[JFFS2] Fix hanging close for /dev/mtd character device.
[safe/jmp/linux-2.6] / fs / jffs2 / erase.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/slab.h>
14 #include <linux/mtd/mtd.h>
15 #include <linux/compiler.h>
16 #include <linux/crc32.h>
17 #include <linux/sched.h>
18 #include <linux/pagemap.h>
19 #include "nodelist.h"
20
21 struct erase_priv_struct {
22         struct jffs2_eraseblock *jeb;
23         struct jffs2_sb_info *c;
24 };
25
26 #ifndef __ECOS
27 static void jffs2_erase_callback(struct erase_info *);
28 #endif
29 static void jffs2_erase_failed(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, uint32_t bad_offset);
30 static void jffs2_erase_succeeded(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb);
31 static void jffs2_mark_erased_block(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb);
32
33 static void jffs2_erase_block(struct jffs2_sb_info *c,
34                               struct jffs2_eraseblock *jeb)
35 {
36         int ret;
37         uint32_t bad_offset;
38 #ifdef __ECOS
39        ret = jffs2_flash_erase(c, jeb);
40        if (!ret) {
41                jffs2_erase_succeeded(c, jeb);
42                return;
43        }
44        bad_offset = jeb->offset;
45 #else /* Linux */
46         struct erase_info *instr;
47
48         D1(printk(KERN_DEBUG "jffs2_erase_block(): erase block %#08x (range %#08x-%#08x)\n",
49                                 jeb->offset, jeb->offset, jeb->offset + c->sector_size));
50         instr = kmalloc(sizeof(struct erase_info) + sizeof(struct erase_priv_struct), GFP_KERNEL);
51         if (!instr) {
52                 printk(KERN_WARNING "kmalloc for struct erase_info in jffs2_erase_block failed. Refiling block for later\n");
53                 spin_lock(&c->erase_completion_lock);
54                 list_move(&jeb->list, &c->erase_pending_list);
55                 c->erasing_size -= c->sector_size;
56                 c->dirty_size += c->sector_size;
57                 jeb->dirty_size = c->sector_size;
58                 spin_unlock(&c->erase_completion_lock);
59                 return;
60         }
61
62         memset(instr, 0, sizeof(*instr));
63
64         instr->mtd = c->mtd;
65         instr->addr = jeb->offset;
66         instr->len = c->sector_size;
67         instr->callback = jffs2_erase_callback;
68         instr->priv = (unsigned long)(&instr[1]);
69         instr->fail_addr = 0xffffffff;
70
71         ((struct erase_priv_struct *)instr->priv)->jeb = jeb;
72         ((struct erase_priv_struct *)instr->priv)->c = c;
73
74         ret = c->mtd->erase(c->mtd, instr);
75         if (!ret)
76                 return;
77
78         bad_offset = instr->fail_addr;
79         kfree(instr);
80 #endif /* __ECOS */
81
82         if (ret == -ENOMEM || ret == -EAGAIN) {
83                 /* Erase failed immediately. Refile it on the list */
84                 D1(printk(KERN_DEBUG "Erase at 0x%08x failed: %d. Refiling on erase_pending_list\n", jeb->offset, ret));
85                 spin_lock(&c->erase_completion_lock);
86                 list_move(&jeb->list, &c->erase_pending_list);
87                 c->erasing_size -= c->sector_size;
88                 c->dirty_size += c->sector_size;
89                 jeb->dirty_size = c->sector_size;
90                 spin_unlock(&c->erase_completion_lock);
91                 return;
92         }
93
94         if (ret == -EROFS)
95                 printk(KERN_WARNING "Erase at 0x%08x failed immediately: -EROFS. Is the sector locked?\n", jeb->offset);
96         else
97                 printk(KERN_WARNING "Erase at 0x%08x failed immediately: errno %d\n", jeb->offset, ret);
98
99         jffs2_erase_failed(c, jeb, bad_offset);
100 }
101
102 void jffs2_erase_pending_blocks(struct jffs2_sb_info *c, int count)
103 {
104         struct jffs2_eraseblock *jeb;
105
106         down(&c->erase_free_sem);
107
108         spin_lock(&c->erase_completion_lock);
109
110         while (!list_empty(&c->erase_complete_list) ||
111                !list_empty(&c->erase_pending_list)) {
112
113                 if (!list_empty(&c->erase_complete_list)) {
114                         jeb = list_entry(c->erase_complete_list.next, struct jffs2_eraseblock, list);
115                         list_del(&jeb->list);
116                         spin_unlock(&c->erase_completion_lock);
117                         jffs2_mark_erased_block(c, jeb);
118
119                         if (!--count) {
120                                 D1(printk(KERN_DEBUG "Count reached. jffs2_erase_pending_blocks leaving\n"));
121                                 goto done;
122                         }
123
124                 } else if (!list_empty(&c->erase_pending_list)) {
125                         jeb = list_entry(c->erase_pending_list.next, struct jffs2_eraseblock, list);
126                         D1(printk(KERN_DEBUG "Starting erase of pending block 0x%08x\n", jeb->offset));
127                         list_del(&jeb->list);
128                         c->erasing_size += c->sector_size;
129                         c->wasted_size -= jeb->wasted_size;
130                         c->free_size -= jeb->free_size;
131                         c->used_size -= jeb->used_size;
132                         c->dirty_size -= jeb->dirty_size;
133                         jeb->wasted_size = jeb->used_size = jeb->dirty_size = jeb->free_size = 0;
134                         jffs2_free_jeb_node_refs(c, jeb);
135                         list_add(&jeb->list, &c->erasing_list);
136                         spin_unlock(&c->erase_completion_lock);
137
138                         jffs2_erase_block(c, jeb);
139
140                 } else {
141                         BUG();
142                 }
143
144                 /* Be nice */
145                 yield();
146                 spin_lock(&c->erase_completion_lock);
147         }
148
149         spin_unlock(&c->erase_completion_lock);
150  done:
151         D1(printk(KERN_DEBUG "jffs2_erase_pending_blocks completed\n"));
152
153         up(&c->erase_free_sem);
154 }
155
156 static void jffs2_erase_succeeded(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
157 {
158         D1(printk(KERN_DEBUG "Erase completed successfully at 0x%08x\n", jeb->offset));
159         spin_lock(&c->erase_completion_lock);
160         list_move_tail(&jeb->list, &c->erase_complete_list);
161         spin_unlock(&c->erase_completion_lock);
162         /* Ensure that kupdated calls us again to mark them clean */
163         jffs2_erase_pending_trigger(c);
164 }
165
166 static void jffs2_erase_failed(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, uint32_t bad_offset)
167 {
168         /* For NAND, if the failure did not occur at the device level for a
169            specific physical page, don't bother updating the bad block table. */
170         if (jffs2_cleanmarker_oob(c) && (bad_offset != 0xffffffff)) {
171                 /* We had a device-level failure to erase.  Let's see if we've
172                    failed too many times. */
173                 if (!jffs2_write_nand_badblock(c, jeb, bad_offset)) {
174                         /* We'd like to give this block another try. */
175                         spin_lock(&c->erase_completion_lock);
176                         list_move(&jeb->list, &c->erase_pending_list);
177                         c->erasing_size -= c->sector_size;
178                         c->dirty_size += c->sector_size;
179                         jeb->dirty_size = c->sector_size;
180                         spin_unlock(&c->erase_completion_lock);
181                         return;
182                 }
183         }
184
185         spin_lock(&c->erase_completion_lock);
186         c->erasing_size -= c->sector_size;
187         c->bad_size += c->sector_size;
188         list_move(&jeb->list, &c->bad_list);
189         c->nr_erasing_blocks--;
190         spin_unlock(&c->erase_completion_lock);
191         wake_up(&c->erase_wait);
192 }
193
194 #ifndef __ECOS
195 static void jffs2_erase_callback(struct erase_info *instr)
196 {
197         struct erase_priv_struct *priv = (void *)instr->priv;
198
199         if(instr->state != MTD_ERASE_DONE) {
200                 printk(KERN_WARNING "Erase at 0x%08x finished, but state != MTD_ERASE_DONE. State is 0x%x instead.\n", instr->addr, instr->state);
201                 jffs2_erase_failed(priv->c, priv->jeb, instr->fail_addr);
202         } else {
203                 jffs2_erase_succeeded(priv->c, priv->jeb);
204         }
205         kfree(instr);
206 }
207 #endif /* !__ECOS */
208
209 /* Hmmm. Maybe we should accept the extra space it takes and make
210    this a standard doubly-linked list? */
211 static inline void jffs2_remove_node_refs_from_ino_list(struct jffs2_sb_info *c,
212                         struct jffs2_raw_node_ref *ref, struct jffs2_eraseblock *jeb)
213 {
214         struct jffs2_inode_cache *ic = NULL;
215         struct jffs2_raw_node_ref **prev;
216
217         prev = &ref->next_in_ino;
218
219         /* Walk the inode's list once, removing any nodes from this eraseblock */
220         while (1) {
221                 if (!(*prev)->next_in_ino) {
222                         /* We're looking at the jffs2_inode_cache, which is
223                            at the end of the linked list. Stash it and continue
224                            from the beginning of the list */
225                         ic = (struct jffs2_inode_cache *)(*prev);
226                         prev = &ic->nodes;
227                         continue;
228                 }
229
230                 if (SECTOR_ADDR((*prev)->flash_offset) == jeb->offset) {
231                         /* It's in the block we're erasing */
232                         struct jffs2_raw_node_ref *this;
233
234                         this = *prev;
235                         *prev = this->next_in_ino;
236                         this->next_in_ino = NULL;
237
238                         if (this == ref)
239                                 break;
240
241                         continue;
242                 }
243                 /* Not to be deleted. Skip */
244                 prev = &((*prev)->next_in_ino);
245         }
246
247         /* PARANOIA */
248         if (!ic) {
249                 JFFS2_WARNING("inode_cache/xattr_datum/xattr_ref"
250                               " not found in remove_node_refs()!!\n");
251                 return;
252         }
253
254         D1(printk(KERN_DEBUG "Removed nodes in range 0x%08x-0x%08x from ino #%u\n",
255                   jeb->offset, jeb->offset + c->sector_size, ic->ino));
256
257         D2({
258                 int i=0;
259                 struct jffs2_raw_node_ref *this;
260                 printk(KERN_DEBUG "After remove_node_refs_from_ino_list: \n" KERN_DEBUG);
261
262                 this = ic->nodes;
263
264                 while(this) {
265                         printk( "0x%08x(%d)->", ref_offset(this), ref_flags(this));
266                         if (++i == 5) {
267                                 printk("\n" KERN_DEBUG);
268                                 i=0;
269                         }
270                         this = this->next_in_ino;
271                 }
272                 printk("\n");
273         });
274
275         switch (ic->class) {
276 #ifdef CONFIG_JFFS2_FS_XATTR
277                 case RAWNODE_CLASS_XATTR_DATUM:
278                         jffs2_release_xattr_datum(c, (struct jffs2_xattr_datum *)ic);
279                         break;
280                 case RAWNODE_CLASS_XATTR_REF:
281                         jffs2_release_xattr_ref(c, (struct jffs2_xattr_ref *)ic);
282                         break;
283 #endif
284                 default:
285                         if (ic->nodes == (void *)ic && ic->nlink == 0)
286                                 jffs2_del_ino_cache(c, ic);
287         }
288 }
289
290 void jffs2_free_jeb_node_refs(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
291 {
292         struct jffs2_raw_node_ref *block, *ref;
293         D1(printk(KERN_DEBUG "Freeing all node refs for eraseblock offset 0x%08x\n", jeb->offset));
294
295         block = ref = jeb->first_node;
296
297         while (ref) {
298                 if (ref->flash_offset == REF_LINK_NODE) {
299                         ref = ref->next_in_ino;
300                         jffs2_free_refblock(block);
301                         block = ref;
302                         continue;
303                 }
304                 if (ref->flash_offset != REF_EMPTY_NODE && ref->next_in_ino)
305                         jffs2_remove_node_refs_from_ino_list(c, ref, jeb);
306                 /* else it was a non-inode node or already removed, so don't bother */
307
308                 ref++;
309         }
310         jeb->first_node = jeb->last_node = NULL;
311 }
312
313 static int jffs2_block_check_erase(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, uint32_t *bad_offset)
314 {
315         void *ebuf;
316         uint32_t ofs;
317         size_t retlen;
318         int ret = -EIO;
319
320         ebuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
321         if (!ebuf) {
322                 printk(KERN_WARNING "Failed to allocate page buffer for verifying erase at 0x%08x. Refiling\n", jeb->offset);
323                 return -EAGAIN;
324         }
325
326         D1(printk(KERN_DEBUG "Verifying erase at 0x%08x\n", jeb->offset));
327
328         for (ofs = jeb->offset; ofs < jeb->offset + c->sector_size; ) {
329                 uint32_t readlen = min((uint32_t)PAGE_SIZE, jeb->offset + c->sector_size - ofs);
330                 int i;
331
332                 *bad_offset = ofs;
333
334                 ret = c->mtd->read(c->mtd, ofs, readlen, &retlen, ebuf);
335                 if (ret) {
336                         printk(KERN_WARNING "Read of newly-erased block at 0x%08x failed: %d. Putting on bad_list\n", ofs, ret);
337                         goto fail;
338                 }
339                 if (retlen != readlen) {
340                         printk(KERN_WARNING "Short read from newly-erased block at 0x%08x. Wanted %d, got %zd\n", ofs, readlen, retlen);
341                         goto fail;
342                 }
343                 for (i=0; i<readlen; i += sizeof(unsigned long)) {
344                         /* It's OK. We know it's properly aligned */
345                         unsigned long *datum = ebuf + i;
346                         if (*datum + 1) {
347                                 *bad_offset += i;
348                                 printk(KERN_WARNING "Newly-erased block contained word 0x%lx at offset 0x%08x\n", *datum, *bad_offset);
349                                 goto fail;
350                         }
351                 }
352                 ofs += readlen;
353                 cond_resched();
354         }
355         ret = 0;
356 fail:
357         kfree(ebuf);
358         return ret;
359 }
360
361 static void jffs2_mark_erased_block(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
362 {
363         size_t retlen;
364         int ret;
365         uint32_t bad_offset;
366
367         switch (jffs2_block_check_erase(c, jeb, &bad_offset)) {
368         case -EAGAIN:   goto refile;
369         case -EIO:      goto filebad;
370         }
371
372         /* Write the erase complete marker */
373         D1(printk(KERN_DEBUG "Writing erased marker to block at 0x%08x\n", jeb->offset));
374         bad_offset = jeb->offset;
375
376         /* Cleanmarker in oob area or no cleanmarker at all ? */
377         if (jffs2_cleanmarker_oob(c) || c->cleanmarker_size == 0) {
378
379                 if (jffs2_cleanmarker_oob(c)) {
380                         if (jffs2_write_nand_cleanmarker(c, jeb))
381                                 goto filebad;
382                 }
383
384                 /* Everything else got zeroed before the erase */
385                 jeb->free_size = c->sector_size;
386         } else {
387
388                 struct kvec vecs[1];
389                 struct jffs2_unknown_node marker = {
390                         .magic =        cpu_to_je16(JFFS2_MAGIC_BITMASK),
391                         .nodetype =     cpu_to_je16(JFFS2_NODETYPE_CLEANMARKER),
392                         .totlen =       cpu_to_je32(c->cleanmarker_size)
393                 };
394
395                 jffs2_prealloc_raw_node_refs(c, jeb, 1);
396
397                 marker.hdr_crc = cpu_to_je32(crc32(0, &marker, sizeof(struct jffs2_unknown_node)-4));
398
399                 vecs[0].iov_base = (unsigned char *) &marker;
400                 vecs[0].iov_len = sizeof(marker);
401                 ret = jffs2_flash_direct_writev(c, vecs, 1, jeb->offset, &retlen);
402
403                 if (ret || retlen != sizeof(marker)) {
404                         if (ret)
405                                 printk(KERN_WARNING "Write clean marker to block at 0x%08x failed: %d\n",
406                                        jeb->offset, ret);
407                         else
408                                 printk(KERN_WARNING "Short write to newly-erased block at 0x%08x: Wanted %zd, got %zd\n",
409                                        jeb->offset, sizeof(marker), retlen);
410
411                         goto filebad;
412                 }
413
414                 /* Everything else got zeroed before the erase */
415                 jeb->free_size = c->sector_size;
416                 /* FIXME Special case for cleanmarker in empty block */
417                 jffs2_link_node_ref(c, jeb, jeb->offset | REF_NORMAL, c->cleanmarker_size, NULL);
418         }
419
420         spin_lock(&c->erase_completion_lock);
421         c->erasing_size -= c->sector_size;
422         c->free_size += jeb->free_size;
423         c->used_size += jeb->used_size;
424
425         jffs2_dbg_acct_sanity_check_nolock(c,jeb);
426         jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
427
428         list_add_tail(&jeb->list, &c->free_list);
429         c->nr_erasing_blocks--;
430         c->nr_free_blocks++;
431         spin_unlock(&c->erase_completion_lock);
432         wake_up(&c->erase_wait);
433         return;
434
435 filebad:
436         spin_lock(&c->erase_completion_lock);
437         /* Stick it on a list (any list) so erase_failed can take it
438            right off again.  Silly, but shouldn't happen often. */
439         list_add(&jeb->list, &c->erasing_list);
440         spin_unlock(&c->erase_completion_lock);
441         jffs2_erase_failed(c, jeb, bad_offset);
442         return;
443
444 refile:
445         /* Stick it back on the list from whence it came and come back later */
446         jffs2_erase_pending_trigger(c);
447         spin_lock(&c->erase_completion_lock);
448         list_add(&jeb->list, &c->erase_complete_list);
449         spin_unlock(&c->erase_completion_lock);
450         return;
451 }