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 <linux/mutex.h>
  11
  12 #include <asm/uaccess.h>
  13
  14 int simple_getattr(struct vfsmount *mnt, struct dentry *dentry,
  15                    struct kstat *stat)
  16 {
  17         struct inode *inode = dentry->d_inode;
  18         generic_fillattr(inode, stat);
  19         stat->blocks = inode->i_mapping->nrpages << (PAGE_CACHE_SHIFT - 9);
  20         return 0;
  21 }
  22
  23 int simple_statfs(struct dentry *dentry, struct kstatfs *buf)
  24 {
  25         buf->f_type = dentry->d_sb->s_magic;
  26         buf->f_bsize = PAGE_CACHE_SIZE;
  27         buf->f_namelen = NAME_MAX;
  28         return 0;
  29 }
  30
  31 /*
  32  * Retaining negative dentries for an in-memory filesystem just wastes
  33  * memory and lookup time: arrange for them to be deleted immediately.
  34  */
  35 static int simple_delete_dentry(struct dentry *dentry)
  36 {
  37         return 1;
  38 }
  39
  40 /*
  41  * Lookup the data. This is trivial - if the dentry didn't already
  42  * exist, we know it is negative.  Set d_op to delete negative dentries.
  43  */
  44 struct dentry *simple_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
  45 {
  46         static struct dentry_operations simple_dentry_operations = {
  47                 .d_delete = simple_delete_dentry,
  48         };
  49
  50         if (dentry->d_name.len > NAME_MAX)
  51                 return ERR_PTR(-ENAMETOOLONG);
  52         dentry->d_op = &simple_dentry_operations;
  53         d_add(dentry, NULL);
  54         return NULL;
  55 }
  56
  57 int simple_sync_file(struct file * file, struct dentry *dentry, int datasync)
  58 {
  59         return 0;
  60 }
  61
  62 int dcache_dir_open(struct inode *inode, struct file *file)
  63 {
  64         static struct qstr cursor_name = {.len = 1, .name = "."};
  65
  66         file->private_data = d_alloc(file->f_dentry, &cursor_name);
  67
  68         return file->private_data ? 0 : -ENOMEM;
  69 }
  70
  71 int dcache_dir_close(struct inode *inode, struct file *file)
  72 {
  73         dput(file->private_data);
  74         return 0;
  75 }
  76
  77 loff_t dcache_dir_lseek(struct file *file, loff_t offset, int origin)
  78 {
  79         mutex_lock(&file->f_dentry->d_inode->i_mutex);
  80         switch (origin) {
  81                 case 1:
  82                         offset += file->f_pos;
  83                 case 0:
  84                         if (offset >= 0)
  85                                 break;
  86                 default:
  87                         mutex_unlock(&file->f_dentry->d_inode->i_mutex);
  88                         return -EINVAL;
  89         }
  90         if (offset != file->f_pos) {
  91                 file->f_pos = offset;
  92                 if (file->f_pos >= 2) {
  93                         struct list_head *p;
  94                         struct dentry *cursor = file->private_data;
  95                         loff_t n = file->f_pos - 2;
  96
  97                         spin_lock(&dcache_lock);
  98                         list_del(&cursor->d_u.d_child);
  99                         p = file->f_dentry->d_subdirs.next;
 100                         while (n && p != &file->f_dentry->d_subdirs) {
 101                                 struct dentry *next;
 102                                 next = list_entry(p, struct dentry, d_u.d_child);
 103                                 if (!d_unhashed(next) && next->d_inode)
 104                                         n--;
 105                                 p = p->next;
 106                         }
 107                         list_add_tail(&cursor->d_u.d_child, p);
 108                         spin_unlock(&dcache_lock);
 109                 }
 110         }
 111         mutex_unlock(&file->f_dentry->d_inode->i_mutex);
 112         return offset;
 113 }
 114
 115 /* Relationship between i_mode and the DT_xxx types */
 116 static inline unsigned char dt_type(struct inode *inode)
 117 {
 118         return (inode->i_mode >> 12) & 15;
 119 }
 120
 121 /*
 122  * Directory is locked and all positive dentries in it are safe, since
 123  * for ramfs-type trees they can't go away without unlink() or rmdir(),
 124  * both impossible due to the lock on directory.
 125  */
 126
 127 static inline int do_dcache_readdir_filter(struct file * filp,
 128         void * dirent, filldir_t filldir, int (*filter)(struct dentry *dentry))
 129 {
 130         struct dentry *dentry = filp->f_dentry;
 131         struct dentry *cursor = filp->private_data;
 132         struct list_head *p, *q = &cursor->d_u.d_child;
 133         ino_t ino;
 134         int i = filp->f_pos;
 135
 136         switch (i) {
 137                 case 0:
 138                         ino = dentry->d_inode->i_ino;
 139                         if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
 140                                 break;
 141                         filp->f_pos++;
 142                         i++;
 143                         /* fallthrough */
 144                 case 1:
 145                         ino = parent_ino(dentry);
 146                         if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
 147                                 break;
 148                         filp->f_pos++;
 149                         i++;
 150                         /* fallthrough */
 151                 default:
 152                         spin_lock(&dcache_lock);
 153                         if (filp->f_pos == 2)
 154                                 list_move(q, &dentry->d_subdirs);
 155
 156                         for (p=q->next; p != &dentry->d_subdirs; p=p->next) {
 157                                 struct dentry *next;
 158                                 next = list_entry(p, struct dentry, d_u.d_child);
 159                                 if (d_unhashed(next) || !next->d_inode)
 160                                         continue;
 161                                 if (filter && !filter(next))
 162                                         continue;
 163
 164                                 spin_unlock(&dcache_lock);
 165                                 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)
 166                                         return 0;
 167                                 spin_lock(&dcache_lock);
 168                                 /* next is still alive */
 169                                 list_move(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 const 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         .fsync          = simple_sync_file,
 202 };
 203
 204 struct inode_operations simple_dir_inode_operations = {
 205         .lookup         = simple_lookup,
 206 };
 207
 208 /*
 209  * Common helper for pseudo-filesystems (sockfs, pipefs, bdev - stuff that
 210  * will never be mountable)
 211  */
 212 int get_sb_pseudo(struct file_system_type *fs_type, char *name,
 213         struct super_operations *ops, unsigned long magic,
 214         struct vfsmount *mnt)
 215 {
 216         struct super_block *s = sget(fs_type, NULL, set_anon_super, NULL);
 217         static struct super_operations default_ops = {.statfs = simple_statfs};
 218         struct dentry *dentry;
 219         struct inode *root;
 220         struct qstr d_name = {.name = name, .len = strlen(name)};
 221
 222         if (IS_ERR(s))
 223                 return PTR_ERR(s);
 224
 225         s->s_flags = MS_NOUSER;
 226         s->s_maxbytes = ~0ULL;
 227         s->s_blocksize = 1024;
 228         s->s_blocksize_bits = 10;
 229         s->s_magic = magic;
 230         s->s_op = ops ? ops : &default_ops;
 231         s->s_time_gran = 1;
 232         root = new_inode(s);
 233         if (!root)
 234                 goto Enomem;
 235         root->i_mode = S_IFDIR | S_IRUSR | S_IWUSR;
 236         root->i_uid = root->i_gid = 0;
 237         root->i_atime = root->i_mtime = root->i_ctime = CURRENT_TIME;
 238         dentry = d_alloc(NULL, &d_name);
 239         if (!dentry) {
 240                 iput(root);
 241                 goto Enomem;
 242         }
 243         dentry->d_sb = s;
 244         dentry->d_parent = dentry;
 245         d_instantiate(dentry, root);
 246         s->s_root = dentry;
 247         s->s_flags |= MS_ACTIVE;
 248         return simple_set_mnt(mnt, s);
 249
 250 Enomem:
 251         up_write(&s->s_umount);
 252         deactivate_super(s);
 253         return -ENOMEM;
 254 }
 255
 256 int simple_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
 257 {
 258         struct inode *inode = old_dentry->d_inode;
 259
 260         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 261         inode->i_nlink++;
 262         atomic_inc(&inode->i_count);
 263         dget(dentry);
 264         d_instantiate(dentry, inode);
 265         return 0;
 266 }
 267
 268 static inline int simple_positive(struct dentry *dentry)
 269 {
 270         return dentry->d_inode && !d_unhashed(dentry);
 271 }
 272
 273 int simple_empty(struct dentry *dentry)
 274 {
 275         struct dentry *child;
 276         int ret = 0;
 277
 278         spin_lock(&dcache_lock);
 279         list_for_each_entry(child, &dentry->d_subdirs, d_u.d_child)
 280                 if (simple_positive(child))
 281                         goto out;
 282         ret = 1;
 283 out:
 284         spin_unlock(&dcache_lock);
 285         return ret;
 286 }
 287
 288 int simple_unlink(struct inode *dir, struct dentry *dentry)
 289 {
 290         struct inode *inode = dentry->d_inode;
 291
 292         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 293         inode->i_nlink--;
 294         dput(dentry);
 295         return 0;
 296 }
 297
 298 int simple_rmdir(struct inode *dir, struct dentry *dentry)
 299 {
 300         if (!simple_empty(dentry))
 301                 return -ENOTEMPTY;
 302
 303         dentry->d_inode->i_nlink--;
 304         simple_unlink(dir, dentry);
 305         dir->i_nlink--;
 306         return 0;
 307 }
 308
 309 int simple_rename(struct inode *old_dir, struct dentry *old_dentry,
 310                 struct inode *new_dir, struct dentry *new_dentry)
 311 {
 312         struct inode *inode = old_dentry->d_inode;
 313         int they_are_dirs = S_ISDIR(old_dentry->d_inode->i_mode);
 314
 315         if (!simple_empty(new_dentry))
 316                 return -ENOTEMPTY;
 317
 318         if (new_dentry->d_inode) {
 319                 simple_unlink(new_dir, new_dentry);
 320                 if (they_are_dirs)
 321                         old_dir->i_nlink--;
 322         } else if (they_are_dirs) {
 323                 old_dir->i_nlink--;
 324                 new_dir->i_nlink++;
 325         }
 326
 327         old_dir->i_ctime = old_dir->i_mtime = new_dir->i_ctime =
 328                 new_dir->i_mtime = inode->i_ctime = CURRENT_TIME;
 329
 330         return 0;
 331 }
 332
 333 int simple_readpage(struct file *file, struct page *page)
 334 {
 335         void *kaddr;
 336
 337         if (PageUptodate(page))
 338                 goto out;
 339
 340         kaddr = kmap_atomic(page, KM_USER0);
 341         memset(kaddr, 0, PAGE_CACHE_SIZE);
 342         kunmap_atomic(kaddr, KM_USER0);
 343         flush_dcache_page(page);
 344         SetPageUptodate(page);
 345 out:
 346         unlock_page(page);
 347         return 0;
 348 }
 349
 350 int simple_prepare_write(struct file *file, struct page *page,
 351                         unsigned from, unsigned to)
 352 {
 353         if (!PageUptodate(page)) {
 354                 if (to - from != PAGE_CACHE_SIZE) {
 355                         void *kaddr = kmap_atomic(page, KM_USER0);
 356                         memset(kaddr, 0, from);
 357                         memset(kaddr + to, 0, PAGE_CACHE_SIZE - to);
 358                         flush_dcache_page(page);
 359                         kunmap_atomic(kaddr, KM_USER0);
 360                 }
 361                 SetPageUptodate(page);
 362         }
 363         return 0;
 364 }
 365
 366 int simple_commit_write(struct file *file, struct page *page,
 367                         unsigned offset, unsigned to)
 368 {
 369         struct inode *inode = page->mapping->host;
 370         loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
 371
 372         /*
 373          * No need to use i_size_read() here, the i_size
 374          * cannot change under us because we hold the i_mutex.
 375          */
 376         if (pos > inode->i_size)
 377                 i_size_write(inode, pos);
 378         set_page_dirty(page);
 379         return 0;
 380 }
 381
 382 int simple_fill_super(struct super_block *s, int magic, struct tree_descr *files)
 383 {
 384         static struct super_operations s_ops = {.statfs = simple_statfs};
 385         struct inode *inode;
 386         struct dentry *root;
 387         struct dentry *dentry;
 388         int i;
 389
 390         s->s_blocksize = PAGE_CACHE_SIZE;
 391         s->s_blocksize_bits = PAGE_CACHE_SHIFT;
 392         s->s_magic = magic;
 393         s->s_op = &s_ops;
 394         s->s_time_gran = 1;
 395
 396         inode = new_inode(s);
 397         if (!inode)
 398                 return -ENOMEM;
 399         inode->i_mode = S_IFDIR | 0755;
 400         inode->i_uid = inode->i_gid = 0;
 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         inode->i_nlink = 2;
 406         root = d_alloc_root(inode);
 407         if (!root) {
 408                 iput(inode);
 409                 return -ENOMEM;
 410         }
 411         for (i = 0; !files->name || files->name[0]; i++, files++) {
 412                 if (!files->name)
 413                         continue;
 414                 dentry = d_alloc_name(root, files->name);
 415                 if (!dentry)
 416                         goto out;
 417                 inode = new_inode(s);
 418                 if (!inode)
 419                         goto out;
 420                 inode->i_mode = S_IFREG | files->mode;
 421                 inode->i_uid = inode->i_gid = 0;
 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(struct file_system_type *type, 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 = vfs_kern_mount(type, 0, type->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 /* Simple attribute files */
 538
 539 struct simple_attr {
 540         u64 (*get)(void *);
 541         void (*set)(void *, u64);
 542         char get_buf[24];       /* enough to store a u64 and "\n\0" */
 543         char set_buf[24];
 544         void *data;
 545         const char *fmt;        /* format for read operation */
 546         struct mutex mutex;     /* protects access to these buffers */
 547 };
 548
 549 /* simple_attr_open is called by an actual attribute open file operation
 550  * to set the attribute specific access operations. */
 551 int simple_attr_open(struct inode *inode, struct file *file,
 552                      u64 (*get)(void *), void (*set)(void *, u64),
 553                      const char *fmt)
 554 {
 555         struct simple_attr *attr;
 556
 557         attr = kmalloc(sizeof(*attr), GFP_KERNEL);
 558         if (!attr)
 559                 return -ENOMEM;
 560
 561         attr->get = get;
 562         attr->set = set;
 563         attr->data = inode->i_private;
 564         attr->fmt = fmt;
 565         mutex_init(&attr->mutex);
 566
 567         file->private_data = attr;
 568
 569         return nonseekable_open(inode, file);
 570 }
 571
 572 int simple_attr_close(struct inode *inode, struct file *file)
 573 {
 574         kfree(file->private_data);
 575         return 0;
 576 }
 577
 578 /* read from the buffer that is filled with the get function */
 579 ssize_t simple_attr_read(struct file *file, char __user *buf,
 580                          size_t len, loff_t *ppos)
 581 {
 582         struct simple_attr *attr;
 583         size_t size;
 584         ssize_t ret;
 585
 586         attr = file->private_data;
 587
 588         if (!attr->get)
 589                 return -EACCES;
 590
 591         mutex_lock(&attr->mutex);
 592         if (*ppos) /* continued read */
 593                 size = strlen(attr->get_buf);
 594         else      /* first read */
 595                 size = scnprintf(attr->get_buf, sizeof(attr->get_buf),
 596                                  attr->fmt,
 597                                  (unsigned long long)attr->get(attr->data));
 598
 599         ret = simple_read_from_buffer(buf, len, ppos, attr->get_buf, size);
 600         mutex_unlock(&attr->mutex);
 601         return ret;
 602 }
 603
 604 /* interpret the buffer as a number to call the set function with */
 605 ssize_t simple_attr_write(struct file *file, const char __user *buf,
 606                           size_t len, loff_t *ppos)
 607 {
 608         struct simple_attr *attr;
 609         u64 val;
 610         size_t size;
 611         ssize_t ret;
 612
 613         attr = file->private_data;
 614
 615         if (!attr->set)
 616                 return -EACCES;
 617
 618         mutex_lock(&attr->mutex);
 619         ret = -EFAULT;
 620         size = min(sizeof(attr->set_buf) - 1, len);
 621         if (copy_from_user(attr->set_buf, buf, size))
 622                 goto out;
 623
 624         ret = len; /* claim we got the whole input */
 625         attr->set_buf[size] = '\0';
 626         val = simple_strtol(attr->set_buf, NULL, 0);
 627         attr->set(attr->data, val);
 628 out:
 629         mutex_unlock(&attr->mutex);
 630         return ret;
 631 }
 632
 633 EXPORT_SYMBOL(dcache_dir_close);
 634 EXPORT_SYMBOL(dcache_dir_lseek);
 635 EXPORT_SYMBOL(dcache_dir_open);
 636 EXPORT_SYMBOL(dcache_readdir);
 637 EXPORT_SYMBOL(dcache_readdir_filter);
 638 EXPORT_SYMBOL(generic_read_dir);
 639 EXPORT_SYMBOL(get_sb_pseudo);
 640 EXPORT_SYMBOL(simple_commit_write);
 641 EXPORT_SYMBOL(simple_dir_inode_operations);
 642 EXPORT_SYMBOL(simple_dir_operations);
 643 EXPORT_SYMBOL(simple_empty);
 644 EXPORT_SYMBOL(d_alloc_name);
 645 EXPORT_SYMBOL(simple_fill_super);
 646 EXPORT_SYMBOL(simple_getattr);
 647 EXPORT_SYMBOL(simple_link);
 648 EXPORT_SYMBOL(simple_lookup);
 649 EXPORT_SYMBOL(simple_pin_fs);
 650 EXPORT_SYMBOL(simple_prepare_write);
 651 EXPORT_SYMBOL(simple_readpage);
 652 EXPORT_SYMBOL(simple_release_fs);
 653 EXPORT_SYMBOL(simple_rename);
 654 EXPORT_SYMBOL(simple_rmdir);
 655 EXPORT_SYMBOL(simple_statfs);
 656 EXPORT_SYMBOL(simple_sync_file);
 657 EXPORT_SYMBOL(simple_unlink);
 658 EXPORT_SYMBOL(simple_read_from_buffer);
 659 EXPORT_SYMBOL(simple_transaction_get);
 660 EXPORT_SYMBOL(simple_transaction_read);
 661 EXPORT_SYMBOL(simple_transaction_release);
 662 EXPORT_SYMBOL_GPL(simple_attr_open);
 663 EXPORT_SYMBOL_GPL(simple_attr_close);
 664 EXPORT_SYMBOL_GPL(simple_attr_read);
 665 EXPORT_SYMBOL_GPL(simple_attr_write);