net/ceph/ceph_common.c

   1
   2 #include <linux/ceph/ceph_debug.h>
   3 #include <linux/backing-dev.h>
   4 #include <linux/ctype.h>
   5 #include <linux/fs.h>
   6 #include <linux/inet.h>
   7 #include <linux/in6.h>
   8 #include <linux/key.h>
   9 #include <keys/ceph-type.h>
  10 #include <linux/module.h>
  11 #include <linux/mount.h>
  12 #include <linux/parser.h>
  13 #include <linux/sched.h>
  14 #include <linux/seq_file.h>
  15 #include <linux/slab.h>
  16 #include <linux/statfs.h>
  17 #include <linux/string.h>
  18
  19
  20 #include <linux/ceph/libceph.h>
  21 #include <linux/ceph/debugfs.h>
  22 #include <linux/ceph/decode.h>
  23 #include <linux/ceph/mon_client.h>
  24 #include <linux/ceph/auth.h>
  25 #include "crypto.h"
  26
  27
  28
  29 /*
  30  * find filename portion of a path (/foo/bar/baz -> baz)
  31  */
  32 const char *ceph_file_part(const char *s, int len)
  33 {
  34         const char *e = s + len;
  35
  36         while (e != s && *(e-1) != '/')
  37                 e--;
  38         return e;
  39 }
  40 EXPORT_SYMBOL(ceph_file_part);
  41
  42 const char *ceph_msg_type_name(int type)
  43 {
  44         switch (type) {
  45         case CEPH_MSG_SHUTDOWN: return "shutdown";
  46         case CEPH_MSG_PING: return "ping";
  47         case CEPH_MSG_AUTH: return "auth";
  48         case CEPH_MSG_AUTH_REPLY: return "auth_reply";
  49         case CEPH_MSG_MON_MAP: return "mon_map";
  50         case CEPH_MSG_MON_GET_MAP: return "mon_get_map";
  51         case CEPH_MSG_MON_SUBSCRIBE: return "mon_subscribe";
  52         case CEPH_MSG_MON_SUBSCRIBE_ACK: return "mon_subscribe_ack";
  53         case CEPH_MSG_STATFS: return "statfs";
  54         case CEPH_MSG_STATFS_REPLY: return "statfs_reply";
  55         case CEPH_MSG_MDS_MAP: return "mds_map";
  56         case CEPH_MSG_CLIENT_SESSION: return "client_session";
  57         case CEPH_MSG_CLIENT_RECONNECT: return "client_reconnect";
  58         case CEPH_MSG_CLIENT_REQUEST: return "client_request";
  59         case CEPH_MSG_CLIENT_REQUEST_FORWARD: return "client_request_forward";
  60         case CEPH_MSG_CLIENT_REPLY: return "client_reply";
  61         case CEPH_MSG_CLIENT_CAPS: return "client_caps";
  62         case CEPH_MSG_CLIENT_CAPRELEASE: return "client_cap_release";
  63         case CEPH_MSG_CLIENT_SNAP: return "client_snap";
  64         case CEPH_MSG_CLIENT_LEASE: return "client_lease";
  65         case CEPH_MSG_OSD_MAP: return "osd_map";
  66         case CEPH_MSG_OSD_OP: return "osd_op";
  67         case CEPH_MSG_OSD_OPREPLY: return "osd_opreply";
  68         case CEPH_MSG_WATCH_NOTIFY: return "watch_notify";
  69         default: return "unknown";
  70         }
  71 }
  72 EXPORT_SYMBOL(ceph_msg_type_name);
  73
  74 /*
  75  * Initially learn our fsid, or verify an fsid matches.
  76  */
  77 int ceph_check_fsid(struct ceph_client *client, struct ceph_fsid *fsid)
  78 {
  79         if (client->have_fsid) {
  80                 if (ceph_fsid_compare(&client->fsid, fsid)) {
  81                         pr_err("bad fsid, had %pU got %pU",
  82                                &client->fsid, fsid);
  83                         return -1;
  84                 }
  85         } else {
  86                 pr_info("client%lld fsid %pU\n", ceph_client_id(client), fsid);
  87                 memcpy(&client->fsid, fsid, sizeof(*fsid));
  88                 ceph_debugfs_client_init(client);
  89                 client->have_fsid = true;
  90         }
  91         return 0;
  92 }
  93 EXPORT_SYMBOL(ceph_check_fsid);
  94
  95 static int strcmp_null(const char *s1, const char *s2)
  96 {
  97         if (!s1 && !s2)
  98                 return 0;
  99         if (s1 && !s2)
 100                 return -1;
 101         if (!s1 && s2)
 102                 return 1;
 103         return strcmp(s1, s2);
 104 }
 105
 106 int ceph_compare_options(struct ceph_options *new_opt,
 107                          struct ceph_client *client)
 108 {
 109         struct ceph_options *opt1 = new_opt;
 110         struct ceph_options *opt2 = client->options;
 111         int ofs = offsetof(struct ceph_options, mon_addr);
 112         int i;
 113         int ret;
 114
 115         ret = memcmp(opt1, opt2, ofs);
 116         if (ret)
 117                 return ret;
 118
 119         ret = strcmp_null(opt1->name, opt2->name);
 120         if (ret)
 121                 return ret;
 122
 123         if (opt1->key && !opt2->key)
 124                 return -1;
 125         if (!opt1->key && opt2->key)
 126                 return 1;
 127         if (opt1->key && opt2->key) {
 128                 if (opt1->key->type != opt2->key->type)
 129                         return -1;
 130                 if (opt1->key->created.tv_sec != opt2->key->created.tv_sec)
 131                         return -1;
 132                 if (opt1->key->created.tv_nsec != opt2->key->created.tv_nsec)
 133                         return -1;
 134                 if (opt1->key->len != opt2->key->len)
 135                         return -1;
 136                 if (opt1->key->key && !opt2->key->key)
 137                         return -1;
 138                 if (!opt1->key->key && opt2->key->key)
 139                         return 1;
 140                 if (opt1->key->key && opt2->key->key) {
 141                         ret = memcmp(opt1->key->key, opt2->key->key, opt1->key->len);
 142                         if (ret)
 143                                 return ret;
 144                 }
 145         }
 146
 147         /* any matching mon ip implies a match */
 148         for (i = 0; i < opt1->num_mon; i++) {
 149                 if (ceph_monmap_contains(client->monc.monmap,
 150                                  &opt1->mon_addr[i]))
 151                         return 0;
 152         }
 153         return -1;
 154 }
 155 EXPORT_SYMBOL(ceph_compare_options);
 156
 157
 158 static int parse_fsid(const char *str, struct ceph_fsid *fsid)
 159 {
 160         int i = 0;
 161         char tmp[3];
 162         int err = -EINVAL;
 163         int d;
 164
 165         dout("parse_fsid '%s'\n", str);
 166         tmp[2] = 0;
 167         while (*str && i < 16) {
 168                 if (ispunct(*str)) {
 169                         str++;
 170                         continue;
 171                 }
 172                 if (!isxdigit(str[0]) || !isxdigit(str[1]))
 173                         break;
 174                 tmp[0] = str[0];
 175                 tmp[1] = str[1];
 176                 if (sscanf(tmp, "%x", &d) < 1)
 177                         break;
 178                 fsid->fsid[i] = d & 0xff;
 179                 i++;
 180                 str += 2;
 181         }
 182
 183         if (i == 16)
 184                 err = 0;
 185         dout("parse_fsid ret %d got fsid %pU", err, fsid);
 186         return err;
 187 }
 188
 189 /*
 190  * ceph options
 191  */
 192 enum {
 193         Opt_osdtimeout,
 194         Opt_osdkeepalivetimeout,
 195         Opt_mount_timeout,
 196         Opt_osd_idle_ttl,
 197         Opt_last_int,
 198         /* int args above */
 199         Opt_fsid,
 200         Opt_name,
 201         Opt_secret,
 202         Opt_key,
 203         Opt_ip,
 204         Opt_last_string,
 205         /* string args above */
 206         Opt_noshare,
 207         Opt_nocrc,
 208 };
 209
 210 static match_table_t opt_tokens = {
 211         {Opt_osdtimeout, "osdtimeout=%d"},
 212         {Opt_osdkeepalivetimeout, "osdkeepalive=%d"},
 213         {Opt_mount_timeout, "mount_timeout=%d"},
 214         {Opt_osd_idle_ttl, "osd_idle_ttl=%d"},
 215         /* int args above */
 216         {Opt_fsid, "fsid=%s"},
 217         {Opt_name, "name=%s"},
 218         {Opt_secret, "secret=%s"},
 219         {Opt_key, "key=%s"},
 220         {Opt_ip, "ip=%s"},
 221         /* string args above */
 222         {Opt_noshare, "noshare"},
 223         {Opt_nocrc, "nocrc"},
 224         {-1, NULL}
 225 };
 226
 227 void ceph_destroy_options(struct ceph_options *opt)
 228 {
 229         dout("destroy_options %p\n", opt);
 230         kfree(opt->name);
 231         if (opt->key) {
 232                 ceph_crypto_key_destroy(opt->key);
 233                 kfree(opt->key);
 234         }
 235         kfree(opt);
 236 }
 237 EXPORT_SYMBOL(ceph_destroy_options);
 238
 239 /* get secret from key store */
 240 static int get_secret(struct ceph_crypto_key *dst, const char *name) {
 241         struct key *ukey;
 242         int key_err;
 243         int err = 0;
 244         struct ceph_crypto_key *ckey;
 245
 246         ukey = request_key(&key_type_ceph, name, NULL);
 247         if (!ukey || IS_ERR(ukey)) {
 248                 /* request_key errors don't map nicely to mount(2)
 249                    errors; don't even try, but still printk */
 250                 key_err = PTR_ERR(ukey);
 251                 switch (key_err) {
 252                 case -ENOKEY:
 253                         pr_warning("ceph: Mount failed due to key not found: %s\n", name);
 254                         break;
 255                 case -EKEYEXPIRED:
 256                         pr_warning("ceph: Mount failed due to expired key: %s\n", name);
 257                         break;
 258                 case -EKEYREVOKED:
 259                         pr_warning("ceph: Mount failed due to revoked key: %s\n", name);
 260                         break;
 261                 default:
 262                         pr_warning("ceph: Mount failed due to unknown key error"
 263                                " %d: %s\n", key_err, name);
 264                 }
 265                 err = -EPERM;
 266                 goto out;
 267         }
 268
 269         ckey = ukey->payload.data;
 270         err = ceph_crypto_key_clone(dst, ckey);
 271         if (err)
 272                 goto out_key;
 273         /* pass through, err is 0 */
 274
 275 out_key:
 276         key_put(ukey);
 277 out:
 278         return err;
 279 }
 280
 281 int ceph_parse_options(struct ceph_options **popt, char *options,
 282                        const char *dev_name, const char *dev_name_end,
 283                        int (*parse_extra_token)(char *c, void *private),
 284                        void *private)
 285 {
 286         struct ceph_options *opt;
 287         const char *c;
 288         int err = -ENOMEM;
 289         substring_t argstr[MAX_OPT_ARGS];
 290
 291         opt = kzalloc(sizeof(*opt), GFP_KERNEL);
 292         if (!opt)
 293                 return err;
 294         opt->mon_addr = kcalloc(CEPH_MAX_MON, sizeof(*opt->mon_addr),
 295                                 GFP_KERNEL);
 296         if (!opt->mon_addr)
 297                 goto out;
 298
 299         dout("parse_options %p options '%s' dev_name '%s'\n", opt, options,
 300              dev_name);
 301
 302         /* start with defaults */
 303         opt->flags = CEPH_OPT_DEFAULT;
 304         opt->osd_timeout = CEPH_OSD_TIMEOUT_DEFAULT;
 305         opt->osd_keepalive_timeout = CEPH_OSD_KEEPALIVE_DEFAULT;
 306         opt->mount_timeout = CEPH_MOUNT_TIMEOUT_DEFAULT; /* seconds */
 307         opt->osd_idle_ttl = CEPH_OSD_IDLE_TTL_DEFAULT;   /* seconds */
 308
 309         /* get mon ip(s) */
 310         /* ip1[:port1][,ip2[:port2]...] */
 311         err = ceph_parse_ips(dev_name, dev_name_end, opt->mon_addr,
 312                              CEPH_MAX_MON, &opt->num_mon);
 313         if (err < 0)
 314                 goto out;
 315
 316         /* parse mount options */
 317         while ((c = strsep(&options, ",")) != NULL) {
 318                 int token, intval, ret;
 319                 if (!*c)
 320                         continue;
 321                 err = -EINVAL;
 322                 token = match_token((char *)c, opt_tokens, argstr);
 323                 if (token < 0 && parse_extra_token) {
 324                         /* extra? */
 325                         err = parse_extra_token((char *)c, private);
 326                         if (err < 0) {
 327                                 pr_err("bad option at '%s'\n", c);
 328                                 goto out;
 329                         }
 330                         continue;
 331                 }
 332                 if (token < Opt_last_int) {
 333                         ret = match_int(&argstr[0], &intval);
 334                         if (ret < 0) {
 335                                 pr_err("bad mount option arg (not int) "
 336                                        "at '%s'\n", c);
 337                                 continue;
 338                         }
 339                         dout("got int token %d val %d\n", token, intval);
 340                 } else if (token > Opt_last_int && token < Opt_last_string) {
 341                         dout("got string token %d val %s\n", token,
 342                              argstr[0].from);
 343                 } else {
 344                         dout("got token %d\n", token);
 345                 }
 346                 switch (token) {
 347                 case Opt_ip:
 348                         err = ceph_parse_ips(argstr[0].from,
 349                                              argstr[0].to,
 350                                              &opt->my_addr,
 351                                              1, NULL);
 352                         if (err < 0)
 353                                 goto out;
 354                         opt->flags |= CEPH_OPT_MYIP;
 355                         break;
 356
 357                 case Opt_fsid:
 358                         err = parse_fsid(argstr[0].from, &opt->fsid);
 359                         if (err == 0)
 360                                 opt->flags |= CEPH_OPT_FSID;
 361                         break;
 362                 case Opt_name:
 363                         opt->name = kstrndup(argstr[0].from,
 364                                               argstr[0].to-argstr[0].from,
 365                                               GFP_KERNEL);
 366                         break;
 367                 case Opt_secret:
 368                         opt->key = kzalloc(sizeof(*opt->key), GFP_KERNEL);
 369                         if (!opt->key) {
 370                                 err = -ENOMEM;
 371                                 goto out;
 372                         }
 373                         err = ceph_crypto_key_unarmor(opt->key, argstr[0].from);
 374                         if (err < 0)
 375                                 goto out;
 376                         break;
 377                 case Opt_key:
 378                         opt->key = kzalloc(sizeof(*opt->key), GFP_KERNEL);
 379                         if (!opt->key) {
 380                                 err = -ENOMEM;
 381                                 goto out;
 382                         }
 383                         err = get_secret(opt->key, argstr[0].from);
 384                         if (err < 0)
 385                                 goto out;
 386                         break;
 387
 388                         /* misc */
 389                 case Opt_osdtimeout:
 390                         opt->osd_timeout = intval;
 391                         break;
 392                 case Opt_osdkeepalivetimeout:
 393                         opt->osd_keepalive_timeout = intval;
 394                         break;
 395                 case Opt_osd_idle_ttl:
 396                         opt->osd_idle_ttl = intval;
 397                         break;
 398                 case Opt_mount_timeout:
 399                         opt->mount_timeout = intval;
 400                         break;
 401
 402                 case Opt_noshare:
 403                         opt->flags |= CEPH_OPT_NOSHARE;
 404                         break;
 405
 406                 case Opt_nocrc:
 407                         opt->flags |= CEPH_OPT_NOCRC;
 408                         break;
 409
 410                 default:
 411                         BUG_ON(token);
 412                 }
 413         }
 414
 415         /* success */
 416         *popt = opt;
 417         return 0;
 418
 419 out:
 420         ceph_destroy_options(opt);
 421         return err;
 422 }
 423 EXPORT_SYMBOL(ceph_parse_options);
 424
 425 u64 ceph_client_id(struct ceph_client *client)
 426 {
 427         return client->monc.auth->global_id;
 428 }
 429 EXPORT_SYMBOL(ceph_client_id);
 430
 431 /*
 432  * create a fresh client instance
 433  */
 434 struct ceph_client *ceph_create_client(struct ceph_options *opt, void *private)
 435 {
 436         struct ceph_client *client;
 437         int err = -ENOMEM;
 438
 439         client = kzalloc(sizeof(*client), GFP_KERNEL);
 440         if (client == NULL)
 441                 return ERR_PTR(-ENOMEM);
 442
 443         client->private = private;
 444         client->options = opt;
 445
 446         mutex_init(&client->mount_mutex);
 447         init_waitqueue_head(&client->auth_wq);
 448         client->auth_err = 0;
 449
 450         client->extra_mon_dispatch = NULL;
 451         client->supported_features = CEPH_FEATURE_SUPPORTED_DEFAULT;
 452         client->required_features = CEPH_FEATURE_REQUIRED_DEFAULT;
 453
 454         client->msgr = NULL;
 455
 456         /* subsystems */
 457         err = ceph_monc_init(&client->monc, client);
 458         if (err < 0)
 459                 goto fail;
 460         err = ceph_osdc_init(&client->osdc, client);
 461         if (err < 0)
 462                 goto fail_monc;
 463
 464         return client;
 465
 466 fail_monc:
 467         ceph_monc_stop(&client->monc);
 468 fail:
 469         kfree(client);
 470         return ERR_PTR(err);
 471 }
 472 EXPORT_SYMBOL(ceph_create_client);
 473
 474 void ceph_destroy_client(struct ceph_client *client)
 475 {
 476         dout("destroy_client %p\n", client);
 477
 478         /* unmount */
 479         ceph_osdc_stop(&client->osdc);
 480
 481         /*
 482          * make sure osd connections close out before destroying the
 483          * auth module, which is needed to free those connections'
 484          * ceph_authorizers.
 485          */
 486         ceph_msgr_flush();
 487
 488         ceph_monc_stop(&client->monc);
 489
 490         ceph_debugfs_client_cleanup(client);
 491
 492         if (client->msgr)
 493                 ceph_messenger_destroy(client->msgr);
 494
 495         ceph_destroy_options(client->options);
 496
 497         kfree(client);
 498         dout("destroy_client %p done\n", client);
 499 }
 500 EXPORT_SYMBOL(ceph_destroy_client);
 501
 502 /*
 503  * true if we have the mon map (and have thus joined the cluster)
 504  */
 505 static int have_mon_and_osd_map(struct ceph_client *client)
 506 {
 507         return client->monc.monmap && client->monc.monmap->epoch &&
 508                client->osdc.osdmap && client->osdc.osdmap->epoch;
 509 }
 510
 511 /*
 512  * mount: join the ceph cluster, and open root directory.
 513  */
 514 int __ceph_open_session(struct ceph_client *client, unsigned long started)
 515 {
 516         struct ceph_entity_addr *myaddr = NULL;
 517         int err;
 518         unsigned long timeout = client->options->mount_timeout * HZ;
 519
 520         /* initialize the messenger */
 521         if (client->msgr == NULL) {
 522                 if (ceph_test_opt(client, MYIP))
 523                         myaddr = &client->options->my_addr;
 524                 client->msgr = ceph_messenger_create(myaddr,
 525                                         client->supported_features,
 526                                         client->required_features);
 527                 if (IS_ERR(client->msgr)) {
 528                         client->msgr = NULL;
 529                         return PTR_ERR(client->msgr);
 530                 }
 531                 client->msgr->nocrc = ceph_test_opt(client, NOCRC);
 532         }
 533
 534         /* open session, and wait for mon and osd maps */
 535         err = ceph_monc_open_session(&client->monc);
 536         if (err < 0)
 537                 return err;
 538
 539         while (!have_mon_and_osd_map(client)) {
 540                 err = -EIO;
 541                 if (timeout && time_after_eq(jiffies, started + timeout))
 542                         return err;
 543
 544                 /* wait */
 545                 dout("mount waiting for mon_map\n");
 546                 err = wait_event_interruptible_timeout(client->auth_wq,
 547                         have_mon_and_osd_map(client) || (client->auth_err < 0),
 548                         timeout);
 549                 if (err == -EINTR || err == -ERESTARTSYS)
 550                         return err;
 551                 if (client->auth_err < 0)
 552                         return client->auth_err;
 553         }
 554
 555         return 0;
 556 }
 557 EXPORT_SYMBOL(__ceph_open_session);
 558
 559
 560 int ceph_open_session(struct ceph_client *client)
 561 {
 562         int ret;
 563         unsigned long started = jiffies;  /* note the start time */
 564
 565         dout("open_session start\n");
 566         mutex_lock(&client->mount_mutex);
 567
 568         ret = __ceph_open_session(client, started);
 569
 570         mutex_unlock(&client->mount_mutex);
 571         return ret;
 572 }
 573 EXPORT_SYMBOL(ceph_open_session);
 574
 575
 576 static int __init init_ceph_lib(void)
 577 {
 578         int ret = 0;
 579
 580         ret = ceph_debugfs_init();
 581         if (ret < 0)
 582                 goto out;
 583
 584         ret = ceph_crypto_init();
 585         if (ret < 0)
 586                 goto out_debugfs;
 587
 588         ret = ceph_msgr_init();
 589         if (ret < 0)
 590                 goto out_crypto;
 591
 592         pr_info("loaded (mon/osd proto %d/%d, osdmap %d/%d %d/%d)\n",
 593                 CEPH_MONC_PROTOCOL, CEPH_OSDC_PROTOCOL,
 594                 CEPH_OSDMAP_VERSION, CEPH_OSDMAP_VERSION_EXT,
 595                 CEPH_OSDMAP_INC_VERSION, CEPH_OSDMAP_INC_VERSION_EXT);
 596
 597         return 0;
 598
 599 out_crypto:
 600         ceph_crypto_shutdown();
 601 out_debugfs:
 602         ceph_debugfs_cleanup();
 603 out:
 604         return ret;
 605 }
 606
 607 static void __exit exit_ceph_lib(void)
 608 {
 609         dout("exit_ceph_lib\n");
 610         ceph_msgr_exit();
 611         ceph_crypto_shutdown();
 612         ceph_debugfs_cleanup();
 613 }
 614
 615 module_init(init_ceph_lib);
 616 module_exit(exit_ceph_lib);
 617
 618 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
 619 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
 620 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
 621 MODULE_DESCRIPTION("Ceph filesystem for Linux");
 622 MODULE_LICENSE("GPL");