blob: 0875b60b4bf700cc8fda2fad76437d0dc9ea7420 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * JFFS2 -- Journalling Flash File System, Version 2.
3 *
David Woodhousec00c3102007-04-25 14:16:47 +01004 * Copyright © 2001-2007 Red Hat, Inc.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005 *
6 * Created by David Woodhouse <dwmw2@infradead.org>
7 *
8 * For licensing information, see the file 'LICENCE' in this directory.
9 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070010 */
11
12#include <linux/kernel.h>
13#include <linux/slab.h>
14#include <linux/mtd/mtd.h>
15#include <linux/compiler.h>
16#include <linux/sched.h> /* For cond_resched() */
17#include "nodelist.h"
Ferenc Havasie631ddb2005-09-07 09:35:26 +010018#include "debug.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070019
20/**
21 * jffs2_reserve_space - request physical space to write nodes to flash
22 * @c: superblock info
23 * @minsize: Minimum acceptable size of allocation
Linus Torvalds1da177e2005-04-16 15:20:36 -070024 * @len: Returned value of allocation length
25 * @prio: Allocation type - ALLOC_{NORMAL,DELETION}
26 *
27 * Requests a block of physical space on the flash. Returns zero for success
David Woodhouse9fe48542006-05-23 00:38:06 +010028 * and puts 'len' into the appropriate place, or returns -ENOSPC or other
29 * error if appropriate. Doesn't return len since that's
Linus Torvalds1da177e2005-04-16 15:20:36 -070030 *
31 * If it returns zero, jffs2_reserve_space() also downs the per-filesystem
32 * allocation semaphore, to prevent more than one allocation from being
33 * active at any time. The semaphore is later released by jffs2_commit_allocation()
34 *
35 * jffs2_reserve_space() may trigger garbage collection in order to make room
36 * for the requested allocation.
37 */
38
Ferenc Havasie631ddb2005-09-07 09:35:26 +010039static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
David Woodhouse9fe48542006-05-23 00:38:06 +010040 uint32_t *len, uint32_t sumsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
David Woodhouse9fe48542006-05-23 00:38:06 +010042int jffs2_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
Ferenc Havasie631ddb2005-09-07 09:35:26 +010043 uint32_t *len, int prio, uint32_t sumsize)
Linus Torvalds1da177e2005-04-16 15:20:36 -070044{
45 int ret = -EAGAIN;
46 int blocksneeded = c->resv_blocks_write;
47 /* align it */
48 minsize = PAD(minsize);
49
50 D1(printk(KERN_DEBUG "jffs2_reserve_space(): Requested 0x%x bytes\n", minsize));
David Woodhouseced22072008-04-22 15:13:40 +010051 mutex_lock(&c->alloc_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
53 D1(printk(KERN_DEBUG "jffs2_reserve_space(): alloc sem got\n"));
54
55 spin_lock(&c->erase_completion_lock);
56
57 /* this needs a little more thought (true <tglx> :)) */
58 while(ret == -EAGAIN) {
59 while(c->nr_free_blocks + c->nr_erasing_blocks < blocksneeded) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070060 uint32_t dirty, avail;
61
62 /* calculate real dirty size
63 * dirty_size contains blocks on erase_pending_list
64 * those blocks are counted in c->nr_erasing_blocks.
65 * If one block is actually erased, it is not longer counted as dirty_space
66 * but it is counted in c->nr_erasing_blocks, so we add it and subtract it
67 * with c->nr_erasing_blocks * c->sector_size again.
68 * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks
69 * This helps us to force gc and pick eventually a clean block to spread the load.
70 * We add unchecked_size here, as we hopefully will find some space to use.
71 * This will affect the sum only once, as gc first finishes checking
72 * of nodes.
73 */
74 dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size + c->unchecked_size;
75 if (dirty < c->nospc_dirty_size) {
76 if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) {
Artem B. Bityuckiy4132ace2005-05-06 10:30:30 +010077 D1(printk(KERN_NOTICE "jffs2_reserve_space(): Low on dirty space to GC, but it's a deletion. Allowing...\n"));
Linus Torvalds1da177e2005-04-16 15:20:36 -070078 break;
79 }
80 D1(printk(KERN_DEBUG "dirty size 0x%08x + unchecked_size 0x%08x < nospc_dirty_size 0x%08x, returning -ENOSPC\n",
81 dirty, c->unchecked_size, c->sector_size));
82
83 spin_unlock(&c->erase_completion_lock);
David Woodhouseced22072008-04-22 15:13:40 +010084 mutex_unlock(&c->alloc_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -070085 return -ENOSPC;
86 }
Thomas Gleixner182ec4e2005-11-07 11:16:07 +000087
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 /* Calc possibly available space. Possibly available means that we
89 * don't know, if unchecked size contains obsoleted nodes, which could give us some
90 * more usable space. This will affect the sum only once, as gc first finishes checking
91 * of nodes.
Thomas Gleixner182ec4e2005-11-07 11:16:07 +000092 + Return -ENOSPC, if the maximum possibly available space is less or equal than
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 * blocksneeded * sector_size.
94 * This blocks endless gc looping on a filesystem, which is nearly full, even if
95 * the check above passes.
96 */
97 avail = c->free_size + c->dirty_size + c->erasing_size + c->unchecked_size;
98 if ( (avail / c->sector_size) <= blocksneeded) {
99 if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) {
Artem B. Bityuckiy4132ace2005-05-06 10:30:30 +0100100 D1(printk(KERN_NOTICE "jffs2_reserve_space(): Low on possibly available space, but it's a deletion. Allowing...\n"));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 break;
102 }
103
104 D1(printk(KERN_DEBUG "max. available size 0x%08x < blocksneeded * sector_size 0x%08x, returning -ENOSPC\n",
105 avail, blocksneeded * c->sector_size));
106 spin_unlock(&c->erase_completion_lock);
David Woodhouseced22072008-04-22 15:13:40 +0100107 mutex_unlock(&c->alloc_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 return -ENOSPC;
109 }
110
David Woodhouseced22072008-04-22 15:13:40 +0100111 mutex_unlock(&c->alloc_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112
113 D1(printk(KERN_DEBUG "Triggering GC pass. nr_free_blocks %d, nr_erasing_blocks %d, free_size 0x%08x, dirty_size 0x%08x, wasted_size 0x%08x, used_size 0x%08x, erasing_size 0x%08x, bad_size 0x%08x (total 0x%08x of 0x%08x)\n",
114 c->nr_free_blocks, c->nr_erasing_blocks, c->free_size, c->dirty_size, c->wasted_size, c->used_size, c->erasing_size, c->bad_size,
115 c->free_size + c->dirty_size + c->wasted_size + c->used_size + c->erasing_size + c->bad_size, c->flash_size));
116 spin_unlock(&c->erase_completion_lock);
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000117
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 ret = jffs2_garbage_collect_pass(c);
David Woodhouse422b1202008-04-23 15:40:52 +0100119
120 if (ret == -EAGAIN)
121 jffs2_erase_pending_blocks(c, 1);
122 else if (ret)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 return ret;
124
125 cond_resched();
126
127 if (signal_pending(current))
128 return -EINTR;
129
David Woodhouseced22072008-04-22 15:13:40 +0100130 mutex_lock(&c->alloc_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 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)
David Woodhouseced22072008-04-22 15:13:40 +0100143 mutex_unlock(&c->alloc_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 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) {
David Woodhouseef53cb02007-07-10 10:01:22 +0100159 D1(printk(KERN_DEBUG "jffs2_reserve_space_gc: looping, ret is %d\n", ret));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 }
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
Adrian Hunter99c25942007-03-29 11:00:47 +0300175 if (c->nextblock == NULL) {
176 D1(printk(KERN_DEBUG "jffs2_close_nextblock: Erase block at 0x%08x has already been placed in a list\n",
177 jeb->offset));
178 return;
179 }
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100180 /* Check, if we have a dirty block now, or if it was dirty already */
181 if (ISDIRTY (jeb->wasted_size + jeb->dirty_size)) {
182 c->dirty_size += jeb->wasted_size;
183 c->wasted_size -= jeb->wasted_size;
184 jeb->dirty_size += jeb->wasted_size;
185 jeb->wasted_size = 0;
186 if (VERYDIRTY(c, jeb->dirty_size)) {
187 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",
188 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
189 list_add_tail(&jeb->list, &c->very_dirty_list);
190 } else {
191 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to dirty_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->dirty_list);
194 }
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000195 } else {
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100196 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
197 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
198 list_add_tail(&jeb->list, &c->clean_list);
199 }
200 c->nextblock = NULL;
201
202}
203
204/* Select a new jeb for nextblock */
205
206static int jffs2_find_nextblock(struct jffs2_sb_info *c)
207{
208 struct list_head *next;
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000209
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100210 /* Take the next block off the 'free' list */
211
212 if (list_empty(&c->free_list)) {
213
214 if (!c->nr_erasing_blocks &&
215 !list_empty(&c->erasable_list)) {
216 struct jffs2_eraseblock *ejeb;
217
218 ejeb = list_entry(c->erasable_list.next, struct jffs2_eraseblock, list);
Akinobu Mitaf1166292006-06-26 00:24:46 -0700219 list_move_tail(&ejeb->list, &c->erase_pending_list);
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100220 c->nr_erasing_blocks++;
221 jffs2_erase_pending_trigger(c);
222 D1(printk(KERN_DEBUG "jffs2_find_nextblock: Triggering erase of erasable block at 0x%08x\n",
223 ejeb->offset));
224 }
225
226 if (!c->nr_erasing_blocks &&
227 !list_empty(&c->erasable_pending_wbuf_list)) {
228 D1(printk(KERN_DEBUG "jffs2_find_nextblock: Flushing write buffer\n"));
229 /* c->nextblock is NULL, no update to c->nextblock allowed */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 spin_unlock(&c->erase_completion_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 jffs2_flush_wbuf_pad(c);
232 spin_lock(&c->erase_completion_lock);
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100233 /* Have another go. It'll be on the erasable_list now */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 return -EAGAIN;
235 }
236
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100237 if (!c->nr_erasing_blocks) {
238 /* Ouch. We're in GC, or we wouldn't have got here.
239 And there's no space left. At all. */
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000240 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",
241 c->nr_erasing_blocks, c->nr_free_blocks, list_empty(&c->erasable_list)?"yes":"no",
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100242 list_empty(&c->erasing_list)?"yes":"no", list_empty(&c->erase_pending_list)?"yes":"no");
243 return -ENOSPC;
244 }
245
246 spin_unlock(&c->erase_completion_lock);
247 /* Don't wait for it; just erase one right now */
248 jffs2_erase_pending_blocks(c, 1);
249 spin_lock(&c->erase_completion_lock);
250
251 /* An erase may have failed, decreasing the
252 amount of free space available. So we must
253 restart from the beginning */
254 return -EAGAIN;
255 }
256
257 next = c->free_list.next;
258 list_del(next);
259 c->nextblock = list_entry(next, struct jffs2_eraseblock, list);
260 c->nr_free_blocks--;
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000261
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100262 jffs2_sum_reset_collected(c->summary); /* reset collected summary */
263
Alexander Belyakov5bf17232008-10-17 19:19:13 +0400264 /* adjust write buffer offset, else we get a non contiguous write bug */
265 if (!(c->wbuf_ofs % c->sector_size) && !c->wbuf_len)
266 c->wbuf_ofs = 0xffffffff;
267
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100268 D1(printk(KERN_DEBUG "jffs2_find_nextblock(): new nextblock = 0x%08x\n", c->nextblock->offset));
269
270 return 0;
271}
272
273/* Called with alloc sem _and_ erase_completion_lock */
David Woodhouse9fe48542006-05-23 00:38:06 +0100274static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
275 uint32_t *len, uint32_t sumsize)
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100276{
277 struct jffs2_eraseblock *jeb = c->nextblock;
David Woodhouse9fe48542006-05-23 00:38:06 +0100278 uint32_t reserved_size; /* for summary information at the end of the jeb */
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100279 int ret;
280
281 restart:
282 reserved_size = 0;
283
284 if (jffs2_sum_active() && (sumsize != JFFS2_SUMMARY_NOSUM_SIZE)) {
285 /* NOSUM_SIZE means not to generate summary */
286
287 if (jeb) {
288 reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE);
Artem B. Bityutskiy733802d2005-09-22 12:25:00 +0100289 dbg_summary("minsize=%d , jeb->free=%d ,"
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100290 "summary->size=%d , sumsize=%d\n",
291 minsize, jeb->free_size,
292 c->summary->sum_size, sumsize);
293 }
294
295 /* Is there enough space for writing out the current node, or we have to
296 write out summary information now, close this jeb and select new nextblock? */
297 if (jeb && (PAD(minsize) + PAD(c->summary->sum_size + sumsize +
298 JFFS2_SUMMARY_FRAME_SIZE) > jeb->free_size)) {
299
300 /* Has summary been disabled for this jeb? */
301 if (jffs2_sum_is_disabled(c->summary)) {
302 sumsize = JFFS2_SUMMARY_NOSUM_SIZE;
303 goto restart;
304 }
305
306 /* Writing out the collected summary information */
Artem B. Bityutskiy733802d2005-09-22 12:25:00 +0100307 dbg_summary("generating summary for 0x%08x.\n", jeb->offset);
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100308 ret = jffs2_sum_write_sumnode(c);
309
310 if (ret)
311 return ret;
312
313 if (jffs2_sum_is_disabled(c->summary)) {
314 /* jffs2_write_sumnode() couldn't write out the summary information
315 diabling summary for this jeb and free the collected information
316 */
317 sumsize = JFFS2_SUMMARY_NOSUM_SIZE;
318 goto restart;
319 }
320
321 jffs2_close_nextblock(c, jeb);
322 jeb = NULL;
Ferenc Havasi34c0e902005-09-16 13:58:20 +0100323 /* keep always valid value in reserved_size */
324 reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE);
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100325 }
326 } else {
327 if (jeb && minsize > jeb->free_size) {
David Woodhousefc6612f2006-06-18 18:35:10 +0100328 uint32_t waste;
329
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100330 /* Skip the end of this block and file it as having some dirty space */
331 /* If there's a pending write to it, flush now */
332
333 if (jffs2_wbuf_dirty(c)) {
334 spin_unlock(&c->erase_completion_lock);
335 D1(printk(KERN_DEBUG "jffs2_do_reserve_space: Flushing write buffer\n"));
336 jffs2_flush_wbuf_pad(c);
337 spin_lock(&c->erase_completion_lock);
338 jeb = c->nextblock;
339 goto restart;
340 }
341
David Woodhousefc6612f2006-06-18 18:35:10 +0100342 spin_unlock(&c->erase_completion_lock);
343
344 ret = jffs2_prealloc_raw_node_refs(c, jeb, 1);
345 if (ret)
346 return ret;
347 /* Just lock it again and continue. Nothing much can change because
348 we hold c->alloc_sem anyway. In fact, it's not entirely clear why
349 we hold c->erase_completion_lock in the majority of this function...
350 but that's a question for another (more caffeine-rich) day. */
351 spin_lock(&c->erase_completion_lock);
352
353 waste = jeb->free_size;
354 jffs2_link_node_ref(c, jeb,
355 (jeb->offset + c->sector_size - waste) | REF_OBSOLETE,
356 waste, NULL);
357 /* FIXME: that made it count as dirty. Convert to wasted */
358 jeb->dirty_size -= waste;
359 c->dirty_size -= waste;
360 jeb->wasted_size += waste;
361 c->wasted_size += waste;
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100362
363 jffs2_close_nextblock(c, jeb);
364 jeb = NULL;
365 }
366 }
367
368 if (!jeb) {
369
370 ret = jffs2_find_nextblock(c);
371 if (ret)
372 return ret;
373
374 jeb = c->nextblock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375
376 if (jeb->free_size != c->sector_size - c->cleanmarker_size) {
377 printk(KERN_WARNING "Eep. Block 0x%08x taken from free_list had free_size of 0x%08x!!\n", jeb->offset, jeb->free_size);
378 goto restart;
379 }
380 }
381 /* OK, jeb (==c->nextblock) is now pointing at a block which definitely has
382 enough space */
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100383 *len = jeb->free_size - reserved_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384
385 if (c->cleanmarker_size && jeb->used_size == c->cleanmarker_size &&
386 !jeb->first_node->next_in_ino) {
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000387 /* Only node in it beforehand was a CLEANMARKER node (we think).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 So mark it obsolete now that there's going to be another node
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000389 in the block. This will reduce used_size to zero but We've
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 already set c->nextblock so that jffs2_mark_node_obsolete()
391 won't try to refile it to the dirty_list.
392 */
393 spin_unlock(&c->erase_completion_lock);
394 jffs2_mark_node_obsolete(c, jeb->first_node);
395 spin_lock(&c->erase_completion_lock);
396 }
397
David Woodhouse9fe48542006-05-23 00:38:06 +0100398 D1(printk(KERN_DEBUG "jffs2_do_reserve_space(): Giving 0x%x bytes at 0x%x\n",
399 *len, jeb->offset + (c->sector_size - jeb->free_size)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 return 0;
401}
402
403/**
404 * jffs2_add_physical_node_ref - add a physical node reference to the list
405 * @c: superblock info
406 * @new: new node reference to add
407 * @len: length of this physical node
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 *
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000409 * Should only be used to report nodes for which space has been allocated
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 * by jffs2_reserve_space.
411 *
412 * Must be called with the alloc_sem held.
413 */
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000414
David Woodhouse2f785402006-05-24 02:04:45 +0100415struct jffs2_raw_node_ref *jffs2_add_physical_node_ref(struct jffs2_sb_info *c,
416 uint32_t ofs, uint32_t len,
417 struct jffs2_inode_cache *ic)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418{
419 struct jffs2_eraseblock *jeb;
David Woodhouse2f785402006-05-24 02:04:45 +0100420 struct jffs2_raw_node_ref *new;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421
David Woodhouse2f785402006-05-24 02:04:45 +0100422 jeb = &c->blocks[ofs / c->sector_size];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423
David Woodhouse2f785402006-05-24 02:04:45 +0100424 D1(printk(KERN_DEBUG "jffs2_add_physical_node_ref(): Node at 0x%x(%d), size 0x%x\n",
425 ofs & ~3, ofs & 3, len));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426#if 1
David Woodhouse2f785402006-05-24 02:04:45 +0100427 /* Allow non-obsolete nodes only to be added at the end of c->nextblock,
428 if c->nextblock is set. Note that wbuf.c will file obsolete nodes
429 even after refiling c->nextblock */
430 if ((c->nextblock || ((ofs & 3) != REF_OBSOLETE))
431 && (jeb != c->nextblock || (ofs & ~3) != jeb->offset + (c->sector_size - jeb->free_size))) {
David Woodhouse66bfaea2007-06-28 19:03:11 +0100432 printk(KERN_WARNING "argh. node added in wrong place at 0x%08x(%d)\n", ofs & ~3, ofs & 3);
433 if (c->nextblock)
434 printk(KERN_WARNING "nextblock 0x%08x", c->nextblock->offset);
435 else
436 printk(KERN_WARNING "No nextblock");
437 printk(", expected at %08x\n", jeb->offset + (c->sector_size - jeb->free_size));
David Woodhouse2f785402006-05-24 02:04:45 +0100438 return ERR_PTR(-EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 }
440#endif
441 spin_lock(&c->erase_completion_lock);
442
David Woodhouse2f785402006-05-24 02:04:45 +0100443 new = jffs2_link_node_ref(c, jeb, ofs, len, ic);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444
Estelle Hammache9b88f472005-01-28 18:53:05 +0000445 if (!jeb->free_size && !jeb->dirty_size && !ISDIRTY(jeb->wasted_size)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 /* If it lives on the dirty_list, jffs2_reserve_space will put it there */
447 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
448 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
449 if (jffs2_wbuf_dirty(c)) {
450 /* Flush the last write in the block if it's outstanding */
451 spin_unlock(&c->erase_completion_lock);
452 jffs2_flush_wbuf_pad(c);
453 spin_lock(&c->erase_completion_lock);
454 }
455
456 list_add_tail(&jeb->list, &c->clean_list);
457 c->nextblock = NULL;
458 }
Artem B. Bityutskiye0c8e422005-07-24 16:14:17 +0100459 jffs2_dbg_acct_sanity_check_nolock(c,jeb);
460 jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461
462 spin_unlock(&c->erase_completion_lock);
463
David Woodhouse2f785402006-05-24 02:04:45 +0100464 return new;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465}
466
467
468void jffs2_complete_reservation(struct jffs2_sb_info *c)
469{
470 D1(printk(KERN_DEBUG "jffs2_complete_reservation()\n"));
471 jffs2_garbage_collect_trigger(c);
David Woodhouseced22072008-04-22 15:13:40 +0100472 mutex_unlock(&c->alloc_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473}
474
475static inline int on_list(struct list_head *obj, struct list_head *head)
476{
477 struct list_head *this;
478
479 list_for_each(this, head) {
480 if (this == obj) {
481 D1(printk("%p is on list at %p\n", obj, head));
482 return 1;
483
484 }
485 }
486 return 0;
487}
488
489void jffs2_mark_node_obsolete(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref)
490{
491 struct jffs2_eraseblock *jeb;
492 int blocknr;
493 struct jffs2_unknown_node n;
494 int ret, addedsize;
495 size_t retlen;
David Woodhouse1417fc42006-05-20 16:20:19 +0100496 uint32_t freed_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497
David Woodhouse9bfeb692006-05-26 21:19:05 +0100498 if(unlikely(!ref)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 printk(KERN_NOTICE "EEEEEK. jffs2_mark_node_obsolete called with NULL node\n");
500 return;
501 }
502 if (ref_obsolete(ref)) {
503 D1(printk(KERN_DEBUG "jffs2_mark_node_obsolete called with already obsolete node at 0x%08x\n", ref_offset(ref)));
504 return;
505 }
506 blocknr = ref->flash_offset / c->sector_size;
507 if (blocknr >= c->nr_blocks) {
508 printk(KERN_NOTICE "raw node at 0x%08x is off the end of device!\n", ref->flash_offset);
509 BUG();
510 }
511 jeb = &c->blocks[blocknr];
512
513 if (jffs2_can_mark_obsolete(c) && !jffs2_is_readonly(c) &&
Artem B. Bityuckiy31fbdf72005-02-28 08:21:09 +0000514 !(c->flags & (JFFS2_SB_FLAG_SCANNING | JFFS2_SB_FLAG_BUILDING))) {
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000515 /* Hm. This may confuse static lock analysis. If any of the above
516 three conditions is false, we're going to return from this
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 function without actually obliterating any nodes or freeing
518 any jffs2_raw_node_refs. So we don't need to stop erases from
519 happening, or protect against people holding an obsolete
520 jffs2_raw_node_ref without the erase_completion_lock. */
David Woodhouseced22072008-04-22 15:13:40 +0100521 mutex_lock(&c->erase_free_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 }
523
524 spin_lock(&c->erase_completion_lock);
525
David Woodhouse1417fc42006-05-20 16:20:19 +0100526 freed_len = ref_totlen(c, jeb, ref);
527
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 if (ref_flags(ref) == REF_UNCHECKED) {
David Woodhouse1417fc42006-05-20 16:20:19 +0100529 D1(if (unlikely(jeb->unchecked_size < freed_len)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 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 +0100531 freed_len, blocknr, ref->flash_offset, jeb->used_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 BUG();
533 })
David Woodhouse1417fc42006-05-20 16:20:19 +0100534 D1(printk(KERN_DEBUG "Obsoleting previously unchecked node at 0x%08x of len %x: ", ref_offset(ref), freed_len));
535 jeb->unchecked_size -= freed_len;
536 c->unchecked_size -= freed_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 } else {
David Woodhouse1417fc42006-05-20 16:20:19 +0100538 D1(if (unlikely(jeb->used_size < freed_len)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 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 +0100540 freed_len, blocknr, ref->flash_offset, jeb->used_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 BUG();
542 })
David Woodhouse1417fc42006-05-20 16:20:19 +0100543 D1(printk(KERN_DEBUG "Obsoleting node at 0x%08x of len %#x: ", ref_offset(ref), freed_len));
544 jeb->used_size -= freed_len;
545 c->used_size -= freed_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 }
547
548 // Take care, that wasted size is taken into concern
David Woodhouse1417fc42006-05-20 16:20:19 +0100549 if ((jeb->dirty_size || ISDIRTY(jeb->wasted_size + freed_len)) && jeb != c->nextblock) {
David Woodhousec7c16c82006-05-24 14:24:02 +0100550 D1(printk("Dirtying\n"));
David Woodhouse1417fc42006-05-20 16:20:19 +0100551 addedsize = freed_len;
552 jeb->dirty_size += freed_len;
553 c->dirty_size += freed_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554
555 /* Convert wasted space to dirty, if not a bad block */
556 if (jeb->wasted_size) {
557 if (on_list(&jeb->list, &c->bad_used_list)) {
558 D1(printk(KERN_DEBUG "Leaving block at %08x on the bad_used_list\n",
559 jeb->offset));
560 addedsize = 0; /* To fool the refiling code later */
561 } else {
562 D1(printk(KERN_DEBUG "Converting %d bytes of wasted space to dirty in block at %08x\n",
563 jeb->wasted_size, jeb->offset));
564 addedsize += jeb->wasted_size;
565 jeb->dirty_size += jeb->wasted_size;
566 c->dirty_size += jeb->wasted_size;
567 c->wasted_size -= jeb->wasted_size;
568 jeb->wasted_size = 0;
569 }
570 }
571 } else {
David Woodhousec7c16c82006-05-24 14:24:02 +0100572 D1(printk("Wasting\n"));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 addedsize = 0;
David Woodhouse1417fc42006-05-20 16:20:19 +0100574 jeb->wasted_size += freed_len;
575 c->wasted_size += freed_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 }
577 ref->flash_offset = ref_offset(ref) | REF_OBSOLETE;
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000578
Artem B. Bityutskiye0c8e422005-07-24 16:14:17 +0100579 jffs2_dbg_acct_sanity_check_nolock(c, jeb);
580 jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581
Artem B. Bityuckiy31fbdf72005-02-28 08:21:09 +0000582 if (c->flags & JFFS2_SB_FLAG_SCANNING) {
583 /* Flash scanning is in progress. Don't muck about with the block
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 lists because they're not ready yet, and don't actually
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000585 obliterate nodes that look obsolete. If they weren't
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 marked obsolete on the flash at the time they _became_
587 obsolete, there was probably a reason for that. */
588 spin_unlock(&c->erase_completion_lock);
589 /* We didn't lock the erase_free_sem */
590 return;
591 }
592
593 if (jeb == c->nextblock) {
594 D2(printk(KERN_DEBUG "Not moving nextblock 0x%08x to dirty/erase_pending list\n", jeb->offset));
595 } else if (!jeb->used_size && !jeb->unchecked_size) {
596 if (jeb == c->gcblock) {
597 D1(printk(KERN_DEBUG "gcblock at 0x%08x completely dirtied. Clearing gcblock...\n", jeb->offset));
598 c->gcblock = NULL;
599 } else {
600 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x completely dirtied. Removing from (dirty?) list...\n", jeb->offset));
601 list_del(&jeb->list);
602 }
603 if (jffs2_wbuf_dirty(c)) {
604 D1(printk(KERN_DEBUG "...and adding to erasable_pending_wbuf_list\n"));
605 list_add_tail(&jeb->list, &c->erasable_pending_wbuf_list);
606 } else {
607 if (jiffies & 127) {
608 /* Most of the time, we just erase it immediately. Otherwise we
609 spend ages scanning it on mount, etc. */
610 D1(printk(KERN_DEBUG "...and adding to erase_pending_list\n"));
611 list_add_tail(&jeb->list, &c->erase_pending_list);
612 c->nr_erasing_blocks++;
613 jffs2_erase_pending_trigger(c);
614 } else {
615 /* Sometimes, however, we leave it elsewhere so it doesn't get
616 immediately reused, and we spread the load a bit. */
617 D1(printk(KERN_DEBUG "...and adding to erasable_list\n"));
618 list_add_tail(&jeb->list, &c->erasable_list);
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000619 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 }
621 D1(printk(KERN_DEBUG "Done OK\n"));
622 } else if (jeb == c->gcblock) {
623 D2(printk(KERN_DEBUG "Not moving gcblock 0x%08x to dirty_list\n", jeb->offset));
624 } else if (ISDIRTY(jeb->dirty_size) && !ISDIRTY(jeb->dirty_size - addedsize)) {
625 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x is freshly dirtied. Removing from clean list...\n", jeb->offset));
626 list_del(&jeb->list);
627 D1(printk(KERN_DEBUG "...and adding to dirty_list\n"));
628 list_add_tail(&jeb->list, &c->dirty_list);
629 } else if (VERYDIRTY(c, jeb->dirty_size) &&
630 !VERYDIRTY(c, jeb->dirty_size - addedsize)) {
631 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x is now very dirty. Removing from dirty list...\n", jeb->offset));
632 list_del(&jeb->list);
633 D1(printk(KERN_DEBUG "...and adding to very_dirty_list\n"));
634 list_add_tail(&jeb->list, &c->very_dirty_list);
635 } else {
636 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 +0000637 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
638 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639
640 spin_unlock(&c->erase_completion_lock);
641
Artem B. Bityuckiy31fbdf72005-02-28 08:21:09 +0000642 if (!jffs2_can_mark_obsolete(c) || jffs2_is_readonly(c) ||
643 (c->flags & JFFS2_SB_FLAG_BUILDING)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 /* We didn't lock the erase_free_sem */
645 return;
646 }
647
648 /* The erase_free_sem is locked, and has been since before we marked the node obsolete
649 and potentially put its eraseblock onto the erase_pending_list. Thus, we know that
650 the block hasn't _already_ been erased, and that 'ref' itself hasn't been freed yet
David Woodhousec38c1b62006-05-25 01:38:27 +0100651 by jffs2_free_jeb_node_refs() in erase.c. Which is nice. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652
653 D1(printk(KERN_DEBUG "obliterating obsoleted node at 0x%08x\n", ref_offset(ref)));
654 ret = jffs2_flash_read(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
655 if (ret) {
656 printk(KERN_WARNING "Read error reading from obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret);
657 goto out_erase_sem;
658 }
659 if (retlen != sizeof(n)) {
660 printk(KERN_WARNING "Short read from obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen);
661 goto out_erase_sem;
662 }
David Woodhouse1417fc42006-05-20 16:20:19 +0100663 if (PAD(je32_to_cpu(n.totlen)) != PAD(freed_len)) {
664 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 -0700665 goto out_erase_sem;
666 }
667 if (!(je16_to_cpu(n.nodetype) & JFFS2_NODE_ACCURATE)) {
668 D1(printk(KERN_DEBUG "Node at 0x%08x was already marked obsolete (nodetype 0x%04x)\n", ref_offset(ref), je16_to_cpu(n.nodetype)));
669 goto out_erase_sem;
670 }
671 /* XXX FIXME: This is ugly now */
672 n.nodetype = cpu_to_je16(je16_to_cpu(n.nodetype) & ~JFFS2_NODE_ACCURATE);
673 ret = jffs2_flash_write(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
674 if (ret) {
675 printk(KERN_WARNING "Write error in obliterating obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret);
676 goto out_erase_sem;
677 }
678 if (retlen != sizeof(n)) {
679 printk(KERN_WARNING "Short write in obliterating obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen);
680 goto out_erase_sem;
681 }
682
683 /* Nodes which have been marked obsolete no longer need to be
684 associated with any inode. Remove them from the per-inode list.
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000685
686 Note we can't do this for NAND at the moment because we need
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 obsolete dirent nodes to stay on the lists, because of the
688 horridness in jffs2_garbage_collect_deletion_dirent(). Also
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000689 because we delete the inocache, and on NAND we need that to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 stay around until all the nodes are actually erased, in order
691 to stop us from giving the same inode number to another newly
692 created inode. */
693 if (ref->next_in_ino) {
694 struct jffs2_inode_cache *ic;
695 struct jffs2_raw_node_ref **p;
696
697 spin_lock(&c->erase_completion_lock);
698
699 ic = jffs2_raw_ref_to_ic(ref);
700 for (p = &ic->nodes; (*p) != ref; p = &((*p)->next_in_ino))
701 ;
702
703 *p = ref->next_in_ino;
704 ref->next_in_ino = NULL;
705
KaiGai Koheic9f700f2006-06-11 10:35:15 +0900706 switch (ic->class) {
707#ifdef CONFIG_JFFS2_FS_XATTR
708 case RAWNODE_CLASS_XATTR_DATUM:
709 jffs2_release_xattr_datum(c, (struct jffs2_xattr_datum *)ic);
710 break;
711 case RAWNODE_CLASS_XATTR_REF:
712 jffs2_release_xattr_ref(c, (struct jffs2_xattr_ref *)ic);
713 break;
714#endif
715 default:
David Woodhouse27c72b02008-05-01 18:47:17 +0100716 if (ic->nodes == (void *)ic && ic->pino_nlink == 0)
KaiGai Koheic9f700f2006-06-11 10:35:15 +0900717 jffs2_del_ino_cache(c, ic);
718 break;
719 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 spin_unlock(&c->erase_completion_lock);
721 }
722
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 out_erase_sem:
David Woodhouseced22072008-04-22 15:13:40 +0100724 mutex_unlock(&c->erase_free_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725}
726
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727int jffs2_thread_should_wake(struct jffs2_sb_info *c)
728{
729 int ret = 0;
730 uint32_t dirty;
David Woodhouse8fb870df2007-10-06 15:12:58 -0400731 int nr_very_dirty = 0;
732 struct jffs2_eraseblock *jeb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733
734 if (c->unchecked_size) {
735 D1(printk(KERN_DEBUG "jffs2_thread_should_wake(): unchecked_size %d, checked_ino #%d\n",
736 c->unchecked_size, c->checked_ino));
737 return 1;
738 }
739
740 /* dirty_size contains blocks on erase_pending_list
741 * those blocks are counted in c->nr_erasing_blocks.
742 * If one block is actually erased, it is not longer counted as dirty_space
743 * but it is counted in c->nr_erasing_blocks, so we add it and subtract it
744 * with c->nr_erasing_blocks * c->sector_size again.
745 * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks
746 * This helps us to force gc and pick eventually a clean block to spread the load.
747 */
748 dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size;
749
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000750 if (c->nr_free_blocks + c->nr_erasing_blocks < c->resv_blocks_gctrigger &&
751 (dirty > c->nospc_dirty_size))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 ret = 1;
753
David Woodhouse8fb870df2007-10-06 15:12:58 -0400754 list_for_each_entry(jeb, &c->very_dirty_list, list) {
755 nr_very_dirty++;
756 if (nr_very_dirty == c->vdirty_blocks_gctrigger) {
757 ret = 1;
David Woodhousea8c68f32007-10-13 11:32:16 +0100758 /* In debug mode, actually go through and count them all */
759 D1(continue);
760 break;
David Woodhouse8fb870df2007-10-06 15:12:58 -0400761 }
762 }
763
764 D1(printk(KERN_DEBUG "jffs2_thread_should_wake(): nr_free_blocks %d, nr_erasing_blocks %d, dirty_size 0x%x, vdirty_blocks %d: %s\n",
765 c->nr_free_blocks, c->nr_erasing_blocks, c->dirty_size, nr_very_dirty, ret?"yes":"no"));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766
767 return ret;
768}