Merge branch 'bkl/procfs' of git://git.kernel.org/pub/scm/linux/kernel/git/frederic...
[linux-2.6-block.git] / fs / logfs / inode.c
CommitLineData
5db53f3e
JE
1/*
2 * fs/logfs/inode.c - inode handling code
3 *
4 * As should be obvious for Linux kernel code, license is GPLv2
5 *
6 * Copyright (c) 2005-2008 Joern Engel <joern@logfs.org>
7 */
8#include "logfs.h"
5a0e3ad6 9#include <linux/slab.h>
5db53f3e
JE
10#include <linux/writeback.h>
11#include <linux/backing-dev.h>
12
13/*
14 * How soon to reuse old inode numbers? LogFS doesn't store deleted inodes
15 * on the medium. It therefore also lacks a method to store the previous
16 * generation number for deleted inodes. Instead a single generation number
17 * is stored which will be used for new inodes. Being just a 32bit counter,
18 * this can obvious wrap relatively quickly. So we only reuse inodes if we
19 * know that a fair number of inodes can be created before we have to increment
20 * the generation again - effectively adding some bits to the counter.
21 * But being too aggressive here means we keep a very large and very sparse
22 * inode file, wasting space on indirect blocks.
23 * So what is a good value? Beats me. 64k seems moderately bad on both
24 * fronts, so let's use that for now...
25 *
26 * NFS sucks, as everyone already knows.
27 */
28#define INOS_PER_WRAP (0x10000)
29
30/*
31 * Logfs' requirement to read inodes for garbage collection makes life a bit
32 * harder. GC may have to read inodes that are in I_FREEING state, when they
33 * are being written out - and waiting for GC to make progress, naturally.
34 *
35 * So we cannot just call iget() or some variant of it, but first have to check
36 * wether the inode in question might be in I_FREEING state. Therefore we
37 * maintain our own per-sb list of "almost deleted" inodes and check against
38 * that list first. Normally this should be at most 1-2 entries long.
39 *
40 * Also, inodes have logfs-specific reference counting on top of what the vfs
41 * does. When .destroy_inode is called, normally the reference count will drop
42 * to zero and the inode gets deleted. But if GC accessed the inode, its
43 * refcount will remain nonzero and final deletion will have to wait.
44 *
45 * As a result we have two sets of functions to get/put inodes:
46 * logfs_safe_iget/logfs_safe_iput - safe to call from GC context
47 * logfs_iget/iput - normal version
48 */
49static struct kmem_cache *logfs_inode_cache;
50
51static DEFINE_SPINLOCK(logfs_inode_lock);
52
53static void logfs_inode_setops(struct inode *inode)
54{
55 switch (inode->i_mode & S_IFMT) {
56 case S_IFDIR:
57 inode->i_op = &logfs_dir_iops;
58 inode->i_fop = &logfs_dir_fops;
59 inode->i_mapping->a_ops = &logfs_reg_aops;
60 break;
61 case S_IFREG:
62 inode->i_op = &logfs_reg_iops;
63 inode->i_fop = &logfs_reg_fops;
64 inode->i_mapping->a_ops = &logfs_reg_aops;
65 break;
66 case S_IFLNK:
67 inode->i_op = &logfs_symlink_iops;
68 inode->i_mapping->a_ops = &logfs_reg_aops;
69 break;
70 case S_IFSOCK: /* fall through */
71 case S_IFBLK: /* fall through */
72 case S_IFCHR: /* fall through */
73 case S_IFIFO:
74 init_special_inode(inode, inode->i_mode, inode->i_rdev);
75 break;
76 default:
77 BUG();
78 }
79}
80
81static struct inode *__logfs_iget(struct super_block *sb, ino_t ino)
82{
83 struct inode *inode = iget_locked(sb, ino);
84 int err;
85
86 if (!inode)
87 return ERR_PTR(-ENOMEM);
88 if (!(inode->i_state & I_NEW))
89 return inode;
90
91 err = logfs_read_inode(inode);
92 if (err || inode->i_nlink == 0) {
93 /* inode->i_nlink == 0 can be true when called from
94 * block validator */
95 /* set i_nlink to 0 to prevent caching */
96 inode->i_nlink = 0;
97 logfs_inode(inode)->li_flags |= LOGFS_IF_ZOMBIE;
98 iget_failed(inode);
99 if (!err)
100 err = -ENOENT;
101 return ERR_PTR(err);
102 }
103
104 logfs_inode_setops(inode);
105 unlock_new_inode(inode);
106 return inode;
107}
108
109struct inode *logfs_iget(struct super_block *sb, ino_t ino)
110{
111 BUG_ON(ino == LOGFS_INO_MASTER);
112 BUG_ON(ino == LOGFS_INO_SEGFILE);
113 return __logfs_iget(sb, ino);
114}
115
116/*
117 * is_cached is set to 1 if we hand out a cached inode, 0 otherwise.
118 * this allows logfs_iput to do the right thing later
119 */
120struct inode *logfs_safe_iget(struct super_block *sb, ino_t ino, int *is_cached)
121{
122 struct logfs_super *super = logfs_super(sb);
123 struct logfs_inode *li;
124
125 if (ino == LOGFS_INO_MASTER)
126 return super->s_master_inode;
127 if (ino == LOGFS_INO_SEGFILE)
128 return super->s_segfile_inode;
129
130 spin_lock(&logfs_inode_lock);
131 list_for_each_entry(li, &super->s_freeing_list, li_freeing_list)
132 if (li->vfs_inode.i_ino == ino) {
133 li->li_refcount++;
134 spin_unlock(&logfs_inode_lock);
135 *is_cached = 1;
136 return &li->vfs_inode;
137 }
138 spin_unlock(&logfs_inode_lock);
139
140 *is_cached = 0;
141 return __logfs_iget(sb, ino);
142}
143
144static void __logfs_destroy_inode(struct inode *inode)
145{
146 struct logfs_inode *li = logfs_inode(inode);
147
148 BUG_ON(li->li_block);
149 list_del(&li->li_freeing_list);
150 kmem_cache_free(logfs_inode_cache, li);
151}
152
153static void logfs_destroy_inode(struct inode *inode)
154{
155 struct logfs_inode *li = logfs_inode(inode);
156
157 BUG_ON(list_empty(&li->li_freeing_list));
158 spin_lock(&logfs_inode_lock);
159 li->li_refcount--;
160 if (li->li_refcount == 0)
161 __logfs_destroy_inode(inode);
162 spin_unlock(&logfs_inode_lock);
163}
164
165void logfs_safe_iput(struct inode *inode, int is_cached)
166{
167 if (inode->i_ino == LOGFS_INO_MASTER)
168 return;
169 if (inode->i_ino == LOGFS_INO_SEGFILE)
170 return;
171
172 if (is_cached) {
173 logfs_destroy_inode(inode);
174 return;
175 }
176
177 iput(inode);
178}
179
180static void logfs_init_inode(struct super_block *sb, struct inode *inode)
181{
182 struct logfs_inode *li = logfs_inode(inode);
183 int i;
184
185 li->li_flags = 0;
186 li->li_height = 0;
187 li->li_used_bytes = 0;
188 li->li_block = NULL;
189 inode->i_uid = 0;
190 inode->i_gid = 0;
191 inode->i_size = 0;
192 inode->i_blocks = 0;
193 inode->i_ctime = CURRENT_TIME;
194 inode->i_mtime = CURRENT_TIME;
195 inode->i_nlink = 1;
24797535 196 li->li_refcount = 1;
5db53f3e
JE
197 INIT_LIST_HEAD(&li->li_freeing_list);
198
199 for (i = 0; i < LOGFS_EMBEDDED_FIELDS; i++)
200 li->li_data[i] = 0;
201
202 return;
203}
204
205static struct inode *logfs_alloc_inode(struct super_block *sb)
206{
207 struct logfs_inode *li;
208
209 li = kmem_cache_alloc(logfs_inode_cache, GFP_NOFS);
210 if (!li)
211 return NULL;
212 logfs_init_inode(sb, &li->vfs_inode);
213 return &li->vfs_inode;
214}
215
216/*
217 * In logfs inodes are written to an inode file. The inode file, like any
218 * other file, is managed with a inode. The inode file's inode, aka master
219 * inode, requires special handling in several respects. First, it cannot be
220 * written to the inode file, so it is stored in the journal instead.
221 *
222 * Secondly, this inode cannot be written back and destroyed before all other
223 * inodes have been written. The ordering is important. Linux' VFS is happily
224 * unaware of the ordering constraint and would ordinarily destroy the master
225 * inode at umount time while other inodes are still in use and dirty. Not
226 * good.
227 *
228 * So logfs makes sure the master inode is not written until all other inodes
229 * have been destroyed. Sadly, this method has another side-effect. The VFS
230 * will notice one remaining inode and print a frightening warning message.
231 * Worse, it is impossible to judge whether such a warning was caused by the
232 * master inode or any other inodes have leaked as well.
233 *
234 * Our attempt of solving this is with logfs_new_meta_inode() below. Its
235 * purpose is to create a new inode that will not trigger the warning if such
236 * an inode is still in use. An ugly hack, no doubt. Suggections for
237 * improvement are welcome.
238 */
239struct inode *logfs_new_meta_inode(struct super_block *sb, u64 ino)
240{
241 struct inode *inode;
242
243 inode = logfs_alloc_inode(sb);
244 if (!inode)
245 return ERR_PTR(-ENOMEM);
246
247 inode->i_mode = S_IFREG;
248 inode->i_ino = ino;
249 inode->i_sb = sb;
250
251 /* This is a blatant copy of alloc_inode code. We'd need alloc_inode
252 * to be nonstatic, alas. */
253 {
254 struct address_space * const mapping = &inode->i_data;
255
256 mapping->a_ops = &logfs_reg_aops;
257 mapping->host = inode;
258 mapping->flags = 0;
259 mapping_set_gfp_mask(mapping, GFP_NOFS);
260 mapping->assoc_mapping = NULL;
261 mapping->backing_dev_info = &default_backing_dev_info;
262 inode->i_mapping = mapping;
263 inode->i_nlink = 1;
264 }
265
266 return inode;
267}
268
269struct inode *logfs_read_meta_inode(struct super_block *sb, u64 ino)
270{
271 struct inode *inode;
272 int err;
273
274 inode = logfs_new_meta_inode(sb, ino);
275 if (IS_ERR(inode))
276 return inode;
277
278 err = logfs_read_inode(inode);
279 if (err) {
280 destroy_meta_inode(inode);
281 return ERR_PTR(err);
282 }
283 logfs_inode_setops(inode);
284 return inode;
285}
286
66b89159 287static int logfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5db53f3e
JE
288{
289 int ret;
290 long flags = WF_LOCK;
291
292 /* Can only happen if creat() failed. Safe to skip. */
293 if (logfs_inode(inode)->li_flags & LOGFS_IF_STILLBORN)
294 return 0;
295
296 ret = __logfs_write_inode(inode, flags);
297 LOGFS_BUG_ON(ret, inode->i_sb);
298 return ret;
299}
300
301void destroy_meta_inode(struct inode *inode)
302{
303 if (inode) {
304 if (inode->i_data.nrpages)
305 truncate_inode_pages(&inode->i_data, 0);
306 logfs_clear_inode(inode);
307 kmem_cache_free(logfs_inode_cache, logfs_inode(inode));
308 }
309}
310
311/* called with inode_lock held */
312static void logfs_drop_inode(struct inode *inode)
313{
314 struct logfs_super *super = logfs_super(inode->i_sb);
315 struct logfs_inode *li = logfs_inode(inode);
316
317 spin_lock(&logfs_inode_lock);
318 list_move(&li->li_freeing_list, &super->s_freeing_list);
319 spin_unlock(&logfs_inode_lock);
320 generic_drop_inode(inode);
321}
322
323static void logfs_set_ino_generation(struct super_block *sb,
324 struct inode *inode)
325{
326 struct logfs_super *super = logfs_super(sb);
327 u64 ino;
328
329 mutex_lock(&super->s_journal_mutex);
ccc0197b 330 ino = logfs_seek_hole(super->s_master_inode, super->s_last_ino + 1);
5db53f3e
JE
331 super->s_last_ino = ino;
332 super->s_inos_till_wrap--;
333 if (super->s_inos_till_wrap < 0) {
334 super->s_last_ino = LOGFS_RESERVED_INOS;
335 super->s_generation++;
336 super->s_inos_till_wrap = INOS_PER_WRAP;
337 }
338 inode->i_ino = ino;
339 inode->i_generation = super->s_generation;
340 mutex_unlock(&super->s_journal_mutex);
341}
342
343struct inode *logfs_new_inode(struct inode *dir, int mode)
344{
345 struct super_block *sb = dir->i_sb;
346 struct inode *inode;
347
348 inode = new_inode(sb);
349 if (!inode)
350 return ERR_PTR(-ENOMEM);
351
352 logfs_init_inode(sb, inode);
353
354 /* inherit parent flags */
355 logfs_inode(inode)->li_flags |=
356 logfs_inode(dir)->li_flags & LOGFS_FL_INHERITED;
357
358 inode->i_mode = mode;
359 logfs_set_ino_generation(sb, inode);
360
361 inode->i_uid = current_fsuid();
362 inode->i_gid = current_fsgid();
363 if (dir->i_mode & S_ISGID) {
364 inode->i_gid = dir->i_gid;
365 if (S_ISDIR(mode))
366 inode->i_mode |= S_ISGID;
367 }
368
369 logfs_inode_setops(inode);
370 insert_inode_hash(inode);
371
372 return inode;
373}
374
375static void logfs_init_once(void *_li)
376{
377 struct logfs_inode *li = _li;
378 int i;
379
380 li->li_flags = 0;
381 li->li_used_bytes = 0;
382 li->li_refcount = 1;
383 for (i = 0; i < LOGFS_EMBEDDED_FIELDS; i++)
384 li->li_data[i] = 0;
385 inode_init_once(&li->vfs_inode);
386}
387
388static int logfs_sync_fs(struct super_block *sb, int wait)
389{
c0c79c31 390 logfs_write_anchor(sb);
5db53f3e
JE
391 return 0;
392}
393
394const struct super_operations logfs_super_operations = {
395 .alloc_inode = logfs_alloc_inode,
396 .clear_inode = logfs_clear_inode,
397 .delete_inode = logfs_delete_inode,
398 .destroy_inode = logfs_destroy_inode,
399 .drop_inode = logfs_drop_inode,
400 .write_inode = logfs_write_inode,
401 .statfs = logfs_statfs,
402 .sync_fs = logfs_sync_fs,
403};
404
405int logfs_init_inode_cache(void)
406{
407 logfs_inode_cache = kmem_cache_create("logfs_inode_cache",
408 sizeof(struct logfs_inode), 0, SLAB_RECLAIM_ACCOUNT,
409 logfs_init_once);
410 if (!logfs_inode_cache)
411 return -ENOMEM;
412 return 0;
413}
414
415void logfs_destroy_inode_cache(void)
416{
417 kmem_cache_destroy(logfs_inode_cache);
418}