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> | |
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" | |
e631ddba | 18 | #include "debug.h" |
1da177e4 LT |
19 | |
20 | /** | |
21 | * jffs2_reserve_space - request physical space to write nodes to flash | |
22 | * @c: superblock info | |
23 | * @minsize: Minimum acceptable size of allocation | |
1da177e4 LT |
24 | * @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 | |
9fe4854c DW |
28 | * and puts 'len' into the appropriate place, or returns -ENOSPC or other |
29 | * error if appropriate. Doesn't return len since that's | |
1da177e4 LT |
30 | * |
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 | ||
e631ddba | 39 | static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize, |
9fe4854c | 40 | uint32_t *len, uint32_t sumsize); |
1da177e4 | 41 | |
9fe4854c | 42 | int jffs2_reserve_space(struct jffs2_sb_info *c, uint32_t minsize, |
e631ddba | 43 | uint32_t *len, int prio, uint32_t sumsize) |
1da177e4 LT |
44 | { |
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)); | |
ced22070 | 51 | mutex_lock(&c->alloc_sem); |
1da177e4 LT |
52 | |
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) { | |
1da177e4 LT |
60 | 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) { | |
4132ace8 | 77 | D1(printk(KERN_NOTICE "jffs2_reserve_space(): Low on dirty space to GC, but it's a deletion. Allowing...\n")); |
1da177e4 LT |
78 | 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); | |
ced22070 | 84 | mutex_unlock(&c->alloc_sem); |
1da177e4 LT |
85 | return -ENOSPC; |
86 | } | |
182ec4ee | 87 | |
1da177e4 LT |
88 | /* 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. | |
182ec4ee | 92 | + Return -ENOSPC, if the maximum possibly available space is less or equal than |
1da177e4 LT |
93 | * 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) { | |
4132ace8 | 100 | D1(printk(KERN_NOTICE "jffs2_reserve_space(): Low on possibly available space, but it's a deletion. Allowing...\n")); |
1da177e4 LT |
101 | 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); | |
ced22070 | 107 | mutex_unlock(&c->alloc_sem); |
1da177e4 LT |
108 | return -ENOSPC; |
109 | } | |
110 | ||
ced22070 | 111 | mutex_unlock(&c->alloc_sem); |
1da177e4 LT |
112 | |
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); | |
182ec4ee | 117 | |
1da177e4 LT |
118 | ret = jffs2_garbage_collect_pass(c); |
119 | if (ret) | |
120 | return ret; | |
121 | ||
122 | cond_resched(); | |
123 | ||
124 | if (signal_pending(current)) | |
125 | return -EINTR; | |
126 | ||
ced22070 | 127 | mutex_lock(&c->alloc_sem); |
1da177e4 LT |
128 | spin_lock(&c->erase_completion_lock); |
129 | } | |
130 | ||
9fe4854c | 131 | ret = jffs2_do_reserve_space(c, minsize, len, sumsize); |
1da177e4 LT |
132 | if (ret) { |
133 | D1(printk(KERN_DEBUG "jffs2_reserve_space: ret is %d\n", ret)); | |
134 | } | |
135 | } | |
136 | spin_unlock(&c->erase_completion_lock); | |
2f785402 | 137 | if (!ret) |
046b8b98 | 138 | ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1); |
1da177e4 | 139 | if (ret) |
ced22070 | 140 | mutex_unlock(&c->alloc_sem); |
1da177e4 LT |
141 | return ret; |
142 | } | |
143 | ||
9fe4854c DW |
144 | int jffs2_reserve_space_gc(struct jffs2_sb_info *c, uint32_t minsize, |
145 | uint32_t *len, uint32_t sumsize) | |
1da177e4 LT |
146 | { |
147 | int ret = -EAGAIN; | |
148 | minsize = PAD(minsize); | |
149 | ||
150 | D1(printk(KERN_DEBUG "jffs2_reserve_space_gc(): Requested 0x%x bytes\n", minsize)); | |
151 | ||
152 | spin_lock(&c->erase_completion_lock); | |
153 | while(ret == -EAGAIN) { | |
9fe4854c | 154 | ret = jffs2_do_reserve_space(c, minsize, len, sumsize); |
1da177e4 | 155 | if (ret) { |
ef53cb02 | 156 | D1(printk(KERN_DEBUG "jffs2_reserve_space_gc: looping, ret is %d\n", 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); |
2f785402 | 162 | |
1da177e4 LT |
163 | return ret; |
164 | } | |
165 | ||
e631ddba FH |
166 | |
167 | /* Classify nextblock (clean, dirty of verydirty) and force to select an other one */ | |
168 | ||
169 | static void jffs2_close_nextblock(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb) | |
1da177e4 | 170 | { |
e631ddba | 171 | |
99c2594f AH |
172 | if (c->nextblock == NULL) { |
173 | D1(printk(KERN_DEBUG "jffs2_close_nextblock: Erase block at 0x%08x has already been placed in a list\n", | |
174 | jeb->offset)); | |
175 | return; | |
176 | } | |
e631ddba FH |
177 | /* Check, if we have a dirty block now, or if it was dirty already */ |
178 | if (ISDIRTY (jeb->wasted_size + jeb->dirty_size)) { | |
179 | c->dirty_size += jeb->wasted_size; | |
180 | c->wasted_size -= jeb->wasted_size; | |
181 | jeb->dirty_size += jeb->wasted_size; | |
182 | jeb->wasted_size = 0; | |
183 | if (VERYDIRTY(c, jeb->dirty_size)) { | |
184 | 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", | |
185 | jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size)); | |
186 | list_add_tail(&jeb->list, &c->very_dirty_list); | |
187 | } else { | |
188 | D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to dirty_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n", | |
189 | jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size)); | |
190 | list_add_tail(&jeb->list, &c->dirty_list); | |
191 | } | |
182ec4ee | 192 | } else { |
e631ddba FH |
193 | D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n", |
194 | jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size)); | |
195 | list_add_tail(&jeb->list, &c->clean_list); | |
196 | } | |
197 | c->nextblock = NULL; | |
198 | ||
199 | } | |
200 | ||
201 | /* Select a new jeb for nextblock */ | |
202 | ||
203 | static int jffs2_find_nextblock(struct jffs2_sb_info *c) | |
204 | { | |
205 | struct list_head *next; | |
182ec4ee | 206 | |
e631ddba FH |
207 | /* Take the next block off the 'free' list */ |
208 | ||
209 | if (list_empty(&c->free_list)) { | |
210 | ||
211 | if (!c->nr_erasing_blocks && | |
212 | !list_empty(&c->erasable_list)) { | |
213 | struct jffs2_eraseblock *ejeb; | |
214 | ||
215 | ejeb = list_entry(c->erasable_list.next, struct jffs2_eraseblock, list); | |
f116629d | 216 | list_move_tail(&ejeb->list, &c->erase_pending_list); |
e631ddba FH |
217 | c->nr_erasing_blocks++; |
218 | jffs2_erase_pending_trigger(c); | |
219 | D1(printk(KERN_DEBUG "jffs2_find_nextblock: Triggering erase of erasable block at 0x%08x\n", | |
220 | ejeb->offset)); | |
221 | } | |
222 | ||
223 | if (!c->nr_erasing_blocks && | |
224 | !list_empty(&c->erasable_pending_wbuf_list)) { | |
225 | D1(printk(KERN_DEBUG "jffs2_find_nextblock: Flushing write buffer\n")); | |
226 | /* c->nextblock is NULL, no update to c->nextblock allowed */ | |
1da177e4 | 227 | spin_unlock(&c->erase_completion_lock); |
1da177e4 LT |
228 | jffs2_flush_wbuf_pad(c); |
229 | spin_lock(&c->erase_completion_lock); | |
e631ddba FH |
230 | /* Have another go. It'll be on the erasable_list now */ |
231 | return -EAGAIN; | |
1da177e4 | 232 | } |
e631ddba FH |
233 | |
234 | if (!c->nr_erasing_blocks) { | |
235 | /* Ouch. We're in GC, or we wouldn't have got here. | |
236 | And there's no space left. At all. */ | |
182ec4ee TG |
237 | 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", |
238 | c->nr_erasing_blocks, c->nr_free_blocks, list_empty(&c->erasable_list)?"yes":"no", | |
e631ddba FH |
239 | list_empty(&c->erasing_list)?"yes":"no", list_empty(&c->erase_pending_list)?"yes":"no"); |
240 | return -ENOSPC; | |
1da177e4 | 241 | } |
e631ddba FH |
242 | |
243 | spin_unlock(&c->erase_completion_lock); | |
244 | /* Don't wait for it; just erase one right now */ | |
245 | jffs2_erase_pending_blocks(c, 1); | |
246 | spin_lock(&c->erase_completion_lock); | |
247 | ||
248 | /* An erase may have failed, decreasing the | |
249 | amount of free space available. So we must | |
250 | restart from the beginning */ | |
251 | return -EAGAIN; | |
1da177e4 | 252 | } |
e631ddba FH |
253 | |
254 | next = c->free_list.next; | |
255 | list_del(next); | |
256 | c->nextblock = list_entry(next, struct jffs2_eraseblock, list); | |
257 | c->nr_free_blocks--; | |
182ec4ee | 258 | |
e631ddba FH |
259 | jffs2_sum_reset_collected(c->summary); /* reset collected summary */ |
260 | ||
261 | D1(printk(KERN_DEBUG "jffs2_find_nextblock(): new nextblock = 0x%08x\n", c->nextblock->offset)); | |
262 | ||
263 | return 0; | |
264 | } | |
265 | ||
266 | /* Called with alloc sem _and_ erase_completion_lock */ | |
9fe4854c DW |
267 | static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize, |
268 | uint32_t *len, uint32_t sumsize) | |
e631ddba FH |
269 | { |
270 | struct jffs2_eraseblock *jeb = c->nextblock; | |
9fe4854c | 271 | uint32_t reserved_size; /* for summary information at the end of the jeb */ |
e631ddba FH |
272 | int ret; |
273 | ||
274 | restart: | |
275 | reserved_size = 0; | |
276 | ||
277 | if (jffs2_sum_active() && (sumsize != JFFS2_SUMMARY_NOSUM_SIZE)) { | |
278 | /* NOSUM_SIZE means not to generate summary */ | |
279 | ||
280 | if (jeb) { | |
281 | reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE); | |
733802d9 | 282 | dbg_summary("minsize=%d , jeb->free=%d ," |
e631ddba FH |
283 | "summary->size=%d , sumsize=%d\n", |
284 | minsize, jeb->free_size, | |
285 | c->summary->sum_size, sumsize); | |
286 | } | |
287 | ||
288 | /* Is there enough space for writing out the current node, or we have to | |
289 | write out summary information now, close this jeb and select new nextblock? */ | |
290 | if (jeb && (PAD(minsize) + PAD(c->summary->sum_size + sumsize + | |
291 | JFFS2_SUMMARY_FRAME_SIZE) > jeb->free_size)) { | |
292 | ||
293 | /* Has summary been disabled for this jeb? */ | |
294 | if (jffs2_sum_is_disabled(c->summary)) { | |
295 | sumsize = JFFS2_SUMMARY_NOSUM_SIZE; | |
296 | goto restart; | |
1da177e4 LT |
297 | } |
298 | ||
e631ddba | 299 | /* Writing out the collected summary information */ |
733802d9 | 300 | dbg_summary("generating summary for 0x%08x.\n", jeb->offset); |
e631ddba FH |
301 | ret = jffs2_sum_write_sumnode(c); |
302 | ||
303 | if (ret) | |
304 | return ret; | |
305 | ||
306 | if (jffs2_sum_is_disabled(c->summary)) { | |
307 | /* jffs2_write_sumnode() couldn't write out the summary information | |
308 | diabling summary for this jeb and free the collected information | |
309 | */ | |
310 | sumsize = JFFS2_SUMMARY_NOSUM_SIZE; | |
311 | goto restart; | |
312 | } | |
313 | ||
314 | jffs2_close_nextblock(c, jeb); | |
315 | jeb = NULL; | |
34c0e906 FH |
316 | /* keep always valid value in reserved_size */ |
317 | reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE); | |
e631ddba FH |
318 | } |
319 | } else { | |
320 | if (jeb && minsize > jeb->free_size) { | |
fc6612f6 DW |
321 | uint32_t waste; |
322 | ||
e631ddba FH |
323 | /* Skip the end of this block and file it as having some dirty space */ |
324 | /* If there's a pending write to it, flush now */ | |
325 | ||
326 | if (jffs2_wbuf_dirty(c)) { | |
1da177e4 | 327 | spin_unlock(&c->erase_completion_lock); |
e631ddba | 328 | D1(printk(KERN_DEBUG "jffs2_do_reserve_space: Flushing write buffer\n")); |
1da177e4 LT |
329 | jffs2_flush_wbuf_pad(c); |
330 | spin_lock(&c->erase_completion_lock); | |
e631ddba FH |
331 | jeb = c->nextblock; |
332 | goto restart; | |
1da177e4 LT |
333 | } |
334 | ||
fc6612f6 DW |
335 | spin_unlock(&c->erase_completion_lock); |
336 | ||
337 | ret = jffs2_prealloc_raw_node_refs(c, jeb, 1); | |
338 | if (ret) | |
339 | return ret; | |
340 | /* Just lock it again and continue. Nothing much can change because | |
341 | we hold c->alloc_sem anyway. In fact, it's not entirely clear why | |
342 | we hold c->erase_completion_lock in the majority of this function... | |
343 | but that's a question for another (more caffeine-rich) day. */ | |
344 | spin_lock(&c->erase_completion_lock); | |
345 | ||
346 | waste = jeb->free_size; | |
347 | jffs2_link_node_ref(c, jeb, | |
348 | (jeb->offset + c->sector_size - waste) | REF_OBSOLETE, | |
349 | waste, NULL); | |
350 | /* FIXME: that made it count as dirty. Convert to wasted */ | |
351 | jeb->dirty_size -= waste; | |
352 | c->dirty_size -= waste; | |
353 | jeb->wasted_size += waste; | |
354 | c->wasted_size += waste; | |
1da177e4 | 355 | |
e631ddba FH |
356 | jffs2_close_nextblock(c, jeb); |
357 | jeb = NULL; | |
1da177e4 | 358 | } |
e631ddba FH |
359 | } |
360 | ||
361 | if (!jeb) { | |
362 | ||
363 | ret = jffs2_find_nextblock(c); | |
364 | if (ret) | |
365 | return ret; | |
1da177e4 | 366 | |
e631ddba | 367 | jeb = c->nextblock; |
1da177e4 LT |
368 | |
369 | if (jeb->free_size != c->sector_size - c->cleanmarker_size) { | |
370 | printk(KERN_WARNING "Eep. Block 0x%08x taken from free_list had free_size of 0x%08x!!\n", jeb->offset, jeb->free_size); | |
371 | goto restart; | |
372 | } | |
373 | } | |
374 | /* OK, jeb (==c->nextblock) is now pointing at a block which definitely has | |
375 | enough space */ | |
e631ddba | 376 | *len = jeb->free_size - reserved_size; |
1da177e4 LT |
377 | |
378 | if (c->cleanmarker_size && jeb->used_size == c->cleanmarker_size && | |
379 | !jeb->first_node->next_in_ino) { | |
182ec4ee | 380 | /* Only node in it beforehand was a CLEANMARKER node (we think). |
1da177e4 | 381 | So mark it obsolete now that there's going to be another node |
182ec4ee | 382 | in the block. This will reduce used_size to zero but We've |
1da177e4 LT |
383 | already set c->nextblock so that jffs2_mark_node_obsolete() |
384 | won't try to refile it to the dirty_list. | |
385 | */ | |
386 | spin_unlock(&c->erase_completion_lock); | |
387 | jffs2_mark_node_obsolete(c, jeb->first_node); | |
388 | spin_lock(&c->erase_completion_lock); | |
389 | } | |
390 | ||
9fe4854c DW |
391 | D1(printk(KERN_DEBUG "jffs2_do_reserve_space(): Giving 0x%x bytes at 0x%x\n", |
392 | *len, jeb->offset + (c->sector_size - jeb->free_size))); | |
1da177e4 LT |
393 | return 0; |
394 | } | |
395 | ||
396 | /** | |
397 | * jffs2_add_physical_node_ref - add a physical node reference to the list | |
398 | * @c: superblock info | |
399 | * @new: new node reference to add | |
400 | * @len: length of this physical node | |
1da177e4 | 401 | * |
182ec4ee | 402 | * Should only be used to report nodes for which space has been allocated |
1da177e4 LT |
403 | * by jffs2_reserve_space. |
404 | * | |
405 | * Must be called with the alloc_sem held. | |
406 | */ | |
182ec4ee | 407 | |
2f785402 DW |
408 | struct jffs2_raw_node_ref *jffs2_add_physical_node_ref(struct jffs2_sb_info *c, |
409 | uint32_t ofs, uint32_t len, | |
410 | struct jffs2_inode_cache *ic) | |
1da177e4 LT |
411 | { |
412 | struct jffs2_eraseblock *jeb; | |
2f785402 | 413 | struct jffs2_raw_node_ref *new; |
1da177e4 | 414 | |
2f785402 | 415 | jeb = &c->blocks[ofs / c->sector_size]; |
1da177e4 | 416 | |
2f785402 DW |
417 | D1(printk(KERN_DEBUG "jffs2_add_physical_node_ref(): Node at 0x%x(%d), size 0x%x\n", |
418 | ofs & ~3, ofs & 3, len)); | |
1da177e4 | 419 | #if 1 |
2f785402 DW |
420 | /* Allow non-obsolete nodes only to be added at the end of c->nextblock, |
421 | if c->nextblock is set. Note that wbuf.c will file obsolete nodes | |
422 | even after refiling c->nextblock */ | |
423 | if ((c->nextblock || ((ofs & 3) != REF_OBSOLETE)) | |
424 | && (jeb != c->nextblock || (ofs & ~3) != jeb->offset + (c->sector_size - jeb->free_size))) { | |
66bfaeaa DW |
425 | printk(KERN_WARNING "argh. node added in wrong place at 0x%08x(%d)\n", ofs & ~3, ofs & 3); |
426 | if (c->nextblock) | |
427 | printk(KERN_WARNING "nextblock 0x%08x", c->nextblock->offset); | |
428 | else | |
429 | printk(KERN_WARNING "No nextblock"); | |
430 | printk(", expected at %08x\n", jeb->offset + (c->sector_size - jeb->free_size)); | |
2f785402 | 431 | return ERR_PTR(-EINVAL); |
1da177e4 LT |
432 | } |
433 | #endif | |
434 | spin_lock(&c->erase_completion_lock); | |
435 | ||
2f785402 | 436 | new = jffs2_link_node_ref(c, jeb, ofs, len, ic); |
1da177e4 | 437 | |
9b88f473 | 438 | if (!jeb->free_size && !jeb->dirty_size && !ISDIRTY(jeb->wasted_size)) { |
1da177e4 LT |
439 | /* If it lives on the dirty_list, jffs2_reserve_space will put it there */ |
440 | D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n", | |
441 | jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size)); | |
442 | if (jffs2_wbuf_dirty(c)) { | |
443 | /* Flush the last write in the block if it's outstanding */ | |
444 | spin_unlock(&c->erase_completion_lock); | |
445 | jffs2_flush_wbuf_pad(c); | |
446 | spin_lock(&c->erase_completion_lock); | |
447 | } | |
448 | ||
449 | list_add_tail(&jeb->list, &c->clean_list); | |
450 | c->nextblock = NULL; | |
451 | } | |
e0c8e42f AB |
452 | jffs2_dbg_acct_sanity_check_nolock(c,jeb); |
453 | jffs2_dbg_acct_paranoia_check_nolock(c, jeb); | |
1da177e4 LT |
454 | |
455 | spin_unlock(&c->erase_completion_lock); | |
456 | ||
2f785402 | 457 | return new; |
1da177e4 LT |
458 | } |
459 | ||
460 | ||
461 | void jffs2_complete_reservation(struct jffs2_sb_info *c) | |
462 | { | |
463 | D1(printk(KERN_DEBUG "jffs2_complete_reservation()\n")); | |
464 | jffs2_garbage_collect_trigger(c); | |
ced22070 | 465 | mutex_unlock(&c->alloc_sem); |
1da177e4 LT |
466 | } |
467 | ||
468 | static inline int on_list(struct list_head *obj, struct list_head *head) | |
469 | { | |
470 | struct list_head *this; | |
471 | ||
472 | list_for_each(this, head) { | |
473 | if (this == obj) { | |
474 | D1(printk("%p is on list at %p\n", obj, head)); | |
475 | return 1; | |
476 | ||
477 | } | |
478 | } | |
479 | return 0; | |
480 | } | |
481 | ||
482 | void jffs2_mark_node_obsolete(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref) | |
483 | { | |
484 | struct jffs2_eraseblock *jeb; | |
485 | int blocknr; | |
486 | struct jffs2_unknown_node n; | |
487 | int ret, addedsize; | |
488 | size_t retlen; | |
1417fc44 | 489 | uint32_t freed_len; |
1da177e4 | 490 | |
9bfeb691 | 491 | if(unlikely(!ref)) { |
1da177e4 LT |
492 | printk(KERN_NOTICE "EEEEEK. jffs2_mark_node_obsolete called with NULL node\n"); |
493 | return; | |
494 | } | |
495 | if (ref_obsolete(ref)) { | |
496 | D1(printk(KERN_DEBUG "jffs2_mark_node_obsolete called with already obsolete node at 0x%08x\n", ref_offset(ref))); | |
497 | return; | |
498 | } | |
499 | blocknr = ref->flash_offset / c->sector_size; | |
500 | if (blocknr >= c->nr_blocks) { | |
501 | printk(KERN_NOTICE "raw node at 0x%08x is off the end of device!\n", ref->flash_offset); | |
502 | BUG(); | |
503 | } | |
504 | jeb = &c->blocks[blocknr]; | |
505 | ||
506 | if (jffs2_can_mark_obsolete(c) && !jffs2_is_readonly(c) && | |
31fbdf7a | 507 | !(c->flags & (JFFS2_SB_FLAG_SCANNING | JFFS2_SB_FLAG_BUILDING))) { |
182ec4ee TG |
508 | /* Hm. This may confuse static lock analysis. If any of the above |
509 | three conditions is false, we're going to return from this | |
1da177e4 LT |
510 | function without actually obliterating any nodes or freeing |
511 | any jffs2_raw_node_refs. So we don't need to stop erases from | |
512 | happening, or protect against people holding an obsolete | |
513 | jffs2_raw_node_ref without the erase_completion_lock. */ | |
ced22070 | 514 | mutex_lock(&c->erase_free_sem); |
1da177e4 LT |
515 | } |
516 | ||
517 | spin_lock(&c->erase_completion_lock); | |
518 | ||
1417fc44 DW |
519 | freed_len = ref_totlen(c, jeb, ref); |
520 | ||
1da177e4 | 521 | if (ref_flags(ref) == REF_UNCHECKED) { |
1417fc44 | 522 | D1(if (unlikely(jeb->unchecked_size < freed_len)) { |
1da177e4 | 523 | 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 | 524 | freed_len, blocknr, ref->flash_offset, jeb->used_size); |
1da177e4 LT |
525 | BUG(); |
526 | }) | |
1417fc44 DW |
527 | D1(printk(KERN_DEBUG "Obsoleting previously unchecked node at 0x%08x of len %x: ", ref_offset(ref), freed_len)); |
528 | jeb->unchecked_size -= freed_len; | |
529 | c->unchecked_size -= freed_len; | |
1da177e4 | 530 | } else { |
1417fc44 | 531 | D1(if (unlikely(jeb->used_size < freed_len)) { |
1da177e4 | 532 | 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 | 533 | freed_len, blocknr, ref->flash_offset, jeb->used_size); |
1da177e4 LT |
534 | BUG(); |
535 | }) | |
1417fc44 DW |
536 | D1(printk(KERN_DEBUG "Obsoleting node at 0x%08x of len %#x: ", ref_offset(ref), freed_len)); |
537 | jeb->used_size -= freed_len; | |
538 | c->used_size -= freed_len; | |
1da177e4 LT |
539 | } |
540 | ||
541 | // Take care, that wasted size is taken into concern | |
1417fc44 | 542 | if ((jeb->dirty_size || ISDIRTY(jeb->wasted_size + freed_len)) && jeb != c->nextblock) { |
c7c16c8e | 543 | D1(printk("Dirtying\n")); |
1417fc44 DW |
544 | addedsize = freed_len; |
545 | jeb->dirty_size += freed_len; | |
546 | c->dirty_size += freed_len; | |
1da177e4 LT |
547 | |
548 | /* Convert wasted space to dirty, if not a bad block */ | |
549 | if (jeb->wasted_size) { | |
550 | if (on_list(&jeb->list, &c->bad_used_list)) { | |
551 | D1(printk(KERN_DEBUG "Leaving block at %08x on the bad_used_list\n", | |
552 | jeb->offset)); | |
553 | addedsize = 0; /* To fool the refiling code later */ | |
554 | } else { | |
555 | D1(printk(KERN_DEBUG "Converting %d bytes of wasted space to dirty in block at %08x\n", | |
556 | jeb->wasted_size, jeb->offset)); | |
557 | addedsize += jeb->wasted_size; | |
558 | jeb->dirty_size += jeb->wasted_size; | |
559 | c->dirty_size += jeb->wasted_size; | |
560 | c->wasted_size -= jeb->wasted_size; | |
561 | jeb->wasted_size = 0; | |
562 | } | |
563 | } | |
564 | } else { | |
c7c16c8e | 565 | D1(printk("Wasting\n")); |
1da177e4 | 566 | addedsize = 0; |
1417fc44 DW |
567 | jeb->wasted_size += freed_len; |
568 | c->wasted_size += freed_len; | |
1da177e4 LT |
569 | } |
570 | ref->flash_offset = ref_offset(ref) | REF_OBSOLETE; | |
182ec4ee | 571 | |
e0c8e42f AB |
572 | jffs2_dbg_acct_sanity_check_nolock(c, jeb); |
573 | jffs2_dbg_acct_paranoia_check_nolock(c, jeb); | |
1da177e4 | 574 | |
31fbdf7a AB |
575 | if (c->flags & JFFS2_SB_FLAG_SCANNING) { |
576 | /* Flash scanning is in progress. Don't muck about with the block | |
1da177e4 | 577 | lists because they're not ready yet, and don't actually |
182ec4ee | 578 | obliterate nodes that look obsolete. If they weren't |
1da177e4 LT |
579 | marked obsolete on the flash at the time they _became_ |
580 | obsolete, there was probably a reason for that. */ | |
581 | spin_unlock(&c->erase_completion_lock); | |
582 | /* We didn't lock the erase_free_sem */ | |
583 | return; | |
584 | } | |
585 | ||
586 | if (jeb == c->nextblock) { | |
587 | D2(printk(KERN_DEBUG "Not moving nextblock 0x%08x to dirty/erase_pending list\n", jeb->offset)); | |
588 | } else if (!jeb->used_size && !jeb->unchecked_size) { | |
589 | if (jeb == c->gcblock) { | |
590 | D1(printk(KERN_DEBUG "gcblock at 0x%08x completely dirtied. Clearing gcblock...\n", jeb->offset)); | |
591 | c->gcblock = NULL; | |
592 | } else { | |
593 | D1(printk(KERN_DEBUG "Eraseblock at 0x%08x completely dirtied. Removing from (dirty?) list...\n", jeb->offset)); | |
594 | list_del(&jeb->list); | |
595 | } | |
596 | if (jffs2_wbuf_dirty(c)) { | |
597 | D1(printk(KERN_DEBUG "...and adding to erasable_pending_wbuf_list\n")); | |
598 | list_add_tail(&jeb->list, &c->erasable_pending_wbuf_list); | |
599 | } else { | |
600 | if (jiffies & 127) { | |
601 | /* Most of the time, we just erase it immediately. Otherwise we | |
602 | spend ages scanning it on mount, etc. */ | |
603 | D1(printk(KERN_DEBUG "...and adding to erase_pending_list\n")); | |
604 | list_add_tail(&jeb->list, &c->erase_pending_list); | |
605 | c->nr_erasing_blocks++; | |
606 | jffs2_erase_pending_trigger(c); | |
607 | } else { | |
608 | /* Sometimes, however, we leave it elsewhere so it doesn't get | |
609 | immediately reused, and we spread the load a bit. */ | |
610 | D1(printk(KERN_DEBUG "...and adding to erasable_list\n")); | |
611 | list_add_tail(&jeb->list, &c->erasable_list); | |
182ec4ee | 612 | } |
1da177e4 LT |
613 | } |
614 | D1(printk(KERN_DEBUG "Done OK\n")); | |
615 | } else if (jeb == c->gcblock) { | |
616 | D2(printk(KERN_DEBUG "Not moving gcblock 0x%08x to dirty_list\n", jeb->offset)); | |
617 | } else if (ISDIRTY(jeb->dirty_size) && !ISDIRTY(jeb->dirty_size - addedsize)) { | |
618 | D1(printk(KERN_DEBUG "Eraseblock at 0x%08x is freshly dirtied. Removing from clean list...\n", jeb->offset)); | |
619 | list_del(&jeb->list); | |
620 | D1(printk(KERN_DEBUG "...and adding to dirty_list\n")); | |
621 | list_add_tail(&jeb->list, &c->dirty_list); | |
622 | } else if (VERYDIRTY(c, jeb->dirty_size) && | |
623 | !VERYDIRTY(c, jeb->dirty_size - addedsize)) { | |
624 | D1(printk(KERN_DEBUG "Eraseblock at 0x%08x is now very dirty. Removing from dirty list...\n", jeb->offset)); | |
625 | list_del(&jeb->list); | |
626 | D1(printk(KERN_DEBUG "...and adding to very_dirty_list\n")); | |
627 | list_add_tail(&jeb->list, &c->very_dirty_list); | |
628 | } else { | |
629 | D1(printk(KERN_DEBUG "Eraseblock at 0x%08x not moved anywhere. (free 0x%08x, dirty 0x%08x, used 0x%08x)\n", | |
182ec4ee TG |
630 | jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size)); |
631 | } | |
1da177e4 LT |
632 | |
633 | spin_unlock(&c->erase_completion_lock); | |
634 | ||
31fbdf7a AB |
635 | if (!jffs2_can_mark_obsolete(c) || jffs2_is_readonly(c) || |
636 | (c->flags & JFFS2_SB_FLAG_BUILDING)) { | |
1da177e4 LT |
637 | /* We didn't lock the erase_free_sem */ |
638 | return; | |
639 | } | |
640 | ||
641 | /* The erase_free_sem is locked, and has been since before we marked the node obsolete | |
642 | and potentially put its eraseblock onto the erase_pending_list. Thus, we know that | |
643 | the block hasn't _already_ been erased, and that 'ref' itself hasn't been freed yet | |
c38c1b61 | 644 | by jffs2_free_jeb_node_refs() in erase.c. Which is nice. */ |
1da177e4 LT |
645 | |
646 | D1(printk(KERN_DEBUG "obliterating obsoleted node at 0x%08x\n", ref_offset(ref))); | |
647 | ret = jffs2_flash_read(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n); | |
648 | if (ret) { | |
649 | printk(KERN_WARNING "Read error reading from obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret); | |
650 | goto out_erase_sem; | |
651 | } | |
652 | if (retlen != sizeof(n)) { | |
653 | printk(KERN_WARNING "Short read from obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen); | |
654 | goto out_erase_sem; | |
655 | } | |
1417fc44 DW |
656 | if (PAD(je32_to_cpu(n.totlen)) != PAD(freed_len)) { |
657 | 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 |
658 | goto out_erase_sem; |
659 | } | |
660 | if (!(je16_to_cpu(n.nodetype) & JFFS2_NODE_ACCURATE)) { | |
661 | D1(printk(KERN_DEBUG "Node at 0x%08x was already marked obsolete (nodetype 0x%04x)\n", ref_offset(ref), je16_to_cpu(n.nodetype))); | |
662 | goto out_erase_sem; | |
663 | } | |
664 | /* XXX FIXME: This is ugly now */ | |
665 | n.nodetype = cpu_to_je16(je16_to_cpu(n.nodetype) & ~JFFS2_NODE_ACCURATE); | |
666 | ret = jffs2_flash_write(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n); | |
667 | if (ret) { | |
668 | printk(KERN_WARNING "Write error in obliterating obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret); | |
669 | goto out_erase_sem; | |
670 | } | |
671 | if (retlen != sizeof(n)) { | |
672 | printk(KERN_WARNING "Short write in obliterating obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen); | |
673 | goto out_erase_sem; | |
674 | } | |
675 | ||
676 | /* Nodes which have been marked obsolete no longer need to be | |
677 | associated with any inode. Remove them from the per-inode list. | |
182ec4ee TG |
678 | |
679 | Note we can't do this for NAND at the moment because we need | |
1da177e4 LT |
680 | obsolete dirent nodes to stay on the lists, because of the |
681 | horridness in jffs2_garbage_collect_deletion_dirent(). Also | |
182ec4ee | 682 | because we delete the inocache, and on NAND we need that to |
1da177e4 LT |
683 | stay around until all the nodes are actually erased, in order |
684 | to stop us from giving the same inode number to another newly | |
685 | created inode. */ | |
686 | if (ref->next_in_ino) { | |
687 | struct jffs2_inode_cache *ic; | |
688 | struct jffs2_raw_node_ref **p; | |
689 | ||
690 | spin_lock(&c->erase_completion_lock); | |
691 | ||
692 | ic = jffs2_raw_ref_to_ic(ref); | |
693 | for (p = &ic->nodes; (*p) != ref; p = &((*p)->next_in_ino)) | |
694 | ; | |
695 | ||
696 | *p = ref->next_in_ino; | |
697 | ref->next_in_ino = NULL; | |
698 | ||
c9f700f8 KK |
699 | switch (ic->class) { |
700 | #ifdef CONFIG_JFFS2_FS_XATTR | |
701 | case RAWNODE_CLASS_XATTR_DATUM: | |
702 | jffs2_release_xattr_datum(c, (struct jffs2_xattr_datum *)ic); | |
703 | break; | |
704 | case RAWNODE_CLASS_XATTR_REF: | |
705 | jffs2_release_xattr_ref(c, (struct jffs2_xattr_ref *)ic); | |
706 | break; | |
707 | #endif | |
708 | default: | |
709 | if (ic->nodes == (void *)ic && ic->nlink == 0) | |
710 | jffs2_del_ino_cache(c, ic); | |
711 | break; | |
712 | } | |
1da177e4 LT |
713 | spin_unlock(&c->erase_completion_lock); |
714 | } | |
715 | ||
1da177e4 | 716 | out_erase_sem: |
ced22070 | 717 | mutex_unlock(&c->erase_free_sem); |
1da177e4 LT |
718 | } |
719 | ||
1da177e4 LT |
720 | int jffs2_thread_should_wake(struct jffs2_sb_info *c) |
721 | { | |
722 | int ret = 0; | |
723 | uint32_t dirty; | |
8fb870df DW |
724 | int nr_very_dirty = 0; |
725 | struct jffs2_eraseblock *jeb; | |
1da177e4 LT |
726 | |
727 | if (c->unchecked_size) { | |
728 | D1(printk(KERN_DEBUG "jffs2_thread_should_wake(): unchecked_size %d, checked_ino #%d\n", | |
729 | c->unchecked_size, c->checked_ino)); | |
730 | return 1; | |
731 | } | |
732 | ||
733 | /* dirty_size contains blocks on erase_pending_list | |
734 | * those blocks are counted in c->nr_erasing_blocks. | |
735 | * If one block is actually erased, it is not longer counted as dirty_space | |
736 | * but it is counted in c->nr_erasing_blocks, so we add it and subtract it | |
737 | * with c->nr_erasing_blocks * c->sector_size again. | |
738 | * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks | |
739 | * This helps us to force gc and pick eventually a clean block to spread the load. | |
740 | */ | |
741 | dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size; | |
742 | ||
182ec4ee TG |
743 | if (c->nr_free_blocks + c->nr_erasing_blocks < c->resv_blocks_gctrigger && |
744 | (dirty > c->nospc_dirty_size)) | |
1da177e4 LT |
745 | ret = 1; |
746 | ||
8fb870df DW |
747 | list_for_each_entry(jeb, &c->very_dirty_list, list) { |
748 | nr_very_dirty++; | |
749 | if (nr_very_dirty == c->vdirty_blocks_gctrigger) { | |
750 | ret = 1; | |
a8c68f32 DW |
751 | /* In debug mode, actually go through and count them all */ |
752 | D1(continue); | |
753 | break; | |
8fb870df DW |
754 | } |
755 | } | |
756 | ||
757 | 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", | |
758 | c->nr_free_blocks, c->nr_erasing_blocks, c->dirty_size, nr_very_dirty, ret?"yes":"no")); | |
1da177e4 LT |
759 | |
760 | return ret; | |
761 | } |