blob: d88376992ed9b094cc23f7b75f7b8c07870468ba [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * JFFS2 -- Journalling Flash File System, Version 2.
3 *
4 * Copyright (C) 2001-2003 Red Hat, Inc.
5 *
6 * Created by David Woodhouse <dwmw2@infradead.org>
7 *
8 * For licensing information, see the file 'LICENCE' in this directory.
9 *
Artem B. Bityutskiy733802d2005-09-22 12:25:00 +010010 * $Id: nodemgmt.c,v 1.127 2005/09/20 15:49:12 dedekind Exp $
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 *
12 */
13
14#include <linux/kernel.h>
15#include <linux/slab.h>
16#include <linux/mtd/mtd.h>
17#include <linux/compiler.h>
18#include <linux/sched.h> /* For cond_resched() */
19#include "nodelist.h"
Ferenc Havasie631ddb2005-09-07 09:35:26 +010020#include "debug.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070021
22/**
23 * jffs2_reserve_space - request physical space to write nodes to flash
24 * @c: superblock info
25 * @minsize: Minimum acceptable size of allocation
Linus Torvalds1da177e2005-04-16 15:20:36 -070026 * @len: Returned value of allocation length
27 * @prio: Allocation type - ALLOC_{NORMAL,DELETION}
28 *
29 * Requests a block of physical space on the flash. Returns zero for success
David Woodhouse9fe48542006-05-23 00:38:06 +010030 * and puts 'len' into the appropriate place, or returns -ENOSPC or other
31 * error if appropriate. Doesn't return len since that's
Linus Torvalds1da177e2005-04-16 15:20:36 -070032 *
33 * If it returns zero, jffs2_reserve_space() also downs the per-filesystem
34 * allocation semaphore, to prevent more than one allocation from being
35 * active at any time. The semaphore is later released by jffs2_commit_allocation()
36 *
37 * jffs2_reserve_space() may trigger garbage collection in order to make room
38 * for the requested allocation.
39 */
40
Ferenc Havasie631ddb2005-09-07 09:35:26 +010041static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
David Woodhouse9fe48542006-05-23 00:38:06 +010042 uint32_t *len, uint32_t sumsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
David Woodhouse9fe48542006-05-23 00:38:06 +010044int jffs2_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
Ferenc Havasie631ddb2005-09-07 09:35:26 +010045 uint32_t *len, int prio, uint32_t sumsize)
Linus Torvalds1da177e2005-04-16 15:20:36 -070046{
47 int ret = -EAGAIN;
48 int blocksneeded = c->resv_blocks_write;
49 /* align it */
50 minsize = PAD(minsize);
51
52 D1(printk(KERN_DEBUG "jffs2_reserve_space(): Requested 0x%x bytes\n", minsize));
53 down(&c->alloc_sem);
54
55 D1(printk(KERN_DEBUG "jffs2_reserve_space(): alloc sem got\n"));
56
57 spin_lock(&c->erase_completion_lock);
58
59 /* this needs a little more thought (true <tglx> :)) */
60 while(ret == -EAGAIN) {
61 while(c->nr_free_blocks + c->nr_erasing_blocks < blocksneeded) {
62 int ret;
63 uint32_t dirty, avail;
64
65 /* calculate real dirty size
66 * dirty_size contains blocks on erase_pending_list
67 * those blocks are counted in c->nr_erasing_blocks.
68 * If one block is actually erased, it is not longer counted as dirty_space
69 * but it is counted in c->nr_erasing_blocks, so we add it and subtract it
70 * with c->nr_erasing_blocks * c->sector_size again.
71 * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks
72 * This helps us to force gc and pick eventually a clean block to spread the load.
73 * We add unchecked_size here, as we hopefully will find some space to use.
74 * This will affect the sum only once, as gc first finishes checking
75 * of nodes.
76 */
77 dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size + c->unchecked_size;
78 if (dirty < c->nospc_dirty_size) {
79 if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) {
Artem B. Bityuckiy4132ace2005-05-06 10:30:30 +010080 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 -070081 break;
82 }
83 D1(printk(KERN_DEBUG "dirty size 0x%08x + unchecked_size 0x%08x < nospc_dirty_size 0x%08x, returning -ENOSPC\n",
84 dirty, c->unchecked_size, c->sector_size));
85
86 spin_unlock(&c->erase_completion_lock);
87 up(&c->alloc_sem);
88 return -ENOSPC;
89 }
Thomas Gleixner182ec4e2005-11-07 11:16:07 +000090
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 /* Calc possibly available space. Possibly available means that we
92 * don't know, if unchecked size contains obsoleted nodes, which could give us some
93 * more usable space. This will affect the sum only once, as gc first finishes checking
94 * of nodes.
Thomas Gleixner182ec4e2005-11-07 11:16:07 +000095 + Return -ENOSPC, if the maximum possibly available space is less or equal than
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 * blocksneeded * sector_size.
97 * This blocks endless gc looping on a filesystem, which is nearly full, even if
98 * the check above passes.
99 */
100 avail = c->free_size + c->dirty_size + c->erasing_size + c->unchecked_size;
101 if ( (avail / c->sector_size) <= blocksneeded) {
102 if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) {
Artem B. Bityuckiy4132ace2005-05-06 10:30:30 +0100103 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 -0700104 break;
105 }
106
107 D1(printk(KERN_DEBUG "max. available size 0x%08x < blocksneeded * sector_size 0x%08x, returning -ENOSPC\n",
108 avail, blocksneeded * c->sector_size));
109 spin_unlock(&c->erase_completion_lock);
110 up(&c->alloc_sem);
111 return -ENOSPC;
112 }
113
114 up(&c->alloc_sem);
115
116 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",
117 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,
118 c->free_size + c->dirty_size + c->wasted_size + c->used_size + c->erasing_size + c->bad_size, c->flash_size));
119 spin_unlock(&c->erase_completion_lock);
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000120
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121 ret = jffs2_garbage_collect_pass(c);
122 if (ret)
123 return ret;
124
125 cond_resched();
126
127 if (signal_pending(current))
128 return -EINTR;
129
130 down(&c->alloc_sem);
131 spin_lock(&c->erase_completion_lock);
132 }
133
David Woodhouse9fe48542006-05-23 00:38:06 +0100134 ret = jffs2_do_reserve_space(c, minsize, len, sumsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135 if (ret) {
136 D1(printk(KERN_DEBUG "jffs2_reserve_space: ret is %d\n", ret));
137 }
138 }
139 spin_unlock(&c->erase_completion_lock);
David Woodhouse2f785402006-05-24 02:04:45 +0100140 if (!ret)
David Woodhouse046b8b92006-05-25 01:50:35 +0100141 ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 if (ret)
143 up(&c->alloc_sem);
144 return ret;
145}
146
David Woodhouse9fe48542006-05-23 00:38:06 +0100147int jffs2_reserve_space_gc(struct jffs2_sb_info *c, uint32_t minsize,
148 uint32_t *len, uint32_t sumsize)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149{
150 int ret = -EAGAIN;
151 minsize = PAD(minsize);
152
153 D1(printk(KERN_DEBUG "jffs2_reserve_space_gc(): Requested 0x%x bytes\n", minsize));
154
155 spin_lock(&c->erase_completion_lock);
156 while(ret == -EAGAIN) {
David Woodhouse9fe48542006-05-23 00:38:06 +0100157 ret = jffs2_do_reserve_space(c, minsize, len, sumsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 if (ret) {
159 D1(printk(KERN_DEBUG "jffs2_reserve_space_gc: looping, ret is %d\n", ret));
160 }
161 }
162 spin_unlock(&c->erase_completion_lock);
David Woodhouse2f785402006-05-24 02:04:45 +0100163 if (!ret)
David Woodhouse046b8b92006-05-25 01:50:35 +0100164 ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1);
David Woodhouse2f785402006-05-24 02:04:45 +0100165
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 return ret;
167}
168
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100169
170/* Classify nextblock (clean, dirty of verydirty) and force to select an other one */
171
172static void jffs2_close_nextblock(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173{
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100174
175 /* Check, if we have a dirty block now, or if it was dirty already */
176 if (ISDIRTY (jeb->wasted_size + jeb->dirty_size)) {
177 c->dirty_size += jeb->wasted_size;
178 c->wasted_size -= jeb->wasted_size;
179 jeb->dirty_size += jeb->wasted_size;
180 jeb->wasted_size = 0;
181 if (VERYDIRTY(c, jeb->dirty_size)) {
182 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",
183 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
184 list_add_tail(&jeb->list, &c->very_dirty_list);
185 } else {
186 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to dirty_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
187 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
188 list_add_tail(&jeb->list, &c->dirty_list);
189 }
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000190 } else {
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100191 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
192 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
193 list_add_tail(&jeb->list, &c->clean_list);
194 }
195 c->nextblock = NULL;
196
197}
198
199/* Select a new jeb for nextblock */
200
201static int jffs2_find_nextblock(struct jffs2_sb_info *c)
202{
203 struct list_head *next;
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000204
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100205 /* Take the next block off the 'free' list */
206
207 if (list_empty(&c->free_list)) {
208
209 if (!c->nr_erasing_blocks &&
210 !list_empty(&c->erasable_list)) {
211 struct jffs2_eraseblock *ejeb;
212
213 ejeb = list_entry(c->erasable_list.next, struct jffs2_eraseblock, list);
Akinobu Mitaf1166292006-06-26 00:24:46 -0700214 list_move_tail(&ejeb->list, &c->erase_pending_list);
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100215 c->nr_erasing_blocks++;
216 jffs2_erase_pending_trigger(c);
217 D1(printk(KERN_DEBUG "jffs2_find_nextblock: Triggering erase of erasable block at 0x%08x\n",
218 ejeb->offset));
219 }
220
221 if (!c->nr_erasing_blocks &&
222 !list_empty(&c->erasable_pending_wbuf_list)) {
223 D1(printk(KERN_DEBUG "jffs2_find_nextblock: Flushing write buffer\n"));
224 /* c->nextblock is NULL, no update to c->nextblock allowed */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 spin_unlock(&c->erase_completion_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 jffs2_flush_wbuf_pad(c);
227 spin_lock(&c->erase_completion_lock);
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100228 /* Have another go. It'll be on the erasable_list now */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 return -EAGAIN;
230 }
231
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100232 if (!c->nr_erasing_blocks) {
233 /* Ouch. We're in GC, or we wouldn't have got here.
234 And there's no space left. At all. */
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000235 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",
236 c->nr_erasing_blocks, c->nr_free_blocks, list_empty(&c->erasable_list)?"yes":"no",
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100237 list_empty(&c->erasing_list)?"yes":"no", list_empty(&c->erase_pending_list)?"yes":"no");
238 return -ENOSPC;
239 }
240
241 spin_unlock(&c->erase_completion_lock);
242 /* Don't wait for it; just erase one right now */
243 jffs2_erase_pending_blocks(c, 1);
244 spin_lock(&c->erase_completion_lock);
245
246 /* An erase may have failed, decreasing the
247 amount of free space available. So we must
248 restart from the beginning */
249 return -EAGAIN;
250 }
251
252 next = c->free_list.next;
253 list_del(next);
254 c->nextblock = list_entry(next, struct jffs2_eraseblock, list);
255 c->nr_free_blocks--;
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000256
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100257 jffs2_sum_reset_collected(c->summary); /* reset collected summary */
258
259 D1(printk(KERN_DEBUG "jffs2_find_nextblock(): new nextblock = 0x%08x\n", c->nextblock->offset));
260
261 return 0;
262}
263
264/* Called with alloc sem _and_ erase_completion_lock */
David Woodhouse9fe48542006-05-23 00:38:06 +0100265static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
266 uint32_t *len, uint32_t sumsize)
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100267{
268 struct jffs2_eraseblock *jeb = c->nextblock;
David Woodhouse9fe48542006-05-23 00:38:06 +0100269 uint32_t reserved_size; /* for summary information at the end of the jeb */
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100270 int ret;
271
272 restart:
273 reserved_size = 0;
274
275 if (jffs2_sum_active() && (sumsize != JFFS2_SUMMARY_NOSUM_SIZE)) {
276 /* NOSUM_SIZE means not to generate summary */
277
278 if (jeb) {
279 reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE);
Artem B. Bityutskiy733802d2005-09-22 12:25:00 +0100280 dbg_summary("minsize=%d , jeb->free=%d ,"
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100281 "summary->size=%d , sumsize=%d\n",
282 minsize, jeb->free_size,
283 c->summary->sum_size, sumsize);
284 }
285
286 /* Is there enough space for writing out the current node, or we have to
287 write out summary information now, close this jeb and select new nextblock? */
288 if (jeb && (PAD(minsize) + PAD(c->summary->sum_size + sumsize +
289 JFFS2_SUMMARY_FRAME_SIZE) > jeb->free_size)) {
290
291 /* Has summary been disabled for this jeb? */
292 if (jffs2_sum_is_disabled(c->summary)) {
293 sumsize = JFFS2_SUMMARY_NOSUM_SIZE;
294 goto restart;
295 }
296
297 /* Writing out the collected summary information */
Artem B. Bityutskiy733802d2005-09-22 12:25:00 +0100298 dbg_summary("generating summary for 0x%08x.\n", jeb->offset);
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100299 ret = jffs2_sum_write_sumnode(c);
300
301 if (ret)
302 return ret;
303
304 if (jffs2_sum_is_disabled(c->summary)) {
305 /* jffs2_write_sumnode() couldn't write out the summary information
306 diabling summary for this jeb and free the collected information
307 */
308 sumsize = JFFS2_SUMMARY_NOSUM_SIZE;
309 goto restart;
310 }
311
312 jffs2_close_nextblock(c, jeb);
313 jeb = NULL;
Ferenc Havasi34c0e902005-09-16 13:58:20 +0100314 /* keep always valid value in reserved_size */
315 reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE);
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100316 }
317 } else {
318 if (jeb && minsize > jeb->free_size) {
David Woodhousefc6612f2006-06-18 18:35:10 +0100319 uint32_t waste;
320
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100321 /* Skip the end of this block and file it as having some dirty space */
322 /* If there's a pending write to it, flush now */
323
324 if (jffs2_wbuf_dirty(c)) {
325 spin_unlock(&c->erase_completion_lock);
326 D1(printk(KERN_DEBUG "jffs2_do_reserve_space: Flushing write buffer\n"));
327 jffs2_flush_wbuf_pad(c);
328 spin_lock(&c->erase_completion_lock);
329 jeb = c->nextblock;
330 goto restart;
331 }
332
David Woodhousefc6612f2006-06-18 18:35:10 +0100333 spin_unlock(&c->erase_completion_lock);
334
335 ret = jffs2_prealloc_raw_node_refs(c, jeb, 1);
336 if (ret)
337 return ret;
338 /* Just lock it again and continue. Nothing much can change because
339 we hold c->alloc_sem anyway. In fact, it's not entirely clear why
340 we hold c->erase_completion_lock in the majority of this function...
341 but that's a question for another (more caffeine-rich) day. */
342 spin_lock(&c->erase_completion_lock);
343
344 waste = jeb->free_size;
345 jffs2_link_node_ref(c, jeb,
346 (jeb->offset + c->sector_size - waste) | REF_OBSOLETE,
347 waste, NULL);
348 /* FIXME: that made it count as dirty. Convert to wasted */
349 jeb->dirty_size -= waste;
350 c->dirty_size -= waste;
351 jeb->wasted_size += waste;
352 c->wasted_size += waste;
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100353
354 jffs2_close_nextblock(c, jeb);
355 jeb = NULL;
356 }
357 }
358
359 if (!jeb) {
360
361 ret = jffs2_find_nextblock(c);
362 if (ret)
363 return ret;
364
365 jeb = c->nextblock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366
367 if (jeb->free_size != c->sector_size - c->cleanmarker_size) {
368 printk(KERN_WARNING "Eep. Block 0x%08x taken from free_list had free_size of 0x%08x!!\n", jeb->offset, jeb->free_size);
369 goto restart;
370 }
371 }
372 /* OK, jeb (==c->nextblock) is now pointing at a block which definitely has
373 enough space */
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100374 *len = jeb->free_size - reserved_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375
376 if (c->cleanmarker_size && jeb->used_size == c->cleanmarker_size &&
377 !jeb->first_node->next_in_ino) {
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000378 /* Only node in it beforehand was a CLEANMARKER node (we think).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 So mark it obsolete now that there's going to be another node
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000380 in the block. This will reduce used_size to zero but We've
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 already set c->nextblock so that jffs2_mark_node_obsolete()
382 won't try to refile it to the dirty_list.
383 */
384 spin_unlock(&c->erase_completion_lock);
385 jffs2_mark_node_obsolete(c, jeb->first_node);
386 spin_lock(&c->erase_completion_lock);
387 }
388
David Woodhouse9fe48542006-05-23 00:38:06 +0100389 D1(printk(KERN_DEBUG "jffs2_do_reserve_space(): Giving 0x%x bytes at 0x%x\n",
390 *len, jeb->offset + (c->sector_size - jeb->free_size)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 return 0;
392}
393
394/**
395 * jffs2_add_physical_node_ref - add a physical node reference to the list
396 * @c: superblock info
397 * @new: new node reference to add
398 * @len: length of this physical node
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 *
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000400 * Should only be used to report nodes for which space has been allocated
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 * by jffs2_reserve_space.
402 *
403 * Must be called with the alloc_sem held.
404 */
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000405
David Woodhouse2f785402006-05-24 02:04:45 +0100406struct jffs2_raw_node_ref *jffs2_add_physical_node_ref(struct jffs2_sb_info *c,
407 uint32_t ofs, uint32_t len,
408 struct jffs2_inode_cache *ic)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409{
410 struct jffs2_eraseblock *jeb;
David Woodhouse2f785402006-05-24 02:04:45 +0100411 struct jffs2_raw_node_ref *new;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412
David Woodhouse2f785402006-05-24 02:04:45 +0100413 jeb = &c->blocks[ofs / c->sector_size];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414
David Woodhouse2f785402006-05-24 02:04:45 +0100415 D1(printk(KERN_DEBUG "jffs2_add_physical_node_ref(): Node at 0x%x(%d), size 0x%x\n",
416 ofs & ~3, ofs & 3, len));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417#if 1
David Woodhouse2f785402006-05-24 02:04:45 +0100418 /* Allow non-obsolete nodes only to be added at the end of c->nextblock,
419 if c->nextblock is set. Note that wbuf.c will file obsolete nodes
420 even after refiling c->nextblock */
421 if ((c->nextblock || ((ofs & 3) != REF_OBSOLETE))
422 && (jeb != c->nextblock || (ofs & ~3) != jeb->offset + (c->sector_size - jeb->free_size))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 printk(KERN_WARNING "argh. node added in wrong place\n");
David Woodhouse2f785402006-05-24 02:04:45 +0100424 return ERR_PTR(-EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 }
426#endif
427 spin_lock(&c->erase_completion_lock);
428
David Woodhouse2f785402006-05-24 02:04:45 +0100429 new = jffs2_link_node_ref(c, jeb, ofs, len, ic);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430
Estelle Hammache9b88f472005-01-28 18:53:05 +0000431 if (!jeb->free_size && !jeb->dirty_size && !ISDIRTY(jeb->wasted_size)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 /* If it lives on the dirty_list, jffs2_reserve_space will put it there */
433 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
434 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
435 if (jffs2_wbuf_dirty(c)) {
436 /* Flush the last write in the block if it's outstanding */
437 spin_unlock(&c->erase_completion_lock);
438 jffs2_flush_wbuf_pad(c);
439 spin_lock(&c->erase_completion_lock);
440 }
441
442 list_add_tail(&jeb->list, &c->clean_list);
443 c->nextblock = NULL;
444 }
Artem B. Bityutskiye0c8e422005-07-24 16:14:17 +0100445 jffs2_dbg_acct_sanity_check_nolock(c,jeb);
446 jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447
448 spin_unlock(&c->erase_completion_lock);
449
David Woodhouse2f785402006-05-24 02:04:45 +0100450 return new;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451}
452
453
454void jffs2_complete_reservation(struct jffs2_sb_info *c)
455{
456 D1(printk(KERN_DEBUG "jffs2_complete_reservation()\n"));
457 jffs2_garbage_collect_trigger(c);
458 up(&c->alloc_sem);
459}
460
461static inline int on_list(struct list_head *obj, struct list_head *head)
462{
463 struct list_head *this;
464
465 list_for_each(this, head) {
466 if (this == obj) {
467 D1(printk("%p is on list at %p\n", obj, head));
468 return 1;
469
470 }
471 }
472 return 0;
473}
474
475void jffs2_mark_node_obsolete(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref)
476{
477 struct jffs2_eraseblock *jeb;
478 int blocknr;
479 struct jffs2_unknown_node n;
480 int ret, addedsize;
481 size_t retlen;
David Woodhouse1417fc42006-05-20 16:20:19 +0100482 uint32_t freed_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483
David Woodhouse9bfeb692006-05-26 21:19:05 +0100484 if(unlikely(!ref)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 printk(KERN_NOTICE "EEEEEK. jffs2_mark_node_obsolete called with NULL node\n");
486 return;
487 }
488 if (ref_obsolete(ref)) {
489 D1(printk(KERN_DEBUG "jffs2_mark_node_obsolete called with already obsolete node at 0x%08x\n", ref_offset(ref)));
490 return;
491 }
492 blocknr = ref->flash_offset / c->sector_size;
493 if (blocknr >= c->nr_blocks) {
494 printk(KERN_NOTICE "raw node at 0x%08x is off the end of device!\n", ref->flash_offset);
495 BUG();
496 }
497 jeb = &c->blocks[blocknr];
498
499 if (jffs2_can_mark_obsolete(c) && !jffs2_is_readonly(c) &&
Artem B. Bityuckiy31fbdf72005-02-28 08:21:09 +0000500 !(c->flags & (JFFS2_SB_FLAG_SCANNING | JFFS2_SB_FLAG_BUILDING))) {
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000501 /* Hm. This may confuse static lock analysis. If any of the above
502 three conditions is false, we're going to return from this
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 function without actually obliterating any nodes or freeing
504 any jffs2_raw_node_refs. So we don't need to stop erases from
505 happening, or protect against people holding an obsolete
506 jffs2_raw_node_ref without the erase_completion_lock. */
507 down(&c->erase_free_sem);
508 }
509
510 spin_lock(&c->erase_completion_lock);
511
David Woodhouse1417fc42006-05-20 16:20:19 +0100512 freed_len = ref_totlen(c, jeb, ref);
513
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 if (ref_flags(ref) == REF_UNCHECKED) {
David Woodhouse1417fc42006-05-20 16:20:19 +0100515 D1(if (unlikely(jeb->unchecked_size < freed_len)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 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 +0100517 freed_len, blocknr, ref->flash_offset, jeb->used_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 BUG();
519 })
David Woodhouse1417fc42006-05-20 16:20:19 +0100520 D1(printk(KERN_DEBUG "Obsoleting previously unchecked node at 0x%08x of len %x: ", ref_offset(ref), freed_len));
521 jeb->unchecked_size -= freed_len;
522 c->unchecked_size -= freed_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 } else {
David Woodhouse1417fc42006-05-20 16:20:19 +0100524 D1(if (unlikely(jeb->used_size < freed_len)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 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 +0100526 freed_len, blocknr, ref->flash_offset, jeb->used_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 BUG();
528 })
David Woodhouse1417fc42006-05-20 16:20:19 +0100529 D1(printk(KERN_DEBUG "Obsoleting node at 0x%08x of len %#x: ", ref_offset(ref), freed_len));
530 jeb->used_size -= freed_len;
531 c->used_size -= freed_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 }
533
534 // Take care, that wasted size is taken into concern
David Woodhouse1417fc42006-05-20 16:20:19 +0100535 if ((jeb->dirty_size || ISDIRTY(jeb->wasted_size + freed_len)) && jeb != c->nextblock) {
David Woodhousec7c16c82006-05-24 14:24:02 +0100536 D1(printk("Dirtying\n"));
David Woodhouse1417fc42006-05-20 16:20:19 +0100537 addedsize = freed_len;
538 jeb->dirty_size += freed_len;
539 c->dirty_size += freed_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540
541 /* Convert wasted space to dirty, if not a bad block */
542 if (jeb->wasted_size) {
543 if (on_list(&jeb->list, &c->bad_used_list)) {
544 D1(printk(KERN_DEBUG "Leaving block at %08x on the bad_used_list\n",
545 jeb->offset));
546 addedsize = 0; /* To fool the refiling code later */
547 } else {
548 D1(printk(KERN_DEBUG "Converting %d bytes of wasted space to dirty in block at %08x\n",
549 jeb->wasted_size, jeb->offset));
550 addedsize += jeb->wasted_size;
551 jeb->dirty_size += jeb->wasted_size;
552 c->dirty_size += jeb->wasted_size;
553 c->wasted_size -= jeb->wasted_size;
554 jeb->wasted_size = 0;
555 }
556 }
557 } else {
David Woodhousec7c16c82006-05-24 14:24:02 +0100558 D1(printk("Wasting\n"));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 addedsize = 0;
David Woodhouse1417fc42006-05-20 16:20:19 +0100560 jeb->wasted_size += freed_len;
561 c->wasted_size += freed_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 }
563 ref->flash_offset = ref_offset(ref) | REF_OBSOLETE;
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000564
Artem B. Bityutskiye0c8e422005-07-24 16:14:17 +0100565 jffs2_dbg_acct_sanity_check_nolock(c, jeb);
566 jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567
Artem B. Bityuckiy31fbdf72005-02-28 08:21:09 +0000568 if (c->flags & JFFS2_SB_FLAG_SCANNING) {
569 /* Flash scanning is in progress. Don't muck about with the block
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 lists because they're not ready yet, and don't actually
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000571 obliterate nodes that look obsolete. If they weren't
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 marked obsolete on the flash at the time they _became_
573 obsolete, there was probably a reason for that. */
574 spin_unlock(&c->erase_completion_lock);
575 /* We didn't lock the erase_free_sem */
576 return;
577 }
578
579 if (jeb == c->nextblock) {
580 D2(printk(KERN_DEBUG "Not moving nextblock 0x%08x to dirty/erase_pending list\n", jeb->offset));
581 } else if (!jeb->used_size && !jeb->unchecked_size) {
582 if (jeb == c->gcblock) {
583 D1(printk(KERN_DEBUG "gcblock at 0x%08x completely dirtied. Clearing gcblock...\n", jeb->offset));
584 c->gcblock = NULL;
585 } else {
586 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x completely dirtied. Removing from (dirty?) list...\n", jeb->offset));
587 list_del(&jeb->list);
588 }
589 if (jffs2_wbuf_dirty(c)) {
590 D1(printk(KERN_DEBUG "...and adding to erasable_pending_wbuf_list\n"));
591 list_add_tail(&jeb->list, &c->erasable_pending_wbuf_list);
592 } else {
593 if (jiffies & 127) {
594 /* Most of the time, we just erase it immediately. Otherwise we
595 spend ages scanning it on mount, etc. */
596 D1(printk(KERN_DEBUG "...and adding to erase_pending_list\n"));
597 list_add_tail(&jeb->list, &c->erase_pending_list);
598 c->nr_erasing_blocks++;
599 jffs2_erase_pending_trigger(c);
600 } else {
601 /* Sometimes, however, we leave it elsewhere so it doesn't get
602 immediately reused, and we spread the load a bit. */
603 D1(printk(KERN_DEBUG "...and adding to erasable_list\n"));
604 list_add_tail(&jeb->list, &c->erasable_list);
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000605 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 }
607 D1(printk(KERN_DEBUG "Done OK\n"));
608 } else if (jeb == c->gcblock) {
609 D2(printk(KERN_DEBUG "Not moving gcblock 0x%08x to dirty_list\n", jeb->offset));
610 } else if (ISDIRTY(jeb->dirty_size) && !ISDIRTY(jeb->dirty_size - addedsize)) {
611 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x is freshly dirtied. Removing from clean list...\n", jeb->offset));
612 list_del(&jeb->list);
613 D1(printk(KERN_DEBUG "...and adding to dirty_list\n"));
614 list_add_tail(&jeb->list, &c->dirty_list);
615 } else if (VERYDIRTY(c, jeb->dirty_size) &&
616 !VERYDIRTY(c, jeb->dirty_size - addedsize)) {
617 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x is now very dirty. Removing from dirty list...\n", jeb->offset));
618 list_del(&jeb->list);
619 D1(printk(KERN_DEBUG "...and adding to very_dirty_list\n"));
620 list_add_tail(&jeb->list, &c->very_dirty_list);
621 } else {
622 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 +0000623 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
624 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625
626 spin_unlock(&c->erase_completion_lock);
627
Artem B. Bityuckiy31fbdf72005-02-28 08:21:09 +0000628 if (!jffs2_can_mark_obsolete(c) || jffs2_is_readonly(c) ||
629 (c->flags & JFFS2_SB_FLAG_BUILDING)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 /* We didn't lock the erase_free_sem */
631 return;
632 }
633
634 /* The erase_free_sem is locked, and has been since before we marked the node obsolete
635 and potentially put its eraseblock onto the erase_pending_list. Thus, we know that
636 the block hasn't _already_ been erased, and that 'ref' itself hasn't been freed yet
David Woodhousec38c1b62006-05-25 01:38:27 +0100637 by jffs2_free_jeb_node_refs() in erase.c. Which is nice. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638
639 D1(printk(KERN_DEBUG "obliterating obsoleted node at 0x%08x\n", ref_offset(ref)));
640 ret = jffs2_flash_read(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
641 if (ret) {
642 printk(KERN_WARNING "Read error reading from obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret);
643 goto out_erase_sem;
644 }
645 if (retlen != sizeof(n)) {
646 printk(KERN_WARNING "Short read from obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen);
647 goto out_erase_sem;
648 }
David Woodhouse1417fc42006-05-20 16:20:19 +0100649 if (PAD(je32_to_cpu(n.totlen)) != PAD(freed_len)) {
650 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 -0700651 goto out_erase_sem;
652 }
653 if (!(je16_to_cpu(n.nodetype) & JFFS2_NODE_ACCURATE)) {
654 D1(printk(KERN_DEBUG "Node at 0x%08x was already marked obsolete (nodetype 0x%04x)\n", ref_offset(ref), je16_to_cpu(n.nodetype)));
655 goto out_erase_sem;
656 }
657 /* XXX FIXME: This is ugly now */
658 n.nodetype = cpu_to_je16(je16_to_cpu(n.nodetype) & ~JFFS2_NODE_ACCURATE);
659 ret = jffs2_flash_write(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
660 if (ret) {
661 printk(KERN_WARNING "Write error in obliterating obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret);
662 goto out_erase_sem;
663 }
664 if (retlen != sizeof(n)) {
665 printk(KERN_WARNING "Short write in obliterating obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen);
666 goto out_erase_sem;
667 }
668
669 /* Nodes which have been marked obsolete no longer need to be
670 associated with any inode. Remove them from the per-inode list.
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000671
672 Note we can't do this for NAND at the moment because we need
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 obsolete dirent nodes to stay on the lists, because of the
674 horridness in jffs2_garbage_collect_deletion_dirent(). Also
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000675 because we delete the inocache, and on NAND we need that to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 stay around until all the nodes are actually erased, in order
677 to stop us from giving the same inode number to another newly
678 created inode. */
679 if (ref->next_in_ino) {
680 struct jffs2_inode_cache *ic;
681 struct jffs2_raw_node_ref **p;
682
683 spin_lock(&c->erase_completion_lock);
684
685 ic = jffs2_raw_ref_to_ic(ref);
686 for (p = &ic->nodes; (*p) != ref; p = &((*p)->next_in_ino))
687 ;
688
689 *p = ref->next_in_ino;
690 ref->next_in_ino = NULL;
691
KaiGai Koheic9f700f2006-06-11 10:35:15 +0900692 switch (ic->class) {
693#ifdef CONFIG_JFFS2_FS_XATTR
694 case RAWNODE_CLASS_XATTR_DATUM:
695 jffs2_release_xattr_datum(c, (struct jffs2_xattr_datum *)ic);
696 break;
697 case RAWNODE_CLASS_XATTR_REF:
698 jffs2_release_xattr_ref(c, (struct jffs2_xattr_ref *)ic);
699 break;
700#endif
701 default:
702 if (ic->nodes == (void *)ic && ic->nlink == 0)
703 jffs2_del_ino_cache(c, ic);
704 break;
705 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 spin_unlock(&c->erase_completion_lock);
707 }
708
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 out_erase_sem:
710 up(&c->erase_free_sem);
711}
712
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713int jffs2_thread_should_wake(struct jffs2_sb_info *c)
714{
715 int ret = 0;
716 uint32_t dirty;
717
718 if (c->unchecked_size) {
719 D1(printk(KERN_DEBUG "jffs2_thread_should_wake(): unchecked_size %d, checked_ino #%d\n",
720 c->unchecked_size, c->checked_ino));
721 return 1;
722 }
723
724 /* dirty_size contains blocks on erase_pending_list
725 * those blocks are counted in c->nr_erasing_blocks.
726 * If one block is actually erased, it is not longer counted as dirty_space
727 * but it is counted in c->nr_erasing_blocks, so we add it and subtract it
728 * with c->nr_erasing_blocks * c->sector_size again.
729 * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks
730 * This helps us to force gc and pick eventually a clean block to spread the load.
731 */
732 dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size;
733
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000734 if (c->nr_free_blocks + c->nr_erasing_blocks < c->resv_blocks_gctrigger &&
735 (dirty > c->nospc_dirty_size))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 ret = 1;
737
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000738 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 -0700739 c->nr_free_blocks, c->nr_erasing_blocks, c->dirty_size, ret?"yes":"no"));
740
741 return ret;
742}