blob: 49127a1f045861324bab8e79570b895b7b1ab4c5 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * JFFS2 -- Journalling Flash File System, Version 2.
3 *
4 * Copyright (C) 2001-2003 Red Hat, Inc.
5 *
6 * Created by David Woodhouse <dwmw2@infradead.org>
7 *
8 * For licensing information, see the file 'LICENCE' in this directory.
9 *
Artem B. Bityutskiy733802d2005-09-22 12:25:00 +010010 * $Id: nodemgmt.c,v 1.127 2005/09/20 15:49:12 dedekind Exp $
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 *
12 */
13
14#include <linux/kernel.h>
15#include <linux/slab.h>
16#include <linux/mtd/mtd.h>
17#include <linux/compiler.h>
18#include <linux/sched.h> /* For cond_resched() */
19#include "nodelist.h"
Ferenc Havasie631ddb2005-09-07 09:35:26 +010020#include "debug.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070021
22/**
23 * jffs2_reserve_space - request physical space to write nodes to flash
24 * @c: superblock info
25 * @minsize: Minimum acceptable size of allocation
26 * @ofs: Returned value of node offset
27 * @len: Returned value of allocation length
28 * @prio: Allocation type - ALLOC_{NORMAL,DELETION}
29 *
30 * Requests a block of physical space on the flash. Returns zero for success
31 * and puts 'ofs' and 'len' into the appriopriate place, or returns -ENOSPC
32 * or other error if appropriate.
33 *
34 * If it returns zero, jffs2_reserve_space() also downs the per-filesystem
35 * allocation semaphore, to prevent more than one allocation from being
36 * active at any time. The semaphore is later released by jffs2_commit_allocation()
37 *
38 * jffs2_reserve_space() may trigger garbage collection in order to make room
39 * for the requested allocation.
40 */
41
Ferenc Havasie631ddb2005-09-07 09:35:26 +010042static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
43 uint32_t *ofs, uint32_t *len, uint32_t sumsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
Ferenc Havasie631ddb2005-09-07 09:35:26 +010045int jffs2_reserve_space(struct jffs2_sb_info *c, uint32_t minsize, uint32_t *ofs,
46 uint32_t *len, int prio, uint32_t sumsize)
Linus Torvalds1da177e2005-04-16 15:20:36 -070047{
48 int ret = -EAGAIN;
49 int blocksneeded = c->resv_blocks_write;
50 /* align it */
51 minsize = PAD(minsize);
52
53 D1(printk(KERN_DEBUG "jffs2_reserve_space(): Requested 0x%x bytes\n", minsize));
54 down(&c->alloc_sem);
55
56 D1(printk(KERN_DEBUG "jffs2_reserve_space(): alloc sem got\n"));
57
58 spin_lock(&c->erase_completion_lock);
59
60 /* this needs a little more thought (true <tglx> :)) */
61 while(ret == -EAGAIN) {
62 while(c->nr_free_blocks + c->nr_erasing_blocks < blocksneeded) {
63 int ret;
64 uint32_t dirty, avail;
65
66 /* calculate real dirty size
67 * dirty_size contains blocks on erase_pending_list
68 * those blocks are counted in c->nr_erasing_blocks.
69 * If one block is actually erased, it is not longer counted as dirty_space
70 * but it is counted in c->nr_erasing_blocks, so we add it and subtract it
71 * with c->nr_erasing_blocks * c->sector_size again.
72 * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks
73 * This helps us to force gc and pick eventually a clean block to spread the load.
74 * We add unchecked_size here, as we hopefully will find some space to use.
75 * This will affect the sum only once, as gc first finishes checking
76 * of nodes.
77 */
78 dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size + c->unchecked_size;
79 if (dirty < c->nospc_dirty_size) {
80 if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) {
Artem B. Bityuckiy4132ace2005-05-06 10:30:30 +010081 D1(printk(KERN_NOTICE "jffs2_reserve_space(): Low on dirty space to GC, but it's a deletion. Allowing...\n"));
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 break;
83 }
84 D1(printk(KERN_DEBUG "dirty size 0x%08x + unchecked_size 0x%08x < nospc_dirty_size 0x%08x, returning -ENOSPC\n",
85 dirty, c->unchecked_size, c->sector_size));
86
87 spin_unlock(&c->erase_completion_lock);
88 up(&c->alloc_sem);
89 return -ENOSPC;
90 }
Thomas Gleixner182ec4e2005-11-07 11:16:07 +000091
Linus Torvalds1da177e2005-04-16 15:20:36 -070092 /* Calc possibly available space. Possibly available means that we
93 * don't know, if unchecked size contains obsoleted nodes, which could give us some
94 * more usable space. This will affect the sum only once, as gc first finishes checking
95 * of nodes.
Thomas Gleixner182ec4e2005-11-07 11:16:07 +000096 + Return -ENOSPC, if the maximum possibly available space is less or equal than
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 * blocksneeded * sector_size.
98 * This blocks endless gc looping on a filesystem, which is nearly full, even if
99 * the check above passes.
100 */
101 avail = c->free_size + c->dirty_size + c->erasing_size + c->unchecked_size;
102 if ( (avail / c->sector_size) <= blocksneeded) {
103 if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) {
Artem B. Bityuckiy4132ace2005-05-06 10:30:30 +0100104 D1(printk(KERN_NOTICE "jffs2_reserve_space(): Low on possibly available space, but it's a deletion. Allowing...\n"));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 break;
106 }
107
108 D1(printk(KERN_DEBUG "max. available size 0x%08x < blocksneeded * sector_size 0x%08x, returning -ENOSPC\n",
109 avail, blocksneeded * c->sector_size));
110 spin_unlock(&c->erase_completion_lock);
111 up(&c->alloc_sem);
112 return -ENOSPC;
113 }
114
115 up(&c->alloc_sem);
116
117 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",
118 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,
119 c->free_size + c->dirty_size + c->wasted_size + c->used_size + c->erasing_size + c->bad_size, c->flash_size));
120 spin_unlock(&c->erase_completion_lock);
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000121
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 ret = jffs2_garbage_collect_pass(c);
123 if (ret)
124 return ret;
125
126 cond_resched();
127
128 if (signal_pending(current))
129 return -EINTR;
130
131 down(&c->alloc_sem);
132 spin_lock(&c->erase_completion_lock);
133 }
134
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100135 ret = jffs2_do_reserve_space(c, minsize, ofs, len, sumsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 if (ret) {
137 D1(printk(KERN_DEBUG "jffs2_reserve_space: ret is %d\n", ret));
138 }
139 }
140 spin_unlock(&c->erase_completion_lock);
141 if (ret)
142 up(&c->alloc_sem);
143 return ret;
144}
145
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100146int jffs2_reserve_space_gc(struct jffs2_sb_info *c, uint32_t minsize, uint32_t *ofs,
147 uint32_t *len, uint32_t sumsize)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148{
149 int ret = -EAGAIN;
150 minsize = PAD(minsize);
151
152 D1(printk(KERN_DEBUG "jffs2_reserve_space_gc(): Requested 0x%x bytes\n", minsize));
153
154 spin_lock(&c->erase_completion_lock);
155 while(ret == -EAGAIN) {
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100156 ret = jffs2_do_reserve_space(c, minsize, ofs, len, sumsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 if (ret) {
158 D1(printk(KERN_DEBUG "jffs2_reserve_space_gc: looping, ret is %d\n", ret));
159 }
160 }
161 spin_unlock(&c->erase_completion_lock);
162 return ret;
163}
164
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100165
166/* Classify nextblock (clean, dirty of verydirty) and force to select an other one */
167
168static void jffs2_close_nextblock(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169{
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100170
171 /* Check, if we have a dirty block now, or if it was dirty already */
172 if (ISDIRTY (jeb->wasted_size + jeb->dirty_size)) {
173 c->dirty_size += jeb->wasted_size;
174 c->wasted_size -= jeb->wasted_size;
175 jeb->dirty_size += jeb->wasted_size;
176 jeb->wasted_size = 0;
177 if (VERYDIRTY(c, jeb->dirty_size)) {
178 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",
179 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
180 list_add_tail(&jeb->list, &c->very_dirty_list);
181 } else {
182 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to dirty_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
183 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
184 list_add_tail(&jeb->list, &c->dirty_list);
185 }
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000186 } else {
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100187 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
188 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
189 list_add_tail(&jeb->list, &c->clean_list);
190 }
191 c->nextblock = NULL;
192
193}
194
195/* Select a new jeb for nextblock */
196
197static int jffs2_find_nextblock(struct jffs2_sb_info *c)
198{
199 struct list_head *next;
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000200
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100201 /* Take the next block off the 'free' list */
202
203 if (list_empty(&c->free_list)) {
204
205 if (!c->nr_erasing_blocks &&
206 !list_empty(&c->erasable_list)) {
207 struct jffs2_eraseblock *ejeb;
208
209 ejeb = list_entry(c->erasable_list.next, struct jffs2_eraseblock, list);
210 list_del(&ejeb->list);
211 list_add_tail(&ejeb->list, &c->erase_pending_list);
212 c->nr_erasing_blocks++;
213 jffs2_erase_pending_trigger(c);
214 D1(printk(KERN_DEBUG "jffs2_find_nextblock: Triggering erase of erasable block at 0x%08x\n",
215 ejeb->offset));
216 }
217
218 if (!c->nr_erasing_blocks &&
219 !list_empty(&c->erasable_pending_wbuf_list)) {
220 D1(printk(KERN_DEBUG "jffs2_find_nextblock: Flushing write buffer\n"));
221 /* c->nextblock is NULL, no update to c->nextblock allowed */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 spin_unlock(&c->erase_completion_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 jffs2_flush_wbuf_pad(c);
224 spin_lock(&c->erase_completion_lock);
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100225 /* Have another go. It'll be on the erasable_list now */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 return -EAGAIN;
227 }
228
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100229 if (!c->nr_erasing_blocks) {
230 /* Ouch. We're in GC, or we wouldn't have got here.
231 And there's no space left. At all. */
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000232 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",
233 c->nr_erasing_blocks, c->nr_free_blocks, list_empty(&c->erasable_list)?"yes":"no",
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100234 list_empty(&c->erasing_list)?"yes":"no", list_empty(&c->erase_pending_list)?"yes":"no");
235 return -ENOSPC;
236 }
237
238 spin_unlock(&c->erase_completion_lock);
239 /* Don't wait for it; just erase one right now */
240 jffs2_erase_pending_blocks(c, 1);
241 spin_lock(&c->erase_completion_lock);
242
243 /* An erase may have failed, decreasing the
244 amount of free space available. So we must
245 restart from the beginning */
246 return -EAGAIN;
247 }
248
249 next = c->free_list.next;
250 list_del(next);
251 c->nextblock = list_entry(next, struct jffs2_eraseblock, list);
252 c->nr_free_blocks--;
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000253
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100254 jffs2_sum_reset_collected(c->summary); /* reset collected summary */
255
256 D1(printk(KERN_DEBUG "jffs2_find_nextblock(): new nextblock = 0x%08x\n", c->nextblock->offset));
257
258 return 0;
259}
260
261/* Called with alloc sem _and_ erase_completion_lock */
262static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize, uint32_t *ofs, uint32_t *len, uint32_t sumsize)
263{
264 struct jffs2_eraseblock *jeb = c->nextblock;
265 uint32_t reserved_size; /* for summary information at the end of the jeb */
266 int ret;
267
268 restart:
269 reserved_size = 0;
270
271 if (jffs2_sum_active() && (sumsize != JFFS2_SUMMARY_NOSUM_SIZE)) {
272 /* NOSUM_SIZE means not to generate summary */
273
274 if (jeb) {
275 reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE);
Artem B. Bityutskiy733802d2005-09-22 12:25:00 +0100276 dbg_summary("minsize=%d , jeb->free=%d ,"
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100277 "summary->size=%d , sumsize=%d\n",
278 minsize, jeb->free_size,
279 c->summary->sum_size, sumsize);
280 }
281
282 /* Is there enough space for writing out the current node, or we have to
283 write out summary information now, close this jeb and select new nextblock? */
284 if (jeb && (PAD(minsize) + PAD(c->summary->sum_size + sumsize +
285 JFFS2_SUMMARY_FRAME_SIZE) > jeb->free_size)) {
286
287 /* Has summary been disabled for this jeb? */
288 if (jffs2_sum_is_disabled(c->summary)) {
289 sumsize = JFFS2_SUMMARY_NOSUM_SIZE;
290 goto restart;
291 }
292
293 /* Writing out the collected summary information */
Artem B. Bityutskiy733802d2005-09-22 12:25:00 +0100294 dbg_summary("generating summary for 0x%08x.\n", jeb->offset);
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100295 ret = jffs2_sum_write_sumnode(c);
296
297 if (ret)
298 return ret;
299
300 if (jffs2_sum_is_disabled(c->summary)) {
301 /* jffs2_write_sumnode() couldn't write out the summary information
302 diabling summary for this jeb and free the collected information
303 */
304 sumsize = JFFS2_SUMMARY_NOSUM_SIZE;
305 goto restart;
306 }
307
308 jffs2_close_nextblock(c, jeb);
309 jeb = NULL;
Ferenc Havasi34c0e902005-09-16 13:58:20 +0100310 /* keep always valid value in reserved_size */
311 reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE);
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100312 }
313 } else {
314 if (jeb && minsize > jeb->free_size) {
315 /* Skip the end of this block and file it as having some dirty space */
316 /* If there's a pending write to it, flush now */
317
318 if (jffs2_wbuf_dirty(c)) {
319 spin_unlock(&c->erase_completion_lock);
320 D1(printk(KERN_DEBUG "jffs2_do_reserve_space: Flushing write buffer\n"));
321 jffs2_flush_wbuf_pad(c);
322 spin_lock(&c->erase_completion_lock);
323 jeb = c->nextblock;
324 goto restart;
325 }
326
327 c->wasted_size += jeb->free_size;
328 c->free_size -= jeb->free_size;
329 jeb->wasted_size += jeb->free_size;
330 jeb->free_size = 0;
331
332 jffs2_close_nextblock(c, jeb);
333 jeb = NULL;
334 }
335 }
336
337 if (!jeb) {
338
339 ret = jffs2_find_nextblock(c);
340 if (ret)
341 return ret;
342
343 jeb = c->nextblock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344
345 if (jeb->free_size != c->sector_size - c->cleanmarker_size) {
346 printk(KERN_WARNING "Eep. Block 0x%08x taken from free_list had free_size of 0x%08x!!\n", jeb->offset, jeb->free_size);
347 goto restart;
348 }
349 }
350 /* OK, jeb (==c->nextblock) is now pointing at a block which definitely has
351 enough space */
352 *ofs = jeb->offset + (c->sector_size - jeb->free_size);
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100353 *len = jeb->free_size - reserved_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354
355 if (c->cleanmarker_size && jeb->used_size == c->cleanmarker_size &&
356 !jeb->first_node->next_in_ino) {
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000357 /* Only node in it beforehand was a CLEANMARKER node (we think).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 So mark it obsolete now that there's going to be another node
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000359 in the block. This will reduce used_size to zero but We've
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 already set c->nextblock so that jffs2_mark_node_obsolete()
361 won't try to refile it to the dirty_list.
362 */
363 spin_unlock(&c->erase_completion_lock);
364 jffs2_mark_node_obsolete(c, jeb->first_node);
365 spin_lock(&c->erase_completion_lock);
366 }
367
368 D1(printk(KERN_DEBUG "jffs2_do_reserve_space(): Giving 0x%x bytes at 0x%x\n", *len, *ofs));
369 return 0;
370}
371
372/**
373 * jffs2_add_physical_node_ref - add a physical node reference to the list
374 * @c: superblock info
375 * @new: new node reference to add
376 * @len: length of this physical node
377 * @dirty: dirty flag for new node
378 *
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000379 * Should only be used to report nodes for which space has been allocated
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 * by jffs2_reserve_space.
381 *
382 * Must be called with the alloc_sem held.
383 */
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000384
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385int jffs2_add_physical_node_ref(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *new)
386{
387 struct jffs2_eraseblock *jeb;
388 uint32_t len;
389
390 jeb = &c->blocks[new->flash_offset / c->sector_size];
391 len = ref_totlen(c, jeb, new);
392
393 D1(printk(KERN_DEBUG "jffs2_add_physical_node_ref(): Node at 0x%x(%d), size 0x%x\n", ref_offset(new), ref_flags(new), len));
394#if 1
Estelle Hammache3118db32005-01-24 21:30:25 +0000395 /* we could get some obsolete nodes after nextblock was refiled
396 in wbuf.c */
Estelle Hammache9b88f472005-01-28 18:53:05 +0000397 if ((c->nextblock || !ref_obsolete(new))
398 &&(jeb != c->nextblock || ref_offset(new) != jeb->offset + (c->sector_size - jeb->free_size))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 printk(KERN_WARNING "argh. node added in wrong place\n");
400 jffs2_free_raw_node_ref(new);
401 return -EINVAL;
402 }
403#endif
404 spin_lock(&c->erase_completion_lock);
405
406 if (!jeb->first_node)
407 jeb->first_node = new;
408 if (jeb->last_node)
409 jeb->last_node->next_phys = new;
410 jeb->last_node = new;
411
412 jeb->free_size -= len;
413 c->free_size -= len;
414 if (ref_obsolete(new)) {
415 jeb->dirty_size += len;
416 c->dirty_size += len;
417 } else {
418 jeb->used_size += len;
419 c->used_size += len;
420 }
421
Estelle Hammache9b88f472005-01-28 18:53:05 +0000422 if (!jeb->free_size && !jeb->dirty_size && !ISDIRTY(jeb->wasted_size)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 /* If it lives on the dirty_list, jffs2_reserve_space will put it there */
424 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
425 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
426 if (jffs2_wbuf_dirty(c)) {
427 /* Flush the last write in the block if it's outstanding */
428 spin_unlock(&c->erase_completion_lock);
429 jffs2_flush_wbuf_pad(c);
430 spin_lock(&c->erase_completion_lock);
431 }
432
433 list_add_tail(&jeb->list, &c->clean_list);
434 c->nextblock = NULL;
435 }
Artem B. Bityutskiye0c8e422005-07-24 16:14:17 +0100436 jffs2_dbg_acct_sanity_check_nolock(c,jeb);
437 jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438
439 spin_unlock(&c->erase_completion_lock);
440
441 return 0;
442}
443
444
445void jffs2_complete_reservation(struct jffs2_sb_info *c)
446{
447 D1(printk(KERN_DEBUG "jffs2_complete_reservation()\n"));
448 jffs2_garbage_collect_trigger(c);
449 up(&c->alloc_sem);
450}
451
452static inline int on_list(struct list_head *obj, struct list_head *head)
453{
454 struct list_head *this;
455
456 list_for_each(this, head) {
457 if (this == obj) {
458 D1(printk("%p is on list at %p\n", obj, head));
459 return 1;
460
461 }
462 }
463 return 0;
464}
465
466void jffs2_mark_node_obsolete(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref)
467{
468 struct jffs2_eraseblock *jeb;
469 int blocknr;
470 struct jffs2_unknown_node n;
471 int ret, addedsize;
472 size_t retlen;
473
474 if(!ref) {
475 printk(KERN_NOTICE "EEEEEK. jffs2_mark_node_obsolete called with NULL node\n");
476 return;
477 }
478 if (ref_obsolete(ref)) {
479 D1(printk(KERN_DEBUG "jffs2_mark_node_obsolete called with already obsolete node at 0x%08x\n", ref_offset(ref)));
480 return;
481 }
482 blocknr = ref->flash_offset / c->sector_size;
483 if (blocknr >= c->nr_blocks) {
484 printk(KERN_NOTICE "raw node at 0x%08x is off the end of device!\n", ref->flash_offset);
485 BUG();
486 }
487 jeb = &c->blocks[blocknr];
488
489 if (jffs2_can_mark_obsolete(c) && !jffs2_is_readonly(c) &&
Artem B. Bityuckiy31fbdf72005-02-28 08:21:09 +0000490 !(c->flags & (JFFS2_SB_FLAG_SCANNING | JFFS2_SB_FLAG_BUILDING))) {
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000491 /* Hm. This may confuse static lock analysis. If any of the above
492 three conditions is false, we're going to return from this
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 function without actually obliterating any nodes or freeing
494 any jffs2_raw_node_refs. So we don't need to stop erases from
495 happening, or protect against people holding an obsolete
496 jffs2_raw_node_ref without the erase_completion_lock. */
497 down(&c->erase_free_sem);
498 }
499
500 spin_lock(&c->erase_completion_lock);
501
502 if (ref_flags(ref) == REF_UNCHECKED) {
503 D1(if (unlikely(jeb->unchecked_size < ref_totlen(c, jeb, ref))) {
504 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",
505 ref_totlen(c, jeb, ref), blocknr, ref->flash_offset, jeb->used_size);
506 BUG();
507 })
508 D1(printk(KERN_DEBUG "Obsoleting previously unchecked node at 0x%08x of len %x: ", ref_offset(ref), ref_totlen(c, jeb, ref)));
509 jeb->unchecked_size -= ref_totlen(c, jeb, ref);
510 c->unchecked_size -= ref_totlen(c, jeb, ref);
511 } else {
512 D1(if (unlikely(jeb->used_size < ref_totlen(c, jeb, ref))) {
513 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",
514 ref_totlen(c, jeb, ref), blocknr, ref->flash_offset, jeb->used_size);
515 BUG();
516 })
Artem B. Bityutskiye0c8e422005-07-24 16:14:17 +0100517 D1(printk(KERN_DEBUG "Obsoleting node at 0x%08x of len %#x: ", ref_offset(ref), ref_totlen(c, jeb, ref)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 jeb->used_size -= ref_totlen(c, jeb, ref);
519 c->used_size -= ref_totlen(c, jeb, ref);
520 }
521
522 // Take care, that wasted size is taken into concern
523 if ((jeb->dirty_size || ISDIRTY(jeb->wasted_size + ref_totlen(c, jeb, ref))) && jeb != c->nextblock) {
Artem B. Bityuckiy6f401a402005-04-06 17:02:55 +0100524 D1(printk(KERN_DEBUG "Dirtying\n"));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 addedsize = ref_totlen(c, jeb, ref);
526 jeb->dirty_size += ref_totlen(c, jeb, ref);
527 c->dirty_size += ref_totlen(c, jeb, ref);
528
529 /* Convert wasted space to dirty, if not a bad block */
530 if (jeb->wasted_size) {
531 if (on_list(&jeb->list, &c->bad_used_list)) {
532 D1(printk(KERN_DEBUG "Leaving block at %08x on the bad_used_list\n",
533 jeb->offset));
534 addedsize = 0; /* To fool the refiling code later */
535 } else {
536 D1(printk(KERN_DEBUG "Converting %d bytes of wasted space to dirty in block at %08x\n",
537 jeb->wasted_size, jeb->offset));
538 addedsize += jeb->wasted_size;
539 jeb->dirty_size += jeb->wasted_size;
540 c->dirty_size += jeb->wasted_size;
541 c->wasted_size -= jeb->wasted_size;
542 jeb->wasted_size = 0;
543 }
544 }
545 } else {
Artem B. Bityuckiy6f401a402005-04-06 17:02:55 +0100546 D1(printk(KERN_DEBUG "Wasting\n"));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 addedsize = 0;
548 jeb->wasted_size += ref_totlen(c, jeb, ref);
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000549 c->wasted_size += ref_totlen(c, jeb, ref);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 }
551 ref->flash_offset = ref_offset(ref) | REF_OBSOLETE;
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000552
Artem B. Bityutskiye0c8e422005-07-24 16:14:17 +0100553 jffs2_dbg_acct_sanity_check_nolock(c, jeb);
554 jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555
Artem B. Bityuckiy31fbdf72005-02-28 08:21:09 +0000556 if (c->flags & JFFS2_SB_FLAG_SCANNING) {
557 /* Flash scanning is in progress. Don't muck about with the block
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 lists because they're not ready yet, and don't actually
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000559 obliterate nodes that look obsolete. If they weren't
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 marked obsolete on the flash at the time they _became_
561 obsolete, there was probably a reason for that. */
562 spin_unlock(&c->erase_completion_lock);
563 /* We didn't lock the erase_free_sem */
564 return;
565 }
566
567 if (jeb == c->nextblock) {
568 D2(printk(KERN_DEBUG "Not moving nextblock 0x%08x to dirty/erase_pending list\n", jeb->offset));
569 } else if (!jeb->used_size && !jeb->unchecked_size) {
570 if (jeb == c->gcblock) {
571 D1(printk(KERN_DEBUG "gcblock at 0x%08x completely dirtied. Clearing gcblock...\n", jeb->offset));
572 c->gcblock = NULL;
573 } else {
574 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x completely dirtied. Removing from (dirty?) list...\n", jeb->offset));
575 list_del(&jeb->list);
576 }
577 if (jffs2_wbuf_dirty(c)) {
578 D1(printk(KERN_DEBUG "...and adding to erasable_pending_wbuf_list\n"));
579 list_add_tail(&jeb->list, &c->erasable_pending_wbuf_list);
580 } else {
581 if (jiffies & 127) {
582 /* Most of the time, we just erase it immediately. Otherwise we
583 spend ages scanning it on mount, etc. */
584 D1(printk(KERN_DEBUG "...and adding to erase_pending_list\n"));
585 list_add_tail(&jeb->list, &c->erase_pending_list);
586 c->nr_erasing_blocks++;
587 jffs2_erase_pending_trigger(c);
588 } else {
589 /* Sometimes, however, we leave it elsewhere so it doesn't get
590 immediately reused, and we spread the load a bit. */
591 D1(printk(KERN_DEBUG "...and adding to erasable_list\n"));
592 list_add_tail(&jeb->list, &c->erasable_list);
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000593 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 }
595 D1(printk(KERN_DEBUG "Done OK\n"));
596 } else if (jeb == c->gcblock) {
597 D2(printk(KERN_DEBUG "Not moving gcblock 0x%08x to dirty_list\n", jeb->offset));
598 } else if (ISDIRTY(jeb->dirty_size) && !ISDIRTY(jeb->dirty_size - addedsize)) {
599 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x is freshly dirtied. Removing from clean list...\n", jeb->offset));
600 list_del(&jeb->list);
601 D1(printk(KERN_DEBUG "...and adding to dirty_list\n"));
602 list_add_tail(&jeb->list, &c->dirty_list);
603 } else if (VERYDIRTY(c, jeb->dirty_size) &&
604 !VERYDIRTY(c, jeb->dirty_size - addedsize)) {
605 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x is now very dirty. Removing from dirty list...\n", jeb->offset));
606 list_del(&jeb->list);
607 D1(printk(KERN_DEBUG "...and adding to very_dirty_list\n"));
608 list_add_tail(&jeb->list, &c->very_dirty_list);
609 } else {
610 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x not moved anywhere. (free 0x%08x, dirty 0x%08x, used 0x%08x)\n",
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000611 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
612 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613
614 spin_unlock(&c->erase_completion_lock);
615
Artem B. Bityuckiy31fbdf72005-02-28 08:21:09 +0000616 if (!jffs2_can_mark_obsolete(c) || jffs2_is_readonly(c) ||
617 (c->flags & JFFS2_SB_FLAG_BUILDING)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 /* We didn't lock the erase_free_sem */
619 return;
620 }
621
622 /* The erase_free_sem is locked, and has been since before we marked the node obsolete
623 and potentially put its eraseblock onto the erase_pending_list. Thus, we know that
624 the block hasn't _already_ been erased, and that 'ref' itself hasn't been freed yet
625 by jffs2_free_all_node_refs() in erase.c. Which is nice. */
626
627 D1(printk(KERN_DEBUG "obliterating obsoleted node at 0x%08x\n", ref_offset(ref)));
628 ret = jffs2_flash_read(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
629 if (ret) {
630 printk(KERN_WARNING "Read error reading from obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret);
631 goto out_erase_sem;
632 }
633 if (retlen != sizeof(n)) {
634 printk(KERN_WARNING "Short read from obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen);
635 goto out_erase_sem;
636 }
637 if (PAD(je32_to_cpu(n.totlen)) != PAD(ref_totlen(c, jeb, ref))) {
638 printk(KERN_WARNING "Node totlen on flash (0x%08x) != totlen from node ref (0x%08x)\n", je32_to_cpu(n.totlen), ref_totlen(c, jeb, ref));
639 goto out_erase_sem;
640 }
641 if (!(je16_to_cpu(n.nodetype) & JFFS2_NODE_ACCURATE)) {
642 D1(printk(KERN_DEBUG "Node at 0x%08x was already marked obsolete (nodetype 0x%04x)\n", ref_offset(ref), je16_to_cpu(n.nodetype)));
643 goto out_erase_sem;
644 }
645 /* XXX FIXME: This is ugly now */
646 n.nodetype = cpu_to_je16(je16_to_cpu(n.nodetype) & ~JFFS2_NODE_ACCURATE);
647 ret = jffs2_flash_write(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
648 if (ret) {
649 printk(KERN_WARNING "Write error in obliterating obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret);
650 goto out_erase_sem;
651 }
652 if (retlen != sizeof(n)) {
653 printk(KERN_WARNING "Short write in obliterating obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen);
654 goto out_erase_sem;
655 }
656
657 /* Nodes which have been marked obsolete no longer need to be
658 associated with any inode. Remove them from the per-inode list.
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000659
660 Note we can't do this for NAND at the moment because we need
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 obsolete dirent nodes to stay on the lists, because of the
662 horridness in jffs2_garbage_collect_deletion_dirent(). Also
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000663 because we delete the inocache, and on NAND we need that to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 stay around until all the nodes are actually erased, in order
665 to stop us from giving the same inode number to another newly
666 created inode. */
667 if (ref->next_in_ino) {
668 struct jffs2_inode_cache *ic;
669 struct jffs2_raw_node_ref **p;
670
671 spin_lock(&c->erase_completion_lock);
672
673 ic = jffs2_raw_ref_to_ic(ref);
674 for (p = &ic->nodes; (*p) != ref; p = &((*p)->next_in_ino))
675 ;
676
677 *p = ref->next_in_ino;
678 ref->next_in_ino = NULL;
679
Artem B. Bityuckiy437316d2005-03-20 17:46:23 +0000680 if (ic->nodes == (void *)ic && ic->nlink == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 jffs2_del_ino_cache(c, ic);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682
683 spin_unlock(&c->erase_completion_lock);
684 }
685
686
687 /* Merge with the next node in the physical list, if there is one
688 and if it's also obsolete and if it doesn't belong to any inode */
689 if (ref->next_phys && ref_obsolete(ref->next_phys) &&
690 !ref->next_phys->next_in_ino) {
691 struct jffs2_raw_node_ref *n = ref->next_phys;
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000692
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 spin_lock(&c->erase_completion_lock);
694
695 ref->__totlen += n->__totlen;
696 ref->next_phys = n->next_phys;
697 if (jeb->last_node == n) jeb->last_node = ref;
698 if (jeb->gc_node == n) {
699 /* gc will be happy continuing gc on this node */
700 jeb->gc_node=ref;
701 }
702 spin_unlock(&c->erase_completion_lock);
703
704 jffs2_free_raw_node_ref(n);
705 }
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000706
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 /* Also merge with the previous node in the list, if there is one
708 and that one is obsolete */
709 if (ref != jeb->first_node ) {
710 struct jffs2_raw_node_ref *p = jeb->first_node;
711
712 spin_lock(&c->erase_completion_lock);
713
714 while (p->next_phys != ref)
715 p = p->next_phys;
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000716
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 if (ref_obsolete(p) && !ref->next_in_ino) {
718 p->__totlen += ref->__totlen;
719 if (jeb->last_node == ref) {
720 jeb->last_node = p;
721 }
722 if (jeb->gc_node == ref) {
723 /* gc will be happy continuing gc on this node */
724 jeb->gc_node=p;
725 }
726 p->next_phys = ref->next_phys;
727 jffs2_free_raw_node_ref(ref);
728 }
729 spin_unlock(&c->erase_completion_lock);
730 }
731 out_erase_sem:
732 up(&c->erase_free_sem);
733}
734
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735int jffs2_thread_should_wake(struct jffs2_sb_info *c)
736{
737 int ret = 0;
738 uint32_t dirty;
739
740 if (c->unchecked_size) {
741 D1(printk(KERN_DEBUG "jffs2_thread_should_wake(): unchecked_size %d, checked_ino #%d\n",
742 c->unchecked_size, c->checked_ino));
743 return 1;
744 }
745
746 /* dirty_size contains blocks on erase_pending_list
747 * those blocks are counted in c->nr_erasing_blocks.
748 * If one block is actually erased, it is not longer counted as dirty_space
749 * but it is counted in c->nr_erasing_blocks, so we add it and subtract it
750 * with c->nr_erasing_blocks * c->sector_size again.
751 * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks
752 * This helps us to force gc and pick eventually a clean block to spread the load.
753 */
754 dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size;
755
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000756 if (c->nr_free_blocks + c->nr_erasing_blocks < c->resv_blocks_gctrigger &&
757 (dirty > c->nospc_dirty_size))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 ret = 1;
759
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000760 D1(printk(KERN_DEBUG "jffs2_thread_should_wake(): nr_free_blocks %d, nr_erasing_blocks %d, dirty_size 0x%x: %s\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 c->nr_free_blocks, c->nr_erasing_blocks, c->dirty_size, ret?"yes":"no"));
762
763 return ret;
764}