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 /*
 154  * Remove an inode from the cache
 155  */
 156 static void exofs_destroy_inode(struct inode *inode)
 157 {
 158         kmem_cache_free(exofs_inode_cachep, exofs_i(inode));
 159 }
 160
 161 /*
 162  * Initialize the inode
 163  */
 164 static void exofs_init_once(void *foo)
 165 {
 166         struct exofs_i_info *oi = foo;
 167
 168         inode_init_once(&oi->vfs_inode);
 169 }
 170
 171 /*
 172  * Create and initialize the inode cache
 173  */
 174 static int init_inodecache(void)
 175 {
 176         exofs_inode_cachep = kmem_cache_create("exofs_inode_cache",
 177                                 sizeof(struct exofs_i_info), 0,
 178                                 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
 179                                 exofs_init_once);
 180         if (exofs_inode_cachep == NULL)
 181                 return -ENOMEM;
 182         return 0;
 183 }
 184
 185 /*
 186  * Destroy the inode cache
 187  */
 188 static void destroy_inodecache(void)
 189 {
 190         kmem_cache_destroy(exofs_inode_cachep);
 191 }
 192
 193 /******************************************************************************
 194  * SUPERBLOCK FUNCTIONS
 195  *****************************************************************************/
 196 static const struct super_operations exofs_sops;
 197 static const struct export_operations exofs_export_ops;
 198
 199 /*
 200  * Write the superblock to the OSD
 201  */
 202 int exofs_sync_fs(struct super_block *sb, int wait)
 203 {
 204         struct exofs_sb_info *sbi;
 205         struct exofs_fscb *fscb;
 206         struct exofs_io_state *ios;
 207         int ret = -ENOMEM;
 208
 209         lock_super(sb);
 210         sbi = sb->s_fs_info;
 211         fscb = &sbi->s_fscb;
 212
 213         ret = exofs_get_io_state(&sbi->layout, &ios);
 214         if (ret)
 215                 goto out;
 216
 217         /* Note: We only write the changing part of the fscb. .i.e upto the
 218          *       the fscb->s_dev_table_oid member. There is no read-modify-write
 219          *       here.
 220          */
 221         ios->length = offsetof(struct exofs_fscb, s_dev_table_oid);
 222         memset(fscb, 0, ios->length);
 223         fscb->s_nextid = cpu_to_le64(sbi->s_nextid);
 224         fscb->s_numfiles = cpu_to_le32(sbi->s_numfiles);
 225         fscb->s_magic = cpu_to_le16(sb->s_magic);
 226         fscb->s_newfs = 0;
 227         fscb->s_version = EXOFS_FSCB_VER;
 228
 229         ios->obj.id = EXOFS_SUPER_ID;
 230         ios->offset = 0;
 231         ios->kern_buff = fscb;
 232         ios->cred = sbi->s_cred;
 233
 234         ret = exofs_sbi_write(ios);
 235         if (unlikely(ret)) {
 236                 EXOFS_ERR("%s: exofs_sbi_write failed.\n", __func__);
 237                 goto out;
 238         }
 239         sb->s_dirt = 0;
 240
 241 out:
 242         EXOFS_DBGMSG("s_nextid=0x%llx ret=%d\n", _LLU(sbi->s_nextid), ret);
 243         exofs_put_io_state(ios);
 244         unlock_super(sb);
 245         return ret;
 246 }
 247
 248 static void exofs_write_super(struct super_block *sb)
 249 {
 250         if (!(sb->s_flags & MS_RDONLY))
 251                 exofs_sync_fs(sb, 1);
 252         else
 253                 sb->s_dirt = 0;
 254 }
 255
 256 static void _exofs_print_device(const char *msg, const char *dev_path,
 257                                 struct osd_dev *od, u64 pid)
 258 {
 259         const struct osd_dev_info *odi = osduld_device_info(od);
 260
 261         printk(KERN_NOTICE "exofs: %s %s osd_name-%s pid-0x%llx\n",
 262                 msg, dev_path ?: "", odi->osdname, _LLU(pid));
 263 }
 264
 265 void exofs_free_sbi(struct exofs_sb_info *sbi)
 266 {
 267         while (sbi->layout.s_numdevs) {
 268                 int i = --sbi->layout.s_numdevs;
 269                 struct osd_dev *od = sbi->layout.s_ods[i];
 270
 271                 if (od) {
 272                         sbi->layout.s_ods[i] = NULL;
 273                         osduld_put_device(od);
 274                 }
 275         }
 276         kfree(sbi);
 277 }
 278
 279 /*
 280  * This function is called when the vfs is freeing the superblock.  We just
 281  * need to free our own part.
 282  */
 283 static void exofs_put_super(struct super_block *sb)
 284 {
 285         int num_pend;
 286         struct exofs_sb_info *sbi = sb->s_fs_info;
 287
 288         if (sb->s_dirt)
 289                 exofs_write_super(sb);
 290
 291         /* make sure there are no pending commands */
 292         for (num_pend = atomic_read(&sbi->s_curr_pending); num_pend > 0;
 293              num_pend = atomic_read(&sbi->s_curr_pending)) {
 294                 wait_queue_head_t wq;
 295                 init_waitqueue_head(&wq);
 296                 wait_event_timeout(wq,
 297                                   (atomic_read(&sbi->s_curr_pending) == 0),
 298                                   msecs_to_jiffies(100));
 299         }
 300
 301         _exofs_print_device("Unmounting", NULL, sbi->layout.s_ods[0],
 302                             sbi->layout.s_pid);
 303
 304         bdi_destroy(&sbi->bdi);
 305         exofs_free_sbi(sbi);
 306         sb->s_fs_info = NULL;
 307 }
 308
 309 static int _read_and_match_data_map(struct exofs_sb_info *sbi, unsigned numdevs,
 310                                     struct exofs_device_table *dt)
 311 {
 312         u64 stripe_length;
 313
 314         sbi->data_map.odm_num_comps   =
 315                                 le32_to_cpu(dt->dt_data_map.cb_num_comps);
 316         sbi->data_map.odm_stripe_unit =
 317                                 le64_to_cpu(dt->dt_data_map.cb_stripe_unit);
 318         sbi->data_map.odm_group_width =
 319                                 le32_to_cpu(dt->dt_data_map.cb_group_width);
 320         sbi->data_map.odm_group_depth =
 321                                 le32_to_cpu(dt->dt_data_map.cb_group_depth);
 322         sbi->data_map.odm_mirror_cnt  =
 323                                 le32_to_cpu(dt->dt_data_map.cb_mirror_cnt);
 324         sbi->data_map.odm_raid_algorithm  =
 325                                 le32_to_cpu(dt->dt_data_map.cb_raid_algorithm);
 326
 327 /* FIXME: Only raid0 for now. if not so, do not mount */
 328         if (sbi->data_map.odm_num_comps != numdevs) {
 329                 EXOFS_ERR("odm_num_comps(%u) != numdevs(%u)\n",
 330                           sbi->data_map.odm_num_comps, numdevs);
 331                 return -EINVAL;
 332         }
 333         if (sbi->data_map.odm_raid_algorithm != PNFS_OSD_RAID_0) {
 334                 EXOFS_ERR("Only RAID_0 for now\n");
 335                 return -EINVAL;
 336         }
 337         if (0 != (numdevs % (sbi->data_map.odm_mirror_cnt + 1))) {
 338                 EXOFS_ERR("Data Map wrong, numdevs=%d mirrors=%d\n",
 339                           numdevs, sbi->data_map.odm_mirror_cnt);
 340                 return -EINVAL;
 341         }
 342
 343         if (0 != (sbi->data_map.odm_stripe_unit & ~PAGE_MASK)) {
 344                 EXOFS_ERR("Stripe Unit(0x%llx)"
 345                           " must be Multples of PAGE_SIZE(0x%lx)\n",
 346                           _LLU(sbi->data_map.odm_stripe_unit), PAGE_SIZE);
 347                 return -EINVAL;
 348         }
 349
 350         sbi->layout.stripe_unit = sbi->data_map.odm_stripe_unit;
 351         sbi->layout.mirrors_p1 = sbi->data_map.odm_mirror_cnt + 1;
 352
 353         if (sbi->data_map.odm_group_width) {
 354                 sbi->layout.group_width = sbi->data_map.odm_group_width;
 355                 sbi->layout.group_depth = sbi->data_map.odm_group_depth;
 356                 if (!sbi->layout.group_depth) {
 357                         EXOFS_ERR("group_depth == 0 && group_width != 0\n");
 358                         return -EINVAL;
 359                 }
 360                 sbi->layout.group_count = sbi->data_map.odm_num_comps /
 361                                                 sbi->layout.mirrors_p1 /
 362                                                 sbi->data_map.odm_group_width;
 363         } else {
 364                 if (sbi->data_map.odm_group_depth) {
 365                         printk(KERN_NOTICE "Warning: group_depth ignored "
 366                                 "group_width == 0 && group_depth == %d\n",
 367                                 sbi->data_map.odm_group_depth);
 368                         sbi->data_map.odm_group_depth = 0;
 369                 }
 370                 sbi->layout.group_width = sbi->data_map.odm_num_comps /
 371                                                         sbi->layout.mirrors_p1;
 372                 sbi->layout.group_depth = -1;
 373                 sbi->layout.group_count = 1;
 374         }
 375
 376         stripe_length = (u64)sbi->layout.group_width * sbi->layout.stripe_unit;
 377         if (stripe_length >= (1ULL << 32)) {
 378                 EXOFS_ERR("Total Stripe length(0x%llx)"
 379                           " >= 32bit is not supported\n", _LLU(stripe_length));
 380                 return -EINVAL;
 381         }
 382
 383         return 0;
 384 }
 385
 386 /* @odi is valid only as long as @fscb_dev is valid */
 387 static int exofs_devs_2_odi(struct exofs_dt_device_info *dt_dev,
 388                              struct osd_dev_info *odi)
 389 {
 390         odi->systemid_len = le32_to_cpu(dt_dev->systemid_len);
 391         memcpy(odi->systemid, dt_dev->systemid, odi->systemid_len);
 392
 393         odi->osdname_len = le32_to_cpu(dt_dev->osdname_len);
 394         odi->osdname = dt_dev->osdname;
 395
 396         /* FIXME support long names. Will need a _put function */
 397         if (dt_dev->long_name_offset)
 398                 return -EINVAL;
 399
 400         /* Make sure osdname is printable!
 401          * mkexofs should give us space for a null-terminator else the
 402          * device-table is invalid.
 403          */
 404         if (unlikely(odi->osdname_len >= sizeof(dt_dev->osdname)))
 405                 odi->osdname_len = sizeof(dt_dev->osdname) - 1;
 406         dt_dev->osdname[odi->osdname_len] = 0;
 407
 408         /* If it's all zeros something is bad we read past end-of-obj */
 409         return !(odi->systemid_len || odi->osdname_len);
 410 }
 411
 412 static int exofs_read_lookup_dev_table(struct exofs_sb_info **psbi,
 413                                        unsigned table_count)
 414 {
 415         struct exofs_sb_info *sbi = *psbi;
 416         struct osd_dev *fscb_od;
 417         struct osd_obj_id obj = {.partition = sbi->layout.s_pid,
 418                                  .id = EXOFS_DEVTABLE_ID};
 419         struct exofs_device_table *dt;
 420         unsigned table_bytes = table_count * sizeof(dt->dt_dev_table[0]) +
 421                                              sizeof(*dt);
 422         unsigned numdevs, i;
 423         int ret;
 424
 425         dt = kmalloc(table_bytes, GFP_KERNEL);
 426         if (unlikely(!dt)) {
 427                 EXOFS_ERR("ERROR: allocating %x bytes for device table\n",
 428                           table_bytes);
 429                 return -ENOMEM;
 430         }
 431
 432         fscb_od = sbi->layout.s_ods[0];
 433         sbi->layout.s_ods[0] = NULL;
 434         sbi->layout.s_numdevs = 0;
 435         ret = exofs_read_kern(fscb_od, sbi->s_cred, &obj, 0, dt, table_bytes);
 436         if (unlikely(ret)) {
 437                 EXOFS_ERR("ERROR: reading device table\n");
 438                 goto out;
 439         }
 440
 441         numdevs = le64_to_cpu(dt->dt_num_devices);
 442         if (unlikely(!numdevs)) {
 443                 ret = -EINVAL;
 444                 goto out;
 445         }
 446         WARN_ON(table_count != numdevs);
 447
 448         ret = _read_and_match_data_map(sbi, numdevs, dt);
 449         if (unlikely(ret))
 450                 goto out;
 451
 452         if (likely(numdevs > 1)) {
 453                 unsigned size = numdevs * sizeof(sbi->layout.s_ods[0]);
 454
 455                 sbi = krealloc(sbi, sizeof(*sbi) + size, GFP_KERNEL);
 456                 if (unlikely(!sbi)) {
 457                         ret = -ENOMEM;
 458                         goto out;
 459                 }
 460                 memset(&sbi->layout.s_ods[1], 0,
 461                        size - sizeof(sbi->layout.s_ods[0]));
 462                 *psbi = sbi;
 463         }
 464
 465         for (i = 0; i < numdevs; i++) {
 466                 struct exofs_fscb fscb;
 467                 struct osd_dev_info odi;
 468                 struct osd_dev *od;
 469
 470                 if (exofs_devs_2_odi(&dt->dt_dev_table[i], &odi)) {
 471                         EXOFS_ERR("ERROR: Read all-zeros device entry\n");
 472                         ret = -EINVAL;
 473                         goto out;
 474                 }
 475
 476                 printk(KERN_NOTICE "Add device[%d]: osd_name-%s\n",
 477                        i, odi.osdname);
 478
 479                 /* On all devices the device table is identical. The user can
 480                  * specify any one of the participating devices on the command
 481                  * line. We always keep them in device-table order.
 482                  */
 483                 if (fscb_od && osduld_device_same(fscb_od, &odi)) {
 484                         sbi->layout.s_ods[i] = fscb_od;
 485                         ++sbi->layout.s_numdevs;
 486                         fscb_od = NULL;
 487                         continue;
 488                 }
 489
 490                 od = osduld_info_lookup(&odi);
 491                 if (unlikely(IS_ERR(od))) {
 492                         ret = PTR_ERR(od);
 493                         EXOFS_ERR("ERROR: device requested is not found "
 494                                   "osd_name-%s =>%d\n", odi.osdname, ret);
 495                         goto out;
 496                 }
 497
 498                 sbi->layout.s_ods[i] = od;
 499                 ++sbi->layout.s_numdevs;
 500
 501                 /* Read the fscb of the other devices to make sure the FS
 502                  * partition is there.
 503                  */
 504                 ret = exofs_read_kern(od, sbi->s_cred, &obj, 0, &fscb,
 505                                       sizeof(fscb));
 506                 if (unlikely(ret)) {
 507                         EXOFS_ERR("ERROR: Malformed participating device "
 508                                   "error reading fscb osd_name-%s\n",
 509                                   odi.osdname);
 510                         goto out;
 511                 }
 512
 513                 /* TODO: verify other information is correct and FS-uuid
 514                  *       matches. Benny what did you say about device table
 515                  *       generation and old devices?
 516                  */
 517         }
 518
 519 out:
 520         kfree(dt);
 521         if (unlikely(!ret && fscb_od)) {
 522                 EXOFS_ERR(
 523                       "ERROR: Bad device-table container device not present\n");
 524                 osduld_put_device(fscb_od);
 525                 ret = -EINVAL;
 526         }
 527
 528         return ret;
 529 }
 530
 531 /*
 532  * Read the superblock from the OSD and fill in the fields
 533  */
 534 static int exofs_fill_super(struct super_block *sb, void *data, int silent)
 535 {
 536         struct inode *root;
 537         struct exofs_mountopt *opts = data;
 538         struct exofs_sb_info *sbi;      /*extended info                  */
 539         struct osd_dev *od;             /* Master device                 */
 540         struct exofs_fscb fscb;         /*on-disk superblock info        */
 541         struct osd_obj_id obj;
 542         unsigned table_count;
 543         int ret;
 544
 545         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
 546         if (!sbi)
 547                 return -ENOMEM;
 548
 549         ret = bdi_setup_and_register(&sbi->bdi, "exofs", BDI_CAP_MAP_COPY);
 550         if (ret)
 551                 goto free_bdi;
 552
 553         /* use mount options to fill superblock */
 554         od = osduld_path_lookup(opts->dev_name);
 555         if (IS_ERR(od)) {
 556                 ret = PTR_ERR(od);
 557                 goto free_sbi;
 558         }
 559
 560         /* Default layout in case we do not have a device-table */
 561         sbi->layout.stripe_unit = PAGE_SIZE;
 562         sbi->layout.mirrors_p1 = 1;
 563         sbi->layout.group_width = 1;
 564         sbi->layout.group_depth = -1;
 565         sbi->layout.group_count = 1;
 566         sbi->layout.s_ods[0] = od;
 567         sbi->layout.s_numdevs = 1;
 568         sbi->layout.s_pid = opts->pid;
 569         sbi->s_timeout = opts->timeout;
 570
 571         /* fill in some other data by hand */
 572         memset(sb->s_id, 0, sizeof(sb->s_id));
 573         strcpy(sb->s_id, "exofs");
 574         sb->s_blocksize = EXOFS_BLKSIZE;
 575         sb->s_blocksize_bits = EXOFS_BLKSHIFT;
 576         sb->s_maxbytes = MAX_LFS_FILESIZE;
 577         atomic_set(&sbi->s_curr_pending, 0);
 578         sb->s_bdev = NULL;
 579         sb->s_dev = 0;
 580
 581         obj.partition = sbi->layout.s_pid;
 582         obj.id = EXOFS_SUPER_ID;
 583         exofs_make_credential(sbi->s_cred, &obj);
 584
 585         ret = exofs_read_kern(od, sbi->s_cred, &obj, 0, &fscb, sizeof(fscb));
 586         if (unlikely(ret))
 587                 goto free_sbi;
 588
 589         sb->s_magic = le16_to_cpu(fscb.s_magic);
 590         sbi->s_nextid = le64_to_cpu(fscb.s_nextid);
 591         sbi->s_numfiles = le32_to_cpu(fscb.s_numfiles);
 592
 593         /* make sure what we read from the object store is correct */
 594         if (sb->s_magic != EXOFS_SUPER_MAGIC) {
 595                 if (!silent)
 596                         EXOFS_ERR("ERROR: Bad magic value\n");
 597                 ret = -EINVAL;
 598                 goto free_sbi;
 599         }
 600         if (le32_to_cpu(fscb.s_version) != EXOFS_FSCB_VER) {
 601                 EXOFS_ERR("ERROR: Bad FSCB version expected-%d got-%d\n",
 602                           EXOFS_FSCB_VER, le32_to_cpu(fscb.s_version));
 603                 ret = -EINVAL;
 604                 goto free_sbi;
 605         }
 606
 607         /* start generation numbers from a random point */
 608         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
 609         spin_lock_init(&sbi->s_next_gen_lock);
 610
 611         table_count = le64_to_cpu(fscb.s_dev_table_count);
 612         if (table_count) {
 613                 ret = exofs_read_lookup_dev_table(&sbi, table_count);
 614                 if (unlikely(ret))
 615                         goto free_sbi;
 616         }
 617
 618         /* set up operation vectors */
 619         sb->s_bdi = &sbi->bdi;
 620         sb->s_fs_info = sbi;
 621         sb->s_op = &exofs_sops;
 622         sb->s_export_op = &exofs_export_ops;
 623         root = exofs_iget(sb, EXOFS_ROOT_ID - EXOFS_OBJ_OFF);
 624         if (IS_ERR(root)) {
 625                 EXOFS_ERR("ERROR: exofs_iget failed\n");
 626                 ret = PTR_ERR(root);
 627                 goto free_sbi;
 628         }
 629         sb->s_root = d_alloc_root(root);
 630         if (!sb->s_root) {
 631                 iput(root);
 632                 EXOFS_ERR("ERROR: get root inode failed\n");
 633                 ret = -ENOMEM;
 634                 goto free_sbi;
 635         }
 636
 637         if (!S_ISDIR(root->i_mode)) {
 638                 dput(sb->s_root);
 639                 sb->s_root = NULL;
 640                 EXOFS_ERR("ERROR: corrupt root inode (mode = %hd)\n",
 641                        root->i_mode);
 642                 ret = -EINVAL;
 643                 goto free_sbi;
 644         }
 645
 646         _exofs_print_device("Mounting", opts->dev_name, sbi->layout.s_ods[0],
 647                             sbi->layout.s_pid);
 648         return 0;
 649
 650 free_sbi:
 651         bdi_destroy(&sbi->bdi);
 652 free_bdi:
 653         EXOFS_ERR("Unable to mount exofs on %s pid=0x%llx err=%d\n",
 654                   opts->dev_name, sbi->layout.s_pid, ret);
 655         exofs_free_sbi(sbi);
 656         return ret;
 657 }
 658
 659 /*
 660  * Set up the superblock (calls exofs_fill_super eventually)
 661  */
 662 static int exofs_get_sb(struct file_system_type *type,
 663                           int flags, const char *dev_name,
 664                           void *data, struct vfsmount *mnt)
 665 {
 666         struct exofs_mountopt opts;
 667         int ret;
 668
 669         ret = parse_options(data, &opts);
 670         if (ret)
 671                 return ret;
 672
 673         opts.dev_name = dev_name;
 674         return get_sb_nodev(type, flags, &opts, exofs_fill_super, mnt);
 675 }
 676
 677 /*
 678  * Return information about the file system state in the buffer.  This is used
 679  * by the 'df' command, for example.
 680  */
 681 static int exofs_statfs(struct dentry *dentry, struct kstatfs *buf)
 682 {
 683         struct super_block *sb = dentry->d_sb;
 684         struct exofs_sb_info *sbi = sb->s_fs_info;
 685         struct exofs_io_state *ios;
 686         struct osd_attr attrs[] = {
 687                 ATTR_DEF(OSD_APAGE_PARTITION_QUOTAS,
 688                         OSD_ATTR_PQ_CAPACITY_QUOTA, sizeof(__be64)),
 689                 ATTR_DEF(OSD_APAGE_PARTITION_INFORMATION,
 690                         OSD_ATTR_PI_USED_CAPACITY, sizeof(__be64)),
 691         };
 692         uint64_t capacity = ULLONG_MAX;
 693         uint64_t used = ULLONG_MAX;
 694         uint8_t cred_a[OSD_CAP_LEN];
 695         int ret;
 696
 697         ret = exofs_get_io_state(&sbi->layout, &ios);
 698         if (ret) {
 699                 EXOFS_DBGMSG("exofs_get_io_state failed.\n");
 700                 return ret;
 701         }
 702
 703         exofs_make_credential(cred_a, &ios->obj);
 704         ios->cred = sbi->s_cred;
 705         ios->in_attr = attrs;
 706         ios->in_attr_len = ARRAY_SIZE(attrs);
 707
 708         ret = exofs_sbi_read(ios);
 709         if (unlikely(ret))
 710                 goto out;
 711
 712         ret = extract_attr_from_ios(ios, &attrs[0]);
 713         if (likely(!ret)) {
 714                 capacity = get_unaligned_be64(attrs[0].val_ptr);
 715                 if (unlikely(!capacity))
 716                         capacity = ULLONG_MAX;
 717         } else
 718                 EXOFS_DBGMSG("exofs_statfs: get capacity failed.\n");
 719
 720         ret = extract_attr_from_ios(ios, &attrs[1]);
 721         if (likely(!ret))
 722                 used = get_unaligned_be64(attrs[1].val_ptr);
 723         else
 724                 EXOFS_DBGMSG("exofs_statfs: get used-space failed.\n");
 725
 726         /* fill in the stats buffer */
 727         buf->f_type = EXOFS_SUPER_MAGIC;
 728         buf->f_bsize = EXOFS_BLKSIZE;
 729         buf->f_blocks = capacity >> 9;
 730         buf->f_bfree = (capacity - used) >> 9;
 731         buf->f_bavail = buf->f_bfree;
 732         buf->f_files = sbi->s_numfiles;
 733         buf->f_ffree = EXOFS_MAX_ID - sbi->s_numfiles;
 734         buf->f_namelen = EXOFS_NAME_LEN;
 735
 736 out:
 737         exofs_put_io_state(ios);
 738         return ret;
 739 }
 740
 741 static const struct super_operations exofs_sops = {
 742         .alloc_inode    = exofs_alloc_inode,
 743         .destroy_inode  = exofs_destroy_inode,
 744         .write_inode    = exofs_write_inode,
 745         .evict_inode    = exofs_evict_inode,
 746         .put_super      = exofs_put_super,
 747         .write_super    = exofs_write_super,
 748         .sync_fs        = exofs_sync_fs,
 749         .statfs         = exofs_statfs,
 750 };
 751
 752 /******************************************************************************
 753  * EXPORT OPERATIONS
 754  *****************************************************************************/
 755
 756 struct dentry *exofs_get_parent(struct dentry *child)
 757 {
 758         unsigned long ino = exofs_parent_ino(child);
 759
 760         if (!ino)
 761                 return NULL;
 762
 763         return d_obtain_alias(exofs_iget(child->d_inode->i_sb, ino));
 764 }
 765
 766 static struct inode *exofs_nfs_get_inode(struct super_block *sb,
 767                 u64 ino, u32 generation)
 768 {
 769         struct inode *inode;
 770
 771         inode = exofs_iget(sb, ino);
 772         if (IS_ERR(inode))
 773                 return ERR_CAST(inode);
 774         if (generation && inode->i_generation != generation) {
 775                 /* we didn't find the right inode.. */
 776                 iput(inode);
 777                 return ERR_PTR(-ESTALE);
 778         }
 779         return inode;
 780 }
 781
 782 static struct dentry *exofs_fh_to_dentry(struct super_block *sb,
 783                                 struct fid *fid, int fh_len, int fh_type)
 784 {
 785         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 786                                     exofs_nfs_get_inode);
 787 }
 788
 789 static struct dentry *exofs_fh_to_parent(struct super_block *sb,
 790                                 struct fid *fid, int fh_len, int fh_type)
 791 {
 792         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 793                                     exofs_nfs_get_inode);
 794 }
 795
 796 static const struct export_operations exofs_export_ops = {
 797         .fh_to_dentry = exofs_fh_to_dentry,
 798         .fh_to_parent = exofs_fh_to_parent,
 799         .get_parent = exofs_get_parent,
 800 };
 801
 802 /******************************************************************************
 803  * INSMOD/RMMOD
 804  *****************************************************************************/
 805
 806 /*
 807  * struct that describes this file system
 808  */
 809 static struct file_system_type exofs_type = {
 810         .owner          = THIS_MODULE,
 811         .name           = "exofs",
 812         .get_sb         = exofs_get_sb,
 813         .kill_sb        = generic_shutdown_super,
 814 };
 815
 816 static int __init init_exofs(void)
 817 {
 818         int err;
 819
 820         err = init_inodecache();
 821         if (err)
 822                 goto out;
 823
 824         err = register_filesystem(&exofs_type);
 825         if (err)
 826                 goto out_d;
 827
 828         return 0;
 829 out_d:
 830         destroy_inodecache();
 831 out:
 832         return err;
 833 }
 834
 835 static void __exit exit_exofs(void)
 836 {
 837         unregister_filesystem(&exofs_type);
 838         destroy_inodecache();
 839 }
 840
 841 MODULE_AUTHOR("Avishay Traeger <avishay@gmail.com>");
 842 MODULE_DESCRIPTION("exofs");
 843 MODULE_LICENSE("GPL");
 844
 845 module_init(init_exofs)
 846 module_exit(exit_exofs)