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                         kfree(opts->dev_name);
 105                         opts->dev_name = match_strdup(&args[0]);
 106                         if (unlikely(!opts->dev_name)) {
 107                                 EXOFS_ERR("Error allocating dev_name");
 108                                 return -ENOMEM;
 109                         }
 110                         opts->is_osdname = true;
 111                         break;
 112                 case Opt_pid:
 113                         if (0 == match_strlcpy(str, &args[0], sizeof(str)))
 114                                 return -EINVAL;
 115                         opts->pid = simple_strtoull(str, NULL, 0);
 116                         if (opts->pid < EXOFS_MIN_PID) {
 117                                 EXOFS_ERR("Partition ID must be >= %u",
 118                                           EXOFS_MIN_PID);
 119                                 return -EINVAL;
 120                         }
 121                         s_pid = true;
 122                         break;
 123                 case Opt_to:
 124                         if (match_int(&args[0], &option))
 125                                 return -EINVAL;
 126                         if (option <= 0) {
 127                                 EXOFS_ERR("Timeout must be > 0");
 128                                 return -EINVAL;
 129                         }
 130                         opts->timeout = option * HZ;
 131                         break;
 132                 }
 133         }
 134
 135         if (!s_pid) {
 136                 EXOFS_ERR("Need to specify the following options:\n");
 137                 EXOFS_ERR("    -o pid=pid_no_to_use\n");
 138                 return -EINVAL;
 139         }
 140
 141         return 0;
 142 }
 143
 144 /******************************************************************************
 145  * INODE CACHE
 146  *****************************************************************************/
 147
 148 /*
 149  * Our inode cache.  Isn't it pretty?
 150  */
 151 static struct kmem_cache *exofs_inode_cachep;
 152
 153 /*
 154  * Allocate an inode in the cache
 155  */
 156 static struct inode *exofs_alloc_inode(struct super_block *sb)
 157 {
 158         struct exofs_i_info *oi;
 159
 160         oi = kmem_cache_alloc(exofs_inode_cachep, GFP_KERNEL);
 161         if (!oi)
 162                 return NULL;
 163
 164         inode_set_iversion(&oi->vfs_inode, 1);
 165         return &oi->vfs_inode;
 166 }
 167
 168 static void exofs_i_callback(struct rcu_head *head)
 169 {
 170         struct inode *inode = container_of(head, struct inode, i_rcu);
 171         kmem_cache_free(exofs_inode_cachep, exofs_i(inode));
 172 }
 173
 174 /*
 175  * Remove an inode from the cache
 176  */
 177 static void exofs_destroy_inode(struct inode *inode)
 178 {
 179         call_rcu(&inode->i_rcu, exofs_i_callback);
 180 }
 181
 182 /*
 183  * Initialize the inode
 184  */
 185 static void exofs_init_once(void *foo)
 186 {
 187         struct exofs_i_info *oi = foo;
 188
 189         inode_init_once(&oi->vfs_inode);
 190 }
 191
 192 /*
 193  * Create and initialize the inode cache
 194  */
 195 static int init_inodecache(void)
 196 {
 197         exofs_inode_cachep = kmem_cache_create_usercopy("exofs_inode_cache",
 198                                 sizeof(struct exofs_i_info), 0,
 199                                 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD |
 200                                 SLAB_ACCOUNT,
 201                                 offsetof(struct exofs_i_info, i_data),
 202                                 sizeof_field(struct exofs_i_info, i_data),
 203                                 exofs_init_once);
 204         if (exofs_inode_cachep == NULL)
 205                 return -ENOMEM;
 206         return 0;
 207 }
 208
 209 /*
 210  * Destroy the inode cache
 211  */
 212 static void destroy_inodecache(void)
 213 {
 214         /*
 215          * Make sure all delayed rcu free inodes are flushed before we
 216          * destroy cache.
 217          */
 218         rcu_barrier();
 219         kmem_cache_destroy(exofs_inode_cachep);
 220 }
 221
 222 /******************************************************************************
 223  * Some osd helpers
 224  *****************************************************************************/
 225 void exofs_make_credential(u8 cred_a[OSD_CAP_LEN], const struct osd_obj_id *obj)
 226 {
 227         osd_sec_init_nosec_doall_caps(cred_a, obj, false, true);
 228 }
 229
 230 static int exofs_read_kern(struct osd_dev *od, u8 *cred, struct osd_obj_id *obj,
 231                     u64 offset, void *p, unsigned length)
 232 {
 233         struct osd_request *or = osd_start_request(od);
 234 /*      struct osd_sense_info osi = {.key = 0};*/
 235         int ret;
 236
 237         if (unlikely(!or)) {
 238                 EXOFS_DBGMSG("%s: osd_start_request failed.\n", __func__);
 239                 return -ENOMEM;
 240         }
 241         ret = osd_req_read_kern(or, obj, offset, p, length);
 242         if (unlikely(ret)) {
 243                 EXOFS_DBGMSG("%s: osd_req_read_kern failed.\n", __func__);
 244                 goto out;
 245         }
 246
 247         ret = osd_finalize_request(or, 0, cred, NULL);
 248         if (unlikely(ret)) {
 249                 EXOFS_DBGMSG("Failed to osd_finalize_request() => %d\n", ret);
 250                 goto out;
 251         }
 252
 253         ret = osd_execute_request(or);
 254         if (unlikely(ret))
 255                 EXOFS_DBGMSG("osd_execute_request() => %d\n", ret);
 256         /* osd_req_decode_sense(or, ret); */
 257
 258 out:
 259         osd_end_request(or);
 260         EXOFS_DBGMSG2("read_kern(0x%llx) offset=0x%llx "
 261                       "length=0x%llx dev=%p ret=>%d\n",
 262                       _LLU(obj->id), _LLU(offset), _LLU(length), od, ret);
 263         return ret;
 264 }
 265
 266 static const struct osd_attr g_attr_sb_stats = ATTR_DEF(
 267         EXOFS_APAGE_SB_DATA,
 268         EXOFS_ATTR_SB_STATS,
 269         sizeof(struct exofs_sb_stats));
 270
 271 static int __sbi_read_stats(struct exofs_sb_info *sbi)
 272 {
 273         struct osd_attr attrs[] = {
 274                 [0] = g_attr_sb_stats,
 275         };
 276         struct ore_io_state *ios;
 277         int ret;
 278
 279         ret = ore_get_io_state(&sbi->layout, &sbi->oc, &ios);
 280         if (unlikely(ret)) {
 281                 EXOFS_ERR("%s: ore_get_io_state failed.\n", __func__);
 282                 return ret;
 283         }
 284
 285         ios->in_attr = attrs;
 286         ios->in_attr_len = ARRAY_SIZE(attrs);
 287
 288         ret = ore_read(ios);
 289         if (unlikely(ret)) {
 290                 EXOFS_ERR("Error reading super_block stats => %d\n", ret);
 291                 goto out;
 292         }
 293
 294         ret = extract_attr_from_ios(ios, &attrs[0]);
 295         if (ret) {
 296                 EXOFS_ERR("%s: extract_attr of sb_stats failed\n", __func__);
 297                 goto out;
 298         }
 299         if (attrs[0].len) {
 300                 struct exofs_sb_stats *ess;
 301
 302                 if (unlikely(attrs[0].len != sizeof(*ess))) {
 303                         EXOFS_ERR("%s: Wrong version of exofs_sb_stats "
 304                                   "size(%d) != expected(%zd)\n",
 305                                   __func__, attrs[0].len, sizeof(*ess));
 306                         goto out;
 307                 }
 308
 309                 ess = attrs[0].val_ptr;
 310                 sbi->s_nextid = le64_to_cpu(ess->s_nextid);
 311                 sbi->s_numfiles = le32_to_cpu(ess->s_numfiles);
 312         }
 313
 314 out:
 315         ore_put_io_state(ios);
 316         return ret;
 317 }
 318
 319 static void stats_done(struct ore_io_state *ios, void *p)
 320 {
 321         ore_put_io_state(ios);
 322         /* Good thanks nothing to do anymore */
 323 }
 324
 325 /* Asynchronously write the stats attribute */
 326 int exofs_sbi_write_stats(struct exofs_sb_info *sbi)
 327 {
 328         struct osd_attr attrs[] = {
 329                 [0] = g_attr_sb_stats,
 330         };
 331         struct ore_io_state *ios;
 332         int ret;
 333
 334         ret = ore_get_io_state(&sbi->layout, &sbi->oc, &ios);
 335         if (unlikely(ret)) {
 336                 EXOFS_ERR("%s: ore_get_io_state failed.\n", __func__);
 337                 return ret;
 338         }
 339
 340         sbi->s_ess.s_nextid   = cpu_to_le64(sbi->s_nextid);
 341         sbi->s_ess.s_numfiles = cpu_to_le64(sbi->s_numfiles);
 342         attrs[0].val_ptr = &sbi->s_ess;
 343
 344
 345         ios->done = stats_done;
 346         ios->private = sbi;
 347         ios->out_attr = attrs;
 348         ios->out_attr_len = ARRAY_SIZE(attrs);
 349
 350         ret = ore_write(ios);
 351         if (unlikely(ret)) {
 352                 EXOFS_ERR("%s: ore_write failed.\n", __func__);
 353                 ore_put_io_state(ios);
 354         }
 355
 356         return ret;
 357 }
 358
 359 /******************************************************************************
 360  * SUPERBLOCK FUNCTIONS
 361  *****************************************************************************/
 362 static const struct super_operations exofs_sops;
 363 static const struct export_operations exofs_export_ops;
 364
 365 /*
 366  * Write the superblock to the OSD
 367  */
 368 static int exofs_sync_fs(struct super_block *sb, int wait)
 369 {
 370         struct exofs_sb_info *sbi;
 371         struct exofs_fscb *fscb;
 372         struct ore_comp one_comp;
 373         struct ore_components oc;
 374         struct ore_io_state *ios;
 375         int ret = -ENOMEM;
 376
 377         fscb = kmalloc(sizeof(*fscb), GFP_KERNEL);
 378         if (unlikely(!fscb))
 379                 return -ENOMEM;
 380
 381         sbi = sb->s_fs_info;
 382
 383         /* NOTE: We no longer dirty the super_block anywhere in exofs. The
 384          * reason we write the fscb here on unmount is so we can stay backwards
 385          * compatible with fscb->s_version == 1. (What we are not compatible
 386          * with is if a new version FS crashed and then we try to mount an old
 387          * version). Otherwise the exofs_fscb is read-only from mkfs time. All
 388          * the writeable info is set in exofs_sbi_write_stats() above.
 389          */
 390
 391         exofs_init_comps(&oc, &one_comp, sbi, EXOFS_SUPER_ID);
 392
 393         ret = ore_get_io_state(&sbi->layout, &oc, &ios);
 394         if (unlikely(ret))
 395                 goto out;
 396
 397         ios->length = offsetof(struct exofs_fscb, s_dev_table_oid);
 398         memset(fscb, 0, ios->length);
 399         fscb->s_nextid = cpu_to_le64(sbi->s_nextid);
 400         fscb->s_numfiles = cpu_to_le64(sbi->s_numfiles);
 401         fscb->s_magic = cpu_to_le16(sb->s_magic);
 402         fscb->s_newfs = 0;
 403         fscb->s_version = EXOFS_FSCB_VER;
 404
 405         ios->offset = 0;
 406         ios->kern_buff = fscb;
 407
 408         ret = ore_write(ios);
 409         if (unlikely(ret))
 410                 EXOFS_ERR("%s: ore_write failed.\n", __func__);
 411
 412 out:
 413         EXOFS_DBGMSG("s_nextid=0x%llx ret=%d\n", _LLU(sbi->s_nextid), ret);
 414         ore_put_io_state(ios);
 415         kfree(fscb);
 416         return ret;
 417 }
 418
 419 static void _exofs_print_device(const char *msg, const char *dev_path,
 420                                 struct osd_dev *od, u64 pid)
 421 {
 422         const struct osd_dev_info *odi = osduld_device_info(od);
 423
 424         printk(KERN_NOTICE "exofs: %s %s osd_name-%s pid-0x%llx\n",
 425                 msg, dev_path ?: "", odi->osdname, _LLU(pid));
 426 }
 427
 428 static void exofs_free_sbi(struct exofs_sb_info *sbi)
 429 {
 430         unsigned numdevs = sbi->oc.numdevs;
 431
 432         while (numdevs) {
 433                 unsigned i = --numdevs;
 434                 struct osd_dev *od = ore_comp_dev(&sbi->oc, i);
 435
 436                 if (od) {
 437                         ore_comp_set_dev(&sbi->oc, i, NULL);
 438                         osduld_put_device(od);
 439                 }
 440         }
 441         kfree(sbi->oc.ods);
 442         kfree(sbi);
 443 }
 444
 445 /*
 446  * This function is called when the vfs is freeing the superblock.  We just
 447  * need to free our own part.
 448  */
 449 static void exofs_put_super(struct super_block *sb)
 450 {
 451         int num_pend;
 452         struct exofs_sb_info *sbi = sb->s_fs_info;
 453
 454         /* make sure there are no pending commands */
 455         for (num_pend = atomic_read(&sbi->s_curr_pending); num_pend > 0;
 456              num_pend = atomic_read(&sbi->s_curr_pending)) {
 457                 wait_queue_head_t wq;
 458
 459                 printk(KERN_NOTICE "%s: !!Pending operations in flight. "
 460                        "This is a BUG. please report to osd-dev@open-osd.org\n",
 461                        __func__);
 462                 init_waitqueue_head(&wq);
 463                 wait_event_timeout(wq,
 464                                   (atomic_read(&sbi->s_curr_pending) == 0),
 465                                   msecs_to_jiffies(100));
 466         }
 467
 468         _exofs_print_device("Unmounting", NULL, ore_comp_dev(&sbi->oc, 0),
 469                             sbi->one_comp.obj.partition);
 470
 471         exofs_sysfs_sb_del(sbi);
 472         exofs_free_sbi(sbi);
 473         sb->s_fs_info = NULL;
 474 }
 475
 476 static int _read_and_match_data_map(struct exofs_sb_info *sbi, unsigned numdevs,
 477                                     struct exofs_device_table *dt)
 478 {
 479         int ret;
 480
 481         sbi->layout.stripe_unit =
 482                                 le64_to_cpu(dt->dt_data_map.cb_stripe_unit);
 483         sbi->layout.group_width =
 484                                 le32_to_cpu(dt->dt_data_map.cb_group_width);
 485         sbi->layout.group_depth =
 486                                 le32_to_cpu(dt->dt_data_map.cb_group_depth);
 487         sbi->layout.mirrors_p1  =
 488                                 le32_to_cpu(dt->dt_data_map.cb_mirror_cnt) + 1;
 489         sbi->layout.raid_algorithm  =
 490                                 le32_to_cpu(dt->dt_data_map.cb_raid_algorithm);
 491
 492         ret = ore_verify_layout(numdevs, &sbi->layout);
 493
 494         EXOFS_DBGMSG("exofs: layout: "
 495                 "num_comps=%u stripe_unit=0x%x group_width=%u "
 496                 "group_depth=0x%llx mirrors_p1=%u raid_algorithm=%u\n",
 497                 numdevs,
 498                 sbi->layout.stripe_unit,
 499                 sbi->layout.group_width,
 500                 _LLU(sbi->layout.group_depth),
 501                 sbi->layout.mirrors_p1,
 502                 sbi->layout.raid_algorithm);
 503         return ret;
 504 }
 505
 506 static unsigned __ra_pages(struct ore_layout *layout)
 507 {
 508         const unsigned _MIN_RA = 32; /* min 128K read-ahead */
 509         unsigned ra_pages = layout->group_width * layout->stripe_unit /
 510                                 PAGE_SIZE;
 511         unsigned max_io_pages = exofs_max_io_pages(layout, ~0);
 512
 513         ra_pages *= 2; /* two stripes */
 514         if (ra_pages < _MIN_RA)
 515                 ra_pages = roundup(_MIN_RA, ra_pages / 2);
 516
 517         if (ra_pages > max_io_pages)
 518                 ra_pages = max_io_pages;
 519
 520         return ra_pages;
 521 }
 522
 523 /* @odi is valid only as long as @fscb_dev is valid */
 524 static int exofs_devs_2_odi(struct exofs_dt_device_info *dt_dev,
 525                              struct osd_dev_info *odi)
 526 {
 527         odi->systemid_len = le32_to_cpu(dt_dev->systemid_len);
 528         if (likely(odi->systemid_len))
 529                 memcpy(odi->systemid, dt_dev->systemid, OSD_SYSTEMID_LEN);
 530
 531         odi->osdname_len = le32_to_cpu(dt_dev->osdname_len);
 532         odi->osdname = dt_dev->osdname;
 533
 534         /* FIXME support long names. Will need a _put function */
 535         if (dt_dev->long_name_offset)
 536                 return -EINVAL;
 537
 538         /* Make sure osdname is printable!
 539          * mkexofs should give us space for a null-terminator else the
 540          * device-table is invalid.
 541          */
 542         if (unlikely(odi->osdname_len >= sizeof(dt_dev->osdname)))
 543                 odi->osdname_len = sizeof(dt_dev->osdname) - 1;
 544         dt_dev->osdname[odi->osdname_len] = 0;
 545
 546         /* If it's all zeros something is bad we read past end-of-obj */
 547         return !(odi->systemid_len || odi->osdname_len);
 548 }
 549
 550 static int __alloc_dev_table(struct exofs_sb_info *sbi, unsigned numdevs,
 551                       struct exofs_dev **peds)
 552 {
 553         /* Twice bigger table: See exofs_init_comps() and comment at
 554          * exofs_read_lookup_dev_table()
 555          */
 556         const size_t numores = numdevs * 2 - 1;
 557         struct exofs_dev *eds;
 558         unsigned i;
 559
 560         sbi->oc.ods = kzalloc(numores * sizeof(struct ore_dev *) +
 561                               numdevs * sizeof(struct exofs_dev), GFP_KERNEL);
 562         if (unlikely(!sbi->oc.ods)) {
 563                 EXOFS_ERR("ERROR: failed allocating Device array[%d]\n",
 564                           numdevs);
 565                 return -ENOMEM;
 566         }
 567
 568         /* Start of allocated struct exofs_dev entries */
 569         *peds = eds = (void *)sbi->oc.ods[numores];
 570         /* Initialize pointers into struct exofs_dev */
 571         for (i = 0; i < numdevs; ++i)
 572                 sbi->oc.ods[i] = &eds[i].ored;
 573         return 0;
 574 }
 575
 576 static int exofs_read_lookup_dev_table(struct exofs_sb_info *sbi,
 577                                        struct osd_dev *fscb_od,
 578                                        unsigned table_count)
 579 {
 580         struct ore_comp comp;
 581         struct exofs_device_table *dt;
 582         struct exofs_dev *eds;
 583         unsigned table_bytes = table_count * sizeof(dt->dt_dev_table[0]) +
 584                                              sizeof(*dt);
 585         unsigned numdevs, i;
 586         int ret;
 587
 588         dt = kmalloc(table_bytes, GFP_KERNEL);
 589         if (unlikely(!dt)) {
 590                 EXOFS_ERR("ERROR: allocating %x bytes for device table\n",
 591                           table_bytes);
 592                 return -ENOMEM;
 593         }
 594
 595         sbi->oc.numdevs = 0;
 596
 597         comp.obj.partition = sbi->one_comp.obj.partition;
 598         comp.obj.id = EXOFS_DEVTABLE_ID;
 599         exofs_make_credential(comp.cred, &comp.obj);
 600
 601         ret = exofs_read_kern(fscb_od, comp.cred, &comp.obj, 0, dt,
 602                               table_bytes);
 603         if (unlikely(ret)) {
 604                 EXOFS_ERR("ERROR: reading device table\n");
 605                 goto out;
 606         }
 607
 608         numdevs = le64_to_cpu(dt->dt_num_devices);
 609         if (unlikely(!numdevs)) {
 610                 ret = -EINVAL;
 611                 goto out;
 612         }
 613         WARN_ON(table_count != numdevs);
 614
 615         ret = _read_and_match_data_map(sbi, numdevs, dt);
 616         if (unlikely(ret))
 617                 goto out;
 618
 619         ret = __alloc_dev_table(sbi, numdevs, &eds);
 620         if (unlikely(ret))
 621                 goto out;
 622         /* exofs round-robins the device table view according to inode
 623          * number. We hold a: twice bigger table hence inodes can point
 624          * to any device and have a sequential view of the table
 625          * starting at this device. See exofs_init_comps()
 626          */
 627         memcpy(&sbi->oc.ods[numdevs], &sbi->oc.ods[0],
 628                 (numdevs - 1) * sizeof(sbi->oc.ods[0]));
 629
 630         /* create sysfs subdir under which we put the device table
 631          * And cluster layout. A Superblock is identified by the string:
 632          *      "dev[0].osdname"_"pid"
 633          */
 634         exofs_sysfs_sb_add(sbi, &dt->dt_dev_table[0]);
 635
 636         for (i = 0; i < numdevs; i++) {
 637                 struct exofs_fscb fscb;
 638                 struct osd_dev_info odi;
 639                 struct osd_dev *od;
 640
 641                 if (exofs_devs_2_odi(&dt->dt_dev_table[i], &odi)) {
 642                         EXOFS_ERR("ERROR: Read all-zeros device entry\n");
 643                         ret = -EINVAL;
 644                         goto out;
 645                 }
 646
 647                 printk(KERN_NOTICE "Add device[%d]: osd_name-%s\n",
 648                        i, odi.osdname);
 649
 650                 /* the exofs id is currently the table index */
 651                 eds[i].did = i;
 652
 653                 /* On all devices the device table is identical. The user can
 654                  * specify any one of the participating devices on the command
 655                  * line. We always keep them in device-table order.
 656                  */
 657                 if (fscb_od && osduld_device_same(fscb_od, &odi)) {
 658                         eds[i].ored.od = fscb_od;
 659                         ++sbi->oc.numdevs;
 660                         fscb_od = NULL;
 661                         exofs_sysfs_odev_add(&eds[i], sbi);
 662                         continue;
 663                 }
 664
 665                 od = osduld_info_lookup(&odi);
 666                 if (IS_ERR(od)) {
 667                         ret = PTR_ERR(od);
 668                         EXOFS_ERR("ERROR: device requested is not found "
 669                                   "osd_name-%s =>%d\n", odi.osdname, ret);
 670                         goto out;
 671                 }
 672
 673                 eds[i].ored.od = od;
 674                 ++sbi->oc.numdevs;
 675
 676                 /* Read the fscb of the other devices to make sure the FS
 677                  * partition is there.
 678                  */
 679                 ret = exofs_read_kern(od, comp.cred, &comp.obj, 0, &fscb,
 680                                       sizeof(fscb));
 681                 if (unlikely(ret)) {
 682                         EXOFS_ERR("ERROR: Malformed participating device "
 683                                   "error reading fscb osd_name-%s\n",
 684                                   odi.osdname);
 685                         goto out;
 686                 }
 687                 exofs_sysfs_odev_add(&eds[i], sbi);
 688
 689                 /* TODO: verify other information is correct and FS-uuid
 690                  *       matches. Benny what did you say about device table
 691                  *       generation and old devices?
 692                  */
 693         }
 694
 695 out:
 696         kfree(dt);
 697         if (unlikely(fscb_od && !ret)) {
 698                         EXOFS_ERR("ERROR: Bad device-table container device not present\n");
 699                         osduld_put_device(fscb_od);
 700                         return -EINVAL;
 701         }
 702         return ret;
 703 }
 704
 705 /*
 706  * Read the superblock from the OSD and fill in the fields
 707  */
 708 static int exofs_fill_super(struct super_block *sb, void *data, int silent)
 709 {
 710         struct inode *root;
 711         struct exofs_mountopt *opts = data;
 712         struct exofs_sb_info *sbi;      /*extended info                  */
 713         struct osd_dev *od;             /* Master device                 */
 714         struct exofs_fscb fscb;         /*on-disk superblock info        */
 715         struct ore_comp comp;
 716         unsigned table_count;
 717         int ret;
 718
 719         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
 720         if (!sbi)
 721                 return -ENOMEM;
 722
 723         /* use mount options to fill superblock */
 724         if (opts->is_osdname) {
 725                 struct osd_dev_info odi = {.systemid_len = 0};
 726
 727                 odi.osdname_len = strlen(opts->dev_name);
 728                 odi.osdname = (u8 *)opts->dev_name;
 729                 od = osduld_info_lookup(&odi);
 730                 kfree(opts->dev_name);
 731                 opts->dev_name = NULL;
 732         } else {
 733                 od = osduld_path_lookup(opts->dev_name);
 734         }
 735         if (IS_ERR(od)) {
 736                 ret = -EINVAL;
 737                 goto free_sbi;
 738         }
 739
 740         /* Default layout in case we do not have a device-table */
 741         sbi->layout.stripe_unit = PAGE_SIZE;
 742         sbi->layout.mirrors_p1 = 1;
 743         sbi->layout.group_width = 1;
 744         sbi->layout.group_depth = -1;
 745         sbi->layout.group_count = 1;
 746         sbi->s_timeout = opts->timeout;
 747
 748         sbi->one_comp.obj.partition = opts->pid;
 749         sbi->one_comp.obj.id = 0;
 750         exofs_make_credential(sbi->one_comp.cred, &sbi->one_comp.obj);
 751         sbi->oc.single_comp = EC_SINGLE_COMP;
 752         sbi->oc.comps = &sbi->one_comp;
 753
 754         /* fill in some other data by hand */
 755         memset(sb->s_id, 0, sizeof(sb->s_id));
 756         strcpy(sb->s_id, "exofs");
 757         sb->s_blocksize = EXOFS_BLKSIZE;
 758         sb->s_blocksize_bits = EXOFS_BLKSHIFT;
 759         sb->s_maxbytes = MAX_LFS_FILESIZE;
 760         sb->s_max_links = EXOFS_LINK_MAX;
 761         atomic_set(&sbi->s_curr_pending, 0);
 762         sb->s_bdev = NULL;
 763         sb->s_dev = 0;
 764
 765         comp.obj.partition = sbi->one_comp.obj.partition;
 766         comp.obj.id = EXOFS_SUPER_ID;
 767         exofs_make_credential(comp.cred, &comp.obj);
 768
 769         ret = exofs_read_kern(od, comp.cred, &comp.obj, 0, &fscb, sizeof(fscb));
 770         if (unlikely(ret))
 771                 goto free_sbi;
 772
 773         sb->s_magic = le16_to_cpu(fscb.s_magic);
 774         /* NOTE: we read below to be backward compatible with old versions */
 775         sbi->s_nextid = le64_to_cpu(fscb.s_nextid);
 776         sbi->s_numfiles = le32_to_cpu(fscb.s_numfiles);
 777
 778         /* make sure what we read from the object store is correct */
 779         if (sb->s_magic != EXOFS_SUPER_MAGIC) {
 780                 if (!silent)
 781                         EXOFS_ERR("ERROR: Bad magic value\n");
 782                 ret = -EINVAL;
 783                 goto free_sbi;
 784         }
 785         if (le32_to_cpu(fscb.s_version) > EXOFS_FSCB_VER) {
 786                 EXOFS_ERR("ERROR: Bad FSCB version expected-%d got-%d\n",
 787                           EXOFS_FSCB_VER, le32_to_cpu(fscb.s_version));
 788                 ret = -EINVAL;
 789                 goto free_sbi;
 790         }
 791
 792         /* start generation numbers from a random point */
 793         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
 794         spin_lock_init(&sbi->s_next_gen_lock);
 795
 796         table_count = le64_to_cpu(fscb.s_dev_table_count);
 797         if (table_count) {
 798                 ret = exofs_read_lookup_dev_table(sbi, od, table_count);
 799                 if (unlikely(ret))
 800                         goto free_sbi;
 801         } else {
 802                 struct exofs_dev *eds;
 803
 804                 ret = __alloc_dev_table(sbi, 1, &eds);
 805                 if (unlikely(ret))
 806                         goto free_sbi;
 807
 808                 ore_comp_set_dev(&sbi->oc, 0, od);
 809                 sbi->oc.numdevs = 1;
 810         }
 811
 812         __sbi_read_stats(sbi);
 813
 814         /* set up operation vectors */
 815         ret = super_setup_bdi(sb);
 816         if (ret) {
 817                 EXOFS_DBGMSG("Failed to super_setup_bdi\n");
 818                 goto free_sbi;
 819         }
 820         sb->s_bdi->ra_pages = __ra_pages(&sbi->layout);
 821         sb->s_fs_info = sbi;
 822         sb->s_op = &exofs_sops;
 823         sb->s_export_op = &exofs_export_ops;
 824         root = exofs_iget(sb, EXOFS_ROOT_ID - EXOFS_OBJ_OFF);
 825         if (IS_ERR(root)) {
 826                 EXOFS_ERR("ERROR: exofs_iget failed\n");
 827                 ret = PTR_ERR(root);
 828                 goto free_sbi;
 829         }
 830         sb->s_root = d_make_root(root);
 831         if (!sb->s_root) {
 832                 EXOFS_ERR("ERROR: get root inode failed\n");
 833                 ret = -ENOMEM;
 834                 goto free_sbi;
 835         }
 836
 837         if (!S_ISDIR(root->i_mode)) {
 838                 dput(sb->s_root);
 839                 sb->s_root = NULL;
 840                 EXOFS_ERR("ERROR: corrupt root inode (mode = %hd)\n",
 841                        root->i_mode);
 842                 ret = -EINVAL;
 843                 goto free_sbi;
 844         }
 845
 846         exofs_sysfs_dbg_print();
 847         _exofs_print_device("Mounting", opts->dev_name,
 848                             ore_comp_dev(&sbi->oc, 0),
 849                             sbi->one_comp.obj.partition);
 850         return 0;
 851
 852 free_sbi:
 853         EXOFS_ERR("Unable to mount exofs on %s pid=0x%llx err=%d\n",
 854                   opts->dev_name, sbi->one_comp.obj.partition, ret);
 855         exofs_free_sbi(sbi);
 856         return ret;
 857 }
 858
 859 /*
 860  * Set up the superblock (calls exofs_fill_super eventually)
 861  */
 862 static struct dentry *exofs_mount(struct file_system_type *type,
 863                           int flags, const char *dev_name,
 864                           void *data)
 865 {
 866         struct exofs_mountopt opts;
 867         int ret;
 868
 869         ret = parse_options(data, &opts);
 870         if (ret) {
 871                 kfree(opts.dev_name);
 872                 return ERR_PTR(ret);
 873         }
 874
 875         if (!opts.dev_name)
 876                 opts.dev_name = dev_name;
 877         return mount_nodev(type, flags, &opts, exofs_fill_super);
 878 }
 879
 880 /*
 881  * Return information about the file system state in the buffer.  This is used
 882  * by the 'df' command, for example.
 883  */
 884 static int exofs_statfs(struct dentry *dentry, struct kstatfs *buf)
 885 {
 886         struct super_block *sb = dentry->d_sb;
 887         struct exofs_sb_info *sbi = sb->s_fs_info;
 888         struct ore_io_state *ios;
 889         struct osd_attr attrs[] = {
 890                 ATTR_DEF(OSD_APAGE_PARTITION_QUOTAS,
 891                         OSD_ATTR_PQ_CAPACITY_QUOTA, sizeof(__be64)),
 892                 ATTR_DEF(OSD_APAGE_PARTITION_INFORMATION,
 893                         OSD_ATTR_PI_USED_CAPACITY, sizeof(__be64)),
 894         };
 895         uint64_t capacity = ULLONG_MAX;
 896         uint64_t used = ULLONG_MAX;
 897         int ret;
 898
 899         ret = ore_get_io_state(&sbi->layout, &sbi->oc, &ios);
 900         if (ret) {
 901                 EXOFS_DBGMSG("ore_get_io_state failed.\n");
 902                 return ret;
 903         }
 904
 905         ios->in_attr = attrs;
 906         ios->in_attr_len = ARRAY_SIZE(attrs);
 907
 908         ret = ore_read(ios);
 909         if (unlikely(ret))
 910                 goto out;
 911
 912         ret = extract_attr_from_ios(ios, &attrs[0]);
 913         if (likely(!ret)) {
 914                 capacity = get_unaligned_be64(attrs[0].val_ptr);
 915                 if (unlikely(!capacity))
 916                         capacity = ULLONG_MAX;
 917         } else
 918                 EXOFS_DBGMSG("exofs_statfs: get capacity failed.\n");
 919
 920         ret = extract_attr_from_ios(ios, &attrs[1]);
 921         if (likely(!ret))
 922                 used = get_unaligned_be64(attrs[1].val_ptr);
 923         else
 924                 EXOFS_DBGMSG("exofs_statfs: get used-space failed.\n");
 925
 926         /* fill in the stats buffer */
 927         buf->f_type = EXOFS_SUPER_MAGIC;
 928         buf->f_bsize = EXOFS_BLKSIZE;
 929         buf->f_blocks = capacity >> 9;
 930         buf->f_bfree = (capacity - used) >> 9;
 931         buf->f_bavail = buf->f_bfree;
 932         buf->f_files = sbi->s_numfiles;
 933         buf->f_ffree = EXOFS_MAX_ID - sbi->s_numfiles;
 934         buf->f_namelen = EXOFS_NAME_LEN;
 935
 936 out:
 937         ore_put_io_state(ios);
 938         return ret;
 939 }
 940
 941 static const struct super_operations exofs_sops = {
 942         .alloc_inode    = exofs_alloc_inode,
 943         .destroy_inode  = exofs_destroy_inode,
 944         .write_inode    = exofs_write_inode,
 945         .evict_inode    = exofs_evict_inode,
 946         .put_super      = exofs_put_super,
 947         .sync_fs        = exofs_sync_fs,
 948         .statfs         = exofs_statfs,
 949 };
 950
 951 /******************************************************************************
 952  * EXPORT OPERATIONS
 953  *****************************************************************************/
 954
 955 static struct dentry *exofs_get_parent(struct dentry *child)
 956 {
 957         unsigned long ino = exofs_parent_ino(child);
 958
 959         if (!ino)
 960                 return ERR_PTR(-ESTALE);
 961
 962         return d_obtain_alias(exofs_iget(child->d_sb, ino));
 963 }
 964
 965 static struct inode *exofs_nfs_get_inode(struct super_block *sb,
 966                 u64 ino, u32 generation)
 967 {
 968         struct inode *inode;
 969
 970         inode = exofs_iget(sb, ino);
 971         if (IS_ERR(inode))
 972                 return ERR_CAST(inode);
 973         if (generation && inode->i_generation != generation) {
 974                 /* we didn't find the right inode.. */
 975                 iput(inode);
 976                 return ERR_PTR(-ESTALE);
 977         }
 978         return inode;
 979 }
 980
 981 static struct dentry *exofs_fh_to_dentry(struct super_block *sb,
 982                                 struct fid *fid, int fh_len, int fh_type)
 983 {
 984         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 985                                     exofs_nfs_get_inode);
 986 }
 987
 988 static struct dentry *exofs_fh_to_parent(struct super_block *sb,
 989                                 struct fid *fid, int fh_len, int fh_type)
 990 {
 991         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 992                                     exofs_nfs_get_inode);
 993 }
 994
 995 static const struct export_operations exofs_export_ops = {
 996         .fh_to_dentry = exofs_fh_to_dentry,
 997         .fh_to_parent = exofs_fh_to_parent,
 998         .get_parent = exofs_get_parent,
 999 };
1000
1001 /******************************************************************************
1002  * INSMOD/RMMOD
1003  *****************************************************************************/
1004
1005 /*
1006  * struct that describes this file system
1007  */
1008 static struct file_system_type exofs_type = {
1009         .owner          = THIS_MODULE,
1010         .name           = "exofs",
1011         .mount          = exofs_mount,
1012         .kill_sb        = generic_shutdown_super,
1013 };
1014 MODULE_ALIAS_FS("exofs");
1015
1016 static int __init init_exofs(void)
1017 {
1018         int err;
1019
1020         err = init_inodecache();
1021         if (err)
1022                 goto out;
1023
1024         err = register_filesystem(&exofs_type);
1025         if (err)
1026                 goto out_d;
1027
1028         /* We don't fail if sysfs creation failed */
1029         exofs_sysfs_init();
1030
1031         return 0;
1032 out_d:
1033         destroy_inodecache();
1034 out:
1035         return err;
1036 }
1037
1038 static void __exit exit_exofs(void)
1039 {
1040         exofs_sysfs_uninit();
1041         unregister_filesystem(&exofs_type);
1042         destroy_inodecache();
1043 }
1044
1045 MODULE_AUTHOR("Avishay Traeger <avishay@gmail.com>");
1046 MODULE_DESCRIPTION("exofs");
1047 MODULE_LICENSE("GPL");
1048
1049 module_init(init_exofs)
1050 module_exit(exit_exofs)