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         bool is_osdname;
  52         const char *dev_name;
  53         uint64_t pid;
  54         int timeout;
  55 };
  56
  57 /*
  58  * exofs-specific mount-time options.
  59  */
  60 enum { Opt_name, Opt_pid, Opt_to, Opt_err };
  61
  62 /*
  63  * Our mount-time options.  These should ideally be 64-bit unsigned, but the
  64  * kernel's parsing functions do not currently support that.  32-bit should be
  65  * sufficient for most applications now.
  66  */
  67 static match_table_t tokens = {
  68         {Opt_name, "osdname=%s"},
  69         {Opt_pid, "pid=%u"},
  70         {Opt_to, "to=%u"},
  71         {Opt_err, NULL}
  72 };
  73
  74 /*
  75  * The main option parsing method.  Also makes sure that all of the mandatory
  76  * mount options were set.
  77  */
  78 static int parse_options(char *options, struct exofs_mountopt *opts)
  79 {
  80         char *p;
  81         substring_t args[MAX_OPT_ARGS];
  82         int option;
  83         bool s_pid = false;
  84
  85         EXOFS_DBGMSG("parse_options %s\n", options);
  86         /* defaults */
  87         memset(opts, 0, sizeof(*opts));
  88         opts->timeout = BLK_DEFAULT_SG_TIMEOUT;
  89
  90         while ((p = strsep(&options, ",")) != NULL) {
  91                 int token;
  92                 char str[32];
  93
  94                 if (!*p)
  95                         continue;
  96
  97                 token = match_token(p, tokens, args);
  98                 switch (token) {
  99                 case Opt_name:
 100                         opts->dev_name = match_strdup(&args[0]);
 101                         if (unlikely(!opts->dev_name)) {
 102                                 EXOFS_ERR("Error allocating dev_name");
 103                                 return -ENOMEM;
 104                         }
 105                         opts->is_osdname = true;
 106                         break;
 107                 case Opt_pid:
 108                         if (0 == match_strlcpy(str, &args[0], sizeof(str)))
 109                                 return -EINVAL;
 110                         opts->pid = simple_strtoull(str, NULL, 0);
 111                         if (opts->pid < EXOFS_MIN_PID) {
 112                                 EXOFS_ERR("Partition ID must be >= %u",
 113                                           EXOFS_MIN_PID);
 114                                 return -EINVAL;
 115                         }
 116                         s_pid = 1;
 117                         break;
 118                 case Opt_to:
 119                         if (match_int(&args[0], &option))
 120                                 return -EINVAL;
 121                         if (option <= 0) {
 122                                 EXOFS_ERR("Timout must be > 0");
 123                                 return -EINVAL;
 124                         }
 125                         opts->timeout = option * HZ;
 126                         break;
 127                 }
 128         }
 129
 130         if (!s_pid) {
 131                 EXOFS_ERR("Need to specify the following options:\n");
 132                 EXOFS_ERR("    -o pid=pid_no_to_use\n");
 133                 return -EINVAL;
 134         }
 135
 136         return 0;
 137 }
 138
 139 /******************************************************************************
 140  * INODE CACHE
 141  *****************************************************************************/
 142
 143 /*
 144  * Our inode cache.  Isn't it pretty?
 145  */
 146 static struct kmem_cache *exofs_inode_cachep;
 147
 148 /*
 149  * Allocate an inode in the cache
 150  */
 151 static struct inode *exofs_alloc_inode(struct super_block *sb)
 152 {
 153         struct exofs_i_info *oi;
 154
 155         oi = kmem_cache_alloc(exofs_inode_cachep, GFP_KERNEL);
 156         if (!oi)
 157                 return NULL;
 158
 159         oi->vfs_inode.i_version = 1;
 160         return &oi->vfs_inode;
 161 }
 162
 163 static void exofs_i_callback(struct rcu_head *head)
 164 {
 165         struct inode *inode = container_of(head, struct inode, i_rcu);
 166         INIT_LIST_HEAD(&inode->i_dentry);
 167         kmem_cache_free(exofs_inode_cachep, exofs_i(inode));
 168 }
 169
 170 /*
 171  * Remove an inode from the cache
 172  */
 173 static void exofs_destroy_inode(struct inode *inode)
 174 {
 175         call_rcu(&inode->i_rcu, exofs_i_callback);
 176 }
 177
 178 /*
 179  * Initialize the inode
 180  */
 181 static void exofs_init_once(void *foo)
 182 {
 183         struct exofs_i_info *oi = foo;
 184
 185         inode_init_once(&oi->vfs_inode);
 186 }
 187
 188 /*
 189  * Create and initialize the inode cache
 190  */
 191 static int init_inodecache(void)
 192 {
 193         exofs_inode_cachep = kmem_cache_create("exofs_inode_cache",
 194                                 sizeof(struct exofs_i_info), 0,
 195                                 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
 196                                 exofs_init_once);
 197         if (exofs_inode_cachep == NULL)
 198                 return -ENOMEM;
 199         return 0;
 200 }
 201
 202 /*
 203  * Destroy the inode cache
 204  */
 205 static void destroy_inodecache(void)
 206 {
 207         kmem_cache_destroy(exofs_inode_cachep);
 208 }
 209
 210 /******************************************************************************
 211  * SUPERBLOCK FUNCTIONS
 212  *****************************************************************************/
 213 static const struct super_operations exofs_sops;
 214 static const struct export_operations exofs_export_ops;
 215
 216 static const struct osd_attr g_attr_sb_stats = ATTR_DEF(
 217         EXOFS_APAGE_SB_DATA,
 218         EXOFS_ATTR_SB_STATS,
 219         sizeof(struct exofs_sb_stats));
 220
 221 static int __sbi_read_stats(struct exofs_sb_info *sbi)
 222 {
 223         struct osd_attr attrs[] = {
 224                 [0] = g_attr_sb_stats,
 225         };
 226         struct exofs_io_state *ios;
 227         int ret;
 228
 229         ret = exofs_get_io_state(&sbi->layout, &ios);
 230         if (unlikely(ret)) {
 231                 EXOFS_ERR("%s: exofs_get_io_state failed.\n", __func__);
 232                 return ret;
 233         }
 234
 235         ios->cred = sbi->s_cred;
 236
 237         ios->in_attr = attrs;
 238         ios->in_attr_len = ARRAY_SIZE(attrs);
 239
 240         ret = exofs_sbi_read(ios);
 241         if (unlikely(ret)) {
 242                 EXOFS_ERR("Error reading super_block stats => %d\n", ret);
 243                 goto out;
 244         }
 245
 246         ret = extract_attr_from_ios(ios, &attrs[0]);
 247         if (ret) {
 248                 EXOFS_ERR("%s: extract_attr of sb_stats failed\n", __func__);
 249                 goto out;
 250         }
 251         if (attrs[0].len) {
 252                 struct exofs_sb_stats *ess;
 253
 254                 if (unlikely(attrs[0].len != sizeof(*ess))) {
 255                         EXOFS_ERR("%s: Wrong version of exofs_sb_stats "
 256                                   "size(%d) != expected(%zd)\n",
 257                                   __func__, attrs[0].len, sizeof(*ess));
 258                         goto out;
 259                 }
 260
 261                 ess = attrs[0].val_ptr;
 262                 sbi->s_nextid = le64_to_cpu(ess->s_nextid);
 263                 sbi->s_numfiles = le32_to_cpu(ess->s_numfiles);
 264         }
 265
 266 out:
 267         exofs_put_io_state(ios);
 268         return ret;
 269 }
 270
 271 static void stats_done(struct exofs_io_state *ios, void *p)
 272 {
 273         exofs_put_io_state(ios);
 274         /* Good thanks nothing to do anymore */
 275 }
 276
 277 /* Asynchronously write the stats attribute */
 278 int exofs_sbi_write_stats(struct exofs_sb_info *sbi)
 279 {
 280         struct osd_attr attrs[] = {
 281                 [0] = g_attr_sb_stats,
 282         };
 283         struct exofs_io_state *ios;
 284         int ret;
 285
 286         ret = exofs_get_io_state(&sbi->layout, &ios);
 287         if (unlikely(ret)) {
 288                 EXOFS_ERR("%s: exofs_get_io_state failed.\n", __func__);
 289                 return ret;
 290         }
 291
 292         sbi->s_ess.s_nextid   = cpu_to_le64(sbi->s_nextid);
 293         sbi->s_ess.s_numfiles = cpu_to_le64(sbi->s_numfiles);
 294         attrs[0].val_ptr = &sbi->s_ess;
 295
 296         ios->cred = sbi->s_cred;
 297         ios->done = stats_done;
 298         ios->private = sbi;
 299         ios->out_attr = attrs;
 300         ios->out_attr_len = ARRAY_SIZE(attrs);
 301
 302         ret = exofs_sbi_write(ios);
 303         if (unlikely(ret)) {
 304                 EXOFS_ERR("%s: exofs_sbi_write failed.\n", __func__);
 305                 exofs_put_io_state(ios);
 306         }
 307
 308         return ret;
 309 }
 310
 311 /*
 312  * Write the superblock to the OSD
 313  */
 314 int exofs_sync_fs(struct super_block *sb, int wait)
 315 {
 316         struct exofs_sb_info *sbi;
 317         struct exofs_fscb *fscb;
 318         struct exofs_io_state *ios;
 319         int ret = -ENOMEM;
 320
 321         fscb = kmalloc(sizeof(*fscb), GFP_KERNEL);
 322         if (unlikely(!fscb))
 323                 return -ENOMEM;
 324
 325         sbi = sb->s_fs_info;
 326
 327         /* NOTE: We no longer dirty the super_block anywhere in exofs. The
 328          * reason we write the fscb here on unmount is so we can stay backwards
 329          * compatible with fscb->s_version == 1. (What we are not compatible
 330          * with is if a new version FS crashed and then we try to mount an old
 331          * version). Otherwise the exofs_fscb is read-only from mkfs time. All
 332          * the writeable info is set in exofs_sbi_write_stats() above.
 333          */
 334         ret = exofs_get_io_state(&sbi->layout, &ios);
 335         if (unlikely(ret))
 336                 goto out;
 337
 338         lock_super(sb);
 339
 340         ios->length = offsetof(struct exofs_fscb, s_dev_table_oid);
 341         memset(fscb, 0, ios->length);
 342         fscb->s_nextid = cpu_to_le64(sbi->s_nextid);
 343         fscb->s_numfiles = cpu_to_le32(sbi->s_numfiles);
 344         fscb->s_magic = cpu_to_le16(sb->s_magic);
 345         fscb->s_newfs = 0;
 346         fscb->s_version = EXOFS_FSCB_VER;
 347
 348         ios->obj.id = EXOFS_SUPER_ID;
 349         ios->offset = 0;
 350         ios->kern_buff = fscb;
 351         ios->cred = sbi->s_cred;
 352
 353         ret = exofs_sbi_write(ios);
 354         if (unlikely(ret))
 355                 EXOFS_ERR("%s: exofs_sbi_write failed.\n", __func__);
 356         else
 357                 sb->s_dirt = 0;
 358
 359
 360         unlock_super(sb);
 361 out:
 362         EXOFS_DBGMSG("s_nextid=0x%llx ret=%d\n", _LLU(sbi->s_nextid), ret);
 363         exofs_put_io_state(ios);
 364         kfree(fscb);
 365         return ret;
 366 }
 367
 368 static void exofs_write_super(struct super_block *sb)
 369 {
 370         if (!(sb->s_flags & MS_RDONLY))
 371                 exofs_sync_fs(sb, 1);
 372         else
 373                 sb->s_dirt = 0;
 374 }
 375
 376 static void _exofs_print_device(const char *msg, const char *dev_path,
 377                                 struct osd_dev *od, u64 pid)
 378 {
 379         const struct osd_dev_info *odi = osduld_device_info(od);
 380
 381         printk(KERN_NOTICE "exofs: %s %s osd_name-%s pid-0x%llx\n",
 382                 msg, dev_path ?: "", odi->osdname, _LLU(pid));
 383 }
 384
 385 void exofs_free_sbi(struct exofs_sb_info *sbi)
 386 {
 387         while (sbi->layout.s_numdevs) {
 388                 int i = --sbi->layout.s_numdevs;
 389                 struct osd_dev *od = sbi->layout.s_ods[i];
 390
 391                 if (od) {
 392                         sbi->layout.s_ods[i] = NULL;
 393                         osduld_put_device(od);
 394                 }
 395         }
 396         kfree(sbi);
 397 }
 398
 399 /*
 400  * This function is called when the vfs is freeing the superblock.  We just
 401  * need to free our own part.
 402  */
 403 static void exofs_put_super(struct super_block *sb)
 404 {
 405         int num_pend;
 406         struct exofs_sb_info *sbi = sb->s_fs_info;
 407
 408         /* make sure there are no pending commands */
 409         for (num_pend = atomic_read(&sbi->s_curr_pending); num_pend > 0;
 410              num_pend = atomic_read(&sbi->s_curr_pending)) {
 411                 wait_queue_head_t wq;
 412
 413                 printk(KERN_NOTICE "%s: !!Pending operations in flight. "
 414                        "This is a BUG. please report to osd-dev@open-osd.org\n",
 415                        __func__);
 416                 init_waitqueue_head(&wq);
 417                 wait_event_timeout(wq,
 418                                   (atomic_read(&sbi->s_curr_pending) == 0),
 419                                   msecs_to_jiffies(100));
 420         }
 421
 422         _exofs_print_device("Unmounting", NULL, sbi->layout.s_ods[0],
 423                             sbi->layout.s_pid);
 424
 425         bdi_destroy(&sbi->bdi);
 426         exofs_free_sbi(sbi);
 427         sb->s_fs_info = NULL;
 428 }
 429
 430 static int _read_and_match_data_map(struct exofs_sb_info *sbi, unsigned numdevs,
 431                                     struct exofs_device_table *dt)
 432 {
 433         u64 stripe_length;
 434
 435         sbi->data_map.odm_num_comps   =
 436                                 le32_to_cpu(dt->dt_data_map.cb_num_comps);
 437         sbi->data_map.odm_stripe_unit =
 438                                 le64_to_cpu(dt->dt_data_map.cb_stripe_unit);
 439         sbi->data_map.odm_group_width =
 440                                 le32_to_cpu(dt->dt_data_map.cb_group_width);
 441         sbi->data_map.odm_group_depth =
 442                                 le32_to_cpu(dt->dt_data_map.cb_group_depth);
 443         sbi->data_map.odm_mirror_cnt  =
 444                                 le32_to_cpu(dt->dt_data_map.cb_mirror_cnt);
 445         sbi->data_map.odm_raid_algorithm  =
 446                                 le32_to_cpu(dt->dt_data_map.cb_raid_algorithm);
 447
 448 /* FIXME: Only raid0 for now. if not so, do not mount */
 449         if (sbi->data_map.odm_num_comps != numdevs) {
 450                 EXOFS_ERR("odm_num_comps(%u) != numdevs(%u)\n",
 451                           sbi->data_map.odm_num_comps, numdevs);
 452                 return -EINVAL;
 453         }
 454         if (sbi->data_map.odm_raid_algorithm != PNFS_OSD_RAID_0) {
 455                 EXOFS_ERR("Only RAID_0 for now\n");
 456                 return -EINVAL;
 457         }
 458         if (0 != (numdevs % (sbi->data_map.odm_mirror_cnt + 1))) {
 459                 EXOFS_ERR("Data Map wrong, numdevs=%d mirrors=%d\n",
 460                           numdevs, sbi->data_map.odm_mirror_cnt);
 461                 return -EINVAL;
 462         }
 463
 464         if (0 != (sbi->data_map.odm_stripe_unit & ~PAGE_MASK)) {
 465                 EXOFS_ERR("Stripe Unit(0x%llx)"
 466                           " must be Multples of PAGE_SIZE(0x%lx)\n",
 467                           _LLU(sbi->data_map.odm_stripe_unit), PAGE_SIZE);
 468                 return -EINVAL;
 469         }
 470
 471         sbi->layout.stripe_unit = sbi->data_map.odm_stripe_unit;
 472         sbi->layout.mirrors_p1 = sbi->data_map.odm_mirror_cnt + 1;
 473
 474         if (sbi->data_map.odm_group_width) {
 475                 sbi->layout.group_width = sbi->data_map.odm_group_width;
 476                 sbi->layout.group_depth = sbi->data_map.odm_group_depth;
 477                 if (!sbi->layout.group_depth) {
 478                         EXOFS_ERR("group_depth == 0 && group_width != 0\n");
 479                         return -EINVAL;
 480                 }
 481                 sbi->layout.group_count = sbi->data_map.odm_num_comps /
 482                                                 sbi->layout.mirrors_p1 /
 483                                                 sbi->data_map.odm_group_width;
 484         } else {
 485                 if (sbi->data_map.odm_group_depth) {
 486                         printk(KERN_NOTICE "Warning: group_depth ignored "
 487                                 "group_width == 0 && group_depth == %d\n",
 488                                 sbi->data_map.odm_group_depth);
 489                         sbi->data_map.odm_group_depth = 0;
 490                 }
 491                 sbi->layout.group_width = sbi->data_map.odm_num_comps /
 492                                                         sbi->layout.mirrors_p1;
 493                 sbi->layout.group_depth = -1;
 494                 sbi->layout.group_count = 1;
 495         }
 496
 497         stripe_length = (u64)sbi->layout.group_width * sbi->layout.stripe_unit;
 498         if (stripe_length >= (1ULL << 32)) {
 499                 EXOFS_ERR("Total Stripe length(0x%llx)"
 500                           " >= 32bit is not supported\n", _LLU(stripe_length));
 501                 return -EINVAL;
 502         }
 503
 504         return 0;
 505 }
 506
 507 static unsigned __ra_pages(struct exofs_layout *layout)
 508 {
 509         const unsigned _MIN_RA = 32; /* min 128K read-ahead */
 510         unsigned ra_pages = layout->group_width * layout->stripe_unit /
 511                                 PAGE_SIZE;
 512         unsigned max_io_pages = exofs_max_io_pages(layout, ~0);
 513
 514         ra_pages *= 2; /* two stripes */
 515         if (ra_pages < _MIN_RA)
 516                 ra_pages = roundup(_MIN_RA, ra_pages / 2);
 517
 518         if (ra_pages > max_io_pages)
 519                 ra_pages = max_io_pages;
 520
 521         return ra_pages;
 522 }
 523
 524 /* @odi is valid only as long as @fscb_dev is valid */
 525 static int exofs_devs_2_odi(struct exofs_dt_device_info *dt_dev,
 526                              struct osd_dev_info *odi)
 527 {
 528         odi->systemid_len = le32_to_cpu(dt_dev->systemid_len);
 529         memcpy(odi->systemid, dt_dev->systemid, odi->systemid_len);
 530
 531         odi->osdname_len = le32_to_cpu(dt_dev->osdname_len);
 532         odi->osdname = dt_dev->osdname;
 533
 534         /* FIXME support long names. Will need a _put function */
 535         if (dt_dev->long_name_offset)
 536                 return -EINVAL;
 537
 538         /* Make sure osdname is printable!
 539          * mkexofs should give us space for a null-terminator else the
 540          * device-table is invalid.
 541          */
 542         if (unlikely(odi->osdname_len >= sizeof(dt_dev->osdname)))
 543                 odi->osdname_len = sizeof(dt_dev->osdname) - 1;
 544         dt_dev->osdname[odi->osdname_len] = 0;
 545
 546         /* If it's all zeros something is bad we read past end-of-obj */
 547         return !(odi->systemid_len || odi->osdname_len);
 548 }
 549
 550 static int exofs_read_lookup_dev_table(struct exofs_sb_info **psbi,
 551                                        unsigned table_count)
 552 {
 553         struct exofs_sb_info *sbi = *psbi;
 554         struct osd_dev *fscb_od;
 555         struct osd_obj_id obj = {.partition = sbi->layout.s_pid,
 556                                  .id = EXOFS_DEVTABLE_ID};
 557         struct exofs_device_table *dt;
 558         unsigned table_bytes = table_count * sizeof(dt->dt_dev_table[0]) +
 559                                              sizeof(*dt);
 560         unsigned numdevs, i;
 561         int ret;
 562
 563         dt = kmalloc(table_bytes, GFP_KERNEL);
 564         if (unlikely(!dt)) {
 565                 EXOFS_ERR("ERROR: allocating %x bytes for device table\n",
 566                           table_bytes);
 567                 return -ENOMEM;
 568         }
 569
 570         fscb_od = sbi->layout.s_ods[0];
 571         sbi->layout.s_ods[0] = NULL;
 572         sbi->layout.s_numdevs = 0;
 573         ret = exofs_read_kern(fscb_od, sbi->s_cred, &obj, 0, dt, table_bytes);
 574         if (unlikely(ret)) {
 575                 EXOFS_ERR("ERROR: reading device table\n");
 576                 goto out;
 577         }
 578
 579         numdevs = le64_to_cpu(dt->dt_num_devices);
 580         if (unlikely(!numdevs)) {
 581                 ret = -EINVAL;
 582                 goto out;
 583         }
 584         WARN_ON(table_count != numdevs);
 585
 586         ret = _read_and_match_data_map(sbi, numdevs, dt);
 587         if (unlikely(ret))
 588                 goto out;
 589
 590         if (likely(numdevs > 1)) {
 591                 unsigned size = numdevs * sizeof(sbi->layout.s_ods[0]);
 592
 593                 sbi = krealloc(sbi, sizeof(*sbi) + size, GFP_KERNEL);
 594                 if (unlikely(!sbi)) {
 595                         ret = -ENOMEM;
 596                         goto out;
 597                 }
 598                 memset(&sbi->layout.s_ods[1], 0,
 599                        size - sizeof(sbi->layout.s_ods[0]));
 600                 *psbi = sbi;
 601         }
 602
 603         for (i = 0; i < numdevs; i++) {
 604                 struct exofs_fscb fscb;
 605                 struct osd_dev_info odi;
 606                 struct osd_dev *od;
 607
 608                 if (exofs_devs_2_odi(&dt->dt_dev_table[i], &odi)) {
 609                         EXOFS_ERR("ERROR: Read all-zeros device entry\n");
 610                         ret = -EINVAL;
 611                         goto out;
 612                 }
 613
 614                 printk(KERN_NOTICE "Add device[%d]: osd_name-%s\n",
 615                        i, odi.osdname);
 616
 617                 /* On all devices the device table is identical. The user can
 618                  * specify any one of the participating devices on the command
 619                  * line. We always keep them in device-table order.
 620                  */
 621                 if (fscb_od && osduld_device_same(fscb_od, &odi)) {
 622                         sbi->layout.s_ods[i] = fscb_od;
 623                         ++sbi->layout.s_numdevs;
 624                         fscb_od = NULL;
 625                         continue;
 626                 }
 627
 628                 od = osduld_info_lookup(&odi);
 629                 if (IS_ERR(od)) {
 630                         ret = PTR_ERR(od);
 631                         EXOFS_ERR("ERROR: device requested is not found "
 632                                   "osd_name-%s =>%d\n", odi.osdname, ret);
 633                         goto out;
 634                 }
 635
 636                 sbi->layout.s_ods[i] = od;
 637                 ++sbi->layout.s_numdevs;
 638
 639                 /* Read the fscb of the other devices to make sure the FS
 640                  * partition is there.
 641                  */
 642                 ret = exofs_read_kern(od, sbi->s_cred, &obj, 0, &fscb,
 643                                       sizeof(fscb));
 644                 if (unlikely(ret)) {
 645                         EXOFS_ERR("ERROR: Malformed participating device "
 646                                   "error reading fscb osd_name-%s\n",
 647                                   odi.osdname);
 648                         goto out;
 649                 }
 650
 651                 /* TODO: verify other information is correct and FS-uuid
 652                  *       matches. Benny what did you say about device table
 653                  *       generation and old devices?
 654                  */
 655         }
 656
 657 out:
 658         kfree(dt);
 659         if (unlikely(!ret && fscb_od)) {
 660                 EXOFS_ERR(
 661                       "ERROR: Bad device-table container device not present\n");
 662                 osduld_put_device(fscb_od);
 663                 ret = -EINVAL;
 664         }
 665
 666         return ret;
 667 }
 668
 669 /*
 670  * Read the superblock from the OSD and fill in the fields
 671  */
 672 static int exofs_fill_super(struct super_block *sb, void *data, int silent)
 673 {
 674         struct inode *root;
 675         struct exofs_mountopt *opts = data;
 676         struct exofs_sb_info *sbi;      /*extended info                  */
 677         struct osd_dev *od;             /* Master device                 */
 678         struct exofs_fscb fscb;         /*on-disk superblock info        */
 679         struct osd_obj_id obj;
 680         unsigned table_count;
 681         int ret;
 682
 683         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
 684         if (!sbi)
 685                 return -ENOMEM;
 686
 687         ret = bdi_setup_and_register(&sbi->bdi, "exofs", BDI_CAP_MAP_COPY);
 688         if (ret)
 689                 goto free_bdi;
 690
 691         /* use mount options to fill superblock */
 692         if (opts->is_osdname) {
 693                 struct osd_dev_info odi = {.systemid_len = 0};
 694
 695                 odi.osdname_len = strlen(opts->dev_name);
 696                 odi.osdname = (u8 *)opts->dev_name;
 697                 od = osduld_info_lookup(&odi);
 698         } else {
 699                 od = osduld_path_lookup(opts->dev_name);
 700         }
 701         if (IS_ERR(od)) {
 702                 ret = -EINVAL;
 703                 goto free_sbi;
 704         }
 705
 706         /* Default layout in case we do not have a device-table */
 707         sbi->layout.stripe_unit = PAGE_SIZE;
 708         sbi->layout.mirrors_p1 = 1;
 709         sbi->layout.group_width = 1;
 710         sbi->layout.group_depth = -1;
 711         sbi->layout.group_count = 1;
 712         sbi->layout.s_ods[0] = od;
 713         sbi->layout.s_numdevs = 1;
 714         sbi->layout.s_pid = opts->pid;
 715         sbi->s_timeout = opts->timeout;
 716
 717         /* fill in some other data by hand */
 718         memset(sb->s_id, 0, sizeof(sb->s_id));
 719         strcpy(sb->s_id, "exofs");
 720         sb->s_blocksize = EXOFS_BLKSIZE;
 721         sb->s_blocksize_bits = EXOFS_BLKSHIFT;
 722         sb->s_maxbytes = MAX_LFS_FILESIZE;
 723         atomic_set(&sbi->s_curr_pending, 0);
 724         sb->s_bdev = NULL;
 725         sb->s_dev = 0;
 726
 727         obj.partition = sbi->layout.s_pid;
 728         obj.id = EXOFS_SUPER_ID;
 729         exofs_make_credential(sbi->s_cred, &obj);
 730
 731         ret = exofs_read_kern(od, sbi->s_cred, &obj, 0, &fscb, sizeof(fscb));
 732         if (unlikely(ret))
 733                 goto free_sbi;
 734
 735         sb->s_magic = le16_to_cpu(fscb.s_magic);
 736         /* NOTE: we read below to be backward compatible with old versions */
 737         sbi->s_nextid = le64_to_cpu(fscb.s_nextid);
 738         sbi->s_numfiles = le32_to_cpu(fscb.s_numfiles);
 739
 740         /* make sure what we read from the object store is correct */
 741         if (sb->s_magic != EXOFS_SUPER_MAGIC) {
 742                 if (!silent)
 743                         EXOFS_ERR("ERROR: Bad magic value\n");
 744                 ret = -EINVAL;
 745                 goto free_sbi;
 746         }
 747         if (le32_to_cpu(fscb.s_version) > EXOFS_FSCB_VER) {
 748                 EXOFS_ERR("ERROR: Bad FSCB version expected-%d got-%d\n",
 749                           EXOFS_FSCB_VER, le32_to_cpu(fscb.s_version));
 750                 ret = -EINVAL;
 751                 goto free_sbi;
 752         }
 753
 754         /* start generation numbers from a random point */
 755         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
 756         spin_lock_init(&sbi->s_next_gen_lock);
 757
 758         table_count = le64_to_cpu(fscb.s_dev_table_count);
 759         if (table_count) {
 760                 ret = exofs_read_lookup_dev_table(&sbi, table_count);
 761                 if (unlikely(ret))
 762                         goto free_sbi;
 763         }
 764
 765         __sbi_read_stats(sbi);
 766
 767         /* set up operation vectors */
 768         sbi->bdi.ra_pages = __ra_pages(&sbi->layout);
 769         sb->s_bdi = &sbi->bdi;
 770         sb->s_fs_info = sbi;
 771         sb->s_op = &exofs_sops;
 772         sb->s_export_op = &exofs_export_ops;
 773         root = exofs_iget(sb, EXOFS_ROOT_ID - EXOFS_OBJ_OFF);
 774         if (IS_ERR(root)) {
 775                 EXOFS_ERR("ERROR: exofs_iget failed\n");
 776                 ret = PTR_ERR(root);
 777                 goto free_sbi;
 778         }
 779         sb->s_root = d_alloc_root(root);
 780         if (!sb->s_root) {
 781                 iput(root);
 782                 EXOFS_ERR("ERROR: get root inode failed\n");
 783                 ret = -ENOMEM;
 784                 goto free_sbi;
 785         }
 786
 787         if (!S_ISDIR(root->i_mode)) {
 788                 dput(sb->s_root);
 789                 sb->s_root = NULL;
 790                 EXOFS_ERR("ERROR: corrupt root inode (mode = %hd)\n",
 791                        root->i_mode);
 792                 ret = -EINVAL;
 793                 goto free_sbi;
 794         }
 795
 796         _exofs_print_device("Mounting", opts->dev_name, sbi->layout.s_ods[0],
 797                             sbi->layout.s_pid);
 798         if (opts->is_osdname)
 799                 kfree(opts->dev_name);
 800         return 0;
 801
 802 free_sbi:
 803         bdi_destroy(&sbi->bdi);
 804 free_bdi:
 805         EXOFS_ERR("Unable to mount exofs on %s pid=0x%llx err=%d\n",
 806                   opts->dev_name, sbi->layout.s_pid, ret);
 807         exofs_free_sbi(sbi);
 808         if (opts->is_osdname)
 809                 kfree(opts->dev_name);
 810         return ret;
 811 }
 812
 813 /*
 814  * Set up the superblock (calls exofs_fill_super eventually)
 815  */
 816 static struct dentry *exofs_mount(struct file_system_type *type,
 817                           int flags, const char *dev_name,
 818                           void *data)
 819 {
 820         struct exofs_mountopt opts;
 821         int ret;
 822
 823         ret = parse_options(data, &opts);
 824         if (ret)
 825                 return ERR_PTR(ret);
 826
 827         if (!opts.dev_name)
 828                 opts.dev_name = dev_name;
 829         return mount_nodev(type, flags, &opts, exofs_fill_super);
 830 }
 831
 832 /*
 833  * Return information about the file system state in the buffer.  This is used
 834  * by the 'df' command, for example.
 835  */
 836 static int exofs_statfs(struct dentry *dentry, struct kstatfs *buf)
 837 {
 838         struct super_block *sb = dentry->d_sb;
 839         struct exofs_sb_info *sbi = sb->s_fs_info;
 840         struct exofs_io_state *ios;
 841         struct osd_attr attrs[] = {
 842                 ATTR_DEF(OSD_APAGE_PARTITION_QUOTAS,
 843                         OSD_ATTR_PQ_CAPACITY_QUOTA, sizeof(__be64)),
 844                 ATTR_DEF(OSD_APAGE_PARTITION_INFORMATION,
 845                         OSD_ATTR_PI_USED_CAPACITY, sizeof(__be64)),
 846         };
 847         uint64_t capacity = ULLONG_MAX;
 848         uint64_t used = ULLONG_MAX;
 849         uint8_t cred_a[OSD_CAP_LEN];
 850         int ret;
 851
 852         ret = exofs_get_io_state(&sbi->layout, &ios);
 853         if (ret) {
 854                 EXOFS_DBGMSG("exofs_get_io_state failed.\n");
 855                 return ret;
 856         }
 857
 858         exofs_make_credential(cred_a, &ios->obj);
 859         ios->cred = sbi->s_cred;
 860         ios->in_attr = attrs;
 861         ios->in_attr_len = ARRAY_SIZE(attrs);
 862
 863         ret = exofs_sbi_read(ios);
 864         if (unlikely(ret))
 865                 goto out;
 866
 867         ret = extract_attr_from_ios(ios, &attrs[0]);
 868         if (likely(!ret)) {
 869                 capacity = get_unaligned_be64(attrs[0].val_ptr);
 870                 if (unlikely(!capacity))
 871                         capacity = ULLONG_MAX;
 872         } else
 873                 EXOFS_DBGMSG("exofs_statfs: get capacity failed.\n");
 874
 875         ret = extract_attr_from_ios(ios, &attrs[1]);
 876         if (likely(!ret))
 877                 used = get_unaligned_be64(attrs[1].val_ptr);
 878         else
 879                 EXOFS_DBGMSG("exofs_statfs: get used-space failed.\n");
 880
 881         /* fill in the stats buffer */
 882         buf->f_type = EXOFS_SUPER_MAGIC;
 883         buf->f_bsize = EXOFS_BLKSIZE;
 884         buf->f_blocks = capacity >> 9;
 885         buf->f_bfree = (capacity - used) >> 9;
 886         buf->f_bavail = buf->f_bfree;
 887         buf->f_files = sbi->s_numfiles;
 888         buf->f_ffree = EXOFS_MAX_ID - sbi->s_numfiles;
 889         buf->f_namelen = EXOFS_NAME_LEN;
 890
 891 out:
 892         exofs_put_io_state(ios);
 893         return ret;
 894 }
 895
 896 static const struct super_operations exofs_sops = {
 897         .alloc_inode    = exofs_alloc_inode,
 898         .destroy_inode  = exofs_destroy_inode,
 899         .write_inode    = exofs_write_inode,
 900         .evict_inode    = exofs_evict_inode,
 901         .put_super      = exofs_put_super,
 902         .write_super    = exofs_write_super,
 903         .sync_fs        = exofs_sync_fs,
 904         .statfs         = exofs_statfs,
 905 };
 906
 907 /******************************************************************************
 908  * EXPORT OPERATIONS
 909  *****************************************************************************/
 910
 911 struct dentry *exofs_get_parent(struct dentry *child)
 912 {
 913         unsigned long ino = exofs_parent_ino(child);
 914
 915         if (!ino)
 916                 return ERR_PTR(-ESTALE);
 917
 918         return d_obtain_alias(exofs_iget(child->d_inode->i_sb, ino));
 919 }
 920
 921 static struct inode *exofs_nfs_get_inode(struct super_block *sb,
 922                 u64 ino, u32 generation)
 923 {
 924         struct inode *inode;
 925
 926         inode = exofs_iget(sb, ino);
 927         if (IS_ERR(inode))
 928                 return ERR_CAST(inode);
 929         if (generation && inode->i_generation != generation) {
 930                 /* we didn't find the right inode.. */
 931                 iput(inode);
 932                 return ERR_PTR(-ESTALE);
 933         }
 934         return inode;
 935 }
 936
 937 static struct dentry *exofs_fh_to_dentry(struct super_block *sb,
 938                                 struct fid *fid, int fh_len, int fh_type)
 939 {
 940         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 941                                     exofs_nfs_get_inode);
 942 }
 943
 944 static struct dentry *exofs_fh_to_parent(struct super_block *sb,
 945                                 struct fid *fid, int fh_len, int fh_type)
 946 {
 947         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 948                                     exofs_nfs_get_inode);
 949 }
 950
 951 static const struct export_operations exofs_export_ops = {
 952         .fh_to_dentry = exofs_fh_to_dentry,
 953         .fh_to_parent = exofs_fh_to_parent,
 954         .get_parent = exofs_get_parent,
 955 };
 956
 957 /******************************************************************************
 958  * INSMOD/RMMOD
 959  *****************************************************************************/
 960
 961 /*
 962  * struct that describes this file system
 963  */
 964 static struct file_system_type exofs_type = {
 965         .owner          = THIS_MODULE,
 966         .name           = "exofs",
 967         .mount          = exofs_mount,
 968         .kill_sb        = generic_shutdown_super,
 969 };
 970
 971 static int __init init_exofs(void)
 972 {
 973         int err;
 974
 975         err = init_inodecache();
 976         if (err)
 977                 goto out;
 978
 979         err = register_filesystem(&exofs_type);
 980         if (err)
 981                 goto out_d;
 982
 983         return 0;
 984 out_d:
 985         destroy_inodecache();
 986 out:
 987         return err;
 988 }
 989
 990 static void __exit exit_exofs(void)
 991 {
 992         unregister_filesystem(&exofs_type);
 993         destroy_inodecache();
 994 }
 995
 996 MODULE_AUTHOR("Avishay Traeger <avishay@gmail.com>");
 997 MODULE_DESCRIPTION("exofs");
 998 MODULE_LICENSE("GPL");
 999
1000 module_init(init_exofs)
1001 module_exit(exit_exofs)