completion(&osdc->map_waiters); |
| 1275 | osdc->last_requested_map = 0; |
| 1276 | mutex_init(&osdc->request_mutex); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1277 | osdc->last_tid = 0; |
| 1278 | osdc->osds = RB_ROOT; |
Yehuda Sadeh | f5a2041 | 2010-02-03 11:00:26 -0800 | [diff] [blame] | 1279 | INIT_LIST_HEAD(&osdc->osd_lru); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1280 | osdc->requests = RB_ROOT; |
Yehuda Sadeh | 422d2cb | 2010-02-26 15:32:31 -0800 | [diff] [blame] | 1281 | INIT_LIST_HEAD(&osdc->req_lru); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1282 | osdc->num_requests = 0; |
| 1283 | INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout); |
Yehuda Sadeh | f5a2041 | 2010-02-03 11:00:26 -0800 | [diff] [blame] | 1284 | INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout); |
| 1285 | |
| 1286 | schedule_delayed_work(&osdc->osds_timeout_work, |
| 1287 | round_jiffies_relative(osdc->client->mount_args->osd_idle_ttl * HZ)); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1288 | |
Sage Weil | 5f44f14 | 2009-11-18 14:52:18 -0800 | [diff] [blame] | 1289 | err = -ENOMEM; |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1290 | osdc->req_mempool = mempool_create_kmalloc_pool(10, |
| 1291 | sizeof(struct ceph_osd_request)); |
| 1292 | if (!osdc->req_mempool) |
Sage Weil | 5f44f14 | 2009-11-18 14:52:18 -0800 | [diff] [blame] | 1293 | goto out; |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1294 | |
Sage Weil | 4f48280 | 2010-04-24 09:56:35 -0700 | [diff] [blame] | 1295 | err = ceph_msgpool_init(&osdc->msgpool_op, OSD_OP_FRONT_LEN, 10, true, |
| 1296 | "osd_op"); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1297 | if (err < 0) |
Sage Weil | 5f44f14 | 2009-11-18 14:52:18 -0800 | [diff] [blame] | 1298 | goto out_mempool; |
Sage Weil | c16e786 | 2010-03-01 13:02:00 -0800 | [diff] [blame] | 1299 | err = ceph_msgpool_init(&osdc->msgpool_op_reply, |
Sage Weil | 4f48280 | 2010-04-24 09:56:35 -0700 | [diff] [blame] | 1300 | OSD_OPREPLY_FRONT_LEN, 10, true, |
| 1301 | "osd_op_reply"); |
Sage Weil | c16e786 | 2010-03-01 13:02:00 -0800 | [diff] [blame] | 1302 | if (err < 0) |
| 1303 | goto out_msgpool; |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1304 | return 0; |
Sage Weil | 5f44f14 | 2009-11-18 14:52:18 -0800 | [diff] [blame] | 1305 | |
Sage Weil | c16e786 | 2010-03-01 13:02:00 -0800 | [diff] [blame] | 1306 | out_msgpool: |
| 1307 | ceph_msgpool_destroy(&osdc->msgpool_op); |
Sage Weil | 5f44f14 | 2009-11-18 14:52:18 -0800 | [diff] [blame] | 1308 | out_mempool: |
| 1309 | mempool_destroy(osdc->req_mempool); |
| 1310 | out: |
| 1311 | return err; |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1312 | } |
| 1313 | |
| 1314 | void ceph_osdc_stop(struct ceph_osd_client *osdc) |
| 1315 | { |
| 1316 | cancel_delayed_work_sync(&osdc->timeout_work); |
Yehuda Sadeh | f5a2041 | 2010-02-03 11:00:26 -0800 | [diff] [blame] | 1317 | cancel_delayed_work_sync(&osdc->osds_timeout_work); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1318 | if (osdc->osdmap) { |
| 1319 | ceph_osdmap_destroy(osdc->osdmap); |
| 1320 | osdc->osdmap = NULL; |
| 1321 | } |
Yehuda Sadeh | f5a2041 | 2010-02-03 11:00:26 -0800 | [diff] [blame] | 1322 | remove_old_osds(osdc, 1); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1323 | mempool_destroy(osdc->req_mempool); |
| 1324 | ceph_msgpool_destroy(&osdc->msgpool_op); |
Sage Weil | c16e786 | 2010-03-01 13:02:00 -0800 | [diff] [blame] | 1325 | ceph_msgpool_destroy(&osdc->msgpool_op_reply); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1326 | } |
| 1327 | |
| 1328 | /* |
| 1329 | * Read some contiguous pages. If we cross a stripe boundary, shorten |
| 1330 | * *plen. Return number of bytes read, or error. |
| 1331 | */ |
| 1332 | int ceph_osdc_readpages(struct ceph_osd_client *osdc, |
| 1333 | struct ceph_vino vino, struct ceph_file_layout *layout, |
| 1334 | u64 off, u64 *plen, |
| 1335 | u32 truncate_seq, u64 truncate_size, |
| 1336 | struct page **pages, int num_pages) |
| 1337 | { |
| 1338 | struct ceph_osd_request *req; |
| 1339 | int rc = 0; |
| 1340 | |
| 1341 | dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino, |
| 1342 | vino.snap, off, *plen); |
| 1343 | req = ceph_osdc_new_request(osdc, layout, vino, off, plen, |
| 1344 | CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ, |
| 1345 | NULL, 0, truncate_seq, truncate_size, NULL, |
| 1346 | false, 1); |
Sage Weil | a79832f | 2010-04-01 16:06:19 -0700 | [diff] [blame] | 1347 | if (!req) |
| 1348 | return -ENOMEM; |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1349 | |
| 1350 | /* it may be a short read due to an object boundary */ |
| 1351 | req->r_pages = pages; |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1352 | |
| 1353 | dout("readpages final extent is %llu~%llu (%d pages)\n", |
| 1354 | off, *plen, req->r_num_pages); |
| 1355 | |
| 1356 | rc = ceph_osdc_start_request(osdc, req, false); |
| 1357 | if (!rc) |
| 1358 | rc = ceph_osdc_wait_request(osdc, req); |
| 1359 | |
| 1360 | ceph_osdc_put_request(req); |
| 1361 | dout("readpages result %d\n", rc); |
| 1362 | return rc; |
| 1363 | } |
| 1364 | |
| 1365 | /* |
| 1366 | * do a synchronous write on N pages |
| 1367 | */ |
| 1368 | int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino, |
| 1369 | struct ceph_file_layout *layout, |
| 1370 | struct ceph_snap_context *snapc, |
| 1371 | u64 off, u64 len, |
| 1372 | u32 truncate_seq, u64 truncate_size, |
| 1373 | struct timespec *mtime, |
| 1374 | struct page **pages, int num_pages, |
| 1375 | int flags, int do_sync, bool nofail) |
| 1376 | { |
| 1377 | struct ceph_osd_request *req; |
| 1378 | int rc = 0; |
| 1379 | |
| 1380 | BUG_ON(vino.snap != CEPH_NOSNAP); |
| 1381 | req = ceph_osdc_new_request(osdc, layout, vino, off, &len, |
| 1382 | CEPH_OSD_OP_WRITE, |
| 1383 | flags | CEPH_OSD_FLAG_ONDISK | |
| 1384 | CEPH_OSD_FLAG_WRITE, |
| 1385 | snapc, do_sync, |
| 1386 | truncate_seq, truncate_size, mtime, |
| 1387 | nofail, 1); |
Sage Weil | a79832f | 2010-04-01 16:06:19 -0700 | [diff] [blame] | 1388 | if (!req) |
| 1389 | return -ENOMEM; |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1390 | |
| 1391 | /* it may be a short write due to an object boundary */ |
| 1392 | req->r_pages = pages; |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1393 | dout("writepages %llu~%llu (%d pages)\n", off, len, |
| 1394 | req->r_num_pages); |
| 1395 | |
| 1396 | rc = ceph_osdc_start_request(osdc, req, nofail); |
| 1397 | if (!rc) |
| 1398 | rc = ceph_osdc_wait_request(osdc, req); |
| 1399 | |
| 1400 | ceph_osdc_put_request(req); |
| 1401 | if (rc == 0) |
| 1402 | rc = len; |
| 1403 | dout("writepages result %d\n", rc); |
| 1404 | return rc; |
| 1405 | } |
| 1406 | |
| 1407 | /* |
| 1408 | * handle incoming message |
| 1409 | */ |
| 1410 | static void dispatch(struct ceph_connection *con, struct ceph_msg *msg) |
| 1411 | { |
| 1412 | struct ceph_osd *osd = con->private; |
Julia Lawall | 32c895e | 2009-11-21 16:53:16 +0100 | [diff] [blame] | 1413 | struct ceph_osd_client *osdc; |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1414 | int type = le16_to_cpu(msg->hdr.type); |
| 1415 | |
| 1416 | if (!osd) |
Sage Weil | 4a32f93 | 2010-06-13 10:27:53 -0700 | [diff] [blame] | 1417 | goto out; |
Julia Lawall | 32c895e | 2009-11-21 16:53:16 +0100 | [diff] [blame] | 1418 | osdc = osd->o_osdc; |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1419 | |
| 1420 | switch (type) { |
| 1421 | case CEPH_MSG_OSD_MAP: |
| 1422 | ceph_osdc_handle_map(osdc, msg); |
| 1423 | break; |
| 1424 | case CEPH_MSG_OSD_OPREPLY: |
Sage Weil | 350b1c3 | 2009-12-22 10:45:45 -0800 | [diff] [blame] | 1425 | handle_reply(osdc, msg, con); |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1426 | break; |
| 1427 | |
| 1428 | default: |
| 1429 | pr_err("received unknown message type %d %s\n", type, |
| 1430 | ceph_msg_type_name(type)); |
| 1431 | } |
Sage Weil | 4a32f93 | 2010-06-13 10:27:53 -0700 | [diff] [blame] | 1432 | out: |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1433 | ceph_msg_put(msg); |
| 1434 | } |
| 1435 | |
Sage Weil | 5b3a4db | 2010-02-19 21:43:23 -0800 | [diff] [blame] | 1436 | /* |
Sage Weil | 21b667f | 2010-03-04 10:22:59 -0800 | [diff] [blame] | 1437 | * lookup and return message for incoming reply. set up reply message |
| 1438 | * pages. |
Sage Weil | 5b3a4db | 2010-02-19 21:43:23 -0800 | [diff] [blame] | 1439 | */ |
| 1440 | static struct ceph_msg *get_reply(struct ceph_connection *con, |
Yehuda Sadeh | 2450418 | 2010-01-08 13:58:34 -0800 | [diff] [blame] | 1441 | struct ceph_msg_header *hdr, |
| 1442 | int *skip) |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1443 | { |
| 1444 | struct ceph_osd *osd = con->private; |
| 1445 | struct ceph_osd_client *osdc = osd->o_osdc; |
Yehuda Sadeh | 2450418 | 2010-01-08 13:58:34 -0800 | [diff] [blame] | 1446 | struct ceph_msg *m; |
Yehuda Sadeh | 0547a9b | 2010-01-11 14:47:13 -0800 | [diff] [blame] | 1447 | struct ceph_osd_request *req; |
Sage Weil | 5b3a4db | 2010-02-19 21:43:23 -0800 | [diff] [blame] | 1448 | int front = le32_to_cpu(hdr->front_len); |
| 1449 | int data_len = le32_to_cpu(hdr->data_len); |
Yehuda Sadeh | 0547a9b | 2010-01-11 14:47:13 -0800 | [diff] [blame] | 1450 | u64 tid; |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1451 | |
Yehuda Sadeh | 0547a9b | 2010-01-11 14:47:13 -0800 | [diff] [blame] | 1452 | tid = le64_to_cpu(hdr->tid); |
| 1453 | mutex_lock(&osdc->request_mutex); |
| 1454 | req = __lookup_request(osdc, tid); |
| 1455 | if (!req) { |
| 1456 | *skip = 1; |
| 1457 | m = NULL; |
Sage Weil | c16e786 | 2010-03-01 13:02:00 -0800 | [diff] [blame] | 1458 | pr_info("get_reply unknown tid %llu from osd%d\n", tid, |
Sage Weil | 5b3a4db | 2010-02-19 21:43:23 -0800 | [diff] [blame] | 1459 | osd->o_osd); |
Yehuda Sadeh | 0547a9b | 2010-01-11 14:47:13 -0800 | [diff] [blame] | 1460 | goto out; |
| 1461 | } |
Sage Weil | c16e786 | 2010-03-01 13:02:00 -0800 | [diff] [blame] | 1462 | |
| 1463 | if (req->r_con_filling_msg) { |
| 1464 | dout("get_reply revoking msg %p from old con %p\n", |
| 1465 | req->r_reply, req->r_con_filling_msg); |
| 1466 | ceph_con_revoke_message(req->r_con_filling_msg, req->r_reply); |
| 1467 | ceph_con_put(req->r_con_filling_msg); |
Sage Weil | 6f46cb2 | 2010-03-24 21:30:19 -0700 | [diff] [blame] | 1468 | req->r_con_filling_msg = NULL; |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1469 | } |
Yehuda Sadeh | 2450418 | 2010-01-08 13:58:34 -0800 | [diff] [blame] | 1470 | |
Sage Weil | c16e786 | 2010-03-01 13:02:00 -0800 | [diff] [blame] | 1471 | if (front > req->r_reply->front.iov_len) { |
| 1472 | pr_warning("get_reply front %d > preallocated %d\n", |
| 1473 | front, (int)req->r_reply->front.iov_len); |
Yehuda Sadeh | 34d2376 | 2010-04-06 14:33:58 -0700 | [diff] [blame] | 1474 | m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front, GFP_NOFS); |
Sage Weil | a79832f | 2010-04-01 16:06:19 -0700 | [diff] [blame] | 1475 | if (!m) |
Sage Weil | c16e786 | 2010-03-01 13:02:00 -0800 | [diff] [blame] | 1476 | goto out; |
| 1477 | ceph_msg_put(req->r_reply); |
| 1478 | req->r_reply = m; |
| 1479 | } |
| 1480 | m = ceph_msg_get(req->r_reply); |
| 1481 | |
Yehuda Sadeh | 0547a9b | 2010-01-11 14:47:13 -0800 | [diff] [blame] | 1482 | if (data_len > 0) { |
Sage Weil | 21b667f | 2010-03-04 10:22:59 -0800 | [diff] [blame] | 1483 | unsigned data_off = le16_to_cpu(hdr->data_off); |
| 1484 | int want = calc_pages_for(data_off & ~PAGE_MASK, data_len); |
| 1485 | |
| 1486 | if (unlikely(req->r_num_pages < want)) { |
| 1487 | pr_warning("tid %lld reply %d > expected %d pages\n", |
| 1488 | tid, want, m->nr_pages); |
Yehuda Sadeh | 0547a9b | 2010-01-11 14:47:13 -0800 | [diff] [blame] | 1489 | *skip = 1; |
| 1490 | ceph_msg_put(m); |
Sage Weil | a79832f | 2010-04-01 16:06:19 -0700 | [diff] [blame] | 1491 | m = NULL; |
Sage Weil | 21b667f | 2010-03-04 10:22:59 -0800 | [diff] [blame] | 1492 | goto out; |
Yehuda Sadeh | 0547a9b | 2010-01-11 14:47:13 -0800 | [diff] [blame] | 1493 | } |
Sage Weil | 21b667f | 2010-03-04 10:22:59 -0800 | [diff] [blame] | 1494 | m->pages = req->r_pages; |
| 1495 | m->nr_pages = req->r_num_pages; |
Yehuda Sadeh | 0547a9b | 2010-01-11 14:47:13 -0800 | [diff] [blame] | 1496 | } |
Sage Weil | 5b3a4db | 2010-02-19 21:43:23 -0800 | [diff] [blame] | 1497 | *skip = 0; |
Sage Weil | c16e786 | 2010-03-01 13:02:00 -0800 | [diff] [blame] | 1498 | req->r_con_filling_msg = ceph_con_get(con); |
| 1499 | dout("get_reply tid %lld %p\n", tid, m); |
Yehuda Sadeh | 0547a9b | 2010-01-11 14:47:13 -0800 | [diff] [blame] | 1500 | |
| 1501 | out: |
| 1502 | mutex_unlock(&osdc->request_mutex); |
Yehuda Sadeh | 2450418 | 2010-01-08 13:58:34 -0800 | [diff] [blame] | 1503 | return m; |
Sage Weil | 5b3a4db | 2010-02-19 21:43:23 -0800 | [diff] [blame] | 1504 | |
| 1505 | } |
| 1506 | |
| 1507 | static struct ceph_msg *alloc_msg(struct ceph_connection *con, |
| 1508 | struct ceph_msg_header *hdr, |
| 1509 | int *skip) |
| 1510 | { |
| 1511 | struct ceph_osd *osd = con->private; |
| 1512 | int type = le16_to_cpu(hdr->type); |
| 1513 | int front = le32_to_cpu(hdr->front_len); |
| 1514 | |
| 1515 | switch (type) { |
| 1516 | case CEPH_MSG_OSD_MAP: |
Yehuda Sadeh | 34d2376 | 2010-04-06 14:33:58 -0700 | [diff] [blame] | 1517 | return ceph_msg_new(type, front, GFP_NOFS); |
Sage Weil | 5b3a4db | 2010-02-19 21:43:23 -0800 | [diff] [blame] | 1518 | case CEPH_MSG_OSD_OPREPLY: |
| 1519 | return get_reply(con, hdr, skip); |
| 1520 | default: |
| 1521 | pr_info("alloc_msg unexpected msg type %d from osd%d\n", type, |
| 1522 | osd->o_osd); |
| 1523 | *skip = 1; |
| 1524 | return NULL; |
| 1525 | } |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1526 | } |
| 1527 | |
| 1528 | /* |
| 1529 | * Wrappers to refcount containing ceph_osd struct |
| 1530 | */ |
| 1531 | static struct ceph_connection *get_osd_con(struct ceph_connection *con) |
| 1532 | { |
| 1533 | struct ceph_osd *osd = con->private; |
| 1534 | if (get_osd(osd)) |
| 1535 | return con; |
| 1536 | return NULL; |
| 1537 | } |
| 1538 | |
| 1539 | static void put_osd_con(struct ceph_connection *con) |
| 1540 | { |
| 1541 | struct ceph_osd *osd = con->private; |
| 1542 | put_osd(osd); |
| 1543 | } |
| 1544 | |
Sage Weil | 4e7a5dc | 2009-11-18 16:19:57 -0800 | [diff] [blame] | 1545 | /* |
| 1546 | * authentication |
| 1547 | */ |
| 1548 | static int get_authorizer(struct ceph_connection *con, |
Sage Weil | 213c99e | 2010-08-03 10:25:11 -0700 | [diff] [blame] | 1549 | void **buf, int *len, int *proto, |
| 1550 | void **reply_buf, int *reply_len, int force_new) |
Sage Weil | 4e7a5dc | 2009-11-18 16:19:57 -0800 | [diff] [blame] | 1551 | { |
| 1552 | struct ceph_osd *o = con->private; |
| 1553 | struct ceph_osd_client *osdc = o->o_osdc; |
| 1554 | struct ceph_auth_client *ac = osdc->client->monc.auth; |
| 1555 | int ret = 0; |
| 1556 | |
| 1557 | if (force_new && o->o_authorizer) { |
| 1558 | ac->ops->destroy_authorizer(ac, o->o_authorizer); |
| 1559 | o->o_authorizer = NULL; |
| 1560 | } |
| 1561 | if (o->o_authorizer == NULL) { |
| 1562 | ret = ac->ops->create_authorizer( |
| 1563 | ac, CEPH_ENTITY_TYPE_OSD, |
| 1564 | &o->o_authorizer, |
| 1565 | &o->o_authorizer_buf, |
| 1566 | &o->o_authorizer_buf_len, |
| 1567 | &o->o_authorizer_reply_buf, |
| 1568 | &o->o_authorizer_reply_buf_len); |
| 1569 | if (ret) |
Sage Weil | 213c99e | 2010-08-03 10:25:11 -0700 | [diff] [blame] | 1570 | return ret; |
Sage Weil | 4e7a5dc | 2009-11-18 16:19:57 -0800 | [diff] [blame] | 1571 | } |
| 1572 | |
| 1573 | *proto = ac->protocol; |
| 1574 | *buf = o->o_authorizer_buf; |
| 1575 | *len = o->o_authorizer_buf_len; |
| 1576 | *reply_buf = o->o_authorizer_reply_buf; |
| 1577 | *reply_len = o->o_authorizer_reply_buf_len; |
| 1578 | return 0; |
| 1579 | } |
| 1580 | |
| 1581 | |
| 1582 | static int verify_authorizer_reply(struct ceph_connection *con, int len) |
| 1583 | { |
| 1584 | struct ceph_osd *o = con->private; |
| 1585 | struct ceph_osd_client *osdc = o->o_osdc; |
| 1586 | struct ceph_auth_client *ac = osdc->client->monc.auth; |
| 1587 | |
| 1588 | return ac->ops->verify_authorizer_reply(ac, o->o_authorizer, len); |
| 1589 | } |
| 1590 | |
Sage Weil | 9bd2e6f | 2010-02-02 16:21:06 -0800 | [diff] [blame] | 1591 | static int invalidate_authorizer(struct ceph_connection *con) |
| 1592 | { |
| 1593 | struct ceph_osd *o = con->private; |
| 1594 | struct ceph_osd_client *osdc = o->o_osdc; |
| 1595 | struct ceph_auth_client *ac = osdc->client->monc.auth; |
| 1596 | |
| 1597 | if (ac->ops->invalidate_authorizer) |
| 1598 | ac->ops->invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD); |
| 1599 | |
| 1600 | return ceph_monc_validate_auth(&osdc->client->monc); |
| 1601 | } |
Sage Weil | 4e7a5dc | 2009-11-18 16:19:57 -0800 | [diff] [blame] | 1602 | |
Tobias Klauser | 9e32789 | 2010-05-20 10:40:19 +0200 | [diff] [blame] | 1603 | static const struct ceph_connection_operations osd_con_ops = { |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1604 | .get = get_osd_con, |
| 1605 | .put = put_osd_con, |
| 1606 | .dispatch = dispatch, |
Sage Weil | 4e7a5dc | 2009-11-18 16:19:57 -0800 | [diff] [blame] | 1607 | .get_authorizer = get_authorizer, |
| 1608 | .verify_authorizer_reply = verify_authorizer_reply, |
Sage Weil | 9bd2e6f | 2010-02-02 16:21:06 -0800 | [diff] [blame] | 1609 | .invalidate_authorizer = invalidate_authorizer, |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1610 | .alloc_msg = alloc_msg, |
Sage Weil | 81b024e | 2009-10-09 10:29:18 -0700 | [diff] [blame] | 1611 | .fault = osd_reset, |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 1612 | }; |