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,
  37                                        int mode,
  38                                        struct file_operations *fops,
  39                                        void *data)
  40 {
  41         struct inode *inode;
  42
  43         inode = new_inode(sb);
  44         if (!inode)
  45                 return NULL;
  46
  47         inode->i_mode = mode;
  48         inode->i_uid = 0;
  49         inode->i_gid = 0;
  50         inode->i_blksize = PAGE_CACHE_SIZE;
  51         inode->i_blocks = 0;
  52         inode->i_mapping->backing_dev_info = &relayfs_backing_dev_info;
  53         inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
  54         switch (mode & S_IFMT) {
  55         case S_IFREG:
  56                 inode->i_fop = fops;
  57                 if (data)
  58                         inode->u.generic_ip = data;
  59                 break;
  60         case S_IFDIR:
  61                 inode->i_op = &simple_dir_inode_operations;
  62                 inode->i_fop = &simple_dir_operations;
  63
  64                 /* directory inodes start off with i_nlink == 2 (for "." entry) */
  65                 inode->i_nlink++;
  66                 break;
  67         default:
  68                 break;
  69         }
  70
  71         return inode;
  72 }
  73
  74 /**
  75  *      relayfs_create_entry - create a relayfs directory or file
  76  *      @name: the name of the file to create
  77  *      @parent: parent directory
  78  *      @mode: mode
  79  *      @fops: file operations to use for the file
  80  *      @data: user-associated data for this file
  81  *
  82  *      Returns the new dentry, NULL on failure
  83  *
  84  *      Creates a file or directory with the specifed permissions.
  85  */
  86 static struct dentry *relayfs_create_entry(const char *name,
  87                                            struct dentry *parent,
  88                                            int mode,
  89                                            struct file_operations *fops,
  90                                            void *data)
  91 {
  92         struct dentry *d;
  93         struct inode *inode;
  94         int error = 0;
  95
  96         BUG_ON(!name || !(S_ISREG(mode) || S_ISDIR(mode)));
  97
  98         error = simple_pin_fs("relayfs", &relayfs_mount, &relayfs_mount_count);
  99         if (error) {
 100                 printk(KERN_ERR "Couldn't mount relayfs: errcode %d\n", error);
 101                 return NULL;
 102         }
 103
 104         if (!parent && relayfs_mount && relayfs_mount->mnt_sb)
 105                 parent = relayfs_mount->mnt_sb->s_root;
 106
 107         if (!parent) {
 108                 simple_release_fs(&relayfs_mount, &relayfs_mount_count);
 109                 return NULL;
 110         }
 111
 112         parent = dget(parent);
 113         down(&parent->d_inode->i_sem);
 114         d = lookup_one_len(name, parent, strlen(name));
 115         if (IS_ERR(d)) {
 116                 d = NULL;
 117                 goto release_mount;
 118         }
 119
 120         if (d->d_inode) {
 121                 d = NULL;
 122                 goto release_mount;
 123         }
 124
 125         inode = relayfs_get_inode(parent->d_inode->i_sb, mode, fops, data);
 126         if (!inode) {
 127                 d = NULL;
 128                 goto release_mount;
 129         }
 130
 131         d_instantiate(d, inode);
 132         dget(d);        /* Extra count - pin the dentry in core */
 133
 134         if (S_ISDIR(mode))
 135                 parent->d_inode->i_nlink++;
 136
 137         goto exit;
 138
 139 release_mount:
 140         simple_release_fs(&relayfs_mount, &relayfs_mount_count);
 141
 142 exit:
 143         up(&parent->d_inode->i_sem);
 144         dput(parent);
 145         return d;
 146 }
 147
 148 /**
 149  *      relayfs_create_file - create a file in the relay filesystem
 150  *      @name: the name of the file to create
 151  *      @parent: parent directory
 152  *      @mode: mode, if not specied the default perms are used
 153  *      @fops: file operations to use for the file
 154  *      @data: user-associated data for this file
 155  *
 156  *      Returns file dentry if successful, NULL otherwise.
 157  *
 158  *      The file will be created user r on behalf of current user.
 159  */
 160 struct dentry *relayfs_create_file(const char *name,
 161                                    struct dentry *parent,
 162                                    int mode,
 163                                    struct file_operations *fops,
 164                                    void *data)
 165 {
 166         BUG_ON(!fops);
 167
 168         if (!mode)
 169                 mode = S_IRUSR;
 170         mode = (mode & S_IALLUGO) | S_IFREG;
 171
 172         return relayfs_create_entry(name, parent, mode, fops, data);
 173 }
 174
 175 /**
 176  *      relayfs_create_dir - create a directory in the relay filesystem
 177  *      @name: the name of the directory to create
 178  *      @parent: parent directory, NULL if parent should be fs root
 179  *
 180  *      Returns directory dentry if successful, NULL otherwise.
 181  *
 182  *      The directory will be created world rwx on behalf of current user.
 183  */
 184 struct dentry *relayfs_create_dir(const char *name, struct dentry *parent)
 185 {
 186         int mode = S_IFDIR | S_IRWXU | S_IRUGO | S_IXUGO;
 187         return relayfs_create_entry(name, parent, mode, NULL, NULL);
 188 }
 189
 190 /**
 191  *      relayfs_remove - remove a file or directory in the relay filesystem
 192  *      @dentry: file or directory dentry
 193  *
 194  *      Returns 0 if successful, negative otherwise.
 195  */
 196 int relayfs_remove(struct dentry *dentry)
 197 {
 198         struct dentry *parent;
 199         int error = 0;
 200
 201         if (!dentry)
 202                 return -EINVAL;
 203         parent = dentry->d_parent;
 204         if (!parent)
 205                 return -EINVAL;
 206
 207         parent = dget(parent);
 208         down(&parent->d_inode->i_sem);
 209         if (dentry->d_inode) {
 210                 if (S_ISDIR(dentry->d_inode->i_mode))
 211                         error = simple_rmdir(parent->d_inode, dentry);
 212                 else
 213                         error = simple_unlink(parent->d_inode, dentry);
 214                 if (!error)
 215                         d_delete(dentry);
 216         }
 217         if (!error)
 218                 dput(dentry);
 219         up(&parent->d_inode->i_sem);
 220         dput(parent);
 221
 222         if (!error)
 223                 simple_release_fs(&relayfs_mount, &relayfs_mount_count);
 224
 225         return error;
 226 }
 227
 228 /**
 229  *      relayfs_remove_file - remove a file from relay filesystem
 230  *      @dentry: directory dentry
 231  *
 232  *      Returns 0 if successful, negative otherwise.
 233  */
 234 int relayfs_remove_file(struct dentry *dentry)
 235 {
 236         return relayfs_remove(dentry);
 237 }
 238
 239 /**
 240  *      relayfs_remove_dir - remove a directory in the relay filesystem
 241  *      @dentry: directory dentry
 242  *
 243  *      Returns 0 if successful, negative otherwise.
 244  */
 245 int relayfs_remove_dir(struct dentry *dentry)
 246 {
 247         return relayfs_remove(dentry);
 248 }
 249
 250 /**
 251  *      relayfs_open - open file op for relayfs files
 252  *      @inode: the inode
 253  *      @filp: the file
 254  *
 255  *      Increments the channel buffer refcount.
 256  */
 257 static int relayfs_open(struct inode *inode, struct file *filp)
 258 {
 259         struct rchan_buf *buf = inode->u.generic_ip;
 260         kref_get(&buf->kref);
 261         filp->private_data = buf;
 262
 263         return 0;
 264 }
 265
 266 /**
 267  *      relayfs_mmap - mmap file op for relayfs files
 268  *      @filp: the file
 269  *      @vma: the vma describing what to map
 270  *
 271  *      Calls upon relay_mmap_buf to map the file into user space.
 272  */
 273 static int relayfs_mmap(struct file *filp, struct vm_area_struct *vma)
 274 {
 275         struct rchan_buf *buf = filp->private_data;
 276         return relay_mmap_buf(buf, vma);
 277 }
 278
 279 /**
 280  *      relayfs_poll - poll file op for relayfs files
 281  *      @filp: the file
 282  *      @wait: poll table
 283  *
 284  *      Poll implemention.
 285  */
 286 static unsigned int relayfs_poll(struct file *filp, poll_table *wait)
 287 {
 288         unsigned int mask = 0;
 289         struct rchan_buf *buf = filp->private_data;
 290
 291         if (buf->finalized)
 292                 return POLLERR;
 293
 294         if (filp->f_mode & FMODE_READ) {
 295                 poll_wait(filp, &buf->read_wait, wait);
 296                 if (!relay_buf_empty(buf))
 297                         mask |= POLLIN | POLLRDNORM;
 298         }
 299
 300         return mask;
 301 }
 302
 303 /**
 304  *      relayfs_release - release file op for relayfs files
 305  *      @inode: the inode
 306  *      @filp: the file
 307  *
 308  *      Decrements the channel refcount, as the filesystem is
 309  *      no longer using it.
 310  */
 311 static int relayfs_release(struct inode *inode, struct file *filp)
 312 {
 313         struct rchan_buf *buf = filp->private_data;
 314         kref_put(&buf->kref, relay_remove_buf);
 315
 316         return 0;
 317 }
 318
 319 /**
 320  *      relayfs_read_consume - update the consumed count for the buffer
 321  */
 322 static void relayfs_read_consume(struct rchan_buf *buf,
 323                                  size_t read_pos,
 324                                  size_t bytes_consumed)
 325 {
 326         size_t subbuf_size = buf->chan->subbuf_size;
 327         size_t n_subbufs = buf->chan->n_subbufs;
 328         size_t read_subbuf;
 329
 330         if (buf->bytes_consumed + bytes_consumed > subbuf_size) {
 331                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
 332                 buf->bytes_consumed = 0;
 333         }
 334
 335         buf->bytes_consumed += bytes_consumed;
 336         read_subbuf = read_pos / buf->chan->subbuf_size;
 337         if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) {
 338                 if ((read_subbuf == buf->subbufs_produced % n_subbufs) &&
 339                     (buf->offset == subbuf_size))
 340                         return;
 341                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
 342                 buf->bytes_consumed = 0;
 343         }
 344 }
 345
 346 /**
 347  *      relayfs_read_avail - boolean, are there unconsumed bytes available?
 348  */
 349 static int relayfs_read_avail(struct rchan_buf *buf, size_t read_pos)
 350 {
 351         size_t bytes_produced, bytes_consumed, write_offset;
 352         size_t subbuf_size = buf->chan->subbuf_size;
 353         size_t n_subbufs = buf->chan->n_subbufs;
 354         size_t produced = buf->subbufs_produced % n_subbufs;
 355         size_t consumed = buf->subbufs_consumed % n_subbufs;
 356
 357         write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
 358
 359         if (consumed > produced) {
 360                 if ((produced > n_subbufs) &&
 361                     (produced + n_subbufs - consumed <= n_subbufs))
 362                         produced += n_subbufs;
 363         } else if (consumed == produced) {
 364                 if (buf->offset > subbuf_size) {
 365                         produced += n_subbufs;
 366                         if (buf->subbufs_produced == buf->subbufs_consumed)
 367                                 consumed += n_subbufs;
 368                 }
 369         }
 370
 371         if (buf->offset > subbuf_size)
 372                 bytes_produced = (produced - 1) * subbuf_size + write_offset;
 373         else
 374                 bytes_produced = produced * subbuf_size + write_offset;
 375         bytes_consumed = consumed * subbuf_size + buf->bytes_consumed;
 376
 377         if (bytes_produced == bytes_consumed)
 378                 return 0;
 379
 380         relayfs_read_consume(buf, read_pos, 0);
 381
 382         return 1;
 383 }
 384
 385 /**
 386  *      relayfs_read_subbuf_avail - return bytes available in sub-buffer
 387  */
 388 static size_t relayfs_read_subbuf_avail(size_t read_pos,
 389                                         struct rchan_buf *buf)
 390 {
 391         size_t padding, avail = 0;
 392         size_t read_subbuf, read_offset, write_subbuf, write_offset;
 393         size_t subbuf_size = buf->chan->subbuf_size;
 394
 395         write_subbuf = (buf->data - buf->start) / subbuf_size;
 396         write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
 397         read_subbuf = read_pos / subbuf_size;
 398         read_offset = read_pos % subbuf_size;
 399         padding = buf->padding[read_subbuf];
 400
 401         if (read_subbuf == write_subbuf) {
 402                 if (read_offset + padding < write_offset)
 403                         avail = write_offset - (read_offset + padding);
 404         } else
 405                 avail = (subbuf_size - padding) - read_offset;
 406
 407         return avail;
 408 }
 409
 410 /**
 411  *      relayfs_read_start_pos - find the first available byte to read
 412  *
 413  *      If the read_pos is in the middle of padding, return the
 414  *      position of the first actually available byte, otherwise
 415  *      return the original value.
 416  */
 417 static size_t relayfs_read_start_pos(size_t read_pos,
 418                                      struct rchan_buf *buf)
 419 {
 420         size_t read_subbuf, padding, padding_start, padding_end;
 421         size_t subbuf_size = buf->chan->subbuf_size;
 422         size_t n_subbufs = buf->chan->n_subbufs;
 423
 424         read_subbuf = read_pos / subbuf_size;
 425         padding = buf->padding[read_subbuf];
 426         padding_start = (read_subbuf + 1) * subbuf_size - padding;
 427         padding_end = (read_subbuf + 1) * subbuf_size;
 428         if (read_pos >= padding_start && read_pos < padding_end) {
 429                 read_subbuf = (read_subbuf + 1) % n_subbufs;
 430                 read_pos = read_subbuf * subbuf_size;
 431         }
 432
 433         return read_pos;
 434 }
 435
 436 /**
 437  *      relayfs_read_end_pos - return the new read position
 438  */
 439 static size_t relayfs_read_end_pos(struct rchan_buf *buf,
 440                                    size_t read_pos,
 441                                    size_t count)
 442 {
 443         size_t read_subbuf, padding, end_pos;
 444         size_t subbuf_size = buf->chan->subbuf_size;
 445         size_t n_subbufs = buf->chan->n_subbufs;
 446
 447         read_subbuf = read_pos / subbuf_size;
 448         padding = buf->padding[read_subbuf];
 449         if (read_pos % subbuf_size + count + padding == subbuf_size)
 450                 end_pos = (read_subbuf + 1) * subbuf_size;
 451         else
 452                 end_pos = read_pos + count;
 453         if (end_pos >= subbuf_size * n_subbufs)
 454                 end_pos = 0;
 455
 456         return end_pos;
 457 }
 458
 459 /**
 460  *      relayfs_read - read file op for relayfs files
 461  *      @filp: the file
 462  *      @buffer: the userspace buffer
 463  *      @count: number of bytes to read
 464  *      @ppos: position to read from
 465  *
 466  *      Reads count bytes or the number of bytes available in the
 467  *      current sub-buffer being read, whichever is smaller.
 468  */
 469 static ssize_t relayfs_read(struct file *filp,
 470                             char __user *buffer,
 471                             size_t count,
 472                             loff_t *ppos)
 473 {
 474         struct rchan_buf *buf = filp->private_data;
 475         struct inode *inode = filp->f_dentry->d_inode;
 476         size_t read_start, avail;
 477         ssize_t ret = 0;
 478         void *from;
 479
 480         down(&inode->i_sem);
 481         if(!relayfs_read_avail(buf, *ppos))
 482                 goto out;
 483
 484         read_start = relayfs_read_start_pos(*ppos, buf);
 485         avail = relayfs_read_subbuf_avail(read_start, buf);
 486         if (!avail)
 487                 goto out;
 488
 489         from = buf->start + read_start;
 490         ret = count = min(count, avail);
 491         if (copy_to_user(buffer, from, count)) {
 492                 ret = -EFAULT;
 493                 goto out;
 494         }
 495         relayfs_read_consume(buf, read_start, count);
 496         *ppos = relayfs_read_end_pos(buf, read_start, count);
 497 out:
 498         up(&inode->i_sem);
 499         return ret;
 500 }
 501
 502 /**
 503  *      relayfs alloc_inode() implementation
 504  */
 505 static struct inode *relayfs_alloc_inode(struct super_block *sb)
 506 {
 507         struct relayfs_inode_info *p = kmem_cache_alloc(relayfs_inode_cachep, SLAB_KERNEL);
 508         if (!p)
 509                 return NULL;
 510         p->data = NULL;
 511
 512         return &p->vfs_inode;
 513 }
 514
 515 /**
 516  *      relayfs destroy_inode() implementation
 517  */
 518 static void relayfs_destroy_inode(struct inode *inode)
 519 {
 520         kmem_cache_free(relayfs_inode_cachep, RELAYFS_I(inode));
 521 }
 522
 523 static void init_once(void *p, kmem_cache_t *cachep, unsigned long flags)
 524 {
 525         struct relayfs_inode_info *i = p;
 526         if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) == SLAB_CTOR_CONSTRUCTOR)
 527                 inode_init_once(&i->vfs_inode);
 528 }
 529
 530 struct file_operations relayfs_file_operations = {
 531         .open           = relayfs_open,
 532         .poll           = relayfs_poll,
 533         .mmap           = relayfs_mmap,
 534         .read           = relayfs_read,
 535         .llseek         = no_llseek,
 536         .release        = relayfs_release,
 537 };
 538
 539 static struct super_operations relayfs_ops = {
 540         .statfs         = simple_statfs,
 541         .drop_inode     = generic_delete_inode,
 542         .alloc_inode    = relayfs_alloc_inode,
 543         .destroy_inode  = relayfs_destroy_inode,
 544 };
 545
 546 static int relayfs_fill_super(struct super_block * sb, void * data, int silent)
 547 {
 548         struct inode *inode;
 549         struct dentry *root;
 550         int mode = S_IFDIR | S_IRWXU | S_IRUGO | S_IXUGO;
 551
 552         sb->s_blocksize = PAGE_CACHE_SIZE;
 553         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
 554         sb->s_magic = RELAYFS_MAGIC;
 555         sb->s_op = &relayfs_ops;
 556         inode = relayfs_get_inode(sb, mode, NULL, NULL);
 557
 558         if (!inode)
 559                 return -ENOMEM;
 560
 561         root = d_alloc_root(inode);
 562         if (!root) {
 563                 iput(inode);
 564                 return -ENOMEM;
 565         }
 566         sb->s_root = root;
 567
 568         return 0;
 569 }
 570
 571 static struct super_block * relayfs_get_sb(struct file_system_type *fs_type,
 572                                            int flags, const char *dev_name,
 573                                            void *data)
 574 {
 575         return get_sb_single(fs_type, flags, data, relayfs_fill_super);
 576 }
 577
 578 static struct file_system_type relayfs_fs_type = {
 579         .owner          = THIS_MODULE,
 580         .name           = "relayfs",
 581         .get_sb         = relayfs_get_sb,
 582         .kill_sb        = kill_litter_super,
 583 };
 584
 585 static int __init init_relayfs_fs(void)
 586 {
 587         int err;
 588
 589         relayfs_inode_cachep = kmem_cache_create("relayfs_inode_cache",
 590                                 sizeof(struct relayfs_inode_info), 0,
 591                                 0, init_once, NULL);
 592         if (!relayfs_inode_cachep)
 593                 return -ENOMEM;
 594
 595         err = register_filesystem(&relayfs_fs_type);
 596         if (err)
 597                 kmem_cache_destroy(relayfs_inode_cachep);
 598
 599         return err;
 600 }
 601
 602 static void __exit exit_relayfs_fs(void)
 603 {
 604         unregister_filesystem(&relayfs_fs_type);
 605         kmem_cache_destroy(relayfs_inode_cachep);
 606 }
 607
 608 module_init(init_relayfs_fs)
 609 module_exit(exit_relayfs_fs)
 610
 611 EXPORT_SYMBOL_GPL(relayfs_file_operations);
 612 EXPORT_SYMBOL_GPL(relayfs_create_dir);
 613 EXPORT_SYMBOL_GPL(relayfs_remove_dir);
 614 EXPORT_SYMBOL_GPL(relayfs_create_file);
 615 EXPORT_SYMBOL_GPL(relayfs_remove_file);
 616
 617 MODULE_AUTHOR("Tom Zanussi <zanussi@us.ibm.com> and Karim Yaghmour <karim@opersys.com>");
 618 MODULE_DESCRIPTION("Relay Filesystem");
 619 MODULE_LICENSE("GPL");
 620