VServer 1.9.2 (patch-2.6.8.1-vs1.9.2.diff)
[linux-2.6.git] / fs / jffs2 / gc.c
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@redhat.com>
7  *
8  * For licensing information, see the file 'LICENCE' in this directory.
9  *
10  * $Id: gc.c,v 1.136 2004/05/27 19:06:09 gleixner Exp $
11  *
12  */
13
14 #include <linux/kernel.h>
15 #include <linux/mtd/mtd.h>
16 #include <linux/slab.h>
17 #include <linux/pagemap.h>
18 #include <linux/crc32.h>
19 #include <linux/compiler.h>
20 #include <linux/stat.h>
21 #include "nodelist.h"
22 #include "compr.h"
23
24 static int jffs2_garbage_collect_pristine(struct jffs2_sb_info *c, 
25                                           struct jffs2_inode_cache *ic,
26                                           struct jffs2_raw_node_ref *raw);
27 static int jffs2_garbage_collect_metadata(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, 
28                                         struct jffs2_inode_info *f, struct jffs2_full_dnode *fd);
29 static int jffs2_garbage_collect_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, 
30                                         struct jffs2_inode_info *f, struct jffs2_full_dirent *fd);
31 static int jffs2_garbage_collect_deletion_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, 
32                                         struct jffs2_inode_info *f, struct jffs2_full_dirent *fd);
33 static int jffs2_garbage_collect_hole(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
34                                       struct jffs2_inode_info *f, struct jffs2_full_dnode *fn,
35                                       uint32_t start, uint32_t end);
36 static int jffs2_garbage_collect_dnode(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
37                                        struct jffs2_inode_info *f, struct jffs2_full_dnode *fn,
38                                        uint32_t start, uint32_t end);
39 static int jffs2_garbage_collect_live(struct jffs2_sb_info *c,  struct jffs2_eraseblock *jeb,
40                                struct jffs2_raw_node_ref *raw, struct jffs2_inode_info *f);
41
42 /* Called with erase_completion_lock held */
43 static struct jffs2_eraseblock *jffs2_find_gc_block(struct jffs2_sb_info *c)
44 {
45         struct jffs2_eraseblock *ret;
46         struct list_head *nextlist = NULL;
47         int n = jiffies % 128;
48
49         /* Pick an eraseblock to garbage collect next. This is where we'll
50            put the clever wear-levelling algorithms. Eventually.  */
51         /* We possibly want to favour the dirtier blocks more when the
52            number of free blocks is low. */
53         if (!list_empty(&c->bad_used_list) && c->nr_free_blocks > c->resv_blocks_gcbad) {
54                 D1(printk(KERN_DEBUG "Picking block from bad_used_list to GC next\n"));
55                 nextlist = &c->bad_used_list;
56         } else if (n < 50 && !list_empty(&c->erasable_list)) {
57                 /* Note that most of them will have gone directly to be erased. 
58                    So don't favour the erasable_list _too_ much. */
59                 D1(printk(KERN_DEBUG "Picking block from erasable_list to GC next\n"));
60                 nextlist = &c->erasable_list;
61         } else if (n < 110 && !list_empty(&c->very_dirty_list)) {
62                 /* Most of the time, pick one off the very_dirty list */
63                 D1(printk(KERN_DEBUG "Picking block from very_dirty_list to GC next\n"));
64                 nextlist = &c->very_dirty_list;
65         } else if (n < 126 && !list_empty(&c->dirty_list)) {
66                 D1(printk(KERN_DEBUG "Picking block from dirty_list to GC next\n"));
67                 nextlist = &c->dirty_list;
68         } else if (!list_empty(&c->clean_list)) {
69                 D1(printk(KERN_DEBUG "Picking block from clean_list to GC next\n"));
70                 nextlist = &c->clean_list;
71         } else if (!list_empty(&c->dirty_list)) {
72                 D1(printk(KERN_DEBUG "Picking block from dirty_list to GC next (clean_list was empty)\n"));
73
74                 nextlist = &c->dirty_list;
75         } else if (!list_empty(&c->very_dirty_list)) {
76                 D1(printk(KERN_DEBUG "Picking block from very_dirty_list to GC next (clean_list and dirty_list were empty)\n"));
77                 nextlist = &c->very_dirty_list;
78         } else if (!list_empty(&c->erasable_list)) {
79                 D1(printk(KERN_DEBUG "Picking block from erasable_list to GC next (clean_list and {very_,}dirty_list were empty)\n"));
80
81                 nextlist = &c->erasable_list;
82         } else {
83                 /* Eep. All were empty */
84                 D1(printk(KERN_NOTICE "jffs2: No clean, dirty _or_ erasable blocks to GC from! Where are they all?\n"));
85                 return NULL;
86         }
87
88         ret = list_entry(nextlist->next, struct jffs2_eraseblock, list);
89         list_del(&ret->list);
90         c->gcblock = ret;
91         ret->gc_node = ret->first_node;
92         if (!ret->gc_node) {
93                 printk(KERN_WARNING "Eep. ret->gc_node for block at 0x%08x is NULL\n", ret->offset);
94                 BUG();
95         }
96         
97         /* Have we accidentally picked a clean block with wasted space ? */
98         if (ret->wasted_size) {
99                 D1(printk(KERN_DEBUG "Converting wasted_size %08x to dirty_size\n", ret->wasted_size));
100                 ret->dirty_size += ret->wasted_size;
101                 c->wasted_size -= ret->wasted_size;
102                 c->dirty_size += ret->wasted_size;
103                 ret->wasted_size = 0;
104         }
105
106         D1(jffs2_dump_block_lists(c));
107         return ret;
108 }
109
110 /* jffs2_garbage_collect_pass
111  * Make a single attempt to progress GC. Move one node, and possibly
112  * start erasing one eraseblock.
113  */
114 int jffs2_garbage_collect_pass(struct jffs2_sb_info *c)
115 {
116         struct jffs2_inode_info *f;
117         struct jffs2_inode_cache *ic;
118         struct jffs2_eraseblock *jeb;
119         struct jffs2_raw_node_ref *raw;
120         int ret = 0, inum, nlink;
121
122         if (down_interruptible(&c->alloc_sem))
123                 return -EINTR;
124
125         for (;;) {
126                 spin_lock(&c->erase_completion_lock);
127                 if (!c->unchecked_size)
128                         break;
129
130                 /* We can't start doing GC yet. We haven't finished checking
131                    the node CRCs etc. Do it now. */
132                 
133                 /* checked_ino is protected by the alloc_sem */
134                 if (c->checked_ino > c->highest_ino) {
135                         printk(KERN_CRIT "Checked all inodes but still 0x%x bytes of unchecked space?\n",
136                                c->unchecked_size);
137                         D1(jffs2_dump_block_lists(c));
138                         spin_unlock(&c->erase_completion_lock);
139                         BUG();
140                 }
141
142                 spin_unlock(&c->erase_completion_lock);
143
144                 spin_lock(&c->inocache_lock);
145
146                 ic = jffs2_get_ino_cache(c, c->checked_ino++);
147
148                 if (!ic) {
149                         spin_unlock(&c->inocache_lock);
150                         continue;
151                 }
152
153                 if (!ic->nlink) {
154                         D1(printk(KERN_DEBUG "Skipping check of ino #%d with nlink zero\n",
155                                   ic->ino));
156                         spin_unlock(&c->inocache_lock);
157                         continue;
158                 }
159                 switch(ic->state) {
160                 case INO_STATE_CHECKEDABSENT:
161                 case INO_STATE_PRESENT:
162                         D1(printk(KERN_DEBUG "Skipping ino #%u already checked\n", ic->ino));
163                         spin_unlock(&c->inocache_lock);
164                         continue;
165
166                 case INO_STATE_GC:
167                 case INO_STATE_CHECKING:
168                         printk(KERN_WARNING "Inode #%u is in state %d during CRC check phase!\n", ic->ino, ic->state);
169                         spin_unlock(&c->inocache_lock);
170                         BUG();
171
172                 case INO_STATE_READING:
173                         /* We need to wait for it to finish, lest we move on
174                            and trigger the BUG() above while we haven't yet 
175                            finished checking all its nodes */
176                         D1(printk(KERN_DEBUG "Waiting for ino #%u to finish reading\n", ic->ino));
177                         up(&c->alloc_sem);
178                         sleep_on_spinunlock(&c->inocache_wq, &c->inocache_lock);
179                         return 0;
180
181                 default:
182                         BUG();
183
184                 case INO_STATE_UNCHECKED:
185                         ;
186                 }
187                 ic->state = INO_STATE_CHECKING;
188                 spin_unlock(&c->inocache_lock);
189
190                 D1(printk(KERN_DEBUG "jffs2_garbage_collect_pass() triggering inode scan of ino#%u\n", ic->ino));
191
192                 ret = jffs2_do_crccheck_inode(c, ic);
193                 if (ret)
194                         printk(KERN_WARNING "Returned error for crccheck of ino #%u. Expect badness...\n", ic->ino);
195
196                 jffs2_set_inocache_state(c, ic, INO_STATE_CHECKEDABSENT);
197                 up(&c->alloc_sem);
198                 return ret;
199         }
200
201         /* First, work out which block we're garbage-collecting */
202         jeb = c->gcblock;
203
204         if (!jeb)
205                 jeb = jffs2_find_gc_block(c);
206
207         if (!jeb) {
208                 D1 (printk(KERN_NOTICE "jffs2: Couldn't find erase block to garbage collect!\n"));
209                 spin_unlock(&c->erase_completion_lock);
210                 up(&c->alloc_sem);
211                 return -EIO;
212         }
213
214         D1(printk(KERN_DEBUG "GC from block %08x, used_size %08x, dirty_size %08x, free_size %08x\n", jeb->offset, jeb->used_size, jeb->dirty_size, jeb->free_size));
215         D1(if (c->nextblock)
216            printk(KERN_DEBUG "Nextblock at  %08x, used_size %08x, dirty_size %08x, wasted_size %08x, free_size %08x\n", c->nextblock->offset, c->nextblock->used_size, c->nextblock->dirty_size, c->nextblock->wasted_size, c->nextblock->free_size));
217
218         if (!jeb->used_size) {
219                 up(&c->alloc_sem);
220                 goto eraseit;
221         }
222
223         raw = jeb->gc_node;
224                         
225         while(ref_obsolete(raw)) {
226                 D1(printk(KERN_DEBUG "Node at 0x%08x is obsolete... skipping\n", ref_offset(raw)));
227                 raw = raw->next_phys;
228                 if (unlikely(!raw)) {
229                         printk(KERN_WARNING "eep. End of raw list while still supposedly nodes to GC\n");
230                         printk(KERN_WARNING "erase block at 0x%08x. free_size 0x%08x, dirty_size 0x%08x, used_size 0x%08x\n", 
231                                jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size);
232                         jeb->gc_node = raw;
233                         spin_unlock(&c->erase_completion_lock);
234                         up(&c->alloc_sem);
235                         BUG();
236                 }
237         }
238         jeb->gc_node = raw;
239
240         D1(printk(KERN_DEBUG "Going to garbage collect node at 0x%08x\n", ref_offset(raw)));
241
242         if (!raw->next_in_ino) {
243                 /* Inode-less node. Clean marker, snapshot or something like that */
244                 /* FIXME: If it's something that needs to be copied, including something
245                    we don't grok that has JFFS2_NODETYPE_RWCOMPAT_COPY, we should do so */
246                 spin_unlock(&c->erase_completion_lock);
247                 jffs2_mark_node_obsolete(c, raw);
248                 up(&c->alloc_sem);
249                 goto eraseit_lock;
250         }
251
252         ic = jffs2_raw_ref_to_ic(raw);
253
254         /* We need to hold the inocache. Either the erase_completion_lock or
255            the inocache_lock are sufficient; we trade down since the inocache_lock 
256            causes less contention. */
257         spin_lock(&c->inocache_lock);
258
259         spin_unlock(&c->erase_completion_lock);
260
261         D1(printk(KERN_DEBUG "jffs2_garbage_collect_pass collecting from block @0x%08x. Node @0x%08x(%d), ino #%u\n", jeb->offset, ref_offset(raw), ref_flags(raw), ic->ino));
262
263         /* Three possibilities:
264            1. Inode is already in-core. We must iget it and do proper
265               updating to its fragtree, etc.
266            2. Inode is not in-core, node is REF_PRISTINE. We lock the
267               inocache to prevent a read_inode(), copy the node intact.
268            3. Inode is not in-core, node is not pristine. We must iget()
269               and take the slow path.
270         */
271
272         switch(ic->state) {
273         case INO_STATE_CHECKEDABSENT:
274                 /* It's been checked, but it's not currently in-core. 
275                    We can just copy any pristine nodes, but have
276                    to prevent anyone else from doing read_inode() while
277                    we're at it, so we set the state accordingly */
278                 if (ref_flags(raw) == REF_PRISTINE)
279                         ic->state = INO_STATE_GC;
280                 else {
281                         D1(printk(KERN_DEBUG "Ino #%u is absent but node not REF_PRISTINE. Reading.\n", 
282                                   ic->ino));
283                 }
284                 break;
285
286         case INO_STATE_PRESENT:
287                 /* It's in-core. GC must iget() it. */
288                 break;
289
290         case INO_STATE_UNCHECKED:
291         case INO_STATE_CHECKING:
292         case INO_STATE_GC:
293                 /* Should never happen. We should have finished checking
294                    by the time we actually start doing any GC, and since 
295                    we're holding the alloc_sem, no other garbage collection 
296                    can happen.
297                 */
298                 printk(KERN_CRIT "Inode #%u already in state %d in jffs2_garbage_collect_pass()!\n",
299                        ic->ino, ic->state);
300                 up(&c->alloc_sem);
301                 spin_unlock(&c->inocache_lock);
302                 BUG();
303
304         case INO_STATE_READING:
305                 /* Someone's currently trying to read it. We must wait for
306                    them to finish and then go through the full iget() route
307                    to do the GC. However, sometimes read_inode() needs to get
308                    the alloc_sem() (for marking nodes invalid) so we must
309                    drop the alloc_sem before sleeping. */
310
311                 up(&c->alloc_sem);
312                 D1(printk(KERN_DEBUG "jffs2_garbage_collect_pass() waiting for ino #%u in state %d\n",
313                           ic->ino, ic->state));
314                 sleep_on_spinunlock(&c->inocache_wq, &c->inocache_lock);
315                 /* And because we dropped the alloc_sem we must start again from the 
316                    beginning. Ponder chance of livelock here -- we're returning success
317                    without actually making any progress.
318
319                    Q: What are the chances that the inode is back in INO_STATE_READING 
320                    again by the time we next enter this function? And that this happens
321                    enough times to cause a real delay?
322
323                    A: Small enough that I don't care :) 
324                 */
325                 return 0;
326         }
327
328         /* OK. Now if the inode is in state INO_STATE_GC, we are going to copy the
329            node intact, and we don't have to muck about with the fragtree etc. 
330            because we know it's not in-core. If it _was_ in-core, we go through
331            all the iget() crap anyway */
332
333         if (ic->state == INO_STATE_GC) {
334                 spin_unlock(&c->inocache_lock);
335
336                 ret = jffs2_garbage_collect_pristine(c, ic, raw);
337
338                 spin_lock(&c->inocache_lock);
339                 ic->state = INO_STATE_CHECKEDABSENT;
340                 wake_up(&c->inocache_wq);
341
342                 if (ret != -EBADFD) {
343                         spin_unlock(&c->inocache_lock);
344                         goto release_sem;
345                 }
346
347                 /* Fall through if it wanted us to, with inocache_lock held */
348         }
349
350         /* Prevent the fairly unlikely race where the gcblock is
351            entirely obsoleted by the final close of a file which had
352            the only valid nodes in the block, followed by erasure,
353            followed by freeing of the ic because the erased block(s)
354            held _all_ the nodes of that inode.... never been seen but
355            it's vaguely possible. */
356
357         inum = ic->ino;
358         nlink = ic->nlink;
359         spin_unlock(&c->inocache_lock);
360
361         f = jffs2_gc_fetch_inode(c, inum, nlink);
362         if (IS_ERR(f))
363                 return PTR_ERR(f);
364         if (!f)
365                 return 0;
366
367         ret = jffs2_garbage_collect_live(c, jeb, raw, f);
368
369         jffs2_gc_release_inode(c, f);
370
371  release_sem:
372         up(&c->alloc_sem);
373
374  eraseit_lock:
375         /* If we've finished this block, start it erasing */
376         spin_lock(&c->erase_completion_lock);
377
378  eraseit:
379         if (c->gcblock && !c->gcblock->used_size) {
380                 D1(printk(KERN_DEBUG "Block at 0x%08x completely obsoleted by GC. Moving to erase_pending_list\n", c->gcblock->offset));
381                 /* We're GC'ing an empty block? */
382                 list_add_tail(&c->gcblock->list, &c->erase_pending_list);
383                 c->gcblock = NULL;
384                 c->nr_erasing_blocks++;
385                 jffs2_erase_pending_trigger(c);
386         }
387         spin_unlock(&c->erase_completion_lock);
388
389         return ret;
390 }
391
392 static int jffs2_garbage_collect_live(struct jffs2_sb_info *c,  struct jffs2_eraseblock *jeb,
393                                       struct jffs2_raw_node_ref *raw, struct jffs2_inode_info *f)
394 {
395         struct jffs2_node_frag *frag;
396         struct jffs2_full_dnode *fn = NULL;
397         struct jffs2_full_dirent *fd;
398         uint32_t start = 0, end = 0, nrfrags = 0;
399         int ret = 0;
400
401         down(&f->sem);
402
403         /* Now we have the lock for this inode. Check that it's still the one at the head
404            of the list. */
405
406         spin_lock(&c->erase_completion_lock);
407
408         if (c->gcblock != jeb) {
409                 spin_unlock(&c->erase_completion_lock);
410                 D1(printk(KERN_DEBUG "GC block is no longer gcblock. Restart\n"));
411                 goto upnout;
412         }
413         if (ref_obsolete(raw)) {
414                 spin_unlock(&c->erase_completion_lock);
415                 D1(printk(KERN_DEBUG "node to be GC'd was obsoleted in the meantime.\n"));
416                 /* They'll call again */
417                 goto upnout;
418         }
419         spin_unlock(&c->erase_completion_lock);
420
421         /* OK. Looks safe. And nobody can get us now because we have the semaphore. Move the block */
422         if (f->metadata && f->metadata->raw == raw) {
423                 fn = f->metadata;
424                 ret = jffs2_garbage_collect_metadata(c, jeb, f, fn);
425                 goto upnout;
426         }
427
428         /* FIXME. Read node and do lookup? */
429         for (frag = frag_first(&f->fragtree); frag; frag = frag_next(frag)) {
430                 if (frag->node && frag->node->raw == raw) {
431                         fn = frag->node;
432                         end = frag->ofs + frag->size;
433                         if (!nrfrags++)
434                                 start = frag->ofs;
435                         if (nrfrags == frag->node->frags)
436                                 break; /* We've found them all */
437                 }
438         }
439         if (fn) {
440                 if (ref_flags(raw) == REF_PRISTINE) {
441                         ret = jffs2_garbage_collect_pristine(c, f->inocache, raw);
442                         if (!ret) {
443                                 /* Urgh. Return it sensibly. */
444                                 frag->node->raw = f->inocache->nodes;
445                         }       
446                         if (ret != -EBADFD)
447                                 goto upnout;
448                 }
449                 /* We found a datanode. Do the GC */
450                 if((start >> PAGE_CACHE_SHIFT) < ((end-1) >> PAGE_CACHE_SHIFT)) {
451                         /* It crosses a page boundary. Therefore, it must be a hole. */
452                         ret = jffs2_garbage_collect_hole(c, jeb, f, fn, start, end);
453                 } else {
454                         /* It could still be a hole. But we GC the page this way anyway */
455                         ret = jffs2_garbage_collect_dnode(c, jeb, f, fn, start, end);
456                 }
457                 goto upnout;
458         }
459         
460         /* Wasn't a dnode. Try dirent */
461         for (fd = f->dents; fd; fd=fd->next) {
462                 if (fd->raw == raw)
463                         break;
464         }
465
466         if (fd && fd->ino) {
467                 ret = jffs2_garbage_collect_dirent(c, jeb, f, fd);
468         } else if (fd) {
469                 ret = jffs2_garbage_collect_deletion_dirent(c, jeb, f, fd);
470         } else {
471                 printk(KERN_WARNING "Raw node at 0x%08x wasn't in node lists for ino #%u\n",
472                        ref_offset(raw), f->inocache->ino);
473                 if (ref_obsolete(raw)) {
474                         printk(KERN_WARNING "But it's obsolete so we don't mind too much\n");
475                 } else {
476                         ret = -EIO;
477                 }
478         }
479  upnout:
480         up(&f->sem);
481
482         return ret;
483 }
484
485 static int jffs2_garbage_collect_pristine(struct jffs2_sb_info *c, 
486                                           struct jffs2_inode_cache *ic,
487                                           struct jffs2_raw_node_ref *raw)
488 {
489         union jffs2_node_union *node;
490         struct jffs2_raw_node_ref *nraw;
491         size_t retlen;
492         int ret;
493         uint32_t phys_ofs, alloclen;
494         uint32_t crc, rawlen;
495         int retried = 0;
496
497         D1(printk(KERN_DEBUG "Going to GC REF_PRISTINE node at 0x%08x\n", ref_offset(raw)));
498
499         rawlen = ref_totlen(c, c->gcblock, raw);
500
501         /* Ask for a small amount of space (or the totlen if smaller) because we
502            don't want to force wastage of the end of a block if splitting would
503            work. */
504         ret = jffs2_reserve_space_gc(c, min_t(uint32_t, sizeof(struct jffs2_raw_inode) + JFFS2_MIN_DATA_LEN, 
505                                               rawlen), &phys_ofs, &alloclen);
506         if (ret)
507                 return ret;
508
509         if (alloclen < rawlen) {
510                 /* Doesn't fit untouched. We'll go the old route and split it */
511                 return -EBADFD;
512         }
513
514         node = kmalloc(rawlen, GFP_KERNEL);
515         if (!node)
516                return -ENOMEM;
517
518         ret = jffs2_flash_read(c, ref_offset(raw), rawlen, &retlen, (char *)node);
519         if (!ret && retlen != rawlen)
520                 ret = -EIO;
521         if (ret)
522                 goto out_node;
523
524         crc = crc32(0, node, sizeof(struct jffs2_unknown_node)-4);
525         if (je32_to_cpu(node->u.hdr_crc) != crc) {
526                 printk(KERN_WARNING "Header CRC failed on REF_PRISTINE node at 0x%08x: Read 0x%08x, calculated 0x%08x\n",
527                        ref_offset(raw), je32_to_cpu(node->u.hdr_crc), crc);
528                 goto bail;
529         }
530
531         switch(je16_to_cpu(node->u.nodetype)) {
532         case JFFS2_NODETYPE_INODE:
533                 crc = crc32(0, node, sizeof(node->i)-8);
534                 if (je32_to_cpu(node->i.node_crc) != crc) {
535                         printk(KERN_WARNING "Node CRC failed on REF_PRISTINE data node at 0x%08x: Read 0x%08x, calculated 0x%08x\n",
536                                ref_offset(raw), je32_to_cpu(node->i.node_crc), crc);
537                         goto bail;
538                 }
539
540                 if (je32_to_cpu(node->i.dsize)) {
541                         crc = crc32(0, node->i.data, je32_to_cpu(node->i.csize));
542                         if (je32_to_cpu(node->i.data_crc) != crc) {
543                                 printk(KERN_WARNING "Data CRC failed on REF_PRISTINE data node at 0x%08x: Read 0x%08x, calculated 0x%08x\n",
544                                        ref_offset(raw), je32_to_cpu(node->i.data_crc), crc);
545                                 goto bail;
546                         }
547                 }
548                 break;
549
550         case JFFS2_NODETYPE_DIRENT:
551                 crc = crc32(0, node, sizeof(node->d)-8);
552                 if (je32_to_cpu(node->d.node_crc) != crc) {
553                         printk(KERN_WARNING "Node CRC failed on REF_PRISTINE dirent node at 0x%08x: Read 0x%08x, calculated 0x%08x\n",
554                                ref_offset(raw), je32_to_cpu(node->d.node_crc), crc);
555                         goto bail;
556                 }
557
558                 if (node->d.nsize) {
559                         crc = crc32(0, node->d.name, node->d.nsize);
560                         if (je32_to_cpu(node->d.name_crc) != crc) {
561                                 printk(KERN_WARNING "Name CRC failed on REF_PRISTINE dirent ode at 0x%08x: Read 0x%08x, calculated 0x%08x\n",
562                                        ref_offset(raw), je32_to_cpu(node->d.name_crc), crc);
563                                 goto bail;
564                         }
565                 }
566                 break;
567         default:
568                 printk(KERN_WARNING "Unknown node type for REF_PRISTINE node at 0x%08x: 0x%04x\n", 
569                        ref_offset(raw), je16_to_cpu(node->u.nodetype));
570                 goto bail;
571         }
572
573         nraw = jffs2_alloc_raw_node_ref();
574         if (!nraw) {
575                 ret = -ENOMEM;
576                 goto out_node;
577         }
578
579         /* OK, all the CRCs are good; this node can just be copied as-is. */
580  retry:
581         nraw->flash_offset = phys_ofs;
582         nraw->__totlen = rawlen;
583         nraw->next_phys = NULL;
584
585         ret = jffs2_flash_write(c, phys_ofs, rawlen, &retlen, (char *)node);
586
587         if (ret || (retlen != rawlen)) {
588                 printk(KERN_NOTICE "Write of %d bytes at 0x%08x failed. returned %d, retlen %zd\n",
589                        rawlen, phys_ofs, ret, retlen);
590                 if (retlen) {
591                         /* Doesn't belong to any inode */
592                         nraw->next_in_ino = NULL;
593
594                         nraw->flash_offset |= REF_OBSOLETE;
595                         jffs2_add_physical_node_ref(c, nraw);
596                         jffs2_mark_node_obsolete(c, nraw);
597                 } else {
598                         printk(KERN_NOTICE "Not marking the space at 0x%08x as dirty because the flash driver returned retlen zero\n", nraw->flash_offset);
599                         jffs2_free_raw_node_ref(nraw);
600                 }
601                 if (!retried && (nraw == jffs2_alloc_raw_node_ref())) {
602                         /* Try to reallocate space and retry */
603                         uint32_t dummy;
604                         struct jffs2_eraseblock *jeb = &c->blocks[phys_ofs / c->sector_size];
605
606                         retried = 1;
607
608                         D1(printk(KERN_DEBUG "Retrying failed write of REF_PRISTINE node.\n"));
609                         
610                         ACCT_SANITY_CHECK(c,jeb);
611                         D1(ACCT_PARANOIA_CHECK(jeb));
612
613                         ret = jffs2_reserve_space_gc(c, rawlen, &phys_ofs, &dummy);
614
615                         if (!ret) {
616                                 D1(printk(KERN_DEBUG "Allocated space at 0x%08x to retry failed write.\n", phys_ofs));
617
618                                 ACCT_SANITY_CHECK(c,jeb);
619                                 D1(ACCT_PARANOIA_CHECK(jeb));
620
621                                 goto retry;
622                         }
623                         D1(printk(KERN_DEBUG "Failed to allocate space to retry failed write: %d!\n", ret));
624                         jffs2_free_raw_node_ref(nraw);
625                 }
626
627                 if (!ret)
628                         ret = -EIO;
629                 goto out_node;
630         }
631         nraw->flash_offset |= REF_PRISTINE;
632         jffs2_add_physical_node_ref(c, nraw);
633
634         /* Link into per-inode list. This is safe because of the ic
635            state being INO_STATE_GC. Note that if we're doing this
636            for an inode which is in-code, the 'nraw' pointer is then
637            going to be fetched from ic->nodes by our caller. */
638         nraw->next_in_ino = ic->nodes;
639         ic->nodes = nraw;
640
641         jffs2_mark_node_obsolete(c, raw);
642         D1(printk(KERN_DEBUG "WHEEE! GC REF_PRISTINE node at 0x%08x succeeded\n", ref_offset(raw)));
643
644  out_node:
645         kfree(node);
646         return ret;
647  bail:
648         ret = -EBADFD;
649         goto out_node;
650 }
651
652 static int jffs2_garbage_collect_metadata(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, 
653                                         struct jffs2_inode_info *f, struct jffs2_full_dnode *fn)
654 {
655         struct jffs2_full_dnode *new_fn;
656         struct jffs2_raw_inode ri;
657         jint16_t dev;
658         char *mdata = NULL, mdatalen = 0;
659         uint32_t alloclen, phys_ofs;
660         int ret;
661
662         if (S_ISBLK(JFFS2_F_I_MODE(f)) ||
663             S_ISCHR(JFFS2_F_I_MODE(f)) ) {
664                 /* For these, we don't actually need to read the old node */
665                 /* FIXME: for minor or major > 255. */
666                 dev = cpu_to_je16(((JFFS2_F_I_RDEV_MAJ(f) << 8) | 
667                         JFFS2_F_I_RDEV_MIN(f)));
668                 mdata = (char *)&dev;
669                 mdatalen = sizeof(dev);
670                 D1(printk(KERN_DEBUG "jffs2_garbage_collect_metadata(): Writing %d bytes of kdev_t\n", mdatalen));
671         } else if (S_ISLNK(JFFS2_F_I_MODE(f))) {
672                 mdatalen = fn->size;
673                 mdata = kmalloc(fn->size, GFP_KERNEL);
674                 if (!mdata) {
675                         printk(KERN_WARNING "kmalloc of mdata failed in jffs2_garbage_collect_metadata()\n");
676                         return -ENOMEM;
677                 }
678                 ret = jffs2_read_dnode(c, f, fn, mdata, 0, mdatalen);
679                 if (ret) {
680                         printk(KERN_WARNING "read of old metadata failed in jffs2_garbage_collect_metadata(): %d\n", ret);
681                         kfree(mdata);
682                         return ret;
683                 }
684                 D1(printk(KERN_DEBUG "jffs2_garbage_collect_metadata(): Writing %d bites of symlink target\n", mdatalen));
685
686         }
687         
688         ret = jffs2_reserve_space_gc(c, sizeof(ri) + mdatalen, &phys_ofs, &alloclen);
689         if (ret) {
690                 printk(KERN_WARNING "jffs2_reserve_space_gc of %zd bytes for garbage_collect_metadata failed: %d\n",
691                        sizeof(ri)+ mdatalen, ret);
692                 goto out;
693         }
694         
695         memset(&ri, 0, sizeof(ri));
696         ri.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK);
697         ri.nodetype = cpu_to_je16(JFFS2_NODETYPE_INODE);
698         ri.totlen = cpu_to_je32(sizeof(ri) + mdatalen);
699         ri.hdr_crc = cpu_to_je32(crc32(0, &ri, sizeof(struct jffs2_unknown_node)-4));
700
701         ri.ino = cpu_to_je32(f->inocache->ino);
702         ri.version = cpu_to_je32(++f->highest_version);
703         ri.mode = cpu_to_jemode(JFFS2_F_I_MODE(f));
704         ri.uid = cpu_to_je16(JFFS2_F_I_UID(f));
705         ri.gid = cpu_to_je16(JFFS2_F_I_GID(f));
706         ri.isize = cpu_to_je32(JFFS2_F_I_SIZE(f));
707         ri.atime = cpu_to_je32(JFFS2_F_I_ATIME(f));
708         ri.ctime = cpu_to_je32(JFFS2_F_I_CTIME(f));
709         ri.mtime = cpu_to_je32(JFFS2_F_I_MTIME(f));
710         ri.offset = cpu_to_je32(0);
711         ri.csize = cpu_to_je32(mdatalen);
712         ri.dsize = cpu_to_je32(mdatalen);
713         ri.compr = JFFS2_COMPR_NONE;
714         ri.node_crc = cpu_to_je32(crc32(0, &ri, sizeof(ri)-8));
715         ri.data_crc = cpu_to_je32(crc32(0, mdata, mdatalen));
716
717         new_fn = jffs2_write_dnode(c, f, &ri, mdata, mdatalen, phys_ofs, ALLOC_GC);
718
719         if (IS_ERR(new_fn)) {
720                 printk(KERN_WARNING "Error writing new dnode: %ld\n", PTR_ERR(new_fn));
721                 ret = PTR_ERR(new_fn);
722                 goto out;
723         }
724         jffs2_mark_node_obsolete(c, fn->raw);
725         jffs2_free_full_dnode(fn);
726         f->metadata = new_fn;
727  out:
728         if (S_ISLNK(JFFS2_F_I_MODE(f)))
729                 kfree(mdata);
730         return ret;
731 }
732
733 static int jffs2_garbage_collect_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, 
734                                         struct jffs2_inode_info *f, struct jffs2_full_dirent *fd)
735 {
736         struct jffs2_full_dirent *new_fd;
737         struct jffs2_raw_dirent rd;
738         uint32_t alloclen, phys_ofs;
739         int ret;
740
741         rd.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK);
742         rd.nodetype = cpu_to_je16(JFFS2_NODETYPE_DIRENT);
743         rd.nsize = strlen(fd->name);
744         rd.totlen = cpu_to_je32(sizeof(rd) + rd.nsize);
745         rd.hdr_crc = cpu_to_je32(crc32(0, &rd, sizeof(struct jffs2_unknown_node)-4));
746
747         rd.pino = cpu_to_je32(f->inocache->ino);
748         rd.version = cpu_to_je32(++f->highest_version);
749         rd.ino = cpu_to_je32(fd->ino);
750         rd.mctime = cpu_to_je32(max(JFFS2_F_I_MTIME(f), JFFS2_F_I_CTIME(f)));
751         rd.type = fd->type;
752         rd.node_crc = cpu_to_je32(crc32(0, &rd, sizeof(rd)-8));
753         rd.name_crc = cpu_to_je32(crc32(0, fd->name, rd.nsize));
754         
755         ret = jffs2_reserve_space_gc(c, sizeof(rd)+rd.nsize, &phys_ofs, &alloclen);
756         if (ret) {
757                 printk(KERN_WARNING "jffs2_reserve_space_gc of %zd bytes for garbage_collect_dirent failed: %d\n",
758                        sizeof(rd)+rd.nsize, ret);
759                 return ret;
760         }
761         new_fd = jffs2_write_dirent(c, f, &rd, fd->name, rd.nsize, phys_ofs, ALLOC_GC);
762
763         if (IS_ERR(new_fd)) {
764                 printk(KERN_WARNING "jffs2_write_dirent in garbage_collect_dirent failed: %ld\n", PTR_ERR(new_fd));
765                 return PTR_ERR(new_fd);
766         }
767         jffs2_add_fd_to_list(c, new_fd, &f->dents);
768         return 0;
769 }
770
771 static int jffs2_garbage_collect_deletion_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, 
772                                         struct jffs2_inode_info *f, struct jffs2_full_dirent *fd)
773 {
774         struct jffs2_full_dirent **fdp = &f->dents;
775         int found = 0;
776
777         /* On a medium where we can't actually mark nodes obsolete
778            pernamently, such as NAND flash, we need to work out
779            whether this deletion dirent is still needed to actively
780            delete a 'real' dirent with the same name that's still
781            somewhere else on the flash. */
782         if (!jffs2_can_mark_obsolete(c)) {
783                 struct jffs2_raw_dirent *rd;
784                 struct jffs2_raw_node_ref *raw;
785                 int ret;
786                 size_t retlen;
787                 int name_len = strlen(fd->name);
788                 uint32_t name_crc = crc32(0, fd->name, name_len);
789                 uint32_t rawlen = ref_totlen(c, jeb, fd->raw);
790
791                 rd = kmalloc(rawlen, GFP_KERNEL);
792                 if (!rd)
793                         return -ENOMEM;
794
795                 /* Prevent the erase code from nicking the obsolete node refs while
796                    we're looking at them. I really don't like this extra lock but
797                    can't see any alternative. Suggestions on a postcard to... */
798                 down(&c->erase_free_sem);
799
800                 for (raw = f->inocache->nodes; raw != (void *)f->inocache; raw = raw->next_in_ino) {
801
802                         /* We only care about obsolete ones */
803                         if (!(ref_obsolete(raw)))
804                                 continue;
805
806                         /* Any dirent with the same name is going to have the same length... */
807                         if (ref_totlen(c, NULL, raw) != rawlen)
808                                 continue;
809
810                         /* Doesn't matter if there's one in the same erase block. We're going to 
811                            delete it too at the same time. */
812                         if ((raw->flash_offset & ~(c->sector_size-1)) ==
813                             (fd->raw->flash_offset & ~(c->sector_size-1)))
814                                 continue;
815
816                         D1(printk(KERN_DEBUG "Check potential deletion dirent at %08x\n", ref_offset(raw)));
817
818                         /* This is an obsolete node belonging to the same directory, and it's of the right
819                            length. We need to take a closer look...*/
820                         ret = jffs2_flash_read(c, ref_offset(raw), rawlen, &retlen, (char *)rd);
821                         if (ret) {
822                                 printk(KERN_WARNING "jffs2_g_c_deletion_dirent(): Read error (%d) reading obsolete node at %08x\n", ret, ref_offset(raw));
823                                 /* If we can't read it, we don't need to continue to obsolete it. Continue */
824                                 continue;
825                         }
826                         if (retlen != rawlen) {
827                                 printk(KERN_WARNING "jffs2_g_c_deletion_dirent(): Short read (%zd not %u) reading header from obsolete node at %08x\n",
828                                        retlen, rawlen, ref_offset(raw));
829                                 continue;
830                         }
831
832                         if (je16_to_cpu(rd->nodetype) != JFFS2_NODETYPE_DIRENT)
833                                 continue;
834
835                         /* If the name CRC doesn't match, skip */
836                         if (je32_to_cpu(rd->name_crc) != name_crc)
837                                 continue;
838
839                         /* If the name length doesn't match, or it's another deletion dirent, skip */
840                         if (rd->nsize != name_len || !je32_to_cpu(rd->ino))
841                                 continue;
842
843                         /* OK, check the actual name now */
844                         if (memcmp(rd->name, fd->name, name_len))
845                                 continue;
846
847                         /* OK. The name really does match. There really is still an older node on
848                            the flash which our deletion dirent obsoletes. So we have to write out
849                            a new deletion dirent to replace it */
850                         up(&c->erase_free_sem);
851
852                         D1(printk(KERN_DEBUG "Deletion dirent at %08x still obsoletes real dirent \"%s\" at %08x for ino #%u\n",
853                                   ref_offset(fd->raw), fd->name, ref_offset(raw), je32_to_cpu(rd->ino)));
854                         kfree(rd);
855
856                         return jffs2_garbage_collect_dirent(c, jeb, f, fd);
857                 }
858
859                 up(&c->erase_free_sem);
860                 kfree(rd);
861         }
862
863         /* No need for it any more. Just mark it obsolete and remove it from the list */
864         while (*fdp) {
865                 if ((*fdp) == fd) {
866                         found = 1;
867                         *fdp = fd->next;
868                         break;
869                 }
870                 fdp = &(*fdp)->next;
871         }
872         if (!found) {
873                 printk(KERN_WARNING "Deletion dirent \"%s\" not found in list for ino #%u\n", fd->name, f->inocache->ino);
874         }
875         jffs2_mark_node_obsolete(c, fd->raw);
876         jffs2_free_full_dirent(fd);
877         return 0;
878 }
879
880 static int jffs2_garbage_collect_hole(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
881                                       struct jffs2_inode_info *f, struct jffs2_full_dnode *fn,
882                                       uint32_t start, uint32_t end)
883 {
884         struct jffs2_raw_inode ri;
885         struct jffs2_node_frag *frag;
886         struct jffs2_full_dnode *new_fn;
887         uint32_t alloclen, phys_ofs;
888         int ret;
889
890         D1(printk(KERN_DEBUG "Writing replacement hole node for ino #%u from offset 0x%x to 0x%x\n",
891                   f->inocache->ino, start, end));
892         
893         memset(&ri, 0, sizeof(ri));
894
895         if(fn->frags > 1) {
896                 size_t readlen;
897                 uint32_t crc;
898                 /* It's partially obsoleted by a later write. So we have to 
899                    write it out again with the _same_ version as before */
900                 ret = jffs2_flash_read(c, ref_offset(fn->raw), sizeof(ri), &readlen, (char *)&ri);
901                 if (readlen != sizeof(ri) || ret) {
902                         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);
903                         goto fill;
904                 }
905                 if (je16_to_cpu(ri.nodetype) != JFFS2_NODETYPE_INODE) {
906                         printk(KERN_WARNING "jffs2_garbage_collect_hole: Node at 0x%08x had node type 0x%04x instead of JFFS2_NODETYPE_INODE(0x%04x)\n",
907                                ref_offset(fn->raw),
908                                je16_to_cpu(ri.nodetype), JFFS2_NODETYPE_INODE);
909                         return -EIO;
910                 }
911                 if (je32_to_cpu(ri.totlen) != sizeof(ri)) {
912                         printk(KERN_WARNING "jffs2_garbage_collect_hole: Node at 0x%08x had totlen 0x%x instead of expected 0x%zx\n",
913                                ref_offset(fn->raw),
914                                je32_to_cpu(ri.totlen), sizeof(ri));
915                         return -EIO;
916                 }
917                 crc = crc32(0, &ri, sizeof(ri)-8);
918                 if (crc != je32_to_cpu(ri.node_crc)) {
919                         printk(KERN_WARNING "jffs2_garbage_collect_hole: Node at 0x%08x had CRC 0x%08x which doesn't match calculated CRC 0x%08x\n",
920                                ref_offset(fn->raw), 
921                                je32_to_cpu(ri.node_crc), crc);
922                         /* FIXME: We could possibly deal with this by writing new holes for each frag */
923                         printk(KERN_WARNING "Data in the range 0x%08x to 0x%08x of inode #%u will be lost\n", 
924                                start, end, f->inocache->ino);
925                         goto fill;
926                 }
927                 if (ri.compr != JFFS2_COMPR_ZERO) {
928                         printk(KERN_WARNING "jffs2_garbage_collect_hole: Node 0x%08x wasn't a hole node!\n", ref_offset(fn->raw));
929                         printk(KERN_WARNING "Data in the range 0x%08x to 0x%08x of inode #%u will be lost\n", 
930                                start, end, f->inocache->ino);
931                         goto fill;
932                 }
933         } else {
934         fill:
935                 ri.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK);
936                 ri.nodetype = cpu_to_je16(JFFS2_NODETYPE_INODE);
937                 ri.totlen = cpu_to_je32(sizeof(ri));
938                 ri.hdr_crc = cpu_to_je32(crc32(0, &ri, sizeof(struct jffs2_unknown_node)-4));
939
940                 ri.ino = cpu_to_je32(f->inocache->ino);
941                 ri.version = cpu_to_je32(++f->highest_version);
942                 ri.offset = cpu_to_je32(start);
943                 ri.dsize = cpu_to_je32(end - start);
944                 ri.csize = cpu_to_je32(0);
945                 ri.compr = JFFS2_COMPR_ZERO;
946         }
947         ri.mode = cpu_to_jemode(JFFS2_F_I_MODE(f));
948         ri.uid = cpu_to_je16(JFFS2_F_I_UID(f));
949         ri.gid = cpu_to_je16(JFFS2_F_I_GID(f));
950         ri.isize = cpu_to_je32(JFFS2_F_I_SIZE(f));
951         ri.atime = cpu_to_je32(JFFS2_F_I_ATIME(f));
952         ri.ctime = cpu_to_je32(JFFS2_F_I_CTIME(f));
953         ri.mtime = cpu_to_je32(JFFS2_F_I_MTIME(f));
954         ri.data_crc = cpu_to_je32(0);
955         ri.node_crc = cpu_to_je32(crc32(0, &ri, sizeof(ri)-8));
956
957         ret = jffs2_reserve_space_gc(c, sizeof(ri), &phys_ofs, &alloclen);
958         if (ret) {
959                 printk(KERN_WARNING "jffs2_reserve_space_gc of %zd bytes for garbage_collect_hole failed: %d\n",
960                        sizeof(ri), ret);
961                 return ret;
962         }
963         new_fn = jffs2_write_dnode(c, f, &ri, NULL, 0, phys_ofs, ALLOC_GC);
964
965         if (IS_ERR(new_fn)) {
966                 printk(KERN_WARNING "Error writing new hole node: %ld\n", PTR_ERR(new_fn));
967                 return PTR_ERR(new_fn);
968         }
969         if (je32_to_cpu(ri.version) == f->highest_version) {
970                 jffs2_add_full_dnode_to_inode(c, f, new_fn);
971                 if (f->metadata) {
972                         jffs2_mark_node_obsolete(c, f->metadata->raw);
973                         jffs2_free_full_dnode(f->metadata);
974                         f->metadata = NULL;
975                 }
976                 return 0;
977         }
978
979         /* 
980          * We should only get here in the case where the node we are
981          * replacing had more than one frag, so we kept the same version
982          * number as before. (Except in case of error -- see 'goto fill;' 
983          * above.)
984          */
985         D1(if(unlikely(fn->frags <= 1)) {
986                 printk(KERN_WARNING "jffs2_garbage_collect_hole: Replacing fn with %d frag(s) but new ver %d != highest_version %d of ino #%d\n",
987                        fn->frags, je32_to_cpu(ri.version), f->highest_version,
988                        je32_to_cpu(ri.ino));
989         });
990
991         /* This is a partially-overlapped hole node. Mark it REF_NORMAL not REF_PRISTINE */
992         mark_ref_normal(new_fn->raw);
993
994         for (frag = jffs2_lookup_node_frag(&f->fragtree, fn->ofs); 
995              frag; frag = frag_next(frag)) {
996                 if (frag->ofs > fn->size + fn->ofs)
997                         break;
998                 if (frag->node == fn) {
999                         frag->node = new_fn;
1000                         new_fn->frags++;
1001                         fn->frags--;
1002                 }
1003         }
1004         if (fn->frags) {
1005                 printk(KERN_WARNING "jffs2_garbage_collect_hole: Old node still has frags!\n");
1006                 BUG();
1007         }
1008         if (!new_fn->frags) {
1009                 printk(KERN_WARNING "jffs2_garbage_collect_hole: New node has no frags!\n");
1010                 BUG();
1011         }
1012                 
1013         jffs2_mark_node_obsolete(c, fn->raw);
1014         jffs2_free_full_dnode(fn);
1015         
1016         return 0;
1017 }
1018
1019 static int jffs2_garbage_collect_dnode(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
1020                                        struct jffs2_inode_info *f, struct jffs2_full_dnode *fn,
1021                                        uint32_t start, uint32_t end)
1022 {
1023         struct jffs2_full_dnode *new_fn;
1024         struct jffs2_raw_inode ri;
1025         uint32_t alloclen, phys_ofs, offset, orig_end, orig_start;      
1026         int ret = 0;
1027         unsigned char *comprbuf = NULL, *writebuf;
1028         unsigned long pg;
1029         unsigned char *pg_ptr;
1030  
1031         memset(&ri, 0, sizeof(ri));
1032
1033         D1(printk(KERN_DEBUG "Writing replacement dnode for ino #%u from offset 0x%x to 0x%x\n",
1034                   f->inocache->ino, start, end));
1035
1036         orig_end = end;
1037         orig_start = start;
1038
1039         if (c->nr_free_blocks + c->nr_erasing_blocks > c->resv_blocks_gcmerge) {
1040                 /* Attempt to do some merging. But only expand to cover logically
1041                    adjacent frags if the block containing them is already considered
1042                    to be dirty. Otherwise we end up with GC just going round in 
1043                    circles dirtying the nodes it already wrote out, especially 
1044                    on NAND where we have small eraseblocks and hence a much higher
1045                    chance of nodes having to be split to cross boundaries. */
1046
1047                 struct jffs2_node_frag *frag;
1048                 uint32_t min, max;
1049
1050                 min = start & ~(PAGE_CACHE_SIZE-1);
1051                 max = min + PAGE_CACHE_SIZE;
1052
1053                 frag = jffs2_lookup_node_frag(&f->fragtree, start);
1054
1055                 /* BUG_ON(!frag) but that'll happen anyway... */
1056
1057                 BUG_ON(frag->ofs != start);
1058
1059                 /* First grow down... */
1060                 while((frag = frag_prev(frag)) && frag->ofs >= min) {
1061
1062                         /* If the previous frag doesn't even reach the beginning, there's
1063                            excessive fragmentation. Just merge. */
1064                         if (frag->ofs > min) {
1065                                 D1(printk(KERN_DEBUG "Expanding down to cover partial frag (0x%x-0x%x)\n",
1066                                           frag->ofs, frag->ofs+frag->size));
1067                                 start = frag->ofs;
1068                                 continue;
1069                         }
1070                         /* OK. This frag holds the first byte of the page. */
1071                         if (!frag->node || !frag->node->raw) {
1072                                 D1(printk(KERN_DEBUG "First frag in page is hole (0x%x-0x%x). Not expanding down.\n",
1073                                           frag->ofs, frag->ofs+frag->size));
1074                                 break;
1075                         } else {
1076
1077                                 /* OK, it's a frag which extends to the beginning of the page. Does it live 
1078                                    in a block which is still considered clean? If so, don't obsolete it.
1079                                    If not, cover it anyway. */
1080
1081                                 struct jffs2_raw_node_ref *raw = frag->node->raw;
1082                                 struct jffs2_eraseblock *jeb;
1083
1084                                 jeb = &c->blocks[raw->flash_offset / c->sector_size];
1085
1086                                 if (jeb == c->gcblock) {
1087                                         D1(printk(KERN_DEBUG "Expanding down to cover frag (0x%x-0x%x) in gcblock at %08x\n",
1088                                                   frag->ofs, frag->ofs+frag->size, ref_offset(raw)));
1089                                         start = frag->ofs;
1090                                         break;
1091                                 }
1092                                 if (!ISDIRTY(jeb->dirty_size + jeb->wasted_size)) {
1093                                         D1(printk(KERN_DEBUG "Not expanding down to cover frag (0x%x-0x%x) in clean block %08x\n",
1094                                                   frag->ofs, frag->ofs+frag->size, jeb->offset));
1095                                         break;
1096                                 }
1097
1098                                 D1(printk(KERN_DEBUG "Expanding down to cover frag (0x%x-0x%x) in dirty block %08x\n",
1099                                                   frag->ofs, frag->ofs+frag->size, jeb->offset));
1100                                 start = frag->ofs;
1101                                 break;
1102                         }
1103                 }
1104
1105                 /* ... then up */
1106
1107                 /* Find last frag which is actually part of the node we're to GC. */
1108                 frag = jffs2_lookup_node_frag(&f->fragtree, end-1);
1109
1110                 while((frag = frag_next(frag)) && frag->ofs+frag->size <= max) {
1111
1112                         /* If the previous frag doesn't even reach the beginning, there's lots
1113                            of fragmentation. Just merge. */
1114                         if (frag->ofs+frag->size < max) {
1115                                 D1(printk(KERN_DEBUG "Expanding up to cover partial frag (0x%x-0x%x)\n",
1116                                           frag->ofs, frag->ofs+frag->size));
1117                                 end = frag->ofs + frag->size;
1118                                 continue;
1119                         }
1120
1121                         if (!frag->node || !frag->node->raw) {
1122                                 D1(printk(KERN_DEBUG "Last frag in page is hole (0x%x-0x%x). Not expanding up.\n",
1123                                           frag->ofs, frag->ofs+frag->size));
1124                                 break;
1125                         } else {
1126
1127                                 /* OK, it's a frag which extends to the beginning of the page. Does it live 
1128                                    in a block which is still considered clean? If so, don't obsolete it.
1129                                    If not, cover it anyway. */
1130
1131                                 struct jffs2_raw_node_ref *raw = frag->node->raw;
1132                                 struct jffs2_eraseblock *jeb;
1133
1134                                 jeb = &c->blocks[raw->flash_offset / c->sector_size];
1135
1136                                 if (jeb == c->gcblock) {
1137                                         D1(printk(KERN_DEBUG "Expanding up to cover frag (0x%x-0x%x) in gcblock at %08x\n",
1138                                                   frag->ofs, frag->ofs+frag->size, ref_offset(raw)));
1139                                         end = frag->ofs + frag->size;
1140                                         break;
1141                                 }
1142                                 if (!ISDIRTY(jeb->dirty_size + jeb->wasted_size)) {
1143                                         D1(printk(KERN_DEBUG "Not expanding up to cover frag (0x%x-0x%x) in clean block %08x\n",
1144                                                   frag->ofs, frag->ofs+frag->size, jeb->offset));
1145                                         break;
1146                                 }
1147
1148                                 D1(printk(KERN_DEBUG "Expanding up to cover frag (0x%x-0x%x) in dirty block %08x\n",
1149                                                   frag->ofs, frag->ofs+frag->size, jeb->offset));
1150                                 end = frag->ofs + frag->size;
1151                                 break;
1152                         }
1153                 }
1154                 D1(printk(KERN_DEBUG "Expanded dnode to write from (0x%x-0x%x) to (0x%x-0x%x)\n", 
1155                           orig_start, orig_end, start, end));
1156
1157                 BUG_ON(end > JFFS2_F_I_SIZE(f));
1158                 BUG_ON(end < orig_end);
1159                 BUG_ON(start > orig_start);
1160         }
1161         
1162         /* First, use readpage() to read the appropriate page into the page cache */
1163         /* Q: What happens if we actually try to GC the _same_ page for which commit_write()
1164          *    triggered garbage collection in the first place?
1165          * A: I _think_ it's OK. read_cache_page shouldn't deadlock, we'll write out the
1166          *    page OK. We'll actually write it out again in commit_write, which is a little
1167          *    suboptimal, but at least we're correct.
1168          */
1169         pg_ptr = jffs2_gc_fetch_page(c, f, start, &pg);
1170
1171         if (IS_ERR(pg_ptr)) {
1172                 printk(KERN_WARNING "read_cache_page() returned error: %ld\n", PTR_ERR(pg_ptr));
1173                 return PTR_ERR(pg_ptr);
1174         }
1175
1176         offset = start;
1177         while(offset < orig_end) {
1178                 uint32_t datalen;
1179                 uint32_t cdatalen;
1180                 uint16_t comprtype = JFFS2_COMPR_NONE;
1181
1182                 ret = jffs2_reserve_space_gc(c, sizeof(ri) + JFFS2_MIN_DATA_LEN, &phys_ofs, &alloclen);
1183
1184                 if (ret) {
1185                         printk(KERN_WARNING "jffs2_reserve_space_gc of %zd bytes for garbage_collect_dnode failed: %d\n",
1186                                sizeof(ri)+ JFFS2_MIN_DATA_LEN, ret);
1187                         break;
1188                 }
1189                 cdatalen = min_t(uint32_t, alloclen - sizeof(ri), end - offset);
1190                 datalen = end - offset;
1191
1192                 writebuf = pg_ptr + (offset & (PAGE_CACHE_SIZE -1));
1193
1194                 comprtype = jffs2_compress(c, f, writebuf, &comprbuf, &datalen, &cdatalen);
1195
1196                 ri.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK);
1197                 ri.nodetype = cpu_to_je16(JFFS2_NODETYPE_INODE);
1198                 ri.totlen = cpu_to_je32(sizeof(ri) + cdatalen);
1199                 ri.hdr_crc = cpu_to_je32(crc32(0, &ri, sizeof(struct jffs2_unknown_node)-4));
1200
1201                 ri.ino = cpu_to_je32(f->inocache->ino);
1202                 ri.version = cpu_to_je32(++f->highest_version);
1203                 ri.mode = cpu_to_jemode(JFFS2_F_I_MODE(f));
1204                 ri.uid = cpu_to_je16(JFFS2_F_I_UID(f));
1205                 ri.gid = cpu_to_je16(JFFS2_F_I_GID(f));
1206                 ri.isize = cpu_to_je32(JFFS2_F_I_SIZE(f));
1207                 ri.atime = cpu_to_je32(JFFS2_F_I_ATIME(f));
1208                 ri.ctime = cpu_to_je32(JFFS2_F_I_CTIME(f));
1209                 ri.mtime = cpu_to_je32(JFFS2_F_I_MTIME(f));
1210                 ri.offset = cpu_to_je32(offset);
1211                 ri.csize = cpu_to_je32(cdatalen);
1212                 ri.dsize = cpu_to_je32(datalen);
1213                 ri.compr = comprtype & 0xff;
1214                 ri.usercompr = (comprtype >> 8) & 0xff;
1215                 ri.node_crc = cpu_to_je32(crc32(0, &ri, sizeof(ri)-8));
1216                 ri.data_crc = cpu_to_je32(crc32(0, comprbuf, cdatalen));
1217         
1218                 new_fn = jffs2_write_dnode(c, f, &ri, comprbuf, cdatalen, phys_ofs, ALLOC_GC);
1219
1220                 jffs2_free_comprbuf(comprbuf, writebuf);
1221
1222                 if (IS_ERR(new_fn)) {
1223                         printk(KERN_WARNING "Error writing new dnode: %ld\n", PTR_ERR(new_fn));
1224                         ret = PTR_ERR(new_fn);
1225                         break;
1226                 }
1227                 ret = jffs2_add_full_dnode_to_inode(c, f, new_fn);
1228                 offset += datalen;
1229                 if (f->metadata) {
1230                         jffs2_mark_node_obsolete(c, f->metadata->raw);
1231                         jffs2_free_full_dnode(f->metadata);
1232                         f->metadata = NULL;
1233                 }
1234         }
1235
1236         jffs2_gc_release_page(c, pg_ptr, &pg);
1237         return ret;
1238 }
1239