]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - fs/jffs2/nodemgmt.c
jffs2: remove from wait queue after schedule()
[mirror_ubuntu-artful-kernel.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
5a528957
JP
12#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13
1da177e4 14#include <linux/kernel.h>
1da177e4
LT
15#include <linux/mtd/mtd.h>
16#include <linux/compiler.h>
17#include <linux/sched.h> /* For cond_resched() */
18#include "nodelist.h"
e631ddba 19#include "debug.h"
1da177e4 20
8da8ba2e
DD
21/*
22 * Check whether the user is allowed to write.
23 */
24static 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
1da177e4
LT
52/**
53 * jffs2_reserve_space - request physical space to write nodes to flash
54 * @c: superblock info
55 * @minsize: Minimum acceptable size of allocation
1da177e4
LT
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
9fe4854c
DW
60 * and puts 'len' into the appropriate place, or returns -ENOSPC or other
61 * error if appropriate. Doesn't return len since that's
1da177e4
LT
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
e631ddba 71static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
9fe4854c 72 uint32_t *len, uint32_t sumsize);
1da177e4 73
9fe4854c 74int jffs2_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
e631ddba 75 uint32_t *len, int prio, uint32_t sumsize)
1da177e4
LT
76{
77 int ret = -EAGAIN;
78 int blocksneeded = c->resv_blocks_write;
79 /* align it */
80 minsize = PAD(minsize);
81
9c261b33 82 jffs2_dbg(1, "%s(): Requested 0x%x bytes\n", __func__, minsize);
ced22070 83 mutex_lock(&c->alloc_sem);
1da177e4 84
9c261b33 85 jffs2_dbg(1, "%s(): alloc sem got\n", __func__);
1da177e4
LT
86
87 spin_lock(&c->erase_completion_lock);
88
8da8ba2e
DD
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
1da177e4
LT
98 /* this needs a little more thought (true <tglx> :)) */
99 while(ret == -EAGAIN) {
100 while(c->nr_free_blocks + c->nr_erasing_blocks < blocksneeded) {
1da177e4
LT
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) {
9c261b33
JP
118 jffs2_dbg(1, "%s(): Low on dirty space to GC, but it's a deletion. Allowing...\n",
119 __func__);
1da177e4
LT
120 break;
121 }
9c261b33
JP
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);
1da177e4
LT
125
126 spin_unlock(&c->erase_completion_lock);
ced22070 127 mutex_unlock(&c->alloc_sem);
1da177e4
LT
128 return -ENOSPC;
129 }
182ec4ee 130
1da177e4
LT
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.
182ec4ee 135 + Return -ENOSPC, if the maximum possibly available space is less or equal than
1da177e4
LT
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) {
9c261b33
JP
143 jffs2_dbg(1, "%s(): Low on possibly available space, but it's a deletion. Allowing...\n",
144 __func__);
1da177e4
LT
145 break;
146 }
147
9c261b33
JP
148 jffs2_dbg(1, "max. available size 0x%08x < blocksneeded * sector_size 0x%08x, returning -ENOSPC\n",
149 avail, blocksneeded * c->sector_size);
1da177e4 150 spin_unlock(&c->erase_completion_lock);
ced22070 151 mutex_unlock(&c->alloc_sem);
1da177e4
LT
152 return -ENOSPC;
153 }
154
ced22070 155 mutex_unlock(&c->alloc_sem);
1da177e4 156
9c261b33
JP
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);
1da177e4 165 spin_unlock(&c->erase_completion_lock);
182ec4ee 166
1da177e4 167 ret = jffs2_garbage_collect_pass(c);
422b1202 168
0717bf84
DW
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);
9c261b33
JP
177 jffs2_dbg(1, "%s waiting for erase to complete\n",
178 __func__);
0717bf84
DW
179 spin_unlock(&c->erase_completion_lock);
180
181 schedule();
3ead9578 182 remove_wait_queue(&c->erase_wait, &wait);
0717bf84
DW
183 } else
184 spin_unlock(&c->erase_completion_lock);
185 } else if (ret)
1da177e4
LT
186 return ret;
187
188 cond_resched();
189
190 if (signal_pending(current))
191 return -EINTR;
192
ced22070 193 mutex_lock(&c->alloc_sem);
1da177e4
LT
194 spin_lock(&c->erase_completion_lock);
195 }
196
9fe4854c 197 ret = jffs2_do_reserve_space(c, minsize, len, sumsize);
1da177e4 198 if (ret) {
9c261b33 199 jffs2_dbg(1, "%s(): ret is %d\n", __func__, ret);
1da177e4
LT
200 }
201 }
8da8ba2e
DD
202
203out:
1da177e4 204 spin_unlock(&c->erase_completion_lock);
2f785402 205 if (!ret)
046b8b98 206 ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1);
1da177e4 207 if (ret)
ced22070 208 mutex_unlock(&c->alloc_sem);
1da177e4
LT
209 return ret;
210}
211
9fe4854c
DW
212int jffs2_reserve_space_gc(struct jffs2_sb_info *c, uint32_t minsize,
213 uint32_t *len, uint32_t sumsize)
1da177e4
LT
214{
215 int ret = -EAGAIN;
216 minsize = PAD(minsize);
217
9c261b33 218 jffs2_dbg(1, "%s(): Requested 0x%x bytes\n", __func__, minsize);
1da177e4
LT
219
220 spin_lock(&c->erase_completion_lock);
221 while(ret == -EAGAIN) {
9fe4854c 222 ret = jffs2_do_reserve_space(c, minsize, len, sumsize);
1da177e4 223 if (ret) {
9c261b33
JP
224 jffs2_dbg(1, "%s(): looping, ret is %d\n",
225 __func__, ret);
1da177e4
LT
226 }
227 }
228 spin_unlock(&c->erase_completion_lock);
2f785402 229 if (!ret)
046b8b98 230 ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1);
2f785402 231
1da177e4
LT
232 return ret;
233}
234
e631ddba
FH
235
236/* Classify nextblock (clean, dirty of verydirty) and force to select an other one */
237
238static void jffs2_close_nextblock(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
1da177e4 239{
e631ddba 240
99c2594f 241 if (c->nextblock == NULL) {
9c261b33
JP
242 jffs2_dbg(1, "%s(): Erase block at 0x%08x has already been placed in a list\n",
243 __func__, jeb->offset);
99c2594f
AH
244 return;
245 }
e631ddba
FH
246 /* Check, if we have a dirty block now, or if it was dirty already */
247 if (ISDIRTY (jeb->wasted_size + jeb->dirty_size)) {
248 c->dirty_size += jeb->wasted_size;
249 c->wasted_size -= jeb->wasted_size;
250 jeb->dirty_size += jeb->wasted_size;
251 jeb->wasted_size = 0;
252 if (VERYDIRTY(c, jeb->dirty_size)) {
9c261b33
JP
253 jffs2_dbg(1, "Adding full erase block at 0x%08x to very_dirty_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
254 jeb->offset, jeb->free_size, jeb->dirty_size,
255 jeb->used_size);
e631ddba
FH
256 list_add_tail(&jeb->list, &c->very_dirty_list);
257 } else {
9c261b33
JP
258 jffs2_dbg(1, "Adding full erase block at 0x%08x to dirty_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
259 jeb->offset, jeb->free_size, jeb->dirty_size,
260 jeb->used_size);
e631ddba
FH
261 list_add_tail(&jeb->list, &c->dirty_list);
262 }
182ec4ee 263 } else {
9c261b33
JP
264 jffs2_dbg(1, "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
265 jeb->offset, jeb->free_size, jeb->dirty_size,
266 jeb->used_size);
e631ddba
FH
267 list_add_tail(&jeb->list, &c->clean_list);
268 }
269 c->nextblock = NULL;
270
271}
272
273/* Select a new jeb for nextblock */
274
275static int jffs2_find_nextblock(struct jffs2_sb_info *c)
276{
277 struct list_head *next;
182ec4ee 278
e631ddba
FH
279 /* Take the next block off the 'free' list */
280
281 if (list_empty(&c->free_list)) {
282
283 if (!c->nr_erasing_blocks &&
284 !list_empty(&c->erasable_list)) {
285 struct jffs2_eraseblock *ejeb;
286
287 ejeb = list_entry(c->erasable_list.next, struct jffs2_eraseblock, list);
f116629d 288 list_move_tail(&ejeb->list, &c->erase_pending_list);
e631ddba 289 c->nr_erasing_blocks++;
ae3b6ba0 290 jffs2_garbage_collect_trigger(c);
9c261b33
JP
291 jffs2_dbg(1, "%s(): Triggering erase of erasable block at 0x%08x\n",
292 __func__, ejeb->offset);
e631ddba
FH
293 }
294
295 if (!c->nr_erasing_blocks &&
296 !list_empty(&c->erasable_pending_wbuf_list)) {
9c261b33
JP
297 jffs2_dbg(1, "%s(): Flushing write buffer\n",
298 __func__);
e631ddba 299 /* c->nextblock is NULL, no update to c->nextblock allowed */
1da177e4 300 spin_unlock(&c->erase_completion_lock);
1da177e4
LT
301 jffs2_flush_wbuf_pad(c);
302 spin_lock(&c->erase_completion_lock);
e631ddba
FH
303 /* Have another go. It'll be on the erasable_list now */
304 return -EAGAIN;
1da177e4 305 }
e631ddba
FH
306
307 if (!c->nr_erasing_blocks) {
308 /* Ouch. We're in GC, or we wouldn't have got here.
309 And there's no space left. At all. */
da320f05
JP
310 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",
311 c->nr_erasing_blocks, c->nr_free_blocks,
312 list_empty(&c->erasable_list) ? "yes" : "no",
313 list_empty(&c->erasing_list) ? "yes" : "no",
314 list_empty(&c->erase_pending_list) ? "yes" : "no");
e631ddba 315 return -ENOSPC;
1da177e4 316 }
e631ddba
FH
317
318 spin_unlock(&c->erase_completion_lock);
319 /* Don't wait for it; just erase one right now */
320 jffs2_erase_pending_blocks(c, 1);
321 spin_lock(&c->erase_completion_lock);
322
323 /* An erase may have failed, decreasing the
324 amount of free space available. So we must
325 restart from the beginning */
326 return -EAGAIN;
1da177e4 327 }
e631ddba
FH
328
329 next = c->free_list.next;
330 list_del(next);
331 c->nextblock = list_entry(next, struct jffs2_eraseblock, list);
332 c->nr_free_blocks--;
182ec4ee 333
e631ddba
FH
334 jffs2_sum_reset_collected(c->summary); /* reset collected summary */
335
f04de505 336#ifdef CONFIG_JFFS2_FS_WRITEBUFFER
5bf17237
AB
337 /* adjust write buffer offset, else we get a non contiguous write bug */
338 if (!(c->wbuf_ofs % c->sector_size) && !c->wbuf_len)
339 c->wbuf_ofs = 0xffffffff;
f04de505 340#endif
5bf17237 341
9c261b33
JP
342 jffs2_dbg(1, "%s(): new nextblock = 0x%08x\n",
343 __func__, c->nextblock->offset);
e631ddba
FH
344
345 return 0;
346}
347
348/* Called with alloc sem _and_ erase_completion_lock */
9fe4854c
DW
349static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
350 uint32_t *len, uint32_t sumsize)
e631ddba
FH
351{
352 struct jffs2_eraseblock *jeb = c->nextblock;
9fe4854c 353 uint32_t reserved_size; /* for summary information at the end of the jeb */
e631ddba
FH
354 int ret;
355
356 restart:
357 reserved_size = 0;
358
359 if (jffs2_sum_active() && (sumsize != JFFS2_SUMMARY_NOSUM_SIZE)) {
360 /* NOSUM_SIZE means not to generate summary */
361
362 if (jeb) {
363 reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE);
733802d9 364 dbg_summary("minsize=%d , jeb->free=%d ,"
e631ddba
FH
365 "summary->size=%d , sumsize=%d\n",
366 minsize, jeb->free_size,
367 c->summary->sum_size, sumsize);
368 }
369
370 /* Is there enough space for writing out the current node, or we have to
371 write out summary information now, close this jeb and select new nextblock? */
372 if (jeb && (PAD(minsize) + PAD(c->summary->sum_size + sumsize +
373 JFFS2_SUMMARY_FRAME_SIZE) > jeb->free_size)) {
374
375 /* Has summary been disabled for this jeb? */
376 if (jffs2_sum_is_disabled(c->summary)) {
377 sumsize = JFFS2_SUMMARY_NOSUM_SIZE;
378 goto restart;
1da177e4
LT
379 }
380
e631ddba 381 /* Writing out the collected summary information */
733802d9 382 dbg_summary("generating summary for 0x%08x.\n", jeb->offset);
e631ddba
FH
383 ret = jffs2_sum_write_sumnode(c);
384
385 if (ret)
386 return ret;
387
388 if (jffs2_sum_is_disabled(c->summary)) {
389 /* jffs2_write_sumnode() couldn't write out the summary information
390 diabling summary for this jeb and free the collected information
391 */
392 sumsize = JFFS2_SUMMARY_NOSUM_SIZE;
393 goto restart;
394 }
395
396 jffs2_close_nextblock(c, jeb);
397 jeb = NULL;
34c0e906
FH
398 /* keep always valid value in reserved_size */
399 reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE);
e631ddba
FH
400 }
401 } else {
402 if (jeb && minsize > jeb->free_size) {
fc6612f6
DW
403 uint32_t waste;
404
e631ddba
FH
405 /* Skip the end of this block and file it as having some dirty space */
406 /* If there's a pending write to it, flush now */
407
408 if (jffs2_wbuf_dirty(c)) {
1da177e4 409 spin_unlock(&c->erase_completion_lock);
9c261b33
JP
410 jffs2_dbg(1, "%s(): Flushing write buffer\n",
411 __func__);
1da177e4
LT
412 jffs2_flush_wbuf_pad(c);
413 spin_lock(&c->erase_completion_lock);
e631ddba
FH
414 jeb = c->nextblock;
415 goto restart;
1da177e4
LT
416 }
417
fc6612f6
DW
418 spin_unlock(&c->erase_completion_lock);
419
420 ret = jffs2_prealloc_raw_node_refs(c, jeb, 1);
2cbba75a 421
fc6612f6
DW
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
2cbba75a
AK
428 if (ret)
429 return ret;
430
fc6612f6
DW
431 waste = jeb->free_size;
432 jffs2_link_node_ref(c, jeb,
433 (jeb->offset + c->sector_size - waste) | REF_OBSOLETE,
434 waste, NULL);
435 /* FIXME: that made it count as dirty. Convert to wasted */
436 jeb->dirty_size -= waste;
437 c->dirty_size -= waste;
438 jeb->wasted_size += waste;
439 c->wasted_size += waste;
1da177e4 440
e631ddba
FH
441 jffs2_close_nextblock(c, jeb);
442 jeb = NULL;
1da177e4 443 }
e631ddba
FH
444 }
445
446 if (!jeb) {
447
448 ret = jffs2_find_nextblock(c);
449 if (ret)
450 return ret;
1da177e4 451
e631ddba 452 jeb = c->nextblock;
1da177e4
LT
453
454 if (jeb->free_size != c->sector_size - c->cleanmarker_size) {
da320f05
JP
455 pr_warn("Eep. Block 0x%08x taken from free_list had free_size of 0x%08x!!\n",
456 jeb->offset, jeb->free_size);
1da177e4
LT
457 goto restart;
458 }
459 }
460 /* OK, jeb (==c->nextblock) is now pointing at a block which definitely has
461 enough space */
e631ddba 462 *len = jeb->free_size - reserved_size;
1da177e4
LT
463
464 if (c->cleanmarker_size && jeb->used_size == c->cleanmarker_size &&
465 !jeb->first_node->next_in_ino) {
182ec4ee 466 /* Only node in it beforehand was a CLEANMARKER node (we think).
1da177e4 467 So mark it obsolete now that there's going to be another node
182ec4ee 468 in the block. This will reduce used_size to zero but We've
1da177e4
LT
469 already set c->nextblock so that jffs2_mark_node_obsolete()
470 won't try to refile it to the dirty_list.
471 */
472 spin_unlock(&c->erase_completion_lock);
473 jffs2_mark_node_obsolete(c, jeb->first_node);
474 spin_lock(&c->erase_completion_lock);
475 }
476
9c261b33
JP
477 jffs2_dbg(1, "%s(): Giving 0x%x bytes at 0x%x\n",
478 __func__,
479 *len, jeb->offset + (c->sector_size - jeb->free_size));
1da177e4
LT
480 return 0;
481}
482
483/**
484 * jffs2_add_physical_node_ref - add a physical node reference to the list
485 * @c: superblock info
486 * @new: new node reference to add
487 * @len: length of this physical node
1da177e4 488 *
182ec4ee 489 * Should only be used to report nodes for which space has been allocated
1da177e4
LT
490 * by jffs2_reserve_space.
491 *
492 * Must be called with the alloc_sem held.
493 */
182ec4ee 494
2f785402
DW
495struct jffs2_raw_node_ref *jffs2_add_physical_node_ref(struct jffs2_sb_info *c,
496 uint32_t ofs, uint32_t len,
497 struct jffs2_inode_cache *ic)
1da177e4
LT
498{
499 struct jffs2_eraseblock *jeb;
2f785402 500 struct jffs2_raw_node_ref *new;
1da177e4 501
2f785402 502 jeb = &c->blocks[ofs / c->sector_size];
1da177e4 503
9c261b33
JP
504 jffs2_dbg(1, "%s(): Node at 0x%x(%d), size 0x%x\n",
505 __func__, ofs & ~3, ofs & 3, len);
1da177e4 506#if 1
2f785402
DW
507 /* Allow non-obsolete nodes only to be added at the end of c->nextblock,
508 if c->nextblock is set. Note that wbuf.c will file obsolete nodes
509 even after refiling c->nextblock */
510 if ((c->nextblock || ((ofs & 3) != REF_OBSOLETE))
511 && (jeb != c->nextblock || (ofs & ~3) != jeb->offset + (c->sector_size - jeb->free_size))) {
da320f05
JP
512 pr_warn("argh. node added in wrong place at 0x%08x(%d)\n",
513 ofs & ~3, ofs & 3);
66bfaeaa 514 if (c->nextblock)
da320f05 515 pr_warn("nextblock 0x%08x", c->nextblock->offset);
66bfaeaa 516 else
da320f05
JP
517 pr_warn("No nextblock");
518 pr_cont(", expected at %08x\n",
519 jeb->offset + (c->sector_size - jeb->free_size));
2f785402 520 return ERR_PTR(-EINVAL);
1da177e4
LT
521 }
522#endif
523 spin_lock(&c->erase_completion_lock);
524
2f785402 525 new = jffs2_link_node_ref(c, jeb, ofs, len, ic);
1da177e4 526
9b88f473 527 if (!jeb->free_size && !jeb->dirty_size && !ISDIRTY(jeb->wasted_size)) {
1da177e4 528 /* If it lives on the dirty_list, jffs2_reserve_space will put it there */
9c261b33
JP
529 jffs2_dbg(1, "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
530 jeb->offset, jeb->free_size, jeb->dirty_size,
531 jeb->used_size);
1da177e4
LT
532 if (jffs2_wbuf_dirty(c)) {
533 /* Flush the last write in the block if it's outstanding */
534 spin_unlock(&c->erase_completion_lock);
535 jffs2_flush_wbuf_pad(c);
536 spin_lock(&c->erase_completion_lock);
537 }
538
539 list_add_tail(&jeb->list, &c->clean_list);
540 c->nextblock = NULL;
541 }
e0c8e42f
AB
542 jffs2_dbg_acct_sanity_check_nolock(c,jeb);
543 jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
1da177e4
LT
544
545 spin_unlock(&c->erase_completion_lock);
546
2f785402 547 return new;
1da177e4
LT
548}
549
550
551void jffs2_complete_reservation(struct jffs2_sb_info *c)
552{
9c261b33 553 jffs2_dbg(1, "jffs2_complete_reservation()\n");
acb64a43 554 spin_lock(&c->erase_completion_lock);
1da177e4 555 jffs2_garbage_collect_trigger(c);
acb64a43 556 spin_unlock(&c->erase_completion_lock);
ced22070 557 mutex_unlock(&c->alloc_sem);
1da177e4
LT
558}
559
560static inline int on_list(struct list_head *obj, struct list_head *head)
561{
562 struct list_head *this;
563
564 list_for_each(this, head) {
565 if (this == obj) {
9c261b33 566 jffs2_dbg(1, "%p is on list at %p\n", obj, head);
1da177e4
LT
567 return 1;
568
569 }
570 }
571 return 0;
572}
573
574void jffs2_mark_node_obsolete(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref)
575{
576 struct jffs2_eraseblock *jeb;
577 int blocknr;
578 struct jffs2_unknown_node n;
579 int ret, addedsize;
580 size_t retlen;
1417fc44 581 uint32_t freed_len;
1da177e4 582
9bfeb691 583 if(unlikely(!ref)) {
da320f05 584 pr_notice("EEEEEK. jffs2_mark_node_obsolete called with NULL node\n");
1da177e4
LT
585 return;
586 }
587 if (ref_obsolete(ref)) {
9c261b33
JP
588 jffs2_dbg(1, "%s(): called with already obsolete node at 0x%08x\n",
589 __func__, ref_offset(ref));
1da177e4
LT
590 return;
591 }
592 blocknr = ref->flash_offset / c->sector_size;
593 if (blocknr >= c->nr_blocks) {
da320f05
JP
594 pr_notice("raw node at 0x%08x is off the end of device!\n",
595 ref->flash_offset);
1da177e4
LT
596 BUG();
597 }
598 jeb = &c->blocks[blocknr];
599
600 if (jffs2_can_mark_obsolete(c) && !jffs2_is_readonly(c) &&
31fbdf7a 601 !(c->flags & (JFFS2_SB_FLAG_SCANNING | JFFS2_SB_FLAG_BUILDING))) {
182ec4ee
TG
602 /* Hm. This may confuse static lock analysis. If any of the above
603 three conditions is false, we're going to return from this
1da177e4
LT
604 function without actually obliterating any nodes or freeing
605 any jffs2_raw_node_refs. So we don't need to stop erases from
606 happening, or protect against people holding an obsolete
607 jffs2_raw_node_ref without the erase_completion_lock. */
ced22070 608 mutex_lock(&c->erase_free_sem);
1da177e4
LT
609 }
610
611 spin_lock(&c->erase_completion_lock);
612
1417fc44
DW
613 freed_len = ref_totlen(c, jeb, ref);
614
1da177e4 615 if (ref_flags(ref) == REF_UNCHECKED) {
1417fc44 616 D1(if (unlikely(jeb->unchecked_size < freed_len)) {
da320f05
JP
617 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",
618 freed_len, blocknr,
619 ref->flash_offset, jeb->used_size);
1da177e4
LT
620 BUG();
621 })
9c261b33
JP
622 jffs2_dbg(1, "Obsoleting previously unchecked node at 0x%08x of len %x\n",
623 ref_offset(ref), freed_len);
1417fc44
DW
624 jeb->unchecked_size -= freed_len;
625 c->unchecked_size -= freed_len;
1da177e4 626 } else {
1417fc44 627 D1(if (unlikely(jeb->used_size < freed_len)) {
da320f05
JP
628 pr_notice("raw node of size 0x%08x freed from erase block %d at 0x%08x, but used_size was already 0x%08x\n",
629 freed_len, blocknr,
630 ref->flash_offset, jeb->used_size);
1da177e4
LT
631 BUG();
632 })
9c261b33
JP
633 jffs2_dbg(1, "Obsoleting node at 0x%08x of len %#x: ",
634 ref_offset(ref), freed_len);
1417fc44
DW
635 jeb->used_size -= freed_len;
636 c->used_size -= freed_len;
1da177e4
LT
637 }
638
639 // Take care, that wasted size is taken into concern
1417fc44 640 if ((jeb->dirty_size || ISDIRTY(jeb->wasted_size + freed_len)) && jeb != c->nextblock) {
9c261b33 641 jffs2_dbg(1, "Dirtying\n");
1417fc44
DW
642 addedsize = freed_len;
643 jeb->dirty_size += freed_len;
644 c->dirty_size += freed_len;
1da177e4
LT
645
646 /* Convert wasted space to dirty, if not a bad block */
647 if (jeb->wasted_size) {
648 if (on_list(&jeb->list, &c->bad_used_list)) {
9c261b33
JP
649 jffs2_dbg(1, "Leaving block at %08x on the bad_used_list\n",
650 jeb->offset);
1da177e4
LT
651 addedsize = 0; /* To fool the refiling code later */
652 } else {
9c261b33
JP
653 jffs2_dbg(1, "Converting %d bytes of wasted space to dirty in block at %08x\n",
654 jeb->wasted_size, jeb->offset);
1da177e4
LT
655 addedsize += jeb->wasted_size;
656 jeb->dirty_size += jeb->wasted_size;
657 c->dirty_size += jeb->wasted_size;
658 c->wasted_size -= jeb->wasted_size;
659 jeb->wasted_size = 0;
660 }
661 }
662 } else {
9c261b33 663 jffs2_dbg(1, "Wasting\n");
1da177e4 664 addedsize = 0;
1417fc44
DW
665 jeb->wasted_size += freed_len;
666 c->wasted_size += freed_len;
1da177e4
LT
667 }
668 ref->flash_offset = ref_offset(ref) | REF_OBSOLETE;
182ec4ee 669
e0c8e42f
AB
670 jffs2_dbg_acct_sanity_check_nolock(c, jeb);
671 jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
1da177e4 672
31fbdf7a
AB
673 if (c->flags & JFFS2_SB_FLAG_SCANNING) {
674 /* Flash scanning is in progress. Don't muck about with the block
1da177e4 675 lists because they're not ready yet, and don't actually
182ec4ee 676 obliterate nodes that look obsolete. If they weren't
1da177e4
LT
677 marked obsolete on the flash at the time they _became_
678 obsolete, there was probably a reason for that. */
679 spin_unlock(&c->erase_completion_lock);
680 /* We didn't lock the erase_free_sem */
681 return;
682 }
683
684 if (jeb == c->nextblock) {
9c261b33
JP
685 jffs2_dbg(2, "Not moving nextblock 0x%08x to dirty/erase_pending list\n",
686 jeb->offset);
1da177e4
LT
687 } else if (!jeb->used_size && !jeb->unchecked_size) {
688 if (jeb == c->gcblock) {
9c261b33
JP
689 jffs2_dbg(1, "gcblock at 0x%08x completely dirtied. Clearing gcblock...\n",
690 jeb->offset);
1da177e4
LT
691 c->gcblock = NULL;
692 } else {
9c261b33
JP
693 jffs2_dbg(1, "Eraseblock at 0x%08x completely dirtied. Removing from (dirty?) list...\n",
694 jeb->offset);
1da177e4
LT
695 list_del(&jeb->list);
696 }
697 if (jffs2_wbuf_dirty(c)) {
9c261b33 698 jffs2_dbg(1, "...and adding to erasable_pending_wbuf_list\n");
1da177e4
LT
699 list_add_tail(&jeb->list, &c->erasable_pending_wbuf_list);
700 } else {
701 if (jiffies & 127) {
702 /* Most of the time, we just erase it immediately. Otherwise we
703 spend ages scanning it on mount, etc. */
9c261b33 704 jffs2_dbg(1, "...and adding to erase_pending_list\n");
1da177e4
LT
705 list_add_tail(&jeb->list, &c->erase_pending_list);
706 c->nr_erasing_blocks++;
ae3b6ba0 707 jffs2_garbage_collect_trigger(c);
1da177e4
LT
708 } else {
709 /* Sometimes, however, we leave it elsewhere so it doesn't get
710 immediately reused, and we spread the load a bit. */
9c261b33 711 jffs2_dbg(1, "...and adding to erasable_list\n");
1da177e4 712 list_add_tail(&jeb->list, &c->erasable_list);
182ec4ee 713 }
1da177e4 714 }
9c261b33 715 jffs2_dbg(1, "Done OK\n");
1da177e4 716 } else if (jeb == c->gcblock) {
9c261b33
JP
717 jffs2_dbg(2, "Not moving gcblock 0x%08x to dirty_list\n",
718 jeb->offset);
1da177e4 719 } else if (ISDIRTY(jeb->dirty_size) && !ISDIRTY(jeb->dirty_size - addedsize)) {
9c261b33
JP
720 jffs2_dbg(1, "Eraseblock at 0x%08x is freshly dirtied. Removing from clean list...\n",
721 jeb->offset);
1da177e4 722 list_del(&jeb->list);
9c261b33 723 jffs2_dbg(1, "...and adding to dirty_list\n");
1da177e4
LT
724 list_add_tail(&jeb->list, &c->dirty_list);
725 } else if (VERYDIRTY(c, jeb->dirty_size) &&
726 !VERYDIRTY(c, jeb->dirty_size - addedsize)) {
9c261b33
JP
727 jffs2_dbg(1, "Eraseblock at 0x%08x is now very dirty. Removing from dirty list...\n",
728 jeb->offset);
1da177e4 729 list_del(&jeb->list);
9c261b33 730 jffs2_dbg(1, "...and adding to very_dirty_list\n");
1da177e4
LT
731 list_add_tail(&jeb->list, &c->very_dirty_list);
732 } else {
9c261b33
JP
733 jffs2_dbg(1, "Eraseblock at 0x%08x not moved anywhere. (free 0x%08x, dirty 0x%08x, used 0x%08x)\n",
734 jeb->offset, jeb->free_size, jeb->dirty_size,
735 jeb->used_size);
182ec4ee 736 }
1da177e4
LT
737
738 spin_unlock(&c->erase_completion_lock);
739
31fbdf7a
AB
740 if (!jffs2_can_mark_obsolete(c) || jffs2_is_readonly(c) ||
741 (c->flags & JFFS2_SB_FLAG_BUILDING)) {
1da177e4
LT
742 /* We didn't lock the erase_free_sem */
743 return;
744 }
745
746 /* The erase_free_sem is locked, and has been since before we marked the node obsolete
747 and potentially put its eraseblock onto the erase_pending_list. Thus, we know that
748 the block hasn't _already_ been erased, and that 'ref' itself hasn't been freed yet
c38c1b61 749 by jffs2_free_jeb_node_refs() in erase.c. Which is nice. */
1da177e4 750
9c261b33
JP
751 jffs2_dbg(1, "obliterating obsoleted node at 0x%08x\n",
752 ref_offset(ref));
1da177e4
LT
753 ret = jffs2_flash_read(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
754 if (ret) {
da320f05
JP
755 pr_warn("Read error reading from obsoleted node at 0x%08x: %d\n",
756 ref_offset(ref), ret);
1da177e4
LT
757 goto out_erase_sem;
758 }
759 if (retlen != sizeof(n)) {
da320f05
JP
760 pr_warn("Short read from obsoleted node at 0x%08x: %zd\n",
761 ref_offset(ref), retlen);
1da177e4
LT
762 goto out_erase_sem;
763 }
1417fc44 764 if (PAD(je32_to_cpu(n.totlen)) != PAD(freed_len)) {
da320f05
JP
765 pr_warn("Node totlen on flash (0x%08x) != totlen from node ref (0x%08x)\n",
766 je32_to_cpu(n.totlen), freed_len);
1da177e4
LT
767 goto out_erase_sem;
768 }
769 if (!(je16_to_cpu(n.nodetype) & JFFS2_NODE_ACCURATE)) {
9c261b33
JP
770 jffs2_dbg(1, "Node at 0x%08x was already marked obsolete (nodetype 0x%04x)\n",
771 ref_offset(ref), je16_to_cpu(n.nodetype));
1da177e4
LT
772 goto out_erase_sem;
773 }
774 /* XXX FIXME: This is ugly now */
775 n.nodetype = cpu_to_je16(je16_to_cpu(n.nodetype) & ~JFFS2_NODE_ACCURATE);
776 ret = jffs2_flash_write(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
777 if (ret) {
da320f05
JP
778 pr_warn("Write error in obliterating obsoleted node at 0x%08x: %d\n",
779 ref_offset(ref), ret);
1da177e4
LT
780 goto out_erase_sem;
781 }
782 if (retlen != sizeof(n)) {
da320f05
JP
783 pr_warn("Short write in obliterating obsoleted node at 0x%08x: %zd\n",
784 ref_offset(ref), retlen);
1da177e4
LT
785 goto out_erase_sem;
786 }
787
788 /* Nodes which have been marked obsolete no longer need to be
789 associated with any inode. Remove them from the per-inode list.
182ec4ee
TG
790
791 Note we can't do this for NAND at the moment because we need
1da177e4
LT
792 obsolete dirent nodes to stay on the lists, because of the
793 horridness in jffs2_garbage_collect_deletion_dirent(). Also
182ec4ee 794 because we delete the inocache, and on NAND we need that to
1da177e4
LT
795 stay around until all the nodes are actually erased, in order
796 to stop us from giving the same inode number to another newly
797 created inode. */
798 if (ref->next_in_ino) {
799 struct jffs2_inode_cache *ic;
800 struct jffs2_raw_node_ref **p;
801
802 spin_lock(&c->erase_completion_lock);
803
804 ic = jffs2_raw_ref_to_ic(ref);
805 for (p = &ic->nodes; (*p) != ref; p = &((*p)->next_in_ino))
806 ;
807
808 *p = ref->next_in_ino;
809 ref->next_in_ino = NULL;
810
c9f700f8
KK
811 switch (ic->class) {
812#ifdef CONFIG_JFFS2_FS_XATTR
813 case RAWNODE_CLASS_XATTR_DATUM:
814 jffs2_release_xattr_datum(c, (struct jffs2_xattr_datum *)ic);
815 break;
816 case RAWNODE_CLASS_XATTR_REF:
817 jffs2_release_xattr_ref(c, (struct jffs2_xattr_ref *)ic);
818 break;
819#endif
820 default:
27c72b04 821 if (ic->nodes == (void *)ic && ic->pino_nlink == 0)
c9f700f8
KK
822 jffs2_del_ino_cache(c, ic);
823 break;
824 }
1da177e4
LT
825 spin_unlock(&c->erase_completion_lock);
826 }
827
1da177e4 828 out_erase_sem:
ced22070 829 mutex_unlock(&c->erase_free_sem);
1da177e4
LT
830}
831
1da177e4
LT
832int jffs2_thread_should_wake(struct jffs2_sb_info *c)
833{
834 int ret = 0;
835 uint32_t dirty;
8fb870df
DW
836 int nr_very_dirty = 0;
837 struct jffs2_eraseblock *jeb;
1da177e4 838
d6ce1710
JT
839 if (!list_empty(&c->erase_complete_list) ||
840 !list_empty(&c->erase_pending_list))
841 return 1;
842
1da177e4 843 if (c->unchecked_size) {
9c261b33
JP
844 jffs2_dbg(1, "jffs2_thread_should_wake(): unchecked_size %d, checked_ino #%d\n",
845 c->unchecked_size, c->checked_ino);
1da177e4
LT
846 return 1;
847 }
848
849 /* dirty_size contains blocks on erase_pending_list
850 * those blocks are counted in c->nr_erasing_blocks.
851 * If one block is actually erased, it is not longer counted as dirty_space
852 * but it is counted in c->nr_erasing_blocks, so we add it and subtract it
853 * with c->nr_erasing_blocks * c->sector_size again.
854 * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks
855 * This helps us to force gc and pick eventually a clean block to spread the load.
856 */
857 dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size;
858
182ec4ee
TG
859 if (c->nr_free_blocks + c->nr_erasing_blocks < c->resv_blocks_gctrigger &&
860 (dirty > c->nospc_dirty_size))
1da177e4
LT
861 ret = 1;
862
8fb870df
DW
863 list_for_each_entry(jeb, &c->very_dirty_list, list) {
864 nr_very_dirty++;
865 if (nr_very_dirty == c->vdirty_blocks_gctrigger) {
866 ret = 1;
a8c68f32
DW
867 /* In debug mode, actually go through and count them all */
868 D1(continue);
869 break;
8fb870df
DW
870 }
871 }
872
9c261b33
JP
873 jffs2_dbg(1, "%s(): nr_free_blocks %d, nr_erasing_blocks %d, dirty_size 0x%x, vdirty_blocks %d: %s\n",
874 __func__, c->nr_free_blocks, c->nr_erasing_blocks,
875 c->dirty_size, nr_very_dirty, ret ? "yes" : "no");
1da177e4
LT
876
877 return ret;
878}