Linux-libre 3.0.78-gnu1
[librecmc/linux-libre.git] / fs / jffs2 / nodemgmt.c
1 /*
2  * JFFS2 -- Journalling Flash File System, Version 2.
3  *
4  * Copyright © 2001-2007 Red Hat, Inc.
5  *
6  * Created by David Woodhouse <dwmw2@infradead.org>
7  *
8  * For licensing information, see the file 'LICENCE' in this directory.
9  *
10  */
11
12 #include <linux/kernel.h>
13 #include <linux/mtd/mtd.h>
14 #include <linux/compiler.h>
15 #include <linux/sched.h> /* For cond_resched() */
16 #include "nodelist.h"
17 #include "debug.h"
18
19 /**
20  *      jffs2_reserve_space - request physical space to write nodes to flash
21  *      @c: superblock info
22  *      @minsize: Minimum acceptable size of allocation
23  *      @len: Returned value of allocation length
24  *      @prio: Allocation type - ALLOC_{NORMAL,DELETION}
25  *
26  *      Requests a block of physical space on the flash. Returns zero for success
27  *      and puts 'len' into the appropriate place, or returns -ENOSPC or other 
28  *      error if appropriate. Doesn't return len since that's 
29  *
30  *      If it returns zero, jffs2_reserve_space() also downs the per-filesystem
31  *      allocation semaphore, to prevent more than one allocation from being
32  *      active at any time. The semaphore is later released by jffs2_commit_allocation()
33  *
34  *      jffs2_reserve_space() may trigger garbage collection in order to make room
35  *      for the requested allocation.
36  */
37
38 static int jffs2_do_reserve_space(struct jffs2_sb_info *c,  uint32_t minsize,
39                                   uint32_t *len, uint32_t sumsize);
40
41 int jffs2_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
42                         uint32_t *len, int prio, uint32_t sumsize)
43 {
44         int ret = -EAGAIN;
45         int blocksneeded = c->resv_blocks_write;
46         /* align it */
47         minsize = PAD(minsize);
48
49         D1(printk(KERN_DEBUG "jffs2_reserve_space(): Requested 0x%x bytes\n", minsize));
50         mutex_lock(&c->alloc_sem);
51
52         D1(printk(KERN_DEBUG "jffs2_reserve_space(): alloc sem got\n"));
53
54         spin_lock(&c->erase_completion_lock);
55
56         /* this needs a little more thought (true <tglx> :)) */
57         while(ret == -EAGAIN) {
58                 while(c->nr_free_blocks + c->nr_erasing_blocks < blocksneeded) {
59                         uint32_t dirty, avail;
60
61                         /* calculate real dirty size
62                          * dirty_size contains blocks on erase_pending_list
63                          * those blocks are counted in c->nr_erasing_blocks.
64                          * If one block is actually erased, it is not longer counted as dirty_space
65                          * but it is counted in c->nr_erasing_blocks, so we add it and subtract it
66                          * with c->nr_erasing_blocks * c->sector_size again.
67                          * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks
68                          * This helps us to force gc and pick eventually a clean block to spread the load.
69                          * We add unchecked_size here, as we hopefully will find some space to use.
70                          * This will affect the sum only once, as gc first finishes checking
71                          * of nodes.
72                          */
73                         dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size + c->unchecked_size;
74                         if (dirty < c->nospc_dirty_size) {
75                                 if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) {
76                                         D1(printk(KERN_NOTICE "jffs2_reserve_space(): Low on dirty space to GC, but it's a deletion. Allowing...\n"));
77                                         break;
78                                 }
79                                 D1(printk(KERN_DEBUG "dirty size 0x%08x + unchecked_size 0x%08x < nospc_dirty_size 0x%08x, returning -ENOSPC\n",
80                                           dirty, c->unchecked_size, c->sector_size));
81
82                                 spin_unlock(&c->erase_completion_lock);
83                                 mutex_unlock(&c->alloc_sem);
84                                 return -ENOSPC;
85                         }
86
87                         /* Calc possibly available space. Possibly available means that we
88                          * don't know, if unchecked size contains obsoleted nodes, which could give us some
89                          * more usable space. This will affect the sum only once, as gc first finishes checking
90                          * of nodes.
91                          + Return -ENOSPC, if the maximum possibly available space is less or equal than
92                          * blocksneeded * sector_size.
93                          * This blocks endless gc looping on a filesystem, which is nearly full, even if
94                          * the check above passes.
95                          */
96                         avail = c->free_size + c->dirty_size + c->erasing_size + c->unchecked_size;
97                         if ( (avail / c->sector_size) <= blocksneeded) {
98                                 if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) {
99                                         D1(printk(KERN_NOTICE "jffs2_reserve_space(): Low on possibly available space, but it's a deletion. Allowing...\n"));
100                                         break;
101                                 }
102
103                                 D1(printk(KERN_DEBUG "max. available size 0x%08x  < blocksneeded * sector_size 0x%08x, returning -ENOSPC\n",
104                                           avail, blocksneeded * c->sector_size));
105                                 spin_unlock(&c->erase_completion_lock);
106                                 mutex_unlock(&c->alloc_sem);
107                                 return -ENOSPC;
108                         }
109
110                         mutex_unlock(&c->alloc_sem);
111
112                         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",
113                                   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,
114                                   c->free_size + c->dirty_size + c->wasted_size + c->used_size + c->erasing_size + c->bad_size, c->flash_size));
115                         spin_unlock(&c->erase_completion_lock);
116
117                         ret = jffs2_garbage_collect_pass(c);
118
119                         if (ret == -EAGAIN) {
120                                 spin_lock(&c->erase_completion_lock);
121                                 if (c->nr_erasing_blocks &&
122                                     list_empty(&c->erase_pending_list) &&
123                                     list_empty(&c->erase_complete_list)) {
124                                         DECLARE_WAITQUEUE(wait, current);
125                                         set_current_state(TASK_UNINTERRUPTIBLE);
126                                         add_wait_queue(&c->erase_wait, &wait);
127                                         D1(printk(KERN_DEBUG "%s waiting for erase to complete\n", __func__));
128                                         spin_unlock(&c->erase_completion_lock);
129
130                                         schedule();
131                                 } else
132                                         spin_unlock(&c->erase_completion_lock);
133                         } else if (ret)
134                                 return ret;
135
136                         cond_resched();
137
138                         if (signal_pending(current))
139                                 return -EINTR;
140
141                         mutex_lock(&c->alloc_sem);
142                         spin_lock(&c->erase_completion_lock);
143                 }
144
145                 ret = jffs2_do_reserve_space(c, minsize, len, sumsize);
146                 if (ret) {
147                         D1(printk(KERN_DEBUG "jffs2_reserve_space: ret is %d\n", ret));
148                 }
149         }
150         spin_unlock(&c->erase_completion_lock);
151         if (!ret)
152                 ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1);
153         if (ret)
154                 mutex_unlock(&c->alloc_sem);
155         return ret;
156 }
157
158 int jffs2_reserve_space_gc(struct jffs2_sb_info *c, uint32_t minsize,
159                            uint32_t *len, uint32_t sumsize)
160 {
161         int ret = -EAGAIN;
162         minsize = PAD(minsize);
163
164         D1(printk(KERN_DEBUG "jffs2_reserve_space_gc(): Requested 0x%x bytes\n", minsize));
165
166         spin_lock(&c->erase_completion_lock);
167         while(ret == -EAGAIN) {
168                 ret = jffs2_do_reserve_space(c, minsize, len, sumsize);
169                 if (ret) {
170                         D1(printk(KERN_DEBUG "jffs2_reserve_space_gc: looping, ret is %d\n", ret));
171                 }
172         }
173         spin_unlock(&c->erase_completion_lock);
174         if (!ret)
175                 ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1);
176
177         return ret;
178 }
179
180
181 /* Classify nextblock (clean, dirty of verydirty) and force to select an other one */
182
183 static void jffs2_close_nextblock(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
184 {
185
186         if (c->nextblock == NULL) {
187                 D1(printk(KERN_DEBUG "jffs2_close_nextblock: Erase block at 0x%08x has already been placed in a list\n",
188                   jeb->offset));
189                 return;
190         }
191         /* Check, if we have a dirty block now, or if it was dirty already */
192         if (ISDIRTY (jeb->wasted_size + jeb->dirty_size)) {
193                 c->dirty_size += jeb->wasted_size;
194                 c->wasted_size -= jeb->wasted_size;
195                 jeb->dirty_size += jeb->wasted_size;
196                 jeb->wasted_size = 0;
197                 if (VERYDIRTY(c, jeb->dirty_size)) {
198                         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",
199                           jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
200                         list_add_tail(&jeb->list, &c->very_dirty_list);
201                 } else {
202                         D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to dirty_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
203                           jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
204                         list_add_tail(&jeb->list, &c->dirty_list);
205                 }
206         } else {
207                 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
208                   jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
209                 list_add_tail(&jeb->list, &c->clean_list);
210         }
211         c->nextblock = NULL;
212
213 }
214
215 /* Select a new jeb for nextblock */
216
217 static int jffs2_find_nextblock(struct jffs2_sb_info *c)
218 {
219         struct list_head *next;
220
221         /* Take the next block off the 'free' list */
222
223         if (list_empty(&c->free_list)) {
224
225                 if (!c->nr_erasing_blocks &&
226                         !list_empty(&c->erasable_list)) {
227                         struct jffs2_eraseblock *ejeb;
228
229                         ejeb = list_entry(c->erasable_list.next, struct jffs2_eraseblock, list);
230                         list_move_tail(&ejeb->list, &c->erase_pending_list);
231                         c->nr_erasing_blocks++;
232                         jffs2_garbage_collect_trigger(c);
233                         D1(printk(KERN_DEBUG "jffs2_find_nextblock: Triggering erase of erasable block at 0x%08x\n",
234                                   ejeb->offset));
235                 }
236
237                 if (!c->nr_erasing_blocks &&
238                         !list_empty(&c->erasable_pending_wbuf_list)) {
239                         D1(printk(KERN_DEBUG "jffs2_find_nextblock: Flushing write buffer\n"));
240                         /* c->nextblock is NULL, no update to c->nextblock allowed */
241                         spin_unlock(&c->erase_completion_lock);
242                         jffs2_flush_wbuf_pad(c);
243                         spin_lock(&c->erase_completion_lock);
244                         /* Have another go. It'll be on the erasable_list now */
245                         return -EAGAIN;
246                 }
247
248                 if (!c->nr_erasing_blocks) {
249                         /* Ouch. We're in GC, or we wouldn't have got here.
250                            And there's no space left. At all. */
251                         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",
252                                    c->nr_erasing_blocks, c->nr_free_blocks, list_empty(&c->erasable_list)?"yes":"no",
253                                    list_empty(&c->erasing_list)?"yes":"no", list_empty(&c->erase_pending_list)?"yes":"no");
254                         return -ENOSPC;
255                 }
256
257                 spin_unlock(&c->erase_completion_lock);
258                 /* Don't wait for it; just erase one right now */
259                 jffs2_erase_pending_blocks(c, 1);
260                 spin_lock(&c->erase_completion_lock);
261
262                 /* An erase may have failed, decreasing the
263                    amount of free space available. So we must
264                    restart from the beginning */
265                 return -EAGAIN;
266         }
267
268         next = c->free_list.next;
269         list_del(next);
270         c->nextblock = list_entry(next, struct jffs2_eraseblock, list);
271         c->nr_free_blocks--;
272
273         jffs2_sum_reset_collected(c->summary); /* reset collected summary */
274
275 #ifdef CONFIG_JFFS2_FS_WRITEBUFFER
276         /* adjust write buffer offset, else we get a non contiguous write bug */
277         if (!(c->wbuf_ofs % c->sector_size) && !c->wbuf_len)
278                 c->wbuf_ofs = 0xffffffff;
279 #endif
280
281         D1(printk(KERN_DEBUG "jffs2_find_nextblock(): new nextblock = 0x%08x\n", c->nextblock->offset));
282
283         return 0;
284 }
285
286 /* Called with alloc sem _and_ erase_completion_lock */
287 static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
288                                   uint32_t *len, uint32_t sumsize)
289 {
290         struct jffs2_eraseblock *jeb = c->nextblock;
291         uint32_t reserved_size;                         /* for summary information at the end of the jeb */
292         int ret;
293
294  restart:
295         reserved_size = 0;
296
297         if (jffs2_sum_active() && (sumsize != JFFS2_SUMMARY_NOSUM_SIZE)) {
298                                                         /* NOSUM_SIZE means not to generate summary */
299
300                 if (jeb) {
301                         reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE);
302                         dbg_summary("minsize=%d , jeb->free=%d ,"
303                                                 "summary->size=%d , sumsize=%d\n",
304                                                 minsize, jeb->free_size,
305                                                 c->summary->sum_size, sumsize);
306                 }
307
308                 /* Is there enough space for writing out the current node, or we have to
309                    write out summary information now, close this jeb and select new nextblock? */
310                 if (jeb && (PAD(minsize) + PAD(c->summary->sum_size + sumsize +
311                                         JFFS2_SUMMARY_FRAME_SIZE) > jeb->free_size)) {
312
313                         /* Has summary been disabled for this jeb? */
314                         if (jffs2_sum_is_disabled(c->summary)) {
315                                 sumsize = JFFS2_SUMMARY_NOSUM_SIZE;
316                                 goto restart;
317                         }
318
319                         /* Writing out the collected summary information */
320                         dbg_summary("generating summary for 0x%08x.\n", jeb->offset);
321                         ret = jffs2_sum_write_sumnode(c);
322
323                         if (ret)
324                                 return ret;
325
326                         if (jffs2_sum_is_disabled(c->summary)) {
327                                 /* jffs2_write_sumnode() couldn't write out the summary information
328                                    diabling summary for this jeb and free the collected information
329                                  */
330                                 sumsize = JFFS2_SUMMARY_NOSUM_SIZE;
331                                 goto restart;
332                         }
333
334                         jffs2_close_nextblock(c, jeb);
335                         jeb = NULL;
336                         /* keep always valid value in reserved_size */
337                         reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE);
338                 }
339         } else {
340                 if (jeb && minsize > jeb->free_size) {
341                         uint32_t waste;
342
343                         /* Skip the end of this block and file it as having some dirty space */
344                         /* If there's a pending write to it, flush now */
345
346                         if (jffs2_wbuf_dirty(c)) {
347                                 spin_unlock(&c->erase_completion_lock);
348                                 D1(printk(KERN_DEBUG "jffs2_do_reserve_space: Flushing write buffer\n"));
349                                 jffs2_flush_wbuf_pad(c);
350                                 spin_lock(&c->erase_completion_lock);
351                                 jeb = c->nextblock;
352                                 goto restart;
353                         }
354
355                         spin_unlock(&c->erase_completion_lock);
356
357                         ret = jffs2_prealloc_raw_node_refs(c, jeb, 1);
358
359                         /* Just lock it again and continue. Nothing much can change because
360                            we hold c->alloc_sem anyway. In fact, it's not entirely clear why
361                            we hold c->erase_completion_lock in the majority of this function...
362                            but that's a question for another (more caffeine-rich) day. */
363                         spin_lock(&c->erase_completion_lock);
364
365                         if (ret)
366                                 return ret;
367
368                         waste = jeb->free_size;
369                         jffs2_link_node_ref(c, jeb,
370                                             (jeb->offset + c->sector_size - waste) | REF_OBSOLETE,
371                                             waste, NULL);
372                         /* FIXME: that made it count as dirty. Convert to wasted */
373                         jeb->dirty_size -= waste;
374                         c->dirty_size -= waste;
375                         jeb->wasted_size += waste;
376                         c->wasted_size += waste;
377
378                         jffs2_close_nextblock(c, jeb);
379                         jeb = NULL;
380                 }
381         }
382
383         if (!jeb) {
384
385                 ret = jffs2_find_nextblock(c);
386                 if (ret)
387                         return ret;
388
389                 jeb = c->nextblock;
390
391                 if (jeb->free_size != c->sector_size - c->cleanmarker_size) {
392                         printk(KERN_WARNING "Eep. Block 0x%08x taken from free_list had free_size of 0x%08x!!\n", jeb->offset, jeb->free_size);
393                         goto restart;
394                 }
395         }
396         /* OK, jeb (==c->nextblock) is now pointing at a block which definitely has
397            enough space */
398         *len = jeb->free_size - reserved_size;
399
400         if (c->cleanmarker_size && jeb->used_size == c->cleanmarker_size &&
401             !jeb->first_node->next_in_ino) {
402                 /* Only node in it beforehand was a CLEANMARKER node (we think).
403                    So mark it obsolete now that there's going to be another node
404                    in the block. This will reduce used_size to zero but We've
405                    already set c->nextblock so that jffs2_mark_node_obsolete()
406                    won't try to refile it to the dirty_list.
407                 */
408                 spin_unlock(&c->erase_completion_lock);
409                 jffs2_mark_node_obsolete(c, jeb->first_node);
410                 spin_lock(&c->erase_completion_lock);
411         }
412
413         D1(printk(KERN_DEBUG "jffs2_do_reserve_space(): Giving 0x%x bytes at 0x%x\n",
414                   *len, jeb->offset + (c->sector_size - jeb->free_size)));
415         return 0;
416 }
417
418 /**
419  *      jffs2_add_physical_node_ref - add a physical node reference to the list
420  *      @c: superblock info
421  *      @new: new node reference to add
422  *      @len: length of this physical node
423  *
424  *      Should only be used to report nodes for which space has been allocated
425  *      by jffs2_reserve_space.
426  *
427  *      Must be called with the alloc_sem held.
428  */
429
430 struct jffs2_raw_node_ref *jffs2_add_physical_node_ref(struct jffs2_sb_info *c,
431                                                        uint32_t ofs, uint32_t len,
432                                                        struct jffs2_inode_cache *ic)
433 {
434         struct jffs2_eraseblock *jeb;
435         struct jffs2_raw_node_ref *new;
436
437         jeb = &c->blocks[ofs / c->sector_size];
438
439         D1(printk(KERN_DEBUG "jffs2_add_physical_node_ref(): Node at 0x%x(%d), size 0x%x\n",
440                   ofs & ~3, ofs & 3, len));
441 #if 1
442         /* Allow non-obsolete nodes only to be added at the end of c->nextblock, 
443            if c->nextblock is set. Note that wbuf.c will file obsolete nodes
444            even after refiling c->nextblock */
445         if ((c->nextblock || ((ofs & 3) != REF_OBSOLETE))
446             && (jeb != c->nextblock || (ofs & ~3) != jeb->offset + (c->sector_size - jeb->free_size))) {
447                 printk(KERN_WARNING "argh. node added in wrong place at 0x%08x(%d)\n", ofs & ~3, ofs & 3);
448                 if (c->nextblock)
449                         printk(KERN_WARNING "nextblock 0x%08x", c->nextblock->offset);
450                 else
451                         printk(KERN_WARNING "No nextblock");
452                 printk(", expected at %08x\n", jeb->offset + (c->sector_size - jeb->free_size));
453                 return ERR_PTR(-EINVAL);
454         }
455 #endif
456         spin_lock(&c->erase_completion_lock);
457
458         new = jffs2_link_node_ref(c, jeb, ofs, len, ic);
459
460         if (!jeb->free_size && !jeb->dirty_size && !ISDIRTY(jeb->wasted_size)) {
461                 /* If it lives on the dirty_list, jffs2_reserve_space will put it there */
462                 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
463                           jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
464                 if (jffs2_wbuf_dirty(c)) {
465                         /* Flush the last write in the block if it's outstanding */
466                         spin_unlock(&c->erase_completion_lock);
467                         jffs2_flush_wbuf_pad(c);
468                         spin_lock(&c->erase_completion_lock);
469                 }
470
471                 list_add_tail(&jeb->list, &c->clean_list);
472                 c->nextblock = NULL;
473         }
474         jffs2_dbg_acct_sanity_check_nolock(c,jeb);
475         jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
476
477         spin_unlock(&c->erase_completion_lock);
478
479         return new;
480 }
481
482
483 void jffs2_complete_reservation(struct jffs2_sb_info *c)
484 {
485         D1(printk(KERN_DEBUG "jffs2_complete_reservation()\n"));
486         spin_lock(&c->erase_completion_lock);
487         jffs2_garbage_collect_trigger(c);
488         spin_unlock(&c->erase_completion_lock);
489         mutex_unlock(&c->alloc_sem);
490 }
491
492 static inline int on_list(struct list_head *obj, struct list_head *head)
493 {
494         struct list_head *this;
495
496         list_for_each(this, head) {
497                 if (this == obj) {
498                         D1(printk("%p is on list at %p\n", obj, head));
499                         return 1;
500
501                 }
502         }
503         return 0;
504 }
505
506 void jffs2_mark_node_obsolete(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref)
507 {
508         struct jffs2_eraseblock *jeb;
509         int blocknr;
510         struct jffs2_unknown_node n;
511         int ret, addedsize;
512         size_t retlen;
513         uint32_t freed_len;
514
515         if(unlikely(!ref)) {
516                 printk(KERN_NOTICE "EEEEEK. jffs2_mark_node_obsolete called with NULL node\n");
517                 return;
518         }
519         if (ref_obsolete(ref)) {
520                 D1(printk(KERN_DEBUG "jffs2_mark_node_obsolete called with already obsolete node at 0x%08x\n", ref_offset(ref)));
521                 return;
522         }
523         blocknr = ref->flash_offset / c->sector_size;
524         if (blocknr >= c->nr_blocks) {
525                 printk(KERN_NOTICE "raw node at 0x%08x is off the end of device!\n", ref->flash_offset);
526                 BUG();
527         }
528         jeb = &c->blocks[blocknr];
529
530         if (jffs2_can_mark_obsolete(c) && !jffs2_is_readonly(c) &&
531             !(c->flags & (JFFS2_SB_FLAG_SCANNING | JFFS2_SB_FLAG_BUILDING))) {
532                 /* Hm. This may confuse static lock analysis. If any of the above
533                    three conditions is false, we're going to return from this
534                    function without actually obliterating any nodes or freeing
535                    any jffs2_raw_node_refs. So we don't need to stop erases from
536                    happening, or protect against people holding an obsolete
537                    jffs2_raw_node_ref without the erase_completion_lock. */
538                 mutex_lock(&c->erase_free_sem);
539         }
540
541         spin_lock(&c->erase_completion_lock);
542
543         freed_len = ref_totlen(c, jeb, ref);
544
545         if (ref_flags(ref) == REF_UNCHECKED) {
546                 D1(if (unlikely(jeb->unchecked_size < freed_len)) {
547                         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",
548                                freed_len, blocknr, ref->flash_offset, jeb->used_size);
549                         BUG();
550                 })
551                 D1(printk(KERN_DEBUG "Obsoleting previously unchecked node at 0x%08x of len %x: ", ref_offset(ref), freed_len));
552                 jeb->unchecked_size -= freed_len;
553                 c->unchecked_size -= freed_len;
554         } else {
555                 D1(if (unlikely(jeb->used_size < freed_len)) {
556                         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",
557                                freed_len, blocknr, ref->flash_offset, jeb->used_size);
558                         BUG();
559                 })
560                 D1(printk(KERN_DEBUG "Obsoleting node at 0x%08x of len %#x: ", ref_offset(ref), freed_len));
561                 jeb->used_size -= freed_len;
562                 c->used_size -= freed_len;
563         }
564
565         // Take care, that wasted size is taken into concern
566         if ((jeb->dirty_size || ISDIRTY(jeb->wasted_size + freed_len)) && jeb != c->nextblock) {
567                 D1(printk("Dirtying\n"));
568                 addedsize = freed_len;
569                 jeb->dirty_size += freed_len;
570                 c->dirty_size += freed_len;
571
572                 /* Convert wasted space to dirty, if not a bad block */
573                 if (jeb->wasted_size) {
574                         if (on_list(&jeb->list, &c->bad_used_list)) {
575                                 D1(printk(KERN_DEBUG "Leaving block at %08x on the bad_used_list\n",
576                                           jeb->offset));
577                                 addedsize = 0; /* To fool the refiling code later */
578                         } else {
579                                 D1(printk(KERN_DEBUG "Converting %d bytes of wasted space to dirty in block at %08x\n",
580                                           jeb->wasted_size, jeb->offset));
581                                 addedsize += jeb->wasted_size;
582                                 jeb->dirty_size += jeb->wasted_size;
583                                 c->dirty_size += jeb->wasted_size;
584                                 c->wasted_size -= jeb->wasted_size;
585                                 jeb->wasted_size = 0;
586                         }
587                 }
588         } else {
589                 D1(printk("Wasting\n"));
590                 addedsize = 0;
591                 jeb->wasted_size += freed_len;
592                 c->wasted_size += freed_len;
593         }
594         ref->flash_offset = ref_offset(ref) | REF_OBSOLETE;
595
596         jffs2_dbg_acct_sanity_check_nolock(c, jeb);
597         jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
598
599         if (c->flags & JFFS2_SB_FLAG_SCANNING) {
600                 /* Flash scanning is in progress. Don't muck about with the block
601                    lists because they're not ready yet, and don't actually
602                    obliterate nodes that look obsolete. If they weren't
603                    marked obsolete on the flash at the time they _became_
604                    obsolete, there was probably a reason for that. */
605                 spin_unlock(&c->erase_completion_lock);
606                 /* We didn't lock the erase_free_sem */
607                 return;
608         }
609
610         if (jeb == c->nextblock) {
611                 D2(printk(KERN_DEBUG "Not moving nextblock 0x%08x to dirty/erase_pending list\n", jeb->offset));
612         } else if (!jeb->used_size && !jeb->unchecked_size) {
613                 if (jeb == c->gcblock) {
614                         D1(printk(KERN_DEBUG "gcblock at 0x%08x completely dirtied. Clearing gcblock...\n", jeb->offset));
615                         c->gcblock = NULL;
616                 } else {
617                         D1(printk(KERN_DEBUG "Eraseblock at 0x%08x completely dirtied. Removing from (dirty?) list...\n", jeb->offset));
618                         list_del(&jeb->list);
619                 }
620                 if (jffs2_wbuf_dirty(c)) {
621                         D1(printk(KERN_DEBUG "...and adding to erasable_pending_wbuf_list\n"));
622                         list_add_tail(&jeb->list, &c->erasable_pending_wbuf_list);
623                 } else {
624                         if (jiffies & 127) {
625                                 /* Most of the time, we just erase it immediately. Otherwise we
626                                    spend ages scanning it on mount, etc. */
627                                 D1(printk(KERN_DEBUG "...and adding to erase_pending_list\n"));
628                                 list_add_tail(&jeb->list, &c->erase_pending_list);
629                                 c->nr_erasing_blocks++;
630                                 jffs2_garbage_collect_trigger(c);
631                         } else {
632                                 /* Sometimes, however, we leave it elsewhere so it doesn't get
633                                    immediately reused, and we spread the load a bit. */
634                                 D1(printk(KERN_DEBUG "...and adding to erasable_list\n"));
635                                 list_add_tail(&jeb->list, &c->erasable_list);
636                         }
637                 }
638                 D1(printk(KERN_DEBUG "Done OK\n"));
639         } else if (jeb == c->gcblock) {
640                 D2(printk(KERN_DEBUG "Not moving gcblock 0x%08x to dirty_list\n", jeb->offset));
641         } else if (ISDIRTY(jeb->dirty_size) && !ISDIRTY(jeb->dirty_size - addedsize)) {
642                 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x is freshly dirtied. Removing from clean list...\n", jeb->offset));
643                 list_del(&jeb->list);
644                 D1(printk(KERN_DEBUG "...and adding to dirty_list\n"));
645                 list_add_tail(&jeb->list, &c->dirty_list);
646         } else if (VERYDIRTY(c, jeb->dirty_size) &&
647                    !VERYDIRTY(c, jeb->dirty_size - addedsize)) {
648                 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x is now very dirty. Removing from dirty list...\n", jeb->offset));
649                 list_del(&jeb->list);
650                 D1(printk(KERN_DEBUG "...and adding to very_dirty_list\n"));
651                 list_add_tail(&jeb->list, &c->very_dirty_list);
652         } else {
653                 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x not moved anywhere. (free 0x%08x, dirty 0x%08x, used 0x%08x)\n",
654                           jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
655         }
656
657         spin_unlock(&c->erase_completion_lock);
658
659         if (!jffs2_can_mark_obsolete(c) || jffs2_is_readonly(c) ||
660                 (c->flags & JFFS2_SB_FLAG_BUILDING)) {
661                 /* We didn't lock the erase_free_sem */
662                 return;
663         }
664
665         /* The erase_free_sem is locked, and has been since before we marked the node obsolete
666            and potentially put its eraseblock onto the erase_pending_list. Thus, we know that
667            the block hasn't _already_ been erased, and that 'ref' itself hasn't been freed yet
668            by jffs2_free_jeb_node_refs() in erase.c. Which is nice. */
669
670         D1(printk(KERN_DEBUG "obliterating obsoleted node at 0x%08x\n", ref_offset(ref)));
671         ret = jffs2_flash_read(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
672         if (ret) {
673                 printk(KERN_WARNING "Read error reading from obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret);
674                 goto out_erase_sem;
675         }
676         if (retlen != sizeof(n)) {
677                 printk(KERN_WARNING "Short read from obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen);
678                 goto out_erase_sem;
679         }
680         if (PAD(je32_to_cpu(n.totlen)) != PAD(freed_len)) {
681                 printk(KERN_WARNING "Node totlen on flash (0x%08x) != totlen from node ref (0x%08x)\n", je32_to_cpu(n.totlen), freed_len);
682                 goto out_erase_sem;
683         }
684         if (!(je16_to_cpu(n.nodetype) & JFFS2_NODE_ACCURATE)) {
685                 D1(printk(KERN_DEBUG "Node at 0x%08x was already marked obsolete (nodetype 0x%04x)\n", ref_offset(ref), je16_to_cpu(n.nodetype)));
686                 goto out_erase_sem;
687         }
688         /* XXX FIXME: This is ugly now */
689         n.nodetype = cpu_to_je16(je16_to_cpu(n.nodetype) & ~JFFS2_NODE_ACCURATE);
690         ret = jffs2_flash_write(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
691         if (ret) {
692                 printk(KERN_WARNING "Write error in obliterating obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret);
693                 goto out_erase_sem;
694         }
695         if (retlen != sizeof(n)) {
696                 printk(KERN_WARNING "Short write in obliterating obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen);
697                 goto out_erase_sem;
698         }
699
700         /* Nodes which have been marked obsolete no longer need to be
701            associated with any inode. Remove them from the per-inode list.
702
703            Note we can't do this for NAND at the moment because we need
704            obsolete dirent nodes to stay on the lists, because of the
705            horridness in jffs2_garbage_collect_deletion_dirent(). Also
706            because we delete the inocache, and on NAND we need that to
707            stay around until all the nodes are actually erased, in order
708            to stop us from giving the same inode number to another newly
709            created inode. */
710         if (ref->next_in_ino) {
711                 struct jffs2_inode_cache *ic;
712                 struct jffs2_raw_node_ref **p;
713
714                 spin_lock(&c->erase_completion_lock);
715
716                 ic = jffs2_raw_ref_to_ic(ref);
717                 for (p = &ic->nodes; (*p) != ref; p = &((*p)->next_in_ino))
718                         ;
719
720                 *p = ref->next_in_ino;
721                 ref->next_in_ino = NULL;
722
723                 switch (ic->class) {
724 #ifdef CONFIG_JFFS2_FS_XATTR
725                         case RAWNODE_CLASS_XATTR_DATUM:
726                                 jffs2_release_xattr_datum(c, (struct jffs2_xattr_datum *)ic);
727                                 break;
728                         case RAWNODE_CLASS_XATTR_REF:
729                                 jffs2_release_xattr_ref(c, (struct jffs2_xattr_ref *)ic);
730                                 break;
731 #endif
732                         default:
733                                 if (ic->nodes == (void *)ic && ic->pino_nlink == 0)
734                                         jffs2_del_ino_cache(c, ic);
735                                 break;
736                 }
737                 spin_unlock(&c->erase_completion_lock);
738         }
739
740  out_erase_sem:
741         mutex_unlock(&c->erase_free_sem);
742 }
743
744 int jffs2_thread_should_wake(struct jffs2_sb_info *c)
745 {
746         int ret = 0;
747         uint32_t dirty;
748         int nr_very_dirty = 0;
749         struct jffs2_eraseblock *jeb;
750
751         if (!list_empty(&c->erase_complete_list) ||
752             !list_empty(&c->erase_pending_list))
753                 return 1;
754
755         if (c->unchecked_size) {
756                 D1(printk(KERN_DEBUG "jffs2_thread_should_wake(): unchecked_size %d, checked_ino #%d\n",
757                           c->unchecked_size, c->checked_ino));
758                 return 1;
759         }
760
761         /* dirty_size contains blocks on erase_pending_list
762          * those blocks are counted in c->nr_erasing_blocks.
763          * If one block is actually erased, it is not longer counted as dirty_space
764          * but it is counted in c->nr_erasing_blocks, so we add it and subtract it
765          * with c->nr_erasing_blocks * c->sector_size again.
766          * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks
767          * This helps us to force gc and pick eventually a clean block to spread the load.
768          */
769         dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size;
770
771         if (c->nr_free_blocks + c->nr_erasing_blocks < c->resv_blocks_gctrigger &&
772                         (dirty > c->nospc_dirty_size))
773                 ret = 1;
774
775         list_for_each_entry(jeb, &c->very_dirty_list, list) {
776                 nr_very_dirty++;
777                 if (nr_very_dirty == c->vdirty_blocks_gctrigger) {
778                         ret = 1;
779                         /* In debug mode, actually go through and count them all */
780                         D1(continue);
781                         break;
782                 }
783         }
784
785         D1(printk(KERN_DEBUG "jffs2_thread_should_wake(): nr_free_blocks %d, nr_erasing_blocks %d, dirty_size 0x%x, vdirty_blocks %d: %s\n",
786                   c->nr_free_blocks, c->nr_erasing_blocks, c->dirty_size, nr_very_dirty, ret?"yes":"no"));
787
788         return ret;
789 }