sched: guest CPU accounting: add guest-CPU /proc/<pid>/stat fields
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / jffs2 / nodemgmt.c
bloba0313fa8748e375d5ee1c2438794e5052faa360f
1 /*
2 * JFFS2 -- Journalling Flash File System, Version 2.
4 * Copyright © 2001-2007 Red Hat, Inc.
6 * Created by David Woodhouse <dwmw2@infradead.org>
8 * For licensing information, see the file 'LICENCE' in this directory.
12 #include <linux/kernel.h>
13 #include <linux/slab.h>
14 #include <linux/mtd/mtd.h>
15 #include <linux/compiler.h>
16 #include <linux/sched.h> /* For cond_resched() */
17 #include "nodelist.h"
18 #include "debug.h"
20 /**
21 * jffs2_reserve_space - request physical space to write nodes to flash
22 * @c: superblock info
23 * @minsize: Minimum acceptable size of allocation
24 * @len: Returned value of allocation length
25 * @prio: Allocation type - ALLOC_{NORMAL,DELETION}
27 * Requests a block of physical space on the flash. Returns zero for success
28 * and puts 'len' into the appropriate place, or returns -ENOSPC or other
29 * error if appropriate. Doesn't return len since that's
31 * If it returns zero, jffs2_reserve_space() also downs the per-filesystem
32 * allocation semaphore, to prevent more than one allocation from being
33 * active at any time. The semaphore is later released by jffs2_commit_allocation()
35 * jffs2_reserve_space() may trigger garbage collection in order to make room
36 * for the requested allocation.
39 static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
40 uint32_t *len, uint32_t sumsize);
42 int jffs2_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
43 uint32_t *len, int prio, uint32_t sumsize)
45 int ret = -EAGAIN;
46 int blocksneeded = c->resv_blocks_write;
47 /* align it */
48 minsize = PAD(minsize);
50 D1(printk(KERN_DEBUG "jffs2_reserve_space(): Requested 0x%x bytes\n", minsize));
51 down(&c->alloc_sem);
53 D1(printk(KERN_DEBUG "jffs2_reserve_space(): alloc sem got\n"));
55 spin_lock(&c->erase_completion_lock);
57 /* this needs a little more thought (true <tglx> :)) */
58 while(ret == -EAGAIN) {
59 while(c->nr_free_blocks + c->nr_erasing_blocks < blocksneeded) {
60 int ret;
61 uint32_t dirty, avail;
63 /* calculate real dirty size
64 * dirty_size contains blocks on erase_pending_list
65 * those blocks are counted in c->nr_erasing_blocks.
66 * If one block is actually erased, it is not longer counted as dirty_space
67 * but it is counted in c->nr_erasing_blocks, so we add it and subtract it
68 * with c->nr_erasing_blocks * c->sector_size again.
69 * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks
70 * This helps us to force gc and pick eventually a clean block to spread the load.
71 * We add unchecked_size here, as we hopefully will find some space to use.
72 * This will affect the sum only once, as gc first finishes checking
73 * of nodes.
75 dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size + c->unchecked_size;
76 if (dirty < c->nospc_dirty_size) {
77 if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) {
78 D1(printk(KERN_NOTICE "jffs2_reserve_space(): Low on dirty space to GC, but it's a deletion. Allowing...\n"));
79 break;
81 D1(printk(KERN_DEBUG "dirty size 0x%08x + unchecked_size 0x%08x < nospc_dirty_size 0x%08x, returning -ENOSPC\n",
82 dirty, c->unchecked_size, c->sector_size));
84 spin_unlock(&c->erase_completion_lock);
85 up(&c->alloc_sem);
86 return -ENOSPC;
89 /* Calc possibly available space. Possibly available means that we
90 * don't know, if unchecked size contains obsoleted nodes, which could give us some
91 * more usable space. This will affect the sum only once, as gc first finishes checking
92 * of nodes.
93 + Return -ENOSPC, if the maximum possibly available space is less or equal than
94 * blocksneeded * sector_size.
95 * This blocks endless gc looping on a filesystem, which is nearly full, even if
96 * the check above passes.
98 avail = c->free_size + c->dirty_size + c->erasing_size + c->unchecked_size;
99 if ( (avail / c->sector_size) <= blocksneeded) {
100 if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) {
101 D1(printk(KERN_NOTICE "jffs2_reserve_space(): Low on possibly available space, but it's a deletion. Allowing...\n"));
102 break;
105 D1(printk(KERN_DEBUG "max. available size 0x%08x < blocksneeded * sector_size 0x%08x, returning -ENOSPC\n",
106 avail, blocksneeded * c->sector_size));
107 spin_unlock(&c->erase_completion_lock);
108 up(&c->alloc_sem);
109 return -ENOSPC;
112 up(&c->alloc_sem);
114 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",
115 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,
116 c->free_size + c->dirty_size + c->wasted_size + c->used_size + c->erasing_size + c->bad_size, c->flash_size));
117 spin_unlock(&c->erase_completion_lock);
119 ret = jffs2_garbage_collect_pass(c);
120 if (ret)
121 return ret;
123 cond_resched();
125 if (signal_pending(current))
126 return -EINTR;
128 down(&c->alloc_sem);
129 spin_lock(&c->erase_completion_lock);
132 ret = jffs2_do_reserve_space(c, minsize, len, sumsize);
133 if (ret) {
134 D1(printk(KERN_DEBUG "jffs2_reserve_space: ret is %d\n", ret));
137 spin_unlock(&c->erase_completion_lock);
138 if (!ret)
139 ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1);
140 if (ret)
141 up(&c->alloc_sem);
142 return ret;
145 int jffs2_reserve_space_gc(struct jffs2_sb_info *c, uint32_t minsize,
146 uint32_t *len, uint32_t sumsize)
148 int ret = -EAGAIN;
149 minsize = PAD(minsize);
151 D1(printk(KERN_DEBUG "jffs2_reserve_space_gc(): Requested 0x%x bytes\n", minsize));
153 spin_lock(&c->erase_completion_lock);
154 while(ret == -EAGAIN) {
155 ret = jffs2_do_reserve_space(c, minsize, len, sumsize);
156 if (ret) {
157 D1(printk(KERN_DEBUG "jffs2_reserve_space_gc: looping, ret is %d\n", ret));
160 spin_unlock(&c->erase_completion_lock);
161 if (!ret)
162 ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1);
164 return ret;
168 /* Classify nextblock (clean, dirty of verydirty) and force to select an other one */
170 static void jffs2_close_nextblock(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
173 if (c->nextblock == NULL) {
174 D1(printk(KERN_DEBUG "jffs2_close_nextblock: Erase block at 0x%08x has already been placed in a list\n",
175 jeb->offset));
176 return;
178 /* Check, if we have a dirty block now, or if it was dirty already */
179 if (ISDIRTY (jeb->wasted_size + jeb->dirty_size)) {
180 c->dirty_size += jeb->wasted_size;
181 c->wasted_size -= jeb->wasted_size;
182 jeb->dirty_size += jeb->wasted_size;
183 jeb->wasted_size = 0;
184 if (VERYDIRTY(c, jeb->dirty_size)) {
185 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",
186 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
187 list_add_tail(&jeb->list, &c->very_dirty_list);
188 } else {
189 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to dirty_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
190 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
191 list_add_tail(&jeb->list, &c->dirty_list);
193 } else {
194 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
195 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
196 list_add_tail(&jeb->list, &c->clean_list);
198 c->nextblock = NULL;
202 /* Select a new jeb for nextblock */
204 static int jffs2_find_nextblock(struct jffs2_sb_info *c)
206 struct list_head *next;
208 /* Take the next block off the 'free' list */
210 if (list_empty(&c->free_list)) {
212 if (!c->nr_erasing_blocks &&
213 !list_empty(&c->erasable_list)) {
214 struct jffs2_eraseblock *ejeb;
216 ejeb = list_entry(c->erasable_list.next, struct jffs2_eraseblock, list);
217 list_move_tail(&ejeb->list, &c->erase_pending_list);
218 c->nr_erasing_blocks++;
219 jffs2_erase_pending_trigger(c);
220 D1(printk(KERN_DEBUG "jffs2_find_nextblock: Triggering erase of erasable block at 0x%08x\n",
221 ejeb->offset));
224 if (!c->nr_erasing_blocks &&
225 !list_empty(&c->erasable_pending_wbuf_list)) {
226 D1(printk(KERN_DEBUG "jffs2_find_nextblock: Flushing write buffer\n"));
227 /* c->nextblock is NULL, no update to c->nextblock allowed */
228 spin_unlock(&c->erase_completion_lock);
229 jffs2_flush_wbuf_pad(c);
230 spin_lock(&c->erase_completion_lock);
231 /* Have another go. It'll be on the erasable_list now */
232 return -EAGAIN;
235 if (!c->nr_erasing_blocks) {
236 /* Ouch. We're in GC, or we wouldn't have got here.
237 And there's no space left. At all. */
238 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",
239 c->nr_erasing_blocks, c->nr_free_blocks, list_empty(&c->erasable_list)?"yes":"no",
240 list_empty(&c->erasing_list)?"yes":"no", list_empty(&c->erase_pending_list)?"yes":"no");
241 return -ENOSPC;
244 spin_unlock(&c->erase_completion_lock);
245 /* Don't wait for it; just erase one right now */
246 jffs2_erase_pending_blocks(c, 1);
247 spin_lock(&c->erase_completion_lock);
249 /* An erase may have failed, decreasing the
250 amount of free space available. So we must
251 restart from the beginning */
252 return -EAGAIN;
255 next = c->free_list.next;
256 list_del(next);
257 c->nextblock = list_entry(next, struct jffs2_eraseblock, list);
258 c->nr_free_blocks--;
260 jffs2_sum_reset_collected(c->summary); /* reset collected summary */
262 D1(printk(KERN_DEBUG "jffs2_find_nextblock(): new nextblock = 0x%08x\n", c->nextblock->offset));
264 return 0;
267 /* Called with alloc sem _and_ erase_completion_lock */
268 static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
269 uint32_t *len, uint32_t sumsize)
271 struct jffs2_eraseblock *jeb = c->nextblock;
272 uint32_t reserved_size; /* for summary information at the end of the jeb */
273 int ret;
275 restart:
276 reserved_size = 0;
278 if (jffs2_sum_active() && (sumsize != JFFS2_SUMMARY_NOSUM_SIZE)) {
279 /* NOSUM_SIZE means not to generate summary */
281 if (jeb) {
282 reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE);
283 dbg_summary("minsize=%d , jeb->free=%d ,"
284 "summary->size=%d , sumsize=%d\n",
285 minsize, jeb->free_size,
286 c->summary->sum_size, sumsize);
289 /* Is there enough space for writing out the current node, or we have to
290 write out summary information now, close this jeb and select new nextblock? */
291 if (jeb && (PAD(minsize) + PAD(c->summary->sum_size + sumsize +
292 JFFS2_SUMMARY_FRAME_SIZE) > jeb->free_size)) {
294 /* Has summary been disabled for this jeb? */
295 if (jffs2_sum_is_disabled(c->summary)) {
296 sumsize = JFFS2_SUMMARY_NOSUM_SIZE;
297 goto restart;
300 /* Writing out the collected summary information */
301 dbg_summary("generating summary for 0x%08x.\n", jeb->offset);
302 ret = jffs2_sum_write_sumnode(c);
304 if (ret)
305 return ret;
307 if (jffs2_sum_is_disabled(c->summary)) {
308 /* jffs2_write_sumnode() couldn't write out the summary information
309 diabling summary for this jeb and free the collected information
311 sumsize = JFFS2_SUMMARY_NOSUM_SIZE;
312 goto restart;
315 jffs2_close_nextblock(c, jeb);
316 jeb = NULL;
317 /* keep always valid value in reserved_size */
318 reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE);
320 } else {
321 if (jeb && minsize > jeb->free_size) {
322 uint32_t waste;
324 /* Skip the end of this block and file it as having some dirty space */
325 /* If there's a pending write to it, flush now */
327 if (jffs2_wbuf_dirty(c)) {
328 spin_unlock(&c->erase_completion_lock);
329 D1(printk(KERN_DEBUG "jffs2_do_reserve_space: Flushing write buffer\n"));
330 jffs2_flush_wbuf_pad(c);
331 spin_lock(&c->erase_completion_lock);
332 jeb = c->nextblock;
333 goto restart;
336 spin_unlock(&c->erase_completion_lock);
338 ret = jffs2_prealloc_raw_node_refs(c, jeb, 1);
339 if (ret)
340 return ret;
341 /* Just lock it again and continue. Nothing much can change because
342 we hold c->alloc_sem anyway. In fact, it's not entirely clear why
343 we hold c->erase_completion_lock in the majority of this function...
344 but that's a question for another (more caffeine-rich) day. */
345 spin_lock(&c->erase_completion_lock);
347 waste = jeb->free_size;
348 jffs2_link_node_ref(c, jeb,
349 (jeb->offset + c->sector_size - waste) | REF_OBSOLETE,
350 waste, NULL);
351 /* FIXME: that made it count as dirty. Convert to wasted */
352 jeb->dirty_size -= waste;
353 c->dirty_size -= waste;
354 jeb->wasted_size += waste;
355 c->wasted_size += waste;
357 jffs2_close_nextblock(c, jeb);
358 jeb = NULL;
362 if (!jeb) {
364 ret = jffs2_find_nextblock(c);
365 if (ret)
366 return ret;
368 jeb = c->nextblock;
370 if (jeb->free_size != c->sector_size - c->cleanmarker_size) {
371 printk(KERN_WARNING "Eep. Block 0x%08x taken from free_list had free_size of 0x%08x!!\n", jeb->offset, jeb->free_size);
372 goto restart;
375 /* OK, jeb (==c->nextblock) is now pointing at a block which definitely has
376 enough space */
377 *len = jeb->free_size - reserved_size;
379 if (c->cleanmarker_size && jeb->used_size == c->cleanmarker_size &&
380 !jeb->first_node->next_in_ino) {
381 /* Only node in it beforehand was a CLEANMARKER node (we think).
382 So mark it obsolete now that there's going to be another node
383 in the block. This will reduce used_size to zero but We've
384 already set c->nextblock so that jffs2_mark_node_obsolete()
385 won't try to refile it to the dirty_list.
387 spin_unlock(&c->erase_completion_lock);
388 jffs2_mark_node_obsolete(c, jeb->first_node);
389 spin_lock(&c->erase_completion_lock);
392 D1(printk(KERN_DEBUG "jffs2_do_reserve_space(): Giving 0x%x bytes at 0x%x\n",
393 *len, jeb->offset + (c->sector_size - jeb->free_size)));
394 return 0;
398 * jffs2_add_physical_node_ref - add a physical node reference to the list
399 * @c: superblock info
400 * @new: new node reference to add
401 * @len: length of this physical node
403 * Should only be used to report nodes for which space has been allocated
404 * by jffs2_reserve_space.
406 * Must be called with the alloc_sem held.
409 struct jffs2_raw_node_ref *jffs2_add_physical_node_ref(struct jffs2_sb_info *c,
410 uint32_t ofs, uint32_t len,
411 struct jffs2_inode_cache *ic)
413 struct jffs2_eraseblock *jeb;
414 struct jffs2_raw_node_ref *new;
416 jeb = &c->blocks[ofs / c->sector_size];
418 D1(printk(KERN_DEBUG "jffs2_add_physical_node_ref(): Node at 0x%x(%d), size 0x%x\n",
419 ofs & ~3, ofs & 3, len));
420 #if 1
421 /* Allow non-obsolete nodes only to be added at the end of c->nextblock,
422 if c->nextblock is set. Note that wbuf.c will file obsolete nodes
423 even after refiling c->nextblock */
424 if ((c->nextblock || ((ofs & 3) != REF_OBSOLETE))
425 && (jeb != c->nextblock || (ofs & ~3) != jeb->offset + (c->sector_size - jeb->free_size))) {
426 printk(KERN_WARNING "argh. node added in wrong place at 0x%08x(%d)\n", ofs & ~3, ofs & 3);
427 if (c->nextblock)
428 printk(KERN_WARNING "nextblock 0x%08x", c->nextblock->offset);
429 else
430 printk(KERN_WARNING "No nextblock");
431 printk(", expected at %08x\n", jeb->offset + (c->sector_size - jeb->free_size));
432 return ERR_PTR(-EINVAL);
434 #endif
435 spin_lock(&c->erase_completion_lock);
437 new = jffs2_link_node_ref(c, jeb, ofs, len, ic);
439 if (!jeb->free_size && !jeb->dirty_size && !ISDIRTY(jeb->wasted_size)) {
440 /* If it lives on the dirty_list, jffs2_reserve_space will put it there */
441 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
442 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
443 if (jffs2_wbuf_dirty(c)) {
444 /* Flush the last write in the block if it's outstanding */
445 spin_unlock(&c->erase_completion_lock);
446 jffs2_flush_wbuf_pad(c);
447 spin_lock(&c->erase_completion_lock);
450 list_add_tail(&jeb->list, &c->clean_list);
451 c->nextblock = NULL;
453 jffs2_dbg_acct_sanity_check_nolock(c,jeb);
454 jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
456 spin_unlock(&c->erase_completion_lock);
458 return new;
462 void jffs2_complete_reservation(struct jffs2_sb_info *c)
464 D1(printk(KERN_DEBUG "jffs2_complete_reservation()\n"));
465 jffs2_garbage_collect_trigger(c);
466 up(&c->alloc_sem);
469 static inline int on_list(struct list_head *obj, struct list_head *head)
471 struct list_head *this;
473 list_for_each(this, head) {
474 if (this == obj) {
475 D1(printk("%p is on list at %p\n", obj, head));
476 return 1;
480 return 0;
483 void jffs2_mark_node_obsolete(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref)
485 struct jffs2_eraseblock *jeb;
486 int blocknr;
487 struct jffs2_unknown_node n;
488 int ret, addedsize;
489 size_t retlen;
490 uint32_t freed_len;
492 if(unlikely(!ref)) {
493 printk(KERN_NOTICE "EEEEEK. jffs2_mark_node_obsolete called with NULL node\n");
494 return;
496 if (ref_obsolete(ref)) {
497 D1(printk(KERN_DEBUG "jffs2_mark_node_obsolete called with already obsolete node at 0x%08x\n", ref_offset(ref)));
498 return;
500 blocknr = ref->flash_offset / c->sector_size;
501 if (blocknr >= c->nr_blocks) {
502 printk(KERN_NOTICE "raw node at 0x%08x is off the end of device!\n", ref->flash_offset);
503 BUG();
505 jeb = &c->blocks[blocknr];
507 if (jffs2_can_mark_obsolete(c) && !jffs2_is_readonly(c) &&
508 !(c->flags & (JFFS2_SB_FLAG_SCANNING | JFFS2_SB_FLAG_BUILDING))) {
509 /* Hm. This may confuse static lock analysis. If any of the above
510 three conditions is false, we're going to return from this
511 function without actually obliterating any nodes or freeing
512 any jffs2_raw_node_refs. So we don't need to stop erases from
513 happening, or protect against people holding an obsolete
514 jffs2_raw_node_ref without the erase_completion_lock. */
515 down(&c->erase_free_sem);
518 spin_lock(&c->erase_completion_lock);
520 freed_len = ref_totlen(c, jeb, ref);
522 if (ref_flags(ref) == REF_UNCHECKED) {
523 D1(if (unlikely(jeb->unchecked_size < freed_len)) {
524 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",
525 freed_len, blocknr, ref->flash_offset, jeb->used_size);
526 BUG();
528 D1(printk(KERN_DEBUG "Obsoleting previously unchecked node at 0x%08x of len %x: ", ref_offset(ref), freed_len));
529 jeb->unchecked_size -= freed_len;
530 c->unchecked_size -= freed_len;
531 } else {
532 D1(if (unlikely(jeb->used_size < freed_len)) {
533 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",
534 freed_len, blocknr, ref->flash_offset, jeb->used_size);
535 BUG();
537 D1(printk(KERN_DEBUG "Obsoleting node at 0x%08x of len %#x: ", ref_offset(ref), freed_len));
538 jeb->used_size -= freed_len;
539 c->used_size -= freed_len;
542 // Take care, that wasted size is taken into concern
543 if ((jeb->dirty_size || ISDIRTY(jeb->wasted_size + freed_len)) && jeb != c->nextblock) {
544 D1(printk("Dirtying\n"));
545 addedsize = freed_len;
546 jeb->dirty_size += freed_len;
547 c->dirty_size += freed_len;
549 /* Convert wasted space to dirty, if not a bad block */
550 if (jeb->wasted_size) {
551 if (on_list(&jeb->list, &c->bad_used_list)) {
552 D1(printk(KERN_DEBUG "Leaving block at %08x on the bad_used_list\n",
553 jeb->offset));
554 addedsize = 0; /* To fool the refiling code later */
555 } else {
556 D1(printk(KERN_DEBUG "Converting %d bytes of wasted space to dirty in block at %08x\n",
557 jeb->wasted_size, jeb->offset));
558 addedsize += jeb->wasted_size;
559 jeb->dirty_size += jeb->wasted_size;
560 c->dirty_size += jeb->wasted_size;
561 c->wasted_size -= jeb->wasted_size;
562 jeb->wasted_size = 0;
565 } else {
566 D1(printk("Wasting\n"));
567 addedsize = 0;
568 jeb->wasted_size += freed_len;
569 c->wasted_size += freed_len;
571 ref->flash_offset = ref_offset(ref) | REF_OBSOLETE;
573 jffs2_dbg_acct_sanity_check_nolock(c, jeb);
574 jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
576 if (c->flags & JFFS2_SB_FLAG_SCANNING) {
577 /* Flash scanning is in progress. Don't muck about with the block
578 lists because they're not ready yet, and don't actually
579 obliterate nodes that look obsolete. If they weren't
580 marked obsolete on the flash at the time they _became_
581 obsolete, there was probably a reason for that. */
582 spin_unlock(&c->erase_completion_lock);
583 /* We didn't lock the erase_free_sem */
584 return;
587 if (jeb == c->nextblock) {
588 D2(printk(KERN_DEBUG "Not moving nextblock 0x%08x to dirty/erase_pending list\n", jeb->offset));
589 } else if (!jeb->used_size && !jeb->unchecked_size) {
590 if (jeb == c->gcblock) {
591 D1(printk(KERN_DEBUG "gcblock at 0x%08x completely dirtied. Clearing gcblock...\n", jeb->offset));
592 c->gcblock = NULL;
593 } else {
594 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x completely dirtied. Removing from (dirty?) list...\n", jeb->offset));
595 list_del(&jeb->list);
597 if (jffs2_wbuf_dirty(c)) {
598 D1(printk(KERN_DEBUG "...and adding to erasable_pending_wbuf_list\n"));
599 list_add_tail(&jeb->list, &c->erasable_pending_wbuf_list);
600 } else {
601 if (jiffies & 127) {
602 /* Most of the time, we just erase it immediately. Otherwise we
603 spend ages scanning it on mount, etc. */
604 D1(printk(KERN_DEBUG "...and adding to erase_pending_list\n"));
605 list_add_tail(&jeb->list, &c->erase_pending_list);
606 c->nr_erasing_blocks++;
607 jffs2_erase_pending_trigger(c);
608 } else {
609 /* Sometimes, however, we leave it elsewhere so it doesn't get
610 immediately reused, and we spread the load a bit. */
611 D1(printk(KERN_DEBUG "...and adding to erasable_list\n"));
612 list_add_tail(&jeb->list, &c->erasable_list);
615 D1(printk(KERN_DEBUG "Done OK\n"));
616 } else if (jeb == c->gcblock) {
617 D2(printk(KERN_DEBUG "Not moving gcblock 0x%08x to dirty_list\n", jeb->offset));
618 } else if (ISDIRTY(jeb->dirty_size) && !ISDIRTY(jeb->dirty_size - addedsize)) {
619 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x is freshly dirtied. Removing from clean list...\n", jeb->offset));
620 list_del(&jeb->list);
621 D1(printk(KERN_DEBUG "...and adding to dirty_list\n"));
622 list_add_tail(&jeb->list, &c->dirty_list);
623 } else if (VERYDIRTY(c, jeb->dirty_size) &&
624 !VERYDIRTY(c, jeb->dirty_size - addedsize)) {
625 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x is now very dirty. Removing from dirty list...\n", jeb->offset));
626 list_del(&jeb->list);
627 D1(printk(KERN_DEBUG "...and adding to very_dirty_list\n"));
628 list_add_tail(&jeb->list, &c->very_dirty_list);
629 } else {
630 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x not moved anywhere. (free 0x%08x, dirty 0x%08x, used 0x%08x)\n",
631 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
634 spin_unlock(&c->erase_completion_lock);
636 if (!jffs2_can_mark_obsolete(c) || jffs2_is_readonly(c) ||
637 (c->flags & JFFS2_SB_FLAG_BUILDING)) {
638 /* We didn't lock the erase_free_sem */
639 return;
642 /* The erase_free_sem is locked, and has been since before we marked the node obsolete
643 and potentially put its eraseblock onto the erase_pending_list. Thus, we know that
644 the block hasn't _already_ been erased, and that 'ref' itself hasn't been freed yet
645 by jffs2_free_jeb_node_refs() in erase.c. Which is nice. */
647 D1(printk(KERN_DEBUG "obliterating obsoleted node at 0x%08x\n", ref_offset(ref)));
648 ret = jffs2_flash_read(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
649 if (ret) {
650 printk(KERN_WARNING "Read error reading from obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret);
651 goto out_erase_sem;
653 if (retlen != sizeof(n)) {
654 printk(KERN_WARNING "Short read from obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen);
655 goto out_erase_sem;
657 if (PAD(je32_to_cpu(n.totlen)) != PAD(freed_len)) {
658 printk(KERN_WARNING "Node totlen on flash (0x%08x) != totlen from node ref (0x%08x)\n", je32_to_cpu(n.totlen), freed_len);
659 goto out_erase_sem;
661 if (!(je16_to_cpu(n.nodetype) & JFFS2_NODE_ACCURATE)) {
662 D1(printk(KERN_DEBUG "Node at 0x%08x was already marked obsolete (nodetype 0x%04x)\n", ref_offset(ref), je16_to_cpu(n.nodetype)));
663 goto out_erase_sem;
665 /* XXX FIXME: This is ugly now */
666 n.nodetype = cpu_to_je16(je16_to_cpu(n.nodetype) & ~JFFS2_NODE_ACCURATE);
667 ret = jffs2_flash_write(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
668 if (ret) {
669 printk(KERN_WARNING "Write error in obliterating obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret);
670 goto out_erase_sem;
672 if (retlen != sizeof(n)) {
673 printk(KERN_WARNING "Short write in obliterating obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen);
674 goto out_erase_sem;
677 /* Nodes which have been marked obsolete no longer need to be
678 associated with any inode. Remove them from the per-inode list.
680 Note we can't do this for NAND at the moment because we need
681 obsolete dirent nodes to stay on the lists, because of the
682 horridness in jffs2_garbage_collect_deletion_dirent(). Also
683 because we delete the inocache, and on NAND we need that to
684 stay around until all the nodes are actually erased, in order
685 to stop us from giving the same inode number to another newly
686 created inode. */
687 if (ref->next_in_ino) {
688 struct jffs2_inode_cache *ic;
689 struct jffs2_raw_node_ref **p;
691 spin_lock(&c->erase_completion_lock);
693 ic = jffs2_raw_ref_to_ic(ref);
694 for (p = &ic->nodes; (*p) != ref; p = &((*p)->next_in_ino))
697 *p = ref->next_in_ino;
698 ref->next_in_ino = NULL;
700 switch (ic->class) {
701 #ifdef CONFIG_JFFS2_FS_XATTR
702 case RAWNODE_CLASS_XATTR_DATUM:
703 jffs2_release_xattr_datum(c, (struct jffs2_xattr_datum *)ic);
704 break;
705 case RAWNODE_CLASS_XATTR_REF:
706 jffs2_release_xattr_ref(c, (struct jffs2_xattr_ref *)ic);
707 break;
708 #endif
709 default:
710 if (ic->nodes == (void *)ic && ic->nlink == 0)
711 jffs2_del_ino_cache(c, ic);
712 break;
714 spin_unlock(&c->erase_completion_lock);
717 out_erase_sem:
718 up(&c->erase_free_sem);
721 int jffs2_thread_should_wake(struct jffs2_sb_info *c)
723 int ret = 0;
724 uint32_t dirty;
725 int nr_very_dirty = 0;
726 struct jffs2_eraseblock *jeb;
728 if (c->unchecked_size) {
729 D1(printk(KERN_DEBUG "jffs2_thread_should_wake(): unchecked_size %d, checked_ino #%d\n",
730 c->unchecked_size, c->checked_ino));
731 return 1;
734 /* dirty_size contains blocks on erase_pending_list
735 * those blocks are counted in c->nr_erasing_blocks.
736 * If one block is actually erased, it is not longer counted as dirty_space
737 * but it is counted in c->nr_erasing_blocks, so we add it and subtract it
738 * with c->nr_erasing_blocks * c->sector_size again.
739 * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks
740 * This helps us to force gc and pick eventually a clean block to spread the load.
742 dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size;
744 if (c->nr_free_blocks + c->nr_erasing_blocks < c->resv_blocks_gctrigger &&
745 (dirty > c->nospc_dirty_size))
746 ret = 1;
748 list_for_each_entry(jeb, &c->very_dirty_list, list) {
749 nr_very_dirty++;
750 if (nr_very_dirty == c->vdirty_blocks_gctrigger) {
751 ret = 1;
752 /* In debug mode, actually go through and count them all */
753 D1(continue);
754 break;
758 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",
759 c->nr_free_blocks, c->nr_erasing_blocks, c->dirty_size, nr_very_dirty, ret?"yes":"no"));
761 return ret;