Merge branch 'for-linus' of git://git.infradead.org/users/eparis/notify
[pandora-kernel.git] / fs / jffs2 / erase.c
1 /*
2  * JFFS2 -- Journalling Flash File System, Version 2.
3  *
4  * Copyright © 2001-2007 Red Hat, Inc.
5  * Copyright © 2004-2010 David Woodhouse <dwmw2@infradead.org>
6  *
7  * Created by David Woodhouse <dwmw2@infradead.org>
8  *
9  * For licensing information, see the file 'LICENCE' in this directory.
10  *
11  */
12
13 #include <linux/kernel.h>
14 #include <linux/slab.h>
15 #include <linux/mtd/mtd.h>
16 #include <linux/compiler.h>
17 #include <linux/crc32.h>
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include "nodelist.h"
21
22 struct erase_priv_struct {
23         struct jffs2_eraseblock *jeb;
24         struct jffs2_sb_info *c;
25 };
26
27 #ifndef __ECOS
28 static void jffs2_erase_callback(struct erase_info *);
29 #endif
30 static void jffs2_erase_failed(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, uint32_t bad_offset);
31 static void jffs2_erase_succeeded(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb);
32 static void jffs2_mark_erased_block(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb);
33
34 static void jffs2_erase_block(struct jffs2_sb_info *c,
35                               struct jffs2_eraseblock *jeb)
36 {
37         int ret;
38         uint32_t bad_offset;
39 #ifdef __ECOS
40        ret = jffs2_flash_erase(c, jeb);
41        if (!ret) {
42                jffs2_erase_succeeded(c, jeb);
43                return;
44        }
45        bad_offset = jeb->offset;
46 #else /* Linux */
47         struct erase_info *instr;
48
49         D1(printk(KERN_DEBUG "jffs2_erase_block(): erase block %#08x (range %#08x-%#08x)\n",
50                                 jeb->offset, jeb->offset, jeb->offset + c->sector_size));
51         instr = kmalloc(sizeof(struct erase_info) + sizeof(struct erase_priv_struct), GFP_KERNEL);
52         if (!instr) {
53                 printk(KERN_WARNING "kmalloc for struct erase_info in jffs2_erase_block failed. Refiling block for later\n");
54                 mutex_lock(&c->erase_free_sem);
55                 spin_lock(&c->erase_completion_lock);
56                 list_move(&jeb->list, &c->erase_pending_list);
57                 c->erasing_size -= c->sector_size;
58                 c->dirty_size += c->sector_size;
59                 jeb->dirty_size = c->sector_size;
60                 spin_unlock(&c->erase_completion_lock);
61                 mutex_unlock(&c->erase_free_sem);
62                 return;
63         }
64
65         memset(instr, 0, sizeof(*instr));
66
67         instr->mtd = c->mtd;
68         instr->addr = jeb->offset;
69         instr->len = c->sector_size;
70         instr->callback = jffs2_erase_callback;
71         instr->priv = (unsigned long)(&instr[1]);
72         instr->fail_addr = MTD_FAIL_ADDR_UNKNOWN;
73
74         ((struct erase_priv_struct *)instr->priv)->jeb = jeb;
75         ((struct erase_priv_struct *)instr->priv)->c = c;
76
77         ret = c->mtd->erase(c->mtd, instr);
78         if (!ret)
79                 return;
80
81         bad_offset = instr->fail_addr;
82         kfree(instr);
83 #endif /* __ECOS */
84
85         if (ret == -ENOMEM || ret == -EAGAIN) {
86                 /* Erase failed immediately. Refile it on the list */
87                 D1(printk(KERN_DEBUG "Erase at 0x%08x failed: %d. Refiling on erase_pending_list\n", jeb->offset, ret));
88                 mutex_lock(&c->erase_free_sem);
89                 spin_lock(&c->erase_completion_lock);
90                 list_move(&jeb->list, &c->erase_pending_list);
91                 c->erasing_size -= c->sector_size;
92                 c->dirty_size += c->sector_size;
93                 jeb->dirty_size = c->sector_size;
94                 spin_unlock(&c->erase_completion_lock);
95                 mutex_unlock(&c->erase_free_sem);
96                 return;
97         }
98
99         if (ret == -EROFS)
100                 printk(KERN_WARNING "Erase at 0x%08x failed immediately: -EROFS. Is the sector locked?\n", jeb->offset);
101         else
102                 printk(KERN_WARNING "Erase at 0x%08x failed immediately: errno %d\n", jeb->offset, ret);
103
104         jffs2_erase_failed(c, jeb, bad_offset);
105 }
106
107 int jffs2_erase_pending_blocks(struct jffs2_sb_info *c, int count)
108 {
109         struct jffs2_eraseblock *jeb;
110         int work_done = 0;
111
112         mutex_lock(&c->erase_free_sem);
113
114         spin_lock(&c->erase_completion_lock);
115
116         while (!list_empty(&c->erase_complete_list) ||
117                !list_empty(&c->erase_pending_list)) {
118
119                 if (!list_empty(&c->erase_complete_list)) {
120                         jeb = list_entry(c->erase_complete_list.next, struct jffs2_eraseblock, list);
121                         list_move(&jeb->list, &c->erase_checking_list);
122                         spin_unlock(&c->erase_completion_lock);
123                         mutex_unlock(&c->erase_free_sem);
124                         jffs2_mark_erased_block(c, jeb);
125
126                         work_done++;
127                         if (!--count) {
128                                 D1(printk(KERN_DEBUG "Count reached. jffs2_erase_pending_blocks leaving\n"));
129                                 goto done;
130                         }
131
132                 } else if (!list_empty(&c->erase_pending_list)) {
133                         jeb = list_entry(c->erase_pending_list.next, struct jffs2_eraseblock, list);
134                         D1(printk(KERN_DEBUG "Starting erase of pending block 0x%08x\n", jeb->offset));
135                         list_del(&jeb->list);
136                         c->erasing_size += c->sector_size;
137                         c->wasted_size -= jeb->wasted_size;
138                         c->free_size -= jeb->free_size;
139                         c->used_size -= jeb->used_size;
140                         c->dirty_size -= jeb->dirty_size;
141                         jeb->wasted_size = jeb->used_size = jeb->dirty_size = jeb->free_size = 0;
142                         jffs2_free_jeb_node_refs(c, jeb);
143                         list_add(&jeb->list, &c->erasing_list);
144                         spin_unlock(&c->erase_completion_lock);
145                         mutex_unlock(&c->erase_free_sem);
146
147                         jffs2_erase_block(c, jeb);
148
149                 } else {
150                         BUG();
151                 }
152
153                 /* Be nice */
154                 cond_resched();
155                 mutex_lock(&c->erase_free_sem);
156                 spin_lock(&c->erase_completion_lock);
157         }
158
159         spin_unlock(&c->erase_completion_lock);
160         mutex_unlock(&c->erase_free_sem);
161  done:
162         D1(printk(KERN_DEBUG "jffs2_erase_pending_blocks completed\n"));
163         return work_done;
164 }
165
166 static void jffs2_erase_succeeded(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
167 {
168         D1(printk(KERN_DEBUG "Erase completed successfully at 0x%08x\n", jeb->offset));
169         mutex_lock(&c->erase_free_sem);
170         spin_lock(&c->erase_completion_lock);
171         list_move_tail(&jeb->list, &c->erase_complete_list);
172         /* Wake the GC thread to mark them clean */
173         jffs2_garbage_collect_trigger(c);
174         spin_unlock(&c->erase_completion_lock);
175         mutex_unlock(&c->erase_free_sem);
176         wake_up(&c->erase_wait);
177 }
178
179 static void jffs2_erase_failed(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, uint32_t bad_offset)
180 {
181         /* For NAND, if the failure did not occur at the device level for a
182            specific physical page, don't bother updating the bad block table. */
183         if (jffs2_cleanmarker_oob(c) && (bad_offset != (uint32_t)MTD_FAIL_ADDR_UNKNOWN)) {
184                 /* We had a device-level failure to erase.  Let's see if we've
185                    failed too many times. */
186                 if (!jffs2_write_nand_badblock(c, jeb, bad_offset)) {
187                         /* We'd like to give this block another try. */
188                         mutex_lock(&c->erase_free_sem);
189                         spin_lock(&c->erase_completion_lock);
190                         list_move(&jeb->list, &c->erase_pending_list);
191                         c->erasing_size -= c->sector_size;
192                         c->dirty_size += c->sector_size;
193                         jeb->dirty_size = c->sector_size;
194                         spin_unlock(&c->erase_completion_lock);
195                         mutex_unlock(&c->erase_free_sem);
196                         return;
197                 }
198         }
199
200         mutex_lock(&c->erase_free_sem);
201         spin_lock(&c->erase_completion_lock);
202         c->erasing_size -= c->sector_size;
203         c->bad_size += c->sector_size;
204         list_move(&jeb->list, &c->bad_list);
205         c->nr_erasing_blocks--;
206         spin_unlock(&c->erase_completion_lock);
207         mutex_unlock(&c->erase_free_sem);
208         wake_up(&c->erase_wait);
209 }
210
211 #ifndef __ECOS
212 static void jffs2_erase_callback(struct erase_info *instr)
213 {
214         struct erase_priv_struct *priv = (void *)instr->priv;
215
216         if(instr->state != MTD_ERASE_DONE) {
217                 printk(KERN_WARNING "Erase at 0x%08llx finished, but state != MTD_ERASE_DONE. State is 0x%x instead.\n",
218                         (unsigned long long)instr->addr, instr->state);
219                 jffs2_erase_failed(priv->c, priv->jeb, instr->fail_addr);
220         } else {
221                 jffs2_erase_succeeded(priv->c, priv->jeb);
222         }
223         kfree(instr);
224 }
225 #endif /* !__ECOS */
226
227 /* Hmmm. Maybe we should accept the extra space it takes and make
228    this a standard doubly-linked list? */
229 static inline void jffs2_remove_node_refs_from_ino_list(struct jffs2_sb_info *c,
230                         struct jffs2_raw_node_ref *ref, struct jffs2_eraseblock *jeb)
231 {
232         struct jffs2_inode_cache *ic = NULL;
233         struct jffs2_raw_node_ref **prev;
234
235         prev = &ref->next_in_ino;
236
237         /* Walk the inode's list once, removing any nodes from this eraseblock */
238         while (1) {
239                 if (!(*prev)->next_in_ino) {
240                         /* We're looking at the jffs2_inode_cache, which is
241                            at the end of the linked list. Stash it and continue
242                            from the beginning of the list */
243                         ic = (struct jffs2_inode_cache *)(*prev);
244                         prev = &ic->nodes;
245                         continue;
246                 }
247
248                 if (SECTOR_ADDR((*prev)->flash_offset) == jeb->offset) {
249                         /* It's in the block we're erasing */
250                         struct jffs2_raw_node_ref *this;
251
252                         this = *prev;
253                         *prev = this->next_in_ino;
254                         this->next_in_ino = NULL;
255
256                         if (this == ref)
257                                 break;
258
259                         continue;
260                 }
261                 /* Not to be deleted. Skip */
262                 prev = &((*prev)->next_in_ino);
263         }
264
265         /* PARANOIA */
266         if (!ic) {
267                 JFFS2_WARNING("inode_cache/xattr_datum/xattr_ref"
268                               " not found in remove_node_refs()!!\n");
269                 return;
270         }
271
272         D1(printk(KERN_DEBUG "Removed nodes in range 0x%08x-0x%08x from ino #%u\n",
273                   jeb->offset, jeb->offset + c->sector_size, ic->ino));
274
275         D2({
276                 int i=0;
277                 struct jffs2_raw_node_ref *this;
278                 printk(KERN_DEBUG "After remove_node_refs_from_ino_list: \n");
279
280                 this = ic->nodes;
281
282                 printk(KERN_DEBUG);
283                 while(this) {
284                         printk(KERN_CONT "0x%08x(%d)->",
285                                ref_offset(this), ref_flags(this));
286                         if (++i == 5) {
287                                 printk(KERN_DEBUG);
288                                 i=0;
289                         }
290                         this = this->next_in_ino;
291                 }
292                 printk(KERN_CONT "\n");
293         });
294
295         switch (ic->class) {
296 #ifdef CONFIG_JFFS2_FS_XATTR
297                 case RAWNODE_CLASS_XATTR_DATUM:
298                         jffs2_release_xattr_datum(c, (struct jffs2_xattr_datum *)ic);
299                         break;
300                 case RAWNODE_CLASS_XATTR_REF:
301                         jffs2_release_xattr_ref(c, (struct jffs2_xattr_ref *)ic);
302                         break;
303 #endif
304                 default:
305                         if (ic->nodes == (void *)ic && ic->pino_nlink == 0)
306                                 jffs2_del_ino_cache(c, ic);
307         }
308 }
309
310 void jffs2_free_jeb_node_refs(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
311 {
312         struct jffs2_raw_node_ref *block, *ref;
313         D1(printk(KERN_DEBUG "Freeing all node refs for eraseblock offset 0x%08x\n", jeb->offset));
314
315         block = ref = jeb->first_node;
316
317         while (ref) {
318                 if (ref->flash_offset == REF_LINK_NODE) {
319                         ref = ref->next_in_ino;
320                         jffs2_free_refblock(block);
321                         block = ref;
322                         continue;
323                 }
324                 if (ref->flash_offset != REF_EMPTY_NODE && ref->next_in_ino)
325                         jffs2_remove_node_refs_from_ino_list(c, ref, jeb);
326                 /* else it was a non-inode node or already removed, so don't bother */
327
328                 ref++;
329         }
330         jeb->first_node = jeb->last_node = NULL;
331 }
332
333 static int jffs2_block_check_erase(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, uint32_t *bad_offset)
334 {
335         void *ebuf;
336         uint32_t ofs;
337         size_t retlen;
338         int ret = -EIO;
339
340         if (c->mtd->point) {
341                 unsigned long *wordebuf;
342
343                 ret = c->mtd->point(c->mtd, jeb->offset, c->sector_size,
344                                     &retlen, &ebuf, NULL);
345                 if (ret) {
346                         D1(printk(KERN_DEBUG "MTD point failed %d\n", ret));
347                         goto do_flash_read;
348                 }
349                 if (retlen < c->sector_size) {
350                         /* Don't muck about if it won't let us point to the whole erase sector */
351                         D1(printk(KERN_DEBUG "MTD point returned len too short: 0x%zx\n", retlen));
352                         c->mtd->unpoint(c->mtd, jeb->offset, retlen);
353                         goto do_flash_read;
354                 }
355                 wordebuf = ebuf-sizeof(*wordebuf);
356                 retlen /= sizeof(*wordebuf);
357                 do {
358                    if (*++wordebuf != ~0)
359                            break;
360                 } while(--retlen);
361                 c->mtd->unpoint(c->mtd, jeb->offset, c->sector_size);
362                 if (retlen) {
363                         printk(KERN_WARNING "Newly-erased block contained word 0x%lx at offset 0x%08tx\n",
364                                *wordebuf, jeb->offset + c->sector_size-retlen*sizeof(*wordebuf));
365                         return -EIO;
366                 }
367                 return 0;
368         }
369  do_flash_read:
370         ebuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
371         if (!ebuf) {
372                 printk(KERN_WARNING "Failed to allocate page buffer for verifying erase at 0x%08x. Refiling\n", jeb->offset);
373                 return -EAGAIN;
374         }
375
376         D1(printk(KERN_DEBUG "Verifying erase at 0x%08x\n", jeb->offset));
377
378         for (ofs = jeb->offset; ofs < jeb->offset + c->sector_size; ) {
379                 uint32_t readlen = min((uint32_t)PAGE_SIZE, jeb->offset + c->sector_size - ofs);
380                 int i;
381
382                 *bad_offset = ofs;
383
384                 ret = c->mtd->read(c->mtd, ofs, readlen, &retlen, ebuf);
385                 if (ret) {
386                         printk(KERN_WARNING "Read of newly-erased block at 0x%08x failed: %d. Putting on bad_list\n", ofs, ret);
387                         ret = -EIO;
388                         goto fail;
389                 }
390                 if (retlen != readlen) {
391                         printk(KERN_WARNING "Short read from newly-erased block at 0x%08x. Wanted %d, got %zd\n", ofs, readlen, retlen);
392                         ret = -EIO;
393                         goto fail;
394                 }
395                 for (i=0; i<readlen; i += sizeof(unsigned long)) {
396                         /* It's OK. We know it's properly aligned */
397                         unsigned long *datum = ebuf + i;
398                         if (*datum + 1) {
399                                 *bad_offset += i;
400                                 printk(KERN_WARNING "Newly-erased block contained word 0x%lx at offset 0x%08x\n", *datum, *bad_offset);
401                                 ret = -EIO;
402                                 goto fail;
403                         }
404                 }
405                 ofs += readlen;
406                 cond_resched();
407         }
408         ret = 0;
409 fail:
410         kfree(ebuf);
411         return ret;
412 }
413
414 static void jffs2_mark_erased_block(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
415 {
416         size_t retlen;
417         int ret;
418         uint32_t uninitialized_var(bad_offset);
419
420         switch (jffs2_block_check_erase(c, jeb, &bad_offset)) {
421         case -EAGAIN:   goto refile;
422         case -EIO:      goto filebad;
423         }
424
425         /* Write the erase complete marker */
426         D1(printk(KERN_DEBUG "Writing erased marker to block at 0x%08x\n", jeb->offset));
427         bad_offset = jeb->offset;
428
429         /* Cleanmarker in oob area or no cleanmarker at all ? */
430         if (jffs2_cleanmarker_oob(c) || c->cleanmarker_size == 0) {
431
432                 if (jffs2_cleanmarker_oob(c)) {
433                         if (jffs2_write_nand_cleanmarker(c, jeb))
434                                 goto filebad;
435                 }
436         } else {
437
438                 struct kvec vecs[1];
439                 struct jffs2_unknown_node marker = {
440                         .magic =        cpu_to_je16(JFFS2_MAGIC_BITMASK),
441                         .nodetype =     cpu_to_je16(JFFS2_NODETYPE_CLEANMARKER),
442                         .totlen =       cpu_to_je32(c->cleanmarker_size)
443                 };
444
445                 jffs2_prealloc_raw_node_refs(c, jeb, 1);
446
447                 marker.hdr_crc = cpu_to_je32(crc32(0, &marker, sizeof(struct jffs2_unknown_node)-4));
448
449                 vecs[0].iov_base = (unsigned char *) &marker;
450                 vecs[0].iov_len = sizeof(marker);
451                 ret = jffs2_flash_direct_writev(c, vecs, 1, jeb->offset, &retlen);
452
453                 if (ret || retlen != sizeof(marker)) {
454                         if (ret)
455                                 printk(KERN_WARNING "Write clean marker to block at 0x%08x failed: %d\n",
456                                        jeb->offset, ret);
457                         else
458                                 printk(KERN_WARNING "Short write to newly-erased block at 0x%08x: Wanted %zd, got %zd\n",
459                                        jeb->offset, sizeof(marker), retlen);
460
461                         goto filebad;
462                 }
463         }
464         /* Everything else got zeroed before the erase */
465         jeb->free_size = c->sector_size;
466
467         mutex_lock(&c->erase_free_sem);
468         spin_lock(&c->erase_completion_lock);
469
470         c->erasing_size -= c->sector_size;
471         c->free_size += c->sector_size;
472
473         /* Account for cleanmarker now, if it's in-band */
474         if (c->cleanmarker_size && !jffs2_cleanmarker_oob(c))
475                 jffs2_link_node_ref(c, jeb, jeb->offset | REF_NORMAL, c->cleanmarker_size, NULL);
476
477         list_move_tail(&jeb->list, &c->free_list);
478         c->nr_erasing_blocks--;
479         c->nr_free_blocks++;
480
481         jffs2_dbg_acct_sanity_check_nolock(c, jeb);
482         jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
483
484         spin_unlock(&c->erase_completion_lock);
485         mutex_unlock(&c->erase_free_sem);
486         wake_up(&c->erase_wait);
487         return;
488
489 filebad:
490         jffs2_erase_failed(c, jeb, bad_offset);
491         return;
492
493 refile:
494         /* Stick it back on the list from whence it came and come back later */
495         mutex_lock(&c->erase_free_sem);
496         spin_lock(&c->erase_completion_lock);
497         jffs2_garbage_collect_trigger(c);
498         list_move(&jeb->list, &c->erase_complete_list);
499         spin_unlock(&c->erase_completion_lock);
500         mutex_unlock(&c->erase_free_sem);
501         return;
502 }