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