Loading...
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/mtd/mtd.h>
14#include <linux/compiler.h>
15#include <linux/sched.h> /* For cond_resched() */
16#include "nodelist.h"
17#include "debug.h"
18
19/**
20 * jffs2_reserve_space - request physical space to write nodes to flash
21 * @c: superblock info
22 * @minsize: Minimum acceptable size of allocation
23 * @len: Returned value of allocation length
24 * @prio: Allocation type - ALLOC_{NORMAL,DELETION}
25 *
26 * Requests a block of physical space on the flash. Returns zero for success
27 * and puts 'len' into the appropriate place, or returns -ENOSPC or other
28 * error if appropriate. Doesn't return len since that's
29 *
30 * If it returns zero, jffs2_reserve_space() also downs the per-filesystem
31 * allocation semaphore, to prevent more than one allocation from being
32 * active at any time. The semaphore is later released by jffs2_commit_allocation()
33 *
34 * jffs2_reserve_space() may trigger garbage collection in order to make room
35 * for the requested allocation.
36 */
37
38static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
39 uint32_t *len, uint32_t sumsize);
40
41int jffs2_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
42 uint32_t *len, int prio, uint32_t sumsize)
43{
44 int ret = -EAGAIN;
45 int blocksneeded = c->resv_blocks_write;
46 /* align it */
47 minsize = PAD(minsize);
48
49 D1(printk(KERN_DEBUG "jffs2_reserve_space(): Requested 0x%x bytes\n", minsize));
50 mutex_lock(&c->alloc_sem);
51
52 D1(printk(KERN_DEBUG "jffs2_reserve_space(): alloc sem got\n"));
53
54 spin_lock(&c->erase_completion_lock);
55
56 /* this needs a little more thought (true <tglx> :)) */
57 while(ret == -EAGAIN) {
58 while(c->nr_free_blocks + c->nr_erasing_blocks < blocksneeded) {
59 uint32_t dirty, avail;
60
61 /* calculate real dirty size
62 * dirty_size contains blocks on erase_pending_list
63 * those blocks are counted in c->nr_erasing_blocks.
64 * If one block is actually erased, it is not longer counted as dirty_space
65 * but it is counted in c->nr_erasing_blocks, so we add it and subtract it
66 * with c->nr_erasing_blocks * c->sector_size again.
67 * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks
68 * This helps us to force gc and pick eventually a clean block to spread the load.
69 * We add unchecked_size here, as we hopefully will find some space to use.
70 * This will affect the sum only once, as gc first finishes checking
71 * of nodes.
72 */
73 dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size + c->unchecked_size;
74 if (dirty < c->nospc_dirty_size) {
75 if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) {
76 D1(printk(KERN_NOTICE "jffs2_reserve_space(): Low on dirty space to GC, but it's a deletion. Allowing...\n"));
77 break;
78 }
79 D1(printk(KERN_DEBUG "dirty size 0x%08x + unchecked_size 0x%08x < nospc_dirty_size 0x%08x, returning -ENOSPC\n",
80 dirty, c->unchecked_size, c->sector_size));
81
82 spin_unlock(&c->erase_completion_lock);
83 mutex_unlock(&c->alloc_sem);
84 return -ENOSPC;
85 }
86
87 /* Calc possibly available space. Possibly available means that we
88 * don't know, if unchecked size contains obsoleted nodes, which could give us some
89 * more usable space. This will affect the sum only once, as gc first finishes checking
90 * of nodes.
91 + Return -ENOSPC, if the maximum possibly available space is less or equal than
92 * blocksneeded * sector_size.
93 * This blocks endless gc looping on a filesystem, which is nearly full, even if
94 * the check above passes.
95 */
96 avail = c->free_size + c->dirty_size + c->erasing_size + c->unchecked_size;
97 if ( (avail / c->sector_size) <= blocksneeded) {
98 if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) {
99 D1(printk(KERN_NOTICE "jffs2_reserve_space(): Low on possibly available space, but it's a deletion. Allowing...\n"));
100 break;
101 }
102
103 D1(printk(KERN_DEBUG "max. available size 0x%08x < blocksneeded * sector_size 0x%08x, returning -ENOSPC\n",
104 avail, blocksneeded * c->sector_size));
105 spin_unlock(&c->erase_completion_lock);
106 mutex_unlock(&c->alloc_sem);
107 return -ENOSPC;
108 }
109
110 mutex_unlock(&c->alloc_sem);
111
112 D1(printk(KERN_DEBUG "Triggering GC pass. nr_free_blocks %d, nr_erasing_blocks %d, free_size 0x%08x, dirty_size 0x%08x, wasted_size 0x%08x, used_size 0x%08x, erasing_size 0x%08x, bad_size 0x%08x (total 0x%08x of 0x%08x)\n",
113 c->nr_free_blocks, c->nr_erasing_blocks, c->free_size, c->dirty_size, c->wasted_size, c->used_size, c->erasing_size, c->bad_size,
114 c->free_size + c->dirty_size + c->wasted_size + c->used_size + c->erasing_size + c->bad_size, c->flash_size));
115 spin_unlock(&c->erase_completion_lock);
116
117 ret = jffs2_garbage_collect_pass(c);
118
119 if (ret == -EAGAIN) {
120 spin_lock(&c->erase_completion_lock);
121 if (c->nr_erasing_blocks &&
122 list_empty(&c->erase_pending_list) &&
123 list_empty(&c->erase_complete_list)) {
124 DECLARE_WAITQUEUE(wait, current);
125 set_current_state(TASK_UNINTERRUPTIBLE);
126 add_wait_queue(&c->erase_wait, &wait);
127 D1(printk(KERN_DEBUG "%s waiting for erase to complete\n", __func__));
128 spin_unlock(&c->erase_completion_lock);
129
130 schedule();
131 } else
132 spin_unlock(&c->erase_completion_lock);
133 } else if (ret)
134 return ret;
135
136 cond_resched();
137
138 if (signal_pending(current))
139 return -EINTR;
140
141 mutex_lock(&c->alloc_sem);
142 spin_lock(&c->erase_completion_lock);
143 }
144
145 ret = jffs2_do_reserve_space(c, minsize, len, sumsize);
146 if (ret) {
147 D1(printk(KERN_DEBUG "jffs2_reserve_space: ret is %d\n", ret));
148 }
149 }
150 spin_unlock(&c->erase_completion_lock);
151 if (!ret)
152 ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1);
153 if (ret)
154 mutex_unlock(&c->alloc_sem);
155 return ret;
156}
157
158int jffs2_reserve_space_gc(struct jffs2_sb_info *c, uint32_t minsize,
159 uint32_t *len, uint32_t sumsize)
160{
161 int ret = -EAGAIN;
162 minsize = PAD(minsize);
163
164 D1(printk(KERN_DEBUG "jffs2_reserve_space_gc(): Requested 0x%x bytes\n", minsize));
165
166 spin_lock(&c->erase_completion_lock);
167 while(ret == -EAGAIN) {
168 ret = jffs2_do_reserve_space(c, minsize, len, sumsize);
169 if (ret) {
170 D1(printk(KERN_DEBUG "jffs2_reserve_space_gc: looping, ret is %d\n", ret));
171 }
172 }
173 spin_unlock(&c->erase_completion_lock);
174 if (!ret)
175 ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1);
176
177 return ret;
178}
179
180
181/* Classify nextblock (clean, dirty of verydirty) and force to select an other one */
182
183static void jffs2_close_nextblock(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
184{
185
186 if (c->nextblock == NULL) {
187 D1(printk(KERN_DEBUG "jffs2_close_nextblock: Erase block at 0x%08x has already been placed in a list\n",
188 jeb->offset));
189 return;
190 }
191 /* Check, if we have a dirty block now, or if it was dirty already */
192 if (ISDIRTY (jeb->wasted_size + jeb->dirty_size)) {
193 c->dirty_size += jeb->wasted_size;
194 c->wasted_size -= jeb->wasted_size;
195 jeb->dirty_size += jeb->wasted_size;
196 jeb->wasted_size = 0;
197 if (VERYDIRTY(c, jeb->dirty_size)) {
198 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to very_dirty_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
199 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
200 list_add_tail(&jeb->list, &c->very_dirty_list);
201 } else {
202 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to dirty_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
203 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
204 list_add_tail(&jeb->list, &c->dirty_list);
205 }
206 } else {
207 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
208 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
209 list_add_tail(&jeb->list, &c->clean_list);
210 }
211 c->nextblock = NULL;
212
213}
214
215/* Select a new jeb for nextblock */
216
217static int jffs2_find_nextblock(struct jffs2_sb_info *c)
218{
219 struct list_head *next;
220
221 /* Take the next block off the 'free' list */
222
223 if (list_empty(&c->free_list)) {
224
225 if (!c->nr_erasing_blocks &&
226 !list_empty(&c->erasable_list)) {
227 struct jffs2_eraseblock *ejeb;
228
229 ejeb = list_entry(c->erasable_list.next, struct jffs2_eraseblock, list);
230 list_move_tail(&ejeb->list, &c->erase_pending_list);
231 c->nr_erasing_blocks++;
232 jffs2_garbage_collect_trigger(c);
233 D1(printk(KERN_DEBUG "jffs2_find_nextblock: Triggering erase of erasable block at 0x%08x\n",
234 ejeb->offset));
235 }
236
237 if (!c->nr_erasing_blocks &&
238 !list_empty(&c->erasable_pending_wbuf_list)) {
239 D1(printk(KERN_DEBUG "jffs2_find_nextblock: Flushing write buffer\n"));
240 /* c->nextblock is NULL, no update to c->nextblock allowed */
241 spin_unlock(&c->erase_completion_lock);
242 jffs2_flush_wbuf_pad(c);
243 spin_lock(&c->erase_completion_lock);
244 /* Have another go. It'll be on the erasable_list now */
245 return -EAGAIN;
246 }
247
248 if (!c->nr_erasing_blocks) {
249 /* Ouch. We're in GC, or we wouldn't have got here.
250 And there's no space left. At all. */
251 printk(KERN_CRIT "Argh. No free space left for GC. nr_erasing_blocks is %d. nr_free_blocks is %d. (erasableempty: %s, erasingempty: %s, erasependingempty: %s)\n",
252 c->nr_erasing_blocks, c->nr_free_blocks, list_empty(&c->erasable_list)?"yes":"no",
253 list_empty(&c->erasing_list)?"yes":"no", list_empty(&c->erase_pending_list)?"yes":"no");
254 return -ENOSPC;
255 }
256
257 spin_unlock(&c->erase_completion_lock);
258 /* Don't wait for it; just erase one right now */
259 jffs2_erase_pending_blocks(c, 1);
260 spin_lock(&c->erase_completion_lock);
261
262 /* An erase may have failed, decreasing the
263 amount of free space available. So we must
264 restart from the beginning */
265 return -EAGAIN;
266 }
267
268 next = c->free_list.next;
269 list_del(next);
270 c->nextblock = list_entry(next, struct jffs2_eraseblock, list);
271 c->nr_free_blocks--;
272
273 jffs2_sum_reset_collected(c->summary); /* reset collected summary */
274
275#ifdef CONFIG_JFFS2_FS_WRITEBUFFER
276 /* adjust write buffer offset, else we get a non contiguous write bug */
277 if (!(c->wbuf_ofs % c->sector_size) && !c->wbuf_len)
278 c->wbuf_ofs = 0xffffffff;
279#endif
280
281 D1(printk(KERN_DEBUG "jffs2_find_nextblock(): new nextblock = 0x%08x\n", c->nextblock->offset));
282
283 return 0;
284}
285
286/* Called with alloc sem _and_ erase_completion_lock */
287static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
288 uint32_t *len, uint32_t sumsize)
289{
290 struct jffs2_eraseblock *jeb = c->nextblock;
291 uint32_t reserved_size; /* for summary information at the end of the jeb */
292 int ret;
293
294 restart:
295 reserved_size = 0;
296
297 if (jffs2_sum_active() && (sumsize != JFFS2_SUMMARY_NOSUM_SIZE)) {
298 /* NOSUM_SIZE means not to generate summary */
299
300 if (jeb) {
301 reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE);
302 dbg_summary("minsize=%d , jeb->free=%d ,"
303 "summary->size=%d , sumsize=%d\n",
304 minsize, jeb->free_size,
305 c->summary->sum_size, sumsize);
306 }
307
308 /* Is there enough space for writing out the current node, or we have to
309 write out summary information now, close this jeb and select new nextblock? */
310 if (jeb && (PAD(minsize) + PAD(c->summary->sum_size + sumsize +
311 JFFS2_SUMMARY_FRAME_SIZE) > jeb->free_size)) {
312
313 /* Has summary been disabled for this jeb? */
314 if (jffs2_sum_is_disabled(c->summary)) {
315 sumsize = JFFS2_SUMMARY_NOSUM_SIZE;
316 goto restart;
317 }
318
319 /* Writing out the collected summary information */
320 dbg_summary("generating summary for 0x%08x.\n", jeb->offset);
321 ret = jffs2_sum_write_sumnode(c);
322
323 if (ret)
324 return ret;
325
326 if (jffs2_sum_is_disabled(c->summary)) {
327 /* jffs2_write_sumnode() couldn't write out the summary information
328 diabling summary for this jeb and free the collected information
329 */
330 sumsize = JFFS2_SUMMARY_NOSUM_SIZE;
331 goto restart;
332 }
333
334 jffs2_close_nextblock(c, jeb);
335 jeb = NULL;
336 /* keep always valid value in reserved_size */
337 reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE);
338 }
339 } else {
340 if (jeb && minsize > jeb->free_size) {
341 uint32_t waste;
342
343 /* Skip the end of this block and file it as having some dirty space */
344 /* If there's a pending write to it, flush now */
345
346 if (jffs2_wbuf_dirty(c)) {
347 spin_unlock(&c->erase_completion_lock);
348 D1(printk(KERN_DEBUG "jffs2_do_reserve_space: Flushing write buffer\n"));
349 jffs2_flush_wbuf_pad(c);
350 spin_lock(&c->erase_completion_lock);
351 jeb = c->nextblock;
352 goto restart;
353 }
354
355 spin_unlock(&c->erase_completion_lock);
356
357 ret = jffs2_prealloc_raw_node_refs(c, jeb, 1);
358 if (ret)
359 return ret;
360 /* Just lock it again and continue. Nothing much can change because
361 we hold c->alloc_sem anyway. In fact, it's not entirely clear why
362 we hold c->erase_completion_lock in the majority of this function...
363 but that's a question for another (more caffeine-rich) day. */
364 spin_lock(&c->erase_completion_lock);
365
366 waste = jeb->free_size;
367 jffs2_link_node_ref(c, jeb,
368 (jeb->offset + c->sector_size - waste) | REF_OBSOLETE,
369 waste, NULL);
370 /* FIXME: that made it count as dirty. Convert to wasted */
371 jeb->dirty_size -= waste;
372 c->dirty_size -= waste;
373 jeb->wasted_size += waste;
374 c->wasted_size += waste;
375
376 jffs2_close_nextblock(c, jeb);
377 jeb = NULL;
378 }
379 }
380
381 if (!jeb) {
382
383 ret = jffs2_find_nextblock(c);
384 if (ret)
385 return ret;
386
387 jeb = c->nextblock;
388
389 if (jeb->free_size != c->sector_size - c->cleanmarker_size) {
390 printk(KERN_WARNING "Eep. Block 0x%08x taken from free_list had free_size of 0x%08x!!\n", jeb->offset, jeb->free_size);
391 goto restart;
392 }
393 }
394 /* OK, jeb (==c->nextblock) is now pointing at a block which definitely has
395 enough space */
396 *len = jeb->free_size - reserved_size;
397
398 if (c->cleanmarker_size && jeb->used_size == c->cleanmarker_size &&
399 !jeb->first_node->next_in_ino) {
400 /* Only node in it beforehand was a CLEANMARKER node (we think).
401 So mark it obsolete now that there's going to be another node
402 in the block. This will reduce used_size to zero but We've
403 already set c->nextblock so that jffs2_mark_node_obsolete()
404 won't try to refile it to the dirty_list.
405 */
406 spin_unlock(&c->erase_completion_lock);
407 jffs2_mark_node_obsolete(c, jeb->first_node);
408 spin_lock(&c->erase_completion_lock);
409 }
410
411 D1(printk(KERN_DEBUG "jffs2_do_reserve_space(): Giving 0x%x bytes at 0x%x\n",
412 *len, jeb->offset + (c->sector_size - jeb->free_size)));
413 return 0;
414}
415
416/**
417 * jffs2_add_physical_node_ref - add a physical node reference to the list
418 * @c: superblock info
419 * @new: new node reference to add
420 * @len: length of this physical node
421 *
422 * Should only be used to report nodes for which space has been allocated
423 * by jffs2_reserve_space.
424 *
425 * Must be called with the alloc_sem held.
426 */
427
428struct jffs2_raw_node_ref *jffs2_add_physical_node_ref(struct jffs2_sb_info *c,
429 uint32_t ofs, uint32_t len,
430 struct jffs2_inode_cache *ic)
431{
432 struct jffs2_eraseblock *jeb;
433 struct jffs2_raw_node_ref *new;
434
435 jeb = &c->blocks[ofs / c->sector_size];
436
437 D1(printk(KERN_DEBUG "jffs2_add_physical_node_ref(): Node at 0x%x(%d), size 0x%x\n",
438 ofs & ~3, ofs & 3, len));
439#if 1
440 /* Allow non-obsolete nodes only to be added at the end of c->nextblock,
441 if c->nextblock is set. Note that wbuf.c will file obsolete nodes
442 even after refiling c->nextblock */
443 if ((c->nextblock || ((ofs & 3) != REF_OBSOLETE))
444 && (jeb != c->nextblock || (ofs & ~3) != jeb->offset + (c->sector_size - jeb->free_size))) {
445 printk(KERN_WARNING "argh. node added in wrong place at 0x%08x(%d)\n", ofs & ~3, ofs & 3);
446 if (c->nextblock)
447 printk(KERN_WARNING "nextblock 0x%08x", c->nextblock->offset);
448 else
449 printk(KERN_WARNING "No nextblock");
450 printk(", expected at %08x\n", jeb->offset + (c->sector_size - jeb->free_size));
451 return ERR_PTR(-EINVAL);
452 }
453#endif
454 spin_lock(&c->erase_completion_lock);
455
456 new = jffs2_link_node_ref(c, jeb, ofs, len, ic);
457
458 if (!jeb->free_size && !jeb->dirty_size && !ISDIRTY(jeb->wasted_size)) {
459 /* If it lives on the dirty_list, jffs2_reserve_space will put it there */
460 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
461 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
462 if (jffs2_wbuf_dirty(c)) {
463 /* Flush the last write in the block if it's outstanding */
464 spin_unlock(&c->erase_completion_lock);
465 jffs2_flush_wbuf_pad(c);
466 spin_lock(&c->erase_completion_lock);
467 }
468
469 list_add_tail(&jeb->list, &c->clean_list);
470 c->nextblock = NULL;
471 }
472 jffs2_dbg_acct_sanity_check_nolock(c,jeb);
473 jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
474
475 spin_unlock(&c->erase_completion_lock);
476
477 return new;
478}
479
480
481void jffs2_complete_reservation(struct jffs2_sb_info *c)
482{
483 D1(printk(KERN_DEBUG "jffs2_complete_reservation()\n"));
484 spin_lock(&c->erase_completion_lock);
485 jffs2_garbage_collect_trigger(c);
486 spin_unlock(&c->erase_completion_lock);
487 mutex_unlock(&c->alloc_sem);
488}
489
490static inline int on_list(struct list_head *obj, struct list_head *head)
491{
492 struct list_head *this;
493
494 list_for_each(this, head) {
495 if (this == obj) {
496 D1(printk("%p is on list at %p\n", obj, head));
497 return 1;
498
499 }
500 }
501 return 0;
502}
503
504void jffs2_mark_node_obsolete(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref)
505{
506 struct jffs2_eraseblock *jeb;
507 int blocknr;
508 struct jffs2_unknown_node n;
509 int ret, addedsize;
510 size_t retlen;
511 uint32_t freed_len;
512
513 if(unlikely(!ref)) {
514 printk(KERN_NOTICE "EEEEEK. jffs2_mark_node_obsolete called with NULL node\n");
515 return;
516 }
517 if (ref_obsolete(ref)) {
518 D1(printk(KERN_DEBUG "jffs2_mark_node_obsolete called with already obsolete node at 0x%08x\n", ref_offset(ref)));
519 return;
520 }
521 blocknr = ref->flash_offset / c->sector_size;
522 if (blocknr >= c->nr_blocks) {
523 printk(KERN_NOTICE "raw node at 0x%08x is off the end of device!\n", ref->flash_offset);
524 BUG();
525 }
526 jeb = &c->blocks[blocknr];
527
528 if (jffs2_can_mark_obsolete(c) && !jffs2_is_readonly(c) &&
529 !(c->flags & (JFFS2_SB_FLAG_SCANNING | JFFS2_SB_FLAG_BUILDING))) {
530 /* Hm. This may confuse static lock analysis. If any of the above
531 three conditions is false, we're going to return from this
532 function without actually obliterating any nodes or freeing
533 any jffs2_raw_node_refs. So we don't need to stop erases from
534 happening, or protect against people holding an obsolete
535 jffs2_raw_node_ref without the erase_completion_lock. */
536 mutex_lock(&c->erase_free_sem);
537 }
538
539 spin_lock(&c->erase_completion_lock);
540
541 freed_len = ref_totlen(c, jeb, ref);
542
543 if (ref_flags(ref) == REF_UNCHECKED) {
544 D1(if (unlikely(jeb->unchecked_size < freed_len)) {
545 printk(KERN_NOTICE "raw unchecked node of size 0x%08x freed from erase block %d at 0x%08x, but unchecked_size was already 0x%08x\n",
546 freed_len, blocknr, ref->flash_offset, jeb->used_size);
547 BUG();
548 })
549 D1(printk(KERN_DEBUG "Obsoleting previously unchecked node at 0x%08x of len %x: ", ref_offset(ref), freed_len));
550 jeb->unchecked_size -= freed_len;
551 c->unchecked_size -= freed_len;
552 } else {
553 D1(if (unlikely(jeb->used_size < freed_len)) {
554 printk(KERN_NOTICE "raw node of size 0x%08x freed from erase block %d at 0x%08x, but used_size was already 0x%08x\n",
555 freed_len, blocknr, ref->flash_offset, jeb->used_size);
556 BUG();
557 })
558 D1(printk(KERN_DEBUG "Obsoleting node at 0x%08x of len %#x: ", ref_offset(ref), freed_len));
559 jeb->used_size -= freed_len;
560 c->used_size -= freed_len;
561 }
562
563 // Take care, that wasted size is taken into concern
564 if ((jeb->dirty_size || ISDIRTY(jeb->wasted_size + freed_len)) && jeb != c->nextblock) {
565 D1(printk("Dirtying\n"));
566 addedsize = freed_len;
567 jeb->dirty_size += freed_len;
568 c->dirty_size += freed_len;
569
570 /* Convert wasted space to dirty, if not a bad block */
571 if (jeb->wasted_size) {
572 if (on_list(&jeb->list, &c->bad_used_list)) {
573 D1(printk(KERN_DEBUG "Leaving block at %08x on the bad_used_list\n",
574 jeb->offset));
575 addedsize = 0; /* To fool the refiling code later */
576 } else {
577 D1(printk(KERN_DEBUG "Converting %d bytes of wasted space to dirty in block at %08x\n",
578 jeb->wasted_size, jeb->offset));
579 addedsize += jeb->wasted_size;
580 jeb->dirty_size += jeb->wasted_size;
581 c->dirty_size += jeb->wasted_size;
582 c->wasted_size -= jeb->wasted_size;
583 jeb->wasted_size = 0;
584 }
585 }
586 } else {
587 D1(printk("Wasting\n"));
588 addedsize = 0;
589 jeb->wasted_size += freed_len;
590 c->wasted_size += freed_len;
591 }
592 ref->flash_offset = ref_offset(ref) | REF_OBSOLETE;
593
594 jffs2_dbg_acct_sanity_check_nolock(c, jeb);
595 jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
596
597 if (c->flags & JFFS2_SB_FLAG_SCANNING) {
598 /* Flash scanning is in progress. Don't muck about with the block
599 lists because they're not ready yet, and don't actually
600 obliterate nodes that look obsolete. If they weren't
601 marked obsolete on the flash at the time they _became_
602 obsolete, there was probably a reason for that. */
603 spin_unlock(&c->erase_completion_lock);
604 /* We didn't lock the erase_free_sem */
605 return;
606 }
607
608 if (jeb == c->nextblock) {
609 D2(printk(KERN_DEBUG "Not moving nextblock 0x%08x to dirty/erase_pending list\n", jeb->offset));
610 } else if (!jeb->used_size && !jeb->unchecked_size) {
611 if (jeb == c->gcblock) {
612 D1(printk(KERN_DEBUG "gcblock at 0x%08x completely dirtied. Clearing gcblock...\n", jeb->offset));
613 c->gcblock = NULL;
614 } else {
615 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x completely dirtied. Removing from (dirty?) list...\n", jeb->offset));
616 list_del(&jeb->list);
617 }
618 if (jffs2_wbuf_dirty(c)) {
619 D1(printk(KERN_DEBUG "...and adding to erasable_pending_wbuf_list\n"));
620 list_add_tail(&jeb->list, &c->erasable_pending_wbuf_list);
621 } else {
622 if (jiffies & 127) {
623 /* Most of the time, we just erase it immediately. Otherwise we
624 spend ages scanning it on mount, etc. */
625 D1(printk(KERN_DEBUG "...and adding to erase_pending_list\n"));
626 list_add_tail(&jeb->list, &c->erase_pending_list);
627 c->nr_erasing_blocks++;
628 jffs2_garbage_collect_trigger(c);
629 } else {
630 /* Sometimes, however, we leave it elsewhere so it doesn't get
631 immediately reused, and we spread the load a bit. */
632 D1(printk(KERN_DEBUG "...and adding to erasable_list\n"));
633 list_add_tail(&jeb->list, &c->erasable_list);
634 }
635 }
636 D1(printk(KERN_DEBUG "Done OK\n"));
637 } else if (jeb == c->gcblock) {
638 D2(printk(KERN_DEBUG "Not moving gcblock 0x%08x to dirty_list\n", jeb->offset));
639 } else if (ISDIRTY(jeb->dirty_size) && !ISDIRTY(jeb->dirty_size - addedsize)) {
640 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x is freshly dirtied. Removing from clean list...\n", jeb->offset));
641 list_del(&jeb->list);
642 D1(printk(KERN_DEBUG "...and adding to dirty_list\n"));
643 list_add_tail(&jeb->list, &c->dirty_list);
644 } else if (VERYDIRTY(c, jeb->dirty_size) &&
645 !VERYDIRTY(c, jeb->dirty_size - addedsize)) {
646 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x is now very dirty. Removing from dirty list...\n", jeb->offset));
647 list_del(&jeb->list);
648 D1(printk(KERN_DEBUG "...and adding to very_dirty_list\n"));
649 list_add_tail(&jeb->list, &c->very_dirty_list);
650 } else {
651 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x not moved anywhere. (free 0x%08x, dirty 0x%08x, used 0x%08x)\n",
652 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
653 }
654
655 spin_unlock(&c->erase_completion_lock);
656
657 if (!jffs2_can_mark_obsolete(c) || jffs2_is_readonly(c) ||
658 (c->flags & JFFS2_SB_FLAG_BUILDING)) {
659 /* We didn't lock the erase_free_sem */
660 return;
661 }
662
663 /* The erase_free_sem is locked, and has been since before we marked the node obsolete
664 and potentially put its eraseblock onto the erase_pending_list. Thus, we know that
665 the block hasn't _already_ been erased, and that 'ref' itself hasn't been freed yet
666 by jffs2_free_jeb_node_refs() in erase.c. Which is nice. */
667
668 D1(printk(KERN_DEBUG "obliterating obsoleted node at 0x%08x\n", ref_offset(ref)));
669 ret = jffs2_flash_read(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
670 if (ret) {
671 printk(KERN_WARNING "Read error reading from obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret);
672 goto out_erase_sem;
673 }
674 if (retlen != sizeof(n)) {
675 printk(KERN_WARNING "Short read from obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen);
676 goto out_erase_sem;
677 }
678 if (PAD(je32_to_cpu(n.totlen)) != PAD(freed_len)) {
679 printk(KERN_WARNING "Node totlen on flash (0x%08x) != totlen from node ref (0x%08x)\n", je32_to_cpu(n.totlen), freed_len);
680 goto out_erase_sem;
681 }
682 if (!(je16_to_cpu(n.nodetype) & JFFS2_NODE_ACCURATE)) {
683 D1(printk(KERN_DEBUG "Node at 0x%08x was already marked obsolete (nodetype 0x%04x)\n", ref_offset(ref), je16_to_cpu(n.nodetype)));
684 goto out_erase_sem;
685 }
686 /* XXX FIXME: This is ugly now */
687 n.nodetype = cpu_to_je16(je16_to_cpu(n.nodetype) & ~JFFS2_NODE_ACCURATE);
688 ret = jffs2_flash_write(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
689 if (ret) {
690 printk(KERN_WARNING "Write error in obliterating obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret);
691 goto out_erase_sem;
692 }
693 if (retlen != sizeof(n)) {
694 printk(KERN_WARNING "Short write in obliterating obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen);
695 goto out_erase_sem;
696 }
697
698 /* Nodes which have been marked obsolete no longer need to be
699 associated with any inode. Remove them from the per-inode list.
700
701 Note we can't do this for NAND at the moment because we need
702 obsolete dirent nodes to stay on the lists, because of the
703 horridness in jffs2_garbage_collect_deletion_dirent(). Also
704 because we delete the inocache, and on NAND we need that to
705 stay around until all the nodes are actually erased, in order
706 to stop us from giving the same inode number to another newly
707 created inode. */
708 if (ref->next_in_ino) {
709 struct jffs2_inode_cache *ic;
710 struct jffs2_raw_node_ref **p;
711
712 spin_lock(&c->erase_completion_lock);
713
714 ic = jffs2_raw_ref_to_ic(ref);
715 for (p = &ic->nodes; (*p) != ref; p = &((*p)->next_in_ino))
716 ;
717
718 *p = ref->next_in_ino;
719 ref->next_in_ino = NULL;
720
721 switch (ic->class) {
722#ifdef CONFIG_JFFS2_FS_XATTR
723 case RAWNODE_CLASS_XATTR_DATUM:
724 jffs2_release_xattr_datum(c, (struct jffs2_xattr_datum *)ic);
725 break;
726 case RAWNODE_CLASS_XATTR_REF:
727 jffs2_release_xattr_ref(c, (struct jffs2_xattr_ref *)ic);
728 break;
729#endif
730 default:
731 if (ic->nodes == (void *)ic && ic->pino_nlink == 0)
732 jffs2_del_ino_cache(c, ic);
733 break;
734 }
735 spin_unlock(&c->erase_completion_lock);
736 }
737
738 out_erase_sem:
739 mutex_unlock(&c->erase_free_sem);
740}
741
742int jffs2_thread_should_wake(struct jffs2_sb_info *c)
743{
744 int ret = 0;
745 uint32_t dirty;
746 int nr_very_dirty = 0;
747 struct jffs2_eraseblock *jeb;
748
749 if (!list_empty(&c->erase_complete_list) ||
750 !list_empty(&c->erase_pending_list))
751 return 1;
752
753 if (c->unchecked_size) {
754 D1(printk(KERN_DEBUG "jffs2_thread_should_wake(): unchecked_size %d, checked_ino #%d\n",
755 c->unchecked_size, c->checked_ino));
756 return 1;
757 }
758
759 /* dirty_size contains blocks on erase_pending_list
760 * those blocks are counted in c->nr_erasing_blocks.
761 * If one block is actually erased, it is not longer counted as dirty_space
762 * but it is counted in c->nr_erasing_blocks, so we add it and subtract it
763 * with c->nr_erasing_blocks * c->sector_size again.
764 * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks
765 * This helps us to force gc and pick eventually a clean block to spread the load.
766 */
767 dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size;
768
769 if (c->nr_free_blocks + c->nr_erasing_blocks < c->resv_blocks_gctrigger &&
770 (dirty > c->nospc_dirty_size))
771 ret = 1;
772
773 list_for_each_entry(jeb, &c->very_dirty_list, list) {
774 nr_very_dirty++;
775 if (nr_very_dirty == c->vdirty_blocks_gctrigger) {
776 ret = 1;
777 /* In debug mode, actually go through and count them all */
778 D1(continue);
779 break;
780 }
781 }
782
783 D1(printk(KERN_DEBUG "jffs2_thread_should_wake(): nr_free_blocks %d, nr_erasing_blocks %d, dirty_size 0x%x, vdirty_blocks %d: %s\n",
784 c->nr_free_blocks, c->nr_erasing_blocks, c->dirty_size, nr_very_dirty, ret?"yes":"no"));
785
786 return ret;
787}
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#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13
14#include <linux/kernel.h>
15#include <linux/mtd/mtd.h>
16#include <linux/compiler.h>
17#include <linux/sched/signal.h>
18#include <linux/string_choices.h>
19#include "nodelist.h"
20#include "debug.h"
21
22/*
23 * Check whether the user is allowed to write.
24 */
25static int jffs2_rp_can_write(struct jffs2_sb_info *c)
26{
27 uint32_t avail;
28 struct jffs2_mount_opts *opts = &c->mount_opts;
29
30 avail = c->dirty_size + c->free_size + c->unchecked_size +
31 c->erasing_size - c->resv_blocks_write * c->sector_size
32 - c->nospc_dirty_size;
33
34 if (avail < 2 * opts->rp_size)
35 jffs2_dbg(1, "rpsize %u, dirty_size %u, free_size %u, "
36 "erasing_size %u, unchecked_size %u, "
37 "nr_erasing_blocks %u, avail %u, resrv %u\n",
38 opts->rp_size, c->dirty_size, c->free_size,
39 c->erasing_size, c->unchecked_size,
40 c->nr_erasing_blocks, avail, c->nospc_dirty_size);
41
42 if (avail > opts->rp_size)
43 return 1;
44
45 /* Always allow root */
46 if (capable(CAP_SYS_RESOURCE))
47 return 1;
48
49 jffs2_dbg(1, "forbid writing\n");
50 return 0;
51}
52
53static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
54 uint32_t *len, uint32_t sumsize);
55
56/**
57 * jffs2_reserve_space - request physical space to write nodes to flash
58 * @c: superblock info
59 * @minsize: Minimum acceptable size of allocation
60 * @len: Returned value of allocation length
61 * @prio: Allocation type - ALLOC_{NORMAL,DELETION}
62 * @sumsize: summary size requested or JFFS2_SUMMARY_NOSUM_SIZE for no summary
63 *
64 * Requests a block of physical space on the flash.
65 *
66 * Returns: %0 for success and puts 'len' into the appropriate place,
67 * or returns -ENOSPC or other error if appropriate.
68 * Doesn't return len since that's already returned in @len.
69 *
70 * If it returns %0, jffs2_reserve_space() also downs the per-filesystem
71 * allocation semaphore, to prevent more than one allocation from being
72 * active at any time. The semaphore is later released by jffs2_commit_allocation().
73 *
74 * jffs2_reserve_space() may trigger garbage collection in order to make room
75 * for the requested allocation.
76 */
77
78int jffs2_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
79 uint32_t *len, int prio, uint32_t sumsize)
80{
81 int ret = -EAGAIN;
82 int blocksneeded = c->resv_blocks_write;
83 /* align it */
84 minsize = PAD(minsize);
85
86 jffs2_dbg(1, "%s(): Requested 0x%x bytes\n", __func__, minsize);
87 mutex_lock(&c->alloc_sem);
88
89 jffs2_dbg(1, "%s(): alloc sem got\n", __func__);
90
91 spin_lock(&c->erase_completion_lock);
92
93 /*
94 * Check if the free space is greater then size of the reserved pool.
95 * If not, only allow root to proceed with writing.
96 */
97 if (prio != ALLOC_DELETION && !jffs2_rp_can_write(c)) {
98 ret = -ENOSPC;
99 goto out;
100 }
101
102 /* this needs a little more thought (true <tglx> :)) */
103 while(ret == -EAGAIN) {
104 while(c->nr_free_blocks + c->nr_erasing_blocks < blocksneeded) {
105 uint32_t dirty, avail;
106
107 /* calculate real dirty size
108 * dirty_size contains blocks on erase_pending_list
109 * those blocks are counted in c->nr_erasing_blocks.
110 * If one block is actually erased, it is not longer counted as dirty_space
111 * but it is counted in c->nr_erasing_blocks, so we add it and subtract it
112 * with c->nr_erasing_blocks * c->sector_size again.
113 * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks
114 * This helps us to force gc and pick eventually a clean block to spread the load.
115 * We add unchecked_size here, as we hopefully will find some space to use.
116 * This will affect the sum only once, as gc first finishes checking
117 * of nodes.
118 */
119 dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size + c->unchecked_size;
120 if (dirty < c->nospc_dirty_size) {
121 if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) {
122 jffs2_dbg(1, "%s(): Low on dirty space to GC, but it's a deletion. Allowing...\n",
123 __func__);
124 break;
125 }
126 jffs2_dbg(1, "dirty size 0x%08x + unchecked_size 0x%08x < nospc_dirty_size 0x%08x, returning -ENOSPC\n",
127 dirty, c->unchecked_size,
128 c->sector_size);
129
130 spin_unlock(&c->erase_completion_lock);
131 mutex_unlock(&c->alloc_sem);
132 return -ENOSPC;
133 }
134
135 /* Calc possibly available space. Possibly available means that we
136 * don't know, if unchecked size contains obsoleted nodes, which could give us some
137 * more usable space. This will affect the sum only once, as gc first finishes checking
138 * of nodes.
139 + Return -ENOSPC, if the maximum possibly available space is less or equal than
140 * blocksneeded * sector_size.
141 * This blocks endless gc looping on a filesystem, which is nearly full, even if
142 * the check above passes.
143 */
144 avail = c->free_size + c->dirty_size + c->erasing_size + c->unchecked_size;
145 if ( (avail / c->sector_size) <= blocksneeded) {
146 if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) {
147 jffs2_dbg(1, "%s(): Low on possibly available space, but it's a deletion. Allowing...\n",
148 __func__);
149 break;
150 }
151
152 jffs2_dbg(1, "max. available size 0x%08x < blocksneeded * sector_size 0x%08x, returning -ENOSPC\n",
153 avail, blocksneeded * c->sector_size);
154 spin_unlock(&c->erase_completion_lock);
155 mutex_unlock(&c->alloc_sem);
156 return -ENOSPC;
157 }
158
159 mutex_unlock(&c->alloc_sem);
160
161 jffs2_dbg(1, "Triggering GC pass. nr_free_blocks %d, nr_erasing_blocks %d, free_size 0x%08x, dirty_size 0x%08x, wasted_size 0x%08x, used_size 0x%08x, erasing_size 0x%08x, bad_size 0x%08x (total 0x%08x of 0x%08x)\n",
162 c->nr_free_blocks, c->nr_erasing_blocks,
163 c->free_size, c->dirty_size, c->wasted_size,
164 c->used_size, c->erasing_size, c->bad_size,
165 c->free_size + c->dirty_size +
166 c->wasted_size + c->used_size +
167 c->erasing_size + c->bad_size,
168 c->flash_size);
169 spin_unlock(&c->erase_completion_lock);
170
171 ret = jffs2_garbage_collect_pass(c);
172
173 if (ret == -EAGAIN) {
174 spin_lock(&c->erase_completion_lock);
175 if (c->nr_erasing_blocks &&
176 list_empty(&c->erase_pending_list) &&
177 list_empty(&c->erase_complete_list)) {
178 DECLARE_WAITQUEUE(wait, current);
179 set_current_state(TASK_UNINTERRUPTIBLE);
180 add_wait_queue(&c->erase_wait, &wait);
181 jffs2_dbg(1, "%s waiting for erase to complete\n",
182 __func__);
183 spin_unlock(&c->erase_completion_lock);
184
185 schedule();
186 remove_wait_queue(&c->erase_wait, &wait);
187 } else
188 spin_unlock(&c->erase_completion_lock);
189 } else if (ret)
190 return ret;
191
192 cond_resched();
193
194 if (signal_pending(current))
195 return -EINTR;
196
197 mutex_lock(&c->alloc_sem);
198 spin_lock(&c->erase_completion_lock);
199 }
200
201 ret = jffs2_do_reserve_space(c, minsize, len, sumsize);
202 if (ret) {
203 jffs2_dbg(1, "%s(): ret is %d\n", __func__, ret);
204 }
205 }
206
207out:
208 spin_unlock(&c->erase_completion_lock);
209 if (!ret)
210 ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1);
211 if (ret)
212 mutex_unlock(&c->alloc_sem);
213 return ret;
214}
215
216int jffs2_reserve_space_gc(struct jffs2_sb_info *c, uint32_t minsize,
217 uint32_t *len, uint32_t sumsize)
218{
219 int ret;
220 minsize = PAD(minsize);
221
222 jffs2_dbg(1, "%s(): Requested 0x%x bytes\n", __func__, minsize);
223
224 while (true) {
225 spin_lock(&c->erase_completion_lock);
226 ret = jffs2_do_reserve_space(c, minsize, len, sumsize);
227 if (ret) {
228 jffs2_dbg(1, "%s(): looping, ret is %d\n",
229 __func__, ret);
230 }
231 spin_unlock(&c->erase_completion_lock);
232
233 if (ret == -EAGAIN)
234 cond_resched();
235 else
236 break;
237 }
238 if (!ret)
239 ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1);
240
241 return ret;
242}
243
244
245/* Classify nextblock (clean, dirty of verydirty) and force to select an other one */
246
247static void jffs2_close_nextblock(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
248{
249
250 if (c->nextblock == NULL) {
251 jffs2_dbg(1, "%s(): Erase block at 0x%08x has already been placed in a list\n",
252 __func__, jeb->offset);
253 return;
254 }
255 /* Check, if we have a dirty block now, or if it was dirty already */
256 if (ISDIRTY (jeb->wasted_size + jeb->dirty_size)) {
257 c->dirty_size += jeb->wasted_size;
258 c->wasted_size -= jeb->wasted_size;
259 jeb->dirty_size += jeb->wasted_size;
260 jeb->wasted_size = 0;
261 if (VERYDIRTY(c, jeb->dirty_size)) {
262 jffs2_dbg(1, "Adding full erase block at 0x%08x to very_dirty_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
263 jeb->offset, jeb->free_size, jeb->dirty_size,
264 jeb->used_size);
265 list_add_tail(&jeb->list, &c->very_dirty_list);
266 } else {
267 jffs2_dbg(1, "Adding full erase block at 0x%08x to dirty_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
268 jeb->offset, jeb->free_size, jeb->dirty_size,
269 jeb->used_size);
270 list_add_tail(&jeb->list, &c->dirty_list);
271 }
272 } else {
273 jffs2_dbg(1, "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
274 jeb->offset, jeb->free_size, jeb->dirty_size,
275 jeb->used_size);
276 list_add_tail(&jeb->list, &c->clean_list);
277 }
278 c->nextblock = NULL;
279
280}
281
282/* Select a new jeb for nextblock */
283
284static int jffs2_find_nextblock(struct jffs2_sb_info *c)
285{
286 struct list_head *next;
287
288 /* Take the next block off the 'free' list */
289
290 if (list_empty(&c->free_list)) {
291
292 if (!c->nr_erasing_blocks &&
293 !list_empty(&c->erasable_list)) {
294 struct jffs2_eraseblock *ejeb;
295
296 ejeb = list_entry(c->erasable_list.next, struct jffs2_eraseblock, list);
297 list_move_tail(&ejeb->list, &c->erase_pending_list);
298 c->nr_erasing_blocks++;
299 jffs2_garbage_collect_trigger(c);
300 jffs2_dbg(1, "%s(): Triggering erase of erasable block at 0x%08x\n",
301 __func__, ejeb->offset);
302 }
303
304 if (!c->nr_erasing_blocks &&
305 !list_empty(&c->erasable_pending_wbuf_list)) {
306 jffs2_dbg(1, "%s(): Flushing write buffer\n",
307 __func__);
308 /* c->nextblock is NULL, no update to c->nextblock allowed */
309 spin_unlock(&c->erase_completion_lock);
310 jffs2_flush_wbuf_pad(c);
311 spin_lock(&c->erase_completion_lock);
312 /* Have another go. It'll be on the erasable_list now */
313 return -EAGAIN;
314 }
315
316 if (!c->nr_erasing_blocks) {
317 /* Ouch. We're in GC, or we wouldn't have got here.
318 And there's no space left. At all. */
319 pr_crit("Argh. No free space left for GC. nr_erasing_blocks is %d. nr_free_blocks is %d. (erasableempty: %s, erasingempty: %s, erasependingempty: %s)\n",
320 c->nr_erasing_blocks, c->nr_free_blocks,
321 str_yes_no(list_empty(&c->erasable_list)),
322 str_yes_no(list_empty(&c->erasing_list)),
323 str_yes_no(list_empty(&c->erase_pending_list)));
324 return -ENOSPC;
325 }
326
327 spin_unlock(&c->erase_completion_lock);
328 /* Don't wait for it; just erase one right now */
329 jffs2_erase_pending_blocks(c, 1);
330 spin_lock(&c->erase_completion_lock);
331
332 /* An erase may have failed, decreasing the
333 amount of free space available. So we must
334 restart from the beginning */
335 return -EAGAIN;
336 }
337
338 next = c->free_list.next;
339 list_del(next);
340 c->nextblock = list_entry(next, struct jffs2_eraseblock, list);
341 c->nr_free_blocks--;
342
343 jffs2_sum_reset_collected(c->summary); /* reset collected summary */
344
345#ifdef CONFIG_JFFS2_FS_WRITEBUFFER
346 /* adjust write buffer offset, else we get a non contiguous write bug */
347 if (!(c->wbuf_ofs % c->sector_size) && !c->wbuf_len)
348 c->wbuf_ofs = 0xffffffff;
349#endif
350
351 jffs2_dbg(1, "%s(): new nextblock = 0x%08x\n",
352 __func__, c->nextblock->offset);
353
354 return 0;
355}
356
357/* Called with alloc sem _and_ erase_completion_lock */
358static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
359 uint32_t *len, uint32_t sumsize)
360{
361 struct jffs2_eraseblock *jeb = c->nextblock;
362 uint32_t reserved_size; /* for summary information at the end of the jeb */
363 int ret;
364
365 restart:
366 reserved_size = 0;
367
368 if (jffs2_sum_active() && (sumsize != JFFS2_SUMMARY_NOSUM_SIZE)) {
369 /* NOSUM_SIZE means not to generate summary */
370
371 if (jeb) {
372 reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE);
373 dbg_summary("minsize=%d , jeb->free=%d ,"
374 "summary->size=%d , sumsize=%d\n",
375 minsize, jeb->free_size,
376 c->summary->sum_size, sumsize);
377 }
378
379 /* Is there enough space for writing out the current node, or we have to
380 write out summary information now, close this jeb and select new nextblock? */
381 if (jeb && (PAD(minsize) + PAD(c->summary->sum_size + sumsize +
382 JFFS2_SUMMARY_FRAME_SIZE) > jeb->free_size)) {
383
384 /* Has summary been disabled for this jeb? */
385 if (jffs2_sum_is_disabled(c->summary)) {
386 sumsize = JFFS2_SUMMARY_NOSUM_SIZE;
387 goto restart;
388 }
389
390 /* Writing out the collected summary information */
391 dbg_summary("generating summary for 0x%08x.\n", jeb->offset);
392 ret = jffs2_sum_write_sumnode(c);
393
394 if (ret)
395 return ret;
396
397 if (jffs2_sum_is_disabled(c->summary)) {
398 /* jffs2_write_sumnode() couldn't write out the summary information
399 diabling summary for this jeb and free the collected information
400 */
401 sumsize = JFFS2_SUMMARY_NOSUM_SIZE;
402 goto restart;
403 }
404
405 jffs2_close_nextblock(c, jeb);
406 jeb = NULL;
407 /* keep always valid value in reserved_size */
408 reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE);
409 }
410 } else {
411 if (jeb && minsize > jeb->free_size) {
412 uint32_t waste;
413
414 /* Skip the end of this block and file it as having some dirty space */
415 /* If there's a pending write to it, flush now */
416
417 if (jffs2_wbuf_dirty(c)) {
418 spin_unlock(&c->erase_completion_lock);
419 jffs2_dbg(1, "%s(): Flushing write buffer\n",
420 __func__);
421 jffs2_flush_wbuf_pad(c);
422 spin_lock(&c->erase_completion_lock);
423 jeb = c->nextblock;
424 goto restart;
425 }
426
427 spin_unlock(&c->erase_completion_lock);
428
429 ret = jffs2_prealloc_raw_node_refs(c, jeb, 1);
430
431 /* Just lock it again and continue. Nothing much can change because
432 we hold c->alloc_sem anyway. In fact, it's not entirely clear why
433 we hold c->erase_completion_lock in the majority of this function...
434 but that's a question for another (more caffeine-rich) day. */
435 spin_lock(&c->erase_completion_lock);
436
437 if (ret)
438 return ret;
439
440 waste = jeb->free_size;
441 jffs2_link_node_ref(c, jeb,
442 (jeb->offset + c->sector_size - waste) | REF_OBSOLETE,
443 waste, NULL);
444 /* FIXME: that made it count as dirty. Convert to wasted */
445 jeb->dirty_size -= waste;
446 c->dirty_size -= waste;
447 jeb->wasted_size += waste;
448 c->wasted_size += waste;
449
450 jffs2_close_nextblock(c, jeb);
451 jeb = NULL;
452 }
453 }
454
455 if (!jeb) {
456
457 ret = jffs2_find_nextblock(c);
458 if (ret)
459 return ret;
460
461 jeb = c->nextblock;
462
463 if (jeb->free_size != c->sector_size - c->cleanmarker_size) {
464 pr_warn("Eep. Block 0x%08x taken from free_list had free_size of 0x%08x!!\n",
465 jeb->offset, jeb->free_size);
466 goto restart;
467 }
468 }
469 /* OK, jeb (==c->nextblock) is now pointing at a block which definitely has
470 enough space */
471 *len = jeb->free_size - reserved_size;
472
473 if (c->cleanmarker_size && jeb->used_size == c->cleanmarker_size &&
474 !jeb->first_node->next_in_ino) {
475 /* Only node in it beforehand was a CLEANMARKER node (we think).
476 So mark it obsolete now that there's going to be another node
477 in the block. This will reduce used_size to zero but We've
478 already set c->nextblock so that jffs2_mark_node_obsolete()
479 won't try to refile it to the dirty_list.
480 */
481 spin_unlock(&c->erase_completion_lock);
482 jffs2_mark_node_obsolete(c, jeb->first_node);
483 spin_lock(&c->erase_completion_lock);
484 }
485
486 jffs2_dbg(1, "%s(): Giving 0x%x bytes at 0x%x\n",
487 __func__,
488 *len, jeb->offset + (c->sector_size - jeb->free_size));
489 return 0;
490}
491
492/**
493 * jffs2_add_physical_node_ref - add a physical node reference to the list
494 * @c: superblock info
495 * @ofs: offset in the block
496 * @len: length of this physical node
497 * @ic: inode cache pointer
498 *
499 * Should only be used to report nodes for which space has been allocated
500 * by jffs2_reserve_space.
501 *
502 * Must be called with the alloc_sem held.
503 *
504 * Returns: pointer to new node on success or -errno code on error
505 */
506
507struct jffs2_raw_node_ref *jffs2_add_physical_node_ref(struct jffs2_sb_info *c,
508 uint32_t ofs, uint32_t len,
509 struct jffs2_inode_cache *ic)
510{
511 struct jffs2_eraseblock *jeb;
512 struct jffs2_raw_node_ref *new;
513
514 jeb = &c->blocks[ofs / c->sector_size];
515
516 jffs2_dbg(1, "%s(): Node at 0x%x(%d), size 0x%x\n",
517 __func__, ofs & ~3, ofs & 3, len);
518#if 1
519 /* Allow non-obsolete nodes only to be added at the end of c->nextblock,
520 if c->nextblock is set. Note that wbuf.c will file obsolete nodes
521 even after refiling c->nextblock */
522 if ((c->nextblock || ((ofs & 3) != REF_OBSOLETE))
523 && (jeb != c->nextblock || (ofs & ~3) != jeb->offset + (c->sector_size - jeb->free_size))) {
524 pr_warn("argh. node added in wrong place at 0x%08x(%d)\n",
525 ofs & ~3, ofs & 3);
526 if (c->nextblock)
527 pr_warn("nextblock 0x%08x", c->nextblock->offset);
528 else
529 pr_warn("No nextblock");
530 pr_cont(", expected at %08x\n",
531 jeb->offset + (c->sector_size - jeb->free_size));
532 return ERR_PTR(-EINVAL);
533 }
534#endif
535 spin_lock(&c->erase_completion_lock);
536
537 new = jffs2_link_node_ref(c, jeb, ofs, len, ic);
538
539 if (!jeb->free_size && !jeb->dirty_size && !ISDIRTY(jeb->wasted_size)) {
540 /* If it lives on the dirty_list, jffs2_reserve_space will put it there */
541 jffs2_dbg(1, "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
542 jeb->offset, jeb->free_size, jeb->dirty_size,
543 jeb->used_size);
544 if (jffs2_wbuf_dirty(c)) {
545 /* Flush the last write in the block if it's outstanding */
546 spin_unlock(&c->erase_completion_lock);
547 jffs2_flush_wbuf_pad(c);
548 spin_lock(&c->erase_completion_lock);
549 }
550
551 list_add_tail(&jeb->list, &c->clean_list);
552 c->nextblock = NULL;
553 }
554 jffs2_dbg_acct_sanity_check_nolock(c,jeb);
555 jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
556
557 spin_unlock(&c->erase_completion_lock);
558
559 return new;
560}
561
562
563void jffs2_complete_reservation(struct jffs2_sb_info *c)
564{
565 jffs2_dbg(1, "jffs2_complete_reservation()\n");
566 spin_lock(&c->erase_completion_lock);
567 jffs2_garbage_collect_trigger(c);
568 spin_unlock(&c->erase_completion_lock);
569 mutex_unlock(&c->alloc_sem);
570}
571
572static inline int on_list(struct list_head *obj, struct list_head *head)
573{
574 struct list_head *this;
575
576 list_for_each(this, head) {
577 if (this == obj) {
578 jffs2_dbg(1, "%p is on list at %p\n", obj, head);
579 return 1;
580
581 }
582 }
583 return 0;
584}
585
586void jffs2_mark_node_obsolete(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref)
587{
588 struct jffs2_eraseblock *jeb;
589 int blocknr;
590 struct jffs2_unknown_node n;
591 int ret, addedsize;
592 size_t retlen;
593 uint32_t freed_len;
594
595 if(unlikely(!ref)) {
596 pr_notice("EEEEEK. jffs2_mark_node_obsolete called with NULL node\n");
597 return;
598 }
599 if (ref_obsolete(ref)) {
600 jffs2_dbg(1, "%s(): called with already obsolete node at 0x%08x\n",
601 __func__, ref_offset(ref));
602 return;
603 }
604 blocknr = ref->flash_offset / c->sector_size;
605 if (blocknr >= c->nr_blocks) {
606 pr_notice("raw node at 0x%08x is off the end of device!\n",
607 ref->flash_offset);
608 BUG();
609 }
610 jeb = &c->blocks[blocknr];
611
612 if (jffs2_can_mark_obsolete(c) && !jffs2_is_readonly(c) &&
613 !(c->flags & (JFFS2_SB_FLAG_SCANNING | JFFS2_SB_FLAG_BUILDING))) {
614 /* Hm. This may confuse static lock analysis. If any of the above
615 three conditions is false, we're going to return from this
616 function without actually obliterating any nodes or freeing
617 any jffs2_raw_node_refs. So we don't need to stop erases from
618 happening, or protect against people holding an obsolete
619 jffs2_raw_node_ref without the erase_completion_lock. */
620 mutex_lock(&c->erase_free_sem);
621 }
622
623 spin_lock(&c->erase_completion_lock);
624
625 freed_len = ref_totlen(c, jeb, ref);
626
627 if (ref_flags(ref) == REF_UNCHECKED) {
628 D1(if (unlikely(jeb->unchecked_size < freed_len)) {
629 pr_notice("raw unchecked node of size 0x%08x freed from erase block %d at 0x%08x, but unchecked_size was already 0x%08x\n",
630 freed_len, blocknr,
631 ref->flash_offset, jeb->used_size);
632 BUG();
633 })
634 jffs2_dbg(1, "Obsoleting previously unchecked node at 0x%08x of len %x\n",
635 ref_offset(ref), freed_len);
636 jeb->unchecked_size -= freed_len;
637 c->unchecked_size -= freed_len;
638 } else {
639 D1(if (unlikely(jeb->used_size < freed_len)) {
640 pr_notice("raw node of size 0x%08x freed from erase block %d at 0x%08x, but used_size was already 0x%08x\n",
641 freed_len, blocknr,
642 ref->flash_offset, jeb->used_size);
643 BUG();
644 })
645 jffs2_dbg(1, "Obsoleting node at 0x%08x of len %#x: ",
646 ref_offset(ref), freed_len);
647 jeb->used_size -= freed_len;
648 c->used_size -= freed_len;
649 }
650
651 // Take care, that wasted size is taken into concern
652 if ((jeb->dirty_size || ISDIRTY(jeb->wasted_size + freed_len)) && jeb != c->nextblock) {
653 jffs2_dbg(1, "Dirtying\n");
654 addedsize = freed_len;
655 jeb->dirty_size += freed_len;
656 c->dirty_size += freed_len;
657
658 /* Convert wasted space to dirty, if not a bad block */
659 if (jeb->wasted_size) {
660 if (on_list(&jeb->list, &c->bad_used_list)) {
661 jffs2_dbg(1, "Leaving block at %08x on the bad_used_list\n",
662 jeb->offset);
663 addedsize = 0; /* To fool the refiling code later */
664 } else {
665 jffs2_dbg(1, "Converting %d bytes of wasted space to dirty in block at %08x\n",
666 jeb->wasted_size, jeb->offset);
667 addedsize += jeb->wasted_size;
668 jeb->dirty_size += jeb->wasted_size;
669 c->dirty_size += jeb->wasted_size;
670 c->wasted_size -= jeb->wasted_size;
671 jeb->wasted_size = 0;
672 }
673 }
674 } else {
675 jffs2_dbg(1, "Wasting\n");
676 addedsize = 0;
677 jeb->wasted_size += freed_len;
678 c->wasted_size += freed_len;
679 }
680 ref->flash_offset = ref_offset(ref) | REF_OBSOLETE;
681
682 jffs2_dbg_acct_sanity_check_nolock(c, jeb);
683 jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
684
685 if (c->flags & JFFS2_SB_FLAG_SCANNING) {
686 /* Flash scanning is in progress. Don't muck about with the block
687 lists because they're not ready yet, and don't actually
688 obliterate nodes that look obsolete. If they weren't
689 marked obsolete on the flash at the time they _became_
690 obsolete, there was probably a reason for that. */
691 spin_unlock(&c->erase_completion_lock);
692 /* We didn't lock the erase_free_sem */
693 return;
694 }
695
696 if (jeb == c->nextblock) {
697 jffs2_dbg(2, "Not moving nextblock 0x%08x to dirty/erase_pending list\n",
698 jeb->offset);
699 } else if (!jeb->used_size && !jeb->unchecked_size) {
700 if (jeb == c->gcblock) {
701 jffs2_dbg(1, "gcblock at 0x%08x completely dirtied. Clearing gcblock...\n",
702 jeb->offset);
703 c->gcblock = NULL;
704 } else {
705 jffs2_dbg(1, "Eraseblock at 0x%08x completely dirtied. Removing from (dirty?) list...\n",
706 jeb->offset);
707 list_del(&jeb->list);
708 }
709 if (jffs2_wbuf_dirty(c)) {
710 jffs2_dbg(1, "...and adding to erasable_pending_wbuf_list\n");
711 list_add_tail(&jeb->list, &c->erasable_pending_wbuf_list);
712 } else {
713 if (jiffies & 127) {
714 /* Most of the time, we just erase it immediately. Otherwise we
715 spend ages scanning it on mount, etc. */
716 jffs2_dbg(1, "...and adding to erase_pending_list\n");
717 list_add_tail(&jeb->list, &c->erase_pending_list);
718 c->nr_erasing_blocks++;
719 jffs2_garbage_collect_trigger(c);
720 } else {
721 /* Sometimes, however, we leave it elsewhere so it doesn't get
722 immediately reused, and we spread the load a bit. */
723 jffs2_dbg(1, "...and adding to erasable_list\n");
724 list_add_tail(&jeb->list, &c->erasable_list);
725 }
726 }
727 jffs2_dbg(1, "Done OK\n");
728 } else if (jeb == c->gcblock) {
729 jffs2_dbg(2, "Not moving gcblock 0x%08x to dirty_list\n",
730 jeb->offset);
731 } else if (ISDIRTY(jeb->dirty_size) && !ISDIRTY(jeb->dirty_size - addedsize)) {
732 jffs2_dbg(1, "Eraseblock at 0x%08x is freshly dirtied. Removing from clean list...\n",
733 jeb->offset);
734 list_del(&jeb->list);
735 jffs2_dbg(1, "...and adding to dirty_list\n");
736 list_add_tail(&jeb->list, &c->dirty_list);
737 } else if (VERYDIRTY(c, jeb->dirty_size) &&
738 !VERYDIRTY(c, jeb->dirty_size - addedsize)) {
739 jffs2_dbg(1, "Eraseblock at 0x%08x is now very dirty. Removing from dirty list...\n",
740 jeb->offset);
741 list_del(&jeb->list);
742 jffs2_dbg(1, "...and adding to very_dirty_list\n");
743 list_add_tail(&jeb->list, &c->very_dirty_list);
744 } else {
745 jffs2_dbg(1, "Eraseblock at 0x%08x not moved anywhere. (free 0x%08x, dirty 0x%08x, used 0x%08x)\n",
746 jeb->offset, jeb->free_size, jeb->dirty_size,
747 jeb->used_size);
748 }
749
750 spin_unlock(&c->erase_completion_lock);
751
752 if (!jffs2_can_mark_obsolete(c) || jffs2_is_readonly(c) ||
753 (c->flags & JFFS2_SB_FLAG_BUILDING)) {
754 /* We didn't lock the erase_free_sem */
755 return;
756 }
757
758 /* The erase_free_sem is locked, and has been since before we marked the node obsolete
759 and potentially put its eraseblock onto the erase_pending_list. Thus, we know that
760 the block hasn't _already_ been erased, and that 'ref' itself hasn't been freed yet
761 by jffs2_free_jeb_node_refs() in erase.c. Which is nice. */
762
763 jffs2_dbg(1, "obliterating obsoleted node at 0x%08x\n",
764 ref_offset(ref));
765 ret = jffs2_flash_read(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
766 if (ret) {
767 pr_warn("Read error reading from obsoleted node at 0x%08x: %d\n",
768 ref_offset(ref), ret);
769 goto out_erase_sem;
770 }
771 if (retlen != sizeof(n)) {
772 pr_warn("Short read from obsoleted node at 0x%08x: %zd\n",
773 ref_offset(ref), retlen);
774 goto out_erase_sem;
775 }
776 if (PAD(je32_to_cpu(n.totlen)) != PAD(freed_len)) {
777 pr_warn("Node totlen on flash (0x%08x) != totlen from node ref (0x%08x)\n",
778 je32_to_cpu(n.totlen), freed_len);
779 goto out_erase_sem;
780 }
781 if (!(je16_to_cpu(n.nodetype) & JFFS2_NODE_ACCURATE)) {
782 jffs2_dbg(1, "Node at 0x%08x was already marked obsolete (nodetype 0x%04x)\n",
783 ref_offset(ref), je16_to_cpu(n.nodetype));
784 goto out_erase_sem;
785 }
786 /* XXX FIXME: This is ugly now */
787 n.nodetype = cpu_to_je16(je16_to_cpu(n.nodetype) & ~JFFS2_NODE_ACCURATE);
788 ret = jffs2_flash_write(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
789 if (ret) {
790 pr_warn("Write error in obliterating obsoleted node at 0x%08x: %d\n",
791 ref_offset(ref), ret);
792 goto out_erase_sem;
793 }
794 if (retlen != sizeof(n)) {
795 pr_warn("Short write in obliterating obsoleted node at 0x%08x: %zd\n",
796 ref_offset(ref), retlen);
797 goto out_erase_sem;
798 }
799
800 /* Nodes which have been marked obsolete no longer need to be
801 associated with any inode. Remove them from the per-inode list.
802
803 Note we can't do this for NAND at the moment because we need
804 obsolete dirent nodes to stay on the lists, because of the
805 horridness in jffs2_garbage_collect_deletion_dirent(). Also
806 because we delete the inocache, and on NAND we need that to
807 stay around until all the nodes are actually erased, in order
808 to stop us from giving the same inode number to another newly
809 created inode. */
810 if (ref->next_in_ino) {
811 struct jffs2_inode_cache *ic;
812 struct jffs2_raw_node_ref **p;
813
814 spin_lock(&c->erase_completion_lock);
815
816 ic = jffs2_raw_ref_to_ic(ref);
817 for (p = &ic->nodes; (*p) != ref; p = &((*p)->next_in_ino))
818 ;
819
820 *p = ref->next_in_ino;
821 ref->next_in_ino = NULL;
822
823 switch (ic->class) {
824#ifdef CONFIG_JFFS2_FS_XATTR
825 case RAWNODE_CLASS_XATTR_DATUM:
826 jffs2_release_xattr_datum(c, (struct jffs2_xattr_datum *)ic);
827 break;
828 case RAWNODE_CLASS_XATTR_REF:
829 jffs2_release_xattr_ref(c, (struct jffs2_xattr_ref *)ic);
830 break;
831#endif
832 default:
833 if (ic->nodes == (void *)ic && ic->pino_nlink == 0)
834 jffs2_del_ino_cache(c, ic);
835 break;
836 }
837 spin_unlock(&c->erase_completion_lock);
838 }
839
840 out_erase_sem:
841 mutex_unlock(&c->erase_free_sem);
842}
843
844int jffs2_thread_should_wake(struct jffs2_sb_info *c)
845{
846 int ret = 0;
847 uint32_t dirty;
848 int nr_very_dirty = 0;
849 struct jffs2_eraseblock *jeb;
850
851 if (!list_empty(&c->erase_complete_list) ||
852 !list_empty(&c->erase_pending_list))
853 return 1;
854
855 if (c->unchecked_size) {
856 jffs2_dbg(1, "jffs2_thread_should_wake(): unchecked_size %d, check_ino #%d\n",
857 c->unchecked_size, c->check_ino);
858 return 1;
859 }
860
861 /* dirty_size contains blocks on erase_pending_list
862 * those blocks are counted in c->nr_erasing_blocks.
863 * If one block is actually erased, it is not longer counted as dirty_space
864 * but it is counted in c->nr_erasing_blocks, so we add it and subtract it
865 * with c->nr_erasing_blocks * c->sector_size again.
866 * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks
867 * This helps us to force gc and pick eventually a clean block to spread the load.
868 */
869 dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size;
870
871 if (c->nr_free_blocks + c->nr_erasing_blocks < c->resv_blocks_gctrigger &&
872 (dirty > c->nospc_dirty_size))
873 ret = 1;
874
875 list_for_each_entry(jeb, &c->very_dirty_list, list) {
876 nr_very_dirty++;
877 if (nr_very_dirty == c->vdirty_blocks_gctrigger) {
878 ret = 1;
879 /* In debug mode, actually go through and count them all */
880 D1(continue);
881 break;
882 }
883 }
884
885 jffs2_dbg(1, "%s(): nr_free_blocks %d, nr_erasing_blocks %d, dirty_size 0x%x, vdirty_blocks %d: %s\n",
886 __func__, c->nr_free_blocks, c->nr_erasing_blocks,
887 c->dirty_size, nr_very_dirty, str_yes_no(ret));
888
889 return ret;
890}