Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 1 | #include "ceph_debug.h" |
| 2 | |
| 3 | #include <linux/wait.h> |
| 4 | #include <linux/sched.h> |
| 5 | |
| 6 | #include "mds_client.h" |
| 7 | #include "mon_client.h" |
| 8 | #include "super.h" |
| 9 | #include "messenger.h" |
| 10 | #include "decode.h" |
Sage Weil | 4e7a5dc | 2009-11-18 16:19:57 -0800 | [diff] [blame] | 11 | #include "auth.h" |
Sage Weil | 93cea5b | 2009-12-23 12:21:51 -0800 | [diff] [blame] | 12 | #include "pagelist.h" |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 13 | |
| 14 | /* |
| 15 | * A cluster of MDS (metadata server) daemons is responsible for |
| 16 | * managing the file system namespace (the directory hierarchy and |
| 17 | * inodes) and for coordinating shared access to storage. Metadata is |
| 18 | * partitioning hierarchically across a number of servers, and that |
| 19 | * partition varies over time as the cluster adjusts the distribution |
| 20 | * in order to balance load. |
| 21 | * |
| 22 | * The MDS client is primarily responsible to managing synchronous |
| 23 | * metadata requests for operations like open, unlink, and so forth. |
| 24 | * If there is a MDS failure, we find out about it when we (possibly |
| 25 | * request and) receive a new MDS map, and can resubmit affected |
| 26 | * requests. |
| 27 | * |
| 28 | * For the most part, though, we take advantage of a lossless |
| 29 | * communications channel to the MDS, and do not need to worry about |
| 30 | * timing out or resubmitting requests. |
| 31 | * |
| 32 | * We maintain a stateful "session" with each MDS we interact with. |
| 33 | * Within each session, we sent periodic heartbeat messages to ensure |
| 34 | * any capabilities or leases we have been issues remain valid. If |
| 35 | * the session times out and goes stale, our leases and capabilities |
| 36 | * are no longer valid. |
| 37 | */ |
| 38 | |
| 39 | static void __wake_requests(struct ceph_mds_client *mdsc, |
| 40 | struct list_head *head); |
| 41 | |
| 42 | const static struct ceph_connection_operations mds_con_ops; |
| 43 | |
| 44 | |
| 45 | /* |
| 46 | * mds reply parsing |
| 47 | */ |
| 48 | |
| 49 | /* |
| 50 | * parse individual inode info |
| 51 | */ |
| 52 | static int parse_reply_info_in(void **p, void *end, |
| 53 | struct ceph_mds_reply_info_in *info) |
| 54 | { |
| 55 | int err = -EIO; |
| 56 | |
| 57 | info->in = *p; |
| 58 | *p += sizeof(struct ceph_mds_reply_inode) + |
| 59 | sizeof(*info->in->fragtree.splits) * |
| 60 | le32_to_cpu(info->in->fragtree.nsplits); |
| 61 | |
| 62 | ceph_decode_32_safe(p, end, info->symlink_len, bad); |
| 63 | ceph_decode_need(p, end, info->symlink_len, bad); |
| 64 | info->symlink = *p; |
| 65 | *p += info->symlink_len; |
| 66 | |
| 67 | ceph_decode_32_safe(p, end, info->xattr_len, bad); |
| 68 | ceph_decode_need(p, end, info->xattr_len, bad); |
| 69 | info->xattr_data = *p; |
| 70 | *p += info->xattr_len; |
| 71 | return 0; |
| 72 | bad: |
| 73 | return err; |
| 74 | } |
| 75 | |
| 76 | /* |
| 77 | * parse a normal reply, which may contain a (dir+)dentry and/or a |
| 78 | * target inode. |
| 79 | */ |
| 80 | static int parse_reply_info_trace(void **p, void *end, |
| 81 | struct ceph_mds_reply_info_parsed *info) |
| 82 | { |
| 83 | int err; |
| 84 | |
| 85 | if (info->head->is_dentry) { |
| 86 | err = parse_reply_info_in(p, end, &info->diri); |
| 87 | if (err < 0) |
| 88 | goto out_bad; |
| 89 | |
| 90 | if (unlikely(*p + sizeof(*info->dirfrag) > end)) |
| 91 | goto bad; |
| 92 | info->dirfrag = *p; |
| 93 | *p += sizeof(*info->dirfrag) + |
| 94 | sizeof(u32)*le32_to_cpu(info->dirfrag->ndist); |
| 95 | if (unlikely(*p > end)) |
| 96 | goto bad; |
| 97 | |
| 98 | ceph_decode_32_safe(p, end, info->dname_len, bad); |
| 99 | ceph_decode_need(p, end, info->dname_len, bad); |
| 100 | info->dname = *p; |
| 101 | *p += info->dname_len; |
| 102 | info->dlease = *p; |
| 103 | *p += sizeof(*info->dlease); |
| 104 | } |
| 105 | |
| 106 | if (info->head->is_target) { |
| 107 | err = parse_reply_info_in(p, end, &info->targeti); |
| 108 | if (err < 0) |
| 109 | goto out_bad; |
| 110 | } |
| 111 | |
| 112 | if (unlikely(*p != end)) |
| 113 | goto bad; |
| 114 | return 0; |
| 115 | |
| 116 | bad: |
| 117 | err = -EIO; |
| 118 | out_bad: |
| 119 | pr_err("problem parsing mds trace %d\n", err); |
| 120 | return err; |
| 121 | } |
| 122 | |
| 123 | /* |
| 124 | * parse readdir results |
| 125 | */ |
| 126 | static int parse_reply_info_dir(void **p, void *end, |
| 127 | struct ceph_mds_reply_info_parsed *info) |
| 128 | { |
| 129 | u32 num, i = 0; |
| 130 | int err; |
| 131 | |
| 132 | info->dir_dir = *p; |
| 133 | if (*p + sizeof(*info->dir_dir) > end) |
| 134 | goto bad; |
| 135 | *p += sizeof(*info->dir_dir) + |
| 136 | sizeof(u32)*le32_to_cpu(info->dir_dir->ndist); |
| 137 | if (*p > end) |
| 138 | goto bad; |
| 139 | |
| 140 | ceph_decode_need(p, end, sizeof(num) + 2, bad); |
Sage Weil | c89136e | 2009-10-14 09:59:09 -0700 | [diff] [blame] | 141 | num = ceph_decode_32(p); |
| 142 | info->dir_end = ceph_decode_8(p); |
| 143 | info->dir_complete = ceph_decode_8(p); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 144 | if (num == 0) |
| 145 | goto done; |
| 146 | |
| 147 | /* alloc large array */ |
| 148 | info->dir_nr = num; |
| 149 | info->dir_in = kcalloc(num, sizeof(*info->dir_in) + |
| 150 | sizeof(*info->dir_dname) + |
| 151 | sizeof(*info->dir_dname_len) + |
| 152 | sizeof(*info->dir_dlease), |
| 153 | GFP_NOFS); |
| 154 | if (info->dir_in == NULL) { |
| 155 | err = -ENOMEM; |
| 156 | goto out_bad; |
| 157 | } |
| 158 | info->dir_dname = (void *)(info->dir_in + num); |
| 159 | info->dir_dname_len = (void *)(info->dir_dname + num); |
| 160 | info->dir_dlease = (void *)(info->dir_dname_len + num); |
| 161 | |
| 162 | while (num) { |
| 163 | /* dentry */ |
| 164 | ceph_decode_need(p, end, sizeof(u32)*2, bad); |
Sage Weil | c89136e | 2009-10-14 09:59:09 -0700 | [diff] [blame] | 165 | info->dir_dname_len[i] = ceph_decode_32(p); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 166 | ceph_decode_need(p, end, info->dir_dname_len[i], bad); |
| 167 | info->dir_dname[i] = *p; |
| 168 | *p += info->dir_dname_len[i]; |
| 169 | dout("parsed dir dname '%.*s'\n", info->dir_dname_len[i], |
| 170 | info->dir_dname[i]); |
| 171 | info->dir_dlease[i] = *p; |
| 172 | *p += sizeof(struct ceph_mds_reply_lease); |
| 173 | |
| 174 | /* inode */ |
| 175 | err = parse_reply_info_in(p, end, &info->dir_in[i]); |
| 176 | if (err < 0) |
| 177 | goto out_bad; |
| 178 | i++; |
| 179 | num--; |
| 180 | } |
| 181 | |
| 182 | done: |
| 183 | if (*p != end) |
| 184 | goto bad; |
| 185 | return 0; |
| 186 | |
| 187 | bad: |
| 188 | err = -EIO; |
| 189 | out_bad: |
| 190 | pr_err("problem parsing dir contents %d\n", err); |
| 191 | return err; |
| 192 | } |
| 193 | |
| 194 | /* |
| 195 | * parse entire mds reply |
| 196 | */ |
| 197 | static int parse_reply_info(struct ceph_msg *msg, |
| 198 | struct ceph_mds_reply_info_parsed *info) |
| 199 | { |
| 200 | void *p, *end; |
| 201 | u32 len; |
| 202 | int err; |
| 203 | |
| 204 | info->head = msg->front.iov_base; |
| 205 | p = msg->front.iov_base + sizeof(struct ceph_mds_reply_head); |
| 206 | end = p + msg->front.iov_len - sizeof(struct ceph_mds_reply_head); |
| 207 | |
| 208 | /* trace */ |
| 209 | ceph_decode_32_safe(&p, end, len, bad); |
| 210 | if (len > 0) { |
| 211 | err = parse_reply_info_trace(&p, p+len, info); |
| 212 | if (err < 0) |
| 213 | goto out_bad; |
| 214 | } |
| 215 | |
| 216 | /* dir content */ |
| 217 | ceph_decode_32_safe(&p, end, len, bad); |
| 218 | if (len > 0) { |
| 219 | err = parse_reply_info_dir(&p, p+len, info); |
| 220 | if (err < 0) |
| 221 | goto out_bad; |
| 222 | } |
| 223 | |
| 224 | /* snap blob */ |
| 225 | ceph_decode_32_safe(&p, end, len, bad); |
| 226 | info->snapblob_len = len; |
| 227 | info->snapblob = p; |
| 228 | p += len; |
| 229 | |
| 230 | if (p != end) |
| 231 | goto bad; |
| 232 | return 0; |
| 233 | |
| 234 | bad: |
| 235 | err = -EIO; |
| 236 | out_bad: |
| 237 | pr_err("mds parse_reply err %d\n", err); |
| 238 | return err; |
| 239 | } |
| 240 | |
| 241 | static void destroy_reply_info(struct ceph_mds_reply_info_parsed *info) |
| 242 | { |
| 243 | kfree(info->dir_in); |
| 244 | } |
| 245 | |
| 246 | |
| 247 | /* |
| 248 | * sessions |
| 249 | */ |
| 250 | static const char *session_state_name(int s) |
| 251 | { |
| 252 | switch (s) { |
| 253 | case CEPH_MDS_SESSION_NEW: return "new"; |
| 254 | case CEPH_MDS_SESSION_OPENING: return "opening"; |
| 255 | case CEPH_MDS_SESSION_OPEN: return "open"; |
| 256 | case CEPH_MDS_SESSION_HUNG: return "hung"; |
| 257 | case CEPH_MDS_SESSION_CLOSING: return "closing"; |
Sage Weil | 44ca18f | 2010-02-15 12:08:46 -0800 | [diff] [blame] | 258 | case CEPH_MDS_SESSION_RESTARTING: return "restarting"; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 259 | case CEPH_MDS_SESSION_RECONNECTING: return "reconnecting"; |
| 260 | default: return "???"; |
| 261 | } |
| 262 | } |
| 263 | |
| 264 | static struct ceph_mds_session *get_session(struct ceph_mds_session *s) |
| 265 | { |
| 266 | if (atomic_inc_not_zero(&s->s_ref)) { |
| 267 | dout("mdsc get_session %p %d -> %d\n", s, |
| 268 | atomic_read(&s->s_ref)-1, atomic_read(&s->s_ref)); |
| 269 | return s; |
| 270 | } else { |
| 271 | dout("mdsc get_session %p 0 -- FAIL", s); |
| 272 | return NULL; |
| 273 | } |
| 274 | } |
| 275 | |
| 276 | void ceph_put_mds_session(struct ceph_mds_session *s) |
| 277 | { |
| 278 | dout("mdsc put_session %p %d -> %d\n", s, |
| 279 | atomic_read(&s->s_ref), atomic_read(&s->s_ref)-1); |
Sage Weil | 4e7a5dc | 2009-11-18 16:19:57 -0800 | [diff] [blame] | 280 | if (atomic_dec_and_test(&s->s_ref)) { |
| 281 | if (s->s_authorizer) |
| 282 | s->s_mdsc->client->monc.auth->ops->destroy_authorizer( |
| 283 | s->s_mdsc->client->monc.auth, s->s_authorizer); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 284 | kfree(s); |
Sage Weil | 4e7a5dc | 2009-11-18 16:19:57 -0800 | [diff] [blame] | 285 | } |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 286 | } |
| 287 | |
| 288 | /* |
| 289 | * called under mdsc->mutex |
| 290 | */ |
| 291 | struct ceph_mds_session *__ceph_lookup_mds_session(struct ceph_mds_client *mdsc, |
| 292 | int mds) |
| 293 | { |
| 294 | struct ceph_mds_session *session; |
| 295 | |
| 296 | if (mds >= mdsc->max_sessions || mdsc->sessions[mds] == NULL) |
| 297 | return NULL; |
| 298 | session = mdsc->sessions[mds]; |
| 299 | dout("lookup_mds_session %p %d\n", session, |
| 300 | atomic_read(&session->s_ref)); |
| 301 | get_session(session); |
| 302 | return session; |
| 303 | } |
| 304 | |
| 305 | static bool __have_session(struct ceph_mds_client *mdsc, int mds) |
| 306 | { |
| 307 | if (mds >= mdsc->max_sessions) |
| 308 | return false; |
| 309 | return mdsc->sessions[mds]; |
| 310 | } |
| 311 | |
Sage Weil | 2600d2d | 2010-02-22 15:12:16 -0800 | [diff] [blame] | 312 | static int __verify_registered_session(struct ceph_mds_client *mdsc, |
| 313 | struct ceph_mds_session *s) |
| 314 | { |
| 315 | if (s->s_mds >= mdsc->max_sessions || |
| 316 | mdsc->sessions[s->s_mds] != s) |
| 317 | return -ENOENT; |
| 318 | return 0; |
| 319 | } |
| 320 | |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 321 | /* |
| 322 | * create+register a new session for given mds. |
| 323 | * called under mdsc->mutex. |
| 324 | */ |
| 325 | static struct ceph_mds_session *register_session(struct ceph_mds_client *mdsc, |
| 326 | int mds) |
| 327 | { |
| 328 | struct ceph_mds_session *s; |
| 329 | |
| 330 | s = kzalloc(sizeof(*s), GFP_NOFS); |
| 331 | s->s_mdsc = mdsc; |
| 332 | s->s_mds = mds; |
| 333 | s->s_state = CEPH_MDS_SESSION_NEW; |
| 334 | s->s_ttl = 0; |
| 335 | s->s_seq = 0; |
| 336 | mutex_init(&s->s_mutex); |
| 337 | |
| 338 | ceph_con_init(mdsc->client->msgr, &s->s_con); |
| 339 | s->s_con.private = s; |
| 340 | s->s_con.ops = &mds_con_ops; |
| 341 | s->s_con.peer_name.type = CEPH_ENTITY_TYPE_MDS; |
| 342 | s->s_con.peer_name.num = cpu_to_le64(mds); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 343 | |
| 344 | spin_lock_init(&s->s_cap_lock); |
| 345 | s->s_cap_gen = 0; |
| 346 | s->s_cap_ttl = 0; |
| 347 | s->s_renew_requested = 0; |
| 348 | s->s_renew_seq = 0; |
| 349 | INIT_LIST_HEAD(&s->s_caps); |
| 350 | s->s_nr_caps = 0; |
Sage Weil | 5dacf09 | 2009-12-21 20:40:34 -0800 | [diff] [blame] | 351 | s->s_trim_caps = 0; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 352 | atomic_set(&s->s_ref, 1); |
| 353 | INIT_LIST_HEAD(&s->s_waiting); |
| 354 | INIT_LIST_HEAD(&s->s_unsafe); |
| 355 | s->s_num_cap_releases = 0; |
Sage Weil | 7c1332b | 2010-02-16 11:39:45 -0800 | [diff] [blame] | 356 | s->s_cap_iterator = NULL; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 357 | INIT_LIST_HEAD(&s->s_cap_releases); |
| 358 | INIT_LIST_HEAD(&s->s_cap_releases_done); |
| 359 | INIT_LIST_HEAD(&s->s_cap_flushing); |
| 360 | INIT_LIST_HEAD(&s->s_cap_snaps_flushing); |
| 361 | |
| 362 | dout("register_session mds%d\n", mds); |
| 363 | if (mds >= mdsc->max_sessions) { |
| 364 | int newmax = 1 << get_count_order(mds+1); |
| 365 | struct ceph_mds_session **sa; |
| 366 | |
| 367 | dout("register_session realloc to %d\n", newmax); |
| 368 | sa = kcalloc(newmax, sizeof(void *), GFP_NOFS); |
| 369 | if (sa == NULL) |
Sage Weil | 42ce56e | 2009-11-18 11:22:36 -0800 | [diff] [blame] | 370 | goto fail_realloc; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 371 | if (mdsc->sessions) { |
| 372 | memcpy(sa, mdsc->sessions, |
| 373 | mdsc->max_sessions * sizeof(void *)); |
| 374 | kfree(mdsc->sessions); |
| 375 | } |
| 376 | mdsc->sessions = sa; |
| 377 | mdsc->max_sessions = newmax; |
| 378 | } |
| 379 | mdsc->sessions[mds] = s; |
| 380 | atomic_inc(&s->s_ref); /* one ref to sessions[], one to caller */ |
Sage Weil | 42ce56e | 2009-11-18 11:22:36 -0800 | [diff] [blame] | 381 | |
| 382 | ceph_con_open(&s->s_con, ceph_mdsmap_get_addr(mdsc->mdsmap, mds)); |
| 383 | |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 384 | return s; |
Sage Weil | 42ce56e | 2009-11-18 11:22:36 -0800 | [diff] [blame] | 385 | |
| 386 | fail_realloc: |
| 387 | kfree(s); |
| 388 | return ERR_PTR(-ENOMEM); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 389 | } |
| 390 | |
| 391 | /* |
| 392 | * called under mdsc->mutex |
| 393 | */ |
Sage Weil | 2600d2d | 2010-02-22 15:12:16 -0800 | [diff] [blame] | 394 | static void __unregister_session(struct ceph_mds_client *mdsc, |
Sage Weil | 42ce56e | 2009-11-18 11:22:36 -0800 | [diff] [blame] | 395 | struct ceph_mds_session *s) |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 396 | { |
Sage Weil | 2600d2d | 2010-02-22 15:12:16 -0800 | [diff] [blame] | 397 | dout("__unregister_session mds%d %p\n", s->s_mds, s); |
| 398 | BUG_ON(mdsc->sessions[s->s_mds] != s); |
Sage Weil | 42ce56e | 2009-11-18 11:22:36 -0800 | [diff] [blame] | 399 | mdsc->sessions[s->s_mds] = NULL; |
| 400 | ceph_con_close(&s->s_con); |
| 401 | ceph_put_mds_session(s); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 402 | } |
| 403 | |
| 404 | /* |
| 405 | * drop session refs in request. |
| 406 | * |
| 407 | * should be last request ref, or hold mdsc->mutex |
| 408 | */ |
| 409 | static void put_request_session(struct ceph_mds_request *req) |
| 410 | { |
| 411 | if (req->r_session) { |
| 412 | ceph_put_mds_session(req->r_session); |
| 413 | req->r_session = NULL; |
| 414 | } |
| 415 | } |
| 416 | |
Sage Weil | 153c8e6 | 2009-12-07 12:31:09 -0800 | [diff] [blame] | 417 | void ceph_mdsc_release_request(struct kref *kref) |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 418 | { |
Sage Weil | 153c8e6 | 2009-12-07 12:31:09 -0800 | [diff] [blame] | 419 | struct ceph_mds_request *req = container_of(kref, |
| 420 | struct ceph_mds_request, |
| 421 | r_kref); |
| 422 | if (req->r_request) |
| 423 | ceph_msg_put(req->r_request); |
| 424 | if (req->r_reply) { |
| 425 | ceph_msg_put(req->r_reply); |
| 426 | destroy_reply_info(&req->r_reply_info); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 427 | } |
Sage Weil | 153c8e6 | 2009-12-07 12:31:09 -0800 | [diff] [blame] | 428 | if (req->r_inode) { |
| 429 | ceph_put_cap_refs(ceph_inode(req->r_inode), |
| 430 | CEPH_CAP_PIN); |
| 431 | iput(req->r_inode); |
| 432 | } |
| 433 | if (req->r_locked_dir) |
| 434 | ceph_put_cap_refs(ceph_inode(req->r_locked_dir), |
| 435 | CEPH_CAP_PIN); |
| 436 | if (req->r_target_inode) |
| 437 | iput(req->r_target_inode); |
| 438 | if (req->r_dentry) |
| 439 | dput(req->r_dentry); |
| 440 | if (req->r_old_dentry) { |
| 441 | ceph_put_cap_refs( |
| 442 | ceph_inode(req->r_old_dentry->d_parent->d_inode), |
| 443 | CEPH_CAP_PIN); |
| 444 | dput(req->r_old_dentry); |
| 445 | } |
| 446 | kfree(req->r_path1); |
| 447 | kfree(req->r_path2); |
| 448 | put_request_session(req); |
| 449 | ceph_unreserve_caps(&req->r_caps_reservation); |
| 450 | kfree(req); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 451 | } |
| 452 | |
| 453 | /* |
| 454 | * lookup session, bump ref if found. |
| 455 | * |
| 456 | * called under mdsc->mutex. |
| 457 | */ |
| 458 | static struct ceph_mds_request *__lookup_request(struct ceph_mds_client *mdsc, |
| 459 | u64 tid) |
| 460 | { |
| 461 | struct ceph_mds_request *req; |
Sage Weil | 44ca18f | 2010-02-15 12:08:46 -0800 | [diff] [blame] | 462 | struct rb_node *n = mdsc->request_tree.rb_node; |
| 463 | |
| 464 | while (n) { |
| 465 | req = rb_entry(n, struct ceph_mds_request, r_node); |
| 466 | if (tid < req->r_tid) |
| 467 | n = n->rb_left; |
| 468 | else if (tid > req->r_tid) |
| 469 | n = n->rb_right; |
| 470 | else { |
| 471 | ceph_mdsc_get_request(req); |
| 472 | return req; |
| 473 | } |
| 474 | } |
| 475 | return NULL; |
| 476 | } |
| 477 | |
| 478 | static void __insert_request(struct ceph_mds_client *mdsc, |
| 479 | struct ceph_mds_request *new) |
| 480 | { |
| 481 | struct rb_node **p = &mdsc->request_tree.rb_node; |
| 482 | struct rb_node *parent = NULL; |
| 483 | struct ceph_mds_request *req = NULL; |
| 484 | |
| 485 | while (*p) { |
| 486 | parent = *p; |
| 487 | req = rb_entry(parent, struct ceph_mds_request, r_node); |
| 488 | if (new->r_tid < req->r_tid) |
| 489 | p = &(*p)->rb_left; |
| 490 | else if (new->r_tid > req->r_tid) |
| 491 | p = &(*p)->rb_right; |
| 492 | else |
| 493 | BUG(); |
| 494 | } |
| 495 | |
| 496 | rb_link_node(&new->r_node, parent, p); |
| 497 | rb_insert_color(&new->r_node, &mdsc->request_tree); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 498 | } |
| 499 | |
| 500 | /* |
| 501 | * Register an in-flight request, and assign a tid. Link to directory |
| 502 | * are modifying (if any). |
| 503 | * |
| 504 | * Called under mdsc->mutex. |
| 505 | */ |
| 506 | static void __register_request(struct ceph_mds_client *mdsc, |
| 507 | struct ceph_mds_request *req, |
| 508 | struct inode *dir) |
| 509 | { |
| 510 | req->r_tid = ++mdsc->last_tid; |
| 511 | if (req->r_num_caps) |
| 512 | ceph_reserve_caps(&req->r_caps_reservation, req->r_num_caps); |
| 513 | dout("__register_request %p tid %lld\n", req, req->r_tid); |
| 514 | ceph_mdsc_get_request(req); |
Sage Weil | 44ca18f | 2010-02-15 12:08:46 -0800 | [diff] [blame] | 515 | __insert_request(mdsc, req); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 516 | |
| 517 | if (dir) { |
| 518 | struct ceph_inode_info *ci = ceph_inode(dir); |
| 519 | |
| 520 | spin_lock(&ci->i_unsafe_lock); |
| 521 | req->r_unsafe_dir = dir; |
| 522 | list_add_tail(&req->r_unsafe_dir_item, &ci->i_unsafe_dirops); |
| 523 | spin_unlock(&ci->i_unsafe_lock); |
| 524 | } |
| 525 | } |
| 526 | |
| 527 | static void __unregister_request(struct ceph_mds_client *mdsc, |
| 528 | struct ceph_mds_request *req) |
| 529 | { |
| 530 | dout("__unregister_request %p tid %lld\n", req, req->r_tid); |
Sage Weil | 44ca18f | 2010-02-15 12:08:46 -0800 | [diff] [blame] | 531 | rb_erase(&req->r_node, &mdsc->request_tree); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 532 | ceph_mdsc_put_request(req); |
| 533 | |
| 534 | if (req->r_unsafe_dir) { |
| 535 | struct ceph_inode_info *ci = ceph_inode(req->r_unsafe_dir); |
| 536 | |
| 537 | spin_lock(&ci->i_unsafe_lock); |
| 538 | list_del_init(&req->r_unsafe_dir_item); |
| 539 | spin_unlock(&ci->i_unsafe_lock); |
| 540 | } |
| 541 | } |
| 542 | |
| 543 | /* |
| 544 | * Choose mds to send request to next. If there is a hint set in the |
| 545 | * request (e.g., due to a prior forward hint from the mds), use that. |
| 546 | * Otherwise, consult frag tree and/or caps to identify the |
| 547 | * appropriate mds. If all else fails, choose randomly. |
| 548 | * |
| 549 | * Called under mdsc->mutex. |
| 550 | */ |
| 551 | static int __choose_mds(struct ceph_mds_client *mdsc, |
| 552 | struct ceph_mds_request *req) |
| 553 | { |
| 554 | struct inode *inode; |
| 555 | struct ceph_inode_info *ci; |
| 556 | struct ceph_cap *cap; |
| 557 | int mode = req->r_direct_mode; |
| 558 | int mds = -1; |
| 559 | u32 hash = req->r_direct_hash; |
| 560 | bool is_hash = req->r_direct_is_hash; |
| 561 | |
| 562 | /* |
| 563 | * is there a specific mds we should try? ignore hint if we have |
| 564 | * no session and the mds is not up (active or recovering). |
| 565 | */ |
| 566 | if (req->r_resend_mds >= 0 && |
| 567 | (__have_session(mdsc, req->r_resend_mds) || |
| 568 | ceph_mdsmap_get_state(mdsc->mdsmap, req->r_resend_mds) > 0)) { |
| 569 | dout("choose_mds using resend_mds mds%d\n", |
| 570 | req->r_resend_mds); |
| 571 | return req->r_resend_mds; |
| 572 | } |
| 573 | |
| 574 | if (mode == USE_RANDOM_MDS) |
| 575 | goto random; |
| 576 | |
| 577 | inode = NULL; |
| 578 | if (req->r_inode) { |
| 579 | inode = req->r_inode; |
| 580 | } else if (req->r_dentry) { |
| 581 | if (req->r_dentry->d_inode) { |
| 582 | inode = req->r_dentry->d_inode; |
| 583 | } else { |
| 584 | inode = req->r_dentry->d_parent->d_inode; |
| 585 | hash = req->r_dentry->d_name.hash; |
| 586 | is_hash = true; |
| 587 | } |
| 588 | } |
| 589 | dout("__choose_mds %p is_hash=%d (%d) mode %d\n", inode, (int)is_hash, |
| 590 | (int)hash, mode); |
| 591 | if (!inode) |
| 592 | goto random; |
| 593 | ci = ceph_inode(inode); |
| 594 | |
| 595 | if (is_hash && S_ISDIR(inode->i_mode)) { |
| 596 | struct ceph_inode_frag frag; |
| 597 | int found; |
| 598 | |
| 599 | ceph_choose_frag(ci, hash, &frag, &found); |
| 600 | if (found) { |
| 601 | if (mode == USE_ANY_MDS && frag.ndist > 0) { |
| 602 | u8 r; |
| 603 | |
| 604 | /* choose a random replica */ |
| 605 | get_random_bytes(&r, 1); |
| 606 | r %= frag.ndist; |
| 607 | mds = frag.dist[r]; |
| 608 | dout("choose_mds %p %llx.%llx " |
| 609 | "frag %u mds%d (%d/%d)\n", |
| 610 | inode, ceph_vinop(inode), |
| 611 | frag.frag, frag.mds, |
| 612 | (int)r, frag.ndist); |
| 613 | return mds; |
| 614 | } |
| 615 | |
| 616 | /* since this file/dir wasn't known to be |
| 617 | * replicated, then we want to look for the |
| 618 | * authoritative mds. */ |
| 619 | mode = USE_AUTH_MDS; |
| 620 | if (frag.mds >= 0) { |
| 621 | /* choose auth mds */ |
| 622 | mds = frag.mds; |
| 623 | dout("choose_mds %p %llx.%llx " |
| 624 | "frag %u mds%d (auth)\n", |
| 625 | inode, ceph_vinop(inode), frag.frag, mds); |
| 626 | return mds; |
| 627 | } |
| 628 | } |
| 629 | } |
| 630 | |
| 631 | spin_lock(&inode->i_lock); |
| 632 | cap = NULL; |
| 633 | if (mode == USE_AUTH_MDS) |
| 634 | cap = ci->i_auth_cap; |
| 635 | if (!cap && !RB_EMPTY_ROOT(&ci->i_caps)) |
| 636 | cap = rb_entry(rb_first(&ci->i_caps), struct ceph_cap, ci_node); |
| 637 | if (!cap) { |
| 638 | spin_unlock(&inode->i_lock); |
| 639 | goto random; |
| 640 | } |
| 641 | mds = cap->session->s_mds; |
| 642 | dout("choose_mds %p %llx.%llx mds%d (%scap %p)\n", |
| 643 | inode, ceph_vinop(inode), mds, |
| 644 | cap == ci->i_auth_cap ? "auth " : "", cap); |
| 645 | spin_unlock(&inode->i_lock); |
| 646 | return mds; |
| 647 | |
| 648 | random: |
| 649 | mds = ceph_mdsmap_get_random_mds(mdsc->mdsmap); |
| 650 | dout("choose_mds chose random mds%d\n", mds); |
| 651 | return mds; |
| 652 | } |
| 653 | |
| 654 | |
| 655 | /* |
| 656 | * session messages |
| 657 | */ |
| 658 | static struct ceph_msg *create_session_msg(u32 op, u64 seq) |
| 659 | { |
| 660 | struct ceph_msg *msg; |
| 661 | struct ceph_mds_session_head *h; |
| 662 | |
| 663 | msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h), 0, 0, NULL); |
| 664 | if (IS_ERR(msg)) { |
| 665 | pr_err("create_session_msg ENOMEM creating msg\n"); |
| 666 | return ERR_PTR(PTR_ERR(msg)); |
| 667 | } |
| 668 | h = msg->front.iov_base; |
| 669 | h->op = cpu_to_le32(op); |
| 670 | h->seq = cpu_to_le64(seq); |
| 671 | return msg; |
| 672 | } |
| 673 | |
| 674 | /* |
| 675 | * send session open request. |
| 676 | * |
| 677 | * called under mdsc->mutex |
| 678 | */ |
| 679 | static int __open_session(struct ceph_mds_client *mdsc, |
| 680 | struct ceph_mds_session *session) |
| 681 | { |
| 682 | struct ceph_msg *msg; |
| 683 | int mstate; |
| 684 | int mds = session->s_mds; |
| 685 | int err = 0; |
| 686 | |
| 687 | /* wait for mds to go active? */ |
| 688 | mstate = ceph_mdsmap_get_state(mdsc->mdsmap, mds); |
| 689 | dout("open_session to mds%d (%s)\n", mds, |
| 690 | ceph_mds_state_name(mstate)); |
| 691 | session->s_state = CEPH_MDS_SESSION_OPENING; |
| 692 | session->s_renew_requested = jiffies; |
| 693 | |
| 694 | /* send connect message */ |
| 695 | msg = create_session_msg(CEPH_SESSION_REQUEST_OPEN, session->s_seq); |
| 696 | if (IS_ERR(msg)) { |
| 697 | err = PTR_ERR(msg); |
| 698 | goto out; |
| 699 | } |
| 700 | ceph_con_send(&session->s_con, msg); |
| 701 | |
| 702 | out: |
| 703 | return 0; |
| 704 | } |
| 705 | |
| 706 | /* |
| 707 | * session caps |
| 708 | */ |
| 709 | |
| 710 | /* |
| 711 | * Free preallocated cap messages assigned to this session |
| 712 | */ |
| 713 | static void cleanup_cap_releases(struct ceph_mds_session *session) |
| 714 | { |
| 715 | struct ceph_msg *msg; |
| 716 | |
| 717 | spin_lock(&session->s_cap_lock); |
| 718 | while (!list_empty(&session->s_cap_releases)) { |
| 719 | msg = list_first_entry(&session->s_cap_releases, |
| 720 | struct ceph_msg, list_head); |
| 721 | list_del_init(&msg->list_head); |
| 722 | ceph_msg_put(msg); |
| 723 | } |
| 724 | while (!list_empty(&session->s_cap_releases_done)) { |
| 725 | msg = list_first_entry(&session->s_cap_releases_done, |
| 726 | struct ceph_msg, list_head); |
| 727 | list_del_init(&msg->list_head); |
| 728 | ceph_msg_put(msg); |
| 729 | } |
| 730 | spin_unlock(&session->s_cap_lock); |
| 731 | } |
| 732 | |
| 733 | /* |
| 734 | * Helper to safely iterate over all caps associated with a session. |
| 735 | * |
| 736 | * caller must hold session s_mutex |
| 737 | */ |
| 738 | static int iterate_session_caps(struct ceph_mds_session *session, |
| 739 | int (*cb)(struct inode *, struct ceph_cap *, |
| 740 | void *), void *arg) |
| 741 | { |
Sage Weil | 7c1332b | 2010-02-16 11:39:45 -0800 | [diff] [blame] | 742 | struct list_head *p; |
| 743 | struct ceph_cap *cap; |
| 744 | struct inode *inode, *last_inode = NULL; |
| 745 | struct ceph_cap *old_cap = NULL; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 746 | int ret; |
| 747 | |
| 748 | dout("iterate_session_caps %p mds%d\n", session, session->s_mds); |
| 749 | spin_lock(&session->s_cap_lock); |
Sage Weil | 7c1332b | 2010-02-16 11:39:45 -0800 | [diff] [blame] | 750 | p = session->s_caps.next; |
| 751 | while (p != &session->s_caps) { |
| 752 | cap = list_entry(p, struct ceph_cap, session_caps); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 753 | inode = igrab(&cap->ci->vfs_inode); |
Sage Weil | 7c1332b | 2010-02-16 11:39:45 -0800 | [diff] [blame] | 754 | if (!inode) { |
| 755 | p = p->next; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 756 | continue; |
Sage Weil | 7c1332b | 2010-02-16 11:39:45 -0800 | [diff] [blame] | 757 | } |
| 758 | session->s_cap_iterator = cap; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 759 | spin_unlock(&session->s_cap_lock); |
Sage Weil | 7c1332b | 2010-02-16 11:39:45 -0800 | [diff] [blame] | 760 | |
| 761 | if (last_inode) { |
| 762 | iput(last_inode); |
| 763 | last_inode = NULL; |
| 764 | } |
| 765 | if (old_cap) { |
| 766 | ceph_put_cap(old_cap); |
| 767 | old_cap = NULL; |
| 768 | } |
| 769 | |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 770 | ret = cb(inode, cap, arg); |
Sage Weil | 7c1332b | 2010-02-16 11:39:45 -0800 | [diff] [blame] | 771 | last_inode = inode; |
| 772 | |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 773 | spin_lock(&session->s_cap_lock); |
Sage Weil | 7c1332b | 2010-02-16 11:39:45 -0800 | [diff] [blame] | 774 | p = p->next; |
| 775 | if (cap->ci == NULL) { |
| 776 | dout("iterate_session_caps finishing cap %p removal\n", |
| 777 | cap); |
| 778 | BUG_ON(cap->session != session); |
| 779 | list_del_init(&cap->session_caps); |
| 780 | session->s_nr_caps--; |
| 781 | cap->session = NULL; |
| 782 | old_cap = cap; /* put_cap it w/o locks held */ |
| 783 | } |
Sage Weil | 5dacf09 | 2009-12-21 20:40:34 -0800 | [diff] [blame] | 784 | if (ret < 0) |
| 785 | goto out; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 786 | } |
Sage Weil | 5dacf09 | 2009-12-21 20:40:34 -0800 | [diff] [blame] | 787 | ret = 0; |
| 788 | out: |
Sage Weil | 7c1332b | 2010-02-16 11:39:45 -0800 | [diff] [blame] | 789 | session->s_cap_iterator = NULL; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 790 | spin_unlock(&session->s_cap_lock); |
Sage Weil | 7c1332b | 2010-02-16 11:39:45 -0800 | [diff] [blame] | 791 | |
| 792 | if (last_inode) |
| 793 | iput(last_inode); |
| 794 | if (old_cap) |
| 795 | ceph_put_cap(old_cap); |
| 796 | |
Sage Weil | 5dacf09 | 2009-12-21 20:40:34 -0800 | [diff] [blame] | 797 | return ret; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 798 | } |
| 799 | |
| 800 | static int remove_session_caps_cb(struct inode *inode, struct ceph_cap *cap, |
| 801 | void *arg) |
| 802 | { |
| 803 | struct ceph_inode_info *ci = ceph_inode(inode); |
| 804 | dout("removing cap %p, ci is %p, inode is %p\n", |
| 805 | cap, ci, &ci->vfs_inode); |
| 806 | ceph_remove_cap(cap); |
| 807 | return 0; |
| 808 | } |
| 809 | |
| 810 | /* |
| 811 | * caller must hold session s_mutex |
| 812 | */ |
| 813 | static void remove_session_caps(struct ceph_mds_session *session) |
| 814 | { |
| 815 | dout("remove_session_caps on %p\n", session); |
| 816 | iterate_session_caps(session, remove_session_caps_cb, NULL); |
| 817 | BUG_ON(session->s_nr_caps > 0); |
| 818 | cleanup_cap_releases(session); |
| 819 | } |
| 820 | |
| 821 | /* |
| 822 | * wake up any threads waiting on this session's caps. if the cap is |
| 823 | * old (didn't get renewed on the client reconnect), remove it now. |
| 824 | * |
| 825 | * caller must hold s_mutex. |
| 826 | */ |
| 827 | static int wake_up_session_cb(struct inode *inode, struct ceph_cap *cap, |
| 828 | void *arg) |
| 829 | { |
Sage Weil | 0dc2570 | 2009-11-20 13:43:45 -0800 | [diff] [blame] | 830 | struct ceph_inode_info *ci = ceph_inode(inode); |
| 831 | |
| 832 | wake_up(&ci->i_cap_wq); |
| 833 | if (arg) { |
| 834 | spin_lock(&inode->i_lock); |
| 835 | ci->i_wanted_max_size = 0; |
| 836 | ci->i_requested_max_size = 0; |
| 837 | spin_unlock(&inode->i_lock); |
| 838 | } |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 839 | return 0; |
| 840 | } |
| 841 | |
Sage Weil | 0dc2570 | 2009-11-20 13:43:45 -0800 | [diff] [blame] | 842 | static void wake_up_session_caps(struct ceph_mds_session *session, |
| 843 | int reconnect) |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 844 | { |
| 845 | dout("wake_up_session_caps %p mds%d\n", session, session->s_mds); |
Sage Weil | 0dc2570 | 2009-11-20 13:43:45 -0800 | [diff] [blame] | 846 | iterate_session_caps(session, wake_up_session_cb, |
| 847 | (void *)(unsigned long)reconnect); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 848 | } |
| 849 | |
| 850 | /* |
| 851 | * Send periodic message to MDS renewing all currently held caps. The |
| 852 | * ack will reset the expiration for all caps from this session. |
| 853 | * |
| 854 | * caller holds s_mutex |
| 855 | */ |
| 856 | static int send_renew_caps(struct ceph_mds_client *mdsc, |
| 857 | struct ceph_mds_session *session) |
| 858 | { |
| 859 | struct ceph_msg *msg; |
| 860 | int state; |
| 861 | |
| 862 | if (time_after_eq(jiffies, session->s_cap_ttl) && |
| 863 | time_after_eq(session->s_cap_ttl, session->s_renew_requested)) |
| 864 | pr_info("mds%d caps stale\n", session->s_mds); |
| 865 | |
| 866 | /* do not try to renew caps until a recovering mds has reconnected |
| 867 | * with its clients. */ |
| 868 | state = ceph_mdsmap_get_state(mdsc->mdsmap, session->s_mds); |
| 869 | if (state < CEPH_MDS_STATE_RECONNECT) { |
| 870 | dout("send_renew_caps ignoring mds%d (%s)\n", |
| 871 | session->s_mds, ceph_mds_state_name(state)); |
| 872 | return 0; |
| 873 | } |
| 874 | |
| 875 | dout("send_renew_caps to mds%d (%s)\n", session->s_mds, |
| 876 | ceph_mds_state_name(state)); |
| 877 | session->s_renew_requested = jiffies; |
| 878 | msg = create_session_msg(CEPH_SESSION_REQUEST_RENEWCAPS, |
| 879 | ++session->s_renew_seq); |
| 880 | if (IS_ERR(msg)) |
| 881 | return PTR_ERR(msg); |
| 882 | ceph_con_send(&session->s_con, msg); |
| 883 | return 0; |
| 884 | } |
| 885 | |
| 886 | /* |
| 887 | * Note new cap ttl, and any transition from stale -> not stale (fresh?). |
Sage Weil | 0dc2570 | 2009-11-20 13:43:45 -0800 | [diff] [blame] | 888 | * |
| 889 | * Called under session->s_mutex |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 890 | */ |
| 891 | static void renewed_caps(struct ceph_mds_client *mdsc, |
| 892 | struct ceph_mds_session *session, int is_renew) |
| 893 | { |
| 894 | int was_stale; |
| 895 | int wake = 0; |
| 896 | |
| 897 | spin_lock(&session->s_cap_lock); |
| 898 | was_stale = is_renew && (session->s_cap_ttl == 0 || |
| 899 | time_after_eq(jiffies, session->s_cap_ttl)); |
| 900 | |
| 901 | session->s_cap_ttl = session->s_renew_requested + |
| 902 | mdsc->mdsmap->m_session_timeout*HZ; |
| 903 | |
| 904 | if (was_stale) { |
| 905 | if (time_before(jiffies, session->s_cap_ttl)) { |
| 906 | pr_info("mds%d caps renewed\n", session->s_mds); |
| 907 | wake = 1; |
| 908 | } else { |
| 909 | pr_info("mds%d caps still stale\n", session->s_mds); |
| 910 | } |
| 911 | } |
| 912 | dout("renewed_caps mds%d ttl now %lu, was %s, now %s\n", |
| 913 | session->s_mds, session->s_cap_ttl, was_stale ? "stale" : "fresh", |
| 914 | time_before(jiffies, session->s_cap_ttl) ? "stale" : "fresh"); |
| 915 | spin_unlock(&session->s_cap_lock); |
| 916 | |
| 917 | if (wake) |
Sage Weil | 0dc2570 | 2009-11-20 13:43:45 -0800 | [diff] [blame] | 918 | wake_up_session_caps(session, 0); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 919 | } |
| 920 | |
| 921 | /* |
| 922 | * send a session close request |
| 923 | */ |
| 924 | static int request_close_session(struct ceph_mds_client *mdsc, |
| 925 | struct ceph_mds_session *session) |
| 926 | { |
| 927 | struct ceph_msg *msg; |
| 928 | int err = 0; |
| 929 | |
| 930 | dout("request_close_session mds%d state %s seq %lld\n", |
| 931 | session->s_mds, session_state_name(session->s_state), |
| 932 | session->s_seq); |
| 933 | msg = create_session_msg(CEPH_SESSION_REQUEST_CLOSE, session->s_seq); |
| 934 | if (IS_ERR(msg)) |
| 935 | err = PTR_ERR(msg); |
| 936 | else |
| 937 | ceph_con_send(&session->s_con, msg); |
| 938 | return err; |
| 939 | } |
| 940 | |
| 941 | /* |
| 942 | * Called with s_mutex held. |
| 943 | */ |
| 944 | static int __close_session(struct ceph_mds_client *mdsc, |
| 945 | struct ceph_mds_session *session) |
| 946 | { |
| 947 | if (session->s_state >= CEPH_MDS_SESSION_CLOSING) |
| 948 | return 0; |
| 949 | session->s_state = CEPH_MDS_SESSION_CLOSING; |
| 950 | return request_close_session(mdsc, session); |
| 951 | } |
| 952 | |
| 953 | /* |
| 954 | * Trim old(er) caps. |
| 955 | * |
| 956 | * Because we can't cache an inode without one or more caps, we do |
| 957 | * this indirectly: if a cap is unused, we prune its aliases, at which |
| 958 | * point the inode will hopefully get dropped to. |
| 959 | * |
| 960 | * Yes, this is a bit sloppy. Our only real goal here is to respond to |
| 961 | * memory pressure from the MDS, though, so it needn't be perfect. |
| 962 | */ |
| 963 | static int trim_caps_cb(struct inode *inode, struct ceph_cap *cap, void *arg) |
| 964 | { |
| 965 | struct ceph_mds_session *session = arg; |
| 966 | struct ceph_inode_info *ci = ceph_inode(inode); |
| 967 | int used, oissued, mine; |
| 968 | |
| 969 | if (session->s_trim_caps <= 0) |
| 970 | return -1; |
| 971 | |
| 972 | spin_lock(&inode->i_lock); |
| 973 | mine = cap->issued | cap->implemented; |
| 974 | used = __ceph_caps_used(ci); |
| 975 | oissued = __ceph_caps_issued_other(ci, cap); |
| 976 | |
| 977 | dout("trim_caps_cb %p cap %p mine %s oissued %s used %s\n", |
| 978 | inode, cap, ceph_cap_string(mine), ceph_cap_string(oissued), |
| 979 | ceph_cap_string(used)); |
| 980 | if (ci->i_dirty_caps) |
| 981 | goto out; /* dirty caps */ |
| 982 | if ((used & ~oissued) & mine) |
| 983 | goto out; /* we need these caps */ |
| 984 | |
| 985 | session->s_trim_caps--; |
| 986 | if (oissued) { |
| 987 | /* we aren't the only cap.. just remove us */ |
Sage Weil | 7c1332b | 2010-02-16 11:39:45 -0800 | [diff] [blame] | 988 | __ceph_remove_cap(cap); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 989 | } else { |
| 990 | /* try to drop referring dentries */ |
| 991 | spin_unlock(&inode->i_lock); |
| 992 | d_prune_aliases(inode); |
| 993 | dout("trim_caps_cb %p cap %p pruned, count now %d\n", |
| 994 | inode, cap, atomic_read(&inode->i_count)); |
| 995 | return 0; |
| 996 | } |
| 997 | |
| 998 | out: |
| 999 | spin_unlock(&inode->i_lock); |
| 1000 | return 0; |
| 1001 | } |
| 1002 | |
| 1003 | /* |
| 1004 | * Trim session cap count down to some max number. |
| 1005 | */ |
| 1006 | static int trim_caps(struct ceph_mds_client *mdsc, |
| 1007 | struct ceph_mds_session *session, |
| 1008 | int max_caps) |
| 1009 | { |
| 1010 | int trim_caps = session->s_nr_caps - max_caps; |
| 1011 | |
| 1012 | dout("trim_caps mds%d start: %d / %d, trim %d\n", |
| 1013 | session->s_mds, session->s_nr_caps, max_caps, trim_caps); |
| 1014 | if (trim_caps > 0) { |
| 1015 | session->s_trim_caps = trim_caps; |
| 1016 | iterate_session_caps(session, trim_caps_cb, session); |
| 1017 | dout("trim_caps mds%d done: %d / %d, trimmed %d\n", |
| 1018 | session->s_mds, session->s_nr_caps, max_caps, |
| 1019 | trim_caps - session->s_trim_caps); |
Sage Weil | 5dacf09 | 2009-12-21 20:40:34 -0800 | [diff] [blame] | 1020 | session->s_trim_caps = 0; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 1021 | } |
| 1022 | return 0; |
| 1023 | } |
| 1024 | |
| 1025 | /* |
| 1026 | * Allocate cap_release messages. If there is a partially full message |
| 1027 | * in the queue, try to allocate enough to cover it's remainder, so that |
| 1028 | * we can send it immediately. |
| 1029 | * |
| 1030 | * Called under s_mutex. |
| 1031 | */ |
| 1032 | static int add_cap_releases(struct ceph_mds_client *mdsc, |
| 1033 | struct ceph_mds_session *session, |
| 1034 | int extra) |
| 1035 | { |
| 1036 | struct ceph_msg *msg; |
| 1037 | struct ceph_mds_cap_release *head; |
| 1038 | int err = -ENOMEM; |
| 1039 | |
| 1040 | if (extra < 0) |
Sage Weil | 6b80518 | 2009-10-27 11:50:50 -0700 | [diff] [blame] | 1041 | extra = mdsc->client->mount_args->cap_release_safety; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 1042 | |
| 1043 | spin_lock(&session->s_cap_lock); |
| 1044 | |
| 1045 | if (!list_empty(&session->s_cap_releases)) { |
| 1046 | msg = list_first_entry(&session->s_cap_releases, |
| 1047 | struct ceph_msg, |
| 1048 | list_head); |
| 1049 | head = msg->front.iov_base; |
| 1050 | extra += CEPH_CAPS_PER_RELEASE - le32_to_cpu(head->num); |
| 1051 | } |
| 1052 | |
| 1053 | while (session->s_num_cap_releases < session->s_nr_caps + extra) { |
| 1054 | spin_unlock(&session->s_cap_lock); |
| 1055 | msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPRELEASE, PAGE_CACHE_SIZE, |
| 1056 | 0, 0, NULL); |
| 1057 | if (!msg) |
| 1058 | goto out_unlocked; |
| 1059 | dout("add_cap_releases %p msg %p now %d\n", session, msg, |
| 1060 | (int)msg->front.iov_len); |
| 1061 | head = msg->front.iov_base; |
| 1062 | head->num = cpu_to_le32(0); |
| 1063 | msg->front.iov_len = sizeof(*head); |
| 1064 | spin_lock(&session->s_cap_lock); |
| 1065 | list_add(&msg->list_head, &session->s_cap_releases); |
| 1066 | session->s_num_cap_releases += CEPH_CAPS_PER_RELEASE; |
| 1067 | } |
| 1068 | |
| 1069 | if (!list_empty(&session->s_cap_releases)) { |
| 1070 | msg = list_first_entry(&session->s_cap_releases, |
| 1071 | struct ceph_msg, |
| 1072 | list_head); |
| 1073 | head = msg->front.iov_base; |
| 1074 | if (head->num) { |
| 1075 | dout(" queueing non-full %p (%d)\n", msg, |
| 1076 | le32_to_cpu(head->num)); |
| 1077 | list_move_tail(&msg->list_head, |
| 1078 | &session->s_cap_releases_done); |
| 1079 | session->s_num_cap_releases -= |
| 1080 | CEPH_CAPS_PER_RELEASE - le32_to_cpu(head->num); |
| 1081 | } |
| 1082 | } |
| 1083 | err = 0; |
| 1084 | spin_unlock(&session->s_cap_lock); |
| 1085 | out_unlocked: |
| 1086 | return err; |
| 1087 | } |
| 1088 | |
| 1089 | /* |
| 1090 | * flush all dirty inode data to disk. |
| 1091 | * |
| 1092 | * returns true if we've flushed through want_flush_seq |
| 1093 | */ |
| 1094 | static int check_cap_flush(struct ceph_mds_client *mdsc, u64 want_flush_seq) |
| 1095 | { |
| 1096 | int mds, ret = 1; |
| 1097 | |
| 1098 | dout("check_cap_flush want %lld\n", want_flush_seq); |
| 1099 | mutex_lock(&mdsc->mutex); |
| 1100 | for (mds = 0; ret && mds < mdsc->max_sessions; mds++) { |
| 1101 | struct ceph_mds_session *session = mdsc->sessions[mds]; |
| 1102 | |
| 1103 | if (!session) |
| 1104 | continue; |
| 1105 | get_session(session); |
| 1106 | mutex_unlock(&mdsc->mutex); |
| 1107 | |
| 1108 | mutex_lock(&session->s_mutex); |
| 1109 | if (!list_empty(&session->s_cap_flushing)) { |
| 1110 | struct ceph_inode_info *ci = |
| 1111 | list_entry(session->s_cap_flushing.next, |
| 1112 | struct ceph_inode_info, |
| 1113 | i_flushing_item); |
| 1114 | struct inode *inode = &ci->vfs_inode; |
| 1115 | |
| 1116 | spin_lock(&inode->i_lock); |
| 1117 | if (ci->i_cap_flush_seq <= want_flush_seq) { |
| 1118 | dout("check_cap_flush still flushing %p " |
| 1119 | "seq %lld <= %lld to mds%d\n", inode, |
| 1120 | ci->i_cap_flush_seq, want_flush_seq, |
| 1121 | session->s_mds); |
| 1122 | ret = 0; |
| 1123 | } |
| 1124 | spin_unlock(&inode->i_lock); |
| 1125 | } |
| 1126 | mutex_unlock(&session->s_mutex); |
| 1127 | ceph_put_mds_session(session); |
| 1128 | |
| 1129 | if (!ret) |
| 1130 | return ret; |
| 1131 | mutex_lock(&mdsc->mutex); |
| 1132 | } |
| 1133 | |
| 1134 | mutex_unlock(&mdsc->mutex); |
| 1135 | dout("check_cap_flush ok, flushed thru %lld\n", want_flush_seq); |
| 1136 | return ret; |
| 1137 | } |
| 1138 | |
| 1139 | /* |
| 1140 | * called under s_mutex |
| 1141 | */ |
| 1142 | static void send_cap_releases(struct ceph_mds_client *mdsc, |
| 1143 | struct ceph_mds_session *session) |
| 1144 | { |
| 1145 | struct ceph_msg *msg; |
| 1146 | |
| 1147 | dout("send_cap_releases mds%d\n", session->s_mds); |
| 1148 | while (1) { |
| 1149 | spin_lock(&session->s_cap_lock); |
| 1150 | if (list_empty(&session->s_cap_releases_done)) |
| 1151 | break; |
| 1152 | msg = list_first_entry(&session->s_cap_releases_done, |
| 1153 | struct ceph_msg, list_head); |
| 1154 | list_del_init(&msg->list_head); |
| 1155 | spin_unlock(&session->s_cap_lock); |
| 1156 | msg->hdr.front_len = cpu_to_le32(msg->front.iov_len); |
| 1157 | dout("send_cap_releases mds%d %p\n", session->s_mds, msg); |
| 1158 | ceph_con_send(&session->s_con, msg); |
| 1159 | } |
| 1160 | spin_unlock(&session->s_cap_lock); |
| 1161 | } |
| 1162 | |
| 1163 | /* |
| 1164 | * requests |
| 1165 | */ |
| 1166 | |
| 1167 | /* |
| 1168 | * Create an mds request. |
| 1169 | */ |
| 1170 | struct ceph_mds_request * |
| 1171 | ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode) |
| 1172 | { |
| 1173 | struct ceph_mds_request *req = kzalloc(sizeof(*req), GFP_NOFS); |
| 1174 | |
| 1175 | if (!req) |
| 1176 | return ERR_PTR(-ENOMEM); |
| 1177 | |
| 1178 | req->r_started = jiffies; |
| 1179 | req->r_resend_mds = -1; |
| 1180 | INIT_LIST_HEAD(&req->r_unsafe_dir_item); |
| 1181 | req->r_fmode = -1; |
Sage Weil | 153c8e6 | 2009-12-07 12:31:09 -0800 | [diff] [blame] | 1182 | kref_init(&req->r_kref); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 1183 | INIT_LIST_HEAD(&req->r_wait); |
| 1184 | init_completion(&req->r_completion); |
| 1185 | init_completion(&req->r_safe_completion); |
| 1186 | INIT_LIST_HEAD(&req->r_unsafe_item); |
| 1187 | |
| 1188 | req->r_op = op; |
| 1189 | req->r_direct_mode = mode; |
| 1190 | return req; |
| 1191 | } |
| 1192 | |
| 1193 | /* |
Sage Weil | 44ca18f | 2010-02-15 12:08:46 -0800 | [diff] [blame] | 1194 | * return oldest (lowest) request, tid in request tree, 0 if none. |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 1195 | * |
| 1196 | * called under mdsc->mutex. |
| 1197 | */ |
Sage Weil | 44ca18f | 2010-02-15 12:08:46 -0800 | [diff] [blame] | 1198 | static struct ceph_mds_request *__get_oldest_req(struct ceph_mds_client *mdsc) |
| 1199 | { |
| 1200 | if (RB_EMPTY_ROOT(&mdsc->request_tree)) |
| 1201 | return NULL; |
| 1202 | return rb_entry(rb_first(&mdsc->request_tree), |
| 1203 | struct ceph_mds_request, r_node); |
| 1204 | } |
| 1205 | |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 1206 | static u64 __get_oldest_tid(struct ceph_mds_client *mdsc) |
| 1207 | { |
Sage Weil | 44ca18f | 2010-02-15 12:08:46 -0800 | [diff] [blame] | 1208 | struct ceph_mds_request *req = __get_oldest_req(mdsc); |
| 1209 | |
| 1210 | if (req) |
| 1211 | return req->r_tid; |
| 1212 | return 0; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 1213 | } |
| 1214 | |
| 1215 | /* |
| 1216 | * Build a dentry's path. Allocate on heap; caller must kfree. Based |
| 1217 | * on build_path_from_dentry in fs/cifs/dir.c. |
| 1218 | * |
| 1219 | * If @stop_on_nosnap, generate path relative to the first non-snapped |
| 1220 | * inode. |
| 1221 | * |
| 1222 | * Encode hidden .snap dirs as a double /, i.e. |
| 1223 | * foo/.snap/bar -> foo//bar |
| 1224 | */ |
| 1225 | char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base, |
| 1226 | int stop_on_nosnap) |
| 1227 | { |
| 1228 | struct dentry *temp; |
| 1229 | char *path; |
| 1230 | int len, pos; |
| 1231 | |
| 1232 | if (dentry == NULL) |
| 1233 | return ERR_PTR(-EINVAL); |
| 1234 | |
| 1235 | retry: |
| 1236 | len = 0; |
| 1237 | for (temp = dentry; !IS_ROOT(temp);) { |
| 1238 | struct inode *inode = temp->d_inode; |
| 1239 | if (inode && ceph_snap(inode) == CEPH_SNAPDIR) |
| 1240 | len++; /* slash only */ |
| 1241 | else if (stop_on_nosnap && inode && |
| 1242 | ceph_snap(inode) == CEPH_NOSNAP) |
| 1243 | break; |
| 1244 | else |
| 1245 | len += 1 + temp->d_name.len; |
| 1246 | temp = temp->d_parent; |
| 1247 | if (temp == NULL) { |
| 1248 | pr_err("build_path_dentry corrupt dentry %p\n", dentry); |
| 1249 | return ERR_PTR(-EINVAL); |
| 1250 | } |
| 1251 | } |
| 1252 | if (len) |
| 1253 | len--; /* no leading '/' */ |
| 1254 | |
| 1255 | path = kmalloc(len+1, GFP_NOFS); |
| 1256 | if (path == NULL) |
| 1257 | return ERR_PTR(-ENOMEM); |
| 1258 | pos = len; |
| 1259 | path[pos] = 0; /* trailing null */ |
| 1260 | for (temp = dentry; !IS_ROOT(temp) && pos != 0; ) { |
| 1261 | struct inode *inode = temp->d_inode; |
| 1262 | |
| 1263 | if (inode && ceph_snap(inode) == CEPH_SNAPDIR) { |
| 1264 | dout("build_path_dentry path+%d: %p SNAPDIR\n", |
| 1265 | pos, temp); |
| 1266 | } else if (stop_on_nosnap && inode && |
| 1267 | ceph_snap(inode) == CEPH_NOSNAP) { |
| 1268 | break; |
| 1269 | } else { |
| 1270 | pos -= temp->d_name.len; |
| 1271 | if (pos < 0) |
| 1272 | break; |
| 1273 | strncpy(path + pos, temp->d_name.name, |
| 1274 | temp->d_name.len); |
| 1275 | dout("build_path_dentry path+%d: %p '%.*s'\n", |
| 1276 | pos, temp, temp->d_name.len, path + pos); |
| 1277 | } |
| 1278 | if (pos) |
| 1279 | path[--pos] = '/'; |
| 1280 | temp = temp->d_parent; |
| 1281 | if (temp == NULL) { |
| 1282 | pr_err("build_path_dentry corrupt dentry\n"); |
| 1283 | kfree(path); |
| 1284 | return ERR_PTR(-EINVAL); |
| 1285 | } |
| 1286 | } |
| 1287 | if (pos != 0) { |
| 1288 | pr_err("build_path_dentry did not end path lookup where " |
| 1289 | "expected, namelen is %d, pos is %d\n", len, pos); |
| 1290 | /* presumably this is only possible if racing with a |
| 1291 | rename of one of the parent directories (we can not |
| 1292 | lock the dentries above us to prevent this, but |
| 1293 | retrying should be harmless) */ |
| 1294 | kfree(path); |
| 1295 | goto retry; |
| 1296 | } |
| 1297 | |
| 1298 | *base = ceph_ino(temp->d_inode); |
| 1299 | *plen = len; |
| 1300 | dout("build_path_dentry on %p %d built %llx '%.*s'\n", |
| 1301 | dentry, atomic_read(&dentry->d_count), *base, len, path); |
| 1302 | return path; |
| 1303 | } |
| 1304 | |
| 1305 | static int build_dentry_path(struct dentry *dentry, |
| 1306 | const char **ppath, int *ppathlen, u64 *pino, |
| 1307 | int *pfreepath) |
| 1308 | { |
| 1309 | char *path; |
| 1310 | |
| 1311 | if (ceph_snap(dentry->d_parent->d_inode) == CEPH_NOSNAP) { |
| 1312 | *pino = ceph_ino(dentry->d_parent->d_inode); |
| 1313 | *ppath = dentry->d_name.name; |
| 1314 | *ppathlen = dentry->d_name.len; |
| 1315 | return 0; |
| 1316 | } |
| 1317 | path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1); |
| 1318 | if (IS_ERR(path)) |
| 1319 | return PTR_ERR(path); |
| 1320 | *ppath = path; |
| 1321 | *pfreepath = 1; |
| 1322 | return 0; |
| 1323 | } |
| 1324 | |
| 1325 | static int build_inode_path(struct inode *inode, |
| 1326 | const char **ppath, int *ppathlen, u64 *pino, |
| 1327 | int *pfreepath) |
| 1328 | { |
| 1329 | struct dentry *dentry; |
| 1330 | char *path; |
| 1331 | |
| 1332 | if (ceph_snap(inode) == CEPH_NOSNAP) { |
| 1333 | *pino = ceph_ino(inode); |
| 1334 | *ppathlen = 0; |
| 1335 | return 0; |
| 1336 | } |
| 1337 | dentry = d_find_alias(inode); |
| 1338 | path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1); |
| 1339 | dput(dentry); |
| 1340 | if (IS_ERR(path)) |
| 1341 | return PTR_ERR(path); |
| 1342 | *ppath = path; |
| 1343 | *pfreepath = 1; |
| 1344 | return 0; |
| 1345 | } |
| 1346 | |
| 1347 | /* |
| 1348 | * request arguments may be specified via an inode *, a dentry *, or |
| 1349 | * an explicit ino+path. |
| 1350 | */ |
| 1351 | static int set_request_path_attr(struct inode *rinode, struct dentry *rdentry, |
| 1352 | const char *rpath, u64 rino, |
| 1353 | const char **ppath, int *pathlen, |
| 1354 | u64 *ino, int *freepath) |
| 1355 | { |
| 1356 | int r = 0; |
| 1357 | |
| 1358 | if (rinode) { |
| 1359 | r = build_inode_path(rinode, ppath, pathlen, ino, freepath); |
| 1360 | dout(" inode %p %llx.%llx\n", rinode, ceph_ino(rinode), |
| 1361 | ceph_snap(rinode)); |
| 1362 | } else if (rdentry) { |
| 1363 | r = build_dentry_path(rdentry, ppath, pathlen, ino, freepath); |
| 1364 | dout(" dentry %p %llx/%.*s\n", rdentry, *ino, *pathlen, |
| 1365 | *ppath); |
| 1366 | } else if (rpath) { |
| 1367 | *ino = rino; |
| 1368 | *ppath = rpath; |
| 1369 | *pathlen = strlen(rpath); |
| 1370 | dout(" path %.*s\n", *pathlen, rpath); |
| 1371 | } |
| 1372 | |
| 1373 | return r; |
| 1374 | } |
| 1375 | |
| 1376 | /* |
| 1377 | * called under mdsc->mutex |
| 1378 | */ |
| 1379 | static struct ceph_msg *create_request_message(struct ceph_mds_client *mdsc, |
| 1380 | struct ceph_mds_request *req, |
| 1381 | int mds) |
| 1382 | { |
| 1383 | struct ceph_msg *msg; |
| 1384 | struct ceph_mds_request_head *head; |
| 1385 | const char *path1 = NULL; |
| 1386 | const char *path2 = NULL; |
| 1387 | u64 ino1 = 0, ino2 = 0; |
| 1388 | int pathlen1 = 0, pathlen2 = 0; |
| 1389 | int freepath1 = 0, freepath2 = 0; |
| 1390 | int len; |
| 1391 | u16 releases; |
| 1392 | void *p, *end; |
| 1393 | int ret; |
| 1394 | |
| 1395 | ret = set_request_path_attr(req->r_inode, req->r_dentry, |
| 1396 | req->r_path1, req->r_ino1.ino, |
| 1397 | &path1, &pathlen1, &ino1, &freepath1); |
| 1398 | if (ret < 0) { |
| 1399 | msg = ERR_PTR(ret); |
| 1400 | goto out; |
| 1401 | } |
| 1402 | |
| 1403 | ret = set_request_path_attr(NULL, req->r_old_dentry, |
| 1404 | req->r_path2, req->r_ino2.ino, |
| 1405 | &path2, &pathlen2, &ino2, &freepath2); |
| 1406 | if (ret < 0) { |
| 1407 | msg = ERR_PTR(ret); |
| 1408 | goto out_free1; |
| 1409 | } |
| 1410 | |
| 1411 | len = sizeof(*head) + |
Sage Weil | ac8839d | 2010-01-27 14:28:10 -0800 | [diff] [blame] | 1412 | pathlen1 + pathlen2 + 2*(1 + sizeof(u32) + sizeof(u64)); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 1413 | |
| 1414 | /* calculate (max) length for cap releases */ |
| 1415 | len += sizeof(struct ceph_mds_request_release) * |
| 1416 | (!!req->r_inode_drop + !!req->r_dentry_drop + |
| 1417 | !!req->r_old_inode_drop + !!req->r_old_dentry_drop); |
| 1418 | if (req->r_dentry_drop) |
| 1419 | len += req->r_dentry->d_name.len; |
| 1420 | if (req->r_old_dentry_drop) |
| 1421 | len += req->r_old_dentry->d_name.len; |
| 1422 | |
| 1423 | msg = ceph_msg_new(CEPH_MSG_CLIENT_REQUEST, len, 0, 0, NULL); |
| 1424 | if (IS_ERR(msg)) |
| 1425 | goto out_free2; |
| 1426 | |
Sage Weil | 6df058c | 2009-12-22 11:24:33 -0800 | [diff] [blame] | 1427 | msg->hdr.tid = cpu_to_le64(req->r_tid); |
| 1428 | |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 1429 | head = msg->front.iov_base; |
| 1430 | p = msg->front.iov_base + sizeof(*head); |
| 1431 | end = msg->front.iov_base + msg->front.iov_len; |
| 1432 | |
| 1433 | head->mdsmap_epoch = cpu_to_le32(mdsc->mdsmap->m_epoch); |
| 1434 | head->op = cpu_to_le32(req->r_op); |
| 1435 | head->caller_uid = cpu_to_le32(current_fsuid()); |
| 1436 | head->caller_gid = cpu_to_le32(current_fsgid()); |
| 1437 | head->args = req->r_args; |
| 1438 | |
| 1439 | ceph_encode_filepath(&p, end, ino1, path1); |
| 1440 | ceph_encode_filepath(&p, end, ino2, path2); |
| 1441 | |
| 1442 | /* cap releases */ |
| 1443 | releases = 0; |
| 1444 | if (req->r_inode_drop) |
| 1445 | releases += ceph_encode_inode_release(&p, |
| 1446 | req->r_inode ? req->r_inode : req->r_dentry->d_inode, |
| 1447 | mds, req->r_inode_drop, req->r_inode_unless, 0); |
| 1448 | if (req->r_dentry_drop) |
| 1449 | releases += ceph_encode_dentry_release(&p, req->r_dentry, |
| 1450 | mds, req->r_dentry_drop, req->r_dentry_unless); |
| 1451 | if (req->r_old_dentry_drop) |
| 1452 | releases += ceph_encode_dentry_release(&p, req->r_old_dentry, |
| 1453 | mds, req->r_old_dentry_drop, req->r_old_dentry_unless); |
| 1454 | if (req->r_old_inode_drop) |
| 1455 | releases += ceph_encode_inode_release(&p, |
| 1456 | req->r_old_dentry->d_inode, |
| 1457 | mds, req->r_old_inode_drop, req->r_old_inode_unless, 0); |
| 1458 | head->num_releases = cpu_to_le16(releases); |
| 1459 | |
| 1460 | BUG_ON(p > end); |
| 1461 | msg->front.iov_len = p - msg->front.iov_base; |
| 1462 | msg->hdr.front_len = cpu_to_le32(msg->front.iov_len); |
| 1463 | |
| 1464 | msg->pages = req->r_pages; |
| 1465 | msg->nr_pages = req->r_num_pages; |
| 1466 | msg->hdr.data_len = cpu_to_le32(req->r_data_len); |
| 1467 | msg->hdr.data_off = cpu_to_le16(0); |
| 1468 | |
| 1469 | out_free2: |
| 1470 | if (freepath2) |
| 1471 | kfree((char *)path2); |
| 1472 | out_free1: |
| 1473 | if (freepath1) |
| 1474 | kfree((char *)path1); |
| 1475 | out: |
| 1476 | return msg; |
| 1477 | } |
| 1478 | |
| 1479 | /* |
| 1480 | * called under mdsc->mutex if error, under no mutex if |
| 1481 | * success. |
| 1482 | */ |
| 1483 | static void complete_request(struct ceph_mds_client *mdsc, |
| 1484 | struct ceph_mds_request *req) |
| 1485 | { |
| 1486 | if (req->r_callback) |
| 1487 | req->r_callback(mdsc, req); |
| 1488 | else |
| 1489 | complete(&req->r_completion); |
| 1490 | } |
| 1491 | |
| 1492 | /* |
| 1493 | * called under mdsc->mutex |
| 1494 | */ |
| 1495 | static int __prepare_send_request(struct ceph_mds_client *mdsc, |
| 1496 | struct ceph_mds_request *req, |
| 1497 | int mds) |
| 1498 | { |
| 1499 | struct ceph_mds_request_head *rhead; |
| 1500 | struct ceph_msg *msg; |
| 1501 | int flags = 0; |
| 1502 | |
| 1503 | req->r_mds = mds; |
| 1504 | req->r_attempts++; |
| 1505 | dout("prepare_send_request %p tid %lld %s (attempt %d)\n", req, |
| 1506 | req->r_tid, ceph_mds_op_name(req->r_op), req->r_attempts); |
| 1507 | |
| 1508 | if (req->r_request) { |
| 1509 | ceph_msg_put(req->r_request); |
| 1510 | req->r_request = NULL; |
| 1511 | } |
| 1512 | msg = create_request_message(mdsc, req, mds); |
| 1513 | if (IS_ERR(msg)) { |
| 1514 | req->r_reply = ERR_PTR(PTR_ERR(msg)); |
| 1515 | complete_request(mdsc, req); |
| 1516 | return -PTR_ERR(msg); |
| 1517 | } |
| 1518 | req->r_request = msg; |
| 1519 | |
| 1520 | rhead = msg->front.iov_base; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 1521 | rhead->oldest_client_tid = cpu_to_le64(__get_oldest_tid(mdsc)); |
| 1522 | if (req->r_got_unsafe) |
| 1523 | flags |= CEPH_MDS_FLAG_REPLAY; |
| 1524 | if (req->r_locked_dir) |
| 1525 | flags |= CEPH_MDS_FLAG_WANT_DENTRY; |
| 1526 | rhead->flags = cpu_to_le32(flags); |
| 1527 | rhead->num_fwd = req->r_num_fwd; |
| 1528 | rhead->num_retry = req->r_attempts - 1; |
| 1529 | |
| 1530 | dout(" r_locked_dir = %p\n", req->r_locked_dir); |
| 1531 | |
| 1532 | if (req->r_target_inode && req->r_got_unsafe) |
| 1533 | rhead->ino = cpu_to_le64(ceph_ino(req->r_target_inode)); |
| 1534 | else |
| 1535 | rhead->ino = 0; |
| 1536 | return 0; |
| 1537 | } |
| 1538 | |
| 1539 | /* |
| 1540 | * send request, or put it on the appropriate wait list. |
| 1541 | */ |
| 1542 | static int __do_request(struct ceph_mds_client *mdsc, |
| 1543 | struct ceph_mds_request *req) |
| 1544 | { |
| 1545 | struct ceph_mds_session *session = NULL; |
| 1546 | int mds = -1; |
| 1547 | int err = -EAGAIN; |
| 1548 | |
| 1549 | if (req->r_reply) |
| 1550 | goto out; |
| 1551 | |
| 1552 | if (req->r_timeout && |
| 1553 | time_after_eq(jiffies, req->r_started + req->r_timeout)) { |
| 1554 | dout("do_request timed out\n"); |
| 1555 | err = -EIO; |
| 1556 | goto finish; |
| 1557 | } |
| 1558 | |
| 1559 | mds = __choose_mds(mdsc, req); |
| 1560 | if (mds < 0 || |
| 1561 | ceph_mdsmap_get_state(mdsc->mdsmap, mds) < CEPH_MDS_STATE_ACTIVE) { |
| 1562 | dout("do_request no mds or not active, waiting for map\n"); |
| 1563 | list_add(&req->r_wait, &mdsc->waiting_for_map); |
| 1564 | goto out; |
| 1565 | } |
| 1566 | |
| 1567 | /* get, open session */ |
| 1568 | session = __ceph_lookup_mds_session(mdsc, mds); |
| 1569 | if (!session) |
| 1570 | session = register_session(mdsc, mds); |
| 1571 | dout("do_request mds%d session %p state %s\n", mds, session, |
| 1572 | session_state_name(session->s_state)); |
| 1573 | if (session->s_state != CEPH_MDS_SESSION_OPEN && |
| 1574 | session->s_state != CEPH_MDS_SESSION_HUNG) { |
| 1575 | if (session->s_state == CEPH_MDS_SESSION_NEW || |
| 1576 | session->s_state == CEPH_MDS_SESSION_CLOSING) |
| 1577 | __open_session(mdsc, session); |
| 1578 | list_add(&req->r_wait, &session->s_waiting); |
| 1579 | goto out_session; |
| 1580 | } |
| 1581 | |
| 1582 | /* send request */ |
| 1583 | req->r_session = get_session(session); |
| 1584 | req->r_resend_mds = -1; /* forget any previous mds hint */ |
| 1585 | |
| 1586 | if (req->r_request_started == 0) /* note request start time */ |
| 1587 | req->r_request_started = jiffies; |
| 1588 | |
| 1589 | err = __prepare_send_request(mdsc, req, mds); |
| 1590 | if (!err) { |
| 1591 | ceph_msg_get(req->r_request); |
| 1592 | ceph_con_send(&session->s_con, req->r_request); |
| 1593 | } |
| 1594 | |
| 1595 | out_session: |
| 1596 | ceph_put_mds_session(session); |
| 1597 | out: |
| 1598 | return err; |
| 1599 | |
| 1600 | finish: |
| 1601 | req->r_reply = ERR_PTR(err); |
| 1602 | complete_request(mdsc, req); |
| 1603 | goto out; |
| 1604 | } |
| 1605 | |
| 1606 | /* |
| 1607 | * called under mdsc->mutex |
| 1608 | */ |
| 1609 | static void __wake_requests(struct ceph_mds_client *mdsc, |
| 1610 | struct list_head *head) |
| 1611 | { |
| 1612 | struct ceph_mds_request *req, *nreq; |
| 1613 | |
| 1614 | list_for_each_entry_safe(req, nreq, head, r_wait) { |
| 1615 | list_del_init(&req->r_wait); |
| 1616 | __do_request(mdsc, req); |
| 1617 | } |
| 1618 | } |
| 1619 | |
| 1620 | /* |
| 1621 | * Wake up threads with requests pending for @mds, so that they can |
| 1622 | * resubmit their requests to a possibly different mds. If @all is set, |
| 1623 | * wake up if their requests has been forwarded to @mds, too. |
| 1624 | */ |
| 1625 | static void kick_requests(struct ceph_mds_client *mdsc, int mds, int all) |
| 1626 | { |
Sage Weil | 44ca18f | 2010-02-15 12:08:46 -0800 | [diff] [blame] | 1627 | struct ceph_mds_request *req; |
| 1628 | struct rb_node *p; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 1629 | |
| 1630 | dout("kick_requests mds%d\n", mds); |
Sage Weil | 44ca18f | 2010-02-15 12:08:46 -0800 | [diff] [blame] | 1631 | for (p = rb_first(&mdsc->request_tree); p; p = rb_next(p)) { |
| 1632 | req = rb_entry(p, struct ceph_mds_request, r_node); |
| 1633 | if (req->r_got_unsafe) |
| 1634 | continue; |
| 1635 | if (req->r_session && |
| 1636 | req->r_session->s_mds == mds) { |
| 1637 | dout(" kicking tid %llu\n", req->r_tid); |
| 1638 | put_request_session(req); |
| 1639 | __do_request(mdsc, req); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 1640 | } |
| 1641 | } |
| 1642 | } |
| 1643 | |
| 1644 | void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc, |
| 1645 | struct ceph_mds_request *req) |
| 1646 | { |
| 1647 | dout("submit_request on %p\n", req); |
| 1648 | mutex_lock(&mdsc->mutex); |
| 1649 | __register_request(mdsc, req, NULL); |
| 1650 | __do_request(mdsc, req); |
| 1651 | mutex_unlock(&mdsc->mutex); |
| 1652 | } |
| 1653 | |
| 1654 | /* |
| 1655 | * Synchrously perform an mds request. Take care of all of the |
| 1656 | * session setup, forwarding, retry details. |
| 1657 | */ |
| 1658 | int ceph_mdsc_do_request(struct ceph_mds_client *mdsc, |
| 1659 | struct inode *dir, |
| 1660 | struct ceph_mds_request *req) |
| 1661 | { |
| 1662 | int err; |
| 1663 | |
| 1664 | dout("do_request on %p\n", req); |
| 1665 | |
| 1666 | /* take CAP_PIN refs for r_inode, r_locked_dir, r_old_dentry */ |
| 1667 | if (req->r_inode) |
| 1668 | ceph_get_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN); |
| 1669 | if (req->r_locked_dir) |
| 1670 | ceph_get_cap_refs(ceph_inode(req->r_locked_dir), CEPH_CAP_PIN); |
| 1671 | if (req->r_old_dentry) |
| 1672 | ceph_get_cap_refs( |
| 1673 | ceph_inode(req->r_old_dentry->d_parent->d_inode), |
| 1674 | CEPH_CAP_PIN); |
| 1675 | |
| 1676 | /* issue */ |
| 1677 | mutex_lock(&mdsc->mutex); |
| 1678 | __register_request(mdsc, req, dir); |
| 1679 | __do_request(mdsc, req); |
| 1680 | |
| 1681 | /* wait */ |
| 1682 | if (!req->r_reply) { |
| 1683 | mutex_unlock(&mdsc->mutex); |
| 1684 | if (req->r_timeout) { |
Sage Weil | e2885f0 | 2009-12-15 10:27:48 -0800 | [diff] [blame] | 1685 | err = (long)wait_for_completion_interruptible_timeout( |
| 1686 | &req->r_completion, req->r_timeout); |
| 1687 | if (err == 0) |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 1688 | req->r_reply = ERR_PTR(-EIO); |
Sage Weil | e2885f0 | 2009-12-15 10:27:48 -0800 | [diff] [blame] | 1689 | else if (err < 0) |
| 1690 | req->r_reply = ERR_PTR(err); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 1691 | } else { |
Sage Weil | e2885f0 | 2009-12-15 10:27:48 -0800 | [diff] [blame] | 1692 | err = wait_for_completion_interruptible( |
| 1693 | &req->r_completion); |
| 1694 | if (err) |
| 1695 | req->r_reply = ERR_PTR(err); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 1696 | } |
| 1697 | mutex_lock(&mdsc->mutex); |
| 1698 | } |
| 1699 | |
| 1700 | if (IS_ERR(req->r_reply)) { |
| 1701 | err = PTR_ERR(req->r_reply); |
| 1702 | req->r_reply = NULL; |
| 1703 | |
Sage Weil | 5b1daec | 2010-01-25 11:33:08 -0800 | [diff] [blame] | 1704 | if (err == -ERESTARTSYS) { |
| 1705 | /* aborted */ |
| 1706 | req->r_aborted = true; |
| 1707 | |
| 1708 | if (req->r_locked_dir && |
| 1709 | (req->r_op & CEPH_MDS_OP_WRITE)) { |
| 1710 | struct ceph_inode_info *ci = |
| 1711 | ceph_inode(req->r_locked_dir); |
| 1712 | |
| 1713 | dout("aborted, clearing I_COMPLETE on %p\n", |
| 1714 | req->r_locked_dir); |
| 1715 | spin_lock(&req->r_locked_dir->i_lock); |
| 1716 | ci->i_ceph_flags &= ~CEPH_I_COMPLETE; |
| 1717 | ci->i_release_count++; |
| 1718 | spin_unlock(&req->r_locked_dir->i_lock); |
| 1719 | } |
| 1720 | } else { |
| 1721 | /* clean up this request */ |
| 1722 | __unregister_request(mdsc, req); |
| 1723 | if (!list_empty(&req->r_unsafe_item)) |
| 1724 | list_del_init(&req->r_unsafe_item); |
| 1725 | complete(&req->r_safe_completion); |
| 1726 | } |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 1727 | } else if (req->r_err) { |
| 1728 | err = req->r_err; |
| 1729 | } else { |
| 1730 | err = le32_to_cpu(req->r_reply_info.head->result); |
| 1731 | } |
| 1732 | mutex_unlock(&mdsc->mutex); |
| 1733 | |
| 1734 | dout("do_request %p done, result %d\n", req, err); |
| 1735 | return err; |
| 1736 | } |
| 1737 | |
| 1738 | /* |
| 1739 | * Handle mds reply. |
| 1740 | * |
| 1741 | * We take the session mutex and parse and process the reply immediately. |
| 1742 | * This preserves the logical ordering of replies, capabilities, etc., sent |
| 1743 | * by the MDS as they are applied to our local cache. |
| 1744 | */ |
| 1745 | static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg) |
| 1746 | { |
| 1747 | struct ceph_mds_client *mdsc = session->s_mdsc; |
| 1748 | struct ceph_mds_request *req; |
| 1749 | struct ceph_mds_reply_head *head = msg->front.iov_base; |
| 1750 | struct ceph_mds_reply_info_parsed *rinfo; /* parsed reply info */ |
| 1751 | u64 tid; |
| 1752 | int err, result; |
Sage Weil | 2600d2d | 2010-02-22 15:12:16 -0800 | [diff] [blame] | 1753 | int mds = session->s_mds; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 1754 | |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 1755 | if (msg->front.iov_len < sizeof(*head)) { |
| 1756 | pr_err("mdsc_handle_reply got corrupt (short) reply\n"); |
Sage Weil | 9ec7cab | 2009-12-14 15:13:47 -0800 | [diff] [blame] | 1757 | ceph_msg_dump(msg); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 1758 | return; |
| 1759 | } |
| 1760 | |
| 1761 | /* get request, session */ |
Sage Weil | 6df058c | 2009-12-22 11:24:33 -0800 | [diff] [blame] | 1762 | tid = le64_to_cpu(msg->hdr.tid); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 1763 | mutex_lock(&mdsc->mutex); |
| 1764 | req = __lookup_request(mdsc, tid); |
| 1765 | if (!req) { |
| 1766 | dout("handle_reply on unknown tid %llu\n", tid); |
| 1767 | mutex_unlock(&mdsc->mutex); |
| 1768 | return; |
| 1769 | } |
| 1770 | dout("handle_reply %p\n", req); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 1771 | |
| 1772 | /* correct session? */ |
| 1773 | if (!req->r_session && req->r_session != session) { |
| 1774 | pr_err("mdsc_handle_reply got %llu on session mds%d" |
| 1775 | " not mds%d\n", tid, session->s_mds, |
| 1776 | req->r_session ? req->r_session->s_mds : -1); |
| 1777 | mutex_unlock(&mdsc->mutex); |
| 1778 | goto out; |
| 1779 | } |
| 1780 | |
| 1781 | /* dup? */ |
| 1782 | if ((req->r_got_unsafe && !head->safe) || |
| 1783 | (req->r_got_safe && head->safe)) { |
| 1784 | pr_warning("got a dup %s reply on %llu from mds%d\n", |
| 1785 | head->safe ? "safe" : "unsafe", tid, mds); |
| 1786 | mutex_unlock(&mdsc->mutex); |
| 1787 | goto out; |
| 1788 | } |
| 1789 | |
| 1790 | result = le32_to_cpu(head->result); |
| 1791 | |
| 1792 | /* |
| 1793 | * Tolerate 2 consecutive ESTALEs from the same mds. |
| 1794 | * FIXME: we should be looking at the cap migrate_seq. |
| 1795 | */ |
| 1796 | if (result == -ESTALE) { |
| 1797 | req->r_direct_mode = USE_AUTH_MDS; |
| 1798 | req->r_num_stale++; |
| 1799 | if (req->r_num_stale <= 2) { |
| 1800 | __do_request(mdsc, req); |
| 1801 | mutex_unlock(&mdsc->mutex); |
| 1802 | goto out; |
| 1803 | } |
| 1804 | } else { |
| 1805 | req->r_num_stale = 0; |
| 1806 | } |
| 1807 | |
| 1808 | if (head->safe) { |
| 1809 | req->r_got_safe = true; |
| 1810 | __unregister_request(mdsc, req); |
| 1811 | complete(&req->r_safe_completion); |
| 1812 | |
| 1813 | if (req->r_got_unsafe) { |
| 1814 | /* |
| 1815 | * We already handled the unsafe response, now do the |
| 1816 | * cleanup. No need to examine the response; the MDS |
| 1817 | * doesn't include any result info in the safe |
| 1818 | * response. And even if it did, there is nothing |
| 1819 | * useful we could do with a revised return value. |
| 1820 | */ |
| 1821 | dout("got safe reply %llu, mds%d\n", tid, mds); |
| 1822 | list_del_init(&req->r_unsafe_item); |
| 1823 | |
| 1824 | /* last unsafe request during umount? */ |
Sage Weil | 44ca18f | 2010-02-15 12:08:46 -0800 | [diff] [blame] | 1825 | if (mdsc->stopping && !__get_oldest_req(mdsc)) |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 1826 | complete(&mdsc->safe_umount_waiters); |
| 1827 | mutex_unlock(&mdsc->mutex); |
| 1828 | goto out; |
| 1829 | } |
| 1830 | } |
| 1831 | |
| 1832 | BUG_ON(req->r_reply); |
| 1833 | |
| 1834 | if (!head->safe) { |
| 1835 | req->r_got_unsafe = true; |
| 1836 | list_add_tail(&req->r_unsafe_item, &req->r_session->s_unsafe); |
| 1837 | } |
| 1838 | |
| 1839 | dout("handle_reply tid %lld result %d\n", tid, result); |
| 1840 | rinfo = &req->r_reply_info; |
| 1841 | err = parse_reply_info(msg, rinfo); |
| 1842 | mutex_unlock(&mdsc->mutex); |
| 1843 | |
| 1844 | mutex_lock(&session->s_mutex); |
| 1845 | if (err < 0) { |
| 1846 | pr_err("mdsc_handle_reply got corrupt reply mds%d\n", mds); |
Sage Weil | 9ec7cab | 2009-12-14 15:13:47 -0800 | [diff] [blame] | 1847 | ceph_msg_dump(msg); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 1848 | goto out_err; |
| 1849 | } |
| 1850 | |
| 1851 | /* snap trace */ |
| 1852 | if (rinfo->snapblob_len) { |
| 1853 | down_write(&mdsc->snap_rwsem); |
| 1854 | ceph_update_snap_trace(mdsc, rinfo->snapblob, |
| 1855 | rinfo->snapblob + rinfo->snapblob_len, |
| 1856 | le32_to_cpu(head->op) == CEPH_MDS_OP_RMSNAP); |
| 1857 | downgrade_write(&mdsc->snap_rwsem); |
| 1858 | } else { |
| 1859 | down_read(&mdsc->snap_rwsem); |
| 1860 | } |
| 1861 | |
| 1862 | /* insert trace into our cache */ |
| 1863 | err = ceph_fill_trace(mdsc->client->sb, req, req->r_session); |
| 1864 | if (err == 0) { |
| 1865 | if (result == 0 && rinfo->dir_nr) |
| 1866 | ceph_readdir_prepopulate(req, req->r_session); |
| 1867 | ceph_unreserve_caps(&req->r_caps_reservation); |
| 1868 | } |
| 1869 | |
| 1870 | up_read(&mdsc->snap_rwsem); |
| 1871 | out_err: |
| 1872 | if (err) { |
| 1873 | req->r_err = err; |
| 1874 | } else { |
| 1875 | req->r_reply = msg; |
| 1876 | ceph_msg_get(msg); |
| 1877 | } |
| 1878 | |
| 1879 | add_cap_releases(mdsc, req->r_session, -1); |
| 1880 | mutex_unlock(&session->s_mutex); |
| 1881 | |
| 1882 | /* kick calling process */ |
| 1883 | complete_request(mdsc, req); |
| 1884 | out: |
| 1885 | ceph_mdsc_put_request(req); |
| 1886 | return; |
| 1887 | } |
| 1888 | |
| 1889 | |
| 1890 | |
| 1891 | /* |
| 1892 | * handle mds notification that our request has been forwarded. |
| 1893 | */ |
Sage Weil | 2600d2d | 2010-02-22 15:12:16 -0800 | [diff] [blame] | 1894 | static void handle_forward(struct ceph_mds_client *mdsc, |
| 1895 | struct ceph_mds_session *session, |
| 1896 | struct ceph_msg *msg) |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 1897 | { |
| 1898 | struct ceph_mds_request *req; |
Sage Weil | a1ea787 | 2010-02-23 14:02:44 -0800 | [diff] [blame] | 1899 | u64 tid = le64_to_cpu(msg->hdr.tid); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 1900 | u32 next_mds; |
| 1901 | u32 fwd_seq; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 1902 | int err = -EINVAL; |
| 1903 | void *p = msg->front.iov_base; |
| 1904 | void *end = p + msg->front.iov_len; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 1905 | |
Sage Weil | a1ea787 | 2010-02-23 14:02:44 -0800 | [diff] [blame] | 1906 | ceph_decode_need(&p, end, 2*sizeof(u32), bad); |
Sage Weil | c89136e | 2009-10-14 09:59:09 -0700 | [diff] [blame] | 1907 | next_mds = ceph_decode_32(&p); |
| 1908 | fwd_seq = ceph_decode_32(&p); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 1909 | |
| 1910 | mutex_lock(&mdsc->mutex); |
| 1911 | req = __lookup_request(mdsc, tid); |
| 1912 | if (!req) { |
Sage Weil | 080af17 | 2010-02-25 16:40:07 -0800 | [diff] [blame] | 1913 | dout("forward %llu to mds%d - req dne\n", tid, next_mds); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 1914 | goto out; /* dup reply? */ |
| 1915 | } |
| 1916 | |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 1917 | if (fwd_seq <= req->r_num_fwd) { |
| 1918 | dout("forward %llu to mds%d - old seq %d <= %d\n", |
| 1919 | tid, next_mds, req->r_num_fwd, fwd_seq); |
| 1920 | } else { |
| 1921 | /* resend. forward race not possible; mds would drop */ |
| 1922 | dout("forward %llu to mds%d (we resend)\n", tid, next_mds); |
| 1923 | req->r_num_fwd = fwd_seq; |
| 1924 | req->r_resend_mds = next_mds; |
| 1925 | put_request_session(req); |
| 1926 | __do_request(mdsc, req); |
| 1927 | } |
| 1928 | ceph_mdsc_put_request(req); |
| 1929 | out: |
| 1930 | mutex_unlock(&mdsc->mutex); |
| 1931 | return; |
| 1932 | |
| 1933 | bad: |
| 1934 | pr_err("mdsc_handle_forward decode error err=%d\n", err); |
| 1935 | } |
| 1936 | |
| 1937 | /* |
| 1938 | * handle a mds session control message |
| 1939 | */ |
| 1940 | static void handle_session(struct ceph_mds_session *session, |
| 1941 | struct ceph_msg *msg) |
| 1942 | { |
| 1943 | struct ceph_mds_client *mdsc = session->s_mdsc; |
| 1944 | u32 op; |
| 1945 | u64 seq; |
Sage Weil | 2600d2d | 2010-02-22 15:12:16 -0800 | [diff] [blame] | 1946 | int mds = session->s_mds; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 1947 | struct ceph_mds_session_head *h = msg->front.iov_base; |
| 1948 | int wake = 0; |
| 1949 | |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 1950 | /* decode */ |
| 1951 | if (msg->front.iov_len != sizeof(*h)) |
| 1952 | goto bad; |
| 1953 | op = le32_to_cpu(h->op); |
| 1954 | seq = le64_to_cpu(h->seq); |
| 1955 | |
| 1956 | mutex_lock(&mdsc->mutex); |
Sage Weil | 2600d2d | 2010-02-22 15:12:16 -0800 | [diff] [blame] | 1957 | if (op == CEPH_SESSION_CLOSE) |
| 1958 | __unregister_session(mdsc, session); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 1959 | /* FIXME: this ttl calculation is generous */ |
| 1960 | session->s_ttl = jiffies + HZ*mdsc->mdsmap->m_session_autoclose; |
| 1961 | mutex_unlock(&mdsc->mutex); |
| 1962 | |
| 1963 | mutex_lock(&session->s_mutex); |
| 1964 | |
| 1965 | dout("handle_session mds%d %s %p state %s seq %llu\n", |
| 1966 | mds, ceph_session_op_name(op), session, |
| 1967 | session_state_name(session->s_state), seq); |
| 1968 | |
| 1969 | if (session->s_state == CEPH_MDS_SESSION_HUNG) { |
| 1970 | session->s_state = CEPH_MDS_SESSION_OPEN; |
| 1971 | pr_info("mds%d came back\n", session->s_mds); |
| 1972 | } |
| 1973 | |
| 1974 | switch (op) { |
| 1975 | case CEPH_SESSION_OPEN: |
| 1976 | session->s_state = CEPH_MDS_SESSION_OPEN; |
| 1977 | renewed_caps(mdsc, session, 0); |
| 1978 | wake = 1; |
| 1979 | if (mdsc->stopping) |
| 1980 | __close_session(mdsc, session); |
| 1981 | break; |
| 1982 | |
| 1983 | case CEPH_SESSION_RENEWCAPS: |
| 1984 | if (session->s_renew_seq == seq) |
| 1985 | renewed_caps(mdsc, session, 1); |
| 1986 | break; |
| 1987 | |
| 1988 | case CEPH_SESSION_CLOSE: |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 1989 | remove_session_caps(session); |
| 1990 | wake = 1; /* for good measure */ |
| 1991 | complete(&mdsc->session_close_waiters); |
| 1992 | kick_requests(mdsc, mds, 0); /* cur only */ |
| 1993 | break; |
| 1994 | |
| 1995 | case CEPH_SESSION_STALE: |
| 1996 | pr_info("mds%d caps went stale, renewing\n", |
| 1997 | session->s_mds); |
| 1998 | spin_lock(&session->s_cap_lock); |
| 1999 | session->s_cap_gen++; |
| 2000 | session->s_cap_ttl = 0; |
| 2001 | spin_unlock(&session->s_cap_lock); |
| 2002 | send_renew_caps(mdsc, session); |
| 2003 | break; |
| 2004 | |
| 2005 | case CEPH_SESSION_RECALL_STATE: |
| 2006 | trim_caps(mdsc, session, le32_to_cpu(h->max_caps)); |
| 2007 | break; |
| 2008 | |
| 2009 | default: |
| 2010 | pr_err("mdsc_handle_session bad op %d mds%d\n", op, mds); |
| 2011 | WARN_ON(1); |
| 2012 | } |
| 2013 | |
| 2014 | mutex_unlock(&session->s_mutex); |
| 2015 | if (wake) { |
| 2016 | mutex_lock(&mdsc->mutex); |
| 2017 | __wake_requests(mdsc, &session->s_waiting); |
| 2018 | mutex_unlock(&mdsc->mutex); |
| 2019 | } |
| 2020 | return; |
| 2021 | |
| 2022 | bad: |
| 2023 | pr_err("mdsc_handle_session corrupt message mds%d len %d\n", mds, |
| 2024 | (int)msg->front.iov_len); |
Sage Weil | 9ec7cab | 2009-12-14 15:13:47 -0800 | [diff] [blame] | 2025 | ceph_msg_dump(msg); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2026 | return; |
| 2027 | } |
| 2028 | |
| 2029 | |
| 2030 | /* |
| 2031 | * called under session->mutex. |
| 2032 | */ |
| 2033 | static void replay_unsafe_requests(struct ceph_mds_client *mdsc, |
| 2034 | struct ceph_mds_session *session) |
| 2035 | { |
| 2036 | struct ceph_mds_request *req, *nreq; |
| 2037 | int err; |
| 2038 | |
| 2039 | dout("replay_unsafe_requests mds%d\n", session->s_mds); |
| 2040 | |
| 2041 | mutex_lock(&mdsc->mutex); |
| 2042 | list_for_each_entry_safe(req, nreq, &session->s_unsafe, r_unsafe_item) { |
| 2043 | err = __prepare_send_request(mdsc, req, session->s_mds); |
| 2044 | if (!err) { |
| 2045 | ceph_msg_get(req->r_request); |
| 2046 | ceph_con_send(&session->s_con, req->r_request); |
| 2047 | } |
| 2048 | } |
| 2049 | mutex_unlock(&mdsc->mutex); |
| 2050 | } |
| 2051 | |
| 2052 | /* |
| 2053 | * Encode information about a cap for a reconnect with the MDS. |
| 2054 | */ |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2055 | static int encode_caps_cb(struct inode *inode, struct ceph_cap *cap, |
| 2056 | void *arg) |
| 2057 | { |
Sage Weil | 93cea5b | 2009-12-23 12:21:51 -0800 | [diff] [blame] | 2058 | struct ceph_mds_cap_reconnect rec; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2059 | struct ceph_inode_info *ci; |
Sage Weil | 93cea5b | 2009-12-23 12:21:51 -0800 | [diff] [blame] | 2060 | struct ceph_pagelist *pagelist = arg; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2061 | char *path; |
| 2062 | int pathlen, err; |
| 2063 | u64 pathbase; |
| 2064 | struct dentry *dentry; |
| 2065 | |
| 2066 | ci = cap->ci; |
| 2067 | |
| 2068 | dout(" adding %p ino %llx.%llx cap %p %lld %s\n", |
| 2069 | inode, ceph_vinop(inode), cap, cap->cap_id, |
| 2070 | ceph_cap_string(cap->issued)); |
Sage Weil | 93cea5b | 2009-12-23 12:21:51 -0800 | [diff] [blame] | 2071 | err = ceph_pagelist_encode_64(pagelist, ceph_ino(inode)); |
| 2072 | if (err) |
| 2073 | return err; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2074 | |
| 2075 | dentry = d_find_alias(inode); |
| 2076 | if (dentry) { |
| 2077 | path = ceph_mdsc_build_path(dentry, &pathlen, &pathbase, 0); |
| 2078 | if (IS_ERR(path)) { |
| 2079 | err = PTR_ERR(path); |
| 2080 | BUG_ON(err); |
| 2081 | } |
| 2082 | } else { |
| 2083 | path = NULL; |
| 2084 | pathlen = 0; |
| 2085 | } |
Sage Weil | 93cea5b | 2009-12-23 12:21:51 -0800 | [diff] [blame] | 2086 | err = ceph_pagelist_encode_string(pagelist, path, pathlen); |
| 2087 | if (err) |
| 2088 | goto out; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2089 | |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2090 | spin_lock(&inode->i_lock); |
| 2091 | cap->seq = 0; /* reset cap seq */ |
| 2092 | cap->issue_seq = 0; /* and issue_seq */ |
Sage Weil | 93cea5b | 2009-12-23 12:21:51 -0800 | [diff] [blame] | 2093 | rec.cap_id = cpu_to_le64(cap->cap_id); |
| 2094 | rec.pathbase = cpu_to_le64(pathbase); |
| 2095 | rec.wanted = cpu_to_le32(__ceph_caps_wanted(ci)); |
| 2096 | rec.issued = cpu_to_le32(cap->issued); |
| 2097 | rec.size = cpu_to_le64(inode->i_size); |
| 2098 | ceph_encode_timespec(&rec.mtime, &inode->i_mtime); |
| 2099 | ceph_encode_timespec(&rec.atime, &inode->i_atime); |
| 2100 | rec.snaprealm = cpu_to_le64(ci->i_snap_realm->ino); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2101 | spin_unlock(&inode->i_lock); |
| 2102 | |
Sage Weil | 93cea5b | 2009-12-23 12:21:51 -0800 | [diff] [blame] | 2103 | err = ceph_pagelist_append(pagelist, &rec, sizeof(rec)); |
| 2104 | |
| 2105 | out: |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2106 | kfree(path); |
| 2107 | dput(dentry); |
Sage Weil | 93cea5b | 2009-12-23 12:21:51 -0800 | [diff] [blame] | 2108 | return err; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2109 | } |
| 2110 | |
| 2111 | |
| 2112 | /* |
| 2113 | * If an MDS fails and recovers, clients need to reconnect in order to |
| 2114 | * reestablish shared state. This includes all caps issued through |
| 2115 | * this session _and_ the snap_realm hierarchy. Because it's not |
| 2116 | * clear which snap realms the mds cares about, we send everything we |
| 2117 | * know about.. that ensures we'll then get any new info the |
| 2118 | * recovering MDS might have. |
| 2119 | * |
| 2120 | * This is a relatively heavyweight operation, but it's rare. |
| 2121 | * |
| 2122 | * called with mdsc->mutex held. |
| 2123 | */ |
| 2124 | static void send_mds_reconnect(struct ceph_mds_client *mdsc, int mds) |
| 2125 | { |
Sage Weil | 93cea5b | 2009-12-23 12:21:51 -0800 | [diff] [blame] | 2126 | struct ceph_mds_session *session = NULL; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2127 | struct ceph_msg *reply; |
Sage Weil | a105f00 | 2010-02-15 14:37:55 -0800 | [diff] [blame] | 2128 | struct rb_node *p; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2129 | int err; |
Sage Weil | 93cea5b | 2009-12-23 12:21:51 -0800 | [diff] [blame] | 2130 | struct ceph_pagelist *pagelist; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2131 | |
| 2132 | pr_info("reconnect to recovering mds%d\n", mds); |
| 2133 | |
Sage Weil | 93cea5b | 2009-12-23 12:21:51 -0800 | [diff] [blame] | 2134 | pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS); |
| 2135 | if (!pagelist) |
| 2136 | goto fail_nopagelist; |
| 2137 | ceph_pagelist_init(pagelist); |
| 2138 | |
| 2139 | reply = ceph_msg_new(CEPH_MSG_CLIENT_RECONNECT, 0, 0, 0, NULL); |
| 2140 | if (IS_ERR(reply)) { |
| 2141 | err = PTR_ERR(reply); |
| 2142 | goto fail_nomsg; |
| 2143 | } |
| 2144 | |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2145 | /* find session */ |
| 2146 | session = __ceph_lookup_mds_session(mdsc, mds); |
| 2147 | mutex_unlock(&mdsc->mutex); /* drop lock for duration */ |
| 2148 | |
| 2149 | if (session) { |
| 2150 | mutex_lock(&session->s_mutex); |
| 2151 | |
| 2152 | session->s_state = CEPH_MDS_SESSION_RECONNECTING; |
| 2153 | session->s_seq = 0; |
| 2154 | |
| 2155 | ceph_con_open(&session->s_con, |
| 2156 | ceph_mdsmap_get_addr(mdsc->mdsmap, mds)); |
| 2157 | |
| 2158 | /* replay unsafe requests */ |
| 2159 | replay_unsafe_requests(mdsc, session); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2160 | } else { |
| 2161 | dout("no session for mds%d, will send short reconnect\n", |
| 2162 | mds); |
| 2163 | } |
| 2164 | |
| 2165 | down_read(&mdsc->snap_rwsem); |
| 2166 | |
Sage Weil | 93cea5b | 2009-12-23 12:21:51 -0800 | [diff] [blame] | 2167 | if (!session) |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2168 | goto send; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2169 | dout("session %p state %s\n", session, |
| 2170 | session_state_name(session->s_state)); |
| 2171 | |
| 2172 | /* traverse this session's caps */ |
Sage Weil | 93cea5b | 2009-12-23 12:21:51 -0800 | [diff] [blame] | 2173 | err = ceph_pagelist_encode_32(pagelist, session->s_nr_caps); |
| 2174 | if (err) |
| 2175 | goto fail; |
| 2176 | err = iterate_session_caps(session, encode_caps_cb, pagelist); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2177 | if (err < 0) |
| 2178 | goto out; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2179 | |
| 2180 | /* |
| 2181 | * snaprealms. we provide mds with the ino, seq (version), and |
| 2182 | * parent for all of our realms. If the mds has any newer info, |
| 2183 | * it will tell us. |
| 2184 | */ |
Sage Weil | a105f00 | 2010-02-15 14:37:55 -0800 | [diff] [blame] | 2185 | for (p = rb_first(&mdsc->snap_realms); p; p = rb_next(p)) { |
| 2186 | struct ceph_snap_realm *realm = |
| 2187 | rb_entry(p, struct ceph_snap_realm, node); |
Sage Weil | 93cea5b | 2009-12-23 12:21:51 -0800 | [diff] [blame] | 2188 | struct ceph_mds_snaprealm_reconnect sr_rec; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2189 | |
| 2190 | dout(" adding snap realm %llx seq %lld parent %llx\n", |
| 2191 | realm->ino, realm->seq, realm->parent_ino); |
Sage Weil | 93cea5b | 2009-12-23 12:21:51 -0800 | [diff] [blame] | 2192 | sr_rec.ino = cpu_to_le64(realm->ino); |
| 2193 | sr_rec.seq = cpu_to_le64(realm->seq); |
| 2194 | sr_rec.parent = cpu_to_le64(realm->parent_ino); |
| 2195 | err = ceph_pagelist_append(pagelist, &sr_rec, sizeof(sr_rec)); |
| 2196 | if (err) |
| 2197 | goto fail; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2198 | } |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2199 | |
| 2200 | send: |
Sage Weil | 93cea5b | 2009-12-23 12:21:51 -0800 | [diff] [blame] | 2201 | reply->pagelist = pagelist; |
| 2202 | reply->hdr.data_len = cpu_to_le32(pagelist->length); |
| 2203 | reply->nr_pages = calc_pages_for(0, pagelist->length); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2204 | ceph_con_send(&session->s_con, reply); |
| 2205 | |
| 2206 | if (session) { |
| 2207 | session->s_state = CEPH_MDS_SESSION_OPEN; |
| 2208 | __wake_requests(mdsc, &session->s_waiting); |
| 2209 | } |
| 2210 | |
| 2211 | out: |
| 2212 | up_read(&mdsc->snap_rwsem); |
| 2213 | if (session) { |
| 2214 | mutex_unlock(&session->s_mutex); |
| 2215 | ceph_put_mds_session(session); |
| 2216 | } |
| 2217 | mutex_lock(&mdsc->mutex); |
| 2218 | return; |
| 2219 | |
Sage Weil | 93cea5b | 2009-12-23 12:21:51 -0800 | [diff] [blame] | 2220 | fail: |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2221 | ceph_msg_put(reply); |
Sage Weil | 93cea5b | 2009-12-23 12:21:51 -0800 | [diff] [blame] | 2222 | fail_nomsg: |
| 2223 | ceph_pagelist_release(pagelist); |
| 2224 | kfree(pagelist); |
| 2225 | fail_nopagelist: |
| 2226 | pr_err("ENOMEM preparing reconnect for mds%d\n", mds); |
| 2227 | goto out; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2228 | } |
| 2229 | |
| 2230 | |
| 2231 | /* |
| 2232 | * compare old and new mdsmaps, kicking requests |
| 2233 | * and closing out old connections as necessary |
| 2234 | * |
| 2235 | * called under mdsc->mutex. |
| 2236 | */ |
| 2237 | static void check_new_map(struct ceph_mds_client *mdsc, |
| 2238 | struct ceph_mdsmap *newmap, |
| 2239 | struct ceph_mdsmap *oldmap) |
| 2240 | { |
| 2241 | int i; |
| 2242 | int oldstate, newstate; |
| 2243 | struct ceph_mds_session *s; |
| 2244 | |
| 2245 | dout("check_new_map new %u old %u\n", |
| 2246 | newmap->m_epoch, oldmap->m_epoch); |
| 2247 | |
| 2248 | for (i = 0; i < oldmap->m_max_mds && i < mdsc->max_sessions; i++) { |
| 2249 | if (mdsc->sessions[i] == NULL) |
| 2250 | continue; |
| 2251 | s = mdsc->sessions[i]; |
| 2252 | oldstate = ceph_mdsmap_get_state(oldmap, i); |
| 2253 | newstate = ceph_mdsmap_get_state(newmap, i); |
| 2254 | |
| 2255 | dout("check_new_map mds%d state %s -> %s (session %s)\n", |
| 2256 | i, ceph_mds_state_name(oldstate), |
| 2257 | ceph_mds_state_name(newstate), |
| 2258 | session_state_name(s->s_state)); |
| 2259 | |
| 2260 | if (memcmp(ceph_mdsmap_get_addr(oldmap, i), |
| 2261 | ceph_mdsmap_get_addr(newmap, i), |
| 2262 | sizeof(struct ceph_entity_addr))) { |
| 2263 | if (s->s_state == CEPH_MDS_SESSION_OPENING) { |
| 2264 | /* the session never opened, just close it |
| 2265 | * out now */ |
| 2266 | __wake_requests(mdsc, &s->s_waiting); |
Sage Weil | 2600d2d | 2010-02-22 15:12:16 -0800 | [diff] [blame] | 2267 | __unregister_session(mdsc, s); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2268 | } else { |
| 2269 | /* just close it */ |
| 2270 | mutex_unlock(&mdsc->mutex); |
| 2271 | mutex_lock(&s->s_mutex); |
| 2272 | mutex_lock(&mdsc->mutex); |
| 2273 | ceph_con_close(&s->s_con); |
| 2274 | mutex_unlock(&s->s_mutex); |
| 2275 | s->s_state = CEPH_MDS_SESSION_RESTARTING; |
| 2276 | } |
| 2277 | |
| 2278 | /* kick any requests waiting on the recovering mds */ |
| 2279 | kick_requests(mdsc, i, 1); |
| 2280 | } else if (oldstate == newstate) { |
| 2281 | continue; /* nothing new with this mds */ |
| 2282 | } |
| 2283 | |
| 2284 | /* |
| 2285 | * send reconnect? |
| 2286 | */ |
| 2287 | if (s->s_state == CEPH_MDS_SESSION_RESTARTING && |
| 2288 | newstate >= CEPH_MDS_STATE_RECONNECT) |
| 2289 | send_mds_reconnect(mdsc, i); |
| 2290 | |
| 2291 | /* |
| 2292 | * kick requests on any mds that has gone active. |
| 2293 | * |
| 2294 | * kick requests on cur or forwarder: we may have sent |
| 2295 | * the request to mds1, mds1 told us it forwarded it |
| 2296 | * to mds2, but then we learn mds1 failed and can't be |
| 2297 | * sure it successfully forwarded our request before |
| 2298 | * it died. |
| 2299 | */ |
| 2300 | if (oldstate < CEPH_MDS_STATE_ACTIVE && |
| 2301 | newstate >= CEPH_MDS_STATE_ACTIVE) { |
Sage Weil | fef320f | 2009-11-11 15:50:12 -0800 | [diff] [blame] | 2302 | pr_info("mds%d reconnect completed\n", s->s_mds); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2303 | kick_requests(mdsc, i, 1); |
| 2304 | ceph_kick_flushing_caps(mdsc, s); |
Sage Weil | 0dc2570 | 2009-11-20 13:43:45 -0800 | [diff] [blame] | 2305 | wake_up_session_caps(s, 1); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2306 | } |
| 2307 | } |
| 2308 | } |
| 2309 | |
| 2310 | |
| 2311 | |
| 2312 | /* |
| 2313 | * leases |
| 2314 | */ |
| 2315 | |
| 2316 | /* |
| 2317 | * caller must hold session s_mutex, dentry->d_lock |
| 2318 | */ |
| 2319 | void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry) |
| 2320 | { |
| 2321 | struct ceph_dentry_info *di = ceph_dentry(dentry); |
| 2322 | |
| 2323 | ceph_put_mds_session(di->lease_session); |
| 2324 | di->lease_session = NULL; |
| 2325 | } |
| 2326 | |
Sage Weil | 2600d2d | 2010-02-22 15:12:16 -0800 | [diff] [blame] | 2327 | static void handle_lease(struct ceph_mds_client *mdsc, |
| 2328 | struct ceph_mds_session *session, |
| 2329 | struct ceph_msg *msg) |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2330 | { |
| 2331 | struct super_block *sb = mdsc->client->sb; |
| 2332 | struct inode *inode; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2333 | struct ceph_inode_info *ci; |
| 2334 | struct dentry *parent, *dentry; |
| 2335 | struct ceph_dentry_info *di; |
Sage Weil | 2600d2d | 2010-02-22 15:12:16 -0800 | [diff] [blame] | 2336 | int mds = session->s_mds; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2337 | struct ceph_mds_lease *h = msg->front.iov_base; |
| 2338 | struct ceph_vino vino; |
| 2339 | int mask; |
| 2340 | struct qstr dname; |
| 2341 | int release = 0; |
| 2342 | |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2343 | dout("handle_lease from mds%d\n", mds); |
| 2344 | |
| 2345 | /* decode */ |
| 2346 | if (msg->front.iov_len < sizeof(*h) + sizeof(u32)) |
| 2347 | goto bad; |
| 2348 | vino.ino = le64_to_cpu(h->ino); |
| 2349 | vino.snap = CEPH_NOSNAP; |
| 2350 | mask = le16_to_cpu(h->mask); |
| 2351 | dname.name = (void *)h + sizeof(*h) + sizeof(u32); |
| 2352 | dname.len = msg->front.iov_len - sizeof(*h) - sizeof(u32); |
| 2353 | if (dname.len != get_unaligned_le32(h+1)) |
| 2354 | goto bad; |
| 2355 | |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2356 | mutex_lock(&session->s_mutex); |
| 2357 | session->s_seq++; |
| 2358 | |
| 2359 | /* lookup inode */ |
| 2360 | inode = ceph_find_inode(sb, vino); |
| 2361 | dout("handle_lease '%s', mask %d, ino %llx %p\n", |
| 2362 | ceph_lease_op_name(h->action), mask, vino.ino, inode); |
| 2363 | if (inode == NULL) { |
| 2364 | dout("handle_lease no inode %llx\n", vino.ino); |
| 2365 | goto release; |
| 2366 | } |
| 2367 | ci = ceph_inode(inode); |
| 2368 | |
| 2369 | /* dentry */ |
| 2370 | parent = d_find_alias(inode); |
| 2371 | if (!parent) { |
| 2372 | dout("no parent dentry on inode %p\n", inode); |
| 2373 | WARN_ON(1); |
| 2374 | goto release; /* hrm... */ |
| 2375 | } |
| 2376 | dname.hash = full_name_hash(dname.name, dname.len); |
| 2377 | dentry = d_lookup(parent, &dname); |
| 2378 | dput(parent); |
| 2379 | if (!dentry) |
| 2380 | goto release; |
| 2381 | |
| 2382 | spin_lock(&dentry->d_lock); |
| 2383 | di = ceph_dentry(dentry); |
| 2384 | switch (h->action) { |
| 2385 | case CEPH_MDS_LEASE_REVOKE: |
| 2386 | if (di && di->lease_session == session) { |
| 2387 | h->seq = cpu_to_le32(di->lease_seq); |
| 2388 | __ceph_mdsc_drop_dentry_lease(dentry); |
| 2389 | } |
| 2390 | release = 1; |
| 2391 | break; |
| 2392 | |
| 2393 | case CEPH_MDS_LEASE_RENEW: |
| 2394 | if (di && di->lease_session == session && |
| 2395 | di->lease_gen == session->s_cap_gen && |
| 2396 | di->lease_renew_from && |
| 2397 | di->lease_renew_after == 0) { |
| 2398 | unsigned long duration = |
| 2399 | le32_to_cpu(h->duration_ms) * HZ / 1000; |
| 2400 | |
| 2401 | di->lease_seq = le32_to_cpu(h->seq); |
| 2402 | dentry->d_time = di->lease_renew_from + duration; |
| 2403 | di->lease_renew_after = di->lease_renew_from + |
| 2404 | (duration >> 1); |
| 2405 | di->lease_renew_from = 0; |
| 2406 | } |
| 2407 | break; |
| 2408 | } |
| 2409 | spin_unlock(&dentry->d_lock); |
| 2410 | dput(dentry); |
| 2411 | |
| 2412 | if (!release) |
| 2413 | goto out; |
| 2414 | |
| 2415 | release: |
| 2416 | /* let's just reuse the same message */ |
| 2417 | h->action = CEPH_MDS_LEASE_REVOKE_ACK; |
| 2418 | ceph_msg_get(msg); |
| 2419 | ceph_con_send(&session->s_con, msg); |
| 2420 | |
| 2421 | out: |
| 2422 | iput(inode); |
| 2423 | mutex_unlock(&session->s_mutex); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2424 | return; |
| 2425 | |
| 2426 | bad: |
| 2427 | pr_err("corrupt lease message\n"); |
Sage Weil | 9ec7cab | 2009-12-14 15:13:47 -0800 | [diff] [blame] | 2428 | ceph_msg_dump(msg); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2429 | } |
| 2430 | |
| 2431 | void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session, |
| 2432 | struct inode *inode, |
| 2433 | struct dentry *dentry, char action, |
| 2434 | u32 seq) |
| 2435 | { |
| 2436 | struct ceph_msg *msg; |
| 2437 | struct ceph_mds_lease *lease; |
| 2438 | int len = sizeof(*lease) + sizeof(u32); |
| 2439 | int dnamelen = 0; |
| 2440 | |
| 2441 | dout("lease_send_msg inode %p dentry %p %s to mds%d\n", |
| 2442 | inode, dentry, ceph_lease_op_name(action), session->s_mds); |
| 2443 | dnamelen = dentry->d_name.len; |
| 2444 | len += dnamelen; |
| 2445 | |
| 2446 | msg = ceph_msg_new(CEPH_MSG_CLIENT_LEASE, len, 0, 0, NULL); |
| 2447 | if (IS_ERR(msg)) |
| 2448 | return; |
| 2449 | lease = msg->front.iov_base; |
| 2450 | lease->action = action; |
| 2451 | lease->mask = cpu_to_le16(CEPH_LOCK_DN); |
| 2452 | lease->ino = cpu_to_le64(ceph_vino(inode).ino); |
| 2453 | lease->first = lease->last = cpu_to_le64(ceph_vino(inode).snap); |
| 2454 | lease->seq = cpu_to_le32(seq); |
| 2455 | put_unaligned_le32(dnamelen, lease + 1); |
| 2456 | memcpy((void *)(lease + 1) + 4, dentry->d_name.name, dnamelen); |
| 2457 | |
| 2458 | /* |
| 2459 | * if this is a preemptive lease RELEASE, no need to |
| 2460 | * flush request stream, since the actual request will |
| 2461 | * soon follow. |
| 2462 | */ |
| 2463 | msg->more_to_follow = (action == CEPH_MDS_LEASE_RELEASE); |
| 2464 | |
| 2465 | ceph_con_send(&session->s_con, msg); |
| 2466 | } |
| 2467 | |
| 2468 | /* |
| 2469 | * Preemptively release a lease we expect to invalidate anyway. |
| 2470 | * Pass @inode always, @dentry is optional. |
| 2471 | */ |
| 2472 | void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc, struct inode *inode, |
| 2473 | struct dentry *dentry, int mask) |
| 2474 | { |
| 2475 | struct ceph_dentry_info *di; |
| 2476 | struct ceph_mds_session *session; |
| 2477 | u32 seq; |
| 2478 | |
| 2479 | BUG_ON(inode == NULL); |
| 2480 | BUG_ON(dentry == NULL); |
| 2481 | BUG_ON(mask != CEPH_LOCK_DN); |
| 2482 | |
| 2483 | /* is dentry lease valid? */ |
| 2484 | spin_lock(&dentry->d_lock); |
| 2485 | di = ceph_dentry(dentry); |
| 2486 | if (!di || !di->lease_session || |
| 2487 | di->lease_session->s_mds < 0 || |
| 2488 | di->lease_gen != di->lease_session->s_cap_gen || |
| 2489 | !time_before(jiffies, dentry->d_time)) { |
| 2490 | dout("lease_release inode %p dentry %p -- " |
| 2491 | "no lease on %d\n", |
| 2492 | inode, dentry, mask); |
| 2493 | spin_unlock(&dentry->d_lock); |
| 2494 | return; |
| 2495 | } |
| 2496 | |
| 2497 | /* we do have a lease on this dentry; note mds and seq */ |
| 2498 | session = ceph_get_mds_session(di->lease_session); |
| 2499 | seq = di->lease_seq; |
| 2500 | __ceph_mdsc_drop_dentry_lease(dentry); |
| 2501 | spin_unlock(&dentry->d_lock); |
| 2502 | |
| 2503 | dout("lease_release inode %p dentry %p mask %d to mds%d\n", |
| 2504 | inode, dentry, mask, session->s_mds); |
| 2505 | ceph_mdsc_lease_send_msg(session, inode, dentry, |
| 2506 | CEPH_MDS_LEASE_RELEASE, seq); |
| 2507 | ceph_put_mds_session(session); |
| 2508 | } |
| 2509 | |
| 2510 | /* |
| 2511 | * drop all leases (and dentry refs) in preparation for umount |
| 2512 | */ |
| 2513 | static void drop_leases(struct ceph_mds_client *mdsc) |
| 2514 | { |
| 2515 | int i; |
| 2516 | |
| 2517 | dout("drop_leases\n"); |
| 2518 | mutex_lock(&mdsc->mutex); |
| 2519 | for (i = 0; i < mdsc->max_sessions; i++) { |
| 2520 | struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i); |
| 2521 | if (!s) |
| 2522 | continue; |
| 2523 | mutex_unlock(&mdsc->mutex); |
| 2524 | mutex_lock(&s->s_mutex); |
| 2525 | mutex_unlock(&s->s_mutex); |
| 2526 | ceph_put_mds_session(s); |
| 2527 | mutex_lock(&mdsc->mutex); |
| 2528 | } |
| 2529 | mutex_unlock(&mdsc->mutex); |
| 2530 | } |
| 2531 | |
| 2532 | |
| 2533 | |
| 2534 | /* |
| 2535 | * delayed work -- periodically trim expired leases, renew caps with mds |
| 2536 | */ |
| 2537 | static void schedule_delayed(struct ceph_mds_client *mdsc) |
| 2538 | { |
| 2539 | int delay = 5; |
| 2540 | unsigned hz = round_jiffies_relative(HZ * delay); |
| 2541 | schedule_delayed_work(&mdsc->delayed_work, hz); |
| 2542 | } |
| 2543 | |
| 2544 | static void delayed_work(struct work_struct *work) |
| 2545 | { |
| 2546 | int i; |
| 2547 | struct ceph_mds_client *mdsc = |
| 2548 | container_of(work, struct ceph_mds_client, delayed_work.work); |
| 2549 | int renew_interval; |
| 2550 | int renew_caps; |
| 2551 | |
| 2552 | dout("mdsc delayed_work\n"); |
Sage Weil | afcdaea | 2009-10-14 14:27:38 -0700 | [diff] [blame] | 2553 | ceph_check_delayed_caps(mdsc); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2554 | |
| 2555 | mutex_lock(&mdsc->mutex); |
| 2556 | renew_interval = mdsc->mdsmap->m_session_timeout >> 2; |
| 2557 | renew_caps = time_after_eq(jiffies, HZ*renew_interval + |
| 2558 | mdsc->last_renew_caps); |
| 2559 | if (renew_caps) |
| 2560 | mdsc->last_renew_caps = jiffies; |
| 2561 | |
| 2562 | for (i = 0; i < mdsc->max_sessions; i++) { |
| 2563 | struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i); |
| 2564 | if (s == NULL) |
| 2565 | continue; |
| 2566 | if (s->s_state == CEPH_MDS_SESSION_CLOSING) { |
| 2567 | dout("resending session close request for mds%d\n", |
| 2568 | s->s_mds); |
| 2569 | request_close_session(mdsc, s); |
| 2570 | ceph_put_mds_session(s); |
| 2571 | continue; |
| 2572 | } |
| 2573 | if (s->s_ttl && time_after(jiffies, s->s_ttl)) { |
| 2574 | if (s->s_state == CEPH_MDS_SESSION_OPEN) { |
| 2575 | s->s_state = CEPH_MDS_SESSION_HUNG; |
| 2576 | pr_info("mds%d hung\n", s->s_mds); |
| 2577 | } |
| 2578 | } |
| 2579 | if (s->s_state < CEPH_MDS_SESSION_OPEN) { |
| 2580 | /* this mds is failed or recovering, just wait */ |
| 2581 | ceph_put_mds_session(s); |
| 2582 | continue; |
| 2583 | } |
| 2584 | mutex_unlock(&mdsc->mutex); |
| 2585 | |
| 2586 | mutex_lock(&s->s_mutex); |
| 2587 | if (renew_caps) |
| 2588 | send_renew_caps(mdsc, s); |
| 2589 | else |
| 2590 | ceph_con_keepalive(&s->s_con); |
| 2591 | add_cap_releases(mdsc, s, -1); |
| 2592 | send_cap_releases(mdsc, s); |
| 2593 | mutex_unlock(&s->s_mutex); |
| 2594 | ceph_put_mds_session(s); |
| 2595 | |
| 2596 | mutex_lock(&mdsc->mutex); |
| 2597 | } |
| 2598 | mutex_unlock(&mdsc->mutex); |
| 2599 | |
| 2600 | schedule_delayed(mdsc); |
| 2601 | } |
| 2602 | |
| 2603 | |
Sage Weil | 5f44f14 | 2009-11-18 14:52:18 -0800 | [diff] [blame] | 2604 | int ceph_mdsc_init(struct ceph_mds_client *mdsc, struct ceph_client *client) |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2605 | { |
| 2606 | mdsc->client = client; |
| 2607 | mutex_init(&mdsc->mutex); |
| 2608 | mdsc->mdsmap = kzalloc(sizeof(*mdsc->mdsmap), GFP_NOFS); |
| 2609 | init_completion(&mdsc->safe_umount_waiters); |
| 2610 | init_completion(&mdsc->session_close_waiters); |
| 2611 | INIT_LIST_HEAD(&mdsc->waiting_for_map); |
| 2612 | mdsc->sessions = NULL; |
| 2613 | mdsc->max_sessions = 0; |
| 2614 | mdsc->stopping = 0; |
| 2615 | init_rwsem(&mdsc->snap_rwsem); |
Sage Weil | a105f00 | 2010-02-15 14:37:55 -0800 | [diff] [blame] | 2616 | mdsc->snap_realms = RB_ROOT; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2617 | INIT_LIST_HEAD(&mdsc->snap_empty); |
| 2618 | spin_lock_init(&mdsc->snap_empty_lock); |
| 2619 | mdsc->last_tid = 0; |
Sage Weil | 44ca18f | 2010-02-15 12:08:46 -0800 | [diff] [blame] | 2620 | mdsc->request_tree = RB_ROOT; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2621 | INIT_DELAYED_WORK(&mdsc->delayed_work, delayed_work); |
| 2622 | mdsc->last_renew_caps = jiffies; |
| 2623 | INIT_LIST_HEAD(&mdsc->cap_delay_list); |
| 2624 | spin_lock_init(&mdsc->cap_delay_lock); |
| 2625 | INIT_LIST_HEAD(&mdsc->snap_flush_list); |
| 2626 | spin_lock_init(&mdsc->snap_flush_lock); |
| 2627 | mdsc->cap_flush_seq = 0; |
| 2628 | INIT_LIST_HEAD(&mdsc->cap_dirty); |
| 2629 | mdsc->num_cap_flushing = 0; |
| 2630 | spin_lock_init(&mdsc->cap_dirty_lock); |
| 2631 | init_waitqueue_head(&mdsc->cap_flushing_wq); |
| 2632 | spin_lock_init(&mdsc->dentry_lru_lock); |
| 2633 | INIT_LIST_HEAD(&mdsc->dentry_lru); |
Sage Weil | 5f44f14 | 2009-11-18 14:52:18 -0800 | [diff] [blame] | 2634 | return 0; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2635 | } |
| 2636 | |
| 2637 | /* |
| 2638 | * Wait for safe replies on open mds requests. If we time out, drop |
| 2639 | * all requests from the tree to avoid dangling dentry refs. |
| 2640 | */ |
| 2641 | static void wait_requests(struct ceph_mds_client *mdsc) |
| 2642 | { |
| 2643 | struct ceph_mds_request *req; |
| 2644 | struct ceph_client *client = mdsc->client; |
| 2645 | |
| 2646 | mutex_lock(&mdsc->mutex); |
Sage Weil | 44ca18f | 2010-02-15 12:08:46 -0800 | [diff] [blame] | 2647 | if (__get_oldest_req(mdsc)) { |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2648 | mutex_unlock(&mdsc->mutex); |
Sage Weil | 44ca18f | 2010-02-15 12:08:46 -0800 | [diff] [blame] | 2649 | |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2650 | dout("wait_requests waiting for requests\n"); |
| 2651 | wait_for_completion_timeout(&mdsc->safe_umount_waiters, |
Sage Weil | 6b80518 | 2009-10-27 11:50:50 -0700 | [diff] [blame] | 2652 | client->mount_args->mount_timeout * HZ); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2653 | |
| 2654 | /* tear down remaining requests */ |
Sage Weil | 44ca18f | 2010-02-15 12:08:46 -0800 | [diff] [blame] | 2655 | mutex_lock(&mdsc->mutex); |
| 2656 | while ((req = __get_oldest_req(mdsc))) { |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2657 | dout("wait_requests timed out on tid %llu\n", |
| 2658 | req->r_tid); |
Sage Weil | 44ca18f | 2010-02-15 12:08:46 -0800 | [diff] [blame] | 2659 | __unregister_request(mdsc, req); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2660 | } |
| 2661 | } |
| 2662 | mutex_unlock(&mdsc->mutex); |
| 2663 | dout("wait_requests done\n"); |
| 2664 | } |
| 2665 | |
| 2666 | /* |
| 2667 | * called before mount is ro, and before dentries are torn down. |
| 2668 | * (hmm, does this still race with new lookups?) |
| 2669 | */ |
| 2670 | void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc) |
| 2671 | { |
| 2672 | dout("pre_umount\n"); |
| 2673 | mdsc->stopping = 1; |
| 2674 | |
| 2675 | drop_leases(mdsc); |
Sage Weil | afcdaea | 2009-10-14 14:27:38 -0700 | [diff] [blame] | 2676 | ceph_flush_dirty_caps(mdsc); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2677 | wait_requests(mdsc); |
| 2678 | } |
| 2679 | |
| 2680 | /* |
| 2681 | * wait for all write mds requests to flush. |
| 2682 | */ |
| 2683 | static void wait_unsafe_requests(struct ceph_mds_client *mdsc, u64 want_tid) |
| 2684 | { |
Sage Weil | 44ca18f | 2010-02-15 12:08:46 -0800 | [diff] [blame] | 2685 | struct ceph_mds_request *req = NULL; |
| 2686 | struct rb_node *n; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2687 | |
| 2688 | mutex_lock(&mdsc->mutex); |
| 2689 | dout("wait_unsafe_requests want %lld\n", want_tid); |
Sage Weil | 44ca18f | 2010-02-15 12:08:46 -0800 | [diff] [blame] | 2690 | req = __get_oldest_req(mdsc); |
| 2691 | while (req && req->r_tid <= want_tid) { |
| 2692 | if ((req->r_op & CEPH_MDS_OP_WRITE)) { |
| 2693 | /* write op */ |
| 2694 | ceph_mdsc_get_request(req); |
| 2695 | mutex_unlock(&mdsc->mutex); |
| 2696 | dout("wait_unsafe_requests wait on %llu (want %llu)\n", |
| 2697 | req->r_tid, want_tid); |
| 2698 | wait_for_completion(&req->r_safe_completion); |
| 2699 | mutex_lock(&mdsc->mutex); |
| 2700 | n = rb_next(&req->r_node); |
| 2701 | ceph_mdsc_put_request(req); |
| 2702 | } else { |
| 2703 | n = rb_next(&req->r_node); |
| 2704 | } |
| 2705 | if (!n) |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2706 | break; |
Sage Weil | 44ca18f | 2010-02-15 12:08:46 -0800 | [diff] [blame] | 2707 | req = rb_entry(n, struct ceph_mds_request, r_node); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2708 | } |
| 2709 | mutex_unlock(&mdsc->mutex); |
| 2710 | dout("wait_unsafe_requests done\n"); |
| 2711 | } |
| 2712 | |
| 2713 | void ceph_mdsc_sync(struct ceph_mds_client *mdsc) |
| 2714 | { |
| 2715 | u64 want_tid, want_flush; |
| 2716 | |
| 2717 | dout("sync\n"); |
| 2718 | mutex_lock(&mdsc->mutex); |
| 2719 | want_tid = mdsc->last_tid; |
| 2720 | want_flush = mdsc->cap_flush_seq; |
| 2721 | mutex_unlock(&mdsc->mutex); |
| 2722 | dout("sync want tid %lld flush_seq %lld\n", want_tid, want_flush); |
| 2723 | |
Sage Weil | afcdaea | 2009-10-14 14:27:38 -0700 | [diff] [blame] | 2724 | ceph_flush_dirty_caps(mdsc); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2725 | |
| 2726 | wait_unsafe_requests(mdsc, want_tid); |
| 2727 | wait_event(mdsc->cap_flushing_wq, check_cap_flush(mdsc, want_flush)); |
| 2728 | } |
| 2729 | |
| 2730 | |
| 2731 | /* |
| 2732 | * called after sb is ro. |
| 2733 | */ |
| 2734 | void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc) |
| 2735 | { |
| 2736 | struct ceph_mds_session *session; |
| 2737 | int i; |
| 2738 | int n; |
| 2739 | struct ceph_client *client = mdsc->client; |
Sage Weil | 6b80518 | 2009-10-27 11:50:50 -0700 | [diff] [blame] | 2740 | unsigned long started, timeout = client->mount_args->mount_timeout * HZ; |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2741 | |
| 2742 | dout("close_sessions\n"); |
| 2743 | |
| 2744 | mutex_lock(&mdsc->mutex); |
| 2745 | |
| 2746 | /* close sessions */ |
| 2747 | started = jiffies; |
| 2748 | while (time_before(jiffies, started + timeout)) { |
| 2749 | dout("closing sessions\n"); |
| 2750 | n = 0; |
| 2751 | for (i = 0; i < mdsc->max_sessions; i++) { |
| 2752 | session = __ceph_lookup_mds_session(mdsc, i); |
| 2753 | if (!session) |
| 2754 | continue; |
| 2755 | mutex_unlock(&mdsc->mutex); |
| 2756 | mutex_lock(&session->s_mutex); |
| 2757 | __close_session(mdsc, session); |
| 2758 | mutex_unlock(&session->s_mutex); |
| 2759 | ceph_put_mds_session(session); |
| 2760 | mutex_lock(&mdsc->mutex); |
| 2761 | n++; |
| 2762 | } |
| 2763 | if (n == 0) |
| 2764 | break; |
| 2765 | |
| 2766 | if (client->mount_state == CEPH_MOUNT_SHUTDOWN) |
| 2767 | break; |
| 2768 | |
| 2769 | dout("waiting for sessions to close\n"); |
| 2770 | mutex_unlock(&mdsc->mutex); |
| 2771 | wait_for_completion_timeout(&mdsc->session_close_waiters, |
| 2772 | timeout); |
| 2773 | mutex_lock(&mdsc->mutex); |
| 2774 | } |
| 2775 | |
| 2776 | /* tear down remaining sessions */ |
| 2777 | for (i = 0; i < mdsc->max_sessions; i++) { |
| 2778 | if (mdsc->sessions[i]) { |
| 2779 | session = get_session(mdsc->sessions[i]); |
Sage Weil | 2600d2d | 2010-02-22 15:12:16 -0800 | [diff] [blame] | 2780 | __unregister_session(mdsc, session); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2781 | mutex_unlock(&mdsc->mutex); |
| 2782 | mutex_lock(&session->s_mutex); |
| 2783 | remove_session_caps(session); |
| 2784 | mutex_unlock(&session->s_mutex); |
| 2785 | ceph_put_mds_session(session); |
| 2786 | mutex_lock(&mdsc->mutex); |
| 2787 | } |
| 2788 | } |
| 2789 | |
| 2790 | WARN_ON(!list_empty(&mdsc->cap_delay_list)); |
| 2791 | |
| 2792 | mutex_unlock(&mdsc->mutex); |
| 2793 | |
| 2794 | ceph_cleanup_empty_realms(mdsc); |
| 2795 | |
| 2796 | cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */ |
| 2797 | |
| 2798 | dout("stopped\n"); |
| 2799 | } |
| 2800 | |
| 2801 | void ceph_mdsc_stop(struct ceph_mds_client *mdsc) |
| 2802 | { |
| 2803 | dout("stop\n"); |
| 2804 | cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */ |
| 2805 | if (mdsc->mdsmap) |
| 2806 | ceph_mdsmap_destroy(mdsc->mdsmap); |
| 2807 | kfree(mdsc->sessions); |
| 2808 | } |
| 2809 | |
| 2810 | |
| 2811 | /* |
| 2812 | * handle mds map update. |
| 2813 | */ |
| 2814 | void ceph_mdsc_handle_map(struct ceph_mds_client *mdsc, struct ceph_msg *msg) |
| 2815 | { |
| 2816 | u32 epoch; |
| 2817 | u32 maplen; |
| 2818 | void *p = msg->front.iov_base; |
| 2819 | void *end = p + msg->front.iov_len; |
| 2820 | struct ceph_mdsmap *newmap, *oldmap; |
| 2821 | struct ceph_fsid fsid; |
| 2822 | int err = -EINVAL; |
| 2823 | |
| 2824 | ceph_decode_need(&p, end, sizeof(fsid)+2*sizeof(u32), bad); |
| 2825 | ceph_decode_copy(&p, &fsid, sizeof(fsid)); |
Sage Weil | 0743304 | 2009-11-18 16:50:41 -0800 | [diff] [blame] | 2826 | if (ceph_check_fsid(mdsc->client, &fsid) < 0) |
| 2827 | return; |
Sage Weil | c89136e | 2009-10-14 09:59:09 -0700 | [diff] [blame] | 2828 | epoch = ceph_decode_32(&p); |
| 2829 | maplen = ceph_decode_32(&p); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2830 | dout("handle_map epoch %u len %d\n", epoch, (int)maplen); |
| 2831 | |
| 2832 | /* do we need it? */ |
| 2833 | ceph_monc_got_mdsmap(&mdsc->client->monc, epoch); |
| 2834 | mutex_lock(&mdsc->mutex); |
| 2835 | if (mdsc->mdsmap && epoch <= mdsc->mdsmap->m_epoch) { |
| 2836 | dout("handle_map epoch %u <= our %u\n", |
| 2837 | epoch, mdsc->mdsmap->m_epoch); |
| 2838 | mutex_unlock(&mdsc->mutex); |
| 2839 | return; |
| 2840 | } |
| 2841 | |
| 2842 | newmap = ceph_mdsmap_decode(&p, end); |
| 2843 | if (IS_ERR(newmap)) { |
| 2844 | err = PTR_ERR(newmap); |
| 2845 | goto bad_unlock; |
| 2846 | } |
| 2847 | |
| 2848 | /* swap into place */ |
| 2849 | if (mdsc->mdsmap) { |
| 2850 | oldmap = mdsc->mdsmap; |
| 2851 | mdsc->mdsmap = newmap; |
| 2852 | check_new_map(mdsc, newmap, oldmap); |
| 2853 | ceph_mdsmap_destroy(oldmap); |
| 2854 | } else { |
| 2855 | mdsc->mdsmap = newmap; /* first mds map */ |
| 2856 | } |
| 2857 | mdsc->client->sb->s_maxbytes = mdsc->mdsmap->m_max_file_size; |
| 2858 | |
| 2859 | __wake_requests(mdsc, &mdsc->waiting_for_map); |
| 2860 | |
| 2861 | mutex_unlock(&mdsc->mutex); |
| 2862 | schedule_delayed(mdsc); |
| 2863 | return; |
| 2864 | |
| 2865 | bad_unlock: |
| 2866 | mutex_unlock(&mdsc->mutex); |
| 2867 | bad: |
| 2868 | pr_err("error decoding mdsmap %d\n", err); |
| 2869 | return; |
| 2870 | } |
| 2871 | |
| 2872 | static struct ceph_connection *con_get(struct ceph_connection *con) |
| 2873 | { |
| 2874 | struct ceph_mds_session *s = con->private; |
| 2875 | |
| 2876 | if (get_session(s)) { |
Sage Weil | 2600d2d | 2010-02-22 15:12:16 -0800 | [diff] [blame] | 2877 | dout("mdsc con_get %p ok (%d)\n", s, atomic_read(&s->s_ref)); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2878 | return con; |
| 2879 | } |
| 2880 | dout("mdsc con_get %p FAIL\n", s); |
| 2881 | return NULL; |
| 2882 | } |
| 2883 | |
| 2884 | static void con_put(struct ceph_connection *con) |
| 2885 | { |
| 2886 | struct ceph_mds_session *s = con->private; |
| 2887 | |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2888 | ceph_put_mds_session(s); |
Sage Weil | 2600d2d | 2010-02-22 15:12:16 -0800 | [diff] [blame] | 2889 | dout("mdsc con_put %p (%d)\n", s, atomic_read(&s->s_ref)); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2890 | } |
| 2891 | |
| 2892 | /* |
| 2893 | * if the client is unresponsive for long enough, the mds will kill |
| 2894 | * the session entirely. |
| 2895 | */ |
| 2896 | static void peer_reset(struct ceph_connection *con) |
| 2897 | { |
| 2898 | struct ceph_mds_session *s = con->private; |
| 2899 | |
| 2900 | pr_err("mds%d gave us the boot. IMPLEMENT RECONNECT.\n", |
| 2901 | s->s_mds); |
| 2902 | } |
| 2903 | |
| 2904 | static void dispatch(struct ceph_connection *con, struct ceph_msg *msg) |
| 2905 | { |
| 2906 | struct ceph_mds_session *s = con->private; |
| 2907 | struct ceph_mds_client *mdsc = s->s_mdsc; |
| 2908 | int type = le16_to_cpu(msg->hdr.type); |
| 2909 | |
Sage Weil | 2600d2d | 2010-02-22 15:12:16 -0800 | [diff] [blame] | 2910 | mutex_lock(&mdsc->mutex); |
| 2911 | if (__verify_registered_session(mdsc, s) < 0) { |
| 2912 | mutex_unlock(&mdsc->mutex); |
| 2913 | goto out; |
| 2914 | } |
| 2915 | mutex_unlock(&mdsc->mutex); |
| 2916 | |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2917 | switch (type) { |
| 2918 | case CEPH_MSG_MDS_MAP: |
| 2919 | ceph_mdsc_handle_map(mdsc, msg); |
| 2920 | break; |
| 2921 | case CEPH_MSG_CLIENT_SESSION: |
| 2922 | handle_session(s, msg); |
| 2923 | break; |
| 2924 | case CEPH_MSG_CLIENT_REPLY: |
| 2925 | handle_reply(s, msg); |
| 2926 | break; |
| 2927 | case CEPH_MSG_CLIENT_REQUEST_FORWARD: |
Sage Weil | 2600d2d | 2010-02-22 15:12:16 -0800 | [diff] [blame] | 2928 | handle_forward(mdsc, s, msg); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2929 | break; |
| 2930 | case CEPH_MSG_CLIENT_CAPS: |
| 2931 | ceph_handle_caps(s, msg); |
| 2932 | break; |
| 2933 | case CEPH_MSG_CLIENT_SNAP: |
Sage Weil | 2600d2d | 2010-02-22 15:12:16 -0800 | [diff] [blame] | 2934 | ceph_handle_snap(mdsc, s, msg); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2935 | break; |
| 2936 | case CEPH_MSG_CLIENT_LEASE: |
Sage Weil | 2600d2d | 2010-02-22 15:12:16 -0800 | [diff] [blame] | 2937 | handle_lease(mdsc, s, msg); |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2938 | break; |
| 2939 | |
| 2940 | default: |
| 2941 | pr_err("received unknown message type %d %s\n", type, |
| 2942 | ceph_msg_type_name(type)); |
| 2943 | } |
Sage Weil | 2600d2d | 2010-02-22 15:12:16 -0800 | [diff] [blame] | 2944 | out: |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 2945 | ceph_msg_put(msg); |
| 2946 | } |
| 2947 | |
Sage Weil | 4e7a5dc | 2009-11-18 16:19:57 -0800 | [diff] [blame] | 2948 | /* |
| 2949 | * authentication |
| 2950 | */ |
| 2951 | static int get_authorizer(struct ceph_connection *con, |
| 2952 | void **buf, int *len, int *proto, |
| 2953 | void **reply_buf, int *reply_len, int force_new) |
| 2954 | { |
| 2955 | struct ceph_mds_session *s = con->private; |
| 2956 | struct ceph_mds_client *mdsc = s->s_mdsc; |
| 2957 | struct ceph_auth_client *ac = mdsc->client->monc.auth; |
| 2958 | int ret = 0; |
| 2959 | |
| 2960 | if (force_new && s->s_authorizer) { |
| 2961 | ac->ops->destroy_authorizer(ac, s->s_authorizer); |
| 2962 | s->s_authorizer = NULL; |
| 2963 | } |
| 2964 | if (s->s_authorizer == NULL) { |
| 2965 | if (ac->ops->create_authorizer) { |
| 2966 | ret = ac->ops->create_authorizer( |
| 2967 | ac, CEPH_ENTITY_TYPE_MDS, |
| 2968 | &s->s_authorizer, |
| 2969 | &s->s_authorizer_buf, |
| 2970 | &s->s_authorizer_buf_len, |
| 2971 | &s->s_authorizer_reply_buf, |
| 2972 | &s->s_authorizer_reply_buf_len); |
| 2973 | if (ret) |
| 2974 | return ret; |
| 2975 | } |
| 2976 | } |
| 2977 | |
| 2978 | *proto = ac->protocol; |
| 2979 | *buf = s->s_authorizer_buf; |
| 2980 | *len = s->s_authorizer_buf_len; |
| 2981 | *reply_buf = s->s_authorizer_reply_buf; |
| 2982 | *reply_len = s->s_authorizer_reply_buf_len; |
| 2983 | return 0; |
| 2984 | } |
| 2985 | |
| 2986 | |
| 2987 | static int verify_authorizer_reply(struct ceph_connection *con, int len) |
| 2988 | { |
| 2989 | struct ceph_mds_session *s = con->private; |
| 2990 | struct ceph_mds_client *mdsc = s->s_mdsc; |
| 2991 | struct ceph_auth_client *ac = mdsc->client->monc.auth; |
| 2992 | |
| 2993 | return ac->ops->verify_authorizer_reply(ac, s->s_authorizer, len); |
| 2994 | } |
| 2995 | |
Sage Weil | 9bd2e6f | 2010-02-02 16:21:06 -0800 | [diff] [blame] | 2996 | static int invalidate_authorizer(struct ceph_connection *con) |
| 2997 | { |
| 2998 | struct ceph_mds_session *s = con->private; |
| 2999 | struct ceph_mds_client *mdsc = s->s_mdsc; |
| 3000 | struct ceph_auth_client *ac = mdsc->client->monc.auth; |
| 3001 | |
| 3002 | if (ac->ops->invalidate_authorizer) |
| 3003 | ac->ops->invalidate_authorizer(ac, CEPH_ENTITY_TYPE_MDS); |
| 3004 | |
| 3005 | return ceph_monc_validate_auth(&mdsc->client->monc); |
| 3006 | } |
| 3007 | |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 3008 | const static struct ceph_connection_operations mds_con_ops = { |
| 3009 | .get = con_get, |
| 3010 | .put = con_put, |
| 3011 | .dispatch = dispatch, |
Sage Weil | 4e7a5dc | 2009-11-18 16:19:57 -0800 | [diff] [blame] | 3012 | .get_authorizer = get_authorizer, |
| 3013 | .verify_authorizer_reply = verify_authorizer_reply, |
Sage Weil | 9bd2e6f | 2010-02-02 16:21:06 -0800 | [diff] [blame] | 3014 | .invalidate_authorizer = invalidate_authorizer, |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 3015 | .peer_reset = peer_reset, |
Sage Weil | 2f2dc05 | 2009-10-06 11:31:09 -0700 | [diff] [blame] | 3016 | }; |
| 3017 | |
| 3018 | |
| 3019 | |
| 3020 | |
| 3021 | /* eof */ |