1 2 #include <linux/ceph/ceph_debug.h> 3 4 #include <linux/backing-dev.h> 5 #include <linux/ctype.h> 6 #include <linux/fs.h> 7 #include <linux/inet.h> 8 #include <linux/in6.h> 9 #include <linux/module.h> 10 #include <linux/mount.h> 11 #include <linux/parser.h> 12 #include <linux/sched.h> 13 #include <linux/seq_file.h> 14 #include <linux/slab.h> 15 #include <linux/statfs.h> 16 #include <linux/string.h> 17 18 #include "super.h" 19 #include "mds_client.h" 20 #include "cache.h" 21 22 #include <linux/ceph/ceph_features.h> 23 #include <linux/ceph/decode.h> 24 #include <linux/ceph/mon_client.h> 25 #include <linux/ceph/auth.h> 26 #include <linux/ceph/debugfs.h> 27 28 /* 29 * Ceph superblock operations 30 * 31 * Handle the basics of mounting, unmounting. 32 */ 33 34 /* 35 * super ops 36 */ 37 static void ceph_put_super(struct super_block *s) 38 { 39 struct ceph_fs_client *fsc = ceph_sb_to_client(s); 40 41 dout("put_super\n"); 42 ceph_mdsc_close_sessions(fsc->mdsc); 43 } 44 45 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf) 46 { 47 struct ceph_fs_client *fsc = ceph_inode_to_client(d_inode(dentry)); 48 struct ceph_mon_client *monc = &fsc->client->monc; 49 struct ceph_statfs st; 50 u64 fsid; 51 int err; 52 u64 data_pool; 53 54 if (fsc->mdsc->mdsmap->m_num_data_pg_pools == 1) { 55 data_pool = fsc->mdsc->mdsmap->m_data_pg_pools[0]; 56 } else { 57 data_pool = CEPH_NOPOOL; 58 } 59 60 dout("statfs\n"); 61 err = ceph_monc_do_statfs(monc, data_pool, &st); 62 if (err < 0) 63 return err; 64 65 /* fill in kstatfs */ 66 buf->f_type = CEPH_SUPER_MAGIC; /* ?? */ 67 68 /* 69 * express utilization in terms of large blocks to avoid 70 * overflow on 32-bit machines. 71 * 72 * NOTE: for the time being, we make bsize == frsize to humor 73 * not-yet-ancient versions of glibc that are broken. 74 * Someday, we will probably want to report a real block 75 * size... whatever that may mean for a network file system! 76 */ 77 buf->f_bsize = 1 << CEPH_BLOCK_SHIFT; 78 buf->f_frsize = 1 << CEPH_BLOCK_SHIFT; 79 80 /* 81 * By default use root quota for stats; fallback to overall filesystem 82 * usage if using 'noquotadf' mount option or if the root dir doesn't 83 * have max_bytes quota set. 84 */ 85 if (ceph_test_mount_opt(fsc, NOQUOTADF) || 86 !ceph_quota_update_statfs(fsc, buf)) { 87 buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10); 88 buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10); 89 buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10); 90 } 91 92 buf->f_files = le64_to_cpu(st.num_objects); 93 buf->f_ffree = -1; 94 buf->f_namelen = NAME_MAX; 95 96 /* Must convert the fsid, for consistent values across arches */ 97 mutex_lock(&monc->mutex); 98 fsid = le64_to_cpu(*(__le64 *)(&monc->monmap->fsid)) ^ 99 le64_to_cpu(*((__le64 *)&monc->monmap->fsid + 1)); 100 mutex_unlock(&monc->mutex); 101 102 buf->f_fsid.val[0] = fsid & 0xffffffff; 103 buf->f_fsid.val[1] = fsid >> 32; 104 105 return 0; 106 } 107 108 109 static int ceph_sync_fs(struct super_block *sb, int wait) 110 { 111 struct ceph_fs_client *fsc = ceph_sb_to_client(sb); 112 113 if (!wait) { 114 dout("sync_fs (non-blocking)\n"); 115 ceph_flush_dirty_caps(fsc->mdsc); 116 dout("sync_fs (non-blocking) done\n"); 117 return 0; 118 } 119 120 dout("sync_fs (blocking)\n"); 121 ceph_osdc_sync(&fsc->client->osdc); 122 ceph_mdsc_sync(fsc->mdsc); 123 dout("sync_fs (blocking) done\n"); 124 return 0; 125 } 126 127 /* 128 * mount options 129 */ 130 enum { 131 Opt_wsize, 132 Opt_rsize, 133 Opt_rasize, 134 Opt_caps_wanted_delay_min, 135 Opt_caps_wanted_delay_max, 136 Opt_readdir_max_entries, 137 Opt_readdir_max_bytes, 138 Opt_congestion_kb, 139 Opt_last_int, 140 /* int args above */ 141 Opt_snapdirname, 142 Opt_mds_namespace, 143 Opt_fscache_uniq, 144 Opt_last_string, 145 /* string args above */ 146 Opt_dirstat, 147 Opt_nodirstat, 148 Opt_rbytes, 149 Opt_norbytes, 150 Opt_asyncreaddir, 151 Opt_noasyncreaddir, 152 Opt_dcache, 153 Opt_nodcache, 154 Opt_ino32, 155 Opt_noino32, 156 Opt_fscache, 157 Opt_nofscache, 158 Opt_poolperm, 159 Opt_nopoolperm, 160 Opt_require_active_mds, 161 Opt_norequire_active_mds, 162 #ifdef CONFIG_CEPH_FS_POSIX_ACL 163 Opt_acl, 164 #endif 165 Opt_noacl, 166 Opt_quotadf, 167 Opt_noquotadf, 168 }; 169 170 static match_table_t fsopt_tokens = { 171 {Opt_wsize, "wsize=%d"}, 172 {Opt_rsize, "rsize=%d"}, 173 {Opt_rasize, "rasize=%d"}, 174 {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"}, 175 {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"}, 176 {Opt_readdir_max_entries, "readdir_max_entries=%d"}, 177 {Opt_readdir_max_bytes, "readdir_max_bytes=%d"}, 178 {Opt_congestion_kb, "write_congestion_kb=%d"}, 179 /* int args above */ 180 {Opt_snapdirname, "snapdirname=%s"}, 181 {Opt_mds_namespace, "mds_namespace=%s"}, 182 {Opt_fscache_uniq, "fsc=%s"}, 183 /* string args above */ 184 {Opt_dirstat, "dirstat"}, 185 {Opt_nodirstat, "nodirstat"}, 186 {Opt_rbytes, "rbytes"}, 187 {Opt_norbytes, "norbytes"}, 188 {Opt_asyncreaddir, "asyncreaddir"}, 189 {Opt_noasyncreaddir, "noasyncreaddir"}, 190 {Opt_dcache, "dcache"}, 191 {Opt_nodcache, "nodcache"}, 192 {Opt_ino32, "ino32"}, 193 {Opt_noino32, "noino32"}, 194 {Opt_fscache, "fsc"}, 195 {Opt_nofscache, "nofsc"}, 196 {Opt_poolperm, "poolperm"}, 197 {Opt_nopoolperm, "nopoolperm"}, 198 {Opt_require_active_mds, "require_active_mds"}, 199 {Opt_norequire_active_mds, "norequire_active_mds"}, 200 #ifdef CONFIG_CEPH_FS_POSIX_ACL 201 {Opt_acl, "acl"}, 202 #endif 203 {Opt_noacl, "noacl"}, 204 {Opt_quotadf, "quotadf"}, 205 {Opt_noquotadf, "noquotadf"}, 206 {-1, NULL} 207 }; 208 209 static int parse_fsopt_token(char *c, void *private) 210 { 211 struct ceph_mount_options *fsopt = private; 212 substring_t argstr[MAX_OPT_ARGS]; 213 int token, intval, ret; 214 215 token = match_token((char *)c, fsopt_tokens, argstr); 216 if (token < 0) 217 return -EINVAL; 218 219 if (token < Opt_last_int) { 220 ret = match_int(&argstr[0], &intval); 221 if (ret < 0) { 222 pr_err("bad mount option arg (not int) " 223 "at '%s'\n", c); 224 return ret; 225 } 226 dout("got int token %d val %d\n", token, intval); 227 } else if (token > Opt_last_int && token < Opt_last_string) { 228 dout("got string token %d val %s\n", token, 229 argstr[0].from); 230 } else { 231 dout("got token %d\n", token); 232 } 233 234 switch (token) { 235 case Opt_snapdirname: 236 kfree(fsopt->snapdir_name); 237 fsopt->snapdir_name = kstrndup(argstr[0].from, 238 argstr[0].to-argstr[0].from, 239 GFP_KERNEL); 240 if (!fsopt->snapdir_name) 241 return -ENOMEM; 242 break; 243 case Opt_mds_namespace: 244 kfree(fsopt->mds_namespace); 245 fsopt->mds_namespace = kstrndup(argstr[0].from, 246 argstr[0].to-argstr[0].from, 247 GFP_KERNEL); 248 if (!fsopt->mds_namespace) 249 return -ENOMEM; 250 break; 251 case Opt_fscache_uniq: 252 kfree(fsopt->fscache_uniq); 253 fsopt->fscache_uniq = kstrndup(argstr[0].from, 254 argstr[0].to-argstr[0].from, 255 GFP_KERNEL); 256 if (!fsopt->fscache_uniq) 257 return -ENOMEM; 258 fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE; 259 break; 260 /* misc */ 261 case Opt_wsize: 262 if (intval < (int)PAGE_SIZE || intval > CEPH_MAX_WRITE_SIZE) 263 return -EINVAL; 264 fsopt->wsize = ALIGN(intval, PAGE_SIZE); 265 break; 266 case Opt_rsize: 267 if (intval < (int)PAGE_SIZE || intval > CEPH_MAX_READ_SIZE) 268 return -EINVAL; 269 fsopt->rsize = ALIGN(intval, PAGE_SIZE); 270 break; 271 case Opt_rasize: 272 if (intval < 0) 273 return -EINVAL; 274 fsopt->rasize = ALIGN(intval, PAGE_SIZE); 275 break; 276 case Opt_caps_wanted_delay_min: 277 if (intval < 1) 278 return -EINVAL; 279 fsopt->caps_wanted_delay_min = intval; 280 break; 281 case Opt_caps_wanted_delay_max: 282 if (intval < 1) 283 return -EINVAL; 284 fsopt->caps_wanted_delay_max = intval; 285 break; 286 case Opt_readdir_max_entries: 287 if (intval < 1) 288 return -EINVAL; 289 fsopt->max_readdir = intval; 290 break; 291 case Opt_readdir_max_bytes: 292 if (intval < (int)PAGE_SIZE && intval != 0) 293 return -EINVAL; 294 fsopt->max_readdir_bytes = intval; 295 break; 296 case Opt_congestion_kb: 297 if (intval < 1024) /* at least 1M */ 298 return -EINVAL; 299 fsopt->congestion_kb = intval; 300 break; 301 case Opt_dirstat: 302 fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT; 303 break; 304 case Opt_nodirstat: 305 fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT; 306 break; 307 case Opt_rbytes: 308 fsopt->flags |= CEPH_MOUNT_OPT_RBYTES; 309 break; 310 case Opt_norbytes: 311 fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES; 312 break; 313 case Opt_asyncreaddir: 314 fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR; 315 break; 316 case Opt_noasyncreaddir: 317 fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR; 318 break; 319 case Opt_dcache: 320 fsopt->flags |= CEPH_MOUNT_OPT_DCACHE; 321 break; 322 case Opt_nodcache: 323 fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE; 324 break; 325 case Opt_ino32: 326 fsopt->flags |= CEPH_MOUNT_OPT_INO32; 327 break; 328 case Opt_noino32: 329 fsopt->flags &= ~CEPH_MOUNT_OPT_INO32; 330 break; 331 case Opt_fscache: 332 fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE; 333 kfree(fsopt->fscache_uniq); 334 fsopt->fscache_uniq = NULL; 335 break; 336 case Opt_nofscache: 337 fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE; 338 kfree(fsopt->fscache_uniq); 339 fsopt->fscache_uniq = NULL; 340 break; 341 case Opt_poolperm: 342 fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM; 343 break; 344 case Opt_nopoolperm: 345 fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM; 346 break; 347 case Opt_require_active_mds: 348 fsopt->flags &= ~CEPH_MOUNT_OPT_MOUNTWAIT; 349 break; 350 case Opt_norequire_active_mds: 351 fsopt->flags |= CEPH_MOUNT_OPT_MOUNTWAIT; 352 break; 353 case Opt_quotadf: 354 fsopt->flags &= ~CEPH_MOUNT_OPT_NOQUOTADF; 355 break; 356 case Opt_noquotadf: 357 fsopt->flags |= CEPH_MOUNT_OPT_NOQUOTADF; 358 break; 359 #ifdef CONFIG_CEPH_FS_POSIX_ACL 360 case Opt_acl: 361 fsopt->sb_flags |= SB_POSIXACL; 362 break; 363 #endif 364 case Opt_noacl: 365 fsopt->sb_flags &= ~SB_POSIXACL; 366 break; 367 default: 368 BUG_ON(token); 369 } 370 return 0; 371 } 372 373 static void destroy_mount_options(struct ceph_mount_options *args) 374 { 375 dout("destroy_mount_options %p\n", args); 376 kfree(args->snapdir_name); 377 kfree(args->mds_namespace); 378 kfree(args->server_path); 379 kfree(args->fscache_uniq); 380 kfree(args); 381 } 382 383 static int strcmp_null(const char *s1, const char *s2) 384 { 385 if (!s1 && !s2) 386 return 0; 387 if (s1 && !s2) 388 return -1; 389 if (!s1 && s2) 390 return 1; 391 return strcmp(s1, s2); 392 } 393 394 static int compare_mount_options(struct ceph_mount_options *new_fsopt, 395 struct ceph_options *new_opt, 396 struct ceph_fs_client *fsc) 397 { 398 struct ceph_mount_options *fsopt1 = new_fsopt; 399 struct ceph_mount_options *fsopt2 = fsc->mount_options; 400 int ofs = offsetof(struct ceph_mount_options, snapdir_name); 401 int ret; 402 403 ret = memcmp(fsopt1, fsopt2, ofs); 404 if (ret) 405 return ret; 406 407 ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name); 408 if (ret) 409 return ret; 410 ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace); 411 if (ret) 412 return ret; 413 ret = strcmp_null(fsopt1->server_path, fsopt2->server_path); 414 if (ret) 415 return ret; 416 ret = strcmp_null(fsopt1->fscache_uniq, fsopt2->fscache_uniq); 417 if (ret) 418 return ret; 419 420 return ceph_compare_options(new_opt, fsc->client); 421 } 422 423 static int parse_mount_options(struct ceph_mount_options **pfsopt, 424 struct ceph_options **popt, 425 int flags, char *options, 426 const char *dev_name) 427 { 428 struct ceph_mount_options *fsopt; 429 const char *dev_name_end; 430 int err; 431 432 if (!dev_name || !*dev_name) 433 return -EINVAL; 434 435 fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL); 436 if (!fsopt) 437 return -ENOMEM; 438 439 dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name); 440 441 fsopt->sb_flags = flags; 442 fsopt->flags = CEPH_MOUNT_OPT_DEFAULT; 443 444 fsopt->wsize = CEPH_MAX_WRITE_SIZE; 445 fsopt->rsize = CEPH_MAX_READ_SIZE; 446 fsopt->rasize = CEPH_RASIZE_DEFAULT; 447 fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL); 448 if (!fsopt->snapdir_name) { 449 err = -ENOMEM; 450 goto out; 451 } 452 453 fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT; 454 fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT; 455 fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT; 456 fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT; 457 fsopt->congestion_kb = default_congestion_kb(); 458 459 /* 460 * Distinguish the server list from the path in "dev_name". 461 * Internally we do not include the leading '/' in the path. 462 * 463 * "dev_name" will look like: 464 * <server_spec>[,<server_spec>...]:[<path>] 465 * where 466 * <server_spec> is <ip>[:<port>] 467 * <path> is optional, but if present must begin with '/' 468 */ 469 dev_name_end = strchr(dev_name, '/'); 470 if (dev_name_end) { 471 if (strlen(dev_name_end) > 1) { 472 fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL); 473 if (!fsopt->server_path) { 474 err = -ENOMEM; 475 goto out; 476 } 477 } 478 } else { 479 dev_name_end = dev_name + strlen(dev_name); 480 } 481 err = -EINVAL; 482 dev_name_end--; /* back up to ':' separator */ 483 if (dev_name_end < dev_name || *dev_name_end != ':') { 484 pr_err("device name is missing path (no : separator in %s)\n", 485 dev_name); 486 goto out; 487 } 488 dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name); 489 if (fsopt->server_path) 490 dout("server path '%s'\n", fsopt->server_path); 491 492 *popt = ceph_parse_options(options, dev_name, dev_name_end, 493 parse_fsopt_token, (void *)fsopt); 494 if (IS_ERR(*popt)) { 495 err = PTR_ERR(*popt); 496 goto out; 497 } 498 499 /* success */ 500 *pfsopt = fsopt; 501 return 0; 502 503 out: 504 destroy_mount_options(fsopt); 505 return err; 506 } 507 508 /** 509 * ceph_show_options - Show mount options in /proc/mounts 510 * @m: seq_file to write to 511 * @root: root of that (sub)tree 512 */ 513 static int ceph_show_options(struct seq_file *m, struct dentry *root) 514 { 515 struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb); 516 struct ceph_mount_options *fsopt = fsc->mount_options; 517 size_t pos; 518 int ret; 519 520 /* a comma between MNT/MS and client options */ 521 seq_putc(m, ','); 522 pos = m->count; 523 524 ret = ceph_print_client_options(m, fsc->client); 525 if (ret) 526 return ret; 527 528 /* retract our comma if no client options */ 529 if (m->count == pos) 530 m->count--; 531 532 if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT) 533 seq_puts(m, ",dirstat"); 534 if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES)) 535 seq_puts(m, ",rbytes"); 536 if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR) 537 seq_puts(m, ",noasyncreaddir"); 538 if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0) 539 seq_puts(m, ",nodcache"); 540 if (fsopt->flags & CEPH_MOUNT_OPT_INO32) 541 seq_puts(m, ",ino32"); 542 if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) { 543 seq_show_option(m, "fsc", fsopt->fscache_uniq); 544 } 545 if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM) 546 seq_puts(m, ",nopoolperm"); 547 if (fsopt->flags & CEPH_MOUNT_OPT_NOQUOTADF) 548 seq_puts(m, ",noquotadf"); 549 550 #ifdef CONFIG_CEPH_FS_POSIX_ACL 551 if (fsopt->sb_flags & SB_POSIXACL) 552 seq_puts(m, ",acl"); 553 else 554 seq_puts(m, ",noacl"); 555 #endif 556 557 if (fsopt->mds_namespace) 558 seq_show_option(m, "mds_namespace", fsopt->mds_namespace); 559 if (fsopt->wsize != CEPH_MAX_WRITE_SIZE) 560 seq_printf(m, ",wsize=%d", fsopt->wsize); 561 if (fsopt->rsize != CEPH_MAX_READ_SIZE) 562 seq_printf(m, ",rsize=%d", fsopt->rsize); 563 if (fsopt->rasize != CEPH_RASIZE_DEFAULT) 564 seq_printf(m, ",rasize=%d", fsopt->rasize); 565 if (fsopt->congestion_kb != default_congestion_kb()) 566 seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb); 567 if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT) 568 seq_printf(m, ",caps_wanted_delay_min=%d", 569 fsopt->caps_wanted_delay_min); 570 if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT) 571 seq_printf(m, ",caps_wanted_delay_max=%d", 572 fsopt->caps_wanted_delay_max); 573 if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT) 574 seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir); 575 if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT) 576 seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes); 577 if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT)) 578 seq_show_option(m, "snapdirname", fsopt->snapdir_name); 579 580 return 0; 581 } 582 583 /* 584 * handle any mon messages the standard library doesn't understand. 585 * return error if we don't either. 586 */ 587 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg) 588 { 589 struct ceph_fs_client *fsc = client->private; 590 int type = le16_to_cpu(msg->hdr.type); 591 592 switch (type) { 593 case CEPH_MSG_MDS_MAP: 594 ceph_mdsc_handle_mdsmap(fsc->mdsc, msg); 595 return 0; 596 case CEPH_MSG_FS_MAP_USER: 597 ceph_mdsc_handle_fsmap(fsc->mdsc, msg); 598 return 0; 599 default: 600 return -1; 601 } 602 } 603 604 /* 605 * create a new fs client 606 */ 607 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt, 608 struct ceph_options *opt) 609 { 610 struct ceph_fs_client *fsc; 611 int page_count; 612 size_t size; 613 int err = -ENOMEM; 614 615 fsc = kzalloc(sizeof(*fsc), GFP_KERNEL); 616 if (!fsc) 617 return ERR_PTR(-ENOMEM); 618 619 fsc->client = ceph_create_client(opt, fsc); 620 if (IS_ERR(fsc->client)) { 621 err = PTR_ERR(fsc->client); 622 goto fail; 623 } 624 625 fsc->client->extra_mon_dispatch = extra_mon_dispatch; 626 fsc->client->osdc.abort_on_full = true; 627 628 if (!fsopt->mds_namespace) { 629 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP, 630 0, true); 631 } else { 632 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP, 633 0, false); 634 } 635 636 fsc->mount_options = fsopt; 637 638 fsc->sb = NULL; 639 fsc->mount_state = CEPH_MOUNT_MOUNTING; 640 641 atomic_long_set(&fsc->writeback_count, 0); 642 643 err = -ENOMEM; 644 /* 645 * The number of concurrent works can be high but they don't need 646 * to be processed in parallel, limit concurrency. 647 */ 648 fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1); 649 if (!fsc->wb_wq) 650 goto fail_client; 651 fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1); 652 if (!fsc->pg_inv_wq) 653 goto fail_wb_wq; 654 fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1); 655 if (!fsc->trunc_wq) 656 goto fail_pg_inv_wq; 657 658 /* set up mempools */ 659 err = -ENOMEM; 660 page_count = fsc->mount_options->wsize >> PAGE_SHIFT; 661 size = sizeof (struct page *) * (page_count ? page_count : 1); 662 fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size); 663 if (!fsc->wb_pagevec_pool) 664 goto fail_trunc_wq; 665 666 /* caps */ 667 fsc->min_caps = fsopt->max_readdir; 668 669 return fsc; 670 671 fail_trunc_wq: 672 destroy_workqueue(fsc->trunc_wq); 673 fail_pg_inv_wq: 674 destroy_workqueue(fsc->pg_inv_wq); 675 fail_wb_wq: 676 destroy_workqueue(fsc->wb_wq); 677 fail_client: 678 ceph_destroy_client(fsc->client); 679 fail: 680 kfree(fsc); 681 return ERR_PTR(err); 682 } 683 684 static void flush_fs_workqueues(struct ceph_fs_client *fsc) 685 { 686 flush_workqueue(fsc->wb_wq); 687 flush_workqueue(fsc->pg_inv_wq); 688 flush_workqueue(fsc->trunc_wq); 689 } 690 691 static void destroy_fs_client(struct ceph_fs_client *fsc) 692 { 693 dout("destroy_fs_client %p\n", fsc); 694 695 destroy_workqueue(fsc->wb_wq); 696 destroy_workqueue(fsc->pg_inv_wq); 697 destroy_workqueue(fsc->trunc_wq); 698 699 mempool_destroy(fsc->wb_pagevec_pool); 700 701 destroy_mount_options(fsc->mount_options); 702 703 ceph_destroy_client(fsc->client); 704 705 kfree(fsc); 706 dout("destroy_fs_client %p done\n", fsc); 707 } 708 709 /* 710 * caches 711 */ 712 struct kmem_cache *ceph_inode_cachep; 713 struct kmem_cache *ceph_cap_cachep; 714 struct kmem_cache *ceph_cap_flush_cachep; 715 struct kmem_cache *ceph_dentry_cachep; 716 struct kmem_cache *ceph_file_cachep; 717 struct kmem_cache *ceph_dir_file_cachep; 718 719 static void ceph_inode_init_once(void *foo) 720 { 721 struct ceph_inode_info *ci = foo; 722 inode_init_once(&ci->vfs_inode); 723 } 724 725 static int __init init_caches(void) 726 { 727 int error = -ENOMEM; 728 729 ceph_inode_cachep = kmem_cache_create("ceph_inode_info", 730 sizeof(struct ceph_inode_info), 731 __alignof__(struct ceph_inode_info), 732 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD| 733 SLAB_ACCOUNT, ceph_inode_init_once); 734 if (!ceph_inode_cachep) 735 return -ENOMEM; 736 737 ceph_cap_cachep = KMEM_CACHE(ceph_cap, SLAB_MEM_SPREAD); 738 if (!ceph_cap_cachep) 739 goto bad_cap; 740 ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush, 741 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD); 742 if (!ceph_cap_flush_cachep) 743 goto bad_cap_flush; 744 745 ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info, 746 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD); 747 if (!ceph_dentry_cachep) 748 goto bad_dentry; 749 750 ceph_file_cachep = KMEM_CACHE(ceph_file_info, SLAB_MEM_SPREAD); 751 if (!ceph_file_cachep) 752 goto bad_file; 753 754 ceph_dir_file_cachep = KMEM_CACHE(ceph_dir_file_info, SLAB_MEM_SPREAD); 755 if (!ceph_dir_file_cachep) 756 goto bad_dir_file; 757 758 error = ceph_fscache_register(); 759 if (error) 760 goto bad_fscache; 761 762 return 0; 763 764 bad_fscache: 765 kmem_cache_destroy(ceph_dir_file_cachep); 766 bad_dir_file: 767 kmem_cache_destroy(ceph_file_cachep); 768 bad_file: 769 kmem_cache_destroy(ceph_dentry_cachep); 770 bad_dentry: 771 kmem_cache_destroy(ceph_cap_flush_cachep); 772 bad_cap_flush: 773 kmem_cache_destroy(ceph_cap_cachep); 774 bad_cap: 775 kmem_cache_destroy(ceph_inode_cachep); 776 return error; 777 } 778 779 static void destroy_caches(void) 780 { 781 /* 782 * Make sure all delayed rcu free inodes are flushed before we 783 * destroy cache. 784 */ 785 rcu_barrier(); 786 787 kmem_cache_destroy(ceph_inode_cachep); 788 kmem_cache_destroy(ceph_cap_cachep); 789 kmem_cache_destroy(ceph_cap_flush_cachep); 790 kmem_cache_destroy(ceph_dentry_cachep); 791 kmem_cache_destroy(ceph_file_cachep); 792 kmem_cache_destroy(ceph_dir_file_cachep); 793 794 ceph_fscache_unregister(); 795 } 796 797 798 /* 799 * ceph_umount_begin - initiate forced umount. Tear down down the 800 * mount, skipping steps that may hang while waiting for server(s). 801 */ 802 static void ceph_umount_begin(struct super_block *sb) 803 { 804 struct ceph_fs_client *fsc = ceph_sb_to_client(sb); 805 806 dout("ceph_umount_begin - starting forced umount\n"); 807 if (!fsc) 808 return; 809 fsc->mount_state = CEPH_MOUNT_SHUTDOWN; 810 ceph_osdc_abort_requests(&fsc->client->osdc, -EIO); 811 ceph_mdsc_force_umount(fsc->mdsc); 812 return; 813 } 814 815 static const struct super_operations ceph_super_ops = { 816 .alloc_inode = ceph_alloc_inode, 817 .destroy_inode = ceph_destroy_inode, 818 .write_inode = ceph_write_inode, 819 .drop_inode = ceph_drop_inode, 820 .sync_fs = ceph_sync_fs, 821 .put_super = ceph_put_super, 822 .show_options = ceph_show_options, 823 .statfs = ceph_statfs, 824 .umount_begin = ceph_umount_begin, 825 }; 826 827 /* 828 * Bootstrap mount by opening the root directory. Note the mount 829 * @started time from caller, and time out if this takes too long. 830 */ 831 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc, 832 const char *path, 833 unsigned long started) 834 { 835 struct ceph_mds_client *mdsc = fsc->mdsc; 836 struct ceph_mds_request *req = NULL; 837 int err; 838 struct dentry *root; 839 840 /* open dir */ 841 dout("open_root_inode opening '%s'\n", path); 842 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS); 843 if (IS_ERR(req)) 844 return ERR_CAST(req); 845 req->r_path1 = kstrdup(path, GFP_NOFS); 846 if (!req->r_path1) { 847 root = ERR_PTR(-ENOMEM); 848 goto out; 849 } 850 851 req->r_ino1.ino = CEPH_INO_ROOT; 852 req->r_ino1.snap = CEPH_NOSNAP; 853 req->r_started = started; 854 req->r_timeout = fsc->client->options->mount_timeout; 855 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE); 856 req->r_num_caps = 2; 857 err = ceph_mdsc_do_request(mdsc, NULL, req); 858 if (err == 0) { 859 struct inode *inode = req->r_target_inode; 860 req->r_target_inode = NULL; 861 dout("open_root_inode success\n"); 862 root = d_make_root(inode); 863 if (!root) { 864 root = ERR_PTR(-ENOMEM); 865 goto out; 866 } 867 dout("open_root_inode success, root dentry is %p\n", root); 868 } else { 869 root = ERR_PTR(err); 870 } 871 out: 872 ceph_mdsc_put_request(req); 873 return root; 874 } 875 876 877 878 879 /* 880 * mount: join the ceph cluster, and open root directory. 881 */ 882 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc) 883 { 884 int err; 885 unsigned long started = jiffies; /* note the start time */ 886 struct dentry *root; 887 888 dout("mount start %p\n", fsc); 889 mutex_lock(&fsc->client->mount_mutex); 890 891 if (!fsc->sb->s_root) { 892 const char *path; 893 err = __ceph_open_session(fsc->client, started); 894 if (err < 0) 895 goto out; 896 897 /* setup fscache */ 898 if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) { 899 err = ceph_fscache_register_fs(fsc); 900 if (err < 0) 901 goto out; 902 } 903 904 if (!fsc->mount_options->server_path) { 905 path = ""; 906 dout("mount opening path \\t\n"); 907 } else { 908 path = fsc->mount_options->server_path + 1; 909 dout("mount opening path %s\n", path); 910 } 911 912 err = ceph_fs_debugfs_init(fsc); 913 if (err < 0) 914 goto out; 915 916 root = open_root_dentry(fsc, path, started); 917 if (IS_ERR(root)) { 918 err = PTR_ERR(root); 919 goto out; 920 } 921 fsc->sb->s_root = dget(root); 922 } else { 923 root = dget(fsc->sb->s_root); 924 } 925 926 fsc->mount_state = CEPH_MOUNT_MOUNTED; 927 dout("mount success\n"); 928 mutex_unlock(&fsc->client->mount_mutex); 929 return root; 930 931 out: 932 mutex_unlock(&fsc->client->mount_mutex); 933 return ERR_PTR(err); 934 } 935 936 static int ceph_set_super(struct super_block *s, void *data) 937 { 938 struct ceph_fs_client *fsc = data; 939 int ret; 940 941 dout("set_super %p data %p\n", s, data); 942 943 s->s_flags = fsc->mount_options->sb_flags; 944 s->s_maxbytes = 1ULL << 40; /* temp value until we get mdsmap */ 945 946 s->s_xattr = ceph_xattr_handlers; 947 s->s_fs_info = fsc; 948 fsc->sb = s; 949 950 s->s_op = &ceph_super_ops; 951 s->s_d_op = &ceph_dentry_ops; 952 s->s_export_op = &ceph_export_ops; 953 954 s->s_time_gran = 1000; /* 1000 ns == 1 us */ 955 956 ret = set_anon_super(s, NULL); /* what is that second arg for? */ 957 if (ret != 0) 958 goto fail; 959 960 return ret; 961 962 fail: 963 s->s_fs_info = NULL; 964 fsc->sb = NULL; 965 return ret; 966 } 967 968 /* 969 * share superblock if same fs AND options 970 */ 971 static int ceph_compare_super(struct super_block *sb, void *data) 972 { 973 struct ceph_fs_client *new = data; 974 struct ceph_mount_options *fsopt = new->mount_options; 975 struct ceph_options *opt = new->client->options; 976 struct ceph_fs_client *other = ceph_sb_to_client(sb); 977 978 dout("ceph_compare_super %p\n", sb); 979 980 if (compare_mount_options(fsopt, opt, other)) { 981 dout("monitor(s)/mount options don't match\n"); 982 return 0; 983 } 984 if ((opt->flags & CEPH_OPT_FSID) && 985 ceph_fsid_compare(&opt->fsid, &other->client->fsid)) { 986 dout("fsid doesn't match\n"); 987 return 0; 988 } 989 if (fsopt->sb_flags != other->mount_options->sb_flags) { 990 dout("flags differ\n"); 991 return 0; 992 } 993 return 1; 994 } 995 996 /* 997 * construct our own bdi so we can control readahead, etc. 998 */ 999 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0); 1000 1001 static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc) 1002 { 1003 int err; 1004 1005 err = super_setup_bdi_name(sb, "ceph-%ld", 1006 atomic_long_inc_return(&bdi_seq)); 1007 if (err) 1008 return err; 1009 1010 /* set ra_pages based on rasize mount option? */ 1011 sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT; 1012 1013 /* set io_pages based on max osd read size */ 1014 sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT; 1015 1016 return 0; 1017 } 1018 1019 static struct dentry *ceph_mount(struct file_system_type *fs_type, 1020 int flags, const char *dev_name, void *data) 1021 { 1022 struct super_block *sb; 1023 struct ceph_fs_client *fsc; 1024 struct dentry *res; 1025 int err; 1026 int (*compare_super)(struct super_block *, void *) = ceph_compare_super; 1027 struct ceph_mount_options *fsopt = NULL; 1028 struct ceph_options *opt = NULL; 1029 1030 dout("ceph_mount\n"); 1031 1032 #ifdef CONFIG_CEPH_FS_POSIX_ACL 1033 flags |= SB_POSIXACL; 1034 #endif 1035 err = parse_mount_options(&fsopt, &opt, flags, data, dev_name); 1036 if (err < 0) { 1037 res = ERR_PTR(err); 1038 goto out_final; 1039 } 1040 1041 /* create client (which we may/may not use) */ 1042 fsc = create_fs_client(fsopt, opt); 1043 if (IS_ERR(fsc)) { 1044 res = ERR_CAST(fsc); 1045 destroy_mount_options(fsopt); 1046 ceph_destroy_options(opt); 1047 goto out_final; 1048 } 1049 1050 err = ceph_mdsc_init(fsc); 1051 if (err < 0) { 1052 res = ERR_PTR(err); 1053 goto out; 1054 } 1055 1056 if (ceph_test_opt(fsc->client, NOSHARE)) 1057 compare_super = NULL; 1058 sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc); 1059 if (IS_ERR(sb)) { 1060 res = ERR_CAST(sb); 1061 goto out; 1062 } 1063 1064 if (ceph_sb_to_client(sb) != fsc) { 1065 ceph_mdsc_destroy(fsc); 1066 destroy_fs_client(fsc); 1067 fsc = ceph_sb_to_client(sb); 1068 dout("get_sb got existing client %p\n", fsc); 1069 } else { 1070 dout("get_sb using new client %p\n", fsc); 1071 err = ceph_setup_bdi(sb, fsc); 1072 if (err < 0) { 1073 res = ERR_PTR(err); 1074 goto out_splat; 1075 } 1076 } 1077 1078 res = ceph_real_mount(fsc); 1079 if (IS_ERR(res)) 1080 goto out_splat; 1081 dout("root %p inode %p ino %llx.%llx\n", res, 1082 d_inode(res), ceph_vinop(d_inode(res))); 1083 return res; 1084 1085 out_splat: 1086 ceph_mdsc_close_sessions(fsc->mdsc); 1087 deactivate_locked_super(sb); 1088 goto out_final; 1089 1090 out: 1091 ceph_mdsc_destroy(fsc); 1092 destroy_fs_client(fsc); 1093 out_final: 1094 dout("ceph_mount fail %ld\n", PTR_ERR(res)); 1095 return res; 1096 } 1097 1098 static void ceph_kill_sb(struct super_block *s) 1099 { 1100 struct ceph_fs_client *fsc = ceph_sb_to_client(s); 1101 dev_t dev = s->s_dev; 1102 1103 dout("kill_sb %p\n", s); 1104 1105 ceph_mdsc_pre_umount(fsc->mdsc); 1106 flush_fs_workqueues(fsc); 1107 1108 generic_shutdown_super(s); 1109 1110 fsc->client->extra_mon_dispatch = NULL; 1111 ceph_fs_debugfs_cleanup(fsc); 1112 1113 ceph_fscache_unregister_fs(fsc); 1114 1115 ceph_mdsc_destroy(fsc); 1116 1117 destroy_fs_client(fsc); 1118 free_anon_bdev(dev); 1119 } 1120 1121 static struct file_system_type ceph_fs_type = { 1122 .owner = THIS_MODULE, 1123 .name = "ceph", 1124 .mount = ceph_mount, 1125 .kill_sb = ceph_kill_sb, 1126 .fs_flags = FS_RENAME_DOES_D_MOVE, 1127 }; 1128 MODULE_ALIAS_FS("ceph"); 1129 1130 static int __init init_ceph(void) 1131 { 1132 int ret = init_caches(); 1133 if (ret) 1134 goto out; 1135 1136 ceph_flock_init(); 1137 ceph_xattr_init(); 1138 ret = register_filesystem(&ceph_fs_type); 1139 if (ret) 1140 goto out_xattr; 1141 1142 pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL); 1143 1144 return 0; 1145 1146 out_xattr: 1147 ceph_xattr_exit(); 1148 destroy_caches(); 1149 out: 1150 return ret; 1151 } 1152 1153 static void __exit exit_ceph(void) 1154 { 1155 dout("exit_ceph\n"); 1156 unregister_filesystem(&ceph_fs_type); 1157 ceph_xattr_exit(); 1158 destroy_caches(); 1159 } 1160 1161 module_init(init_ceph); 1162 module_exit(exit_ceph); 1163 1164 MODULE_AUTHOR("Sage Weil <sage@newdream.net>"); 1165 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>"); 1166 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>"); 1167 MODULE_DESCRIPTION("Ceph filesystem for Linux"); 1168 MODULE_LICENSE("GPL"); 1169