[JFFS2] Handle dirents on the flash with embedded zero bytes in names.
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / fs / jffs2 / gc.c
1 /*
2 * JFFS2 -- Journalling Flash File System, Version 2.
3 *
4 * Copyright © 2001-2007 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 */
11
12 #include <linux/kernel.h>
13 #include <linux/mtd/mtd.h>
14 #include <linux/slab.h>
15 #include <linux/pagemap.h>
16 #include <linux/crc32.h>
17 #include <linux/compiler.h>
18 #include <linux/stat.h>
19 #include "nodelist.h"
20 #include "compr.h"
21
22 static int jffs2_garbage_collect_pristine(struct jffs2_sb_info *c,
23 struct jffs2_inode_cache *ic,
24 struct jffs2_raw_node_ref *raw);
25 static int jffs2_garbage_collect_metadata(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
26 struct jffs2_inode_info *f, struct jffs2_full_dnode *fd);
27 static int jffs2_garbage_collect_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
28 struct jffs2_inode_info *f, struct jffs2_full_dirent *fd);
29 static int jffs2_garbage_collect_deletion_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_hole(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
32 struct jffs2_inode_info *f, struct jffs2_full_dnode *fn,
33 uint32_t start, uint32_t end);
34 static int jffs2_garbage_collect_dnode(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
35 struct jffs2_inode_info *f, struct jffs2_full_dnode *fn,
36 uint32_t start, uint32_t end);
37 static int jffs2_garbage_collect_live(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
38 struct jffs2_raw_node_ref *raw, struct jffs2_inode_info *f);
39
40 /* Called with erase_completion_lock held */
41 static struct jffs2_eraseblock *jffs2_find_gc_block(struct jffs2_sb_info *c)
42 {
43 struct jffs2_eraseblock *ret;
44 struct list_head *nextlist = NULL;
45 int n = jiffies % 128;
46
47 /* Pick an eraseblock to garbage collect next. This is where we'll
48 put the clever wear-levelling algorithms. Eventually. */
49 /* We possibly want to favour the dirtier blocks more when the
50 number of free blocks is low. */
51 again:
52 if (!list_empty(&c->bad_used_list) && c->nr_free_blocks > c->resv_blocks_gcbad) {
53 D1(printk(KERN_DEBUG "Picking block from bad_used_list to GC next\n"));
54 nextlist = &c->bad_used_list;
55 } else if (n < 50 && !list_empty(&c->erasable_list)) {
56 /* Note that most of them will have gone directly to be erased.
57 So don't favour the erasable_list _too_ much. */
58 D1(printk(KERN_DEBUG "Picking block from erasable_list to GC next\n"));
59 nextlist = &c->erasable_list;
60 } else if (n < 110 && !list_empty(&c->very_dirty_list)) {
61 /* Most of the time, pick one off the very_dirty list */
62 D1(printk(KERN_DEBUG "Picking block from very_dirty_list to GC next\n"));
63 nextlist = &c->very_dirty_list;
64 } else if (n < 126 && !list_empty(&c->dirty_list)) {
65 D1(printk(KERN_DEBUG "Picking block from dirty_list to GC next\n"));
66 nextlist = &c->dirty_list;
67 } else if (!list_empty(&c->clean_list)) {
68 D1(printk(KERN_DEBUG "Picking block from clean_list to GC next\n"));
69 nextlist = &c->clean_list;
70 } else if (!list_empty(&c->dirty_list)) {
71 D1(printk(KERN_DEBUG "Picking block from dirty_list to GC next (clean_list was empty)\n"));
72
73 nextlist = &c->dirty_list;
74 } else if (!list_empty(&c->very_dirty_list)) {
75 D1(printk(KERN_DEBUG "Picking block from very_dirty_list to GC next (clean_list and dirty_list were empty)\n"));
76 nextlist = &c->very_dirty_list;
77 } else if (!list_empty(&c->erasable_list)) {
78 D1(printk(KERN_DEBUG "Picking block from erasable_list to GC next (clean_list and {very_,}dirty_list were empty)\n"));
79
80 nextlist = &c->erasable_list;
81 } else if (!list_empty(&c->erasable_pending_wbuf_list)) {
82 /* There are blocks are wating for the wbuf sync */
83 D1(printk(KERN_DEBUG "Synching wbuf in order to reuse erasable_pending_wbuf_list blocks\n"));
84 spin_unlock(&c->erase_completion_lock);
85 jffs2_flush_wbuf_pad(c);
86 spin_lock(&c->erase_completion_lock);
87 goto again;
88 } else {
89 /* Eep. All were empty */
90 D1(printk(KERN_NOTICE "jffs2: No clean, dirty _or_ erasable blocks to GC from! Where are they all?\n"));
91 return NULL;
92 }
93
94 ret = list_entry(nextlist->next, struct jffs2_eraseblock, list);
95 list_del(&ret->list);
96 c->gcblock = ret;
97 ret->gc_node = ret->first_node;
98 if (!ret->gc_node) {
99 printk(KERN_WARNING "Eep. ret->gc_node for block at 0x%08x is NULL\n", ret->offset);
100 BUG();
101 }
102
103 /* Have we accidentally picked a clean block with wasted space ? */
104 if (ret->wasted_size) {
105 D1(printk(KERN_DEBUG "Converting wasted_size %08x to dirty_size\n", ret->wasted_size));
106 ret->dirty_size += ret->wasted_size;
107 c->wasted_size -= ret->wasted_size;
108 c->dirty_size += ret->wasted_size;
109 ret->wasted_size = 0;
110 }
111
112 return ret;
113 }
114
115 /* jffs2_garbage_collect_pass
116 * Make a single attempt to progress GC. Move one node, and possibly
117 * start erasing one eraseblock.
118 */
119 int jffs2_garbage_collect_pass(struct jffs2_sb_info *c)
120 {
121 struct jffs2_inode_info *f;
122 struct jffs2_inode_cache *ic;
123 struct jffs2_eraseblock *jeb;
124 struct jffs2_raw_node_ref *raw;
125 uint32_t gcblock_dirty;
126 int ret = 0, inum, nlink;
127 int xattr = 0;
128
129 if (down_interruptible(&c->alloc_sem))
130 return -EINTR;
131
132 for (;;) {
133 spin_lock(&c->erase_completion_lock);
134 if (!c->unchecked_size)
135 break;
136
137 /* We can't start doing GC yet. We haven't finished checking
138 the node CRCs etc. Do it now. */
139
140 /* checked_ino is protected by the alloc_sem */
141 if (c->checked_ino > c->highest_ino && xattr) {
142 printk(KERN_CRIT "Checked all inodes but still 0x%x bytes of unchecked space?\n",
143 c->unchecked_size);
144 jffs2_dbg_dump_block_lists_nolock(c);
145 spin_unlock(&c->erase_completion_lock);
146 up(&c->alloc_sem);
147 return -ENOSPC;
148 }
149
150 spin_unlock(&c->erase_completion_lock);
151
152 if (!xattr)
153 xattr = jffs2_verify_xattr(c);
154
155 spin_lock(&c->inocache_lock);
156
157 ic = jffs2_get_ino_cache(c, c->checked_ino++);
158
159 if (!ic) {
160 spin_unlock(&c->inocache_lock);
161 continue;
162 }
163
164 if (!ic->nlink) {
165 D1(printk(KERN_DEBUG "Skipping check of ino #%d with nlink zero\n",
166 ic->ino));
167 spin_unlock(&c->inocache_lock);
168 jffs2_xattr_delete_inode(c, ic);
169 continue;
170 }
171 switch(ic->state) {
172 case INO_STATE_CHECKEDABSENT:
173 case INO_STATE_PRESENT:
174 D1(printk(KERN_DEBUG "Skipping ino #%u already checked\n", ic->ino));
175 spin_unlock(&c->inocache_lock);
176 continue;
177
178 case INO_STATE_GC:
179 case INO_STATE_CHECKING:
180 printk(KERN_WARNING "Inode #%u is in state %d during CRC check phase!\n", ic->ino, ic->state);
181 spin_unlock(&c->inocache_lock);
182 BUG();
183
184 case INO_STATE_READING:
185 /* We need to wait for it to finish, lest we move on
186 and trigger the BUG() above while we haven't yet
187 finished checking all its nodes */
188 D1(printk(KERN_DEBUG "Waiting for ino #%u to finish reading\n", ic->ino));
189 /* We need to come back again for the _same_ inode. We've
190 made no progress in this case, but that should be OK */
191 c->checked_ino--;
192
193 up(&c->alloc_sem);
194 sleep_on_spinunlock(&c->inocache_wq, &c->inocache_lock);
195 return 0;
196
197 default:
198 BUG();
199
200 case INO_STATE_UNCHECKED:
201 ;
202 }
203 ic->state = INO_STATE_CHECKING;
204 spin_unlock(&c->inocache_lock);
205
206 D1(printk(KERN_DEBUG "jffs2_garbage_collect_pass() triggering inode scan of ino#%u\n", ic->ino));
207
208 ret = jffs2_do_crccheck_inode(c, ic);
209 if (ret)
210 printk(KERN_WARNING "Returned error for crccheck of ino #%u. Expect badness...\n", ic->ino);
211
212 jffs2_set_inocache_state(c, ic, INO_STATE_CHECKEDABSENT);
213 up(&c->alloc_sem);
214 return ret;
215 }
216
217 /* First, work out which block we're garbage-collecting */
218 jeb = c->gcblock;
219
220 if (!jeb)
221 jeb = jffs2_find_gc_block(c);
222
223 if (!jeb) {
224 D1 (printk(KERN_NOTICE "jffs2: Couldn't find erase block to garbage collect!\n"));
225 spin_unlock(&c->erase_completion_lock);
226 up(&c->alloc_sem);
227 return -EIO;
228 }
229
230 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));
231 D1(if (c->nextblock)
232 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));
233
234 if (!jeb->used_size) {
235 up(&c->alloc_sem);
236 goto eraseit;
237 }
238
239 raw = jeb->gc_node;
240 gcblock_dirty = jeb->dirty_size;
241
242 while(ref_obsolete(raw)) {
243 D1(printk(KERN_DEBUG "Node at 0x%08x is obsolete... skipping\n", ref_offset(raw)));
244 raw = ref_next(raw);
245 if (unlikely(!raw)) {
246 printk(KERN_WARNING "eep. End of raw list while still supposedly nodes to GC\n");
247 printk(KERN_WARNING "erase block at 0x%08x. free_size 0x%08x, dirty_size 0x%08x, used_size 0x%08x\n",
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 */
261 spin_unlock(&c->erase_completion_lock);
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 }
269 up(&c->alloc_sem);
270 goto eraseit_lock;
271 }
272
273 ic = jffs2_raw_ref_to_ic(raw);
274
275 #ifdef CONFIG_JFFS2_FS_XATTR
276 /* When 'ic' refers xattr_datum/xattr_ref, this node is GCed as xattr.
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) {
280 spin_unlock(&c->erase_completion_lock);
281
282 if (ic->class == RAWNODE_CLASS_XATTR_DATUM) {
283 ret = jffs2_garbage_collect_xattr_datum(c, (struct jffs2_xattr_datum *)ic, raw);
284 } else {
285 ret = jffs2_garbage_collect_xattr_ref(c, (struct jffs2_xattr_ref *)ic, raw);
286 }
287 goto test_gcnode;
288 }
289 #endif
290
291 /* We need to hold the inocache. Either the erase_completion_lock or
292 the inocache_lock are sufficient; we trade down since the inocache_lock
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:
311 /* It's been checked, but it's not currently in-core.
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 {
318 D1(printk(KERN_DEBUG "Ino #%u is absent but node not REF_PRISTINE. Reading.\n",
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
331 by the time we actually start doing any GC, and since
332 we're holding the alloc_sem, no other garbage collection
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);
352 /* And because we dropped the alloc_sem we must start again from the
353 beginning. Ponder chance of livelock here -- we're returning success
354 without actually making any progress.
355
356 Q: What are the chances that the inode is back in INO_STATE_READING
357 again by the time we next enter this function? And that this happens
358 enough times to cause a real delay?
359
360 A: Small enough that I don't care :)
361 */
362 return 0;
363 }
364
365 /* OK. Now if the inode is in state INO_STATE_GC, we are going to copy the
366 node intact, and we don't have to muck about with the fragtree etc.
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 test_gcnode;
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 test_gcnode:
413 if (jeb->dirty_size == gcblock_dirty && !ref_obsolete(jeb->gc_node)) {
414 /* Eep. This really should never happen. GC is broken */
415 printk(KERN_ERR "Error garbage collecting node at %08x!\n", ref_offset(jeb->gc_node));
416 ret = -ENOSPC;
417 } else if (ref_offset(jeb->gc_node) == 0x1c616bdc)
418 printk(KERN_ERR "Wheee. Correctly GC'd node at %08x\n", ref_offset(jeb->gc_node));
419
420 release_sem:
421 up(&c->alloc_sem);
422
423 eraseit_lock:
424 /* If we've finished this block, start it erasing */
425 spin_lock(&c->erase_completion_lock);
426
427 eraseit:
428 if (c->gcblock && !c->gcblock->used_size) {
429 D1(printk(KERN_DEBUG "Block at 0x%08x completely obsoleted by GC. Moving to erase_pending_list\n", c->gcblock->offset));
430 /* We're GC'ing an empty block? */
431 list_add_tail(&c->gcblock->list, &c->erase_pending_list);
432 c->gcblock = NULL;
433 c->nr_erasing_blocks++;
434 jffs2_erase_pending_trigger(c);
435 }
436 spin_unlock(&c->erase_completion_lock);
437
438 return ret;
439 }
440
441 static int jffs2_garbage_collect_live(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
442 struct jffs2_raw_node_ref *raw, struct jffs2_inode_info *f)
443 {
444 struct jffs2_node_frag *frag;
445 struct jffs2_full_dnode *fn = NULL;
446 struct jffs2_full_dirent *fd;
447 uint32_t start = 0, end = 0, nrfrags = 0;
448 int ret = 0;
449
450 down(&f->sem);
451
452 /* Now we have the lock for this inode. Check that it's still the one at the head
453 of the list. */
454
455 spin_lock(&c->erase_completion_lock);
456
457 if (c->gcblock != jeb) {
458 spin_unlock(&c->erase_completion_lock);
459 D1(printk(KERN_DEBUG "GC block is no longer gcblock. Restart\n"));
460 goto upnout;
461 }
462 if (ref_obsolete(raw)) {
463 spin_unlock(&c->erase_completion_lock);
464 D1(printk(KERN_DEBUG "node to be GC'd was obsoleted in the meantime.\n"));
465 /* They'll call again */
466 goto upnout;
467 }
468 spin_unlock(&c->erase_completion_lock);
469
470 /* OK. Looks safe. And nobody can get us now because we have the semaphore. Move the block */
471 if (f->metadata && f->metadata->raw == raw) {
472 fn = f->metadata;
473 ret = jffs2_garbage_collect_metadata(c, jeb, f, fn);
474 goto upnout;
475 }
476
477 /* FIXME. Read node and do lookup? */
478 for (frag = frag_first(&f->fragtree); frag; frag = frag_next(frag)) {
479 if (frag->node && frag->node->raw == raw) {
480 fn = frag->node;
481 end = frag->ofs + frag->size;
482 if (!nrfrags++)
483 start = frag->ofs;
484 if (nrfrags == frag->node->frags)
485 break; /* We've found them all */
486 }
487 }
488 if (fn) {
489 if (ref_flags(raw) == REF_PRISTINE) {
490 ret = jffs2_garbage_collect_pristine(c, f->inocache, raw);
491 if (!ret) {
492 /* Urgh. Return it sensibly. */
493 frag->node->raw = f->inocache->nodes;
494 }
495 if (ret != -EBADFD)
496 goto upnout;
497 }
498 /* We found a datanode. Do the GC */
499 if((start >> PAGE_CACHE_SHIFT) < ((end-1) >> PAGE_CACHE_SHIFT)) {
500 /* It crosses a page boundary. Therefore, it must be a hole. */
501 ret = jffs2_garbage_collect_hole(c, jeb, f, fn, start, end);
502 } else {
503 /* It could still be a hole. But we GC the page this way anyway */
504 ret = jffs2_garbage_collect_dnode(c, jeb, f, fn, start, end);
505 }
506 goto upnout;
507 }
508
509 /* Wasn't a dnode. Try dirent */
510 for (fd = f->dents; fd; fd=fd->next) {
511 if (fd->raw == raw)
512 break;
513 }
514
515 if (fd && fd->ino) {
516 ret = jffs2_garbage_collect_dirent(c, jeb, f, fd);
517 } else if (fd) {
518 ret = jffs2_garbage_collect_deletion_dirent(c, jeb, f, fd);
519 } else {
520 printk(KERN_WARNING "Raw node at 0x%08x wasn't in node lists for ino #%u\n",
521 ref_offset(raw), f->inocache->ino);
522 if (ref_obsolete(raw)) {
523 printk(KERN_WARNING "But it's obsolete so we don't mind too much\n");
524 } else {
525 jffs2_dbg_dump_node(c, ref_offset(raw));
526 BUG();
527 }
528 }
529 upnout:
530 up(&f->sem);
531
532 return ret;
533 }
534
535 static int jffs2_garbage_collect_pristine(struct jffs2_sb_info *c,
536 struct jffs2_inode_cache *ic,
537 struct jffs2_raw_node_ref *raw)
538 {
539 union jffs2_node_union *node;
540 size_t retlen;
541 int ret;
542 uint32_t phys_ofs, alloclen;
543 uint32_t crc, rawlen;
544 int retried = 0;
545
546 D1(printk(KERN_DEBUG "Going to GC REF_PRISTINE node at 0x%08x\n", ref_offset(raw)));
547
548 alloclen = rawlen = ref_totlen(c, c->gcblock, raw);
549
550 /* Ask for a small amount of space (or the totlen if smaller) because we
551 don't want to force wastage of the end of a block if splitting would
552 work. */
553 if (ic && alloclen > sizeof(struct jffs2_raw_inode) + JFFS2_MIN_DATA_LEN)
554 alloclen = sizeof(struct jffs2_raw_inode) + JFFS2_MIN_DATA_LEN;
555
556 ret = jffs2_reserve_space_gc(c, alloclen, &alloclen, rawlen);
557 /* 'rawlen' is not the exact summary size; it is only an upper estimation */
558
559 if (ret)
560 return ret;
561
562 if (alloclen < rawlen) {
563 /* Doesn't fit untouched. We'll go the old route and split it */
564 return -EBADFD;
565 }
566
567 node = kmalloc(rawlen, GFP_KERNEL);
568 if (!node)
569 return -ENOMEM;
570
571 ret = jffs2_flash_read(c, ref_offset(raw), rawlen, &retlen, (char *)node);
572 if (!ret && retlen != rawlen)
573 ret = -EIO;
574 if (ret)
575 goto out_node;
576
577 crc = crc32(0, node, sizeof(struct jffs2_unknown_node)-4);
578 if (je32_to_cpu(node->u.hdr_crc) != crc) {
579 printk(KERN_WARNING "Header CRC failed on REF_PRISTINE node at 0x%08x: Read 0x%08x, calculated 0x%08x\n",
580 ref_offset(raw), je32_to_cpu(node->u.hdr_crc), crc);
581 goto bail;
582 }
583
584 switch(je16_to_cpu(node->u.nodetype)) {
585 case JFFS2_NODETYPE_INODE:
586 crc = crc32(0, node, sizeof(node->i)-8);
587 if (je32_to_cpu(node->i.node_crc) != crc) {
588 printk(KERN_WARNING "Node 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.node_crc), crc);
590 goto bail;
591 }
592
593 if (je32_to_cpu(node->i.dsize)) {
594 crc = crc32(0, node->i.data, je32_to_cpu(node->i.csize));
595 if (je32_to_cpu(node->i.data_crc) != crc) {
596 printk(KERN_WARNING "Data CRC failed on REF_PRISTINE data node at 0x%08x: Read 0x%08x, calculated 0x%08x\n",
597 ref_offset(raw), je32_to_cpu(node->i.data_crc), crc);
598 goto bail;
599 }
600 }
601 break;
602
603 case JFFS2_NODETYPE_DIRENT:
604 crc = crc32(0, node, sizeof(node->d)-8);
605 if (je32_to_cpu(node->d.node_crc) != crc) {
606 printk(KERN_WARNING "Node CRC failed on REF_PRISTINE dirent node at 0x%08x: Read 0x%08x, calculated 0x%08x\n",
607 ref_offset(raw), je32_to_cpu(node->d.node_crc), crc);
608 goto bail;
609 }
610
611 if (strnlen(node->d.name, node->d.nsize) != node->d.nsize) {
612 printk(KERN_WARNING "Name in dirent node at 0x%08x contains zeroes\n", ref_offset(raw));
613 goto bail;
614 }
615
616 if (node->d.nsize) {
617 crc = crc32(0, node->d.name, node->d.nsize);
618 if (je32_to_cpu(node->d.name_crc) != crc) {
619 printk(KERN_WARNING "Name CRC failed on REF_PRISTINE dirent node at 0x%08x: Read 0x%08x, calculated 0x%08x\n",
620 ref_offset(raw), je32_to_cpu(node->d.name_crc), crc);
621 goto bail;
622 }
623 }
624 break;
625 default:
626 /* If it's inode-less, we don't _know_ what it is. Just copy it intact */
627 if (ic) {
628 printk(KERN_WARNING "Unknown node type for REF_PRISTINE node at 0x%08x: 0x%04x\n",
629 ref_offset(raw), je16_to_cpu(node->u.nodetype));
630 goto bail;
631 }
632 }
633
634 /* OK, all the CRCs are good; this node can just be copied as-is. */
635 retry:
636 phys_ofs = write_ofs(c);
637
638 ret = jffs2_flash_write(c, phys_ofs, rawlen, &retlen, (char *)node);
639
640 if (ret || (retlen != rawlen)) {
641 printk(KERN_NOTICE "Write of %d bytes at 0x%08x failed. returned %d, retlen %zd\n",
642 rawlen, phys_ofs, ret, retlen);
643 if (retlen) {
644 jffs2_add_physical_node_ref(c, phys_ofs | REF_OBSOLETE, rawlen, NULL);
645 } else {
646 printk(KERN_NOTICE "Not marking the space at 0x%08x as dirty because the flash driver returned retlen zero\n", phys_ofs);
647 }
648 if (!retried) {
649 /* Try to reallocate space and retry */
650 uint32_t dummy;
651 struct jffs2_eraseblock *jeb = &c->blocks[phys_ofs / c->sector_size];
652
653 retried = 1;
654
655 D1(printk(KERN_DEBUG "Retrying failed write of REF_PRISTINE node.\n"));
656
657 jffs2_dbg_acct_sanity_check(c,jeb);
658 jffs2_dbg_acct_paranoia_check(c, jeb);
659
660 ret = jffs2_reserve_space_gc(c, rawlen, &dummy, rawlen);
661 /* this is not the exact summary size of it,
662 it is only an upper estimation */
663
664 if (!ret) {
665 D1(printk(KERN_DEBUG "Allocated space at 0x%08x to retry failed write.\n", phys_ofs));
666
667 jffs2_dbg_acct_sanity_check(c,jeb);
668 jffs2_dbg_acct_paranoia_check(c, jeb);
669
670 goto retry;
671 }
672 D1(printk(KERN_DEBUG "Failed to allocate space to retry failed write: %d!\n", ret));
673 }
674
675 if (!ret)
676 ret = -EIO;
677 goto out_node;
678 }
679 jffs2_add_physical_node_ref(c, phys_ofs | REF_PRISTINE, rawlen, ic);
680
681 jffs2_mark_node_obsolete(c, raw);
682 D1(printk(KERN_DEBUG "WHEEE! GC REF_PRISTINE node at 0x%08x succeeded\n", ref_offset(raw)));
683
684 out_node:
685 kfree(node);
686 return ret;
687 bail:
688 ret = -EBADFD;
689 goto out_node;
690 }
691
692 static int jffs2_garbage_collect_metadata(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
693 struct jffs2_inode_info *f, struct jffs2_full_dnode *fn)
694 {
695 struct jffs2_full_dnode *new_fn;
696 struct jffs2_raw_inode ri;
697 struct jffs2_node_frag *last_frag;
698 union jffs2_device_node dev;
699 char *mdata = NULL, mdatalen = 0;
700 uint32_t alloclen, ilen;
701 int ret;
702
703 if (S_ISBLK(JFFS2_F_I_MODE(f)) ||
704 S_ISCHR(JFFS2_F_I_MODE(f)) ) {
705 /* For these, we don't actually need to read the old node */
706 mdatalen = jffs2_encode_dev(&dev, JFFS2_F_I_RDEV(f));
707 mdata = (char *)&dev;
708 D1(printk(KERN_DEBUG "jffs2_garbage_collect_metadata(): Writing %d bytes of kdev_t\n", mdatalen));
709 } else if (S_ISLNK(JFFS2_F_I_MODE(f))) {
710 mdatalen = fn->size;
711 mdata = kmalloc(fn->size, GFP_KERNEL);
712 if (!mdata) {
713 printk(KERN_WARNING "kmalloc of mdata failed in jffs2_garbage_collect_metadata()\n");
714 return -ENOMEM;
715 }
716 ret = jffs2_read_dnode(c, f, fn, mdata, 0, mdatalen);
717 if (ret) {
718 printk(KERN_WARNING "read of old metadata failed in jffs2_garbage_collect_metadata(): %d\n", ret);
719 kfree(mdata);
720 return ret;
721 }
722 D1(printk(KERN_DEBUG "jffs2_garbage_collect_metadata(): Writing %d bites of symlink target\n", mdatalen));
723
724 }
725
726 ret = jffs2_reserve_space_gc(c, sizeof(ri) + mdatalen, &alloclen,
727 JFFS2_SUMMARY_INODE_SIZE);
728 if (ret) {
729 printk(KERN_WARNING "jffs2_reserve_space_gc of %zd bytes for garbage_collect_metadata failed: %d\n",
730 sizeof(ri)+ mdatalen, ret);
731 goto out;
732 }
733
734 last_frag = frag_last(&f->fragtree);
735 if (last_frag)
736 /* Fetch the inode length from the fragtree rather then
737 * from i_size since i_size may have not been updated yet */
738 ilen = last_frag->ofs + last_frag->size;
739 else
740 ilen = JFFS2_F_I_SIZE(f);
741
742 memset(&ri, 0, sizeof(ri));
743 ri.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK);
744 ri.nodetype = cpu_to_je16(JFFS2_NODETYPE_INODE);
745 ri.totlen = cpu_to_je32(sizeof(ri) + mdatalen);
746 ri.hdr_crc = cpu_to_je32(crc32(0, &ri, sizeof(struct jffs2_unknown_node)-4));
747
748 ri.ino = cpu_to_je32(f->inocache->ino);
749 ri.version = cpu_to_je32(++f->highest_version);
750 ri.mode = cpu_to_jemode(JFFS2_F_I_MODE(f));
751 ri.uid = cpu_to_je16(JFFS2_F_I_UID(f));
752 ri.gid = cpu_to_je16(JFFS2_F_I_GID(f));
753 ri.isize = cpu_to_je32(ilen);
754 ri.atime = cpu_to_je32(JFFS2_F_I_ATIME(f));
755 ri.ctime = cpu_to_je32(JFFS2_F_I_CTIME(f));
756 ri.mtime = cpu_to_je32(JFFS2_F_I_MTIME(f));
757 ri.offset = cpu_to_je32(0);
758 ri.csize = cpu_to_je32(mdatalen);
759 ri.dsize = cpu_to_je32(mdatalen);
760 ri.compr = JFFS2_COMPR_NONE;
761 ri.node_crc = cpu_to_je32(crc32(0, &ri, sizeof(ri)-8));
762 ri.data_crc = cpu_to_je32(crc32(0, mdata, mdatalen));
763
764 new_fn = jffs2_write_dnode(c, f, &ri, mdata, mdatalen, ALLOC_GC);
765
766 if (IS_ERR(new_fn)) {
767 printk(KERN_WARNING "Error writing new dnode: %ld\n", PTR_ERR(new_fn));
768 ret = PTR_ERR(new_fn);
769 goto out;
770 }
771 jffs2_mark_node_obsolete(c, fn->raw);
772 jffs2_free_full_dnode(fn);
773 f->metadata = new_fn;
774 out:
775 if (S_ISLNK(JFFS2_F_I_MODE(f)))
776 kfree(mdata);
777 return ret;
778 }
779
780 static int jffs2_garbage_collect_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
781 struct jffs2_inode_info *f, struct jffs2_full_dirent *fd)
782 {
783 struct jffs2_full_dirent *new_fd;
784 struct jffs2_raw_dirent rd;
785 uint32_t alloclen;
786 int ret;
787
788 rd.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK);
789 rd.nodetype = cpu_to_je16(JFFS2_NODETYPE_DIRENT);
790 rd.nsize = strlen(fd->name);
791 rd.totlen = cpu_to_je32(sizeof(rd) + rd.nsize);
792 rd.hdr_crc = cpu_to_je32(crc32(0, &rd, sizeof(struct jffs2_unknown_node)-4));
793
794 rd.pino = cpu_to_je32(f->inocache->ino);
795 rd.version = cpu_to_je32(++f->highest_version);
796 rd.ino = cpu_to_je32(fd->ino);
797 /* If the times on this inode were set by explicit utime() they can be different,
798 so refrain from splatting them. */
799 if (JFFS2_F_I_MTIME(f) == JFFS2_F_I_CTIME(f))
800 rd.mctime = cpu_to_je32(JFFS2_F_I_MTIME(f));
801 else
802 rd.mctime = cpu_to_je32(0);
803 rd.type = fd->type;
804 rd.node_crc = cpu_to_je32(crc32(0, &rd, sizeof(rd)-8));
805 rd.name_crc = cpu_to_je32(crc32(0, fd->name, rd.nsize));
806
807 ret = jffs2_reserve_space_gc(c, sizeof(rd)+rd.nsize, &alloclen,
808 JFFS2_SUMMARY_DIRENT_SIZE(rd.nsize));
809 if (ret) {
810 printk(KERN_WARNING "jffs2_reserve_space_gc of %zd bytes for garbage_collect_dirent failed: %d\n",
811 sizeof(rd)+rd.nsize, ret);
812 return ret;
813 }
814 new_fd = jffs2_write_dirent(c, f, &rd, fd->name, rd.nsize, ALLOC_GC);
815
816 if (IS_ERR(new_fd)) {
817 printk(KERN_WARNING "jffs2_write_dirent in garbage_collect_dirent failed: %ld\n", PTR_ERR(new_fd));
818 return PTR_ERR(new_fd);
819 }
820 jffs2_add_fd_to_list(c, new_fd, &f->dents);
821 return 0;
822 }
823
824 static int jffs2_garbage_collect_deletion_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
825 struct jffs2_inode_info *f, struct jffs2_full_dirent *fd)
826 {
827 struct jffs2_full_dirent **fdp = &f->dents;
828 int found = 0;
829
830 /* On a medium where we can't actually mark nodes obsolete
831 pernamently, such as NAND flash, we need to work out
832 whether this deletion dirent is still needed to actively
833 delete a 'real' dirent with the same name that's still
834 somewhere else on the flash. */
835 if (!jffs2_can_mark_obsolete(c)) {
836 struct jffs2_raw_dirent *rd;
837 struct jffs2_raw_node_ref *raw;
838 int ret;
839 size_t retlen;
840 int name_len = strlen(fd->name);
841 uint32_t name_crc = crc32(0, fd->name, name_len);
842 uint32_t rawlen = ref_totlen(c, jeb, fd->raw);
843
844 rd = kmalloc(rawlen, GFP_KERNEL);
845 if (!rd)
846 return -ENOMEM;
847
848 /* Prevent the erase code from nicking the obsolete node refs while
849 we're looking at them. I really don't like this extra lock but
850 can't see any alternative. Suggestions on a postcard to... */
851 down(&c->erase_free_sem);
852
853 for (raw = f->inocache->nodes; raw != (void *)f->inocache; raw = raw->next_in_ino) {
854
855 cond_resched();
856
857 /* We only care about obsolete ones */
858 if (!(ref_obsolete(raw)))
859 continue;
860
861 /* Any dirent with the same name is going to have the same length... */
862 if (ref_totlen(c, NULL, raw) != rawlen)
863 continue;
864
865 /* Doesn't matter if there's one in the same erase block. We're going to
866 delete it too at the same time. */
867 if (SECTOR_ADDR(raw->flash_offset) == SECTOR_ADDR(fd->raw->flash_offset))
868 continue;
869
870 D1(printk(KERN_DEBUG "Check potential deletion dirent at %08x\n", ref_offset(raw)));
871
872 /* This is an obsolete node belonging to the same directory, and it's of the right
873 length. We need to take a closer look...*/
874 ret = jffs2_flash_read(c, ref_offset(raw), rawlen, &retlen, (char *)rd);
875 if (ret) {
876 printk(KERN_WARNING "jffs2_g_c_deletion_dirent(): Read error (%d) reading obsolete node at %08x\n", ret, ref_offset(raw));
877 /* If we can't read it, we don't need to continue to obsolete it. Continue */
878 continue;
879 }
880 if (retlen != rawlen) {
881 printk(KERN_WARNING "jffs2_g_c_deletion_dirent(): Short read (%zd not %u) reading header from obsolete node at %08x\n",
882 retlen, rawlen, ref_offset(raw));
883 continue;
884 }
885
886 if (je16_to_cpu(rd->nodetype) != JFFS2_NODETYPE_DIRENT)
887 continue;
888
889 /* If the name CRC doesn't match, skip */
890 if (je32_to_cpu(rd->name_crc) != name_crc)
891 continue;
892
893 /* If the name length doesn't match, or it's another deletion dirent, skip */
894 if (rd->nsize != name_len || !je32_to_cpu(rd->ino))
895 continue;
896
897 /* OK, check the actual name now */
898 if (memcmp(rd->name, fd->name, name_len))
899 continue;
900
901 /* OK. The name really does match. There really is still an older node on
902 the flash which our deletion dirent obsoletes. So we have to write out
903 a new deletion dirent to replace it */
904 up(&c->erase_free_sem);
905
906 D1(printk(KERN_DEBUG "Deletion dirent at %08x still obsoletes real dirent \"%s\" at %08x for ino #%u\n",
907 ref_offset(fd->raw), fd->name, ref_offset(raw), je32_to_cpu(rd->ino)));
908 kfree(rd);
909
910 return jffs2_garbage_collect_dirent(c, jeb, f, fd);
911 }
912
913 up(&c->erase_free_sem);
914 kfree(rd);
915 }
916
917 /* FIXME: If we're deleting a dirent which contains the current mtime and ctime,
918 we should update the metadata node with those times accordingly */
919
920 /* No need for it any more. Just mark it obsolete and remove it from the list */
921 while (*fdp) {
922 if ((*fdp) == fd) {
923 found = 1;
924 *fdp = fd->next;
925 break;
926 }
927 fdp = &(*fdp)->next;
928 }
929 if (!found) {
930 printk(KERN_WARNING "Deletion dirent \"%s\" not found in list for ino #%u\n", fd->name, f->inocache->ino);
931 }
932 jffs2_mark_node_obsolete(c, fd->raw);
933 jffs2_free_full_dirent(fd);
934 return 0;
935 }
936
937 static int jffs2_garbage_collect_hole(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
938 struct jffs2_inode_info *f, struct jffs2_full_dnode *fn,
939 uint32_t start, uint32_t end)
940 {
941 struct jffs2_raw_inode ri;
942 struct jffs2_node_frag *frag;
943 struct jffs2_full_dnode *new_fn;
944 uint32_t alloclen, ilen;
945 int ret;
946
947 D1(printk(KERN_DEBUG "Writing replacement hole node for ino #%u from offset 0x%x to 0x%x\n",
948 f->inocache->ino, start, end));
949
950 memset(&ri, 0, sizeof(ri));
951
952 if(fn->frags > 1) {
953 size_t readlen;
954 uint32_t crc;
955 /* It's partially obsoleted by a later write. So we have to
956 write it out again with the _same_ version as before */
957 ret = jffs2_flash_read(c, ref_offset(fn->raw), sizeof(ri), &readlen, (char *)&ri);
958 if (readlen != sizeof(ri) || ret) {
959 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);
960 goto fill;
961 }
962 if (je16_to_cpu(ri.nodetype) != JFFS2_NODETYPE_INODE) {
963 printk(KERN_WARNING "jffs2_garbage_collect_hole: Node at 0x%08x had node type 0x%04x instead of JFFS2_NODETYPE_INODE(0x%04x)\n",
964 ref_offset(fn->raw),
965 je16_to_cpu(ri.nodetype), JFFS2_NODETYPE_INODE);
966 return -EIO;
967 }
968 if (je32_to_cpu(ri.totlen) != sizeof(ri)) {
969 printk(KERN_WARNING "jffs2_garbage_collect_hole: Node at 0x%08x had totlen 0x%x instead of expected 0x%zx\n",
970 ref_offset(fn->raw),
971 je32_to_cpu(ri.totlen), sizeof(ri));
972 return -EIO;
973 }
974 crc = crc32(0, &ri, sizeof(ri)-8);
975 if (crc != je32_to_cpu(ri.node_crc)) {
976 printk(KERN_WARNING "jffs2_garbage_collect_hole: Node at 0x%08x had CRC 0x%08x which doesn't match calculated CRC 0x%08x\n",
977 ref_offset(fn->raw),
978 je32_to_cpu(ri.node_crc), crc);
979 /* FIXME: We could possibly deal with this by writing new holes for each frag */
980 printk(KERN_WARNING "Data in the range 0x%08x to 0x%08x of inode #%u will be lost\n",
981 start, end, f->inocache->ino);
982 goto fill;
983 }
984 if (ri.compr != JFFS2_COMPR_ZERO) {
985 printk(KERN_WARNING "jffs2_garbage_collect_hole: Node 0x%08x wasn't a hole node!\n", ref_offset(fn->raw));
986 printk(KERN_WARNING "Data in the range 0x%08x to 0x%08x of inode #%u will be lost\n",
987 start, end, f->inocache->ino);
988 goto fill;
989 }
990 } else {
991 fill:
992 ri.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK);
993 ri.nodetype = cpu_to_je16(JFFS2_NODETYPE_INODE);
994 ri.totlen = cpu_to_je32(sizeof(ri));
995 ri.hdr_crc = cpu_to_je32(crc32(0, &ri, sizeof(struct jffs2_unknown_node)-4));
996
997 ri.ino = cpu_to_je32(f->inocache->ino);
998 ri.version = cpu_to_je32(++f->highest_version);
999 ri.offset = cpu_to_je32(start);
1000 ri.dsize = cpu_to_je32(end - start);
1001 ri.csize = cpu_to_je32(0);
1002 ri.compr = JFFS2_COMPR_ZERO;
1003 }
1004
1005 frag = frag_last(&f->fragtree);
1006 if (frag)
1007 /* Fetch the inode length from the fragtree rather then
1008 * from i_size since i_size may have not been updated yet */
1009 ilen = frag->ofs + frag->size;
1010 else
1011 ilen = JFFS2_F_I_SIZE(f);
1012
1013 ri.mode = cpu_to_jemode(JFFS2_F_I_MODE(f));
1014 ri.uid = cpu_to_je16(JFFS2_F_I_UID(f));
1015 ri.gid = cpu_to_je16(JFFS2_F_I_GID(f));
1016 ri.isize = cpu_to_je32(ilen);
1017 ri.atime = cpu_to_je32(JFFS2_F_I_ATIME(f));
1018 ri.ctime = cpu_to_je32(JFFS2_F_I_CTIME(f));
1019 ri.mtime = cpu_to_je32(JFFS2_F_I_MTIME(f));
1020 ri.data_crc = cpu_to_je32(0);
1021 ri.node_crc = cpu_to_je32(crc32(0, &ri, sizeof(ri)-8));
1022
1023 ret = jffs2_reserve_space_gc(c, sizeof(ri), &alloclen,
1024 JFFS2_SUMMARY_INODE_SIZE);
1025 if (ret) {
1026 printk(KERN_WARNING "jffs2_reserve_space_gc of %zd bytes for garbage_collect_hole failed: %d\n",
1027 sizeof(ri), ret);
1028 return ret;
1029 }
1030 new_fn = jffs2_write_dnode(c, f, &ri, NULL, 0, ALLOC_GC);
1031
1032 if (IS_ERR(new_fn)) {
1033 printk(KERN_WARNING "Error writing new hole node: %ld\n", PTR_ERR(new_fn));
1034 return PTR_ERR(new_fn);
1035 }
1036 if (je32_to_cpu(ri.version) == f->highest_version) {
1037 jffs2_add_full_dnode_to_inode(c, f, new_fn);
1038 if (f->metadata) {
1039 jffs2_mark_node_obsolete(c, f->metadata->raw);
1040 jffs2_free_full_dnode(f->metadata);
1041 f->metadata = NULL;
1042 }
1043 return 0;
1044 }
1045
1046 /*
1047 * We should only get here in the case where the node we are
1048 * replacing had more than one frag, so we kept the same version
1049 * number as before. (Except in case of error -- see 'goto fill;'
1050 * above.)
1051 */
1052 D1(if(unlikely(fn->frags <= 1)) {
1053 printk(KERN_WARNING "jffs2_garbage_collect_hole: Replacing fn with %d frag(s) but new ver %d != highest_version %d of ino #%d\n",
1054 fn->frags, je32_to_cpu(ri.version), f->highest_version,
1055 je32_to_cpu(ri.ino));
1056 });
1057
1058 /* This is a partially-overlapped hole node. Mark it REF_NORMAL not REF_PRISTINE */
1059 mark_ref_normal(new_fn->raw);
1060
1061 for (frag = jffs2_lookup_node_frag(&f->fragtree, fn->ofs);
1062 frag; frag = frag_next(frag)) {
1063 if (frag->ofs > fn->size + fn->ofs)
1064 break;
1065 if (frag->node == fn) {
1066 frag->node = new_fn;
1067 new_fn->frags++;
1068 fn->frags--;
1069 }
1070 }
1071 if (fn->frags) {
1072 printk(KERN_WARNING "jffs2_garbage_collect_hole: Old node still has frags!\n");
1073 BUG();
1074 }
1075 if (!new_fn->frags) {
1076 printk(KERN_WARNING "jffs2_garbage_collect_hole: New node has no frags!\n");
1077 BUG();
1078 }
1079
1080 jffs2_mark_node_obsolete(c, fn->raw);
1081 jffs2_free_full_dnode(fn);
1082
1083 return 0;
1084 }
1085
1086 static int jffs2_garbage_collect_dnode(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
1087 struct jffs2_inode_info *f, struct jffs2_full_dnode *fn,
1088 uint32_t start, uint32_t end)
1089 {
1090 struct jffs2_full_dnode *new_fn;
1091 struct jffs2_raw_inode ri;
1092 uint32_t alloclen, offset, orig_end, orig_start;
1093 int ret = 0;
1094 unsigned char *comprbuf = NULL, *writebuf;
1095 unsigned long pg;
1096 unsigned char *pg_ptr;
1097
1098 memset(&ri, 0, sizeof(ri));
1099
1100 D1(printk(KERN_DEBUG "Writing replacement dnode for ino #%u from offset 0x%x to 0x%x\n",
1101 f->inocache->ino, start, end));
1102
1103 orig_end = end;
1104 orig_start = start;
1105
1106 if (c->nr_free_blocks + c->nr_erasing_blocks > c->resv_blocks_gcmerge) {
1107 /* Attempt to do some merging. But only expand to cover logically
1108 adjacent frags if the block containing them is already considered
1109 to be dirty. Otherwise we end up with GC just going round in
1110 circles dirtying the nodes it already wrote out, especially
1111 on NAND where we have small eraseblocks and hence a much higher
1112 chance of nodes having to be split to cross boundaries. */
1113
1114 struct jffs2_node_frag *frag;
1115 uint32_t min, max;
1116
1117 min = start & ~(PAGE_CACHE_SIZE-1);
1118 max = min + PAGE_CACHE_SIZE;
1119
1120 frag = jffs2_lookup_node_frag(&f->fragtree, start);
1121
1122 /* BUG_ON(!frag) but that'll happen anyway... */
1123
1124 BUG_ON(frag->ofs != start);
1125
1126 /* First grow down... */
1127 while((frag = frag_prev(frag)) && frag->ofs >= min) {
1128
1129 /* If the previous frag doesn't even reach the beginning, there's
1130 excessive fragmentation. Just merge. */
1131 if (frag->ofs > min) {
1132 D1(printk(KERN_DEBUG "Expanding down to cover partial frag (0x%x-0x%x)\n",
1133 frag->ofs, frag->ofs+frag->size));
1134 start = frag->ofs;
1135 continue;
1136 }
1137 /* OK. This frag holds the first byte of the page. */
1138 if (!frag->node || !frag->node->raw) {
1139 D1(printk(KERN_DEBUG "First frag in page is hole (0x%x-0x%x). Not expanding down.\n",
1140 frag->ofs, frag->ofs+frag->size));
1141 break;
1142 } else {
1143
1144 /* OK, it's a frag which extends to the beginning of the page. Does it live
1145 in a block which is still considered clean? If so, don't obsolete it.
1146 If not, cover it anyway. */
1147
1148 struct jffs2_raw_node_ref *raw = frag->node->raw;
1149 struct jffs2_eraseblock *jeb;
1150
1151 jeb = &c->blocks[raw->flash_offset / c->sector_size];
1152
1153 if (jeb == c->gcblock) {
1154 D1(printk(KERN_DEBUG "Expanding down to cover frag (0x%x-0x%x) in gcblock at %08x\n",
1155 frag->ofs, frag->ofs+frag->size, ref_offset(raw)));
1156 start = frag->ofs;
1157 break;
1158 }
1159 if (!ISDIRTY(jeb->dirty_size + jeb->wasted_size)) {
1160 D1(printk(KERN_DEBUG "Not expanding down to cover frag (0x%x-0x%x) in clean block %08x\n",
1161 frag->ofs, frag->ofs+frag->size, jeb->offset));
1162 break;
1163 }
1164
1165 D1(printk(KERN_DEBUG "Expanding down to cover frag (0x%x-0x%x) in dirty block %08x\n",
1166 frag->ofs, frag->ofs+frag->size, jeb->offset));
1167 start = frag->ofs;
1168 break;
1169 }
1170 }
1171
1172 /* ... then up */
1173
1174 /* Find last frag which is actually part of the node we're to GC. */
1175 frag = jffs2_lookup_node_frag(&f->fragtree, end-1);
1176
1177 while((frag = frag_next(frag)) && frag->ofs+frag->size <= max) {
1178
1179 /* If the previous frag doesn't even reach the beginning, there's lots
1180 of fragmentation. Just merge. */
1181 if (frag->ofs+frag->size < max) {
1182 D1(printk(KERN_DEBUG "Expanding up to cover partial frag (0x%x-0x%x)\n",
1183 frag->ofs, frag->ofs+frag->size));
1184 end = frag->ofs + frag->size;
1185 continue;
1186 }
1187
1188 if (!frag->node || !frag->node->raw) {
1189 D1(printk(KERN_DEBUG "Last frag in page is hole (0x%x-0x%x). Not expanding up.\n",
1190 frag->ofs, frag->ofs+frag->size));
1191 break;
1192 } else {
1193
1194 /* OK, it's a frag which extends to the beginning of the page. Does it live
1195 in a block which is still considered clean? If so, don't obsolete it.
1196 If not, cover it anyway. */
1197
1198 struct jffs2_raw_node_ref *raw = frag->node->raw;
1199 struct jffs2_eraseblock *jeb;
1200
1201 jeb = &c->blocks[raw->flash_offset / c->sector_size];
1202
1203 if (jeb == c->gcblock) {
1204 D1(printk(KERN_DEBUG "Expanding up to cover frag (0x%x-0x%x) in gcblock at %08x\n",
1205 frag->ofs, frag->ofs+frag->size, ref_offset(raw)));
1206 end = frag->ofs + frag->size;
1207 break;
1208 }
1209 if (!ISDIRTY(jeb->dirty_size + jeb->wasted_size)) {
1210 D1(printk(KERN_DEBUG "Not expanding up to cover frag (0x%x-0x%x) in clean block %08x\n",
1211 frag->ofs, frag->ofs+frag->size, jeb->offset));
1212 break;
1213 }
1214
1215 D1(printk(KERN_DEBUG "Expanding up to cover frag (0x%x-0x%x) in dirty block %08x\n",
1216 frag->ofs, frag->ofs+frag->size, jeb->offset));
1217 end = frag->ofs + frag->size;
1218 break;
1219 }
1220 }
1221 D1(printk(KERN_DEBUG "Expanded dnode to write from (0x%x-0x%x) to (0x%x-0x%x)\n",
1222 orig_start, orig_end, start, end));
1223
1224 D1(BUG_ON(end > frag_last(&f->fragtree)->ofs + frag_last(&f->fragtree)->size));
1225 BUG_ON(end < orig_end);
1226 BUG_ON(start > orig_start);
1227 }
1228
1229 /* First, use readpage() to read the appropriate page into the page cache */
1230 /* Q: What happens if we actually try to GC the _same_ page for which commit_write()
1231 * triggered garbage collection in the first place?
1232 * A: I _think_ it's OK. read_cache_page shouldn't deadlock, we'll write out the
1233 * page OK. We'll actually write it out again in commit_write, which is a little
1234 * suboptimal, but at least we're correct.
1235 */
1236 pg_ptr = jffs2_gc_fetch_page(c, f, start, &pg);
1237
1238 if (IS_ERR(pg_ptr)) {
1239 printk(KERN_WARNING "read_cache_page() returned error: %ld\n", PTR_ERR(pg_ptr));
1240 return PTR_ERR(pg_ptr);
1241 }
1242
1243 offset = start;
1244 while(offset < orig_end) {
1245 uint32_t datalen;
1246 uint32_t cdatalen;
1247 uint16_t comprtype = JFFS2_COMPR_NONE;
1248
1249 ret = jffs2_reserve_space_gc(c, sizeof(ri) + JFFS2_MIN_DATA_LEN,
1250 &alloclen, JFFS2_SUMMARY_INODE_SIZE);
1251
1252 if (ret) {
1253 printk(KERN_WARNING "jffs2_reserve_space_gc of %zd bytes for garbage_collect_dnode failed: %d\n",
1254 sizeof(ri)+ JFFS2_MIN_DATA_LEN, ret);
1255 break;
1256 }
1257 cdatalen = min_t(uint32_t, alloclen - sizeof(ri), end - offset);
1258 datalen = end - offset;
1259
1260 writebuf = pg_ptr + (offset & (PAGE_CACHE_SIZE -1));
1261
1262 comprtype = jffs2_compress(c, f, writebuf, &comprbuf, &datalen, &cdatalen);
1263
1264 ri.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK);
1265 ri.nodetype = cpu_to_je16(JFFS2_NODETYPE_INODE);
1266 ri.totlen = cpu_to_je32(sizeof(ri) + cdatalen);
1267 ri.hdr_crc = cpu_to_je32(crc32(0, &ri, sizeof(struct jffs2_unknown_node)-4));
1268
1269 ri.ino = cpu_to_je32(f->inocache->ino);
1270 ri.version = cpu_to_je32(++f->highest_version);
1271 ri.mode = cpu_to_jemode(JFFS2_F_I_MODE(f));
1272 ri.uid = cpu_to_je16(JFFS2_F_I_UID(f));
1273 ri.gid = cpu_to_je16(JFFS2_F_I_GID(f));
1274 ri.isize = cpu_to_je32(JFFS2_F_I_SIZE(f));
1275 ri.atime = cpu_to_je32(JFFS2_F_I_ATIME(f));
1276 ri.ctime = cpu_to_je32(JFFS2_F_I_CTIME(f));
1277 ri.mtime = cpu_to_je32(JFFS2_F_I_MTIME(f));
1278 ri.offset = cpu_to_je32(offset);
1279 ri.csize = cpu_to_je32(cdatalen);
1280 ri.dsize = cpu_to_je32(datalen);
1281 ri.compr = comprtype & 0xff;
1282 ri.usercompr = (comprtype >> 8) & 0xff;
1283 ri.node_crc = cpu_to_je32(crc32(0, &ri, sizeof(ri)-8));
1284 ri.data_crc = cpu_to_je32(crc32(0, comprbuf, cdatalen));
1285
1286 new_fn = jffs2_write_dnode(c, f, &ri, comprbuf, cdatalen, ALLOC_GC);
1287
1288 jffs2_free_comprbuf(comprbuf, writebuf);
1289
1290 if (IS_ERR(new_fn)) {
1291 printk(KERN_WARNING "Error writing new dnode: %ld\n", PTR_ERR(new_fn));
1292 ret = PTR_ERR(new_fn);
1293 break;
1294 }
1295 ret = jffs2_add_full_dnode_to_inode(c, f, new_fn);
1296 offset += datalen;
1297 if (f->metadata) {
1298 jffs2_mark_node_obsolete(c, f->metadata->raw);
1299 jffs2_free_full_dnode(f->metadata);
1300 f->metadata = NULL;
1301 }
1302 }
1303
1304 jffs2_gc_release_page(c, pg_ptr, &pg);
1305 return ret;
1306 }