fs/ceph/super.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2
   3 #include <linux/ceph/ceph_debug.h>
   4
   5 #include <linux/backing-dev.h>
   6 #include <linux/ctype.h>
   7 #include <linux/fs.h>
   8 #include <linux/inet.h>
   9 #include <linux/in6.h>
  10 #include <linux/module.h>
  11 #include <linux/mount.h>
  12 #include <linux/fs_context.h>
  13 #include <linux/fs_parser.h>
  14 #include <linux/sched.h>
  15 #include <linux/seq_file.h>
  16 #include <linux/slab.h>
  17 #include <linux/statfs.h>
  18 #include <linux/string.h>
  19
  20 #include "super.h"
  21 #include "mds_client.h"
  22 #include "cache.h"
  23 #include "crypto.h"
  24
  25 #include <linux/ceph/ceph_features.h>
  26 #include <linux/ceph/decode.h>
  27 #include <linux/ceph/mon_client.h>
  28 #include <linux/ceph/auth.h>
  29 #include <linux/ceph/debugfs.h>
  30
  31 #include <uapi/linux/magic.h>
  32
  33 static DEFINE_SPINLOCK(ceph_fsc_lock);
  34 static LIST_HEAD(ceph_fsc_list);
  35
  36 /*
  37  * Ceph superblock operations
  38  *
  39  * Handle the basics of mounting, unmounting.
  40  */
  41
  42 /*
  43  * super ops
  44  */
  45 static void ceph_put_super(struct super_block *s)
  46 {
  47         struct ceph_fs_client *fsc = ceph_sb_to_fs_client(s);
  48
  49         doutc(fsc->client, "begin\n");
  50         ceph_fscrypt_free_dummy_policy(fsc);
  51         ceph_mdsc_close_sessions(fsc->mdsc);
  52         doutc(fsc->client, "done\n");
  53 }
  54
  55 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
  56 {
  57         struct ceph_fs_client *fsc = ceph_inode_to_fs_client(d_inode(dentry));
  58         struct ceph_mon_client *monc = &fsc->client->monc;
  59         struct ceph_statfs st;
  60         int i, err;
  61         u64 data_pool;
  62
  63         doutc(fsc->client, "begin\n");
  64         if (fsc->mdsc->mdsmap->m_num_data_pg_pools == 1) {
  65                 data_pool = fsc->mdsc->mdsmap->m_data_pg_pools[0];
  66         } else {
  67                 data_pool = CEPH_NOPOOL;
  68         }
  69
  70         err = ceph_monc_do_statfs(monc, data_pool, &st);
  71         if (err < 0)
  72                 return err;
  73
  74         /* fill in kstatfs */
  75         buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
  76
  77         /*
  78          * Express utilization in terms of large blocks to avoid
  79          * overflow on 32-bit machines.
  80          */
  81         buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
  82
  83         /*
  84          * By default use root quota for stats; fallback to overall filesystem
  85          * usage if using 'noquotadf' mount option or if the root dir doesn't
  86          * have max_bytes quota set.
  87          */
  88         if (ceph_test_mount_opt(fsc, NOQUOTADF) ||
  89             !ceph_quota_update_statfs(fsc, buf)) {
  90                 buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
  91                 buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
  92                 buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
  93         }
  94
  95         /*
  96          * NOTE: for the time being, we make bsize == frsize to humor
  97          * not-yet-ancient versions of glibc that are broken.
  98          * Someday, we will probably want to report a real block
  99          * size...  whatever that may mean for a network file system!
 100          */
 101         buf->f_bsize = buf->f_frsize;
 102
 103         buf->f_files = le64_to_cpu(st.num_objects);
 104         buf->f_ffree = -1;
 105         buf->f_namelen = NAME_MAX;
 106
 107         /* Must convert the fsid, for consistent values across arches */
 108         buf->f_fsid.val[0] = 0;
 109         mutex_lock(&monc->mutex);
 110         for (i = 0 ; i < sizeof(monc->monmap->fsid) / sizeof(__le32) ; ++i)
 111                 buf->f_fsid.val[0] ^= le32_to_cpu(((__le32 *)&monc->monmap->fsid)[i]);
 112         mutex_unlock(&monc->mutex);
 113
 114         /* fold the fs_cluster_id into the upper bits */
 115         buf->f_fsid.val[1] = monc->fs_cluster_id;
 116
 117         doutc(fsc->client, "done\n");
 118         return 0;
 119 }
 120
 121 static int ceph_sync_fs(struct super_block *sb, int wait)
 122 {
 123         struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
 124         struct ceph_client *cl = fsc->client;
 125
 126         if (!wait) {
 127                 doutc(cl, "(non-blocking)\n");
 128                 ceph_flush_dirty_caps(fsc->mdsc);
 129                 ceph_flush_cap_releases(fsc->mdsc);
 130                 doutc(cl, "(non-blocking) done\n");
 131                 return 0;
 132         }
 133
 134         doutc(cl, "(blocking)\n");
 135         ceph_osdc_sync(&fsc->client->osdc);
 136         ceph_mdsc_sync(fsc->mdsc);
 137         doutc(cl, "(blocking) done\n");
 138         return 0;
 139 }
 140
 141 /*
 142  * mount options
 143  */
 144 enum {
 145         Opt_wsize,
 146         Opt_rsize,
 147         Opt_rasize,
 148         Opt_caps_wanted_delay_min,
 149         Opt_caps_wanted_delay_max,
 150         Opt_caps_max,
 151         Opt_readdir_max_entries,
 152         Opt_readdir_max_bytes,
 153         Opt_congestion_kb,
 154         /* int args above */
 155         Opt_snapdirname,
 156         Opt_mds_namespace,
 157         Opt_recover_session,
 158         Opt_source,
 159         Opt_mon_addr,
 160         Opt_test_dummy_encryption,
 161         /* string args above */
 162         Opt_dirstat,
 163         Opt_rbytes,
 164         Opt_asyncreaddir,
 165         Opt_dcache,
 166         Opt_ino32,
 167         Opt_fscache,
 168         Opt_poolperm,
 169         Opt_require_active_mds,
 170         Opt_acl,
 171         Opt_quotadf,
 172         Opt_copyfrom,
 173         Opt_wsync,
 174         Opt_pagecache,
 175         Opt_sparseread,
 176 };
 177
 178 enum ceph_recover_session_mode {
 179         ceph_recover_session_no,
 180         ceph_recover_session_clean
 181 };
 182
 183 static const struct constant_table ceph_param_recover[] = {
 184         { "no",         ceph_recover_session_no },
 185         { "clean",      ceph_recover_session_clean },
 186         {}
 187 };
 188
 189 static const struct fs_parameter_spec ceph_mount_parameters[] = {
 190         fsparam_flag_no ("acl",                         Opt_acl),
 191         fsparam_flag_no ("asyncreaddir",                Opt_asyncreaddir),
 192         fsparam_s32     ("caps_max",                    Opt_caps_max),
 193         fsparam_u32     ("caps_wanted_delay_max",       Opt_caps_wanted_delay_max),
 194         fsparam_u32     ("caps_wanted_delay_min",       Opt_caps_wanted_delay_min),
 195         fsparam_u32     ("write_congestion_kb",         Opt_congestion_kb),
 196         fsparam_flag_no ("copyfrom",                    Opt_copyfrom),
 197         fsparam_flag_no ("dcache",                      Opt_dcache),
 198         fsparam_flag_no ("dirstat",                     Opt_dirstat),
 199         fsparam_flag_no ("fsc",                         Opt_fscache), // fsc|nofsc
 200         fsparam_string  ("fsc",                         Opt_fscache), // fsc=...
 201         fsparam_flag_no ("ino32",                       Opt_ino32),
 202         fsparam_string  ("mds_namespace",               Opt_mds_namespace),
 203         fsparam_string  ("mon_addr",                    Opt_mon_addr),
 204         fsparam_flag_no ("poolperm",                    Opt_poolperm),
 205         fsparam_flag_no ("quotadf",                     Opt_quotadf),
 206         fsparam_u32     ("rasize",                      Opt_rasize),
 207         fsparam_flag_no ("rbytes",                      Opt_rbytes),
 208         fsparam_u32     ("readdir_max_bytes",           Opt_readdir_max_bytes),
 209         fsparam_u32     ("readdir_max_entries",         Opt_readdir_max_entries),
 210         fsparam_enum    ("recover_session",             Opt_recover_session, ceph_param_recover),
 211         fsparam_flag_no ("require_active_mds",          Opt_require_active_mds),
 212         fsparam_u32     ("rsize",                       Opt_rsize),
 213         fsparam_string  ("snapdirname",                 Opt_snapdirname),
 214         fsparam_string  ("source",                      Opt_source),
 215         fsparam_flag    ("test_dummy_encryption",       Opt_test_dummy_encryption),
 216         fsparam_string  ("test_dummy_encryption",       Opt_test_dummy_encryption),
 217         fsparam_u32     ("wsize",                       Opt_wsize),
 218         fsparam_flag_no ("wsync",                       Opt_wsync),
 219         fsparam_flag_no ("pagecache",                   Opt_pagecache),
 220         fsparam_flag_no ("sparseread",                  Opt_sparseread),
 221         {}
 222 };
 223
 224 struct ceph_parse_opts_ctx {
 225         struct ceph_options             *copts;
 226         struct ceph_mount_options       *opts;
 227 };
 228
 229 /*
 230  * Remove adjacent slashes and then the trailing slash, unless it is
 231  * the only remaining character.
 232  *
 233  * E.g. "//dir1////dir2///" --> "/dir1/dir2", "///" --> "/".
 234  */
 235 static void canonicalize_path(char *path)
 236 {
 237         int i, j = 0;
 238
 239         for (i = 0; path[i] != '\0'; i++) {
 240                 if (path[i] != '/' || j < 1 || path[j - 1] != '/')
 241                         path[j++] = path[i];
 242         }
 243
 244         if (j > 1 && path[j - 1] == '/')
 245                 j--;
 246         path[j] = '\0';
 247 }
 248
 249 /*
 250  * Check if the mds namespace in ceph_mount_options matches
 251  * the passed in namespace string. First time match (when
 252  * ->mds_namespace is NULL) is treated specially, since
 253  * ->mds_namespace needs to be initialized by the caller.
 254  */
 255 static int namespace_equals(struct ceph_mount_options *fsopt,
 256                             const char *namespace, size_t len)
 257 {
 258         return !(fsopt->mds_namespace &&
 259                  (strlen(fsopt->mds_namespace) != len ||
 260                   strncmp(fsopt->mds_namespace, namespace, len)));
 261 }
 262
 263 static int ceph_parse_old_source(const char *dev_name, const char *dev_name_end,
 264                                  struct fs_context *fc)
 265 {
 266         int r;
 267         struct ceph_parse_opts_ctx *pctx = fc->fs_private;
 268         struct ceph_mount_options *fsopt = pctx->opts;
 269
 270         if (*dev_name_end != ':')
 271                 return invalfc(fc, "separator ':' missing in source");
 272
 273         r = ceph_parse_mon_ips(dev_name, dev_name_end - dev_name,
 274                                pctx->copts, fc->log.log, ',');
 275         if (r)
 276                 return r;
 277
 278         fsopt->new_dev_syntax = false;
 279         return 0;
 280 }
 281
 282 static int ceph_parse_new_source(const char *dev_name, const char *dev_name_end,
 283                                  struct fs_context *fc)
 284 {
 285         size_t len;
 286         struct ceph_fsid fsid;
 287         struct ceph_parse_opts_ctx *pctx = fc->fs_private;
 288         struct ceph_options *opts = pctx->copts;
 289         struct ceph_mount_options *fsopt = pctx->opts;
 290         const char *name_start = dev_name;
 291         const char *fsid_start, *fs_name_start;
 292
 293         if (*dev_name_end != '=') {
 294                 dout("separator '=' missing in source");
 295                 return -EINVAL;
 296         }
 297
 298         fsid_start = strchr(dev_name, '@');
 299         if (!fsid_start)
 300                 return invalfc(fc, "missing cluster fsid");
 301         len = fsid_start - name_start;
 302         kfree(opts->name);
 303         opts->name = kstrndup(name_start, len, GFP_KERNEL);
 304         if (!opts->name)
 305                 return -ENOMEM;
 306         dout("using %s entity name", opts->name);
 307
 308         ++fsid_start; /* start of cluster fsid */
 309         fs_name_start = strchr(fsid_start, '.');
 310         if (!fs_name_start)
 311                 return invalfc(fc, "missing file system name");
 312
 313         if (ceph_parse_fsid(fsid_start, &fsid))
 314                 return invalfc(fc, "Invalid FSID");
 315
 316         ++fs_name_start; /* start of file system name */
 317         len = dev_name_end - fs_name_start;
 318
 319         if (!namespace_equals(fsopt, fs_name_start, len))
 320                 return invalfc(fc, "Mismatching mds_namespace");
 321         kfree(fsopt->mds_namespace);
 322         fsopt->mds_namespace = kstrndup(fs_name_start, len, GFP_KERNEL);
 323         if (!fsopt->mds_namespace)
 324                 return -ENOMEM;
 325         dout("file system (mds namespace) '%s'\n", fsopt->mds_namespace);
 326
 327         fsopt->new_dev_syntax = true;
 328         return 0;
 329 }
 330
 331 /*
 332  * Parse the source parameter for new device format. Distinguish the device
 333  * spec from the path. Try parsing new device format and fallback to old
 334  * format if needed.
 335  *
 336  * New device syntax will looks like:
 337  *     <device_spec>=/<path>
 338  * where
 339  *     <device_spec> is [email protected]
 340  *     <path> is optional, but if present must begin with '/'
 341  * (monitor addresses are passed via mount option)
 342  *
 343  * Old device syntax is:
 344  *     <server_spec>[,<server_spec>...]:[<path>]
 345  * where
 346  *     <server_spec> is <ip>[:<port>]
 347  *     <path> is optional, but if present must begin with '/'
 348  */
 349 static int ceph_parse_source(struct fs_parameter *param, struct fs_context *fc)
 350 {
 351         struct ceph_parse_opts_ctx *pctx = fc->fs_private;
 352         struct ceph_mount_options *fsopt = pctx->opts;
 353         char *dev_name = param->string, *dev_name_end;
 354         int ret;
 355
 356         dout("'%s'\n", dev_name);
 357         if (!dev_name || !*dev_name)
 358                 return invalfc(fc, "Empty source");
 359
 360         dev_name_end = strchr(dev_name, '/');
 361         if (dev_name_end) {
 362                 /*
 363                  * The server_path will include the whole chars from userland
 364                  * including the leading '/'.
 365                  */
 366                 kfree(fsopt->server_path);
 367                 fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
 368                 if (!fsopt->server_path)
 369                         return -ENOMEM;
 370
 371                 canonicalize_path(fsopt->server_path);
 372         } else {
 373                 dev_name_end = dev_name + strlen(dev_name);
 374         }
 375
 376         dev_name_end--;         /* back up to separator */
 377         if (dev_name_end < dev_name)
 378                 return invalfc(fc, "Path missing in source");
 379
 380         dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
 381         if (fsopt->server_path)
 382                 dout("server path '%s'\n", fsopt->server_path);
 383
 384         dout("trying new device syntax");
 385         ret = ceph_parse_new_source(dev_name, dev_name_end, fc);
 386         if (ret) {
 387                 if (ret != -EINVAL)
 388                         return ret;
 389                 dout("trying old device syntax");
 390                 ret = ceph_parse_old_source(dev_name, dev_name_end, fc);
 391                 if (ret)
 392                         return ret;
 393         }
 394
 395         fc->source = param->string;
 396         param->string = NULL;
 397         return 0;
 398 }
 399
 400 static int ceph_parse_mon_addr(struct fs_parameter *param,
 401                                struct fs_context *fc)
 402 {
 403         struct ceph_parse_opts_ctx *pctx = fc->fs_private;
 404         struct ceph_mount_options *fsopt = pctx->opts;
 405
 406         kfree(fsopt->mon_addr);
 407         fsopt->mon_addr = param->string;
 408         param->string = NULL;
 409
 410         return ceph_parse_mon_ips(fsopt->mon_addr, strlen(fsopt->mon_addr),
 411                                   pctx->copts, fc->log.log, '/');
 412 }
 413
 414 static int ceph_parse_mount_param(struct fs_context *fc,
 415                                   struct fs_parameter *param)
 416 {
 417         struct ceph_parse_opts_ctx *pctx = fc->fs_private;
 418         struct ceph_mount_options *fsopt = pctx->opts;
 419         struct fs_parse_result result;
 420         unsigned int mode;
 421         int token, ret;
 422
 423         ret = ceph_parse_param(param, pctx->copts, fc->log.log);
 424         if (ret != -ENOPARAM)
 425                 return ret;
 426
 427         token = fs_parse(fc, ceph_mount_parameters, param, &result);
 428         dout("%s: fs_parse '%s' token %d\n",__func__, param->key, token);
 429         if (token < 0)
 430                 return token;
 431
 432         switch (token) {
 433         case Opt_snapdirname:
 434                 kfree(fsopt->snapdir_name);
 435                 fsopt->snapdir_name = param->string;
 436                 param->string = NULL;
 437                 break;
 438         case Opt_mds_namespace:
 439                 if (!namespace_equals(fsopt, param->string, strlen(param->string)))
 440                         return invalfc(fc, "Mismatching mds_namespace");
 441                 kfree(fsopt->mds_namespace);
 442                 fsopt->mds_namespace = param->string;
 443                 param->string = NULL;
 444                 break;
 445         case Opt_recover_session:
 446                 mode = result.uint_32;
 447                 if (mode == ceph_recover_session_no)
 448                         fsopt->flags &= ~CEPH_MOUNT_OPT_CLEANRECOVER;
 449                 else if (mode == ceph_recover_session_clean)
 450                         fsopt->flags |= CEPH_MOUNT_OPT_CLEANRECOVER;
 451                 else
 452                         BUG();
 453                 break;
 454         case Opt_source:
 455                 if (fc->source)
 456                         return invalfc(fc, "Multiple sources specified");
 457                 return ceph_parse_source(param, fc);
 458         case Opt_mon_addr:
 459                 return ceph_parse_mon_addr(param, fc);
 460         case Opt_wsize:
 461                 if (result.uint_32 < PAGE_SIZE ||
 462                     result.uint_32 > CEPH_MAX_WRITE_SIZE)
 463                         goto out_of_range;
 464                 fsopt->wsize = ALIGN(result.uint_32, PAGE_SIZE);
 465                 break;
 466         case Opt_rsize:
 467                 if (result.uint_32 < PAGE_SIZE ||
 468                     result.uint_32 > CEPH_MAX_READ_SIZE)
 469                         goto out_of_range;
 470                 fsopt->rsize = ALIGN(result.uint_32, PAGE_SIZE);
 471                 break;
 472         case Opt_rasize:
 473                 fsopt->rasize = ALIGN(result.uint_32, PAGE_SIZE);
 474                 break;
 475         case Opt_caps_wanted_delay_min:
 476                 if (result.uint_32 < 1)
 477                         goto out_of_range;
 478                 fsopt->caps_wanted_delay_min = result.uint_32;
 479                 break;
 480         case Opt_caps_wanted_delay_max:
 481                 if (result.uint_32 < 1)
 482                         goto out_of_range;
 483                 fsopt->caps_wanted_delay_max = result.uint_32;
 484                 break;
 485         case Opt_caps_max:
 486                 if (result.int_32 < 0)
 487                         goto out_of_range;
 488                 fsopt->caps_max = result.int_32;
 489                 break;
 490         case Opt_readdir_max_entries:
 491                 if (result.uint_32 < 1)
 492                         goto out_of_range;
 493                 fsopt->max_readdir = result.uint_32;
 494                 break;
 495         case Opt_readdir_max_bytes:
 496                 if (result.uint_32 < PAGE_SIZE && result.uint_32 != 0)
 497                         goto out_of_range;
 498                 fsopt->max_readdir_bytes = result.uint_32;
 499                 break;
 500         case Opt_congestion_kb:
 501                 if (result.uint_32 < 1024) /* at least 1M */
 502                         goto out_of_range;
 503                 fsopt->congestion_kb = result.uint_32;
 504                 break;
 505         case Opt_dirstat:
 506                 if (!result.negated)
 507                         fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
 508                 else
 509                         fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
 510                 break;
 511         case Opt_rbytes:
 512                 if (!result.negated)
 513                         fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
 514                 else
 515                         fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
 516                 break;
 517         case Opt_asyncreaddir:
 518                 if (!result.negated)
 519                         fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
 520                 else
 521                         fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
 522                 break;
 523         case Opt_dcache:
 524                 if (!result.negated)
 525                         fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
 526                 else
 527                         fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
 528                 break;
 529         case Opt_ino32:
 530                 if (!result.negated)
 531                         fsopt->flags |= CEPH_MOUNT_OPT_INO32;
 532                 else
 533                         fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
 534                 break;
 535
 536         case Opt_fscache:
 537 #ifdef CONFIG_CEPH_FSCACHE
 538                 kfree(fsopt->fscache_uniq);
 539                 fsopt->fscache_uniq = NULL;
 540                 if (result.negated) {
 541                         fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
 542                 } else {
 543                         fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
 544                         fsopt->fscache_uniq = param->string;
 545                         param->string = NULL;
 546                 }
 547                 break;
 548 #else
 549                 return invalfc(fc, "fscache support is disabled");
 550 #endif
 551         case Opt_poolperm:
 552                 if (!result.negated)
 553                         fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
 554                 else
 555                         fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
 556                 break;
 557         case Opt_require_active_mds:
 558                 if (!result.negated)
 559                         fsopt->flags &= ~CEPH_MOUNT_OPT_MOUNTWAIT;
 560                 else
 561                         fsopt->flags |= CEPH_MOUNT_OPT_MOUNTWAIT;
 562                 break;
 563         case Opt_quotadf:
 564                 if (!result.negated)
 565                         fsopt->flags &= ~CEPH_MOUNT_OPT_NOQUOTADF;
 566                 else
 567                         fsopt->flags |= CEPH_MOUNT_OPT_NOQUOTADF;
 568                 break;
 569         case Opt_copyfrom:
 570                 if (!result.negated)
 571                         fsopt->flags &= ~CEPH_MOUNT_OPT_NOCOPYFROM;
 572                 else
 573                         fsopt->flags |= CEPH_MOUNT_OPT_NOCOPYFROM;
 574                 break;
 575         case Opt_acl:
 576                 if (!result.negated) {
 577 #ifdef CONFIG_CEPH_FS_POSIX_ACL
 578                         fc->sb_flags |= SB_POSIXACL;
 579 #else
 580                         return invalfc(fc, "POSIX ACL support is disabled");
 581 #endif
 582                 } else {
 583                         fc->sb_flags &= ~SB_POSIXACL;
 584                 }
 585                 break;
 586         case Opt_wsync:
 587                 if (!result.negated)
 588                         fsopt->flags &= ~CEPH_MOUNT_OPT_ASYNC_DIROPS;
 589                 else
 590                         fsopt->flags |= CEPH_MOUNT_OPT_ASYNC_DIROPS;
 591                 break;
 592         case Opt_pagecache:
 593                 if (result.negated)
 594                         fsopt->flags |= CEPH_MOUNT_OPT_NOPAGECACHE;
 595                 else
 596                         fsopt->flags &= ~CEPH_MOUNT_OPT_NOPAGECACHE;
 597                 break;
 598         case Opt_sparseread:
 599                 if (result.negated)
 600                         fsopt->flags &= ~CEPH_MOUNT_OPT_SPARSEREAD;
 601                 else
 602                         fsopt->flags |= CEPH_MOUNT_OPT_SPARSEREAD;
 603                 break;
 604         case Opt_test_dummy_encryption:
 605 #ifdef CONFIG_FS_ENCRYPTION
 606                 fscrypt_free_dummy_policy(&fsopt->dummy_enc_policy);
 607                 ret = fscrypt_parse_test_dummy_encryption(param,
 608                                                 &fsopt->dummy_enc_policy);
 609                 if (ret == -EINVAL) {
 610                         warnfc(fc, "Value of option \"%s\" is unrecognized",
 611                                param->key);
 612                 } else if (ret == -EEXIST) {
 613                         warnfc(fc, "Conflicting test_dummy_encryption options");
 614                         ret = -EINVAL;
 615                 }
 616 #else
 617                 warnfc(fc,
 618                        "FS encryption not supported: test_dummy_encryption mount option ignored");
 619 #endif
 620                 break;
 621         default:
 622                 BUG();
 623         }
 624         return 0;
 625
 626 out_of_range:
 627         return invalfc(fc, "%s out of range", param->key);
 628 }
 629
 630 static void destroy_mount_options(struct ceph_mount_options *args)
 631 {
 632         dout("destroy_mount_options %p\n", args);
 633         if (!args)
 634                 return;
 635
 636         kfree(args->snapdir_name);
 637         kfree(args->mds_namespace);
 638         kfree(args->server_path);
 639         kfree(args->fscache_uniq);
 640         kfree(args->mon_addr);
 641         fscrypt_free_dummy_policy(&args->dummy_enc_policy);
 642         kfree(args);
 643 }
 644
 645 static int strcmp_null(const char *s1, const char *s2)
 646 {
 647         if (!s1 && !s2)
 648                 return 0;
 649         if (s1 && !s2)
 650                 return -1;
 651         if (!s1 && s2)
 652                 return 1;
 653         return strcmp(s1, s2);
 654 }
 655
 656 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
 657                                  struct ceph_options *new_opt,
 658                                  struct ceph_fs_client *fsc)
 659 {
 660         struct ceph_mount_options *fsopt1 = new_fsopt;
 661         struct ceph_mount_options *fsopt2 = fsc->mount_options;
 662         int ofs = offsetof(struct ceph_mount_options, snapdir_name);
 663         int ret;
 664
 665         ret = memcmp(fsopt1, fsopt2, ofs);
 666         if (ret)
 667                 return ret;
 668
 669         ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
 670         if (ret)
 671                 return ret;
 672
 673         ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
 674         if (ret)
 675                 return ret;
 676
 677         ret = strcmp_null(fsopt1->server_path, fsopt2->server_path);
 678         if (ret)
 679                 return ret;
 680
 681         ret = strcmp_null(fsopt1->fscache_uniq, fsopt2->fscache_uniq);
 682         if (ret)
 683                 return ret;
 684
 685         ret = strcmp_null(fsopt1->mon_addr, fsopt2->mon_addr);
 686         if (ret)
 687                 return ret;
 688
 689         return ceph_compare_options(new_opt, fsc->client);
 690 }
 691
 692 /**
 693  * ceph_show_options - Show mount options in /proc/mounts
 694  * @m: seq_file to write to
 695  * @root: root of that (sub)tree
 696  */
 697 static int ceph_show_options(struct seq_file *m, struct dentry *root)
 698 {
 699         struct ceph_fs_client *fsc = ceph_sb_to_fs_client(root->d_sb);
 700         struct ceph_mount_options *fsopt = fsc->mount_options;
 701         size_t pos;
 702         int ret;
 703
 704         /* a comma between MNT/MS and client options */
 705         seq_putc(m, ',');
 706         pos = m->count;
 707
 708         ret = ceph_print_client_options(m, fsc->client, false);
 709         if (ret)
 710                 return ret;
 711
 712         /* retract our comma if no client options */
 713         if (m->count == pos)
 714                 m->count--;
 715
 716         if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
 717                 seq_puts(m, ",dirstat");
 718         if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
 719                 seq_puts(m, ",rbytes");
 720         if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
 721                 seq_puts(m, ",noasyncreaddir");
 722         if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
 723                 seq_puts(m, ",nodcache");
 724         if (fsopt->flags & CEPH_MOUNT_OPT_INO32)
 725                 seq_puts(m, ",ino32");
 726         if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) {
 727                 seq_show_option(m, "fsc", fsopt->fscache_uniq);
 728         }
 729         if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
 730                 seq_puts(m, ",nopoolperm");
 731         if (fsopt->flags & CEPH_MOUNT_OPT_NOQUOTADF)
 732                 seq_puts(m, ",noquotadf");
 733
 734 #ifdef CONFIG_CEPH_FS_POSIX_ACL
 735         if (root->d_sb->s_flags & SB_POSIXACL)
 736                 seq_puts(m, ",acl");
 737         else
 738                 seq_puts(m, ",noacl");
 739 #endif
 740
 741         if ((fsopt->flags & CEPH_MOUNT_OPT_NOCOPYFROM) == 0)
 742                 seq_puts(m, ",copyfrom");
 743
 744         /* dump mds_namespace when old device syntax is in use */
 745         if (fsopt->mds_namespace && !fsopt->new_dev_syntax)
 746                 seq_show_option(m, "mds_namespace", fsopt->mds_namespace);
 747
 748         if (fsopt->mon_addr)
 749                 seq_printf(m, ",mon_addr=%s", fsopt->mon_addr);
 750
 751         if (fsopt->flags & CEPH_MOUNT_OPT_CLEANRECOVER)
 752                 seq_show_option(m, "recover_session", "clean");
 753
 754         if (!(fsopt->flags & CEPH_MOUNT_OPT_ASYNC_DIROPS))
 755                 seq_puts(m, ",wsync");
 756         if (fsopt->flags & CEPH_MOUNT_OPT_NOPAGECACHE)
 757                 seq_puts(m, ",nopagecache");
 758         if (fsopt->flags & CEPH_MOUNT_OPT_SPARSEREAD)
 759                 seq_puts(m, ",sparseread");
 760
 761         fscrypt_show_test_dummy_encryption(m, ',', root->d_sb);
 762
 763         if (fsopt->wsize != CEPH_MAX_WRITE_SIZE)
 764                 seq_printf(m, ",wsize=%u", fsopt->wsize);
 765         if (fsopt->rsize != CEPH_MAX_READ_SIZE)
 766                 seq_printf(m, ",rsize=%u", fsopt->rsize);
 767         if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
 768                 seq_printf(m, ",rasize=%u", fsopt->rasize);
 769         if (fsopt->congestion_kb != default_congestion_kb())
 770                 seq_printf(m, ",write_congestion_kb=%u", fsopt->congestion_kb);
 771         if (fsopt->caps_max)
 772                 seq_printf(m, ",caps_max=%d", fsopt->caps_max);
 773         if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
 774                 seq_printf(m, ",caps_wanted_delay_min=%u",
 775                          fsopt->caps_wanted_delay_min);
 776         if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
 777                 seq_printf(m, ",caps_wanted_delay_max=%u",
 778                            fsopt->caps_wanted_delay_max);
 779         if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
 780                 seq_printf(m, ",readdir_max_entries=%u", fsopt->max_readdir);
 781         if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
 782                 seq_printf(m, ",readdir_max_bytes=%u", fsopt->max_readdir_bytes);
 783         if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
 784                 seq_show_option(m, "snapdirname", fsopt->snapdir_name);
 785
 786         return 0;
 787 }
 788
 789 /*
 790  * handle any mon messages the standard library doesn't understand.
 791  * return error if we don't either.
 792  */
 793 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
 794 {
 795         struct ceph_fs_client *fsc = client->private;
 796         int type = le16_to_cpu(msg->hdr.type);
 797
 798         switch (type) {
 799         case CEPH_MSG_MDS_MAP:
 800                 ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
 801                 return 0;
 802         case CEPH_MSG_FS_MAP_USER:
 803                 ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
 804                 return 0;
 805         default:
 806                 return -1;
 807         }
 808 }
 809
 810 /*
 811  * create a new fs client
 812  *
 813  * Success or not, this function consumes @fsopt and @opt.
 814  */
 815 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
 816                                         struct ceph_options *opt)
 817 {
 818         struct ceph_fs_client *fsc;
 819         int err;
 820
 821         fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
 822         if (!fsc) {
 823                 err = -ENOMEM;
 824                 goto fail;
 825         }
 826
 827         fsc->client = ceph_create_client(opt, fsc);
 828         if (IS_ERR(fsc->client)) {
 829                 err = PTR_ERR(fsc->client);
 830                 goto fail;
 831         }
 832         opt = NULL; /* fsc->client now owns this */
 833
 834         fsc->client->extra_mon_dispatch = extra_mon_dispatch;
 835         ceph_set_opt(fsc->client, ABORT_ON_FULL);
 836
 837         if (!fsopt->mds_namespace) {
 838                 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
 839                                    0, true);
 840         } else {
 841                 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
 842                                    0, false);
 843         }
 844
 845         fsc->mount_options = fsopt;
 846
 847         fsc->sb = NULL;
 848         fsc->mount_state = CEPH_MOUNT_MOUNTING;
 849         fsc->filp_gen = 1;
 850         fsc->have_copy_from2 = true;
 851
 852         atomic_long_set(&fsc->writeback_count, 0);
 853         fsc->write_congested = false;
 854
 855         err = -ENOMEM;
 856         /*
 857          * The number of concurrent works can be high but they don't need
 858          * to be processed in parallel, limit concurrency.
 859          */
 860         fsc->inode_wq = alloc_workqueue("ceph-inode", WQ_UNBOUND, 0);
 861         if (!fsc->inode_wq)
 862                 goto fail_client;
 863         fsc->cap_wq = alloc_workqueue("ceph-cap", 0, 1);
 864         if (!fsc->cap_wq)
 865                 goto fail_inode_wq;
 866
 867         hash_init(fsc->async_unlink_conflict);
 868         spin_lock_init(&fsc->async_unlink_conflict_lock);
 869
 870         spin_lock(&ceph_fsc_lock);
 871         list_add_tail(&fsc->metric_wakeup, &ceph_fsc_list);
 872         spin_unlock(&ceph_fsc_lock);
 873
 874         return fsc;
 875
 876 fail_inode_wq:
 877         destroy_workqueue(fsc->inode_wq);
 878 fail_client:
 879         ceph_destroy_client(fsc->client);
 880 fail:
 881         kfree(fsc);
 882         if (opt)
 883                 ceph_destroy_options(opt);
 884         destroy_mount_options(fsopt);
 885         return ERR_PTR(err);
 886 }
 887
 888 static void flush_fs_workqueues(struct ceph_fs_client *fsc)
 889 {
 890         flush_workqueue(fsc->inode_wq);
 891         flush_workqueue(fsc->cap_wq);
 892 }
 893
 894 static void destroy_fs_client(struct ceph_fs_client *fsc)
 895 {
 896         doutc(fsc->client, "%p\n", fsc);
 897
 898         spin_lock(&ceph_fsc_lock);
 899         list_del(&fsc->metric_wakeup);
 900         spin_unlock(&ceph_fsc_lock);
 901
 902         ceph_mdsc_destroy(fsc);
 903         destroy_workqueue(fsc->inode_wq);
 904         destroy_workqueue(fsc->cap_wq);
 905
 906         destroy_mount_options(fsc->mount_options);
 907
 908         ceph_destroy_client(fsc->client);
 909
 910         kfree(fsc);
 911         dout("%s: %p done\n", __func__, fsc);
 912 }
 913
 914 /*
 915  * caches
 916  */
 917 struct kmem_cache *ceph_inode_cachep;
 918 struct kmem_cache *ceph_cap_cachep;
 919 struct kmem_cache *ceph_cap_snap_cachep;
 920 struct kmem_cache *ceph_cap_flush_cachep;
 921 struct kmem_cache *ceph_dentry_cachep;
 922 struct kmem_cache *ceph_file_cachep;
 923 struct kmem_cache *ceph_dir_file_cachep;
 924 struct kmem_cache *ceph_mds_request_cachep;
 925 mempool_t *ceph_wb_pagevec_pool;
 926
 927 static void ceph_inode_init_once(void *foo)
 928 {
 929         struct ceph_inode_info *ci = foo;
 930         inode_init_once(&ci->netfs.inode);
 931 }
 932
 933 static int __init init_caches(void)
 934 {
 935         int error = -ENOMEM;
 936
 937         ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
 938                                       sizeof(struct ceph_inode_info),
 939                                       __alignof__(struct ceph_inode_info),
 940                                       SLAB_RECLAIM_ACCOUNT | SLAB_ACCOUNT,
 941                                       ceph_inode_init_once);
 942         if (!ceph_inode_cachep)
 943                 return -ENOMEM;
 944
 945         ceph_cap_cachep = KMEM_CACHE(ceph_cap, 0);
 946         if (!ceph_cap_cachep)
 947                 goto bad_cap;
 948         ceph_cap_snap_cachep = KMEM_CACHE(ceph_cap_snap, 0);
 949         if (!ceph_cap_snap_cachep)
 950                 goto bad_cap_snap;
 951         ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
 952                                            SLAB_RECLAIM_ACCOUNT);
 953         if (!ceph_cap_flush_cachep)
 954                 goto bad_cap_flush;
 955
 956         ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
 957                                         SLAB_RECLAIM_ACCOUNT);
 958         if (!ceph_dentry_cachep)
 959                 goto bad_dentry;
 960
 961         ceph_file_cachep = KMEM_CACHE(ceph_file_info, 0);
 962         if (!ceph_file_cachep)
 963                 goto bad_file;
 964
 965         ceph_dir_file_cachep = KMEM_CACHE(ceph_dir_file_info, 0);
 966         if (!ceph_dir_file_cachep)
 967                 goto bad_dir_file;
 968
 969         ceph_mds_request_cachep = KMEM_CACHE(ceph_mds_request, 0);
 970         if (!ceph_mds_request_cachep)
 971                 goto bad_mds_req;
 972
 973         ceph_wb_pagevec_pool = mempool_create_kmalloc_pool(10,
 974             (CEPH_MAX_WRITE_SIZE >> PAGE_SHIFT) * sizeof(struct page *));
 975         if (!ceph_wb_pagevec_pool)
 976                 goto bad_pagevec_pool;
 977
 978         return 0;
 979
 980 bad_pagevec_pool:
 981         kmem_cache_destroy(ceph_mds_request_cachep);
 982 bad_mds_req:
 983         kmem_cache_destroy(ceph_dir_file_cachep);
 984 bad_dir_file:
 985         kmem_cache_destroy(ceph_file_cachep);
 986 bad_file:
 987         kmem_cache_destroy(ceph_dentry_cachep);
 988 bad_dentry:
 989         kmem_cache_destroy(ceph_cap_flush_cachep);
 990 bad_cap_flush:
 991         kmem_cache_destroy(ceph_cap_snap_cachep);
 992 bad_cap_snap:
 993         kmem_cache_destroy(ceph_cap_cachep);
 994 bad_cap:
 995         kmem_cache_destroy(ceph_inode_cachep);
 996         return error;
 997 }
 998
 999 static void destroy_caches(void)
1000 {
1001         /*
1002          * Make sure all delayed rcu free inodes are flushed before we
1003          * destroy cache.
1004          */
1005         rcu_barrier();
1006
1007         kmem_cache_destroy(ceph_inode_cachep);
1008         kmem_cache_destroy(ceph_cap_cachep);
1009         kmem_cache_destroy(ceph_cap_snap_cachep);
1010         kmem_cache_destroy(ceph_cap_flush_cachep);
1011         kmem_cache_destroy(ceph_dentry_cachep);
1012         kmem_cache_destroy(ceph_file_cachep);
1013         kmem_cache_destroy(ceph_dir_file_cachep);
1014         kmem_cache_destroy(ceph_mds_request_cachep);
1015         mempool_destroy(ceph_wb_pagevec_pool);
1016 }
1017
1018 static void __ceph_umount_begin(struct ceph_fs_client *fsc)
1019 {
1020         ceph_osdc_abort_requests(&fsc->client->osdc, -EIO);
1021         ceph_mdsc_force_umount(fsc->mdsc);
1022         fsc->filp_gen++; // invalidate open files
1023 }
1024
1025 /*
1026  * ceph_umount_begin - initiate forced umount.  Tear down the
1027  * mount, skipping steps that may hang while waiting for server(s).
1028  */
1029 void ceph_umount_begin(struct super_block *sb)
1030 {
1031         struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1032
1033         doutc(fsc->client, "starting forced umount\n");
1034         if (!fsc)
1035                 return;
1036         fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
1037         __ceph_umount_begin(fsc);
1038 }
1039
1040 static const struct super_operations ceph_super_ops = {
1041         .alloc_inode    = ceph_alloc_inode,
1042         .free_inode     = ceph_free_inode,
1043         .write_inode    = ceph_write_inode,
1044         .drop_inode     = generic_delete_inode,
1045         .evict_inode    = ceph_evict_inode,
1046         .sync_fs        = ceph_sync_fs,
1047         .put_super      = ceph_put_super,
1048         .show_options   = ceph_show_options,
1049         .statfs         = ceph_statfs,
1050         .umount_begin   = ceph_umount_begin,
1051 };
1052
1053 /*
1054  * Bootstrap mount by opening the root directory.  Note the mount
1055  * @started time from caller, and time out if this takes too long.
1056  */
1057 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
1058                                        const char *path,
1059                                        unsigned long started)
1060 {
1061         struct ceph_client *cl = fsc->client;
1062         struct ceph_mds_client *mdsc = fsc->mdsc;
1063         struct ceph_mds_request *req = NULL;
1064         int err;
1065         struct dentry *root;
1066
1067         /* open dir */
1068         doutc(cl, "opening '%s'\n", path);
1069         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1070         if (IS_ERR(req))
1071                 return ERR_CAST(req);
1072         req->r_path1 = kstrdup(path, GFP_NOFS);
1073         if (!req->r_path1) {
1074                 root = ERR_PTR(-ENOMEM);
1075                 goto out;
1076         }
1077
1078         req->r_ino1.ino = CEPH_INO_ROOT;
1079         req->r_ino1.snap = CEPH_NOSNAP;
1080         req->r_started = started;
1081         req->r_timeout = fsc->client->options->mount_timeout;
1082         req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
1083         req->r_num_caps = 2;
1084         err = ceph_mdsc_do_request(mdsc, NULL, req);
1085         if (err == 0) {
1086                 struct inode *inode = req->r_target_inode;
1087                 req->r_target_inode = NULL;
1088                 doutc(cl, "success\n");
1089                 root = d_make_root(inode);
1090                 if (!root) {
1091                         root = ERR_PTR(-ENOMEM);
1092                         goto out;
1093                 }
1094                 doutc(cl, "success, root dentry is %p\n", root);
1095         } else {
1096                 root = ERR_PTR(err);
1097         }
1098 out:
1099         ceph_mdsc_put_request(req);
1100         return root;
1101 }
1102
1103 #ifdef CONFIG_FS_ENCRYPTION
1104 static int ceph_apply_test_dummy_encryption(struct super_block *sb,
1105                                             struct fs_context *fc,
1106                                             struct ceph_mount_options *fsopt)
1107 {
1108         struct ceph_fs_client *fsc = sb->s_fs_info;
1109
1110         if (!fscrypt_is_dummy_policy_set(&fsopt->dummy_enc_policy))
1111                 return 0;
1112
1113         /* No changing encryption context on remount. */
1114         if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE &&
1115             !fscrypt_is_dummy_policy_set(&fsc->fsc_dummy_enc_policy)) {
1116                 if (fscrypt_dummy_policies_equal(&fsopt->dummy_enc_policy,
1117                                                  &fsc->fsc_dummy_enc_policy))
1118                         return 0;
1119                 errorfc(fc, "Can't set test_dummy_encryption on remount");
1120                 return -EINVAL;
1121         }
1122
1123         /* Also make sure fsopt doesn't contain a conflicting value. */
1124         if (fscrypt_is_dummy_policy_set(&fsc->fsc_dummy_enc_policy)) {
1125                 if (fscrypt_dummy_policies_equal(&fsopt->dummy_enc_policy,
1126                                                  &fsc->fsc_dummy_enc_policy))
1127                         return 0;
1128                 errorfc(fc, "Conflicting test_dummy_encryption options");
1129                 return -EINVAL;
1130         }
1131
1132         fsc->fsc_dummy_enc_policy = fsopt->dummy_enc_policy;
1133         memset(&fsopt->dummy_enc_policy, 0, sizeof(fsopt->dummy_enc_policy));
1134
1135         warnfc(fc, "test_dummy_encryption mode enabled");
1136         return 0;
1137 }
1138 #else
1139 static int ceph_apply_test_dummy_encryption(struct super_block *sb,
1140                                             struct fs_context *fc,
1141                                             struct ceph_mount_options *fsopt)
1142 {
1143         return 0;
1144 }
1145 #endif
1146
1147 /*
1148  * mount: join the ceph cluster, and open root directory.
1149  */
1150 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
1151                                       struct fs_context *fc)
1152 {
1153         struct ceph_client *cl = fsc->client;
1154         int err;
1155         unsigned long started = jiffies;  /* note the start time */
1156         struct dentry *root;
1157
1158         doutc(cl, "mount start %p\n", fsc);
1159         mutex_lock(&fsc->client->mount_mutex);
1160
1161         if (!fsc->sb->s_root) {
1162                 const char *path = fsc->mount_options->server_path ?
1163                                      fsc->mount_options->server_path + 1 : "";
1164
1165                 err = __ceph_open_session(fsc->client, started);
1166                 if (err < 0)
1167                         goto out;
1168
1169                 /* setup fscache */
1170                 if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
1171                         err = ceph_fscache_register_fs(fsc, fc);
1172                         if (err < 0)
1173                                 goto out;
1174                 }
1175
1176                 err = ceph_apply_test_dummy_encryption(fsc->sb, fc,
1177                                                        fsc->mount_options);
1178                 if (err)
1179                         goto out;
1180
1181                 doutc(cl, "mount opening path '%s'\n", path);
1182
1183                 ceph_fs_debugfs_init(fsc);
1184
1185                 root = open_root_dentry(fsc, path, started);
1186                 if (IS_ERR(root)) {
1187                         err = PTR_ERR(root);
1188                         goto out;
1189                 }
1190                 fsc->sb->s_root = dget(root);
1191         } else {
1192                 root = dget(fsc->sb->s_root);
1193         }
1194
1195         fsc->mount_state = CEPH_MOUNT_MOUNTED;
1196         doutc(cl, "mount success\n");
1197         mutex_unlock(&fsc->client->mount_mutex);
1198         return root;
1199
1200 out:
1201         mutex_unlock(&fsc->client->mount_mutex);
1202         ceph_fscrypt_free_dummy_policy(fsc);
1203         return ERR_PTR(err);
1204 }
1205
1206 static int ceph_set_super(struct super_block *s, struct fs_context *fc)
1207 {
1208         struct ceph_fs_client *fsc = s->s_fs_info;
1209         struct ceph_client *cl = fsc->client;
1210         int ret;
1211
1212         doutc(cl, "%p\n", s);
1213
1214         s->s_maxbytes = MAX_LFS_FILESIZE;
1215
1216         s->s_xattr = ceph_xattr_handlers;
1217         fsc->sb = s;
1218         fsc->max_file_size = 1ULL << 40; /* temp value until we get mdsmap */
1219
1220         s->s_op = &ceph_super_ops;
1221         s->s_d_op = &ceph_dentry_ops;
1222         s->s_export_op = &ceph_export_ops;
1223
1224         s->s_time_gran = 1;
1225         s->s_time_min = 0;
1226         s->s_time_max = U32_MAX;
1227         s->s_flags |= SB_NODIRATIME | SB_NOATIME;
1228
1229         ceph_fscrypt_set_ops(s);
1230
1231         ret = set_anon_super_fc(s, fc);
1232         if (ret != 0)
1233                 fsc->sb = NULL;
1234         return ret;
1235 }
1236
1237 /*
1238  * share superblock if same fs AND options
1239  */
1240 static int ceph_compare_super(struct super_block *sb, struct fs_context *fc)
1241 {
1242         struct ceph_fs_client *new = fc->s_fs_info;
1243         struct ceph_mount_options *fsopt = new->mount_options;
1244         struct ceph_options *opt = new->client->options;
1245         struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1246         struct ceph_client *cl = fsc->client;
1247
1248         doutc(cl, "%p\n", sb);
1249
1250         if (compare_mount_options(fsopt, opt, fsc)) {
1251                 doutc(cl, "monitor(s)/mount options don't match\n");
1252                 return 0;
1253         }
1254         if ((opt->flags & CEPH_OPT_FSID) &&
1255             ceph_fsid_compare(&opt->fsid, &fsc->client->fsid)) {
1256                 doutc(cl, "fsid doesn't match\n");
1257                 return 0;
1258         }
1259         if (fc->sb_flags != (sb->s_flags & ~SB_BORN)) {
1260                 doutc(cl, "flags differ\n");
1261                 return 0;
1262         }
1263
1264         if (fsc->blocklisted && !ceph_test_mount_opt(fsc, CLEANRECOVER)) {
1265                 doutc(cl, "client is blocklisted (and CLEANRECOVER is not set)\n");
1266                 return 0;
1267         }
1268
1269         if (fsc->mount_state == CEPH_MOUNT_SHUTDOWN) {
1270                 doutc(cl, "client has been forcibly unmounted\n");
1271                 return 0;
1272         }
1273
1274         return 1;
1275 }
1276
1277 /*
1278  * construct our own bdi so we can control readahead, etc.
1279  */
1280 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1281
1282 static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
1283 {
1284         int err;
1285
1286         err = super_setup_bdi_name(sb, "ceph-%ld",
1287                                    atomic_long_inc_return(&bdi_seq));
1288         if (err)
1289                 return err;
1290
1291         /* set ra_pages based on rasize mount option? */
1292         sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
1293
1294         /* set io_pages based on max osd read size */
1295         sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
1296
1297         return 0;
1298 }
1299
1300 static int ceph_get_tree(struct fs_context *fc)
1301 {
1302         struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1303         struct ceph_mount_options *fsopt = pctx->opts;
1304         struct super_block *sb;
1305         struct ceph_fs_client *fsc;
1306         struct dentry *res;
1307         int (*compare_super)(struct super_block *, struct fs_context *) =
1308                 ceph_compare_super;
1309         int err;
1310
1311         dout("ceph_get_tree\n");
1312
1313         if (!fc->source)
1314                 return invalfc(fc, "No source");
1315         if (fsopt->new_dev_syntax && !fsopt->mon_addr)
1316                 return invalfc(fc, "No monitor address");
1317
1318         /* create client (which we may/may not use) */
1319         fsc = create_fs_client(pctx->opts, pctx->copts);
1320         pctx->opts = NULL;
1321         pctx->copts = NULL;
1322         if (IS_ERR(fsc)) {
1323                 err = PTR_ERR(fsc);
1324                 goto out_final;
1325         }
1326
1327         err = ceph_mdsc_init(fsc);
1328         if (err < 0)
1329                 goto out;
1330
1331         if (ceph_test_opt(fsc->client, NOSHARE))
1332                 compare_super = NULL;
1333
1334         fc->s_fs_info = fsc;
1335         sb = sget_fc(fc, compare_super, ceph_set_super);
1336         fc->s_fs_info = NULL;
1337         if (IS_ERR(sb)) {
1338                 err = PTR_ERR(sb);
1339                 goto out;
1340         }
1341
1342         if (ceph_sb_to_fs_client(sb) != fsc) {
1343                 destroy_fs_client(fsc);
1344                 fsc = ceph_sb_to_fs_client(sb);
1345                 dout("get_sb got existing client %p\n", fsc);
1346         } else {
1347                 dout("get_sb using new client %p\n", fsc);
1348                 err = ceph_setup_bdi(sb, fsc);
1349                 if (err < 0)
1350                         goto out_splat;
1351         }
1352
1353         res = ceph_real_mount(fsc, fc);
1354         if (IS_ERR(res)) {
1355                 err = PTR_ERR(res);
1356                 goto out_splat;
1357         }
1358
1359         doutc(fsc->client, "root %p inode %p ino %llx.%llx\n", res,
1360                     d_inode(res), ceph_vinop(d_inode(res)));
1361         fc->root = fsc->sb->s_root;
1362         return 0;
1363
1364 out_splat:
1365         if (!ceph_mdsmap_is_cluster_available(fsc->mdsc->mdsmap)) {
1366                 pr_info("No mds server is up or the cluster is laggy\n");
1367                 err = -EHOSTUNREACH;
1368         }
1369
1370         ceph_mdsc_close_sessions(fsc->mdsc);
1371         deactivate_locked_super(sb);
1372         goto out_final;
1373
1374 out:
1375         destroy_fs_client(fsc);
1376 out_final:
1377         dout("ceph_get_tree fail %d\n", err);
1378         return err;
1379 }
1380
1381 static void ceph_free_fc(struct fs_context *fc)
1382 {
1383         struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1384
1385         if (pctx) {
1386                 destroy_mount_options(pctx->opts);
1387                 ceph_destroy_options(pctx->copts);
1388                 kfree(pctx);
1389         }
1390 }
1391
1392 static int ceph_reconfigure_fc(struct fs_context *fc)
1393 {
1394         int err;
1395         struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1396         struct ceph_mount_options *fsopt = pctx->opts;
1397         struct super_block *sb = fc->root->d_sb;
1398         struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1399
1400         err = ceph_apply_test_dummy_encryption(sb, fc, fsopt);
1401         if (err)
1402                 return err;
1403
1404         if (fsopt->flags & CEPH_MOUNT_OPT_ASYNC_DIROPS)
1405                 ceph_set_mount_opt(fsc, ASYNC_DIROPS);
1406         else
1407                 ceph_clear_mount_opt(fsc, ASYNC_DIROPS);
1408
1409         if (fsopt->flags & CEPH_MOUNT_OPT_SPARSEREAD)
1410                 ceph_set_mount_opt(fsc, SPARSEREAD);
1411         else
1412                 ceph_clear_mount_opt(fsc, SPARSEREAD);
1413
1414         if (strcmp_null(fsc->mount_options->mon_addr, fsopt->mon_addr)) {
1415                 kfree(fsc->mount_options->mon_addr);
1416                 fsc->mount_options->mon_addr = fsopt->mon_addr;
1417                 fsopt->mon_addr = NULL;
1418                 pr_notice_client(fsc->client,
1419                         "monitor addresses recorded, but not used for reconnection");
1420         }
1421
1422         sync_filesystem(sb);
1423         return 0;
1424 }
1425
1426 static const struct fs_context_operations ceph_context_ops = {
1427         .free           = ceph_free_fc,
1428         .parse_param    = ceph_parse_mount_param,
1429         .get_tree       = ceph_get_tree,
1430         .reconfigure    = ceph_reconfigure_fc,
1431 };
1432
1433 /*
1434  * Set up the filesystem mount context.
1435  */
1436 static int ceph_init_fs_context(struct fs_context *fc)
1437 {
1438         struct ceph_parse_opts_ctx *pctx;
1439         struct ceph_mount_options *fsopt;
1440
1441         pctx = kzalloc(sizeof(*pctx), GFP_KERNEL);
1442         if (!pctx)
1443                 return -ENOMEM;
1444
1445         pctx->copts = ceph_alloc_options();
1446         if (!pctx->copts)
1447                 goto nomem;
1448
1449         pctx->opts = kzalloc(sizeof(*pctx->opts), GFP_KERNEL);
1450         if (!pctx->opts)
1451                 goto nomem;
1452
1453         fsopt = pctx->opts;
1454         fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
1455
1456         fsopt->wsize = CEPH_MAX_WRITE_SIZE;
1457         fsopt->rsize = CEPH_MAX_READ_SIZE;
1458         fsopt->rasize = CEPH_RASIZE_DEFAULT;
1459         fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
1460         if (!fsopt->snapdir_name)
1461                 goto nomem;
1462
1463         fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
1464         fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
1465         fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
1466         fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
1467         fsopt->congestion_kb = default_congestion_kb();
1468
1469 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1470         fc->sb_flags |= SB_POSIXACL;
1471 #endif
1472
1473         fc->fs_private = pctx;
1474         fc->ops = &ceph_context_ops;
1475         return 0;
1476
1477 nomem:
1478         destroy_mount_options(pctx->opts);
1479         ceph_destroy_options(pctx->copts);
1480         kfree(pctx);
1481         return -ENOMEM;
1482 }
1483
1484 /*
1485  * Return true if it successfully increases the blocker counter,
1486  * or false if the mdsc is in stopping and flushed state.
1487  */
1488 static bool __inc_stopping_blocker(struct ceph_mds_client *mdsc)
1489 {
1490         spin_lock(&mdsc->stopping_lock);
1491         if (mdsc->stopping >= CEPH_MDSC_STOPPING_FLUSHING) {
1492                 spin_unlock(&mdsc->stopping_lock);
1493                 return false;
1494         }
1495         atomic_inc(&mdsc->stopping_blockers);
1496         spin_unlock(&mdsc->stopping_lock);
1497         return true;
1498 }
1499
1500 static void __dec_stopping_blocker(struct ceph_mds_client *mdsc)
1501 {
1502         spin_lock(&mdsc->stopping_lock);
1503         if (!atomic_dec_return(&mdsc->stopping_blockers) &&
1504             mdsc->stopping >= CEPH_MDSC_STOPPING_FLUSHING)
1505                 complete_all(&mdsc->stopping_waiter);
1506         spin_unlock(&mdsc->stopping_lock);
1507 }
1508
1509 /* For metadata IO requests */
1510 bool ceph_inc_mds_stopping_blocker(struct ceph_mds_client *mdsc,
1511                                    struct ceph_mds_session *session)
1512 {
1513         mutex_lock(&session->s_mutex);
1514         inc_session_sequence(session);
1515         mutex_unlock(&session->s_mutex);
1516
1517         return __inc_stopping_blocker(mdsc);
1518 }
1519
1520 void ceph_dec_mds_stopping_blocker(struct ceph_mds_client *mdsc)
1521 {
1522         __dec_stopping_blocker(mdsc);
1523 }
1524
1525 /* For data IO requests */
1526 bool ceph_inc_osd_stopping_blocker(struct ceph_mds_client *mdsc)
1527 {
1528         return __inc_stopping_blocker(mdsc);
1529 }
1530
1531 void ceph_dec_osd_stopping_blocker(struct ceph_mds_client *mdsc)
1532 {
1533         __dec_stopping_blocker(mdsc);
1534 }
1535
1536 static void ceph_kill_sb(struct super_block *s)
1537 {
1538         struct ceph_fs_client *fsc = ceph_sb_to_fs_client(s);
1539         struct ceph_client *cl = fsc->client;
1540         struct ceph_mds_client *mdsc = fsc->mdsc;
1541         bool wait;
1542
1543         doutc(cl, "%p\n", s);
1544
1545         ceph_mdsc_pre_umount(mdsc);
1546         flush_fs_workqueues(fsc);
1547
1548         /*
1549          * Though the kill_anon_super() will finally trigger the
1550          * sync_filesystem() anyway, we still need to do it here and
1551          * then bump the stage of shutdown. This will allow us to
1552          * drop any further message, which will increase the inodes'
1553          * i_count reference counters but makes no sense any more,
1554          * from MDSs.
1555          *
1556          * Without this when evicting the inodes it may fail in the
1557          * kill_anon_super(), which will trigger a warning when
1558          * destroying the fscrypt keyring and then possibly trigger
1559          * a further crash in ceph module when the iput() tries to
1560          * evict the inodes later.
1561          */
1562         sync_filesystem(s);
1563
1564         spin_lock(&mdsc->stopping_lock);
1565         mdsc->stopping = CEPH_MDSC_STOPPING_FLUSHING;
1566         wait = !!atomic_read(&mdsc->stopping_blockers);
1567         spin_unlock(&mdsc->stopping_lock);
1568
1569         if (wait && atomic_read(&mdsc->stopping_blockers)) {
1570                 long timeleft = wait_for_completion_killable_timeout(
1571                                         &mdsc->stopping_waiter,
1572                                         fsc->client->options->mount_timeout);
1573                 if (!timeleft) /* timed out */
1574                         pr_warn_client(cl, "umount timed out, %ld\n", timeleft);
1575                 else if (timeleft < 0) /* killed */
1576                         pr_warn_client(cl, "umount was killed, %ld\n", timeleft);
1577         }
1578
1579         mdsc->stopping = CEPH_MDSC_STOPPING_FLUSHED;
1580         kill_anon_super(s);
1581
1582         fsc->client->extra_mon_dispatch = NULL;
1583         ceph_fs_debugfs_cleanup(fsc);
1584
1585         ceph_fscache_unregister_fs(fsc);
1586
1587         destroy_fs_client(fsc);
1588 }
1589
1590 static struct file_system_type ceph_fs_type = {
1591         .owner          = THIS_MODULE,
1592         .name           = "ceph",
1593         .init_fs_context = ceph_init_fs_context,
1594         .kill_sb        = ceph_kill_sb,
1595         .fs_flags       = FS_RENAME_DOES_D_MOVE | FS_ALLOW_IDMAP,
1596 };
1597 MODULE_ALIAS_FS("ceph");
1598
1599 int ceph_force_reconnect(struct super_block *sb)
1600 {
1601         struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1602         int err = 0;
1603
1604         fsc->mount_state = CEPH_MOUNT_RECOVER;
1605         __ceph_umount_begin(fsc);
1606
1607         /* Make sure all page caches get invalidated.
1608          * see remove_session_caps_cb() */
1609         flush_workqueue(fsc->inode_wq);
1610
1611         /* In case that we were blocklisted. This also reset
1612          * all mon/osd connections */
1613         ceph_reset_client_addr(fsc->client);
1614
1615         ceph_osdc_clear_abort_err(&fsc->client->osdc);
1616
1617         fsc->blocklisted = false;
1618         fsc->mount_state = CEPH_MOUNT_MOUNTED;
1619
1620         if (sb->s_root) {
1621                 err = __ceph_do_getattr(d_inode(sb->s_root), NULL,
1622                                         CEPH_STAT_CAP_INODE, true);
1623         }
1624         return err;
1625 }
1626
1627 static int __init init_ceph(void)
1628 {
1629         int ret = init_caches();
1630         if (ret)
1631                 goto out;
1632
1633         ceph_flock_init();
1634         ret = register_filesystem(&ceph_fs_type);
1635         if (ret)
1636                 goto out_caches;
1637
1638         pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1639
1640         return 0;
1641
1642 out_caches:
1643         destroy_caches();
1644 out:
1645         return ret;
1646 }
1647
1648 static void __exit exit_ceph(void)
1649 {
1650         dout("exit_ceph\n");
1651         unregister_filesystem(&ceph_fs_type);
1652         destroy_caches();
1653 }
1654
1655 static int param_set_metrics(const char *val, const struct kernel_param *kp)
1656 {
1657         struct ceph_fs_client *fsc;
1658         int ret;
1659
1660         ret = param_set_bool(val, kp);
1661         if (ret) {
1662                 pr_err("Failed to parse sending metrics switch value '%s'\n",
1663                        val);
1664                 return ret;
1665         } else if (!disable_send_metrics) {
1666                 // wake up all the mds clients
1667                 spin_lock(&ceph_fsc_lock);
1668                 list_for_each_entry(fsc, &ceph_fsc_list, metric_wakeup) {
1669                         metric_schedule_delayed(&fsc->mdsc->metric);
1670                 }
1671                 spin_unlock(&ceph_fsc_lock);
1672         }
1673
1674         return 0;
1675 }
1676
1677 static const struct kernel_param_ops param_ops_metrics = {
1678         .set = param_set_metrics,
1679         .get = param_get_bool,
1680 };
1681
1682 bool disable_send_metrics = false;
1683 module_param_cb(disable_send_metrics, &param_ops_metrics, &disable_send_metrics, 0644);
1684 MODULE_PARM_DESC(disable_send_metrics, "Enable sending perf metrics to ceph cluster (default: on)");
1685
1686 /* for both v1 and v2 syntax */
1687 static bool mount_support = true;
1688 static const struct kernel_param_ops param_ops_mount_syntax = {
1689         .get = param_get_bool,
1690 };
1691 module_param_cb(mount_syntax_v1, &param_ops_mount_syntax, &mount_support, 0444);
1692 module_param_cb(mount_syntax_v2, &param_ops_mount_syntax, &mount_support, 0444);
1693
1694 bool enable_unsafe_idmap = false;
1695 module_param(enable_unsafe_idmap, bool, 0644);
1696 MODULE_PARM_DESC(enable_unsafe_idmap,
1697                  "Allow to use idmapped mounts with MDS without CEPHFS_FEATURE_HAS_OWNER_UIDGID");
1698
1699 module_init(init_ceph);
1700 module_exit(exit_ceph);
1701
1702 MODULE_AUTHOR("Sage Weil <[email protected]>");
1703 MODULE_AUTHOR("Yehuda Sadeh <[email protected]>");
1704 MODULE_AUTHOR("Patience Warnick <[email protected]>");
1705 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1706 MODULE_LICENSE("GPL");