xref: /linux/fs/jffs2/nodemgmt.c (revision 8da8ba2e)
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.h> /* For cond_resched() */
18 #include "nodelist.h"
19 #include "debug.h"
20 
21 /*
22  * Check whether the user is allowed to write.
23  */
24 static int jffs2_rp_can_write(struct jffs2_sb_info *c)
25 {
26 	uint32_t avail;
27 	struct jffs2_mount_opts *opts = &c->mount_opts;
28 
29 	avail = c->dirty_size + c->free_size + c->unchecked_size +
30 		c->erasing_size - c->resv_blocks_write * c->sector_size
31 		- c->nospc_dirty_size;
32 
33 	if (avail < 2 * opts->rp_size)
34 		jffs2_dbg(1, "rpsize %u, dirty_size %u, free_size %u, "
35 			  "erasing_size %u, unchecked_size %u, "
36 			  "nr_erasing_blocks %u, avail %u, resrv %u\n",
37 			  opts->rp_size, c->dirty_size, c->free_size,
38 			  c->erasing_size, c->unchecked_size,
39 			  c->nr_erasing_blocks, avail, c->nospc_dirty_size);
40 
41 	if (avail > opts->rp_size)
42 		return 1;
43 
44 	/* Always allow root */
45 	if (capable(CAP_SYS_RESOURCE))
46 		return 1;
47 
48 	jffs2_dbg(1, "forbid writing\n");
49 	return 0;
50 }
51 
52 /**
53  *	jffs2_reserve_space - request physical space to write nodes to flash
54  *	@c: superblock info
55  *	@minsize: Minimum acceptable size of allocation
56  *	@len: Returned value of allocation length
57  *	@prio: Allocation type - ALLOC_{NORMAL,DELETION}
58  *
59  *	Requests a block of physical space on the flash. Returns zero for success
60  *	and puts 'len' into the appropriate place, or returns -ENOSPC or other
61  *	error if appropriate. Doesn't return len since that's
62  *
63  *	If it returns zero, jffs2_reserve_space() also downs the per-filesystem
64  *	allocation semaphore, to prevent more than one allocation from being
65  *	active at any time. The semaphore is later released by jffs2_commit_allocation()
66  *
67  *	jffs2_reserve_space() may trigger garbage collection in order to make room
68  *	for the requested allocation.
69  */
70 
71 static int jffs2_do_reserve_space(struct jffs2_sb_info *c,  uint32_t minsize,
72 				  uint32_t *len, uint32_t sumsize);
73 
74 int jffs2_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
75 			uint32_t *len, int prio, uint32_t sumsize)
76 {
77 	int ret = -EAGAIN;
78 	int blocksneeded = c->resv_blocks_write;
79 	/* align it */
80 	minsize = PAD(minsize);
81 
82 	jffs2_dbg(1, "%s(): Requested 0x%x bytes\n", __func__, minsize);
83 	mutex_lock(&c->alloc_sem);
84 
85 	jffs2_dbg(1, "%s(): alloc sem got\n", __func__);
86 
87 	spin_lock(&c->erase_completion_lock);
88 
89 	/*
90 	 * Check if the free space is greater then size of the reserved pool.
91 	 * If not, only allow root to proceed with writing.
92 	 */
93 	if (prio != ALLOC_DELETION && !jffs2_rp_can_write(c)) {
94 		ret = -ENOSPC;
95 		goto out;
96 	}
97 
98 	/* this needs a little more thought (true <tglx> :)) */
99 	while(ret == -EAGAIN) {
100 		while(c->nr_free_blocks + c->nr_erasing_blocks < blocksneeded) {
101 			uint32_t dirty, avail;
102 
103 			/* calculate real dirty size
104 			 * dirty_size contains blocks on erase_pending_list
105 			 * those blocks are counted in c->nr_erasing_blocks.
106 			 * If one block is actually erased, it is not longer counted as dirty_space
107 			 * but it is counted in c->nr_erasing_blocks, so we add it and subtract it
108 			 * with c->nr_erasing_blocks * c->sector_size again.
109 			 * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks
110 			 * This helps us to force gc and pick eventually a clean block to spread the load.
111 			 * We add unchecked_size here, as we hopefully will find some space to use.
112 			 * This will affect the sum only once, as gc first finishes checking
113 			 * of nodes.
114 			 */
115 			dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size + c->unchecked_size;
116 			if (dirty < c->nospc_dirty_size) {
117 				if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) {
118 					jffs2_dbg(1, "%s(): Low on dirty space to GC, but it's a deletion. Allowing...\n",
119 						  __func__);
120 					break;
121 				}
122 				jffs2_dbg(1, "dirty size 0x%08x + unchecked_size 0x%08x < nospc_dirty_size 0x%08x, returning -ENOSPC\n",
123 					  dirty, c->unchecked_size,
124 					  c->sector_size);
125 
126 				spin_unlock(&c->erase_completion_lock);
127 				mutex_unlock(&c->alloc_sem);
128 				return -ENOSPC;
129 			}
130 
131 			/* Calc possibly available space. Possibly available means that we
132 			 * don't know, if unchecked size contains obsoleted nodes, which could give us some
133 			 * more usable space. This will affect the sum only once, as gc first finishes checking
134 			 * of nodes.
135 			 + Return -ENOSPC, if the maximum possibly available space is less or equal than
136 			 * blocksneeded * sector_size.
137 			 * This blocks endless gc looping on a filesystem, which is nearly full, even if
138 			 * the check above passes.
139 			 */
140 			avail = c->free_size + c->dirty_size + c->erasing_size + c->unchecked_size;
141 			if ( (avail / c->sector_size) <= blocksneeded) {
142 				if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) {
143 					jffs2_dbg(1, "%s(): Low on possibly available space, but it's a deletion. Allowing...\n",
144 						  __func__);
145 					break;
146 				}
147 
148 				jffs2_dbg(1, "max. available size 0x%08x  < blocksneeded * sector_size 0x%08x, returning -ENOSPC\n",
149 					  avail, blocksneeded * c->sector_size);
150 				spin_unlock(&c->erase_completion_lock);
151 				mutex_unlock(&c->alloc_sem);
152 				return -ENOSPC;
153 			}
154 
155 			mutex_unlock(&c->alloc_sem);
156 
157 			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",
158 				  c->nr_free_blocks, c->nr_erasing_blocks,
159 				  c->free_size, c->dirty_size, c->wasted_size,
160 				  c->used_size, c->erasing_size, c->bad_size,
161 				  c->free_size + c->dirty_size +
162 				  c->wasted_size + c->used_size +
163 				  c->erasing_size + c->bad_size,
164 				  c->flash_size);
165 			spin_unlock(&c->erase_completion_lock);
166 
167 			ret = jffs2_garbage_collect_pass(c);
168 
169 			if (ret == -EAGAIN) {
170 				spin_lock(&c->erase_completion_lock);
171 				if (c->nr_erasing_blocks &&
172 				    list_empty(&c->erase_pending_list) &&
173 				    list_empty(&c->erase_complete_list)) {
174 					DECLARE_WAITQUEUE(wait, current);
175 					set_current_state(TASK_UNINTERRUPTIBLE);
176 					add_wait_queue(&c->erase_wait, &wait);
177 					jffs2_dbg(1, "%s waiting for erase to complete\n",
178 						  __func__);
179 					spin_unlock(&c->erase_completion_lock);
180 
181 					schedule();
182 				} else
183 					spin_unlock(&c->erase_completion_lock);
184 			} else if (ret)
185 				return ret;
186 
187 			cond_resched();
188 
189 			if (signal_pending(current))
190 				return -EINTR;
191 
192 			mutex_lock(&c->alloc_sem);
193 			spin_lock(&c->erase_completion_lock);
194 		}
195 
196 		ret = jffs2_do_reserve_space(c, minsize, len, sumsize);
197 		if (ret) {
198 			jffs2_dbg(1, "%s(): ret is %d\n", __func__, ret);
199 		}
200 	}
201 
202 out:
203 	spin_unlock(&c->erase_completion_lock);
204 	if (!ret)
205 		ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1);
206 	if (ret)
207 		mutex_unlock(&c->alloc_sem);
208 	return ret;
209 }
210 
211 int jffs2_reserve_space_gc(struct jffs2_sb_info *c, uint32_t minsize,
212 			   uint32_t *len, uint32_t sumsize)
213 {
214 	int ret = -EAGAIN;
215 	minsize = PAD(minsize);
216 
217 	jffs2_dbg(1, "%s(): Requested 0x%x bytes\n", __func__, minsize);
218 
219 	spin_lock(&c->erase_completion_lock);
220 	while(ret == -EAGAIN) {
221 		ret = jffs2_do_reserve_space(c, minsize, len, sumsize);
222 		if (ret) {
223 			jffs2_dbg(1, "%s(): looping, ret is %d\n",
224 				  __func__, ret);
225 		}
226 	}
227 	spin_unlock(&c->erase_completion_lock);
228 	if (!ret)
229 		ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1);
230 
231 	return ret;
232 }
233 
234 
235 /* Classify nextblock (clean, dirty of verydirty) and force to select an other one */
236 
237 static void jffs2_close_nextblock(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
238 {
239 
240 	if (c->nextblock == NULL) {
241 		jffs2_dbg(1, "%s(): Erase block at 0x%08x has already been placed in a list\n",
242 			  __func__, jeb->offset);
243 		return;
244 	}
245 	/* Check, if we have a dirty block now, or if it was dirty already */
246 	if (ISDIRTY (jeb->wasted_size + jeb->dirty_size)) {
247 		c->dirty_size += jeb->wasted_size;
248 		c->wasted_size -= jeb->wasted_size;
249 		jeb->dirty_size += jeb->wasted_size;
250 		jeb->wasted_size = 0;
251 		if (VERYDIRTY(c, jeb->dirty_size)) {
252 			jffs2_dbg(1, "Adding full erase block at 0x%08x to very_dirty_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
253 				  jeb->offset, jeb->free_size, jeb->dirty_size,
254 				  jeb->used_size);
255 			list_add_tail(&jeb->list, &c->very_dirty_list);
256 		} else {
257 			jffs2_dbg(1, "Adding full erase block at 0x%08x to dirty_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
258 				  jeb->offset, jeb->free_size, jeb->dirty_size,
259 				  jeb->used_size);
260 			list_add_tail(&jeb->list, &c->dirty_list);
261 		}
262 	} else {
263 		jffs2_dbg(1, "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
264 			  jeb->offset, jeb->free_size, jeb->dirty_size,
265 			  jeb->used_size);
266 		list_add_tail(&jeb->list, &c->clean_list);
267 	}
268 	c->nextblock = NULL;
269 
270 }
271 
272 /* Select a new jeb for nextblock */
273 
274 static int jffs2_find_nextblock(struct jffs2_sb_info *c)
275 {
276 	struct list_head *next;
277 
278 	/* Take the next block off the 'free' list */
279 
280 	if (list_empty(&c->free_list)) {
281 
282 		if (!c->nr_erasing_blocks &&
283 			!list_empty(&c->erasable_list)) {
284 			struct jffs2_eraseblock *ejeb;
285 
286 			ejeb = list_entry(c->erasable_list.next, struct jffs2_eraseblock, list);
287 			list_move_tail(&ejeb->list, &c->erase_pending_list);
288 			c->nr_erasing_blocks++;
289 			jffs2_garbage_collect_trigger(c);
290 			jffs2_dbg(1, "%s(): Triggering erase of erasable block at 0x%08x\n",
291 				  __func__, ejeb->offset);
292 		}
293 
294 		if (!c->nr_erasing_blocks &&
295 			!list_empty(&c->erasable_pending_wbuf_list)) {
296 			jffs2_dbg(1, "%s(): Flushing write buffer\n",
297 				  __func__);
298 			/* c->nextblock is NULL, no update to c->nextblock allowed */
299 			spin_unlock(&c->erase_completion_lock);
300 			jffs2_flush_wbuf_pad(c);
301 			spin_lock(&c->erase_completion_lock);
302 			/* Have another go. It'll be on the erasable_list now */
303 			return -EAGAIN;
304 		}
305 
306 		if (!c->nr_erasing_blocks) {
307 			/* Ouch. We're in GC, or we wouldn't have got here.
308 			   And there's no space left. At all. */
309 			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",
310 				c->nr_erasing_blocks, c->nr_free_blocks,
311 				list_empty(&c->erasable_list) ? "yes" : "no",
312 				list_empty(&c->erasing_list) ? "yes" : "no",
313 				list_empty(&c->erase_pending_list) ? "yes" : "no");
314 			return -ENOSPC;
315 		}
316 
317 		spin_unlock(&c->erase_completion_lock);
318 		/* Don't wait for it; just erase one right now */
319 		jffs2_erase_pending_blocks(c, 1);
320 		spin_lock(&c->erase_completion_lock);
321 
322 		/* An erase may have failed, decreasing the
323 		   amount of free space available. So we must
324 		   restart from the beginning */
325 		return -EAGAIN;
326 	}
327 
328 	next = c->free_list.next;
329 	list_del(next);
330 	c->nextblock = list_entry(next, struct jffs2_eraseblock, list);
331 	c->nr_free_blocks--;
332 
333 	jffs2_sum_reset_collected(c->summary); /* reset collected summary */
334 
335 #ifdef CONFIG_JFFS2_FS_WRITEBUFFER
336 	/* adjust write buffer offset, else we get a non contiguous write bug */
337 	if (!(c->wbuf_ofs % c->sector_size) && !c->wbuf_len)
338 		c->wbuf_ofs = 0xffffffff;
339 #endif
340 
341 	jffs2_dbg(1, "%s(): new nextblock = 0x%08x\n",
342 		  __func__, c->nextblock->offset);
343 
344 	return 0;
345 }
346 
347 /* Called with alloc sem _and_ erase_completion_lock */
348 static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
349 				  uint32_t *len, uint32_t sumsize)
350 {
351 	struct jffs2_eraseblock *jeb = c->nextblock;
352 	uint32_t reserved_size;				/* for summary information at the end of the jeb */
353 	int ret;
354 
355  restart:
356 	reserved_size = 0;
357 
358 	if (jffs2_sum_active() && (sumsize != JFFS2_SUMMARY_NOSUM_SIZE)) {
359 							/* NOSUM_SIZE means not to generate summary */
360 
361 		if (jeb) {
362 			reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE);
363 			dbg_summary("minsize=%d , jeb->free=%d ,"
364 						"summary->size=%d , sumsize=%d\n",
365 						minsize, jeb->free_size,
366 						c->summary->sum_size, sumsize);
367 		}
368 
369 		/* Is there enough space for writing out the current node, or we have to
370 		   write out summary information now, close this jeb and select new nextblock? */
371 		if (jeb && (PAD(minsize) + PAD(c->summary->sum_size + sumsize +
372 					JFFS2_SUMMARY_FRAME_SIZE) > jeb->free_size)) {
373 
374 			/* Has summary been disabled for this jeb? */
375 			if (jffs2_sum_is_disabled(c->summary)) {
376 				sumsize = JFFS2_SUMMARY_NOSUM_SIZE;
377 				goto restart;
378 			}
379 
380 			/* Writing out the collected summary information */
381 			dbg_summary("generating summary for 0x%08x.\n", jeb->offset);
382 			ret = jffs2_sum_write_sumnode(c);
383 
384 			if (ret)
385 				return ret;
386 
387 			if (jffs2_sum_is_disabled(c->summary)) {
388 				/* jffs2_write_sumnode() couldn't write out the summary information
389 				   diabling summary for this jeb and free the collected information
390 				 */
391 				sumsize = JFFS2_SUMMARY_NOSUM_SIZE;
392 				goto restart;
393 			}
394 
395 			jffs2_close_nextblock(c, jeb);
396 			jeb = NULL;
397 			/* keep always valid value in reserved_size */
398 			reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE);
399 		}
400 	} else {
401 		if (jeb && minsize > jeb->free_size) {
402 			uint32_t waste;
403 
404 			/* Skip the end of this block and file it as having some dirty space */
405 			/* If there's a pending write to it, flush now */
406 
407 			if (jffs2_wbuf_dirty(c)) {
408 				spin_unlock(&c->erase_completion_lock);
409 				jffs2_dbg(1, "%s(): Flushing write buffer\n",
410 					  __func__);
411 				jffs2_flush_wbuf_pad(c);
412 				spin_lock(&c->erase_completion_lock);
413 				jeb = c->nextblock;
414 				goto restart;
415 			}
416 
417 			spin_unlock(&c->erase_completion_lock);
418 
419 			ret = jffs2_prealloc_raw_node_refs(c, jeb, 1);
420 			if (ret)
421 				return ret;
422 			/* Just lock it again and continue. Nothing much can change because
423 			   we hold c->alloc_sem anyway. In fact, it's not entirely clear why
424 			   we hold c->erase_completion_lock in the majority of this function...
425 			   but that's a question for another (more caffeine-rich) day. */
426 			spin_lock(&c->erase_completion_lock);
427 
428 			waste = jeb->free_size;
429 			jffs2_link_node_ref(c, jeb,
430 					    (jeb->offset + c->sector_size - waste) | REF_OBSOLETE,
431 					    waste, NULL);
432 			/* FIXME: that made it count as dirty. Convert to wasted */
433 			jeb->dirty_size -= waste;
434 			c->dirty_size -= waste;
435 			jeb->wasted_size += waste;
436 			c->wasted_size += waste;
437 
438 			jffs2_close_nextblock(c, jeb);
439 			jeb = NULL;
440 		}
441 	}
442 
443 	if (!jeb) {
444 
445 		ret = jffs2_find_nextblock(c);
446 		if (ret)
447 			return ret;
448 
449 		jeb = c->nextblock;
450 
451 		if (jeb->free_size != c->sector_size - c->cleanmarker_size) {
452 			pr_warn("Eep. Block 0x%08x taken from free_list had free_size of 0x%08x!!\n",
453 				jeb->offset, jeb->free_size);
454 			goto restart;
455 		}
456 	}
457 	/* OK, jeb (==c->nextblock) is now pointing at a block which definitely has
458 	   enough space */
459 	*len = jeb->free_size - reserved_size;
460 
461 	if (c->cleanmarker_size && jeb->used_size == c->cleanmarker_size &&
462 	    !jeb->first_node->next_in_ino) {
463 		/* Only node in it beforehand was a CLEANMARKER node (we think).
464 		   So mark it obsolete now that there's going to be another node
465 		   in the block. This will reduce used_size to zero but We've
466 		   already set c->nextblock so that jffs2_mark_node_obsolete()
467 		   won't try to refile it to the dirty_list.
468 		*/
469 		spin_unlock(&c->erase_completion_lock);
470 		jffs2_mark_node_obsolete(c, jeb->first_node);
471 		spin_lock(&c->erase_completion_lock);
472 	}
473 
474 	jffs2_dbg(1, "%s(): Giving 0x%x bytes at 0x%x\n",
475 		  __func__,
476 		  *len, jeb->offset + (c->sector_size - jeb->free_size));
477 	return 0;
478 }
479 
480 /**
481  *	jffs2_add_physical_node_ref - add a physical node reference to the list
482  *	@c: superblock info
483  *	@new: new node reference to add
484  *	@len: length of this physical node
485  *
486  *	Should only be used to report nodes for which space has been allocated
487  *	by jffs2_reserve_space.
488  *
489  *	Must be called with the alloc_sem held.
490  */
491 
492 struct jffs2_raw_node_ref *jffs2_add_physical_node_ref(struct jffs2_sb_info *c,
493 						       uint32_t ofs, uint32_t len,
494 						       struct jffs2_inode_cache *ic)
495 {
496 	struct jffs2_eraseblock *jeb;
497 	struct jffs2_raw_node_ref *new;
498 
499 	jeb = &c->blocks[ofs / c->sector_size];
500 
501 	jffs2_dbg(1, "%s(): Node at 0x%x(%d), size 0x%x\n",
502 		  __func__, ofs & ~3, ofs & 3, len);
503 #if 1
504 	/* Allow non-obsolete nodes only to be added at the end of c->nextblock,
505 	   if c->nextblock is set. Note that wbuf.c will file obsolete nodes
506 	   even after refiling c->nextblock */
507 	if ((c->nextblock || ((ofs & 3) != REF_OBSOLETE))
508 	    && (jeb != c->nextblock || (ofs & ~3) != jeb->offset + (c->sector_size - jeb->free_size))) {
509 		pr_warn("argh. node added in wrong place at 0x%08x(%d)\n",
510 			ofs & ~3, ofs & 3);
511 		if (c->nextblock)
512 			pr_warn("nextblock 0x%08x", c->nextblock->offset);
513 		else
514 			pr_warn("No nextblock");
515 		pr_cont(", expected at %08x\n",
516 			jeb->offset + (c->sector_size - jeb->free_size));
517 		return ERR_PTR(-EINVAL);
518 	}
519 #endif
520 	spin_lock(&c->erase_completion_lock);
521 
522 	new = jffs2_link_node_ref(c, jeb, ofs, len, ic);
523 
524 	if (!jeb->free_size && !jeb->dirty_size && !ISDIRTY(jeb->wasted_size)) {
525 		/* If it lives on the dirty_list, jffs2_reserve_space will put it there */
526 		jffs2_dbg(1, "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
527 			  jeb->offset, jeb->free_size, jeb->dirty_size,
528 			  jeb->used_size);
529 		if (jffs2_wbuf_dirty(c)) {
530 			/* Flush the last write in the block if it's outstanding */
531 			spin_unlock(&c->erase_completion_lock);
532 			jffs2_flush_wbuf_pad(c);
533 			spin_lock(&c->erase_completion_lock);
534 		}
535 
536 		list_add_tail(&jeb->list, &c->clean_list);
537 		c->nextblock = NULL;
538 	}
539 	jffs2_dbg_acct_sanity_check_nolock(c,jeb);
540 	jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
541 
542 	spin_unlock(&c->erase_completion_lock);
543 
544 	return new;
545 }
546 
547 
548 void jffs2_complete_reservation(struct jffs2_sb_info *c)
549 {
550 	jffs2_dbg(1, "jffs2_complete_reservation()\n");
551 	spin_lock(&c->erase_completion_lock);
552 	jffs2_garbage_collect_trigger(c);
553 	spin_unlock(&c->erase_completion_lock);
554 	mutex_unlock(&c->alloc_sem);
555 }
556 
557 static inline int on_list(struct list_head *obj, struct list_head *head)
558 {
559 	struct list_head *this;
560 
561 	list_for_each(this, head) {
562 		if (this == obj) {
563 			jffs2_dbg(1, "%p is on list at %p\n", obj, head);
564 			return 1;
565 
566 		}
567 	}
568 	return 0;
569 }
570 
571 void jffs2_mark_node_obsolete(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref)
572 {
573 	struct jffs2_eraseblock *jeb;
574 	int blocknr;
575 	struct jffs2_unknown_node n;
576 	int ret, addedsize;
577 	size_t retlen;
578 	uint32_t freed_len;
579 
580 	if(unlikely(!ref)) {
581 		pr_notice("EEEEEK. jffs2_mark_node_obsolete called with NULL node\n");
582 		return;
583 	}
584 	if (ref_obsolete(ref)) {
585 		jffs2_dbg(1, "%s(): called with already obsolete node at 0x%08x\n",
586 			  __func__, ref_offset(ref));
587 		return;
588 	}
589 	blocknr = ref->flash_offset / c->sector_size;
590 	if (blocknr >= c->nr_blocks) {
591 		pr_notice("raw node at 0x%08x is off the end of device!\n",
592 			  ref->flash_offset);
593 		BUG();
594 	}
595 	jeb = &c->blocks[blocknr];
596 
597 	if (jffs2_can_mark_obsolete(c) && !jffs2_is_readonly(c) &&
598 	    !(c->flags & (JFFS2_SB_FLAG_SCANNING | JFFS2_SB_FLAG_BUILDING))) {
599 		/* Hm. This may confuse static lock analysis. If any of the above
600 		   three conditions is false, we're going to return from this
601 		   function without actually obliterating any nodes or freeing
602 		   any jffs2_raw_node_refs. So we don't need to stop erases from
603 		   happening, or protect against people holding an obsolete
604 		   jffs2_raw_node_ref without the erase_completion_lock. */
605 		mutex_lock(&c->erase_free_sem);
606 	}
607 
608 	spin_lock(&c->erase_completion_lock);
609 
610 	freed_len = ref_totlen(c, jeb, ref);
611 
612 	if (ref_flags(ref) == REF_UNCHECKED) {
613 		D1(if (unlikely(jeb->unchecked_size < freed_len)) {
614 				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",
615 					  freed_len, blocknr,
616 					  ref->flash_offset, jeb->used_size);
617 			BUG();
618 		})
619 			jffs2_dbg(1, "Obsoleting previously unchecked node at 0x%08x of len %x\n",
620 				  ref_offset(ref), freed_len);
621 		jeb->unchecked_size -= freed_len;
622 		c->unchecked_size -= freed_len;
623 	} else {
624 		D1(if (unlikely(jeb->used_size < freed_len)) {
625 				pr_notice("raw node of size 0x%08x freed from erase block %d at 0x%08x, but used_size was already 0x%08x\n",
626 					  freed_len, blocknr,
627 					  ref->flash_offset, jeb->used_size);
628 			BUG();
629 		})
630 			jffs2_dbg(1, "Obsoleting node at 0x%08x of len %#x: ",
631 				  ref_offset(ref), freed_len);
632 		jeb->used_size -= freed_len;
633 		c->used_size -= freed_len;
634 	}
635 
636 	// Take care, that wasted size is taken into concern
637 	if ((jeb->dirty_size || ISDIRTY(jeb->wasted_size + freed_len)) && jeb != c->nextblock) {
638 		jffs2_dbg(1, "Dirtying\n");
639 		addedsize = freed_len;
640 		jeb->dirty_size += freed_len;
641 		c->dirty_size += freed_len;
642 
643 		/* Convert wasted space to dirty, if not a bad block */
644 		if (jeb->wasted_size) {
645 			if (on_list(&jeb->list, &c->bad_used_list)) {
646 				jffs2_dbg(1, "Leaving block at %08x on the bad_used_list\n",
647 					  jeb->offset);
648 				addedsize = 0; /* To fool the refiling code later */
649 			} else {
650 				jffs2_dbg(1, "Converting %d bytes of wasted space to dirty in block at %08x\n",
651 					  jeb->wasted_size, jeb->offset);
652 				addedsize += jeb->wasted_size;
653 				jeb->dirty_size += jeb->wasted_size;
654 				c->dirty_size += jeb->wasted_size;
655 				c->wasted_size -= jeb->wasted_size;
656 				jeb->wasted_size = 0;
657 			}
658 		}
659 	} else {
660 		jffs2_dbg(1, "Wasting\n");
661 		addedsize = 0;
662 		jeb->wasted_size += freed_len;
663 		c->wasted_size += freed_len;
664 	}
665 	ref->flash_offset = ref_offset(ref) | REF_OBSOLETE;
666 
667 	jffs2_dbg_acct_sanity_check_nolock(c, jeb);
668 	jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
669 
670 	if (c->flags & JFFS2_SB_FLAG_SCANNING) {
671 		/* Flash scanning is in progress. Don't muck about with the block
672 		   lists because they're not ready yet, and don't actually
673 		   obliterate nodes that look obsolete. If they weren't
674 		   marked obsolete on the flash at the time they _became_
675 		   obsolete, there was probably a reason for that. */
676 		spin_unlock(&c->erase_completion_lock);
677 		/* We didn't lock the erase_free_sem */
678 		return;
679 	}
680 
681 	if (jeb == c->nextblock) {
682 		jffs2_dbg(2, "Not moving nextblock 0x%08x to dirty/erase_pending list\n",
683 			  jeb->offset);
684 	} else if (!jeb->used_size && !jeb->unchecked_size) {
685 		if (jeb == c->gcblock) {
686 			jffs2_dbg(1, "gcblock at 0x%08x completely dirtied. Clearing gcblock...\n",
687 				  jeb->offset);
688 			c->gcblock = NULL;
689 		} else {
690 			jffs2_dbg(1, "Eraseblock at 0x%08x completely dirtied. Removing from (dirty?) list...\n",
691 				  jeb->offset);
692 			list_del(&jeb->list);
693 		}
694 		if (jffs2_wbuf_dirty(c)) {
695 			jffs2_dbg(1, "...and adding to erasable_pending_wbuf_list\n");
696 			list_add_tail(&jeb->list, &c->erasable_pending_wbuf_list);
697 		} else {
698 			if (jiffies & 127) {
699 				/* Most of the time, we just erase it immediately. Otherwise we
700 				   spend ages scanning it on mount, etc. */
701 				jffs2_dbg(1, "...and adding to erase_pending_list\n");
702 				list_add_tail(&jeb->list, &c->erase_pending_list);
703 				c->nr_erasing_blocks++;
704 				jffs2_garbage_collect_trigger(c);
705 			} else {
706 				/* Sometimes, however, we leave it elsewhere so it doesn't get
707 				   immediately reused, and we spread the load a bit. */
708 				jffs2_dbg(1, "...and adding to erasable_list\n");
709 				list_add_tail(&jeb->list, &c->erasable_list);
710 			}
711 		}
712 		jffs2_dbg(1, "Done OK\n");
713 	} else if (jeb == c->gcblock) {
714 		jffs2_dbg(2, "Not moving gcblock 0x%08x to dirty_list\n",
715 			  jeb->offset);
716 	} else if (ISDIRTY(jeb->dirty_size) && !ISDIRTY(jeb->dirty_size - addedsize)) {
717 		jffs2_dbg(1, "Eraseblock at 0x%08x is freshly dirtied. Removing from clean list...\n",
718 			  jeb->offset);
719 		list_del(&jeb->list);
720 		jffs2_dbg(1, "...and adding to dirty_list\n");
721 		list_add_tail(&jeb->list, &c->dirty_list);
722 	} else if (VERYDIRTY(c, jeb->dirty_size) &&
723 		   !VERYDIRTY(c, jeb->dirty_size - addedsize)) {
724 		jffs2_dbg(1, "Eraseblock at 0x%08x is now very dirty. Removing from dirty list...\n",
725 			  jeb->offset);
726 		list_del(&jeb->list);
727 		jffs2_dbg(1, "...and adding to very_dirty_list\n");
728 		list_add_tail(&jeb->list, &c->very_dirty_list);
729 	} else {
730 		jffs2_dbg(1, "Eraseblock at 0x%08x not moved anywhere. (free 0x%08x, dirty 0x%08x, used 0x%08x)\n",
731 			  jeb->offset, jeb->free_size, jeb->dirty_size,
732 			  jeb->used_size);
733 	}
734 
735 	spin_unlock(&c->erase_completion_lock);
736 
737 	if (!jffs2_can_mark_obsolete(c) || jffs2_is_readonly(c) ||
738 		(c->flags & JFFS2_SB_FLAG_BUILDING)) {
739 		/* We didn't lock the erase_free_sem */
740 		return;
741 	}
742 
743 	/* The erase_free_sem is locked, and has been since before we marked the node obsolete
744 	   and potentially put its eraseblock onto the erase_pending_list. Thus, we know that
745 	   the block hasn't _already_ been erased, and that 'ref' itself hasn't been freed yet
746 	   by jffs2_free_jeb_node_refs() in erase.c. Which is nice. */
747 
748 	jffs2_dbg(1, "obliterating obsoleted node at 0x%08x\n",
749 		  ref_offset(ref));
750 	ret = jffs2_flash_read(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
751 	if (ret) {
752 		pr_warn("Read error reading from obsoleted node at 0x%08x: %d\n",
753 			ref_offset(ref), ret);
754 		goto out_erase_sem;
755 	}
756 	if (retlen != sizeof(n)) {
757 		pr_warn("Short read from obsoleted node at 0x%08x: %zd\n",
758 			ref_offset(ref), retlen);
759 		goto out_erase_sem;
760 	}
761 	if (PAD(je32_to_cpu(n.totlen)) != PAD(freed_len)) {
762 		pr_warn("Node totlen on flash (0x%08x) != totlen from node ref (0x%08x)\n",
763 			je32_to_cpu(n.totlen), freed_len);
764 		goto out_erase_sem;
765 	}
766 	if (!(je16_to_cpu(n.nodetype) & JFFS2_NODE_ACCURATE)) {
767 		jffs2_dbg(1, "Node at 0x%08x was already marked obsolete (nodetype 0x%04x)\n",
768 			  ref_offset(ref), je16_to_cpu(n.nodetype));
769 		goto out_erase_sem;
770 	}
771 	/* XXX FIXME: This is ugly now */
772 	n.nodetype = cpu_to_je16(je16_to_cpu(n.nodetype) & ~JFFS2_NODE_ACCURATE);
773 	ret = jffs2_flash_write(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
774 	if (ret) {
775 		pr_warn("Write error in obliterating obsoleted node at 0x%08x: %d\n",
776 			ref_offset(ref), ret);
777 		goto out_erase_sem;
778 	}
779 	if (retlen != sizeof(n)) {
780 		pr_warn("Short write in obliterating obsoleted node at 0x%08x: %zd\n",
781 			ref_offset(ref), retlen);
782 		goto out_erase_sem;
783 	}
784 
785 	/* Nodes which have been marked obsolete no longer need to be
786 	   associated with any inode. Remove them from the per-inode list.
787 
788 	   Note we can't do this for NAND at the moment because we need
789 	   obsolete dirent nodes to stay on the lists, because of the
790 	   horridness in jffs2_garbage_collect_deletion_dirent(). Also
791 	   because we delete the inocache, and on NAND we need that to
792 	   stay around until all the nodes are actually erased, in order
793 	   to stop us from giving the same inode number to another newly
794 	   created inode. */
795 	if (ref->next_in_ino) {
796 		struct jffs2_inode_cache *ic;
797 		struct jffs2_raw_node_ref **p;
798 
799 		spin_lock(&c->erase_completion_lock);
800 
801 		ic = jffs2_raw_ref_to_ic(ref);
802 		for (p = &ic->nodes; (*p) != ref; p = &((*p)->next_in_ino))
803 			;
804 
805 		*p = ref->next_in_ino;
806 		ref->next_in_ino = NULL;
807 
808 		switch (ic->class) {
809 #ifdef CONFIG_JFFS2_FS_XATTR
810 			case RAWNODE_CLASS_XATTR_DATUM:
811 				jffs2_release_xattr_datum(c, (struct jffs2_xattr_datum *)ic);
812 				break;
813 			case RAWNODE_CLASS_XATTR_REF:
814 				jffs2_release_xattr_ref(c, (struct jffs2_xattr_ref *)ic);
815 				break;
816 #endif
817 			default:
818 				if (ic->nodes == (void *)ic && ic->pino_nlink == 0)
819 					jffs2_del_ino_cache(c, ic);
820 				break;
821 		}
822 		spin_unlock(&c->erase_completion_lock);
823 	}
824 
825  out_erase_sem:
826 	mutex_unlock(&c->erase_free_sem);
827 }
828 
829 int jffs2_thread_should_wake(struct jffs2_sb_info *c)
830 {
831 	int ret = 0;
832 	uint32_t dirty;
833 	int nr_very_dirty = 0;
834 	struct jffs2_eraseblock *jeb;
835 
836 	if (!list_empty(&c->erase_complete_list) ||
837 	    !list_empty(&c->erase_pending_list))
838 		return 1;
839 
840 	if (c->unchecked_size) {
841 		jffs2_dbg(1, "jffs2_thread_should_wake(): unchecked_size %d, checked_ino #%d\n",
842 			  c->unchecked_size, c->checked_ino);
843 		return 1;
844 	}
845 
846 	/* dirty_size contains blocks on erase_pending_list
847 	 * those blocks are counted in c->nr_erasing_blocks.
848 	 * If one block is actually erased, it is not longer counted as dirty_space
849 	 * but it is counted in c->nr_erasing_blocks, so we add it and subtract it
850 	 * with c->nr_erasing_blocks * c->sector_size again.
851 	 * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks
852 	 * This helps us to force gc and pick eventually a clean block to spread the load.
853 	 */
854 	dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size;
855 
856 	if (c->nr_free_blocks + c->nr_erasing_blocks < c->resv_blocks_gctrigger &&
857 			(dirty > c->nospc_dirty_size))
858 		ret = 1;
859 
860 	list_for_each_entry(jeb, &c->very_dirty_list, list) {
861 		nr_very_dirty++;
862 		if (nr_very_dirty == c->vdirty_blocks_gctrigger) {
863 			ret = 1;
864 			/* In debug mode, actually go through and count them all */
865 			D1(continue);
866 			break;
867 		}
868 	}
869 
870 	jffs2_dbg(1, "%s(): nr_free_blocks %d, nr_erasing_blocks %d, dirty_size 0x%x, vdirty_blocks %d: %s\n",
871 		  __func__, c->nr_free_blocks, c->nr_erasing_blocks,
872 		  c->dirty_size, nr_very_dirty, ret ? "yes" : "no");
873 
874 	return ret;
875 }
876