blob: dbc908ad622b10d89c3e6b740ec4b063fd6c60e9 [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));
51 down(&c->alloc_sem);
52
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) {
60 int ret;
61 uint32_t dirty, avail;
62
63 /* calculate real dirty size
64 * dirty_size contains blocks on erase_pending_list
65 * those blocks are counted in c->nr_erasing_blocks.
66 * If one block is actually erased, it is not longer counted as dirty_space
67 * but it is counted in c->nr_erasing_blocks, so we add it and subtract it
68 * with c->nr_erasing_blocks * c->sector_size again.
69 * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks
70 * This helps us to force gc and pick eventually a clean block to spread the load.
71 * We add unchecked_size here, as we hopefully will find some space to use.
72 * This will affect the sum only once, as gc first finishes checking
73 * of nodes.
74 */
75 dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size + c->unchecked_size;
76 if (dirty < c->nospc_dirty_size) {
77 if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) {
Artem B. Bityuckiy4132ace2005-05-06 10:30:30 +010078 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 -070079 break;
80 }
81 D1(printk(KERN_DEBUG "dirty size 0x%08x + unchecked_size 0x%08x < nospc_dirty_size 0x%08x, returning -ENOSPC\n",
82 dirty, c->unchecked_size, c->sector_size));
83
84 spin_unlock(&c->erase_completion_lock);
85 up(&c->alloc_sem);
86 return -ENOSPC;
87 }
Thomas Gleixner182ec4e2005-11-07 11:16:07 +000088
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 /* Calc possibly available space. Possibly available means that we
90 * don't know, if unchecked size contains obsoleted nodes, which could give us some
91 * more usable space. This will affect the sum only once, as gc first finishes checking
92 * of nodes.
Thomas Gleixner182ec4e2005-11-07 11:16:07 +000093 + Return -ENOSPC, if the maximum possibly available space is less or equal than
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 * blocksneeded * sector_size.
95 * This blocks endless gc looping on a filesystem, which is nearly full, even if
96 * the check above passes.
97 */
98 avail = c->free_size + c->dirty_size + c->erasing_size + c->unchecked_size;
99 if ( (avail / c->sector_size) <= blocksneeded) {
100 if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) {
Artem B. Bityuckiy4132ace2005-05-06 10:30:30 +0100101 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 -0700102 break;
103 }
104
105 D1(printk(KERN_DEBUG "max. available size 0x%08x < blocksneeded * sector_size 0x%08x, returning -ENOSPC\n",
106 avail, blocksneeded * c->sector_size));
107 spin_unlock(&c->erase_completion_lock);
108 up(&c->alloc_sem);
109 return -ENOSPC;
110 }
111
112 up(&c->alloc_sem);
113
114 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",
115 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,
116 c->free_size + c->dirty_size + c->wasted_size + c->used_size + c->erasing_size + c->bad_size, c->flash_size));
117 spin_unlock(&c->erase_completion_lock);
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000118
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 ret = jffs2_garbage_collect_pass(c);
120 if (ret)
121 return ret;
122
123 cond_resched();
124
125 if (signal_pending(current))
126 return -EINTR;
127
128 down(&c->alloc_sem);
129 spin_lock(&c->erase_completion_lock);
130 }
131
David Woodhouse9fe48542006-05-23 00:38:06 +0100132 ret = jffs2_do_reserve_space(c, minsize, len, sumsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 if (ret) {
134 D1(printk(KERN_DEBUG "jffs2_reserve_space: ret is %d\n", ret));
135 }
136 }
137 spin_unlock(&c->erase_completion_lock);
David Woodhouse2f785402006-05-24 02:04:45 +0100138 if (!ret)
David Woodhouse046b8b92006-05-25 01:50:35 +0100139 ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 if (ret)
141 up(&c->alloc_sem);
142 return ret;
143}
144
David Woodhouse9fe48542006-05-23 00:38:06 +0100145int jffs2_reserve_space_gc(struct jffs2_sb_info *c, uint32_t minsize,
146 uint32_t *len, uint32_t sumsize)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147{
148 int ret = -EAGAIN;
149 minsize = PAD(minsize);
150
151 D1(printk(KERN_DEBUG "jffs2_reserve_space_gc(): Requested 0x%x bytes\n", minsize));
152
153 spin_lock(&c->erase_completion_lock);
154 while(ret == -EAGAIN) {
David Woodhouse9fe48542006-05-23 00:38:06 +0100155 ret = jffs2_do_reserve_space(c, minsize, len, sumsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 if (ret) {
157 D1(printk(KERN_DEBUG "jffs2_reserve_space_gc: looping, ret is %d\n", ret));
158 }
159 }
160 spin_unlock(&c->erase_completion_lock);
David Woodhouse2f785402006-05-24 02:04:45 +0100161 if (!ret)
David Woodhouse046b8b92006-05-25 01:50:35 +0100162 ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1);
David Woodhouse2f785402006-05-24 02:04:45 +0100163
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 return ret;
165}
166
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100167
168/* Classify nextblock (clean, dirty of verydirty) and force to select an other one */
169
170static void jffs2_close_nextblock(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171{
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100172
Adrian Hunter99c25942007-03-29 11:00:47 +0300173 if (c->nextblock == NULL) {
174 D1(printk(KERN_DEBUG "jffs2_close_nextblock: Erase block at 0x%08x has already been placed in a list\n",
175 jeb->offset));
176 return;
177 }
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100178 /* Check, if we have a dirty block now, or if it was dirty already */
179 if (ISDIRTY (jeb->wasted_size + jeb->dirty_size)) {
180 c->dirty_size += jeb->wasted_size;
181 c->wasted_size -= jeb->wasted_size;
182 jeb->dirty_size += jeb->wasted_size;
183 jeb->wasted_size = 0;
184 if (VERYDIRTY(c, jeb->dirty_size)) {
185 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",
186 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
187 list_add_tail(&jeb->list, &c->very_dirty_list);
188 } else {
189 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to dirty_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
190 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
191 list_add_tail(&jeb->list, &c->dirty_list);
192 }
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000193 } else {
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100194 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
195 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
196 list_add_tail(&jeb->list, &c->clean_list);
197 }
198 c->nextblock = NULL;
199
200}
201
202/* Select a new jeb for nextblock */
203
204static int jffs2_find_nextblock(struct jffs2_sb_info *c)
205{
206 struct list_head *next;
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000207
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100208 /* Take the next block off the 'free' list */
209
210 if (list_empty(&c->free_list)) {
211
212 if (!c->nr_erasing_blocks &&
213 !list_empty(&c->erasable_list)) {
214 struct jffs2_eraseblock *ejeb;
215
216 ejeb = list_entry(c->erasable_list.next, struct jffs2_eraseblock, list);
Akinobu Mitaf1166292006-06-26 00:24:46 -0700217 list_move_tail(&ejeb->list, &c->erase_pending_list);
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100218 c->nr_erasing_blocks++;
219 jffs2_erase_pending_trigger(c);
220 D1(printk(KERN_DEBUG "jffs2_find_nextblock: Triggering erase of erasable block at 0x%08x\n",
221 ejeb->offset));
222 }
223
224 if (!c->nr_erasing_blocks &&
225 !list_empty(&c->erasable_pending_wbuf_list)) {
226 D1(printk(KERN_DEBUG "jffs2_find_nextblock: Flushing write buffer\n"));
227 /* c->nextblock is NULL, no update to c->nextblock allowed */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 spin_unlock(&c->erase_completion_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 jffs2_flush_wbuf_pad(c);
230 spin_lock(&c->erase_completion_lock);
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100231 /* Have another go. It'll be on the erasable_list now */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 return -EAGAIN;
233 }
234
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100235 if (!c->nr_erasing_blocks) {
236 /* Ouch. We're in GC, or we wouldn't have got here.
237 And there's no space left. At all. */
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000238 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",
239 c->nr_erasing_blocks, c->nr_free_blocks, list_empty(&c->erasable_list)?"yes":"no",
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100240 list_empty(&c->erasing_list)?"yes":"no", list_empty(&c->erase_pending_list)?"yes":"no");
241 return -ENOSPC;
242 }
243
244 spin_unlock(&c->erase_completion_lock);
245 /* Don't wait for it; just erase one right now */
246 jffs2_erase_pending_blocks(c, 1);
247 spin_lock(&c->erase_completion_lock);
248
249 /* An erase may have failed, decreasing the
250 amount of free space available. So we must
251 restart from the beginning */
252 return -EAGAIN;
253 }
254
255 next = c->free_list.next;
256 list_del(next);
257 c->nextblock = list_entry(next, struct jffs2_eraseblock, list);
258 c->nr_free_blocks--;
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000259
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100260 jffs2_sum_reset_collected(c->summary); /* reset collected summary */
261
262 D1(printk(KERN_DEBUG "jffs2_find_nextblock(): new nextblock = 0x%08x\n", c->nextblock->offset));
263
264 return 0;
265}
266
267/* Called with alloc sem _and_ erase_completion_lock */
David Woodhouse9fe48542006-05-23 00:38:06 +0100268static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
269 uint32_t *len, uint32_t sumsize)
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100270{
271 struct jffs2_eraseblock *jeb = c->nextblock;
David Woodhouse9fe48542006-05-23 00:38:06 +0100272 uint32_t reserved_size; /* for summary information at the end of the jeb */
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100273 int ret;
274
275 restart:
276 reserved_size = 0;
277
278 if (jffs2_sum_active() && (sumsize != JFFS2_SUMMARY_NOSUM_SIZE)) {
279 /* NOSUM_SIZE means not to generate summary */
280
281 if (jeb) {
282 reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE);
Artem B. Bityutskiy733802d2005-09-22 12:25:00 +0100283 dbg_summary("minsize=%d , jeb->free=%d ,"
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100284 "summary->size=%d , sumsize=%d\n",
285 minsize, jeb->free_size,
286 c->summary->sum_size, sumsize);
287 }
288
289 /* Is there enough space for writing out the current node, or we have to
290 write out summary information now, close this jeb and select new nextblock? */
291 if (jeb && (PAD(minsize) + PAD(c->summary->sum_size + sumsize +
292 JFFS2_SUMMARY_FRAME_SIZE) > jeb->free_size)) {
293
294 /* Has summary been disabled for this jeb? */
295 if (jffs2_sum_is_disabled(c->summary)) {
296 sumsize = JFFS2_SUMMARY_NOSUM_SIZE;
297 goto restart;
298 }
299
300 /* Writing out the collected summary information */
Artem B. Bityutskiy733802d2005-09-22 12:25:00 +0100301 dbg_summary("generating summary for 0x%08x.\n", jeb->offset);
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100302 ret = jffs2_sum_write_sumnode(c);
303
304 if (ret)
305 return ret;
306
307 if (jffs2_sum_is_disabled(c->summary)) {
308 /* jffs2_write_sumnode() couldn't write out the summary information
309 diabling summary for this jeb and free the collected information
310 */
311 sumsize = JFFS2_SUMMARY_NOSUM_SIZE;
312 goto restart;
313 }
314
315 jffs2_close_nextblock(c, jeb);
316 jeb = NULL;
Ferenc Havasi34c0e902005-09-16 13:58:20 +0100317 /* keep always valid value in reserved_size */
318 reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE);
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100319 }
320 } else {
321 if (jeb && minsize > jeb->free_size) {
David Woodhousefc6612f2006-06-18 18:35:10 +0100322 uint32_t waste;
323
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100324 /* Skip the end of this block and file it as having some dirty space */
325 /* If there's a pending write to it, flush now */
326
327 if (jffs2_wbuf_dirty(c)) {
328 spin_unlock(&c->erase_completion_lock);
329 D1(printk(KERN_DEBUG "jffs2_do_reserve_space: Flushing write buffer\n"));
330 jffs2_flush_wbuf_pad(c);
331 spin_lock(&c->erase_completion_lock);
332 jeb = c->nextblock;
333 goto restart;
334 }
335
David Woodhousefc6612f2006-06-18 18:35:10 +0100336 spin_unlock(&c->erase_completion_lock);
337
338 ret = jffs2_prealloc_raw_node_refs(c, jeb, 1);
339 if (ret)
340 return ret;
341 /* Just lock it again and continue. Nothing much can change because
342 we hold c->alloc_sem anyway. In fact, it's not entirely clear why
343 we hold c->erase_completion_lock in the majority of this function...
344 but that's a question for another (more caffeine-rich) day. */
345 spin_lock(&c->erase_completion_lock);
346
347 waste = jeb->free_size;
348 jffs2_link_node_ref(c, jeb,
349 (jeb->offset + c->sector_size - waste) | REF_OBSOLETE,
350 waste, NULL);
351 /* FIXME: that made it count as dirty. Convert to wasted */
352 jeb->dirty_size -= waste;
353 c->dirty_size -= waste;
354 jeb->wasted_size += waste;
355 c->wasted_size += waste;
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100356
357 jffs2_close_nextblock(c, jeb);
358 jeb = NULL;
359 }
360 }
361
362 if (!jeb) {
363
364 ret = jffs2_find_nextblock(c);
365 if (ret)
366 return ret;
367
368 jeb = c->nextblock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369
370 if (jeb->free_size != c->sector_size - c->cleanmarker_size) {
371 printk(KERN_WARNING "Eep. Block 0x%08x taken from free_list had free_size of 0x%08x!!\n", jeb->offset, jeb->free_size);
372 goto restart;
373 }
374 }
375 /* OK, jeb (==c->nextblock) is now pointing at a block which definitely has
376 enough space */
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100377 *len = jeb->free_size - reserved_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378
379 if (c->cleanmarker_size && jeb->used_size == c->cleanmarker_size &&
380 !jeb->first_node->next_in_ino) {
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000381 /* Only node in it beforehand was a CLEANMARKER node (we think).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 So mark it obsolete now that there's going to be another node
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000383 in the block. This will reduce used_size to zero but We've
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 already set c->nextblock so that jffs2_mark_node_obsolete()
385 won't try to refile it to the dirty_list.
386 */
387 spin_unlock(&c->erase_completion_lock);
388 jffs2_mark_node_obsolete(c, jeb->first_node);
389 spin_lock(&c->erase_completion_lock);
390 }
391
David Woodhouse9fe48542006-05-23 00:38:06 +0100392 D1(printk(KERN_DEBUG "jffs2_do_reserve_space(): Giving 0x%x bytes at 0x%x\n",
393 *len, jeb->offset + (c->sector_size - jeb->free_size)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 return 0;
395}
396
397/**
398 * jffs2_add_physical_node_ref - add a physical node reference to the list
399 * @c: superblock info
400 * @new: new node reference to add
401 * @len: length of this physical node
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 *
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000403 * Should only be used to report nodes for which space has been allocated
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 * by jffs2_reserve_space.
405 *
406 * Must be called with the alloc_sem held.
407 */
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000408
David Woodhouse2f785402006-05-24 02:04:45 +0100409struct jffs2_raw_node_ref *jffs2_add_physical_node_ref(struct jffs2_sb_info *c,
410 uint32_t ofs, uint32_t len,
411 struct jffs2_inode_cache *ic)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412{
413 struct jffs2_eraseblock *jeb;
David Woodhouse2f785402006-05-24 02:04:45 +0100414 struct jffs2_raw_node_ref *new;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415
David Woodhouse2f785402006-05-24 02:04:45 +0100416 jeb = &c->blocks[ofs / c->sector_size];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417
David Woodhouse2f785402006-05-24 02:04:45 +0100418 D1(printk(KERN_DEBUG "jffs2_add_physical_node_ref(): Node at 0x%x(%d), size 0x%x\n",
419 ofs & ~3, ofs & 3, len));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420#if 1
David Woodhouse2f785402006-05-24 02:04:45 +0100421 /* Allow non-obsolete nodes only to be added at the end of c->nextblock,
422 if c->nextblock is set. Note that wbuf.c will file obsolete nodes
423 even after refiling c->nextblock */
424 if ((c->nextblock || ((ofs & 3) != REF_OBSOLETE))
425 && (jeb != c->nextblock || (ofs & ~3) != jeb->offset + (c->sector_size - jeb->free_size))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 printk(KERN_WARNING "argh. node added in wrong place\n");
David Woodhouse2f785402006-05-24 02:04:45 +0100427 return ERR_PTR(-EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428 }
429#endif
430 spin_lock(&c->erase_completion_lock);
431
David Woodhouse2f785402006-05-24 02:04:45 +0100432 new = jffs2_link_node_ref(c, jeb, ofs, len, ic);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433
Estelle Hammache9b88f472005-01-28 18:53:05 +0000434 if (!jeb->free_size && !jeb->dirty_size && !ISDIRTY(jeb->wasted_size)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 /* If it lives on the dirty_list, jffs2_reserve_space will put it there */
436 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
437 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
438 if (jffs2_wbuf_dirty(c)) {
439 /* Flush the last write in the block if it's outstanding */
440 spin_unlock(&c->erase_completion_lock);
441 jffs2_flush_wbuf_pad(c);
442 spin_lock(&c->erase_completion_lock);
443 }
444
445 list_add_tail(&jeb->list, &c->clean_list);
446 c->nextblock = NULL;
447 }
Artem B. Bityutskiye0c8e422005-07-24 16:14:17 +0100448 jffs2_dbg_acct_sanity_check_nolock(c,jeb);
449 jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450
451 spin_unlock(&c->erase_completion_lock);
452
David Woodhouse2f785402006-05-24 02:04:45 +0100453 return new;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454}
455
456
457void jffs2_complete_reservation(struct jffs2_sb_info *c)
458{
459 D1(printk(KERN_DEBUG "jffs2_complete_reservation()\n"));
460 jffs2_garbage_collect_trigger(c);
461 up(&c->alloc_sem);
462}
463
464static inline int on_list(struct list_head *obj, struct list_head *head)
465{
466 struct list_head *this;
467
468 list_for_each(this, head) {
469 if (this == obj) {
470 D1(printk("%p is on list at %p\n", obj, head));
471 return 1;
472
473 }
474 }
475 return 0;
476}
477
478void jffs2_mark_node_obsolete(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref)
479{
480 struct jffs2_eraseblock *jeb;
481 int blocknr;
482 struct jffs2_unknown_node n;
483 int ret, addedsize;
484 size_t retlen;
David Woodhouse1417fc42006-05-20 16:20:19 +0100485 uint32_t freed_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486
David Woodhouse9bfeb692006-05-26 21:19:05 +0100487 if(unlikely(!ref)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 printk(KERN_NOTICE "EEEEEK. jffs2_mark_node_obsolete called with NULL node\n");
489 return;
490 }
491 if (ref_obsolete(ref)) {
492 D1(printk(KERN_DEBUG "jffs2_mark_node_obsolete called with already obsolete node at 0x%08x\n", ref_offset(ref)));
493 return;
494 }
495 blocknr = ref->flash_offset / c->sector_size;
496 if (blocknr >= c->nr_blocks) {
497 printk(KERN_NOTICE "raw node at 0x%08x is off the end of device!\n", ref->flash_offset);
498 BUG();
499 }
500 jeb = &c->blocks[blocknr];
501
502 if (jffs2_can_mark_obsolete(c) && !jffs2_is_readonly(c) &&
Artem B. Bityuckiy31fbdf72005-02-28 08:21:09 +0000503 !(c->flags & (JFFS2_SB_FLAG_SCANNING | JFFS2_SB_FLAG_BUILDING))) {
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000504 /* Hm. This may confuse static lock analysis. If any of the above
505 three conditions is false, we're going to return from this
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 function without actually obliterating any nodes or freeing
507 any jffs2_raw_node_refs. So we don't need to stop erases from
508 happening, or protect against people holding an obsolete
509 jffs2_raw_node_ref without the erase_completion_lock. */
510 down(&c->erase_free_sem);
511 }
512
513 spin_lock(&c->erase_completion_lock);
514
David Woodhouse1417fc42006-05-20 16:20:19 +0100515 freed_len = ref_totlen(c, jeb, ref);
516
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 if (ref_flags(ref) == REF_UNCHECKED) {
David Woodhouse1417fc42006-05-20 16:20:19 +0100518 D1(if (unlikely(jeb->unchecked_size < freed_len)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 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 +0100520 freed_len, blocknr, ref->flash_offset, jeb->used_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 BUG();
522 })
David Woodhouse1417fc42006-05-20 16:20:19 +0100523 D1(printk(KERN_DEBUG "Obsoleting previously unchecked node at 0x%08x of len %x: ", ref_offset(ref), freed_len));
524 jeb->unchecked_size -= freed_len;
525 c->unchecked_size -= freed_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 } else {
David Woodhouse1417fc42006-05-20 16:20:19 +0100527 D1(if (unlikely(jeb->used_size < freed_len)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 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 +0100529 freed_len, blocknr, ref->flash_offset, jeb->used_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 BUG();
531 })
David Woodhouse1417fc42006-05-20 16:20:19 +0100532 D1(printk(KERN_DEBUG "Obsoleting node at 0x%08x of len %#x: ", ref_offset(ref), freed_len));
533 jeb->used_size -= freed_len;
534 c->used_size -= freed_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 }
536
537 // Take care, that wasted size is taken into concern
David Woodhouse1417fc42006-05-20 16:20:19 +0100538 if ((jeb->dirty_size || ISDIRTY(jeb->wasted_size + freed_len)) && jeb != c->nextblock) {
David Woodhousec7c16c82006-05-24 14:24:02 +0100539 D1(printk("Dirtying\n"));
David Woodhouse1417fc42006-05-20 16:20:19 +0100540 addedsize = freed_len;
541 jeb->dirty_size += freed_len;
542 c->dirty_size += freed_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543
544 /* Convert wasted space to dirty, if not a bad block */
545 if (jeb->wasted_size) {
546 if (on_list(&jeb->list, &c->bad_used_list)) {
547 D1(printk(KERN_DEBUG "Leaving block at %08x on the bad_used_list\n",
548 jeb->offset));
549 addedsize = 0; /* To fool the refiling code later */
550 } else {
551 D1(printk(KERN_DEBUG "Converting %d bytes of wasted space to dirty in block at %08x\n",
552 jeb->wasted_size, jeb->offset));
553 addedsize += jeb->wasted_size;
554 jeb->dirty_size += jeb->wasted_size;
555 c->dirty_size += jeb->wasted_size;
556 c->wasted_size -= jeb->wasted_size;
557 jeb->wasted_size = 0;
558 }
559 }
560 } else {
David Woodhousec7c16c82006-05-24 14:24:02 +0100561 D1(printk("Wasting\n"));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 addedsize = 0;
David Woodhouse1417fc42006-05-20 16:20:19 +0100563 jeb->wasted_size += freed_len;
564 c->wasted_size += freed_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 }
566 ref->flash_offset = ref_offset(ref) | REF_OBSOLETE;
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000567
Artem B. Bityutskiye0c8e422005-07-24 16:14:17 +0100568 jffs2_dbg_acct_sanity_check_nolock(c, jeb);
569 jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570
Artem B. Bityuckiy31fbdf72005-02-28 08:21:09 +0000571 if (c->flags & JFFS2_SB_FLAG_SCANNING) {
572 /* Flash scanning is in progress. Don't muck about with the block
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 lists because they're not ready yet, and don't actually
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000574 obliterate nodes that look obsolete. If they weren't
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 marked obsolete on the flash at the time they _became_
576 obsolete, there was probably a reason for that. */
577 spin_unlock(&c->erase_completion_lock);
578 /* We didn't lock the erase_free_sem */
579 return;
580 }
581
582 if (jeb == c->nextblock) {
583 D2(printk(KERN_DEBUG "Not moving nextblock 0x%08x to dirty/erase_pending list\n", jeb->offset));
584 } else if (!jeb->used_size && !jeb->unchecked_size) {
585 if (jeb == c->gcblock) {
586 D1(printk(KERN_DEBUG "gcblock at 0x%08x completely dirtied. Clearing gcblock...\n", jeb->offset));
587 c->gcblock = NULL;
588 } else {
589 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x completely dirtied. Removing from (dirty?) list...\n", jeb->offset));
590 list_del(&jeb->list);
591 }
592 if (jffs2_wbuf_dirty(c)) {
593 D1(printk(KERN_DEBUG "...and adding to erasable_pending_wbuf_list\n"));
594 list_add_tail(&jeb->list, &c->erasable_pending_wbuf_list);
595 } else {
596 if (jiffies & 127) {
597 /* Most of the time, we just erase it immediately. Otherwise we
598 spend ages scanning it on mount, etc. */
599 D1(printk(KERN_DEBUG "...and adding to erase_pending_list\n"));
600 list_add_tail(&jeb->list, &c->erase_pending_list);
601 c->nr_erasing_blocks++;
602 jffs2_erase_pending_trigger(c);
603 } else {
604 /* Sometimes, however, we leave it elsewhere so it doesn't get
605 immediately reused, and we spread the load a bit. */
606 D1(printk(KERN_DEBUG "...and adding to erasable_list\n"));
607 list_add_tail(&jeb->list, &c->erasable_list);
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000608 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 }
610 D1(printk(KERN_DEBUG "Done OK\n"));
611 } else if (jeb == c->gcblock) {
612 D2(printk(KERN_DEBUG "Not moving gcblock 0x%08x to dirty_list\n", jeb->offset));
613 } else if (ISDIRTY(jeb->dirty_size) && !ISDIRTY(jeb->dirty_size - addedsize)) {
614 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x is freshly dirtied. Removing from clean list...\n", jeb->offset));
615 list_del(&jeb->list);
616 D1(printk(KERN_DEBUG "...and adding to dirty_list\n"));
617 list_add_tail(&jeb->list, &c->dirty_list);
618 } else if (VERYDIRTY(c, jeb->dirty_size) &&
619 !VERYDIRTY(c, jeb->dirty_size - addedsize)) {
620 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x is now very dirty. Removing from dirty list...\n", jeb->offset));
621 list_del(&jeb->list);
622 D1(printk(KERN_DEBUG "...and adding to very_dirty_list\n"));
623 list_add_tail(&jeb->list, &c->very_dirty_list);
624 } else {
625 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 +0000626 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
627 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628
629 spin_unlock(&c->erase_completion_lock);
630
Artem B. Bityuckiy31fbdf72005-02-28 08:21:09 +0000631 if (!jffs2_can_mark_obsolete(c) || jffs2_is_readonly(c) ||
632 (c->flags & JFFS2_SB_FLAG_BUILDING)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 /* We didn't lock the erase_free_sem */
634 return;
635 }
636
637 /* The erase_free_sem is locked, and has been since before we marked the node obsolete
638 and potentially put its eraseblock onto the erase_pending_list. Thus, we know that
639 the block hasn't _already_ been erased, and that 'ref' itself hasn't been freed yet
David Woodhousec38c1b62006-05-25 01:38:27 +0100640 by jffs2_free_jeb_node_refs() in erase.c. Which is nice. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641
642 D1(printk(KERN_DEBUG "obliterating obsoleted node at 0x%08x\n", ref_offset(ref)));
643 ret = jffs2_flash_read(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
644 if (ret) {
645 printk(KERN_WARNING "Read error reading from obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret);
646 goto out_erase_sem;
647 }
648 if (retlen != sizeof(n)) {
649 printk(KERN_WARNING "Short read from obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen);
650 goto out_erase_sem;
651 }
David Woodhouse1417fc42006-05-20 16:20:19 +0100652 if (PAD(je32_to_cpu(n.totlen)) != PAD(freed_len)) {
653 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 -0700654 goto out_erase_sem;
655 }
656 if (!(je16_to_cpu(n.nodetype) & JFFS2_NODE_ACCURATE)) {
657 D1(printk(KERN_DEBUG "Node at 0x%08x was already marked obsolete (nodetype 0x%04x)\n", ref_offset(ref), je16_to_cpu(n.nodetype)));
658 goto out_erase_sem;
659 }
660 /* XXX FIXME: This is ugly now */
661 n.nodetype = cpu_to_je16(je16_to_cpu(n.nodetype) & ~JFFS2_NODE_ACCURATE);
662 ret = jffs2_flash_write(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
663 if (ret) {
664 printk(KERN_WARNING "Write error in obliterating obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret);
665 goto out_erase_sem;
666 }
667 if (retlen != sizeof(n)) {
668 printk(KERN_WARNING "Short write in obliterating obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen);
669 goto out_erase_sem;
670 }
671
672 /* Nodes which have been marked obsolete no longer need to be
673 associated with any inode. Remove them from the per-inode list.
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000674
675 Note we can't do this for NAND at the moment because we need
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 obsolete dirent nodes to stay on the lists, because of the
677 horridness in jffs2_garbage_collect_deletion_dirent(). Also
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000678 because we delete the inocache, and on NAND we need that to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 stay around until all the nodes are actually erased, in order
680 to stop us from giving the same inode number to another newly
681 created inode. */
682 if (ref->next_in_ino) {
683 struct jffs2_inode_cache *ic;
684 struct jffs2_raw_node_ref **p;
685
686 spin_lock(&c->erase_completion_lock);
687
688 ic = jffs2_raw_ref_to_ic(ref);
689 for (p = &ic->nodes; (*p) != ref; p = &((*p)->next_in_ino))
690 ;
691
692 *p = ref->next_in_ino;
693 ref->next_in_ino = NULL;
694
KaiGai Koheic9f700f2006-06-11 10:35:15 +0900695 switch (ic->class) {
696#ifdef CONFIG_JFFS2_FS_XATTR
697 case RAWNODE_CLASS_XATTR_DATUM:
698 jffs2_release_xattr_datum(c, (struct jffs2_xattr_datum *)ic);
699 break;
700 case RAWNODE_CLASS_XATTR_REF:
701 jffs2_release_xattr_ref(c, (struct jffs2_xattr_ref *)ic);
702 break;
703#endif
704 default:
705 if (ic->nodes == (void *)ic && ic->nlink == 0)
706 jffs2_del_ino_cache(c, ic);
707 break;
708 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 spin_unlock(&c->erase_completion_lock);
710 }
711
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 out_erase_sem:
713 up(&c->erase_free_sem);
714}
715
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716int jffs2_thread_should_wake(struct jffs2_sb_info *c)
717{
718 int ret = 0;
719 uint32_t dirty;
720
721 if (c->unchecked_size) {
722 D1(printk(KERN_DEBUG "jffs2_thread_should_wake(): unchecked_size %d, checked_ino #%d\n",
723 c->unchecked_size, c->checked_ino));
724 return 1;
725 }
726
727 /* dirty_size contains blocks on erase_pending_list
728 * those blocks are counted in c->nr_erasing_blocks.
729 * If one block is actually erased, it is not longer counted as dirty_space
730 * but it is counted in c->nr_erasing_blocks, so we add it and subtract it
731 * with c->nr_erasing_blocks * c->sector_size again.
732 * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks
733 * This helps us to force gc and pick eventually a clean block to spread the load.
734 */
735 dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size;
736
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000737 if (c->nr_free_blocks + c->nr_erasing_blocks < c->resv_blocks_gctrigger &&
738 (dirty > c->nospc_dirty_size))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 ret = 1;
740
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000741 D1(printk(KERN_DEBUG "jffs2_thread_should_wake(): nr_free_blocks %d, nr_erasing_blocks %d, dirty_size 0x%x: %s\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 c->nr_free_blocks, c->nr_erasing_blocks, c->dirty_size, ret?"yes":"no"));
743
744 return ret;
745}