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