]> bbs.cooldavid.org Git - net-next-2.6.git/blame - fs/jffs2/nodemgmt.c
xps: Transmit Packet Steering
[net-next-2.6.git] / fs / jffs2 / nodemgmt.c
CommitLineData
1da177e4
LT
1/*
2 * JFFS2 -- Journalling Flash File System, Version 2.
3 *
c00c310e 4 * Copyright © 2001-2007 Red Hat, Inc.
1da177e4
LT
5 *
6 * Created by David Woodhouse <dwmw2@infradead.org>
7 *
8 * For licensing information, see the file 'LICENCE' in this directory.
9 *
1da177e4
LT
10 */
11
12#include <linux/kernel.h>
1da177e4
LT
13#include <linux/mtd/mtd.h>
14#include <linux/compiler.h>
15#include <linux/sched.h> /* For cond_resched() */
16#include "nodelist.h"
e631ddba 17#include "debug.h"
1da177e4
LT
18
19/**
20 * jffs2_reserve_space - request physical space to write nodes to flash
21 * @c: superblock info
22 * @minsize: Minimum acceptable size of allocation
1da177e4
LT
23 * @len: Returned value of allocation length
24 * @prio: Allocation type - ALLOC_{NORMAL,DELETION}
25 *
26 * Requests a block of physical space on the flash. Returns zero for success
9fe4854c
DW
27 * and puts 'len' into the appropriate place, or returns -ENOSPC or other
28 * error if appropriate. Doesn't return len since that's
1da177e4
LT
29 *
30 * If it returns zero, jffs2_reserve_space() also downs the per-filesystem
31 * allocation semaphore, to prevent more than one allocation from being
32 * active at any time. The semaphore is later released by jffs2_commit_allocation()
33 *
34 * jffs2_reserve_space() may trigger garbage collection in order to make room
35 * for the requested allocation.
36 */
37
e631ddba 38static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
9fe4854c 39 uint32_t *len, uint32_t sumsize);
1da177e4 40
9fe4854c 41int jffs2_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
e631ddba 42 uint32_t *len, int prio, uint32_t sumsize)
1da177e4
LT
43{
44 int ret = -EAGAIN;
45 int blocksneeded = c->resv_blocks_write;
46 /* align it */
47 minsize = PAD(minsize);
48
49 D1(printk(KERN_DEBUG "jffs2_reserve_space(): Requested 0x%x bytes\n", minsize));
ced22070 50 mutex_lock(&c->alloc_sem);
1da177e4
LT
51
52 D1(printk(KERN_DEBUG "jffs2_reserve_space(): alloc sem got\n"));
53
54 spin_lock(&c->erase_completion_lock);
55
56 /* this needs a little more thought (true <tglx> :)) */
57 while(ret == -EAGAIN) {
58 while(c->nr_free_blocks + c->nr_erasing_blocks < blocksneeded) {
1da177e4
LT
59 uint32_t dirty, avail;
60
61 /* calculate real dirty size
62 * dirty_size contains blocks on erase_pending_list
63 * those blocks are counted in c->nr_erasing_blocks.
64 * If one block is actually erased, it is not longer counted as dirty_space
65 * but it is counted in c->nr_erasing_blocks, so we add it and subtract it
66 * with c->nr_erasing_blocks * c->sector_size again.
67 * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks
68 * This helps us to force gc and pick eventually a clean block to spread the load.
69 * We add unchecked_size here, as we hopefully will find some space to use.
70 * This will affect the sum only once, as gc first finishes checking
71 * of nodes.
72 */
73 dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size + c->unchecked_size;
74 if (dirty < c->nospc_dirty_size) {
75 if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) {
4132ace8 76 D1(printk(KERN_NOTICE "jffs2_reserve_space(): Low on dirty space to GC, but it's a deletion. Allowing...\n"));
1da177e4
LT
77 break;
78 }
79 D1(printk(KERN_DEBUG "dirty size 0x%08x + unchecked_size 0x%08x < nospc_dirty_size 0x%08x, returning -ENOSPC\n",
80 dirty, c->unchecked_size, c->sector_size));
81
82 spin_unlock(&c->erase_completion_lock);
ced22070 83 mutex_unlock(&c->alloc_sem);
1da177e4
LT
84 return -ENOSPC;
85 }
182ec4ee 86
1da177e4
LT
87 /* Calc possibly available space. Possibly available means that we
88 * don't know, if unchecked size contains obsoleted nodes, which could give us some
89 * more usable space. This will affect the sum only once, as gc first finishes checking
90 * of nodes.
182ec4ee 91 + Return -ENOSPC, if the maximum possibly available space is less or equal than
1da177e4
LT
92 * blocksneeded * sector_size.
93 * This blocks endless gc looping on a filesystem, which is nearly full, even if
94 * the check above passes.
95 */
96 avail = c->free_size + c->dirty_size + c->erasing_size + c->unchecked_size;
97 if ( (avail / c->sector_size) <= blocksneeded) {
98 if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) {
4132ace8 99 D1(printk(KERN_NOTICE "jffs2_reserve_space(): Low on possibly available space, but it's a deletion. Allowing...\n"));
1da177e4
LT
100 break;
101 }
102
103 D1(printk(KERN_DEBUG "max. available size 0x%08x < blocksneeded * sector_size 0x%08x, returning -ENOSPC\n",
104 avail, blocksneeded * c->sector_size));
105 spin_unlock(&c->erase_completion_lock);
ced22070 106 mutex_unlock(&c->alloc_sem);
1da177e4
LT
107 return -ENOSPC;
108 }
109
ced22070 110 mutex_unlock(&c->alloc_sem);
1da177e4
LT
111
112 D1(printk(KERN_DEBUG "Triggering GC pass. nr_free_blocks %d, nr_erasing_blocks %d, free_size 0x%08x, dirty_size 0x%08x, wasted_size 0x%08x, used_size 0x%08x, erasing_size 0x%08x, bad_size 0x%08x (total 0x%08x of 0x%08x)\n",
113 c->nr_free_blocks, c->nr_erasing_blocks, c->free_size, c->dirty_size, c->wasted_size, c->used_size, c->erasing_size, c->bad_size,
114 c->free_size + c->dirty_size + c->wasted_size + c->used_size + c->erasing_size + c->bad_size, c->flash_size));
115 spin_unlock(&c->erase_completion_lock);
182ec4ee 116
1da177e4 117 ret = jffs2_garbage_collect_pass(c);
422b1202 118
0717bf84
DW
119 if (ret == -EAGAIN) {
120 spin_lock(&c->erase_completion_lock);
121 if (c->nr_erasing_blocks &&
122 list_empty(&c->erase_pending_list) &&
123 list_empty(&c->erase_complete_list)) {
124 DECLARE_WAITQUEUE(wait, current);
125 set_current_state(TASK_UNINTERRUPTIBLE);
126 add_wait_queue(&c->erase_wait, &wait);
127 D1(printk(KERN_DEBUG "%s waiting for erase to complete\n", __func__));
128 spin_unlock(&c->erase_completion_lock);
129
130 schedule();
131 } else
132 spin_unlock(&c->erase_completion_lock);
133 } else if (ret)
1da177e4
LT
134 return ret;
135
136 cond_resched();
137
138 if (signal_pending(current))
139 return -EINTR;
140
ced22070 141 mutex_lock(&c->alloc_sem);
1da177e4
LT
142 spin_lock(&c->erase_completion_lock);
143 }
144
9fe4854c 145 ret = jffs2_do_reserve_space(c, minsize, len, sumsize);
1da177e4
LT
146 if (ret) {
147 D1(printk(KERN_DEBUG "jffs2_reserve_space: ret is %d\n", ret));
148 }
149 }
150 spin_unlock(&c->erase_completion_lock);
2f785402 151 if (!ret)
046b8b98 152 ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1);
1da177e4 153 if (ret)
ced22070 154 mutex_unlock(&c->alloc_sem);
1da177e4
LT
155 return ret;
156}
157
9fe4854c
DW
158int jffs2_reserve_space_gc(struct jffs2_sb_info *c, uint32_t minsize,
159 uint32_t *len, uint32_t sumsize)
1da177e4
LT
160{
161 int ret = -EAGAIN;
162 minsize = PAD(minsize);
163
164 D1(printk(KERN_DEBUG "jffs2_reserve_space_gc(): Requested 0x%x bytes\n", minsize));
165
166 spin_lock(&c->erase_completion_lock);
167 while(ret == -EAGAIN) {
9fe4854c 168 ret = jffs2_do_reserve_space(c, minsize, len, sumsize);
1da177e4 169 if (ret) {
ef53cb02 170 D1(printk(KERN_DEBUG "jffs2_reserve_space_gc: looping, ret is %d\n", ret));
1da177e4
LT
171 }
172 }
173 spin_unlock(&c->erase_completion_lock);
2f785402 174 if (!ret)
046b8b98 175 ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1);
2f785402 176
1da177e4
LT
177 return ret;
178}
179
e631ddba
FH
180
181/* Classify nextblock (clean, dirty of verydirty) and force to select an other one */
182
183static void jffs2_close_nextblock(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
1da177e4 184{
e631ddba 185
99c2594f
AH
186 if (c->nextblock == NULL) {
187 D1(printk(KERN_DEBUG "jffs2_close_nextblock: Erase block at 0x%08x has already been placed in a list\n",
188 jeb->offset));
189 return;
190 }
e631ddba
FH
191 /* Check, if we have a dirty block now, or if it was dirty already */
192 if (ISDIRTY (jeb->wasted_size + jeb->dirty_size)) {
193 c->dirty_size += jeb->wasted_size;
194 c->wasted_size -= jeb->wasted_size;
195 jeb->dirty_size += jeb->wasted_size;
196 jeb->wasted_size = 0;
197 if (VERYDIRTY(c, jeb->dirty_size)) {
198 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to very_dirty_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
199 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
200 list_add_tail(&jeb->list, &c->very_dirty_list);
201 } else {
202 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to dirty_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
203 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
204 list_add_tail(&jeb->list, &c->dirty_list);
205 }
182ec4ee 206 } else {
e631ddba
FH
207 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
208 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
209 list_add_tail(&jeb->list, &c->clean_list);
210 }
211 c->nextblock = NULL;
212
213}
214
215/* Select a new jeb for nextblock */
216
217static int jffs2_find_nextblock(struct jffs2_sb_info *c)
218{
219 struct list_head *next;
182ec4ee 220
e631ddba
FH
221 /* Take the next block off the 'free' list */
222
223 if (list_empty(&c->free_list)) {
224
225 if (!c->nr_erasing_blocks &&
226 !list_empty(&c->erasable_list)) {
227 struct jffs2_eraseblock *ejeb;
228
229 ejeb = list_entry(c->erasable_list.next, struct jffs2_eraseblock, list);
f116629d 230 list_move_tail(&ejeb->list, &c->erase_pending_list);
e631ddba 231 c->nr_erasing_blocks++;
ae3b6ba0 232 jffs2_garbage_collect_trigger(c);
e631ddba
FH
233 D1(printk(KERN_DEBUG "jffs2_find_nextblock: Triggering erase of erasable block at 0x%08x\n",
234 ejeb->offset));
235 }
236
237 if (!c->nr_erasing_blocks &&
238 !list_empty(&c->erasable_pending_wbuf_list)) {
239 D1(printk(KERN_DEBUG "jffs2_find_nextblock: Flushing write buffer\n"));
240 /* c->nextblock is NULL, no update to c->nextblock allowed */
1da177e4 241 spin_unlock(&c->erase_completion_lock);
1da177e4
LT
242 jffs2_flush_wbuf_pad(c);
243 spin_lock(&c->erase_completion_lock);
e631ddba
FH
244 /* Have another go. It'll be on the erasable_list now */
245 return -EAGAIN;
1da177e4 246 }
e631ddba
FH
247
248 if (!c->nr_erasing_blocks) {
249 /* Ouch. We're in GC, or we wouldn't have got here.
250 And there's no space left. At all. */
182ec4ee
TG
251 printk(KERN_CRIT "Argh. No free space left for GC. nr_erasing_blocks is %d. nr_free_blocks is %d. (erasableempty: %s, erasingempty: %s, erasependingempty: %s)\n",
252 c->nr_erasing_blocks, c->nr_free_blocks, list_empty(&c->erasable_list)?"yes":"no",
e631ddba
FH
253 list_empty(&c->erasing_list)?"yes":"no", list_empty(&c->erase_pending_list)?"yes":"no");
254 return -ENOSPC;
1da177e4 255 }
e631ddba
FH
256
257 spin_unlock(&c->erase_completion_lock);
258 /* Don't wait for it; just erase one right now */
259 jffs2_erase_pending_blocks(c, 1);
260 spin_lock(&c->erase_completion_lock);
261
262 /* An erase may have failed, decreasing the
263 amount of free space available. So we must
264 restart from the beginning */
265 return -EAGAIN;
1da177e4 266 }
e631ddba
FH
267
268 next = c->free_list.next;
269 list_del(next);
270 c->nextblock = list_entry(next, struct jffs2_eraseblock, list);
271 c->nr_free_blocks--;
182ec4ee 272
e631ddba
FH
273 jffs2_sum_reset_collected(c->summary); /* reset collected summary */
274
f04de505 275#ifdef CONFIG_JFFS2_FS_WRITEBUFFER
5bf17237
AB
276 /* adjust write buffer offset, else we get a non contiguous write bug */
277 if (!(c->wbuf_ofs % c->sector_size) && !c->wbuf_len)
278 c->wbuf_ofs = 0xffffffff;
f04de505 279#endif
5bf17237 280
e631ddba
FH
281 D1(printk(KERN_DEBUG "jffs2_find_nextblock(): new nextblock = 0x%08x\n", c->nextblock->offset));
282
283 return 0;
284}
285
286/* Called with alloc sem _and_ erase_completion_lock */
9fe4854c
DW
287static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
288 uint32_t *len, uint32_t sumsize)
e631ddba
FH
289{
290 struct jffs2_eraseblock *jeb = c->nextblock;
9fe4854c 291 uint32_t reserved_size; /* for summary information at the end of the jeb */
e631ddba
FH
292 int ret;
293
294 restart:
295 reserved_size = 0;
296
297 if (jffs2_sum_active() && (sumsize != JFFS2_SUMMARY_NOSUM_SIZE)) {
298 /* NOSUM_SIZE means not to generate summary */
299
300 if (jeb) {
301 reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE);
733802d9 302 dbg_summary("minsize=%d , jeb->free=%d ,"
e631ddba
FH
303 "summary->size=%d , sumsize=%d\n",
304 minsize, jeb->free_size,
305 c->summary->sum_size, sumsize);
306 }
307
308 /* Is there enough space for writing out the current node, or we have to
309 write out summary information now, close this jeb and select new nextblock? */
310 if (jeb && (PAD(minsize) + PAD(c->summary->sum_size + sumsize +
311 JFFS2_SUMMARY_FRAME_SIZE) > jeb->free_size)) {
312
313 /* Has summary been disabled for this jeb? */
314 if (jffs2_sum_is_disabled(c->summary)) {
315 sumsize = JFFS2_SUMMARY_NOSUM_SIZE;
316 goto restart;
1da177e4
LT
317 }
318
e631ddba 319 /* Writing out the collected summary information */
733802d9 320 dbg_summary("generating summary for 0x%08x.\n", jeb->offset);
e631ddba
FH
321 ret = jffs2_sum_write_sumnode(c);
322
323 if (ret)
324 return ret;
325
326 if (jffs2_sum_is_disabled(c->summary)) {
327 /* jffs2_write_sumnode() couldn't write out the summary information
328 diabling summary for this jeb and free the collected information
329 */
330 sumsize = JFFS2_SUMMARY_NOSUM_SIZE;
331 goto restart;
332 }
333
334 jffs2_close_nextblock(c, jeb);
335 jeb = NULL;
34c0e906
FH
336 /* keep always valid value in reserved_size */
337 reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE);
e631ddba
FH
338 }
339 } else {
340 if (jeb && minsize > jeb->free_size) {
fc6612f6
DW
341 uint32_t waste;
342
e631ddba
FH
343 /* Skip the end of this block and file it as having some dirty space */
344 /* If there's a pending write to it, flush now */
345
346 if (jffs2_wbuf_dirty(c)) {
1da177e4 347 spin_unlock(&c->erase_completion_lock);
e631ddba 348 D1(printk(KERN_DEBUG "jffs2_do_reserve_space: Flushing write buffer\n"));
1da177e4
LT
349 jffs2_flush_wbuf_pad(c);
350 spin_lock(&c->erase_completion_lock);
e631ddba
FH
351 jeb = c->nextblock;
352 goto restart;
1da177e4
LT
353 }
354
fc6612f6
DW
355 spin_unlock(&c->erase_completion_lock);
356
357 ret = jffs2_prealloc_raw_node_refs(c, jeb, 1);
358 if (ret)
359 return ret;
360 /* Just lock it again and continue. Nothing much can change because
361 we hold c->alloc_sem anyway. In fact, it's not entirely clear why
362 we hold c->erase_completion_lock in the majority of this function...
363 but that's a question for another (more caffeine-rich) day. */
364 spin_lock(&c->erase_completion_lock);
365
366 waste = jeb->free_size;
367 jffs2_link_node_ref(c, jeb,
368 (jeb->offset + c->sector_size - waste) | REF_OBSOLETE,
369 waste, NULL);
370 /* FIXME: that made it count as dirty. Convert to wasted */
371 jeb->dirty_size -= waste;
372 c->dirty_size -= waste;
373 jeb->wasted_size += waste;
374 c->wasted_size += waste;
1da177e4 375
e631ddba
FH
376 jffs2_close_nextblock(c, jeb);
377 jeb = NULL;
1da177e4 378 }
e631ddba
FH
379 }
380
381 if (!jeb) {
382
383 ret = jffs2_find_nextblock(c);
384 if (ret)
385 return ret;
1da177e4 386
e631ddba 387 jeb = c->nextblock;
1da177e4
LT
388
389 if (jeb->free_size != c->sector_size - c->cleanmarker_size) {
390 printk(KERN_WARNING "Eep. Block 0x%08x taken from free_list had free_size of 0x%08x!!\n", jeb->offset, jeb->free_size);
391 goto restart;
392 }
393 }
394 /* OK, jeb (==c->nextblock) is now pointing at a block which definitely has
395 enough space */
e631ddba 396 *len = jeb->free_size - reserved_size;
1da177e4
LT
397
398 if (c->cleanmarker_size && jeb->used_size == c->cleanmarker_size &&
399 !jeb->first_node->next_in_ino) {
182ec4ee 400 /* Only node in it beforehand was a CLEANMARKER node (we think).
1da177e4 401 So mark it obsolete now that there's going to be another node
182ec4ee 402 in the block. This will reduce used_size to zero but We've
1da177e4
LT
403 already set c->nextblock so that jffs2_mark_node_obsolete()
404 won't try to refile it to the dirty_list.
405 */
406 spin_unlock(&c->erase_completion_lock);
407 jffs2_mark_node_obsolete(c, jeb->first_node);
408 spin_lock(&c->erase_completion_lock);
409 }
410
9fe4854c
DW
411 D1(printk(KERN_DEBUG "jffs2_do_reserve_space(): Giving 0x%x bytes at 0x%x\n",
412 *len, jeb->offset + (c->sector_size - jeb->free_size)));
1da177e4
LT
413 return 0;
414}
415
416/**
417 * jffs2_add_physical_node_ref - add a physical node reference to the list
418 * @c: superblock info
419 * @new: new node reference to add
420 * @len: length of this physical node
1da177e4 421 *
182ec4ee 422 * Should only be used to report nodes for which space has been allocated
1da177e4
LT
423 * by jffs2_reserve_space.
424 *
425 * Must be called with the alloc_sem held.
426 */
182ec4ee 427
2f785402
DW
428struct jffs2_raw_node_ref *jffs2_add_physical_node_ref(struct jffs2_sb_info *c,
429 uint32_t ofs, uint32_t len,
430 struct jffs2_inode_cache *ic)
1da177e4
LT
431{
432 struct jffs2_eraseblock *jeb;
2f785402 433 struct jffs2_raw_node_ref *new;
1da177e4 434
2f785402 435 jeb = &c->blocks[ofs / c->sector_size];
1da177e4 436
2f785402
DW
437 D1(printk(KERN_DEBUG "jffs2_add_physical_node_ref(): Node at 0x%x(%d), size 0x%x\n",
438 ofs & ~3, ofs & 3, len));
1da177e4 439#if 1
2f785402
DW
440 /* Allow non-obsolete nodes only to be added at the end of c->nextblock,
441 if c->nextblock is set. Note that wbuf.c will file obsolete nodes
442 even after refiling c->nextblock */
443 if ((c->nextblock || ((ofs & 3) != REF_OBSOLETE))
444 && (jeb != c->nextblock || (ofs & ~3) != jeb->offset + (c->sector_size - jeb->free_size))) {
66bfaeaa
DW
445 printk(KERN_WARNING "argh. node added in wrong place at 0x%08x(%d)\n", ofs & ~3, ofs & 3);
446 if (c->nextblock)
447 printk(KERN_WARNING "nextblock 0x%08x", c->nextblock->offset);
448 else
449 printk(KERN_WARNING "No nextblock");
450 printk(", expected at %08x\n", jeb->offset + (c->sector_size - jeb->free_size));
2f785402 451 return ERR_PTR(-EINVAL);
1da177e4
LT
452 }
453#endif
454 spin_lock(&c->erase_completion_lock);
455
2f785402 456 new = jffs2_link_node_ref(c, jeb, ofs, len, ic);
1da177e4 457
9b88f473 458 if (!jeb->free_size && !jeb->dirty_size && !ISDIRTY(jeb->wasted_size)) {
1da177e4
LT
459 /* If it lives on the dirty_list, jffs2_reserve_space will put it there */
460 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
461 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
462 if (jffs2_wbuf_dirty(c)) {
463 /* Flush the last write in the block if it's outstanding */
464 spin_unlock(&c->erase_completion_lock);
465 jffs2_flush_wbuf_pad(c);
466 spin_lock(&c->erase_completion_lock);
467 }
468
469 list_add_tail(&jeb->list, &c->clean_list);
470 c->nextblock = NULL;
471 }
e0c8e42f
AB
472 jffs2_dbg_acct_sanity_check_nolock(c,jeb);
473 jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
1da177e4
LT
474
475 spin_unlock(&c->erase_completion_lock);
476
2f785402 477 return new;
1da177e4
LT
478}
479
480
481void jffs2_complete_reservation(struct jffs2_sb_info *c)
482{
483 D1(printk(KERN_DEBUG "jffs2_complete_reservation()\n"));
acb64a43 484 spin_lock(&c->erase_completion_lock);
1da177e4 485 jffs2_garbage_collect_trigger(c);
acb64a43 486 spin_unlock(&c->erase_completion_lock);
ced22070 487 mutex_unlock(&c->alloc_sem);
1da177e4
LT
488}
489
490static inline int on_list(struct list_head *obj, struct list_head *head)
491{
492 struct list_head *this;
493
494 list_for_each(this, head) {
495 if (this == obj) {
496 D1(printk("%p is on list at %p\n", obj, head));
497 return 1;
498
499 }
500 }
501 return 0;
502}
503
504void jffs2_mark_node_obsolete(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref)
505{
506 struct jffs2_eraseblock *jeb;
507 int blocknr;
508 struct jffs2_unknown_node n;
509 int ret, addedsize;
510 size_t retlen;
1417fc44 511 uint32_t freed_len;
1da177e4 512
9bfeb691 513 if(unlikely(!ref)) {
1da177e4
LT
514 printk(KERN_NOTICE "EEEEEK. jffs2_mark_node_obsolete called with NULL node\n");
515 return;
516 }
517 if (ref_obsolete(ref)) {
518 D1(printk(KERN_DEBUG "jffs2_mark_node_obsolete called with already obsolete node at 0x%08x\n", ref_offset(ref)));
519 return;
520 }
521 blocknr = ref->flash_offset / c->sector_size;
522 if (blocknr >= c->nr_blocks) {
523 printk(KERN_NOTICE "raw node at 0x%08x is off the end of device!\n", ref->flash_offset);
524 BUG();
525 }
526 jeb = &c->blocks[blocknr];
527
528 if (jffs2_can_mark_obsolete(c) && !jffs2_is_readonly(c) &&
31fbdf7a 529 !(c->flags & (JFFS2_SB_FLAG_SCANNING | JFFS2_SB_FLAG_BUILDING))) {
182ec4ee
TG
530 /* Hm. This may confuse static lock analysis. If any of the above
531 three conditions is false, we're going to return from this
1da177e4
LT
532 function without actually obliterating any nodes or freeing
533 any jffs2_raw_node_refs. So we don't need to stop erases from
534 happening, or protect against people holding an obsolete
535 jffs2_raw_node_ref without the erase_completion_lock. */
ced22070 536 mutex_lock(&c->erase_free_sem);
1da177e4
LT
537 }
538
539 spin_lock(&c->erase_completion_lock);
540
1417fc44
DW
541 freed_len = ref_totlen(c, jeb, ref);
542
1da177e4 543 if (ref_flags(ref) == REF_UNCHECKED) {
1417fc44 544 D1(if (unlikely(jeb->unchecked_size < freed_len)) {
1da177e4 545 printk(KERN_NOTICE "raw unchecked node of size 0x%08x freed from erase block %d at 0x%08x, but unchecked_size was already 0x%08x\n",
1417fc44 546 freed_len, blocknr, ref->flash_offset, jeb->used_size);
1da177e4
LT
547 BUG();
548 })
1417fc44
DW
549 D1(printk(KERN_DEBUG "Obsoleting previously unchecked node at 0x%08x of len %x: ", ref_offset(ref), freed_len));
550 jeb->unchecked_size -= freed_len;
551 c->unchecked_size -= freed_len;
1da177e4 552 } else {
1417fc44 553 D1(if (unlikely(jeb->used_size < freed_len)) {
1da177e4 554 printk(KERN_NOTICE "raw node of size 0x%08x freed from erase block %d at 0x%08x, but used_size was already 0x%08x\n",
1417fc44 555 freed_len, blocknr, ref->flash_offset, jeb->used_size);
1da177e4
LT
556 BUG();
557 })
1417fc44
DW
558 D1(printk(KERN_DEBUG "Obsoleting node at 0x%08x of len %#x: ", ref_offset(ref), freed_len));
559 jeb->used_size -= freed_len;
560 c->used_size -= freed_len;
1da177e4
LT
561 }
562
563 // Take care, that wasted size is taken into concern
1417fc44 564 if ((jeb->dirty_size || ISDIRTY(jeb->wasted_size + freed_len)) && jeb != c->nextblock) {
c7c16c8e 565 D1(printk("Dirtying\n"));
1417fc44
DW
566 addedsize = freed_len;
567 jeb->dirty_size += freed_len;
568 c->dirty_size += freed_len;
1da177e4
LT
569
570 /* Convert wasted space to dirty, if not a bad block */
571 if (jeb->wasted_size) {
572 if (on_list(&jeb->list, &c->bad_used_list)) {
573 D1(printk(KERN_DEBUG "Leaving block at %08x on the bad_used_list\n",
574 jeb->offset));
575 addedsize = 0; /* To fool the refiling code later */
576 } else {
577 D1(printk(KERN_DEBUG "Converting %d bytes of wasted space to dirty in block at %08x\n",
578 jeb->wasted_size, jeb->offset));
579 addedsize += jeb->wasted_size;
580 jeb->dirty_size += jeb->wasted_size;
581 c->dirty_size += jeb->wasted_size;
582 c->wasted_size -= jeb->wasted_size;
583 jeb->wasted_size = 0;
584 }
585 }
586 } else {
c7c16c8e 587 D1(printk("Wasting\n"));
1da177e4 588 addedsize = 0;
1417fc44
DW
589 jeb->wasted_size += freed_len;
590 c->wasted_size += freed_len;
1da177e4
LT
591 }
592 ref->flash_offset = ref_offset(ref) | REF_OBSOLETE;
182ec4ee 593
e0c8e42f
AB
594 jffs2_dbg_acct_sanity_check_nolock(c, jeb);
595 jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
1da177e4 596
31fbdf7a
AB
597 if (c->flags & JFFS2_SB_FLAG_SCANNING) {
598 /* Flash scanning is in progress. Don't muck about with the block
1da177e4 599 lists because they're not ready yet, and don't actually
182ec4ee 600 obliterate nodes that look obsolete. If they weren't
1da177e4
LT
601 marked obsolete on the flash at the time they _became_
602 obsolete, there was probably a reason for that. */
603 spin_unlock(&c->erase_completion_lock);
604 /* We didn't lock the erase_free_sem */
605 return;
606 }
607
608 if (jeb == c->nextblock) {
609 D2(printk(KERN_DEBUG "Not moving nextblock 0x%08x to dirty/erase_pending list\n", jeb->offset));
610 } else if (!jeb->used_size && !jeb->unchecked_size) {
611 if (jeb == c->gcblock) {
612 D1(printk(KERN_DEBUG "gcblock at 0x%08x completely dirtied. Clearing gcblock...\n", jeb->offset));
613 c->gcblock = NULL;
614 } else {
615 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x completely dirtied. Removing from (dirty?) list...\n", jeb->offset));
616 list_del(&jeb->list);
617 }
618 if (jffs2_wbuf_dirty(c)) {
619 D1(printk(KERN_DEBUG "...and adding to erasable_pending_wbuf_list\n"));
620 list_add_tail(&jeb->list, &c->erasable_pending_wbuf_list);
621 } else {
622 if (jiffies & 127) {
623 /* Most of the time, we just erase it immediately. Otherwise we
624 spend ages scanning it on mount, etc. */
625 D1(printk(KERN_DEBUG "...and adding to erase_pending_list\n"));
626 list_add_tail(&jeb->list, &c->erase_pending_list);
627 c->nr_erasing_blocks++;
ae3b6ba0 628 jffs2_garbage_collect_trigger(c);
1da177e4
LT
629 } else {
630 /* Sometimes, however, we leave it elsewhere so it doesn't get
631 immediately reused, and we spread the load a bit. */
632 D1(printk(KERN_DEBUG "...and adding to erasable_list\n"));
633 list_add_tail(&jeb->list, &c->erasable_list);
182ec4ee 634 }
1da177e4
LT
635 }
636 D1(printk(KERN_DEBUG "Done OK\n"));
637 } else if (jeb == c->gcblock) {
638 D2(printk(KERN_DEBUG "Not moving gcblock 0x%08x to dirty_list\n", jeb->offset));
639 } else if (ISDIRTY(jeb->dirty_size) && !ISDIRTY(jeb->dirty_size - addedsize)) {
640 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x is freshly dirtied. Removing from clean list...\n", jeb->offset));
641 list_del(&jeb->list);
642 D1(printk(KERN_DEBUG "...and adding to dirty_list\n"));
643 list_add_tail(&jeb->list, &c->dirty_list);
644 } else if (VERYDIRTY(c, jeb->dirty_size) &&
645 !VERYDIRTY(c, jeb->dirty_size - addedsize)) {
646 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x is now very dirty. Removing from dirty list...\n", jeb->offset));
647 list_del(&jeb->list);
648 D1(printk(KERN_DEBUG "...and adding to very_dirty_list\n"));
649 list_add_tail(&jeb->list, &c->very_dirty_list);
650 } else {
651 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x not moved anywhere. (free 0x%08x, dirty 0x%08x, used 0x%08x)\n",
182ec4ee
TG
652 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
653 }
1da177e4
LT
654
655 spin_unlock(&c->erase_completion_lock);
656
31fbdf7a
AB
657 if (!jffs2_can_mark_obsolete(c) || jffs2_is_readonly(c) ||
658 (c->flags & JFFS2_SB_FLAG_BUILDING)) {
1da177e4
LT
659 /* We didn't lock the erase_free_sem */
660 return;
661 }
662
663 /* The erase_free_sem is locked, and has been since before we marked the node obsolete
664 and potentially put its eraseblock onto the erase_pending_list. Thus, we know that
665 the block hasn't _already_ been erased, and that 'ref' itself hasn't been freed yet
c38c1b61 666 by jffs2_free_jeb_node_refs() in erase.c. Which is nice. */
1da177e4
LT
667
668 D1(printk(KERN_DEBUG "obliterating obsoleted node at 0x%08x\n", ref_offset(ref)));
669 ret = jffs2_flash_read(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
670 if (ret) {
671 printk(KERN_WARNING "Read error reading from obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret);
672 goto out_erase_sem;
673 }
674 if (retlen != sizeof(n)) {
675 printk(KERN_WARNING "Short read from obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen);
676 goto out_erase_sem;
677 }
1417fc44
DW
678 if (PAD(je32_to_cpu(n.totlen)) != PAD(freed_len)) {
679 printk(KERN_WARNING "Node totlen on flash (0x%08x) != totlen from node ref (0x%08x)\n", je32_to_cpu(n.totlen), freed_len);
1da177e4
LT
680 goto out_erase_sem;
681 }
682 if (!(je16_to_cpu(n.nodetype) & JFFS2_NODE_ACCURATE)) {
683 D1(printk(KERN_DEBUG "Node at 0x%08x was already marked obsolete (nodetype 0x%04x)\n", ref_offset(ref), je16_to_cpu(n.nodetype)));
684 goto out_erase_sem;
685 }
686 /* XXX FIXME: This is ugly now */
687 n.nodetype = cpu_to_je16(je16_to_cpu(n.nodetype) & ~JFFS2_NODE_ACCURATE);
688 ret = jffs2_flash_write(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
689 if (ret) {
690 printk(KERN_WARNING "Write error in obliterating obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret);
691 goto out_erase_sem;
692 }
693 if (retlen != sizeof(n)) {
694 printk(KERN_WARNING "Short write in obliterating obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen);
695 goto out_erase_sem;
696 }
697
698 /* Nodes which have been marked obsolete no longer need to be
699 associated with any inode. Remove them from the per-inode list.
182ec4ee
TG
700
701 Note we can't do this for NAND at the moment because we need
1da177e4
LT
702 obsolete dirent nodes to stay on the lists, because of the
703 horridness in jffs2_garbage_collect_deletion_dirent(). Also
182ec4ee 704 because we delete the inocache, and on NAND we need that to
1da177e4
LT
705 stay around until all the nodes are actually erased, in order
706 to stop us from giving the same inode number to another newly
707 created inode. */
708 if (ref->next_in_ino) {
709 struct jffs2_inode_cache *ic;
710 struct jffs2_raw_node_ref **p;
711
712 spin_lock(&c->erase_completion_lock);
713
714 ic = jffs2_raw_ref_to_ic(ref);
715 for (p = &ic->nodes; (*p) != ref; p = &((*p)->next_in_ino))
716 ;
717
718 *p = ref->next_in_ino;
719 ref->next_in_ino = NULL;
720
c9f700f8
KK
721 switch (ic->class) {
722#ifdef CONFIG_JFFS2_FS_XATTR
723 case RAWNODE_CLASS_XATTR_DATUM:
724 jffs2_release_xattr_datum(c, (struct jffs2_xattr_datum *)ic);
725 break;
726 case RAWNODE_CLASS_XATTR_REF:
727 jffs2_release_xattr_ref(c, (struct jffs2_xattr_ref *)ic);
728 break;
729#endif
730 default:
27c72b04 731 if (ic->nodes == (void *)ic && ic->pino_nlink == 0)
c9f700f8
KK
732 jffs2_del_ino_cache(c, ic);
733 break;
734 }
1da177e4
LT
735 spin_unlock(&c->erase_completion_lock);
736 }
737
1da177e4 738 out_erase_sem:
ced22070 739 mutex_unlock(&c->erase_free_sem);
1da177e4
LT
740}
741
1da177e4
LT
742int jffs2_thread_should_wake(struct jffs2_sb_info *c)
743{
744 int ret = 0;
745 uint32_t dirty;
8fb870df
DW
746 int nr_very_dirty = 0;
747 struct jffs2_eraseblock *jeb;
1da177e4 748
d6ce1710
JT
749 if (!list_empty(&c->erase_complete_list) ||
750 !list_empty(&c->erase_pending_list))
751 return 1;
752
1da177e4
LT
753 if (c->unchecked_size) {
754 D1(printk(KERN_DEBUG "jffs2_thread_should_wake(): unchecked_size %d, checked_ino #%d\n",
755 c->unchecked_size, c->checked_ino));
756 return 1;
757 }
758
759 /* dirty_size contains blocks on erase_pending_list
760 * those blocks are counted in c->nr_erasing_blocks.
761 * If one block is actually erased, it is not longer counted as dirty_space
762 * but it is counted in c->nr_erasing_blocks, so we add it and subtract it
763 * with c->nr_erasing_blocks * c->sector_size again.
764 * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks
765 * This helps us to force gc and pick eventually a clean block to spread the load.
766 */
767 dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size;
768
182ec4ee
TG
769 if (c->nr_free_blocks + c->nr_erasing_blocks < c->resv_blocks_gctrigger &&
770 (dirty > c->nospc_dirty_size))
1da177e4
LT
771 ret = 1;
772
8fb870df
DW
773 list_for_each_entry(jeb, &c->very_dirty_list, list) {
774 nr_very_dirty++;
775 if (nr_very_dirty == c->vdirty_blocks_gctrigger) {
776 ret = 1;
a8c68f32
DW
777 /* In debug mode, actually go through and count them all */
778 D1(continue);
779 break;
8fb870df
DW
780 }
781 }
782
783 D1(printk(KERN_DEBUG "jffs2_thread_should_wake(): nr_free_blocks %d, nr_erasing_blocks %d, dirty_size 0x%x, vdirty_blocks %d: %s\n",
784 c->nr_free_blocks, c->nr_erasing_blocks, c->dirty_size, nr_very_dirty, ret?"yes":"no"));
1da177e4
LT
785
786 return ret;
787}