Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * JFFS2 -- Journalling Flash File System, Version 2. | |
3 | * | |
c00c310e | 4 | * Copyright © 2001-2007 Red Hat, Inc. |
1da177e4 LT |
5 | * |
6 | * Created by David Woodhouse <dwmw2@infradead.org> | |
7 | * | |
8 | * For licensing information, see the file 'LICENCE' in this directory. | |
9 | * | |
1da177e4 LT |
10 | */ |
11 | ||
12 | #include <linux/kernel.h> | |
1da177e4 LT |
13 | #include <linux/mtd/mtd.h> |
14 | #include <linux/compiler.h> | |
15 | #include <linux/sched.h> /* For cond_resched() */ | |
16 | #include "nodelist.h" | |
e631ddba | 17 | #include "debug.h" |
1da177e4 LT |
18 | |
19 | /** | |
20 | * jffs2_reserve_space - request physical space to write nodes to flash | |
21 | * @c: superblock info | |
22 | * @minsize: Minimum acceptable size of allocation | |
1da177e4 LT |
23 | * @len: Returned value of allocation length |
24 | * @prio: Allocation type - ALLOC_{NORMAL,DELETION} | |
25 | * | |
26 | * Requests a block of physical space on the flash. Returns zero for success | |
9fe4854c DW |
27 | * and puts 'len' into the appropriate place, or returns -ENOSPC or other |
28 | * error if appropriate. Doesn't return len since that's | |
1da177e4 LT |
29 | * |
30 | * If it returns zero, jffs2_reserve_space() also downs the per-filesystem | |
31 | * allocation semaphore, to prevent more than one allocation from being | |
32 | * active at any time. The semaphore is later released by jffs2_commit_allocation() | |
33 | * | |
34 | * jffs2_reserve_space() may trigger garbage collection in order to make room | |
35 | * for the requested allocation. | |
36 | */ | |
37 | ||
e631ddba | 38 | static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize, |
9fe4854c | 39 | uint32_t *len, uint32_t sumsize); |
1da177e4 | 40 | |
9fe4854c | 41 | int jffs2_reserve_space(struct jffs2_sb_info *c, uint32_t minsize, |
e631ddba | 42 | uint32_t *len, int prio, uint32_t sumsize) |
1da177e4 LT |
43 | { |
44 | int ret = -EAGAIN; | |
45 | int blocksneeded = c->resv_blocks_write; | |
46 | /* align it */ | |
47 | minsize = PAD(minsize); | |
48 | ||
9c261b33 | 49 | jffs2_dbg(1, "%s(): Requested 0x%x bytes\n", __func__, minsize); |
ced22070 | 50 | mutex_lock(&c->alloc_sem); |
1da177e4 | 51 | |
9c261b33 | 52 | jffs2_dbg(1, "%s(): alloc sem got\n", __func__); |
1da177e4 LT |
53 | |
54 | spin_lock(&c->erase_completion_lock); | |
55 | ||
56 | /* this needs a little more thought (true <tglx> :)) */ | |
57 | while(ret == -EAGAIN) { | |
58 | while(c->nr_free_blocks + c->nr_erasing_blocks < blocksneeded) { | |
1da177e4 LT |
59 | uint32_t dirty, avail; |
60 | ||
61 | /* calculate real dirty size | |
62 | * dirty_size contains blocks on erase_pending_list | |
63 | * those blocks are counted in c->nr_erasing_blocks. | |
64 | * If one block is actually erased, it is not longer counted as dirty_space | |
65 | * but it is counted in c->nr_erasing_blocks, so we add it and subtract it | |
66 | * with c->nr_erasing_blocks * c->sector_size again. | |
67 | * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks | |
68 | * This helps us to force gc and pick eventually a clean block to spread the load. | |
69 | * We add unchecked_size here, as we hopefully will find some space to use. | |
70 | * This will affect the sum only once, as gc first finishes checking | |
71 | * of nodes. | |
72 | */ | |
73 | dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size + c->unchecked_size; | |
74 | if (dirty < c->nospc_dirty_size) { | |
75 | if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) { | |
9c261b33 JP |
76 | jffs2_dbg(1, "%s(): Low on dirty space to GC, but it's a deletion. Allowing...\n", |
77 | __func__); | |
1da177e4 LT |
78 | break; |
79 | } | |
9c261b33 JP |
80 | jffs2_dbg(1, "dirty size 0x%08x + unchecked_size 0x%08x < nospc_dirty_size 0x%08x, returning -ENOSPC\n", |
81 | dirty, c->unchecked_size, | |
82 | c->sector_size); | |
1da177e4 LT |
83 | |
84 | spin_unlock(&c->erase_completion_lock); | |
ced22070 | 85 | mutex_unlock(&c->alloc_sem); |
1da177e4 LT |
86 | return -ENOSPC; |
87 | } | |
182ec4ee | 88 | |
1da177e4 LT |
89 | /* Calc possibly available space. Possibly available means that we |
90 | * don't know, if unchecked size contains obsoleted nodes, which could give us some | |
91 | * more usable space. This will affect the sum only once, as gc first finishes checking | |
92 | * of nodes. | |
182ec4ee | 93 | + Return -ENOSPC, if the maximum possibly available space is less or equal than |
1da177e4 LT |
94 | * blocksneeded * sector_size. |
95 | * This blocks endless gc looping on a filesystem, which is nearly full, even if | |
96 | * the check above passes. | |
97 | */ | |
98 | avail = c->free_size + c->dirty_size + c->erasing_size + c->unchecked_size; | |
99 | if ( (avail / c->sector_size) <= blocksneeded) { | |
100 | if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) { | |
9c261b33 JP |
101 | jffs2_dbg(1, "%s(): Low on possibly available space, but it's a deletion. Allowing...\n", |
102 | __func__); | |
1da177e4 LT |
103 | break; |
104 | } | |
105 | ||
9c261b33 JP |
106 | jffs2_dbg(1, "max. available size 0x%08x < blocksneeded * sector_size 0x%08x, returning -ENOSPC\n", |
107 | avail, blocksneeded * c->sector_size); | |
1da177e4 | 108 | spin_unlock(&c->erase_completion_lock); |
ced22070 | 109 | mutex_unlock(&c->alloc_sem); |
1da177e4 LT |
110 | return -ENOSPC; |
111 | } | |
112 | ||
ced22070 | 113 | mutex_unlock(&c->alloc_sem); |
1da177e4 | 114 | |
9c261b33 JP |
115 | jffs2_dbg(1, "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", |
116 | c->nr_free_blocks, c->nr_erasing_blocks, | |
117 | c->free_size, c->dirty_size, c->wasted_size, | |
118 | c->used_size, c->erasing_size, c->bad_size, | |
119 | c->free_size + c->dirty_size + | |
120 | c->wasted_size + c->used_size + | |
121 | c->erasing_size + c->bad_size, | |
122 | c->flash_size); | |
1da177e4 | 123 | spin_unlock(&c->erase_completion_lock); |
182ec4ee | 124 | |
1da177e4 | 125 | ret = jffs2_garbage_collect_pass(c); |
422b1202 | 126 | |
0717bf84 DW |
127 | if (ret == -EAGAIN) { |
128 | spin_lock(&c->erase_completion_lock); | |
129 | if (c->nr_erasing_blocks && | |
130 | list_empty(&c->erase_pending_list) && | |
131 | list_empty(&c->erase_complete_list)) { | |
132 | DECLARE_WAITQUEUE(wait, current); | |
133 | set_current_state(TASK_UNINTERRUPTIBLE); | |
134 | add_wait_queue(&c->erase_wait, &wait); | |
9c261b33 JP |
135 | jffs2_dbg(1, "%s waiting for erase to complete\n", |
136 | __func__); | |
0717bf84 DW |
137 | spin_unlock(&c->erase_completion_lock); |
138 | ||
139 | schedule(); | |
140 | } else | |
141 | spin_unlock(&c->erase_completion_lock); | |
142 | } else if (ret) | |
1da177e4 LT |
143 | return ret; |
144 | ||
145 | cond_resched(); | |
146 | ||
147 | if (signal_pending(current)) | |
148 | return -EINTR; | |
149 | ||
ced22070 | 150 | mutex_lock(&c->alloc_sem); |
1da177e4 LT |
151 | spin_lock(&c->erase_completion_lock); |
152 | } | |
153 | ||
9fe4854c | 154 | ret = jffs2_do_reserve_space(c, minsize, len, sumsize); |
1da177e4 | 155 | if (ret) { |
9c261b33 | 156 | jffs2_dbg(1, "%s(): ret is %d\n", __func__, ret); |
1da177e4 LT |
157 | } |
158 | } | |
159 | spin_unlock(&c->erase_completion_lock); | |
2f785402 | 160 | if (!ret) |
046b8b98 | 161 | ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1); |
1da177e4 | 162 | if (ret) |
ced22070 | 163 | mutex_unlock(&c->alloc_sem); |
1da177e4 LT |
164 | return ret; |
165 | } | |
166 | ||
9fe4854c DW |
167 | int jffs2_reserve_space_gc(struct jffs2_sb_info *c, uint32_t minsize, |
168 | uint32_t *len, uint32_t sumsize) | |
1da177e4 LT |
169 | { |
170 | int ret = -EAGAIN; | |
171 | minsize = PAD(minsize); | |
172 | ||
9c261b33 | 173 | jffs2_dbg(1, "%s(): Requested 0x%x bytes\n", __func__, minsize); |
1da177e4 LT |
174 | |
175 | spin_lock(&c->erase_completion_lock); | |
176 | while(ret == -EAGAIN) { | |
9fe4854c | 177 | ret = jffs2_do_reserve_space(c, minsize, len, sumsize); |
1da177e4 | 178 | if (ret) { |
9c261b33 JP |
179 | jffs2_dbg(1, "%s(): looping, ret is %d\n", |
180 | __func__, ret); | |
1da177e4 LT |
181 | } |
182 | } | |
183 | spin_unlock(&c->erase_completion_lock); | |
2f785402 | 184 | if (!ret) |
046b8b98 | 185 | ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1); |
2f785402 | 186 | |
1da177e4 LT |
187 | return ret; |
188 | } | |
189 | ||
e631ddba FH |
190 | |
191 | /* Classify nextblock (clean, dirty of verydirty) and force to select an other one */ | |
192 | ||
193 | static void jffs2_close_nextblock(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb) | |
1da177e4 | 194 | { |
e631ddba | 195 | |
99c2594f | 196 | if (c->nextblock == NULL) { |
9c261b33 JP |
197 | jffs2_dbg(1, "%s(): Erase block at 0x%08x has already been placed in a list\n", |
198 | __func__, jeb->offset); | |
99c2594f AH |
199 | return; |
200 | } | |
e631ddba FH |
201 | /* Check, if we have a dirty block now, or if it was dirty already */ |
202 | if (ISDIRTY (jeb->wasted_size + jeb->dirty_size)) { | |
203 | c->dirty_size += jeb->wasted_size; | |
204 | c->wasted_size -= jeb->wasted_size; | |
205 | jeb->dirty_size += jeb->wasted_size; | |
206 | jeb->wasted_size = 0; | |
207 | if (VERYDIRTY(c, jeb->dirty_size)) { | |
9c261b33 JP |
208 | jffs2_dbg(1, "Adding full erase block at 0x%08x to very_dirty_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n", |
209 | jeb->offset, jeb->free_size, jeb->dirty_size, | |
210 | jeb->used_size); | |
e631ddba FH |
211 | list_add_tail(&jeb->list, &c->very_dirty_list); |
212 | } else { | |
9c261b33 JP |
213 | jffs2_dbg(1, "Adding full erase block at 0x%08x to dirty_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n", |
214 | jeb->offset, jeb->free_size, jeb->dirty_size, | |
215 | jeb->used_size); | |
e631ddba FH |
216 | list_add_tail(&jeb->list, &c->dirty_list); |
217 | } | |
182ec4ee | 218 | } else { |
9c261b33 JP |
219 | jffs2_dbg(1, "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n", |
220 | jeb->offset, jeb->free_size, jeb->dirty_size, | |
221 | jeb->used_size); | |
e631ddba FH |
222 | list_add_tail(&jeb->list, &c->clean_list); |
223 | } | |
224 | c->nextblock = NULL; | |
225 | ||
226 | } | |
227 | ||
228 | /* Select a new jeb for nextblock */ | |
229 | ||
230 | static int jffs2_find_nextblock(struct jffs2_sb_info *c) | |
231 | { | |
232 | struct list_head *next; | |
182ec4ee | 233 | |
e631ddba FH |
234 | /* Take the next block off the 'free' list */ |
235 | ||
236 | if (list_empty(&c->free_list)) { | |
237 | ||
238 | if (!c->nr_erasing_blocks && | |
239 | !list_empty(&c->erasable_list)) { | |
240 | struct jffs2_eraseblock *ejeb; | |
241 | ||
242 | ejeb = list_entry(c->erasable_list.next, struct jffs2_eraseblock, list); | |
f116629d | 243 | list_move_tail(&ejeb->list, &c->erase_pending_list); |
e631ddba | 244 | c->nr_erasing_blocks++; |
ae3b6ba0 | 245 | jffs2_garbage_collect_trigger(c); |
9c261b33 JP |
246 | jffs2_dbg(1, "%s(): Triggering erase of erasable block at 0x%08x\n", |
247 | __func__, ejeb->offset); | |
e631ddba FH |
248 | } |
249 | ||
250 | if (!c->nr_erasing_blocks && | |
251 | !list_empty(&c->erasable_pending_wbuf_list)) { | |
9c261b33 JP |
252 | jffs2_dbg(1, "%s(): Flushing write buffer\n", |
253 | __func__); | |
e631ddba | 254 | /* c->nextblock is NULL, no update to c->nextblock allowed */ |
1da177e4 | 255 | spin_unlock(&c->erase_completion_lock); |
1da177e4 LT |
256 | jffs2_flush_wbuf_pad(c); |
257 | spin_lock(&c->erase_completion_lock); | |
e631ddba FH |
258 | /* Have another go. It'll be on the erasable_list now */ |
259 | return -EAGAIN; | |
1da177e4 | 260 | } |
e631ddba FH |
261 | |
262 | if (!c->nr_erasing_blocks) { | |
263 | /* Ouch. We're in GC, or we wouldn't have got here. | |
264 | And there's no space left. At all. */ | |
182ec4ee TG |
265 | 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", |
266 | c->nr_erasing_blocks, c->nr_free_blocks, list_empty(&c->erasable_list)?"yes":"no", | |
e631ddba FH |
267 | list_empty(&c->erasing_list)?"yes":"no", list_empty(&c->erase_pending_list)?"yes":"no"); |
268 | return -ENOSPC; | |
1da177e4 | 269 | } |
e631ddba FH |
270 | |
271 | spin_unlock(&c->erase_completion_lock); | |
272 | /* Don't wait for it; just erase one right now */ | |
273 | jffs2_erase_pending_blocks(c, 1); | |
274 | spin_lock(&c->erase_completion_lock); | |
275 | ||
276 | /* An erase may have failed, decreasing the | |
277 | amount of free space available. So we must | |
278 | restart from the beginning */ | |
279 | return -EAGAIN; | |
1da177e4 | 280 | } |
e631ddba FH |
281 | |
282 | next = c->free_list.next; | |
283 | list_del(next); | |
284 | c->nextblock = list_entry(next, struct jffs2_eraseblock, list); | |
285 | c->nr_free_blocks--; | |
182ec4ee | 286 | |
e631ddba FH |
287 | jffs2_sum_reset_collected(c->summary); /* reset collected summary */ |
288 | ||
f04de505 | 289 | #ifdef CONFIG_JFFS2_FS_WRITEBUFFER |
5bf17237 AB |
290 | /* adjust write buffer offset, else we get a non contiguous write bug */ |
291 | if (!(c->wbuf_ofs % c->sector_size) && !c->wbuf_len) | |
292 | c->wbuf_ofs = 0xffffffff; | |
f04de505 | 293 | #endif |
5bf17237 | 294 | |
9c261b33 JP |
295 | jffs2_dbg(1, "%s(): new nextblock = 0x%08x\n", |
296 | __func__, c->nextblock->offset); | |
e631ddba FH |
297 | |
298 | return 0; | |
299 | } | |
300 | ||
301 | /* Called with alloc sem _and_ erase_completion_lock */ | |
9fe4854c DW |
302 | static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize, |
303 | uint32_t *len, uint32_t sumsize) | |
e631ddba FH |
304 | { |
305 | struct jffs2_eraseblock *jeb = c->nextblock; | |
9fe4854c | 306 | uint32_t reserved_size; /* for summary information at the end of the jeb */ |
e631ddba FH |
307 | int ret; |
308 | ||
309 | restart: | |
310 | reserved_size = 0; | |
311 | ||
312 | if (jffs2_sum_active() && (sumsize != JFFS2_SUMMARY_NOSUM_SIZE)) { | |
313 | /* NOSUM_SIZE means not to generate summary */ | |
314 | ||
315 | if (jeb) { | |
316 | reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE); | |
733802d9 | 317 | dbg_summary("minsize=%d , jeb->free=%d ," |
e631ddba FH |
318 | "summary->size=%d , sumsize=%d\n", |
319 | minsize, jeb->free_size, | |
320 | c->summary->sum_size, sumsize); | |
321 | } | |
322 | ||
323 | /* Is there enough space for writing out the current node, or we have to | |
324 | write out summary information now, close this jeb and select new nextblock? */ | |
325 | if (jeb && (PAD(minsize) + PAD(c->summary->sum_size + sumsize + | |
326 | JFFS2_SUMMARY_FRAME_SIZE) > jeb->free_size)) { | |
327 | ||
328 | /* Has summary been disabled for this jeb? */ | |
329 | if (jffs2_sum_is_disabled(c->summary)) { | |
330 | sumsize = JFFS2_SUMMARY_NOSUM_SIZE; | |
331 | goto restart; | |
1da177e4 LT |
332 | } |
333 | ||
e631ddba | 334 | /* Writing out the collected summary information */ |
733802d9 | 335 | dbg_summary("generating summary for 0x%08x.\n", jeb->offset); |
e631ddba FH |
336 | ret = jffs2_sum_write_sumnode(c); |
337 | ||
338 | if (ret) | |
339 | return ret; | |
340 | ||
341 | if (jffs2_sum_is_disabled(c->summary)) { | |
342 | /* jffs2_write_sumnode() couldn't write out the summary information | |
343 | diabling summary for this jeb and free the collected information | |
344 | */ | |
345 | sumsize = JFFS2_SUMMARY_NOSUM_SIZE; | |
346 | goto restart; | |
347 | } | |
348 | ||
349 | jffs2_close_nextblock(c, jeb); | |
350 | jeb = NULL; | |
34c0e906 FH |
351 | /* keep always valid value in reserved_size */ |
352 | reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE); | |
e631ddba FH |
353 | } |
354 | } else { | |
355 | if (jeb && minsize > jeb->free_size) { | |
fc6612f6 DW |
356 | uint32_t waste; |
357 | ||
e631ddba FH |
358 | /* Skip the end of this block and file it as having some dirty space */ |
359 | /* If there's a pending write to it, flush now */ | |
360 | ||
361 | if (jffs2_wbuf_dirty(c)) { | |
1da177e4 | 362 | spin_unlock(&c->erase_completion_lock); |
9c261b33 JP |
363 | jffs2_dbg(1, "%s(): Flushing write buffer\n", |
364 | __func__); | |
1da177e4 LT |
365 | jffs2_flush_wbuf_pad(c); |
366 | spin_lock(&c->erase_completion_lock); | |
e631ddba FH |
367 | jeb = c->nextblock; |
368 | goto restart; | |
1da177e4 LT |
369 | } |
370 | ||
fc6612f6 DW |
371 | spin_unlock(&c->erase_completion_lock); |
372 | ||
373 | ret = jffs2_prealloc_raw_node_refs(c, jeb, 1); | |
374 | if (ret) | |
375 | return ret; | |
376 | /* Just lock it again and continue. Nothing much can change because | |
377 | we hold c->alloc_sem anyway. In fact, it's not entirely clear why | |
378 | we hold c->erase_completion_lock in the majority of this function... | |
379 | but that's a question for another (more caffeine-rich) day. */ | |
380 | spin_lock(&c->erase_completion_lock); | |
381 | ||
382 | waste = jeb->free_size; | |
383 | jffs2_link_node_ref(c, jeb, | |
384 | (jeb->offset + c->sector_size - waste) | REF_OBSOLETE, | |
385 | waste, NULL); | |
386 | /* FIXME: that made it count as dirty. Convert to wasted */ | |
387 | jeb->dirty_size -= waste; | |
388 | c->dirty_size -= waste; | |
389 | jeb->wasted_size += waste; | |
390 | c->wasted_size += waste; | |
1da177e4 | 391 | |
e631ddba FH |
392 | jffs2_close_nextblock(c, jeb); |
393 | jeb = NULL; | |
1da177e4 | 394 | } |
e631ddba FH |
395 | } |
396 | ||
397 | if (!jeb) { | |
398 | ||
399 | ret = jffs2_find_nextblock(c); | |
400 | if (ret) | |
401 | return ret; | |
1da177e4 | 402 | |
e631ddba | 403 | jeb = c->nextblock; |
1da177e4 LT |
404 | |
405 | if (jeb->free_size != c->sector_size - c->cleanmarker_size) { | |
406 | printk(KERN_WARNING "Eep. Block 0x%08x taken from free_list had free_size of 0x%08x!!\n", jeb->offset, jeb->free_size); | |
407 | goto restart; | |
408 | } | |
409 | } | |
410 | /* OK, jeb (==c->nextblock) is now pointing at a block which definitely has | |
411 | enough space */ | |
e631ddba | 412 | *len = jeb->free_size - reserved_size; |
1da177e4 LT |
413 | |
414 | if (c->cleanmarker_size && jeb->used_size == c->cleanmarker_size && | |
415 | !jeb->first_node->next_in_ino) { | |
182ec4ee | 416 | /* Only node in it beforehand was a CLEANMARKER node (we think). |
1da177e4 | 417 | So mark it obsolete now that there's going to be another node |
182ec4ee | 418 | in the block. This will reduce used_size to zero but We've |
1da177e4 LT |
419 | already set c->nextblock so that jffs2_mark_node_obsolete() |
420 | won't try to refile it to the dirty_list. | |
421 | */ | |
422 | spin_unlock(&c->erase_completion_lock); | |
423 | jffs2_mark_node_obsolete(c, jeb->first_node); | |
424 | spin_lock(&c->erase_completion_lock); | |
425 | } | |
426 | ||
9c261b33 JP |
427 | jffs2_dbg(1, "%s(): Giving 0x%x bytes at 0x%x\n", |
428 | __func__, | |
429 | *len, jeb->offset + (c->sector_size - jeb->free_size)); | |
1da177e4 LT |
430 | return 0; |
431 | } | |
432 | ||
433 | /** | |
434 | * jffs2_add_physical_node_ref - add a physical node reference to the list | |
435 | * @c: superblock info | |
436 | * @new: new node reference to add | |
437 | * @len: length of this physical node | |
1da177e4 | 438 | * |
182ec4ee | 439 | * Should only be used to report nodes for which space has been allocated |
1da177e4 LT |
440 | * by jffs2_reserve_space. |
441 | * | |
442 | * Must be called with the alloc_sem held. | |
443 | */ | |
182ec4ee | 444 | |
2f785402 DW |
445 | struct jffs2_raw_node_ref *jffs2_add_physical_node_ref(struct jffs2_sb_info *c, |
446 | uint32_t ofs, uint32_t len, | |
447 | struct jffs2_inode_cache *ic) | |
1da177e4 LT |
448 | { |
449 | struct jffs2_eraseblock *jeb; | |
2f785402 | 450 | struct jffs2_raw_node_ref *new; |
1da177e4 | 451 | |
2f785402 | 452 | jeb = &c->blocks[ofs / c->sector_size]; |
1da177e4 | 453 | |
9c261b33 JP |
454 | jffs2_dbg(1, "%s(): Node at 0x%x(%d), size 0x%x\n", |
455 | __func__, ofs & ~3, ofs & 3, len); | |
1da177e4 | 456 | #if 1 |
2f785402 DW |
457 | /* Allow non-obsolete nodes only to be added at the end of c->nextblock, |
458 | if c->nextblock is set. Note that wbuf.c will file obsolete nodes | |
459 | even after refiling c->nextblock */ | |
460 | if ((c->nextblock || ((ofs & 3) != REF_OBSOLETE)) | |
461 | && (jeb != c->nextblock || (ofs & ~3) != jeb->offset + (c->sector_size - jeb->free_size))) { | |
66bfaeaa DW |
462 | printk(KERN_WARNING "argh. node added in wrong place at 0x%08x(%d)\n", ofs & ~3, ofs & 3); |
463 | if (c->nextblock) | |
464 | printk(KERN_WARNING "nextblock 0x%08x", c->nextblock->offset); | |
465 | else | |
466 | printk(KERN_WARNING "No nextblock"); | |
467 | printk(", expected at %08x\n", jeb->offset + (c->sector_size - jeb->free_size)); | |
2f785402 | 468 | return ERR_PTR(-EINVAL); |
1da177e4 LT |
469 | } |
470 | #endif | |
471 | spin_lock(&c->erase_completion_lock); | |
472 | ||
2f785402 | 473 | new = jffs2_link_node_ref(c, jeb, ofs, len, ic); |
1da177e4 | 474 | |
9b88f473 | 475 | if (!jeb->free_size && !jeb->dirty_size && !ISDIRTY(jeb->wasted_size)) { |
1da177e4 | 476 | /* If it lives on the dirty_list, jffs2_reserve_space will put it there */ |
9c261b33 JP |
477 | jffs2_dbg(1, "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n", |
478 | jeb->offset, jeb->free_size, jeb->dirty_size, | |
479 | jeb->used_size); | |
1da177e4 LT |
480 | if (jffs2_wbuf_dirty(c)) { |
481 | /* Flush the last write in the block if it's outstanding */ | |
482 | spin_unlock(&c->erase_completion_lock); | |
483 | jffs2_flush_wbuf_pad(c); | |
484 | spin_lock(&c->erase_completion_lock); | |
485 | } | |
486 | ||
487 | list_add_tail(&jeb->list, &c->clean_list); | |
488 | c->nextblock = NULL; | |
489 | } | |
e0c8e42f AB |
490 | jffs2_dbg_acct_sanity_check_nolock(c,jeb); |
491 | jffs2_dbg_acct_paranoia_check_nolock(c, jeb); | |
1da177e4 LT |
492 | |
493 | spin_unlock(&c->erase_completion_lock); | |
494 | ||
2f785402 | 495 | return new; |
1da177e4 LT |
496 | } |
497 | ||
498 | ||
499 | void jffs2_complete_reservation(struct jffs2_sb_info *c) | |
500 | { | |
9c261b33 | 501 | jffs2_dbg(1, "jffs2_complete_reservation()\n"); |
acb64a43 | 502 | spin_lock(&c->erase_completion_lock); |
1da177e4 | 503 | jffs2_garbage_collect_trigger(c); |
acb64a43 | 504 | spin_unlock(&c->erase_completion_lock); |
ced22070 | 505 | mutex_unlock(&c->alloc_sem); |
1da177e4 LT |
506 | } |
507 | ||
508 | static inline int on_list(struct list_head *obj, struct list_head *head) | |
509 | { | |
510 | struct list_head *this; | |
511 | ||
512 | list_for_each(this, head) { | |
513 | if (this == obj) { | |
9c261b33 | 514 | jffs2_dbg(1, "%p is on list at %p\n", obj, head); |
1da177e4 LT |
515 | return 1; |
516 | ||
517 | } | |
518 | } | |
519 | return 0; | |
520 | } | |
521 | ||
522 | void jffs2_mark_node_obsolete(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref) | |
523 | { | |
524 | struct jffs2_eraseblock *jeb; | |
525 | int blocknr; | |
526 | struct jffs2_unknown_node n; | |
527 | int ret, addedsize; | |
528 | size_t retlen; | |
1417fc44 | 529 | uint32_t freed_len; |
1da177e4 | 530 | |
9bfeb691 | 531 | if(unlikely(!ref)) { |
1da177e4 LT |
532 | printk(KERN_NOTICE "EEEEEK. jffs2_mark_node_obsolete called with NULL node\n"); |
533 | return; | |
534 | } | |
535 | if (ref_obsolete(ref)) { | |
9c261b33 JP |
536 | jffs2_dbg(1, "%s(): called with already obsolete node at 0x%08x\n", |
537 | __func__, ref_offset(ref)); | |
1da177e4 LT |
538 | return; |
539 | } | |
540 | blocknr = ref->flash_offset / c->sector_size; | |
541 | if (blocknr >= c->nr_blocks) { | |
542 | printk(KERN_NOTICE "raw node at 0x%08x is off the end of device!\n", ref->flash_offset); | |
543 | BUG(); | |
544 | } | |
545 | jeb = &c->blocks[blocknr]; | |
546 | ||
547 | if (jffs2_can_mark_obsolete(c) && !jffs2_is_readonly(c) && | |
31fbdf7a | 548 | !(c->flags & (JFFS2_SB_FLAG_SCANNING | JFFS2_SB_FLAG_BUILDING))) { |
182ec4ee TG |
549 | /* Hm. This may confuse static lock analysis. If any of the above |
550 | three conditions is false, we're going to return from this | |
1da177e4 LT |
551 | function without actually obliterating any nodes or freeing |
552 | any jffs2_raw_node_refs. So we don't need to stop erases from | |
553 | happening, or protect against people holding an obsolete | |
554 | jffs2_raw_node_ref without the erase_completion_lock. */ | |
ced22070 | 555 | mutex_lock(&c->erase_free_sem); |
1da177e4 LT |
556 | } |
557 | ||
558 | spin_lock(&c->erase_completion_lock); | |
559 | ||
1417fc44 DW |
560 | freed_len = ref_totlen(c, jeb, ref); |
561 | ||
1da177e4 | 562 | if (ref_flags(ref) == REF_UNCHECKED) { |
1417fc44 | 563 | D1(if (unlikely(jeb->unchecked_size < freed_len)) { |
1da177e4 | 564 | 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", |
1417fc44 | 565 | freed_len, blocknr, ref->flash_offset, jeb->used_size); |
1da177e4 LT |
566 | BUG(); |
567 | }) | |
9c261b33 JP |
568 | jffs2_dbg(1, "Obsoleting previously unchecked node at 0x%08x of len %x\n", |
569 | ref_offset(ref), freed_len); | |
1417fc44 DW |
570 | jeb->unchecked_size -= freed_len; |
571 | c->unchecked_size -= freed_len; | |
1da177e4 | 572 | } else { |
1417fc44 | 573 | D1(if (unlikely(jeb->used_size < freed_len)) { |
1da177e4 | 574 | 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", |
1417fc44 | 575 | freed_len, blocknr, ref->flash_offset, jeb->used_size); |
1da177e4 LT |
576 | BUG(); |
577 | }) | |
9c261b33 JP |
578 | jffs2_dbg(1, "Obsoleting node at 0x%08x of len %#x: ", |
579 | ref_offset(ref), freed_len); | |
1417fc44 DW |
580 | jeb->used_size -= freed_len; |
581 | c->used_size -= freed_len; | |
1da177e4 LT |
582 | } |
583 | ||
584 | // Take care, that wasted size is taken into concern | |
1417fc44 | 585 | if ((jeb->dirty_size || ISDIRTY(jeb->wasted_size + freed_len)) && jeb != c->nextblock) { |
9c261b33 | 586 | jffs2_dbg(1, "Dirtying\n"); |
1417fc44 DW |
587 | addedsize = freed_len; |
588 | jeb->dirty_size += freed_len; | |
589 | c->dirty_size += freed_len; | |
1da177e4 LT |
590 | |
591 | /* Convert wasted space to dirty, if not a bad block */ | |
592 | if (jeb->wasted_size) { | |
593 | if (on_list(&jeb->list, &c->bad_used_list)) { | |
9c261b33 JP |
594 | jffs2_dbg(1, "Leaving block at %08x on the bad_used_list\n", |
595 | jeb->offset); | |
1da177e4 LT |
596 | addedsize = 0; /* To fool the refiling code later */ |
597 | } else { | |
9c261b33 JP |
598 | jffs2_dbg(1, "Converting %d bytes of wasted space to dirty in block at %08x\n", |
599 | jeb->wasted_size, jeb->offset); | |
1da177e4 LT |
600 | addedsize += jeb->wasted_size; |
601 | jeb->dirty_size += jeb->wasted_size; | |
602 | c->dirty_size += jeb->wasted_size; | |
603 | c->wasted_size -= jeb->wasted_size; | |
604 | jeb->wasted_size = 0; | |
605 | } | |
606 | } | |
607 | } else { | |
9c261b33 | 608 | jffs2_dbg(1, "Wasting\n"); |
1da177e4 | 609 | addedsize = 0; |
1417fc44 DW |
610 | jeb->wasted_size += freed_len; |
611 | c->wasted_size += freed_len; | |
1da177e4 LT |
612 | } |
613 | ref->flash_offset = ref_offset(ref) | REF_OBSOLETE; | |
182ec4ee | 614 | |
e0c8e42f AB |
615 | jffs2_dbg_acct_sanity_check_nolock(c, jeb); |
616 | jffs2_dbg_acct_paranoia_check_nolock(c, jeb); | |
1da177e4 | 617 | |
31fbdf7a AB |
618 | if (c->flags & JFFS2_SB_FLAG_SCANNING) { |
619 | /* Flash scanning is in progress. Don't muck about with the block | |
1da177e4 | 620 | lists because they're not ready yet, and don't actually |
182ec4ee | 621 | obliterate nodes that look obsolete. If they weren't |
1da177e4 LT |
622 | marked obsolete on the flash at the time they _became_ |
623 | obsolete, there was probably a reason for that. */ | |
624 | spin_unlock(&c->erase_completion_lock); | |
625 | /* We didn't lock the erase_free_sem */ | |
626 | return; | |
627 | } | |
628 | ||
629 | if (jeb == c->nextblock) { | |
9c261b33 JP |
630 | jffs2_dbg(2, "Not moving nextblock 0x%08x to dirty/erase_pending list\n", |
631 | jeb->offset); | |
1da177e4 LT |
632 | } else if (!jeb->used_size && !jeb->unchecked_size) { |
633 | if (jeb == c->gcblock) { | |
9c261b33 JP |
634 | jffs2_dbg(1, "gcblock at 0x%08x completely dirtied. Clearing gcblock...\n", |
635 | jeb->offset); | |
1da177e4 LT |
636 | c->gcblock = NULL; |
637 | } else { | |
9c261b33 JP |
638 | jffs2_dbg(1, "Eraseblock at 0x%08x completely dirtied. Removing from (dirty?) list...\n", |
639 | jeb->offset); | |
1da177e4 LT |
640 | list_del(&jeb->list); |
641 | } | |
642 | if (jffs2_wbuf_dirty(c)) { | |
9c261b33 | 643 | jffs2_dbg(1, "...and adding to erasable_pending_wbuf_list\n"); |
1da177e4 LT |
644 | list_add_tail(&jeb->list, &c->erasable_pending_wbuf_list); |
645 | } else { | |
646 | if (jiffies & 127) { | |
647 | /* Most of the time, we just erase it immediately. Otherwise we | |
648 | spend ages scanning it on mount, etc. */ | |
9c261b33 | 649 | jffs2_dbg(1, "...and adding to erase_pending_list\n"); |
1da177e4 LT |
650 | list_add_tail(&jeb->list, &c->erase_pending_list); |
651 | c->nr_erasing_blocks++; | |
ae3b6ba0 | 652 | jffs2_garbage_collect_trigger(c); |
1da177e4 LT |
653 | } else { |
654 | /* Sometimes, however, we leave it elsewhere so it doesn't get | |
655 | immediately reused, and we spread the load a bit. */ | |
9c261b33 | 656 | jffs2_dbg(1, "...and adding to erasable_list\n"); |
1da177e4 | 657 | list_add_tail(&jeb->list, &c->erasable_list); |
182ec4ee | 658 | } |
1da177e4 | 659 | } |
9c261b33 | 660 | jffs2_dbg(1, "Done OK\n"); |
1da177e4 | 661 | } else if (jeb == c->gcblock) { |
9c261b33 JP |
662 | jffs2_dbg(2, "Not moving gcblock 0x%08x to dirty_list\n", |
663 | jeb->offset); | |
1da177e4 | 664 | } else if (ISDIRTY(jeb->dirty_size) && !ISDIRTY(jeb->dirty_size - addedsize)) { |
9c261b33 JP |
665 | jffs2_dbg(1, "Eraseblock at 0x%08x is freshly dirtied. Removing from clean list...\n", |
666 | jeb->offset); | |
1da177e4 | 667 | list_del(&jeb->list); |
9c261b33 | 668 | jffs2_dbg(1, "...and adding to dirty_list\n"); |
1da177e4 LT |
669 | list_add_tail(&jeb->list, &c->dirty_list); |
670 | } else if (VERYDIRTY(c, jeb->dirty_size) && | |
671 | !VERYDIRTY(c, jeb->dirty_size - addedsize)) { | |
9c261b33 JP |
672 | jffs2_dbg(1, "Eraseblock at 0x%08x is now very dirty. Removing from dirty list...\n", |
673 | jeb->offset); | |
1da177e4 | 674 | list_del(&jeb->list); |
9c261b33 | 675 | jffs2_dbg(1, "...and adding to very_dirty_list\n"); |
1da177e4 LT |
676 | list_add_tail(&jeb->list, &c->very_dirty_list); |
677 | } else { | |
9c261b33 JP |
678 | jffs2_dbg(1, "Eraseblock at 0x%08x not moved anywhere. (free 0x%08x, dirty 0x%08x, used 0x%08x)\n", |
679 | jeb->offset, jeb->free_size, jeb->dirty_size, | |
680 | jeb->used_size); | |
182ec4ee | 681 | } |
1da177e4 LT |
682 | |
683 | spin_unlock(&c->erase_completion_lock); | |
684 | ||
31fbdf7a AB |
685 | if (!jffs2_can_mark_obsolete(c) || jffs2_is_readonly(c) || |
686 | (c->flags & JFFS2_SB_FLAG_BUILDING)) { | |
1da177e4 LT |
687 | /* We didn't lock the erase_free_sem */ |
688 | return; | |
689 | } | |
690 | ||
691 | /* The erase_free_sem is locked, and has been since before we marked the node obsolete | |
692 | and potentially put its eraseblock onto the erase_pending_list. Thus, we know that | |
693 | the block hasn't _already_ been erased, and that 'ref' itself hasn't been freed yet | |
c38c1b61 | 694 | by jffs2_free_jeb_node_refs() in erase.c. Which is nice. */ |
1da177e4 | 695 | |
9c261b33 JP |
696 | jffs2_dbg(1, "obliterating obsoleted node at 0x%08x\n", |
697 | ref_offset(ref)); | |
1da177e4 LT |
698 | ret = jffs2_flash_read(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n); |
699 | if (ret) { | |
700 | printk(KERN_WARNING "Read error reading from obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret); | |
701 | goto out_erase_sem; | |
702 | } | |
703 | if (retlen != sizeof(n)) { | |
704 | printk(KERN_WARNING "Short read from obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen); | |
705 | goto out_erase_sem; | |
706 | } | |
1417fc44 DW |
707 | if (PAD(je32_to_cpu(n.totlen)) != PAD(freed_len)) { |
708 | printk(KERN_WARNING "Node totlen on flash (0x%08x) != totlen from node ref (0x%08x)\n", je32_to_cpu(n.totlen), freed_len); | |
1da177e4 LT |
709 | goto out_erase_sem; |
710 | } | |
711 | if (!(je16_to_cpu(n.nodetype) & JFFS2_NODE_ACCURATE)) { | |
9c261b33 JP |
712 | jffs2_dbg(1, "Node at 0x%08x was already marked obsolete (nodetype 0x%04x)\n", |
713 | ref_offset(ref), je16_to_cpu(n.nodetype)); | |
1da177e4 LT |
714 | goto out_erase_sem; |
715 | } | |
716 | /* XXX FIXME: This is ugly now */ | |
717 | n.nodetype = cpu_to_je16(je16_to_cpu(n.nodetype) & ~JFFS2_NODE_ACCURATE); | |
718 | ret = jffs2_flash_write(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n); | |
719 | if (ret) { | |
720 | printk(KERN_WARNING "Write error in obliterating obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret); | |
721 | goto out_erase_sem; | |
722 | } | |
723 | if (retlen != sizeof(n)) { | |
724 | printk(KERN_WARNING "Short write in obliterating obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen); | |
725 | goto out_erase_sem; | |
726 | } | |
727 | ||
728 | /* Nodes which have been marked obsolete no longer need to be | |
729 | associated with any inode. Remove them from the per-inode list. | |
182ec4ee TG |
730 | |
731 | Note we can't do this for NAND at the moment because we need | |
1da177e4 LT |
732 | obsolete dirent nodes to stay on the lists, because of the |
733 | horridness in jffs2_garbage_collect_deletion_dirent(). Also | |
182ec4ee | 734 | because we delete the inocache, and on NAND we need that to |
1da177e4 LT |
735 | stay around until all the nodes are actually erased, in order |
736 | to stop us from giving the same inode number to another newly | |
737 | created inode. */ | |
738 | if (ref->next_in_ino) { | |
739 | struct jffs2_inode_cache *ic; | |
740 | struct jffs2_raw_node_ref **p; | |
741 | ||
742 | spin_lock(&c->erase_completion_lock); | |
743 | ||
744 | ic = jffs2_raw_ref_to_ic(ref); | |
745 | for (p = &ic->nodes; (*p) != ref; p = &((*p)->next_in_ino)) | |
746 | ; | |
747 | ||
748 | *p = ref->next_in_ino; | |
749 | ref->next_in_ino = NULL; | |
750 | ||
c9f700f8 KK |
751 | switch (ic->class) { |
752 | #ifdef CONFIG_JFFS2_FS_XATTR | |
753 | case RAWNODE_CLASS_XATTR_DATUM: | |
754 | jffs2_release_xattr_datum(c, (struct jffs2_xattr_datum *)ic); | |
755 | break; | |
756 | case RAWNODE_CLASS_XATTR_REF: | |
757 | jffs2_release_xattr_ref(c, (struct jffs2_xattr_ref *)ic); | |
758 | break; | |
759 | #endif | |
760 | default: | |
27c72b04 | 761 | if (ic->nodes == (void *)ic && ic->pino_nlink == 0) |
c9f700f8 KK |
762 | jffs2_del_ino_cache(c, ic); |
763 | break; | |
764 | } | |
1da177e4 LT |
765 | spin_unlock(&c->erase_completion_lock); |
766 | } | |
767 | ||
1da177e4 | 768 | out_erase_sem: |
ced22070 | 769 | mutex_unlock(&c->erase_free_sem); |
1da177e4 LT |
770 | } |
771 | ||
1da177e4 LT |
772 | int jffs2_thread_should_wake(struct jffs2_sb_info *c) |
773 | { | |
774 | int ret = 0; | |
775 | uint32_t dirty; | |
8fb870df DW |
776 | int nr_very_dirty = 0; |
777 | struct jffs2_eraseblock *jeb; | |
1da177e4 | 778 | |
d6ce1710 JT |
779 | if (!list_empty(&c->erase_complete_list) || |
780 | !list_empty(&c->erase_pending_list)) | |
781 | return 1; | |
782 | ||
1da177e4 | 783 | if (c->unchecked_size) { |
9c261b33 JP |
784 | jffs2_dbg(1, "jffs2_thread_should_wake(): unchecked_size %d, checked_ino #%d\n", |
785 | c->unchecked_size, c->checked_ino); | |
1da177e4 LT |
786 | return 1; |
787 | } | |
788 | ||
789 | /* dirty_size contains blocks on erase_pending_list | |
790 | * those blocks are counted in c->nr_erasing_blocks. | |
791 | * If one block is actually erased, it is not longer counted as dirty_space | |
792 | * but it is counted in c->nr_erasing_blocks, so we add it and subtract it | |
793 | * with c->nr_erasing_blocks * c->sector_size again. | |
794 | * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks | |
795 | * This helps us to force gc and pick eventually a clean block to spread the load. | |
796 | */ | |
797 | dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size; | |
798 | ||
182ec4ee TG |
799 | if (c->nr_free_blocks + c->nr_erasing_blocks < c->resv_blocks_gctrigger && |
800 | (dirty > c->nospc_dirty_size)) | |
1da177e4 LT |
801 | ret = 1; |
802 | ||
8fb870df DW |
803 | list_for_each_entry(jeb, &c->very_dirty_list, list) { |
804 | nr_very_dirty++; | |
805 | if (nr_very_dirty == c->vdirty_blocks_gctrigger) { | |
806 | ret = 1; | |
a8c68f32 DW |
807 | /* In debug mode, actually go through and count them all */ |
808 | D1(continue); | |
809 | break; | |
8fb870df DW |
810 | } |
811 | } | |
812 | ||
9c261b33 JP |
813 | jffs2_dbg(1, "%s(): nr_free_blocks %d, nr_erasing_blocks %d, dirty_size 0x%x, vdirty_blocks %d: %s\n", |
814 | __func__, c->nr_free_blocks, c->nr_erasing_blocks, | |
815 | c->dirty_size, nr_very_dirty, ret ? "yes" : "no"); | |
1da177e4 LT |
816 | |
817 | return ret; | |
818 | } |