fs/exofs/super.c

   1 /*
   2  * Copyright (C) 2005, 2006
   3  * Avishay Traeger (avishay@gmail.com)
   4  * Copyright (C) 2008, 2009
   5  * Boaz Harrosh <bharrosh@panasas.com>
   6  *
   7  * Copyrights for code taken from ext2:
   8  *     Copyright (C) 1992, 1993, 1994, 1995
   9  *     Remy Card (card@masi.ibp.fr)
  10  *     Laboratoire MASI - Institut Blaise Pascal
  11  *     Universite Pierre et Marie Curie (Paris VI)
  12  *     from
  13  *     linux/fs/minix/inode.c
  14  *     Copyright (C) 1991, 1992  Linus Torvalds
  15  *
  16  * This file is part of exofs.
  17  *
  18  * exofs is free software; you can redistribute it and/or modify
  19  * it under the terms of the GNU General Public License as published by
  20  * the Free Software Foundation.  Since it is based on ext2, and the only
  21  * valid version of GPL for the Linux kernel is version 2, the only valid
  22  * version of GPL for exofs is version 2.
  23  *
  24  * exofs is distributed in the hope that it will be useful,
  25  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  26  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  27  * GNU General Public License for more details.
  28  *
  29  * You should have received a copy of the GNU General Public License
  30  * along with exofs; if not, write to the Free Software
  31  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  32  */
  33
  34 #include <linux/string.h>
  35 #include <linux/parser.h>
  36 #include <linux/vfs.h>
  37 #include <linux/random.h>
  38 #include <linux/exportfs.h>
  39 #include <linux/slab.h>
  40
  41 #include "exofs.h"
  42
  43 /******************************************************************************
  44  * MOUNT OPTIONS
  45  *****************************************************************************/
  46
  47 /*
  48  * struct to hold what we get from mount options
  49  */
  50 struct exofs_mountopt {
  51         const char *dev_name;
  52         uint64_t pid;
  53         int timeout;
  54 };
  55
  56 /*
  57  * exofs-specific mount-time options.
  58  */
  59 enum { Opt_pid, Opt_to, Opt_mkfs, Opt_format, Opt_err };
  60
  61 /*
  62  * Our mount-time options.  These should ideally be 64-bit unsigned, but the
  63  * kernel's parsing functions do not currently support that.  32-bit should be
  64  * sufficient for most applications now.
  65  */
  66 static match_table_t tokens = {
  67         {Opt_pid, "pid=%u"},
  68         {Opt_to, "to=%u"},
  69         {Opt_err, NULL}
  70 };
  71
  72 /*
  73  * The main option parsing method.  Also makes sure that all of the mandatory
  74  * mount options were set.
  75  */
  76 static int parse_options(char *options, struct exofs_mountopt *opts)
  77 {
  78         char *p;
  79         substring_t args[MAX_OPT_ARGS];
  80         int option;
  81         bool s_pid = false;
  82
  83         EXOFS_DBGMSG("parse_options %s\n", options);
  84         /* defaults */
  85         memset(opts, 0, sizeof(*opts));
  86         opts->timeout = BLK_DEFAULT_SG_TIMEOUT;
  87
  88         while ((p = strsep(&options, ",")) != NULL) {
  89                 int token;
  90                 char str[32];
  91
  92                 if (!*p)
  93                         continue;
  94
  95                 token = match_token(p, tokens, args);
  96                 switch (token) {
  97                 case Opt_pid:
  98                         if (0 == match_strlcpy(str, &args[0], sizeof(str)))
  99                                 return -EINVAL;
 100                         opts->pid = simple_strtoull(str, NULL, 0);
 101                         if (opts->pid < EXOFS_MIN_PID) {
 102                                 EXOFS_ERR("Partition ID must be >= %u",
 103                                           EXOFS_MIN_PID);
 104                                 return -EINVAL;
 105                         }
 106                         s_pid = 1;
 107                         break;
 108                 case Opt_to:
 109                         if (match_int(&args[0], &option))
 110                                 return -EINVAL;
 111                         if (option <= 0) {
 112                                 EXOFS_ERR("Timout must be > 0");
 113                                 return -EINVAL;
 114                         }
 115                         opts->timeout = option * HZ;
 116                         break;
 117                 }
 118         }
 119
 120         if (!s_pid) {
 121                 EXOFS_ERR("Need to specify the following options:\n");
 122                 EXOFS_ERR("    -o pid=pid_no_to_use\n");
 123                 return -EINVAL;
 124         }
 125
 126         return 0;
 127 }
 128
 129 /******************************************************************************
 130  * INODE CACHE
 131  *****************************************************************************/
 132
 133 /*
 134  * Our inode cache.  Isn't it pretty?
 135  */
 136 static struct kmem_cache *exofs_inode_cachep;
 137
 138 /*
 139  * Allocate an inode in the cache
 140  */
 141 static struct inode *exofs_alloc_inode(struct super_block *sb)
 142 {
 143         struct exofs_i_info *oi;
 144
 145         oi = kmem_cache_alloc(exofs_inode_cachep, GFP_KERNEL);
 146         if (!oi)
 147                 return NULL;
 148
 149         oi->vfs_inode.i_version = 1;
 150         return &oi->vfs_inode;
 151 }
 152
 153 static void exofs_i_callback(struct rcu_head *head)
 154 {
 155         struct inode *inode = container_of(head, struct inode, i_rcu);
 156         INIT_LIST_HEAD(&inode->i_dentry);
 157         kmem_cache_free(exofs_inode_cachep, exofs_i(inode));
 158 }
 159
 160 /*
 161  * Remove an inode from the cache
 162  */
 163 static void exofs_destroy_inode(struct inode *inode)
 164 {
 165         call_rcu(&inode->i_rcu, exofs_i_callback);
 166 }
 167
 168 /*
 169  * Initialize the inode
 170  */
 171 static void exofs_init_once(void *foo)
 172 {
 173         struct exofs_i_info *oi = foo;
 174
 175         inode_init_once(&oi->vfs_inode);
 176 }
 177
 178 /*
 179  * Create and initialize the inode cache
 180  */
 181 static int init_inodecache(void)
 182 {
 183         exofs_inode_cachep = kmem_cache_create("exofs_inode_cache",
 184                                 sizeof(struct exofs_i_info), 0,
 185                                 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
 186                                 exofs_init_once);
 187         if (exofs_inode_cachep == NULL)
 188                 return -ENOMEM;
 189         return 0;
 190 }
 191
 192 /*
 193  * Destroy the inode cache
 194  */
 195 static void destroy_inodecache(void)
 196 {
 197         kmem_cache_destroy(exofs_inode_cachep);
 198 }
 199
 200 /******************************************************************************
 201  * SUPERBLOCK FUNCTIONS
 202  *****************************************************************************/
 203 static const struct super_operations exofs_sops;
 204 static const struct export_operations exofs_export_ops;
 205
 206 /*
 207  * Write the superblock to the OSD
 208  */
 209 int exofs_sync_fs(struct super_block *sb, int wait)
 210 {
 211         struct exofs_sb_info *sbi;
 212         struct exofs_fscb *fscb;
 213         struct exofs_io_state *ios;
 214         int ret = -ENOMEM;
 215
 216         lock_super(sb);
 217         sbi = sb->s_fs_info;
 218         fscb = &sbi->s_fscb;
 219
 220         ret = exofs_get_io_state(&sbi->layout, &ios);
 221         if (ret)
 222                 goto out;
 223
 224         /* Note: We only write the changing part of the fscb. .i.e upto the
 225          *       the fscb->s_dev_table_oid member. There is no read-modify-write
 226          *       here.
 227          */
 228         ios->length = offsetof(struct exofs_fscb, s_dev_table_oid);
 229         memset(fscb, 0, ios->length);
 230         fscb->s_nextid = cpu_to_le64(sbi->s_nextid);
 231         fscb->s_numfiles = cpu_to_le32(sbi->s_numfiles);
 232         fscb->s_magic = cpu_to_le16(sb->s_magic);
 233         fscb->s_newfs = 0;
 234         fscb->s_version = EXOFS_FSCB_VER;
 235
 236         ios->obj.id = EXOFS_SUPER_ID;
 237         ios->offset = 0;
 238         ios->kern_buff = fscb;
 239         ios->cred = sbi->s_cred;
 240
 241         ret = exofs_sbi_write(ios);
 242         if (unlikely(ret)) {
 243                 EXOFS_ERR("%s: exofs_sbi_write failed.\n", __func__);
 244                 goto out;
 245         }
 246         sb->s_dirt = 0;
 247
 248 out:
 249         EXOFS_DBGMSG("s_nextid=0x%llx ret=%d\n", _LLU(sbi->s_nextid), ret);
 250         exofs_put_io_state(ios);
 251         unlock_super(sb);
 252         return ret;
 253 }
 254
 255 static void exofs_write_super(struct super_block *sb)
 256 {
 257         if (!(sb->s_flags & MS_RDONLY))
 258                 exofs_sync_fs(sb, 1);
 259         else
 260                 sb->s_dirt = 0;
 261 }
 262
 263 static void _exofs_print_device(const char *msg, const char *dev_path,
 264                                 struct osd_dev *od, u64 pid)
 265 {
 266         const struct osd_dev_info *odi = osduld_device_info(od);
 267
 268         printk(KERN_NOTICE "exofs: %s %s osd_name-%s pid-0x%llx\n",
 269                 msg, dev_path ?: "", odi->osdname, _LLU(pid));
 270 }
 271
 272 void exofs_free_sbi(struct exofs_sb_info *sbi)
 273 {
 274         while (sbi->layout.s_numdevs) {
 275                 int i = --sbi->layout.s_numdevs;
 276                 struct osd_dev *od = sbi->layout.s_ods[i];
 277
 278                 if (od) {
 279                         sbi->layout.s_ods[i] = NULL;
 280                         osduld_put_device(od);
 281                 }
 282         }
 283         kfree(sbi);
 284 }
 285
 286 /*
 287  * This function is called when the vfs is freeing the superblock.  We just
 288  * need to free our own part.
 289  */
 290 static void exofs_put_super(struct super_block *sb)
 291 {
 292         int num_pend;
 293         struct exofs_sb_info *sbi = sb->s_fs_info;
 294
 295         if (sb->s_dirt)
 296                 exofs_write_super(sb);
 297
 298         /* make sure there are no pending commands */
 299         for (num_pend = atomic_read(&sbi->s_curr_pending); num_pend > 0;
 300              num_pend = atomic_read(&sbi->s_curr_pending)) {
 301                 wait_queue_head_t wq;
 302                 init_waitqueue_head(&wq);
 303                 wait_event_timeout(wq,
 304                                   (atomic_read(&sbi->s_curr_pending) == 0),
 305                                   msecs_to_jiffies(100));
 306         }
 307
 308         _exofs_print_device("Unmounting", NULL, sbi->layout.s_ods[0],
 309                             sbi->layout.s_pid);
 310
 311         bdi_destroy(&sbi->bdi);
 312         exofs_free_sbi(sbi);
 313         sb->s_fs_info = NULL;
 314 }
 315
 316 static int _read_and_match_data_map(struct exofs_sb_info *sbi, unsigned numdevs,
 317                                     struct exofs_device_table *dt)
 318 {
 319         u64 stripe_length;
 320
 321         sbi->data_map.odm_num_comps   =
 322                                 le32_to_cpu(dt->dt_data_map.cb_num_comps);
 323         sbi->data_map.odm_stripe_unit =
 324                                 le64_to_cpu(dt->dt_data_map.cb_stripe_unit);
 325         sbi->data_map.odm_group_width =
 326                                 le32_to_cpu(dt->dt_data_map.cb_group_width);
 327         sbi->data_map.odm_group_depth =
 328                                 le32_to_cpu(dt->dt_data_map.cb_group_depth);
 329         sbi->data_map.odm_mirror_cnt  =
 330                                 le32_to_cpu(dt->dt_data_map.cb_mirror_cnt);
 331         sbi->data_map.odm_raid_algorithm  =
 332                                 le32_to_cpu(dt->dt_data_map.cb_raid_algorithm);
 333
 334 /* FIXME: Only raid0 for now. if not so, do not mount */
 335         if (sbi->data_map.odm_num_comps != numdevs) {
 336                 EXOFS_ERR("odm_num_comps(%u) != numdevs(%u)\n",
 337                           sbi->data_map.odm_num_comps, numdevs);
 338                 return -EINVAL;
 339         }
 340         if (sbi->data_map.odm_raid_algorithm != PNFS_OSD_RAID_0) {
 341                 EXOFS_ERR("Only RAID_0 for now\n");
 342                 return -EINVAL;
 343         }
 344         if (0 != (numdevs % (sbi->data_map.odm_mirror_cnt + 1))) {
 345                 EXOFS_ERR("Data Map wrong, numdevs=%d mirrors=%d\n",
 346                           numdevs, sbi->data_map.odm_mirror_cnt);
 347                 return -EINVAL;
 348         }
 349
 350         if (0 != (sbi->data_map.odm_stripe_unit & ~PAGE_MASK)) {
 351                 EXOFS_ERR("Stripe Unit(0x%llx)"
 352                           " must be Multples of PAGE_SIZE(0x%lx)\n",
 353                           _LLU(sbi->data_map.odm_stripe_unit), PAGE_SIZE);
 354                 return -EINVAL;
 355         }
 356
 357         sbi->layout.stripe_unit = sbi->data_map.odm_stripe_unit;
 358         sbi->layout.mirrors_p1 = sbi->data_map.odm_mirror_cnt + 1;
 359
 360         if (sbi->data_map.odm_group_width) {
 361                 sbi->layout.group_width = sbi->data_map.odm_group_width;
 362                 sbi->layout.group_depth = sbi->data_map.odm_group_depth;
 363                 if (!sbi->layout.group_depth) {
 364                         EXOFS_ERR("group_depth == 0 && group_width != 0\n");
 365                         return -EINVAL;
 366                 }
 367                 sbi->layout.group_count = sbi->data_map.odm_num_comps /
 368                                                 sbi->layout.mirrors_p1 /
 369                                                 sbi->data_map.odm_group_width;
 370         } else {
 371                 if (sbi->data_map.odm_group_depth) {
 372                         printk(KERN_NOTICE "Warning: group_depth ignored "
 373                                 "group_width == 0 && group_depth == %d\n",
 374                                 sbi->data_map.odm_group_depth);
 375                         sbi->data_map.odm_group_depth = 0;
 376                 }
 377                 sbi->layout.group_width = sbi->data_map.odm_num_comps /
 378                                                         sbi->layout.mirrors_p1;
 379                 sbi->layout.group_depth = -1;
 380                 sbi->layout.group_count = 1;
 381         }
 382
 383         stripe_length = (u64)sbi->layout.group_width * sbi->layout.stripe_unit;
 384         if (stripe_length >= (1ULL << 32)) {
 385                 EXOFS_ERR("Total Stripe length(0x%llx)"
 386                           " >= 32bit is not supported\n", _LLU(stripe_length));
 387                 return -EINVAL;
 388         }
 389
 390         return 0;
 391 }
 392
 393 /* @odi is valid only as long as @fscb_dev is valid */
 394 static int exofs_devs_2_odi(struct exofs_dt_device_info *dt_dev,
 395                              struct osd_dev_info *odi)
 396 {
 397         odi->systemid_len = le32_to_cpu(dt_dev->systemid_len);
 398         memcpy(odi->systemid, dt_dev->systemid, odi->systemid_len);
 399
 400         odi->osdname_len = le32_to_cpu(dt_dev->osdname_len);
 401         odi->osdname = dt_dev->osdname;
 402
 403         /* FIXME support long names. Will need a _put function */
 404         if (dt_dev->long_name_offset)
 405                 return -EINVAL;
 406
 407         /* Make sure osdname is printable!
 408          * mkexofs should give us space for a null-terminator else the
 409          * device-table is invalid.
 410          */
 411         if (unlikely(odi->osdname_len >= sizeof(dt_dev->osdname)))
 412                 odi->osdname_len = sizeof(dt_dev->osdname) - 1;
 413         dt_dev->osdname[odi->osdname_len] = 0;
 414
 415         /* If it's all zeros something is bad we read past end-of-obj */
 416         return !(odi->systemid_len || odi->osdname_len);
 417 }
 418
 419 static int exofs_read_lookup_dev_table(struct exofs_sb_info **psbi,
 420                                        unsigned table_count)
 421 {
 422         struct exofs_sb_info *sbi = *psbi;
 423         struct osd_dev *fscb_od;
 424         struct osd_obj_id obj = {.partition = sbi->layout.s_pid,
 425                                  .id = EXOFS_DEVTABLE_ID};
 426         struct exofs_device_table *dt;
 427         unsigned table_bytes = table_count * sizeof(dt->dt_dev_table[0]) +
 428                                              sizeof(*dt);
 429         unsigned numdevs, i;
 430         int ret;
 431
 432         dt = kmalloc(table_bytes, GFP_KERNEL);
 433         if (unlikely(!dt)) {
 434                 EXOFS_ERR("ERROR: allocating %x bytes for device table\n",
 435                           table_bytes);
 436                 return -ENOMEM;
 437         }
 438
 439         fscb_od = sbi->layout.s_ods[0];
 440         sbi->layout.s_ods[0] = NULL;
 441         sbi->layout.s_numdevs = 0;
 442         ret = exofs_read_kern(fscb_od, sbi->s_cred, &obj, 0, dt, table_bytes);
 443         if (unlikely(ret)) {
 444                 EXOFS_ERR("ERROR: reading device table\n");
 445                 goto out;
 446         }
 447
 448         numdevs = le64_to_cpu(dt->dt_num_devices);
 449         if (unlikely(!numdevs)) {
 450                 ret = -EINVAL;
 451                 goto out;
 452         }
 453         WARN_ON(table_count != numdevs);
 454
 455         ret = _read_and_match_data_map(sbi, numdevs, dt);
 456         if (unlikely(ret))
 457                 goto out;
 458
 459         if (likely(numdevs > 1)) {
 460                 unsigned size = numdevs * sizeof(sbi->layout.s_ods[0]);
 461
 462                 sbi = krealloc(sbi, sizeof(*sbi) + size, GFP_KERNEL);
 463                 if (unlikely(!sbi)) {
 464                         ret = -ENOMEM;
 465                         goto out;
 466                 }
 467                 memset(&sbi->layout.s_ods[1], 0,
 468                        size - sizeof(sbi->layout.s_ods[0]));
 469                 *psbi = sbi;
 470         }
 471
 472         for (i = 0; i < numdevs; i++) {
 473                 struct exofs_fscb fscb;
 474                 struct osd_dev_info odi;
 475                 struct osd_dev *od;
 476
 477                 if (exofs_devs_2_odi(&dt->dt_dev_table[i], &odi)) {
 478                         EXOFS_ERR("ERROR: Read all-zeros device entry\n");
 479                         ret = -EINVAL;
 480                         goto out;
 481                 }
 482
 483                 printk(KERN_NOTICE "Add device[%d]: osd_name-%s\n",
 484                        i, odi.osdname);
 485
 486                 /* On all devices the device table is identical. The user can
 487                  * specify any one of the participating devices on the command
 488                  * line. We always keep them in device-table order.
 489                  */
 490                 if (fscb_od && osduld_device_same(fscb_od, &odi)) {
 491                         sbi->layout.s_ods[i] = fscb_od;
 492                         ++sbi->layout.s_numdevs;
 493                         fscb_od = NULL;
 494                         continue;
 495                 }
 496
 497                 od = osduld_info_lookup(&odi);
 498                 if (unlikely(IS_ERR(od))) {
 499                         ret = PTR_ERR(od);
 500                         EXOFS_ERR("ERROR: device requested is not found "
 501                                   "osd_name-%s =>%d\n", odi.osdname, ret);
 502                         goto out;
 503                 }
 504
 505                 sbi->layout.s_ods[i] = od;
 506                 ++sbi->layout.s_numdevs;
 507
 508                 /* Read the fscb of the other devices to make sure the FS
 509                  * partition is there.
 510                  */
 511                 ret = exofs_read_kern(od, sbi->s_cred, &obj, 0, &fscb,
 512                                       sizeof(fscb));
 513                 if (unlikely(ret)) {
 514                         EXOFS_ERR("ERROR: Malformed participating device "
 515                                   "error reading fscb osd_name-%s\n",
 516                                   odi.osdname);
 517                         goto out;
 518                 }
 519
 520                 /* TODO: verify other information is correct and FS-uuid
 521                  *       matches. Benny what did you say about device table
 522                  *       generation and old devices?
 523                  */
 524         }
 525
 526 out:
 527         kfree(dt);
 528         if (unlikely(!ret && fscb_od)) {
 529                 EXOFS_ERR(
 530                       "ERROR: Bad device-table container device not present\n");
 531                 osduld_put_device(fscb_od);
 532                 ret = -EINVAL;
 533         }
 534
 535         return ret;
 536 }
 537
 538 /*
 539  * Read the superblock from the OSD and fill in the fields
 540  */
 541 static int exofs_fill_super(struct super_block *sb, void *data, int silent)
 542 {
 543         struct inode *root;
 544         struct exofs_mountopt *opts = data;
 545         struct exofs_sb_info *sbi;      /*extended info                  */
 546         struct osd_dev *od;             /* Master device                 */
 547         struct exofs_fscb fscb;         /*on-disk superblock info        */
 548         struct osd_obj_id obj;
 549         unsigned table_count;
 550         int ret;
 551
 552         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
 553         if (!sbi)
 554                 return -ENOMEM;
 555
 556         ret = bdi_setup_and_register(&sbi->bdi, "exofs", BDI_CAP_MAP_COPY);
 557         if (ret)
 558                 goto free_bdi;
 559
 560         /* use mount options to fill superblock */
 561         od = osduld_path_lookup(opts->dev_name);
 562         if (IS_ERR(od)) {
 563                 ret = PTR_ERR(od);
 564                 goto free_sbi;
 565         }
 566
 567         /* Default layout in case we do not have a device-table */
 568         sbi->layout.stripe_unit = PAGE_SIZE;
 569         sbi->layout.mirrors_p1 = 1;
 570         sbi->layout.group_width = 1;
 571         sbi->layout.group_depth = -1;
 572         sbi->layout.group_count = 1;
 573         sbi->layout.s_ods[0] = od;
 574         sbi->layout.s_numdevs = 1;
 575         sbi->layout.s_pid = opts->pid;
 576         sbi->s_timeout = opts->timeout;
 577
 578         /* fill in some other data by hand */
 579         memset(sb->s_id, 0, sizeof(sb->s_id));
 580         strcpy(sb->s_id, "exofs");
 581         sb->s_blocksize = EXOFS_BLKSIZE;
 582         sb->s_blocksize_bits = EXOFS_BLKSHIFT;
 583         sb->s_maxbytes = MAX_LFS_FILESIZE;
 584         atomic_set(&sbi->s_curr_pending, 0);
 585         sb->s_bdev = NULL;
 586         sb->s_dev = 0;
 587
 588         obj.partition = sbi->layout.s_pid;
 589         obj.id = EXOFS_SUPER_ID;
 590         exofs_make_credential(sbi->s_cred, &obj);
 591
 592         ret = exofs_read_kern(od, sbi->s_cred, &obj, 0, &fscb, sizeof(fscb));
 593         if (unlikely(ret))
 594                 goto free_sbi;
 595
 596         sb->s_magic = le16_to_cpu(fscb.s_magic);
 597         sbi->s_nextid = le64_to_cpu(fscb.s_nextid);
 598         sbi->s_numfiles = le32_to_cpu(fscb.s_numfiles);
 599
 600         /* make sure what we read from the object store is correct */
 601         if (sb->s_magic != EXOFS_SUPER_MAGIC) {
 602                 if (!silent)
 603                         EXOFS_ERR("ERROR: Bad magic value\n");
 604                 ret = -EINVAL;
 605                 goto free_sbi;
 606         }
 607         if (le32_to_cpu(fscb.s_version) != EXOFS_FSCB_VER) {
 608                 EXOFS_ERR("ERROR: Bad FSCB version expected-%d got-%d\n",
 609                           EXOFS_FSCB_VER, le32_to_cpu(fscb.s_version));
 610                 ret = -EINVAL;
 611                 goto free_sbi;
 612         }
 613
 614         /* start generation numbers from a random point */
 615         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
 616         spin_lock_init(&sbi->s_next_gen_lock);
 617
 618         table_count = le64_to_cpu(fscb.s_dev_table_count);
 619         if (table_count) {
 620                 ret = exofs_read_lookup_dev_table(&sbi, table_count);
 621                 if (unlikely(ret))
 622                         goto free_sbi;
 623         }
 624
 625         /* set up operation vectors */
 626         sb->s_bdi = &sbi->bdi;
 627         sb->s_fs_info = sbi;
 628         sb->s_op = &exofs_sops;
 629         sb->s_export_op = &exofs_export_ops;
 630         root = exofs_iget(sb, EXOFS_ROOT_ID - EXOFS_OBJ_OFF);
 631         if (IS_ERR(root)) {
 632                 EXOFS_ERR("ERROR: exofs_iget failed\n");
 633                 ret = PTR_ERR(root);
 634                 goto free_sbi;
 635         }
 636         sb->s_root = d_alloc_root(root);
 637         if (!sb->s_root) {
 638                 iput(root);
 639                 EXOFS_ERR("ERROR: get root inode failed\n");
 640                 ret = -ENOMEM;
 641                 goto free_sbi;
 642         }
 643
 644         if (!S_ISDIR(root->i_mode)) {
 645                 dput(sb->s_root);
 646                 sb->s_root = NULL;
 647                 EXOFS_ERR("ERROR: corrupt root inode (mode = %hd)\n",
 648                        root->i_mode);
 649                 ret = -EINVAL;
 650                 goto free_sbi;
 651         }
 652
 653         _exofs_print_device("Mounting", opts->dev_name, sbi->layout.s_ods[0],
 654                             sbi->layout.s_pid);
 655         return 0;
 656
 657 free_sbi:
 658         bdi_destroy(&sbi->bdi);
 659 free_bdi:
 660         EXOFS_ERR("Unable to mount exofs on %s pid=0x%llx err=%d\n",
 661                   opts->dev_name, sbi->layout.s_pid, ret);
 662         exofs_free_sbi(sbi);
 663         return ret;
 664 }
 665
 666 /*
 667  * Set up the superblock (calls exofs_fill_super eventually)
 668  */
 669 static struct dentry *exofs_mount(struct file_system_type *type,
 670                           int flags, const char *dev_name,
 671                           void *data)
 672 {
 673         struct exofs_mountopt opts;
 674         int ret;
 675
 676         ret = parse_options(data, &opts);
 677         if (ret)
 678                 return ERR_PTR(ret);
 679
 680         opts.dev_name = dev_name;
 681         return mount_nodev(type, flags, &opts, exofs_fill_super);
 682 }
 683
 684 /*
 685  * Return information about the file system state in the buffer.  This is used
 686  * by the 'df' command, for example.
 687  */
 688 static int exofs_statfs(struct dentry *dentry, struct kstatfs *buf)
 689 {
 690         struct super_block *sb = dentry->d_sb;
 691         struct exofs_sb_info *sbi = sb->s_fs_info;
 692         struct exofs_io_state *ios;
 693         struct osd_attr attrs[] = {
 694                 ATTR_DEF(OSD_APAGE_PARTITION_QUOTAS,
 695                         OSD_ATTR_PQ_CAPACITY_QUOTA, sizeof(__be64)),
 696                 ATTR_DEF(OSD_APAGE_PARTITION_INFORMATION,
 697                         OSD_ATTR_PI_USED_CAPACITY, sizeof(__be64)),
 698         };
 699         uint64_t capacity = ULLONG_MAX;
 700         uint64_t used = ULLONG_MAX;
 701         uint8_t cred_a[OSD_CAP_LEN];
 702         int ret;
 703
 704         ret = exofs_get_io_state(&sbi->layout, &ios);
 705         if (ret) {
 706                 EXOFS_DBGMSG("exofs_get_io_state failed.\n");
 707                 return ret;
 708         }
 709
 710         exofs_make_credential(cred_a, &ios->obj);
 711         ios->cred = sbi->s_cred;
 712         ios->in_attr = attrs;
 713         ios->in_attr_len = ARRAY_SIZE(attrs);
 714
 715         ret = exofs_sbi_read(ios);
 716         if (unlikely(ret))
 717                 goto out;
 718
 719         ret = extract_attr_from_ios(ios, &attrs[0]);
 720         if (likely(!ret)) {
 721                 capacity = get_unaligned_be64(attrs[0].val_ptr);
 722                 if (unlikely(!capacity))
 723                         capacity = ULLONG_MAX;
 724         } else
 725                 EXOFS_DBGMSG("exofs_statfs: get capacity failed.\n");
 726
 727         ret = extract_attr_from_ios(ios, &attrs[1]);
 728         if (likely(!ret))
 729                 used = get_unaligned_be64(attrs[1].val_ptr);
 730         else
 731                 EXOFS_DBGMSG("exofs_statfs: get used-space failed.\n");
 732
 733         /* fill in the stats buffer */
 734         buf->f_type = EXOFS_SUPER_MAGIC;
 735         buf->f_bsize = EXOFS_BLKSIZE;
 736         buf->f_blocks = capacity >> 9;
 737         buf->f_bfree = (capacity - used) >> 9;
 738         buf->f_bavail = buf->f_bfree;
 739         buf->f_files = sbi->s_numfiles;
 740         buf->f_ffree = EXOFS_MAX_ID - sbi->s_numfiles;
 741         buf->f_namelen = EXOFS_NAME_LEN;
 742
 743 out:
 744         exofs_put_io_state(ios);
 745         return ret;
 746 }
 747
 748 static const struct super_operations exofs_sops = {
 749         .alloc_inode    = exofs_alloc_inode,
 750         .destroy_inode  = exofs_destroy_inode,
 751         .write_inode    = exofs_write_inode,
 752         .evict_inode    = exofs_evict_inode,
 753         .put_super      = exofs_put_super,
 754         .write_super    = exofs_write_super,
 755         .sync_fs        = exofs_sync_fs,
 756         .statfs         = exofs_statfs,
 757 };
 758
 759 /******************************************************************************
 760  * EXPORT OPERATIONS
 761  *****************************************************************************/
 762
 763 struct dentry *exofs_get_parent(struct dentry *child)
 764 {
 765         unsigned long ino = exofs_parent_ino(child);
 766
 767         if (!ino)
 768                 return NULL;
 769
 770         return d_obtain_alias(exofs_iget(child->d_inode->i_sb, ino));
 771 }
 772
 773 static struct inode *exofs_nfs_get_inode(struct super_block *sb,
 774                 u64 ino, u32 generation)
 775 {
 776         struct inode *inode;
 777
 778         inode = exofs_iget(sb, ino);
 779         if (IS_ERR(inode))
 780                 return ERR_CAST(inode);
 781         if (generation && inode->i_generation != generation) {
 782                 /* we didn't find the right inode.. */
 783                 iput(inode);
 784                 return ERR_PTR(-ESTALE);
 785         }
 786         return inode;
 787 }
 788
 789 static struct dentry *exofs_fh_to_dentry(struct super_block *sb,
 790                                 struct fid *fid, int fh_len, int fh_type)
 791 {
 792         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 793                                     exofs_nfs_get_inode);
 794 }
 795
 796 static struct dentry *exofs_fh_to_parent(struct super_block *sb,
 797                                 struct fid *fid, int fh_len, int fh_type)
 798 {
 799         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 800                                     exofs_nfs_get_inode);
 801 }
 802
 803 static const struct export_operations exofs_export_ops = {
 804         .fh_to_dentry = exofs_fh_to_dentry,
 805         .fh_to_parent = exofs_fh_to_parent,
 806         .get_parent = exofs_get_parent,
 807 };
 808
 809 /******************************************************************************
 810  * INSMOD/RMMOD
 811  *****************************************************************************/
 812
 813 /*
 814  * struct that describes this file system
 815  */
 816 static struct file_system_type exofs_type = {
 817         .owner          = THIS_MODULE,
 818         .name           = "exofs",
 819         .mount          = exofs_mount,
 820         .kill_sb        = generic_shutdown_super,
 821 };
 822
 823 static int __init init_exofs(void)
 824 {
 825         int err;
 826
 827         err = init_inodecache();
 828         if (err)
 829                 goto out;
 830
 831         err = register_filesystem(&exofs_type);
 832         if (err)
 833                 goto out_d;
 834
 835         return 0;
 836 out_d:
 837         destroy_inodecache();
 838 out:
 839         return err;
 840 }
 841
 842 static void __exit exit_exofs(void)
 843 {
 844         unregister_filesystem(&exofs_type);
 845         destroy_inodecache();
 846 }
 847
 848 MODULE_AUTHOR("Avishay Traeger <avishay@gmail.com>");
 849 MODULE_DESCRIPTION("exofs");
 850 MODULE_LICENSE("GPL");
 851
 852 module_init(init_exofs)
 853 module_exit(exit_exofs)