diff options
Diffstat (limited to 'fs/jffs2/gc.c')
-rw-r--r-- | fs/jffs2/gc.c | 1246 |
1 files changed, 1246 insertions, 0 deletions
diff --git a/fs/jffs2/gc.c b/fs/jffs2/gc.c new file mode 100644 index 000000000000..87ec74ff5930 --- /dev/null +++ b/fs/jffs2/gc.c | |||
@@ -0,0 +1,1246 @@ | |||
1 | /* | ||
2 | * JFFS2 -- Journalling Flash File System, Version 2. | ||
3 | * | ||
4 | * Copyright (C) 2001-2003 Red Hat, Inc. | ||
5 | * | ||
6 | * Created by David Woodhouse <dwmw2@infradead.org> | ||
7 | * | ||
8 | * For licensing information, see the file 'LICENCE' in this directory. | ||
9 | * | ||
10 | * $Id: gc.c,v 1.144 2004/12/21 11:18:50 dwmw2 Exp $ | ||
11 | * | ||
12 | */ | ||
13 | |||
14 | #include <linux/kernel.h> | ||
15 | #include <linux/mtd/mtd.h> | ||
16 | #include <linux/slab.h> | ||
17 | #include <linux/pagemap.h> | ||
18 | #include <linux/crc32.h> | ||
19 | #include <linux/compiler.h> | ||
20 | #include <linux/stat.h> | ||
21 | #include "nodelist.h" | ||
22 | #include "compr.h" | ||
23 | |||
24 | static int jffs2_garbage_collect_pristine(struct jffs2_sb_info *c, | ||
25 | struct jffs2_inode_cache *ic, | ||
26 | struct jffs2_raw_node_ref *raw); | ||
27 | static int jffs2_garbage_collect_metadata(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, | ||
28 | struct jffs2_inode_info *f, struct jffs2_full_dnode *fd); | ||
29 | static int jffs2_garbage_collect_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, | ||
30 | struct jffs2_inode_info *f, struct jffs2_full_dirent *fd); | ||
31 | static int jffs2_garbage_collect_deletion_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, | ||
32 | struct jffs2_inode_info *f, struct jffs2_full_dirent *fd); | ||
33 | static int jffs2_garbage_collect_hole(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, | ||
34 | struct jffs2_inode_info *f, struct jffs2_full_dnode *fn, | ||
35 | uint32_t start, uint32_t end); | ||
36 | static int jffs2_garbage_collect_dnode(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, | ||
37 | struct jffs2_inode_info *f, struct jffs2_full_dnode *fn, | ||
38 | uint32_t start, uint32_t end); | ||
39 | static int jffs2_garbage_collect_live(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, | ||
40 | struct jffs2_raw_node_ref *raw, struct jffs2_inode_info *f); | ||
41 | |||
42 | /* Called with erase_completion_lock held */ | ||
43 | static struct jffs2_eraseblock *jffs2_find_gc_block(struct jffs2_sb_info *c) | ||
44 | { | ||
45 | struct jffs2_eraseblock *ret; | ||
46 | struct list_head *nextlist = NULL; | ||
47 | int n = jiffies % 128; | ||
48 | |||
49 | /* Pick an eraseblock to garbage collect next. This is where we'll | ||
50 | put the clever wear-levelling algorithms. Eventually. */ | ||
51 | /* We possibly want to favour the dirtier blocks more when the | ||
52 | number of free blocks is low. */ | ||
53 | if (!list_empty(&c->bad_used_list) && c->nr_free_blocks > c->resv_blocks_gcbad) { | ||
54 | D1(printk(KERN_DEBUG "Picking block from bad_used_list to GC next\n")); | ||
55 | nextlist = &c->bad_used_list; | ||
56 | } else if (n < 50 && !list_empty(&c->erasable_list)) { | ||
57 | /* Note that most of them will have gone directly to be erased. | ||
58 | So don't favour the erasable_list _too_ much. */ | ||
59 | D1(printk(KERN_DEBUG "Picking block from erasable_list to GC next\n")); | ||
60 | nextlist = &c->erasable_list; | ||
61 | } else if (n < 110 && !list_empty(&c->very_dirty_list)) { | ||
62 | /* Most of the time, pick one off the very_dirty list */ | ||
63 | D1(printk(KERN_DEBUG "Picking block from very_dirty_list to GC next\n")); | ||
64 | nextlist = &c->very_dirty_list; | ||
65 | } else if (n < 126 && !list_empty(&c->dirty_list)) { | ||
66 | D1(printk(KERN_DEBUG "Picking block from dirty_list to GC next\n")); | ||
67 | nextlist = &c->dirty_list; | ||
68 | } else if (!list_empty(&c->clean_list)) { | ||
69 | D1(printk(KERN_DEBUG "Picking block from clean_list to GC next\n")); | ||
70 | nextlist = &c->clean_list; | ||
71 | } else if (!list_empty(&c->dirty_list)) { | ||
72 | D1(printk(KERN_DEBUG "Picking block from dirty_list to GC next (clean_list was empty)\n")); | ||
73 | |||
74 | nextlist = &c->dirty_list; | ||
75 | } else if (!list_empty(&c->very_dirty_list)) { | ||
76 | D1(printk(KERN_DEBUG "Picking block from very_dirty_list to GC next (clean_list and dirty_list were empty)\n")); | ||
77 | nextlist = &c->very_dirty_list; | ||
78 | } else if (!list_empty(&c->erasable_list)) { | ||
79 | D1(printk(KERN_DEBUG "Picking block from erasable_list to GC next (clean_list and {very_,}dirty_list were empty)\n")); | ||
80 | |||
81 | nextlist = &c->erasable_list; | ||
82 | } else { | ||
83 | /* Eep. All were empty */ | ||
84 | D1(printk(KERN_NOTICE "jffs2: No clean, dirty _or_ erasable blocks to GC from! Where are they all?\n")); | ||
85 | return NULL; | ||
86 | } | ||
87 | |||
88 | ret = list_entry(nextlist->next, struct jffs2_eraseblock, list); | ||
89 | list_del(&ret->list); | ||
90 | c->gcblock = ret; | ||
91 | ret->gc_node = ret->first_node; | ||
92 | if (!ret->gc_node) { | ||
93 | printk(KERN_WARNING "Eep. ret->gc_node for block at 0x%08x is NULL\n", ret->offset); | ||
94 | BUG(); | ||
95 | } | ||
96 | |||
97 | /* Have we accidentally picked a clean block with wasted space ? */ | ||
98 | if (ret->wasted_size) { | ||
99 | D1(printk(KERN_DEBUG "Converting wasted_size %08x to dirty_size\n", ret->wasted_size)); | ||
100 | ret->dirty_size += ret->wasted_size; | ||
101 | c->wasted_size -= ret->wasted_size; | ||
102 | c->dirty_size += ret->wasted_size; | ||
103 | ret->wasted_size = 0; | ||
104 | } | ||
105 | |||
106 | D2(jffs2_dump_block_lists(c)); | ||
107 | return ret; | ||
108 | } | ||
109 | |||
110 | /* jffs2_garbage_collect_pass | ||
111 | * Make a single attempt to progress GC. Move one node, and possibly | ||
112 | * start erasing one eraseblock. | ||
113 | */ | ||
114 | int jffs2_garbage_collect_pass(struct jffs2_sb_info *c) | ||
115 | { | ||
116 | struct jffs2_inode_info *f; | ||
117 | struct jffs2_inode_cache *ic; | ||
118 | struct jffs2_eraseblock *jeb; | ||
119 | struct jffs2_raw_node_ref *raw; | ||
120 | int ret = 0, inum, nlink; | ||
121 | |||
122 | if (down_interruptible(&c->alloc_sem)) | ||
123 | return -EINTR; | ||
124 | |||
125 | for (;;) { | ||
126 | spin_lock(&c->erase_completion_lock); | ||
127 | if (!c->unchecked_size) | ||
128 | break; | ||
129 | |||
130 | /* We can't start doing GC yet. We haven't finished checking | ||
131 | the node CRCs etc. Do it now. */ | ||
132 | |||
133 | /* checked_ino is protected by the alloc_sem */ | ||
134 | if (c->checked_ino > c->highest_ino) { | ||
135 | printk(KERN_CRIT "Checked all inodes but still 0x%x bytes of unchecked space?\n", | ||
136 | c->unchecked_size); | ||
137 | D2(jffs2_dump_block_lists(c)); | ||
138 | spin_unlock(&c->erase_completion_lock); | ||
139 | BUG(); | ||
140 | } | ||
141 | |||
142 | spin_unlock(&c->erase_completion_lock); | ||
143 | |||
144 | spin_lock(&c->inocache_lock); | ||
145 | |||
146 | ic = jffs2_get_ino_cache(c, c->checked_ino++); | ||
147 | |||
148 | if (!ic) { | ||
149 | spin_unlock(&c->inocache_lock); | ||
150 | continue; | ||
151 | } | ||
152 | |||
153 | if (!ic->nlink) { | ||
154 | D1(printk(KERN_DEBUG "Skipping check of ino #%d with nlink zero\n", | ||
155 | ic->ino)); | ||
156 | spin_unlock(&c->inocache_lock); | ||
157 | continue; | ||
158 | } | ||
159 | switch(ic->state) { | ||
160 | case INO_STATE_CHECKEDABSENT: | ||
161 | case INO_STATE_PRESENT: | ||
162 | D1(printk(KERN_DEBUG "Skipping ino #%u already checked\n", ic->ino)); | ||
163 | spin_unlock(&c->inocache_lock); | ||
164 | continue; | ||
165 | |||
166 | case INO_STATE_GC: | ||
167 | case INO_STATE_CHECKING: | ||
168 | printk(KERN_WARNING "Inode #%u is in state %d during CRC check phase!\n", ic->ino, ic->state); | ||
169 | spin_unlock(&c->inocache_lock); | ||
170 | BUG(); | ||
171 | |||
172 | case INO_STATE_READING: | ||
173 | /* We need to wait for it to finish, lest we move on | ||
174 | and trigger the BUG() above while we haven't yet | ||
175 | finished checking all its nodes */ | ||
176 | D1(printk(KERN_DEBUG "Waiting for ino #%u to finish reading\n", ic->ino)); | ||
177 | up(&c->alloc_sem); | ||
178 | sleep_on_spinunlock(&c->inocache_wq, &c->inocache_lock); | ||
179 | return 0; | ||
180 | |||
181 | default: | ||
182 | BUG(); | ||
183 | |||
184 | case INO_STATE_UNCHECKED: | ||
185 | ; | ||
186 | } | ||
187 | ic->state = INO_STATE_CHECKING; | ||
188 | spin_unlock(&c->inocache_lock); | ||
189 | |||
190 | D1(printk(KERN_DEBUG "jffs2_garbage_collect_pass() triggering inode scan of ino#%u\n", ic->ino)); | ||
191 | |||
192 | ret = jffs2_do_crccheck_inode(c, ic); | ||
193 | if (ret) | ||
194 | printk(KERN_WARNING "Returned error for crccheck of ino #%u. Expect badness...\n", ic->ino); | ||
195 | |||
196 | jffs2_set_inocache_state(c, ic, INO_STATE_CHECKEDABSENT); | ||
197 | up(&c->alloc_sem); | ||
198 | return ret; | ||
199 | } | ||
200 | |||
201 | /* First, work out which block we're garbage-collecting */ | ||
202 | jeb = c->gcblock; | ||
203 | |||
204 | if (!jeb) | ||
205 | jeb = jffs2_find_gc_block(c); | ||
206 | |||
207 | if (!jeb) { | ||
208 | D1 (printk(KERN_NOTICE "jffs2: Couldn't find erase block to garbage collect!\n")); | ||
209 | spin_unlock(&c->erase_completion_lock); | ||
210 | up(&c->alloc_sem); | ||
211 | return -EIO; | ||
212 | } | ||
213 | |||
214 | D1(printk(KERN_DEBUG "GC from block %08x, used_size %08x, dirty_size %08x, free_size %08x\n", jeb->offset, jeb->used_size, jeb->dirty_size, jeb->free_size)); | ||
215 | D1(if (c->nextblock) | ||
216 | printk(KERN_DEBUG "Nextblock at %08x, used_size %08x, dirty_size %08x, wasted_size %08x, free_size %08x\n", c->nextblock->offset, c->nextblock->used_size, c->nextblock->dirty_size, c->nextblock->wasted_size, c->nextblock->free_size)); | ||
217 | |||
218 | if (!jeb->used_size) { | ||
219 | up(&c->alloc_sem); | ||
220 | goto eraseit; | ||
221 | } | ||
222 | |||
223 | raw = jeb->gc_node; | ||
224 | |||
225 | while(ref_obsolete(raw)) { | ||
226 | D1(printk(KERN_DEBUG "Node at 0x%08x is obsolete... skipping\n", ref_offset(raw))); | ||
227 | raw = raw->next_phys; | ||
228 | if (unlikely(!raw)) { | ||
229 | printk(KERN_WARNING "eep. End of raw list while still supposedly nodes to GC\n"); | ||
230 | printk(KERN_WARNING "erase block at 0x%08x. free_size 0x%08x, dirty_size 0x%08x, used_size 0x%08x\n", | ||
231 | jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size); | ||
232 | jeb->gc_node = raw; | ||
233 | spin_unlock(&c->erase_completion_lock); | ||
234 | up(&c->alloc_sem); | ||
235 | BUG(); | ||
236 | } | ||
237 | } | ||
238 | jeb->gc_node = raw; | ||
239 | |||
240 | D1(printk(KERN_DEBUG "Going to garbage collect node at 0x%08x\n", ref_offset(raw))); | ||
241 | |||
242 | if (!raw->next_in_ino) { | ||
243 | /* Inode-less node. Clean marker, snapshot or something like that */ | ||
244 | /* FIXME: If it's something that needs to be copied, including something | ||
245 | we don't grok that has JFFS2_NODETYPE_RWCOMPAT_COPY, we should do so */ | ||
246 | spin_unlock(&c->erase_completion_lock); | ||
247 | jffs2_mark_node_obsolete(c, raw); | ||
248 | up(&c->alloc_sem); | ||
249 | goto eraseit_lock; | ||
250 | } | ||
251 | |||
252 | ic = jffs2_raw_ref_to_ic(raw); | ||
253 | |||
254 | /* We need to hold the inocache. Either the erase_completion_lock or | ||
255 | the inocache_lock are sufficient; we trade down since the inocache_lock | ||
256 | causes less contention. */ | ||
257 | spin_lock(&c->inocache_lock); | ||
258 | |||
259 | spin_unlock(&c->erase_completion_lock); | ||
260 | |||
261 | D1(printk(KERN_DEBUG "jffs2_garbage_collect_pass collecting from block @0x%08x. Node @0x%08x(%d), ino #%u\n", jeb->offset, ref_offset(raw), ref_flags(raw), ic->ino)); | ||
262 | |||
263 | /* Three possibilities: | ||
264 | 1. Inode is already in-core. We must iget it and do proper | ||
265 | updating to its fragtree, etc. | ||
266 | 2. Inode is not in-core, node is REF_PRISTINE. We lock the | ||
267 | inocache to prevent a read_inode(), copy the node intact. | ||
268 | 3. Inode is not in-core, node is not pristine. We must iget() | ||
269 | and take the slow path. | ||
270 | */ | ||
271 | |||
272 | switch(ic->state) { | ||
273 | case INO_STATE_CHECKEDABSENT: | ||
274 | /* It's been checked, but it's not currently in-core. | ||
275 | We can just copy any pristine nodes, but have | ||
276 | to prevent anyone else from doing read_inode() while | ||
277 | we're at it, so we set the state accordingly */ | ||
278 | if (ref_flags(raw) == REF_PRISTINE) | ||
279 | ic->state = INO_STATE_GC; | ||
280 | else { | ||
281 | D1(printk(KERN_DEBUG "Ino #%u is absent but node not REF_PRISTINE. Reading.\n", | ||
282 | ic->ino)); | ||
283 | } | ||
284 | break; | ||
285 | |||
286 | case INO_STATE_PRESENT: | ||
287 | /* It's in-core. GC must iget() it. */ | ||
288 | break; | ||
289 | |||
290 | case INO_STATE_UNCHECKED: | ||
291 | case INO_STATE_CHECKING: | ||
292 | case INO_STATE_GC: | ||
293 | /* Should never happen. We should have finished checking | ||
294 | by the time we actually start doing any GC, and since | ||
295 | we're holding the alloc_sem, no other garbage collection | ||
296 | can happen. | ||
297 | */ | ||
298 | printk(KERN_CRIT "Inode #%u already in state %d in jffs2_garbage_collect_pass()!\n", | ||
299 | ic->ino, ic->state); | ||
300 | up(&c->alloc_sem); | ||
301 | spin_unlock(&c->inocache_lock); | ||
302 | BUG(); | ||
303 | |||
304 | case INO_STATE_READING: | ||
305 | /* Someone's currently trying to read it. We must wait for | ||
306 | them to finish and then go through the full iget() route | ||
307 | to do the GC. However, sometimes read_inode() needs to get | ||
308 | the alloc_sem() (for marking nodes invalid) so we must | ||
309 | drop the alloc_sem before sleeping. */ | ||
310 | |||
311 | up(&c->alloc_sem); | ||
312 | D1(printk(KERN_DEBUG "jffs2_garbage_collect_pass() waiting for ino #%u in state %d\n", | ||
313 | ic->ino, ic->state)); | ||
314 | sleep_on_spinunlock(&c->inocache_wq, &c->inocache_lock); | ||
315 | /* And because we dropped the alloc_sem we must start again from the | ||
316 | beginning. Ponder chance of livelock here -- we're returning success | ||
317 | without actually making any progress. | ||
318 | |||
319 | Q: What are the chances that the inode is back in INO_STATE_READING | ||
320 | again by the time we next enter this function? And that this happens | ||
321 | enough times to cause a real delay? | ||
322 | |||
323 | A: Small enough that I don't care :) | ||
324 | */ | ||
325 | return 0; | ||
326 | } | ||
327 | |||
328 | /* OK. Now if the inode is in state INO_STATE_GC, we are going to copy the | ||
329 | node intact, and we don't have to muck about with the fragtree etc. | ||
330 | because we know it's not in-core. If it _was_ in-core, we go through | ||
331 | all the iget() crap anyway */ | ||
332 | |||
333 | if (ic->state == INO_STATE_GC) { | ||
334 | spin_unlock(&c->inocache_lock); | ||
335 | |||
336 | ret = jffs2_garbage_collect_pristine(c, ic, raw); | ||
337 | |||
338 | spin_lock(&c->inocache_lock); | ||
339 | ic->state = INO_STATE_CHECKEDABSENT; | ||
340 | wake_up(&c->inocache_wq); | ||
341 | |||
342 | if (ret != -EBADFD) { | ||
343 | spin_unlock(&c->inocache_lock); | ||
344 | goto release_sem; | ||
345 | } | ||
346 | |||
347 | /* Fall through if it wanted us to, with inocache_lock held */ | ||
348 | } | ||
349 | |||
350 | /* Prevent the fairly unlikely race where the gcblock is | ||
351 | entirely obsoleted by the final close of a file which had | ||
352 | the only valid nodes in the block, followed by erasure, | ||
353 | followed by freeing of the ic because the erased block(s) | ||
354 | held _all_ the nodes of that inode.... never been seen but | ||
355 | it's vaguely possible. */ | ||
356 | |||
357 | inum = ic->ino; | ||
358 | nlink = ic->nlink; | ||
359 | spin_unlock(&c->inocache_lock); | ||
360 | |||
361 | f = jffs2_gc_fetch_inode(c, inum, nlink); | ||
362 | if (IS_ERR(f)) { | ||
363 | ret = PTR_ERR(f); | ||
364 | goto release_sem; | ||
365 | } | ||
366 | if (!f) { | ||
367 | ret = 0; | ||
368 | goto release_sem; | ||
369 | } | ||
370 | |||
371 | ret = jffs2_garbage_collect_live(c, jeb, raw, f); | ||
372 | |||
373 | jffs2_gc_release_inode(c, f); | ||
374 | |||
375 | release_sem: | ||
376 | up(&c->alloc_sem); | ||
377 | |||
378 | eraseit_lock: | ||
379 | /* If we've finished this block, start it erasing */ | ||
380 | spin_lock(&c->erase_completion_lock); | ||
381 | |||
382 | eraseit: | ||
383 | if (c->gcblock && !c->gcblock->used_size) { | ||
384 | D1(printk(KERN_DEBUG "Block at 0x%08x completely obsoleted by GC. Moving to erase_pending_list\n", c->gcblock->offset)); | ||
385 | /* We're GC'ing an empty block? */ | ||
386 | list_add_tail(&c->gcblock->list, &c->erase_pending_list); | ||
387 | c->gcblock = NULL; | ||
388 | c->nr_erasing_blocks++; | ||
389 | jffs2_erase_pending_trigger(c); | ||
390 | } | ||
391 | spin_unlock(&c->erase_completion_lock); | ||
392 | |||
393 | return ret; | ||
394 | } | ||
395 | |||
396 | static int jffs2_garbage_collect_live(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, | ||
397 | struct jffs2_raw_node_ref *raw, struct jffs2_inode_info *f) | ||
398 | { | ||
399 | struct jffs2_node_frag *frag; | ||
400 | struct jffs2_full_dnode *fn = NULL; | ||
401 | struct jffs2_full_dirent *fd; | ||
402 | uint32_t start = 0, end = 0, nrfrags = 0; | ||
403 | int ret = 0; | ||
404 | |||
405 | down(&f->sem); | ||
406 | |||
407 | /* Now we have the lock for this inode. Check that it's still the one at the head | ||
408 | of the list. */ | ||
409 | |||
410 | spin_lock(&c->erase_completion_lock); | ||
411 | |||
412 | if (c->gcblock != jeb) { | ||
413 | spin_unlock(&c->erase_completion_lock); | ||
414 | D1(printk(KERN_DEBUG "GC block is no longer gcblock. Restart\n")); | ||
415 | goto upnout; | ||
416 | } | ||
417 | if (ref_obsolete(raw)) { | ||
418 | spin_unlock(&c->erase_completion_lock); | ||
419 | D1(printk(KERN_DEBUG "node to be GC'd was obsoleted in the meantime.\n")); | ||
420 | /* They'll call again */ | ||
421 | goto upnout; | ||
422 | } | ||
423 | spin_unlock(&c->erase_completion_lock); | ||
424 | |||
425 | /* OK. Looks safe. And nobody can get us now because we have the semaphore. Move the block */ | ||
426 | if (f->metadata && f->metadata->raw == raw) { | ||
427 | fn = f->metadata; | ||
428 | ret = jffs2_garbage_collect_metadata(c, jeb, f, fn); | ||
429 | goto upnout; | ||
430 | } | ||
431 | |||
432 | /* FIXME. Read node and do lookup? */ | ||
433 | for (frag = frag_first(&f->fragtree); frag; frag = frag_next(frag)) { | ||
434 | if (frag->node && frag->node->raw == raw) { | ||
435 | fn = frag->node; | ||
436 | end = frag->ofs + frag->size; | ||
437 | if (!nrfrags++) | ||
438 | start = frag->ofs; | ||
439 | if (nrfrags == frag->node->frags) | ||
440 | break; /* We've found them all */ | ||
441 | } | ||
442 | } | ||
443 | if (fn) { | ||
444 | if (ref_flags(raw) == REF_PRISTINE) { | ||
445 | ret = jffs2_garbage_collect_pristine(c, f->inocache, raw); | ||
446 | if (!ret) { | ||
447 | /* Urgh. Return it sensibly. */ | ||
448 | frag->node->raw = f->inocache->nodes; | ||
449 | } | ||
450 | if (ret != -EBADFD) | ||
451 | goto upnout; | ||
452 | } | ||
453 | /* We found a datanode. Do the GC */ | ||
454 | if((start >> PAGE_CACHE_SHIFT) < ((end-1) >> PAGE_CACHE_SHIFT)) { | ||
455 | /* It crosses a page boundary. Therefore, it must be a hole. */ | ||
456 | ret = jffs2_garbage_collect_hole(c, jeb, f, fn, start, end); | ||
457 | } else { | ||
458 | /* It could still be a hole. But we GC the page this way anyway */ | ||
459 | ret = jffs2_garbage_collect_dnode(c, jeb, f, fn, start, end); | ||
460 | } | ||
461 | goto upnout; | ||
462 | } | ||
463 | |||
464 | /* Wasn't a dnode. Try dirent */ | ||
465 | for (fd = f->dents; fd; fd=fd->next) { | ||
466 | if (fd->raw == raw) | ||
467 | break; | ||
468 | } | ||
469 | |||
470 | if (fd && fd->ino) { | ||
471 | ret = jffs2_garbage_collect_dirent(c, jeb, f, fd); | ||
472 | } else if (fd) { | ||
473 | ret = jffs2_garbage_collect_deletion_dirent(c, jeb, f, fd); | ||
474 | } else { | ||
475 | printk(KERN_WARNING "Raw node at 0x%08x wasn't in node lists for ino #%u\n", | ||
476 | ref_offset(raw), f->inocache->ino); | ||
477 | if (ref_obsolete(raw)) { | ||
478 | printk(KERN_WARNING "But it's obsolete so we don't mind too much\n"); | ||
479 | } else { | ||
480 | ret = -EIO; | ||
481 | } | ||
482 | } | ||
483 | upnout: | ||
484 | up(&f->sem); | ||
485 | |||
486 | return ret; | ||
487 | } | ||
488 | |||
489 | static int jffs2_garbage_collect_pristine(struct jffs2_sb_info *c, | ||
490 | struct jffs2_inode_cache *ic, | ||
491 | struct jffs2_raw_node_ref *raw) | ||
492 | { | ||
493 | union jffs2_node_union *node; | ||
494 | struct jffs2_raw_node_ref *nraw; | ||
495 | size_t retlen; | ||
496 | int ret; | ||
497 | uint32_t phys_ofs, alloclen; | ||
498 | uint32_t crc, rawlen; | ||
499 | int retried = 0; | ||
500 | |||
501 | D1(printk(KERN_DEBUG "Going to GC REF_PRISTINE node at 0x%08x\n", ref_offset(raw))); | ||
502 | |||
503 | rawlen = ref_totlen(c, c->gcblock, raw); | ||
504 | |||
505 | /* Ask for a small amount of space (or the totlen if smaller) because we | ||
506 | don't want to force wastage of the end of a block if splitting would | ||
507 | work. */ | ||
508 | ret = jffs2_reserve_space_gc(c, min_t(uint32_t, sizeof(struct jffs2_raw_inode) + JFFS2_MIN_DATA_LEN, | ||
509 | rawlen), &phys_ofs, &alloclen); | ||
510 | if (ret) | ||
511 | return ret; | ||
512 | |||
513 | if (alloclen < rawlen) { | ||
514 | /* Doesn't fit untouched. We'll go the old route and split it */ | ||
515 | return -EBADFD; | ||
516 | } | ||
517 | |||
518 | node = kmalloc(rawlen, GFP_KERNEL); | ||
519 | if (!node) | ||
520 | return -ENOMEM; | ||
521 | |||
522 | ret = jffs2_flash_read(c, ref_offset(raw), rawlen, &retlen, (char *)node); | ||
523 | if (!ret && retlen != rawlen) | ||
524 | ret = -EIO; | ||
525 | if (ret) | ||
526 | goto out_node; | ||
527 | |||
528 | crc = crc32(0, node, sizeof(struct jffs2_unknown_node)-4); | ||
529 | if (je32_to_cpu(node->u.hdr_crc) != crc) { | ||
530 | printk(KERN_WARNING "Header CRC failed on REF_PRISTINE node at 0x%08x: Read 0x%08x, calculated 0x%08x\n", | ||
531 | ref_offset(raw), je32_to_cpu(node->u.hdr_crc), crc); | ||
532 | goto bail; | ||
533 | } | ||
534 | |||
535 | switch(je16_to_cpu(node->u.nodetype)) { | ||
536 | case JFFS2_NODETYPE_INODE: | ||
537 | crc = crc32(0, node, sizeof(node->i)-8); | ||
538 | if (je32_to_cpu(node->i.node_crc) != crc) { | ||
539 | printk(KERN_WARNING "Node CRC failed on REF_PRISTINE data node at 0x%08x: Read 0x%08x, calculated 0x%08x\n", | ||
540 | ref_offset(raw), je32_to_cpu(node->i.node_crc), crc); | ||
541 | goto bail; | ||
542 | } | ||
543 | |||
544 | if (je32_to_cpu(node->i.dsize)) { | ||
545 | crc = crc32(0, node->i.data, je32_to_cpu(node->i.csize)); | ||
546 | if (je32_to_cpu(node->i.data_crc) != crc) { | ||
547 | printk(KERN_WARNING "Data CRC failed on REF_PRISTINE data node at 0x%08x: Read 0x%08x, calculated 0x%08x\n", | ||
548 | ref_offset(raw), je32_to_cpu(node->i.data_crc), crc); | ||
549 | goto bail; | ||
550 | } | ||
551 | } | ||
552 | break; | ||
553 | |||
554 | case JFFS2_NODETYPE_DIRENT: | ||
555 | crc = crc32(0, node, sizeof(node->d)-8); | ||
556 | if (je32_to_cpu(node->d.node_crc) != crc) { | ||
557 | printk(KERN_WARNING "Node CRC failed on REF_PRISTINE dirent node at 0x%08x: Read 0x%08x, calculated 0x%08x\n", | ||
558 | ref_offset(raw), je32_to_cpu(node->d.node_crc), crc); | ||
559 | goto bail; | ||
560 | } | ||
561 | |||
562 | if (node->d.nsize) { | ||
563 | crc = crc32(0, node->d.name, node->d.nsize); | ||
564 | if (je32_to_cpu(node->d.name_crc) != crc) { | ||
565 | printk(KERN_WARNING "Name CRC failed on REF_PRISTINE dirent ode at 0x%08x: Read 0x%08x, calculated 0x%08x\n", | ||
566 | ref_offset(raw), je32_to_cpu(node->d.name_crc), crc); | ||
567 | goto bail; | ||
568 | } | ||
569 | } | ||
570 | break; | ||
571 | default: | ||
572 | printk(KERN_WARNING "Unknown node type for REF_PRISTINE node at 0x%08x: 0x%04x\n", | ||
573 | ref_offset(raw), je16_to_cpu(node->u.nodetype)); | ||
574 | goto bail; | ||
575 | } | ||
576 | |||
577 | nraw = jffs2_alloc_raw_node_ref(); | ||
578 | if (!nraw) { | ||
579 | ret = -ENOMEM; | ||
580 | goto out_node; | ||
581 | } | ||
582 | |||
583 | /* OK, all the CRCs are good; this node can just be copied as-is. */ | ||
584 | retry: | ||
585 | nraw->flash_offset = phys_ofs; | ||
586 | nraw->__totlen = rawlen; | ||
587 | nraw->next_phys = NULL; | ||
588 | |||
589 | ret = jffs2_flash_write(c, phys_ofs, rawlen, &retlen, (char *)node); | ||
590 | |||
591 | if (ret || (retlen != rawlen)) { | ||
592 | printk(KERN_NOTICE "Write of %d bytes at 0x%08x failed. returned %d, retlen %zd\n", | ||
593 | rawlen, phys_ofs, ret, retlen); | ||
594 | if (retlen) { | ||
595 | /* Doesn't belong to any inode */ | ||
596 | nraw->next_in_ino = NULL; | ||
597 | |||
598 | nraw->flash_offset |= REF_OBSOLETE; | ||
599 | jffs2_add_physical_node_ref(c, nraw); | ||
600 | jffs2_mark_node_obsolete(c, nraw); | ||
601 | } else { | ||
602 | printk(KERN_NOTICE "Not marking the space at 0x%08x as dirty because the flash driver returned retlen zero\n", nraw->flash_offset); | ||
603 | jffs2_free_raw_node_ref(nraw); | ||
604 | } | ||
605 | if (!retried && (nraw = jffs2_alloc_raw_node_ref())) { | ||
606 | /* Try to reallocate space and retry */ | ||
607 | uint32_t dummy; | ||
608 | struct jffs2_eraseblock *jeb = &c->blocks[phys_ofs / c->sector_size]; | ||
609 | |||
610 | retried = 1; | ||
611 | |||
612 | D1(printk(KERN_DEBUG "Retrying failed write of REF_PRISTINE node.\n")); | ||
613 | |||
614 | ACCT_SANITY_CHECK(c,jeb); | ||
615 | D1(ACCT_PARANOIA_CHECK(jeb)); | ||
616 | |||
617 | ret = jffs2_reserve_space_gc(c, rawlen, &phys_ofs, &dummy); | ||
618 | |||
619 | if (!ret) { | ||
620 | D1(printk(KERN_DEBUG "Allocated space at 0x%08x to retry failed write.\n", phys_ofs)); | ||
621 | |||
622 | ACCT_SANITY_CHECK(c,jeb); | ||
623 | D1(ACCT_PARANOIA_CHECK(jeb)); | ||
624 | |||
625 | goto retry; | ||
626 | } | ||
627 | D1(printk(KERN_DEBUG "Failed to allocate space to retry failed write: %d!\n", ret)); | ||
628 | jffs2_free_raw_node_ref(nraw); | ||
629 | } | ||
630 | |||
631 | jffs2_free_raw_node_ref(nraw); | ||
632 | if (!ret) | ||
633 | ret = -EIO; | ||
634 | goto out_node; | ||
635 | } | ||
636 | nraw->flash_offset |= REF_PRISTINE; | ||
637 | jffs2_add_physical_node_ref(c, nraw); | ||
638 | |||
639 | /* Link into per-inode list. This is safe because of the ic | ||
640 | state being INO_STATE_GC. Note that if we're doing this | ||
641 | for an inode which is in-core, the 'nraw' pointer is then | ||
642 | going to be fetched from ic->nodes by our caller. */ | ||
643 | spin_lock(&c->erase_completion_lock); | ||
644 | nraw->next_in_ino = ic->nodes; | ||
645 | ic->nodes = nraw; | ||
646 | spin_unlock(&c->erase_completion_lock); | ||
647 | |||
648 | jffs2_mark_node_obsolete(c, raw); | ||
649 | D1(printk(KERN_DEBUG "WHEEE! GC REF_PRISTINE node at 0x%08x succeeded\n", ref_offset(raw))); | ||
650 | |||
651 | out_node: | ||
652 | kfree(node); | ||
653 | return ret; | ||
654 | bail: | ||
655 | ret = -EBADFD; | ||
656 | goto out_node; | ||
657 | } | ||
658 | |||
659 | static int jffs2_garbage_collect_metadata(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, | ||
660 | struct jffs2_inode_info *f, struct jffs2_full_dnode *fn) | ||
661 | { | ||
662 | struct jffs2_full_dnode *new_fn; | ||
663 | struct jffs2_raw_inode ri; | ||
664 | jint16_t dev; | ||
665 | char *mdata = NULL, mdatalen = 0; | ||
666 | uint32_t alloclen, phys_ofs; | ||
667 | int ret; | ||
668 | |||
669 | if (S_ISBLK(JFFS2_F_I_MODE(f)) || | ||
670 | S_ISCHR(JFFS2_F_I_MODE(f)) ) { | ||
671 | /* For these, we don't actually need to read the old node */ | ||
672 | /* FIXME: for minor or major > 255. */ | ||
673 | dev = cpu_to_je16(((JFFS2_F_I_RDEV_MAJ(f) << 8) | | ||
674 | JFFS2_F_I_RDEV_MIN(f))); | ||
675 | mdata = (char *)&dev; | ||
676 | mdatalen = sizeof(dev); | ||
677 | D1(printk(KERN_DEBUG "jffs2_garbage_collect_metadata(): Writing %d bytes of kdev_t\n", mdatalen)); | ||
678 | } else if (S_ISLNK(JFFS2_F_I_MODE(f))) { | ||
679 | mdatalen = fn->size; | ||
680 | mdata = kmalloc(fn->size, GFP_KERNEL); | ||
681 | if (!mdata) { | ||
682 | printk(KERN_WARNING "kmalloc of mdata failed in jffs2_garbage_collect_metadata()\n"); | ||
683 | return -ENOMEM; | ||
684 | } | ||
685 | ret = jffs2_read_dnode(c, f, fn, mdata, 0, mdatalen); | ||
686 | if (ret) { | ||
687 | printk(KERN_WARNING "read of old metadata failed in jffs2_garbage_collect_metadata(): %d\n", ret); | ||
688 | kfree(mdata); | ||
689 | return ret; | ||
690 | } | ||
691 | D1(printk(KERN_DEBUG "jffs2_garbage_collect_metadata(): Writing %d bites of symlink target\n", mdatalen)); | ||
692 | |||
693 | } | ||
694 | |||
695 | ret = jffs2_reserve_space_gc(c, sizeof(ri) + mdatalen, &phys_ofs, &alloclen); | ||
696 | if (ret) { | ||
697 | printk(KERN_WARNING "jffs2_reserve_space_gc of %zd bytes for garbage_collect_metadata failed: %d\n", | ||
698 | sizeof(ri)+ mdatalen, ret); | ||
699 | goto out; | ||
700 | } | ||
701 | |||
702 | memset(&ri, 0, sizeof(ri)); | ||
703 | ri.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK); | ||
704 | ri.nodetype = cpu_to_je16(JFFS2_NODETYPE_INODE); | ||
705 | ri.totlen = cpu_to_je32(sizeof(ri) + mdatalen); | ||
706 | ri.hdr_crc = cpu_to_je32(crc32(0, &ri, sizeof(struct jffs2_unknown_node)-4)); | ||
707 | |||
708 | ri.ino = cpu_to_je32(f->inocache->ino); | ||
709 | ri.version = cpu_to_je32(++f->highest_version); | ||
710 | ri.mode = cpu_to_jemode(JFFS2_F_I_MODE(f)); | ||
711 | ri.uid = cpu_to_je16(JFFS2_F_I_UID(f)); | ||
712 | ri.gid = cpu_to_je16(JFFS2_F_I_GID(f)); | ||
713 | ri.isize = cpu_to_je32(JFFS2_F_I_SIZE(f)); | ||
714 | ri.atime = cpu_to_je32(JFFS2_F_I_ATIME(f)); | ||
715 | ri.ctime = cpu_to_je32(JFFS2_F_I_CTIME(f)); | ||
716 | ri.mtime = cpu_to_je32(JFFS2_F_I_MTIME(f)); | ||
717 | ri.offset = cpu_to_je32(0); | ||
718 | ri.csize = cpu_to_je32(mdatalen); | ||
719 | ri.dsize = cpu_to_je32(mdatalen); | ||
720 | ri.compr = JFFS2_COMPR_NONE; | ||
721 | ri.node_crc = cpu_to_je32(crc32(0, &ri, sizeof(ri)-8)); | ||
722 | ri.data_crc = cpu_to_je32(crc32(0, mdata, mdatalen)); | ||
723 | |||
724 | new_fn = jffs2_write_dnode(c, f, &ri, mdata, mdatalen, phys_ofs, ALLOC_GC); | ||
725 | |||
726 | if (IS_ERR(new_fn)) { | ||
727 | printk(KERN_WARNING "Error writing new dnode: %ld\n", PTR_ERR(new_fn)); | ||
728 | ret = PTR_ERR(new_fn); | ||
729 | goto out; | ||
730 | } | ||
731 | jffs2_mark_node_obsolete(c, fn->raw); | ||
732 | jffs2_free_full_dnode(fn); | ||
733 | f->metadata = new_fn; | ||
734 | out: | ||
735 | if (S_ISLNK(JFFS2_F_I_MODE(f))) | ||
736 | kfree(mdata); | ||
737 | return ret; | ||
738 | } | ||
739 | |||
740 | static int jffs2_garbage_collect_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, | ||
741 | struct jffs2_inode_info *f, struct jffs2_full_dirent *fd) | ||
742 | { | ||
743 | struct jffs2_full_dirent *new_fd; | ||
744 | struct jffs2_raw_dirent rd; | ||
745 | uint32_t alloclen, phys_ofs; | ||
746 | int ret; | ||
747 | |||
748 | rd.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK); | ||
749 | rd.nodetype = cpu_to_je16(JFFS2_NODETYPE_DIRENT); | ||
750 | rd.nsize = strlen(fd->name); | ||
751 | rd.totlen = cpu_to_je32(sizeof(rd) + rd.nsize); | ||
752 | rd.hdr_crc = cpu_to_je32(crc32(0, &rd, sizeof(struct jffs2_unknown_node)-4)); | ||
753 | |||
754 | rd.pino = cpu_to_je32(f->inocache->ino); | ||
755 | rd.version = cpu_to_je32(++f->highest_version); | ||
756 | rd.ino = cpu_to_je32(fd->ino); | ||
757 | rd.mctime = cpu_to_je32(max(JFFS2_F_I_MTIME(f), JFFS2_F_I_CTIME(f))); | ||
758 | rd.type = fd->type; | ||
759 | rd.node_crc = cpu_to_je32(crc32(0, &rd, sizeof(rd)-8)); | ||
760 | rd.name_crc = cpu_to_je32(crc32(0, fd->name, rd.nsize)); | ||
761 | |||
762 | ret = jffs2_reserve_space_gc(c, sizeof(rd)+rd.nsize, &phys_ofs, &alloclen); | ||
763 | if (ret) { | ||
764 | printk(KERN_WARNING "jffs2_reserve_space_gc of %zd bytes for garbage_collect_dirent failed: %d\n", | ||
765 | sizeof(rd)+rd.nsize, ret); | ||
766 | return ret; | ||
767 | } | ||
768 | new_fd = jffs2_write_dirent(c, f, &rd, fd->name, rd.nsize, phys_ofs, ALLOC_GC); | ||
769 | |||
770 | if (IS_ERR(new_fd)) { | ||
771 | printk(KERN_WARNING "jffs2_write_dirent in garbage_collect_dirent failed: %ld\n", PTR_ERR(new_fd)); | ||
772 | return PTR_ERR(new_fd); | ||
773 | } | ||
774 | jffs2_add_fd_to_list(c, new_fd, &f->dents); | ||
775 | return 0; | ||
776 | } | ||
777 | |||
778 | static int jffs2_garbage_collect_deletion_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, | ||
779 | struct jffs2_inode_info *f, struct jffs2_full_dirent *fd) | ||
780 | { | ||
781 | struct jffs2_full_dirent **fdp = &f->dents; | ||
782 | int found = 0; | ||
783 | |||
784 | /* On a medium where we can't actually mark nodes obsolete | ||
785 | pernamently, such as NAND flash, we need to work out | ||
786 | whether this deletion dirent is still needed to actively | ||
787 | delete a 'real' dirent with the same name that's still | ||
788 | somewhere else on the flash. */ | ||
789 | if (!jffs2_can_mark_obsolete(c)) { | ||
790 | struct jffs2_raw_dirent *rd; | ||
791 | struct jffs2_raw_node_ref *raw; | ||
792 | int ret; | ||
793 | size_t retlen; | ||
794 | int name_len = strlen(fd->name); | ||
795 | uint32_t name_crc = crc32(0, fd->name, name_len); | ||
796 | uint32_t rawlen = ref_totlen(c, jeb, fd->raw); | ||
797 | |||
798 | rd = kmalloc(rawlen, GFP_KERNEL); | ||
799 | if (!rd) | ||
800 | return -ENOMEM; | ||
801 | |||
802 | /* Prevent the erase code from nicking the obsolete node refs while | ||
803 | we're looking at them. I really don't like this extra lock but | ||
804 | can't see any alternative. Suggestions on a postcard to... */ | ||
805 | down(&c->erase_free_sem); | ||
806 | |||
807 | for (raw = f->inocache->nodes; raw != (void *)f->inocache; raw = raw->next_in_ino) { | ||
808 | |||
809 | /* We only care about obsolete ones */ | ||
810 | if (!(ref_obsolete(raw))) | ||
811 | continue; | ||
812 | |||
813 | /* Any dirent with the same name is going to have the same length... */ | ||
814 | if (ref_totlen(c, NULL, raw) != rawlen) | ||
815 | continue; | ||
816 | |||
817 | /* Doesn't matter if there's one in the same erase block. We're going to | ||
818 | delete it too at the same time. */ | ||
819 | if ((raw->flash_offset & ~(c->sector_size-1)) == | ||
820 | (fd->raw->flash_offset & ~(c->sector_size-1))) | ||
821 | continue; | ||
822 | |||
823 | D1(printk(KERN_DEBUG "Check potential deletion dirent at %08x\n", ref_offset(raw))); | ||
824 | |||
825 | /* This is an obsolete node belonging to the same directory, and it's of the right | ||
826 | length. We need to take a closer look...*/ | ||
827 | ret = jffs2_flash_read(c, ref_offset(raw), rawlen, &retlen, (char *)rd); | ||
828 | if (ret) { | ||
829 | printk(KERN_WARNING "jffs2_g_c_deletion_dirent(): Read error (%d) reading obsolete node at %08x\n", ret, ref_offset(raw)); | ||
830 | /* If we can't read it, we don't need to continue to obsolete it. Continue */ | ||
831 | continue; | ||
832 | } | ||
833 | if (retlen != rawlen) { | ||
834 | printk(KERN_WARNING "jffs2_g_c_deletion_dirent(): Short read (%zd not %u) reading header from obsolete node at %08x\n", | ||
835 | retlen, rawlen, ref_offset(raw)); | ||
836 | continue; | ||
837 | } | ||
838 | |||
839 | if (je16_to_cpu(rd->nodetype) != JFFS2_NODETYPE_DIRENT) | ||
840 | continue; | ||
841 | |||
842 | /* If the name CRC doesn't match, skip */ | ||
843 | if (je32_to_cpu(rd->name_crc) != name_crc) | ||
844 | continue; | ||
845 | |||
846 | /* If the name length doesn't match, or it's another deletion dirent, skip */ | ||
847 | if (rd->nsize != name_len || !je32_to_cpu(rd->ino)) | ||
848 | continue; | ||
849 | |||
850 | /* OK, check the actual name now */ | ||
851 | if (memcmp(rd->name, fd->name, name_len)) | ||
852 | continue; | ||
853 | |||
854 | /* OK. The name really does match. There really is still an older node on | ||
855 | the flash which our deletion dirent obsoletes. So we have to write out | ||
856 | a new deletion dirent to replace it */ | ||
857 | up(&c->erase_free_sem); | ||
858 | |||
859 | D1(printk(KERN_DEBUG "Deletion dirent at %08x still obsoletes real dirent \"%s\" at %08x for ino #%u\n", | ||
860 | ref_offset(fd->raw), fd->name, ref_offset(raw), je32_to_cpu(rd->ino))); | ||
861 | kfree(rd); | ||
862 | |||
863 | return jffs2_garbage_collect_dirent(c, jeb, f, fd); | ||
864 | } | ||
865 | |||
866 | up(&c->erase_free_sem); | ||
867 | kfree(rd); | ||
868 | } | ||
869 | |||
870 | /* No need for it any more. Just mark it obsolete and remove it from the list */ | ||
871 | while (*fdp) { | ||
872 | if ((*fdp) == fd) { | ||
873 | found = 1; | ||
874 | *fdp = fd->next; | ||
875 | break; | ||
876 | } | ||
877 | fdp = &(*fdp)->next; | ||
878 | } | ||
879 | if (!found) { | ||
880 | printk(KERN_WARNING "Deletion dirent \"%s\" not found in list for ino #%u\n", fd->name, f->inocache->ino); | ||
881 | } | ||
882 | jffs2_mark_node_obsolete(c, fd->raw); | ||
883 | jffs2_free_full_dirent(fd); | ||
884 | return 0; | ||
885 | } | ||
886 | |||
887 | static int jffs2_garbage_collect_hole(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, | ||
888 | struct jffs2_inode_info *f, struct jffs2_full_dnode *fn, | ||
889 | uint32_t start, uint32_t end) | ||
890 | { | ||
891 | struct jffs2_raw_inode ri; | ||
892 | struct jffs2_node_frag *frag; | ||
893 | struct jffs2_full_dnode *new_fn; | ||
894 | uint32_t alloclen, phys_ofs; | ||
895 | int ret; | ||
896 | |||
897 | D1(printk(KERN_DEBUG "Writing replacement hole node for ino #%u from offset 0x%x to 0x%x\n", | ||
898 | f->inocache->ino, start, end)); | ||
899 | |||
900 | memset(&ri, 0, sizeof(ri)); | ||
901 | |||
902 | if(fn->frags > 1) { | ||
903 | size_t readlen; | ||
904 | uint32_t crc; | ||
905 | /* It's partially obsoleted by a later write. So we have to | ||
906 | write it out again with the _same_ version as before */ | ||
907 | ret = jffs2_flash_read(c, ref_offset(fn->raw), sizeof(ri), &readlen, (char *)&ri); | ||
908 | if (readlen != sizeof(ri) || ret) { | ||
909 | printk(KERN_WARNING "Node read failed in jffs2_garbage_collect_hole. Ret %d, retlen %zd. Data will be lost by writing new hole node\n", ret, readlen); | ||
910 | goto fill; | ||
911 | } | ||
912 | if (je16_to_cpu(ri.nodetype) != JFFS2_NODETYPE_INODE) { | ||
913 | printk(KERN_WARNING "jffs2_garbage_collect_hole: Node at 0x%08x had node type 0x%04x instead of JFFS2_NODETYPE_INODE(0x%04x)\n", | ||
914 | ref_offset(fn->raw), | ||
915 | je16_to_cpu(ri.nodetype), JFFS2_NODETYPE_INODE); | ||
916 | return -EIO; | ||
917 | } | ||
918 | if (je32_to_cpu(ri.totlen) != sizeof(ri)) { | ||
919 | printk(KERN_WARNING "jffs2_garbage_collect_hole: Node at 0x%08x had totlen 0x%x instead of expected 0x%zx\n", | ||
920 | ref_offset(fn->raw), | ||
921 | je32_to_cpu(ri.totlen), sizeof(ri)); | ||
922 | return -EIO; | ||
923 | } | ||
924 | crc = crc32(0, &ri, sizeof(ri)-8); | ||
925 | if (crc != je32_to_cpu(ri.node_crc)) { | ||
926 | printk(KERN_WARNING "jffs2_garbage_collect_hole: Node at 0x%08x had CRC 0x%08x which doesn't match calculated CRC 0x%08x\n", | ||
927 | ref_offset(fn->raw), | ||
928 | je32_to_cpu(ri.node_crc), crc); | ||
929 | /* FIXME: We could possibly deal with this by writing new holes for each frag */ | ||
930 | printk(KERN_WARNING "Data in the range 0x%08x to 0x%08x of inode #%u will be lost\n", | ||
931 | start, end, f->inocache->ino); | ||
932 | goto fill; | ||
933 | } | ||
934 | if (ri.compr != JFFS2_COMPR_ZERO) { | ||
935 | printk(KERN_WARNING "jffs2_garbage_collect_hole: Node 0x%08x wasn't a hole node!\n", ref_offset(fn->raw)); | ||
936 | printk(KERN_WARNING "Data in the range 0x%08x to 0x%08x of inode #%u will be lost\n", | ||
937 | start, end, f->inocache->ino); | ||
938 | goto fill; | ||
939 | } | ||
940 | } else { | ||
941 | fill: | ||
942 | ri.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK); | ||
943 | ri.nodetype = cpu_to_je16(JFFS2_NODETYPE_INODE); | ||
944 | ri.totlen = cpu_to_je32(sizeof(ri)); | ||
945 | ri.hdr_crc = cpu_to_je32(crc32(0, &ri, sizeof(struct jffs2_unknown_node)-4)); | ||
946 | |||
947 | ri.ino = cpu_to_je32(f->inocache->ino); | ||
948 | ri.version = cpu_to_je32(++f->highest_version); | ||
949 | ri.offset = cpu_to_je32(start); | ||
950 | ri.dsize = cpu_to_je32(end - start); | ||
951 | ri.csize = cpu_to_je32(0); | ||
952 | ri.compr = JFFS2_COMPR_ZERO; | ||
953 | } | ||
954 | ri.mode = cpu_to_jemode(JFFS2_F_I_MODE(f)); | ||
955 | ri.uid = cpu_to_je16(JFFS2_F_I_UID(f)); | ||
956 | ri.gid = cpu_to_je16(JFFS2_F_I_GID(f)); | ||
957 | ri.isize = cpu_to_je32(JFFS2_F_I_SIZE(f)); | ||
958 | ri.atime = cpu_to_je32(JFFS2_F_I_ATIME(f)); | ||
959 | ri.ctime = cpu_to_je32(JFFS2_F_I_CTIME(f)); | ||
960 | ri.mtime = cpu_to_je32(JFFS2_F_I_MTIME(f)); | ||
961 | ri.data_crc = cpu_to_je32(0); | ||
962 | ri.node_crc = cpu_to_je32(crc32(0, &ri, sizeof(ri)-8)); | ||
963 | |||
964 | ret = jffs2_reserve_space_gc(c, sizeof(ri), &phys_ofs, &alloclen); | ||
965 | if (ret) { | ||
966 | printk(KERN_WARNING "jffs2_reserve_space_gc of %zd bytes for garbage_collect_hole failed: %d\n", | ||
967 | sizeof(ri), ret); | ||
968 | return ret; | ||
969 | } | ||
970 | new_fn = jffs2_write_dnode(c, f, &ri, NULL, 0, phys_ofs, ALLOC_GC); | ||
971 | |||
972 | if (IS_ERR(new_fn)) { | ||
973 | printk(KERN_WARNING "Error writing new hole node: %ld\n", PTR_ERR(new_fn)); | ||
974 | return PTR_ERR(new_fn); | ||
975 | } | ||
976 | if (je32_to_cpu(ri.version) == f->highest_version) { | ||
977 | jffs2_add_full_dnode_to_inode(c, f, new_fn); | ||
978 | if (f->metadata) { | ||
979 | jffs2_mark_node_obsolete(c, f->metadata->raw); | ||
980 | jffs2_free_full_dnode(f->metadata); | ||
981 | f->metadata = NULL; | ||
982 | } | ||
983 | return 0; | ||
984 | } | ||
985 | |||
986 | /* | ||
987 | * We should only get here in the case where the node we are | ||
988 | * replacing had more than one frag, so we kept the same version | ||
989 | * number as before. (Except in case of error -- see 'goto fill;' | ||
990 | * above.) | ||
991 | */ | ||
992 | D1(if(unlikely(fn->frags <= 1)) { | ||
993 | printk(KERN_WARNING "jffs2_garbage_collect_hole: Replacing fn with %d frag(s) but new ver %d != highest_version %d of ino #%d\n", | ||
994 | fn->frags, je32_to_cpu(ri.version), f->highest_version, | ||
995 | je32_to_cpu(ri.ino)); | ||
996 | }); | ||
997 | |||
998 | /* This is a partially-overlapped hole node. Mark it REF_NORMAL not REF_PRISTINE */ | ||
999 | mark_ref_normal(new_fn->raw); | ||
1000 | |||
1001 | for (frag = jffs2_lookup_node_frag(&f->fragtree, fn->ofs); | ||
1002 | frag; frag = frag_next(frag)) { | ||
1003 | if (frag->ofs > fn->size + fn->ofs) | ||
1004 | break; | ||
1005 | if (frag->node == fn) { | ||
1006 | frag->node = new_fn; | ||
1007 | new_fn->frags++; | ||
1008 | fn->frags--; | ||
1009 | } | ||
1010 | } | ||
1011 | if (fn->frags) { | ||
1012 | printk(KERN_WARNING "jffs2_garbage_collect_hole: Old node still has frags!\n"); | ||
1013 | BUG(); | ||
1014 | } | ||
1015 | if (!new_fn->frags) { | ||
1016 | printk(KERN_WARNING "jffs2_garbage_collect_hole: New node has no frags!\n"); | ||
1017 | BUG(); | ||
1018 | } | ||
1019 | |||
1020 | jffs2_mark_node_obsolete(c, fn->raw); | ||
1021 | jffs2_free_full_dnode(fn); | ||
1022 | |||
1023 | return 0; | ||
1024 | } | ||
1025 | |||
1026 | static int jffs2_garbage_collect_dnode(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, | ||
1027 | struct jffs2_inode_info *f, struct jffs2_full_dnode *fn, | ||
1028 | uint32_t start, uint32_t end) | ||
1029 | { | ||
1030 | struct jffs2_full_dnode *new_fn; | ||
1031 | struct jffs2_raw_inode ri; | ||
1032 | uint32_t alloclen, phys_ofs, offset, orig_end, orig_start; | ||
1033 | int ret = 0; | ||
1034 | unsigned char *comprbuf = NULL, *writebuf; | ||
1035 | unsigned long pg; | ||
1036 | unsigned char *pg_ptr; | ||
1037 | |||
1038 | memset(&ri, 0, sizeof(ri)); | ||
1039 | |||
1040 | D1(printk(KERN_DEBUG "Writing replacement dnode for ino #%u from offset 0x%x to 0x%x\n", | ||
1041 | f->inocache->ino, start, end)); | ||
1042 | |||
1043 | orig_end = end; | ||
1044 | orig_start = start; | ||
1045 | |||
1046 | if (c->nr_free_blocks + c->nr_erasing_blocks > c->resv_blocks_gcmerge) { | ||
1047 | /* Attempt to do some merging. But only expand to cover logically | ||
1048 | adjacent frags if the block containing them is already considered | ||
1049 | to be dirty. Otherwise we end up with GC just going round in | ||
1050 | circles dirtying the nodes it already wrote out, especially | ||
1051 | on NAND where we have small eraseblocks and hence a much higher | ||
1052 | chance of nodes having to be split to cross boundaries. */ | ||
1053 | |||
1054 | struct jffs2_node_frag *frag; | ||
1055 | uint32_t min, max; | ||
1056 | |||
1057 | min = start & ~(PAGE_CACHE_SIZE-1); | ||
1058 | max = min + PAGE_CACHE_SIZE; | ||
1059 | |||
1060 | frag = jffs2_lookup_node_frag(&f->fragtree, start); | ||
1061 | |||
1062 | /* BUG_ON(!frag) but that'll happen anyway... */ | ||
1063 | |||
1064 | BUG_ON(frag->ofs != start); | ||
1065 | |||
1066 | /* First grow down... */ | ||
1067 | while((frag = frag_prev(frag)) && frag->ofs >= min) { | ||
1068 | |||
1069 | /* If the previous frag doesn't even reach the beginning, there's | ||
1070 | excessive fragmentation. Just merge. */ | ||
1071 | if (frag->ofs > min) { | ||
1072 | D1(printk(KERN_DEBUG "Expanding down to cover partial frag (0x%x-0x%x)\n", | ||
1073 | frag->ofs, frag->ofs+frag->size)); | ||
1074 | start = frag->ofs; | ||
1075 | continue; | ||
1076 | } | ||
1077 | /* OK. This frag holds the first byte of the page. */ | ||
1078 | if (!frag->node || !frag->node->raw) { | ||
1079 | D1(printk(KERN_DEBUG "First frag in page is hole (0x%x-0x%x). Not expanding down.\n", | ||
1080 | frag->ofs, frag->ofs+frag->size)); | ||
1081 | break; | ||
1082 | } else { | ||
1083 | |||
1084 | /* OK, it's a frag which extends to the beginning of the page. Does it live | ||
1085 | in a block which is still considered clean? If so, don't obsolete it. | ||
1086 | If not, cover it anyway. */ | ||
1087 | |||
1088 | struct jffs2_raw_node_ref *raw = frag->node->raw; | ||
1089 | struct jffs2_eraseblock *jeb; | ||
1090 | |||
1091 | jeb = &c->blocks[raw->flash_offset / c->sector_size]; | ||
1092 | |||
1093 | if (jeb == c->gcblock) { | ||
1094 | D1(printk(KERN_DEBUG "Expanding down to cover frag (0x%x-0x%x) in gcblock at %08x\n", | ||
1095 | frag->ofs, frag->ofs+frag->size, ref_offset(raw))); | ||
1096 | start = frag->ofs; | ||
1097 | break; | ||
1098 | } | ||
1099 | if (!ISDIRTY(jeb->dirty_size + jeb->wasted_size)) { | ||
1100 | D1(printk(KERN_DEBUG "Not expanding down to cover frag (0x%x-0x%x) in clean block %08x\n", | ||
1101 | frag->ofs, frag->ofs+frag->size, jeb->offset)); | ||
1102 | break; | ||
1103 | } | ||
1104 | |||
1105 | D1(printk(KERN_DEBUG "Expanding down to cover frag (0x%x-0x%x) in dirty block %08x\n", | ||
1106 | frag->ofs, frag->ofs+frag->size, jeb->offset)); | ||
1107 | start = frag->ofs; | ||
1108 | break; | ||
1109 | } | ||
1110 | } | ||
1111 | |||
1112 | /* ... then up */ | ||
1113 | |||
1114 | /* Find last frag which is actually part of the node we're to GC. */ | ||
1115 | frag = jffs2_lookup_node_frag(&f->fragtree, end-1); | ||
1116 | |||
1117 | while((frag = frag_next(frag)) && frag->ofs+frag->size <= max) { | ||
1118 | |||
1119 | /* If the previous frag doesn't even reach the beginning, there's lots | ||
1120 | of fragmentation. Just merge. */ | ||
1121 | if (frag->ofs+frag->size < max) { | ||
1122 | D1(printk(KERN_DEBUG "Expanding up to cover partial frag (0x%x-0x%x)\n", | ||
1123 | frag->ofs, frag->ofs+frag->size)); | ||
1124 | end = frag->ofs + frag->size; | ||
1125 | continue; | ||
1126 | } | ||
1127 | |||
1128 | if (!frag->node || !frag->node->raw) { | ||
1129 | D1(printk(KERN_DEBUG "Last frag in page is hole (0x%x-0x%x). Not expanding up.\n", | ||
1130 | frag->ofs, frag->ofs+frag->size)); | ||
1131 | break; | ||
1132 | } else { | ||
1133 | |||
1134 | /* OK, it's a frag which extends to the beginning of the page. Does it live | ||
1135 | in a block which is still considered clean? If so, don't obsolete it. | ||
1136 | If not, cover it anyway. */ | ||
1137 | |||
1138 | struct jffs2_raw_node_ref *raw = frag->node->raw; | ||
1139 | struct jffs2_eraseblock *jeb; | ||
1140 | |||
1141 | jeb = &c->blocks[raw->flash_offset / c->sector_size]; | ||
1142 | |||
1143 | if (jeb == c->gcblock) { | ||
1144 | D1(printk(KERN_DEBUG "Expanding up to cover frag (0x%x-0x%x) in gcblock at %08x\n", | ||
1145 | frag->ofs, frag->ofs+frag->size, ref_offset(raw))); | ||
1146 | end = frag->ofs + frag->size; | ||
1147 | break; | ||
1148 | } | ||
1149 | if (!ISDIRTY(jeb->dirty_size + jeb->wasted_size)) { | ||
1150 | D1(printk(KERN_DEBUG "Not expanding up to cover frag (0x%x-0x%x) in clean block %08x\n", | ||
1151 | frag->ofs, frag->ofs+frag->size, jeb->offset)); | ||
1152 | break; | ||
1153 | } | ||
1154 | |||
1155 | D1(printk(KERN_DEBUG "Expanding up to cover frag (0x%x-0x%x) in dirty block %08x\n", | ||
1156 | frag->ofs, frag->ofs+frag->size, jeb->offset)); | ||
1157 | end = frag->ofs + frag->size; | ||
1158 | break; | ||
1159 | } | ||
1160 | } | ||
1161 | D1(printk(KERN_DEBUG "Expanded dnode to write from (0x%x-0x%x) to (0x%x-0x%x)\n", | ||
1162 | orig_start, orig_end, start, end)); | ||
1163 | |||
1164 | BUG_ON(end > JFFS2_F_I_SIZE(f)); | ||
1165 | BUG_ON(end < orig_end); | ||
1166 | BUG_ON(start > orig_start); | ||
1167 | } | ||
1168 | |||
1169 | /* First, use readpage() to read the appropriate page into the page cache */ | ||
1170 | /* Q: What happens if we actually try to GC the _same_ page for which commit_write() | ||
1171 | * triggered garbage collection in the first place? | ||
1172 | * A: I _think_ it's OK. read_cache_page shouldn't deadlock, we'll write out the | ||
1173 | * page OK. We'll actually write it out again in commit_write, which is a little | ||
1174 | * suboptimal, but at least we're correct. | ||
1175 | */ | ||
1176 | pg_ptr = jffs2_gc_fetch_page(c, f, start, &pg); | ||
1177 | |||
1178 | if (IS_ERR(pg_ptr)) { | ||
1179 | printk(KERN_WARNING "read_cache_page() returned error: %ld\n", PTR_ERR(pg_ptr)); | ||
1180 | return PTR_ERR(pg_ptr); | ||
1181 | } | ||
1182 | |||
1183 | offset = start; | ||
1184 | while(offset < orig_end) { | ||
1185 | uint32_t datalen; | ||
1186 | uint32_t cdatalen; | ||
1187 | uint16_t comprtype = JFFS2_COMPR_NONE; | ||
1188 | |||
1189 | ret = jffs2_reserve_space_gc(c, sizeof(ri) + JFFS2_MIN_DATA_LEN, &phys_ofs, &alloclen); | ||
1190 | |||
1191 | if (ret) { | ||
1192 | printk(KERN_WARNING "jffs2_reserve_space_gc of %zd bytes for garbage_collect_dnode failed: %d\n", | ||
1193 | sizeof(ri)+ JFFS2_MIN_DATA_LEN, ret); | ||
1194 | break; | ||
1195 | } | ||
1196 | cdatalen = min_t(uint32_t, alloclen - sizeof(ri), end - offset); | ||
1197 | datalen = end - offset; | ||
1198 | |||
1199 | writebuf = pg_ptr + (offset & (PAGE_CACHE_SIZE -1)); | ||
1200 | |||
1201 | comprtype = jffs2_compress(c, f, writebuf, &comprbuf, &datalen, &cdatalen); | ||
1202 | |||
1203 | ri.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK); | ||
1204 | ri.nodetype = cpu_to_je16(JFFS2_NODETYPE_INODE); | ||
1205 | ri.totlen = cpu_to_je32(sizeof(ri) + cdatalen); | ||
1206 | ri.hdr_crc = cpu_to_je32(crc32(0, &ri, sizeof(struct jffs2_unknown_node)-4)); | ||
1207 | |||
1208 | ri.ino = cpu_to_je32(f->inocache->ino); | ||
1209 | ri.version = cpu_to_je32(++f->highest_version); | ||
1210 | ri.mode = cpu_to_jemode(JFFS2_F_I_MODE(f)); | ||
1211 | ri.uid = cpu_to_je16(JFFS2_F_I_UID(f)); | ||
1212 | ri.gid = cpu_to_je16(JFFS2_F_I_GID(f)); | ||
1213 | ri.isize = cpu_to_je32(JFFS2_F_I_SIZE(f)); | ||
1214 | ri.atime = cpu_to_je32(JFFS2_F_I_ATIME(f)); | ||
1215 | ri.ctime = cpu_to_je32(JFFS2_F_I_CTIME(f)); | ||
1216 | ri.mtime = cpu_to_je32(JFFS2_F_I_MTIME(f)); | ||
1217 | ri.offset = cpu_to_je32(offset); | ||
1218 | ri.csize = cpu_to_je32(cdatalen); | ||
1219 | ri.dsize = cpu_to_je32(datalen); | ||
1220 | ri.compr = comprtype & 0xff; | ||
1221 | ri.usercompr = (comprtype >> 8) & 0xff; | ||
1222 | ri.node_crc = cpu_to_je32(crc32(0, &ri, sizeof(ri)-8)); | ||
1223 | ri.data_crc = cpu_to_je32(crc32(0, comprbuf, cdatalen)); | ||
1224 | |||
1225 | new_fn = jffs2_write_dnode(c, f, &ri, comprbuf, cdatalen, phys_ofs, ALLOC_GC); | ||
1226 | |||
1227 | jffs2_free_comprbuf(comprbuf, writebuf); | ||
1228 | |||
1229 | if (IS_ERR(new_fn)) { | ||
1230 | printk(KERN_WARNING "Error writing new dnode: %ld\n", PTR_ERR(new_fn)); | ||
1231 | ret = PTR_ERR(new_fn); | ||
1232 | break; | ||
1233 | } | ||
1234 | ret = jffs2_add_full_dnode_to_inode(c, f, new_fn); | ||
1235 | offset += datalen; | ||
1236 | if (f->metadata) { | ||
1237 | jffs2_mark_node_obsolete(c, f->metadata->raw); | ||
1238 | jffs2_free_full_dnode(f->metadata); | ||
1239 | f->metadata = NULL; | ||
1240 | } | ||
1241 | } | ||
1242 | |||
1243 | jffs2_gc_release_page(c, pg_ptr, &pg); | ||
1244 | return ret; | ||
1245 | } | ||
1246 | |||