fs/exofs/super.c

   1 /*
   2  * Copyright (C) 2005, 2006
   3  * Avishay Traeger (avishay@gmail.com)
   4  * Copyright (C) 2008, 2009
   5  * Boaz Harrosh <ooo@electrozaur.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/module.h>
  39 #include <linux/exportfs.h>
  40 #include <linux/slab.h>
  41 #include <linux/iversion.h>
  42
  43 #include "exofs.h"
  44
  45 #define EXOFS_DBGMSG2(M...) do {} while (0)
  46
  47 /******************************************************************************
  48  * MOUNT OPTIONS
  49  *****************************************************************************/
  50
  51 /*
  52  * struct to hold what we get from mount options
  53  */
  54 struct exofs_mountopt {
  55         bool is_osdname;
  56         const char *dev_name;
  57         uint64_t pid;
  58         int timeout;
  59 };
  60
  61 /*
  62  * exofs-specific mount-time options.
  63  */
  64 enum { Opt_name, Opt_pid, Opt_to, Opt_err };
  65
  66 /*
  67  * Our mount-time options.  These should ideally be 64-bit unsigned, but the
  68  * kernel's parsing functions do not currently support that.  32-bit should be
  69  * sufficient for most applications now.
  70  */
  71 static match_table_t tokens = {
  72         {Opt_name, "osdname=%s"},
  73         {Opt_pid, "pid=%u"},
  74         {Opt_to, "to=%u"},
  75         {Opt_err, NULL}
  76 };
  77
  78 /*
  79  * The main option parsing method.  Also makes sure that all of the mandatory
  80  * mount options were set.
  81  */
  82 static int parse_options(char *options, struct exofs_mountopt *opts)
  83 {
  84         char *p;
  85         substring_t args[MAX_OPT_ARGS];
  86         int option;
  87         bool s_pid = false;
  88
  89         EXOFS_DBGMSG("parse_options %s\n", options);
  90         /* defaults */
  91         memset(opts, 0, sizeof(*opts));
  92         opts->timeout = BLK_DEFAULT_SG_TIMEOUT;
  93
  94         while ((p = strsep(&options, ",")) != NULL) {
  95                 int token;
  96                 char str[32];
  97
  98                 if (!*p)
  99                         continue;
 100
 101                 token = match_token(p, tokens, args);
 102                 switch (token) {
 103                 case Opt_name:
 104                         opts->dev_name = match_strdup(&args[0]);
 105                         if (unlikely(!opts->dev_name)) {
 106                                 EXOFS_ERR("Error allocating dev_name");
 107                                 return -ENOMEM;
 108                         }
 109                         opts->is_osdname = true;
 110                         break;
 111                 case Opt_pid:
 112                         if (0 == match_strlcpy(str, &args[0], sizeof(str)))
 113                                 return -EINVAL;
 114                         opts->pid = simple_strtoull(str, NULL, 0);
 115                         if (opts->pid < EXOFS_MIN_PID) {
 116                                 EXOFS_ERR("Partition ID must be >= %u",
 117                                           EXOFS_MIN_PID);
 118                                 return -EINVAL;
 119                         }
 120                         s_pid = 1;
 121                         break;
 122                 case Opt_to:
 123                         if (match_int(&args[0], &option))
 124                                 return -EINVAL;
 125                         if (option <= 0) {
 126                                 EXOFS_ERR("Timeout must be > 0");
 127                                 return -EINVAL;
 128                         }
 129                         opts->timeout = option * HZ;
 130                         break;
 131                 }
 132         }
 133
 134         if (!s_pid) {
 135                 EXOFS_ERR("Need to specify the following options:\n");
 136                 EXOFS_ERR("    -o pid=pid_no_to_use\n");
 137                 return -EINVAL;
 138         }
 139
 140         return 0;
 141 }
 142
 143 /******************************************************************************
 144  * INODE CACHE
 145  *****************************************************************************/
 146
 147 /*
 148  * Our inode cache.  Isn't it pretty?
 149  */
 150 static struct kmem_cache *exofs_inode_cachep;
 151
 152 /*
 153  * Allocate an inode in the cache
 154  */
 155 static struct inode *exofs_alloc_inode(struct super_block *sb)
 156 {
 157         struct exofs_i_info *oi;
 158
 159         oi = kmem_cache_alloc(exofs_inode_cachep, GFP_KERNEL);
 160         if (!oi)
 161                 return NULL;
 162
 163         inode_set_iversion(&oi->vfs_inode, 1);
 164         return &oi->vfs_inode;
 165 }
 166
 167 static void exofs_i_callback(struct rcu_head *head)
 168 {
 169         struct inode *inode = container_of(head, struct inode, i_rcu);
 170         kmem_cache_free(exofs_inode_cachep, exofs_i(inode));
 171 }
 172
 173 /*
 174  * Remove an inode from the cache
 175  */
 176 static void exofs_destroy_inode(struct inode *inode)
 177 {
 178         call_rcu(&inode->i_rcu, exofs_i_callback);
 179 }
 180
 181 /*
 182  * Initialize the inode
 183  */
 184 static void exofs_init_once(void *foo)
 185 {
 186         struct exofs_i_info *oi = foo;
 187
 188         inode_init_once(&oi->vfs_inode);
 189 }
 190
 191 /*
 192  * Create and initialize the inode cache
 193  */
 194 static int init_inodecache(void)
 195 {
 196         exofs_inode_cachep = kmem_cache_create_usercopy("exofs_inode_cache",
 197                                 sizeof(struct exofs_i_info), 0,
 198                                 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD |
 199                                 SLAB_ACCOUNT,
 200                                 offsetof(struct exofs_i_info, i_data),
 201                                 sizeof_field(struct exofs_i_info, i_data),
 202                                 exofs_init_once);
 203         if (exofs_inode_cachep == NULL)
 204                 return -ENOMEM;
 205         return 0;
 206 }
 207
 208 /*
 209  * Destroy the inode cache
 210  */
 211 static void destroy_inodecache(void)
 212 {
 213         /*
 214          * Make sure all delayed rcu free inodes are flushed before we
 215          * destroy cache.
 216          */
 217         rcu_barrier();
 218         kmem_cache_destroy(exofs_inode_cachep);
 219 }
 220
 221 /******************************************************************************
 222  * Some osd helpers
 223  *****************************************************************************/
 224 void exofs_make_credential(u8 cred_a[OSD_CAP_LEN], const struct osd_obj_id *obj)
 225 {
 226         osd_sec_init_nosec_doall_caps(cred_a, obj, false, true);
 227 }
 228
 229 static int exofs_read_kern(struct osd_dev *od, u8 *cred, struct osd_obj_id *obj,
 230                     u64 offset, void *p, unsigned length)
 231 {
 232         struct osd_request *or = osd_start_request(od);
 233 /*      struct osd_sense_info osi = {.key = 0};*/
 234         int ret;
 235
 236         if (unlikely(!or)) {
 237                 EXOFS_DBGMSG("%s: osd_start_request failed.\n", __func__);
 238                 return -ENOMEM;
 239         }
 240         ret = osd_req_read_kern(or, obj, offset, p, length);
 241         if (unlikely(ret)) {
 242                 EXOFS_DBGMSG("%s: osd_req_read_kern failed.\n", __func__);
 243                 goto out;
 244         }
 245
 246         ret = osd_finalize_request(or, 0, cred, NULL);
 247         if (unlikely(ret)) {
 248                 EXOFS_DBGMSG("Failed to osd_finalize_request() => %d\n", ret);
 249                 goto out;
 250         }
 251
 252         ret = osd_execute_request(or);
 253         if (unlikely(ret))
 254                 EXOFS_DBGMSG("osd_execute_request() => %d\n", ret);
 255         /* osd_req_decode_sense(or, ret); */
 256
 257 out:
 258         osd_end_request(or);
 259         EXOFS_DBGMSG2("read_kern(0x%llx) offset=0x%llx "
 260                       "length=0x%llx dev=%p ret=>%d\n",
 261                       _LLU(obj->id), _LLU(offset), _LLU(length), od, ret);
 262         return ret;
 263 }
 264
 265 static const struct osd_attr g_attr_sb_stats = ATTR_DEF(
 266         EXOFS_APAGE_SB_DATA,
 267         EXOFS_ATTR_SB_STATS,
 268         sizeof(struct exofs_sb_stats));
 269
 270 static int __sbi_read_stats(struct exofs_sb_info *sbi)
 271 {
 272         struct osd_attr attrs[] = {
 273                 [0] = g_attr_sb_stats,
 274         };
 275         struct ore_io_state *ios;
 276         int ret;
 277
 278         ret = ore_get_io_state(&sbi->layout, &sbi->oc, &ios);
 279         if (unlikely(ret)) {
 280                 EXOFS_ERR("%s: ore_get_io_state failed.\n", __func__);
 281                 return ret;
 282         }
 283
 284         ios->in_attr = attrs;
 285         ios->in_attr_len = ARRAY_SIZE(attrs);
 286
 287         ret = ore_read(ios);
 288         if (unlikely(ret)) {
 289                 EXOFS_ERR("Error reading super_block stats => %d\n", ret);
 290                 goto out;
 291         }
 292
 293         ret = extract_attr_from_ios(ios, &attrs[0]);
 294         if (ret) {
 295                 EXOFS_ERR("%s: extract_attr of sb_stats failed\n", __func__);
 296                 goto out;
 297         }
 298         if (attrs[0].len) {
 299                 struct exofs_sb_stats *ess;
 300
 301                 if (unlikely(attrs[0].len != sizeof(*ess))) {
 302                         EXOFS_ERR("%s: Wrong version of exofs_sb_stats "
 303                                   "size(%d) != expected(%zd)\n",
 304                                   __func__, attrs[0].len, sizeof(*ess));
 305                         goto out;
 306                 }
 307
 308                 ess = attrs[0].val_ptr;
 309                 sbi->s_nextid = le64_to_cpu(ess->s_nextid);
 310                 sbi->s_numfiles = le32_to_cpu(ess->s_numfiles);
 311         }
 312
 313 out:
 314         ore_put_io_state(ios);
 315         return ret;
 316 }
 317
 318 static void stats_done(struct ore_io_state *ios, void *p)
 319 {
 320         ore_put_io_state(ios);
 321         /* Good thanks nothing to do anymore */
 322 }
 323
 324 /* Asynchronously write the stats attribute */
 325 int exofs_sbi_write_stats(struct exofs_sb_info *sbi)
 326 {
 327         struct osd_attr attrs[] = {
 328                 [0] = g_attr_sb_stats,
 329         };
 330         struct ore_io_state *ios;
 331         int ret;
 332
 333         ret = ore_get_io_state(&sbi->layout, &sbi->oc, &ios);
 334         if (unlikely(ret)) {
 335                 EXOFS_ERR("%s: ore_get_io_state failed.\n", __func__);
 336                 return ret;
 337         }
 338
 339         sbi->s_ess.s_nextid   = cpu_to_le64(sbi->s_nextid);
 340         sbi->s_ess.s_numfiles = cpu_to_le64(sbi->s_numfiles);
 341         attrs[0].val_ptr = &sbi->s_ess;
 342
 343
 344         ios->done = stats_done;
 345         ios->private = sbi;
 346         ios->out_attr = attrs;
 347         ios->out_attr_len = ARRAY_SIZE(attrs);
 348
 349         ret = ore_write(ios);
 350         if (unlikely(ret)) {
 351                 EXOFS_ERR("%s: ore_write failed.\n", __func__);
 352                 ore_put_io_state(ios);
 353         }
 354
 355         return ret;
 356 }
 357
 358 /******************************************************************************
 359  * SUPERBLOCK FUNCTIONS
 360  *****************************************************************************/
 361 static const struct super_operations exofs_sops;
 362 static const struct export_operations exofs_export_ops;
 363
 364 /*
 365  * Write the superblock to the OSD
 366  */
 367 static int exofs_sync_fs(struct super_block *sb, int wait)
 368 {
 369         struct exofs_sb_info *sbi;
 370         struct exofs_fscb *fscb;
 371         struct ore_comp one_comp;
 372         struct ore_components oc;
 373         struct ore_io_state *ios;
 374         int ret = -ENOMEM;
 375
 376         fscb = kmalloc(sizeof(*fscb), GFP_KERNEL);
 377         if (unlikely(!fscb))
 378                 return -ENOMEM;
 379
 380         sbi = sb->s_fs_info;
 381
 382         /* NOTE: We no longer dirty the super_block anywhere in exofs. The
 383          * reason we write the fscb here on unmount is so we can stay backwards
 384          * compatible with fscb->s_version == 1. (What we are not compatible
 385          * with is if a new version FS crashed and then we try to mount an old
 386          * version). Otherwise the exofs_fscb is read-only from mkfs time. All
 387          * the writeable info is set in exofs_sbi_write_stats() above.
 388          */
 389
 390         exofs_init_comps(&oc, &one_comp, sbi, EXOFS_SUPER_ID);
 391
 392         ret = ore_get_io_state(&sbi->layout, &oc, &ios);
 393         if (unlikely(ret))
 394                 goto out;
 395
 396         ios->length = offsetof(struct exofs_fscb, s_dev_table_oid);
 397         memset(fscb, 0, ios->length);
 398         fscb->s_nextid = cpu_to_le64(sbi->s_nextid);
 399         fscb->s_numfiles = cpu_to_le64(sbi->s_numfiles);
 400         fscb->s_magic = cpu_to_le16(sb->s_magic);
 401         fscb->s_newfs = 0;
 402         fscb->s_version = EXOFS_FSCB_VER;
 403
 404         ios->offset = 0;
 405         ios->kern_buff = fscb;
 406
 407         ret = ore_write(ios);
 408         if (unlikely(ret))
 409                 EXOFS_ERR("%s: ore_write failed.\n", __func__);
 410
 411 out:
 412         EXOFS_DBGMSG("s_nextid=0x%llx ret=%d\n", _LLU(sbi->s_nextid), ret);
 413         ore_put_io_state(ios);
 414         kfree(fscb);
 415         return ret;
 416 }
 417
 418 static void _exofs_print_device(const char *msg, const char *dev_path,
 419                                 struct osd_dev *od, u64 pid)
 420 {
 421         const struct osd_dev_info *odi = osduld_device_info(od);
 422
 423         printk(KERN_NOTICE "exofs: %s %s osd_name-%s pid-0x%llx\n",
 424                 msg, dev_path ?: "", odi->osdname, _LLU(pid));
 425 }
 426
 427 static void exofs_free_sbi(struct exofs_sb_info *sbi)
 428 {
 429         unsigned numdevs = sbi->oc.numdevs;
 430
 431         while (numdevs) {
 432                 unsigned i = --numdevs;
 433                 struct osd_dev *od = ore_comp_dev(&sbi->oc, i);
 434
 435                 if (od) {
 436                         ore_comp_set_dev(&sbi->oc, i, NULL);
 437                         osduld_put_device(od);
 438                 }
 439         }
 440         kfree(sbi->oc.ods);
 441         kfree(sbi);
 442 }
 443
 444 /*
 445  * This function is called when the vfs is freeing the superblock.  We just
 446  * need to free our own part.
 447  */
 448 static void exofs_put_super(struct super_block *sb)
 449 {
 450         int num_pend;
 451         struct exofs_sb_info *sbi = sb->s_fs_info;
 452
 453         /* make sure there are no pending commands */
 454         for (num_pend = atomic_read(&sbi->s_curr_pending); num_pend > 0;
 455              num_pend = atomic_read(&sbi->s_curr_pending)) {
 456                 wait_queue_head_t wq;
 457
 458                 printk(KERN_NOTICE "%s: !!Pending operations in flight. "
 459                        "This is a BUG. please report to osd-dev@open-osd.org\n",
 460                        __func__);
 461                 init_waitqueue_head(&wq);
 462                 wait_event_timeout(wq,
 463                                   (atomic_read(&sbi->s_curr_pending) == 0),
 464                                   msecs_to_jiffies(100));
 465         }
 466
 467         _exofs_print_device("Unmounting", NULL, ore_comp_dev(&sbi->oc, 0),
 468                             sbi->one_comp.obj.partition);
 469
 470         exofs_sysfs_sb_del(sbi);
 471         exofs_free_sbi(sbi);
 472         sb->s_fs_info = NULL;
 473 }
 474
 475 static int _read_and_match_data_map(struct exofs_sb_info *sbi, unsigned numdevs,
 476                                     struct exofs_device_table *dt)
 477 {
 478         int ret;
 479
 480         sbi->layout.stripe_unit =
 481                                 le64_to_cpu(dt->dt_data_map.cb_stripe_unit);
 482         sbi->layout.group_width =
 483                                 le32_to_cpu(dt->dt_data_map.cb_group_width);
 484         sbi->layout.group_depth =
 485                                 le32_to_cpu(dt->dt_data_map.cb_group_depth);
 486         sbi->layout.mirrors_p1  =
 487                                 le32_to_cpu(dt->dt_data_map.cb_mirror_cnt) + 1;
 488         sbi->layout.raid_algorithm  =
 489                                 le32_to_cpu(dt->dt_data_map.cb_raid_algorithm);
 490
 491         ret = ore_verify_layout(numdevs, &sbi->layout);
 492
 493         EXOFS_DBGMSG("exofs: layout: "
 494                 "num_comps=%u stripe_unit=0x%x group_width=%u "
 495                 "group_depth=0x%llx mirrors_p1=%u raid_algorithm=%u\n",
 496                 numdevs,
 497                 sbi->layout.stripe_unit,
 498                 sbi->layout.group_width,
 499                 _LLU(sbi->layout.group_depth),
 500                 sbi->layout.mirrors_p1,
 501                 sbi->layout.raid_algorithm);
 502         return ret;
 503 }
 504
 505 static unsigned __ra_pages(struct ore_layout *layout)
 506 {
 507         const unsigned _MIN_RA = 32; /* min 128K read-ahead */
 508         unsigned ra_pages = layout->group_width * layout->stripe_unit /
 509                                 PAGE_SIZE;
 510         unsigned max_io_pages = exofs_max_io_pages(layout, ~0);
 511
 512         ra_pages *= 2; /* two stripes */
 513         if (ra_pages < _MIN_RA)
 514                 ra_pages = roundup(_MIN_RA, ra_pages / 2);
 515
 516         if (ra_pages > max_io_pages)
 517                 ra_pages = max_io_pages;
 518
 519         return ra_pages;
 520 }
 521
 522 /* @odi is valid only as long as @fscb_dev is valid */
 523 static int exofs_devs_2_odi(struct exofs_dt_device_info *dt_dev,
 524                              struct osd_dev_info *odi)
 525 {
 526         odi->systemid_len = le32_to_cpu(dt_dev->systemid_len);
 527         if (likely(odi->systemid_len))
 528                 memcpy(odi->systemid, dt_dev->systemid, OSD_SYSTEMID_LEN);
 529
 530         odi->osdname_len = le32_to_cpu(dt_dev->osdname_len);
 531         odi->osdname = dt_dev->osdname;
 532
 533         /* FIXME support long names. Will need a _put function */
 534         if (dt_dev->long_name_offset)
 535                 return -EINVAL;
 536
 537         /* Make sure osdname is printable!
 538          * mkexofs should give us space for a null-terminator else the
 539          * device-table is invalid.
 540          */
 541         if (unlikely(odi->osdname_len >= sizeof(dt_dev->osdname)))
 542                 odi->osdname_len = sizeof(dt_dev->osdname) - 1;
 543         dt_dev->osdname[odi->osdname_len] = 0;
 544
 545         /* If it's all zeros something is bad we read past end-of-obj */
 546         return !(odi->systemid_len || odi->osdname_len);
 547 }
 548
 549 static int __alloc_dev_table(struct exofs_sb_info *sbi, unsigned numdevs,
 550                       struct exofs_dev **peds)
 551 {
 552         /* Twice bigger table: See exofs_init_comps() and comment at
 553          * exofs_read_lookup_dev_table()
 554          */
 555         const size_t numores = numdevs * 2 - 1;
 556         struct exofs_dev *eds;
 557         unsigned i;
 558
 559         sbi->oc.ods = kzalloc(numores * sizeof(struct ore_dev *) +
 560                               numdevs * sizeof(struct exofs_dev), GFP_KERNEL);
 561         if (unlikely(!sbi->oc.ods)) {
 562                 EXOFS_ERR("ERROR: failed allocating Device array[%d]\n",
 563                           numdevs);
 564                 return -ENOMEM;
 565         }
 566
 567         /* Start of allocated struct exofs_dev entries */
 568         *peds = eds = (void *)sbi->oc.ods[numores];
 569         /* Initialize pointers into struct exofs_dev */
 570         for (i = 0; i < numdevs; ++i)
 571                 sbi->oc.ods[i] = &eds[i].ored;
 572         return 0;
 573 }
 574
 575 static int exofs_read_lookup_dev_table(struct exofs_sb_info *sbi,
 576                                        struct osd_dev *fscb_od,
 577                                        unsigned table_count)
 578 {
 579         struct ore_comp comp;
 580         struct exofs_device_table *dt;
 581         struct exofs_dev *eds;
 582         unsigned table_bytes = table_count * sizeof(dt->dt_dev_table[0]) +
 583                                              sizeof(*dt);
 584         unsigned numdevs, i;
 585         int ret;
 586
 587         dt = kmalloc(table_bytes, GFP_KERNEL);
 588         if (unlikely(!dt)) {
 589                 EXOFS_ERR("ERROR: allocating %x bytes for device table\n",
 590                           table_bytes);
 591                 return -ENOMEM;
 592         }
 593
 594         sbi->oc.numdevs = 0;
 595
 596         comp.obj.partition = sbi->one_comp.obj.partition;
 597         comp.obj.id = EXOFS_DEVTABLE_ID;
 598         exofs_make_credential(comp.cred, &comp.obj);
 599
 600         ret = exofs_read_kern(fscb_od, comp.cred, &comp.obj, 0, dt,
 601                               table_bytes);
 602         if (unlikely(ret)) {
 603                 EXOFS_ERR("ERROR: reading device table\n");
 604                 goto out;
 605         }
 606
 607         numdevs = le64_to_cpu(dt->dt_num_devices);
 608         if (unlikely(!numdevs)) {
 609                 ret = -EINVAL;
 610                 goto out;
 611         }
 612         WARN_ON(table_count != numdevs);
 613
 614         ret = _read_and_match_data_map(sbi, numdevs, dt);
 615         if (unlikely(ret))
 616                 goto out;
 617
 618         ret = __alloc_dev_table(sbi, numdevs, &eds);
 619         if (unlikely(ret))
 620                 goto out;
 621         /* exofs round-robins the device table view according to inode
 622          * number. We hold a: twice bigger table hence inodes can point
 623          * to any device and have a sequential view of the table
 624          * starting at this device. See exofs_init_comps()
 625          */
 626         memcpy(&sbi->oc.ods[numdevs], &sbi->oc.ods[0],
 627                 (numdevs - 1) * sizeof(sbi->oc.ods[0]));
 628
 629         /* create sysfs subdir under which we put the device table
 630          * And cluster layout. A Superblock is identified by the string:
 631          *      "dev[0].osdname"_"pid"
 632          */
 633         exofs_sysfs_sb_add(sbi, &dt->dt_dev_table[0]);
 634
 635         for (i = 0; i < numdevs; i++) {
 636                 struct exofs_fscb fscb;
 637                 struct osd_dev_info odi;
 638                 struct osd_dev *od;
 639
 640                 if (exofs_devs_2_odi(&dt->dt_dev_table[i], &odi)) {
 641                         EXOFS_ERR("ERROR: Read all-zeros device entry\n");
 642                         ret = -EINVAL;
 643                         goto out;
 644                 }
 645
 646                 printk(KERN_NOTICE "Add device[%d]: osd_name-%s\n",
 647                        i, odi.osdname);
 648
 649                 /* the exofs id is currently the table index */
 650                 eds[i].did = i;
 651
 652                 /* On all devices the device table is identical. The user can
 653                  * specify any one of the participating devices on the command
 654                  * line. We always keep them in device-table order.
 655                  */
 656                 if (fscb_od && osduld_device_same(fscb_od, &odi)) {
 657                         eds[i].ored.od = fscb_od;
 658                         ++sbi->oc.numdevs;
 659                         fscb_od = NULL;
 660                         exofs_sysfs_odev_add(&eds[i], sbi);
 661                         continue;
 662                 }
 663
 664                 od = osduld_info_lookup(&odi);
 665                 if (IS_ERR(od)) {
 666                         ret = PTR_ERR(od);
 667                         EXOFS_ERR("ERROR: device requested is not found "
 668                                   "osd_name-%s =>%d\n", odi.osdname, ret);
 669                         goto out;
 670                 }
 671
 672                 eds[i].ored.od = od;
 673                 ++sbi->oc.numdevs;
 674
 675                 /* Read the fscb of the other devices to make sure the FS
 676                  * partition is there.
 677                  */
 678                 ret = exofs_read_kern(od, comp.cred, &comp.obj, 0, &fscb,
 679                                       sizeof(fscb));
 680                 if (unlikely(ret)) {
 681                         EXOFS_ERR("ERROR: Malformed participating device "
 682                                   "error reading fscb osd_name-%s\n",
 683                                   odi.osdname);
 684                         goto out;
 685                 }
 686                 exofs_sysfs_odev_add(&eds[i], sbi);
 687
 688                 /* TODO: verify other information is correct and FS-uuid
 689                  *       matches. Benny what did you say about device table
 690                  *       generation and old devices?
 691                  */
 692         }
 693
 694 out:
 695         kfree(dt);
 696         if (unlikely(fscb_od && !ret)) {
 697                         EXOFS_ERR("ERROR: Bad device-table container device not present\n");
 698                         osduld_put_device(fscb_od);
 699                         return -EINVAL;
 700         }
 701         return ret;
 702 }
 703
 704 /*
 705  * Read the superblock from the OSD and fill in the fields
 706  */
 707 static int exofs_fill_super(struct super_block *sb, void *data, int silent)
 708 {
 709         struct inode *root;
 710         struct exofs_mountopt *opts = data;
 711         struct exofs_sb_info *sbi;      /*extended info                  */
 712         struct osd_dev *od;             /* Master device                 */
 713         struct exofs_fscb fscb;         /*on-disk superblock info        */
 714         struct ore_comp comp;
 715         unsigned table_count;
 716         int ret;
 717
 718         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
 719         if (!sbi)
 720                 return -ENOMEM;
 721
 722         /* use mount options to fill superblock */
 723         if (opts->is_osdname) {
 724                 struct osd_dev_info odi = {.systemid_len = 0};
 725
 726                 odi.osdname_len = strlen(opts->dev_name);
 727                 odi.osdname = (u8 *)opts->dev_name;
 728                 od = osduld_info_lookup(&odi);
 729                 kfree(opts->dev_name);
 730                 opts->dev_name = NULL;
 731         } else {
 732                 od = osduld_path_lookup(opts->dev_name);
 733         }
 734         if (IS_ERR(od)) {
 735                 ret = -EINVAL;
 736                 goto free_sbi;
 737         }
 738
 739         /* Default layout in case we do not have a device-table */
 740         sbi->layout.stripe_unit = PAGE_SIZE;
 741         sbi->layout.mirrors_p1 = 1;
 742         sbi->layout.group_width = 1;
 743         sbi->layout.group_depth = -1;
 744         sbi->layout.group_count = 1;
 745         sbi->s_timeout = opts->timeout;
 746
 747         sbi->one_comp.obj.partition = opts->pid;
 748         sbi->one_comp.obj.id = 0;
 749         exofs_make_credential(sbi->one_comp.cred, &sbi->one_comp.obj);
 750         sbi->oc.single_comp = EC_SINGLE_COMP;
 751         sbi->oc.comps = &sbi->one_comp;
 752
 753         /* fill in some other data by hand */
 754         memset(sb->s_id, 0, sizeof(sb->s_id));
 755         strcpy(sb->s_id, "exofs");
 756         sb->s_blocksize = EXOFS_BLKSIZE;
 757         sb->s_blocksize_bits = EXOFS_BLKSHIFT;
 758         sb->s_maxbytes = MAX_LFS_FILESIZE;
 759         sb->s_max_links = EXOFS_LINK_MAX;
 760         atomic_set(&sbi->s_curr_pending, 0);
 761         sb->s_bdev = NULL;
 762         sb->s_dev = 0;
 763
 764         comp.obj.partition = sbi->one_comp.obj.partition;
 765         comp.obj.id = EXOFS_SUPER_ID;
 766         exofs_make_credential(comp.cred, &comp.obj);
 767
 768         ret = exofs_read_kern(od, comp.cred, &comp.obj, 0, &fscb, sizeof(fscb));
 769         if (unlikely(ret))
 770                 goto free_sbi;
 771
 772         sb->s_magic = le16_to_cpu(fscb.s_magic);
 773         /* NOTE: we read below to be backward compatible with old versions */
 774         sbi->s_nextid = le64_to_cpu(fscb.s_nextid);
 775         sbi->s_numfiles = le32_to_cpu(fscb.s_numfiles);
 776
 777         /* make sure what we read from the object store is correct */
 778         if (sb->s_magic != EXOFS_SUPER_MAGIC) {
 779                 if (!silent)
 780                         EXOFS_ERR("ERROR: Bad magic value\n");
 781                 ret = -EINVAL;
 782                 goto free_sbi;
 783         }
 784         if (le32_to_cpu(fscb.s_version) > EXOFS_FSCB_VER) {
 785                 EXOFS_ERR("ERROR: Bad FSCB version expected-%d got-%d\n",
 786                           EXOFS_FSCB_VER, le32_to_cpu(fscb.s_version));
 787                 ret = -EINVAL;
 788                 goto free_sbi;
 789         }
 790
 791         /* start generation numbers from a random point */
 792         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
 793         spin_lock_init(&sbi->s_next_gen_lock);
 794
 795         table_count = le64_to_cpu(fscb.s_dev_table_count);
 796         if (table_count) {
 797                 ret = exofs_read_lookup_dev_table(sbi, od, table_count);
 798                 if (unlikely(ret))
 799                         goto free_sbi;
 800         } else {
 801                 struct exofs_dev *eds;
 802
 803                 ret = __alloc_dev_table(sbi, 1, &eds);
 804                 if (unlikely(ret))
 805                         goto free_sbi;
 806
 807                 ore_comp_set_dev(&sbi->oc, 0, od);
 808                 sbi->oc.numdevs = 1;
 809         }
 810
 811         __sbi_read_stats(sbi);
 812
 813         /* set up operation vectors */
 814         ret = super_setup_bdi(sb);
 815         if (ret) {
 816                 EXOFS_DBGMSG("Failed to super_setup_bdi\n");
 817                 goto free_sbi;
 818         }
 819         sb->s_bdi->ra_pages = __ra_pages(&sbi->layout);
 820         sb->s_fs_info = sbi;
 821         sb->s_op = &exofs_sops;
 822         sb->s_export_op = &exofs_export_ops;
 823         root = exofs_iget(sb, EXOFS_ROOT_ID - EXOFS_OBJ_OFF);
 824         if (IS_ERR(root)) {
 825                 EXOFS_ERR("ERROR: exofs_iget failed\n");
 826                 ret = PTR_ERR(root);
 827                 goto free_sbi;
 828         }
 829         sb->s_root = d_make_root(root);
 830         if (!sb->s_root) {
 831                 EXOFS_ERR("ERROR: get root inode failed\n");
 832                 ret = -ENOMEM;
 833                 goto free_sbi;
 834         }
 835
 836         if (!S_ISDIR(root->i_mode)) {
 837                 dput(sb->s_root);
 838                 sb->s_root = NULL;
 839                 EXOFS_ERR("ERROR: corrupt root inode (mode = %hd)\n",
 840                        root->i_mode);
 841                 ret = -EINVAL;
 842                 goto free_sbi;
 843         }
 844
 845         exofs_sysfs_dbg_print();
 846         _exofs_print_device("Mounting", opts->dev_name,
 847                             ore_comp_dev(&sbi->oc, 0),
 848                             sbi->one_comp.obj.partition);
 849         return 0;
 850
 851 free_sbi:
 852         EXOFS_ERR("Unable to mount exofs on %s pid=0x%llx err=%d\n",
 853                   opts->dev_name, sbi->one_comp.obj.partition, ret);
 854         exofs_free_sbi(sbi);
 855         return ret;
 856 }
 857
 858 /*
 859  * Set up the superblock (calls exofs_fill_super eventually)
 860  */
 861 static struct dentry *exofs_mount(struct file_system_type *type,
 862                           int flags, const char *dev_name,
 863                           void *data)
 864 {
 865         struct exofs_mountopt opts;
 866         int ret;
 867
 868         ret = parse_options(data, &opts);
 869         if (ret)
 870                 return ERR_PTR(ret);
 871
 872         if (!opts.dev_name)
 873                 opts.dev_name = dev_name;
 874         return mount_nodev(type, flags, &opts, exofs_fill_super);
 875 }
 876
 877 /*
 878  * Return information about the file system state in the buffer.  This is used
 879  * by the 'df' command, for example.
 880  */
 881 static int exofs_statfs(struct dentry *dentry, struct kstatfs *buf)
 882 {
 883         struct super_block *sb = dentry->d_sb;
 884         struct exofs_sb_info *sbi = sb->s_fs_info;
 885         struct ore_io_state *ios;
 886         struct osd_attr attrs[] = {
 887                 ATTR_DEF(OSD_APAGE_PARTITION_QUOTAS,
 888                         OSD_ATTR_PQ_CAPACITY_QUOTA, sizeof(__be64)),
 889                 ATTR_DEF(OSD_APAGE_PARTITION_INFORMATION,
 890                         OSD_ATTR_PI_USED_CAPACITY, sizeof(__be64)),
 891         };
 892         uint64_t capacity = ULLONG_MAX;
 893         uint64_t used = ULLONG_MAX;
 894         int ret;
 895
 896         ret = ore_get_io_state(&sbi->layout, &sbi->oc, &ios);
 897         if (ret) {
 898                 EXOFS_DBGMSG("ore_get_io_state failed.\n");
 899                 return ret;
 900         }
 901
 902         ios->in_attr = attrs;
 903         ios->in_attr_len = ARRAY_SIZE(attrs);
 904
 905         ret = ore_read(ios);
 906         if (unlikely(ret))
 907                 goto out;
 908
 909         ret = extract_attr_from_ios(ios, &attrs[0]);
 910         if (likely(!ret)) {
 911                 capacity = get_unaligned_be64(attrs[0].val_ptr);
 912                 if (unlikely(!capacity))
 913                         capacity = ULLONG_MAX;
 914         } else
 915                 EXOFS_DBGMSG("exofs_statfs: get capacity failed.\n");
 916
 917         ret = extract_attr_from_ios(ios, &attrs[1]);
 918         if (likely(!ret))
 919                 used = get_unaligned_be64(attrs[1].val_ptr);
 920         else
 921                 EXOFS_DBGMSG("exofs_statfs: get used-space failed.\n");
 922
 923         /* fill in the stats buffer */
 924         buf->f_type = EXOFS_SUPER_MAGIC;
 925         buf->f_bsize = EXOFS_BLKSIZE;
 926         buf->f_blocks = capacity >> 9;
 927         buf->f_bfree = (capacity - used) >> 9;
 928         buf->f_bavail = buf->f_bfree;
 929         buf->f_files = sbi->s_numfiles;
 930         buf->f_ffree = EXOFS_MAX_ID - sbi->s_numfiles;
 931         buf->f_namelen = EXOFS_NAME_LEN;
 932
 933 out:
 934         ore_put_io_state(ios);
 935         return ret;
 936 }
 937
 938 static const struct super_operations exofs_sops = {
 939         .alloc_inode    = exofs_alloc_inode,
 940         .destroy_inode  = exofs_destroy_inode,
 941         .write_inode    = exofs_write_inode,
 942         .evict_inode    = exofs_evict_inode,
 943         .put_super      = exofs_put_super,
 944         .sync_fs        = exofs_sync_fs,
 945         .statfs         = exofs_statfs,
 946 };
 947
 948 /******************************************************************************
 949  * EXPORT OPERATIONS
 950  *****************************************************************************/
 951
 952 static struct dentry *exofs_get_parent(struct dentry *child)
 953 {
 954         unsigned long ino = exofs_parent_ino(child);
 955
 956         if (!ino)
 957                 return ERR_PTR(-ESTALE);
 958
 959         return d_obtain_alias(exofs_iget(child->d_sb, ino));
 960 }
 961
 962 static struct inode *exofs_nfs_get_inode(struct super_block *sb,
 963                 u64 ino, u32 generation)
 964 {
 965         struct inode *inode;
 966
 967         inode = exofs_iget(sb, ino);
 968         if (IS_ERR(inode))
 969                 return ERR_CAST(inode);
 970         if (generation && inode->i_generation != generation) {
 971                 /* we didn't find the right inode.. */
 972                 iput(inode);
 973                 return ERR_PTR(-ESTALE);
 974         }
 975         return inode;
 976 }
 977
 978 static struct dentry *exofs_fh_to_dentry(struct super_block *sb,
 979                                 struct fid *fid, int fh_len, int fh_type)
 980 {
 981         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 982                                     exofs_nfs_get_inode);
 983 }
 984
 985 static struct dentry *exofs_fh_to_parent(struct super_block *sb,
 986                                 struct fid *fid, int fh_len, int fh_type)
 987 {
 988         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 989                                     exofs_nfs_get_inode);
 990 }
 991
 992 static const struct export_operations exofs_export_ops = {
 993         .fh_to_dentry = exofs_fh_to_dentry,
 994         .fh_to_parent = exofs_fh_to_parent,
 995         .get_parent = exofs_get_parent,
 996 };
 997
 998 /******************************************************************************
 999  * INSMOD/RMMOD
1000  *****************************************************************************/
1001
1002 /*
1003  * struct that describes this file system
1004  */
1005 static struct file_system_type exofs_type = {
1006         .owner          = THIS_MODULE,
1007         .name           = "exofs",
1008         .mount          = exofs_mount,
1009         .kill_sb        = generic_shutdown_super,
1010 };
1011 MODULE_ALIAS_FS("exofs");
1012
1013 static int __init init_exofs(void)
1014 {
1015         int err;
1016
1017         err = init_inodecache();
1018         if (err)
1019                 goto out;
1020
1021         err = register_filesystem(&exofs_type);
1022         if (err)
1023                 goto out_d;
1024
1025         /* We don't fail if sysfs creation failed */
1026         exofs_sysfs_init();
1027
1028         return 0;
1029 out_d:
1030         destroy_inodecache();
1031 out:
1032         return err;
1033 }
1034
1035 static void __exit exit_exofs(void)
1036 {
1037         exofs_sysfs_uninit();
1038         unregister_filesystem(&exofs_type);
1039         destroy_inodecache();
1040 }
1041
1042 MODULE_AUTHOR("Avishay Traeger <avishay@gmail.com>");
1043 MODULE_DESCRIPTION("exofs");
1044 MODULE_LICENSE("GPL");
1045
1046 module_init(init_exofs)
1047 module_exit(exit_exofs)