[PATCH] KVM: fix lockup on 32-bit intel hosts with nx disabled in the bios
[linux-2.6] / 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 1.85 2005/09/20 14:53:15 dedekind Exp $
11  *
12  */
13
14 #include <linux/kernel.h>
15 #include <linux/slab.h>
16 #include <linux/mtd/mtd.h>
17 #include <linux/compiler.h>
18 #include <linux/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_mark_erased_block(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb);
34
35 static void jffs2_erase_block(struct jffs2_sb_info *c,
36                               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         D1(printk(KERN_DEBUG "jffs2_erase_block(): erase block %#08x (range %#08x-%#08x)\n",
51                                 jeb->offset, jeb->offset, jeb->offset + c->sector_size));
52         instr = kmalloc(sizeof(struct erase_info) + sizeof(struct erase_priv_struct), GFP_KERNEL);
53         if (!instr) {
54                 printk(KERN_WARNING "kmalloc for struct erase_info in jffs2_erase_block failed. Refiling block for later\n");
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                 return;
62         }
63
64         memset(instr, 0, sizeof(*instr));
65
66         instr->mtd = c->mtd;
67         instr->addr = jeb->offset;
68         instr->len = c->sector_size;
69         instr->callback = jffs2_erase_callback;
70         instr->priv = (unsigned long)(&instr[1]);
71         instr->fail_addr = 0xffffffff;
72
73         ((struct erase_priv_struct *)instr->priv)->jeb = jeb;
74         ((struct erase_priv_struct *)instr->priv)->c = c;
75
76         ret = c->mtd->erase(c->mtd, instr);
77         if (!ret)
78                 return;
79
80         bad_offset = instr->fail_addr;
81         kfree(instr);
82 #endif /* __ECOS */
83
84         if (ret == -ENOMEM || ret == -EAGAIN) {
85                 /* Erase failed immediately. Refile it on the list */
86                 D1(printk(KERN_DEBUG "Erase at 0x%08x failed: %d. Refiling on erase_pending_list\n", jeb->offset, ret));
87                 spin_lock(&c->erase_completion_lock);
88                 list_move(&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_jeb_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_move_tail(&jeb->list, &c->erase_complete_list);
163         spin_unlock(&c->erase_completion_lock);
164         /* Ensure that kupdated calls us again to mark them clean */
165         jffs2_erase_pending_trigger(c);
166 }
167
168 static void jffs2_erase_failed(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, uint32_t bad_offset)
169 {
170         /* For NAND, if the failure did not occur at the device level for a
171            specific physical page, don't bother updating the bad block table. */
172         if (jffs2_cleanmarker_oob(c) && (bad_offset != 0xffffffff)) {
173                 /* We had a device-level failure to erase.  Let's see if we've
174                    failed too many times. */
175                 if (!jffs2_write_nand_badblock(c, jeb, bad_offset)) {
176                         /* We'd like to give this block another try. */
177                         spin_lock(&c->erase_completion_lock);
178                         list_move(&jeb->list, &c->erase_pending_list);
179                         c->erasing_size -= c->sector_size;
180                         c->dirty_size += c->sector_size;
181                         jeb->dirty_size = c->sector_size;
182                         spin_unlock(&c->erase_completion_lock);
183                         return;
184                 }
185         }
186
187         spin_lock(&c->erase_completion_lock);
188         c->erasing_size -= c->sector_size;
189         c->bad_size += c->sector_size;
190         list_move(&jeb->list, &c->bad_list);
191         c->nr_erasing_blocks--;
192         spin_unlock(&c->erase_completion_lock);
193         wake_up(&c->erase_wait);
194 }
195
196 #ifndef __ECOS
197 static void jffs2_erase_callback(struct erase_info *instr)
198 {
199         struct erase_priv_struct *priv = (void *)instr->priv;
200
201         if(instr->state != MTD_ERASE_DONE) {
202                 printk(KERN_WARNING "Erase at 0x%08x finished, but state != MTD_ERASE_DONE. State is 0x%x instead.\n", instr->addr, instr->state);
203                 jffs2_erase_failed(priv->c, priv->jeb, instr->fail_addr);
204         } else {
205                 jffs2_erase_succeeded(priv->c, priv->jeb);
206         }
207         kfree(instr);
208 }
209 #endif /* !__ECOS */
210
211 /* Hmmm. Maybe we should accept the extra space it takes and make
212    this a standard doubly-linked list? */
213 static inline void jffs2_remove_node_refs_from_ino_list(struct jffs2_sb_info *c,
214                         struct jffs2_raw_node_ref *ref, struct jffs2_eraseblock *jeb)
215 {
216         struct jffs2_inode_cache *ic = NULL;
217         struct jffs2_raw_node_ref **prev;
218
219         prev = &ref->next_in_ino;
220
221         /* Walk the inode's list once, removing any nodes from this eraseblock */
222         while (1) {
223                 if (!(*prev)->next_in_ino) {
224                         /* We're looking at the jffs2_inode_cache, which is
225                            at the end of the linked list. Stash it and continue
226                            from the beginning of the list */
227                         ic = (struct jffs2_inode_cache *)(*prev);
228                         prev = &ic->nodes;
229                         continue;
230                 }
231
232                 if (SECTOR_ADDR((*prev)->flash_offset) == jeb->offset) {
233                         /* It's in the block we're erasing */
234                         struct jffs2_raw_node_ref *this;
235
236                         this = *prev;
237                         *prev = this->next_in_ino;
238                         this->next_in_ino = NULL;
239
240                         if (this == ref)
241                                 break;
242
243                         continue;
244                 }
245                 /* Not to be deleted. Skip */
246                 prev = &((*prev)->next_in_ino);
247         }
248
249         /* PARANOIA */
250         if (!ic) {
251                 JFFS2_WARNING("inode_cache/xattr_datum/xattr_ref"
252                               " not found in remove_node_refs()!!\n");
253                 return;
254         }
255
256         D1(printk(KERN_DEBUG "Removed nodes in range 0x%08x-0x%08x from ino #%u\n",
257                   jeb->offset, jeb->offset + c->sector_size, ic->ino));
258
259         D2({
260                 int i=0;
261                 struct jffs2_raw_node_ref *this;
262                 printk(KERN_DEBUG "After remove_node_refs_from_ino_list: \n" KERN_DEBUG);
263
264                 this = ic->nodes;
265
266                 while(this) {
267                         printk( "0x%08x(%d)->", ref_offset(this), ref_flags(this));
268                         if (++i == 5) {
269                                 printk("\n" KERN_DEBUG);
270                                 i=0;
271                         }
272                         this = this->next_in_ino;
273                 }
274                 printk("\n");
275         });
276
277         switch (ic->class) {
278 #ifdef CONFIG_JFFS2_FS_XATTR
279                 case RAWNODE_CLASS_XATTR_DATUM:
280                         jffs2_release_xattr_datum(c, (struct jffs2_xattr_datum *)ic);
281                         break;
282                 case RAWNODE_CLASS_XATTR_REF:
283                         jffs2_release_xattr_ref(c, (struct jffs2_xattr_ref *)ic);
284                         break;
285 #endif
286                 default:
287                         if (ic->nodes == (void *)ic && ic->nlink == 0)
288                                 jffs2_del_ino_cache(c, ic);
289         }
290 }
291
292 void jffs2_free_jeb_node_refs(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
293 {
294         struct jffs2_raw_node_ref *block, *ref;
295         D1(printk(KERN_DEBUG "Freeing all node refs for eraseblock offset 0x%08x\n", jeb->offset));
296
297         block = ref = jeb->first_node;
298
299         while (ref) {
300                 if (ref->flash_offset == REF_LINK_NODE) {
301                         ref = ref->next_in_ino;
302                         jffs2_free_refblock(block);
303                         block = ref;
304                         continue;
305                 }
306                 if (ref->flash_offset != REF_EMPTY_NODE && ref->next_in_ino)
307                         jffs2_remove_node_refs_from_ino_list(c, ref, jeb);
308                 /* else it was a non-inode node or already removed, so don't bother */
309
310                 ref++;
311         }
312         jeb->first_node = jeb->last_node = NULL;
313 }
314
315 static int jffs2_block_check_erase(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, uint32_t *bad_offset)
316 {
317         void *ebuf;
318         uint32_t ofs;
319         size_t retlen;
320         int ret = -EIO;
321
322         ebuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
323         if (!ebuf) {
324                 printk(KERN_WARNING "Failed to allocate page buffer for verifying erase at 0x%08x. Refiling\n", jeb->offset);
325                 return -EAGAIN;
326         }
327
328         D1(printk(KERN_DEBUG "Verifying erase at 0x%08x\n", jeb->offset));
329
330         for (ofs = jeb->offset; ofs < jeb->offset + c->sector_size; ) {
331                 uint32_t readlen = min((uint32_t)PAGE_SIZE, jeb->offset + c->sector_size - ofs);
332                 int i;
333
334                 *bad_offset = ofs;
335
336                 ret = jffs2_flash_read(c, ofs, readlen, &retlen, ebuf);
337                 if (ret) {
338                         printk(KERN_WARNING "Read of newly-erased block at 0x%08x failed: %d. Putting on bad_list\n", ofs, ret);
339                         goto fail;
340                 }
341                 if (retlen != readlen) {
342                         printk(KERN_WARNING "Short read from newly-erased block at 0x%08x. Wanted %d, got %zd\n", ofs, readlen, retlen);
343                         goto fail;
344                 }
345                 for (i=0; i<readlen; i += sizeof(unsigned long)) {
346                         /* It's OK. We know it's properly aligned */
347                         unsigned long *datum = ebuf + i;
348                         if (*datum + 1) {
349                                 *bad_offset += i;
350                                 printk(KERN_WARNING "Newly-erased block contained word 0x%lx at offset 0x%08x\n", *datum, *bad_offset);
351                                 goto fail;
352                         }
353                 }
354                 ofs += readlen;
355                 cond_resched();
356         }
357         ret = 0;
358 fail:
359         kfree(ebuf);
360         return ret;
361 }
362
363 static void jffs2_mark_erased_block(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
364 {
365         size_t retlen;
366         int ret;
367         uint32_t bad_offset;
368
369         switch (jffs2_block_check_erase(c, jeb, &bad_offset)) {
370         case -EAGAIN:   goto refile;
371         case -EIO:      goto filebad;
372         }
373
374         /* Write the erase complete marker */
375         D1(printk(KERN_DEBUG "Writing erased marker to block at 0x%08x\n", jeb->offset));
376         bad_offset = jeb->offset;
377
378         /* Cleanmarker in oob area or no cleanmarker at all ? */
379         if (jffs2_cleanmarker_oob(c) || c->cleanmarker_size == 0) {
380
381                 if (jffs2_cleanmarker_oob(c)) {
382                         if (jffs2_write_nand_cleanmarker(c, jeb))
383                                 goto filebad;
384                 }
385
386                 /* Everything else got zeroed before the erase */
387                 jeb->free_size = c->sector_size;
388         } else {
389
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                 jffs2_prealloc_raw_node_refs(c, jeb, 1);
398
399                 marker.hdr_crc = cpu_to_je32(crc32(0, &marker, sizeof(struct jffs2_unknown_node)-4));
400
401                 vecs[0].iov_base = (unsigned char *) &marker;
402                 vecs[0].iov_len = sizeof(marker);
403                 ret = jffs2_flash_direct_writev(c, vecs, 1, jeb->offset, &retlen);
404
405                 if (ret || retlen != sizeof(marker)) {
406                         if (ret)
407                                 printk(KERN_WARNING "Write clean marker to block at 0x%08x failed: %d\n",
408                                        jeb->offset, ret);
409                         else
410                                 printk(KERN_WARNING "Short write to newly-erased block at 0x%08x: Wanted %zd, got %zd\n",
411                                        jeb->offset, sizeof(marker), retlen);
412
413                         goto filebad;
414                 }
415
416                 /* Everything else got zeroed before the erase */
417                 jeb->free_size = c->sector_size;
418                 /* FIXME Special case for cleanmarker in empty block */
419                 jffs2_link_node_ref(c, jeb, jeb->offset | REF_NORMAL, c->cleanmarker_size, NULL);
420         }
421
422         spin_lock(&c->erase_completion_lock);
423         c->erasing_size -= c->sector_size;
424         c->free_size += jeb->free_size;
425         c->used_size += jeb->used_size;
426
427         jffs2_dbg_acct_sanity_check_nolock(c,jeb);
428         jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
429
430         list_add_tail(&jeb->list, &c->free_list);
431         c->nr_erasing_blocks--;
432         c->nr_free_blocks++;
433         spin_unlock(&c->erase_completion_lock);
434         wake_up(&c->erase_wait);
435         return;
436
437 filebad:
438         spin_lock(&c->erase_completion_lock);
439         /* Stick it on a list (any list) so erase_failed can take it
440            right off again.  Silly, but shouldn't happen often. */
441         list_add(&jeb->list, &c->erasing_list);
442         spin_unlock(&c->erase_completion_lock);
443         jffs2_erase_failed(c, jeb, bad_offset);
444         return;
445
446 refile:
447         /* Stick it back on the list from whence it came and come back later */
448         jffs2_erase_pending_trigger(c);
449         spin_lock(&c->erase_completion_lock);
450         list_add(&jeb->list, &c->erase_complete_list);
451         spin_unlock(&c->erase_completion_lock);
452         return;
453 }