]>
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 | ||
5a528957 JP |
12 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
13 | ||
1da177e4 | 14 | #include <linux/kernel.h> |
1da177e4 LT |
15 | #include <linux/mtd/mtd.h> |
16 | #include <linux/compiler.h> | |
17 | #include <linux/sched.h> /* For cond_resched() */ | |
18 | #include "nodelist.h" | |
e631ddba | 19 | #include "debug.h" |
1da177e4 LT |
20 | |
21 | /** | |
22 | * jffs2_reserve_space - request physical space to write nodes to flash | |
23 | * @c: superblock info | |
24 | * @minsize: Minimum acceptable size of allocation | |
1da177e4 LT |
25 | * @len: Returned value of allocation length |
26 | * @prio: Allocation type - ALLOC_{NORMAL,DELETION} | |
27 | * | |
28 | * Requests a block of physical space on the flash. Returns zero for success | |
9fe4854c DW |
29 | * and puts 'len' into the appropriate place, or returns -ENOSPC or other |
30 | * error if appropriate. Doesn't return len since that's | |
1da177e4 LT |
31 | * |
32 | * If it returns zero, jffs2_reserve_space() also downs the per-filesystem | |
33 | * allocation semaphore, to prevent more than one allocation from being | |
34 | * active at any time. The semaphore is later released by jffs2_commit_allocation() | |
35 | * | |
36 | * jffs2_reserve_space() may trigger garbage collection in order to make room | |
37 | * for the requested allocation. | |
38 | */ | |
39 | ||
e631ddba | 40 | static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize, |
9fe4854c | 41 | uint32_t *len, uint32_t sumsize); |
1da177e4 | 42 | |
9fe4854c | 43 | int jffs2_reserve_space(struct jffs2_sb_info *c, uint32_t minsize, |
e631ddba | 44 | uint32_t *len, int prio, uint32_t sumsize) |
1da177e4 LT |
45 | { |
46 | int ret = -EAGAIN; | |
47 | int blocksneeded = c->resv_blocks_write; | |
48 | /* align it */ | |
49 | minsize = PAD(minsize); | |
50 | ||
9c261b33 | 51 | jffs2_dbg(1, "%s(): Requested 0x%x bytes\n", __func__, minsize); |
ced22070 | 52 | mutex_lock(&c->alloc_sem); |
1da177e4 | 53 | |
9c261b33 | 54 | jffs2_dbg(1, "%s(): alloc sem got\n", __func__); |
1da177e4 LT |
55 | |
56 | spin_lock(&c->erase_completion_lock); | |
57 | ||
58 | /* this needs a little more thought (true <tglx> :)) */ | |
59 | while(ret == -EAGAIN) { | |
60 | while(c->nr_free_blocks + c->nr_erasing_blocks < blocksneeded) { | |
1da177e4 LT |
61 | uint32_t dirty, avail; |
62 | ||
63 | /* calculate real dirty size | |
64 | * dirty_size contains blocks on erase_pending_list | |
65 | * those blocks are counted in c->nr_erasing_blocks. | |
66 | * If one block is actually erased, it is not longer counted as dirty_space | |
67 | * but it is counted in c->nr_erasing_blocks, so we add it and subtract it | |
68 | * with c->nr_erasing_blocks * c->sector_size again. | |
69 | * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks | |
70 | * This helps us to force gc and pick eventually a clean block to spread the load. | |
71 | * We add unchecked_size here, as we hopefully will find some space to use. | |
72 | * This will affect the sum only once, as gc first finishes checking | |
73 | * of nodes. | |
74 | */ | |
75 | dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size + c->unchecked_size; | |
76 | if (dirty < c->nospc_dirty_size) { | |
77 | if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) { | |
9c261b33 JP |
78 | jffs2_dbg(1, "%s(): Low on dirty space to GC, but it's a deletion. Allowing...\n", |
79 | __func__); | |
1da177e4 LT |
80 | break; |
81 | } | |
9c261b33 JP |
82 | jffs2_dbg(1, "dirty size 0x%08x + unchecked_size 0x%08x < nospc_dirty_size 0x%08x, returning -ENOSPC\n", |
83 | dirty, c->unchecked_size, | |
84 | c->sector_size); | |
1da177e4 LT |
85 | |
86 | spin_unlock(&c->erase_completion_lock); | |
ced22070 | 87 | mutex_unlock(&c->alloc_sem); |
1da177e4 LT |
88 | return -ENOSPC; |
89 | } | |
182ec4ee | 90 | |
1da177e4 LT |
91 | /* 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. | |
182ec4ee | 95 | + Return -ENOSPC, if the maximum possibly available space is less or equal than |
1da177e4 LT |
96 | * 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) { | |
9c261b33 JP |
103 | jffs2_dbg(1, "%s(): Low on possibly available space, but it's a deletion. Allowing...\n", |
104 | __func__); | |
1da177e4 LT |
105 | break; |
106 | } | |
107 | ||
9c261b33 JP |
108 | jffs2_dbg(1, "max. available size 0x%08x < blocksneeded * sector_size 0x%08x, returning -ENOSPC\n", |
109 | avail, blocksneeded * c->sector_size); | |
1da177e4 | 110 | spin_unlock(&c->erase_completion_lock); |
ced22070 | 111 | mutex_unlock(&c->alloc_sem); |
1da177e4 LT |
112 | return -ENOSPC; |
113 | } | |
114 | ||
ced22070 | 115 | mutex_unlock(&c->alloc_sem); |
1da177e4 | 116 | |
9c261b33 JP |
117 | 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", |
118 | c->nr_free_blocks, c->nr_erasing_blocks, | |
119 | c->free_size, c->dirty_size, c->wasted_size, | |
120 | c->used_size, c->erasing_size, c->bad_size, | |
121 | c->free_size + c->dirty_size + | |
122 | c->wasted_size + c->used_size + | |
123 | c->erasing_size + c->bad_size, | |
124 | c->flash_size); | |
1da177e4 | 125 | spin_unlock(&c->erase_completion_lock); |
182ec4ee | 126 | |
1da177e4 | 127 | ret = jffs2_garbage_collect_pass(c); |
422b1202 | 128 | |
0717bf84 DW |
129 | if (ret == -EAGAIN) { |
130 | spin_lock(&c->erase_completion_lock); | |
131 | if (c->nr_erasing_blocks && | |
132 | list_empty(&c->erase_pending_list) && | |
133 | list_empty(&c->erase_complete_list)) { | |
134 | DECLARE_WAITQUEUE(wait, current); | |
135 | set_current_state(TASK_UNINTERRUPTIBLE); | |
136 | add_wait_queue(&c->erase_wait, &wait); | |
9c261b33 JP |
137 | jffs2_dbg(1, "%s waiting for erase to complete\n", |
138 | __func__); | |
0717bf84 DW |
139 | spin_unlock(&c->erase_completion_lock); |
140 | ||
141 | schedule(); | |
142 | } else | |
143 | spin_unlock(&c->erase_completion_lock); | |
144 | } else if (ret) | |
1da177e4 LT |
145 | return ret; |
146 | ||
147 | cond_resched(); | |
148 | ||
149 | if (signal_pending(current)) | |
150 | return -EINTR; | |
151 | ||
ced22070 | 152 | mutex_lock(&c->alloc_sem); |
1da177e4 LT |
153 | spin_lock(&c->erase_completion_lock); |
154 | } | |
155 | ||
9fe4854c | 156 | ret = jffs2_do_reserve_space(c, minsize, len, sumsize); |
1da177e4 | 157 | if (ret) { |
9c261b33 | 158 | jffs2_dbg(1, "%s(): ret is %d\n", __func__, ret); |
1da177e4 LT |
159 | } |
160 | } | |
161 | spin_unlock(&c->erase_completion_lock); | |
2f785402 | 162 | if (!ret) |
046b8b98 | 163 | ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1); |
1da177e4 | 164 | if (ret) |
ced22070 | 165 | mutex_unlock(&c->alloc_sem); |
1da177e4 LT |
166 | return ret; |
167 | } | |
168 | ||
9fe4854c DW |
169 | int jffs2_reserve_space_gc(struct jffs2_sb_info *c, uint32_t minsize, |
170 | uint32_t *len, uint32_t sumsize) | |
1da177e4 LT |
171 | { |
172 | int ret = -EAGAIN; | |
173 | minsize = PAD(minsize); | |
174 | ||
9c261b33 | 175 | jffs2_dbg(1, "%s(): Requested 0x%x bytes\n", __func__, minsize); |
1da177e4 LT |
176 | |
177 | spin_lock(&c->erase_completion_lock); | |
178 | while(ret == -EAGAIN) { | |
9fe4854c | 179 | ret = jffs2_do_reserve_space(c, minsize, len, sumsize); |
1da177e4 | 180 | if (ret) { |
9c261b33 JP |
181 | jffs2_dbg(1, "%s(): looping, ret is %d\n", |
182 | __func__, ret); | |
1da177e4 LT |
183 | } |
184 | } | |
185 | spin_unlock(&c->erase_completion_lock); | |
2f785402 | 186 | if (!ret) |
046b8b98 | 187 | ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1); |
2f785402 | 188 | |
1da177e4 LT |
189 | return ret; |
190 | } | |
191 | ||
e631ddba FH |
192 | |
193 | /* Classify nextblock (clean, dirty of verydirty) and force to select an other one */ | |
194 | ||
195 | static void jffs2_close_nextblock(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb) | |
1da177e4 | 196 | { |
e631ddba | 197 | |
99c2594f | 198 | if (c->nextblock == NULL) { |
9c261b33 JP |
199 | jffs2_dbg(1, "%s(): Erase block at 0x%08x has already been placed in a list\n", |
200 | __func__, jeb->offset); | |
99c2594f AH |
201 | return; |
202 | } | |
e631ddba FH |
203 | /* Check, if we have a dirty block now, or if it was dirty already */ |
204 | if (ISDIRTY (jeb->wasted_size + jeb->dirty_size)) { | |
205 | c->dirty_size += jeb->wasted_size; | |
206 | c->wasted_size -= jeb->wasted_size; | |
207 | jeb->dirty_size += jeb->wasted_size; | |
208 | jeb->wasted_size = 0; | |
209 | if (VERYDIRTY(c, jeb->dirty_size)) { | |
9c261b33 JP |
210 | jffs2_dbg(1, "Adding full erase block at 0x%08x to very_dirty_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n", |
211 | jeb->offset, jeb->free_size, jeb->dirty_size, | |
212 | jeb->used_size); | |
e631ddba FH |
213 | list_add_tail(&jeb->list, &c->very_dirty_list); |
214 | } else { | |
9c261b33 JP |
215 | jffs2_dbg(1, "Adding full erase block at 0x%08x to dirty_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n", |
216 | jeb->offset, jeb->free_size, jeb->dirty_size, | |
217 | jeb->used_size); | |
e631ddba FH |
218 | list_add_tail(&jeb->list, &c->dirty_list); |
219 | } | |
182ec4ee | 220 | } else { |
9c261b33 JP |
221 | jffs2_dbg(1, "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n", |
222 | jeb->offset, jeb->free_size, jeb->dirty_size, | |
223 | jeb->used_size); | |
e631ddba FH |
224 | list_add_tail(&jeb->list, &c->clean_list); |
225 | } | |
226 | c->nextblock = NULL; | |
227 | ||
228 | } | |
229 | ||
230 | /* Select a new jeb for nextblock */ | |
231 | ||
232 | static int jffs2_find_nextblock(struct jffs2_sb_info *c) | |
233 | { | |
234 | struct list_head *next; | |
182ec4ee | 235 | |
e631ddba FH |
236 | /* Take the next block off the 'free' list */ |
237 | ||
238 | if (list_empty(&c->free_list)) { | |
239 | ||
240 | if (!c->nr_erasing_blocks && | |
241 | !list_empty(&c->erasable_list)) { | |
242 | struct jffs2_eraseblock *ejeb; | |
243 | ||
244 | ejeb = list_entry(c->erasable_list.next, struct jffs2_eraseblock, list); | |
f116629d | 245 | list_move_tail(&ejeb->list, &c->erase_pending_list); |
e631ddba | 246 | c->nr_erasing_blocks++; |
ae3b6ba0 | 247 | jffs2_garbage_collect_trigger(c); |
9c261b33 JP |
248 | jffs2_dbg(1, "%s(): Triggering erase of erasable block at 0x%08x\n", |
249 | __func__, ejeb->offset); | |
e631ddba FH |
250 | } |
251 | ||
252 | if (!c->nr_erasing_blocks && | |
253 | !list_empty(&c->erasable_pending_wbuf_list)) { | |
9c261b33 JP |
254 | jffs2_dbg(1, "%s(): Flushing write buffer\n", |
255 | __func__); | |
e631ddba | 256 | /* c->nextblock is NULL, no update to c->nextblock allowed */ |
1da177e4 | 257 | spin_unlock(&c->erase_completion_lock); |
1da177e4 LT |
258 | jffs2_flush_wbuf_pad(c); |
259 | spin_lock(&c->erase_completion_lock); | |
e631ddba FH |
260 | /* Have another go. It'll be on the erasable_list now */ |
261 | return -EAGAIN; | |
1da177e4 | 262 | } |
e631ddba FH |
263 | |
264 | if (!c->nr_erasing_blocks) { | |
265 | /* Ouch. We're in GC, or we wouldn't have got here. | |
266 | And there's no space left. At all. */ | |
da320f05 JP |
267 | pr_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", |
268 | c->nr_erasing_blocks, c->nr_free_blocks, | |
269 | list_empty(&c->erasable_list) ? "yes" : "no", | |
270 | list_empty(&c->erasing_list) ? "yes" : "no", | |
271 | list_empty(&c->erase_pending_list) ? "yes" : "no"); | |
e631ddba | 272 | return -ENOSPC; |
1da177e4 | 273 | } |
e631ddba FH |
274 | |
275 | spin_unlock(&c->erase_completion_lock); | |
276 | /* Don't wait for it; just erase one right now */ | |
277 | jffs2_erase_pending_blocks(c, 1); | |
278 | spin_lock(&c->erase_completion_lock); | |
279 | ||
280 | /* An erase may have failed, decreasing the | |
281 | amount of free space available. So we must | |
282 | restart from the beginning */ | |
283 | return -EAGAIN; | |
1da177e4 | 284 | } |
e631ddba FH |
285 | |
286 | next = c->free_list.next; | |
287 | list_del(next); | |
288 | c->nextblock = list_entry(next, struct jffs2_eraseblock, list); | |
289 | c->nr_free_blocks--; | |
182ec4ee | 290 | |
e631ddba FH |
291 | jffs2_sum_reset_collected(c->summary); /* reset collected summary */ |
292 | ||
f04de505 | 293 | #ifdef CONFIG_JFFS2_FS_WRITEBUFFER |
5bf17237 AB |
294 | /* adjust write buffer offset, else we get a non contiguous write bug */ |
295 | if (!(c->wbuf_ofs % c->sector_size) && !c->wbuf_len) | |
296 | c->wbuf_ofs = 0xffffffff; | |
f04de505 | 297 | #endif |
5bf17237 | 298 | |
9c261b33 JP |
299 | jffs2_dbg(1, "%s(): new nextblock = 0x%08x\n", |
300 | __func__, c->nextblock->offset); | |
e631ddba FH |
301 | |
302 | return 0; | |
303 | } | |
304 | ||
305 | /* Called with alloc sem _and_ erase_completion_lock */ | |
9fe4854c DW |
306 | static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize, |
307 | uint32_t *len, uint32_t sumsize) | |
e631ddba FH |
308 | { |
309 | struct jffs2_eraseblock *jeb = c->nextblock; | |
9fe4854c | 310 | uint32_t reserved_size; /* for summary information at the end of the jeb */ |
e631ddba FH |
311 | int ret; |
312 | ||
313 | restart: | |
314 | reserved_size = 0; | |
315 | ||
316 | if (jffs2_sum_active() && (sumsize != JFFS2_SUMMARY_NOSUM_SIZE)) { | |
317 | /* NOSUM_SIZE means not to generate summary */ | |
318 | ||
319 | if (jeb) { | |
320 | reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE); | |
733802d9 | 321 | dbg_summary("minsize=%d , jeb->free=%d ," |
e631ddba FH |
322 | "summary->size=%d , sumsize=%d\n", |
323 | minsize, jeb->free_size, | |
324 | c->summary->sum_size, sumsize); | |
325 | } | |
326 | ||
327 | /* Is there enough space for writing out the current node, or we have to | |
328 | write out summary information now, close this jeb and select new nextblock? */ | |
329 | if (jeb && (PAD(minsize) + PAD(c->summary->sum_size + sumsize + | |
330 | JFFS2_SUMMARY_FRAME_SIZE) > jeb->free_size)) { | |
331 | ||
332 | /* Has summary been disabled for this jeb? */ | |
333 | if (jffs2_sum_is_disabled(c->summary)) { | |
334 | sumsize = JFFS2_SUMMARY_NOSUM_SIZE; | |
335 | goto restart; | |
1da177e4 LT |
336 | } |
337 | ||
e631ddba | 338 | /* Writing out the collected summary information */ |
733802d9 | 339 | dbg_summary("generating summary for 0x%08x.\n", jeb->offset); |
e631ddba FH |
340 | ret = jffs2_sum_write_sumnode(c); |
341 | ||
342 | if (ret) | |
343 | return ret; | |
344 | ||
345 | if (jffs2_sum_is_disabled(c->summary)) { | |
346 | /* jffs2_write_sumnode() couldn't write out the summary information | |
347 | diabling summary for this jeb and free the collected information | |
348 | */ | |
349 | sumsize = JFFS2_SUMMARY_NOSUM_SIZE; | |
350 | goto restart; | |
351 | } | |
352 | ||
353 | jffs2_close_nextblock(c, jeb); | |
354 | jeb = NULL; | |
34c0e906 FH |
355 | /* keep always valid value in reserved_size */ |
356 | reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE); | |
e631ddba FH |
357 | } |
358 | } else { | |
359 | if (jeb && minsize > jeb->free_size) { | |
fc6612f6 DW |
360 | uint32_t waste; |
361 | ||
e631ddba FH |
362 | /* Skip the end of this block and file it as having some dirty space */ |
363 | /* If there's a pending write to it, flush now */ | |
364 | ||
365 | if (jffs2_wbuf_dirty(c)) { | |
1da177e4 | 366 | spin_unlock(&c->erase_completion_lock); |
9c261b33 JP |
367 | jffs2_dbg(1, "%s(): Flushing write buffer\n", |
368 | __func__); | |
1da177e4 LT |
369 | jffs2_flush_wbuf_pad(c); |
370 | spin_lock(&c->erase_completion_lock); | |
e631ddba FH |
371 | jeb = c->nextblock; |
372 | goto restart; | |
1da177e4 LT |
373 | } |
374 | ||
fc6612f6 DW |
375 | spin_unlock(&c->erase_completion_lock); |
376 | ||
377 | ret = jffs2_prealloc_raw_node_refs(c, jeb, 1); | |
378 | if (ret) | |
379 | return ret; | |
380 | /* Just lock it again and continue. Nothing much can change because | |
381 | we hold c->alloc_sem anyway. In fact, it's not entirely clear why | |
382 | we hold c->erase_completion_lock in the majority of this function... | |
383 | but that's a question for another (more caffeine-rich) day. */ | |
384 | spin_lock(&c->erase_completion_lock); | |
385 | ||
386 | waste = jeb->free_size; | |
387 | jffs2_link_node_ref(c, jeb, | |
388 | (jeb->offset + c->sector_size - waste) | REF_OBSOLETE, | |
389 | waste, NULL); | |
390 | /* FIXME: that made it count as dirty. Convert to wasted */ | |
391 | jeb->dirty_size -= waste; | |
392 | c->dirty_size -= waste; | |
393 | jeb->wasted_size += waste; | |
394 | c->wasted_size += waste; | |
1da177e4 | 395 | |
e631ddba FH |
396 | jffs2_close_nextblock(c, jeb); |
397 | jeb = NULL; | |
1da177e4 | 398 | } |
e631ddba FH |
399 | } |
400 | ||
401 | if (!jeb) { | |
402 | ||
403 | ret = jffs2_find_nextblock(c); | |
404 | if (ret) | |
405 | return ret; | |
1da177e4 | 406 | |
e631ddba | 407 | jeb = c->nextblock; |
1da177e4 LT |
408 | |
409 | if (jeb->free_size != c->sector_size - c->cleanmarker_size) { | |
da320f05 JP |
410 | pr_warn("Eep. Block 0x%08x taken from free_list had free_size of 0x%08x!!\n", |
411 | jeb->offset, jeb->free_size); | |
1da177e4 LT |
412 | goto restart; |
413 | } | |
414 | } | |
415 | /* OK, jeb (==c->nextblock) is now pointing at a block which definitely has | |
416 | enough space */ | |
e631ddba | 417 | *len = jeb->free_size - reserved_size; |
1da177e4 LT |
418 | |
419 | if (c->cleanmarker_size && jeb->used_size == c->cleanmarker_size && | |
420 | !jeb->first_node->next_in_ino) { | |
182ec4ee | 421 | /* Only node in it beforehand was a CLEANMARKER node (we think). |
1da177e4 | 422 | So mark it obsolete now that there's going to be another node |
182ec4ee | 423 | in the block. This will reduce used_size to zero but We've |
1da177e4 LT |
424 | already set c->nextblock so that jffs2_mark_node_obsolete() |
425 | won't try to refile it to the dirty_list. | |
426 | */ | |
427 | spin_unlock(&c->erase_completion_lock); | |
428 | jffs2_mark_node_obsolete(c, jeb->first_node); | |
429 | spin_lock(&c->erase_completion_lock); | |
430 | } | |
431 | ||
9c261b33 JP |
432 | jffs2_dbg(1, "%s(): Giving 0x%x bytes at 0x%x\n", |
433 | __func__, | |
434 | *len, jeb->offset + (c->sector_size - jeb->free_size)); | |
1da177e4 LT |
435 | return 0; |
436 | } | |
437 | ||
438 | /** | |
439 | * jffs2_add_physical_node_ref - add a physical node reference to the list | |
440 | * @c: superblock info | |
441 | * @new: new node reference to add | |
442 | * @len: length of this physical node | |
1da177e4 | 443 | * |
182ec4ee | 444 | * Should only be used to report nodes for which space has been allocated |
1da177e4 LT |
445 | * by jffs2_reserve_space. |
446 | * | |
447 | * Must be called with the alloc_sem held. | |
448 | */ | |
182ec4ee | 449 | |
2f785402 DW |
450 | struct jffs2_raw_node_ref *jffs2_add_physical_node_ref(struct jffs2_sb_info *c, |
451 | uint32_t ofs, uint32_t len, | |
452 | struct jffs2_inode_cache *ic) | |
1da177e4 LT |
453 | { |
454 | struct jffs2_eraseblock *jeb; | |
2f785402 | 455 | struct jffs2_raw_node_ref *new; |
1da177e4 | 456 | |
2f785402 | 457 | jeb = &c->blocks[ofs / c->sector_size]; |
1da177e4 | 458 | |
9c261b33 JP |
459 | jffs2_dbg(1, "%s(): Node at 0x%x(%d), size 0x%x\n", |
460 | __func__, ofs & ~3, ofs & 3, len); | |
1da177e4 | 461 | #if 1 |
2f785402 DW |
462 | /* Allow non-obsolete nodes only to be added at the end of c->nextblock, |
463 | if c->nextblock is set. Note that wbuf.c will file obsolete nodes | |
464 | even after refiling c->nextblock */ | |
465 | if ((c->nextblock || ((ofs & 3) != REF_OBSOLETE)) | |
466 | && (jeb != c->nextblock || (ofs & ~3) != jeb->offset + (c->sector_size - jeb->free_size))) { | |
da320f05 JP |
467 | pr_warn("argh. node added in wrong place at 0x%08x(%d)\n", |
468 | ofs & ~3, ofs & 3); | |
66bfaeaa | 469 | if (c->nextblock) |
da320f05 | 470 | pr_warn("nextblock 0x%08x", c->nextblock->offset); |
66bfaeaa | 471 | else |
da320f05 JP |
472 | pr_warn("No nextblock"); |
473 | pr_cont(", expected at %08x\n", | |
474 | jeb->offset + (c->sector_size - jeb->free_size)); | |
2f785402 | 475 | return ERR_PTR(-EINVAL); |
1da177e4 LT |
476 | } |
477 | #endif | |
478 | spin_lock(&c->erase_completion_lock); | |
479 | ||
2f785402 | 480 | new = jffs2_link_node_ref(c, jeb, ofs, len, ic); |
1da177e4 | 481 | |
9b88f473 | 482 | if (!jeb->free_size && !jeb->dirty_size && !ISDIRTY(jeb->wasted_size)) { |
1da177e4 | 483 | /* If it lives on the dirty_list, jffs2_reserve_space will put it there */ |
9c261b33 JP |
484 | jffs2_dbg(1, "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n", |
485 | jeb->offset, jeb->free_size, jeb->dirty_size, | |
486 | jeb->used_size); | |
1da177e4 LT |
487 | if (jffs2_wbuf_dirty(c)) { |
488 | /* Flush the last write in the block if it's outstanding */ | |
489 | spin_unlock(&c->erase_completion_lock); | |
490 | jffs2_flush_wbuf_pad(c); | |
491 | spin_lock(&c->erase_completion_lock); | |
492 | } | |
493 | ||
494 | list_add_tail(&jeb->list, &c->clean_list); | |
495 | c->nextblock = NULL; | |
496 | } | |
e0c8e42f AB |
497 | jffs2_dbg_acct_sanity_check_nolock(c,jeb); |
498 | jffs2_dbg_acct_paranoia_check_nolock(c, jeb); | |
1da177e4 LT |
499 | |
500 | spin_unlock(&c->erase_completion_lock); | |
501 | ||
2f785402 | 502 | return new; |
1da177e4 LT |
503 | } |
504 | ||
505 | ||
506 | void jffs2_complete_reservation(struct jffs2_sb_info *c) | |
507 | { | |
9c261b33 | 508 | jffs2_dbg(1, "jffs2_complete_reservation()\n"); |
acb64a43 | 509 | spin_lock(&c->erase_completion_lock); |
1da177e4 | 510 | jffs2_garbage_collect_trigger(c); |
acb64a43 | 511 | spin_unlock(&c->erase_completion_lock); |
ced22070 | 512 | mutex_unlock(&c->alloc_sem); |
1da177e4 LT |
513 | } |
514 | ||
515 | static inline int on_list(struct list_head *obj, struct list_head *head) | |
516 | { | |
517 | struct list_head *this; | |
518 | ||
519 | list_for_each(this, head) { | |
520 | if (this == obj) { | |
9c261b33 | 521 | jffs2_dbg(1, "%p is on list at %p\n", obj, head); |
1da177e4 LT |
522 | return 1; |
523 | ||
524 | } | |
525 | } | |
526 | return 0; | |
527 | } | |
528 | ||
529 | void jffs2_mark_node_obsolete(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref) | |
530 | { | |
531 | struct jffs2_eraseblock *jeb; | |
532 | int blocknr; | |
533 | struct jffs2_unknown_node n; | |
534 | int ret, addedsize; | |
535 | size_t retlen; | |
1417fc44 | 536 | uint32_t freed_len; |
1da177e4 | 537 | |
9bfeb691 | 538 | if(unlikely(!ref)) { |
da320f05 | 539 | pr_notice("EEEEEK. jffs2_mark_node_obsolete called with NULL node\n"); |
1da177e4 LT |
540 | return; |
541 | } | |
542 | if (ref_obsolete(ref)) { | |
9c261b33 JP |
543 | jffs2_dbg(1, "%s(): called with already obsolete node at 0x%08x\n", |
544 | __func__, ref_offset(ref)); | |
1da177e4 LT |
545 | return; |
546 | } | |
547 | blocknr = ref->flash_offset / c->sector_size; | |
548 | if (blocknr >= c->nr_blocks) { | |
da320f05 JP |
549 | pr_notice("raw node at 0x%08x is off the end of device!\n", |
550 | ref->flash_offset); | |
1da177e4 LT |
551 | BUG(); |
552 | } | |
553 | jeb = &c->blocks[blocknr]; | |
554 | ||
555 | if (jffs2_can_mark_obsolete(c) && !jffs2_is_readonly(c) && | |
31fbdf7a | 556 | !(c->flags & (JFFS2_SB_FLAG_SCANNING | JFFS2_SB_FLAG_BUILDING))) { |
182ec4ee TG |
557 | /* Hm. This may confuse static lock analysis. If any of the above |
558 | three conditions is false, we're going to return from this | |
1da177e4 LT |
559 | function without actually obliterating any nodes or freeing |
560 | any jffs2_raw_node_refs. So we don't need to stop erases from | |
561 | happening, or protect against people holding an obsolete | |
562 | jffs2_raw_node_ref without the erase_completion_lock. */ | |
ced22070 | 563 | mutex_lock(&c->erase_free_sem); |
1da177e4 LT |
564 | } |
565 | ||
566 | spin_lock(&c->erase_completion_lock); | |
567 | ||
1417fc44 DW |
568 | freed_len = ref_totlen(c, jeb, ref); |
569 | ||
1da177e4 | 570 | if (ref_flags(ref) == REF_UNCHECKED) { |
1417fc44 | 571 | D1(if (unlikely(jeb->unchecked_size < freed_len)) { |
da320f05 JP |
572 | pr_notice("raw unchecked node of size 0x%08x freed from erase block %d at 0x%08x, but unchecked_size was already 0x%08x\n", |
573 | freed_len, blocknr, | |
574 | ref->flash_offset, jeb->used_size); | |
1da177e4 LT |
575 | BUG(); |
576 | }) | |
9c261b33 JP |
577 | jffs2_dbg(1, "Obsoleting previously unchecked node at 0x%08x of len %x\n", |
578 | ref_offset(ref), freed_len); | |
1417fc44 DW |
579 | jeb->unchecked_size -= freed_len; |
580 | c->unchecked_size -= freed_len; | |
1da177e4 | 581 | } else { |
1417fc44 | 582 | D1(if (unlikely(jeb->used_size < freed_len)) { |
da320f05 JP |
583 | pr_notice("raw node of size 0x%08x freed from erase block %d at 0x%08x, but used_size was already 0x%08x\n", |
584 | freed_len, blocknr, | |
585 | ref->flash_offset, jeb->used_size); | |
1da177e4 LT |
586 | BUG(); |
587 | }) | |
9c261b33 JP |
588 | jffs2_dbg(1, "Obsoleting node at 0x%08x of len %#x: ", |
589 | ref_offset(ref), freed_len); | |
1417fc44 DW |
590 | jeb->used_size -= freed_len; |
591 | c->used_size -= freed_len; | |
1da177e4 LT |
592 | } |
593 | ||
594 | // Take care, that wasted size is taken into concern | |
1417fc44 | 595 | if ((jeb->dirty_size || ISDIRTY(jeb->wasted_size + freed_len)) && jeb != c->nextblock) { |
9c261b33 | 596 | jffs2_dbg(1, "Dirtying\n"); |
1417fc44 DW |
597 | addedsize = freed_len; |
598 | jeb->dirty_size += freed_len; | |
599 | c->dirty_size += freed_len; | |
1da177e4 LT |
600 | |
601 | /* Convert wasted space to dirty, if not a bad block */ | |
602 | if (jeb->wasted_size) { | |
603 | if (on_list(&jeb->list, &c->bad_used_list)) { | |
9c261b33 JP |
604 | jffs2_dbg(1, "Leaving block at %08x on the bad_used_list\n", |
605 | jeb->offset); | |
1da177e4 LT |
606 | addedsize = 0; /* To fool the refiling code later */ |
607 | } else { | |
9c261b33 JP |
608 | jffs2_dbg(1, "Converting %d bytes of wasted space to dirty in block at %08x\n", |
609 | jeb->wasted_size, jeb->offset); | |
1da177e4 LT |
610 | addedsize += jeb->wasted_size; |
611 | jeb->dirty_size += jeb->wasted_size; | |
612 | c->dirty_size += jeb->wasted_size; | |
613 | c->wasted_size -= jeb->wasted_size; | |
614 | jeb->wasted_size = 0; | |
615 | } | |
616 | } | |
617 | } else { | |
9c261b33 | 618 | jffs2_dbg(1, "Wasting\n"); |
1da177e4 | 619 | addedsize = 0; |
1417fc44 DW |
620 | jeb->wasted_size += freed_len; |
621 | c->wasted_size += freed_len; | |
1da177e4 LT |
622 | } |
623 | ref->flash_offset = ref_offset(ref) | REF_OBSOLETE; | |
182ec4ee | 624 | |
e0c8e42f AB |
625 | jffs2_dbg_acct_sanity_check_nolock(c, jeb); |
626 | jffs2_dbg_acct_paranoia_check_nolock(c, jeb); | |
1da177e4 | 627 | |
31fbdf7a AB |
628 | if (c->flags & JFFS2_SB_FLAG_SCANNING) { |
629 | /* Flash scanning is in progress. Don't muck about with the block | |
1da177e4 | 630 | lists because they're not ready yet, and don't actually |
182ec4ee | 631 | obliterate nodes that look obsolete. If they weren't |
1da177e4 LT |
632 | marked obsolete on the flash at the time they _became_ |
633 | obsolete, there was probably a reason for that. */ | |
634 | spin_unlock(&c->erase_completion_lock); | |
635 | /* We didn't lock the erase_free_sem */ | |
636 | return; | |
637 | } | |
638 | ||
639 | if (jeb == c->nextblock) { | |
9c261b33 JP |
640 | jffs2_dbg(2, "Not moving nextblock 0x%08x to dirty/erase_pending list\n", |
641 | jeb->offset); | |
1da177e4 LT |
642 | } else if (!jeb->used_size && !jeb->unchecked_size) { |
643 | if (jeb == c->gcblock) { | |
9c261b33 JP |
644 | jffs2_dbg(1, "gcblock at 0x%08x completely dirtied. Clearing gcblock...\n", |
645 | jeb->offset); | |
1da177e4 LT |
646 | c->gcblock = NULL; |
647 | } else { | |
9c261b33 JP |
648 | jffs2_dbg(1, "Eraseblock at 0x%08x completely dirtied. Removing from (dirty?) list...\n", |
649 | jeb->offset); | |
1da177e4 LT |
650 | list_del(&jeb->list); |
651 | } | |
652 | if (jffs2_wbuf_dirty(c)) { | |
9c261b33 | 653 | jffs2_dbg(1, "...and adding to erasable_pending_wbuf_list\n"); |
1da177e4 LT |
654 | list_add_tail(&jeb->list, &c->erasable_pending_wbuf_list); |
655 | } else { | |
656 | if (jiffies & 127) { | |
657 | /* Most of the time, we just erase it immediately. Otherwise we | |
658 | spend ages scanning it on mount, etc. */ | |
9c261b33 | 659 | jffs2_dbg(1, "...and adding to erase_pending_list\n"); |
1da177e4 LT |
660 | list_add_tail(&jeb->list, &c->erase_pending_list); |
661 | c->nr_erasing_blocks++; | |
ae3b6ba0 | 662 | jffs2_garbage_collect_trigger(c); |
1da177e4 LT |
663 | } else { |
664 | /* Sometimes, however, we leave it elsewhere so it doesn't get | |
665 | immediately reused, and we spread the load a bit. */ | |
9c261b33 | 666 | jffs2_dbg(1, "...and adding to erasable_list\n"); |
1da177e4 | 667 | list_add_tail(&jeb->list, &c->erasable_list); |
182ec4ee | 668 | } |
1da177e4 | 669 | } |
9c261b33 | 670 | jffs2_dbg(1, "Done OK\n"); |
1da177e4 | 671 | } else if (jeb == c->gcblock) { |
9c261b33 JP |
672 | jffs2_dbg(2, "Not moving gcblock 0x%08x to dirty_list\n", |
673 | jeb->offset); | |
1da177e4 | 674 | } else if (ISDIRTY(jeb->dirty_size) && !ISDIRTY(jeb->dirty_size - addedsize)) { |
9c261b33 JP |
675 | jffs2_dbg(1, "Eraseblock at 0x%08x is freshly dirtied. Removing from clean list...\n", |
676 | jeb->offset); | |
1da177e4 | 677 | list_del(&jeb->list); |
9c261b33 | 678 | jffs2_dbg(1, "...and adding to dirty_list\n"); |
1da177e4 LT |
679 | list_add_tail(&jeb->list, &c->dirty_list); |
680 | } else if (VERYDIRTY(c, jeb->dirty_size) && | |
681 | !VERYDIRTY(c, jeb->dirty_size - addedsize)) { | |
9c261b33 JP |
682 | jffs2_dbg(1, "Eraseblock at 0x%08x is now very dirty. Removing from dirty list...\n", |
683 | jeb->offset); | |
1da177e4 | 684 | list_del(&jeb->list); |
9c261b33 | 685 | jffs2_dbg(1, "...and adding to very_dirty_list\n"); |
1da177e4 LT |
686 | list_add_tail(&jeb->list, &c->very_dirty_list); |
687 | } else { | |
9c261b33 JP |
688 | jffs2_dbg(1, "Eraseblock at 0x%08x not moved anywhere. (free 0x%08x, dirty 0x%08x, used 0x%08x)\n", |
689 | jeb->offset, jeb->free_size, jeb->dirty_size, | |
690 | jeb->used_size); | |
182ec4ee | 691 | } |
1da177e4 LT |
692 | |
693 | spin_unlock(&c->erase_completion_lock); | |
694 | ||
31fbdf7a AB |
695 | if (!jffs2_can_mark_obsolete(c) || jffs2_is_readonly(c) || |
696 | (c->flags & JFFS2_SB_FLAG_BUILDING)) { | |
1da177e4 LT |
697 | /* We didn't lock the erase_free_sem */ |
698 | return; | |
699 | } | |
700 | ||
701 | /* The erase_free_sem is locked, and has been since before we marked the node obsolete | |
702 | and potentially put its eraseblock onto the erase_pending_list. Thus, we know that | |
703 | the block hasn't _already_ been erased, and that 'ref' itself hasn't been freed yet | |
c38c1b61 | 704 | by jffs2_free_jeb_node_refs() in erase.c. Which is nice. */ |
1da177e4 | 705 | |
9c261b33 JP |
706 | jffs2_dbg(1, "obliterating obsoleted node at 0x%08x\n", |
707 | ref_offset(ref)); | |
1da177e4 LT |
708 | ret = jffs2_flash_read(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n); |
709 | if (ret) { | |
da320f05 JP |
710 | pr_warn("Read error reading from obsoleted node at 0x%08x: %d\n", |
711 | ref_offset(ref), ret); | |
1da177e4 LT |
712 | goto out_erase_sem; |
713 | } | |
714 | if (retlen != sizeof(n)) { | |
da320f05 JP |
715 | pr_warn("Short read from obsoleted node at 0x%08x: %zd\n", |
716 | ref_offset(ref), retlen); | |
1da177e4 LT |
717 | goto out_erase_sem; |
718 | } | |
1417fc44 | 719 | if (PAD(je32_to_cpu(n.totlen)) != PAD(freed_len)) { |
da320f05 JP |
720 | pr_warn("Node totlen on flash (0x%08x) != totlen from node ref (0x%08x)\n", |
721 | je32_to_cpu(n.totlen), freed_len); | |
1da177e4 LT |
722 | goto out_erase_sem; |
723 | } | |
724 | if (!(je16_to_cpu(n.nodetype) & JFFS2_NODE_ACCURATE)) { | |
9c261b33 JP |
725 | jffs2_dbg(1, "Node at 0x%08x was already marked obsolete (nodetype 0x%04x)\n", |
726 | ref_offset(ref), je16_to_cpu(n.nodetype)); | |
1da177e4 LT |
727 | goto out_erase_sem; |
728 | } | |
729 | /* XXX FIXME: This is ugly now */ | |
730 | n.nodetype = cpu_to_je16(je16_to_cpu(n.nodetype) & ~JFFS2_NODE_ACCURATE); | |
731 | ret = jffs2_flash_write(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n); | |
732 | if (ret) { | |
da320f05 JP |
733 | pr_warn("Write error in obliterating obsoleted node at 0x%08x: %d\n", |
734 | ref_offset(ref), ret); | |
1da177e4 LT |
735 | goto out_erase_sem; |
736 | } | |
737 | if (retlen != sizeof(n)) { | |
da320f05 JP |
738 | pr_warn("Short write in obliterating obsoleted node at 0x%08x: %zd\n", |
739 | ref_offset(ref), retlen); | |
1da177e4 LT |
740 | goto out_erase_sem; |
741 | } | |
742 | ||
743 | /* Nodes which have been marked obsolete no longer need to be | |
744 | associated with any inode. Remove them from the per-inode list. | |
182ec4ee TG |
745 | |
746 | Note we can't do this for NAND at the moment because we need | |
1da177e4 LT |
747 | obsolete dirent nodes to stay on the lists, because of the |
748 | horridness in jffs2_garbage_collect_deletion_dirent(). Also | |
182ec4ee | 749 | because we delete the inocache, and on NAND we need that to |
1da177e4 LT |
750 | stay around until all the nodes are actually erased, in order |
751 | to stop us from giving the same inode number to another newly | |
752 | created inode. */ | |
753 | if (ref->next_in_ino) { | |
754 | struct jffs2_inode_cache *ic; | |
755 | struct jffs2_raw_node_ref **p; | |
756 | ||
757 | spin_lock(&c->erase_completion_lock); | |
758 | ||
759 | ic = jffs2_raw_ref_to_ic(ref); | |
760 | for (p = &ic->nodes; (*p) != ref; p = &((*p)->next_in_ino)) | |
761 | ; | |
762 | ||
763 | *p = ref->next_in_ino; | |
764 | ref->next_in_ino = NULL; | |
765 | ||
c9f700f8 KK |
766 | switch (ic->class) { |
767 | #ifdef CONFIG_JFFS2_FS_XATTR | |
768 | case RAWNODE_CLASS_XATTR_DATUM: | |
769 | jffs2_release_xattr_datum(c, (struct jffs2_xattr_datum *)ic); | |
770 | break; | |
771 | case RAWNODE_CLASS_XATTR_REF: | |
772 | jffs2_release_xattr_ref(c, (struct jffs2_xattr_ref *)ic); | |
773 | break; | |
774 | #endif | |
775 | default: | |
27c72b04 | 776 | if (ic->nodes == (void *)ic && ic->pino_nlink == 0) |
c9f700f8 KK |
777 | jffs2_del_ino_cache(c, ic); |
778 | break; | |
779 | } | |
1da177e4 LT |
780 | spin_unlock(&c->erase_completion_lock); |
781 | } | |
782 | ||
1da177e4 | 783 | out_erase_sem: |
ced22070 | 784 | mutex_unlock(&c->erase_free_sem); |
1da177e4 LT |
785 | } |
786 | ||
1da177e4 LT |
787 | int jffs2_thread_should_wake(struct jffs2_sb_info *c) |
788 | { | |
789 | int ret = 0; | |
790 | uint32_t dirty; | |
8fb870df DW |
791 | int nr_very_dirty = 0; |
792 | struct jffs2_eraseblock *jeb; | |
1da177e4 | 793 | |
d6ce1710 JT |
794 | if (!list_empty(&c->erase_complete_list) || |
795 | !list_empty(&c->erase_pending_list)) | |
796 | return 1; | |
797 | ||
1da177e4 | 798 | if (c->unchecked_size) { |
9c261b33 JP |
799 | jffs2_dbg(1, "jffs2_thread_should_wake(): unchecked_size %d, checked_ino #%d\n", |
800 | c->unchecked_size, c->checked_ino); | |
1da177e4 LT |
801 | return 1; |
802 | } | |
803 | ||
804 | /* dirty_size contains blocks on erase_pending_list | |
805 | * those blocks are counted in c->nr_erasing_blocks. | |
806 | * If one block is actually erased, it is not longer counted as dirty_space | |
807 | * but it is counted in c->nr_erasing_blocks, so we add it and subtract it | |
808 | * with c->nr_erasing_blocks * c->sector_size again. | |
809 | * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks | |
810 | * This helps us to force gc and pick eventually a clean block to spread the load. | |
811 | */ | |
812 | dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size; | |
813 | ||
182ec4ee TG |
814 | if (c->nr_free_blocks + c->nr_erasing_blocks < c->resv_blocks_gctrigger && |
815 | (dirty > c->nospc_dirty_size)) | |
1da177e4 LT |
816 | ret = 1; |
817 | ||
8fb870df DW |
818 | list_for_each_entry(jeb, &c->very_dirty_list, list) { |
819 | nr_very_dirty++; | |
820 | if (nr_very_dirty == c->vdirty_blocks_gctrigger) { | |
821 | ret = 1; | |
a8c68f32 DW |
822 | /* In debug mode, actually go through and count them all */ |
823 | D1(continue); | |
824 | break; | |
8fb870df DW |
825 | } |
826 | } | |
827 | ||
9c261b33 JP |
828 | jffs2_dbg(1, "%s(): nr_free_blocks %d, nr_erasing_blocks %d, dirty_size 0x%x, vdirty_blocks %d: %s\n", |
829 | __func__, c->nr_free_blocks, c->nr_erasing_blocks, | |
830 | c->dirty_size, nr_very_dirty, ret ? "yes" : "no"); | |
1da177e4 LT |
831 | |
832 | return ret; | |
833 | } |