fs/nilfs2/super.c

   1 // SPDX-License-Identifier: GPL-2.0+
   2 /*
   3  * super.c - NILFS module and super block management.
   4  *
   5  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
   6  *
   7  * Written by Ryusuke Konishi.
   8  */
   9 /*
  10  *  linux/fs/ext2/super.c
  11  *
  12  * Copyright (C) 1992, 1993, 1994, 1995
  13  * Remy Card (card@masi.ibp.fr)
  14  * Laboratoire MASI - Institut Blaise Pascal
  15  * Universite Pierre et Marie Curie (Paris VI)
  16  *
  17  *  from
  18  *
  19  *  linux/fs/minix/inode.c
  20  *
  21  *  Copyright (C) 1991, 1992  Linus Torvalds
  22  *
  23  *  Big-endian to little-endian byte-swapping/bitmaps by
  24  *        David S. Miller (davem@caip.rutgers.edu), 1995
  25  */
  26
  27 #include <linux/module.h>
  28 #include <linux/string.h>
  29 #include <linux/slab.h>
  30 #include <linux/init.h>
  31 #include <linux/blkdev.h>
  32 #include <linux/parser.h>
  33 #include <linux/crc32.h>
  34 #include <linux/vfs.h>
  35 #include <linux/writeback.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/mount.h>
  38 #include "nilfs.h"
  39 #include "export.h"
  40 #include "mdt.h"
  41 #include "alloc.h"
  42 #include "btree.h"
  43 #include "btnode.h"
  44 #include "page.h"
  45 #include "cpfile.h"
  46 #include "sufile.h" /* nilfs_sufile_resize(), nilfs_sufile_set_alloc_range() */
  47 #include "ifile.h"
  48 #include "dat.h"
  49 #include "segment.h"
  50 #include "segbuf.h"
  51
  52 MODULE_AUTHOR("NTT Corp.");
  53 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
  54                    "(NILFS)");
  55 MODULE_LICENSE("GPL");
  56
  57 static struct kmem_cache *nilfs_inode_cachep;
  58 struct kmem_cache *nilfs_transaction_cachep;
  59 struct kmem_cache *nilfs_segbuf_cachep;
  60 struct kmem_cache *nilfs_btree_path_cache;
  61
  62 static int nilfs_setup_super(struct super_block *sb, int is_mount);
  63 static int nilfs_remount(struct super_block *sb, int *flags, char *data);
  64
  65 void __nilfs_msg(struct super_block *sb, const char *fmt, ...)
  66 {
  67         struct va_format vaf;
  68         va_list args;
  69         int level;
  70
  71         va_start(args, fmt);
  72
  73         level = printk_get_level(fmt);
  74         vaf.fmt = printk_skip_level(fmt);
  75         vaf.va = &args;
  76
  77         if (sb)
  78                 printk("%c%cNILFS (%s): %pV\n",
  79                        KERN_SOH_ASCII, level, sb->s_id, &vaf);
  80         else
  81                 printk("%c%cNILFS: %pV\n",
  82                        KERN_SOH_ASCII, level, &vaf);
  83
  84         va_end(args);
  85 }
  86
  87 static void nilfs_set_error(struct super_block *sb)
  88 {
  89         struct the_nilfs *nilfs = sb->s_fs_info;
  90         struct nilfs_super_block **sbp;
  91
  92         down_write(&nilfs->ns_sem);
  93         if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
  94                 nilfs->ns_mount_state |= NILFS_ERROR_FS;
  95                 sbp = nilfs_prepare_super(sb, 0);
  96                 if (likely(sbp)) {
  97                         sbp[0]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
  98                         if (sbp[1])
  99                                 sbp[1]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
 100                         nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
 101                 }
 102         }
 103         up_write(&nilfs->ns_sem);
 104 }
 105
 106 /**
 107  * __nilfs_error() - report failure condition on a filesystem
 108  *
 109  * __nilfs_error() sets an ERROR_FS flag on the superblock as well as
 110  * reporting an error message.  This function should be called when
 111  * NILFS detects incoherences or defects of meta data on disk.
 112  *
 113  * This implements the body of nilfs_error() macro.  Normally,
 114  * nilfs_error() should be used.  As for sustainable errors such as a
 115  * single-shot I/O error, nilfs_err() should be used instead.
 116  *
 117  * Callers should not add a trailing newline since this will do it.
 118  */
 119 void __nilfs_error(struct super_block *sb, const char *function,
 120                    const char *fmt, ...)
 121 {
 122         struct the_nilfs *nilfs = sb->s_fs_info;
 123         struct va_format vaf;
 124         va_list args;
 125
 126         va_start(args, fmt);
 127
 128         vaf.fmt = fmt;
 129         vaf.va = &args;
 130
 131         printk(KERN_CRIT "NILFS error (device %s): %s: %pV\n",
 132                sb->s_id, function, &vaf);
 133
 134         va_end(args);
 135
 136         if (!sb_rdonly(sb)) {
 137                 nilfs_set_error(sb);
 138
 139                 if (nilfs_test_opt(nilfs, ERRORS_RO)) {
 140                         printk(KERN_CRIT "Remounting filesystem read-only\n");
 141                         sb->s_flags |= SB_RDONLY;
 142                 }
 143         }
 144
 145         if (nilfs_test_opt(nilfs, ERRORS_PANIC))
 146                 panic("NILFS (device %s): panic forced after error\n",
 147                       sb->s_id);
 148 }
 149
 150 struct inode *nilfs_alloc_inode(struct super_block *sb)
 151 {
 152         struct nilfs_inode_info *ii;
 153
 154         ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
 155         if (!ii)
 156                 return NULL;
 157         ii->i_bh = NULL;
 158         ii->i_state = 0;
 159         ii->i_cno = 0;
 160         nilfs_mapping_init(&ii->i_btnode_cache, &ii->vfs_inode);
 161         return &ii->vfs_inode;
 162 }
 163
 164 static void nilfs_free_inode(struct inode *inode)
 165 {
 166         if (nilfs_is_metadata_file_inode(inode))
 167                 nilfs_mdt_destroy(inode);
 168
 169         kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
 170 }
 171
 172 static int nilfs_sync_super(struct super_block *sb, int flag)
 173 {
 174         struct the_nilfs *nilfs = sb->s_fs_info;
 175         int err;
 176
 177  retry:
 178         set_buffer_dirty(nilfs->ns_sbh[0]);
 179         if (nilfs_test_opt(nilfs, BARRIER)) {
 180                 err = __sync_dirty_buffer(nilfs->ns_sbh[0],
 181                                           REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
 182         } else {
 183                 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
 184         }
 185
 186         if (unlikely(err)) {
 187                 nilfs_err(sb, "unable to write superblock: err=%d", err);
 188                 if (err == -EIO && nilfs->ns_sbh[1]) {
 189                         /*
 190                          * sbp[0] points to newer log than sbp[1],
 191                          * so copy sbp[0] to sbp[1] to take over sbp[0].
 192                          */
 193                         memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
 194                                nilfs->ns_sbsize);
 195                         nilfs_fall_back_super_block(nilfs);
 196                         goto retry;
 197                 }
 198         } else {
 199                 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
 200
 201                 nilfs->ns_sbwcount++;
 202
 203                 /*
 204                  * The latest segment becomes trailable from the position
 205                  * written in superblock.
 206                  */
 207                 clear_nilfs_discontinued(nilfs);
 208
 209                 /* update GC protection for recent segments */
 210                 if (nilfs->ns_sbh[1]) {
 211                         if (flag == NILFS_SB_COMMIT_ALL) {
 212                                 set_buffer_dirty(nilfs->ns_sbh[1]);
 213                                 if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
 214                                         goto out;
 215                         }
 216                         if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
 217                             le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
 218                                 sbp = nilfs->ns_sbp[1];
 219                 }
 220
 221                 spin_lock(&nilfs->ns_last_segment_lock);
 222                 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
 223                 spin_unlock(&nilfs->ns_last_segment_lock);
 224         }
 225  out:
 226         return err;
 227 }
 228
 229 void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
 230                           struct the_nilfs *nilfs)
 231 {
 232         sector_t nfreeblocks;
 233
 234         /* nilfs->ns_sem must be locked by the caller. */
 235         nilfs_count_free_blocks(nilfs, &nfreeblocks);
 236         sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
 237
 238         spin_lock(&nilfs->ns_last_segment_lock);
 239         sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
 240         sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
 241         sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
 242         spin_unlock(&nilfs->ns_last_segment_lock);
 243 }
 244
 245 struct nilfs_super_block **nilfs_prepare_super(struct super_block *sb,
 246                                                int flip)
 247 {
 248         struct the_nilfs *nilfs = sb->s_fs_info;
 249         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 250
 251         /* nilfs->ns_sem must be locked by the caller. */
 252         if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
 253                 if (sbp[1] &&
 254                     sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
 255                         memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
 256                 } else {
 257                         nilfs_crit(sb, "superblock broke");
 258                         return NULL;
 259                 }
 260         } else if (sbp[1] &&
 261                    sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
 262                 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 263         }
 264
 265         if (flip && sbp[1])
 266                 nilfs_swap_super_block(nilfs);
 267
 268         return sbp;
 269 }
 270
 271 int nilfs_commit_super(struct super_block *sb, int flag)
 272 {
 273         struct the_nilfs *nilfs = sb->s_fs_info;
 274         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 275         time64_t t;
 276
 277         /* nilfs->ns_sem must be locked by the caller. */
 278         t = ktime_get_real_seconds();
 279         nilfs->ns_sbwtime = t;
 280         sbp[0]->s_wtime = cpu_to_le64(t);
 281         sbp[0]->s_sum = 0;
 282         sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
 283                                              (unsigned char *)sbp[0],
 284                                              nilfs->ns_sbsize));
 285         if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
 286                 sbp[1]->s_wtime = sbp[0]->s_wtime;
 287                 sbp[1]->s_sum = 0;
 288                 sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
 289                                             (unsigned char *)sbp[1],
 290                                             nilfs->ns_sbsize));
 291         }
 292         clear_nilfs_sb_dirty(nilfs);
 293         nilfs->ns_flushed_device = 1;
 294         /* make sure store to ns_flushed_device cannot be reordered */
 295         smp_wmb();
 296         return nilfs_sync_super(sb, flag);
 297 }
 298
 299 /**
 300  * nilfs_cleanup_super() - write filesystem state for cleanup
 301  * @sb: super block instance to be unmounted or degraded to read-only
 302  *
 303  * This function restores state flags in the on-disk super block.
 304  * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
 305  * filesystem was not clean previously.
 306  */
 307 int nilfs_cleanup_super(struct super_block *sb)
 308 {
 309         struct the_nilfs *nilfs = sb->s_fs_info;
 310         struct nilfs_super_block **sbp;
 311         int flag = NILFS_SB_COMMIT;
 312         int ret = -EIO;
 313
 314         sbp = nilfs_prepare_super(sb, 0);
 315         if (sbp) {
 316                 sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
 317                 nilfs_set_log_cursor(sbp[0], nilfs);
 318                 if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
 319                         /*
 320                          * make the "clean" flag also to the opposite
 321                          * super block if both super blocks point to
 322                          * the same checkpoint.
 323                          */
 324                         sbp[1]->s_state = sbp[0]->s_state;
 325                         flag = NILFS_SB_COMMIT_ALL;
 326                 }
 327                 ret = nilfs_commit_super(sb, flag);
 328         }
 329         return ret;
 330 }
 331
 332 /**
 333  * nilfs_move_2nd_super - relocate secondary super block
 334  * @sb: super block instance
 335  * @sb2off: new offset of the secondary super block (in bytes)
 336  */
 337 static int nilfs_move_2nd_super(struct super_block *sb, loff_t sb2off)
 338 {
 339         struct the_nilfs *nilfs = sb->s_fs_info;
 340         struct buffer_head *nsbh;
 341         struct nilfs_super_block *nsbp;
 342         sector_t blocknr, newblocknr;
 343         unsigned long offset;
 344         int sb2i;  /* array index of the secondary superblock */
 345         int ret = 0;
 346
 347         /* nilfs->ns_sem must be locked by the caller. */
 348         if (nilfs->ns_sbh[1] &&
 349             nilfs->ns_sbh[1]->b_blocknr > nilfs->ns_first_data_block) {
 350                 sb2i = 1;
 351                 blocknr = nilfs->ns_sbh[1]->b_blocknr;
 352         } else if (nilfs->ns_sbh[0]->b_blocknr > nilfs->ns_first_data_block) {
 353                 sb2i = 0;
 354                 blocknr = nilfs->ns_sbh[0]->b_blocknr;
 355         } else {
 356                 sb2i = -1;
 357                 blocknr = 0;
 358         }
 359         if (sb2i >= 0 && (u64)blocknr << nilfs->ns_blocksize_bits == sb2off)
 360                 goto out;  /* super block location is unchanged */
 361
 362         /* Get new super block buffer */
 363         newblocknr = sb2off >> nilfs->ns_blocksize_bits;
 364         offset = sb2off & (nilfs->ns_blocksize - 1);
 365         nsbh = sb_getblk(sb, newblocknr);
 366         if (!nsbh) {
 367                 nilfs_warn(sb,
 368                            "unable to move secondary superblock to block %llu",
 369                            (unsigned long long)newblocknr);
 370                 ret = -EIO;
 371                 goto out;
 372         }
 373         nsbp = (void *)nsbh->b_data + offset;
 374         memset(nsbp, 0, nilfs->ns_blocksize);
 375
 376         if (sb2i >= 0) {
 377                 memcpy(nsbp, nilfs->ns_sbp[sb2i], nilfs->ns_sbsize);
 378                 brelse(nilfs->ns_sbh[sb2i]);
 379                 nilfs->ns_sbh[sb2i] = nsbh;
 380                 nilfs->ns_sbp[sb2i] = nsbp;
 381         } else if (nilfs->ns_sbh[0]->b_blocknr < nilfs->ns_first_data_block) {
 382                 /* secondary super block will be restored to index 1 */
 383                 nilfs->ns_sbh[1] = nsbh;
 384                 nilfs->ns_sbp[1] = nsbp;
 385         } else {
 386                 brelse(nsbh);
 387         }
 388 out:
 389         return ret;
 390 }
 391
 392 /**
 393  * nilfs_resize_fs - resize the filesystem
 394  * @sb: super block instance
 395  * @newsize: new size of the filesystem (in bytes)
 396  */
 397 int nilfs_resize_fs(struct super_block *sb, __u64 newsize)
 398 {
 399         struct the_nilfs *nilfs = sb->s_fs_info;
 400         struct nilfs_super_block **sbp;
 401         __u64 devsize, newnsegs;
 402         loff_t sb2off;
 403         int ret;
 404
 405         ret = -ERANGE;
 406         devsize = i_size_read(sb->s_bdev->bd_inode);
 407         if (newsize > devsize)
 408                 goto out;
 409
 410         /*
 411          * Write lock is required to protect some functions depending
 412          * on the number of segments, the number of reserved segments,
 413          * and so forth.
 414          */
 415         down_write(&nilfs->ns_segctor_sem);
 416
 417         sb2off = NILFS_SB2_OFFSET_BYTES(newsize);
 418         newnsegs = sb2off >> nilfs->ns_blocksize_bits;
 419         do_div(newnsegs, nilfs->ns_blocks_per_segment);
 420
 421         ret = nilfs_sufile_resize(nilfs->ns_sufile, newnsegs);
 422         up_write(&nilfs->ns_segctor_sem);
 423         if (ret < 0)
 424                 goto out;
 425
 426         ret = nilfs_construct_segment(sb);
 427         if (ret < 0)
 428                 goto out;
 429
 430         down_write(&nilfs->ns_sem);
 431         nilfs_move_2nd_super(sb, sb2off);
 432         ret = -EIO;
 433         sbp = nilfs_prepare_super(sb, 0);
 434         if (likely(sbp)) {
 435                 nilfs_set_log_cursor(sbp[0], nilfs);
 436                 /*
 437                  * Drop NILFS_RESIZE_FS flag for compatibility with
 438                  * mount-time resize which may be implemented in a
 439                  * future release.
 440                  */
 441                 sbp[0]->s_state = cpu_to_le16(le16_to_cpu(sbp[0]->s_state) &
 442                                               ~NILFS_RESIZE_FS);
 443                 sbp[0]->s_dev_size = cpu_to_le64(newsize);
 444                 sbp[0]->s_nsegments = cpu_to_le64(nilfs->ns_nsegments);
 445                 if (sbp[1])
 446                         memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 447                 ret = nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
 448         }
 449         up_write(&nilfs->ns_sem);
 450
 451         /*
 452          * Reset the range of allocatable segments last.  This order
 453          * is important in the case of expansion because the secondary
 454          * superblock must be protected from log write until migration
 455          * completes.
 456          */
 457         if (!ret)
 458                 nilfs_sufile_set_alloc_range(nilfs->ns_sufile, 0, newnsegs - 1);
 459 out:
 460         return ret;
 461 }
 462
 463 static void nilfs_put_super(struct super_block *sb)
 464 {
 465         struct the_nilfs *nilfs = sb->s_fs_info;
 466
 467         nilfs_detach_log_writer(sb);
 468
 469         if (!sb_rdonly(sb)) {
 470                 down_write(&nilfs->ns_sem);
 471                 nilfs_cleanup_super(sb);
 472                 up_write(&nilfs->ns_sem);
 473         }
 474
 475         iput(nilfs->ns_sufile);
 476         iput(nilfs->ns_cpfile);
 477         iput(nilfs->ns_dat);
 478
 479         destroy_nilfs(nilfs);
 480         sb->s_fs_info = NULL;
 481 }
 482
 483 static int nilfs_sync_fs(struct super_block *sb, int wait)
 484 {
 485         struct the_nilfs *nilfs = sb->s_fs_info;
 486         struct nilfs_super_block **sbp;
 487         int err = 0;
 488
 489         /* This function is called when super block should be written back */
 490         if (wait)
 491                 err = nilfs_construct_segment(sb);
 492
 493         down_write(&nilfs->ns_sem);
 494         if (nilfs_sb_dirty(nilfs)) {
 495                 sbp = nilfs_prepare_super(sb, nilfs_sb_will_flip(nilfs));
 496                 if (likely(sbp)) {
 497                         nilfs_set_log_cursor(sbp[0], nilfs);
 498                         nilfs_commit_super(sb, NILFS_SB_COMMIT);
 499                 }
 500         }
 501         up_write(&nilfs->ns_sem);
 502
 503         if (!err)
 504                 err = nilfs_flush_device(nilfs);
 505
 506         return err;
 507 }
 508
 509 int nilfs_attach_checkpoint(struct super_block *sb, __u64 cno, int curr_mnt,
 510                             struct nilfs_root **rootp)
 511 {
 512         struct the_nilfs *nilfs = sb->s_fs_info;
 513         struct nilfs_root *root;
 514         struct nilfs_checkpoint *raw_cp;
 515         struct buffer_head *bh_cp;
 516         int err = -ENOMEM;
 517
 518         root = nilfs_find_or_create_root(
 519                 nilfs, curr_mnt ? NILFS_CPTREE_CURRENT_CNO : cno);
 520         if (!root)
 521                 return err;
 522
 523         if (root->ifile)
 524                 goto reuse; /* already attached checkpoint */
 525
 526         down_read(&nilfs->ns_segctor_sem);
 527         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
 528                                           &bh_cp);
 529         up_read(&nilfs->ns_segctor_sem);
 530         if (unlikely(err)) {
 531                 if (err == -ENOENT || err == -EINVAL) {
 532                         nilfs_err(sb,
 533                                   "Invalid checkpoint (checkpoint number=%llu)",
 534                                   (unsigned long long)cno);
 535                         err = -EINVAL;
 536                 }
 537                 goto failed;
 538         }
 539
 540         err = nilfs_ifile_read(sb, root, nilfs->ns_inode_size,
 541                                &raw_cp->cp_ifile_inode, &root->ifile);
 542         if (err)
 543                 goto failed_bh;
 544
 545         atomic64_set(&root->inodes_count,
 546                         le64_to_cpu(raw_cp->cp_inodes_count));
 547         atomic64_set(&root->blocks_count,
 548                         le64_to_cpu(raw_cp->cp_blocks_count));
 549
 550         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
 551
 552  reuse:
 553         *rootp = root;
 554         return 0;
 555
 556  failed_bh:
 557         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
 558  failed:
 559         nilfs_put_root(root);
 560
 561         return err;
 562 }
 563
 564 static int nilfs_freeze(struct super_block *sb)
 565 {
 566         struct the_nilfs *nilfs = sb->s_fs_info;
 567         int err;
 568
 569         if (sb_rdonly(sb))
 570                 return 0;
 571
 572         /* Mark super block clean */
 573         down_write(&nilfs->ns_sem);
 574         err = nilfs_cleanup_super(sb);
 575         up_write(&nilfs->ns_sem);
 576         return err;
 577 }
 578
 579 static int nilfs_unfreeze(struct super_block *sb)
 580 {
 581         struct the_nilfs *nilfs = sb->s_fs_info;
 582
 583         if (sb_rdonly(sb))
 584                 return 0;
 585
 586         down_write(&nilfs->ns_sem);
 587         nilfs_setup_super(sb, false);
 588         up_write(&nilfs->ns_sem);
 589         return 0;
 590 }
 591
 592 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 593 {
 594         struct super_block *sb = dentry->d_sb;
 595         struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
 596         struct the_nilfs *nilfs = root->nilfs;
 597         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
 598         unsigned long long blocks;
 599         unsigned long overhead;
 600         unsigned long nrsvblocks;
 601         sector_t nfreeblocks;
 602         u64 nmaxinodes, nfreeinodes;
 603         int err;
 604
 605         /*
 606          * Compute all of the segment blocks
 607          *
 608          * The blocks before first segment and after last segment
 609          * are excluded.
 610          */
 611         blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
 612                 - nilfs->ns_first_data_block;
 613         nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
 614
 615         /*
 616          * Compute the overhead
 617          *
 618          * When distributing meta data blocks outside segment structure,
 619          * We must count them as the overhead.
 620          */
 621         overhead = 0;
 622
 623         err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
 624         if (unlikely(err))
 625                 return err;
 626
 627         err = nilfs_ifile_count_free_inodes(root->ifile,
 628                                             &nmaxinodes, &nfreeinodes);
 629         if (unlikely(err)) {
 630                 nilfs_warn(sb, "failed to count free inodes: err=%d", err);
 631                 if (err == -ERANGE) {
 632                         /*
 633                          * If nilfs_palloc_count_max_entries() returns
 634                          * -ERANGE error code then we simply treat
 635                          * curent inodes count as maximum possible and
 636                          * zero as free inodes value.
 637                          */
 638                         nmaxinodes = atomic64_read(&root->inodes_count);
 639                         nfreeinodes = 0;
 640                         err = 0;
 641                 } else
 642                         return err;
 643         }
 644
 645         buf->f_type = NILFS_SUPER_MAGIC;
 646         buf->f_bsize = sb->s_blocksize;
 647         buf->f_blocks = blocks - overhead;
 648         buf->f_bfree = nfreeblocks;
 649         buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
 650                 (buf->f_bfree - nrsvblocks) : 0;
 651         buf->f_files = nmaxinodes;
 652         buf->f_ffree = nfreeinodes;
 653         buf->f_namelen = NILFS_NAME_LEN;
 654         buf->f_fsid.val[0] = (u32)id;
 655         buf->f_fsid.val[1] = (u32)(id >> 32);
 656
 657         return 0;
 658 }
 659
 660 static int nilfs_show_options(struct seq_file *seq, struct dentry *dentry)
 661 {
 662         struct super_block *sb = dentry->d_sb;
 663         struct the_nilfs *nilfs = sb->s_fs_info;
 664         struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
 665
 666         if (!nilfs_test_opt(nilfs, BARRIER))
 667                 seq_puts(seq, ",nobarrier");
 668         if (root->cno != NILFS_CPTREE_CURRENT_CNO)
 669                 seq_printf(seq, ",cp=%llu", (unsigned long long)root->cno);
 670         if (nilfs_test_opt(nilfs, ERRORS_PANIC))
 671                 seq_puts(seq, ",errors=panic");
 672         if (nilfs_test_opt(nilfs, ERRORS_CONT))
 673                 seq_puts(seq, ",errors=continue");
 674         if (nilfs_test_opt(nilfs, STRICT_ORDER))
 675                 seq_puts(seq, ",order=strict");
 676         if (nilfs_test_opt(nilfs, NORECOVERY))
 677                 seq_puts(seq, ",norecovery");
 678         if (nilfs_test_opt(nilfs, DISCARD))
 679                 seq_puts(seq, ",discard");
 680
 681         return 0;
 682 }
 683
 684 static const struct super_operations nilfs_sops = {
 685         .alloc_inode    = nilfs_alloc_inode,
 686         .free_inode     = nilfs_free_inode,
 687         .dirty_inode    = nilfs_dirty_inode,
 688         .evict_inode    = nilfs_evict_inode,
 689         .put_super      = nilfs_put_super,
 690         .sync_fs        = nilfs_sync_fs,
 691         .freeze_fs      = nilfs_freeze,
 692         .unfreeze_fs    = nilfs_unfreeze,
 693         .statfs         = nilfs_statfs,
 694         .remount_fs     = nilfs_remount,
 695         .show_options = nilfs_show_options
 696 };
 697
 698 enum {
 699         Opt_err_cont, Opt_err_panic, Opt_err_ro,
 700         Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
 701         Opt_discard, Opt_nodiscard, Opt_err,
 702 };
 703
 704 static match_table_t tokens = {
 705         {Opt_err_cont, "errors=continue"},
 706         {Opt_err_panic, "errors=panic"},
 707         {Opt_err_ro, "errors=remount-ro"},
 708         {Opt_barrier, "barrier"},
 709         {Opt_nobarrier, "nobarrier"},
 710         {Opt_snapshot, "cp=%u"},
 711         {Opt_order, "order=%s"},
 712         {Opt_norecovery, "norecovery"},
 713         {Opt_discard, "discard"},
 714         {Opt_nodiscard, "nodiscard"},
 715         {Opt_err, NULL}
 716 };
 717
 718 static int parse_options(char *options, struct super_block *sb, int is_remount)
 719 {
 720         struct the_nilfs *nilfs = sb->s_fs_info;
 721         char *p;
 722         substring_t args[MAX_OPT_ARGS];
 723
 724         if (!options)
 725                 return 1;
 726
 727         while ((p = strsep(&options, ",")) != NULL) {
 728                 int token;
 729
 730                 if (!*p)
 731                         continue;
 732
 733                 token = match_token(p, tokens, args);
 734                 switch (token) {
 735                 case Opt_barrier:
 736                         nilfs_set_opt(nilfs, BARRIER);
 737                         break;
 738                 case Opt_nobarrier:
 739                         nilfs_clear_opt(nilfs, BARRIER);
 740                         break;
 741                 case Opt_order:
 742                         if (strcmp(args[0].from, "relaxed") == 0)
 743                                 /* Ordered data semantics */
 744                                 nilfs_clear_opt(nilfs, STRICT_ORDER);
 745                         else if (strcmp(args[0].from, "strict") == 0)
 746                                 /* Strict in-order semantics */
 747                                 nilfs_set_opt(nilfs, STRICT_ORDER);
 748                         else
 749                                 return 0;
 750                         break;
 751                 case Opt_err_panic:
 752                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_PANIC);
 753                         break;
 754                 case Opt_err_ro:
 755                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_RO);
 756                         break;
 757                 case Opt_err_cont:
 758                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_CONT);
 759                         break;
 760                 case Opt_snapshot:
 761                         if (is_remount) {
 762                                 nilfs_err(sb,
 763                                           "\"%s\" option is invalid for remount",
 764                                           p);
 765                                 return 0;
 766                         }
 767                         break;
 768                 case Opt_norecovery:
 769                         nilfs_set_opt(nilfs, NORECOVERY);
 770                         break;
 771                 case Opt_discard:
 772                         nilfs_set_opt(nilfs, DISCARD);
 773                         break;
 774                 case Opt_nodiscard:
 775                         nilfs_clear_opt(nilfs, DISCARD);
 776                         break;
 777                 default:
 778                         nilfs_err(sb, "unrecognized mount option \"%s\"", p);
 779                         return 0;
 780                 }
 781         }
 782         return 1;
 783 }
 784
 785 static inline void
 786 nilfs_set_default_options(struct super_block *sb,
 787                           struct nilfs_super_block *sbp)
 788 {
 789         struct the_nilfs *nilfs = sb->s_fs_info;
 790
 791         nilfs->ns_mount_opt =
 792                 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
 793 }
 794
 795 static int nilfs_setup_super(struct super_block *sb, int is_mount)
 796 {
 797         struct the_nilfs *nilfs = sb->s_fs_info;
 798         struct nilfs_super_block **sbp;
 799         int max_mnt_count;
 800         int mnt_count;
 801
 802         /* nilfs->ns_sem must be locked by the caller. */
 803         sbp = nilfs_prepare_super(sb, 0);
 804         if (!sbp)
 805                 return -EIO;
 806
 807         if (!is_mount)
 808                 goto skip_mount_setup;
 809
 810         max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
 811         mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
 812
 813         if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
 814                 nilfs_warn(sb, "mounting fs with errors");
 815 #if 0
 816         } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
 817                 nilfs_warn(sb, "maximal mount count reached");
 818 #endif
 819         }
 820         if (!max_mnt_count)
 821                 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
 822
 823         sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
 824         sbp[0]->s_mtime = cpu_to_le64(ktime_get_real_seconds());
 825
 826 skip_mount_setup:
 827         sbp[0]->s_state =
 828                 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
 829         /* synchronize sbp[1] with sbp[0] */
 830         if (sbp[1])
 831                 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 832         return nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
 833 }
 834
 835 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
 836                                                  u64 pos, int blocksize,
 837                                                  struct buffer_head **pbh)
 838 {
 839         unsigned long long sb_index = pos;
 840         unsigned long offset;
 841
 842         offset = do_div(sb_index, blocksize);
 843         *pbh = sb_bread(sb, sb_index);
 844         if (!*pbh)
 845                 return NULL;
 846         return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
 847 }
 848
 849 int nilfs_store_magic_and_option(struct super_block *sb,
 850                                  struct nilfs_super_block *sbp,
 851                                  char *data)
 852 {
 853         struct the_nilfs *nilfs = sb->s_fs_info;
 854
 855         sb->s_magic = le16_to_cpu(sbp->s_magic);
 856
 857         /* FS independent flags */
 858 #ifdef NILFS_ATIME_DISABLE
 859         sb->s_flags |= SB_NOATIME;
 860 #endif
 861
 862         nilfs_set_default_options(sb, sbp);
 863
 864         nilfs->ns_resuid = le16_to_cpu(sbp->s_def_resuid);
 865         nilfs->ns_resgid = le16_to_cpu(sbp->s_def_resgid);
 866         nilfs->ns_interval = le32_to_cpu(sbp->s_c_interval);
 867         nilfs->ns_watermark = le32_to_cpu(sbp->s_c_block_max);
 868
 869         return !parse_options(data, sb, 0) ? -EINVAL : 0;
 870 }
 871
 872 int nilfs_check_feature_compatibility(struct super_block *sb,
 873                                       struct nilfs_super_block *sbp)
 874 {
 875         __u64 features;
 876
 877         features = le64_to_cpu(sbp->s_feature_incompat) &
 878                 ~NILFS_FEATURE_INCOMPAT_SUPP;
 879         if (features) {
 880                 nilfs_err(sb,
 881                           "couldn't mount because of unsupported optional features (%llx)",
 882                           (unsigned long long)features);
 883                 return -EINVAL;
 884         }
 885         features = le64_to_cpu(sbp->s_feature_compat_ro) &
 886                 ~NILFS_FEATURE_COMPAT_RO_SUPP;
 887         if (!sb_rdonly(sb) && features) {
 888                 nilfs_err(sb,
 889                           "couldn't mount RDWR because of unsupported optional features (%llx)",
 890                           (unsigned long long)features);
 891                 return -EINVAL;
 892         }
 893         return 0;
 894 }
 895
 896 static int nilfs_get_root_dentry(struct super_block *sb,
 897                                  struct nilfs_root *root,
 898                                  struct dentry **root_dentry)
 899 {
 900         struct inode *inode;
 901         struct dentry *dentry;
 902         int ret = 0;
 903
 904         inode = nilfs_iget(sb, root, NILFS_ROOT_INO);
 905         if (IS_ERR(inode)) {
 906                 ret = PTR_ERR(inode);
 907                 nilfs_err(sb, "error %d getting root inode", ret);
 908                 goto out;
 909         }
 910         if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
 911                 iput(inode);
 912                 nilfs_err(sb, "corrupt root inode");
 913                 ret = -EINVAL;
 914                 goto out;
 915         }
 916
 917         if (root->cno == NILFS_CPTREE_CURRENT_CNO) {
 918                 dentry = d_find_alias(inode);
 919                 if (!dentry) {
 920                         dentry = d_make_root(inode);
 921                         if (!dentry) {
 922                                 ret = -ENOMEM;
 923                                 goto failed_dentry;
 924                         }
 925                 } else {
 926                         iput(inode);
 927                 }
 928         } else {
 929                 dentry = d_obtain_root(inode);
 930                 if (IS_ERR(dentry)) {
 931                         ret = PTR_ERR(dentry);
 932                         goto failed_dentry;
 933                 }
 934         }
 935         *root_dentry = dentry;
 936  out:
 937         return ret;
 938
 939  failed_dentry:
 940         nilfs_err(sb, "error %d getting root dentry", ret);
 941         goto out;
 942 }
 943
 944 static int nilfs_attach_snapshot(struct super_block *s, __u64 cno,
 945                                  struct dentry **root_dentry)
 946 {
 947         struct the_nilfs *nilfs = s->s_fs_info;
 948         struct nilfs_root *root;
 949         int ret;
 950
 951         mutex_lock(&nilfs->ns_snapshot_mount_mutex);
 952
 953         down_read(&nilfs->ns_segctor_sem);
 954         ret = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, cno);
 955         up_read(&nilfs->ns_segctor_sem);
 956         if (ret < 0) {
 957                 ret = (ret == -ENOENT) ? -EINVAL : ret;
 958                 goto out;
 959         } else if (!ret) {
 960                 nilfs_err(s,
 961                           "The specified checkpoint is not a snapshot (checkpoint number=%llu)",
 962                           (unsigned long long)cno);
 963                 ret = -EINVAL;
 964                 goto out;
 965         }
 966
 967         ret = nilfs_attach_checkpoint(s, cno, false, &root);
 968         if (ret) {
 969                 nilfs_err(s,
 970                           "error %d while loading snapshot (checkpoint number=%llu)",
 971                           ret, (unsigned long long)cno);
 972                 goto out;
 973         }
 974         ret = nilfs_get_root_dentry(s, root, root_dentry);
 975         nilfs_put_root(root);
 976  out:
 977         mutex_unlock(&nilfs->ns_snapshot_mount_mutex);
 978         return ret;
 979 }
 980
 981 /**
 982  * nilfs_tree_is_busy() - try to shrink dentries of a checkpoint
 983  * @root_dentry: root dentry of the tree to be shrunk
 984  *
 985  * This function returns true if the tree was in-use.
 986  */
 987 static bool nilfs_tree_is_busy(struct dentry *root_dentry)
 988 {
 989         shrink_dcache_parent(root_dentry);
 990         return d_count(root_dentry) > 1;
 991 }
 992
 993 int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno)
 994 {
 995         struct the_nilfs *nilfs = sb->s_fs_info;
 996         struct nilfs_root *root;
 997         struct inode *inode;
 998         struct dentry *dentry;
 999         int ret;
1000
1001         if (cno > nilfs->ns_cno)
1002                 return false;
1003
1004         if (cno >= nilfs_last_cno(nilfs))
1005                 return true;    /* protect recent checkpoints */
1006
1007         ret = false;
1008         root = nilfs_lookup_root(nilfs, cno);
1009         if (root) {
1010                 inode = nilfs_ilookup(sb, root, NILFS_ROOT_INO);
1011                 if (inode) {
1012                         dentry = d_find_alias(inode);
1013                         if (dentry) {
1014                                 ret = nilfs_tree_is_busy(dentry);
1015                                 dput(dentry);
1016                         }
1017                         iput(inode);
1018                 }
1019                 nilfs_put_root(root);
1020         }
1021         return ret;
1022 }
1023
1024 /**
1025  * nilfs_fill_super() - initialize a super block instance
1026  * @sb: super_block
1027  * @data: mount options
1028  * @silent: silent mode flag
1029  *
1030  * This function is called exclusively by nilfs->ns_mount_mutex.
1031  * So, the recovery process is protected from other simultaneous mounts.
1032  */
1033 static int
1034 nilfs_fill_super(struct super_block *sb, void *data, int silent)
1035 {
1036         struct the_nilfs *nilfs;
1037         struct nilfs_root *fsroot;
1038         __u64 cno;
1039         int err;
1040
1041         nilfs = alloc_nilfs(sb);
1042         if (!nilfs)
1043                 return -ENOMEM;
1044
1045         sb->s_fs_info = nilfs;
1046
1047         err = init_nilfs(nilfs, sb, (char *)data);
1048         if (err)
1049                 goto failed_nilfs;
1050
1051         sb->s_op = &nilfs_sops;
1052         sb->s_export_op = &nilfs_export_ops;
1053         sb->s_root = NULL;
1054         sb->s_time_gran = 1;
1055         sb->s_max_links = NILFS_LINK_MAX;
1056
1057         sb->s_bdi = bdi_get(sb->s_bdev->bd_bdi);
1058
1059         err = load_nilfs(nilfs, sb);
1060         if (err)
1061                 goto failed_nilfs;
1062
1063         cno = nilfs_last_cno(nilfs);
1064         err = nilfs_attach_checkpoint(sb, cno, true, &fsroot);
1065         if (err) {
1066                 nilfs_err(sb,
1067                           "error %d while loading last checkpoint (checkpoint number=%llu)",
1068                           err, (unsigned long long)cno);
1069                 goto failed_unload;
1070         }
1071
1072         if (!sb_rdonly(sb)) {
1073                 err = nilfs_attach_log_writer(sb, fsroot);
1074                 if (err)
1075                         goto failed_checkpoint;
1076         }
1077
1078         err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
1079         if (err)
1080                 goto failed_segctor;
1081
1082         nilfs_put_root(fsroot);
1083
1084         if (!sb_rdonly(sb)) {
1085                 down_write(&nilfs->ns_sem);
1086                 nilfs_setup_super(sb, true);
1087                 up_write(&nilfs->ns_sem);
1088         }
1089
1090         return 0;
1091
1092  failed_segctor:
1093         nilfs_detach_log_writer(sb);
1094
1095  failed_checkpoint:
1096         nilfs_put_root(fsroot);
1097
1098  failed_unload:
1099         iput(nilfs->ns_sufile);
1100         iput(nilfs->ns_cpfile);
1101         iput(nilfs->ns_dat);
1102
1103  failed_nilfs:
1104         destroy_nilfs(nilfs);
1105         return err;
1106 }
1107
1108 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
1109 {
1110         struct the_nilfs *nilfs = sb->s_fs_info;
1111         unsigned long old_sb_flags;
1112         unsigned long old_mount_opt;
1113         int err;
1114
1115         sync_filesystem(sb);
1116         old_sb_flags = sb->s_flags;
1117         old_mount_opt = nilfs->ns_mount_opt;
1118
1119         if (!parse_options(data, sb, 1)) {
1120                 err = -EINVAL;
1121                 goto restore_opts;
1122         }
1123         sb->s_flags = (sb->s_flags & ~SB_POSIXACL);
1124
1125         err = -EINVAL;
1126
1127         if (!nilfs_valid_fs(nilfs)) {
1128                 nilfs_warn(sb,
1129                            "couldn't remount because the filesystem is in an incomplete recovery state");
1130                 goto restore_opts;
1131         }
1132
1133         if ((bool)(*flags & SB_RDONLY) == sb_rdonly(sb))
1134                 goto out;
1135         if (*flags & SB_RDONLY) {
1136                 /* Shutting down log writer */
1137                 nilfs_detach_log_writer(sb);
1138                 sb->s_flags |= SB_RDONLY;
1139
1140                 /*
1141                  * Remounting a valid RW partition RDONLY, so set
1142                  * the RDONLY flag and then mark the partition as valid again.
1143                  */
1144                 down_write(&nilfs->ns_sem);
1145                 nilfs_cleanup_super(sb);
1146                 up_write(&nilfs->ns_sem);
1147         } else {
1148                 __u64 features;
1149                 struct nilfs_root *root;
1150
1151                 /*
1152                  * Mounting a RDONLY partition read-write, so reread and
1153                  * store the current valid flag.  (It may have been changed
1154                  * by fsck since we originally mounted the partition.)
1155                  */
1156                 down_read(&nilfs->ns_sem);
1157                 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1158                         ~NILFS_FEATURE_COMPAT_RO_SUPP;
1159                 up_read(&nilfs->ns_sem);
1160                 if (features) {
1161                         nilfs_warn(sb,
1162                                    "couldn't remount RDWR because of unsupported optional features (%llx)",
1163                                    (unsigned long long)features);
1164                         err = -EROFS;
1165                         goto restore_opts;
1166                 }
1167
1168                 sb->s_flags &= ~SB_RDONLY;
1169
1170                 root = NILFS_I(d_inode(sb->s_root))->i_root;
1171                 err = nilfs_attach_log_writer(sb, root);
1172                 if (err)
1173                         goto restore_opts;
1174
1175                 down_write(&nilfs->ns_sem);
1176                 nilfs_setup_super(sb, true);
1177                 up_write(&nilfs->ns_sem);
1178         }
1179  out:
1180         return 0;
1181
1182  restore_opts:
1183         sb->s_flags = old_sb_flags;
1184         nilfs->ns_mount_opt = old_mount_opt;
1185         return err;
1186 }
1187
1188 struct nilfs_super_data {
1189         struct block_device *bdev;
1190         __u64 cno;
1191         int flags;
1192 };
1193
1194 static int nilfs_parse_snapshot_option(const char *option,
1195                                        const substring_t *arg,
1196                                        struct nilfs_super_data *sd)
1197 {
1198         unsigned long long val;
1199         const char *msg = NULL;
1200         int err;
1201
1202         if (!(sd->flags & SB_RDONLY)) {
1203                 msg = "read-only option is not specified";
1204                 goto parse_error;
1205         }
1206
1207         err = kstrtoull(arg->from, 0, &val);
1208         if (err) {
1209                 if (err == -ERANGE)
1210                         msg = "too large checkpoint number";
1211                 else
1212                         msg = "malformed argument";
1213                 goto parse_error;
1214         } else if (val == 0) {
1215                 msg = "invalid checkpoint number 0";
1216                 goto parse_error;
1217         }
1218         sd->cno = val;
1219         return 0;
1220
1221 parse_error:
1222         nilfs_err(NULL, "invalid option \"%s\": %s", option, msg);
1223         return 1;
1224 }
1225
1226 /**
1227  * nilfs_identify - pre-read mount options needed to identify mount instance
1228  * @data: mount options
1229  * @sd: nilfs_super_data
1230  */
1231 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1232 {
1233         char *p, *options = data;
1234         substring_t args[MAX_OPT_ARGS];
1235         int token;
1236         int ret = 0;
1237
1238         do {
1239                 p = strsep(&options, ",");
1240                 if (p != NULL && *p) {
1241                         token = match_token(p, tokens, args);
1242                         if (token == Opt_snapshot)
1243                                 ret = nilfs_parse_snapshot_option(p, &args[0],
1244                                                                   sd);
1245                 }
1246                 if (!options)
1247                         break;
1248                 BUG_ON(options == data);
1249                 *(options - 1) = ',';
1250         } while (!ret);
1251         return ret;
1252 }
1253
1254 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1255 {
1256         s->s_bdev = data;
1257         s->s_dev = s->s_bdev->bd_dev;
1258         return 0;
1259 }
1260
1261 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1262 {
1263         return (void *)s->s_bdev == data;
1264 }
1265
1266 static struct dentry *
1267 nilfs_mount(struct file_system_type *fs_type, int flags,
1268              const char *dev_name, void *data)
1269 {
1270         struct nilfs_super_data sd;
1271         struct super_block *s;
1272         fmode_t mode = FMODE_READ | FMODE_EXCL;
1273         struct dentry *root_dentry;
1274         int err, s_new = false;
1275
1276         if (!(flags & SB_RDONLY))
1277                 mode |= FMODE_WRITE;
1278
1279         sd.bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1280         if (IS_ERR(sd.bdev))
1281                 return ERR_CAST(sd.bdev);
1282
1283         sd.cno = 0;
1284         sd.flags = flags;
1285         if (nilfs_identify((char *)data, &sd)) {
1286                 err = -EINVAL;
1287                 goto failed;
1288         }
1289
1290         /*
1291          * once the super is inserted into the list by sget, s_umount
1292          * will protect the lockfs code from trying to start a snapshot
1293          * while we are mounting
1294          */
1295         mutex_lock(&sd.bdev->bd_fsfreeze_mutex);
1296         if (sd.bdev->bd_fsfreeze_count > 0) {
1297                 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1298                 err = -EBUSY;
1299                 goto failed;
1300         }
1301         s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, flags,
1302                  sd.bdev);
1303         mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1304         if (IS_ERR(s)) {
1305                 err = PTR_ERR(s);
1306                 goto failed;
1307         }
1308
1309         if (!s->s_root) {
1310                 s_new = true;
1311
1312                 /* New superblock instance created */
1313                 s->s_mode = mode;
1314                 snprintf(s->s_id, sizeof(s->s_id), "%pg", sd.bdev);
1315                 sb_set_blocksize(s, block_size(sd.bdev));
1316
1317                 err = nilfs_fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1318                 if (err)
1319                         goto failed_super;
1320
1321                 s->s_flags |= SB_ACTIVE;
1322         } else if (!sd.cno) {
1323                 if (nilfs_tree_is_busy(s->s_root)) {
1324                         if ((flags ^ s->s_flags) & SB_RDONLY) {
1325                                 nilfs_err(s,
1326                                           "the device already has a %s mount.",
1327                                           sb_rdonly(s) ? "read-only" : "read/write");
1328                                 err = -EBUSY;
1329                                 goto failed_super;
1330                         }
1331                 } else {
1332                         /*
1333                          * Try remount to setup mount states if the current
1334                          * tree is not mounted and only snapshots use this sb.
1335                          */
1336                         err = nilfs_remount(s, &flags, data);
1337                         if (err)
1338                                 goto failed_super;
1339                 }
1340         }
1341
1342         if (sd.cno) {
1343                 err = nilfs_attach_snapshot(s, sd.cno, &root_dentry);
1344                 if (err)
1345                         goto failed_super;
1346         } else {
1347                 root_dentry = dget(s->s_root);
1348         }
1349
1350         if (!s_new)
1351                 blkdev_put(sd.bdev, mode);
1352
1353         return root_dentry;
1354
1355  failed_super:
1356         deactivate_locked_super(s);
1357
1358  failed:
1359         if (!s_new)
1360                 blkdev_put(sd.bdev, mode);
1361         return ERR_PTR(err);
1362 }
1363
1364 struct file_system_type nilfs_fs_type = {
1365         .owner    = THIS_MODULE,
1366         .name     = "nilfs2",
1367         .mount    = nilfs_mount,
1368         .kill_sb  = kill_block_super,
1369         .fs_flags = FS_REQUIRES_DEV,
1370 };
1371 MODULE_ALIAS_FS("nilfs2");
1372
1373 static void nilfs_inode_init_once(void *obj)
1374 {
1375         struct nilfs_inode_info *ii = obj;
1376
1377         INIT_LIST_HEAD(&ii->i_dirty);
1378 #ifdef CONFIG_NILFS_XATTR
1379         init_rwsem(&ii->xattr_sem);
1380 #endif
1381         address_space_init_once(&ii->i_btnode_cache);
1382         ii->i_bmap = &ii->i_bmap_data;
1383         inode_init_once(&ii->vfs_inode);
1384 }
1385
1386 static void nilfs_segbuf_init_once(void *obj)
1387 {
1388         memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1389 }
1390
1391 static void nilfs_destroy_cachep(void)
1392 {
1393         /*
1394          * Make sure all delayed rcu free inodes are flushed before we
1395          * destroy cache.
1396          */
1397         rcu_barrier();
1398
1399         kmem_cache_destroy(nilfs_inode_cachep);
1400         kmem_cache_destroy(nilfs_transaction_cachep);
1401         kmem_cache_destroy(nilfs_segbuf_cachep);
1402         kmem_cache_destroy(nilfs_btree_path_cache);
1403 }
1404
1405 static int __init nilfs_init_cachep(void)
1406 {
1407         nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1408                         sizeof(struct nilfs_inode_info), 0,
1409                         SLAB_RECLAIM_ACCOUNT|SLAB_ACCOUNT,
1410                         nilfs_inode_init_once);
1411         if (!nilfs_inode_cachep)
1412                 goto fail;
1413
1414         nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1415                         sizeof(struct nilfs_transaction_info), 0,
1416                         SLAB_RECLAIM_ACCOUNT, NULL);
1417         if (!nilfs_transaction_cachep)
1418                 goto fail;
1419
1420         nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1421                         sizeof(struct nilfs_segment_buffer), 0,
1422                         SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1423         if (!nilfs_segbuf_cachep)
1424                 goto fail;
1425
1426         nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1427                         sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1428                         0, 0, NULL);
1429         if (!nilfs_btree_path_cache)
1430                 goto fail;
1431
1432         return 0;
1433
1434 fail:
1435         nilfs_destroy_cachep();
1436         return -ENOMEM;
1437 }
1438
1439 static int __init init_nilfs_fs(void)
1440 {
1441         int err;
1442
1443         err = nilfs_init_cachep();
1444         if (err)
1445                 goto fail;
1446
1447         err = nilfs_sysfs_init();
1448         if (err)
1449                 goto free_cachep;
1450
1451         err = register_filesystem(&nilfs_fs_type);
1452         if (err)
1453                 goto deinit_sysfs_entry;
1454
1455         printk(KERN_INFO "NILFS version 2 loaded\n");
1456         return 0;
1457
1458 deinit_sysfs_entry:
1459         nilfs_sysfs_exit();
1460 free_cachep:
1461         nilfs_destroy_cachep();
1462 fail:
1463         return err;
1464 }
1465
1466 static void __exit exit_nilfs_fs(void)
1467 {
1468         nilfs_destroy_cachep();
1469         nilfs_sysfs_exit();
1470         unregister_filesystem(&nilfs_fs_type);
1471 }
1472
1473 module_init(init_nilfs_fs)
1474 module_exit(exit_nilfs_fs)