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
  42 #include "exofs.h"
  43
  44 #define EXOFS_DBGMSG2(M...) do {} while (0)
  45
  46 /******************************************************************************
  47  * MOUNT OPTIONS
  48  *****************************************************************************/
  49
  50 /*
  51  * struct to hold what we get from mount options
  52  */
  53 struct exofs_mountopt {
  54         bool is_osdname;
  55         const char *dev_name;
  56         uint64_t pid;
  57         int timeout;
  58 };
  59
  60 /*
  61  * exofs-specific mount-time options.
  62  */
  63 enum { Opt_name, Opt_pid, Opt_to, Opt_err };
  64
  65 /*
  66  * Our mount-time options.  These should ideally be 64-bit unsigned, but the
  67  * kernel's parsing functions do not currently support that.  32-bit should be
  68  * sufficient for most applications now.
  69  */
  70 static match_table_t tokens = {
  71         {Opt_name, "osdname=%s"},
  72         {Opt_pid, "pid=%u"},
  73         {Opt_to, "to=%u"},
  74         {Opt_err, NULL}
  75 };
  76
  77 /*
  78  * The main option parsing method.  Also makes sure that all of the mandatory
  79  * mount options were set.
  80  */
  81 static int parse_options(char *options, struct exofs_mountopt *opts)
  82 {
  83         char *p;
  84         substring_t args[MAX_OPT_ARGS];
  85         int option;
  86         bool s_pid = false;
  87
  88         EXOFS_DBGMSG("parse_options %s\n", options);
  89         /* defaults */
  90         memset(opts, 0, sizeof(*opts));
  91         opts->timeout = BLK_DEFAULT_SG_TIMEOUT;
  92
  93         while ((p = strsep(&options, ",")) != NULL) {
  94                 int token;
  95                 char str[32];
  96
  97                 if (!*p)
  98                         continue;
  99
 100                 token = match_token(p, tokens, args);
 101                 switch (token) {
 102                 case Opt_name:
 103                         kfree(opts->dev_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("Timout 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         oi->vfs_inode.i_version = 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                                 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 osd-dev@open-osd.org\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         bdi_destroy(&sbi->bdi);
 469         exofs_free_sbi(sbi);
 470         sb->s_fs_info = NULL;
 471 }
 472
 473 static int _read_and_match_data_map(struct exofs_sb_info *sbi, unsigned numdevs,
 474                                     struct exofs_device_table *dt)
 475 {
 476         int ret;
 477
 478         sbi->layout.stripe_unit =
 479                                 le64_to_cpu(dt->dt_data_map.cb_stripe_unit);
 480         sbi->layout.group_width =
 481                                 le32_to_cpu(dt->dt_data_map.cb_group_width);
 482         sbi->layout.group_depth =
 483                                 le32_to_cpu(dt->dt_data_map.cb_group_depth);
 484         sbi->layout.mirrors_p1  =
 485                                 le32_to_cpu(dt->dt_data_map.cb_mirror_cnt) + 1;
 486         sbi->layout.raid_algorithm  =
 487                                 le32_to_cpu(dt->dt_data_map.cb_raid_algorithm);
 488
 489         ret = ore_verify_layout(numdevs, &sbi->layout);
 490
 491         EXOFS_DBGMSG("exofs: layout: "
 492                 "num_comps=%u stripe_unit=0x%x group_width=%u "
 493                 "group_depth=0x%llx mirrors_p1=%u raid_algorithm=%u\n",
 494                 numdevs,
 495                 sbi->layout.stripe_unit,
 496                 sbi->layout.group_width,
 497                 _LLU(sbi->layout.group_depth),
 498                 sbi->layout.mirrors_p1,
 499                 sbi->layout.raid_algorithm);
 500         return ret;
 501 }
 502
 503 static unsigned __ra_pages(struct ore_layout *layout)
 504 {
 505         const unsigned _MIN_RA = 32; /* min 128K read-ahead */
 506         unsigned ra_pages = layout->group_width * layout->stripe_unit /
 507                                 PAGE_SIZE;
 508         unsigned max_io_pages = exofs_max_io_pages(layout, ~0);
 509
 510         ra_pages *= 2; /* two stripes */
 511         if (ra_pages < _MIN_RA)
 512                 ra_pages = roundup(_MIN_RA, ra_pages / 2);
 513
 514         if (ra_pages > max_io_pages)
 515                 ra_pages = max_io_pages;
 516
 517         return ra_pages;
 518 }
 519
 520 /* @odi is valid only as long as @fscb_dev is valid */
 521 static int exofs_devs_2_odi(struct exofs_dt_device_info *dt_dev,
 522                              struct osd_dev_info *odi)
 523 {
 524         odi->systemid_len = le32_to_cpu(dt_dev->systemid_len);
 525         if (likely(odi->systemid_len))
 526                 memcpy(odi->systemid, dt_dev->systemid, OSD_SYSTEMID_LEN);
 527
 528         odi->osdname_len = le32_to_cpu(dt_dev->osdname_len);
 529         odi->osdname = dt_dev->osdname;
 530
 531         /* FIXME support long names. Will need a _put function */
 532         if (dt_dev->long_name_offset)
 533                 return -EINVAL;
 534
 535         /* Make sure osdname is printable!
 536          * mkexofs should give us space for a null-terminator else the
 537          * device-table is invalid.
 538          */
 539         if (unlikely(odi->osdname_len >= sizeof(dt_dev->osdname)))
 540                 odi->osdname_len = sizeof(dt_dev->osdname) - 1;
 541         dt_dev->osdname[odi->osdname_len] = 0;
 542
 543         /* If it's all zeros something is bad we read past end-of-obj */
 544         return !(odi->systemid_len || odi->osdname_len);
 545 }
 546
 547 static int __alloc_dev_table(struct exofs_sb_info *sbi, unsigned numdevs,
 548                       struct exofs_dev **peds)
 549 {
 550         struct __alloc_ore_devs_and_exofs_devs {
 551                 /* Twice bigger table: See exofs_init_comps() and comment at
 552                  * exofs_read_lookup_dev_table()
 553                  */
 554                 struct ore_dev *oreds[numdevs * 2 - 1];
 555                 struct exofs_dev eds[numdevs];
 556         } *aoded;
 557         struct exofs_dev *eds;
 558         unsigned i;
 559
 560         aoded = kzalloc(sizeof(*aoded), GFP_KERNEL);
 561         if (unlikely(!aoded)) {
 562                 EXOFS_ERR("ERROR: failed allocating Device array[%d]\n",
 563                           numdevs);
 564                 return -ENOMEM;
 565         }
 566
 567         sbi->oc.ods = aoded->oreds;
 568         *peds = eds = aoded->eds;
 569         for (i = 0; i < numdevs; ++i)
 570                 aoded->oreds[i] = &eds[i].ored;
 571         return 0;
 572 }
 573
 574 static int exofs_read_lookup_dev_table(struct exofs_sb_info *sbi,
 575                                        struct osd_dev *fscb_od,
 576                                        unsigned table_count)
 577 {
 578         struct ore_comp comp;
 579         struct exofs_device_table *dt;
 580         struct exofs_dev *eds;
 581         unsigned table_bytes = table_count * sizeof(dt->dt_dev_table[0]) +
 582                                              sizeof(*dt);
 583         unsigned numdevs, i;
 584         int ret;
 585
 586         dt = kmalloc(table_bytes, GFP_KERNEL);
 587         if (unlikely(!dt)) {
 588                 EXOFS_ERR("ERROR: allocating %x bytes for device table\n",
 589                           table_bytes);
 590                 return -ENOMEM;
 591         }
 592
 593         sbi->oc.numdevs = 0;
 594
 595         comp.obj.partition = sbi->one_comp.obj.partition;
 596         comp.obj.id = EXOFS_DEVTABLE_ID;
 597         exofs_make_credential(comp.cred, &comp.obj);
 598
 599         ret = exofs_read_kern(fscb_od, comp.cred, &comp.obj, 0, dt,
 600                               table_bytes);
 601         if (unlikely(ret)) {
 602                 EXOFS_ERR("ERROR: reading device table\n");
 603                 goto out;
 604         }
 605
 606         numdevs = le64_to_cpu(dt->dt_num_devices);
 607         if (unlikely(!numdevs)) {
 608                 ret = -EINVAL;
 609                 goto out;
 610         }
 611         WARN_ON(table_count != numdevs);
 612
 613         ret = _read_and_match_data_map(sbi, numdevs, dt);
 614         if (unlikely(ret))
 615                 goto out;
 616
 617         ret = __alloc_dev_table(sbi, numdevs, &eds);
 618         if (unlikely(ret))
 619                 goto out;
 620         /* exofs round-robins the device table view according to inode
 621          * number. We hold a: twice bigger table hence inodes can point
 622          * to any device and have a sequential view of the table
 623          * starting at this device. See exofs_init_comps()
 624          */
 625         memcpy(&sbi->oc.ods[numdevs], &sbi->oc.ods[0],
 626                 (numdevs - 1) * sizeof(sbi->oc.ods[0]));
 627
 628         /* create sysfs subdir under which we put the device table
 629          * And cluster layout. A Superblock is identified by the string:
 630          *      "dev[0].osdname"_"pid"
 631          */
 632         exofs_sysfs_sb_add(sbi, &dt->dt_dev_table[0]);
 633
 634         for (i = 0; i < numdevs; i++) {
 635                 struct exofs_fscb fscb;
 636                 struct osd_dev_info odi;
 637                 struct osd_dev *od;
 638
 639                 if (exofs_devs_2_odi(&dt->dt_dev_table[i], &odi)) {
 640                         EXOFS_ERR("ERROR: Read all-zeros device entry\n");
 641                         ret = -EINVAL;
 642                         goto out;
 643                 }
 644
 645                 printk(KERN_NOTICE "Add device[%d]: osd_name-%s\n",
 646                        i, odi.osdname);
 647
 648                 /* the exofs id is currently the table index */
 649                 eds[i].did = i;
 650
 651                 /* On all devices the device table is identical. The user can
 652                  * specify any one of the participating devices on the command
 653                  * line. We always keep them in device-table order.
 654                  */
 655                 if (fscb_od && osduld_device_same(fscb_od, &odi)) {
 656                         eds[i].ored.od = fscb_od;
 657                         ++sbi->oc.numdevs;
 658                         fscb_od = NULL;
 659                         exofs_sysfs_odev_add(&eds[i], sbi);
 660                         continue;
 661                 }
 662
 663                 od = osduld_info_lookup(&odi);
 664                 if (IS_ERR(od)) {
 665                         ret = PTR_ERR(od);
 666                         EXOFS_ERR("ERROR: device requested is not found "
 667                                   "osd_name-%s =>%d\n", odi.osdname, ret);
 668                         goto out;
 669                 }
 670
 671                 eds[i].ored.od = od;
 672                 ++sbi->oc.numdevs;
 673
 674                 /* Read the fscb of the other devices to make sure the FS
 675                  * partition is there.
 676                  */
 677                 ret = exofs_read_kern(od, comp.cred, &comp.obj, 0, &fscb,
 678                                       sizeof(fscb));
 679                 if (unlikely(ret)) {
 680                         EXOFS_ERR("ERROR: Malformed participating device "
 681                                   "error reading fscb osd_name-%s\n",
 682                                   odi.osdname);
 683                         goto out;
 684                 }
 685                 exofs_sysfs_odev_add(&eds[i], sbi);
 686
 687                 /* TODO: verify other information is correct and FS-uuid
 688                  *       matches. Benny what did you say about device table
 689                  *       generation and old devices?
 690                  */
 691         }
 692
 693 out:
 694         kfree(dt);
 695         if (unlikely(fscb_od && !ret)) {
 696                         EXOFS_ERR("ERROR: Bad device-table container device not present\n");
 697                         osduld_put_device(fscb_od);
 698                         return -EINVAL;
 699         }
 700         return ret;
 701 }
 702
 703 /*
 704  * Read the superblock from the OSD and fill in the fields
 705  */
 706 static int exofs_fill_super(struct super_block *sb, void *data, int silent)
 707 {
 708         struct inode *root;
 709         struct exofs_mountopt *opts = data;
 710         struct exofs_sb_info *sbi;      /*extended info                  */
 711         struct osd_dev *od;             /* Master device                 */
 712         struct exofs_fscb fscb;         /*on-disk superblock info        */
 713         struct ore_comp comp;
 714         unsigned table_count;
 715         int ret;
 716
 717         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
 718         if (!sbi)
 719                 return -ENOMEM;
 720
 721         /* use mount options to fill superblock */
 722         if (opts->is_osdname) {
 723                 struct osd_dev_info odi = {.systemid_len = 0};
 724
 725                 odi.osdname_len = strlen(opts->dev_name);
 726                 odi.osdname = (u8 *)opts->dev_name;
 727                 od = osduld_info_lookup(&odi);
 728                 kfree(opts->dev_name);
 729                 opts->dev_name = NULL;
 730         } else {
 731                 od = osduld_path_lookup(opts->dev_name);
 732         }
 733         if (IS_ERR(od)) {
 734                 ret = -EINVAL;
 735                 goto free_sbi;
 736         }
 737
 738         /* Default layout in case we do not have a device-table */
 739         sbi->layout.stripe_unit = PAGE_SIZE;
 740         sbi->layout.mirrors_p1 = 1;
 741         sbi->layout.group_width = 1;
 742         sbi->layout.group_depth = -1;
 743         sbi->layout.group_count = 1;
 744         sbi->s_timeout = opts->timeout;
 745
 746         sbi->one_comp.obj.partition = opts->pid;
 747         sbi->one_comp.obj.id = 0;
 748         exofs_make_credential(sbi->one_comp.cred, &sbi->one_comp.obj);
 749         sbi->oc.single_comp = EC_SINGLE_COMP;
 750         sbi->oc.comps = &sbi->one_comp;
 751
 752         /* fill in some other data by hand */
 753         memset(sb->s_id, 0, sizeof(sb->s_id));
 754         strcpy(sb->s_id, "exofs");
 755         sb->s_blocksize = EXOFS_BLKSIZE;
 756         sb->s_blocksize_bits = EXOFS_BLKSHIFT;
 757         sb->s_maxbytes = MAX_LFS_FILESIZE;
 758         sb->s_max_links = EXOFS_LINK_MAX;
 759         atomic_set(&sbi->s_curr_pending, 0);
 760         sb->s_bdev = NULL;
 761         sb->s_dev = 0;
 762
 763         comp.obj.partition = sbi->one_comp.obj.partition;
 764         comp.obj.id = EXOFS_SUPER_ID;
 765         exofs_make_credential(comp.cred, &comp.obj);
 766
 767         ret = exofs_read_kern(od, comp.cred, &comp.obj, 0, &fscb, sizeof(fscb));
 768         if (unlikely(ret))
 769                 goto free_sbi;
 770
 771         sb->s_magic = le16_to_cpu(fscb.s_magic);
 772         /* NOTE: we read below to be backward compatible with old versions */
 773         sbi->s_nextid = le64_to_cpu(fscb.s_nextid);
 774         sbi->s_numfiles = le32_to_cpu(fscb.s_numfiles);
 775
 776         /* make sure what we read from the object store is correct */
 777         if (sb->s_magic != EXOFS_SUPER_MAGIC) {
 778                 if (!silent)
 779                         EXOFS_ERR("ERROR: Bad magic value\n");
 780                 ret = -EINVAL;
 781                 goto free_sbi;
 782         }
 783         if (le32_to_cpu(fscb.s_version) > EXOFS_FSCB_VER) {
 784                 EXOFS_ERR("ERROR: Bad FSCB version expected-%d got-%d\n",
 785                           EXOFS_FSCB_VER, le32_to_cpu(fscb.s_version));
 786                 ret = -EINVAL;
 787                 goto free_sbi;
 788         }
 789
 790         /* start generation numbers from a random point */
 791         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
 792         spin_lock_init(&sbi->s_next_gen_lock);
 793
 794         table_count = le64_to_cpu(fscb.s_dev_table_count);
 795         if (table_count) {
 796                 ret = exofs_read_lookup_dev_table(sbi, od, table_count);
 797                 if (unlikely(ret))
 798                         goto free_sbi;
 799         } else {
 800                 struct exofs_dev *eds;
 801
 802                 ret = __alloc_dev_table(sbi, 1, &eds);
 803                 if (unlikely(ret))
 804                         goto free_sbi;
 805
 806                 ore_comp_set_dev(&sbi->oc, 0, od);
 807                 sbi->oc.numdevs = 1;
 808         }
 809
 810         __sbi_read_stats(sbi);
 811
 812         /* set up operation vectors */
 813         sbi->bdi.ra_pages = __ra_pages(&sbi->layout);
 814         sb->s_bdi = &sbi->bdi;
 815         sb->s_fs_info = sbi;
 816         sb->s_op = &exofs_sops;
 817         sb->s_export_op = &exofs_export_ops;
 818         root = exofs_iget(sb, EXOFS_ROOT_ID - EXOFS_OBJ_OFF);
 819         if (IS_ERR(root)) {
 820                 EXOFS_ERR("ERROR: exofs_iget failed\n");
 821                 ret = PTR_ERR(root);
 822                 goto free_sbi;
 823         }
 824         sb->s_root = d_make_root(root);
 825         if (!sb->s_root) {
 826                 EXOFS_ERR("ERROR: get root inode failed\n");
 827                 ret = -ENOMEM;
 828                 goto free_sbi;
 829         }
 830
 831         if (!S_ISDIR(root->i_mode)) {
 832                 dput(sb->s_root);
 833                 sb->s_root = NULL;
 834                 EXOFS_ERR("ERROR: corrupt root inode (mode = %hd)\n",
 835                        root->i_mode);
 836                 ret = -EINVAL;
 837                 goto free_sbi;
 838         }
 839
 840         ret = bdi_setup_and_register(&sbi->bdi, "exofs");
 841         if (ret) {
 842                 EXOFS_DBGMSG("Failed to bdi_setup_and_register\n");
 843                 dput(sb->s_root);
 844                 sb->s_root = NULL;
 845                 goto free_sbi;
 846         }
 847
 848         exofs_sysfs_dbg_print();
 849         _exofs_print_device("Mounting", opts->dev_name,
 850                             ore_comp_dev(&sbi->oc, 0),
 851                             sbi->one_comp.obj.partition);
 852         return 0;
 853
 854 free_sbi:
 855         EXOFS_ERR("Unable to mount exofs on %s pid=0x%llx err=%d\n",
 856                   opts->dev_name, sbi->one_comp.obj.partition, ret);
 857         exofs_free_sbi(sbi);
 858         return ret;
 859 }
 860
 861 /*
 862  * Set up the superblock (calls exofs_fill_super eventually)
 863  */
 864 static struct dentry *exofs_mount(struct file_system_type *type,
 865                           int flags, const char *dev_name,
 866                           void *data)
 867 {
 868         struct exofs_mountopt opts;
 869         int ret;
 870
 871         ret = parse_options(data, &opts);
 872         if (ret) {
 873                 kfree(opts.dev_name);
 874                 return ERR_PTR(ret);
 875         }
 876
 877         if (!opts.dev_name)
 878                 opts.dev_name = dev_name;
 879         return mount_nodev(type, flags, &opts, exofs_fill_super);
 880 }
 881
 882 /*
 883  * Return information about the file system state in the buffer.  This is used
 884  * by the 'df' command, for example.
 885  */
 886 static int exofs_statfs(struct dentry *dentry, struct kstatfs *buf)
 887 {
 888         struct super_block *sb = dentry->d_sb;
 889         struct exofs_sb_info *sbi = sb->s_fs_info;
 890         struct ore_io_state *ios;
 891         struct osd_attr attrs[] = {
 892                 ATTR_DEF(OSD_APAGE_PARTITION_QUOTAS,
 893                         OSD_ATTR_PQ_CAPACITY_QUOTA, sizeof(__be64)),
 894                 ATTR_DEF(OSD_APAGE_PARTITION_INFORMATION,
 895                         OSD_ATTR_PI_USED_CAPACITY, sizeof(__be64)),
 896         };
 897         uint64_t capacity = ULLONG_MAX;
 898         uint64_t used = ULLONG_MAX;
 899         int ret;
 900
 901         ret = ore_get_io_state(&sbi->layout, &sbi->oc, &ios);
 902         if (ret) {
 903                 EXOFS_DBGMSG("ore_get_io_state failed.\n");
 904                 return ret;
 905         }
 906
 907         ios->in_attr = attrs;
 908         ios->in_attr_len = ARRAY_SIZE(attrs);
 909
 910         ret = ore_read(ios);
 911         if (unlikely(ret))
 912                 goto out;
 913
 914         ret = extract_attr_from_ios(ios, &attrs[0]);
 915         if (likely(!ret)) {
 916                 capacity = get_unaligned_be64(attrs[0].val_ptr);
 917                 if (unlikely(!capacity))
 918                         capacity = ULLONG_MAX;
 919         } else
 920                 EXOFS_DBGMSG("exofs_statfs: get capacity failed.\n");
 921
 922         ret = extract_attr_from_ios(ios, &attrs[1]);
 923         if (likely(!ret))
 924                 used = get_unaligned_be64(attrs[1].val_ptr);
 925         else
 926                 EXOFS_DBGMSG("exofs_statfs: get used-space failed.\n");
 927
 928         /* fill in the stats buffer */
 929         buf->f_type = EXOFS_SUPER_MAGIC;
 930         buf->f_bsize = EXOFS_BLKSIZE;
 931         buf->f_blocks = capacity >> 9;
 932         buf->f_bfree = (capacity - used) >> 9;
 933         buf->f_bavail = buf->f_bfree;
 934         buf->f_files = sbi->s_numfiles;
 935         buf->f_ffree = EXOFS_MAX_ID - sbi->s_numfiles;
 936         buf->f_namelen = EXOFS_NAME_LEN;
 937
 938 out:
 939         ore_put_io_state(ios);
 940         return ret;
 941 }
 942
 943 static const struct super_operations exofs_sops = {
 944         .alloc_inode    = exofs_alloc_inode,
 945         .destroy_inode  = exofs_destroy_inode,
 946         .write_inode    = exofs_write_inode,
 947         .evict_inode    = exofs_evict_inode,
 948         .put_super      = exofs_put_super,
 949         .sync_fs        = exofs_sync_fs,
 950         .statfs         = exofs_statfs,
 951 };
 952
 953 /******************************************************************************
 954  * EXPORT OPERATIONS
 955  *****************************************************************************/
 956
 957 static struct dentry *exofs_get_parent(struct dentry *child)
 958 {
 959         unsigned long ino = exofs_parent_ino(child);
 960
 961         if (!ino)
 962                 return ERR_PTR(-ESTALE);
 963
 964         return d_obtain_alias(exofs_iget(d_inode(child)->i_sb, ino));
 965 }
 966
 967 static struct inode *exofs_nfs_get_inode(struct super_block *sb,
 968                 u64 ino, u32 generation)
 969 {
 970         struct inode *inode;
 971
 972         inode = exofs_iget(sb, ino);
 973         if (IS_ERR(inode))
 974                 return ERR_CAST(inode);
 975         if (generation && inode->i_generation != generation) {
 976                 /* we didn't find the right inode.. */
 977                 iput(inode);
 978                 return ERR_PTR(-ESTALE);
 979         }
 980         return inode;
 981 }
 982
 983 static struct dentry *exofs_fh_to_dentry(struct super_block *sb,
 984                                 struct fid *fid, int fh_len, int fh_type)
 985 {
 986         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 987                                     exofs_nfs_get_inode);
 988 }
 989
 990 static struct dentry *exofs_fh_to_parent(struct super_block *sb,
 991                                 struct fid *fid, int fh_len, int fh_type)
 992 {
 993         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 994                                     exofs_nfs_get_inode);
 995 }
 996
 997 static const struct export_operations exofs_export_ops = {
 998         .fh_to_dentry = exofs_fh_to_dentry,
 999         .fh_to_parent = exofs_fh_to_parent,
1000         .get_parent = exofs_get_parent,
1001 };
1002
1003 /******************************************************************************
1004  * INSMOD/RMMOD
1005  *****************************************************************************/
1006
1007 /*
1008  * struct that describes this file system
1009  */
1010 static struct file_system_type exofs_type = {
1011         .owner          = THIS_MODULE,
1012         .name           = "exofs",
1013         .mount          = exofs_mount,
1014         .kill_sb        = generic_shutdown_super,
1015 };
1016 MODULE_ALIAS_FS("exofs");
1017
1018 static int __init init_exofs(void)
1019 {
1020         int err;
1021
1022         err = init_inodecache();
1023         if (err)
1024                 goto out;
1025
1026         err = register_filesystem(&exofs_type);
1027         if (err)
1028                 goto out_d;
1029
1030         /* We don't fail if sysfs creation failed */
1031         exofs_sysfs_init();
1032
1033         return 0;
1034 out_d:
1035         destroy_inodecache();
1036 out:
1037         return err;
1038 }
1039
1040 static void __exit exit_exofs(void)
1041 {
1042         exofs_sysfs_uninit();
1043         unregister_filesystem(&exofs_type);
1044         destroy_inodecache();
1045 }
1046
1047 MODULE_AUTHOR("Avishay Traeger <avishay@gmail.com>");
1048 MODULE_DESCRIPTION("exofs");
1049 MODULE_LICENSE("GPL");
1050
1051 module_init(init_exofs)
1052 module_exit(exit_exofs)