fs/exofs/super.c

   1 /*
   2  * Copyright (C) 2005, 2006
   3  * Avishay Traeger ([email protected])
   4  * Copyright (C) 2008, 2009
   5  * Boaz Harrosh <[email protected]>
   6  *
   7  * Copyrights for code taken from ext2:
   8  *     Copyright (C) 1992, 1993, 1994, 1995
   9  *     Remy Card ([email protected])
  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("exofs_inode_cache",
 197                                 sizeof(struct exofs_i_info), 0,
 198                                 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD |
 199                                 SLAB_ACCOUNT, exofs_init_once);
 200         if (exofs_inode_cachep == NULL)
 201                 return -ENOMEM;
 202         return 0;
 203 }
 204
 205 /*
 206  * Destroy the inode cache
 207  */
 208 static void destroy_inodecache(void)
 209 {
 210         /*
 211          * Make sure all delayed rcu free inodes are flushed before we
 212          * destroy cache.
 213          */
 214         rcu_barrier();
 215         kmem_cache_destroy(exofs_inode_cachep);
 216 }
 217
 218 /******************************************************************************
 219  * Some osd helpers
 220  *****************************************************************************/
 221 void exofs_make_credential(u8 cred_a[OSD_CAP_LEN], const struct osd_obj_id *obj)
 222 {
 223         osd_sec_init_nosec_doall_caps(cred_a, obj, false, true);
 224 }
 225
 226 static int exofs_read_kern(struct osd_dev *od, u8 *cred, struct osd_obj_id *obj,
 227                     u64 offset, void *p, unsigned length)
 228 {
 229         struct osd_request *or = osd_start_request(od, GFP_KERNEL);
 230 /*      struct osd_sense_info osi = {.key = 0};*/
 231         int ret;
 232
 233         if (unlikely(!or)) {
 234                 EXOFS_DBGMSG("%s: osd_start_request failed.\n", __func__);
 235                 return -ENOMEM;
 236         }
 237         ret = osd_req_read_kern(or, obj, offset, p, length);
 238         if (unlikely(ret)) {
 239                 EXOFS_DBGMSG("%s: osd_req_read_kern failed.\n", __func__);
 240                 goto out;
 241         }
 242
 243         ret = osd_finalize_request(or, 0, cred, NULL);
 244         if (unlikely(ret)) {
 245                 EXOFS_DBGMSG("Failed to osd_finalize_request() => %d\n", ret);
 246                 goto out;
 247         }
 248
 249         ret = osd_execute_request(or);
 250         if (unlikely(ret))
 251                 EXOFS_DBGMSG("osd_execute_request() => %d\n", ret);
 252         /* osd_req_decode_sense(or, ret); */
 253
 254 out:
 255         osd_end_request(or);
 256         EXOFS_DBGMSG2("read_kern(0x%llx) offset=0x%llx "
 257                       "length=0x%llx dev=%p ret=>%d\n",
 258                       _LLU(obj->id), _LLU(offset), _LLU(length), od, ret);
 259         return ret;
 260 }
 261
 262 static const struct osd_attr g_attr_sb_stats = ATTR_DEF(
 263         EXOFS_APAGE_SB_DATA,
 264         EXOFS_ATTR_SB_STATS,
 265         sizeof(struct exofs_sb_stats));
 266
 267 static int __sbi_read_stats(struct exofs_sb_info *sbi)
 268 {
 269         struct osd_attr attrs[] = {
 270                 [0] = g_attr_sb_stats,
 271         };
 272         struct ore_io_state *ios;
 273         int ret;
 274
 275         ret = ore_get_io_state(&sbi->layout, &sbi->oc, &ios);
 276         if (unlikely(ret)) {
 277                 EXOFS_ERR("%s: ore_get_io_state failed.\n", __func__);
 278                 return ret;
 279         }
 280
 281         ios->in_attr = attrs;
 282         ios->in_attr_len = ARRAY_SIZE(attrs);
 283
 284         ret = ore_read(ios);
 285         if (unlikely(ret)) {
 286                 EXOFS_ERR("Error reading super_block stats => %d\n", ret);
 287                 goto out;
 288         }
 289
 290         ret = extract_attr_from_ios(ios, &attrs[0]);
 291         if (ret) {
 292                 EXOFS_ERR("%s: extract_attr of sb_stats failed\n", __func__);
 293                 goto out;
 294         }
 295         if (attrs[0].len) {
 296                 struct exofs_sb_stats *ess;
 297
 298                 if (unlikely(attrs[0].len != sizeof(*ess))) {
 299                         EXOFS_ERR("%s: Wrong version of exofs_sb_stats "
 300                                   "size(%d) != expected(%zd)\n",
 301                                   __func__, attrs[0].len, sizeof(*ess));
 302                         goto out;
 303                 }
 304
 305                 ess = attrs[0].val_ptr;
 306                 sbi->s_nextid = le64_to_cpu(ess->s_nextid);
 307                 sbi->s_numfiles = le32_to_cpu(ess->s_numfiles);
 308         }
 309
 310 out:
 311         ore_put_io_state(ios);
 312         return ret;
 313 }
 314
 315 static void stats_done(struct ore_io_state *ios, void *p)
 316 {
 317         ore_put_io_state(ios);
 318         /* Good thanks nothing to do anymore */
 319 }
 320
 321 /* Asynchronously write the stats attribute */
 322 int exofs_sbi_write_stats(struct exofs_sb_info *sbi)
 323 {
 324         struct osd_attr attrs[] = {
 325                 [0] = g_attr_sb_stats,
 326         };
 327         struct ore_io_state *ios;
 328         int ret;
 329
 330         ret = ore_get_io_state(&sbi->layout, &sbi->oc, &ios);
 331         if (unlikely(ret)) {
 332                 EXOFS_ERR("%s: ore_get_io_state failed.\n", __func__);
 333                 return ret;
 334         }
 335
 336         sbi->s_ess.s_nextid   = cpu_to_le64(sbi->s_nextid);
 337         sbi->s_ess.s_numfiles = cpu_to_le64(sbi->s_numfiles);
 338         attrs[0].val_ptr = &sbi->s_ess;
 339
 340
 341         ios->done = stats_done;
 342         ios->private = sbi;
 343         ios->out_attr = attrs;
 344         ios->out_attr_len = ARRAY_SIZE(attrs);
 345
 346         ret = ore_write(ios);
 347         if (unlikely(ret)) {
 348                 EXOFS_ERR("%s: ore_write failed.\n", __func__);
 349                 ore_put_io_state(ios);
 350         }
 351
 352         return ret;
 353 }
 354
 355 /******************************************************************************
 356  * SUPERBLOCK FUNCTIONS
 357  *****************************************************************************/
 358 static const struct super_operations exofs_sops;
 359 static const struct export_operations exofs_export_ops;
 360
 361 /*
 362  * Write the superblock to the OSD
 363  */
 364 static int exofs_sync_fs(struct super_block *sb, int wait)
 365 {
 366         struct exofs_sb_info *sbi;
 367         struct exofs_fscb *fscb;
 368         struct ore_comp one_comp;
 369         struct ore_components oc;
 370         struct ore_io_state *ios;
 371         int ret = -ENOMEM;
 372
 373         fscb = kmalloc(sizeof(*fscb), GFP_KERNEL);
 374         if (unlikely(!fscb))
 375                 return -ENOMEM;
 376
 377         sbi = sb->s_fs_info;
 378
 379         /* NOTE: We no longer dirty the super_block anywhere in exofs. The
 380          * reason we write the fscb here on unmount is so we can stay backwards
 381          * compatible with fscb->s_version == 1. (What we are not compatible
 382          * with is if a new version FS crashed and then we try to mount an old
 383          * version). Otherwise the exofs_fscb is read-only from mkfs time. All
 384          * the writeable info is set in exofs_sbi_write_stats() above.
 385          */
 386
 387         exofs_init_comps(&oc, &one_comp, sbi, EXOFS_SUPER_ID);
 388
 389         ret = ore_get_io_state(&sbi->layout, &oc, &ios);
 390         if (unlikely(ret))
 391                 goto out;
 392
 393         ios->length = offsetof(struct exofs_fscb, s_dev_table_oid);
 394         memset(fscb, 0, ios->length);
 395         fscb->s_nextid = cpu_to_le64(sbi->s_nextid);
 396         fscb->s_numfiles = cpu_to_le64(sbi->s_numfiles);
 397         fscb->s_magic = cpu_to_le16(sb->s_magic);
 398         fscb->s_newfs = 0;
 399         fscb->s_version = EXOFS_FSCB_VER;
 400
 401         ios->offset = 0;
 402         ios->kern_buff = fscb;
 403
 404         ret = ore_write(ios);
 405         if (unlikely(ret))
 406                 EXOFS_ERR("%s: ore_write failed.\n", __func__);
 407
 408 out:
 409         EXOFS_DBGMSG("s_nextid=0x%llx ret=%d\n", _LLU(sbi->s_nextid), ret);
 410         ore_put_io_state(ios);
 411         kfree(fscb);
 412         return ret;
 413 }
 414
 415 static void _exofs_print_device(const char *msg, const char *dev_path,
 416                                 struct osd_dev *od, u64 pid)
 417 {
 418         const struct osd_dev_info *odi = osduld_device_info(od);
 419
 420         printk(KERN_NOTICE "exofs: %s %s osd_name-%s pid-0x%llx\n",
 421                 msg, dev_path ?: "", odi->osdname, _LLU(pid));
 422 }
 423
 424 static void exofs_free_sbi(struct exofs_sb_info *sbi)
 425 {
 426         unsigned numdevs = sbi->oc.numdevs;
 427
 428         while (numdevs) {
 429                 unsigned i = --numdevs;
 430                 struct osd_dev *od = ore_comp_dev(&sbi->oc, i);
 431
 432                 if (od) {
 433                         ore_comp_set_dev(&sbi->oc, i, NULL);
 434                         osduld_put_device(od);
 435                 }
 436         }
 437         kfree(sbi->oc.ods);
 438         kfree(sbi);
 439 }
 440
 441 /*
 442  * This function is called when the vfs is freeing the superblock.  We just
 443  * need to free our own part.
 444  */
 445 static void exofs_put_super(struct super_block *sb)
 446 {
 447         int num_pend;
 448         struct exofs_sb_info *sbi = sb->s_fs_info;
 449
 450         /* make sure there are no pending commands */
 451         for (num_pend = atomic_read(&sbi->s_curr_pending); num_pend > 0;
 452              num_pend = atomic_read(&sbi->s_curr_pending)) {
 453                 wait_queue_head_t wq;
 454
 455                 printk(KERN_NOTICE "%s: !!Pending operations in flight. "
 456                        "This is a BUG. please report to [email protected]\n",
 457                        __func__);
 458                 init_waitqueue_head(&wq);
 459                 wait_event_timeout(wq,
 460                                   (atomic_read(&sbi->s_curr_pending) == 0),
 461                                   msecs_to_jiffies(100));
 462         }
 463
 464         _exofs_print_device("Unmounting", NULL, ore_comp_dev(&sbi->oc, 0),
 465                             sbi->one_comp.obj.partition);
 466
 467         exofs_sysfs_sb_del(sbi);
 468         exofs_free_sbi(sbi);
 469         sb->s_fs_info = NULL;
 470 }
 471
 472 static int _read_and_match_data_map(struct exofs_sb_info *sbi, unsigned numdevs,
 473                                     struct exofs_device_table *dt)
 474 {
 475         int ret;
 476
 477         sbi->layout.stripe_unit =
 478                                 le64_to_cpu(dt->dt_data_map.cb_stripe_unit);
 479         sbi->layout.group_width =
 480                                 le32_to_cpu(dt->dt_data_map.cb_group_width);
 481         sbi->layout.group_depth =
 482                                 le32_to_cpu(dt->dt_data_map.cb_group_depth);
 483         sbi->layout.mirrors_p1  =
 484                                 le32_to_cpu(dt->dt_data_map.cb_mirror_cnt) + 1;
 485         sbi->layout.raid_algorithm  =
 486                                 le32_to_cpu(dt->dt_data_map.cb_raid_algorithm);
 487
 488         ret = ore_verify_layout(numdevs, &sbi->layout);
 489
 490         EXOFS_DBGMSG("exofs: layout: "
 491                 "num_comps=%u stripe_unit=0x%x group_width=%u "
 492                 "group_depth=0x%llx mirrors_p1=%u raid_algorithm=%u\n",
 493                 numdevs,
 494                 sbi->layout.stripe_unit,
 495                 sbi->layout.group_width,
 496                 _LLU(sbi->layout.group_depth),
 497                 sbi->layout.mirrors_p1,
 498                 sbi->layout.raid_algorithm);
 499         return ret;
 500 }
 501
 502 static unsigned __ra_pages(struct ore_layout *layout)
 503 {
 504         const unsigned _MIN_RA = 32; /* min 128K read-ahead */
 505         unsigned ra_pages = layout->group_width * layout->stripe_unit /
 506                                 PAGE_SIZE;
 507         unsigned max_io_pages = exofs_max_io_pages(layout, ~0);
 508
 509         ra_pages *= 2; /* two stripes */
 510         if (ra_pages < _MIN_RA)
 511                 ra_pages = roundup(_MIN_RA, ra_pages / 2);
 512
 513         if (ra_pages > max_io_pages)
 514                 ra_pages = max_io_pages;
 515
 516         return ra_pages;
 517 }
 518
 519 /* @odi is valid only as long as @fscb_dev is valid */
 520 static int exofs_devs_2_odi(struct exofs_dt_device_info *dt_dev,
 521                              struct osd_dev_info *odi)
 522 {
 523         odi->systemid_len = le32_to_cpu(dt_dev->systemid_len);
 524         if (likely(odi->systemid_len))
 525                 memcpy(odi->systemid, dt_dev->systemid, OSD_SYSTEMID_LEN);
 526
 527         odi->osdname_len = le32_to_cpu(dt_dev->osdname_len);
 528         odi->osdname = dt_dev->osdname;
 529
 530         /* FIXME support long names. Will need a _put function */
 531         if (dt_dev->long_name_offset)
 532                 return -EINVAL;
 533
 534         /* Make sure osdname is printable!
 535          * mkexofs should give us space for a null-terminator else the
 536          * device-table is invalid.
 537          */
 538         if (unlikely(odi->osdname_len >= sizeof(dt_dev->osdname)))
 539                 odi->osdname_len = sizeof(dt_dev->osdname) - 1;
 540         dt_dev->osdname[odi->osdname_len] = 0;
 541
 542         /* If it's all zeros something is bad we read past end-of-obj */
 543         return !(odi->systemid_len || odi->osdname_len);
 544 }
 545
 546 static int __alloc_dev_table(struct exofs_sb_info *sbi, unsigned numdevs,
 547                       struct exofs_dev **peds)
 548 {
 549         struct __alloc_ore_devs_and_exofs_devs {
 550                 /* Twice bigger table: See exofs_init_comps() and comment at
 551                  * exofs_read_lookup_dev_table()
 552                  */
 553                 struct ore_dev *oreds[numdevs * 2 - 1];
 554                 struct exofs_dev eds[numdevs];
 555         } *aoded;
 556         struct exofs_dev *eds;
 557         unsigned i;
 558
 559         aoded = kzalloc(sizeof(*aoded), GFP_KERNEL);
 560         if (unlikely(!aoded)) {
 561                 EXOFS_ERR("ERROR: failed allocating Device array[%d]\n",
 562                           numdevs);
 563                 return -ENOMEM;
 564         }
 565
 566         sbi->oc.ods = aoded->oreds;
 567         *peds = eds = aoded->eds;
 568         for (i = 0; i < numdevs; ++i)
 569                 aoded->oreds[i] = &eds[i].ored;
 570         return 0;
 571 }
 572
 573 static int exofs_read_lookup_dev_table(struct exofs_sb_info *sbi,
 574                                        struct osd_dev *fscb_od,
 575                                        unsigned table_count)
 576 {
 577         struct ore_comp comp;
 578         struct exofs_device_table *dt;
 579         struct exofs_dev *eds;
 580         unsigned table_bytes = table_count * sizeof(dt->dt_dev_table[0]) +
 581                                              sizeof(*dt);
 582         unsigned numdevs, i;
 583         int ret;
 584
 585         dt = kmalloc(table_bytes, GFP_KERNEL);
 586         if (unlikely(!dt)) {
 587                 EXOFS_ERR("ERROR: allocating %x bytes for device table\n",
 588                           table_bytes);
 589                 return -ENOMEM;
 590         }
 591
 592         sbi->oc.numdevs = 0;
 593
 594         comp.obj.partition = sbi->one_comp.obj.partition;
 595         comp.obj.id = EXOFS_DEVTABLE_ID;
 596         exofs_make_credential(comp.cred, &comp.obj);
 597
 598         ret = exofs_read_kern(fscb_od, comp.cred, &comp.obj, 0, dt,
 599                               table_bytes);
 600         if (unlikely(ret)) {
 601                 EXOFS_ERR("ERROR: reading device table\n");
 602                 goto out;
 603         }
 604
 605         numdevs = le64_to_cpu(dt->dt_num_devices);
 606         if (unlikely(!numdevs)) {
 607                 ret = -EINVAL;
 608                 goto out;
 609         }
 610         WARN_ON(table_count != numdevs);
 611
 612         ret = _read_and_match_data_map(sbi, numdevs, dt);
 613         if (unlikely(ret))
 614                 goto out;
 615
 616         ret = __alloc_dev_table(sbi, numdevs, &eds);
 617         if (unlikely(ret))
 618                 goto out;
 619         /* exofs round-robins the device table view according to inode
 620          * number. We hold a: twice bigger table hence inodes can point
 621          * to any device and have a sequential view of the table
 622          * starting at this device. See exofs_init_comps()
 623          */
 624         memcpy(&sbi->oc.ods[numdevs], &sbi->oc.ods[0],
 625                 (numdevs - 1) * sizeof(sbi->oc.ods[0]));
 626
 627         /* create sysfs subdir under which we put the device table
 628          * And cluster layout. A Superblock is identified by the string:
 629          *      "dev[0].osdname"_"pid"
 630          */
 631         exofs_sysfs_sb_add(sbi, &dt->dt_dev_table[0]);
 632
 633         for (i = 0; i < numdevs; i++) {
 634                 struct exofs_fscb fscb;
 635                 struct osd_dev_info odi;
 636                 struct osd_dev *od;
 637
 638                 if (exofs_devs_2_odi(&dt->dt_dev_table[i], &odi)) {
 639                         EXOFS_ERR("ERROR: Read all-zeros device entry\n");
 640                         ret = -EINVAL;
 641                         goto out;
 642                 }
 643
 644                 printk(KERN_NOTICE "Add device[%d]: osd_name-%s\n",
 645                        i, odi.osdname);
 646
 647                 /* the exofs id is currently the table index */
 648                 eds[i].did = i;
 649
 650                 /* On all devices the device table is identical. The user can
 651                  * specify any one of the participating devices on the command
 652                  * line. We always keep them in device-table order.
 653                  */
 654                 if (fscb_od && osduld_device_same(fscb_od, &odi)) {
 655                         eds[i].ored.od = fscb_od;
 656                         ++sbi->oc.numdevs;
 657                         fscb_od = NULL;
 658                         exofs_sysfs_odev_add(&eds[i], sbi);
 659                         continue;
 660                 }
 661
 662                 od = osduld_info_lookup(&odi);
 663                 if (IS_ERR(od)) {
 664                         ret = PTR_ERR(od);
 665                         EXOFS_ERR("ERROR: device requested is not found "
 666                                   "osd_name-%s =>%d\n", odi.osdname, ret);
 667                         goto out;
 668                 }
 669
 670                 eds[i].ored.od = od;
 671                 ++sbi->oc.numdevs;
 672
 673                 /* Read the fscb of the other devices to make sure the FS
 674                  * partition is there.
 675                  */
 676                 ret = exofs_read_kern(od, comp.cred, &comp.obj, 0, &fscb,
 677                                       sizeof(fscb));
 678                 if (unlikely(ret)) {
 679                         EXOFS_ERR("ERROR: Malformed participating device "
 680                                   "error reading fscb osd_name-%s\n",
 681                                   odi.osdname);
 682                         goto out;
 683                 }
 684                 exofs_sysfs_odev_add(&eds[i], sbi);
 685
 686                 /* TODO: verify other information is correct and FS-uuid
 687                  *       matches. Benny what did you say about device table
 688                  *       generation and old devices?
 689                  */
 690         }
 691
 692 out:
 693         kfree(dt);
 694         if (unlikely(fscb_od && !ret)) {
 695                         EXOFS_ERR("ERROR: Bad device-table container device not present\n");
 696                         osduld_put_device(fscb_od);
 697                         return -EINVAL;
 698         }
 699         return ret;
 700 }
 701
 702 /*
 703  * Read the superblock from the OSD and fill in the fields
 704  */
 705 static int exofs_fill_super(struct super_block *sb, void *data, int silent)
 706 {
 707         struct inode *root;
 708         struct exofs_mountopt *opts = data;
 709         struct exofs_sb_info *sbi;      /*extended info                  */
 710         struct osd_dev *od;             /* Master device                 */
 711         struct exofs_fscb fscb;         /*on-disk superblock info        */
 712         struct ore_comp comp;
 713         unsigned table_count;
 714         int ret;
 715
 716         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
 717         if (!sbi)
 718                 return -ENOMEM;
 719
 720         /* use mount options to fill superblock */
 721         if (opts->is_osdname) {
 722                 struct osd_dev_info odi = {.systemid_len = 0};
 723
 724                 odi.osdname_len = strlen(opts->dev_name);
 725                 odi.osdname = (u8 *)opts->dev_name;
 726                 od = osduld_info_lookup(&odi);
 727                 kfree(opts->dev_name);
 728                 opts->dev_name = NULL;
 729         } else {
 730                 od = osduld_path_lookup(opts->dev_name);
 731         }
 732         if (IS_ERR(od)) {
 733                 ret = -EINVAL;
 734                 goto free_sbi;
 735         }
 736
 737         /* Default layout in case we do not have a device-table */
 738         sbi->layout.stripe_unit = PAGE_SIZE;
 739         sbi->layout.mirrors_p1 = 1;
 740         sbi->layout.group_width = 1;
 741         sbi->layout.group_depth = -1;
 742         sbi->layout.group_count = 1;
 743         sbi->s_timeout = opts->timeout;
 744
 745         sbi->one_comp.obj.partition = opts->pid;
 746         sbi->one_comp.obj.id = 0;
 747         exofs_make_credential(sbi->one_comp.cred, &sbi->one_comp.obj);
 748         sbi->oc.single_comp = EC_SINGLE_COMP;
 749         sbi->oc.comps = &sbi->one_comp;
 750
 751         /* fill in some other data by hand */
 752         memset(sb->s_id, 0, sizeof(sb->s_id));
 753         strcpy(sb->s_id, "exofs");
 754         sb->s_blocksize = EXOFS_BLKSIZE;
 755         sb->s_blocksize_bits = EXOFS_BLKSHIFT;
 756         sb->s_maxbytes = MAX_LFS_FILESIZE;
 757         sb->s_max_links = EXOFS_LINK_MAX;
 758         atomic_set(&sbi->s_curr_pending, 0);
 759         sb->s_bdev = NULL;
 760         sb->s_dev = 0;
 761
 762         comp.obj.partition = sbi->one_comp.obj.partition;
 763         comp.obj.id = EXOFS_SUPER_ID;
 764         exofs_make_credential(comp.cred, &comp.obj);
 765
 766         ret = exofs_read_kern(od, comp.cred, &comp.obj, 0, &fscb, sizeof(fscb));
 767         if (unlikely(ret))
 768                 goto free_sbi;
 769
 770         sb->s_magic = le16_to_cpu(fscb.s_magic);
 771         /* NOTE: we read below to be backward compatible with old versions */
 772         sbi->s_nextid = le64_to_cpu(fscb.s_nextid);
 773         sbi->s_numfiles = le32_to_cpu(fscb.s_numfiles);
 774
 775         /* make sure what we read from the object store is correct */
 776         if (sb->s_magic != EXOFS_SUPER_MAGIC) {
 777                 if (!silent)
 778                         EXOFS_ERR("ERROR: Bad magic value\n");
 779                 ret = -EINVAL;
 780                 goto free_sbi;
 781         }
 782         if (le32_to_cpu(fscb.s_version) > EXOFS_FSCB_VER) {
 783                 EXOFS_ERR("ERROR: Bad FSCB version expected-%d got-%d\n",
 784                           EXOFS_FSCB_VER, le32_to_cpu(fscb.s_version));
 785                 ret = -EINVAL;
 786                 goto free_sbi;
 787         }
 788
 789         /* start generation numbers from a random point */
 790         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
 791         spin_lock_init(&sbi->s_next_gen_lock);
 792
 793         table_count = le64_to_cpu(fscb.s_dev_table_count);
 794         if (table_count) {
 795                 ret = exofs_read_lookup_dev_table(sbi, od, table_count);
 796                 if (unlikely(ret))
 797                         goto free_sbi;
 798         } else {
 799                 struct exofs_dev *eds;
 800
 801                 ret = __alloc_dev_table(sbi, 1, &eds);
 802                 if (unlikely(ret))
 803                         goto free_sbi;
 804
 805                 ore_comp_set_dev(&sbi->oc, 0, od);
 806                 sbi->oc.numdevs = 1;
 807         }
 808
 809         __sbi_read_stats(sbi);
 810
 811         /* set up operation vectors */
 812         ret = super_setup_bdi(sb);
 813         if (ret) {
 814                 EXOFS_DBGMSG("Failed to super_setup_bdi\n");
 815                 goto free_sbi;
 816         }
 817         sb->s_bdi->ra_pages = __ra_pages(&sbi->layout);
 818         sb->s_fs_info = sbi;
 819         sb->s_op = &exofs_sops;
 820         sb->s_export_op = &exofs_export_ops;
 821         root = exofs_iget(sb, EXOFS_ROOT_ID - EXOFS_OBJ_OFF);
 822         if (IS_ERR(root)) {
 823                 EXOFS_ERR("ERROR: exofs_iget failed\n");
 824                 ret = PTR_ERR(root);
 825                 goto free_sbi;
 826         }
 827         sb->s_root = d_make_root(root);
 828         if (!sb->s_root) {
 829                 EXOFS_ERR("ERROR: get root inode failed\n");
 830                 ret = -ENOMEM;
 831                 goto free_sbi;
 832         }
 833
 834         if (!S_ISDIR(root->i_mode)) {
 835                 dput(sb->s_root);
 836                 sb->s_root = NULL;
 837                 EXOFS_ERR("ERROR: corrupt root inode (mode = %hd)\n",
 838                        root->i_mode);
 839                 ret = -EINVAL;
 840                 goto free_sbi;
 841         }
 842
 843         exofs_sysfs_dbg_print();
 844         _exofs_print_device("Mounting", opts->dev_name,
 845                             ore_comp_dev(&sbi->oc, 0),
 846                             sbi->one_comp.obj.partition);
 847         return 0;
 848
 849 free_sbi:
 850         EXOFS_ERR("Unable to mount exofs on %s pid=0x%llx err=%d\n",
 851                   opts->dev_name, sbi->one_comp.obj.partition, ret);
 852         exofs_free_sbi(sbi);
 853         return ret;
 854 }
 855
 856 /*
 857  * Set up the superblock (calls exofs_fill_super eventually)
 858  */
 859 static struct dentry *exofs_mount(struct file_system_type *type,
 860                           int flags, const char *dev_name,
 861                           void *data)
 862 {
 863         struct exofs_mountopt opts;
 864         int ret;
 865
 866         ret = parse_options(data, &opts);
 867         if (ret)
 868                 return ERR_PTR(ret);
 869
 870         if (!opts.dev_name)
 871                 opts.dev_name = dev_name;
 872         return mount_nodev(type, flags, &opts, exofs_fill_super);
 873 }
 874
 875 /*
 876  * Return information about the file system state in the buffer.  This is used
 877  * by the 'df' command, for example.
 878  */
 879 static int exofs_statfs(struct dentry *dentry, struct kstatfs *buf)
 880 {
 881         struct super_block *sb = dentry->d_sb;
 882         struct exofs_sb_info *sbi = sb->s_fs_info;
 883         struct ore_io_state *ios;
 884         struct osd_attr attrs[] = {
 885                 ATTR_DEF(OSD_APAGE_PARTITION_QUOTAS,
 886                         OSD_ATTR_PQ_CAPACITY_QUOTA, sizeof(__be64)),
 887                 ATTR_DEF(OSD_APAGE_PARTITION_INFORMATION,
 888                         OSD_ATTR_PI_USED_CAPACITY, sizeof(__be64)),
 889         };
 890         uint64_t capacity = ULLONG_MAX;
 891         uint64_t used = ULLONG_MAX;
 892         int ret;
 893
 894         ret = ore_get_io_state(&sbi->layout, &sbi->oc, &ios);
 895         if (ret) {
 896                 EXOFS_DBGMSG("ore_get_io_state failed.\n");
 897                 return ret;
 898         }
 899
 900         ios->in_attr = attrs;
 901         ios->in_attr_len = ARRAY_SIZE(attrs);
 902
 903         ret = ore_read(ios);
 904         if (unlikely(ret))
 905                 goto out;
 906
 907         ret = extract_attr_from_ios(ios, &attrs[0]);
 908         if (likely(!ret)) {
 909                 capacity = get_unaligned_be64(attrs[0].val_ptr);
 910                 if (unlikely(!capacity))
 911                         capacity = ULLONG_MAX;
 912         } else
 913                 EXOFS_DBGMSG("exofs_statfs: get capacity failed.\n");
 914
 915         ret = extract_attr_from_ios(ios, &attrs[1]);
 916         if (likely(!ret))
 917                 used = get_unaligned_be64(attrs[1].val_ptr);
 918         else
 919                 EXOFS_DBGMSG("exofs_statfs: get used-space failed.\n");
 920
 921         /* fill in the stats buffer */
 922         buf->f_type = EXOFS_SUPER_MAGIC;
 923         buf->f_bsize = EXOFS_BLKSIZE;
 924         buf->f_blocks = capacity >> 9;
 925         buf->f_bfree = (capacity - used) >> 9;
 926         buf->f_bavail = buf->f_bfree;
 927         buf->f_files = sbi->s_numfiles;
 928         buf->f_ffree = EXOFS_MAX_ID - sbi->s_numfiles;
 929         buf->f_namelen = EXOFS_NAME_LEN;
 930
 931 out:
 932         ore_put_io_state(ios);
 933         return ret;
 934 }
 935
 936 static const struct super_operations exofs_sops = {
 937         .alloc_inode    = exofs_alloc_inode,
 938         .destroy_inode  = exofs_destroy_inode,
 939         .write_inode    = exofs_write_inode,
 940         .evict_inode    = exofs_evict_inode,
 941         .put_super      = exofs_put_super,
 942         .sync_fs        = exofs_sync_fs,
 943         .statfs         = exofs_statfs,
 944 };
 945
 946 /******************************************************************************
 947  * EXPORT OPERATIONS
 948  *****************************************************************************/
 949
 950 static struct dentry *exofs_get_parent(struct dentry *child)
 951 {
 952         unsigned long ino = exofs_parent_ino(child);
 953
 954         if (!ino)
 955                 return ERR_PTR(-ESTALE);
 956
 957         return d_obtain_alias(exofs_iget(child->d_sb, ino));
 958 }
 959
 960 static struct inode *exofs_nfs_get_inode(struct super_block *sb,
 961                 u64 ino, u32 generation)
 962 {
 963         struct inode *inode;
 964
 965         inode = exofs_iget(sb, ino);
 966         if (IS_ERR(inode))
 967                 return ERR_CAST(inode);
 968         if (generation && inode->i_generation != generation) {
 969                 /* we didn't find the right inode.. */
 970                 iput(inode);
 971                 return ERR_PTR(-ESTALE);
 972         }
 973         return inode;
 974 }
 975
 976 static struct dentry *exofs_fh_to_dentry(struct super_block *sb,
 977                                 struct fid *fid, int fh_len, int fh_type)
 978 {
 979         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 980                                     exofs_nfs_get_inode);
 981 }
 982
 983 static struct dentry *exofs_fh_to_parent(struct super_block *sb,
 984                                 struct fid *fid, int fh_len, int fh_type)
 985 {
 986         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 987                                     exofs_nfs_get_inode);
 988 }
 989
 990 static const struct export_operations exofs_export_ops = {
 991         .fh_to_dentry = exofs_fh_to_dentry,
 992         .fh_to_parent = exofs_fh_to_parent,
 993         .get_parent = exofs_get_parent,
 994 };
 995
 996 /******************************************************************************
 997  * INSMOD/RMMOD
 998  *****************************************************************************/
 999
1000 /*
1001  * struct that describes this file system
1002  */
1003 static struct file_system_type exofs_type = {
1004         .owner          = THIS_MODULE,
1005         .name           = "exofs",
1006         .mount          = exofs_mount,
1007         .kill_sb        = generic_shutdown_super,
1008 };
1009 MODULE_ALIAS_FS("exofs");
1010
1011 static int __init init_exofs(void)
1012 {
1013         int err;
1014
1015         err = init_inodecache();
1016         if (err)
1017                 goto out;
1018
1019         err = register_filesystem(&exofs_type);
1020         if (err)
1021                 goto out_d;
1022
1023         /* We don't fail if sysfs creation failed */
1024         exofs_sysfs_init();
1025
1026         return 0;
1027 out_d:
1028         destroy_inodecache();
1029 out:
1030         return err;
1031 }
1032
1033 static void __exit exit_exofs(void)
1034 {
1035         exofs_sysfs_uninit();
1036         unregister_filesystem(&exofs_type);
1037         destroy_inodecache();
1038 }
1039
1040 MODULE_AUTHOR("Avishay Traeger <[email protected]>");
1041 MODULE_DESCRIPTION("exofs");
1042 MODULE_LICENSE("GPL");
1043
1044 module_init(init_exofs)
1045 module_exit(exit_exofs)