Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / 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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14
15 #include <linux/kernel.h>
16 #include <linux/slab.h>
17 #include <linux/mtd/mtd.h>
18 #include <linux/compiler.h>
19 #include <linux/crc32.h>
20 #include <linux/sched.h>
21 #include <linux/pagemap.h>
22 #include "nodelist.h"
23
24 struct erase_priv_struct {
25         struct jffs2_eraseblock *jeb;
26         struct jffs2_sb_info *c;
27 };
28
29 #ifndef __ECOS
30 static void jffs2_erase_callback(struct erase_info *);
31 #endif
32 static void jffs2_erase_failed(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, uint32_t bad_offset);
33 static void jffs2_erase_succeeded(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 static void jffs2_erase_block(struct jffs2_sb_info *c,
37                               struct jffs2_eraseblock *jeb)
38 {
39         int ret;
40         uint32_t bad_offset;
41 #ifdef __ECOS
42        ret = jffs2_flash_erase(c, jeb);
43        if (!ret) {
44                jffs2_erase_succeeded(c, jeb);
45                return;
46        }
47        bad_offset = jeb->offset;
48 #else /* Linux */
49         struct erase_info *instr;
50
51         jffs2_dbg(1, "%s(): erase block %#08x (range %#08x-%#08x)\n",
52                   __func__,
53                   jeb->offset, jeb->offset, jeb->offset + c->sector_size);
54         instr = kmalloc(sizeof(struct erase_info) + sizeof(struct erase_priv_struct), GFP_KERNEL);
55         if (!instr) {
56                 pr_warn("kmalloc for struct erase_info in jffs2_erase_block failed. Refiling block for later\n");
57                 mutex_lock(&c->erase_free_sem);
58                 spin_lock(&c->erase_completion_lock);
59                 list_move(&jeb->list, &c->erase_pending_list);
60                 c->erasing_size -= c->sector_size;
61                 c->dirty_size += c->sector_size;
62                 jeb->dirty_size = c->sector_size;
63                 spin_unlock(&c->erase_completion_lock);
64                 mutex_unlock(&c->erase_free_sem);
65                 return;
66         }
67
68         memset(instr, 0, sizeof(*instr));
69
70         instr->mtd = c->mtd;
71         instr->addr = jeb->offset;
72         instr->len = c->sector_size;
73         instr->callback = jffs2_erase_callback;
74         instr->priv = (unsigned long)(&instr[1]);
75
76         ((struct erase_priv_struct *)instr->priv)->jeb = jeb;
77         ((struct erase_priv_struct *)instr->priv)->c = c;
78
79         ret = mtd_erase(c->mtd, instr);
80         if (!ret)
81                 return;
82
83         bad_offset = instr->fail_addr;
84         kfree(instr);
85 #endif /* __ECOS */
86
87         if (ret == -ENOMEM || ret == -EAGAIN) {
88                 /* Erase failed immediately. Refile it on the list */
89                 jffs2_dbg(1, "Erase at 0x%08x failed: %d. Refiling on erase_pending_list\n",
90                           jeb->offset, ret);
91                 mutex_lock(&c->erase_free_sem);
92                 spin_lock(&c->erase_completion_lock);
93                 list_move(&jeb->list, &c->erase_pending_list);
94                 c->erasing_size -= c->sector_size;
95                 c->dirty_size += c->sector_size;
96                 jeb->dirty_size = c->sector_size;
97                 spin_unlock(&c->erase_completion_lock);
98                 mutex_unlock(&c->erase_free_sem);
99                 return;
100         }
101
102         if (ret == -EROFS)
103                 pr_warn("Erase at 0x%08x failed immediately: -EROFS. Is the sector locked?\n",
104                         jeb->offset);
105         else
106                 pr_warn("Erase at 0x%08x failed immediately: errno %d\n",
107                         jeb->offset, ret);
108
109         jffs2_erase_failed(c, jeb, bad_offset);
110 }
111
112 int jffs2_erase_pending_blocks(struct jffs2_sb_info *c, int count)
113 {
114         struct jffs2_eraseblock *jeb;
115         int work_done = 0;
116
117         mutex_lock(&c->erase_free_sem);
118
119         spin_lock(&c->erase_completion_lock);
120
121         while (!list_empty(&c->erase_complete_list) ||
122                !list_empty(&c->erase_pending_list)) {
123
124                 if (!list_empty(&c->erase_complete_list)) {
125                         jeb = list_entry(c->erase_complete_list.next, struct jffs2_eraseblock, list);
126                         list_move(&jeb->list, &c->erase_checking_list);
127                         spin_unlock(&c->erase_completion_lock);
128                         mutex_unlock(&c->erase_free_sem);
129                         jffs2_mark_erased_block(c, jeb);
130
131                         work_done++;
132                         if (!--count) {
133                                 jffs2_dbg(1, "Count reached. jffs2_erase_pending_blocks leaving\n");
134                                 goto done;
135                         }
136
137                 } else if (!list_empty(&c->erase_pending_list)) {
138                         jeb = list_entry(c->erase_pending_list.next, struct jffs2_eraseblock, list);
139                         jffs2_dbg(1, "Starting erase of pending block 0x%08x\n",
140                                   jeb->offset);
141                         list_del(&jeb->list);
142                         c->erasing_size += c->sector_size;
143                         c->wasted_size -= jeb->wasted_size;
144                         c->free_size -= jeb->free_size;
145                         c->used_size -= jeb->used_size;
146                         c->dirty_size -= jeb->dirty_size;
147                         jeb->wasted_size = jeb->used_size = jeb->dirty_size = jeb->free_size = 0;
148                         jffs2_free_jeb_node_refs(c, jeb);
149                         list_add(&jeb->list, &c->erasing_list);
150                         spin_unlock(&c->erase_completion_lock);
151                         mutex_unlock(&c->erase_free_sem);
152
153                         jffs2_erase_block(c, jeb);
154
155                 } else {
156                         BUG();
157                 }
158
159                 /* Be nice */
160                 cond_resched();
161                 mutex_lock(&c->erase_free_sem);
162                 spin_lock(&c->erase_completion_lock);
163         }
164
165         spin_unlock(&c->erase_completion_lock);
166         mutex_unlock(&c->erase_free_sem);
167  done:
168         jffs2_dbg(1, "jffs2_erase_pending_blocks completed\n");
169         return work_done;
170 }
171
172 static void jffs2_erase_succeeded(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
173 {
174         jffs2_dbg(1, "Erase completed successfully at 0x%08x\n", jeb->offset);
175         mutex_lock(&c->erase_free_sem);
176         spin_lock(&c->erase_completion_lock);
177         list_move_tail(&jeb->list, &c->erase_complete_list);
178         /* Wake the GC thread to mark them clean */
179         jffs2_garbage_collect_trigger(c);
180         spin_unlock(&c->erase_completion_lock);
181         mutex_unlock(&c->erase_free_sem);
182         wake_up(&c->erase_wait);
183 }
184
185 static void jffs2_erase_failed(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, uint32_t bad_offset)
186 {
187         /* For NAND, if the failure did not occur at the device level for a
188            specific physical page, don't bother updating the bad block table. */
189         if (jffs2_cleanmarker_oob(c) && (bad_offset != (uint32_t)MTD_FAIL_ADDR_UNKNOWN)) {
190                 /* We had a device-level failure to erase.  Let's see if we've
191                    failed too many times. */
192                 if (!jffs2_write_nand_badblock(c, jeb, bad_offset)) {
193                         /* We'd like to give this block another try. */
194                         mutex_lock(&c->erase_free_sem);
195                         spin_lock(&c->erase_completion_lock);
196                         list_move(&jeb->list, &c->erase_pending_list);
197                         c->erasing_size -= c->sector_size;
198                         c->dirty_size += c->sector_size;
199                         jeb->dirty_size = c->sector_size;
200                         spin_unlock(&c->erase_completion_lock);
201                         mutex_unlock(&c->erase_free_sem);
202                         return;
203                 }
204         }
205
206         mutex_lock(&c->erase_free_sem);
207         spin_lock(&c->erase_completion_lock);
208         c->erasing_size -= c->sector_size;
209         c->bad_size += c->sector_size;
210         list_move(&jeb->list, &c->bad_list);
211         c->nr_erasing_blocks--;
212         spin_unlock(&c->erase_completion_lock);
213         mutex_unlock(&c->erase_free_sem);
214         wake_up(&c->erase_wait);
215 }
216
217 #ifndef __ECOS
218 static void jffs2_erase_callback(struct erase_info *instr)
219 {
220         struct erase_priv_struct *priv = (void *)instr->priv;
221
222         if(instr->state != MTD_ERASE_DONE) {
223                 pr_warn("Erase at 0x%08llx finished, but state != MTD_ERASE_DONE. State is 0x%x instead.\n",
224                         (unsigned long long)instr->addr, instr->state);
225                 jffs2_erase_failed(priv->c, priv->jeb, instr->fail_addr);
226         } else {
227                 jffs2_erase_succeeded(priv->c, priv->jeb);
228         }
229         kfree(instr);
230 }
231 #endif /* !__ECOS */
232
233 /* Hmmm. Maybe we should accept the extra space it takes and make
234    this a standard doubly-linked list? */
235 static inline void jffs2_remove_node_refs_from_ino_list(struct jffs2_sb_info *c,
236                         struct jffs2_raw_node_ref *ref, struct jffs2_eraseblock *jeb)
237 {
238         struct jffs2_inode_cache *ic = NULL;
239         struct jffs2_raw_node_ref **prev;
240
241         prev = &ref->next_in_ino;
242
243         /* Walk the inode's list once, removing any nodes from this eraseblock */
244         while (1) {
245                 if (!(*prev)->next_in_ino) {
246                         /* We're looking at the jffs2_inode_cache, which is
247                            at the end of the linked list. Stash it and continue
248                            from the beginning of the list */
249                         ic = (struct jffs2_inode_cache *)(*prev);
250                         prev = &ic->nodes;
251                         continue;
252                 }
253
254                 if (SECTOR_ADDR((*prev)->flash_offset) == jeb->offset) {
255                         /* It's in the block we're erasing */
256                         struct jffs2_raw_node_ref *this;
257
258                         this = *prev;
259                         *prev = this->next_in_ino;
260                         this->next_in_ino = NULL;
261
262                         if (this == ref)
263                                 break;
264
265                         continue;
266                 }
267                 /* Not to be deleted. Skip */
268                 prev = &((*prev)->next_in_ino);
269         }
270
271         /* PARANOIA */
272         if (!ic) {
273                 JFFS2_WARNING("inode_cache/xattr_datum/xattr_ref"
274                               " not found in remove_node_refs()!!\n");
275                 return;
276         }
277
278         jffs2_dbg(1, "Removed nodes in range 0x%08x-0x%08x from ino #%u\n",
279                   jeb->offset, jeb->offset + c->sector_size, ic->ino);
280
281         D2({
282                 int i=0;
283                 struct jffs2_raw_node_ref *this;
284                 printk(KERN_DEBUG "After remove_node_refs_from_ino_list: \n");
285
286                 this = ic->nodes;
287
288                 printk(KERN_DEBUG);
289                 while(this) {
290                         pr_cont("0x%08x(%d)->",
291                                ref_offset(this), ref_flags(this));
292                         if (++i == 5) {
293                                 printk(KERN_DEBUG);
294                                 i=0;
295                         }
296                         this = this->next_in_ino;
297                 }
298                 pr_cont("\n");
299         });
300
301         switch (ic->class) {
302 #ifdef CONFIG_JFFS2_FS_XATTR
303                 case RAWNODE_CLASS_XATTR_DATUM:
304                         jffs2_release_xattr_datum(c, (struct jffs2_xattr_datum *)ic);
305                         break;
306                 case RAWNODE_CLASS_XATTR_REF:
307                         jffs2_release_xattr_ref(c, (struct jffs2_xattr_ref *)ic);
308                         break;
309 #endif
310                 default:
311                         if (ic->nodes == (void *)ic && ic->pino_nlink == 0)
312                                 jffs2_del_ino_cache(c, ic);
313         }
314 }
315
316 void jffs2_free_jeb_node_refs(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
317 {
318         struct jffs2_raw_node_ref *block, *ref;
319         jffs2_dbg(1, "Freeing all node refs for eraseblock offset 0x%08x\n",
320                   jeb->offset);
321
322         block = ref = jeb->first_node;
323
324         while (ref) {
325                 if (ref->flash_offset == REF_LINK_NODE) {
326                         ref = ref->next_in_ino;
327                         jffs2_free_refblock(block);
328                         block = ref;
329                         continue;
330                 }
331                 if (ref->flash_offset != REF_EMPTY_NODE && ref->next_in_ino)
332                         jffs2_remove_node_refs_from_ino_list(c, ref, jeb);
333                 /* else it was a non-inode node or already removed, so don't bother */
334
335                 ref++;
336         }
337         jeb->first_node = jeb->last_node = NULL;
338 }
339
340 static int jffs2_block_check_erase(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, uint32_t *bad_offset)
341 {
342         void *ebuf;
343         uint32_t ofs;
344         size_t retlen;
345         int ret;
346         unsigned long *wordebuf;
347
348         ret = mtd_point(c->mtd, jeb->offset, c->sector_size, &retlen,
349                         &ebuf, NULL);
350         if (ret != -EOPNOTSUPP) {
351                 if (ret) {
352                         jffs2_dbg(1, "MTD point failed %d\n", ret);
353                         goto do_flash_read;
354                 }
355                 if (retlen < c->sector_size) {
356                         /* Don't muck about if it won't let us point to the whole erase sector */
357                         jffs2_dbg(1, "MTD point returned len too short: 0x%zx\n",
358                                   retlen);
359                         mtd_unpoint(c->mtd, jeb->offset, retlen);
360                         goto do_flash_read;
361                 }
362                 wordebuf = ebuf-sizeof(*wordebuf);
363                 retlen /= sizeof(*wordebuf);
364                 do {
365                    if (*++wordebuf != ~0)
366                            break;
367                 } while(--retlen);
368                 mtd_unpoint(c->mtd, jeb->offset, c->sector_size);
369                 if (retlen) {
370                         pr_warn("Newly-erased block contained word 0x%lx at offset 0x%08tx\n",
371                                 *wordebuf,
372                                 jeb->offset +
373                                 c->sector_size-retlen * sizeof(*wordebuf));
374                         return -EIO;
375                 }
376                 return 0;
377         }
378  do_flash_read:
379         ebuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
380         if (!ebuf) {
381                 pr_warn("Failed to allocate page buffer for verifying erase at 0x%08x. Refiling\n",
382                         jeb->offset);
383                 return -EAGAIN;
384         }
385
386         jffs2_dbg(1, "Verifying erase at 0x%08x\n", jeb->offset);
387
388         for (ofs = jeb->offset; ofs < jeb->offset + c->sector_size; ) {
389                 uint32_t readlen = min((uint32_t)PAGE_SIZE, jeb->offset + c->sector_size - ofs);
390                 int i;
391
392                 *bad_offset = ofs;
393
394                 ret = mtd_read(c->mtd, ofs, readlen, &retlen, ebuf);
395                 if (ret) {
396                         pr_warn("Read of newly-erased block at 0x%08x failed: %d. Putting on bad_list\n",
397                                 ofs, ret);
398                         ret = -EIO;
399                         goto fail;
400                 }
401                 if (retlen != readlen) {
402                         pr_warn("Short read from newly-erased block at 0x%08x. Wanted %d, got %zd\n",
403                                 ofs, readlen, retlen);
404                         ret = -EIO;
405                         goto fail;
406                 }
407                 for (i=0; i<readlen; i += sizeof(unsigned long)) {
408                         /* It's OK. We know it's properly aligned */
409                         unsigned long *datum = ebuf + i;
410                         if (*datum + 1) {
411                                 *bad_offset += i;
412                                 pr_warn("Newly-erased block contained word 0x%lx at offset 0x%08x\n",
413                                         *datum, *bad_offset);
414                                 ret = -EIO;
415                                 goto fail;
416                         }
417                 }
418                 ofs += readlen;
419                 cond_resched();
420         }
421         ret = 0;
422 fail:
423         kfree(ebuf);
424         return ret;
425 }
426
427 static void jffs2_mark_erased_block(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
428 {
429         size_t retlen;
430         int ret;
431         uint32_t uninitialized_var(bad_offset);
432
433         switch (jffs2_block_check_erase(c, jeb, &bad_offset)) {
434         case -EAGAIN:   goto refile;
435         case -EIO:      goto filebad;
436         }
437
438         /* Write the erase complete marker */
439         jffs2_dbg(1, "Writing erased marker to block at 0x%08x\n", jeb->offset);
440         bad_offset = jeb->offset;
441
442         /* Cleanmarker in oob area or no cleanmarker at all ? */
443         if (jffs2_cleanmarker_oob(c) || c->cleanmarker_size == 0) {
444
445                 if (jffs2_cleanmarker_oob(c)) {
446                         if (jffs2_write_nand_cleanmarker(c, jeb))
447                                 goto filebad;
448                 }
449         } else {
450
451                 struct kvec vecs[1];
452                 struct jffs2_unknown_node marker = {
453                         .magic =        cpu_to_je16(JFFS2_MAGIC_BITMASK),
454                         .nodetype =     cpu_to_je16(JFFS2_NODETYPE_CLEANMARKER),
455                         .totlen =       cpu_to_je32(c->cleanmarker_size)
456                 };
457
458                 jffs2_prealloc_raw_node_refs(c, jeb, 1);
459
460                 marker.hdr_crc = cpu_to_je32(crc32(0, &marker, sizeof(struct jffs2_unknown_node)-4));
461
462                 vecs[0].iov_base = (unsigned char *) &marker;
463                 vecs[0].iov_len = sizeof(marker);
464                 ret = jffs2_flash_direct_writev(c, vecs, 1, jeb->offset, &retlen);
465
466                 if (ret || retlen != sizeof(marker)) {
467                         if (ret)
468                                 pr_warn("Write clean marker to block at 0x%08x failed: %d\n",
469                                        jeb->offset, ret);
470                         else
471                                 pr_warn("Short write to newly-erased block at 0x%08x: Wanted %zd, got %zd\n",
472                                        jeb->offset, sizeof(marker), retlen);
473
474                         goto filebad;
475                 }
476         }
477         /* Everything else got zeroed before the erase */
478         jeb->free_size = c->sector_size;
479
480         mutex_lock(&c->erase_free_sem);
481         spin_lock(&c->erase_completion_lock);
482
483         c->erasing_size -= c->sector_size;
484         c->free_size += c->sector_size;
485
486         /* Account for cleanmarker now, if it's in-band */
487         if (c->cleanmarker_size && !jffs2_cleanmarker_oob(c))
488                 jffs2_link_node_ref(c, jeb, jeb->offset | REF_NORMAL, c->cleanmarker_size, NULL);
489
490         list_move_tail(&jeb->list, &c->free_list);
491         c->nr_erasing_blocks--;
492         c->nr_free_blocks++;
493
494         jffs2_dbg_acct_sanity_check_nolock(c, jeb);
495         jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
496
497         spin_unlock(&c->erase_completion_lock);
498         mutex_unlock(&c->erase_free_sem);
499         wake_up(&c->erase_wait);
500         return;
501
502 filebad:
503         jffs2_erase_failed(c, jeb, bad_offset);
504         return;
505
506 refile:
507         /* Stick it back on the list from whence it came and come back later */
508         mutex_lock(&c->erase_free_sem);
509         spin_lock(&c->erase_completion_lock);
510         jffs2_garbage_collect_trigger(c);
511         list_move(&jeb->list, &c->erase_complete_list);
512         spin_unlock(&c->erase_completion_lock);
513         mutex_unlock(&c->erase_free_sem);
514         return;
515 }