fs/libfs.c

   1 /*
   2  *      fs/libfs.c
   3  *      Library for filesystems writers.
   4  */
   5
   6 #include <linux/module.h>
   7 #include <linux/pagemap.h>
   8 #include <linux/mount.h>
   9 #include <linux/vfs.h>
  10 #include <asm/uaccess.h>
  11
  12 int simple_getattr(struct vfsmount *mnt, struct dentry *dentry,
  13                    struct kstat *stat)
  14 {
  15         struct inode *inode = dentry->d_inode;
  16         generic_fillattr(inode, stat);
  17         stat->blocks = inode->i_mapping->nrpages << (PAGE_CACHE_SHIFT - 9);
  18         return 0;
  19 }
  20
  21 int simple_statfs(struct super_block *sb, struct kstatfs *buf)
  22 {
  23         buf->f_type = sb->s_magic;
  24         buf->f_bsize = PAGE_CACHE_SIZE;
  25         buf->f_namelen = NAME_MAX;
  26         return 0;
  27 }
  28
  29 /*
  30  * Retaining negative dentries for an in-memory filesystem just wastes
  31  * memory and lookup time: arrange for them to be deleted immediately.
  32  */
  33 static int simple_delete_dentry(struct dentry *dentry)
  34 {
  35         return 1;
  36 }
  37
  38 /*
  39  * Lookup the data. This is trivial - if the dentry didn't already
  40  * exist, we know it is negative.  Set d_op to delete negative dentries.
  41  */
  42 struct dentry *simple_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
  43 {
  44         static struct dentry_operations simple_dentry_operations = {
  45                 .d_delete = simple_delete_dentry,
  46         };
  47
  48         if (dentry->d_name.len > NAME_MAX)
  49                 return ERR_PTR(-ENAMETOOLONG);
  50         dentry->d_op = &simple_dentry_operations;
  51         d_add(dentry, NULL);
  52         return NULL;
  53 }
  54
  55 int simple_sync_file(struct file * file, struct dentry *dentry, int datasync)
  56 {
  57         return 0;
  58 }
  59
  60 int dcache_dir_open(struct inode *inode, struct file *file)
  61 {
  62         static struct qstr cursor_name = {.len = 1, .name = "."};
  63
  64         file->private_data = d_alloc(file->f_dentry, &cursor_name);
  65
  66         return file->private_data ? 0 : -ENOMEM;
  67 }
  68
  69 int dcache_dir_close(struct inode *inode, struct file *file)
  70 {
  71         dput(file->private_data);
  72         return 0;
  73 }
  74
  75 loff_t dcache_dir_lseek(struct file *file, loff_t offset, int origin)
  76 {
  77         down(&file->f_dentry->d_inode->i_sem);
  78         switch (origin) {
  79                 case 1:
  80                         offset += file->f_pos;
  81                 case 0:
  82                         if (offset >= 0)
  83                                 break;
  84                 default:
  85                         up(&file->f_dentry->d_inode->i_sem);
  86                         return -EINVAL;
  87         }
  88         if (offset != file->f_pos) {
  89                 file->f_pos = offset;
  90                 if (file->f_pos >= 2) {
  91                         struct list_head *p;
  92                         struct dentry *cursor = file->private_data;
  93                         loff_t n = file->f_pos - 2;
  94
  95                         spin_lock(&dcache_lock);
  96                         list_del(&cursor->d_child);
  97                         p = file->f_dentry->d_subdirs.next;
  98                         while (n && p != &file->f_dentry->d_subdirs) {
  99                                 struct dentry *next;
 100                                 next = list_entry(p, struct dentry, d_child);
 101                                 if (!d_unhashed(next) && next->d_inode)
 102                                         n--;
 103                                 p = p->next;
 104                         }
 105                         list_add_tail(&cursor->d_child, p);
 106                         spin_unlock(&dcache_lock);
 107                 }
 108         }
 109         up(&file->f_dentry->d_inode->i_sem);
 110         return offset;
 111 }
 112
 113 /* Relationship between i_mode and the DT_xxx types */
 114 static inline unsigned char dt_type(struct inode *inode)
 115 {
 116         return (inode->i_mode >> 12) & 15;
 117 }
 118
 119 /*
 120  * Directory is locked and all positive dentries in it are safe, since
 121  * for ramfs-type trees they can't go away without unlink() or rmdir(),
 122  * both impossible due to the lock on directory.
 123  */
 124
 125 static inline int do_dcache_readdir_filter(struct file * filp,
 126         void * dirent, filldir_t filldir, int (*filter)(struct dentry *dentry))
 127 {
 128         struct dentry *dentry = filp->f_dentry;
 129         struct dentry *cursor = filp->private_data;
 130         struct list_head *p, *q = &cursor->d_child;
 131         ino_t ino;
 132         int i = filp->f_pos;
 133
 134         switch (i) {
 135                 case 0:
 136                         ino = dentry->d_inode->i_ino;
 137                         if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
 138                                 break;
 139                         filp->f_pos++;
 140                         i++;
 141                         /* fallthrough */
 142                 case 1:
 143                         ino = parent_ino(dentry);
 144                         if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
 145                                 break;
 146                         filp->f_pos++;
 147                         i++;
 148                         /* fallthrough */
 149                 default:
 150                         spin_lock(&dcache_lock);
 151                         if (filp->f_pos == 2) {
 152                                 list_del(q);
 153                                 list_add(q, &dentry->d_subdirs);
 154                         }
 155                         for (p=q->next; p != &dentry->d_subdirs; p=p->next) {
 156                                 struct dentry *next;
 157                                 next = list_entry(p, struct dentry, d_child);
 158                                 if (d_unhashed(next) || !next->d_inode)
 159                                         continue;
 160                                 if (filter && !filter(next))
 161                                         continue;
 162
 163                                 spin_unlock(&dcache_lock);
 164                                 if (filldir(dirent, next->d_name.name, next->d_name.len, filp->f_pos, next->d_inode->i_ino, dt_type(next->d_inode)) < 0)
 165                                         return 0;
 166                                 spin_lock(&dcache_lock);
 167                                 /* next is still alive */
 168                                 list_del(q);
 169                                 list_add(q, p);
 170                                 p = q;
 171                                 filp->f_pos++;
 172                         }
 173                         spin_unlock(&dcache_lock);
 174         }
 175         return 0;
 176 }
 177
 178 int dcache_readdir(struct file * filp, void * dirent, filldir_t filldir)
 179 {
 180         return do_dcache_readdir_filter(filp, dirent, filldir, NULL);
 181 }
 182
 183 int dcache_readdir_filter(struct file * filp, void * dirent, filldir_t filldir,
 184         int (*filter)(struct dentry *))
 185 {
 186         return do_dcache_readdir_filter(filp, dirent, filldir, filter);
 187 }
 188
 189
 190 ssize_t generic_read_dir(struct file *filp, char __user *buf, size_t siz, loff_t *ppos)
 191 {
 192         return -EISDIR;
 193 }
 194
 195 struct file_operations simple_dir_operations = {
 196         .open           = dcache_dir_open,
 197         .release        = dcache_dir_close,
 198         .llseek         = dcache_dir_lseek,
 199         .read           = generic_read_dir,
 200         .readdir        = dcache_readdir,
 201 };
 202
 203 struct inode_operations simple_dir_inode_operations = {
 204         .lookup         = simple_lookup,
 205 };
 206
 207 /*
 208  * Common helper for pseudo-filesystems (sockfs, pipefs, bdev - stuff that
 209  * will never be mountable)
 210  */
 211 struct super_block *
 212 get_sb_pseudo(struct file_system_type *fs_type, char *name,
 213         struct super_operations *ops, unsigned long magic)
 214 {
 215         struct super_block *s = sget(fs_type, NULL, set_anon_super, NULL);
 216         static struct super_operations default_ops = {.statfs = simple_statfs};
 217         struct dentry *dentry;
 218         struct inode *root;
 219         struct qstr d_name = {.name = name, .len = strlen(name)};
 220
 221         if (IS_ERR(s))
 222                 return s;
 223
 224         s->s_flags = MS_NOUSER;
 225         s->s_maxbytes = ~0ULL;
 226         s->s_blocksize = 1024;
 227         s->s_blocksize_bits = 10;
 228         s->s_magic = magic;
 229         s->s_op = ops ? ops : &default_ops;
 230         s->s_time_gran = 1;
 231         root = new_inode(s);
 232         if (!root)
 233                 goto Enomem;
 234         root->i_mode = S_IFDIR | S_IRUSR | S_IWUSR;
 235         root->i_uid = root->i_gid = 0;
 236         root->i_atime = root->i_mtime = root->i_ctime = CURRENT_TIME;
 237         dentry = d_alloc(NULL, &d_name);
 238         if (!dentry) {
 239                 iput(root);
 240                 goto Enomem;
 241         }
 242         dentry->d_sb = s;
 243         dentry->d_parent = dentry;
 244         d_instantiate(dentry, root);
 245         s->s_root = dentry;
 246         s->s_flags |= MS_ACTIVE;
 247         return s;
 248
 249 Enomem:
 250         up_write(&s->s_umount);
 251         deactivate_super(s);
 252         return ERR_PTR(-ENOMEM);
 253 }
 254
 255 int simple_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
 256 {
 257         struct inode *inode = old_dentry->d_inode;
 258
 259         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 260         inode->i_nlink++;
 261         atomic_inc(&inode->i_count);
 262         dget(dentry);
 263         d_instantiate(dentry, inode);
 264         return 0;
 265 }
 266
 267 static inline int simple_positive(struct dentry *dentry)
 268 {
 269         return dentry->d_inode && !d_unhashed(dentry);
 270 }
 271
 272 int simple_empty(struct dentry *dentry)
 273 {
 274         struct dentry *child;
 275         int ret = 0;
 276
 277         spin_lock(&dcache_lock);
 278         list_for_each_entry(child, &dentry->d_subdirs, d_child)
 279                 if (simple_positive(child))
 280                         goto out;
 281         ret = 1;
 282 out:
 283         spin_unlock(&dcache_lock);
 284         return ret;
 285 }
 286
 287 int simple_unlink(struct inode *dir, struct dentry *dentry)
 288 {
 289         struct inode *inode = dentry->d_inode;
 290
 291         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 292         inode->i_nlink--;
 293         dput(dentry);
 294         return 0;
 295 }
 296
 297 int simple_rmdir(struct inode *dir, struct dentry *dentry)
 298 {
 299         if (!simple_empty(dentry))
 300                 return -ENOTEMPTY;
 301
 302         dentry->d_inode->i_nlink--;
 303         simple_unlink(dir, dentry);
 304         dir->i_nlink--;
 305         return 0;
 306 }
 307
 308 int simple_rename(struct inode *old_dir, struct dentry *old_dentry,
 309                 struct inode *new_dir, struct dentry *new_dentry)
 310 {
 311         struct inode *inode = old_dentry->d_inode;
 312         int they_are_dirs = S_ISDIR(old_dentry->d_inode->i_mode);
 313
 314         if (!simple_empty(new_dentry))
 315                 return -ENOTEMPTY;
 316
 317         if (new_dentry->d_inode) {
 318                 simple_unlink(new_dir, new_dentry);
 319                 if (they_are_dirs)
 320                         old_dir->i_nlink--;
 321         } else if (they_are_dirs) {
 322                 old_dir->i_nlink--;
 323                 new_dir->i_nlink++;
 324         }
 325
 326         old_dir->i_ctime = old_dir->i_mtime = new_dir->i_ctime =
 327                 new_dir->i_mtime = inode->i_ctime = CURRENT_TIME;
 328
 329         return 0;
 330 }
 331
 332 int simple_readpage(struct file *file, struct page *page)
 333 {
 334         void *kaddr;
 335
 336         if (PageUptodate(page))
 337                 goto out;
 338
 339         kaddr = kmap_atomic(page, KM_USER0);
 340         memset(kaddr, 0, PAGE_CACHE_SIZE);
 341         kunmap_atomic(kaddr, KM_USER0);
 342         flush_dcache_page(page);
 343         SetPageUptodate(page);
 344 out:
 345         unlock_page(page);
 346         return 0;
 347 }
 348
 349 int simple_prepare_write(struct file *file, struct page *page,
 350                         unsigned from, unsigned to)
 351 {
 352         if (!PageUptodate(page)) {
 353                 if (to - from != PAGE_CACHE_SIZE) {
 354                         void *kaddr = kmap_atomic(page, KM_USER0);
 355                         memset(kaddr, 0, from);
 356                         memset(kaddr + to, 0, PAGE_CACHE_SIZE - to);
 357                         flush_dcache_page(page);
 358                         kunmap_atomic(kaddr, KM_USER0);
 359                 }
 360                 SetPageUptodate(page);
 361         }
 362         return 0;
 363 }
 364
 365 int simple_commit_write(struct file *file, struct page *page,
 366                         unsigned offset, unsigned to)
 367 {
 368         struct inode *inode = page->mapping->host;
 369         loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
 370
 371         /*
 372          * No need to use i_size_read() here, the i_size
 373          * cannot change under us because we hold the i_sem.
 374          */
 375         if (pos > inode->i_size)
 376                 i_size_write(inode, pos);
 377         set_page_dirty(page);
 378         return 0;
 379 }
 380
 381 int simple_fill_super(struct super_block *s, int magic, struct tree_descr *files)
 382 {
 383         static struct super_operations s_ops = {.statfs = simple_statfs};
 384         struct inode *inode;
 385         struct dentry *root;
 386         struct dentry *dentry;
 387         int i;
 388
 389         s->s_blocksize = PAGE_CACHE_SIZE;
 390         s->s_blocksize_bits = PAGE_CACHE_SHIFT;
 391         s->s_magic = magic;
 392         s->s_op = &s_ops;
 393         s->s_time_gran = 1;
 394
 395         inode = new_inode(s);
 396         if (!inode)
 397                 return -ENOMEM;
 398         inode->i_mode = S_IFDIR | 0755;
 399         inode->i_uid = inode->i_gid = 0;
 400         inode->i_blksize = PAGE_CACHE_SIZE;
 401         inode->i_blocks = 0;
 402         inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 403         inode->i_op = &simple_dir_inode_operations;
 404         inode->i_fop = &simple_dir_operations;
 405         root = d_alloc_root(inode);
 406         if (!root) {
 407                 iput(inode);
 408                 return -ENOMEM;
 409         }
 410         for (i = 0; !files->name || files->name[0]; i++, files++) {
 411                 if (!files->name)
 412                         continue;
 413                 dentry = d_alloc_name(root, files->name);
 414                 if (!dentry)
 415                         goto out;
 416                 inode = new_inode(s);
 417                 if (!inode)
 418                         goto out;
 419                 inode->i_mode = S_IFREG | files->mode;
 420                 inode->i_uid = inode->i_gid = 0;
 421                 inode->i_blksize = PAGE_CACHE_SIZE;
 422                 inode->i_blocks = 0;
 423                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 424                 inode->i_fop = files->ops;
 425                 inode->i_ino = i;
 426                 d_add(dentry, inode);
 427         }
 428         s->s_root = root;
 429         return 0;
 430 out:
 431         d_genocide(root);
 432         dput(root);
 433         return -ENOMEM;
 434 }
 435
 436 static DEFINE_SPINLOCK(pin_fs_lock);
 437
 438 int simple_pin_fs(char *name, struct vfsmount **mount, int *count)
 439 {
 440         struct vfsmount *mnt = NULL;
 441         spin_lock(&pin_fs_lock);
 442         if (unlikely(!*mount)) {
 443                 spin_unlock(&pin_fs_lock);
 444                 mnt = do_kern_mount(name, 0, name, NULL);
 445                 if (IS_ERR(mnt))
 446                         return PTR_ERR(mnt);
 447                 spin_lock(&pin_fs_lock);
 448                 if (!*mount)
 449                         *mount = mnt;
 450         }
 451         mntget(*mount);
 452         ++*count;
 453         spin_unlock(&pin_fs_lock);
 454         mntput(mnt);
 455         return 0;
 456 }
 457
 458 void simple_release_fs(struct vfsmount **mount, int *count)
 459 {
 460         struct vfsmount *mnt;
 461         spin_lock(&pin_fs_lock);
 462         mnt = *mount;
 463         if (!--*count)
 464                 *mount = NULL;
 465         spin_unlock(&pin_fs_lock);
 466         mntput(mnt);
 467 }
 468
 469 ssize_t simple_read_from_buffer(void __user *to, size_t count, loff_t *ppos,
 470                                 const void *from, size_t available)
 471 {
 472         loff_t pos = *ppos;
 473         if (pos < 0)
 474                 return -EINVAL;
 475         if (pos >= available)
 476                 return 0;
 477         if (count > available - pos)
 478                 count = available - pos;
 479         if (copy_to_user(to, from + pos, count))
 480                 return -EFAULT;
 481         *ppos = pos + count;
 482         return count;
 483 }
 484
 485 /*
 486  * Transaction based IO.
 487  * The file expects a single write which triggers the transaction, and then
 488  * possibly a read which collects the result - which is stored in a
 489  * file-local buffer.
 490  */
 491 char *simple_transaction_get(struct file *file, const char __user *buf, size_t size)
 492 {
 493         struct simple_transaction_argresp *ar;
 494         static DEFINE_SPINLOCK(simple_transaction_lock);
 495
 496         if (size > SIMPLE_TRANSACTION_LIMIT - 1)
 497                 return ERR_PTR(-EFBIG);
 498
 499         ar = (struct simple_transaction_argresp *)get_zeroed_page(GFP_KERNEL);
 500         if (!ar)
 501                 return ERR_PTR(-ENOMEM);
 502
 503         spin_lock(&simple_transaction_lock);
 504
 505         /* only one write allowed per open */
 506         if (file->private_data) {
 507                 spin_unlock(&simple_transaction_lock);
 508                 free_page((unsigned long)ar);
 509                 return ERR_PTR(-EBUSY);
 510         }
 511
 512         file->private_data = ar;
 513
 514         spin_unlock(&simple_transaction_lock);
 515
 516         if (copy_from_user(ar->data, buf, size))
 517                 return ERR_PTR(-EFAULT);
 518
 519         return ar->data;
 520 }
 521
 522 ssize_t simple_transaction_read(struct file *file, char __user *buf, size_t size, loff_t *pos)
 523 {
 524         struct simple_transaction_argresp *ar = file->private_data;
 525
 526         if (!ar)
 527                 return 0;
 528         return simple_read_from_buffer(buf, size, pos, ar->data, ar->size);
 529 }
 530
 531 int simple_transaction_release(struct inode *inode, struct file *file)
 532 {
 533         free_page((unsigned long)file->private_data);
 534         return 0;
 535 }
 536
 537 EXPORT_SYMBOL(dcache_dir_close);
 538 EXPORT_SYMBOL(dcache_dir_lseek);
 539 EXPORT_SYMBOL(dcache_dir_open);
 540 EXPORT_SYMBOL(dcache_readdir);
 541 EXPORT_SYMBOL(dcache_readdir_filter);
 542 EXPORT_SYMBOL(generic_read_dir);
 543 EXPORT_SYMBOL(get_sb_pseudo);
 544 EXPORT_SYMBOL(simple_commit_write);
 545 EXPORT_SYMBOL(simple_dir_inode_operations);
 546 EXPORT_SYMBOL(simple_dir_operations);
 547 EXPORT_SYMBOL(simple_empty);
 548 EXPORT_SYMBOL(d_alloc_name);
 549 EXPORT_SYMBOL(simple_fill_super);
 550 EXPORT_SYMBOL(simple_getattr);
 551 EXPORT_SYMBOL(simple_link);
 552 EXPORT_SYMBOL(simple_lookup);
 553 EXPORT_SYMBOL(simple_pin_fs);
 554 EXPORT_SYMBOL(simple_prepare_write);
 555 EXPORT_SYMBOL(simple_readpage);
 556 EXPORT_SYMBOL(simple_release_fs);
 557 EXPORT_SYMBOL(simple_rename);
 558 EXPORT_SYMBOL(simple_rmdir);
 559 EXPORT_SYMBOL(simple_statfs);
 560 EXPORT_SYMBOL(simple_sync_file);
 561 EXPORT_SYMBOL(simple_unlink);
 562 EXPORT_SYMBOL(simple_read_from_buffer);
 563 EXPORT_SYMBOL(simple_transaction_get);
 564 EXPORT_SYMBOL(simple_transaction_read);
 565 EXPORT_SYMBOL(simple_transaction_release);