fs/nilfs2/the_nilfs.c

   1 /*
   2  * the_nilfs.c - the_nilfs shared structure.
   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  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  19  *
  20  * Written by Ryusuke Konishi <ryusuke@osrg.net>
  21  *
  22  */
  23
  24 #include <linux/buffer_head.h>
  25 #include <linux/slab.h>
  26 #include <linux/blkdev.h>
  27 #include <linux/backing-dev.h>
  28 #include "nilfs.h"
  29 #include "segment.h"
  30 #include "alloc.h"
  31 #include "cpfile.h"
  32 #include "sufile.h"
  33 #include "dat.h"
  34 #include "seglist.h"
  35 #include "segbuf.h"
  36
  37 void nilfs_set_last_segment(struct the_nilfs *nilfs,
  38                             sector_t start_blocknr, u64 seq, __u64 cno)
  39 {
  40         spin_lock(&nilfs->ns_last_segment_lock);
  41         nilfs->ns_last_pseg = start_blocknr;
  42         nilfs->ns_last_seq = seq;
  43         nilfs->ns_last_cno = cno;
  44         spin_unlock(&nilfs->ns_last_segment_lock);
  45 }
  46
  47 /**
  48  * alloc_nilfs - allocate the_nilfs structure
  49  * @bdev: block device to which the_nilfs is related
  50  *
  51  * alloc_nilfs() allocates memory for the_nilfs and
  52  * initializes its reference count and locks.
  53  *
  54  * Return Value: On success, pointer to the_nilfs is returned.
  55  * On error, NULL is returned.
  56  */
  57 struct the_nilfs *alloc_nilfs(struct block_device *bdev)
  58 {
  59         struct the_nilfs *nilfs;
  60
  61         nilfs = kzalloc(sizeof(*nilfs), GFP_KERNEL);
  62         if (!nilfs)
  63                 return NULL;
  64
  65         nilfs->ns_bdev = bdev;
  66         atomic_set(&nilfs->ns_count, 1);
  67         atomic_set(&nilfs->ns_writer_refcount, -1);
  68         atomic_set(&nilfs->ns_ndirtyblks, 0);
  69         init_rwsem(&nilfs->ns_sem);
  70         mutex_init(&nilfs->ns_writer_mutex);
  71         INIT_LIST_HEAD(&nilfs->ns_supers);
  72         spin_lock_init(&nilfs->ns_last_segment_lock);
  73         nilfs->ns_gc_inodes_h = NULL;
  74         INIT_LIST_HEAD(&nilfs->ns_used_segments);
  75         init_rwsem(&nilfs->ns_segctor_sem);
  76
  77         return nilfs;
  78 }
  79
  80 /**
  81  * put_nilfs - release a reference to the_nilfs
  82  * @nilfs: the_nilfs structure to be released
  83  *
  84  * put_nilfs() decrements a reference counter of the_nilfs.
  85  * If the reference count reaches zero, the_nilfs is freed.
  86  */
  87 void put_nilfs(struct the_nilfs *nilfs)
  88 {
  89         if (!atomic_dec_and_test(&nilfs->ns_count))
  90                 return;
  91         /*
  92          * Increment of ns_count never occur below because the caller
  93          * of get_nilfs() holds at least one reference to the_nilfs.
  94          * Thus its exclusion control is not required here.
  95          */
  96         might_sleep();
  97         if (nilfs_loaded(nilfs)) {
  98                 nilfs_dispose_used_segments(nilfs);
  99                 nilfs_mdt_clear(nilfs->ns_sufile);
 100                 nilfs_mdt_destroy(nilfs->ns_sufile);
 101                 nilfs_mdt_clear(nilfs->ns_cpfile);
 102                 nilfs_mdt_destroy(nilfs->ns_cpfile);
 103                 nilfs_mdt_clear(nilfs->ns_dat);
 104                 nilfs_mdt_destroy(nilfs->ns_dat);
 105                 /* XXX: how and when to clear nilfs->ns_gc_dat? */
 106                 nilfs_mdt_destroy(nilfs->ns_gc_dat);
 107         }
 108         if (nilfs_init(nilfs)) {
 109                 nilfs_destroy_gccache(nilfs);
 110                 brelse(nilfs->ns_sbh);
 111         }
 112         kfree(nilfs);
 113 }
 114
 115 static int nilfs_load_super_root(struct the_nilfs *nilfs,
 116                                  struct nilfs_sb_info *sbi, sector_t sr_block)
 117 {
 118         struct buffer_head *bh_sr;
 119         struct nilfs_super_root *raw_sr;
 120         unsigned dat_entry_size, segment_usage_size, checkpoint_size;
 121         unsigned inode_size;
 122         int err;
 123
 124         err = nilfs_read_super_root_block(sbi->s_super, sr_block, &bh_sr, 1);
 125         if (unlikely(err))
 126                 return err;
 127
 128         down_read(&nilfs->ns_sem);
 129         dat_entry_size = le16_to_cpu(nilfs->ns_sbp->s_dat_entry_size);
 130         checkpoint_size = le16_to_cpu(nilfs->ns_sbp->s_checkpoint_size);
 131         segment_usage_size = le16_to_cpu(nilfs->ns_sbp->s_segment_usage_size);
 132         up_read(&nilfs->ns_sem);
 133
 134         inode_size = nilfs->ns_inode_size;
 135
 136         err = -ENOMEM;
 137         nilfs->ns_dat = nilfs_mdt_new(
 138                 nilfs, NULL, NILFS_DAT_INO, NILFS_DAT_GFP);
 139         if (unlikely(!nilfs->ns_dat))
 140                 goto failed;
 141
 142         nilfs->ns_gc_dat = nilfs_mdt_new(
 143                 nilfs, NULL, NILFS_DAT_INO, NILFS_DAT_GFP);
 144         if (unlikely(!nilfs->ns_gc_dat))
 145                 goto failed_dat;
 146
 147         nilfs->ns_cpfile = nilfs_mdt_new(
 148                 nilfs, NULL, NILFS_CPFILE_INO, NILFS_CPFILE_GFP);
 149         if (unlikely(!nilfs->ns_cpfile))
 150                 goto failed_gc_dat;
 151
 152         nilfs->ns_sufile = nilfs_mdt_new(
 153                 nilfs, NULL, NILFS_SUFILE_INO, NILFS_SUFILE_GFP);
 154         if (unlikely(!nilfs->ns_sufile))
 155                 goto failed_cpfile;
 156
 157         err = nilfs_palloc_init_blockgroup(nilfs->ns_dat, dat_entry_size);
 158         if (unlikely(err))
 159                 goto failed_sufile;
 160
 161         err = nilfs_palloc_init_blockgroup(nilfs->ns_gc_dat, dat_entry_size);
 162         if (unlikely(err))
 163                 goto failed_sufile;
 164
 165         nilfs_mdt_set_shadow(nilfs->ns_dat, nilfs->ns_gc_dat);
 166         nilfs_mdt_set_entry_size(nilfs->ns_cpfile, checkpoint_size,
 167                                  sizeof(struct nilfs_cpfile_header));
 168         nilfs_mdt_set_entry_size(nilfs->ns_sufile, segment_usage_size,
 169                                  sizeof(struct nilfs_sufile_header));
 170
 171         err = nilfs_mdt_read_inode_direct(
 172                 nilfs->ns_dat, bh_sr, NILFS_SR_DAT_OFFSET(inode_size));
 173         if (unlikely(err))
 174                 goto failed_sufile;
 175
 176         err = nilfs_mdt_read_inode_direct(
 177                 nilfs->ns_cpfile, bh_sr, NILFS_SR_CPFILE_OFFSET(inode_size));
 178         if (unlikely(err))
 179                 goto failed_sufile;
 180
 181         err = nilfs_mdt_read_inode_direct(
 182                 nilfs->ns_sufile, bh_sr, NILFS_SR_SUFILE_OFFSET(inode_size));
 183         if (unlikely(err))
 184                 goto failed_sufile;
 185
 186         raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
 187         nilfs->ns_nongc_ctime = le64_to_cpu(raw_sr->sr_nongc_ctime);
 188
 189  failed:
 190         brelse(bh_sr);
 191         return err;
 192
 193  failed_sufile:
 194         nilfs_mdt_destroy(nilfs->ns_sufile);
 195
 196  failed_cpfile:
 197         nilfs_mdt_destroy(nilfs->ns_cpfile);
 198
 199  failed_gc_dat:
 200         nilfs_mdt_destroy(nilfs->ns_gc_dat);
 201
 202  failed_dat:
 203         nilfs_mdt_destroy(nilfs->ns_dat);
 204         goto failed;
 205 }
 206
 207 static void nilfs_init_recovery_info(struct nilfs_recovery_info *ri)
 208 {
 209         memset(ri, 0, sizeof(*ri));
 210         INIT_LIST_HEAD(&ri->ri_used_segments);
 211 }
 212
 213 static void nilfs_clear_recovery_info(struct nilfs_recovery_info *ri)
 214 {
 215         nilfs_dispose_segment_list(&ri->ri_used_segments);
 216 }
 217
 218 /**
 219  * load_nilfs - load and recover the nilfs
 220  * @nilfs: the_nilfs structure to be released
 221  * @sbi: nilfs_sb_info used to recover past segment
 222  *
 223  * load_nilfs() searches and load the latest super root,
 224  * attaches the last segment, and does recovery if needed.
 225  * The caller must call this exclusively for simultaneous mounts.
 226  */
 227 int load_nilfs(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi)
 228 {
 229         struct nilfs_recovery_info ri;
 230         unsigned int s_flags = sbi->s_super->s_flags;
 231         int really_read_only = bdev_read_only(nilfs->ns_bdev);
 232         unsigned valid_fs;
 233         int err = 0;
 234
 235         nilfs_init_recovery_info(&ri);
 236
 237         down_write(&nilfs->ns_sem);
 238         valid_fs = (nilfs->ns_mount_state & NILFS_VALID_FS);
 239         up_write(&nilfs->ns_sem);
 240
 241         if (!valid_fs && (s_flags & MS_RDONLY)) {
 242                 printk(KERN_INFO "NILFS: INFO: recovery "
 243                        "required for readonly filesystem.\n");
 244                 if (really_read_only) {
 245                         printk(KERN_ERR "NILFS: write access "
 246                                "unavailable, cannot proceed.\n");
 247                         err = -EROFS;
 248                         goto failed;
 249                 }
 250                 printk(KERN_INFO "NILFS: write access will "
 251                        "be enabled during recovery.\n");
 252                 sbi->s_super->s_flags &= ~MS_RDONLY;
 253         }
 254
 255         err = nilfs_search_super_root(nilfs, sbi, &ri);
 256         if (unlikely(err)) {
 257                 printk(KERN_ERR "NILFS: error searching super root.\n");
 258                 goto failed;
 259         }
 260
 261         err = nilfs_load_super_root(nilfs, sbi, ri.ri_super_root);
 262         if (unlikely(err)) {
 263                 printk(KERN_ERR "NILFS: error loading super root.\n");
 264                 goto failed;
 265         }
 266
 267         if (!valid_fs) {
 268                 err = nilfs_recover_logical_segments(nilfs, sbi, &ri);
 269                 if (unlikely(err)) {
 270                         nilfs_mdt_destroy(nilfs->ns_cpfile);
 271                         nilfs_mdt_destroy(nilfs->ns_sufile);
 272                         nilfs_mdt_destroy(nilfs->ns_dat);
 273                         goto failed;
 274                 }
 275                 if (ri.ri_need_recovery == NILFS_RECOVERY_SR_UPDATED) {
 276                         down_write(&nilfs->ns_sem);
 277                         nilfs_update_last_segment(sbi, 0);
 278                         up_write(&nilfs->ns_sem);
 279                 }
 280         }
 281
 282         set_nilfs_loaded(nilfs);
 283
 284  failed:
 285         nilfs_clear_recovery_info(&ri);
 286         sbi->s_super->s_flags = s_flags;
 287         return err;
 288 }
 289
 290 static unsigned long long nilfs_max_size(unsigned int blkbits)
 291 {
 292         unsigned int max_bits;
 293         unsigned long long res = MAX_LFS_FILESIZE; /* page cache limit */
 294
 295         max_bits = blkbits + NILFS_BMAP_KEY_BIT; /* bmap size limit */
 296         if (max_bits < 64)
 297                 res = min_t(unsigned long long, res, (1ULL << max_bits) - 1);
 298         return res;
 299 }
 300
 301 static int
 302 nilfs_store_disk_layout(struct the_nilfs *nilfs, struct super_block *sb,
 303                         struct nilfs_super_block *sbp)
 304 {
 305         if (le32_to_cpu(sbp->s_rev_level) != NILFS_CURRENT_REV) {
 306                 printk(KERN_ERR "NILFS: revision mismatch "
 307                        "(superblock rev.=%d.%d, current rev.=%d.%d). "
 308                        "Please check the version of mkfs.nilfs.\n",
 309                        le32_to_cpu(sbp->s_rev_level),
 310                        le16_to_cpu(sbp->s_minor_rev_level),
 311                        NILFS_CURRENT_REV, NILFS_MINOR_REV);
 312                 return -EINVAL;
 313         }
 314         nilfs->ns_inode_size = le16_to_cpu(sbp->s_inode_size);
 315         nilfs->ns_first_ino = le32_to_cpu(sbp->s_first_ino);
 316
 317         nilfs->ns_blocks_per_segment = le32_to_cpu(sbp->s_blocks_per_segment);
 318         if (nilfs->ns_blocks_per_segment < NILFS_SEG_MIN_BLOCKS) {
 319                 printk(KERN_ERR "NILFS: too short segment. \n");
 320                 return -EINVAL;
 321         }
 322
 323         nilfs->ns_first_data_block = le64_to_cpu(sbp->s_first_data_block);
 324         nilfs->ns_nsegments = le64_to_cpu(sbp->s_nsegments);
 325         nilfs->ns_r_segments_percentage =
 326                 le32_to_cpu(sbp->s_r_segments_percentage);
 327         nilfs->ns_nrsvsegs =
 328                 max_t(unsigned long, NILFS_MIN_NRSVSEGS,
 329                       DIV_ROUND_UP(nilfs->ns_nsegments *
 330                                    nilfs->ns_r_segments_percentage, 100));
 331         nilfs->ns_crc_seed = le32_to_cpu(sbp->s_crc_seed);
 332         return 0;
 333 }
 334
 335 /**
 336  * init_nilfs - initialize a NILFS instance.
 337  * @nilfs: the_nilfs structure
 338  * @sbi: nilfs_sb_info
 339  * @sb: super block
 340  * @data: mount options
 341  *
 342  * init_nilfs() performs common initialization per block device (e.g.
 343  * reading the super block, getting disk layout information, initializing
 344  * shared fields in the_nilfs). It takes on some portion of the jobs
 345  * typically done by a fill_super() routine. This division arises from
 346  * the nature that multiple NILFS instances may be simultaneously
 347  * mounted on a device.
 348  * For multiple mounts on the same device, only the first mount
 349  * invokes these tasks.
 350  *
 351  * Return Value: On success, 0 is returned. On error, a negative error
 352  * code is returned.
 353  */
 354 int init_nilfs(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi, char *data)
 355 {
 356         struct super_block *sb = sbi->s_super;
 357         struct buffer_head *sbh;
 358         struct nilfs_super_block *sbp;
 359         struct backing_dev_info *bdi;
 360         int blocksize;
 361         int err = 0;
 362
 363         down_write(&nilfs->ns_sem);
 364         if (nilfs_init(nilfs)) {
 365                 /* Load values from existing the_nilfs */
 366                 sbp = nilfs->ns_sbp;
 367                 err = nilfs_store_magic_and_option(sb, sbp, data);
 368                 if (err)
 369                         goto out;
 370
 371                 blocksize = BLOCK_SIZE << le32_to_cpu(sbp->s_log_block_size);
 372                 if (sb->s_blocksize != blocksize &&
 373                     !sb_set_blocksize(sb, blocksize)) {
 374                         printk(KERN_ERR "NILFS: blocksize %d unfit to device\n",
 375                                blocksize);
 376                         err = -EINVAL;
 377                 }
 378                 sb->s_maxbytes = nilfs_max_size(sb->s_blocksize_bits);
 379                 goto out;
 380         }
 381
 382         sbp = nilfs_load_super_block(sb, &sbh);
 383         if (!sbp) {
 384                 err = -EINVAL;
 385                 goto out;
 386         }
 387         err = nilfs_store_magic_and_option(sb, sbp, data);
 388         if (err)
 389                 goto failed_sbh;
 390
 391         blocksize = BLOCK_SIZE << le32_to_cpu(sbp->s_log_block_size);
 392         if (sb->s_blocksize != blocksize) {
 393                 sbp = nilfs_reload_super_block(sb, &sbh, blocksize);
 394                 if (!sbp) {
 395                         err = -EINVAL;
 396                         goto out;
 397                         /* not failed_sbh; sbh is released automatically
 398                            when reloading fails. */
 399                 }
 400         }
 401         nilfs->ns_blocksize_bits = sb->s_blocksize_bits;
 402
 403         err = nilfs_store_disk_layout(nilfs, sb, sbp);
 404         if (err)
 405                 goto failed_sbh;
 406
 407         sb->s_maxbytes = nilfs_max_size(sb->s_blocksize_bits);
 408
 409         nilfs->ns_mount_state = le16_to_cpu(sbp->s_state);
 410         nilfs->ns_sbh = sbh;
 411         nilfs->ns_sbp = sbp;
 412
 413         bdi = nilfs->ns_bdev->bd_inode_backing_dev_info;
 414         if (!bdi)
 415                 bdi = nilfs->ns_bdev->bd_inode->i_mapping->backing_dev_info;
 416         nilfs->ns_bdi = bdi ? : &default_backing_dev_info;
 417
 418         /* Finding last segment */
 419         nilfs->ns_last_pseg = le64_to_cpu(sbp->s_last_pseg);
 420         nilfs->ns_last_cno = le64_to_cpu(sbp->s_last_cno);
 421         nilfs->ns_last_seq = le64_to_cpu(sbp->s_last_seq);
 422
 423         nilfs->ns_seg_seq = nilfs->ns_last_seq;
 424         nilfs->ns_segnum =
 425                 nilfs_get_segnum_of_block(nilfs, nilfs->ns_last_pseg);
 426         nilfs->ns_cno = nilfs->ns_last_cno + 1;
 427         if (nilfs->ns_segnum >= nilfs->ns_nsegments) {
 428                 printk(KERN_ERR "NILFS invalid last segment number.\n");
 429                 err = -EINVAL;
 430                 goto failed_sbh;
 431         }
 432         /* Dummy values  */
 433         nilfs->ns_free_segments_count =
 434                 nilfs->ns_nsegments - (nilfs->ns_segnum + 1);
 435
 436         /* Initialize gcinode cache */
 437         err = nilfs_init_gccache(nilfs);
 438         if (err)
 439                 goto failed_sbh;
 440
 441         set_nilfs_init(nilfs);
 442         err = 0;
 443  out:
 444         up_write(&nilfs->ns_sem);
 445         return err;
 446
 447  failed_sbh:
 448         brelse(sbh);
 449         goto out;
 450 }
 451
 452 int nilfs_count_free_blocks(struct the_nilfs *nilfs, sector_t *nblocks)
 453 {
 454         struct inode *dat = nilfs_dat_inode(nilfs);
 455         unsigned long ncleansegs;
 456         int err;
 457
 458         down_read(&NILFS_MDT(dat)->mi_sem);     /* XXX */
 459         err = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile, &ncleansegs);
 460         up_read(&NILFS_MDT(dat)->mi_sem);       /* XXX */
 461         if (likely(!err))
 462                 *nblocks = (sector_t)ncleansegs * nilfs->ns_blocks_per_segment;
 463         return err;
 464 }
 465
 466 void nilfs_dispose_used_segments(struct the_nilfs *nilfs)
 467 {
 468         struct nilfs_segment_entry *ent, *n;
 469
 470         /* nilfs->sem must be locked by the caller. */
 471         if (!nilfs_loaded(nilfs))
 472                 return;
 473
 474         list_for_each_entry_safe(ent, n, &nilfs->ns_used_segments, list) {
 475                 list_del_init(&ent->list);
 476                 nilfs_segment_usage_clear_volatile_active(ent->raw_su);
 477                 nilfs_close_segment_entry(ent, nilfs->ns_sufile);
 478                 nilfs_free_segment_entry(ent);
 479         }
 480 }
 481
 482 int nilfs_near_disk_full(struct the_nilfs *nilfs)
 483 {
 484         struct inode *sufile = nilfs->ns_sufile;
 485         unsigned long ncleansegs, nincsegs;
 486         int ret;
 487
 488         ret = nilfs_sufile_get_ncleansegs(sufile, &ncleansegs);
 489         if (likely(!ret)) {
 490                 nincsegs = atomic_read(&nilfs->ns_ndirtyblks) /
 491                         nilfs->ns_blocks_per_segment + 1;
 492                 if (ncleansegs <= nilfs->ns_nrsvsegs + nincsegs)
 493                         ret++;
 494         }
 495         return ret;
 496 }
 497
 498 int nilfs_checkpoint_is_mounted(struct the_nilfs *nilfs, __u64 cno,
 499                                 int snapshot_mount)
 500 {
 501         struct nilfs_sb_info *sbi;
 502         int ret = 0;
 503
 504         down_read(&nilfs->ns_sem);
 505         if (cno == 0 || cno > nilfs->ns_cno)
 506                 goto out_unlock;
 507
 508         list_for_each_entry(sbi, &nilfs->ns_supers, s_list) {
 509                 if (sbi->s_snapshot_cno == cno &&
 510                     (!snapshot_mount || nilfs_test_opt(sbi, SNAPSHOT))) {
 511                                         /* exclude read-only mounts */
 512                         ret++;
 513                         break;
 514                 }
 515         }
 516         /* for protecting recent checkpoints */
 517         if (cno >= nilfs_last_cno(nilfs))
 518                 ret++;
 519
 520  out_unlock:
 521         up_read(&nilfs->ns_sem);
 522         return ret;
 523 }