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" | |
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 | |
48fc7f7e | 36 | * whether the inode in question might be in I_FREEING state. Therefore we |
5db53f3e JE |
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 | */ | |
49 | static struct kmem_cache *logfs_inode_cache; | |
50 | ||
51 | static DEFINE_SPINLOCK(logfs_inode_lock); | |
52 | ||
53 | static 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: | |
fb417f13 | 67 | inode->i_op = &page_symlink_inode_operations; |
21fc61c7 | 68 | inode_nohighmem(inode); |
5db53f3e JE |
69 | inode->i_mapping->a_ops = &logfs_reg_aops; |
70 | break; | |
71 | case S_IFSOCK: /* fall through */ | |
72 | case S_IFBLK: /* fall through */ | |
73 | case S_IFCHR: /* fall through */ | |
74 | case S_IFIFO: | |
75 | init_special_inode(inode, inode->i_mode, inode->i_rdev); | |
76 | break; | |
77 | default: | |
78 | BUG(); | |
79 | } | |
80 | } | |
81 | ||
82 | static struct inode *__logfs_iget(struct super_block *sb, ino_t ino) | |
83 | { | |
84 | struct inode *inode = iget_locked(sb, ino); | |
85 | int err; | |
86 | ||
87 | if (!inode) | |
88 | return ERR_PTR(-ENOMEM); | |
89 | if (!(inode->i_state & I_NEW)) | |
90 | return inode; | |
91 | ||
92 | err = logfs_read_inode(inode); | |
93 | if (err || inode->i_nlink == 0) { | |
94 | /* inode->i_nlink == 0 can be true when called from | |
95 | * block validator */ | |
96 | /* set i_nlink to 0 to prevent caching */ | |
6d6b77f1 | 97 | clear_nlink(inode); |
5db53f3e JE |
98 | logfs_inode(inode)->li_flags |= LOGFS_IF_ZOMBIE; |
99 | iget_failed(inode); | |
100 | if (!err) | |
101 | err = -ENOENT; | |
102 | return ERR_PTR(err); | |
103 | } | |
104 | ||
105 | logfs_inode_setops(inode); | |
106 | unlock_new_inode(inode); | |
107 | return inode; | |
108 | } | |
109 | ||
110 | struct inode *logfs_iget(struct super_block *sb, ino_t ino) | |
111 | { | |
112 | BUG_ON(ino == LOGFS_INO_MASTER); | |
113 | BUG_ON(ino == LOGFS_INO_SEGFILE); | |
114 | return __logfs_iget(sb, ino); | |
115 | } | |
116 | ||
117 | /* | |
118 | * is_cached is set to 1 if we hand out a cached inode, 0 otherwise. | |
119 | * this allows logfs_iput to do the right thing later | |
120 | */ | |
121 | struct inode *logfs_safe_iget(struct super_block *sb, ino_t ino, int *is_cached) | |
122 | { | |
123 | struct logfs_super *super = logfs_super(sb); | |
124 | struct logfs_inode *li; | |
125 | ||
126 | if (ino == LOGFS_INO_MASTER) | |
127 | return super->s_master_inode; | |
128 | if (ino == LOGFS_INO_SEGFILE) | |
129 | return super->s_segfile_inode; | |
130 | ||
131 | spin_lock(&logfs_inode_lock); | |
132 | list_for_each_entry(li, &super->s_freeing_list, li_freeing_list) | |
133 | if (li->vfs_inode.i_ino == ino) { | |
134 | li->li_refcount++; | |
135 | spin_unlock(&logfs_inode_lock); | |
136 | *is_cached = 1; | |
137 | return &li->vfs_inode; | |
138 | } | |
139 | spin_unlock(&logfs_inode_lock); | |
140 | ||
141 | *is_cached = 0; | |
142 | return __logfs_iget(sb, ino); | |
143 | } | |
144 | ||
fa0d7e3d NP |
145 | static void logfs_i_callback(struct rcu_head *head) |
146 | { | |
147 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
148 | kmem_cache_free(logfs_inode_cache, logfs_inode(inode)); |
149 | } | |
150 | ||
5db53f3e JE |
151 | static void __logfs_destroy_inode(struct inode *inode) |
152 | { | |
153 | struct logfs_inode *li = logfs_inode(inode); | |
154 | ||
155 | BUG_ON(li->li_block); | |
156 | list_del(&li->li_freeing_list); | |
fa0d7e3d | 157 | call_rcu(&inode->i_rcu, logfs_i_callback); |
5db53f3e JE |
158 | } |
159 | ||
d2dcd908 PJ |
160 | static void __logfs_destroy_meta_inode(struct inode *inode) |
161 | { | |
162 | struct logfs_inode *li = logfs_inode(inode); | |
163 | BUG_ON(li->li_block); | |
164 | call_rcu(&inode->i_rcu, logfs_i_callback); | |
165 | } | |
166 | ||
5db53f3e JE |
167 | static void logfs_destroy_inode(struct inode *inode) |
168 | { | |
169 | struct logfs_inode *li = logfs_inode(inode); | |
170 | ||
d2dcd908 PJ |
171 | if (inode->i_ino < LOGFS_RESERVED_INOS) { |
172 | /* | |
173 | * The reserved inodes are never destroyed unless we are in | |
174 | * unmont path. | |
175 | */ | |
176 | __logfs_destroy_meta_inode(inode); | |
177 | return; | |
178 | } | |
179 | ||
5db53f3e JE |
180 | BUG_ON(list_empty(&li->li_freeing_list)); |
181 | spin_lock(&logfs_inode_lock); | |
182 | li->li_refcount--; | |
183 | if (li->li_refcount == 0) | |
184 | __logfs_destroy_inode(inode); | |
185 | spin_unlock(&logfs_inode_lock); | |
186 | } | |
187 | ||
188 | void logfs_safe_iput(struct inode *inode, int is_cached) | |
189 | { | |
190 | if (inode->i_ino == LOGFS_INO_MASTER) | |
191 | return; | |
192 | if (inode->i_ino == LOGFS_INO_SEGFILE) | |
193 | return; | |
194 | ||
195 | if (is_cached) { | |
196 | logfs_destroy_inode(inode); | |
197 | return; | |
198 | } | |
199 | ||
200 | iput(inode); | |
201 | } | |
202 | ||
203 | static void logfs_init_inode(struct super_block *sb, struct inode *inode) | |
204 | { | |
205 | struct logfs_inode *li = logfs_inode(inode); | |
206 | int i; | |
207 | ||
208 | li->li_flags = 0; | |
209 | li->li_height = 0; | |
210 | li->li_used_bytes = 0; | |
211 | li->li_block = NULL; | |
1a0a994e EB |
212 | i_uid_write(inode, 0); |
213 | i_gid_write(inode, 0); | |
5db53f3e JE |
214 | inode->i_size = 0; |
215 | inode->i_blocks = 0; | |
216 | inode->i_ctime = CURRENT_TIME; | |
217 | inode->i_mtime = CURRENT_TIME; | |
24797535 | 218 | li->li_refcount = 1; |
5db53f3e JE |
219 | INIT_LIST_HEAD(&li->li_freeing_list); |
220 | ||
221 | for (i = 0; i < LOGFS_EMBEDDED_FIELDS; i++) | |
222 | li->li_data[i] = 0; | |
223 | ||
224 | return; | |
225 | } | |
226 | ||
227 | static struct inode *logfs_alloc_inode(struct super_block *sb) | |
228 | { | |
229 | struct logfs_inode *li; | |
230 | ||
231 | li = kmem_cache_alloc(logfs_inode_cache, GFP_NOFS); | |
232 | if (!li) | |
233 | return NULL; | |
234 | logfs_init_inode(sb, &li->vfs_inode); | |
235 | return &li->vfs_inode; | |
236 | } | |
237 | ||
238 | /* | |
239 | * In logfs inodes are written to an inode file. The inode file, like any | |
240 | * other file, is managed with a inode. The inode file's inode, aka master | |
241 | * inode, requires special handling in several respects. First, it cannot be | |
242 | * written to the inode file, so it is stored in the journal instead. | |
243 | * | |
244 | * Secondly, this inode cannot be written back and destroyed before all other | |
245 | * inodes have been written. The ordering is important. Linux' VFS is happily | |
246 | * unaware of the ordering constraint and would ordinarily destroy the master | |
247 | * inode at umount time while other inodes are still in use and dirty. Not | |
248 | * good. | |
249 | * | |
250 | * So logfs makes sure the master inode is not written until all other inodes | |
251 | * have been destroyed. Sadly, this method has another side-effect. The VFS | |
252 | * will notice one remaining inode and print a frightening warning message. | |
253 | * Worse, it is impossible to judge whether such a warning was caused by the | |
254 | * master inode or any other inodes have leaked as well. | |
255 | * | |
256 | * Our attempt of solving this is with logfs_new_meta_inode() below. Its | |
257 | * purpose is to create a new inode that will not trigger the warning if such | |
258 | * an inode is still in use. An ugly hack, no doubt. Suggections for | |
259 | * improvement are welcome. | |
8e22c1a4 AV |
260 | * |
261 | * AV: that's what ->put_super() is for... | |
5db53f3e JE |
262 | */ |
263 | struct inode *logfs_new_meta_inode(struct super_block *sb, u64 ino) | |
264 | { | |
265 | struct inode *inode; | |
266 | ||
8e22c1a4 | 267 | inode = new_inode(sb); |
5db53f3e JE |
268 | if (!inode) |
269 | return ERR_PTR(-ENOMEM); | |
270 | ||
271 | inode->i_mode = S_IFREG; | |
272 | inode->i_ino = ino; | |
8e22c1a4 AV |
273 | inode->i_data.a_ops = &logfs_reg_aops; |
274 | mapping_set_gfp_mask(&inode->i_data, GFP_NOFS); | |
5db53f3e JE |
275 | |
276 | return inode; | |
277 | } | |
278 | ||
279 | struct inode *logfs_read_meta_inode(struct super_block *sb, u64 ino) | |
280 | { | |
281 | struct inode *inode; | |
282 | int err; | |
283 | ||
284 | inode = logfs_new_meta_inode(sb, ino); | |
285 | if (IS_ERR(inode)) | |
286 | return inode; | |
287 | ||
288 | err = logfs_read_inode(inode); | |
289 | if (err) { | |
8e22c1a4 | 290 | iput(inode); |
5db53f3e JE |
291 | return ERR_PTR(err); |
292 | } | |
293 | logfs_inode_setops(inode); | |
294 | return inode; | |
295 | } | |
296 | ||
66b89159 | 297 | static int logfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
5db53f3e JE |
298 | { |
299 | int ret; | |
300 | long flags = WF_LOCK; | |
301 | ||
302 | /* Can only happen if creat() failed. Safe to skip. */ | |
303 | if (logfs_inode(inode)->li_flags & LOGFS_IF_STILLBORN) | |
304 | return 0; | |
305 | ||
0bd90387 | 306 | ret = __logfs_write_inode(inode, NULL, flags); |
5db53f3e JE |
307 | LOGFS_BUG_ON(ret, inode->i_sb); |
308 | return ret; | |
309 | } | |
310 | ||
f283c86a | 311 | /* called with inode->i_lock held */ |
45321ac5 | 312 | static int logfs_drop_inode(struct inode *inode) |
5db53f3e JE |
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); | |
45321ac5 | 320 | return generic_drop_inode(inode); |
5db53f3e JE |
321 | } |
322 | ||
323 | static 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 | ||
88176446 | 343 | struct inode *logfs_new_inode(struct inode *dir, umode_t mode) |
5db53f3e JE |
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 | ||
ab9a79b9 | 361 | inode_init_owner(inode, dir, mode); |
5db53f3e JE |
362 | logfs_inode_setops(inode); |
363 | insert_inode_hash(inode); | |
364 | ||
365 | return inode; | |
366 | } | |
367 | ||
368 | static void logfs_init_once(void *_li) | |
369 | { | |
370 | struct logfs_inode *li = _li; | |
371 | int i; | |
372 | ||
373 | li->li_flags = 0; | |
374 | li->li_used_bytes = 0; | |
375 | li->li_refcount = 1; | |
376 | for (i = 0; i < LOGFS_EMBEDDED_FIELDS; i++) | |
377 | li->li_data[i] = 0; | |
378 | inode_init_once(&li->vfs_inode); | |
379 | } | |
380 | ||
381 | static int logfs_sync_fs(struct super_block *sb, int wait) | |
382 | { | |
13ced29c | 383 | logfs_get_wblocks(sb, NULL, WF_LOCK); |
c0c79c31 | 384 | logfs_write_anchor(sb); |
13ced29c | 385 | logfs_put_wblocks(sb, NULL, WF_LOCK); |
5db53f3e JE |
386 | return 0; |
387 | } | |
388 | ||
8e22c1a4 AV |
389 | static void logfs_put_super(struct super_block *sb) |
390 | { | |
391 | struct logfs_super *super = logfs_super(sb); | |
392 | /* kill the meta-inodes */ | |
8e22c1a4 | 393 | iput(super->s_segfile_inode); |
41b93bc1 | 394 | iput(super->s_master_inode); |
8e22c1a4 AV |
395 | iput(super->s_mapping_inode); |
396 | } | |
397 | ||
5db53f3e JE |
398 | const struct super_operations logfs_super_operations = { |
399 | .alloc_inode = logfs_alloc_inode, | |
5db53f3e | 400 | .destroy_inode = logfs_destroy_inode, |
7da08fd1 | 401 | .evict_inode = logfs_evict_inode, |
5db53f3e | 402 | .drop_inode = logfs_drop_inode, |
8e22c1a4 | 403 | .put_super = logfs_put_super, |
5db53f3e JE |
404 | .write_inode = logfs_write_inode, |
405 | .statfs = logfs_statfs, | |
406 | .sync_fs = logfs_sync_fs, | |
407 | }; | |
408 | ||
409 | int logfs_init_inode_cache(void) | |
410 | { | |
411 | logfs_inode_cache = kmem_cache_create("logfs_inode_cache", | |
5d097056 VD |
412 | sizeof(struct logfs_inode), 0, |
413 | SLAB_RECLAIM_ACCOUNT|SLAB_ACCOUNT, | |
5db53f3e JE |
414 | logfs_init_once); |
415 | if (!logfs_inode_cache) | |
416 | return -ENOMEM; | |
417 | return 0; | |
418 | } | |
419 | ||
420 | void logfs_destroy_inode_cache(void) | |
421 | { | |
8c0a8537 KS |
422 | /* |
423 | * Make sure all delayed rcu free inodes are flushed before we | |
424 | * destroy cache. | |
425 | */ | |
426 | rcu_barrier(); | |
5db53f3e JE |
427 | kmem_cache_destroy(logfs_inode_cache); |
428 | } |