[JFFS2] Prevent list corruption when handling write errors
[safe/jmp/linux-2.6] / fs / jffs2 / nodemgmt.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@infradead.org>
7  *
8  * For licensing information, see the file 'LICENCE' in this directory.
9  *
10  * $Id: nodemgmt.c,v 1.127 2005/09/20 15:49:12 dedekind Exp $
11  *
12  */
13
14 #include <linux/kernel.h>
15 #include <linux/slab.h>
16 #include <linux/mtd/mtd.h>
17 #include <linux/compiler.h>
18 #include <linux/sched.h> /* For cond_resched() */
19 #include "nodelist.h"
20 #include "debug.h"
21
22 /**
23  *      jffs2_reserve_space - request physical space to write nodes to flash
24  *      @c: superblock info
25  *      @minsize: Minimum acceptable size of allocation
26  *      @len: Returned value of allocation length
27  *      @prio: Allocation type - ALLOC_{NORMAL,DELETION}
28  *
29  *      Requests a block of physical space on the flash. Returns zero for success
30  *      and puts 'len' into the appropriate place, or returns -ENOSPC or other 
31  *      error if appropriate. Doesn't return len since that's 
32  *
33  *      If it returns zero, jffs2_reserve_space() also downs the per-filesystem
34  *      allocation semaphore, to prevent more than one allocation from being
35  *      active at any time. The semaphore is later released by jffs2_commit_allocation()
36  *
37  *      jffs2_reserve_space() may trigger garbage collection in order to make room
38  *      for the requested allocation.
39  */
40
41 static int jffs2_do_reserve_space(struct jffs2_sb_info *c,  uint32_t minsize,
42                                   uint32_t *len, uint32_t sumsize);
43
44 int jffs2_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
45                         uint32_t *len, int prio, uint32_t sumsize)
46 {
47         int ret = -EAGAIN;
48         int blocksneeded = c->resv_blocks_write;
49         /* align it */
50         minsize = PAD(minsize);
51
52         D1(printk(KERN_DEBUG "jffs2_reserve_space(): Requested 0x%x bytes\n", minsize));
53         down(&c->alloc_sem);
54
55         D1(printk(KERN_DEBUG "jffs2_reserve_space(): alloc sem got\n"));
56
57         spin_lock(&c->erase_completion_lock);
58
59         /* this needs a little more thought (true <tglx> :)) */
60         while(ret == -EAGAIN) {
61                 while(c->nr_free_blocks + c->nr_erasing_blocks < blocksneeded) {
62                         int ret;
63                         uint32_t dirty, avail;
64
65                         /* calculate real dirty size
66                          * dirty_size contains blocks on erase_pending_list
67                          * those blocks are counted in c->nr_erasing_blocks.
68                          * If one block is actually erased, it is not longer counted as dirty_space
69                          * but it is counted in c->nr_erasing_blocks, so we add it and subtract it
70                          * with c->nr_erasing_blocks * c->sector_size again.
71                          * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks
72                          * This helps us to force gc and pick eventually a clean block to spread the load.
73                          * We add unchecked_size here, as we hopefully will find some space to use.
74                          * This will affect the sum only once, as gc first finishes checking
75                          * of nodes.
76                          */
77                         dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size + c->unchecked_size;
78                         if (dirty < c->nospc_dirty_size) {
79                                 if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) {
80                                         D1(printk(KERN_NOTICE "jffs2_reserve_space(): Low on dirty space to GC, but it's a deletion. Allowing...\n"));
81                                         break;
82                                 }
83                                 D1(printk(KERN_DEBUG "dirty size 0x%08x + unchecked_size 0x%08x < nospc_dirty_size 0x%08x, returning -ENOSPC\n",
84                                           dirty, c->unchecked_size, c->sector_size));
85
86                                 spin_unlock(&c->erase_completion_lock);
87                                 up(&c->alloc_sem);
88                                 return -ENOSPC;
89                         }
90
91                         /* Calc possibly available space. Possibly available means that we
92                          * don't know, if unchecked size contains obsoleted nodes, which could give us some
93                          * more usable space. This will affect the sum only once, as gc first finishes checking
94                          * of nodes.
95                          + Return -ENOSPC, if the maximum possibly available space is less or equal than
96                          * blocksneeded * sector_size.
97                          * This blocks endless gc looping on a filesystem, which is nearly full, even if
98                          * the check above passes.
99                          */
100                         avail = c->free_size + c->dirty_size + c->erasing_size + c->unchecked_size;
101                         if ( (avail / c->sector_size) <= blocksneeded) {
102                                 if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) {
103                                         D1(printk(KERN_NOTICE "jffs2_reserve_space(): Low on possibly available space, but it's a deletion. Allowing...\n"));
104                                         break;
105                                 }
106
107                                 D1(printk(KERN_DEBUG "max. available size 0x%08x  < blocksneeded * sector_size 0x%08x, returning -ENOSPC\n",
108                                           avail, blocksneeded * c->sector_size));
109                                 spin_unlock(&c->erase_completion_lock);
110                                 up(&c->alloc_sem);
111                                 return -ENOSPC;
112                         }
113
114                         up(&c->alloc_sem);
115
116                         D1(printk(KERN_DEBUG "Triggering GC pass. nr_free_blocks %d, nr_erasing_blocks %d, free_size 0x%08x, dirty_size 0x%08x, wasted_size 0x%08x, used_size 0x%08x, erasing_size 0x%08x, bad_size 0x%08x (total 0x%08x of 0x%08x)\n",
117                                   c->nr_free_blocks, c->nr_erasing_blocks, c->free_size, c->dirty_size, c->wasted_size, c->used_size, c->erasing_size, c->bad_size,
118                                   c->free_size + c->dirty_size + c->wasted_size + c->used_size + c->erasing_size + c->bad_size, c->flash_size));
119                         spin_unlock(&c->erase_completion_lock);
120
121                         ret = jffs2_garbage_collect_pass(c);
122                         if (ret)
123                                 return ret;
124
125                         cond_resched();
126
127                         if (signal_pending(current))
128                                 return -EINTR;
129
130                         down(&c->alloc_sem);
131                         spin_lock(&c->erase_completion_lock);
132                 }
133
134                 ret = jffs2_do_reserve_space(c, minsize, len, sumsize);
135                 if (ret) {
136                         D1(printk(KERN_DEBUG "jffs2_reserve_space: ret is %d\n", ret));
137                 }
138         }
139         spin_unlock(&c->erase_completion_lock);
140         if (!ret)
141                 ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1);
142         if (ret)
143                 up(&c->alloc_sem);
144         return ret;
145 }
146
147 int jffs2_reserve_space_gc(struct jffs2_sb_info *c, uint32_t minsize,
148                            uint32_t *len, uint32_t sumsize)
149 {
150         int ret = -EAGAIN;
151         minsize = PAD(minsize);
152
153         D1(printk(KERN_DEBUG "jffs2_reserve_space_gc(): Requested 0x%x bytes\n", minsize));
154
155         spin_lock(&c->erase_completion_lock);
156         while(ret == -EAGAIN) {
157                 ret = jffs2_do_reserve_space(c, minsize, len, sumsize);
158                 if (ret) {
159                         D1(printk(KERN_DEBUG "jffs2_reserve_space_gc: looping, ret is %d\n", ret));
160                 }
161         }
162         spin_unlock(&c->erase_completion_lock);
163         if (!ret)
164                 ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1);
165
166         return ret;
167 }
168
169
170 /* Classify nextblock (clean, dirty of verydirty) and force to select an other one */
171
172 static void jffs2_close_nextblock(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
173 {
174
175         if (c->nextblock == NULL) {
176                 D1(printk(KERN_DEBUG "jffs2_close_nextblock: Erase block at 0x%08x has already been placed in a list\n",
177                   jeb->offset));
178                 return;
179         }
180         /* Check, if we have a dirty block now, or if it was dirty already */
181         if (ISDIRTY (jeb->wasted_size + jeb->dirty_size)) {
182                 c->dirty_size += jeb->wasted_size;
183                 c->wasted_size -= jeb->wasted_size;
184                 jeb->dirty_size += jeb->wasted_size;
185                 jeb->wasted_size = 0;
186                 if (VERYDIRTY(c, jeb->dirty_size)) {
187                         D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to very_dirty_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
188                           jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
189                         list_add_tail(&jeb->list, &c->very_dirty_list);
190                 } else {
191                         D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to dirty_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
192                           jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
193                         list_add_tail(&jeb->list, &c->dirty_list);
194                 }
195         } else {
196                 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
197                   jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
198                 list_add_tail(&jeb->list, &c->clean_list);
199         }
200         c->nextblock = NULL;
201
202 }
203
204 /* Select a new jeb for nextblock */
205
206 static int jffs2_find_nextblock(struct jffs2_sb_info *c)
207 {
208         struct list_head *next;
209
210         /* Take the next block off the 'free' list */
211
212         if (list_empty(&c->free_list)) {
213
214                 if (!c->nr_erasing_blocks &&
215                         !list_empty(&c->erasable_list)) {
216                         struct jffs2_eraseblock *ejeb;
217
218                         ejeb = list_entry(c->erasable_list.next, struct jffs2_eraseblock, list);
219                         list_move_tail(&ejeb->list, &c->erase_pending_list);
220                         c->nr_erasing_blocks++;
221                         jffs2_erase_pending_trigger(c);
222                         D1(printk(KERN_DEBUG "jffs2_find_nextblock: Triggering erase of erasable block at 0x%08x\n",
223                                   ejeb->offset));
224                 }
225
226                 if (!c->nr_erasing_blocks &&
227                         !list_empty(&c->erasable_pending_wbuf_list)) {
228                         D1(printk(KERN_DEBUG "jffs2_find_nextblock: Flushing write buffer\n"));
229                         /* c->nextblock is NULL, no update to c->nextblock allowed */
230                         spin_unlock(&c->erase_completion_lock);
231                         jffs2_flush_wbuf_pad(c);
232                         spin_lock(&c->erase_completion_lock);
233                         /* Have another go. It'll be on the erasable_list now */
234                         return -EAGAIN;
235                 }
236
237                 if (!c->nr_erasing_blocks) {
238                         /* Ouch. We're in GC, or we wouldn't have got here.
239                            And there's no space left. At all. */
240                         printk(KERN_CRIT "Argh. No free space left for GC. nr_erasing_blocks is %d. nr_free_blocks is %d. (erasableempty: %s, erasingempty: %s, erasependingempty: %s)\n",
241                                    c->nr_erasing_blocks, c->nr_free_blocks, list_empty(&c->erasable_list)?"yes":"no",
242                                    list_empty(&c->erasing_list)?"yes":"no", list_empty(&c->erase_pending_list)?"yes":"no");
243                         return -ENOSPC;
244                 }
245
246                 spin_unlock(&c->erase_completion_lock);
247                 /* Don't wait for it; just erase one right now */
248                 jffs2_erase_pending_blocks(c, 1);
249                 spin_lock(&c->erase_completion_lock);
250
251                 /* An erase may have failed, decreasing the
252                    amount of free space available. So we must
253                    restart from the beginning */
254                 return -EAGAIN;
255         }
256
257         next = c->free_list.next;
258         list_del(next);
259         c->nextblock = list_entry(next, struct jffs2_eraseblock, list);
260         c->nr_free_blocks--;
261
262         jffs2_sum_reset_collected(c->summary); /* reset collected summary */
263
264         D1(printk(KERN_DEBUG "jffs2_find_nextblock(): new nextblock = 0x%08x\n", c->nextblock->offset));
265
266         return 0;
267 }
268
269 /* Called with alloc sem _and_ erase_completion_lock */
270 static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
271                                   uint32_t *len, uint32_t sumsize)
272 {
273         struct jffs2_eraseblock *jeb = c->nextblock;
274         uint32_t reserved_size;                         /* for summary information at the end of the jeb */
275         int ret;
276
277  restart:
278         reserved_size = 0;
279
280         if (jffs2_sum_active() && (sumsize != JFFS2_SUMMARY_NOSUM_SIZE)) {
281                                                         /* NOSUM_SIZE means not to generate summary */
282
283                 if (jeb) {
284                         reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE);
285                         dbg_summary("minsize=%d , jeb->free=%d ,"
286                                                 "summary->size=%d , sumsize=%d\n",
287                                                 minsize, jeb->free_size,
288                                                 c->summary->sum_size, sumsize);
289                 }
290
291                 /* Is there enough space for writing out the current node, or we have to
292                    write out summary information now, close this jeb and select new nextblock? */
293                 if (jeb && (PAD(minsize) + PAD(c->summary->sum_size + sumsize +
294                                         JFFS2_SUMMARY_FRAME_SIZE) > jeb->free_size)) {
295
296                         /* Has summary been disabled for this jeb? */
297                         if (jffs2_sum_is_disabled(c->summary)) {
298                                 sumsize = JFFS2_SUMMARY_NOSUM_SIZE;
299                                 goto restart;
300                         }
301
302                         /* Writing out the collected summary information */
303                         dbg_summary("generating summary for 0x%08x.\n", jeb->offset);
304                         ret = jffs2_sum_write_sumnode(c);
305
306                         if (ret)
307                                 return ret;
308
309                         if (jffs2_sum_is_disabled(c->summary)) {
310                                 /* jffs2_write_sumnode() couldn't write out the summary information
311                                    diabling summary for this jeb and free the collected information
312                                  */
313                                 sumsize = JFFS2_SUMMARY_NOSUM_SIZE;
314                                 goto restart;
315                         }
316
317                         jffs2_close_nextblock(c, jeb);
318                         jeb = NULL;
319                         /* keep always valid value in reserved_size */
320                         reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE);
321                 }
322         } else {
323                 if (jeb && minsize > jeb->free_size) {
324                         uint32_t waste;
325
326                         /* Skip the end of this block and file it as having some dirty space */
327                         /* If there's a pending write to it, flush now */
328
329                         if (jffs2_wbuf_dirty(c)) {
330                                 spin_unlock(&c->erase_completion_lock);
331                                 D1(printk(KERN_DEBUG "jffs2_do_reserve_space: Flushing write buffer\n"));
332                                 jffs2_flush_wbuf_pad(c);
333                                 spin_lock(&c->erase_completion_lock);
334                                 jeb = c->nextblock;
335                                 goto restart;
336                         }
337
338                         spin_unlock(&c->erase_completion_lock);
339
340                         ret = jffs2_prealloc_raw_node_refs(c, jeb, 1);
341                         if (ret)
342                                 return ret;
343                         /* Just lock it again and continue. Nothing much can change because
344                            we hold c->alloc_sem anyway. In fact, it's not entirely clear why
345                            we hold c->erase_completion_lock in the majority of this function...
346                            but that's a question for another (more caffeine-rich) day. */
347                         spin_lock(&c->erase_completion_lock);
348
349                         waste = jeb->free_size;
350                         jffs2_link_node_ref(c, jeb,
351                                             (jeb->offset + c->sector_size - waste) | REF_OBSOLETE,
352                                             waste, NULL);
353                         /* FIXME: that made it count as dirty. Convert to wasted */
354                         jeb->dirty_size -= waste;
355                         c->dirty_size -= waste;
356                         jeb->wasted_size += waste;
357                         c->wasted_size += waste;
358
359                         jffs2_close_nextblock(c, jeb);
360                         jeb = NULL;
361                 }
362         }
363
364         if (!jeb) {
365
366                 ret = jffs2_find_nextblock(c);
367                 if (ret)
368                         return ret;
369
370                 jeb = c->nextblock;
371
372                 if (jeb->free_size != c->sector_size - c->cleanmarker_size) {
373                         printk(KERN_WARNING "Eep. Block 0x%08x taken from free_list had free_size of 0x%08x!!\n", jeb->offset, jeb->free_size);
374                         goto restart;
375                 }
376         }
377         /* OK, jeb (==c->nextblock) is now pointing at a block which definitely has
378            enough space */
379         *len = jeb->free_size - reserved_size;
380
381         if (c->cleanmarker_size && jeb->used_size == c->cleanmarker_size &&
382             !jeb->first_node->next_in_ino) {
383                 /* Only node in it beforehand was a CLEANMARKER node (we think).
384                    So mark it obsolete now that there's going to be another node
385                    in the block. This will reduce used_size to zero but We've
386                    already set c->nextblock so that jffs2_mark_node_obsolete()
387                    won't try to refile it to the dirty_list.
388                 */
389                 spin_unlock(&c->erase_completion_lock);
390                 jffs2_mark_node_obsolete(c, jeb->first_node);
391                 spin_lock(&c->erase_completion_lock);
392         }
393
394         D1(printk(KERN_DEBUG "jffs2_do_reserve_space(): Giving 0x%x bytes at 0x%x\n",
395                   *len, jeb->offset + (c->sector_size - jeb->free_size)));
396         return 0;
397 }
398
399 /**
400  *      jffs2_add_physical_node_ref - add a physical node reference to the list
401  *      @c: superblock info
402  *      @new: new node reference to add
403  *      @len: length of this physical node
404  *
405  *      Should only be used to report nodes for which space has been allocated
406  *      by jffs2_reserve_space.
407  *
408  *      Must be called with the alloc_sem held.
409  */
410
411 struct jffs2_raw_node_ref *jffs2_add_physical_node_ref(struct jffs2_sb_info *c,
412                                                        uint32_t ofs, uint32_t len,
413                                                        struct jffs2_inode_cache *ic)
414 {
415         struct jffs2_eraseblock *jeb;
416         struct jffs2_raw_node_ref *new;
417
418         jeb = &c->blocks[ofs / c->sector_size];
419
420         D1(printk(KERN_DEBUG "jffs2_add_physical_node_ref(): Node at 0x%x(%d), size 0x%x\n",
421                   ofs & ~3, ofs & 3, len));
422 #if 1
423         /* Allow non-obsolete nodes only to be added at the end of c->nextblock, 
424            if c->nextblock is set. Note that wbuf.c will file obsolete nodes
425            even after refiling c->nextblock */
426         if ((c->nextblock || ((ofs & 3) != REF_OBSOLETE))
427             && (jeb != c->nextblock || (ofs & ~3) != jeb->offset + (c->sector_size - jeb->free_size))) {
428                 printk(KERN_WARNING "argh. node added in wrong place\n");
429                 return ERR_PTR(-EINVAL);
430         }
431 #endif
432         spin_lock(&c->erase_completion_lock);
433
434         new = jffs2_link_node_ref(c, jeb, ofs, len, ic);
435
436         if (!jeb->free_size && !jeb->dirty_size && !ISDIRTY(jeb->wasted_size)) {
437                 /* If it lives on the dirty_list, jffs2_reserve_space will put it there */
438                 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
439                           jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
440                 if (jffs2_wbuf_dirty(c)) {
441                         /* Flush the last write in the block if it's outstanding */
442                         spin_unlock(&c->erase_completion_lock);
443                         jffs2_flush_wbuf_pad(c);
444                         spin_lock(&c->erase_completion_lock);
445                 }
446
447                 list_add_tail(&jeb->list, &c->clean_list);
448                 c->nextblock = NULL;
449         }
450         jffs2_dbg_acct_sanity_check_nolock(c,jeb);
451         jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
452
453         spin_unlock(&c->erase_completion_lock);
454
455         return new;
456 }
457
458
459 void jffs2_complete_reservation(struct jffs2_sb_info *c)
460 {
461         D1(printk(KERN_DEBUG "jffs2_complete_reservation()\n"));
462         jffs2_garbage_collect_trigger(c);
463         up(&c->alloc_sem);
464 }
465
466 static inline int on_list(struct list_head *obj, struct list_head *head)
467 {
468         struct list_head *this;
469
470         list_for_each(this, head) {
471                 if (this == obj) {
472                         D1(printk("%p is on list at %p\n", obj, head));
473                         return 1;
474
475                 }
476         }
477         return 0;
478 }
479
480 void jffs2_mark_node_obsolete(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref)
481 {
482         struct jffs2_eraseblock *jeb;
483         int blocknr;
484         struct jffs2_unknown_node n;
485         int ret, addedsize;
486         size_t retlen;
487         uint32_t freed_len;
488
489         if(unlikely(!ref)) {
490                 printk(KERN_NOTICE "EEEEEK. jffs2_mark_node_obsolete called with NULL node\n");
491                 return;
492         }
493         if (ref_obsolete(ref)) {
494                 D1(printk(KERN_DEBUG "jffs2_mark_node_obsolete called with already obsolete node at 0x%08x\n", ref_offset(ref)));
495                 return;
496         }
497         blocknr = ref->flash_offset / c->sector_size;
498         if (blocknr >= c->nr_blocks) {
499                 printk(KERN_NOTICE "raw node at 0x%08x is off the end of device!\n", ref->flash_offset);
500                 BUG();
501         }
502         jeb = &c->blocks[blocknr];
503
504         if (jffs2_can_mark_obsolete(c) && !jffs2_is_readonly(c) &&
505             !(c->flags & (JFFS2_SB_FLAG_SCANNING | JFFS2_SB_FLAG_BUILDING))) {
506                 /* Hm. This may confuse static lock analysis. If any of the above
507                    three conditions is false, we're going to return from this
508                    function without actually obliterating any nodes or freeing
509                    any jffs2_raw_node_refs. So we don't need to stop erases from
510                    happening, or protect against people holding an obsolete
511                    jffs2_raw_node_ref without the erase_completion_lock. */
512                 down(&c->erase_free_sem);
513         }
514
515         spin_lock(&c->erase_completion_lock);
516
517         freed_len = ref_totlen(c, jeb, ref);
518
519         if (ref_flags(ref) == REF_UNCHECKED) {
520                 D1(if (unlikely(jeb->unchecked_size < freed_len)) {
521                         printk(KERN_NOTICE "raw unchecked node of size 0x%08x freed from erase block %d at 0x%08x, but unchecked_size was already 0x%08x\n",
522                                freed_len, blocknr, ref->flash_offset, jeb->used_size);
523                         BUG();
524                 })
525                 D1(printk(KERN_DEBUG "Obsoleting previously unchecked node at 0x%08x of len %x: ", ref_offset(ref), freed_len));
526                 jeb->unchecked_size -= freed_len;
527                 c->unchecked_size -= freed_len;
528         } else {
529                 D1(if (unlikely(jeb->used_size < freed_len)) {
530                         printk(KERN_NOTICE "raw node of size 0x%08x freed from erase block %d at 0x%08x, but used_size was already 0x%08x\n",
531                                freed_len, blocknr, ref->flash_offset, jeb->used_size);
532                         BUG();
533                 })
534                 D1(printk(KERN_DEBUG "Obsoleting node at 0x%08x of len %#x: ", ref_offset(ref), freed_len));
535                 jeb->used_size -= freed_len;
536                 c->used_size -= freed_len;
537         }
538
539         // Take care, that wasted size is taken into concern
540         if ((jeb->dirty_size || ISDIRTY(jeb->wasted_size + freed_len)) && jeb != c->nextblock) {
541                 D1(printk("Dirtying\n"));
542                 addedsize = freed_len;
543                 jeb->dirty_size += freed_len;
544                 c->dirty_size += freed_len;
545
546                 /* Convert wasted space to dirty, if not a bad block */
547                 if (jeb->wasted_size) {
548                         if (on_list(&jeb->list, &c->bad_used_list)) {
549                                 D1(printk(KERN_DEBUG "Leaving block at %08x on the bad_used_list\n",
550                                           jeb->offset));
551                                 addedsize = 0; /* To fool the refiling code later */
552                         } else {
553                                 D1(printk(KERN_DEBUG "Converting %d bytes of wasted space to dirty in block at %08x\n",
554                                           jeb->wasted_size, jeb->offset));
555                                 addedsize += jeb->wasted_size;
556                                 jeb->dirty_size += jeb->wasted_size;
557                                 c->dirty_size += jeb->wasted_size;
558                                 c->wasted_size -= jeb->wasted_size;
559                                 jeb->wasted_size = 0;
560                         }
561                 }
562         } else {
563                 D1(printk("Wasting\n"));
564                 addedsize = 0;
565                 jeb->wasted_size += freed_len;
566                 c->wasted_size += freed_len;
567         }
568         ref->flash_offset = ref_offset(ref) | REF_OBSOLETE;
569
570         jffs2_dbg_acct_sanity_check_nolock(c, jeb);
571         jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
572
573         if (c->flags & JFFS2_SB_FLAG_SCANNING) {
574                 /* Flash scanning is in progress. Don't muck about with the block
575                    lists because they're not ready yet, and don't actually
576                    obliterate nodes that look obsolete. If they weren't
577                    marked obsolete on the flash at the time they _became_
578                    obsolete, there was probably a reason for that. */
579                 spin_unlock(&c->erase_completion_lock);
580                 /* We didn't lock the erase_free_sem */
581                 return;
582         }
583
584         if (jeb == c->nextblock) {
585                 D2(printk(KERN_DEBUG "Not moving nextblock 0x%08x to dirty/erase_pending list\n", jeb->offset));
586         } else if (!jeb->used_size && !jeb->unchecked_size) {
587                 if (jeb == c->gcblock) {
588                         D1(printk(KERN_DEBUG "gcblock at 0x%08x completely dirtied. Clearing gcblock...\n", jeb->offset));
589                         c->gcblock = NULL;
590                 } else {
591                         D1(printk(KERN_DEBUG "Eraseblock at 0x%08x completely dirtied. Removing from (dirty?) list...\n", jeb->offset));
592                         list_del(&jeb->list);
593                 }
594                 if (jffs2_wbuf_dirty(c)) {
595                         D1(printk(KERN_DEBUG "...and adding to erasable_pending_wbuf_list\n"));
596                         list_add_tail(&jeb->list, &c->erasable_pending_wbuf_list);
597                 } else {
598                         if (jiffies & 127) {
599                                 /* Most of the time, we just erase it immediately. Otherwise we
600                                    spend ages scanning it on mount, etc. */
601                                 D1(printk(KERN_DEBUG "...and adding to erase_pending_list\n"));
602                                 list_add_tail(&jeb->list, &c->erase_pending_list);
603                                 c->nr_erasing_blocks++;
604                                 jffs2_erase_pending_trigger(c);
605                         } else {
606                                 /* Sometimes, however, we leave it elsewhere so it doesn't get
607                                    immediately reused, and we spread the load a bit. */
608                                 D1(printk(KERN_DEBUG "...and adding to erasable_list\n"));
609                                 list_add_tail(&jeb->list, &c->erasable_list);
610                         }
611                 }
612                 D1(printk(KERN_DEBUG "Done OK\n"));
613         } else if (jeb == c->gcblock) {
614                 D2(printk(KERN_DEBUG "Not moving gcblock 0x%08x to dirty_list\n", jeb->offset));
615         } else if (ISDIRTY(jeb->dirty_size) && !ISDIRTY(jeb->dirty_size - addedsize)) {
616                 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x is freshly dirtied. Removing from clean list...\n", jeb->offset));
617                 list_del(&jeb->list);
618                 D1(printk(KERN_DEBUG "...and adding to dirty_list\n"));
619                 list_add_tail(&jeb->list, &c->dirty_list);
620         } else if (VERYDIRTY(c, jeb->dirty_size) &&
621                    !VERYDIRTY(c, jeb->dirty_size - addedsize)) {
622                 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x is now very dirty. Removing from dirty list...\n", jeb->offset));
623                 list_del(&jeb->list);
624                 D1(printk(KERN_DEBUG "...and adding to very_dirty_list\n"));
625                 list_add_tail(&jeb->list, &c->very_dirty_list);
626         } else {
627                 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x not moved anywhere. (free 0x%08x, dirty 0x%08x, used 0x%08x)\n",
628                           jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
629         }
630
631         spin_unlock(&c->erase_completion_lock);
632
633         if (!jffs2_can_mark_obsolete(c) || jffs2_is_readonly(c) ||
634                 (c->flags & JFFS2_SB_FLAG_BUILDING)) {
635                 /* We didn't lock the erase_free_sem */
636                 return;
637         }
638
639         /* The erase_free_sem is locked, and has been since before we marked the node obsolete
640            and potentially put its eraseblock onto the erase_pending_list. Thus, we know that
641            the block hasn't _already_ been erased, and that 'ref' itself hasn't been freed yet
642            by jffs2_free_jeb_node_refs() in erase.c. Which is nice. */
643
644         D1(printk(KERN_DEBUG "obliterating obsoleted node at 0x%08x\n", ref_offset(ref)));
645         ret = jffs2_flash_read(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
646         if (ret) {
647                 printk(KERN_WARNING "Read error reading from obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret);
648                 goto out_erase_sem;
649         }
650         if (retlen != sizeof(n)) {
651                 printk(KERN_WARNING "Short read from obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen);
652                 goto out_erase_sem;
653         }
654         if (PAD(je32_to_cpu(n.totlen)) != PAD(freed_len)) {
655                 printk(KERN_WARNING "Node totlen on flash (0x%08x) != totlen from node ref (0x%08x)\n", je32_to_cpu(n.totlen), freed_len);
656                 goto out_erase_sem;
657         }
658         if (!(je16_to_cpu(n.nodetype) & JFFS2_NODE_ACCURATE)) {
659                 D1(printk(KERN_DEBUG "Node at 0x%08x was already marked obsolete (nodetype 0x%04x)\n", ref_offset(ref), je16_to_cpu(n.nodetype)));
660                 goto out_erase_sem;
661         }
662         /* XXX FIXME: This is ugly now */
663         n.nodetype = cpu_to_je16(je16_to_cpu(n.nodetype) & ~JFFS2_NODE_ACCURATE);
664         ret = jffs2_flash_write(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
665         if (ret) {
666                 printk(KERN_WARNING "Write error in obliterating obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret);
667                 goto out_erase_sem;
668         }
669         if (retlen != sizeof(n)) {
670                 printk(KERN_WARNING "Short write in obliterating obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen);
671                 goto out_erase_sem;
672         }
673
674         /* Nodes which have been marked obsolete no longer need to be
675            associated with any inode. Remove them from the per-inode list.
676
677            Note we can't do this for NAND at the moment because we need
678            obsolete dirent nodes to stay on the lists, because of the
679            horridness in jffs2_garbage_collect_deletion_dirent(). Also
680            because we delete the inocache, and on NAND we need that to
681            stay around until all the nodes are actually erased, in order
682            to stop us from giving the same inode number to another newly
683            created inode. */
684         if (ref->next_in_ino) {
685                 struct jffs2_inode_cache *ic;
686                 struct jffs2_raw_node_ref **p;
687
688                 spin_lock(&c->erase_completion_lock);
689
690                 ic = jffs2_raw_ref_to_ic(ref);
691                 for (p = &ic->nodes; (*p) != ref; p = &((*p)->next_in_ino))
692                         ;
693
694                 *p = ref->next_in_ino;
695                 ref->next_in_ino = NULL;
696
697                 switch (ic->class) {
698 #ifdef CONFIG_JFFS2_FS_XATTR
699                         case RAWNODE_CLASS_XATTR_DATUM:
700                                 jffs2_release_xattr_datum(c, (struct jffs2_xattr_datum *)ic);
701                                 break;
702                         case RAWNODE_CLASS_XATTR_REF:
703                                 jffs2_release_xattr_ref(c, (struct jffs2_xattr_ref *)ic);
704                                 break;
705 #endif
706                         default:
707                                 if (ic->nodes == (void *)ic && ic->nlink == 0)
708                                         jffs2_del_ino_cache(c, ic);
709                                 break;
710                 }
711                 spin_unlock(&c->erase_completion_lock);
712         }
713
714  out_erase_sem:
715         up(&c->erase_free_sem);
716 }
717
718 int jffs2_thread_should_wake(struct jffs2_sb_info *c)
719 {
720         int ret = 0;
721         uint32_t dirty;
722
723         if (c->unchecked_size) {
724                 D1(printk(KERN_DEBUG "jffs2_thread_should_wake(): unchecked_size %d, checked_ino #%d\n",
725                           c->unchecked_size, c->checked_ino));
726                 return 1;
727         }
728
729         /* dirty_size contains blocks on erase_pending_list
730          * those blocks are counted in c->nr_erasing_blocks.
731          * If one block is actually erased, it is not longer counted as dirty_space
732          * but it is counted in c->nr_erasing_blocks, so we add it and subtract it
733          * with c->nr_erasing_blocks * c->sector_size again.
734          * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks
735          * This helps us to force gc and pick eventually a clean block to spread the load.
736          */
737         dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size;
738
739         if (c->nr_free_blocks + c->nr_erasing_blocks < c->resv_blocks_gctrigger &&
740                         (dirty > c->nospc_dirty_size))
741                 ret = 1;
742
743         D1(printk(KERN_DEBUG "jffs2_thread_should_wake(): nr_free_blocks %d, nr_erasing_blocks %d, dirty_size 0x%x: %s\n",
744                   c->nr_free_blocks, c->nr_erasing_blocks, c->dirty_size, ret?"yes":"no"));
745
746         return ret;
747 }