blob: 8feb8749bc759d54f8e730afd6d8fae9492ba4fb [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);
140 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);
161 return ret;
162}
163
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100164
165/* Classify nextblock (clean, dirty of verydirty) and force to select an other one */
166
167static void jffs2_close_nextblock(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168{
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100169
170 /* Check, if we have a dirty block now, or if it was dirty already */
171 if (ISDIRTY (jeb->wasted_size + jeb->dirty_size)) {
172 c->dirty_size += jeb->wasted_size;
173 c->wasted_size -= jeb->wasted_size;
174 jeb->dirty_size += jeb->wasted_size;
175 jeb->wasted_size = 0;
176 if (VERYDIRTY(c, jeb->dirty_size)) {
177 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",
178 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
179 list_add_tail(&jeb->list, &c->very_dirty_list);
180 } else {
181 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to dirty_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
182 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
183 list_add_tail(&jeb->list, &c->dirty_list);
184 }
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000185 } else {
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100186 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to clean_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->clean_list);
189 }
190 c->nextblock = NULL;
191
192}
193
194/* Select a new jeb for nextblock */
195
196static int jffs2_find_nextblock(struct jffs2_sb_info *c)
197{
198 struct list_head *next;
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000199
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100200 /* Take the next block off the 'free' list */
201
202 if (list_empty(&c->free_list)) {
203
204 if (!c->nr_erasing_blocks &&
205 !list_empty(&c->erasable_list)) {
206 struct jffs2_eraseblock *ejeb;
207
208 ejeb = list_entry(c->erasable_list.next, struct jffs2_eraseblock, list);
209 list_del(&ejeb->list);
210 list_add_tail(&ejeb->list, &c->erase_pending_list);
211 c->nr_erasing_blocks++;
212 jffs2_erase_pending_trigger(c);
213 D1(printk(KERN_DEBUG "jffs2_find_nextblock: Triggering erase of erasable block at 0x%08x\n",
214 ejeb->offset));
215 }
216
217 if (!c->nr_erasing_blocks &&
218 !list_empty(&c->erasable_pending_wbuf_list)) {
219 D1(printk(KERN_DEBUG "jffs2_find_nextblock: Flushing write buffer\n"));
220 /* c->nextblock is NULL, no update to c->nextblock allowed */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 spin_unlock(&c->erase_completion_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 jffs2_flush_wbuf_pad(c);
223 spin_lock(&c->erase_completion_lock);
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100224 /* Have another go. It'll be on the erasable_list now */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 return -EAGAIN;
226 }
227
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100228 if (!c->nr_erasing_blocks) {
229 /* Ouch. We're in GC, or we wouldn't have got here.
230 And there's no space left. At all. */
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000231 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",
232 c->nr_erasing_blocks, c->nr_free_blocks, list_empty(&c->erasable_list)?"yes":"no",
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100233 list_empty(&c->erasing_list)?"yes":"no", list_empty(&c->erase_pending_list)?"yes":"no");
234 return -ENOSPC;
235 }
236
237 spin_unlock(&c->erase_completion_lock);
238 /* Don't wait for it; just erase one right now */
239 jffs2_erase_pending_blocks(c, 1);
240 spin_lock(&c->erase_completion_lock);
241
242 /* An erase may have failed, decreasing the
243 amount of free space available. So we must
244 restart from the beginning */
245 return -EAGAIN;
246 }
247
248 next = c->free_list.next;
249 list_del(next);
250 c->nextblock = list_entry(next, struct jffs2_eraseblock, list);
251 c->nr_free_blocks--;
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000252
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100253 jffs2_sum_reset_collected(c->summary); /* reset collected summary */
254
255 D1(printk(KERN_DEBUG "jffs2_find_nextblock(): new nextblock = 0x%08x\n", c->nextblock->offset));
256
257 return 0;
258}
259
260/* Called with alloc sem _and_ erase_completion_lock */
David Woodhouse9fe48542006-05-23 00:38:06 +0100261static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
262 uint32_t *len, uint32_t sumsize)
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100263{
264 struct jffs2_eraseblock *jeb = c->nextblock;
David Woodhouse9fe48542006-05-23 00:38:06 +0100265 uint32_t reserved_size; /* for summary information at the end of the jeb */
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100266 int ret;
267
268 restart:
269 reserved_size = 0;
270
271 if (jffs2_sum_active() && (sumsize != JFFS2_SUMMARY_NOSUM_SIZE)) {
272 /* NOSUM_SIZE means not to generate summary */
273
274 if (jeb) {
275 reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE);
Artem B. Bityutskiy733802d2005-09-22 12:25:00 +0100276 dbg_summary("minsize=%d , jeb->free=%d ,"
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100277 "summary->size=%d , sumsize=%d\n",
278 minsize, jeb->free_size,
279 c->summary->sum_size, sumsize);
280 }
281
282 /* Is there enough space for writing out the current node, or we have to
283 write out summary information now, close this jeb and select new nextblock? */
284 if (jeb && (PAD(minsize) + PAD(c->summary->sum_size + sumsize +
285 JFFS2_SUMMARY_FRAME_SIZE) > jeb->free_size)) {
286
287 /* Has summary been disabled for this jeb? */
288 if (jffs2_sum_is_disabled(c->summary)) {
289 sumsize = JFFS2_SUMMARY_NOSUM_SIZE;
290 goto restart;
291 }
292
293 /* Writing out the collected summary information */
Artem B. Bityutskiy733802d2005-09-22 12:25:00 +0100294 dbg_summary("generating summary for 0x%08x.\n", jeb->offset);
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100295 ret = jffs2_sum_write_sumnode(c);
296
297 if (ret)
298 return ret;
299
300 if (jffs2_sum_is_disabled(c->summary)) {
301 /* jffs2_write_sumnode() couldn't write out the summary information
302 diabling summary for this jeb and free the collected information
303 */
304 sumsize = JFFS2_SUMMARY_NOSUM_SIZE;
305 goto restart;
306 }
307
308 jffs2_close_nextblock(c, jeb);
309 jeb = NULL;
Ferenc Havasi34c0e902005-09-16 13:58:20 +0100310 /* keep always valid value in reserved_size */
311 reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE);
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100312 }
313 } else {
314 if (jeb && minsize > jeb->free_size) {
315 /* Skip the end of this block and file it as having some dirty space */
316 /* If there's a pending write to it, flush now */
317
318 if (jffs2_wbuf_dirty(c)) {
319 spin_unlock(&c->erase_completion_lock);
320 D1(printk(KERN_DEBUG "jffs2_do_reserve_space: Flushing write buffer\n"));
321 jffs2_flush_wbuf_pad(c);
322 spin_lock(&c->erase_completion_lock);
323 jeb = c->nextblock;
324 goto restart;
325 }
326
327 c->wasted_size += jeb->free_size;
328 c->free_size -= jeb->free_size;
329 jeb->wasted_size += jeb->free_size;
330 jeb->free_size = 0;
331
332 jffs2_close_nextblock(c, jeb);
333 jeb = NULL;
334 }
335 }
336
337 if (!jeb) {
338
339 ret = jffs2_find_nextblock(c);
340 if (ret)
341 return ret;
342
343 jeb = c->nextblock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344
345 if (jeb->free_size != c->sector_size - c->cleanmarker_size) {
346 printk(KERN_WARNING "Eep. Block 0x%08x taken from free_list had free_size of 0x%08x!!\n", jeb->offset, jeb->free_size);
347 goto restart;
348 }
349 }
350 /* OK, jeb (==c->nextblock) is now pointing at a block which definitely has
351 enough space */
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100352 *len = jeb->free_size - reserved_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353
354 if (c->cleanmarker_size && jeb->used_size == c->cleanmarker_size &&
355 !jeb->first_node->next_in_ino) {
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000356 /* Only node in it beforehand was a CLEANMARKER node (we think).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 So mark it obsolete now that there's going to be another node
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000358 in the block. This will reduce used_size to zero but We've
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 already set c->nextblock so that jffs2_mark_node_obsolete()
360 won't try to refile it to the dirty_list.
361 */
362 spin_unlock(&c->erase_completion_lock);
363 jffs2_mark_node_obsolete(c, jeb->first_node);
364 spin_lock(&c->erase_completion_lock);
365 }
366
David Woodhouse9fe48542006-05-23 00:38:06 +0100367 D1(printk(KERN_DEBUG "jffs2_do_reserve_space(): Giving 0x%x bytes at 0x%x\n",
368 *len, jeb->offset + (c->sector_size - jeb->free_size)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 return 0;
370}
371
372/**
373 * jffs2_add_physical_node_ref - add a physical node reference to the list
374 * @c: superblock info
375 * @new: new node reference to add
376 * @len: length of this physical node
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 *
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000378 * Should only be used to report nodes for which space has been allocated
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 * by jffs2_reserve_space.
380 *
381 * Must be called with the alloc_sem held.
382 */
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000383
David Woodhousefcb75782006-05-22 15:23:10 +0100384int jffs2_add_physical_node_ref(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *new,
385 uint32_t len, struct jffs2_inode_cache *ic)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386{
387 struct jffs2_eraseblock *jeb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388
389 jeb = &c->blocks[new->flash_offset / c->sector_size];
David Woodhouseca89a512006-05-21 13:29:11 +0100390#ifdef TEST_TOTLEN
David Woodhouseb64335f2006-05-21 04:36:45 +0100391 new->__totlen = len;
David Woodhouseca89a512006-05-21 13:29:11 +0100392#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393
394 D1(printk(KERN_DEBUG "jffs2_add_physical_node_ref(): Node at 0x%x(%d), size 0x%x\n", ref_offset(new), ref_flags(new), len));
395#if 1
Estelle Hammache3118db32005-01-24 21:30:25 +0000396 /* we could get some obsolete nodes after nextblock was refiled
397 in wbuf.c */
Estelle Hammache9b88f472005-01-28 18:53:05 +0000398 if ((c->nextblock || !ref_obsolete(new))
399 &&(jeb != c->nextblock || ref_offset(new) != jeb->offset + (c->sector_size - jeb->free_size))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 printk(KERN_WARNING "argh. node added in wrong place\n");
401 jffs2_free_raw_node_ref(new);
402 return -EINVAL;
403 }
404#endif
405 spin_lock(&c->erase_completion_lock);
406
David Woodhousefcb75782006-05-22 15:23:10 +0100407 jffs2_link_node_ref(c, jeb, new, len, ic);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408
Estelle Hammache9b88f472005-01-28 18:53:05 +0000409 if (!jeb->free_size && !jeb->dirty_size && !ISDIRTY(jeb->wasted_size)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 /* If it lives on the dirty_list, jffs2_reserve_space will put it there */
411 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
412 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
413 if (jffs2_wbuf_dirty(c)) {
414 /* Flush the last write in the block if it's outstanding */
415 spin_unlock(&c->erase_completion_lock);
416 jffs2_flush_wbuf_pad(c);
417 spin_lock(&c->erase_completion_lock);
418 }
419
420 list_add_tail(&jeb->list, &c->clean_list);
421 c->nextblock = NULL;
422 }
Artem B. Bityutskiye0c8e422005-07-24 16:14:17 +0100423 jffs2_dbg_acct_sanity_check_nolock(c,jeb);
424 jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425
426 spin_unlock(&c->erase_completion_lock);
427
428 return 0;
429}
430
431
432void jffs2_complete_reservation(struct jffs2_sb_info *c)
433{
434 D1(printk(KERN_DEBUG "jffs2_complete_reservation()\n"));
435 jffs2_garbage_collect_trigger(c);
436 up(&c->alloc_sem);
437}
438
439static inline int on_list(struct list_head *obj, struct list_head *head)
440{
441 struct list_head *this;
442
443 list_for_each(this, head) {
444 if (this == obj) {
445 D1(printk("%p is on list at %p\n", obj, head));
446 return 1;
447
448 }
449 }
450 return 0;
451}
452
453void jffs2_mark_node_obsolete(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref)
454{
455 struct jffs2_eraseblock *jeb;
456 int blocknr;
457 struct jffs2_unknown_node n;
458 int ret, addedsize;
459 size_t retlen;
David Woodhouse1417fc42006-05-20 16:20:19 +0100460 uint32_t freed_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461
462 if(!ref) {
463 printk(KERN_NOTICE "EEEEEK. jffs2_mark_node_obsolete called with NULL node\n");
464 return;
465 }
466 if (ref_obsolete(ref)) {
467 D1(printk(KERN_DEBUG "jffs2_mark_node_obsolete called with already obsolete node at 0x%08x\n", ref_offset(ref)));
468 return;
469 }
470 blocknr = ref->flash_offset / c->sector_size;
471 if (blocknr >= c->nr_blocks) {
472 printk(KERN_NOTICE "raw node at 0x%08x is off the end of device!\n", ref->flash_offset);
473 BUG();
474 }
475 jeb = &c->blocks[blocknr];
476
477 if (jffs2_can_mark_obsolete(c) && !jffs2_is_readonly(c) &&
Artem B. Bityuckiy31fbdf72005-02-28 08:21:09 +0000478 !(c->flags & (JFFS2_SB_FLAG_SCANNING | JFFS2_SB_FLAG_BUILDING))) {
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000479 /* Hm. This may confuse static lock analysis. If any of the above
480 three conditions is false, we're going to return from this
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 function without actually obliterating any nodes or freeing
482 any jffs2_raw_node_refs. So we don't need to stop erases from
483 happening, or protect against people holding an obsolete
484 jffs2_raw_node_ref without the erase_completion_lock. */
485 down(&c->erase_free_sem);
486 }
487
488 spin_lock(&c->erase_completion_lock);
489
David Woodhouse1417fc42006-05-20 16:20:19 +0100490 freed_len = ref_totlen(c, jeb, ref);
491
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 if (ref_flags(ref) == REF_UNCHECKED) {
David Woodhouse1417fc42006-05-20 16:20:19 +0100493 D1(if (unlikely(jeb->unchecked_size < freed_len)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 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 +0100495 freed_len, blocknr, ref->flash_offset, jeb->used_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 BUG();
497 })
David Woodhouse1417fc42006-05-20 16:20:19 +0100498 D1(printk(KERN_DEBUG "Obsoleting previously unchecked node at 0x%08x of len %x: ", ref_offset(ref), freed_len));
499 jeb->unchecked_size -= freed_len;
500 c->unchecked_size -= freed_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 } else {
David Woodhouse1417fc42006-05-20 16:20:19 +0100502 D1(if (unlikely(jeb->used_size < freed_len)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 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 +0100504 freed_len, blocknr, ref->flash_offset, jeb->used_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 BUG();
506 })
David Woodhouse1417fc42006-05-20 16:20:19 +0100507 D1(printk(KERN_DEBUG "Obsoleting node at 0x%08x of len %#x: ", ref_offset(ref), freed_len));
508 jeb->used_size -= freed_len;
509 c->used_size -= freed_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 }
511
512 // Take care, that wasted size is taken into concern
David Woodhouse1417fc42006-05-20 16:20:19 +0100513 if ((jeb->dirty_size || ISDIRTY(jeb->wasted_size + freed_len)) && jeb != c->nextblock) {
Artem B. Bityuckiy6f401a402005-04-06 17:02:55 +0100514 D1(printk(KERN_DEBUG "Dirtying\n"));
David Woodhouse1417fc42006-05-20 16:20:19 +0100515 addedsize = freed_len;
516 jeb->dirty_size += freed_len;
517 c->dirty_size += freed_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518
519 /* Convert wasted space to dirty, if not a bad block */
520 if (jeb->wasted_size) {
521 if (on_list(&jeb->list, &c->bad_used_list)) {
522 D1(printk(KERN_DEBUG "Leaving block at %08x on the bad_used_list\n",
523 jeb->offset));
524 addedsize = 0; /* To fool the refiling code later */
525 } else {
526 D1(printk(KERN_DEBUG "Converting %d bytes of wasted space to dirty in block at %08x\n",
527 jeb->wasted_size, jeb->offset));
528 addedsize += jeb->wasted_size;
529 jeb->dirty_size += jeb->wasted_size;
530 c->dirty_size += jeb->wasted_size;
531 c->wasted_size -= jeb->wasted_size;
532 jeb->wasted_size = 0;
533 }
534 }
535 } else {
Artem B. Bityuckiy6f401a402005-04-06 17:02:55 +0100536 D1(printk(KERN_DEBUG "Wasting\n"));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 addedsize = 0;
David Woodhouse1417fc42006-05-20 16:20:19 +0100538 jeb->wasted_size += freed_len;
539 c->wasted_size += freed_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 }
541 ref->flash_offset = ref_offset(ref) | REF_OBSOLETE;
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000542
Artem B. Bityutskiye0c8e422005-07-24 16:14:17 +0100543 jffs2_dbg_acct_sanity_check_nolock(c, jeb);
544 jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545
Artem B. Bityuckiy31fbdf72005-02-28 08:21:09 +0000546 if (c->flags & JFFS2_SB_FLAG_SCANNING) {
547 /* Flash scanning is in progress. Don't muck about with the block
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 lists because they're not ready yet, and don't actually
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000549 obliterate nodes that look obsolete. If they weren't
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 marked obsolete on the flash at the time they _became_
551 obsolete, there was probably a reason for that. */
552 spin_unlock(&c->erase_completion_lock);
553 /* We didn't lock the erase_free_sem */
554 return;
555 }
556
557 if (jeb == c->nextblock) {
558 D2(printk(KERN_DEBUG "Not moving nextblock 0x%08x to dirty/erase_pending list\n", jeb->offset));
559 } else if (!jeb->used_size && !jeb->unchecked_size) {
560 if (jeb == c->gcblock) {
561 D1(printk(KERN_DEBUG "gcblock at 0x%08x completely dirtied. Clearing gcblock...\n", jeb->offset));
562 c->gcblock = NULL;
563 } else {
564 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x completely dirtied. Removing from (dirty?) list...\n", jeb->offset));
565 list_del(&jeb->list);
566 }
567 if (jffs2_wbuf_dirty(c)) {
568 D1(printk(KERN_DEBUG "...and adding to erasable_pending_wbuf_list\n"));
569 list_add_tail(&jeb->list, &c->erasable_pending_wbuf_list);
570 } else {
571 if (jiffies & 127) {
572 /* Most of the time, we just erase it immediately. Otherwise we
573 spend ages scanning it on mount, etc. */
574 D1(printk(KERN_DEBUG "...and adding to erase_pending_list\n"));
575 list_add_tail(&jeb->list, &c->erase_pending_list);
576 c->nr_erasing_blocks++;
577 jffs2_erase_pending_trigger(c);
578 } else {
579 /* Sometimes, however, we leave it elsewhere so it doesn't get
580 immediately reused, and we spread the load a bit. */
581 D1(printk(KERN_DEBUG "...and adding to erasable_list\n"));
582 list_add_tail(&jeb->list, &c->erasable_list);
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000583 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 }
585 D1(printk(KERN_DEBUG "Done OK\n"));
586 } else if (jeb == c->gcblock) {
587 D2(printk(KERN_DEBUG "Not moving gcblock 0x%08x to dirty_list\n", jeb->offset));
588 } else if (ISDIRTY(jeb->dirty_size) && !ISDIRTY(jeb->dirty_size - addedsize)) {
589 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x is freshly dirtied. Removing from clean list...\n", jeb->offset));
590 list_del(&jeb->list);
591 D1(printk(KERN_DEBUG "...and adding to dirty_list\n"));
592 list_add_tail(&jeb->list, &c->dirty_list);
593 } else if (VERYDIRTY(c, jeb->dirty_size) &&
594 !VERYDIRTY(c, jeb->dirty_size - addedsize)) {
595 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x is now very dirty. Removing from dirty list...\n", jeb->offset));
596 list_del(&jeb->list);
597 D1(printk(KERN_DEBUG "...and adding to very_dirty_list\n"));
598 list_add_tail(&jeb->list, &c->very_dirty_list);
599 } else {
600 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 +0000601 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
602 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603
604 spin_unlock(&c->erase_completion_lock);
605
Artem B. Bityuckiy31fbdf72005-02-28 08:21:09 +0000606 if (!jffs2_can_mark_obsolete(c) || jffs2_is_readonly(c) ||
607 (c->flags & JFFS2_SB_FLAG_BUILDING)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 /* We didn't lock the erase_free_sem */
609 return;
610 }
611
612 /* The erase_free_sem is locked, and has been since before we marked the node obsolete
613 and potentially put its eraseblock onto the erase_pending_list. Thus, we know that
614 the block hasn't _already_ been erased, and that 'ref' itself hasn't been freed yet
615 by jffs2_free_all_node_refs() in erase.c. Which is nice. */
616
617 D1(printk(KERN_DEBUG "obliterating obsoleted node at 0x%08x\n", ref_offset(ref)));
618 ret = jffs2_flash_read(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
619 if (ret) {
620 printk(KERN_WARNING "Read error reading from obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret);
621 goto out_erase_sem;
622 }
623 if (retlen != sizeof(n)) {
624 printk(KERN_WARNING "Short read from obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen);
625 goto out_erase_sem;
626 }
David Woodhouse1417fc42006-05-20 16:20:19 +0100627 if (PAD(je32_to_cpu(n.totlen)) != PAD(freed_len)) {
628 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 -0700629 goto out_erase_sem;
630 }
631 if (!(je16_to_cpu(n.nodetype) & JFFS2_NODE_ACCURATE)) {
632 D1(printk(KERN_DEBUG "Node at 0x%08x was already marked obsolete (nodetype 0x%04x)\n", ref_offset(ref), je16_to_cpu(n.nodetype)));
633 goto out_erase_sem;
634 }
635 /* XXX FIXME: This is ugly now */
636 n.nodetype = cpu_to_je16(je16_to_cpu(n.nodetype) & ~JFFS2_NODE_ACCURATE);
637 ret = jffs2_flash_write(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
638 if (ret) {
639 printk(KERN_WARNING "Write error in obliterating obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret);
640 goto out_erase_sem;
641 }
642 if (retlen != sizeof(n)) {
643 printk(KERN_WARNING "Short write in obliterating obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen);
644 goto out_erase_sem;
645 }
646
647 /* Nodes which have been marked obsolete no longer need to be
648 associated with any inode. Remove them from the per-inode list.
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000649
650 Note we can't do this for NAND at the moment because we need
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 obsolete dirent nodes to stay on the lists, because of the
652 horridness in jffs2_garbage_collect_deletion_dirent(). Also
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000653 because we delete the inocache, and on NAND we need that to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 stay around until all the nodes are actually erased, in order
655 to stop us from giving the same inode number to another newly
656 created inode. */
657 if (ref->next_in_ino) {
658 struct jffs2_inode_cache *ic;
659 struct jffs2_raw_node_ref **p;
660
661 spin_lock(&c->erase_completion_lock);
662
663 ic = jffs2_raw_ref_to_ic(ref);
David Woodhouse0eac9402006-05-22 16:29:23 +0100664 /* It seems we should never call jffs2_mark_node_obsolete() for
665 XATTR nodes.... yet. Make sure we notice if/when we change
666 that :) */
667 BUG_ON(ic->class != RAWNODE_CLASS_INODE_CACHE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 for (p = &ic->nodes; (*p) != ref; p = &((*p)->next_in_ino))
669 ;
670
671 *p = ref->next_in_ino;
672 ref->next_in_ino = NULL;
673
Artem B. Bityuckiy437316d2005-03-20 17:46:23 +0000674 if (ic->nodes == (void *)ic && ic->nlink == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 jffs2_del_ino_cache(c, ic);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676
677 spin_unlock(&c->erase_completion_lock);
678 }
679
680
681 /* Merge with the next node in the physical list, if there is one
682 and if it's also obsolete and if it doesn't belong to any inode */
683 if (ref->next_phys && ref_obsolete(ref->next_phys) &&
684 !ref->next_phys->next_in_ino) {
685 struct jffs2_raw_node_ref *n = ref->next_phys;
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000686
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 spin_lock(&c->erase_completion_lock);
688
David Woodhouseca89a512006-05-21 13:29:11 +0100689#ifdef TEST_TOTLEN
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 ref->__totlen += n->__totlen;
David Woodhouseca89a512006-05-21 13:29:11 +0100691#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 ref->next_phys = n->next_phys;
693 if (jeb->last_node == n) jeb->last_node = ref;
694 if (jeb->gc_node == n) {
695 /* gc will be happy continuing gc on this node */
696 jeb->gc_node=ref;
697 }
698 spin_unlock(&c->erase_completion_lock);
699
700 jffs2_free_raw_node_ref(n);
701 }
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000702
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 /* Also merge with the previous node in the list, if there is one
704 and that one is obsolete */
705 if (ref != jeb->first_node ) {
706 struct jffs2_raw_node_ref *p = jeb->first_node;
707
708 spin_lock(&c->erase_completion_lock);
709
710 while (p->next_phys != ref)
711 p = p->next_phys;
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000712
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 if (ref_obsolete(p) && !ref->next_in_ino) {
David Woodhouseca89a512006-05-21 13:29:11 +0100714#ifdef TEST_TOTLEN
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 p->__totlen += ref->__totlen;
David Woodhouseca89a512006-05-21 13:29:11 +0100716#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 if (jeb->last_node == ref) {
718 jeb->last_node = p;
719 }
720 if (jeb->gc_node == ref) {
721 /* gc will be happy continuing gc on this node */
722 jeb->gc_node=p;
723 }
724 p->next_phys = ref->next_phys;
725 jffs2_free_raw_node_ref(ref);
726 }
727 spin_unlock(&c->erase_completion_lock);
728 }
729 out_erase_sem:
730 up(&c->erase_free_sem);
731}
732
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733int jffs2_thread_should_wake(struct jffs2_sb_info *c)
734{
735 int ret = 0;
736 uint32_t dirty;
737
738 if (c->unchecked_size) {
739 D1(printk(KERN_DEBUG "jffs2_thread_should_wake(): unchecked_size %d, checked_ino #%d\n",
740 c->unchecked_size, c->checked_ino));
741 return 1;
742 }
743
744 /* dirty_size contains blocks on erase_pending_list
745 * those blocks are counted in c->nr_erasing_blocks.
746 * If one block is actually erased, it is not longer counted as dirty_space
747 * but it is counted in c->nr_erasing_blocks, so we add it and subtract it
748 * with c->nr_erasing_blocks * c->sector_size again.
749 * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks
750 * This helps us to force gc and pick eventually a clean block to spread the load.
751 */
752 dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size;
753
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000754 if (c->nr_free_blocks + c->nr_erasing_blocks < c->resv_blocks_gctrigger &&
755 (dirty > c->nospc_dirty_size))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 ret = 1;
757
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000758 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 -0700759 c->nr_free_blocks, c->nr_erasing_blocks, c->dirty_size, ret?"yes":"no"));
760
761 return ret;
762}