fs/logfs/inode.c

   1 /*
   2  * fs/logfs/inode.c     - inode handling code
   3  *
   4  * As should be obvious for Linux kernel code, license is GPLv2
   5  *
   6  * Copyright (c) 2005-2008 Joern Engel <joern@logfs.org>
   7  */
   8 #include "logfs.h"
   9 #include <linux/slab.h>
  10 #include <linux/writeback.h>
  11 #include <linux/backing-dev.h>
  12
  13 /*
  14  * How soon to reuse old inode numbers?  LogFS doesn't store deleted inodes
  15  * on the medium.  It therefore also lacks a method to store the previous
  16  * generation number for deleted inodes.  Instead a single generation number
  17  * is stored which will be used for new inodes.  Being just a 32bit counter,
  18  * this can obvious wrap relatively quickly.  So we only reuse inodes if we
  19  * know that a fair number of inodes can be created before we have to increment
  20  * the generation again - effectively adding some bits to the counter.
  21  * But being too aggressive here means we keep a very large and very sparse
  22  * inode file, wasting space on indirect blocks.
  23  * So what is a good value?  Beats me.  64k seems moderately bad on both
  24  * fronts, so let's use that for now...
  25  *
  26  * NFS sucks, as everyone already knows.
  27  */
  28 #define INOS_PER_WRAP (0x10000)
  29
  30 /*
  31  * Logfs' requirement to read inodes for garbage collection makes life a bit
  32  * harder.  GC may have to read inodes that are in I_FREEING state, when they
  33  * are being written out - and waiting for GC to make progress, naturally.
  34  *
  35  * So we cannot just call iget() or some variant of it, but first have to check
  36  * wether the inode in question might be in I_FREEING state.  Therefore we
  37  * maintain our own per-sb list of "almost deleted" inodes and check against
  38  * that list first.  Normally this should be at most 1-2 entries long.
  39  *
  40  * Also, inodes have logfs-specific reference counting on top of what the vfs
  41  * does.  When .destroy_inode is called, normally the reference count will drop
  42  * to zero and the inode gets deleted.  But if GC accessed the inode, its
  43  * refcount will remain nonzero and final deletion will have to wait.
  44  *
  45  * As a result we have two sets of functions to get/put inodes:
  46  * logfs_safe_iget/logfs_safe_iput      - safe to call from GC context
  47  * logfs_iget/iput                      - normal version
  48  */
  49 static struct kmem_cache *logfs_inode_cache;
  50
  51 static DEFINE_SPINLOCK(logfs_inode_lock);
  52
  53 static void logfs_inode_setops(struct inode *inode)
  54 {
  55         switch (inode->i_mode & S_IFMT) {
  56         case S_IFDIR:
  57                 inode->i_op = &logfs_dir_iops;
  58                 inode->i_fop = &logfs_dir_fops;
  59                 inode->i_mapping->a_ops = &logfs_reg_aops;
  60                 break;
  61         case S_IFREG:
  62                 inode->i_op = &logfs_reg_iops;
  63                 inode->i_fop = &logfs_reg_fops;
  64                 inode->i_mapping->a_ops = &logfs_reg_aops;
  65                 break;
  66         case S_IFLNK:
  67                 inode->i_op = &logfs_symlink_iops;
  68                 inode->i_mapping->a_ops = &logfs_reg_aops;
  69                 break;
  70         case S_IFSOCK:  /* fall through */
  71         case S_IFBLK:   /* fall through */
  72         case S_IFCHR:   /* fall through */
  73         case S_IFIFO:
  74                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
  75                 break;
  76         default:
  77                 BUG();
  78         }
  79 }
  80
  81 static struct inode *__logfs_iget(struct super_block *sb, ino_t ino)
  82 {
  83         struct inode *inode = iget_locked(sb, ino);
  84         int err;
  85
  86         if (!inode)
  87                 return ERR_PTR(-ENOMEM);
  88         if (!(inode->i_state & I_NEW))
  89                 return inode;
  90
  91         err = logfs_read_inode(inode);
  92         if (err || inode->i_nlink == 0) {
  93                 /* inode->i_nlink == 0 can be true when called from
  94                  * block validator */
  95                 /* set i_nlink to 0 to prevent caching */
  96                 clear_nlink(inode);
  97                 logfs_inode(inode)->li_flags |= LOGFS_IF_ZOMBIE;
  98                 iget_failed(inode);
  99                 if (!err)
 100                         err = -ENOENT;
 101                 return ERR_PTR(err);
 102         }
 103
 104         logfs_inode_setops(inode);
 105         unlock_new_inode(inode);
 106         return inode;
 107 }
 108
 109 struct inode *logfs_iget(struct super_block *sb, ino_t ino)
 110 {
 111         BUG_ON(ino == LOGFS_INO_MASTER);
 112         BUG_ON(ino == LOGFS_INO_SEGFILE);
 113         return __logfs_iget(sb, ino);
 114 }
 115
 116 /*
 117  * is_cached is set to 1 if we hand out a cached inode, 0 otherwise.
 118  * this allows logfs_iput to do the right thing later
 119  */
 120 struct inode *logfs_safe_iget(struct super_block *sb, ino_t ino, int *is_cached)
 121 {
 122         struct logfs_super *super = logfs_super(sb);
 123         struct logfs_inode *li;
 124
 125         if (ino == LOGFS_INO_MASTER)
 126                 return super->s_master_inode;
 127         if (ino == LOGFS_INO_SEGFILE)
 128                 return super->s_segfile_inode;
 129
 130         spin_lock(&logfs_inode_lock);
 131         list_for_each_entry(li, &super->s_freeing_list, li_freeing_list)
 132                 if (li->vfs_inode.i_ino == ino) {
 133                         li->li_refcount++;
 134                         spin_unlock(&logfs_inode_lock);
 135                         *is_cached = 1;
 136                         return &li->vfs_inode;
 137                 }
 138         spin_unlock(&logfs_inode_lock);
 139
 140         *is_cached = 0;
 141         return __logfs_iget(sb, ino);
 142 }
 143
 144 static void logfs_i_callback(struct rcu_head *head)
 145 {
 146         struct inode *inode = container_of(head, struct inode, i_rcu);
 147         INIT_LIST_HEAD(&inode->i_dentry);
 148         kmem_cache_free(logfs_inode_cache, logfs_inode(inode));
 149 }
 150
 151 static void __logfs_destroy_inode(struct inode *inode)
 152 {
 153         struct logfs_inode *li = logfs_inode(inode);
 154
 155         BUG_ON(li->li_block);
 156         list_del(&li->li_freeing_list);
 157         call_rcu(&inode->i_rcu, logfs_i_callback);
 158 }
 159
 160 static void logfs_destroy_inode(struct inode *inode)
 161 {
 162         struct logfs_inode *li = logfs_inode(inode);
 163
 164         BUG_ON(list_empty(&li->li_freeing_list));
 165         spin_lock(&logfs_inode_lock);
 166         li->li_refcount--;
 167         if (li->li_refcount == 0)
 168                 __logfs_destroy_inode(inode);
 169         spin_unlock(&logfs_inode_lock);
 170 }
 171
 172 void logfs_safe_iput(struct inode *inode, int is_cached)
 173 {
 174         if (inode->i_ino == LOGFS_INO_MASTER)
 175                 return;
 176         if (inode->i_ino == LOGFS_INO_SEGFILE)
 177                 return;
 178
 179         if (is_cached) {
 180                 logfs_destroy_inode(inode);
 181                 return;
 182         }
 183
 184         iput(inode);
 185 }
 186
 187 static void logfs_init_inode(struct super_block *sb, struct inode *inode)
 188 {
 189         struct logfs_inode *li = logfs_inode(inode);
 190         int i;
 191
 192         li->li_flags    = 0;
 193         li->li_height   = 0;
 194         li->li_used_bytes = 0;
 195         li->li_block    = NULL;
 196         inode->i_uid    = 0;
 197         inode->i_gid    = 0;
 198         inode->i_size   = 0;
 199         inode->i_blocks = 0;
 200         inode->i_ctime  = CURRENT_TIME;
 201         inode->i_mtime  = CURRENT_TIME;
 202         li->li_refcount = 1;
 203         INIT_LIST_HEAD(&li->li_freeing_list);
 204
 205         for (i = 0; i < LOGFS_EMBEDDED_FIELDS; i++)
 206                 li->li_data[i] = 0;
 207
 208         return;
 209 }
 210
 211 static struct inode *logfs_alloc_inode(struct super_block *sb)
 212 {
 213         struct logfs_inode *li;
 214
 215         li = kmem_cache_alloc(logfs_inode_cache, GFP_NOFS);
 216         if (!li)
 217                 return NULL;
 218         logfs_init_inode(sb, &li->vfs_inode);
 219         return &li->vfs_inode;
 220 }
 221
 222 /*
 223  * In logfs inodes are written to an inode file.  The inode file, like any
 224  * other file, is managed with a inode.  The inode file's inode, aka master
 225  * inode, requires special handling in several respects.  First, it cannot be
 226  * written to the inode file, so it is stored in the journal instead.
 227  *
 228  * Secondly, this inode cannot be written back and destroyed before all other
 229  * inodes have been written.  The ordering is important.  Linux' VFS is happily
 230  * unaware of the ordering constraint and would ordinarily destroy the master
 231  * inode at umount time while other inodes are still in use and dirty.  Not
 232  * good.
 233  *
 234  * So logfs makes sure the master inode is not written until all other inodes
 235  * have been destroyed.  Sadly, this method has another side-effect.  The VFS
 236  * will notice one remaining inode and print a frightening warning message.
 237  * Worse, it is impossible to judge whether such a warning was caused by the
 238  * master inode or any other inodes have leaked as well.
 239  *
 240  * Our attempt of solving this is with logfs_new_meta_inode() below.  Its
 241  * purpose is to create a new inode that will not trigger the warning if such
 242  * an inode is still in use.  An ugly hack, no doubt.  Suggections for
 243  * improvement are welcome.
 244  *
 245  * AV: that's what ->put_super() is for...
 246  */
 247 struct inode *logfs_new_meta_inode(struct super_block *sb, u64 ino)
 248 {
 249         struct inode *inode;
 250
 251         inode = new_inode(sb);
 252         if (!inode)
 253                 return ERR_PTR(-ENOMEM);
 254
 255         inode->i_mode = S_IFREG;
 256         inode->i_ino = ino;
 257         inode->i_data.a_ops = &logfs_reg_aops;
 258         mapping_set_gfp_mask(&inode->i_data, GFP_NOFS);
 259
 260         return inode;
 261 }
 262
 263 struct inode *logfs_read_meta_inode(struct super_block *sb, u64 ino)
 264 {
 265         struct inode *inode;
 266         int err;
 267
 268         inode = logfs_new_meta_inode(sb, ino);
 269         if (IS_ERR(inode))
 270                 return inode;
 271
 272         err = logfs_read_inode(inode);
 273         if (err) {
 274                 iput(inode);
 275                 return ERR_PTR(err);
 276         }
 277         logfs_inode_setops(inode);
 278         return inode;
 279 }
 280
 281 static int logfs_write_inode(struct inode *inode, struct writeback_control *wbc)
 282 {
 283         int ret;
 284         long flags = WF_LOCK;
 285
 286         /* Can only happen if creat() failed.  Safe to skip. */
 287         if (logfs_inode(inode)->li_flags & LOGFS_IF_STILLBORN)
 288                 return 0;
 289
 290         ret = __logfs_write_inode(inode, flags);
 291         LOGFS_BUG_ON(ret, inode->i_sb);
 292         return ret;
 293 }
 294
 295 /* called with inode->i_lock held */
 296 static int logfs_drop_inode(struct inode *inode)
 297 {
 298         struct logfs_super *super = logfs_super(inode->i_sb);
 299         struct logfs_inode *li = logfs_inode(inode);
 300
 301         spin_lock(&logfs_inode_lock);
 302         list_move(&li->li_freeing_list, &super->s_freeing_list);
 303         spin_unlock(&logfs_inode_lock);
 304         return generic_drop_inode(inode);
 305 }
 306
 307 static void logfs_set_ino_generation(struct super_block *sb,
 308                 struct inode *inode)
 309 {
 310         struct logfs_super *super = logfs_super(sb);
 311         u64 ino;
 312
 313         mutex_lock(&super->s_journal_mutex);
 314         ino = logfs_seek_hole(super->s_master_inode, super->s_last_ino + 1);
 315         super->s_last_ino = ino;
 316         super->s_inos_till_wrap--;
 317         if (super->s_inos_till_wrap < 0) {
 318                 super->s_last_ino = LOGFS_RESERVED_INOS;
 319                 super->s_generation++;
 320                 super->s_inos_till_wrap = INOS_PER_WRAP;
 321         }
 322         inode->i_ino = ino;
 323         inode->i_generation = super->s_generation;
 324         mutex_unlock(&super->s_journal_mutex);
 325 }
 326
 327 struct inode *logfs_new_inode(struct inode *dir, int mode)
 328 {
 329         struct super_block *sb = dir->i_sb;
 330         struct inode *inode;
 331
 332         inode = new_inode(sb);
 333         if (!inode)
 334                 return ERR_PTR(-ENOMEM);
 335
 336         logfs_init_inode(sb, inode);
 337
 338         /* inherit parent flags */
 339         logfs_inode(inode)->li_flags |=
 340                 logfs_inode(dir)->li_flags & LOGFS_FL_INHERITED;
 341
 342         inode->i_mode = mode;
 343         logfs_set_ino_generation(sb, inode);
 344
 345         inode_init_owner(inode, dir, mode);
 346         logfs_inode_setops(inode);
 347         insert_inode_hash(inode);
 348
 349         return inode;
 350 }
 351
 352 static void logfs_init_once(void *_li)
 353 {
 354         struct logfs_inode *li = _li;
 355         int i;
 356
 357         li->li_flags = 0;
 358         li->li_used_bytes = 0;
 359         li->li_refcount = 1;
 360         for (i = 0; i < LOGFS_EMBEDDED_FIELDS; i++)
 361                 li->li_data[i] = 0;
 362         inode_init_once(&li->vfs_inode);
 363 }
 364
 365 static int logfs_sync_fs(struct super_block *sb, int wait)
 366 {
 367         logfs_write_anchor(sb);
 368         return 0;
 369 }
 370
 371 static void logfs_put_super(struct super_block *sb)
 372 {
 373         struct logfs_super *super = logfs_super(sb);
 374         /* kill the meta-inodes */
 375         iput(super->s_master_inode);
 376         iput(super->s_segfile_inode);
 377         iput(super->s_mapping_inode);
 378 }
 379
 380 const struct super_operations logfs_super_operations = {
 381         .alloc_inode    = logfs_alloc_inode,
 382         .destroy_inode  = logfs_destroy_inode,
 383         .evict_inode    = logfs_evict_inode,
 384         .drop_inode     = logfs_drop_inode,
 385         .put_super      = logfs_put_super,
 386         .write_inode    = logfs_write_inode,
 387         .statfs         = logfs_statfs,
 388         .sync_fs        = logfs_sync_fs,
 389 };
 390
 391 int logfs_init_inode_cache(void)
 392 {
 393         logfs_inode_cache = kmem_cache_create("logfs_inode_cache",
 394                         sizeof(struct logfs_inode), 0, SLAB_RECLAIM_ACCOUNT,
 395                         logfs_init_once);
 396         if (!logfs_inode_cache)
 397                 return -ENOMEM;
 398         return 0;
 399 }
 400
 401 void logfs_destroy_inode_cache(void)
 402 {
 403         kmem_cache_destroy(logfs_inode_cache);
 404 }