blob: 747a73f0aa4d3b8dcf25dfec74fd4a1631fb4a0d [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * JFFS2 -- Journalling Flash File System, Version 2.
3 *
David Woodhousec00c3102007-04-25 14:16:47 +01004 * Copyright © 2001-2007 Red Hat, Inc.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005 *
6 * Created by David Woodhouse <dwmw2@infradead.org>
7 *
8 * For licensing information, see the file 'LICENCE' in this directory.
9 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070010 */
11
12#include <linux/kernel.h>
13#include <linux/slab.h>
14#include <linux/mtd/mtd.h>
15#include <linux/compiler.h>
16#include <linux/sched.h> /* For cond_resched() */
17#include "nodelist.h"
Ferenc Havasie631ddb2005-09-07 09:35:26 +010018#include "debug.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070019
20/**
21 * jffs2_reserve_space - request physical space to write nodes to flash
22 * @c: superblock info
23 * @minsize: Minimum acceptable size of allocation
Linus Torvalds1da177e2005-04-16 15:20:36 -070024 * @len: Returned value of allocation length
25 * @prio: Allocation type - ALLOC_{NORMAL,DELETION}
26 *
27 * Requests a block of physical space on the flash. Returns zero for success
David Woodhouse9fe48542006-05-23 00:38:06 +010028 * and puts 'len' into the appropriate place, or returns -ENOSPC or other
29 * error if appropriate. Doesn't return len since that's
Linus Torvalds1da177e2005-04-16 15:20:36 -070030 *
31 * If it returns zero, jffs2_reserve_space() also downs the per-filesystem
32 * allocation semaphore, to prevent more than one allocation from being
33 * active at any time. The semaphore is later released by jffs2_commit_allocation()
34 *
35 * jffs2_reserve_space() may trigger garbage collection in order to make room
36 * for the requested allocation.
37 */
38
Ferenc Havasie631ddb2005-09-07 09:35:26 +010039static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
David Woodhouse9fe48542006-05-23 00:38:06 +010040 uint32_t *len, uint32_t sumsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
David Woodhouse9fe48542006-05-23 00:38:06 +010042int jffs2_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
Ferenc Havasie631ddb2005-09-07 09:35:26 +010043 uint32_t *len, int prio, uint32_t sumsize)
Linus Torvalds1da177e2005-04-16 15:20:36 -070044{
45 int ret = -EAGAIN;
46 int blocksneeded = c->resv_blocks_write;
47 /* align it */
48 minsize = PAD(minsize);
49
50 D1(printk(KERN_DEBUG "jffs2_reserve_space(): Requested 0x%x bytes\n", minsize));
David Woodhouseced22072008-04-22 15:13:40 +010051 mutex_lock(&c->alloc_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
53 D1(printk(KERN_DEBUG "jffs2_reserve_space(): alloc sem got\n"));
54
55 spin_lock(&c->erase_completion_lock);
56
57 /* this needs a little more thought (true <tglx> :)) */
58 while(ret == -EAGAIN) {
59 while(c->nr_free_blocks + c->nr_erasing_blocks < blocksneeded) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070060 uint32_t dirty, avail;
61
62 /* calculate real dirty size
63 * dirty_size contains blocks on erase_pending_list
64 * those blocks are counted in c->nr_erasing_blocks.
65 * If one block is actually erased, it is not longer counted as dirty_space
66 * but it is counted in c->nr_erasing_blocks, so we add it and subtract it
67 * with c->nr_erasing_blocks * c->sector_size again.
68 * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks
69 * This helps us to force gc and pick eventually a clean block to spread the load.
70 * We add unchecked_size here, as we hopefully will find some space to use.
71 * This will affect the sum only once, as gc first finishes checking
72 * of nodes.
73 */
74 dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size + c->unchecked_size;
75 if (dirty < c->nospc_dirty_size) {
76 if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) {
Artem B. Bityuckiy4132ace2005-05-06 10:30:30 +010077 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 -070078 break;
79 }
80 D1(printk(KERN_DEBUG "dirty size 0x%08x + unchecked_size 0x%08x < nospc_dirty_size 0x%08x, returning -ENOSPC\n",
81 dirty, c->unchecked_size, c->sector_size));
82
83 spin_unlock(&c->erase_completion_lock);
David Woodhouseced22072008-04-22 15:13:40 +010084 mutex_unlock(&c->alloc_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -070085 return -ENOSPC;
86 }
Thomas Gleixner182ec4e2005-11-07 11:16:07 +000087
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 /* Calc possibly available space. Possibly available means that we
89 * don't know, if unchecked size contains obsoleted nodes, which could give us some
90 * more usable space. This will affect the sum only once, as gc first finishes checking
91 * of nodes.
Thomas Gleixner182ec4e2005-11-07 11:16:07 +000092 + Return -ENOSPC, if the maximum possibly available space is less or equal than
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 * blocksneeded * sector_size.
94 * This blocks endless gc looping on a filesystem, which is nearly full, even if
95 * the check above passes.
96 */
97 avail = c->free_size + c->dirty_size + c->erasing_size + c->unchecked_size;
98 if ( (avail / c->sector_size) <= blocksneeded) {
99 if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) {
Artem B. Bityuckiy4132ace2005-05-06 10:30:30 +0100100 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 -0700101 break;
102 }
103
104 D1(printk(KERN_DEBUG "max. available size 0x%08x < blocksneeded * sector_size 0x%08x, returning -ENOSPC\n",
105 avail, blocksneeded * c->sector_size));
106 spin_unlock(&c->erase_completion_lock);
David Woodhouseced22072008-04-22 15:13:40 +0100107 mutex_unlock(&c->alloc_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 return -ENOSPC;
109 }
110
David Woodhouseced22072008-04-22 15:13:40 +0100111 mutex_unlock(&c->alloc_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112
113 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",
114 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,
115 c->free_size + c->dirty_size + c->wasted_size + c->used_size + c->erasing_size + c->bad_size, c->flash_size));
116 spin_unlock(&c->erase_completion_lock);
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000117
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 ret = jffs2_garbage_collect_pass(c);
119 if (ret)
120 return ret;
121
122 cond_resched();
123
124 if (signal_pending(current))
125 return -EINTR;
126
David Woodhouseced22072008-04-22 15:13:40 +0100127 mutex_lock(&c->alloc_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 spin_lock(&c->erase_completion_lock);
129 }
130
David Woodhouse9fe48542006-05-23 00:38:06 +0100131 ret = jffs2_do_reserve_space(c, minsize, len, sumsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 if (ret) {
133 D1(printk(KERN_DEBUG "jffs2_reserve_space: ret is %d\n", ret));
134 }
135 }
136 spin_unlock(&c->erase_completion_lock);
David Woodhouse2f785402006-05-24 02:04:45 +0100137 if (!ret)
David Woodhouse046b8b92006-05-25 01:50:35 +0100138 ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 if (ret)
David Woodhouseced22072008-04-22 15:13:40 +0100140 mutex_unlock(&c->alloc_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 return ret;
142}
143
David Woodhouse9fe48542006-05-23 00:38:06 +0100144int jffs2_reserve_space_gc(struct jffs2_sb_info *c, uint32_t minsize,
145 uint32_t *len, uint32_t sumsize)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146{
147 int ret = -EAGAIN;
148 minsize = PAD(minsize);
149
150 D1(printk(KERN_DEBUG "jffs2_reserve_space_gc(): Requested 0x%x bytes\n", minsize));
151
152 spin_lock(&c->erase_completion_lock);
153 while(ret == -EAGAIN) {
David Woodhouse9fe48542006-05-23 00:38:06 +0100154 ret = jffs2_do_reserve_space(c, minsize, len, sumsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 if (ret) {
David Woodhouseef53cb02007-07-10 10:01:22 +0100156 D1(printk(KERN_DEBUG "jffs2_reserve_space_gc: looping, ret is %d\n", ret));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 }
158 }
159 spin_unlock(&c->erase_completion_lock);
David Woodhouse2f785402006-05-24 02:04:45 +0100160 if (!ret)
David Woodhouse046b8b92006-05-25 01:50:35 +0100161 ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1);
David Woodhouse2f785402006-05-24 02:04:45 +0100162
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 return ret;
164}
165
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100166
167/* Classify nextblock (clean, dirty of verydirty) and force to select an other one */
168
169static void jffs2_close_nextblock(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170{
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100171
Adrian Hunter99c25942007-03-29 11:00:47 +0300172 if (c->nextblock == NULL) {
173 D1(printk(KERN_DEBUG "jffs2_close_nextblock: Erase block at 0x%08x has already been placed in a list\n",
174 jeb->offset));
175 return;
176 }
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100177 /* Check, if we have a dirty block now, or if it was dirty already */
178 if (ISDIRTY (jeb->wasted_size + jeb->dirty_size)) {
179 c->dirty_size += jeb->wasted_size;
180 c->wasted_size -= jeb->wasted_size;
181 jeb->dirty_size += jeb->wasted_size;
182 jeb->wasted_size = 0;
183 if (VERYDIRTY(c, jeb->dirty_size)) {
184 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",
185 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
186 list_add_tail(&jeb->list, &c->very_dirty_list);
187 } else {
188 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to dirty_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
189 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
190 list_add_tail(&jeb->list, &c->dirty_list);
191 }
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000192 } else {
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100193 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
194 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
195 list_add_tail(&jeb->list, &c->clean_list);
196 }
197 c->nextblock = NULL;
198
199}
200
201/* Select a new jeb for nextblock */
202
203static int jffs2_find_nextblock(struct jffs2_sb_info *c)
204{
205 struct list_head *next;
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000206
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100207 /* Take the next block off the 'free' list */
208
209 if (list_empty(&c->free_list)) {
210
211 if (!c->nr_erasing_blocks &&
212 !list_empty(&c->erasable_list)) {
213 struct jffs2_eraseblock *ejeb;
214
215 ejeb = list_entry(c->erasable_list.next, struct jffs2_eraseblock, list);
Akinobu Mitaf1166292006-06-26 00:24:46 -0700216 list_move_tail(&ejeb->list, &c->erase_pending_list);
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100217 c->nr_erasing_blocks++;
218 jffs2_erase_pending_trigger(c);
219 D1(printk(KERN_DEBUG "jffs2_find_nextblock: Triggering erase of erasable block at 0x%08x\n",
220 ejeb->offset));
221 }
222
223 if (!c->nr_erasing_blocks &&
224 !list_empty(&c->erasable_pending_wbuf_list)) {
225 D1(printk(KERN_DEBUG "jffs2_find_nextblock: Flushing write buffer\n"));
226 /* c->nextblock is NULL, no update to c->nextblock allowed */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 spin_unlock(&c->erase_completion_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 jffs2_flush_wbuf_pad(c);
229 spin_lock(&c->erase_completion_lock);
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100230 /* Have another go. It'll be on the erasable_list now */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 return -EAGAIN;
232 }
233
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100234 if (!c->nr_erasing_blocks) {
235 /* Ouch. We're in GC, or we wouldn't have got here.
236 And there's no space left. At all. */
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000237 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",
238 c->nr_erasing_blocks, c->nr_free_blocks, list_empty(&c->erasable_list)?"yes":"no",
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100239 list_empty(&c->erasing_list)?"yes":"no", list_empty(&c->erase_pending_list)?"yes":"no");
240 return -ENOSPC;
241 }
242
243 spin_unlock(&c->erase_completion_lock);
244 /* Don't wait for it; just erase one right now */
245 jffs2_erase_pending_blocks(c, 1);
246 spin_lock(&c->erase_completion_lock);
247
248 /* An erase may have failed, decreasing the
249 amount of free space available. So we must
250 restart from the beginning */
251 return -EAGAIN;
252 }
253
254 next = c->free_list.next;
255 list_del(next);
256 c->nextblock = list_entry(next, struct jffs2_eraseblock, list);
257 c->nr_free_blocks--;
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000258
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100259 jffs2_sum_reset_collected(c->summary); /* reset collected summary */
260
261 D1(printk(KERN_DEBUG "jffs2_find_nextblock(): new nextblock = 0x%08x\n", c->nextblock->offset));
262
263 return 0;
264}
265
266/* Called with alloc sem _and_ erase_completion_lock */
David Woodhouse9fe48542006-05-23 00:38:06 +0100267static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
268 uint32_t *len, uint32_t sumsize)
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100269{
270 struct jffs2_eraseblock *jeb = c->nextblock;
David Woodhouse9fe48542006-05-23 00:38:06 +0100271 uint32_t reserved_size; /* for summary information at the end of the jeb */
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100272 int ret;
273
274 restart:
275 reserved_size = 0;
276
277 if (jffs2_sum_active() && (sumsize != JFFS2_SUMMARY_NOSUM_SIZE)) {
278 /* NOSUM_SIZE means not to generate summary */
279
280 if (jeb) {
281 reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE);
Artem B. Bityutskiy733802d2005-09-22 12:25:00 +0100282 dbg_summary("minsize=%d , jeb->free=%d ,"
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100283 "summary->size=%d , sumsize=%d\n",
284 minsize, jeb->free_size,
285 c->summary->sum_size, sumsize);
286 }
287
288 /* Is there enough space for writing out the current node, or we have to
289 write out summary information now, close this jeb and select new nextblock? */
290 if (jeb && (PAD(minsize) + PAD(c->summary->sum_size + sumsize +
291 JFFS2_SUMMARY_FRAME_SIZE) > jeb->free_size)) {
292
293 /* Has summary been disabled for this jeb? */
294 if (jffs2_sum_is_disabled(c->summary)) {
295 sumsize = JFFS2_SUMMARY_NOSUM_SIZE;
296 goto restart;
297 }
298
299 /* Writing out the collected summary information */
Artem B. Bityutskiy733802d2005-09-22 12:25:00 +0100300 dbg_summary("generating summary for 0x%08x.\n", jeb->offset);
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100301 ret = jffs2_sum_write_sumnode(c);
302
303 if (ret)
304 return ret;
305
306 if (jffs2_sum_is_disabled(c->summary)) {
307 /* jffs2_write_sumnode() couldn't write out the summary information
308 diabling summary for this jeb and free the collected information
309 */
310 sumsize = JFFS2_SUMMARY_NOSUM_SIZE;
311 goto restart;
312 }
313
314 jffs2_close_nextblock(c, jeb);
315 jeb = NULL;
Ferenc Havasi34c0e902005-09-16 13:58:20 +0100316 /* keep always valid value in reserved_size */
317 reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE);
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100318 }
319 } else {
320 if (jeb && minsize > jeb->free_size) {
David Woodhousefc6612f2006-06-18 18:35:10 +0100321 uint32_t waste;
322
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100323 /* Skip the end of this block and file it as having some dirty space */
324 /* If there's a pending write to it, flush now */
325
326 if (jffs2_wbuf_dirty(c)) {
327 spin_unlock(&c->erase_completion_lock);
328 D1(printk(KERN_DEBUG "jffs2_do_reserve_space: Flushing write buffer\n"));
329 jffs2_flush_wbuf_pad(c);
330 spin_lock(&c->erase_completion_lock);
331 jeb = c->nextblock;
332 goto restart;
333 }
334
David Woodhousefc6612f2006-06-18 18:35:10 +0100335 spin_unlock(&c->erase_completion_lock);
336
337 ret = jffs2_prealloc_raw_node_refs(c, jeb, 1);
338 if (ret)
339 return ret;
340 /* Just lock it again and continue. Nothing much can change because
341 we hold c->alloc_sem anyway. In fact, it's not entirely clear why
342 we hold c->erase_completion_lock in the majority of this function...
343 but that's a question for another (more caffeine-rich) day. */
344 spin_lock(&c->erase_completion_lock);
345
346 waste = jeb->free_size;
347 jffs2_link_node_ref(c, jeb,
348 (jeb->offset + c->sector_size - waste) | REF_OBSOLETE,
349 waste, NULL);
350 /* FIXME: that made it count as dirty. Convert to wasted */
351 jeb->dirty_size -= waste;
352 c->dirty_size -= waste;
353 jeb->wasted_size += waste;
354 c->wasted_size += waste;
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100355
356 jffs2_close_nextblock(c, jeb);
357 jeb = NULL;
358 }
359 }
360
361 if (!jeb) {
362
363 ret = jffs2_find_nextblock(c);
364 if (ret)
365 return ret;
366
367 jeb = c->nextblock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368
369 if (jeb->free_size != c->sector_size - c->cleanmarker_size) {
370 printk(KERN_WARNING "Eep. Block 0x%08x taken from free_list had free_size of 0x%08x!!\n", jeb->offset, jeb->free_size);
371 goto restart;
372 }
373 }
374 /* OK, jeb (==c->nextblock) is now pointing at a block which definitely has
375 enough space */
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100376 *len = jeb->free_size - reserved_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377
378 if (c->cleanmarker_size && jeb->used_size == c->cleanmarker_size &&
379 !jeb->first_node->next_in_ino) {
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000380 /* Only node in it beforehand was a CLEANMARKER node (we think).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 So mark it obsolete now that there's going to be another node
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000382 in the block. This will reduce used_size to zero but We've
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 already set c->nextblock so that jffs2_mark_node_obsolete()
384 won't try to refile it to the dirty_list.
385 */
386 spin_unlock(&c->erase_completion_lock);
387 jffs2_mark_node_obsolete(c, jeb->first_node);
388 spin_lock(&c->erase_completion_lock);
389 }
390
David Woodhouse9fe48542006-05-23 00:38:06 +0100391 D1(printk(KERN_DEBUG "jffs2_do_reserve_space(): Giving 0x%x bytes at 0x%x\n",
392 *len, jeb->offset + (c->sector_size - jeb->free_size)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 return 0;
394}
395
396/**
397 * jffs2_add_physical_node_ref - add a physical node reference to the list
398 * @c: superblock info
399 * @new: new node reference to add
400 * @len: length of this physical node
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 *
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000402 * Should only be used to report nodes for which space has been allocated
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 * by jffs2_reserve_space.
404 *
405 * Must be called with the alloc_sem held.
406 */
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000407
David Woodhouse2f785402006-05-24 02:04:45 +0100408struct jffs2_raw_node_ref *jffs2_add_physical_node_ref(struct jffs2_sb_info *c,
409 uint32_t ofs, uint32_t len,
410 struct jffs2_inode_cache *ic)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411{
412 struct jffs2_eraseblock *jeb;
David Woodhouse2f785402006-05-24 02:04:45 +0100413 struct jffs2_raw_node_ref *new;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414
David Woodhouse2f785402006-05-24 02:04:45 +0100415 jeb = &c->blocks[ofs / c->sector_size];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416
David Woodhouse2f785402006-05-24 02:04:45 +0100417 D1(printk(KERN_DEBUG "jffs2_add_physical_node_ref(): Node at 0x%x(%d), size 0x%x\n",
418 ofs & ~3, ofs & 3, len));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419#if 1
David Woodhouse2f785402006-05-24 02:04:45 +0100420 /* Allow non-obsolete nodes only to be added at the end of c->nextblock,
421 if c->nextblock is set. Note that wbuf.c will file obsolete nodes
422 even after refiling c->nextblock */
423 if ((c->nextblock || ((ofs & 3) != REF_OBSOLETE))
424 && (jeb != c->nextblock || (ofs & ~3) != jeb->offset + (c->sector_size - jeb->free_size))) {
David Woodhouse66bfaea2007-06-28 19:03:11 +0100425 printk(KERN_WARNING "argh. node added in wrong place at 0x%08x(%d)\n", ofs & ~3, ofs & 3);
426 if (c->nextblock)
427 printk(KERN_WARNING "nextblock 0x%08x", c->nextblock->offset);
428 else
429 printk(KERN_WARNING "No nextblock");
430 printk(", expected at %08x\n", jeb->offset + (c->sector_size - jeb->free_size));
David Woodhouse2f785402006-05-24 02:04:45 +0100431 return ERR_PTR(-EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 }
433#endif
434 spin_lock(&c->erase_completion_lock);
435
David Woodhouse2f785402006-05-24 02:04:45 +0100436 new = jffs2_link_node_ref(c, jeb, ofs, len, ic);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437
Estelle Hammache9b88f472005-01-28 18:53:05 +0000438 if (!jeb->free_size && !jeb->dirty_size && !ISDIRTY(jeb->wasted_size)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 /* If it lives on the dirty_list, jffs2_reserve_space will put it there */
440 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
441 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
442 if (jffs2_wbuf_dirty(c)) {
443 /* Flush the last write in the block if it's outstanding */
444 spin_unlock(&c->erase_completion_lock);
445 jffs2_flush_wbuf_pad(c);
446 spin_lock(&c->erase_completion_lock);
447 }
448
449 list_add_tail(&jeb->list, &c->clean_list);
450 c->nextblock = NULL;
451 }
Artem B. Bityutskiye0c8e422005-07-24 16:14:17 +0100452 jffs2_dbg_acct_sanity_check_nolock(c,jeb);
453 jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454
455 spin_unlock(&c->erase_completion_lock);
456
David Woodhouse2f785402006-05-24 02:04:45 +0100457 return new;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458}
459
460
461void jffs2_complete_reservation(struct jffs2_sb_info *c)
462{
463 D1(printk(KERN_DEBUG "jffs2_complete_reservation()\n"));
464 jffs2_garbage_collect_trigger(c);
David Woodhouseced22072008-04-22 15:13:40 +0100465 mutex_unlock(&c->alloc_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466}
467
468static inline int on_list(struct list_head *obj, struct list_head *head)
469{
470 struct list_head *this;
471
472 list_for_each(this, head) {
473 if (this == obj) {
474 D1(printk("%p is on list at %p\n", obj, head));
475 return 1;
476
477 }
478 }
479 return 0;
480}
481
482void jffs2_mark_node_obsolete(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref)
483{
484 struct jffs2_eraseblock *jeb;
485 int blocknr;
486 struct jffs2_unknown_node n;
487 int ret, addedsize;
488 size_t retlen;
David Woodhouse1417fc42006-05-20 16:20:19 +0100489 uint32_t freed_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490
David Woodhouse9bfeb692006-05-26 21:19:05 +0100491 if(unlikely(!ref)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 printk(KERN_NOTICE "EEEEEK. jffs2_mark_node_obsolete called with NULL node\n");
493 return;
494 }
495 if (ref_obsolete(ref)) {
496 D1(printk(KERN_DEBUG "jffs2_mark_node_obsolete called with already obsolete node at 0x%08x\n", ref_offset(ref)));
497 return;
498 }
499 blocknr = ref->flash_offset / c->sector_size;
500 if (blocknr >= c->nr_blocks) {
501 printk(KERN_NOTICE "raw node at 0x%08x is off the end of device!\n", ref->flash_offset);
502 BUG();
503 }
504 jeb = &c->blocks[blocknr];
505
506 if (jffs2_can_mark_obsolete(c) && !jffs2_is_readonly(c) &&
Artem B. Bityuckiy31fbdf72005-02-28 08:21:09 +0000507 !(c->flags & (JFFS2_SB_FLAG_SCANNING | JFFS2_SB_FLAG_BUILDING))) {
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000508 /* Hm. This may confuse static lock analysis. If any of the above
509 three conditions is false, we're going to return from this
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 function without actually obliterating any nodes or freeing
511 any jffs2_raw_node_refs. So we don't need to stop erases from
512 happening, or protect against people holding an obsolete
513 jffs2_raw_node_ref without the erase_completion_lock. */
David Woodhouseced22072008-04-22 15:13:40 +0100514 mutex_lock(&c->erase_free_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 }
516
517 spin_lock(&c->erase_completion_lock);
518
David Woodhouse1417fc42006-05-20 16:20:19 +0100519 freed_len = ref_totlen(c, jeb, ref);
520
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 if (ref_flags(ref) == REF_UNCHECKED) {
David Woodhouse1417fc42006-05-20 16:20:19 +0100522 D1(if (unlikely(jeb->unchecked_size < freed_len)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 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",
David Woodhouse1417fc42006-05-20 16:20:19 +0100524 freed_len, blocknr, ref->flash_offset, jeb->used_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 BUG();
526 })
David Woodhouse1417fc42006-05-20 16:20:19 +0100527 D1(printk(KERN_DEBUG "Obsoleting previously unchecked node at 0x%08x of len %x: ", ref_offset(ref), freed_len));
528 jeb->unchecked_size -= freed_len;
529 c->unchecked_size -= freed_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 } else {
David Woodhouse1417fc42006-05-20 16:20:19 +0100531 D1(if (unlikely(jeb->used_size < freed_len)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 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",
David Woodhouse1417fc42006-05-20 16:20:19 +0100533 freed_len, blocknr, ref->flash_offset, jeb->used_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 BUG();
535 })
David Woodhouse1417fc42006-05-20 16:20:19 +0100536 D1(printk(KERN_DEBUG "Obsoleting node at 0x%08x of len %#x: ", ref_offset(ref), freed_len));
537 jeb->used_size -= freed_len;
538 c->used_size -= freed_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 }
540
541 // Take care, that wasted size is taken into concern
David Woodhouse1417fc42006-05-20 16:20:19 +0100542 if ((jeb->dirty_size || ISDIRTY(jeb->wasted_size + freed_len)) && jeb != c->nextblock) {
David Woodhousec7c16c82006-05-24 14:24:02 +0100543 D1(printk("Dirtying\n"));
David Woodhouse1417fc42006-05-20 16:20:19 +0100544 addedsize = freed_len;
545 jeb->dirty_size += freed_len;
546 c->dirty_size += freed_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547
548 /* Convert wasted space to dirty, if not a bad block */
549 if (jeb->wasted_size) {
550 if (on_list(&jeb->list, &c->bad_used_list)) {
551 D1(printk(KERN_DEBUG "Leaving block at %08x on the bad_used_list\n",
552 jeb->offset));
553 addedsize = 0; /* To fool the refiling code later */
554 } else {
555 D1(printk(KERN_DEBUG "Converting %d bytes of wasted space to dirty in block at %08x\n",
556 jeb->wasted_size, jeb->offset));
557 addedsize += jeb->wasted_size;
558 jeb->dirty_size += jeb->wasted_size;
559 c->dirty_size += jeb->wasted_size;
560 c->wasted_size -= jeb->wasted_size;
561 jeb->wasted_size = 0;
562 }
563 }
564 } else {
David Woodhousec7c16c82006-05-24 14:24:02 +0100565 D1(printk("Wasting\n"));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 addedsize = 0;
David Woodhouse1417fc42006-05-20 16:20:19 +0100567 jeb->wasted_size += freed_len;
568 c->wasted_size += freed_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 }
570 ref->flash_offset = ref_offset(ref) | REF_OBSOLETE;
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000571
Artem B. Bityutskiye0c8e422005-07-24 16:14:17 +0100572 jffs2_dbg_acct_sanity_check_nolock(c, jeb);
573 jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574
Artem B. Bityuckiy31fbdf72005-02-28 08:21:09 +0000575 if (c->flags & JFFS2_SB_FLAG_SCANNING) {
576 /* Flash scanning is in progress. Don't muck about with the block
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 lists because they're not ready yet, and don't actually
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000578 obliterate nodes that look obsolete. If they weren't
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 marked obsolete on the flash at the time they _became_
580 obsolete, there was probably a reason for that. */
581 spin_unlock(&c->erase_completion_lock);
582 /* We didn't lock the erase_free_sem */
583 return;
584 }
585
586 if (jeb == c->nextblock) {
587 D2(printk(KERN_DEBUG "Not moving nextblock 0x%08x to dirty/erase_pending list\n", jeb->offset));
588 } else if (!jeb->used_size && !jeb->unchecked_size) {
589 if (jeb == c->gcblock) {
590 D1(printk(KERN_DEBUG "gcblock at 0x%08x completely dirtied. Clearing gcblock...\n", jeb->offset));
591 c->gcblock = NULL;
592 } else {
593 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x completely dirtied. Removing from (dirty?) list...\n", jeb->offset));
594 list_del(&jeb->list);
595 }
596 if (jffs2_wbuf_dirty(c)) {
597 D1(printk(KERN_DEBUG "...and adding to erasable_pending_wbuf_list\n"));
598 list_add_tail(&jeb->list, &c->erasable_pending_wbuf_list);
599 } else {
600 if (jiffies & 127) {
601 /* Most of the time, we just erase it immediately. Otherwise we
602 spend ages scanning it on mount, etc. */
603 D1(printk(KERN_DEBUG "...and adding to erase_pending_list\n"));
604 list_add_tail(&jeb->list, &c->erase_pending_list);
605 c->nr_erasing_blocks++;
606 jffs2_erase_pending_trigger(c);
607 } else {
608 /* Sometimes, however, we leave it elsewhere so it doesn't get
609 immediately reused, and we spread the load a bit. */
610 D1(printk(KERN_DEBUG "...and adding to erasable_list\n"));
611 list_add_tail(&jeb->list, &c->erasable_list);
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000612 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 }
614 D1(printk(KERN_DEBUG "Done OK\n"));
615 } else if (jeb == c->gcblock) {
616 D2(printk(KERN_DEBUG "Not moving gcblock 0x%08x to dirty_list\n", jeb->offset));
617 } else if (ISDIRTY(jeb->dirty_size) && !ISDIRTY(jeb->dirty_size - addedsize)) {
618 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x is freshly dirtied. Removing from clean list...\n", jeb->offset));
619 list_del(&jeb->list);
620 D1(printk(KERN_DEBUG "...and adding to dirty_list\n"));
621 list_add_tail(&jeb->list, &c->dirty_list);
622 } else if (VERYDIRTY(c, jeb->dirty_size) &&
623 !VERYDIRTY(c, jeb->dirty_size - addedsize)) {
624 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x is now very dirty. Removing from dirty list...\n", jeb->offset));
625 list_del(&jeb->list);
626 D1(printk(KERN_DEBUG "...and adding to very_dirty_list\n"));
627 list_add_tail(&jeb->list, &c->very_dirty_list);
628 } else {
629 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 +0000630 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
631 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632
633 spin_unlock(&c->erase_completion_lock);
634
Artem B. Bityuckiy31fbdf72005-02-28 08:21:09 +0000635 if (!jffs2_can_mark_obsolete(c) || jffs2_is_readonly(c) ||
636 (c->flags & JFFS2_SB_FLAG_BUILDING)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 /* We didn't lock the erase_free_sem */
638 return;
639 }
640
641 /* The erase_free_sem is locked, and has been since before we marked the node obsolete
642 and potentially put its eraseblock onto the erase_pending_list. Thus, we know that
643 the block hasn't _already_ been erased, and that 'ref' itself hasn't been freed yet
David Woodhousec38c1b62006-05-25 01:38:27 +0100644 by jffs2_free_jeb_node_refs() in erase.c. Which is nice. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645
646 D1(printk(KERN_DEBUG "obliterating obsoleted node at 0x%08x\n", ref_offset(ref)));
647 ret = jffs2_flash_read(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
648 if (ret) {
649 printk(KERN_WARNING "Read error reading from 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 read from obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen);
654 goto out_erase_sem;
655 }
David Woodhouse1417fc42006-05-20 16:20:19 +0100656 if (PAD(je32_to_cpu(n.totlen)) != PAD(freed_len)) {
657 printk(KERN_WARNING "Node totlen on flash (0x%08x) != totlen from node ref (0x%08x)\n", je32_to_cpu(n.totlen), freed_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 goto out_erase_sem;
659 }
660 if (!(je16_to_cpu(n.nodetype) & JFFS2_NODE_ACCURATE)) {
661 D1(printk(KERN_DEBUG "Node at 0x%08x was already marked obsolete (nodetype 0x%04x)\n", ref_offset(ref), je16_to_cpu(n.nodetype)));
662 goto out_erase_sem;
663 }
664 /* XXX FIXME: This is ugly now */
665 n.nodetype = cpu_to_je16(je16_to_cpu(n.nodetype) & ~JFFS2_NODE_ACCURATE);
666 ret = jffs2_flash_write(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
667 if (ret) {
668 printk(KERN_WARNING "Write error in obliterating obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret);
669 goto out_erase_sem;
670 }
671 if (retlen != sizeof(n)) {
672 printk(KERN_WARNING "Short write in obliterating obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen);
673 goto out_erase_sem;
674 }
675
676 /* Nodes which have been marked obsolete no longer need to be
677 associated with any inode. Remove them from the per-inode list.
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000678
679 Note we can't do this for NAND at the moment because we need
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 obsolete dirent nodes to stay on the lists, because of the
681 horridness in jffs2_garbage_collect_deletion_dirent(). Also
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000682 because we delete the inocache, and on NAND we need that to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 stay around until all the nodes are actually erased, in order
684 to stop us from giving the same inode number to another newly
685 created inode. */
686 if (ref->next_in_ino) {
687 struct jffs2_inode_cache *ic;
688 struct jffs2_raw_node_ref **p;
689
690 spin_lock(&c->erase_completion_lock);
691
692 ic = jffs2_raw_ref_to_ic(ref);
693 for (p = &ic->nodes; (*p) != ref; p = &((*p)->next_in_ino))
694 ;
695
696 *p = ref->next_in_ino;
697 ref->next_in_ino = NULL;
698
KaiGai Koheic9f700f2006-06-11 10:35:15 +0900699 switch (ic->class) {
700#ifdef CONFIG_JFFS2_FS_XATTR
701 case RAWNODE_CLASS_XATTR_DATUM:
702 jffs2_release_xattr_datum(c, (struct jffs2_xattr_datum *)ic);
703 break;
704 case RAWNODE_CLASS_XATTR_REF:
705 jffs2_release_xattr_ref(c, (struct jffs2_xattr_ref *)ic);
706 break;
707#endif
708 default:
709 if (ic->nodes == (void *)ic && ic->nlink == 0)
710 jffs2_del_ino_cache(c, ic);
711 break;
712 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 spin_unlock(&c->erase_completion_lock);
714 }
715
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 out_erase_sem:
David Woodhouseced22072008-04-22 15:13:40 +0100717 mutex_unlock(&c->erase_free_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718}
719
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720int jffs2_thread_should_wake(struct jffs2_sb_info *c)
721{
722 int ret = 0;
723 uint32_t dirty;
David Woodhouse8fb870df2007-10-06 15:12:58 -0400724 int nr_very_dirty = 0;
725 struct jffs2_eraseblock *jeb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726
727 if (c->unchecked_size) {
728 D1(printk(KERN_DEBUG "jffs2_thread_should_wake(): unchecked_size %d, checked_ino #%d\n",
729 c->unchecked_size, c->checked_ino));
730 return 1;
731 }
732
733 /* dirty_size contains blocks on erase_pending_list
734 * those blocks are counted in c->nr_erasing_blocks.
735 * If one block is actually erased, it is not longer counted as dirty_space
736 * but it is counted in c->nr_erasing_blocks, so we add it and subtract it
737 * with c->nr_erasing_blocks * c->sector_size again.
738 * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks
739 * This helps us to force gc and pick eventually a clean block to spread the load.
740 */
741 dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size;
742
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000743 if (c->nr_free_blocks + c->nr_erasing_blocks < c->resv_blocks_gctrigger &&
744 (dirty > c->nospc_dirty_size))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 ret = 1;
746
David Woodhouse8fb870df2007-10-06 15:12:58 -0400747 list_for_each_entry(jeb, &c->very_dirty_list, list) {
748 nr_very_dirty++;
749 if (nr_very_dirty == c->vdirty_blocks_gctrigger) {
750 ret = 1;
David Woodhousea8c68f32007-10-13 11:32:16 +0100751 /* In debug mode, actually go through and count them all */
752 D1(continue);
753 break;
David Woodhouse8fb870df2007-10-06 15:12:58 -0400754 }
755 }
756
757 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",
758 c->nr_free_blocks, c->nr_erasing_blocks, c->dirty_size, nr_very_dirty, ret?"yes":"no"));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759
760 return ret;
761}