fs/nilfs2/super.c

   1 /*
   2  * super.c - NILFS module and super block management.
   3  *
   4  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License as published by
   8  * the Free Software Foundation; either version 2 of the License, or
   9  * (at your option) any later version.
  10  *
  11  * This program is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  * GNU General Public License for more details.
  15  *
  16  * Written by Ryusuke Konishi.
  17  */
  18 /*
  19  *  linux/fs/ext2/super.c
  20  *
  21  * Copyright (C) 1992, 1993, 1994, 1995
  22  * Remy Card (card@masi.ibp.fr)
  23  * Laboratoire MASI - Institut Blaise Pascal
  24  * Universite Pierre et Marie Curie (Paris VI)
  25  *
  26  *  from
  27  *
  28  *  linux/fs/minix/inode.c
  29  *
  30  *  Copyright (C) 1991, 1992  Linus Torvalds
  31  *
  32  *  Big-endian to little-endian byte-swapping/bitmaps by
  33  *        David S. Miller (davem@caip.rutgers.edu), 1995
  34  */
  35
  36 #include <linux/module.h>
  37 #include <linux/string.h>
  38 #include <linux/slab.h>
  39 #include <linux/init.h>
  40 #include <linux/blkdev.h>
  41 #include <linux/parser.h>
  42 #include <linux/crc32.h>
  43 #include <linux/vfs.h>
  44 #include <linux/writeback.h>
  45 #include <linux/seq_file.h>
  46 #include <linux/mount.h>
  47 #include "nilfs.h"
  48 #include "export.h"
  49 #include "mdt.h"
  50 #include "alloc.h"
  51 #include "btree.h"
  52 #include "btnode.h"
  53 #include "page.h"
  54 #include "cpfile.h"
  55 #include "sufile.h" /* nilfs_sufile_resize(), nilfs_sufile_set_alloc_range() */
  56 #include "ifile.h"
  57 #include "dat.h"
  58 #include "segment.h"
  59 #include "segbuf.h"
  60
  61 MODULE_AUTHOR("NTT Corp.");
  62 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
  63                    "(NILFS)");
  64 MODULE_LICENSE("GPL");
  65
  66 static struct kmem_cache *nilfs_inode_cachep;
  67 struct kmem_cache *nilfs_transaction_cachep;
  68 struct kmem_cache *nilfs_segbuf_cachep;
  69 struct kmem_cache *nilfs_btree_path_cache;
  70
  71 static int nilfs_setup_super(struct super_block *sb, int is_mount);
  72 static int nilfs_remount(struct super_block *sb, int *flags, char *data);
  73
  74 void __nilfs_msg(struct super_block *sb, const char *level, const char *fmt,
  75                  ...)
  76 {
  77         struct va_format vaf;
  78         va_list args;
  79
  80         va_start(args, fmt);
  81         vaf.fmt = fmt;
  82         vaf.va = &args;
  83         if (sb)
  84                 printk("%sNILFS (%s): %pV\n", level, sb->s_id, &vaf);
  85         else
  86                 printk("%sNILFS: %pV\n", level, &vaf);
  87         va_end(args);
  88 }
  89
  90 static void nilfs_set_error(struct super_block *sb)
  91 {
  92         struct the_nilfs *nilfs = sb->s_fs_info;
  93         struct nilfs_super_block **sbp;
  94
  95         down_write(&nilfs->ns_sem);
  96         if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
  97                 nilfs->ns_mount_state |= NILFS_ERROR_FS;
  98                 sbp = nilfs_prepare_super(sb, 0);
  99                 if (likely(sbp)) {
 100                         sbp[0]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
 101                         if (sbp[1])
 102                                 sbp[1]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
 103                         nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
 104                 }
 105         }
 106         up_write(&nilfs->ns_sem);
 107 }
 108
 109 /**
 110  * __nilfs_error() - report failure condition on a filesystem
 111  *
 112  * __nilfs_error() sets an ERROR_FS flag on the superblock as well as
 113  * reporting an error message.  This function should be called when
 114  * NILFS detects incoherences or defects of meta data on disk.
 115  *
 116  * This implements the body of nilfs_error() macro.  Normally,
 117  * nilfs_error() should be used.  As for sustainable errors such as a
 118  * single-shot I/O error, nilfs_msg() should be used instead.
 119  *
 120  * Callers should not add a trailing newline since this will do it.
 121  */
 122 void __nilfs_error(struct super_block *sb, const char *function,
 123                    const char *fmt, ...)
 124 {
 125         struct the_nilfs *nilfs = sb->s_fs_info;
 126         struct va_format vaf;
 127         va_list args;
 128
 129         va_start(args, fmt);
 130
 131         vaf.fmt = fmt;
 132         vaf.va = &args;
 133
 134         printk(KERN_CRIT "NILFS error (device %s): %s: %pV\n",
 135                sb->s_id, function, &vaf);
 136
 137         va_end(args);
 138
 139         if (!sb_rdonly(sb)) {
 140                 nilfs_set_error(sb);
 141
 142                 if (nilfs_test_opt(nilfs, ERRORS_RO)) {
 143                         printk(KERN_CRIT "Remounting filesystem read-only\n");
 144                         sb->s_flags |= SB_RDONLY;
 145                 }
 146         }
 147
 148         if (nilfs_test_opt(nilfs, ERRORS_PANIC))
 149                 panic("NILFS (device %s): panic forced after error\n",
 150                       sb->s_id);
 151 }
 152
 153 struct inode *nilfs_alloc_inode(struct super_block *sb)
 154 {
 155         struct nilfs_inode_info *ii;
 156
 157         ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
 158         if (!ii)
 159                 return NULL;
 160         ii->i_bh = NULL;
 161         ii->i_state = 0;
 162         ii->i_cno = 0;
 163         nilfs_mapping_init(&ii->i_btnode_cache, &ii->vfs_inode);
 164         return &ii->vfs_inode;
 165 }
 166
 167 static void nilfs_i_callback(struct rcu_head *head)
 168 {
 169         struct inode *inode = container_of(head, struct inode, i_rcu);
 170
 171         if (nilfs_is_metadata_file_inode(inode))
 172                 nilfs_mdt_destroy(inode);
 173
 174         kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
 175 }
 176
 177 void nilfs_destroy_inode(struct inode *inode)
 178 {
 179         call_rcu(&inode->i_rcu, nilfs_i_callback);
 180 }
 181
 182 static int nilfs_sync_super(struct super_block *sb, int flag)
 183 {
 184         struct the_nilfs *nilfs = sb->s_fs_info;
 185         int err;
 186
 187  retry:
 188         set_buffer_dirty(nilfs->ns_sbh[0]);
 189         if (nilfs_test_opt(nilfs, BARRIER)) {
 190                 err = __sync_dirty_buffer(nilfs->ns_sbh[0],
 191                                           REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
 192         } else {
 193                 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
 194         }
 195
 196         if (unlikely(err)) {
 197                 nilfs_msg(sb, KERN_ERR, "unable to write superblock: err=%d",
 198                           err);
 199                 if (err == -EIO && nilfs->ns_sbh[1]) {
 200                         /*
 201                          * sbp[0] points to newer log than sbp[1],
 202                          * so copy sbp[0] to sbp[1] to take over sbp[0].
 203                          */
 204                         memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
 205                                nilfs->ns_sbsize);
 206                         nilfs_fall_back_super_block(nilfs);
 207                         goto retry;
 208                 }
 209         } else {
 210                 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
 211
 212                 nilfs->ns_sbwcount++;
 213
 214                 /*
 215                  * The latest segment becomes trailable from the position
 216                  * written in superblock.
 217                  */
 218                 clear_nilfs_discontinued(nilfs);
 219
 220                 /* update GC protection for recent segments */
 221                 if (nilfs->ns_sbh[1]) {
 222                         if (flag == NILFS_SB_COMMIT_ALL) {
 223                                 set_buffer_dirty(nilfs->ns_sbh[1]);
 224                                 if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
 225                                         goto out;
 226                         }
 227                         if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
 228                             le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
 229                                 sbp = nilfs->ns_sbp[1];
 230                 }
 231
 232                 spin_lock(&nilfs->ns_last_segment_lock);
 233                 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
 234                 spin_unlock(&nilfs->ns_last_segment_lock);
 235         }
 236  out:
 237         return err;
 238 }
 239
 240 void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
 241                           struct the_nilfs *nilfs)
 242 {
 243         sector_t nfreeblocks;
 244
 245         /* nilfs->ns_sem must be locked by the caller. */
 246         nilfs_count_free_blocks(nilfs, &nfreeblocks);
 247         sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
 248
 249         spin_lock(&nilfs->ns_last_segment_lock);
 250         sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
 251         sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
 252         sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
 253         spin_unlock(&nilfs->ns_last_segment_lock);
 254 }
 255
 256 struct nilfs_super_block **nilfs_prepare_super(struct super_block *sb,
 257                                                int flip)
 258 {
 259         struct the_nilfs *nilfs = sb->s_fs_info;
 260         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 261
 262         /* nilfs->ns_sem must be locked by the caller. */
 263         if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
 264                 if (sbp[1] &&
 265                     sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
 266                         memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
 267                 } else {
 268                         nilfs_msg(sb, KERN_CRIT, "superblock broke");
 269                         return NULL;
 270                 }
 271         } else if (sbp[1] &&
 272                    sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
 273                 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 274         }
 275
 276         if (flip && sbp[1])
 277                 nilfs_swap_super_block(nilfs);
 278
 279         return sbp;
 280 }
 281
 282 int nilfs_commit_super(struct super_block *sb, int flag)
 283 {
 284         struct the_nilfs *nilfs = sb->s_fs_info;
 285         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 286         time64_t t;
 287
 288         /* nilfs->ns_sem must be locked by the caller. */
 289         t = ktime_get_real_seconds();
 290         nilfs->ns_sbwtime = t;
 291         sbp[0]->s_wtime = cpu_to_le64(t);
 292         sbp[0]->s_sum = 0;
 293         sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
 294                                              (unsigned char *)sbp[0],
 295                                              nilfs->ns_sbsize));
 296         if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
 297                 sbp[1]->s_wtime = sbp[0]->s_wtime;
 298                 sbp[1]->s_sum = 0;
 299                 sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
 300                                             (unsigned char *)sbp[1],
 301                                             nilfs->ns_sbsize));
 302         }
 303         clear_nilfs_sb_dirty(nilfs);
 304         nilfs->ns_flushed_device = 1;
 305         /* make sure store to ns_flushed_device cannot be reordered */
 306         smp_wmb();
 307         return nilfs_sync_super(sb, flag);
 308 }
 309
 310 /**
 311  * nilfs_cleanup_super() - write filesystem state for cleanup
 312  * @sb: super block instance to be unmounted or degraded to read-only
 313  *
 314  * This function restores state flags in the on-disk super block.
 315  * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
 316  * filesystem was not clean previously.
 317  */
 318 int nilfs_cleanup_super(struct super_block *sb)
 319 {
 320         struct the_nilfs *nilfs = sb->s_fs_info;
 321         struct nilfs_super_block **sbp;
 322         int flag = NILFS_SB_COMMIT;
 323         int ret = -EIO;
 324
 325         sbp = nilfs_prepare_super(sb, 0);
 326         if (sbp) {
 327                 sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
 328                 nilfs_set_log_cursor(sbp[0], nilfs);
 329                 if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
 330                         /*
 331                          * make the "clean" flag also to the opposite
 332                          * super block if both super blocks point to
 333                          * the same checkpoint.
 334                          */
 335                         sbp[1]->s_state = sbp[0]->s_state;
 336                         flag = NILFS_SB_COMMIT_ALL;
 337                 }
 338                 ret = nilfs_commit_super(sb, flag);
 339         }
 340         return ret;
 341 }
 342
 343 /**
 344  * nilfs_move_2nd_super - relocate secondary super block
 345  * @sb: super block instance
 346  * @sb2off: new offset of the secondary super block (in bytes)
 347  */
 348 static int nilfs_move_2nd_super(struct super_block *sb, loff_t sb2off)
 349 {
 350         struct the_nilfs *nilfs = sb->s_fs_info;
 351         struct buffer_head *nsbh;
 352         struct nilfs_super_block *nsbp;
 353         sector_t blocknr, newblocknr;
 354         unsigned long offset;
 355         int sb2i;  /* array index of the secondary superblock */
 356         int ret = 0;
 357
 358         /* nilfs->ns_sem must be locked by the caller. */
 359         if (nilfs->ns_sbh[1] &&
 360             nilfs->ns_sbh[1]->b_blocknr > nilfs->ns_first_data_block) {
 361                 sb2i = 1;
 362                 blocknr = nilfs->ns_sbh[1]->b_blocknr;
 363         } else if (nilfs->ns_sbh[0]->b_blocknr > nilfs->ns_first_data_block) {
 364                 sb2i = 0;
 365                 blocknr = nilfs->ns_sbh[0]->b_blocknr;
 366         } else {
 367                 sb2i = -1;
 368                 blocknr = 0;
 369         }
 370         if (sb2i >= 0 && (u64)blocknr << nilfs->ns_blocksize_bits == sb2off)
 371                 goto out;  /* super block location is unchanged */
 372
 373         /* Get new super block buffer */
 374         newblocknr = sb2off >> nilfs->ns_blocksize_bits;
 375         offset = sb2off & (nilfs->ns_blocksize - 1);
 376         nsbh = sb_getblk(sb, newblocknr);
 377         if (!nsbh) {
 378                 nilfs_msg(sb, KERN_WARNING,
 379                           "unable to move secondary superblock to block %llu",
 380                           (unsigned long long)newblocknr);
 381                 ret = -EIO;
 382                 goto out;
 383         }
 384         nsbp = (void *)nsbh->b_data + offset;
 385         memset(nsbp, 0, nilfs->ns_blocksize);
 386
 387         if (sb2i >= 0) {
 388                 memcpy(nsbp, nilfs->ns_sbp[sb2i], nilfs->ns_sbsize);
 389                 brelse(nilfs->ns_sbh[sb2i]);
 390                 nilfs->ns_sbh[sb2i] = nsbh;
 391                 nilfs->ns_sbp[sb2i] = nsbp;
 392         } else if (nilfs->ns_sbh[0]->b_blocknr < nilfs->ns_first_data_block) {
 393                 /* secondary super block will be restored to index 1 */
 394                 nilfs->ns_sbh[1] = nsbh;
 395                 nilfs->ns_sbp[1] = nsbp;
 396         } else {
 397                 brelse(nsbh);
 398         }
 399 out:
 400         return ret;
 401 }
 402
 403 /**
 404  * nilfs_resize_fs - resize the filesystem
 405  * @sb: super block instance
 406  * @newsize: new size of the filesystem (in bytes)
 407  */
 408 int nilfs_resize_fs(struct super_block *sb, __u64 newsize)
 409 {
 410         struct the_nilfs *nilfs = sb->s_fs_info;
 411         struct nilfs_super_block **sbp;
 412         __u64 devsize, newnsegs;
 413         loff_t sb2off;
 414         int ret;
 415
 416         ret = -ERANGE;
 417         devsize = i_size_read(sb->s_bdev->bd_inode);
 418         if (newsize > devsize)
 419                 goto out;
 420
 421         /*
 422          * Write lock is required to protect some functions depending
 423          * on the number of segments, the number of reserved segments,
 424          * and so forth.
 425          */
 426         down_write(&nilfs->ns_segctor_sem);
 427
 428         sb2off = NILFS_SB2_OFFSET_BYTES(newsize);
 429         newnsegs = sb2off >> nilfs->ns_blocksize_bits;
 430         do_div(newnsegs, nilfs->ns_blocks_per_segment);
 431
 432         ret = nilfs_sufile_resize(nilfs->ns_sufile, newnsegs);
 433         up_write(&nilfs->ns_segctor_sem);
 434         if (ret < 0)
 435                 goto out;
 436
 437         ret = nilfs_construct_segment(sb);
 438         if (ret < 0)
 439                 goto out;
 440
 441         down_write(&nilfs->ns_sem);
 442         nilfs_move_2nd_super(sb, sb2off);
 443         ret = -EIO;
 444         sbp = nilfs_prepare_super(sb, 0);
 445         if (likely(sbp)) {
 446                 nilfs_set_log_cursor(sbp[0], nilfs);
 447                 /*
 448                  * Drop NILFS_RESIZE_FS flag for compatibility with
 449                  * mount-time resize which may be implemented in a
 450                  * future release.
 451                  */
 452                 sbp[0]->s_state = cpu_to_le16(le16_to_cpu(sbp[0]->s_state) &
 453                                               ~NILFS_RESIZE_FS);
 454                 sbp[0]->s_dev_size = cpu_to_le64(newsize);
 455                 sbp[0]->s_nsegments = cpu_to_le64(nilfs->ns_nsegments);
 456                 if (sbp[1])
 457                         memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 458                 ret = nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
 459         }
 460         up_write(&nilfs->ns_sem);
 461
 462         /*
 463          * Reset the range of allocatable segments last.  This order
 464          * is important in the case of expansion because the secondary
 465          * superblock must be protected from log write until migration
 466          * completes.
 467          */
 468         if (!ret)
 469                 nilfs_sufile_set_alloc_range(nilfs->ns_sufile, 0, newnsegs - 1);
 470 out:
 471         return ret;
 472 }
 473
 474 static void nilfs_put_super(struct super_block *sb)
 475 {
 476         struct the_nilfs *nilfs = sb->s_fs_info;
 477
 478         nilfs_detach_log_writer(sb);
 479
 480         if (!sb_rdonly(sb)) {
 481                 down_write(&nilfs->ns_sem);
 482                 nilfs_cleanup_super(sb);
 483                 up_write(&nilfs->ns_sem);
 484         }
 485
 486         iput(nilfs->ns_sufile);
 487         iput(nilfs->ns_cpfile);
 488         iput(nilfs->ns_dat);
 489
 490         destroy_nilfs(nilfs);
 491         sb->s_fs_info = NULL;
 492 }
 493
 494 static int nilfs_sync_fs(struct super_block *sb, int wait)
 495 {
 496         struct the_nilfs *nilfs = sb->s_fs_info;
 497         struct nilfs_super_block **sbp;
 498         int err = 0;
 499
 500         /* This function is called when super block should be written back */
 501         if (wait)
 502                 err = nilfs_construct_segment(sb);
 503
 504         down_write(&nilfs->ns_sem);
 505         if (nilfs_sb_dirty(nilfs)) {
 506                 sbp = nilfs_prepare_super(sb, nilfs_sb_will_flip(nilfs));
 507                 if (likely(sbp)) {
 508                         nilfs_set_log_cursor(sbp[0], nilfs);
 509                         nilfs_commit_super(sb, NILFS_SB_COMMIT);
 510                 }
 511         }
 512         up_write(&nilfs->ns_sem);
 513
 514         if (!err)
 515                 err = nilfs_flush_device(nilfs);
 516
 517         return err;
 518 }
 519
 520 int nilfs_attach_checkpoint(struct super_block *sb, __u64 cno, int curr_mnt,
 521                             struct nilfs_root **rootp)
 522 {
 523         struct the_nilfs *nilfs = sb->s_fs_info;
 524         struct nilfs_root *root;
 525         struct nilfs_checkpoint *raw_cp;
 526         struct buffer_head *bh_cp;
 527         int err = -ENOMEM;
 528
 529         root = nilfs_find_or_create_root(
 530                 nilfs, curr_mnt ? NILFS_CPTREE_CURRENT_CNO : cno);
 531         if (!root)
 532                 return err;
 533
 534         if (root->ifile)
 535                 goto reuse; /* already attached checkpoint */
 536
 537         down_read(&nilfs->ns_segctor_sem);
 538         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
 539                                           &bh_cp);
 540         up_read(&nilfs->ns_segctor_sem);
 541         if (unlikely(err)) {
 542                 if (err == -ENOENT || err == -EINVAL) {
 543                         nilfs_msg(sb, KERN_ERR,
 544                                   "Invalid checkpoint (checkpoint number=%llu)",
 545                                   (unsigned long long)cno);
 546                         err = -EINVAL;
 547                 }
 548                 goto failed;
 549         }
 550
 551         err = nilfs_ifile_read(sb, root, nilfs->ns_inode_size,
 552                                &raw_cp->cp_ifile_inode, &root->ifile);
 553         if (err)
 554                 goto failed_bh;
 555
 556         atomic64_set(&root->inodes_count,
 557                         le64_to_cpu(raw_cp->cp_inodes_count));
 558         atomic64_set(&root->blocks_count,
 559                         le64_to_cpu(raw_cp->cp_blocks_count));
 560
 561         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
 562
 563  reuse:
 564         *rootp = root;
 565         return 0;
 566
 567  failed_bh:
 568         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
 569  failed:
 570         nilfs_put_root(root);
 571
 572         return err;
 573 }
 574
 575 static int nilfs_freeze(struct super_block *sb)
 576 {
 577         struct the_nilfs *nilfs = sb->s_fs_info;
 578         int err;
 579
 580         if (sb_rdonly(sb))
 581                 return 0;
 582
 583         /* Mark super block clean */
 584         down_write(&nilfs->ns_sem);
 585         err = nilfs_cleanup_super(sb);
 586         up_write(&nilfs->ns_sem);
 587         return err;
 588 }
 589
 590 static int nilfs_unfreeze(struct super_block *sb)
 591 {
 592         struct the_nilfs *nilfs = sb->s_fs_info;
 593
 594         if (sb_rdonly(sb))
 595                 return 0;
 596
 597         down_write(&nilfs->ns_sem);
 598         nilfs_setup_super(sb, false);
 599         up_write(&nilfs->ns_sem);
 600         return 0;
 601 }
 602
 603 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 604 {
 605         struct super_block *sb = dentry->d_sb;
 606         struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
 607         struct the_nilfs *nilfs = root->nilfs;
 608         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
 609         unsigned long long blocks;
 610         unsigned long overhead;
 611         unsigned long nrsvblocks;
 612         sector_t nfreeblocks;
 613         u64 nmaxinodes, nfreeinodes;
 614         int err;
 615
 616         /*
 617          * Compute all of the segment blocks
 618          *
 619          * The blocks before first segment and after last segment
 620          * are excluded.
 621          */
 622         blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
 623                 - nilfs->ns_first_data_block;
 624         nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
 625
 626         /*
 627          * Compute the overhead
 628          *
 629          * When distributing meta data blocks outside segment structure,
 630          * We must count them as the overhead.
 631          */
 632         overhead = 0;
 633
 634         err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
 635         if (unlikely(err))
 636                 return err;
 637
 638         err = nilfs_ifile_count_free_inodes(root->ifile,
 639                                             &nmaxinodes, &nfreeinodes);
 640         if (unlikely(err)) {
 641                 nilfs_msg(sb, KERN_WARNING,
 642                           "failed to count free inodes: err=%d", err);
 643                 if (err == -ERANGE) {
 644                         /*
 645                          * If nilfs_palloc_count_max_entries() returns
 646                          * -ERANGE error code then we simply treat
 647                          * curent inodes count as maximum possible and
 648                          * zero as free inodes value.
 649                          */
 650                         nmaxinodes = atomic64_read(&root->inodes_count);
 651                         nfreeinodes = 0;
 652                         err = 0;
 653                 } else
 654                         return err;
 655         }
 656
 657         buf->f_type = NILFS_SUPER_MAGIC;
 658         buf->f_bsize = sb->s_blocksize;
 659         buf->f_blocks = blocks - overhead;
 660         buf->f_bfree = nfreeblocks;
 661         buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
 662                 (buf->f_bfree - nrsvblocks) : 0;
 663         buf->f_files = nmaxinodes;
 664         buf->f_ffree = nfreeinodes;
 665         buf->f_namelen = NILFS_NAME_LEN;
 666         buf->f_fsid.val[0] = (u32)id;
 667         buf->f_fsid.val[1] = (u32)(id >> 32);
 668
 669         return 0;
 670 }
 671
 672 static int nilfs_show_options(struct seq_file *seq, struct dentry *dentry)
 673 {
 674         struct super_block *sb = dentry->d_sb;
 675         struct the_nilfs *nilfs = sb->s_fs_info;
 676         struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
 677
 678         if (!nilfs_test_opt(nilfs, BARRIER))
 679                 seq_puts(seq, ",nobarrier");
 680         if (root->cno != NILFS_CPTREE_CURRENT_CNO)
 681                 seq_printf(seq, ",cp=%llu", (unsigned long long)root->cno);
 682         if (nilfs_test_opt(nilfs, ERRORS_PANIC))
 683                 seq_puts(seq, ",errors=panic");
 684         if (nilfs_test_opt(nilfs, ERRORS_CONT))
 685                 seq_puts(seq, ",errors=continue");
 686         if (nilfs_test_opt(nilfs, STRICT_ORDER))
 687                 seq_puts(seq, ",order=strict");
 688         if (nilfs_test_opt(nilfs, NORECOVERY))
 689                 seq_puts(seq, ",norecovery");
 690         if (nilfs_test_opt(nilfs, DISCARD))
 691                 seq_puts(seq, ",discard");
 692
 693         return 0;
 694 }
 695
 696 static const struct super_operations nilfs_sops = {
 697         .alloc_inode    = nilfs_alloc_inode,
 698         .destroy_inode  = nilfs_destroy_inode,
 699         .dirty_inode    = nilfs_dirty_inode,
 700         .evict_inode    = nilfs_evict_inode,
 701         .put_super      = nilfs_put_super,
 702         .sync_fs        = nilfs_sync_fs,
 703         .freeze_fs      = nilfs_freeze,
 704         .unfreeze_fs    = nilfs_unfreeze,
 705         .statfs         = nilfs_statfs,
 706         .remount_fs     = nilfs_remount,
 707         .show_options = nilfs_show_options
 708 };
 709
 710 enum {
 711         Opt_err_cont, Opt_err_panic, Opt_err_ro,
 712         Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
 713         Opt_discard, Opt_nodiscard, Opt_err,
 714 };
 715
 716 static match_table_t tokens = {
 717         {Opt_err_cont, "errors=continue"},
 718         {Opt_err_panic, "errors=panic"},
 719         {Opt_err_ro, "errors=remount-ro"},
 720         {Opt_barrier, "barrier"},
 721         {Opt_nobarrier, "nobarrier"},
 722         {Opt_snapshot, "cp=%u"},
 723         {Opt_order, "order=%s"},
 724         {Opt_norecovery, "norecovery"},
 725         {Opt_discard, "discard"},
 726         {Opt_nodiscard, "nodiscard"},
 727         {Opt_err, NULL}
 728 };
 729
 730 static int parse_options(char *options, struct super_block *sb, int is_remount)
 731 {
 732         struct the_nilfs *nilfs = sb->s_fs_info;
 733         char *p;
 734         substring_t args[MAX_OPT_ARGS];
 735
 736         if (!options)
 737                 return 1;
 738
 739         while ((p = strsep(&options, ",")) != NULL) {
 740                 int token;
 741
 742                 if (!*p)
 743                         continue;
 744
 745                 token = match_token(p, tokens, args);
 746                 switch (token) {
 747                 case Opt_barrier:
 748                         nilfs_set_opt(nilfs, BARRIER);
 749                         break;
 750                 case Opt_nobarrier:
 751                         nilfs_clear_opt(nilfs, BARRIER);
 752                         break;
 753                 case Opt_order:
 754                         if (strcmp(args[0].from, "relaxed") == 0)
 755                                 /* Ordered data semantics */
 756                                 nilfs_clear_opt(nilfs, STRICT_ORDER);
 757                         else if (strcmp(args[0].from, "strict") == 0)
 758                                 /* Strict in-order semantics */
 759                                 nilfs_set_opt(nilfs, STRICT_ORDER);
 760                         else
 761                                 return 0;
 762                         break;
 763                 case Opt_err_panic:
 764                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_PANIC);
 765                         break;
 766                 case Opt_err_ro:
 767                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_RO);
 768                         break;
 769                 case Opt_err_cont:
 770                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_CONT);
 771                         break;
 772                 case Opt_snapshot:
 773                         if (is_remount) {
 774                                 nilfs_msg(sb, KERN_ERR,
 775                                           "\"%s\" option is invalid for remount",
 776                                           p);
 777                                 return 0;
 778                         }
 779                         break;
 780                 case Opt_norecovery:
 781                         nilfs_set_opt(nilfs, NORECOVERY);
 782                         break;
 783                 case Opt_discard:
 784                         nilfs_set_opt(nilfs, DISCARD);
 785                         break;
 786                 case Opt_nodiscard:
 787                         nilfs_clear_opt(nilfs, DISCARD);
 788                         break;
 789                 default:
 790                         nilfs_msg(sb, KERN_ERR,
 791                                   "unrecognized mount option \"%s\"", p);
 792                         return 0;
 793                 }
 794         }
 795         return 1;
 796 }
 797
 798 static inline void
 799 nilfs_set_default_options(struct super_block *sb,
 800                           struct nilfs_super_block *sbp)
 801 {
 802         struct the_nilfs *nilfs = sb->s_fs_info;
 803
 804         nilfs->ns_mount_opt =
 805                 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
 806 }
 807
 808 static int nilfs_setup_super(struct super_block *sb, int is_mount)
 809 {
 810         struct the_nilfs *nilfs = sb->s_fs_info;
 811         struct nilfs_super_block **sbp;
 812         int max_mnt_count;
 813         int mnt_count;
 814
 815         /* nilfs->ns_sem must be locked by the caller. */
 816         sbp = nilfs_prepare_super(sb, 0);
 817         if (!sbp)
 818                 return -EIO;
 819
 820         if (!is_mount)
 821                 goto skip_mount_setup;
 822
 823         max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
 824         mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
 825
 826         if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
 827                 nilfs_msg(sb, KERN_WARNING, "mounting fs with errors");
 828 #if 0
 829         } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
 830                 nilfs_msg(sb, KERN_WARNING, "maximal mount count reached");
 831 #endif
 832         }
 833         if (!max_mnt_count)
 834                 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
 835
 836         sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
 837         sbp[0]->s_mtime = cpu_to_le64(get_seconds());
 838
 839 skip_mount_setup:
 840         sbp[0]->s_state =
 841                 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
 842         /* synchronize sbp[1] with sbp[0] */
 843         if (sbp[1])
 844                 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 845         return nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
 846 }
 847
 848 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
 849                                                  u64 pos, int blocksize,
 850                                                  struct buffer_head **pbh)
 851 {
 852         unsigned long long sb_index = pos;
 853         unsigned long offset;
 854
 855         offset = do_div(sb_index, blocksize);
 856         *pbh = sb_bread(sb, sb_index);
 857         if (!*pbh)
 858                 return NULL;
 859         return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
 860 }
 861
 862 int nilfs_store_magic_and_option(struct super_block *sb,
 863                                  struct nilfs_super_block *sbp,
 864                                  char *data)
 865 {
 866         struct the_nilfs *nilfs = sb->s_fs_info;
 867
 868         sb->s_magic = le16_to_cpu(sbp->s_magic);
 869
 870         /* FS independent flags */
 871 #ifdef NILFS_ATIME_DISABLE
 872         sb->s_flags |= SB_NOATIME;
 873 #endif
 874
 875         nilfs_set_default_options(sb, sbp);
 876
 877         nilfs->ns_resuid = le16_to_cpu(sbp->s_def_resuid);
 878         nilfs->ns_resgid = le16_to_cpu(sbp->s_def_resgid);
 879         nilfs->ns_interval = le32_to_cpu(sbp->s_c_interval);
 880         nilfs->ns_watermark = le32_to_cpu(sbp->s_c_block_max);
 881
 882         return !parse_options(data, sb, 0) ? -EINVAL : 0;
 883 }
 884
 885 int nilfs_check_feature_compatibility(struct super_block *sb,
 886                                       struct nilfs_super_block *sbp)
 887 {
 888         __u64 features;
 889
 890         features = le64_to_cpu(sbp->s_feature_incompat) &
 891                 ~NILFS_FEATURE_INCOMPAT_SUPP;
 892         if (features) {
 893                 nilfs_msg(sb, KERN_ERR,
 894                           "couldn't mount because of unsupported optional features (%llx)",
 895                           (unsigned long long)features);
 896                 return -EINVAL;
 897         }
 898         features = le64_to_cpu(sbp->s_feature_compat_ro) &
 899                 ~NILFS_FEATURE_COMPAT_RO_SUPP;
 900         if (!sb_rdonly(sb) && features) {
 901                 nilfs_msg(sb, KERN_ERR,
 902                           "couldn't mount RDWR because of unsupported optional features (%llx)",
 903                           (unsigned long long)features);
 904                 return -EINVAL;
 905         }
 906         return 0;
 907 }
 908
 909 static int nilfs_get_root_dentry(struct super_block *sb,
 910                                  struct nilfs_root *root,
 911                                  struct dentry **root_dentry)
 912 {
 913         struct inode *inode;
 914         struct dentry *dentry;
 915         int ret = 0;
 916
 917         inode = nilfs_iget(sb, root, NILFS_ROOT_INO);
 918         if (IS_ERR(inode)) {
 919                 ret = PTR_ERR(inode);
 920                 nilfs_msg(sb, KERN_ERR, "error %d getting root inode", ret);
 921                 goto out;
 922         }
 923         if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
 924                 iput(inode);
 925                 nilfs_msg(sb, KERN_ERR, "corrupt root inode");
 926                 ret = -EINVAL;
 927                 goto out;
 928         }
 929
 930         if (root->cno == NILFS_CPTREE_CURRENT_CNO) {
 931                 dentry = d_find_alias(inode);
 932                 if (!dentry) {
 933                         dentry = d_make_root(inode);
 934                         if (!dentry) {
 935                                 ret = -ENOMEM;
 936                                 goto failed_dentry;
 937                         }
 938                 } else {
 939                         iput(inode);
 940                 }
 941         } else {
 942                 dentry = d_obtain_root(inode);
 943                 if (IS_ERR(dentry)) {
 944                         ret = PTR_ERR(dentry);
 945                         goto failed_dentry;
 946                 }
 947         }
 948         *root_dentry = dentry;
 949  out:
 950         return ret;
 951
 952  failed_dentry:
 953         nilfs_msg(sb, KERN_ERR, "error %d getting root dentry", ret);
 954         goto out;
 955 }
 956
 957 static int nilfs_attach_snapshot(struct super_block *s, __u64 cno,
 958                                  struct dentry **root_dentry)
 959 {
 960         struct the_nilfs *nilfs = s->s_fs_info;
 961         struct nilfs_root *root;
 962         int ret;
 963
 964         mutex_lock(&nilfs->ns_snapshot_mount_mutex);
 965
 966         down_read(&nilfs->ns_segctor_sem);
 967         ret = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, cno);
 968         up_read(&nilfs->ns_segctor_sem);
 969         if (ret < 0) {
 970                 ret = (ret == -ENOENT) ? -EINVAL : ret;
 971                 goto out;
 972         } else if (!ret) {
 973                 nilfs_msg(s, KERN_ERR,
 974                           "The specified checkpoint is not a snapshot (checkpoint number=%llu)",
 975                           (unsigned long long)cno);
 976                 ret = -EINVAL;
 977                 goto out;
 978         }
 979
 980         ret = nilfs_attach_checkpoint(s, cno, false, &root);
 981         if (ret) {
 982                 nilfs_msg(s, KERN_ERR,
 983                           "error %d while loading snapshot (checkpoint number=%llu)",
 984                           ret, (unsigned long long)cno);
 985                 goto out;
 986         }
 987         ret = nilfs_get_root_dentry(s, root, root_dentry);
 988         nilfs_put_root(root);
 989  out:
 990         mutex_unlock(&nilfs->ns_snapshot_mount_mutex);
 991         return ret;
 992 }
 993
 994 /**
 995  * nilfs_tree_is_busy() - try to shrink dentries of a checkpoint
 996  * @root_dentry: root dentry of the tree to be shrunk
 997  *
 998  * This function returns true if the tree was in-use.
 999  */
1000 static bool nilfs_tree_is_busy(struct dentry *root_dentry)
1001 {
1002         shrink_dcache_parent(root_dentry);
1003         return d_count(root_dentry) > 1;
1004 }
1005
1006 int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno)
1007 {
1008         struct the_nilfs *nilfs = sb->s_fs_info;
1009         struct nilfs_root *root;
1010         struct inode *inode;
1011         struct dentry *dentry;
1012         int ret;
1013
1014         if (cno > nilfs->ns_cno)
1015                 return false;
1016
1017         if (cno >= nilfs_last_cno(nilfs))
1018                 return true;    /* protect recent checkpoints */
1019
1020         ret = false;
1021         root = nilfs_lookup_root(nilfs, cno);
1022         if (root) {
1023                 inode = nilfs_ilookup(sb, root, NILFS_ROOT_INO);
1024                 if (inode) {
1025                         dentry = d_find_alias(inode);
1026                         if (dentry) {
1027                                 ret = nilfs_tree_is_busy(dentry);
1028                                 dput(dentry);
1029                         }
1030                         iput(inode);
1031                 }
1032                 nilfs_put_root(root);
1033         }
1034         return ret;
1035 }
1036
1037 /**
1038  * nilfs_fill_super() - initialize a super block instance
1039  * @sb: super_block
1040  * @data: mount options
1041  * @silent: silent mode flag
1042  *
1043  * This function is called exclusively by nilfs->ns_mount_mutex.
1044  * So, the recovery process is protected from other simultaneous mounts.
1045  */
1046 static int
1047 nilfs_fill_super(struct super_block *sb, void *data, int silent)
1048 {
1049         struct the_nilfs *nilfs;
1050         struct nilfs_root *fsroot;
1051         __u64 cno;
1052         int err;
1053
1054         nilfs = alloc_nilfs(sb);
1055         if (!nilfs)
1056                 return -ENOMEM;
1057
1058         sb->s_fs_info = nilfs;
1059
1060         err = init_nilfs(nilfs, sb, (char *)data);
1061         if (err)
1062                 goto failed_nilfs;
1063
1064         sb->s_op = &nilfs_sops;
1065         sb->s_export_op = &nilfs_export_ops;
1066         sb->s_root = NULL;
1067         sb->s_time_gran = 1;
1068         sb->s_max_links = NILFS_LINK_MAX;
1069
1070         sb->s_bdi = bdi_get(sb->s_bdev->bd_bdi);
1071
1072         err = load_nilfs(nilfs, sb);
1073         if (err)
1074                 goto failed_nilfs;
1075
1076         cno = nilfs_last_cno(nilfs);
1077         err = nilfs_attach_checkpoint(sb, cno, true, &fsroot);
1078         if (err) {
1079                 nilfs_msg(sb, KERN_ERR,
1080                           "error %d while loading last checkpoint (checkpoint number=%llu)",
1081                           err, (unsigned long long)cno);
1082                 goto failed_unload;
1083         }
1084
1085         if (!sb_rdonly(sb)) {
1086                 err = nilfs_attach_log_writer(sb, fsroot);
1087                 if (err)
1088                         goto failed_checkpoint;
1089         }
1090
1091         err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
1092         if (err)
1093                 goto failed_segctor;
1094
1095         nilfs_put_root(fsroot);
1096
1097         if (!sb_rdonly(sb)) {
1098                 down_write(&nilfs->ns_sem);
1099                 nilfs_setup_super(sb, true);
1100                 up_write(&nilfs->ns_sem);
1101         }
1102
1103         return 0;
1104
1105  failed_segctor:
1106         nilfs_detach_log_writer(sb);
1107
1108  failed_checkpoint:
1109         nilfs_put_root(fsroot);
1110
1111  failed_unload:
1112         iput(nilfs->ns_sufile);
1113         iput(nilfs->ns_cpfile);
1114         iput(nilfs->ns_dat);
1115
1116  failed_nilfs:
1117         destroy_nilfs(nilfs);
1118         return err;
1119 }
1120
1121 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
1122 {
1123         struct the_nilfs *nilfs = sb->s_fs_info;
1124         unsigned long old_sb_flags;
1125         unsigned long old_mount_opt;
1126         int err;
1127
1128         sync_filesystem(sb);
1129         old_sb_flags = sb->s_flags;
1130         old_mount_opt = nilfs->ns_mount_opt;
1131
1132         if (!parse_options(data, sb, 1)) {
1133                 err = -EINVAL;
1134                 goto restore_opts;
1135         }
1136         sb->s_flags = (sb->s_flags & ~SB_POSIXACL);
1137
1138         err = -EINVAL;
1139
1140         if (!nilfs_valid_fs(nilfs)) {
1141                 nilfs_msg(sb, KERN_WARNING,
1142                           "couldn't remount because the filesystem is in an incomplete recovery state");
1143                 goto restore_opts;
1144         }
1145
1146         if ((bool)(*flags & SB_RDONLY) == sb_rdonly(sb))
1147                 goto out;
1148         if (*flags & SB_RDONLY) {
1149                 /* Shutting down log writer */
1150                 nilfs_detach_log_writer(sb);
1151                 sb->s_flags |= SB_RDONLY;
1152
1153                 /*
1154                  * Remounting a valid RW partition RDONLY, so set
1155                  * the RDONLY flag and then mark the partition as valid again.
1156                  */
1157                 down_write(&nilfs->ns_sem);
1158                 nilfs_cleanup_super(sb);
1159                 up_write(&nilfs->ns_sem);
1160         } else {
1161                 __u64 features;
1162                 struct nilfs_root *root;
1163
1164                 /*
1165                  * Mounting a RDONLY partition read-write, so reread and
1166                  * store the current valid flag.  (It may have been changed
1167                  * by fsck since we originally mounted the partition.)
1168                  */
1169                 down_read(&nilfs->ns_sem);
1170                 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1171                         ~NILFS_FEATURE_COMPAT_RO_SUPP;
1172                 up_read(&nilfs->ns_sem);
1173                 if (features) {
1174                         nilfs_msg(sb, KERN_WARNING,
1175                                   "couldn't remount RDWR because of unsupported optional features (%llx)",
1176                                   (unsigned long long)features);
1177                         err = -EROFS;
1178                         goto restore_opts;
1179                 }
1180
1181                 sb->s_flags &= ~SB_RDONLY;
1182
1183                 root = NILFS_I(d_inode(sb->s_root))->i_root;
1184                 err = nilfs_attach_log_writer(sb, root);
1185                 if (err)
1186                         goto restore_opts;
1187
1188                 down_write(&nilfs->ns_sem);
1189                 nilfs_setup_super(sb, true);
1190                 up_write(&nilfs->ns_sem);
1191         }
1192  out:
1193         return 0;
1194
1195  restore_opts:
1196         sb->s_flags = old_sb_flags;
1197         nilfs->ns_mount_opt = old_mount_opt;
1198         return err;
1199 }
1200
1201 struct nilfs_super_data {
1202         struct block_device *bdev;
1203         __u64 cno;
1204         int flags;
1205 };
1206
1207 static int nilfs_parse_snapshot_option(const char *option,
1208                                        const substring_t *arg,
1209                                        struct nilfs_super_data *sd)
1210 {
1211         unsigned long long val;
1212         const char *msg = NULL;
1213         int err;
1214
1215         if (!(sd->flags & SB_RDONLY)) {
1216                 msg = "read-only option is not specified";
1217                 goto parse_error;
1218         }
1219
1220         err = kstrtoull(arg->from, 0, &val);
1221         if (err) {
1222                 if (err == -ERANGE)
1223                         msg = "too large checkpoint number";
1224                 else
1225                         msg = "malformed argument";
1226                 goto parse_error;
1227         } else if (val == 0) {
1228                 msg = "invalid checkpoint number 0";
1229                 goto parse_error;
1230         }
1231         sd->cno = val;
1232         return 0;
1233
1234 parse_error:
1235         nilfs_msg(NULL, KERN_ERR, "invalid option \"%s\": %s", option, msg);
1236         return 1;
1237 }
1238
1239 /**
1240  * nilfs_identify - pre-read mount options needed to identify mount instance
1241  * @data: mount options
1242  * @sd: nilfs_super_data
1243  */
1244 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1245 {
1246         char *p, *options = data;
1247         substring_t args[MAX_OPT_ARGS];
1248         int token;
1249         int ret = 0;
1250
1251         do {
1252                 p = strsep(&options, ",");
1253                 if (p != NULL && *p) {
1254                         token = match_token(p, tokens, args);
1255                         if (token == Opt_snapshot)
1256                                 ret = nilfs_parse_snapshot_option(p, &args[0],
1257                                                                   sd);
1258                 }
1259                 if (!options)
1260                         break;
1261                 BUG_ON(options == data);
1262                 *(options - 1) = ',';
1263         } while (!ret);
1264         return ret;
1265 }
1266
1267 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1268 {
1269         s->s_bdev = data;
1270         s->s_dev = s->s_bdev->bd_dev;
1271         return 0;
1272 }
1273
1274 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1275 {
1276         return (void *)s->s_bdev == data;
1277 }
1278
1279 static struct dentry *
1280 nilfs_mount(struct file_system_type *fs_type, int flags,
1281              const char *dev_name, void *data)
1282 {
1283         struct nilfs_super_data sd;
1284         struct super_block *s;
1285         fmode_t mode = FMODE_READ | FMODE_EXCL;
1286         struct dentry *root_dentry;
1287         int err, s_new = false;
1288
1289         if (!(flags & SB_RDONLY))
1290                 mode |= FMODE_WRITE;
1291
1292         sd.bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1293         if (IS_ERR(sd.bdev))
1294                 return ERR_CAST(sd.bdev);
1295
1296         sd.cno = 0;
1297         sd.flags = flags;
1298         if (nilfs_identify((char *)data, &sd)) {
1299                 err = -EINVAL;
1300                 goto failed;
1301         }
1302
1303         /*
1304          * once the super is inserted into the list by sget, s_umount
1305          * will protect the lockfs code from trying to start a snapshot
1306          * while we are mounting
1307          */
1308         mutex_lock(&sd.bdev->bd_fsfreeze_mutex);
1309         if (sd.bdev->bd_fsfreeze_count > 0) {
1310                 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1311                 err = -EBUSY;
1312                 goto failed;
1313         }
1314         s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, flags,
1315                  sd.bdev);
1316         mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1317         if (IS_ERR(s)) {
1318                 err = PTR_ERR(s);
1319                 goto failed;
1320         }
1321
1322         if (!s->s_root) {
1323                 s_new = true;
1324
1325                 /* New superblock instance created */
1326                 s->s_mode = mode;
1327                 snprintf(s->s_id, sizeof(s->s_id), "%pg", sd.bdev);
1328                 sb_set_blocksize(s, block_size(sd.bdev));
1329
1330                 err = nilfs_fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1331                 if (err)
1332                         goto failed_super;
1333
1334                 s->s_flags |= SB_ACTIVE;
1335         } else if (!sd.cno) {
1336                 if (nilfs_tree_is_busy(s->s_root)) {
1337                         if ((flags ^ s->s_flags) & SB_RDONLY) {
1338                                 nilfs_msg(s, KERN_ERR,
1339                                           "the device already has a %s mount.",
1340                                           sb_rdonly(s) ? "read-only" : "read/write");
1341                                 err = -EBUSY;
1342                                 goto failed_super;
1343                         }
1344                 } else {
1345                         /*
1346                          * Try remount to setup mount states if the current
1347                          * tree is not mounted and only snapshots use this sb.
1348                          */
1349                         err = nilfs_remount(s, &flags, data);
1350                         if (err)
1351                                 goto failed_super;
1352                 }
1353         }
1354
1355         if (sd.cno) {
1356                 err = nilfs_attach_snapshot(s, sd.cno, &root_dentry);
1357                 if (err)
1358                         goto failed_super;
1359         } else {
1360                 root_dentry = dget(s->s_root);
1361         }
1362
1363         if (!s_new)
1364                 blkdev_put(sd.bdev, mode);
1365
1366         return root_dentry;
1367
1368  failed_super:
1369         deactivate_locked_super(s);
1370
1371  failed:
1372         if (!s_new)
1373                 blkdev_put(sd.bdev, mode);
1374         return ERR_PTR(err);
1375 }
1376
1377 struct file_system_type nilfs_fs_type = {
1378         .owner    = THIS_MODULE,
1379         .name     = "nilfs2",
1380         .mount    = nilfs_mount,
1381         .kill_sb  = kill_block_super,
1382         .fs_flags = FS_REQUIRES_DEV,
1383 };
1384 MODULE_ALIAS_FS("nilfs2");
1385
1386 static void nilfs_inode_init_once(void *obj)
1387 {
1388         struct nilfs_inode_info *ii = obj;
1389
1390         INIT_LIST_HEAD(&ii->i_dirty);
1391 #ifdef CONFIG_NILFS_XATTR
1392         init_rwsem(&ii->xattr_sem);
1393 #endif
1394         address_space_init_once(&ii->i_btnode_cache);
1395         ii->i_bmap = &ii->i_bmap_data;
1396         inode_init_once(&ii->vfs_inode);
1397 }
1398
1399 static void nilfs_segbuf_init_once(void *obj)
1400 {
1401         memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1402 }
1403
1404 static void nilfs_destroy_cachep(void)
1405 {
1406         /*
1407          * Make sure all delayed rcu free inodes are flushed before we
1408          * destroy cache.
1409          */
1410         rcu_barrier();
1411
1412         kmem_cache_destroy(nilfs_inode_cachep);
1413         kmem_cache_destroy(nilfs_transaction_cachep);
1414         kmem_cache_destroy(nilfs_segbuf_cachep);
1415         kmem_cache_destroy(nilfs_btree_path_cache);
1416 }
1417
1418 static int __init nilfs_init_cachep(void)
1419 {
1420         nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1421                         sizeof(struct nilfs_inode_info), 0,
1422                         SLAB_RECLAIM_ACCOUNT|SLAB_ACCOUNT,
1423                         nilfs_inode_init_once);
1424         if (!nilfs_inode_cachep)
1425                 goto fail;
1426
1427         nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1428                         sizeof(struct nilfs_transaction_info), 0,
1429                         SLAB_RECLAIM_ACCOUNT, NULL);
1430         if (!nilfs_transaction_cachep)
1431                 goto fail;
1432
1433         nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1434                         sizeof(struct nilfs_segment_buffer), 0,
1435                         SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1436         if (!nilfs_segbuf_cachep)
1437                 goto fail;
1438
1439         nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1440                         sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1441                         0, 0, NULL);
1442         if (!nilfs_btree_path_cache)
1443                 goto fail;
1444
1445         return 0;
1446
1447 fail:
1448         nilfs_destroy_cachep();
1449         return -ENOMEM;
1450 }
1451
1452 static int __init init_nilfs_fs(void)
1453 {
1454         int err;
1455
1456         err = nilfs_init_cachep();
1457         if (err)
1458                 goto fail;
1459
1460         err = nilfs_sysfs_init();
1461         if (err)
1462                 goto free_cachep;
1463
1464         err = register_filesystem(&nilfs_fs_type);
1465         if (err)
1466                 goto deinit_sysfs_entry;
1467
1468         printk(KERN_INFO "NILFS version 2 loaded\n");
1469         return 0;
1470
1471 deinit_sysfs_entry:
1472         nilfs_sysfs_exit();
1473 free_cachep:
1474         nilfs_destroy_cachep();
1475 fail:
1476         return err;
1477 }
1478
1479 static void __exit exit_nilfs_fs(void)
1480 {
1481         nilfs_destroy_cachep();
1482         nilfs_sysfs_exit();
1483         unregister_filesystem(&nilfs_fs_type);
1484 }
1485
1486 module_init(init_nilfs_fs)
1487 module_exit(exit_nilfs_fs)