err.no Git - linux-2.6/blob - fs/relayfs/inode.c

   1 /*
   2  * VFS-related code for RelayFS, a high-speed data relay filesystem.
   3  *
   4  * Copyright (C) 2003-2005 - Tom Zanussi <zanussi@us.ibm.com>, IBM Corp
   5  * Copyright (C) 2003-2005 - Karim Yaghmour <karim@opersys.com>
   6  *
   7  * Based on ramfs, Copyright (C) 2002 - Linus Torvalds
   8  *
   9  * This file is released under the GPL.
  10  */
  11
  12 #include <linux/module.h>
  13 #include <linux/fs.h>
  14 #include <linux/mount.h>
  15 #include <linux/pagemap.h>
  16 #include <linux/init.h>
  17 #include <linux/string.h>
  18 #include <linux/backing-dev.h>
  19 #include <linux/namei.h>
  20 #include <linux/poll.h>
  21 #include <linux/relayfs_fs.h>
  22 #include "relay.h"
  23 #include "buffers.h"
  24
  25 #define RELAYFS_MAGIC                   0xF0B4A981
  26
  27 static struct vfsmount *                relayfs_mount;
  28 static int                              relayfs_mount_count;
  29 static kmem_cache_t *                   relayfs_inode_cachep;
  30
  31 static struct backing_dev_info          relayfs_backing_dev_info = {
  32         .ra_pages       = 0,    /* No readahead */
  33         .capabilities   = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
  34 };
  35
  36 static struct inode *relayfs_get_inode(struct super_block *sb, int mode,
  37                                        void *data)
  38 {
  39         struct inode *inode;
  40
  41         inode = new_inode(sb);
  42         if (!inode)
  43                 return NULL;
  44
  45         inode->i_mode = mode;
  46         inode->i_uid = 0;
  47         inode->i_gid = 0;
  48         inode->i_blksize = PAGE_CACHE_SIZE;
  49         inode->i_blocks = 0;
  50         inode->i_mapping->backing_dev_info = &relayfs_backing_dev_info;
  51         inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
  52         switch (mode & S_IFMT) {
  53         case S_IFREG:
  54                 inode->i_fop = &relayfs_file_operations;
  55                 RELAYFS_I(inode)->buf = data;
  56                 break;
  57         case S_IFDIR:
  58                 inode->i_op = &simple_dir_inode_operations;
  59                 inode->i_fop = &simple_dir_operations;
  60
  61                 /* directory inodes start off with i_nlink == 2 (for "." entry) */
  62                 inode->i_nlink++;
  63                 break;
  64         default:
  65                 break;
  66         }
  67
  68         return inode;
  69 }
  70
  71 /**
  72  *      relayfs_create_entry - create a relayfs directory or file
  73  *      @name: the name of the file to create
  74  *      @parent: parent directory
  75  *      @mode: mode
  76  *      @data: user-associated data for this file
  77  *
  78  *      Returns the new dentry, NULL on failure
  79  *
  80  *      Creates a file or directory with the specifed permissions.
  81  */
  82 static struct dentry *relayfs_create_entry(const char *name,
  83                                            struct dentry *parent,
  84                                            int mode,
  85                                            void *data)
  86 {
  87         struct dentry *d;
  88         struct inode *inode;
  89         int error = 0;
  90
  91         BUG_ON(!name || !(S_ISREG(mode) || S_ISDIR(mode)));
  92
  93         error = simple_pin_fs("relayfs", &relayfs_mount, &relayfs_mount_count);
  94         if (error) {
  95                 printk(KERN_ERR "Couldn't mount relayfs: errcode %d\n", error);
  96                 return NULL;
  97         }
  98
  99         if (!parent && relayfs_mount && relayfs_mount->mnt_sb)
 100                 parent = relayfs_mount->mnt_sb->s_root;
 101
 102         if (!parent) {
 103                 simple_release_fs(&relayfs_mount, &relayfs_mount_count);
 104                 return NULL;
 105         }
 106
 107         parent = dget(parent);
 108         down(&parent->d_inode->i_sem);
 109         d = lookup_one_len(name, parent, strlen(name));
 110         if (IS_ERR(d)) {
 111                 d = NULL;
 112                 goto release_mount;
 113         }
 114
 115         if (d->d_inode) {
 116                 d = NULL;
 117                 goto release_mount;
 118         }
 119
 120         inode = relayfs_get_inode(parent->d_inode->i_sb, mode, data);
 121         if (!inode) {
 122                 d = NULL;
 123                 goto release_mount;
 124         }
 125
 126         d_instantiate(d, inode);
 127         dget(d);        /* Extra count - pin the dentry in core */
 128
 129         if (S_ISDIR(mode))
 130                 parent->d_inode->i_nlink++;
 131
 132         goto exit;
 133
 134 release_mount:
 135         simple_release_fs(&relayfs_mount, &relayfs_mount_count);
 136
 137 exit:
 138         up(&parent->d_inode->i_sem);
 139         dput(parent);
 140         return d;
 141 }
 142
 143 /**
 144  *      relayfs_create_file - create a file in the relay filesystem
 145  *      @name: the name of the file to create
 146  *      @parent: parent directory
 147  *      @mode: mode, if not specied the default perms are used
 148  *      @data: user-associated data for this file
 149  *
 150  *      Returns file dentry if successful, NULL otherwise.
 151  *
 152  *      The file will be created user r on behalf of current user.
 153  */
 154 struct dentry *relayfs_create_file(const char *name, struct dentry *parent,
 155                                    int mode, void *data)
 156 {
 157         if (!mode)
 158                 mode = S_IRUSR;
 159         mode = (mode & S_IALLUGO) | S_IFREG;
 160
 161         return relayfs_create_entry(name, parent, mode, data);
 162 }
 163
 164 /**
 165  *      relayfs_create_dir - create a directory in the relay filesystem
 166  *      @name: the name of the directory to create
 167  *      @parent: parent directory, NULL if parent should be fs root
 168  *
 169  *      Returns directory dentry if successful, NULL otherwise.
 170  *
 171  *      The directory will be created world rwx on behalf of current user.
 172  */
 173 struct dentry *relayfs_create_dir(const char *name, struct dentry *parent)
 174 {
 175         int mode = S_IFDIR | S_IRWXU | S_IRUGO | S_IXUGO;
 176         return relayfs_create_entry(name, parent, mode, NULL);
 177 }
 178
 179 /**
 180  *      relayfs_remove - remove a file or directory in the relay filesystem
 181  *      @dentry: file or directory dentry
 182  *
 183  *      Returns 0 if successful, negative otherwise.
 184  */
 185 int relayfs_remove(struct dentry *dentry)
 186 {
 187         struct dentry *parent;
 188         int error = 0;
 189
 190         if (!dentry)
 191                 return -EINVAL;
 192         parent = dentry->d_parent;
 193         if (!parent)
 194                 return -EINVAL;
 195
 196         parent = dget(parent);
 197         down(&parent->d_inode->i_sem);
 198         if (dentry->d_inode) {
 199                 if (S_ISDIR(dentry->d_inode->i_mode))
 200                         error = simple_rmdir(parent->d_inode, dentry);
 201                 else
 202                         error = simple_unlink(parent->d_inode, dentry);
 203                 if (!error)
 204                         d_delete(dentry);
 205         }
 206         if (!error)
 207                 dput(dentry);
 208         up(&parent->d_inode->i_sem);
 209         dput(parent);
 210
 211         if (!error)
 212                 simple_release_fs(&relayfs_mount, &relayfs_mount_count);
 213
 214         return error;
 215 }
 216
 217 /**
 218  *      relayfs_remove_dir - remove a directory in the relay filesystem
 219  *      @dentry: directory dentry
 220  *
 221  *      Returns 0 if successful, negative otherwise.
 222  */
 223 int relayfs_remove_dir(struct dentry *dentry)
 224 {
 225         return relayfs_remove(dentry);
 226 }
 227
 228 /**
 229  *      relayfs_open - open file op for relayfs files
 230  *      @inode: the inode
 231  *      @filp: the file
 232  *
 233  *      Increments the channel buffer refcount.
 234  */
 235 static int relayfs_open(struct inode *inode, struct file *filp)
 236 {
 237         struct rchan_buf *buf = RELAYFS_I(inode)->buf;
 238         kref_get(&buf->kref);
 239
 240         return 0;
 241 }
 242
 243 /**
 244  *      relayfs_mmap - mmap file op for relayfs files
 245  *      @filp: the file
 246  *      @vma: the vma describing what to map
 247  *
 248  *      Calls upon relay_mmap_buf to map the file into user space.
 249  */
 250 static int relayfs_mmap(struct file *filp, struct vm_area_struct *vma)
 251 {
 252         struct inode *inode = filp->f_dentry->d_inode;
 253         return relay_mmap_buf(RELAYFS_I(inode)->buf, vma);
 254 }
 255
 256 /**
 257  *      relayfs_poll - poll file op for relayfs files
 258  *      @filp: the file
 259  *      @wait: poll table
 260  *
 261  *      Poll implemention.
 262  */
 263 static unsigned int relayfs_poll(struct file *filp, poll_table *wait)
 264 {
 265         unsigned int mask = 0;
 266         struct inode *inode = filp->f_dentry->d_inode;
 267         struct rchan_buf *buf = RELAYFS_I(inode)->buf;
 268
 269         if (buf->finalized)
 270                 return POLLERR;
 271
 272         if (filp->f_mode & FMODE_READ) {
 273                 poll_wait(filp, &buf->read_wait, wait);
 274                 if (!relay_buf_empty(buf))
 275                         mask |= POLLIN | POLLRDNORM;
 276         }
 277
 278         return mask;
 279 }
 280
 281 /**
 282  *      relayfs_release - release file op for relayfs files
 283  *      @inode: the inode
 284  *      @filp: the file
 285  *
 286  *      Decrements the channel refcount, as the filesystem is
 287  *      no longer using it.
 288  */
 289 static int relayfs_release(struct inode *inode, struct file *filp)
 290 {
 291         struct rchan_buf *buf = RELAYFS_I(inode)->buf;
 292         kref_put(&buf->kref, relay_remove_buf);
 293
 294         return 0;
 295 }
 296
 297 /**
 298  *      relayfs_read_consume - update the consumed count for the buffer
 299  */
 300 static void relayfs_read_consume(struct rchan_buf *buf,
 301                                  size_t read_pos,
 302                                  size_t bytes_consumed)
 303 {
 304         size_t subbuf_size = buf->chan->subbuf_size;
 305         size_t n_subbufs = buf->chan->n_subbufs;
 306         size_t read_subbuf;
 307
 308         if (buf->bytes_consumed + bytes_consumed > subbuf_size) {
 309                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
 310                 buf->bytes_consumed = 0;
 311         }
 312
 313         buf->bytes_consumed += bytes_consumed;
 314         read_subbuf = read_pos / buf->chan->subbuf_size;
 315         if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) {
 316                 if ((read_subbuf == buf->subbufs_produced % n_subbufs) &&
 317                     (buf->offset == subbuf_size))
 318                         return;
 319                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
 320                 buf->bytes_consumed = 0;
 321         }
 322 }
 323
 324 /**
 325  *      relayfs_read_avail - boolean, are there unconsumed bytes available?
 326  */
 327 static int relayfs_read_avail(struct rchan_buf *buf, size_t read_pos)
 328 {
 329         size_t bytes_produced, bytes_consumed, write_offset;
 330         size_t subbuf_size = buf->chan->subbuf_size;
 331         size_t n_subbufs = buf->chan->n_subbufs;
 332         size_t produced = buf->subbufs_produced % n_subbufs;
 333         size_t consumed = buf->subbufs_consumed % n_subbufs;
 334
 335         write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
 336
 337         if (consumed > produced) {
 338                 if ((produced > n_subbufs) &&
 339                     (produced + n_subbufs - consumed <= n_subbufs))
 340                         produced += n_subbufs;
 341         } else if (consumed == produced) {
 342                 if (buf->offset > subbuf_size) {
 343                         produced += n_subbufs;
 344                         if (buf->subbufs_produced == buf->subbufs_consumed)
 345                                 consumed += n_subbufs;
 346                 }
 347         }
 348
 349         if (buf->offset > subbuf_size)
 350                 bytes_produced = (produced - 1) * subbuf_size + write_offset;
 351         else
 352                 bytes_produced = produced * subbuf_size + write_offset;
 353         bytes_consumed = consumed * subbuf_size + buf->bytes_consumed;
 354
 355         if (bytes_produced == bytes_consumed)
 356                 return 0;
 357
 358         relayfs_read_consume(buf, read_pos, 0);
 359
 360         return 1;
 361 }
 362
 363 /**
 364  *      relayfs_read_subbuf_avail - return bytes available in sub-buffer
 365  */
 366 static size_t relayfs_read_subbuf_avail(size_t read_pos,
 367                                         struct rchan_buf *buf)
 368 {
 369         size_t padding, avail = 0;
 370         size_t read_subbuf, read_offset, write_subbuf, write_offset;
 371         size_t subbuf_size = buf->chan->subbuf_size;
 372
 373         write_subbuf = (buf->data - buf->start) / subbuf_size;
 374         write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
 375         read_subbuf = read_pos / subbuf_size;
 376         read_offset = read_pos % subbuf_size;
 377         padding = buf->padding[read_subbuf];
 378
 379         if (read_subbuf == write_subbuf) {
 380                 if (read_offset + padding < write_offset)
 381                         avail = write_offset - (read_offset + padding);
 382         } else
 383                 avail = (subbuf_size - padding) - read_offset;
 384
 385         return avail;
 386 }
 387
 388 /**
 389  *      relayfs_read_start_pos - find the first available byte to read
 390  *
 391  *      If the read_pos is in the middle of padding, return the
 392  *      position of the first actually available byte, otherwise
 393  *      return the original value.
 394  */
 395 static size_t relayfs_read_start_pos(size_t read_pos,
 396                                      struct rchan_buf *buf)
 397 {
 398         size_t read_subbuf, padding, padding_start, padding_end;
 399         size_t subbuf_size = buf->chan->subbuf_size;
 400         size_t n_subbufs = buf->chan->n_subbufs;
 401
 402         read_subbuf = read_pos / subbuf_size;
 403         padding = buf->padding[read_subbuf];
 404         padding_start = (read_subbuf + 1) * subbuf_size - padding;
 405         padding_end = (read_subbuf + 1) * subbuf_size;
 406         if (read_pos >= padding_start && read_pos < padding_end) {
 407                 read_subbuf = (read_subbuf + 1) % n_subbufs;
 408                 read_pos = read_subbuf * subbuf_size;
 409         }
 410
 411         return read_pos;
 412 }
 413
 414 /**
 415  *      relayfs_read_end_pos - return the new read position
 416  */
 417 static size_t relayfs_read_end_pos(struct rchan_buf *buf,
 418                                    size_t read_pos,
 419                                    size_t count)
 420 {
 421         size_t read_subbuf, padding, end_pos;
 422         size_t subbuf_size = buf->chan->subbuf_size;
 423         size_t n_subbufs = buf->chan->n_subbufs;
 424
 425         read_subbuf = read_pos / subbuf_size;
 426         padding = buf->padding[read_subbuf];
 427         if (read_pos % subbuf_size + count + padding == subbuf_size)
 428                 end_pos = (read_subbuf + 1) * subbuf_size;
 429         else
 430                 end_pos = read_pos + count;
 431         if (end_pos >= subbuf_size * n_subbufs)
 432                 end_pos = 0;
 433
 434         return end_pos;
 435 }
 436
 437 /**
 438  *      relayfs_read - read file op for relayfs files
 439  *      @filp: the file
 440  *      @buffer: the userspace buffer
 441  *      @count: number of bytes to read
 442  *      @ppos: position to read from
 443  *
 444  *      Reads count bytes or the number of bytes available in the
 445  *      current sub-buffer being read, whichever is smaller.
 446  */
 447 static ssize_t relayfs_read(struct file *filp,
 448                             char __user *buffer,
 449                             size_t count,
 450                             loff_t *ppos)
 451 {
 452         struct inode *inode = filp->f_dentry->d_inode;
 453         struct rchan_buf *buf = RELAYFS_I(inode)->buf;
 454         size_t read_start, avail;
 455         ssize_t ret = 0;
 456         void *from;
 457
 458         down(&inode->i_sem);
 459         if(!relayfs_read_avail(buf, *ppos))
 460                 goto out;
 461
 462         read_start = relayfs_read_start_pos(*ppos, buf);
 463         avail = relayfs_read_subbuf_avail(read_start, buf);
 464         if (!avail)
 465                 goto out;
 466
 467         from = buf->start + read_start;
 468         ret = count = min(count, avail);
 469         if (copy_to_user(buffer, from, count)) {
 470                 ret = -EFAULT;
 471                 goto out;
 472         }
 473         relayfs_read_consume(buf, read_start, count);
 474         *ppos = relayfs_read_end_pos(buf, read_start, count);
 475 out:
 476         up(&inode->i_sem);
 477         return ret;
 478 }
 479
 480 /**
 481  *      relayfs alloc_inode() implementation
 482  */
 483 static struct inode *relayfs_alloc_inode(struct super_block *sb)
 484 {
 485         struct relayfs_inode_info *p = kmem_cache_alloc(relayfs_inode_cachep, SLAB_KERNEL);
 486         if (!p)
 487                 return NULL;
 488         p->buf = NULL;
 489
 490         return &p->vfs_inode;
 491 }
 492
 493 /**
 494  *      relayfs destroy_inode() implementation
 495  */
 496 static void relayfs_destroy_inode(struct inode *inode)
 497 {
 498         kmem_cache_free(relayfs_inode_cachep, RELAYFS_I(inode));
 499 }
 500
 501 static void init_once(void *p, kmem_cache_t *cachep, unsigned long flags)
 502 {
 503         struct relayfs_inode_info *i = p;
 504         if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) == SLAB_CTOR_CONSTRUCTOR)
 505                 inode_init_once(&i->vfs_inode);
 506 }
 507
 508 struct file_operations relayfs_file_operations = {
 509         .open           = relayfs_open,
 510         .poll           = relayfs_poll,
 511         .mmap           = relayfs_mmap,
 512         .read           = relayfs_read,
 513         .llseek         = no_llseek,
 514         .release        = relayfs_release,
 515 };
 516
 517 static struct super_operations relayfs_ops = {
 518         .statfs         = simple_statfs,
 519         .drop_inode     = generic_delete_inode,
 520         .alloc_inode    = relayfs_alloc_inode,
 521         .destroy_inode  = relayfs_destroy_inode,
 522 };
 523
 524 static int relayfs_fill_super(struct super_block * sb, void * data, int silent)
 525 {
 526         struct inode *inode;
 527         struct dentry *root;
 528         int mode = S_IFDIR | S_IRWXU | S_IRUGO | S_IXUGO;
 529
 530         sb->s_blocksize = PAGE_CACHE_SIZE;
 531         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
 532         sb->s_magic = RELAYFS_MAGIC;
 533         sb->s_op = &relayfs_ops;
 534         inode = relayfs_get_inode(sb, mode, NULL);
 535
 536         if (!inode)
 537                 return -ENOMEM;
 538
 539         root = d_alloc_root(inode);
 540         if (!root) {
 541                 iput(inode);
 542                 return -ENOMEM;
 543         }
 544         sb->s_root = root;
 545
 546         return 0;
 547 }
 548
 549 static struct super_block * relayfs_get_sb(struct file_system_type *fs_type,
 550                                            int flags, const char *dev_name,
 551                                            void *data)
 552 {
 553         return get_sb_single(fs_type, flags, data, relayfs_fill_super);
 554 }
 555
 556 static struct file_system_type relayfs_fs_type = {
 557         .owner          = THIS_MODULE,
 558         .name           = "relayfs",
 559         .get_sb         = relayfs_get_sb,
 560         .kill_sb        = kill_litter_super,
 561 };
 562
 563 static int __init init_relayfs_fs(void)
 564 {
 565         int err;
 566
 567         relayfs_inode_cachep = kmem_cache_create("relayfs_inode_cache",
 568                                 sizeof(struct relayfs_inode_info), 0,
 569                                 0, init_once, NULL);
 570         if (!relayfs_inode_cachep)
 571                 return -ENOMEM;
 572
 573         err = register_filesystem(&relayfs_fs_type);
 574         if (err)
 575                 kmem_cache_destroy(relayfs_inode_cachep);
 576
 577         return err;
 578 }
 579
 580 static void __exit exit_relayfs_fs(void)
 581 {
 582         unregister_filesystem(&relayfs_fs_type);
 583         kmem_cache_destroy(relayfs_inode_cachep);
 584 }
 585
 586 module_init(init_relayfs_fs)
 587 module_exit(exit_relayfs_fs)
 588
 589 EXPORT_SYMBOL_GPL(relayfs_file_operations);
 590 EXPORT_SYMBOL_GPL(relayfs_create_dir);
 591 EXPORT_SYMBOL_GPL(relayfs_remove_dir);
 592
 593 MODULE_AUTHOR("Tom Zanussi <zanussi@us.ibm.com> and Karim Yaghmour <karim@opersys.com>");
 594 MODULE_DESCRIPTION("Relay Filesystem");
 595 MODULE_LICENSE("GPL");
 596