fs/exofs/super.c

   1 /*
   2  * Copyright (C) 2005, 2006
   3  * Avishay Traeger (avishay@gmail.com)
   4  * Copyright (C) 2008, 2009
   5  * Boaz Harrosh <bharrosh@panasas.com>
   6  *
   7  * Copyrights for code taken from ext2:
   8  *     Copyright (C) 1992, 1993, 1994, 1995
   9  *     Remy Card (card@masi.ibp.fr)
  10  *     Laboratoire MASI - Institut Blaise Pascal
  11  *     Universite Pierre et Marie Curie (Paris VI)
  12  *     from
  13  *     linux/fs/minix/inode.c
  14  *     Copyright (C) 1991, 1992  Linus Torvalds
  15  *
  16  * This file is part of exofs.
  17  *
  18  * exofs is free software; you can redistribute it and/or modify
  19  * it under the terms of the GNU General Public License as published by
  20  * the Free Software Foundation.  Since it is based on ext2, and the only
  21  * valid version of GPL for the Linux kernel is version 2, the only valid
  22  * version of GPL for exofs is version 2.
  23  *
  24  * exofs is distributed in the hope that it will be useful,
  25  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  26  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  27  * GNU General Public License for more details.
  28  *
  29  * You should have received a copy of the GNU General Public License
  30  * along with exofs; if not, write to the Free Software
  31  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  32  */
  33
  34 #include <linux/string.h>
  35 #include <linux/parser.h>
  36 #include <linux/vfs.h>
  37 #include <linux/random.h>
  38 #include <linux/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                         opts->dev_name = match_strdup(&args[0]);
 104                         if (unlikely(!opts->dev_name)) {
 105                                 EXOFS_ERR("Error allocating dev_name");
 106                                 return -ENOMEM;
 107                         }
 108                         opts->is_osdname = true;
 109                         break;
 110                 case Opt_pid:
 111                         if (0 == match_strlcpy(str, &args[0], sizeof(str)))
 112                                 return -EINVAL;
 113                         opts->pid = simple_strtoull(str, NULL, 0);
 114                         if (opts->pid < EXOFS_MIN_PID) {
 115                                 EXOFS_ERR("Partition ID must be >= %u",
 116                                           EXOFS_MIN_PID);
 117                                 return -EINVAL;
 118                         }
 119                         s_pid = 1;
 120                         break;
 121                 case Opt_to:
 122                         if (match_int(&args[0], &option))
 123                                 return -EINVAL;
 124                         if (option <= 0) {
 125                                 EXOFS_ERR("Timout must be > 0");
 126                                 return -EINVAL;
 127                         }
 128                         opts->timeout = option * HZ;
 129                         break;
 130                 }
 131         }
 132
 133         if (!s_pid) {
 134                 EXOFS_ERR("Need to specify the following options:\n");
 135                 EXOFS_ERR("    -o pid=pid_no_to_use\n");
 136                 return -EINVAL;
 137         }
 138
 139         return 0;
 140 }
 141
 142 /******************************************************************************
 143  * INODE CACHE
 144  *****************************************************************************/
 145
 146 /*
 147  * Our inode cache.  Isn't it pretty?
 148  */
 149 static struct kmem_cache *exofs_inode_cachep;
 150
 151 /*
 152  * Allocate an inode in the cache
 153  */
 154 static struct inode *exofs_alloc_inode(struct super_block *sb)
 155 {
 156         struct exofs_i_info *oi;
 157
 158         oi = kmem_cache_alloc(exofs_inode_cachep, GFP_KERNEL);
 159         if (!oi)
 160                 return NULL;
 161
 162         oi->vfs_inode.i_version = 1;
 163         return &oi->vfs_inode;
 164 }
 165
 166 static void exofs_i_callback(struct rcu_head *head)
 167 {
 168         struct inode *inode = container_of(head, struct inode, i_rcu);
 169         kmem_cache_free(exofs_inode_cachep, exofs_i(inode));
 170 }
 171
 172 /*
 173  * Remove an inode from the cache
 174  */
 175 static void exofs_destroy_inode(struct inode *inode)
 176 {
 177         call_rcu(&inode->i_rcu, exofs_i_callback);
 178 }
 179
 180 /*
 181  * Initialize the inode
 182  */
 183 static void exofs_init_once(void *foo)
 184 {
 185         struct exofs_i_info *oi = foo;
 186
 187         inode_init_once(&oi->vfs_inode);
 188 }
 189
 190 /*
 191  * Create and initialize the inode cache
 192  */
 193 static int init_inodecache(void)
 194 {
 195         exofs_inode_cachep = kmem_cache_create("exofs_inode_cache",
 196                                 sizeof(struct exofs_i_info), 0,
 197                                 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
 198                                 exofs_init_once);
 199         if (exofs_inode_cachep == NULL)
 200                 return -ENOMEM;
 201         return 0;
 202 }
 203
 204 /*
 205  * Destroy the inode cache
 206  */
 207 static void destroy_inodecache(void)
 208 {
 209         kmem_cache_destroy(exofs_inode_cachep);
 210 }
 211
 212 /******************************************************************************
 213  * Some osd helpers
 214  *****************************************************************************/
 215 void exofs_make_credential(u8 cred_a[OSD_CAP_LEN], const struct osd_obj_id *obj)
 216 {
 217         osd_sec_init_nosec_doall_caps(cred_a, obj, false, true);
 218 }
 219
 220 static int exofs_read_kern(struct osd_dev *od, u8 *cred, struct osd_obj_id *obj,
 221                     u64 offset, void *p, unsigned length)
 222 {
 223         struct osd_request *or = osd_start_request(od, GFP_KERNEL);
 224 /*      struct osd_sense_info osi = {.key = 0};*/
 225         int ret;
 226
 227         if (unlikely(!or)) {
 228                 EXOFS_DBGMSG("%s: osd_start_request failed.\n", __func__);
 229                 return -ENOMEM;
 230         }
 231         ret = osd_req_read_kern(or, obj, offset, p, length);
 232         if (unlikely(ret)) {
 233                 EXOFS_DBGMSG("%s: osd_req_read_kern failed.\n", __func__);
 234                 goto out;
 235         }
 236
 237         ret = osd_finalize_request(or, 0, cred, NULL);
 238         if (unlikely(ret)) {
 239                 EXOFS_DBGMSG("Failed to osd_finalize_request() => %d\n", ret);
 240                 goto out;
 241         }
 242
 243         ret = osd_execute_request(or);
 244         if (unlikely(ret))
 245                 EXOFS_DBGMSG("osd_execute_request() => %d\n", ret);
 246         /* osd_req_decode_sense(or, ret); */
 247
 248 out:
 249         osd_end_request(or);
 250         EXOFS_DBGMSG2("read_kern(0x%llx) offset=0x%llx "
 251                       "length=0x%llx dev=%p ret=>%d\n",
 252                       _LLU(obj->id), _LLU(offset), _LLU(length), od, ret);
 253         return ret;
 254 }
 255
 256 static const struct osd_attr g_attr_sb_stats = ATTR_DEF(
 257         EXOFS_APAGE_SB_DATA,
 258         EXOFS_ATTR_SB_STATS,
 259         sizeof(struct exofs_sb_stats));
 260
 261 static int __sbi_read_stats(struct exofs_sb_info *sbi)
 262 {
 263         struct osd_attr attrs[] = {
 264                 [0] = g_attr_sb_stats,
 265         };
 266         struct ore_io_state *ios;
 267         int ret;
 268
 269         ret = ore_get_io_state(&sbi->layout, &sbi->oc, &ios);
 270         if (unlikely(ret)) {
 271                 EXOFS_ERR("%s: ore_get_io_state failed.\n", __func__);
 272                 return ret;
 273         }
 274
 275         ios->in_attr = attrs;
 276         ios->in_attr_len = ARRAY_SIZE(attrs);
 277
 278         ret = ore_read(ios);
 279         if (unlikely(ret)) {
 280                 EXOFS_ERR("Error reading super_block stats => %d\n", ret);
 281                 goto out;
 282         }
 283
 284         ret = extract_attr_from_ios(ios, &attrs[0]);
 285         if (ret) {
 286                 EXOFS_ERR("%s: extract_attr of sb_stats failed\n", __func__);
 287                 goto out;
 288         }
 289         if (attrs[0].len) {
 290                 struct exofs_sb_stats *ess;
 291
 292                 if (unlikely(attrs[0].len != sizeof(*ess))) {
 293                         EXOFS_ERR("%s: Wrong version of exofs_sb_stats "
 294                                   "size(%d) != expected(%zd)\n",
 295                                   __func__, attrs[0].len, sizeof(*ess));
 296                         goto out;
 297                 }
 298
 299                 ess = attrs[0].val_ptr;
 300                 sbi->s_nextid = le64_to_cpu(ess->s_nextid);
 301                 sbi->s_numfiles = le32_to_cpu(ess->s_numfiles);
 302         }
 303
 304 out:
 305         ore_put_io_state(ios);
 306         return ret;
 307 }
 308
 309 static void stats_done(struct ore_io_state *ios, void *p)
 310 {
 311         ore_put_io_state(ios);
 312         /* Good thanks nothing to do anymore */
 313 }
 314
 315 /* Asynchronously write the stats attribute */
 316 int exofs_sbi_write_stats(struct exofs_sb_info *sbi)
 317 {
 318         struct osd_attr attrs[] = {
 319                 [0] = g_attr_sb_stats,
 320         };
 321         struct ore_io_state *ios;
 322         int ret;
 323
 324         ret = ore_get_io_state(&sbi->layout, &sbi->oc, &ios);
 325         if (unlikely(ret)) {
 326                 EXOFS_ERR("%s: ore_get_io_state failed.\n", __func__);
 327                 return ret;
 328         }
 329
 330         sbi->s_ess.s_nextid   = cpu_to_le64(sbi->s_nextid);
 331         sbi->s_ess.s_numfiles = cpu_to_le64(sbi->s_numfiles);
 332         attrs[0].val_ptr = &sbi->s_ess;
 333
 334
 335         ios->done = stats_done;
 336         ios->private = sbi;
 337         ios->out_attr = attrs;
 338         ios->out_attr_len = ARRAY_SIZE(attrs);
 339
 340         ret = ore_write(ios);
 341         if (unlikely(ret)) {
 342                 EXOFS_ERR("%s: ore_write failed.\n", __func__);
 343                 ore_put_io_state(ios);
 344         }
 345
 346         return ret;
 347 }
 348
 349 /******************************************************************************
 350  * SUPERBLOCK FUNCTIONS
 351  *****************************************************************************/
 352 static const struct super_operations exofs_sops;
 353 static const struct export_operations exofs_export_ops;
 354
 355 /*
 356  * Write the superblock to the OSD
 357  */
 358 static int exofs_sync_fs(struct super_block *sb, int wait)
 359 {
 360         struct exofs_sb_info *sbi;
 361         struct exofs_fscb *fscb;
 362         struct ore_comp one_comp;
 363         struct ore_components oc;
 364         struct ore_io_state *ios;
 365         int ret = -ENOMEM;
 366
 367         fscb = kmalloc(sizeof(*fscb), GFP_KERNEL);
 368         if (unlikely(!fscb))
 369                 return -ENOMEM;
 370
 371         sbi = sb->s_fs_info;
 372
 373         /* NOTE: We no longer dirty the super_block anywhere in exofs. The
 374          * reason we write the fscb here on unmount is so we can stay backwards
 375          * compatible with fscb->s_version == 1. (What we are not compatible
 376          * with is if a new version FS crashed and then we try to mount an old
 377          * version). Otherwise the exofs_fscb is read-only from mkfs time. All
 378          * the writeable info is set in exofs_sbi_write_stats() above.
 379          */
 380
 381         exofs_init_comps(&oc, &one_comp, sbi, EXOFS_SUPER_ID);
 382
 383         ret = ore_get_io_state(&sbi->layout, &oc, &ios);
 384         if (unlikely(ret))
 385                 goto out;
 386
 387         lock_super(sb);
 388
 389         ios->length = offsetof(struct exofs_fscb, s_dev_table_oid);
 390         memset(fscb, 0, ios->length);
 391         fscb->s_nextid = cpu_to_le64(sbi->s_nextid);
 392         fscb->s_numfiles = cpu_to_le64(sbi->s_numfiles);
 393         fscb->s_magic = cpu_to_le16(sb->s_magic);
 394         fscb->s_newfs = 0;
 395         fscb->s_version = EXOFS_FSCB_VER;
 396
 397         ios->offset = 0;
 398         ios->kern_buff = fscb;
 399
 400         ret = ore_write(ios);
 401         if (unlikely(ret))
 402                 EXOFS_ERR("%s: ore_write failed.\n", __func__);
 403         else
 404                 sb->s_dirt = 0;
 405
 406
 407         unlock_super(sb);
 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_write_super(struct super_block *sb)
 416 {
 417         if (!(sb->s_flags & MS_RDONLY))
 418                 exofs_sync_fs(sb, 1);
 419         else
 420                 sb->s_dirt = 0;
 421 }
 422
 423 static void _exofs_print_device(const char *msg, const char *dev_path,
 424                                 struct osd_dev *od, u64 pid)
 425 {
 426         const struct osd_dev_info *odi = osduld_device_info(od);
 427
 428         printk(KERN_NOTICE "exofs: %s %s osd_name-%s pid-0x%llx\n",
 429                 msg, dev_path ?: "", odi->osdname, _LLU(pid));
 430 }
 431
 432 static void exofs_free_sbi(struct exofs_sb_info *sbi)
 433 {
 434         unsigned numdevs = sbi->oc.numdevs;
 435
 436         while (numdevs) {
 437                 unsigned i = --numdevs;
 438                 struct osd_dev *od = ore_comp_dev(&sbi->oc, i);
 439
 440                 if (od) {
 441                         ore_comp_set_dev(&sbi->oc, i, NULL);
 442                         osduld_put_device(od);
 443                 }
 444         }
 445         kfree(sbi->oc.ods);
 446         kfree(sbi);
 447 }
 448
 449 /*
 450  * This function is called when the vfs is freeing the superblock.  We just
 451  * need to free our own part.
 452  */
 453 static void exofs_put_super(struct super_block *sb)
 454 {
 455         int num_pend;
 456         struct exofs_sb_info *sbi = sb->s_fs_info;
 457
 458         /* make sure there are no pending commands */
 459         for (num_pend = atomic_read(&sbi->s_curr_pending); num_pend > 0;
 460              num_pend = atomic_read(&sbi->s_curr_pending)) {
 461                 wait_queue_head_t wq;
 462
 463                 printk(KERN_NOTICE "%s: !!Pending operations in flight. "
 464                        "This is a BUG. please report to osd-dev@open-osd.org\n",
 465                        __func__);
 466                 init_waitqueue_head(&wq);
 467                 wait_event_timeout(wq,
 468                                   (atomic_read(&sbi->s_curr_pending) == 0),
 469                                   msecs_to_jiffies(100));
 470         }
 471
 472         _exofs_print_device("Unmounting", NULL, ore_comp_dev(&sbi->oc, 0),
 473                             sbi->one_comp.obj.partition);
 474
 475         exofs_sysfs_sb_del(sbi);
 476         bdi_destroy(&sbi->bdi);
 477         exofs_free_sbi(sbi);
 478         sb->s_fs_info = NULL;
 479 }
 480
 481 static int _read_and_match_data_map(struct exofs_sb_info *sbi, unsigned numdevs,
 482                                     struct exofs_device_table *dt)
 483 {
 484         int ret;
 485
 486         sbi->layout.stripe_unit =
 487                                 le64_to_cpu(dt->dt_data_map.cb_stripe_unit);
 488         sbi->layout.group_width =
 489                                 le32_to_cpu(dt->dt_data_map.cb_group_width);
 490         sbi->layout.group_depth =
 491                                 le32_to_cpu(dt->dt_data_map.cb_group_depth);
 492         sbi->layout.mirrors_p1  =
 493                                 le32_to_cpu(dt->dt_data_map.cb_mirror_cnt) + 1;
 494         sbi->layout.raid_algorithm  =
 495                                 le32_to_cpu(dt->dt_data_map.cb_raid_algorithm);
 496
 497         ret = ore_verify_layout(numdevs, &sbi->layout);
 498
 499         EXOFS_DBGMSG("exofs: layout: "
 500                 "num_comps=%u stripe_unit=0x%x group_width=%u "
 501                 "group_depth=0x%llx mirrors_p1=%u raid_algorithm=%u\n",
 502                 numdevs,
 503                 sbi->layout.stripe_unit,
 504                 sbi->layout.group_width,
 505                 _LLU(sbi->layout.group_depth),
 506                 sbi->layout.mirrors_p1,
 507                 sbi->layout.raid_algorithm);
 508         return ret;
 509 }
 510
 511 static unsigned __ra_pages(struct ore_layout *layout)
 512 {
 513         const unsigned _MIN_RA = 32; /* min 128K read-ahead */
 514         unsigned ra_pages = layout->group_width * layout->stripe_unit /
 515                                 PAGE_SIZE;
 516         unsigned max_io_pages = exofs_max_io_pages(layout, ~0);
 517
 518         ra_pages *= 2; /* two stripes */
 519         if (ra_pages < _MIN_RA)
 520                 ra_pages = roundup(_MIN_RA, ra_pages / 2);
 521
 522         if (ra_pages > max_io_pages)
 523                 ra_pages = max_io_pages;
 524
 525         return ra_pages;
 526 }
 527
 528 /* @odi is valid only as long as @fscb_dev is valid */
 529 static int exofs_devs_2_odi(struct exofs_dt_device_info *dt_dev,
 530                              struct osd_dev_info *odi)
 531 {
 532         odi->systemid_len = le32_to_cpu(dt_dev->systemid_len);
 533         if (likely(odi->systemid_len))
 534                 memcpy(odi->systemid, dt_dev->systemid, OSD_SYSTEMID_LEN);
 535
 536         odi->osdname_len = le32_to_cpu(dt_dev->osdname_len);
 537         odi->osdname = dt_dev->osdname;
 538
 539         /* FIXME support long names. Will need a _put function */
 540         if (dt_dev->long_name_offset)
 541                 return -EINVAL;
 542
 543         /* Make sure osdname is printable!
 544          * mkexofs should give us space for a null-terminator else the
 545          * device-table is invalid.
 546          */
 547         if (unlikely(odi->osdname_len >= sizeof(dt_dev->osdname)))
 548                 odi->osdname_len = sizeof(dt_dev->osdname) - 1;
 549         dt_dev->osdname[odi->osdname_len] = 0;
 550
 551         /* If it's all zeros something is bad we read past end-of-obj */
 552         return !(odi->systemid_len || odi->osdname_len);
 553 }
 554
 555 int __alloc_dev_table(struct exofs_sb_info *sbi, unsigned numdevs,
 556                       struct exofs_dev **peds)
 557 {
 558         struct __alloc_ore_devs_and_exofs_devs {
 559                 /* Twice bigger table: See exofs_init_comps() and comment at
 560                  * exofs_read_lookup_dev_table()
 561                  */
 562                 struct ore_dev *oreds[numdevs * 2 - 1];
 563                 struct exofs_dev eds[numdevs];
 564         } *aoded;
 565         struct exofs_dev *eds;
 566         unsigned i;
 567
 568         aoded = kzalloc(sizeof(*aoded), GFP_KERNEL);
 569         if (unlikely(!aoded)) {
 570                 EXOFS_ERR("ERROR: failed allocating Device array[%d]\n",
 571                           numdevs);
 572                 return -ENOMEM;
 573         }
 574
 575         sbi->oc.ods = aoded->oreds;
 576         *peds = eds = aoded->eds;
 577         for (i = 0; i < numdevs; ++i)
 578                 aoded->oreds[i] = &eds[i].ored;
 579         return 0;
 580 }
 581
 582 static int exofs_read_lookup_dev_table(struct exofs_sb_info *sbi,
 583                                        struct osd_dev *fscb_od,
 584                                        unsigned table_count)
 585 {
 586         struct ore_comp comp;
 587         struct exofs_device_table *dt;
 588         struct exofs_dev *eds;
 589         unsigned table_bytes = table_count * sizeof(dt->dt_dev_table[0]) +
 590                                              sizeof(*dt);
 591         unsigned numdevs, i;
 592         int ret;
 593
 594         dt = kmalloc(table_bytes, GFP_KERNEL);
 595         if (unlikely(!dt)) {
 596                 EXOFS_ERR("ERROR: allocating %x bytes for device table\n",
 597                           table_bytes);
 598                 return -ENOMEM;
 599         }
 600
 601         sbi->oc.numdevs = 0;
 602
 603         comp.obj.partition = sbi->one_comp.obj.partition;
 604         comp.obj.id = EXOFS_DEVTABLE_ID;
 605         exofs_make_credential(comp.cred, &comp.obj);
 606
 607         ret = exofs_read_kern(fscb_od, comp.cred, &comp.obj, 0, dt,
 608                               table_bytes);
 609         if (unlikely(ret)) {
 610                 EXOFS_ERR("ERROR: reading device table\n");
 611                 goto out;
 612         }
 613
 614         numdevs = le64_to_cpu(dt->dt_num_devices);
 615         if (unlikely(!numdevs)) {
 616                 ret = -EINVAL;
 617                 goto out;
 618         }
 619         WARN_ON(table_count != numdevs);
 620
 621         ret = _read_and_match_data_map(sbi, numdevs, dt);
 622         if (unlikely(ret))
 623                 goto out;
 624
 625         ret = __alloc_dev_table(sbi, numdevs, &eds);
 626         if (unlikely(ret))
 627                 goto out;
 628         /* exofs round-robins the device table view according to inode
 629          * number. We hold a: twice bigger table hence inodes can point
 630          * to any device and have a sequential view of the table
 631          * starting at this device. See exofs_init_comps()
 632          */
 633         memcpy(&sbi->oc.ods[numdevs], &sbi->oc.ods[0],
 634                 (numdevs - 1) * sizeof(sbi->oc.ods[0]));
 635
 636         /* create sysfs subdir under which we put the device table
 637          * And cluster layout. A Superblock is identified by the string:
 638          *      "dev[0].osdname"_"pid"
 639          */
 640         exofs_sysfs_sb_add(sbi, &dt->dt_dev_table[0]);
 641
 642         for (i = 0; i < numdevs; i++) {
 643                 struct exofs_fscb fscb;
 644                 struct osd_dev_info odi;
 645                 struct osd_dev *od;
 646
 647                 if (exofs_devs_2_odi(&dt->dt_dev_table[i], &odi)) {
 648                         EXOFS_ERR("ERROR: Read all-zeros device entry\n");
 649                         ret = -EINVAL;
 650                         goto out;
 651                 }
 652
 653                 printk(KERN_NOTICE "Add device[%d]: osd_name-%s\n",
 654                        i, odi.osdname);
 655
 656                 /* the exofs id is currently the table index */
 657                 eds[i].did = i;
 658
 659                 /* On all devices the device table is identical. The user can
 660                  * specify any one of the participating devices on the command
 661                  * line. We always keep them in device-table order.
 662                  */
 663                 if (fscb_od && osduld_device_same(fscb_od, &odi)) {
 664                         eds[i].ored.od = fscb_od;
 665                         ++sbi->oc.numdevs;
 666                         fscb_od = NULL;
 667                         exofs_sysfs_odev_add(&eds[i], sbi);
 668                         continue;
 669                 }
 670
 671                 od = osduld_info_lookup(&odi);
 672                 if (IS_ERR(od)) {
 673                         ret = PTR_ERR(od);
 674                         EXOFS_ERR("ERROR: device requested is not found "
 675                                   "osd_name-%s =>%d\n", odi.osdname, ret);
 676                         goto out;
 677                 }
 678
 679                 eds[i].ored.od = od;
 680                 ++sbi->oc.numdevs;
 681
 682                 /* Read the fscb of the other devices to make sure the FS
 683                  * partition is there.
 684                  */
 685                 ret = exofs_read_kern(od, comp.cred, &comp.obj, 0, &fscb,
 686                                       sizeof(fscb));
 687                 if (unlikely(ret)) {
 688                         EXOFS_ERR("ERROR: Malformed participating device "
 689                                   "error reading fscb osd_name-%s\n",
 690                                   odi.osdname);
 691                         goto out;
 692                 }
 693                 exofs_sysfs_odev_add(&eds[i], sbi);
 694
 695                 /* TODO: verify other information is correct and FS-uuid
 696                  *       matches. Benny what did you say about device table
 697                  *       generation and old devices?
 698                  */
 699         }
 700
 701 out:
 702         kfree(dt);
 703         if (unlikely(fscb_od && !ret)) {
 704                         EXOFS_ERR("ERROR: Bad device-table container device not present\n");
 705                         osduld_put_device(fscb_od);
 706                         return -EINVAL;
 707         }
 708         return ret;
 709 }
 710
 711 /*
 712  * Read the superblock from the OSD and fill in the fields
 713  */
 714 static int exofs_fill_super(struct super_block *sb, void *data, int silent)
 715 {
 716         struct inode *root;
 717         struct exofs_mountopt *opts = data;
 718         struct exofs_sb_info *sbi;      /*extended info                  */
 719         struct osd_dev *od;             /* Master device                 */
 720         struct exofs_fscb fscb;         /*on-disk superblock info        */
 721         struct ore_comp comp;
 722         unsigned table_count;
 723         int ret;
 724
 725         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
 726         if (!sbi)
 727                 return -ENOMEM;
 728
 729         /* use mount options to fill superblock */
 730         if (opts->is_osdname) {
 731                 struct osd_dev_info odi = {.systemid_len = 0};
 732
 733                 odi.osdname_len = strlen(opts->dev_name);
 734                 odi.osdname = (u8 *)opts->dev_name;
 735                 od = osduld_info_lookup(&odi);
 736                 kfree(opts->dev_name);
 737                 opts->dev_name = NULL;
 738         } else {
 739                 od = osduld_path_lookup(opts->dev_name);
 740         }
 741         if (IS_ERR(od)) {
 742                 ret = -EINVAL;
 743                 goto free_sbi;
 744         }
 745
 746         /* Default layout in case we do not have a device-table */
 747         sbi->layout.stripe_unit = PAGE_SIZE;
 748         sbi->layout.mirrors_p1 = 1;
 749         sbi->layout.group_width = 1;
 750         sbi->layout.group_depth = -1;
 751         sbi->layout.group_count = 1;
 752         sbi->s_timeout = opts->timeout;
 753
 754         sbi->one_comp.obj.partition = opts->pid;
 755         sbi->one_comp.obj.id = 0;
 756         exofs_make_credential(sbi->one_comp.cred, &sbi->one_comp.obj);
 757         sbi->oc.single_comp = EC_SINGLE_COMP;
 758         sbi->oc.comps = &sbi->one_comp;
 759
 760         /* fill in some other data by hand */
 761         memset(sb->s_id, 0, sizeof(sb->s_id));
 762         strcpy(sb->s_id, "exofs");
 763         sb->s_blocksize = EXOFS_BLKSIZE;
 764         sb->s_blocksize_bits = EXOFS_BLKSHIFT;
 765         sb->s_maxbytes = MAX_LFS_FILESIZE;
 766         sb->s_max_links = EXOFS_LINK_MAX;
 767         atomic_set(&sbi->s_curr_pending, 0);
 768         sb->s_bdev = NULL;
 769         sb->s_dev = 0;
 770
 771         comp.obj.partition = sbi->one_comp.obj.partition;
 772         comp.obj.id = EXOFS_SUPER_ID;
 773         exofs_make_credential(comp.cred, &comp.obj);
 774
 775         ret = exofs_read_kern(od, comp.cred, &comp.obj, 0, &fscb, sizeof(fscb));
 776         if (unlikely(ret))
 777                 goto free_sbi;
 778
 779         sb->s_magic = le16_to_cpu(fscb.s_magic);
 780         /* NOTE: we read below to be backward compatible with old versions */
 781         sbi->s_nextid = le64_to_cpu(fscb.s_nextid);
 782         sbi->s_numfiles = le32_to_cpu(fscb.s_numfiles);
 783
 784         /* make sure what we read from the object store is correct */
 785         if (sb->s_magic != EXOFS_SUPER_MAGIC) {
 786                 if (!silent)
 787                         EXOFS_ERR("ERROR: Bad magic value\n");
 788                 ret = -EINVAL;
 789                 goto free_sbi;
 790         }
 791         if (le32_to_cpu(fscb.s_version) > EXOFS_FSCB_VER) {
 792                 EXOFS_ERR("ERROR: Bad FSCB version expected-%d got-%d\n",
 793                           EXOFS_FSCB_VER, le32_to_cpu(fscb.s_version));
 794                 ret = -EINVAL;
 795                 goto free_sbi;
 796         }
 797
 798         /* start generation numbers from a random point */
 799         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
 800         spin_lock_init(&sbi->s_next_gen_lock);
 801
 802         table_count = le64_to_cpu(fscb.s_dev_table_count);
 803         if (table_count) {
 804                 ret = exofs_read_lookup_dev_table(sbi, od, table_count);
 805                 if (unlikely(ret))
 806                         goto free_sbi;
 807         } else {
 808                 struct exofs_dev *eds;
 809
 810                 ret = __alloc_dev_table(sbi, 1, &eds);
 811                 if (unlikely(ret))
 812                         goto free_sbi;
 813
 814                 ore_comp_set_dev(&sbi->oc, 0, od);
 815                 sbi->oc.numdevs = 1;
 816         }
 817
 818         __sbi_read_stats(sbi);
 819
 820         /* set up operation vectors */
 821         sbi->bdi.ra_pages = __ra_pages(&sbi->layout);
 822         sb->s_bdi = &sbi->bdi;
 823         sb->s_fs_info = sbi;
 824         sb->s_op = &exofs_sops;
 825         sb->s_export_op = &exofs_export_ops;
 826         root = exofs_iget(sb, EXOFS_ROOT_ID - EXOFS_OBJ_OFF);
 827         if (IS_ERR(root)) {
 828                 EXOFS_ERR("ERROR: exofs_iget failed\n");
 829                 ret = PTR_ERR(root);
 830                 goto free_sbi;
 831         }
 832         sb->s_root = d_make_root(root);
 833         if (!sb->s_root) {
 834                 EXOFS_ERR("ERROR: get root inode failed\n");
 835                 ret = -ENOMEM;
 836                 goto free_sbi;
 837         }
 838
 839         if (!S_ISDIR(root->i_mode)) {
 840                 dput(sb->s_root);
 841                 sb->s_root = NULL;
 842                 EXOFS_ERR("ERROR: corrupt root inode (mode = %hd)\n",
 843                        root->i_mode);
 844                 ret = -EINVAL;
 845                 goto free_sbi;
 846         }
 847
 848         ret = bdi_setup_and_register(&sbi->bdi, "exofs", BDI_CAP_MAP_COPY);
 849         if (ret) {
 850                 EXOFS_DBGMSG("Failed to bdi_setup_and_register\n");
 851                 dput(sb->s_root);
 852                 sb->s_root = NULL;
 853                 goto free_sbi;
 854         }
 855
 856         exofs_sysfs_dbg_print();
 857         _exofs_print_device("Mounting", opts->dev_name,
 858                             ore_comp_dev(&sbi->oc, 0),
 859                             sbi->one_comp.obj.partition);
 860         return 0;
 861
 862 free_sbi:
 863         EXOFS_ERR("Unable to mount exofs on %s pid=0x%llx err=%d\n",
 864                   opts->dev_name, sbi->one_comp.obj.partition, ret);
 865         exofs_free_sbi(sbi);
 866         return ret;
 867 }
 868
 869 /*
 870  * Set up the superblock (calls exofs_fill_super eventually)
 871  */
 872 static struct dentry *exofs_mount(struct file_system_type *type,
 873                           int flags, const char *dev_name,
 874                           void *data)
 875 {
 876         struct exofs_mountopt opts;
 877         int ret;
 878
 879         ret = parse_options(data, &opts);
 880         if (ret)
 881                 return ERR_PTR(ret);
 882
 883         if (!opts.dev_name)
 884                 opts.dev_name = dev_name;
 885         return mount_nodev(type, flags, &opts, exofs_fill_super);
 886 }
 887
 888 /*
 889  * Return information about the file system state in the buffer.  This is used
 890  * by the 'df' command, for example.
 891  */
 892 static int exofs_statfs(struct dentry *dentry, struct kstatfs *buf)
 893 {
 894         struct super_block *sb = dentry->d_sb;
 895         struct exofs_sb_info *sbi = sb->s_fs_info;
 896         struct ore_io_state *ios;
 897         struct osd_attr attrs[] = {
 898                 ATTR_DEF(OSD_APAGE_PARTITION_QUOTAS,
 899                         OSD_ATTR_PQ_CAPACITY_QUOTA, sizeof(__be64)),
 900                 ATTR_DEF(OSD_APAGE_PARTITION_INFORMATION,
 901                         OSD_ATTR_PI_USED_CAPACITY, sizeof(__be64)),
 902         };
 903         uint64_t capacity = ULLONG_MAX;
 904         uint64_t used = ULLONG_MAX;
 905         int ret;
 906
 907         ret = ore_get_io_state(&sbi->layout, &sbi->oc, &ios);
 908         if (ret) {
 909                 EXOFS_DBGMSG("ore_get_io_state failed.\n");
 910                 return ret;
 911         }
 912
 913         ios->in_attr = attrs;
 914         ios->in_attr_len = ARRAY_SIZE(attrs);
 915
 916         ret = ore_read(ios);
 917         if (unlikely(ret))
 918                 goto out;
 919
 920         ret = extract_attr_from_ios(ios, &attrs[0]);
 921         if (likely(!ret)) {
 922                 capacity = get_unaligned_be64(attrs[0].val_ptr);
 923                 if (unlikely(!capacity))
 924                         capacity = ULLONG_MAX;
 925         } else
 926                 EXOFS_DBGMSG("exofs_statfs: get capacity failed.\n");
 927
 928         ret = extract_attr_from_ios(ios, &attrs[1]);
 929         if (likely(!ret))
 930                 used = get_unaligned_be64(attrs[1].val_ptr);
 931         else
 932                 EXOFS_DBGMSG("exofs_statfs: get used-space failed.\n");
 933
 934         /* fill in the stats buffer */
 935         buf->f_type = EXOFS_SUPER_MAGIC;
 936         buf->f_bsize = EXOFS_BLKSIZE;
 937         buf->f_blocks = capacity >> 9;
 938         buf->f_bfree = (capacity - used) >> 9;
 939         buf->f_bavail = buf->f_bfree;
 940         buf->f_files = sbi->s_numfiles;
 941         buf->f_ffree = EXOFS_MAX_ID - sbi->s_numfiles;
 942         buf->f_namelen = EXOFS_NAME_LEN;
 943
 944 out:
 945         ore_put_io_state(ios);
 946         return ret;
 947 }
 948
 949 static const struct super_operations exofs_sops = {
 950         .alloc_inode    = exofs_alloc_inode,
 951         .destroy_inode  = exofs_destroy_inode,
 952         .write_inode    = exofs_write_inode,
 953         .evict_inode    = exofs_evict_inode,
 954         .put_super      = exofs_put_super,
 955         .write_super    = exofs_write_super,
 956         .sync_fs        = exofs_sync_fs,
 957         .statfs         = exofs_statfs,
 958 };
 959
 960 /******************************************************************************
 961  * EXPORT OPERATIONS
 962  *****************************************************************************/
 963
 964 static struct dentry *exofs_get_parent(struct dentry *child)
 965 {
 966         unsigned long ino = exofs_parent_ino(child);
 967
 968         if (!ino)
 969                 return ERR_PTR(-ESTALE);
 970
 971         return d_obtain_alias(exofs_iget(child->d_inode->i_sb, ino));
 972 }
 973
 974 static struct inode *exofs_nfs_get_inode(struct super_block *sb,
 975                 u64 ino, u32 generation)
 976 {
 977         struct inode *inode;
 978
 979         inode = exofs_iget(sb, ino);
 980         if (IS_ERR(inode))
 981                 return ERR_CAST(inode);
 982         if (generation && inode->i_generation != generation) {
 983                 /* we didn't find the right inode.. */
 984                 iput(inode);
 985                 return ERR_PTR(-ESTALE);
 986         }
 987         return inode;
 988 }
 989
 990 static struct dentry *exofs_fh_to_dentry(struct super_block *sb,
 991                                 struct fid *fid, int fh_len, int fh_type)
 992 {
 993         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 994                                     exofs_nfs_get_inode);
 995 }
 996
 997 static struct dentry *exofs_fh_to_parent(struct super_block *sb,
 998                                 struct fid *fid, int fh_len, int fh_type)
 999 {
1000         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1001                                     exofs_nfs_get_inode);
1002 }
1003
1004 static const struct export_operations exofs_export_ops = {
1005         .fh_to_dentry = exofs_fh_to_dentry,
1006         .fh_to_parent = exofs_fh_to_parent,
1007         .get_parent = exofs_get_parent,
1008 };
1009
1010 /******************************************************************************
1011  * INSMOD/RMMOD
1012  *****************************************************************************/
1013
1014 /*
1015  * struct that describes this file system
1016  */
1017 static struct file_system_type exofs_type = {
1018         .owner          = THIS_MODULE,
1019         .name           = "exofs",
1020         .mount          = exofs_mount,
1021         .kill_sb        = generic_shutdown_super,
1022 };
1023
1024 static int __init init_exofs(void)
1025 {
1026         int err;
1027
1028         err = init_inodecache();
1029         if (err)
1030                 goto out;
1031
1032         err = register_filesystem(&exofs_type);
1033         if (err)
1034                 goto out_d;
1035
1036         /* We don't fail if sysfs creation failed */
1037         exofs_sysfs_init();
1038
1039         return 0;
1040 out_d:
1041         destroy_inodecache();
1042 out:
1043         return err;
1044 }
1045
1046 static void __exit exit_exofs(void)
1047 {
1048         exofs_sysfs_uninit();
1049         unregister_filesystem(&exofs_type);
1050         destroy_inodecache();
1051 }
1052
1053 MODULE_AUTHOR("Avishay Traeger <avishay@gmail.com>");
1054 MODULE_DESCRIPTION("exofs");
1055 MODULE_LICENSE("GPL");
1056
1057 module_init(init_exofs)
1058 module_exit(exit_exofs)