]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 | * | |
182ec4ee | 10 | * $Id: gc.c,v 1.155 2005/11/07 11:14:39 gleixner Exp $ |
1da177e4 LT |
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 | ||
182ec4ee | 24 | static int jffs2_garbage_collect_pristine(struct jffs2_sb_info *c, |
1da177e4 LT |
25 | struct jffs2_inode_cache *ic, |
26 | struct jffs2_raw_node_ref *raw); | |
182ec4ee | 27 | static int jffs2_garbage_collect_metadata(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, |
1da177e4 | 28 | struct jffs2_inode_info *f, struct jffs2_full_dnode *fd); |
182ec4ee | 29 | static int jffs2_garbage_collect_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, |
1da177e4 | 30 | struct jffs2_inode_info *f, struct jffs2_full_dirent *fd); |
182ec4ee | 31 | static int jffs2_garbage_collect_deletion_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, |
1da177e4 LT |
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. */ | |
a42163d7 | 53 | again: |
1da177e4 LT |
54 | if (!list_empty(&c->bad_used_list) && c->nr_free_blocks > c->resv_blocks_gcbad) { |
55 | D1(printk(KERN_DEBUG "Picking block from bad_used_list to GC next\n")); | |
56 | nextlist = &c->bad_used_list; | |
57 | } else if (n < 50 && !list_empty(&c->erasable_list)) { | |
182ec4ee | 58 | /* Note that most of them will have gone directly to be erased. |
1da177e4 LT |
59 | So don't favour the erasable_list _too_ much. */ |
60 | D1(printk(KERN_DEBUG "Picking block from erasable_list to GC next\n")); | |
61 | nextlist = &c->erasable_list; | |
62 | } else if (n < 110 && !list_empty(&c->very_dirty_list)) { | |
63 | /* Most of the time, pick one off the very_dirty list */ | |
64 | D1(printk(KERN_DEBUG "Picking block from very_dirty_list to GC next\n")); | |
65 | nextlist = &c->very_dirty_list; | |
66 | } else if (n < 126 && !list_empty(&c->dirty_list)) { | |
67 | D1(printk(KERN_DEBUG "Picking block from dirty_list to GC next\n")); | |
68 | nextlist = &c->dirty_list; | |
69 | } else if (!list_empty(&c->clean_list)) { | |
70 | D1(printk(KERN_DEBUG "Picking block from clean_list to GC next\n")); | |
71 | nextlist = &c->clean_list; | |
72 | } else if (!list_empty(&c->dirty_list)) { | |
73 | D1(printk(KERN_DEBUG "Picking block from dirty_list to GC next (clean_list was empty)\n")); | |
74 | ||
75 | nextlist = &c->dirty_list; | |
76 | } else if (!list_empty(&c->very_dirty_list)) { | |
77 | D1(printk(KERN_DEBUG "Picking block from very_dirty_list to GC next (clean_list and dirty_list were empty)\n")); | |
78 | nextlist = &c->very_dirty_list; | |
79 | } else if (!list_empty(&c->erasable_list)) { | |
80 | D1(printk(KERN_DEBUG "Picking block from erasable_list to GC next (clean_list and {very_,}dirty_list were empty)\n")); | |
81 | ||
82 | nextlist = &c->erasable_list; | |
a42163d7 AB |
83 | } else if (!list_empty(&c->erasable_pending_wbuf_list)) { |
84 | /* There are blocks are wating for the wbuf sync */ | |
85 | D1(printk(KERN_DEBUG "Synching wbuf in order to reuse erasable_pending_wbuf_list blocks\n")); | |
3cceb9f6 | 86 | spin_unlock(&c->erase_completion_lock); |
a42163d7 | 87 | jffs2_flush_wbuf_pad(c); |
3cceb9f6 | 88 | spin_lock(&c->erase_completion_lock); |
a42163d7 | 89 | goto again; |
1da177e4 LT |
90 | } else { |
91 | /* Eep. All were empty */ | |
92 | D1(printk(KERN_NOTICE "jffs2: No clean, dirty _or_ erasable blocks to GC from! Where are they all?\n")); | |
93 | return NULL; | |
94 | } | |
95 | ||
96 | ret = list_entry(nextlist->next, struct jffs2_eraseblock, list); | |
97 | list_del(&ret->list); | |
98 | c->gcblock = ret; | |
99 | ret->gc_node = ret->first_node; | |
100 | if (!ret->gc_node) { | |
101 | printk(KERN_WARNING "Eep. ret->gc_node for block at 0x%08x is NULL\n", ret->offset); | |
102 | BUG(); | |
103 | } | |
182ec4ee | 104 | |
1da177e4 LT |
105 | /* Have we accidentally picked a clean block with wasted space ? */ |
106 | if (ret->wasted_size) { | |
107 | D1(printk(KERN_DEBUG "Converting wasted_size %08x to dirty_size\n", ret->wasted_size)); | |
108 | ret->dirty_size += ret->wasted_size; | |
109 | c->wasted_size -= ret->wasted_size; | |
110 | c->dirty_size += ret->wasted_size; | |
111 | ret->wasted_size = 0; | |
112 | } | |
113 | ||
1da177e4 LT |
114 | return ret; |
115 | } | |
116 | ||
117 | /* jffs2_garbage_collect_pass | |
118 | * Make a single attempt to progress GC. Move one node, and possibly | |
119 | * start erasing one eraseblock. | |
120 | */ | |
121 | int jffs2_garbage_collect_pass(struct jffs2_sb_info *c) | |
122 | { | |
123 | struct jffs2_inode_info *f; | |
124 | struct jffs2_inode_cache *ic; | |
125 | struct jffs2_eraseblock *jeb; | |
126 | struct jffs2_raw_node_ref *raw; | |
127 | int ret = 0, inum, nlink; | |
aa98d7cf | 128 | int xattr = 0; |
1da177e4 LT |
129 | |
130 | if (down_interruptible(&c->alloc_sem)) | |
131 | return -EINTR; | |
132 | ||
133 | for (;;) { | |
134 | spin_lock(&c->erase_completion_lock); | |
135 | if (!c->unchecked_size) | |
136 | break; | |
137 | ||
138 | /* We can't start doing GC yet. We haven't finished checking | |
139 | the node CRCs etc. Do it now. */ | |
182ec4ee | 140 | |
1da177e4 | 141 | /* checked_ino is protected by the alloc_sem */ |
aa98d7cf | 142 | if (c->checked_ino > c->highest_ino && xattr) { |
1da177e4 LT |
143 | printk(KERN_CRIT "Checked all inodes but still 0x%x bytes of unchecked space?\n", |
144 | c->unchecked_size); | |
e0c8e42f | 145 | jffs2_dbg_dump_block_lists_nolock(c); |
1da177e4 | 146 | spin_unlock(&c->erase_completion_lock); |
44b998e1 DW |
147 | up(&c->alloc_sem); |
148 | return -ENOSPC; | |
1da177e4 LT |
149 | } |
150 | ||
151 | spin_unlock(&c->erase_completion_lock); | |
152 | ||
aa98d7cf KK |
153 | if (!xattr) |
154 | xattr = jffs2_verify_xattr(c); | |
155 | ||
1da177e4 LT |
156 | spin_lock(&c->inocache_lock); |
157 | ||
158 | ic = jffs2_get_ino_cache(c, c->checked_ino++); | |
159 | ||
160 | if (!ic) { | |
161 | spin_unlock(&c->inocache_lock); | |
162 | continue; | |
163 | } | |
164 | ||
165 | if (!ic->nlink) { | |
166 | D1(printk(KERN_DEBUG "Skipping check of ino #%d with nlink zero\n", | |
167 | ic->ino)); | |
168 | spin_unlock(&c->inocache_lock); | |
355ed4e1 | 169 | jffs2_xattr_delete_inode(c, ic); |
1da177e4 LT |
170 | continue; |
171 | } | |
172 | switch(ic->state) { | |
173 | case INO_STATE_CHECKEDABSENT: | |
174 | case INO_STATE_PRESENT: | |
175 | D1(printk(KERN_DEBUG "Skipping ino #%u already checked\n", ic->ino)); | |
176 | spin_unlock(&c->inocache_lock); | |
177 | continue; | |
178 | ||
179 | case INO_STATE_GC: | |
180 | case INO_STATE_CHECKING: | |
181 | printk(KERN_WARNING "Inode #%u is in state %d during CRC check phase!\n", ic->ino, ic->state); | |
182 | spin_unlock(&c->inocache_lock); | |
183 | BUG(); | |
184 | ||
185 | case INO_STATE_READING: | |
186 | /* We need to wait for it to finish, lest we move on | |
182ec4ee | 187 | and trigger the BUG() above while we haven't yet |
1da177e4 LT |
188 | finished checking all its nodes */ |
189 | D1(printk(KERN_DEBUG "Waiting for ino #%u to finish reading\n", ic->ino)); | |
d96fb997 DW |
190 | /* We need to come back again for the _same_ inode. We've |
191 | made no progress in this case, but that should be OK */ | |
192 | c->checked_ino--; | |
193 | ||
1da177e4 LT |
194 | up(&c->alloc_sem); |
195 | sleep_on_spinunlock(&c->inocache_wq, &c->inocache_lock); | |
196 | return 0; | |
197 | ||
198 | default: | |
199 | BUG(); | |
200 | ||
201 | case INO_STATE_UNCHECKED: | |
202 | ; | |
203 | } | |
204 | ic->state = INO_STATE_CHECKING; | |
205 | spin_unlock(&c->inocache_lock); | |
206 | ||
207 | D1(printk(KERN_DEBUG "jffs2_garbage_collect_pass() triggering inode scan of ino#%u\n", ic->ino)); | |
208 | ||
209 | ret = jffs2_do_crccheck_inode(c, ic); | |
210 | if (ret) | |
211 | printk(KERN_WARNING "Returned error for crccheck of ino #%u. Expect badness...\n", ic->ino); | |
212 | ||
213 | jffs2_set_inocache_state(c, ic, INO_STATE_CHECKEDABSENT); | |
214 | up(&c->alloc_sem); | |
215 | return ret; | |
216 | } | |
217 | ||
218 | /* First, work out which block we're garbage-collecting */ | |
219 | jeb = c->gcblock; | |
220 | ||
221 | if (!jeb) | |
222 | jeb = jffs2_find_gc_block(c); | |
223 | ||
224 | if (!jeb) { | |
225 | D1 (printk(KERN_NOTICE "jffs2: Couldn't find erase block to garbage collect!\n")); | |
226 | spin_unlock(&c->erase_completion_lock); | |
227 | up(&c->alloc_sem); | |
228 | return -EIO; | |
229 | } | |
230 | ||
231 | 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)); | |
232 | D1(if (c->nextblock) | |
233 | 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)); | |
234 | ||
235 | if (!jeb->used_size) { | |
236 | up(&c->alloc_sem); | |
237 | goto eraseit; | |
238 | } | |
239 | ||
240 | raw = jeb->gc_node; | |
182ec4ee | 241 | |
1da177e4 LT |
242 | while(ref_obsolete(raw)) { |
243 | D1(printk(KERN_DEBUG "Node at 0x%08x is obsolete... skipping\n", ref_offset(raw))); | |
99988f7b | 244 | raw = ref_next(raw); |
1da177e4 LT |
245 | if (unlikely(!raw)) { |
246 | printk(KERN_WARNING "eep. End of raw list while still supposedly nodes to GC\n"); | |
182ec4ee | 247 | printk(KERN_WARNING "erase block at 0x%08x. free_size 0x%08x, dirty_size 0x%08x, used_size 0x%08x\n", |
1da177e4 LT |
248 | jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size); |
249 | jeb->gc_node = raw; | |
250 | spin_unlock(&c->erase_completion_lock); | |
251 | up(&c->alloc_sem); | |
252 | BUG(); | |
253 | } | |
254 | } | |
255 | jeb->gc_node = raw; | |
256 | ||
257 | D1(printk(KERN_DEBUG "Going to garbage collect node at 0x%08x\n", ref_offset(raw))); | |
258 | ||
259 | if (!raw->next_in_ino) { | |
260 | /* Inode-less node. Clean marker, snapshot or something like that */ | |
1da177e4 | 261 | spin_unlock(&c->erase_completion_lock); |
6171586a DW |
262 | if (ref_flags(raw) == REF_PRISTINE) { |
263 | /* It's an unknown node with JFFS2_FEATURE_RWCOMPAT_COPY */ | |
264 | jffs2_garbage_collect_pristine(c, NULL, raw); | |
265 | } else { | |
266 | /* Just mark it obsolete */ | |
267 | jffs2_mark_node_obsolete(c, raw); | |
268 | } | |
1da177e4 LT |
269 | up(&c->alloc_sem); |
270 | goto eraseit_lock; | |
271 | } | |
272 | ||
273 | ic = jffs2_raw_ref_to_ic(raw); | |
274 | ||
084702e0 | 275 | #ifdef CONFIG_JFFS2_FS_XATTR |
aa98d7cf | 276 | /* When 'ic' refers xattr_datum/xattr_ref, this node is GCed as xattr. |
084702e0 KK |
277 | * We can decide whether this node is inode or xattr by ic->class. */ |
278 | if (ic->class == RAWNODE_CLASS_XATTR_DATUM | |
279 | || ic->class == RAWNODE_CLASS_XATTR_REF) { | |
084702e0 KK |
280 | spin_unlock(&c->erase_completion_lock); |
281 | ||
282 | if (ic->class == RAWNODE_CLASS_XATTR_DATUM) { | |
c9f700f8 | 283 | ret = jffs2_garbage_collect_xattr_datum(c, (struct jffs2_xattr_datum *)ic, raw); |
084702e0 | 284 | } else { |
c9f700f8 | 285 | ret = jffs2_garbage_collect_xattr_ref(c, (struct jffs2_xattr_ref *)ic, raw); |
084702e0 | 286 | } |
aa98d7cf | 287 | goto release_sem; |
084702e0 KK |
288 | } |
289 | #endif | |
aa98d7cf | 290 | |
1da177e4 | 291 | /* We need to hold the inocache. Either the erase_completion_lock or |
182ec4ee | 292 | the inocache_lock are sufficient; we trade down since the inocache_lock |
1da177e4 LT |
293 | causes less contention. */ |
294 | spin_lock(&c->inocache_lock); | |
295 | ||
296 | spin_unlock(&c->erase_completion_lock); | |
297 | ||
298 | 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)); | |
299 | ||
300 | /* Three possibilities: | |
301 | 1. Inode is already in-core. We must iget it and do proper | |
302 | updating to its fragtree, etc. | |
303 | 2. Inode is not in-core, node is REF_PRISTINE. We lock the | |
304 | inocache to prevent a read_inode(), copy the node intact. | |
305 | 3. Inode is not in-core, node is not pristine. We must iget() | |
306 | and take the slow path. | |
307 | */ | |
308 | ||
309 | switch(ic->state) { | |
310 | case INO_STATE_CHECKEDABSENT: | |
182ec4ee | 311 | /* It's been checked, but it's not currently in-core. |
1da177e4 LT |
312 | We can just copy any pristine nodes, but have |
313 | to prevent anyone else from doing read_inode() while | |
314 | we're at it, so we set the state accordingly */ | |
315 | if (ref_flags(raw) == REF_PRISTINE) | |
316 | ic->state = INO_STATE_GC; | |
317 | else { | |
182ec4ee | 318 | D1(printk(KERN_DEBUG "Ino #%u is absent but node not REF_PRISTINE. Reading.\n", |
1da177e4 LT |
319 | ic->ino)); |
320 | } | |
321 | break; | |
322 | ||
323 | case INO_STATE_PRESENT: | |
324 | /* It's in-core. GC must iget() it. */ | |
325 | break; | |
326 | ||
327 | case INO_STATE_UNCHECKED: | |
328 | case INO_STATE_CHECKING: | |
329 | case INO_STATE_GC: | |
330 | /* Should never happen. We should have finished checking | |
182ec4ee TG |
331 | by the time we actually start doing any GC, and since |
332 | we're holding the alloc_sem, no other garbage collection | |
1da177e4 LT |
333 | can happen. |
334 | */ | |
335 | printk(KERN_CRIT "Inode #%u already in state %d in jffs2_garbage_collect_pass()!\n", | |
336 | ic->ino, ic->state); | |
337 | up(&c->alloc_sem); | |
338 | spin_unlock(&c->inocache_lock); | |
339 | BUG(); | |
340 | ||
341 | case INO_STATE_READING: | |
342 | /* Someone's currently trying to read it. We must wait for | |
343 | them to finish and then go through the full iget() route | |
344 | to do the GC. However, sometimes read_inode() needs to get | |
345 | the alloc_sem() (for marking nodes invalid) so we must | |
346 | drop the alloc_sem before sleeping. */ | |
347 | ||
348 | up(&c->alloc_sem); | |
349 | D1(printk(KERN_DEBUG "jffs2_garbage_collect_pass() waiting for ino #%u in state %d\n", | |
350 | ic->ino, ic->state)); | |
351 | sleep_on_spinunlock(&c->inocache_wq, &c->inocache_lock); | |
182ec4ee | 352 | /* And because we dropped the alloc_sem we must start again from the |
1da177e4 LT |
353 | beginning. Ponder chance of livelock here -- we're returning success |
354 | without actually making any progress. | |
355 | ||
182ec4ee | 356 | Q: What are the chances that the inode is back in INO_STATE_READING |
1da177e4 LT |
357 | again by the time we next enter this function? And that this happens |
358 | enough times to cause a real delay? | |
359 | ||
182ec4ee | 360 | A: Small enough that I don't care :) |
1da177e4 LT |
361 | */ |
362 | return 0; | |
363 | } | |
364 | ||
365 | /* OK. Now if the inode is in state INO_STATE_GC, we are going to copy the | |
182ec4ee | 366 | node intact, and we don't have to muck about with the fragtree etc. |
1da177e4 LT |
367 | because we know it's not in-core. If it _was_ in-core, we go through |
368 | all the iget() crap anyway */ | |
369 | ||
370 | if (ic->state == INO_STATE_GC) { | |
371 | spin_unlock(&c->inocache_lock); | |
372 | ||
373 | ret = jffs2_garbage_collect_pristine(c, ic, raw); | |
374 | ||
375 | spin_lock(&c->inocache_lock); | |
376 | ic->state = INO_STATE_CHECKEDABSENT; | |
377 | wake_up(&c->inocache_wq); | |
378 | ||
379 | if (ret != -EBADFD) { | |
380 | spin_unlock(&c->inocache_lock); | |
381 | goto release_sem; | |
382 | } | |
383 | ||
384 | /* Fall through if it wanted us to, with inocache_lock held */ | |
385 | } | |
386 | ||
387 | /* Prevent the fairly unlikely race where the gcblock is | |
388 | entirely obsoleted by the final close of a file which had | |
389 | the only valid nodes in the block, followed by erasure, | |
390 | followed by freeing of the ic because the erased block(s) | |
391 | held _all_ the nodes of that inode.... never been seen but | |
392 | it's vaguely possible. */ | |
393 | ||
394 | inum = ic->ino; | |
395 | nlink = ic->nlink; | |
396 | spin_unlock(&c->inocache_lock); | |
397 | ||
398 | f = jffs2_gc_fetch_inode(c, inum, nlink); | |
399 | if (IS_ERR(f)) { | |
400 | ret = PTR_ERR(f); | |
401 | goto release_sem; | |
402 | } | |
403 | if (!f) { | |
404 | ret = 0; | |
405 | goto release_sem; | |
406 | } | |
407 | ||
408 | ret = jffs2_garbage_collect_live(c, jeb, raw, f); | |
409 | ||
410 | jffs2_gc_release_inode(c, f); | |
411 | ||
412 | release_sem: | |
413 | up(&c->alloc_sem); | |
414 | ||
415 | eraseit_lock: | |
416 | /* If we've finished this block, start it erasing */ | |
417 | spin_lock(&c->erase_completion_lock); | |
418 | ||
419 | eraseit: | |
420 | if (c->gcblock && !c->gcblock->used_size) { | |
421 | D1(printk(KERN_DEBUG "Block at 0x%08x completely obsoleted by GC. Moving to erase_pending_list\n", c->gcblock->offset)); | |
422 | /* We're GC'ing an empty block? */ | |
423 | list_add_tail(&c->gcblock->list, &c->erase_pending_list); | |
424 | c->gcblock = NULL; | |
425 | c->nr_erasing_blocks++; | |
426 | jffs2_erase_pending_trigger(c); | |
427 | } | |
428 | spin_unlock(&c->erase_completion_lock); | |
429 | ||
430 | return ret; | |
431 | } | |
432 | ||
433 | static int jffs2_garbage_collect_live(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, | |
434 | struct jffs2_raw_node_ref *raw, struct jffs2_inode_info *f) | |
435 | { | |
436 | struct jffs2_node_frag *frag; | |
437 | struct jffs2_full_dnode *fn = NULL; | |
438 | struct jffs2_full_dirent *fd; | |
439 | uint32_t start = 0, end = 0, nrfrags = 0; | |
440 | int ret = 0; | |
441 | ||
442 | down(&f->sem); | |
443 | ||
444 | /* Now we have the lock for this inode. Check that it's still the one at the head | |
445 | of the list. */ | |
446 | ||
447 | spin_lock(&c->erase_completion_lock); | |
448 | ||
449 | if (c->gcblock != jeb) { | |
450 | spin_unlock(&c->erase_completion_lock); | |
451 | D1(printk(KERN_DEBUG "GC block is no longer gcblock. Restart\n")); | |
452 | goto upnout; | |
453 | } | |
454 | if (ref_obsolete(raw)) { | |
455 | spin_unlock(&c->erase_completion_lock); | |
456 | D1(printk(KERN_DEBUG "node to be GC'd was obsoleted in the meantime.\n")); | |
457 | /* They'll call again */ | |
458 | goto upnout; | |
459 | } | |
460 | spin_unlock(&c->erase_completion_lock); | |
461 | ||
462 | /* OK. Looks safe. And nobody can get us now because we have the semaphore. Move the block */ | |
463 | if (f->metadata && f->metadata->raw == raw) { | |
464 | fn = f->metadata; | |
465 | ret = jffs2_garbage_collect_metadata(c, jeb, f, fn); | |
466 | goto upnout; | |
467 | } | |
468 | ||
469 | /* FIXME. Read node and do lookup? */ | |
470 | for (frag = frag_first(&f->fragtree); frag; frag = frag_next(frag)) { | |
471 | if (frag->node && frag->node->raw == raw) { | |
472 | fn = frag->node; | |
473 | end = frag->ofs + frag->size; | |
474 | if (!nrfrags++) | |
475 | start = frag->ofs; | |
476 | if (nrfrags == frag->node->frags) | |
477 | break; /* We've found them all */ | |
478 | } | |
479 | } | |
480 | if (fn) { | |
481 | if (ref_flags(raw) == REF_PRISTINE) { | |
482 | ret = jffs2_garbage_collect_pristine(c, f->inocache, raw); | |
483 | if (!ret) { | |
484 | /* Urgh. Return it sensibly. */ | |
485 | frag->node->raw = f->inocache->nodes; | |
182ec4ee | 486 | } |
1da177e4 LT |
487 | if (ret != -EBADFD) |
488 | goto upnout; | |
489 | } | |
490 | /* We found a datanode. Do the GC */ | |
491 | if((start >> PAGE_CACHE_SHIFT) < ((end-1) >> PAGE_CACHE_SHIFT)) { | |
492 | /* It crosses a page boundary. Therefore, it must be a hole. */ | |
493 | ret = jffs2_garbage_collect_hole(c, jeb, f, fn, start, end); | |
494 | } else { | |
495 | /* It could still be a hole. But we GC the page this way anyway */ | |
496 | ret = jffs2_garbage_collect_dnode(c, jeb, f, fn, start, end); | |
497 | } | |
498 | goto upnout; | |
499 | } | |
182ec4ee | 500 | |
1da177e4 LT |
501 | /* Wasn't a dnode. Try dirent */ |
502 | for (fd = f->dents; fd; fd=fd->next) { | |
503 | if (fd->raw == raw) | |
504 | break; | |
505 | } | |
506 | ||
507 | if (fd && fd->ino) { | |
508 | ret = jffs2_garbage_collect_dirent(c, jeb, f, fd); | |
509 | } else if (fd) { | |
510 | ret = jffs2_garbage_collect_deletion_dirent(c, jeb, f, fd); | |
511 | } else { | |
512 | printk(KERN_WARNING "Raw node at 0x%08x wasn't in node lists for ino #%u\n", | |
513 | ref_offset(raw), f->inocache->ino); | |
514 | if (ref_obsolete(raw)) { | |
515 | printk(KERN_WARNING "But it's obsolete so we don't mind too much\n"); | |
516 | } else { | |
e0c8e42f AB |
517 | jffs2_dbg_dump_node(c, ref_offset(raw)); |
518 | BUG(); | |
1da177e4 LT |
519 | } |
520 | } | |
521 | upnout: | |
522 | up(&f->sem); | |
523 | ||
524 | return ret; | |
525 | } | |
526 | ||
182ec4ee | 527 | static int jffs2_garbage_collect_pristine(struct jffs2_sb_info *c, |
1da177e4 LT |
528 | struct jffs2_inode_cache *ic, |
529 | struct jffs2_raw_node_ref *raw) | |
530 | { | |
531 | union jffs2_node_union *node; | |
1da177e4 LT |
532 | size_t retlen; |
533 | int ret; | |
534 | uint32_t phys_ofs, alloclen; | |
535 | uint32_t crc, rawlen; | |
536 | int retried = 0; | |
537 | ||
538 | D1(printk(KERN_DEBUG "Going to GC REF_PRISTINE node at 0x%08x\n", ref_offset(raw))); | |
539 | ||
6171586a | 540 | alloclen = rawlen = ref_totlen(c, c->gcblock, raw); |
1da177e4 LT |
541 | |
542 | /* Ask for a small amount of space (or the totlen if smaller) because we | |
543 | don't want to force wastage of the end of a block if splitting would | |
544 | work. */ | |
6171586a DW |
545 | if (ic && alloclen > sizeof(struct jffs2_raw_inode) + JFFS2_MIN_DATA_LEN) |
546 | alloclen = sizeof(struct jffs2_raw_inode) + JFFS2_MIN_DATA_LEN; | |
547 | ||
9fe4854c | 548 | ret = jffs2_reserve_space_gc(c, alloclen, &alloclen, rawlen); |
6171586a | 549 | /* 'rawlen' is not the exact summary size; it is only an upper estimation */ |
e631ddba | 550 | |
1da177e4 LT |
551 | if (ret) |
552 | return ret; | |
553 | ||
554 | if (alloclen < rawlen) { | |
555 | /* Doesn't fit untouched. We'll go the old route and split it */ | |
556 | return -EBADFD; | |
557 | } | |
558 | ||
559 | node = kmalloc(rawlen, GFP_KERNEL); | |
560 | if (!node) | |
561 | return -ENOMEM; | |
562 | ||
563 | ret = jffs2_flash_read(c, ref_offset(raw), rawlen, &retlen, (char *)node); | |
564 | if (!ret && retlen != rawlen) | |
565 | ret = -EIO; | |
566 | if (ret) | |
567 | goto out_node; | |
568 | ||
569 | crc = crc32(0, node, sizeof(struct jffs2_unknown_node)-4); | |
570 | if (je32_to_cpu(node->u.hdr_crc) != crc) { | |
571 | printk(KERN_WARNING "Header CRC failed on REF_PRISTINE node at 0x%08x: Read 0x%08x, calculated 0x%08x\n", | |
572 | ref_offset(raw), je32_to_cpu(node->u.hdr_crc), crc); | |
573 | goto bail; | |
574 | } | |
575 | ||
576 | switch(je16_to_cpu(node->u.nodetype)) { | |
577 | case JFFS2_NODETYPE_INODE: | |
578 | crc = crc32(0, node, sizeof(node->i)-8); | |
579 | if (je32_to_cpu(node->i.node_crc) != crc) { | |
580 | printk(KERN_WARNING "Node CRC failed on REF_PRISTINE data node at 0x%08x: Read 0x%08x, calculated 0x%08x\n", | |
581 | ref_offset(raw), je32_to_cpu(node->i.node_crc), crc); | |
582 | goto bail; | |
583 | } | |
584 | ||
585 | if (je32_to_cpu(node->i.dsize)) { | |
586 | crc = crc32(0, node->i.data, je32_to_cpu(node->i.csize)); | |
587 | if (je32_to_cpu(node->i.data_crc) != crc) { | |
588 | printk(KERN_WARNING "Data CRC failed on REF_PRISTINE data node at 0x%08x: Read 0x%08x, calculated 0x%08x\n", | |
589 | ref_offset(raw), je32_to_cpu(node->i.data_crc), crc); | |
590 | goto bail; | |
591 | } | |
592 | } | |
593 | break; | |
594 | ||
595 | case JFFS2_NODETYPE_DIRENT: | |
596 | crc = crc32(0, node, sizeof(node->d)-8); | |
597 | if (je32_to_cpu(node->d.node_crc) != crc) { | |
598 | printk(KERN_WARNING "Node CRC failed on REF_PRISTINE dirent node at 0x%08x: Read 0x%08x, calculated 0x%08x\n", | |
599 | ref_offset(raw), je32_to_cpu(node->d.node_crc), crc); | |
600 | goto bail; | |
601 | } | |
602 | ||
603 | if (node->d.nsize) { | |
604 | crc = crc32(0, node->d.name, node->d.nsize); | |
605 | if (je32_to_cpu(node->d.name_crc) != crc) { | |
606 | printk(KERN_WARNING "Name CRC failed on REF_PRISTINE dirent ode at 0x%08x: Read 0x%08x, calculated 0x%08x\n", | |
607 | ref_offset(raw), je32_to_cpu(node->d.name_crc), crc); | |
608 | goto bail; | |
609 | } | |
610 | } | |
611 | break; | |
612 | default: | |
6171586a DW |
613 | /* If it's inode-less, we don't _know_ what it is. Just copy it intact */ |
614 | if (ic) { | |
615 | printk(KERN_WARNING "Unknown node type for REF_PRISTINE node at 0x%08x: 0x%04x\n", | |
616 | ref_offset(raw), je16_to_cpu(node->u.nodetype)); | |
617 | goto bail; | |
618 | } | |
1da177e4 LT |
619 | } |
620 | ||
1da177e4 LT |
621 | /* OK, all the CRCs are good; this node can just be copied as-is. */ |
622 | retry: | |
2f785402 | 623 | phys_ofs = write_ofs(c); |
1da177e4 LT |
624 | |
625 | ret = jffs2_flash_write(c, phys_ofs, rawlen, &retlen, (char *)node); | |
626 | ||
627 | if (ret || (retlen != rawlen)) { | |
628 | printk(KERN_NOTICE "Write of %d bytes at 0x%08x failed. returned %d, retlen %zd\n", | |
2f785402 | 629 | rawlen, phys_ofs, ret, retlen); |
1da177e4 | 630 | if (retlen) { |
2f785402 | 631 | jffs2_add_physical_node_ref(c, phys_ofs | REF_OBSOLETE, rawlen, NULL); |
1da177e4 | 632 | } else { |
2f785402 | 633 | printk(KERN_NOTICE "Not marking the space at 0x%08x as dirty because the flash driver returned retlen zero\n", phys_ofs); |
1da177e4 | 634 | } |
2f785402 | 635 | if (!retried) { |
1da177e4 LT |
636 | /* Try to reallocate space and retry */ |
637 | uint32_t dummy; | |
638 | struct jffs2_eraseblock *jeb = &c->blocks[phys_ofs / c->sector_size]; | |
639 | ||
640 | retried = 1; | |
641 | ||
642 | D1(printk(KERN_DEBUG "Retrying failed write of REF_PRISTINE node.\n")); | |
182ec4ee | 643 | |
730554d9 AB |
644 | jffs2_dbg_acct_sanity_check(c,jeb); |
645 | jffs2_dbg_acct_paranoia_check(c, jeb); | |
1da177e4 | 646 | |
9fe4854c | 647 | ret = jffs2_reserve_space_gc(c, rawlen, &dummy, rawlen); |
e631ddba FH |
648 | /* this is not the exact summary size of it, |
649 | it is only an upper estimation */ | |
1da177e4 LT |
650 | |
651 | if (!ret) { | |
652 | D1(printk(KERN_DEBUG "Allocated space at 0x%08x to retry failed write.\n", phys_ofs)); | |
653 | ||
730554d9 AB |
654 | jffs2_dbg_acct_sanity_check(c,jeb); |
655 | jffs2_dbg_acct_paranoia_check(c, jeb); | |
1da177e4 LT |
656 | |
657 | goto retry; | |
658 | } | |
659 | D1(printk(KERN_DEBUG "Failed to allocate space to retry failed write: %d!\n", ret)); | |
1da177e4 LT |
660 | } |
661 | ||
1da177e4 LT |
662 | if (!ret) |
663 | ret = -EIO; | |
664 | goto out_node; | |
665 | } | |
2f785402 | 666 | jffs2_add_physical_node_ref(c, phys_ofs | REF_PRISTINE, rawlen, ic); |
1da177e4 | 667 | |
1da177e4 LT |
668 | jffs2_mark_node_obsolete(c, raw); |
669 | D1(printk(KERN_DEBUG "WHEEE! GC REF_PRISTINE node at 0x%08x succeeded\n", ref_offset(raw))); | |
670 | ||
671 | out_node: | |
672 | kfree(node); | |
673 | return ret; | |
674 | bail: | |
675 | ret = -EBADFD; | |
676 | goto out_node; | |
677 | } | |
678 | ||
182ec4ee | 679 | static int jffs2_garbage_collect_metadata(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, |
1da177e4 LT |
680 | struct jffs2_inode_info *f, struct jffs2_full_dnode *fn) |
681 | { | |
682 | struct jffs2_full_dnode *new_fn; | |
683 | struct jffs2_raw_inode ri; | |
8557fd51 | 684 | struct jffs2_node_frag *last_frag; |
aef9ab47 | 685 | union jffs2_device_node dev; |
1da177e4 | 686 | char *mdata = NULL, mdatalen = 0; |
9fe4854c | 687 | uint32_t alloclen, ilen; |
1da177e4 LT |
688 | int ret; |
689 | ||
690 | if (S_ISBLK(JFFS2_F_I_MODE(f)) || | |
691 | S_ISCHR(JFFS2_F_I_MODE(f)) ) { | |
692 | /* For these, we don't actually need to read the old node */ | |
aef9ab47 | 693 | mdatalen = jffs2_encode_dev(&dev, JFFS2_F_I_RDEV(f)); |
1da177e4 | 694 | mdata = (char *)&dev; |
1da177e4 LT |
695 | D1(printk(KERN_DEBUG "jffs2_garbage_collect_metadata(): Writing %d bytes of kdev_t\n", mdatalen)); |
696 | } else if (S_ISLNK(JFFS2_F_I_MODE(f))) { | |
697 | mdatalen = fn->size; | |
698 | mdata = kmalloc(fn->size, GFP_KERNEL); | |
699 | if (!mdata) { | |
700 | printk(KERN_WARNING "kmalloc of mdata failed in jffs2_garbage_collect_metadata()\n"); | |
701 | return -ENOMEM; | |
702 | } | |
703 | ret = jffs2_read_dnode(c, f, fn, mdata, 0, mdatalen); | |
704 | if (ret) { | |
705 | printk(KERN_WARNING "read of old metadata failed in jffs2_garbage_collect_metadata(): %d\n", ret); | |
706 | kfree(mdata); | |
707 | return ret; | |
708 | } | |
709 | D1(printk(KERN_DEBUG "jffs2_garbage_collect_metadata(): Writing %d bites of symlink target\n", mdatalen)); | |
710 | ||
711 | } | |
182ec4ee | 712 | |
9fe4854c | 713 | ret = jffs2_reserve_space_gc(c, sizeof(ri) + mdatalen, &alloclen, |
e631ddba | 714 | JFFS2_SUMMARY_INODE_SIZE); |
1da177e4 LT |
715 | if (ret) { |
716 | printk(KERN_WARNING "jffs2_reserve_space_gc of %zd bytes for garbage_collect_metadata failed: %d\n", | |
717 | sizeof(ri)+ mdatalen, ret); | |
718 | goto out; | |
719 | } | |
182ec4ee | 720 | |
8557fd51 AB |
721 | last_frag = frag_last(&f->fragtree); |
722 | if (last_frag) | |
723 | /* Fetch the inode length from the fragtree rather then | |
724 | * from i_size since i_size may have not been updated yet */ | |
725 | ilen = last_frag->ofs + last_frag->size; | |
726 | else | |
727 | ilen = JFFS2_F_I_SIZE(f); | |
182ec4ee | 728 | |
1da177e4 LT |
729 | memset(&ri, 0, sizeof(ri)); |
730 | ri.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK); | |
731 | ri.nodetype = cpu_to_je16(JFFS2_NODETYPE_INODE); | |
732 | ri.totlen = cpu_to_je32(sizeof(ri) + mdatalen); | |
733 | ri.hdr_crc = cpu_to_je32(crc32(0, &ri, sizeof(struct jffs2_unknown_node)-4)); | |
734 | ||
735 | ri.ino = cpu_to_je32(f->inocache->ino); | |
736 | ri.version = cpu_to_je32(++f->highest_version); | |
737 | ri.mode = cpu_to_jemode(JFFS2_F_I_MODE(f)); | |
738 | ri.uid = cpu_to_je16(JFFS2_F_I_UID(f)); | |
739 | ri.gid = cpu_to_je16(JFFS2_F_I_GID(f)); | |
8557fd51 | 740 | ri.isize = cpu_to_je32(ilen); |
1da177e4 LT |
741 | ri.atime = cpu_to_je32(JFFS2_F_I_ATIME(f)); |
742 | ri.ctime = cpu_to_je32(JFFS2_F_I_CTIME(f)); | |
743 | ri.mtime = cpu_to_je32(JFFS2_F_I_MTIME(f)); | |
744 | ri.offset = cpu_to_je32(0); | |
745 | ri.csize = cpu_to_je32(mdatalen); | |
746 | ri.dsize = cpu_to_je32(mdatalen); | |
747 | ri.compr = JFFS2_COMPR_NONE; | |
748 | ri.node_crc = cpu_to_je32(crc32(0, &ri, sizeof(ri)-8)); | |
749 | ri.data_crc = cpu_to_je32(crc32(0, mdata, mdatalen)); | |
750 | ||
9fe4854c | 751 | new_fn = jffs2_write_dnode(c, f, &ri, mdata, mdatalen, ALLOC_GC); |
1da177e4 LT |
752 | |
753 | if (IS_ERR(new_fn)) { | |
754 | printk(KERN_WARNING "Error writing new dnode: %ld\n", PTR_ERR(new_fn)); | |
755 | ret = PTR_ERR(new_fn); | |
756 | goto out; | |
757 | } | |
758 | jffs2_mark_node_obsolete(c, fn->raw); | |
759 | jffs2_free_full_dnode(fn); | |
760 | f->metadata = new_fn; | |
761 | out: | |
762 | if (S_ISLNK(JFFS2_F_I_MODE(f))) | |
763 | kfree(mdata); | |
764 | return ret; | |
765 | } | |
766 | ||
182ec4ee | 767 | static int jffs2_garbage_collect_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, |
1da177e4 LT |
768 | struct jffs2_inode_info *f, struct jffs2_full_dirent *fd) |
769 | { | |
770 | struct jffs2_full_dirent *new_fd; | |
771 | struct jffs2_raw_dirent rd; | |
9fe4854c | 772 | uint32_t alloclen; |
1da177e4 LT |
773 | int ret; |
774 | ||
775 | rd.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK); | |
776 | rd.nodetype = cpu_to_je16(JFFS2_NODETYPE_DIRENT); | |
777 | rd.nsize = strlen(fd->name); | |
778 | rd.totlen = cpu_to_je32(sizeof(rd) + rd.nsize); | |
779 | rd.hdr_crc = cpu_to_je32(crc32(0, &rd, sizeof(struct jffs2_unknown_node)-4)); | |
780 | ||
781 | rd.pino = cpu_to_je32(f->inocache->ino); | |
782 | rd.version = cpu_to_je32(++f->highest_version); | |
783 | rd.ino = cpu_to_je32(fd->ino); | |
3a69e0cd AB |
784 | /* If the times on this inode were set by explicit utime() they can be different, |
785 | so refrain from splatting them. */ | |
786 | if (JFFS2_F_I_MTIME(f) == JFFS2_F_I_CTIME(f)) | |
787 | rd.mctime = cpu_to_je32(JFFS2_F_I_MTIME(f)); | |
182ec4ee | 788 | else |
3a69e0cd | 789 | rd.mctime = cpu_to_je32(0); |
1da177e4 LT |
790 | rd.type = fd->type; |
791 | rd.node_crc = cpu_to_je32(crc32(0, &rd, sizeof(rd)-8)); | |
792 | rd.name_crc = cpu_to_je32(crc32(0, fd->name, rd.nsize)); | |
182ec4ee | 793 | |
9fe4854c | 794 | ret = jffs2_reserve_space_gc(c, sizeof(rd)+rd.nsize, &alloclen, |
e631ddba | 795 | JFFS2_SUMMARY_DIRENT_SIZE(rd.nsize)); |
1da177e4 LT |
796 | if (ret) { |
797 | printk(KERN_WARNING "jffs2_reserve_space_gc of %zd bytes for garbage_collect_dirent failed: %d\n", | |
798 | sizeof(rd)+rd.nsize, ret); | |
799 | return ret; | |
800 | } | |
9fe4854c | 801 | new_fd = jffs2_write_dirent(c, f, &rd, fd->name, rd.nsize, ALLOC_GC); |
1da177e4 LT |
802 | |
803 | if (IS_ERR(new_fd)) { | |
804 | printk(KERN_WARNING "jffs2_write_dirent in garbage_collect_dirent failed: %ld\n", PTR_ERR(new_fd)); | |
805 | return PTR_ERR(new_fd); | |
806 | } | |
807 | jffs2_add_fd_to_list(c, new_fd, &f->dents); | |
808 | return 0; | |
809 | } | |
810 | ||
182ec4ee | 811 | static int jffs2_garbage_collect_deletion_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, |
1da177e4 LT |
812 | struct jffs2_inode_info *f, struct jffs2_full_dirent *fd) |
813 | { | |
814 | struct jffs2_full_dirent **fdp = &f->dents; | |
815 | int found = 0; | |
816 | ||
817 | /* On a medium where we can't actually mark nodes obsolete | |
818 | pernamently, such as NAND flash, we need to work out | |
819 | whether this deletion dirent is still needed to actively | |
820 | delete a 'real' dirent with the same name that's still | |
821 | somewhere else on the flash. */ | |
822 | if (!jffs2_can_mark_obsolete(c)) { | |
823 | struct jffs2_raw_dirent *rd; | |
824 | struct jffs2_raw_node_ref *raw; | |
825 | int ret; | |
826 | size_t retlen; | |
827 | int name_len = strlen(fd->name); | |
828 | uint32_t name_crc = crc32(0, fd->name, name_len); | |
829 | uint32_t rawlen = ref_totlen(c, jeb, fd->raw); | |
830 | ||
831 | rd = kmalloc(rawlen, GFP_KERNEL); | |
832 | if (!rd) | |
833 | return -ENOMEM; | |
834 | ||
835 | /* Prevent the erase code from nicking the obsolete node refs while | |
836 | we're looking at them. I really don't like this extra lock but | |
837 | can't see any alternative. Suggestions on a postcard to... */ | |
838 | down(&c->erase_free_sem); | |
839 | ||
840 | for (raw = f->inocache->nodes; raw != (void *)f->inocache; raw = raw->next_in_ino) { | |
841 | ||
aba54da3 AB |
842 | cond_resched(); |
843 | ||
1da177e4 LT |
844 | /* We only care about obsolete ones */ |
845 | if (!(ref_obsolete(raw))) | |
846 | continue; | |
847 | ||
848 | /* Any dirent with the same name is going to have the same length... */ | |
849 | if (ref_totlen(c, NULL, raw) != rawlen) | |
850 | continue; | |
851 | ||
182ec4ee | 852 | /* Doesn't matter if there's one in the same erase block. We're going to |
1da177e4 | 853 | delete it too at the same time. */ |
3be36675 | 854 | if (SECTOR_ADDR(raw->flash_offset) == SECTOR_ADDR(fd->raw->flash_offset)) |
1da177e4 LT |
855 | continue; |
856 | ||
857 | D1(printk(KERN_DEBUG "Check potential deletion dirent at %08x\n", ref_offset(raw))); | |
858 | ||
859 | /* This is an obsolete node belonging to the same directory, and it's of the right | |
860 | length. We need to take a closer look...*/ | |
861 | ret = jffs2_flash_read(c, ref_offset(raw), rawlen, &retlen, (char *)rd); | |
862 | if (ret) { | |
863 | printk(KERN_WARNING "jffs2_g_c_deletion_dirent(): Read error (%d) reading obsolete node at %08x\n", ret, ref_offset(raw)); | |
864 | /* If we can't read it, we don't need to continue to obsolete it. Continue */ | |
865 | continue; | |
866 | } | |
867 | if (retlen != rawlen) { | |
868 | printk(KERN_WARNING "jffs2_g_c_deletion_dirent(): Short read (%zd not %u) reading header from obsolete node at %08x\n", | |
869 | retlen, rawlen, ref_offset(raw)); | |
870 | continue; | |
871 | } | |
872 | ||
873 | if (je16_to_cpu(rd->nodetype) != JFFS2_NODETYPE_DIRENT) | |
874 | continue; | |
875 | ||
876 | /* If the name CRC doesn't match, skip */ | |
877 | if (je32_to_cpu(rd->name_crc) != name_crc) | |
878 | continue; | |
879 | ||
880 | /* If the name length doesn't match, or it's another deletion dirent, skip */ | |
881 | if (rd->nsize != name_len || !je32_to_cpu(rd->ino)) | |
882 | continue; | |
883 | ||
884 | /* OK, check the actual name now */ | |
885 | if (memcmp(rd->name, fd->name, name_len)) | |
886 | continue; | |
887 | ||
888 | /* OK. The name really does match. There really is still an older node on | |
889 | the flash which our deletion dirent obsoletes. So we have to write out | |
890 | a new deletion dirent to replace it */ | |
891 | up(&c->erase_free_sem); | |
892 | ||
893 | D1(printk(KERN_DEBUG "Deletion dirent at %08x still obsoletes real dirent \"%s\" at %08x for ino #%u\n", | |
894 | ref_offset(fd->raw), fd->name, ref_offset(raw), je32_to_cpu(rd->ino))); | |
895 | kfree(rd); | |
896 | ||
897 | return jffs2_garbage_collect_dirent(c, jeb, f, fd); | |
898 | } | |
899 | ||
900 | up(&c->erase_free_sem); | |
901 | kfree(rd); | |
902 | } | |
903 | ||
182ec4ee | 904 | /* FIXME: If we're deleting a dirent which contains the current mtime and ctime, |
3a69e0cd AB |
905 | we should update the metadata node with those times accordingly */ |
906 | ||
1da177e4 LT |
907 | /* No need for it any more. Just mark it obsolete and remove it from the list */ |
908 | while (*fdp) { | |
909 | if ((*fdp) == fd) { | |
910 | found = 1; | |
911 | *fdp = fd->next; | |
912 | break; | |
913 | } | |
914 | fdp = &(*fdp)->next; | |
915 | } | |
916 | if (!found) { | |
917 | printk(KERN_WARNING "Deletion dirent \"%s\" not found in list for ino #%u\n", fd->name, f->inocache->ino); | |
918 | } | |
919 | jffs2_mark_node_obsolete(c, fd->raw); | |
920 | jffs2_free_full_dirent(fd); | |
921 | return 0; | |
922 | } | |
923 | ||
924 | static int jffs2_garbage_collect_hole(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, | |
925 | struct jffs2_inode_info *f, struct jffs2_full_dnode *fn, | |
926 | uint32_t start, uint32_t end) | |
927 | { | |
928 | struct jffs2_raw_inode ri; | |
929 | struct jffs2_node_frag *frag; | |
930 | struct jffs2_full_dnode *new_fn; | |
9fe4854c | 931 | uint32_t alloclen, ilen; |
1da177e4 LT |
932 | int ret; |
933 | ||
934 | D1(printk(KERN_DEBUG "Writing replacement hole node for ino #%u from offset 0x%x to 0x%x\n", | |
935 | f->inocache->ino, start, end)); | |
182ec4ee | 936 | |
1da177e4 LT |
937 | memset(&ri, 0, sizeof(ri)); |
938 | ||
939 | if(fn->frags > 1) { | |
940 | size_t readlen; | |
941 | uint32_t crc; | |
182ec4ee | 942 | /* It's partially obsoleted by a later write. So we have to |
1da177e4 LT |
943 | write it out again with the _same_ version as before */ |
944 | ret = jffs2_flash_read(c, ref_offset(fn->raw), sizeof(ri), &readlen, (char *)&ri); | |
945 | if (readlen != sizeof(ri) || ret) { | |
946 | 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); | |
947 | goto fill; | |
948 | } | |
949 | if (je16_to_cpu(ri.nodetype) != JFFS2_NODETYPE_INODE) { | |
950 | printk(KERN_WARNING "jffs2_garbage_collect_hole: Node at 0x%08x had node type 0x%04x instead of JFFS2_NODETYPE_INODE(0x%04x)\n", | |
951 | ref_offset(fn->raw), | |
952 | je16_to_cpu(ri.nodetype), JFFS2_NODETYPE_INODE); | |
953 | return -EIO; | |
954 | } | |
955 | if (je32_to_cpu(ri.totlen) != sizeof(ri)) { | |
956 | printk(KERN_WARNING "jffs2_garbage_collect_hole: Node at 0x%08x had totlen 0x%x instead of expected 0x%zx\n", | |
957 | ref_offset(fn->raw), | |
958 | je32_to_cpu(ri.totlen), sizeof(ri)); | |
959 | return -EIO; | |
960 | } | |
961 | crc = crc32(0, &ri, sizeof(ri)-8); | |
962 | if (crc != je32_to_cpu(ri.node_crc)) { | |
963 | printk(KERN_WARNING "jffs2_garbage_collect_hole: Node at 0x%08x had CRC 0x%08x which doesn't match calculated CRC 0x%08x\n", | |
182ec4ee | 964 | ref_offset(fn->raw), |
1da177e4 LT |
965 | je32_to_cpu(ri.node_crc), crc); |
966 | /* FIXME: We could possibly deal with this by writing new holes for each frag */ | |
182ec4ee | 967 | printk(KERN_WARNING "Data in the range 0x%08x to 0x%08x of inode #%u will be lost\n", |
1da177e4 LT |
968 | start, end, f->inocache->ino); |
969 | goto fill; | |
970 | } | |
971 | if (ri.compr != JFFS2_COMPR_ZERO) { | |
972 | printk(KERN_WARNING "jffs2_garbage_collect_hole: Node 0x%08x wasn't a hole node!\n", ref_offset(fn->raw)); | |
182ec4ee | 973 | printk(KERN_WARNING "Data in the range 0x%08x to 0x%08x of inode #%u will be lost\n", |
1da177e4 LT |
974 | start, end, f->inocache->ino); |
975 | goto fill; | |
976 | } | |
977 | } else { | |
978 | fill: | |
979 | ri.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK); | |
980 | ri.nodetype = cpu_to_je16(JFFS2_NODETYPE_INODE); | |
981 | ri.totlen = cpu_to_je32(sizeof(ri)); | |
982 | ri.hdr_crc = cpu_to_je32(crc32(0, &ri, sizeof(struct jffs2_unknown_node)-4)); | |
983 | ||
984 | ri.ino = cpu_to_je32(f->inocache->ino); | |
985 | ri.version = cpu_to_je32(++f->highest_version); | |
986 | ri.offset = cpu_to_je32(start); | |
987 | ri.dsize = cpu_to_je32(end - start); | |
988 | ri.csize = cpu_to_je32(0); | |
989 | ri.compr = JFFS2_COMPR_ZERO; | |
990 | } | |
182ec4ee | 991 | |
8557fd51 AB |
992 | frag = frag_last(&f->fragtree); |
993 | if (frag) | |
994 | /* Fetch the inode length from the fragtree rather then | |
995 | * from i_size since i_size may have not been updated yet */ | |
996 | ilen = frag->ofs + frag->size; | |
997 | else | |
998 | ilen = JFFS2_F_I_SIZE(f); | |
999 | ||
1da177e4 LT |
1000 | ri.mode = cpu_to_jemode(JFFS2_F_I_MODE(f)); |
1001 | ri.uid = cpu_to_je16(JFFS2_F_I_UID(f)); | |
1002 | ri.gid = cpu_to_je16(JFFS2_F_I_GID(f)); | |
8557fd51 | 1003 | ri.isize = cpu_to_je32(ilen); |
1da177e4 LT |
1004 | ri.atime = cpu_to_je32(JFFS2_F_I_ATIME(f)); |
1005 | ri.ctime = cpu_to_je32(JFFS2_F_I_CTIME(f)); | |
1006 | ri.mtime = cpu_to_je32(JFFS2_F_I_MTIME(f)); | |
1007 | ri.data_crc = cpu_to_je32(0); | |
1008 | ri.node_crc = cpu_to_je32(crc32(0, &ri, sizeof(ri)-8)); | |
1009 | ||
9fe4854c DW |
1010 | ret = jffs2_reserve_space_gc(c, sizeof(ri), &alloclen, |
1011 | JFFS2_SUMMARY_INODE_SIZE); | |
1da177e4 LT |
1012 | if (ret) { |
1013 | printk(KERN_WARNING "jffs2_reserve_space_gc of %zd bytes for garbage_collect_hole failed: %d\n", | |
1014 | sizeof(ri), ret); | |
1015 | return ret; | |
1016 | } | |
9fe4854c | 1017 | new_fn = jffs2_write_dnode(c, f, &ri, NULL, 0, ALLOC_GC); |
1da177e4 LT |
1018 | |
1019 | if (IS_ERR(new_fn)) { | |
1020 | printk(KERN_WARNING "Error writing new hole node: %ld\n", PTR_ERR(new_fn)); | |
1021 | return PTR_ERR(new_fn); | |
1022 | } | |
1023 | if (je32_to_cpu(ri.version) == f->highest_version) { | |
1024 | jffs2_add_full_dnode_to_inode(c, f, new_fn); | |
1025 | if (f->metadata) { | |
1026 | jffs2_mark_node_obsolete(c, f->metadata->raw); | |
1027 | jffs2_free_full_dnode(f->metadata); | |
1028 | f->metadata = NULL; | |
1029 | } | |
1030 | return 0; | |
1031 | } | |
1032 | ||
182ec4ee | 1033 | /* |
1da177e4 LT |
1034 | * We should only get here in the case where the node we are |
1035 | * replacing had more than one frag, so we kept the same version | |
182ec4ee | 1036 | * number as before. (Except in case of error -- see 'goto fill;' |
1da177e4 LT |
1037 | * above.) |
1038 | */ | |
1039 | D1(if(unlikely(fn->frags <= 1)) { | |
1040 | printk(KERN_WARNING "jffs2_garbage_collect_hole: Replacing fn with %d frag(s) but new ver %d != highest_version %d of ino #%d\n", | |
1041 | fn->frags, je32_to_cpu(ri.version), f->highest_version, | |
1042 | je32_to_cpu(ri.ino)); | |
1043 | }); | |
1044 | ||
1045 | /* This is a partially-overlapped hole node. Mark it REF_NORMAL not REF_PRISTINE */ | |
1046 | mark_ref_normal(new_fn->raw); | |
1047 | ||
182ec4ee | 1048 | for (frag = jffs2_lookup_node_frag(&f->fragtree, fn->ofs); |
1da177e4 LT |
1049 | frag; frag = frag_next(frag)) { |
1050 | if (frag->ofs > fn->size + fn->ofs) | |
1051 | break; | |
1052 | if (frag->node == fn) { | |
1053 | frag->node = new_fn; | |
1054 | new_fn->frags++; | |
1055 | fn->frags--; | |
1056 | } | |
1057 | } | |
1058 | if (fn->frags) { | |
1059 | printk(KERN_WARNING "jffs2_garbage_collect_hole: Old node still has frags!\n"); | |
1060 | BUG(); | |
1061 | } | |
1062 | if (!new_fn->frags) { | |
1063 | printk(KERN_WARNING "jffs2_garbage_collect_hole: New node has no frags!\n"); | |
1064 | BUG(); | |
1065 | } | |
182ec4ee | 1066 | |
1da177e4 LT |
1067 | jffs2_mark_node_obsolete(c, fn->raw); |
1068 | jffs2_free_full_dnode(fn); | |
182ec4ee | 1069 | |
1da177e4 LT |
1070 | return 0; |
1071 | } | |
1072 | ||
1073 | static int jffs2_garbage_collect_dnode(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, | |
1074 | struct jffs2_inode_info *f, struct jffs2_full_dnode *fn, | |
1075 | uint32_t start, uint32_t end) | |
1076 | { | |
1077 | struct jffs2_full_dnode *new_fn; | |
1078 | struct jffs2_raw_inode ri; | |
9fe4854c | 1079 | uint32_t alloclen, offset, orig_end, orig_start; |
1da177e4 LT |
1080 | int ret = 0; |
1081 | unsigned char *comprbuf = NULL, *writebuf; | |
1082 | unsigned long pg; | |
1083 | unsigned char *pg_ptr; | |
182ec4ee | 1084 | |
1da177e4 LT |
1085 | memset(&ri, 0, sizeof(ri)); |
1086 | ||
1087 | D1(printk(KERN_DEBUG "Writing replacement dnode for ino #%u from offset 0x%x to 0x%x\n", | |
1088 | f->inocache->ino, start, end)); | |
1089 | ||
1090 | orig_end = end; | |
1091 | orig_start = start; | |
1092 | ||
1093 | if (c->nr_free_blocks + c->nr_erasing_blocks > c->resv_blocks_gcmerge) { | |
1094 | /* Attempt to do some merging. But only expand to cover logically | |
1095 | adjacent frags if the block containing them is already considered | |
182ec4ee TG |
1096 | to be dirty. Otherwise we end up with GC just going round in |
1097 | circles dirtying the nodes it already wrote out, especially | |
1da177e4 LT |
1098 | on NAND where we have small eraseblocks and hence a much higher |
1099 | chance of nodes having to be split to cross boundaries. */ | |
1100 | ||
1101 | struct jffs2_node_frag *frag; | |
1102 | uint32_t min, max; | |
1103 | ||
1104 | min = start & ~(PAGE_CACHE_SIZE-1); | |
1105 | max = min + PAGE_CACHE_SIZE; | |
1106 | ||
1107 | frag = jffs2_lookup_node_frag(&f->fragtree, start); | |
1108 | ||
1109 | /* BUG_ON(!frag) but that'll happen anyway... */ | |
1110 | ||
1111 | BUG_ON(frag->ofs != start); | |
1112 | ||
1113 | /* First grow down... */ | |
1114 | while((frag = frag_prev(frag)) && frag->ofs >= min) { | |
1115 | ||
1116 | /* If the previous frag doesn't even reach the beginning, there's | |
1117 | excessive fragmentation. Just merge. */ | |
1118 | if (frag->ofs > min) { | |
1119 | D1(printk(KERN_DEBUG "Expanding down to cover partial frag (0x%x-0x%x)\n", | |
1120 | frag->ofs, frag->ofs+frag->size)); | |
1121 | start = frag->ofs; | |
1122 | continue; | |
1123 | } | |
1124 | /* OK. This frag holds the first byte of the page. */ | |
1125 | if (!frag->node || !frag->node->raw) { | |
1126 | D1(printk(KERN_DEBUG "First frag in page is hole (0x%x-0x%x). Not expanding down.\n", | |
1127 | frag->ofs, frag->ofs+frag->size)); | |
1128 | break; | |
1129 | } else { | |
1130 | ||
182ec4ee | 1131 | /* OK, it's a frag which extends to the beginning of the page. Does it live |
1da177e4 LT |
1132 | in a block which is still considered clean? If so, don't obsolete it. |
1133 | If not, cover it anyway. */ | |
1134 | ||
1135 | struct jffs2_raw_node_ref *raw = frag->node->raw; | |
1136 | struct jffs2_eraseblock *jeb; | |
1137 | ||
1138 | jeb = &c->blocks[raw->flash_offset / c->sector_size]; | |
1139 | ||
1140 | if (jeb == c->gcblock) { | |
1141 | D1(printk(KERN_DEBUG "Expanding down to cover frag (0x%x-0x%x) in gcblock at %08x\n", | |
1142 | frag->ofs, frag->ofs+frag->size, ref_offset(raw))); | |
1143 | start = frag->ofs; | |
1144 | break; | |
1145 | } | |
1146 | if (!ISDIRTY(jeb->dirty_size + jeb->wasted_size)) { | |
1147 | D1(printk(KERN_DEBUG "Not expanding down to cover frag (0x%x-0x%x) in clean block %08x\n", | |
1148 | frag->ofs, frag->ofs+frag->size, jeb->offset)); | |
1149 | break; | |
1150 | } | |
1151 | ||
1152 | D1(printk(KERN_DEBUG "Expanding down to cover frag (0x%x-0x%x) in dirty block %08x\n", | |
1153 | frag->ofs, frag->ofs+frag->size, jeb->offset)); | |
1154 | start = frag->ofs; | |
1155 | break; | |
1156 | } | |
1157 | } | |
1158 | ||
1159 | /* ... then up */ | |
1160 | ||
1161 | /* Find last frag which is actually part of the node we're to GC. */ | |
1162 | frag = jffs2_lookup_node_frag(&f->fragtree, end-1); | |
1163 | ||
1164 | while((frag = frag_next(frag)) && frag->ofs+frag->size <= max) { | |
1165 | ||
1166 | /* If the previous frag doesn't even reach the beginning, there's lots | |
1167 | of fragmentation. Just merge. */ | |
1168 | if (frag->ofs+frag->size < max) { | |
1169 | D1(printk(KERN_DEBUG "Expanding up to cover partial frag (0x%x-0x%x)\n", | |
1170 | frag->ofs, frag->ofs+frag->size)); | |
1171 | end = frag->ofs + frag->size; | |
1172 | continue; | |
1173 | } | |
1174 | ||
1175 | if (!frag->node || !frag->node->raw) { | |
1176 | D1(printk(KERN_DEBUG "Last frag in page is hole (0x%x-0x%x). Not expanding up.\n", | |
1177 | frag->ofs, frag->ofs+frag->size)); | |
1178 | break; | |
1179 | } else { | |
1180 | ||
182ec4ee | 1181 | /* OK, it's a frag which extends to the beginning of the page. Does it live |
1da177e4 LT |
1182 | in a block which is still considered clean? If so, don't obsolete it. |
1183 | If not, cover it anyway. */ | |
1184 | ||
1185 | struct jffs2_raw_node_ref *raw = frag->node->raw; | |
1186 | struct jffs2_eraseblock *jeb; | |
1187 | ||
1188 | jeb = &c->blocks[raw->flash_offset / c->sector_size]; | |
1189 | ||
1190 | if (jeb == c->gcblock) { | |
1191 | D1(printk(KERN_DEBUG "Expanding up to cover frag (0x%x-0x%x) in gcblock at %08x\n", | |
1192 | frag->ofs, frag->ofs+frag->size, ref_offset(raw))); | |
1193 | end = frag->ofs + frag->size; | |
1194 | break; | |
1195 | } | |
1196 | if (!ISDIRTY(jeb->dirty_size + jeb->wasted_size)) { | |
1197 | D1(printk(KERN_DEBUG "Not expanding up to cover frag (0x%x-0x%x) in clean block %08x\n", | |
1198 | frag->ofs, frag->ofs+frag->size, jeb->offset)); | |
1199 | break; | |
1200 | } | |
1201 | ||
1202 | D1(printk(KERN_DEBUG "Expanding up to cover frag (0x%x-0x%x) in dirty block %08x\n", | |
1203 | frag->ofs, frag->ofs+frag->size, jeb->offset)); | |
1204 | end = frag->ofs + frag->size; | |
1205 | break; | |
1206 | } | |
1207 | } | |
182ec4ee | 1208 | D1(printk(KERN_DEBUG "Expanded dnode to write from (0x%x-0x%x) to (0x%x-0x%x)\n", |
1da177e4 LT |
1209 | orig_start, orig_end, start, end)); |
1210 | ||
8557fd51 | 1211 | D1(BUG_ON(end > frag_last(&f->fragtree)->ofs + frag_last(&f->fragtree)->size)); |
1da177e4 LT |
1212 | BUG_ON(end < orig_end); |
1213 | BUG_ON(start > orig_start); | |
1214 | } | |
182ec4ee | 1215 | |
1da177e4 LT |
1216 | /* First, use readpage() to read the appropriate page into the page cache */ |
1217 | /* Q: What happens if we actually try to GC the _same_ page for which commit_write() | |
1218 | * triggered garbage collection in the first place? | |
1219 | * A: I _think_ it's OK. read_cache_page shouldn't deadlock, we'll write out the | |
1220 | * page OK. We'll actually write it out again in commit_write, which is a little | |
1221 | * suboptimal, but at least we're correct. | |
1222 | */ | |
1223 | pg_ptr = jffs2_gc_fetch_page(c, f, start, &pg); | |
1224 | ||
1225 | if (IS_ERR(pg_ptr)) { | |
1226 | printk(KERN_WARNING "read_cache_page() returned error: %ld\n", PTR_ERR(pg_ptr)); | |
1227 | return PTR_ERR(pg_ptr); | |
1228 | } | |
1229 | ||
1230 | offset = start; | |
1231 | while(offset < orig_end) { | |
1232 | uint32_t datalen; | |
1233 | uint32_t cdatalen; | |
1234 | uint16_t comprtype = JFFS2_COMPR_NONE; | |
1235 | ||
9fe4854c | 1236 | ret = jffs2_reserve_space_gc(c, sizeof(ri) + JFFS2_MIN_DATA_LEN, |
e631ddba | 1237 | &alloclen, JFFS2_SUMMARY_INODE_SIZE); |
1da177e4 LT |
1238 | |
1239 | if (ret) { | |
1240 | printk(KERN_WARNING "jffs2_reserve_space_gc of %zd bytes for garbage_collect_dnode failed: %d\n", | |
1241 | sizeof(ri)+ JFFS2_MIN_DATA_LEN, ret); | |
1242 | break; | |
1243 | } | |
1244 | cdatalen = min_t(uint32_t, alloclen - sizeof(ri), end - offset); | |
1245 | datalen = end - offset; | |
1246 | ||
1247 | writebuf = pg_ptr + (offset & (PAGE_CACHE_SIZE -1)); | |
1248 | ||
1249 | comprtype = jffs2_compress(c, f, writebuf, &comprbuf, &datalen, &cdatalen); | |
1250 | ||
1251 | ri.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK); | |
1252 | ri.nodetype = cpu_to_je16(JFFS2_NODETYPE_INODE); | |
1253 | ri.totlen = cpu_to_je32(sizeof(ri) + cdatalen); | |
1254 | ri.hdr_crc = cpu_to_je32(crc32(0, &ri, sizeof(struct jffs2_unknown_node)-4)); | |
1255 | ||
1256 | ri.ino = cpu_to_je32(f->inocache->ino); | |
1257 | ri.version = cpu_to_je32(++f->highest_version); | |
1258 | ri.mode = cpu_to_jemode(JFFS2_F_I_MODE(f)); | |
1259 | ri.uid = cpu_to_je16(JFFS2_F_I_UID(f)); | |
1260 | ri.gid = cpu_to_je16(JFFS2_F_I_GID(f)); | |
1261 | ri.isize = cpu_to_je32(JFFS2_F_I_SIZE(f)); | |
1262 | ri.atime = cpu_to_je32(JFFS2_F_I_ATIME(f)); | |
1263 | ri.ctime = cpu_to_je32(JFFS2_F_I_CTIME(f)); | |
1264 | ri.mtime = cpu_to_je32(JFFS2_F_I_MTIME(f)); | |
1265 | ri.offset = cpu_to_je32(offset); | |
1266 | ri.csize = cpu_to_je32(cdatalen); | |
1267 | ri.dsize = cpu_to_je32(datalen); | |
1268 | ri.compr = comprtype & 0xff; | |
1269 | ri.usercompr = (comprtype >> 8) & 0xff; | |
1270 | ri.node_crc = cpu_to_je32(crc32(0, &ri, sizeof(ri)-8)); | |
1271 | ri.data_crc = cpu_to_je32(crc32(0, comprbuf, cdatalen)); | |
182ec4ee | 1272 | |
9fe4854c | 1273 | new_fn = jffs2_write_dnode(c, f, &ri, comprbuf, cdatalen, ALLOC_GC); |
1da177e4 LT |
1274 | |
1275 | jffs2_free_comprbuf(comprbuf, writebuf); | |
1276 | ||
1277 | if (IS_ERR(new_fn)) { | |
1278 | printk(KERN_WARNING "Error writing new dnode: %ld\n", PTR_ERR(new_fn)); | |
1279 | ret = PTR_ERR(new_fn); | |
1280 | break; | |
1281 | } | |
1282 | ret = jffs2_add_full_dnode_to_inode(c, f, new_fn); | |
1283 | offset += datalen; | |
1284 | if (f->metadata) { | |
1285 | jffs2_mark_node_obsolete(c, f->metadata->raw); | |
1286 | jffs2_free_full_dnode(f->metadata); | |
1287 | f->metadata = NULL; | |
1288 | } | |
1289 | } | |
1290 | ||
1291 | jffs2_gc_release_page(c, pg_ptr, &pg); | |
1292 | return ret; | |
1293 | } |