Merge branch 'upstream-linus' of git://oss.oracle.com/home/sourcebo/git/ocfs2
[linux-2.6] / 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
30 static struct backing_dev_info          relayfs_backing_dev_info = {
31         .ra_pages       = 0,    /* No readahead */
32         .capabilities   = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
33 };
34
35 static struct inode *relayfs_get_inode(struct super_block *sb,
36                                        int mode,
37                                        struct file_operations *fops,
38                                        void *data)
39 {
40         struct inode *inode;
41
42         inode = new_inode(sb);
43         if (!inode)
44                 return NULL;
45
46         inode->i_mode = mode;
47         inode->i_uid = 0;
48         inode->i_gid = 0;
49         inode->i_blksize = PAGE_CACHE_SIZE;
50         inode->i_blocks = 0;
51         inode->i_mapping->backing_dev_info = &relayfs_backing_dev_info;
52         inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
53         switch (mode & S_IFMT) {
54         case S_IFREG:
55                 inode->i_fop = fops;
56                 if (data)
57                         inode->u.generic_ip = data;
58                 break;
59         case S_IFDIR:
60                 inode->i_op = &simple_dir_inode_operations;
61                 inode->i_fop = &simple_dir_operations;
62
63                 /* directory inodes start off with i_nlink == 2 (for "." entry) */
64                 inode->i_nlink++;
65                 break;
66         default:
67                 break;
68         }
69
70         return inode;
71 }
72
73 /**
74  *      relayfs_create_entry - create a relayfs directory or file
75  *      @name: the name of the file to create
76  *      @parent: parent directory
77  *      @mode: mode
78  *      @fops: file operations to use for the file
79  *      @data: user-associated data for this file
80  *
81  *      Returns the new dentry, NULL on failure
82  *
83  *      Creates a file or directory with the specifed permissions.
84  */
85 static struct dentry *relayfs_create_entry(const char *name,
86                                            struct dentry *parent,
87                                            int mode,
88                                            struct file_operations *fops,
89                                            void *data)
90 {
91         struct dentry *d;
92         struct inode *inode;
93         int error = 0;
94
95         BUG_ON(!name || !(S_ISREG(mode) || S_ISDIR(mode)));
96
97         error = simple_pin_fs("relayfs", &relayfs_mount, &relayfs_mount_count);
98         if (error) {
99                 printk(KERN_ERR "Couldn't mount relayfs: errcode %d\n", error);
100                 return NULL;
101         }
102
103         if (!parent && relayfs_mount && relayfs_mount->mnt_sb)
104                 parent = relayfs_mount->mnt_sb->s_root;
105
106         if (!parent) {
107                 simple_release_fs(&relayfs_mount, &relayfs_mount_count);
108                 return NULL;
109         }
110
111         parent = dget(parent);
112         mutex_lock(&parent->d_inode->i_mutex);
113         d = lookup_one_len(name, parent, strlen(name));
114         if (IS_ERR(d)) {
115                 d = NULL;
116                 goto release_mount;
117         }
118
119         if (d->d_inode) {
120                 d = NULL;
121                 goto release_mount;
122         }
123
124         inode = relayfs_get_inode(parent->d_inode->i_sb, mode, fops, data);
125         if (!inode) {
126                 d = NULL;
127                 goto release_mount;
128         }
129
130         d_instantiate(d, inode);
131         dget(d);        /* Extra count - pin the dentry in core */
132
133         if (S_ISDIR(mode))
134                 parent->d_inode->i_nlink++;
135
136         goto exit;
137
138 release_mount:
139         simple_release_fs(&relayfs_mount, &relayfs_mount_count);
140
141 exit:
142         mutex_unlock(&parent->d_inode->i_mutex);
143         dput(parent);
144         return d;
145 }
146
147 /**
148  *      relayfs_create_file - create a file in the relay filesystem
149  *      @name: the name of the file to create
150  *      @parent: parent directory
151  *      @mode: mode, if not specied the default perms are used
152  *      @fops: file operations to use for the file
153  *      @data: user-associated data for this file
154  *
155  *      Returns file dentry if successful, NULL otherwise.
156  *
157  *      The file will be created user r on behalf of current user.
158  */
159 struct dentry *relayfs_create_file(const char *name,
160                                    struct dentry *parent,
161                                    int mode,
162                                    struct file_operations *fops,
163                                    void *data)
164 {
165         BUG_ON(!fops);
166
167         if (!mode)
168                 mode = S_IRUSR;
169         mode = (mode & S_IALLUGO) | S_IFREG;
170
171         return relayfs_create_entry(name, parent, mode, fops, data);
172 }
173
174 /**
175  *      relayfs_create_dir - create a directory in the relay filesystem
176  *      @name: the name of the directory to create
177  *      @parent: parent directory, NULL if parent should be fs root
178  *
179  *      Returns directory dentry if successful, NULL otherwise.
180  *
181  *      The directory will be created world rwx on behalf of current user.
182  */
183 struct dentry *relayfs_create_dir(const char *name, struct dentry *parent)
184 {
185         int mode = S_IFDIR | S_IRWXU | S_IRUGO | S_IXUGO;
186         return relayfs_create_entry(name, parent, mode, NULL, NULL);
187 }
188
189 /**
190  *      relayfs_remove - remove a file or directory in the relay filesystem
191  *      @dentry: file or directory dentry
192  *
193  *      Returns 0 if successful, negative otherwise.
194  */
195 int relayfs_remove(struct dentry *dentry)
196 {
197         struct dentry *parent;
198         int error = 0;
199
200         if (!dentry)
201                 return -EINVAL;
202         parent = dentry->d_parent;
203         if (!parent)
204                 return -EINVAL;
205
206         parent = dget(parent);
207         mutex_lock(&parent->d_inode->i_mutex);
208         if (dentry->d_inode) {
209                 if (S_ISDIR(dentry->d_inode->i_mode))
210                         error = simple_rmdir(parent->d_inode, dentry);
211                 else
212                         error = simple_unlink(parent->d_inode, dentry);
213                 if (!error)
214                         d_delete(dentry);
215         }
216         if (!error)
217                 dput(dentry);
218         mutex_unlock(&parent->d_inode->i_mutex);
219         dput(parent);
220
221         if (!error)
222                 simple_release_fs(&relayfs_mount, &relayfs_mount_count);
223
224         return error;
225 }
226
227 /**
228  *      relayfs_remove_file - remove a file from relay filesystem
229  *      @dentry: directory dentry
230  *
231  *      Returns 0 if successful, negative otherwise.
232  */
233 int relayfs_remove_file(struct dentry *dentry)
234 {
235         return relayfs_remove(dentry);
236 }
237
238 /**
239  *      relayfs_remove_dir - remove a directory in the relay filesystem
240  *      @dentry: directory dentry
241  *
242  *      Returns 0 if successful, negative otherwise.
243  */
244 int relayfs_remove_dir(struct dentry *dentry)
245 {
246         return relayfs_remove(dentry);
247 }
248
249 /**
250  *      relay_file_open - open file op for relay files
251  *      @inode: the inode
252  *      @filp: the file
253  *
254  *      Increments the channel buffer refcount.
255  */
256 static int relay_file_open(struct inode *inode, struct file *filp)
257 {
258         struct rchan_buf *buf = inode->u.generic_ip;
259         kref_get(&buf->kref);
260         filp->private_data = buf;
261
262         return 0;
263 }
264
265 /**
266  *      relay_file_mmap - mmap file op for relay files
267  *      @filp: the file
268  *      @vma: the vma describing what to map
269  *
270  *      Calls upon relay_mmap_buf to map the file into user space.
271  */
272 static int relay_file_mmap(struct file *filp, struct vm_area_struct *vma)
273 {
274         struct rchan_buf *buf = filp->private_data;
275         return relay_mmap_buf(buf, vma);
276 }
277
278 /**
279  *      relay_file_poll - poll file op for relay files
280  *      @filp: the file
281  *      @wait: poll table
282  *
283  *      Poll implemention.
284  */
285 static unsigned int relay_file_poll(struct file *filp, poll_table *wait)
286 {
287         unsigned int mask = 0;
288         struct rchan_buf *buf = filp->private_data;
289
290         if (buf->finalized)
291                 return POLLERR;
292
293         if (filp->f_mode & FMODE_READ) {
294                 poll_wait(filp, &buf->read_wait, wait);
295                 if (!relay_buf_empty(buf))
296                         mask |= POLLIN | POLLRDNORM;
297         }
298
299         return mask;
300 }
301
302 /**
303  *      relay_file_release - release file op for relay files
304  *      @inode: the inode
305  *      @filp: the file
306  *
307  *      Decrements the channel refcount, as the filesystem is
308  *      no longer using it.
309  */
310 static int relay_file_release(struct inode *inode, struct file *filp)
311 {
312         struct rchan_buf *buf = filp->private_data;
313         kref_put(&buf->kref, relay_remove_buf);
314
315         return 0;
316 }
317
318 /**
319  *      relay_file_read_consume - update the consumed count for the buffer
320  */
321 static void relay_file_read_consume(struct rchan_buf *buf,
322                                     size_t read_pos,
323                                     size_t bytes_consumed)
324 {
325         size_t subbuf_size = buf->chan->subbuf_size;
326         size_t n_subbufs = buf->chan->n_subbufs;
327         size_t read_subbuf;
328
329         if (buf->bytes_consumed + bytes_consumed > subbuf_size) {
330                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
331                 buf->bytes_consumed = 0;
332         }
333
334         buf->bytes_consumed += bytes_consumed;
335         read_subbuf = read_pos / buf->chan->subbuf_size;
336         if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) {
337                 if ((read_subbuf == buf->subbufs_produced % n_subbufs) &&
338                     (buf->offset == subbuf_size))
339                         return;
340                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
341                 buf->bytes_consumed = 0;
342         }
343 }
344
345 /**
346  *      relay_file_read_avail - boolean, are there unconsumed bytes available?
347  */
348 static int relay_file_read_avail(struct rchan_buf *buf, size_t read_pos)
349 {
350         size_t bytes_produced, bytes_consumed, write_offset;
351         size_t subbuf_size = buf->chan->subbuf_size;
352         size_t n_subbufs = buf->chan->n_subbufs;
353         size_t produced = buf->subbufs_produced % n_subbufs;
354         size_t consumed = buf->subbufs_consumed % n_subbufs;
355
356         write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
357
358         if (consumed > produced) {
359                 if ((produced > n_subbufs) &&
360                     (produced + n_subbufs - consumed <= n_subbufs))
361                         produced += n_subbufs;
362         } else if (consumed == produced) {
363                 if (buf->offset > subbuf_size) {
364                         produced += n_subbufs;
365                         if (buf->subbufs_produced == buf->subbufs_consumed)
366                                 consumed += n_subbufs;
367                 }
368         }
369
370         if (buf->offset > subbuf_size)
371                 bytes_produced = (produced - 1) * subbuf_size + write_offset;
372         else
373                 bytes_produced = produced * subbuf_size + write_offset;
374         bytes_consumed = consumed * subbuf_size + buf->bytes_consumed;
375
376         if (bytes_produced == bytes_consumed)
377                 return 0;
378
379         relay_file_read_consume(buf, read_pos, 0);
380
381         return 1;
382 }
383
384 /**
385  *      relay_file_read_subbuf_avail - return bytes available in sub-buffer
386  */
387 static size_t relay_file_read_subbuf_avail(size_t read_pos,
388                                            struct rchan_buf *buf)
389 {
390         size_t padding, avail = 0;
391         size_t read_subbuf, read_offset, write_subbuf, write_offset;
392         size_t subbuf_size = buf->chan->subbuf_size;
393
394         write_subbuf = (buf->data - buf->start) / subbuf_size;
395         write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
396         read_subbuf = read_pos / subbuf_size;
397         read_offset = read_pos % subbuf_size;
398         padding = buf->padding[read_subbuf];
399
400         if (read_subbuf == write_subbuf) {
401                 if (read_offset + padding < write_offset)
402                         avail = write_offset - (read_offset + padding);
403         } else
404                 avail = (subbuf_size - padding) - read_offset;
405
406         return avail;
407 }
408
409 /**
410  *      relay_file_read_start_pos - find the first available byte to read
411  *
412  *      If the read_pos is in the middle of padding, return the
413  *      position of the first actually available byte, otherwise
414  *      return the original value.
415  */
416 static size_t relay_file_read_start_pos(size_t read_pos,
417                                         struct rchan_buf *buf)
418 {
419         size_t read_subbuf, padding, padding_start, padding_end;
420         size_t subbuf_size = buf->chan->subbuf_size;
421         size_t n_subbufs = buf->chan->n_subbufs;
422
423         read_subbuf = read_pos / subbuf_size;
424         padding = buf->padding[read_subbuf];
425         padding_start = (read_subbuf + 1) * subbuf_size - padding;
426         padding_end = (read_subbuf + 1) * subbuf_size;
427         if (read_pos >= padding_start && read_pos < padding_end) {
428                 read_subbuf = (read_subbuf + 1) % n_subbufs;
429                 read_pos = read_subbuf * subbuf_size;
430         }
431
432         return read_pos;
433 }
434
435 /**
436  *      relay_file_read_end_pos - return the new read position
437  */
438 static size_t relay_file_read_end_pos(struct rchan_buf *buf,
439                                       size_t read_pos,
440                                       size_t count)
441 {
442         size_t read_subbuf, padding, end_pos;
443         size_t subbuf_size = buf->chan->subbuf_size;
444         size_t n_subbufs = buf->chan->n_subbufs;
445
446         read_subbuf = read_pos / subbuf_size;
447         padding = buf->padding[read_subbuf];
448         if (read_pos % subbuf_size + count + padding == subbuf_size)
449                 end_pos = (read_subbuf + 1) * subbuf_size;
450         else
451                 end_pos = read_pos + count;
452         if (end_pos >= subbuf_size * n_subbufs)
453                 end_pos = 0;
454
455         return end_pos;
456 }
457
458 /**
459  *      relay_file_read - read file op for relay files
460  *      @filp: the file
461  *      @buffer: the userspace buffer
462  *      @count: number of bytes to read
463  *      @ppos: position to read from
464  *
465  *      Reads count bytes or the number of bytes available in the
466  *      current sub-buffer being read, whichever is smaller.
467  */
468 static ssize_t relay_file_read(struct file *filp,
469                                char __user *buffer,
470                                size_t count,
471                                loff_t *ppos)
472 {
473         struct rchan_buf *buf = filp->private_data;
474         struct inode *inode = filp->f_dentry->d_inode;
475         size_t read_start, avail;
476         ssize_t ret = 0;
477         void *from;
478
479         mutex_lock(&inode->i_mutex);
480         if(!relay_file_read_avail(buf, *ppos))
481                 goto out;
482
483         read_start = relay_file_read_start_pos(*ppos, buf);
484         avail = relay_file_read_subbuf_avail(read_start, buf);
485         if (!avail)
486                 goto out;
487
488         from = buf->start + read_start;
489         ret = count = min(count, avail);
490         if (copy_to_user(buffer, from, count)) {
491                 ret = -EFAULT;
492                 goto out;
493         }
494         relay_file_read_consume(buf, read_start, count);
495         *ppos = relay_file_read_end_pos(buf, read_start, count);
496 out:
497         mutex_unlock(&inode->i_mutex);
498         return ret;
499 }
500
501 struct file_operations relay_file_operations = {
502         .open           = relay_file_open,
503         .poll           = relay_file_poll,
504         .mmap           = relay_file_mmap,
505         .read           = relay_file_read,
506         .llseek         = no_llseek,
507         .release        = relay_file_release,
508 };
509
510 static struct super_operations relayfs_ops = {
511         .statfs         = simple_statfs,
512         .drop_inode     = generic_delete_inode,
513 };
514
515 static int relayfs_fill_super(struct super_block * sb, void * data, int silent)
516 {
517         struct inode *inode;
518         struct dentry *root;
519         int mode = S_IFDIR | S_IRWXU | S_IRUGO | S_IXUGO;
520
521         sb->s_blocksize = PAGE_CACHE_SIZE;
522         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
523         sb->s_magic = RELAYFS_MAGIC;
524         sb->s_op = &relayfs_ops;
525         inode = relayfs_get_inode(sb, mode, NULL, NULL);
526
527         if (!inode)
528                 return -ENOMEM;
529
530         root = d_alloc_root(inode);
531         if (!root) {
532                 iput(inode);
533                 return -ENOMEM;
534         }
535         sb->s_root = root;
536
537         return 0;
538 }
539
540 static struct super_block * relayfs_get_sb(struct file_system_type *fs_type,
541                                            int flags, const char *dev_name,
542                                            void *data)
543 {
544         return get_sb_single(fs_type, flags, data, relayfs_fill_super);
545 }
546
547 static struct file_system_type relayfs_fs_type = {
548         .owner          = THIS_MODULE,
549         .name           = "relayfs",
550         .get_sb         = relayfs_get_sb,
551         .kill_sb        = kill_litter_super,
552 };
553
554 static int __init init_relayfs_fs(void)
555 {
556         return register_filesystem(&relayfs_fs_type);
557 }
558
559 static void __exit exit_relayfs_fs(void)
560 {
561
562
563
564
565
566         unregister_filesystem(&relayfs_fs_type);
567 }
568
569 module_init(init_relayfs_fs)
570 module_exit(exit_relayfs_fs)
571
572 EXPORT_SYMBOL_GPL(relay_file_operations);
573 EXPORT_SYMBOL_GPL(relayfs_create_dir);
574 EXPORT_SYMBOL_GPL(relayfs_remove_dir);
575 EXPORT_SYMBOL_GPL(relayfs_create_file);
576 EXPORT_SYMBOL_GPL(relayfs_remove_file);
577
578 MODULE_AUTHOR("Tom Zanussi <zanussi@us.ibm.com> and Karim Yaghmour <karim@opersys.com>");
579 MODULE_DESCRIPTION("Relay Filesystem");
580 MODULE_LICENSE("GPL");
581