vserver 1.9.5.x5
[linux-2.6.git] / fs / jffs2 / erase.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: erase.c,v 1.66 2004/11/16 20:36:11 dwmw2 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/crc32.h>
19 #include <linux/sched.h>
20 #include <linux/pagemap.h>
21 #include "nodelist.h"
22
23 struct erase_priv_struct {
24         struct jffs2_eraseblock *jeb;
25         struct jffs2_sb_info *c;
26 };
27       
28 #ifndef __ECOS
29 static void jffs2_erase_callback(struct erase_info *);
30 #endif
31 static void jffs2_erase_failed(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, uint32_t bad_offset);
32 static void jffs2_erase_succeeded(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb);
33 static void jffs2_free_all_node_refs(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb);
34 static void jffs2_mark_erased_block(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb);
35
36 void jffs2_erase_block(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
37 {
38         int ret;
39         uint32_t bad_offset;
40 #ifdef __ECOS
41        ret = jffs2_flash_erase(c, jeb);
42        if (!ret) {
43                jffs2_erase_succeeded(c, jeb);
44                return;
45        }
46        bad_offset = jeb->offset;
47 #else /* Linux */
48         struct erase_info *instr;
49
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_del(&jeb->list);
55                 list_add(&jeb->list, &c->erase_pending_list);
56                 c->erasing_size -= c->sector_size;
57                 c->dirty_size += c->sector_size;
58                 jeb->dirty_size = c->sector_size;
59                 spin_unlock(&c->erase_completion_lock);
60                 return;
61         }
62
63         memset(instr, 0, sizeof(*instr));
64
65         instr->mtd = c->mtd;
66         instr->addr = jeb->offset;
67         instr->len = c->sector_size;
68         instr->callback = jffs2_erase_callback;
69         instr->priv = (unsigned long)(&instr[1]);
70         instr->fail_addr = 0xffffffff;
71         
72         ((struct erase_priv_struct *)instr->priv)->jeb = jeb;
73         ((struct erase_priv_struct *)instr->priv)->c = c;
74
75         ret = c->mtd->erase(c->mtd, instr);
76         if (!ret)
77                 return;
78
79         bad_offset = instr->fail_addr;
80         kfree(instr);
81 #endif /* __ECOS */
82
83         if (ret == -ENOMEM || ret == -EAGAIN) {
84                 /* Erase failed immediately. Refile it on the list */
85                 D1(printk(KERN_DEBUG "Erase at 0x%08x failed: %d. Refiling on erase_pending_list\n", jeb->offset, ret));
86                 spin_lock(&c->erase_completion_lock);
87                 list_del(&jeb->list);
88                 list_add(&jeb->list, &c->erase_pending_list);
89                 c->erasing_size -= c->sector_size;
90                 c->dirty_size += c->sector_size;
91                 jeb->dirty_size = c->sector_size;
92                 spin_unlock(&c->erase_completion_lock);
93                 return;
94         }
95
96         if (ret == -EROFS) 
97                 printk(KERN_WARNING "Erase at 0x%08x failed immediately: -EROFS. Is the sector locked?\n", jeb->offset);
98         else
99                 printk(KERN_WARNING "Erase at 0x%08x failed immediately: errno %d\n", jeb->offset, ret);
100
101         jffs2_erase_failed(c, jeb, bad_offset);
102 }
103
104 void jffs2_erase_pending_blocks(struct jffs2_sb_info *c, int count)
105 {
106         struct jffs2_eraseblock *jeb;
107
108         down(&c->erase_free_sem);
109
110         spin_lock(&c->erase_completion_lock);
111
112         while (!list_empty(&c->erase_complete_list) ||
113                !list_empty(&c->erase_pending_list)) {
114
115                 if (!list_empty(&c->erase_complete_list)) {
116                         jeb = list_entry(c->erase_complete_list.next, struct jffs2_eraseblock, list);
117                         list_del(&jeb->list);
118                         spin_unlock(&c->erase_completion_lock);
119                         jffs2_mark_erased_block(c, jeb);
120
121                         if (!--count) {
122                                 D1(printk(KERN_DEBUG "Count reached. jffs2_erase_pending_blocks leaving\n"));
123                                 goto done;
124                         }
125
126                 } else if (!list_empty(&c->erase_pending_list)) {
127                         jeb = list_entry(c->erase_pending_list.next, struct jffs2_eraseblock, list);
128                         D1(printk(KERN_DEBUG "Starting erase of pending block 0x%08x\n", jeb->offset));
129                         list_del(&jeb->list);
130                         c->erasing_size += c->sector_size;
131                         c->wasted_size -= jeb->wasted_size;
132                         c->free_size -= jeb->free_size;
133                         c->used_size -= jeb->used_size;
134                         c->dirty_size -= jeb->dirty_size;
135                         jeb->wasted_size = jeb->used_size = jeb->dirty_size = jeb->free_size = 0;
136                         jffs2_free_all_node_refs(c, jeb);
137                         list_add(&jeb->list, &c->erasing_list);
138                         spin_unlock(&c->erase_completion_lock);
139
140                         jffs2_erase_block(c, jeb);
141
142                 } else {
143                         BUG();
144                 }
145
146                 /* Be nice */
147                 cond_resched();
148                 spin_lock(&c->erase_completion_lock);
149         }
150
151         spin_unlock(&c->erase_completion_lock);
152  done:
153         D1(printk(KERN_DEBUG "jffs2_erase_pending_blocks completed\n"));
154
155         up(&c->erase_free_sem);
156 }
157
158 static void jffs2_erase_succeeded(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
159 {
160         D1(printk(KERN_DEBUG "Erase completed successfully at 0x%08x\n", jeb->offset));
161         spin_lock(&c->erase_completion_lock);
162         list_del(&jeb->list);
163         list_add_tail(&jeb->list, &c->erase_complete_list);
164         spin_unlock(&c->erase_completion_lock);
165         /* Ensure that kupdated calls us again to mark them clean */
166         jffs2_erase_pending_trigger(c);
167 }
168
169 static void jffs2_erase_failed(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, uint32_t bad_offset)
170 {
171         /* For NAND, if the failure did not occur at the device level for a
172            specific physical page, don't bother updating the bad block table. */
173         if (jffs2_cleanmarker_oob(c) && (bad_offset != 0xffffffff)) {
174                 /* We had a device-level failure to erase.  Let's see if we've
175                    failed too many times. */
176                 if (!jffs2_write_nand_badblock(c, jeb, bad_offset)) {
177                         /* We'd like to give this block another try. */
178                         spin_lock(&c->erase_completion_lock);
179                         list_del(&jeb->list);
180                         list_add(&jeb->list, &c->erase_pending_list);
181                         c->erasing_size -= c->sector_size;
182                         c->dirty_size += c->sector_size;
183                         jeb->dirty_size = c->sector_size;
184                         spin_unlock(&c->erase_completion_lock);
185                         return;
186                 }
187         }
188
189         spin_lock(&c->erase_completion_lock);
190         c->erasing_size -= c->sector_size;
191         c->bad_size += c->sector_size;
192         list_del(&jeb->list);
193         list_add(&jeb->list, &c->bad_list);
194         c->nr_erasing_blocks--;
195         spin_unlock(&c->erase_completion_lock);
196         wake_up(&c->erase_wait);
197 }        
198
199 #ifndef __ECOS
200 static void jffs2_erase_callback(struct erase_info *instr)
201 {
202         struct erase_priv_struct *priv = (void *)instr->priv;
203
204         if(instr->state != MTD_ERASE_DONE) {
205                 printk(KERN_WARNING "Erase at 0x%08x finished, but state != MTD_ERASE_DONE. State is 0x%x instead.\n", instr->addr, instr->state);
206                 jffs2_erase_failed(priv->c, priv->jeb, instr->fail_addr);
207         } else {
208                 jffs2_erase_succeeded(priv->c, priv->jeb);
209         }       
210         kfree(instr);
211 }
212 #endif /* !__ECOS */
213
214 /* Hmmm. Maybe we should accept the extra space it takes and make
215    this a standard doubly-linked list? */
216 static inline void jffs2_remove_node_refs_from_ino_list(struct jffs2_sb_info *c,
217                         struct jffs2_raw_node_ref *ref, struct jffs2_eraseblock *jeb)
218 {
219         struct jffs2_inode_cache *ic = NULL;
220         struct jffs2_raw_node_ref **prev;
221
222         prev = &ref->next_in_ino;
223
224         /* Walk the inode's list once, removing any nodes from this eraseblock */
225         while (1) {
226                 if (!(*prev)->next_in_ino) {
227                         /* We're looking at the jffs2_inode_cache, which is 
228                            at the end of the linked list. Stash it and continue
229                            from the beginning of the list */
230                         ic = (struct jffs2_inode_cache *)(*prev);
231                         prev = &ic->nodes;
232                         continue;
233                 } 
234
235                 if (((*prev)->flash_offset & ~(c->sector_size -1)) == jeb->offset) {
236                         /* It's in the block we're erasing */
237                         struct jffs2_raw_node_ref *this;
238
239                         this = *prev;
240                         *prev = this->next_in_ino;
241                         this->next_in_ino = NULL;
242
243                         if (this == ref)
244                                 break;
245
246                         continue;
247                 }
248                 /* Not to be deleted. Skip */
249                 prev = &((*prev)->next_in_ino);
250         }
251
252         /* PARANOIA */
253         if (!ic) {
254                 printk(KERN_WARNING "inode_cache not found in remove_node_refs()!!\n");
255                 return;
256         }
257
258         D1(printk(KERN_DEBUG "Removed nodes in range 0x%08x-0x%08x from ino #%u\n",
259                   jeb->offset, jeb->offset + c->sector_size, ic->ino));
260
261         D2({
262                 int i=0;
263                 struct jffs2_raw_node_ref *this;
264                 printk(KERN_DEBUG "After remove_node_refs_from_ino_list: \n" KERN_DEBUG);
265
266                 this = ic->nodes;
267            
268                 while(this) {
269                         printk( "0x%08x(%d)->", ref_offset(this), ref_flags(this));
270                         if (++i == 5) {
271                                 printk("\n" KERN_DEBUG);
272                                 i=0;
273                         }
274                         this = this->next_in_ino;
275                 }
276                 printk("\n");
277         });
278
279         if (ic->nodes == (void *)ic) {
280                 D1(printk(KERN_DEBUG "inocache for ino #%u is all gone now. Freeing\n", ic->ino));
281                 jffs2_del_ino_cache(c, ic);
282                 jffs2_free_inode_cache(ic);
283         }
284 }
285
286 static void jffs2_free_all_node_refs(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
287 {
288         struct jffs2_raw_node_ref *ref;
289         D1(printk(KERN_DEBUG "Freeing all node refs for eraseblock offset 0x%08x\n", jeb->offset));
290         while(jeb->first_node) {
291                 ref = jeb->first_node;
292                 jeb->first_node = ref->next_phys;
293                 
294                 /* Remove from the inode-list */
295                 if (ref->next_in_ino)
296                         jffs2_remove_node_refs_from_ino_list(c, ref, jeb);
297                 /* else it was a non-inode node or already removed, so don't bother */
298
299                 jffs2_free_raw_node_ref(ref);
300         }
301         jeb->last_node = NULL;
302 }
303
304 static void jffs2_mark_erased_block(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
305 {
306         struct jffs2_raw_node_ref *marker_ref = NULL;
307         unsigned char *ebuf;
308         size_t retlen;
309         int ret;
310         uint32_t bad_offset;
311
312         if (!jffs2_cleanmarker_oob(c)) {
313                 marker_ref = jffs2_alloc_raw_node_ref();
314                 if (!marker_ref) {
315                         printk(KERN_WARNING "Failed to allocate raw node ref for clean marker\n");
316                         /* Stick it back on the list from whence it came and come back later */
317                         jffs2_erase_pending_trigger(c);
318                         spin_lock(&c->erase_completion_lock);
319                         list_add(&jeb->list, &c->erase_complete_list);
320                         spin_unlock(&c->erase_completion_lock);
321                         return;
322                 }
323         }
324         ebuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
325         if (!ebuf) {
326                 printk(KERN_WARNING "Failed to allocate page buffer for verifying erase at 0x%08x. Assuming it worked\n", jeb->offset);
327         } else {
328                 uint32_t ofs = jeb->offset;
329
330                 D1(printk(KERN_DEBUG "Verifying erase at 0x%08x\n", jeb->offset));
331                 while(ofs < jeb->offset + c->sector_size) {
332                         uint32_t readlen = min((uint32_t)PAGE_SIZE, jeb->offset + c->sector_size - ofs);
333                         int i;
334
335                         bad_offset = ofs;
336
337                         ret = jffs2_flash_read(c, ofs, readlen, &retlen, ebuf);
338                         if (ret) {
339                                 printk(KERN_WARNING "Read of newly-erased block at 0x%08x failed: %d. Putting on bad_list\n", ofs, ret);
340                                 goto bad;
341                         }
342                         if (retlen != readlen) {
343                                 printk(KERN_WARNING "Short read from newly-erased block at 0x%08x. Wanted %d, got %zd\n", ofs, readlen, retlen);
344                                 goto bad;
345                         }
346                         for (i=0; i<readlen; i += sizeof(unsigned long)) {
347                                 /* It's OK. We know it's properly aligned */
348                                 unsigned long datum = *(unsigned long *)(&ebuf[i]);
349                                 if (datum + 1) {
350                                         bad_offset += i;
351                                         printk(KERN_WARNING "Newly-erased block contained word 0x%lx at offset 0x%08x\n", datum, bad_offset);
352                                 bad: 
353                                         if (!jffs2_cleanmarker_oob(c))
354                                                 jffs2_free_raw_node_ref(marker_ref);
355                                         kfree(ebuf);
356                                 bad2:
357                                         spin_lock(&c->erase_completion_lock);
358                                         /* Stick it on a list (any list) so
359                                            erase_failed can take it right off
360                                            again.  Silly, but shouldn't happen
361                                            often. */
362                                         list_add(&jeb->list, &c->erasing_list);
363                                         spin_unlock(&c->erase_completion_lock);
364                                         jffs2_erase_failed(c, jeb, bad_offset);
365                                         return;
366                                 }
367                         }
368                         ofs += readlen;
369                         cond_resched();
370                 }
371                 kfree(ebuf);
372         }
373
374         bad_offset = jeb->offset;
375
376         /* Write the erase complete marker */   
377         D1(printk(KERN_DEBUG "Writing erased marker to block at 0x%08x\n", jeb->offset));
378         if (jffs2_cleanmarker_oob(c)) {
379
380                 if (jffs2_write_nand_cleanmarker(c, jeb))
381                         goto bad2;
382                         
383                 jeb->first_node = jeb->last_node = NULL;
384
385                 jeb->free_size = c->sector_size;
386                 jeb->used_size = 0;
387                 jeb->dirty_size = 0;
388                 jeb->wasted_size = 0;
389         } else {
390                 struct kvec vecs[1];
391                 struct jffs2_unknown_node marker = {
392                         .magic =        cpu_to_je16(JFFS2_MAGIC_BITMASK),
393                         .nodetype =     cpu_to_je16(JFFS2_NODETYPE_CLEANMARKER),
394                         .totlen =       cpu_to_je32(c->cleanmarker_size)
395                 };
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) {
404                         printk(KERN_WARNING "Write clean marker to block at 0x%08x failed: %d\n",
405                                jeb->offset, ret);
406                         goto bad2;
407                 }
408                 if (retlen != sizeof(marker)) {
409                         printk(KERN_WARNING "Short write to newly-erased block at 0x%08x: Wanted %zd, got %zd\n",
410                                jeb->offset, sizeof(marker), retlen);
411                         goto bad2;
412                 }
413
414                 marker_ref->next_in_ino = NULL;
415                 marker_ref->next_phys = NULL;
416                 marker_ref->flash_offset = jeb->offset | REF_NORMAL;
417                 marker_ref->__totlen = c->cleanmarker_size;
418                         
419                 jeb->first_node = jeb->last_node = marker_ref;
420                         
421                 jeb->free_size = c->sector_size - c->cleanmarker_size;
422                 jeb->used_size = c->cleanmarker_size;
423                 jeb->dirty_size = 0;
424                 jeb->wasted_size = 0;
425         }
426
427         spin_lock(&c->erase_completion_lock);
428         c->erasing_size -= c->sector_size;
429         c->free_size += jeb->free_size;
430         c->used_size += jeb->used_size;
431
432         ACCT_SANITY_CHECK(c,jeb);
433         D1(ACCT_PARANOIA_CHECK(jeb));
434
435         list_add_tail(&jeb->list, &c->free_list);
436         c->nr_erasing_blocks--;
437         c->nr_free_blocks++;
438         spin_unlock(&c->erase_completion_lock);
439         wake_up(&c->erase_wait);
440 }
441