Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * JFFS2 -- Journalling Flash File System, Version 2. | |
3 | * | |
4 | * Copyright (C) 2001-2003 Red Hat, Inc. | |
5 | * | |
6 | * Created by David Woodhouse <dwmw2@infradead.org> | |
7 | * | |
8 | * For licensing information, see the file 'LICENCE' in this directory. | |
9 | * | |
733802d9 | 10 | * $Id: nodemgmt.c,v 1.127 2005/09/20 15:49:12 dedekind Exp $ |
1da177e4 LT |
11 | * |
12 | */ | |
13 | ||
14 | #include <linux/kernel.h> | |
15 | #include <linux/slab.h> | |
16 | #include <linux/mtd/mtd.h> | |
17 | #include <linux/compiler.h> | |
18 | #include <linux/sched.h> /* For cond_resched() */ | |
19 | #include "nodelist.h" | |
e631ddba | 20 | #include "debug.h" |
1da177e4 LT |
21 | |
22 | /** | |
23 | * jffs2_reserve_space - request physical space to write nodes to flash | |
24 | * @c: superblock info | |
25 | * @minsize: Minimum acceptable size of allocation | |
1da177e4 LT |
26 | * @len: Returned value of allocation length |
27 | * @prio: Allocation type - ALLOC_{NORMAL,DELETION} | |
28 | * | |
29 | * Requests a block of physical space on the flash. Returns zero for success | |
9fe4854c DW |
30 | * and puts 'len' into the appropriate place, or returns -ENOSPC or other |
31 | * error if appropriate. Doesn't return len since that's | |
1da177e4 LT |
32 | * |
33 | * If it returns zero, jffs2_reserve_space() also downs the per-filesystem | |
34 | * allocation semaphore, to prevent more than one allocation from being | |
35 | * active at any time. The semaphore is later released by jffs2_commit_allocation() | |
36 | * | |
37 | * jffs2_reserve_space() may trigger garbage collection in order to make room | |
38 | * for the requested allocation. | |
39 | */ | |
40 | ||
e631ddba | 41 | static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize, |
9fe4854c | 42 | uint32_t *len, uint32_t sumsize); |
1da177e4 | 43 | |
9fe4854c | 44 | int jffs2_reserve_space(struct jffs2_sb_info *c, uint32_t minsize, |
e631ddba | 45 | uint32_t *len, int prio, uint32_t sumsize) |
1da177e4 LT |
46 | { |
47 | int ret = -EAGAIN; | |
48 | int blocksneeded = c->resv_blocks_write; | |
49 | /* align it */ | |
50 | minsize = PAD(minsize); | |
51 | ||
52 | D1(printk(KERN_DEBUG "jffs2_reserve_space(): Requested 0x%x bytes\n", minsize)); | |
53 | down(&c->alloc_sem); | |
54 | ||
55 | D1(printk(KERN_DEBUG "jffs2_reserve_space(): alloc sem got\n")); | |
56 | ||
57 | spin_lock(&c->erase_completion_lock); | |
58 | ||
59 | /* this needs a little more thought (true <tglx> :)) */ | |
60 | while(ret == -EAGAIN) { | |
61 | while(c->nr_free_blocks + c->nr_erasing_blocks < blocksneeded) { | |
62 | int ret; | |
63 | uint32_t dirty, avail; | |
64 | ||
65 | /* calculate real dirty size | |
66 | * dirty_size contains blocks on erase_pending_list | |
67 | * those blocks are counted in c->nr_erasing_blocks. | |
68 | * If one block is actually erased, it is not longer counted as dirty_space | |
69 | * but it is counted in c->nr_erasing_blocks, so we add it and subtract it | |
70 | * with c->nr_erasing_blocks * c->sector_size again. | |
71 | * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks | |
72 | * This helps us to force gc and pick eventually a clean block to spread the load. | |
73 | * We add unchecked_size here, as we hopefully will find some space to use. | |
74 | * This will affect the sum only once, as gc first finishes checking | |
75 | * of nodes. | |
76 | */ | |
77 | dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size + c->unchecked_size; | |
78 | if (dirty < c->nospc_dirty_size) { | |
79 | if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) { | |
4132ace8 | 80 | D1(printk(KERN_NOTICE "jffs2_reserve_space(): Low on dirty space to GC, but it's a deletion. Allowing...\n")); |
1da177e4 LT |
81 | break; |
82 | } | |
83 | D1(printk(KERN_DEBUG "dirty size 0x%08x + unchecked_size 0x%08x < nospc_dirty_size 0x%08x, returning -ENOSPC\n", | |
84 | dirty, c->unchecked_size, c->sector_size)); | |
85 | ||
86 | spin_unlock(&c->erase_completion_lock); | |
87 | up(&c->alloc_sem); | |
88 | return -ENOSPC; | |
89 | } | |
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) { | |
4132ace8 | 103 | D1(printk(KERN_NOTICE "jffs2_reserve_space(): Low on possibly available space, but it's a deletion. Allowing...\n")); |
1da177e4 LT |
104 | break; |
105 | } | |
106 | ||
107 | D1(printk(KERN_DEBUG "max. available size 0x%08x < blocksneeded * sector_size 0x%08x, returning -ENOSPC\n", | |
108 | avail, blocksneeded * c->sector_size)); | |
109 | spin_unlock(&c->erase_completion_lock); | |
110 | up(&c->alloc_sem); | |
111 | return -ENOSPC; | |
112 | } | |
113 | ||
114 | up(&c->alloc_sem); | |
115 | ||
116 | D1(printk(KERN_DEBUG "Triggering GC pass. nr_free_blocks %d, nr_erasing_blocks %d, free_size 0x%08x, dirty_size 0x%08x, wasted_size 0x%08x, used_size 0x%08x, erasing_size 0x%08x, bad_size 0x%08x (total 0x%08x of 0x%08x)\n", | |
117 | c->nr_free_blocks, c->nr_erasing_blocks, c->free_size, c->dirty_size, c->wasted_size, c->used_size, c->erasing_size, c->bad_size, | |
118 | c->free_size + c->dirty_size + c->wasted_size + c->used_size + c->erasing_size + c->bad_size, c->flash_size)); | |
119 | spin_unlock(&c->erase_completion_lock); | |
182ec4ee | 120 | |
1da177e4 LT |
121 | ret = jffs2_garbage_collect_pass(c); |
122 | if (ret) | |
123 | return ret; | |
124 | ||
125 | cond_resched(); | |
126 | ||
127 | if (signal_pending(current)) | |
128 | return -EINTR; | |
129 | ||
130 | down(&c->alloc_sem); | |
131 | spin_lock(&c->erase_completion_lock); | |
132 | } | |
133 | ||
9fe4854c | 134 | ret = jffs2_do_reserve_space(c, minsize, len, sumsize); |
1da177e4 LT |
135 | if (ret) { |
136 | D1(printk(KERN_DEBUG "jffs2_reserve_space: ret is %d\n", ret)); | |
137 | } | |
138 | } | |
139 | spin_unlock(&c->erase_completion_lock); | |
2f785402 | 140 | if (!ret) |
046b8b98 | 141 | ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1); |
1da177e4 LT |
142 | if (ret) |
143 | up(&c->alloc_sem); | |
144 | return ret; | |
145 | } | |
146 | ||
9fe4854c DW |
147 | int jffs2_reserve_space_gc(struct jffs2_sb_info *c, uint32_t minsize, |
148 | uint32_t *len, uint32_t sumsize) | |
1da177e4 LT |
149 | { |
150 | int ret = -EAGAIN; | |
151 | minsize = PAD(minsize); | |
152 | ||
153 | D1(printk(KERN_DEBUG "jffs2_reserve_space_gc(): Requested 0x%x bytes\n", minsize)); | |
154 | ||
155 | spin_lock(&c->erase_completion_lock); | |
156 | while(ret == -EAGAIN) { | |
9fe4854c | 157 | ret = jffs2_do_reserve_space(c, minsize, len, sumsize); |
1da177e4 LT |
158 | if (ret) { |
159 | D1(printk(KERN_DEBUG "jffs2_reserve_space_gc: looping, ret is %d\n", ret)); | |
160 | } | |
161 | } | |
162 | spin_unlock(&c->erase_completion_lock); | |
2f785402 | 163 | if (!ret) |
046b8b98 | 164 | ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1); |
2f785402 | 165 | |
1da177e4 LT |
166 | return ret; |
167 | } | |
168 | ||
e631ddba FH |
169 | |
170 | /* Classify nextblock (clean, dirty of verydirty) and force to select an other one */ | |
171 | ||
172 | static void jffs2_close_nextblock(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb) | |
1da177e4 | 173 | { |
e631ddba FH |
174 | |
175 | /* Check, if we have a dirty block now, or if it was dirty already */ | |
176 | if (ISDIRTY (jeb->wasted_size + jeb->dirty_size)) { | |
177 | c->dirty_size += jeb->wasted_size; | |
178 | c->wasted_size -= jeb->wasted_size; | |
179 | jeb->dirty_size += jeb->wasted_size; | |
180 | jeb->wasted_size = 0; | |
181 | if (VERYDIRTY(c, jeb->dirty_size)) { | |
182 | 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", | |
183 | jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size)); | |
184 | list_add_tail(&jeb->list, &c->very_dirty_list); | |
185 | } else { | |
186 | D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to dirty_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n", | |
187 | jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size)); | |
188 | list_add_tail(&jeb->list, &c->dirty_list); | |
189 | } | |
182ec4ee | 190 | } else { |
e631ddba FH |
191 | D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n", |
192 | jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size)); | |
193 | list_add_tail(&jeb->list, &c->clean_list); | |
194 | } | |
195 | c->nextblock = NULL; | |
196 | ||
197 | } | |
198 | ||
199 | /* Select a new jeb for nextblock */ | |
200 | ||
201 | static int jffs2_find_nextblock(struct jffs2_sb_info *c) | |
202 | { | |
203 | struct list_head *next; | |
182ec4ee | 204 | |
e631ddba FH |
205 | /* Take the next block off the 'free' list */ |
206 | ||
207 | if (list_empty(&c->free_list)) { | |
208 | ||
209 | if (!c->nr_erasing_blocks && | |
210 | !list_empty(&c->erasable_list)) { | |
211 | struct jffs2_eraseblock *ejeb; | |
212 | ||
213 | ejeb = list_entry(c->erasable_list.next, struct jffs2_eraseblock, list); | |
214 | list_del(&ejeb->list); | |
215 | list_add_tail(&ejeb->list, &c->erase_pending_list); | |
216 | c->nr_erasing_blocks++; | |
217 | jffs2_erase_pending_trigger(c); | |
218 | D1(printk(KERN_DEBUG "jffs2_find_nextblock: Triggering erase of erasable block at 0x%08x\n", | |
219 | ejeb->offset)); | |
220 | } | |
221 | ||
222 | if (!c->nr_erasing_blocks && | |
223 | !list_empty(&c->erasable_pending_wbuf_list)) { | |
224 | D1(printk(KERN_DEBUG "jffs2_find_nextblock: Flushing write buffer\n")); | |
225 | /* c->nextblock is NULL, no update to c->nextblock allowed */ | |
1da177e4 | 226 | spin_unlock(&c->erase_completion_lock); |
1da177e4 LT |
227 | jffs2_flush_wbuf_pad(c); |
228 | spin_lock(&c->erase_completion_lock); | |
e631ddba FH |
229 | /* Have another go. It'll be on the erasable_list now */ |
230 | return -EAGAIN; | |
1da177e4 | 231 | } |
e631ddba FH |
232 | |
233 | if (!c->nr_erasing_blocks) { | |
234 | /* Ouch. We're in GC, or we wouldn't have got here. | |
235 | And there's no space left. At all. */ | |
182ec4ee TG |
236 | 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", |
237 | c->nr_erasing_blocks, c->nr_free_blocks, list_empty(&c->erasable_list)?"yes":"no", | |
e631ddba FH |
238 | list_empty(&c->erasing_list)?"yes":"no", list_empty(&c->erase_pending_list)?"yes":"no"); |
239 | return -ENOSPC; | |
1da177e4 | 240 | } |
e631ddba FH |
241 | |
242 | spin_unlock(&c->erase_completion_lock); | |
243 | /* Don't wait for it; just erase one right now */ | |
244 | jffs2_erase_pending_blocks(c, 1); | |
245 | spin_lock(&c->erase_completion_lock); | |
246 | ||
247 | /* An erase may have failed, decreasing the | |
248 | amount of free space available. So we must | |
249 | restart from the beginning */ | |
250 | return -EAGAIN; | |
1da177e4 | 251 | } |
e631ddba FH |
252 | |
253 | next = c->free_list.next; | |
254 | list_del(next); | |
255 | c->nextblock = list_entry(next, struct jffs2_eraseblock, list); | |
256 | c->nr_free_blocks--; | |
182ec4ee | 257 | |
e631ddba FH |
258 | jffs2_sum_reset_collected(c->summary); /* reset collected summary */ |
259 | ||
260 | D1(printk(KERN_DEBUG "jffs2_find_nextblock(): new nextblock = 0x%08x\n", c->nextblock->offset)); | |
261 | ||
262 | return 0; | |
263 | } | |
264 | ||
265 | /* Called with alloc sem _and_ erase_completion_lock */ | |
9fe4854c DW |
266 | static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize, |
267 | uint32_t *len, uint32_t sumsize) | |
e631ddba FH |
268 | { |
269 | struct jffs2_eraseblock *jeb = c->nextblock; | |
9fe4854c | 270 | uint32_t reserved_size; /* for summary information at the end of the jeb */ |
e631ddba FH |
271 | int ret; |
272 | ||
273 | restart: | |
274 | reserved_size = 0; | |
275 | ||
276 | if (jffs2_sum_active() && (sumsize != JFFS2_SUMMARY_NOSUM_SIZE)) { | |
277 | /* NOSUM_SIZE means not to generate summary */ | |
278 | ||
279 | if (jeb) { | |
280 | reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE); | |
733802d9 | 281 | dbg_summary("minsize=%d , jeb->free=%d ," |
e631ddba FH |
282 | "summary->size=%d , sumsize=%d\n", |
283 | minsize, jeb->free_size, | |
284 | c->summary->sum_size, sumsize); | |
285 | } | |
286 | ||
287 | /* Is there enough space for writing out the current node, or we have to | |
288 | write out summary information now, close this jeb and select new nextblock? */ | |
289 | if (jeb && (PAD(minsize) + PAD(c->summary->sum_size + sumsize + | |
290 | JFFS2_SUMMARY_FRAME_SIZE) > jeb->free_size)) { | |
291 | ||
292 | /* Has summary been disabled for this jeb? */ | |
293 | if (jffs2_sum_is_disabled(c->summary)) { | |
294 | sumsize = JFFS2_SUMMARY_NOSUM_SIZE; | |
295 | goto restart; | |
1da177e4 LT |
296 | } |
297 | ||
e631ddba | 298 | /* Writing out the collected summary information */ |
733802d9 | 299 | dbg_summary("generating summary for 0x%08x.\n", jeb->offset); |
e631ddba FH |
300 | ret = jffs2_sum_write_sumnode(c); |
301 | ||
302 | if (ret) | |
303 | return ret; | |
304 | ||
305 | if (jffs2_sum_is_disabled(c->summary)) { | |
306 | /* jffs2_write_sumnode() couldn't write out the summary information | |
307 | diabling summary for this jeb and free the collected information | |
308 | */ | |
309 | sumsize = JFFS2_SUMMARY_NOSUM_SIZE; | |
310 | goto restart; | |
311 | } | |
312 | ||
313 | jffs2_close_nextblock(c, jeb); | |
314 | jeb = NULL; | |
34c0e906 FH |
315 | /* keep always valid value in reserved_size */ |
316 | reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE); | |
e631ddba FH |
317 | } |
318 | } else { | |
319 | if (jeb && minsize > jeb->free_size) { | |
320 | /* Skip the end of this block and file it as having some dirty space */ | |
321 | /* If there's a pending write to it, flush now */ | |
322 | ||
323 | if (jffs2_wbuf_dirty(c)) { | |
1da177e4 | 324 | spin_unlock(&c->erase_completion_lock); |
e631ddba | 325 | D1(printk(KERN_DEBUG "jffs2_do_reserve_space: Flushing write buffer\n")); |
1da177e4 LT |
326 | jffs2_flush_wbuf_pad(c); |
327 | spin_lock(&c->erase_completion_lock); | |
e631ddba FH |
328 | jeb = c->nextblock; |
329 | goto restart; | |
1da177e4 LT |
330 | } |
331 | ||
e631ddba FH |
332 | c->wasted_size += jeb->free_size; |
333 | c->free_size -= jeb->free_size; | |
334 | jeb->wasted_size += jeb->free_size; | |
335 | jeb->free_size = 0; | |
1da177e4 | 336 | |
e631ddba FH |
337 | jffs2_close_nextblock(c, jeb); |
338 | jeb = NULL; | |
1da177e4 | 339 | } |
e631ddba FH |
340 | } |
341 | ||
342 | if (!jeb) { | |
343 | ||
344 | ret = jffs2_find_nextblock(c); | |
345 | if (ret) | |
346 | return ret; | |
1da177e4 | 347 | |
e631ddba | 348 | jeb = c->nextblock; |
1da177e4 LT |
349 | |
350 | if (jeb->free_size != c->sector_size - c->cleanmarker_size) { | |
351 | printk(KERN_WARNING "Eep. Block 0x%08x taken from free_list had free_size of 0x%08x!!\n", jeb->offset, jeb->free_size); | |
352 | goto restart; | |
353 | } | |
354 | } | |
355 | /* OK, jeb (==c->nextblock) is now pointing at a block which definitely has | |
356 | enough space */ | |
e631ddba | 357 | *len = jeb->free_size - reserved_size; |
1da177e4 LT |
358 | |
359 | if (c->cleanmarker_size && jeb->used_size == c->cleanmarker_size && | |
360 | !jeb->first_node->next_in_ino) { | |
182ec4ee | 361 | /* Only node in it beforehand was a CLEANMARKER node (we think). |
1da177e4 | 362 | So mark it obsolete now that there's going to be another node |
182ec4ee | 363 | in the block. This will reduce used_size to zero but We've |
1da177e4 LT |
364 | already set c->nextblock so that jffs2_mark_node_obsolete() |
365 | won't try to refile it to the dirty_list. | |
366 | */ | |
367 | spin_unlock(&c->erase_completion_lock); | |
368 | jffs2_mark_node_obsolete(c, jeb->first_node); | |
369 | spin_lock(&c->erase_completion_lock); | |
370 | } | |
371 | ||
9fe4854c DW |
372 | D1(printk(KERN_DEBUG "jffs2_do_reserve_space(): Giving 0x%x bytes at 0x%x\n", |
373 | *len, jeb->offset + (c->sector_size - jeb->free_size))); | |
1da177e4 LT |
374 | return 0; |
375 | } | |
376 | ||
377 | /** | |
378 | * jffs2_add_physical_node_ref - add a physical node reference to the list | |
379 | * @c: superblock info | |
380 | * @new: new node reference to add | |
381 | * @len: length of this physical node | |
1da177e4 | 382 | * |
182ec4ee | 383 | * Should only be used to report nodes for which space has been allocated |
1da177e4 LT |
384 | * by jffs2_reserve_space. |
385 | * | |
386 | * Must be called with the alloc_sem held. | |
387 | */ | |
182ec4ee | 388 | |
2f785402 DW |
389 | struct jffs2_raw_node_ref *jffs2_add_physical_node_ref(struct jffs2_sb_info *c, |
390 | uint32_t ofs, uint32_t len, | |
391 | struct jffs2_inode_cache *ic) | |
1da177e4 LT |
392 | { |
393 | struct jffs2_eraseblock *jeb; | |
2f785402 | 394 | struct jffs2_raw_node_ref *new; |
1da177e4 | 395 | |
2f785402 | 396 | jeb = &c->blocks[ofs / c->sector_size]; |
1da177e4 | 397 | |
2f785402 DW |
398 | D1(printk(KERN_DEBUG "jffs2_add_physical_node_ref(): Node at 0x%x(%d), size 0x%x\n", |
399 | ofs & ~3, ofs & 3, len)); | |
1da177e4 | 400 | #if 1 |
2f785402 DW |
401 | /* Allow non-obsolete nodes only to be added at the end of c->nextblock, |
402 | if c->nextblock is set. Note that wbuf.c will file obsolete nodes | |
403 | even after refiling c->nextblock */ | |
404 | if ((c->nextblock || ((ofs & 3) != REF_OBSOLETE)) | |
405 | && (jeb != c->nextblock || (ofs & ~3) != jeb->offset + (c->sector_size - jeb->free_size))) { | |
1da177e4 | 406 | printk(KERN_WARNING "argh. node added in wrong place\n"); |
2f785402 | 407 | return ERR_PTR(-EINVAL); |
1da177e4 LT |
408 | } |
409 | #endif | |
410 | spin_lock(&c->erase_completion_lock); | |
411 | ||
2f785402 | 412 | new = jffs2_link_node_ref(c, jeb, ofs, len, ic); |
1da177e4 | 413 | |
9b88f473 | 414 | if (!jeb->free_size && !jeb->dirty_size && !ISDIRTY(jeb->wasted_size)) { |
1da177e4 LT |
415 | /* If it lives on the dirty_list, jffs2_reserve_space will put it there */ |
416 | D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n", | |
417 | jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size)); | |
418 | if (jffs2_wbuf_dirty(c)) { | |
419 | /* Flush the last write in the block if it's outstanding */ | |
420 | spin_unlock(&c->erase_completion_lock); | |
421 | jffs2_flush_wbuf_pad(c); | |
422 | spin_lock(&c->erase_completion_lock); | |
423 | } | |
424 | ||
425 | list_add_tail(&jeb->list, &c->clean_list); | |
426 | c->nextblock = NULL; | |
427 | } | |
e0c8e42f AB |
428 | jffs2_dbg_acct_sanity_check_nolock(c,jeb); |
429 | jffs2_dbg_acct_paranoia_check_nolock(c, jeb); | |
1da177e4 LT |
430 | |
431 | spin_unlock(&c->erase_completion_lock); | |
432 | ||
2f785402 | 433 | return new; |
1da177e4 LT |
434 | } |
435 | ||
436 | ||
437 | void jffs2_complete_reservation(struct jffs2_sb_info *c) | |
438 | { | |
439 | D1(printk(KERN_DEBUG "jffs2_complete_reservation()\n")); | |
440 | jffs2_garbage_collect_trigger(c); | |
441 | up(&c->alloc_sem); | |
442 | } | |
443 | ||
444 | static inline int on_list(struct list_head *obj, struct list_head *head) | |
445 | { | |
446 | struct list_head *this; | |
447 | ||
448 | list_for_each(this, head) { | |
449 | if (this == obj) { | |
450 | D1(printk("%p is on list at %p\n", obj, head)); | |
451 | return 1; | |
452 | ||
453 | } | |
454 | } | |
455 | return 0; | |
456 | } | |
457 | ||
458 | void jffs2_mark_node_obsolete(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref) | |
459 | { | |
460 | struct jffs2_eraseblock *jeb; | |
99988f7b | 461 | struct jffs2_raw_node_ref *next_ref; |
1da177e4 LT |
462 | int blocknr; |
463 | struct jffs2_unknown_node n; | |
464 | int ret, addedsize; | |
465 | size_t retlen; | |
1417fc44 | 466 | uint32_t freed_len; |
1da177e4 LT |
467 | |
468 | if(!ref) { | |
469 | printk(KERN_NOTICE "EEEEEK. jffs2_mark_node_obsolete called with NULL node\n"); | |
470 | return; | |
471 | } | |
472 | if (ref_obsolete(ref)) { | |
473 | D1(printk(KERN_DEBUG "jffs2_mark_node_obsolete called with already obsolete node at 0x%08x\n", ref_offset(ref))); | |
474 | return; | |
475 | } | |
476 | blocknr = ref->flash_offset / c->sector_size; | |
477 | if (blocknr >= c->nr_blocks) { | |
478 | printk(KERN_NOTICE "raw node at 0x%08x is off the end of device!\n", ref->flash_offset); | |
479 | BUG(); | |
480 | } | |
481 | jeb = &c->blocks[blocknr]; | |
482 | ||
483 | if (jffs2_can_mark_obsolete(c) && !jffs2_is_readonly(c) && | |
31fbdf7a | 484 | !(c->flags & (JFFS2_SB_FLAG_SCANNING | JFFS2_SB_FLAG_BUILDING))) { |
182ec4ee TG |
485 | /* Hm. This may confuse static lock analysis. If any of the above |
486 | three conditions is false, we're going to return from this | |
1da177e4 LT |
487 | function without actually obliterating any nodes or freeing |
488 | any jffs2_raw_node_refs. So we don't need to stop erases from | |
489 | happening, or protect against people holding an obsolete | |
490 | jffs2_raw_node_ref without the erase_completion_lock. */ | |
491 | down(&c->erase_free_sem); | |
492 | } | |
493 | ||
494 | spin_lock(&c->erase_completion_lock); | |
495 | ||
1417fc44 DW |
496 | freed_len = ref_totlen(c, jeb, ref); |
497 | ||
1da177e4 | 498 | if (ref_flags(ref) == REF_UNCHECKED) { |
1417fc44 | 499 | D1(if (unlikely(jeb->unchecked_size < freed_len)) { |
1da177e4 | 500 | 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 | 501 | freed_len, blocknr, ref->flash_offset, jeb->used_size); |
1da177e4 LT |
502 | BUG(); |
503 | }) | |
1417fc44 DW |
504 | D1(printk(KERN_DEBUG "Obsoleting previously unchecked node at 0x%08x of len %x: ", ref_offset(ref), freed_len)); |
505 | jeb->unchecked_size -= freed_len; | |
506 | c->unchecked_size -= freed_len; | |
1da177e4 | 507 | } else { |
1417fc44 | 508 | D1(if (unlikely(jeb->used_size < freed_len)) { |
1da177e4 | 509 | 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 | 510 | freed_len, blocknr, ref->flash_offset, jeb->used_size); |
1da177e4 LT |
511 | BUG(); |
512 | }) | |
1417fc44 DW |
513 | D1(printk(KERN_DEBUG "Obsoleting node at 0x%08x of len %#x: ", ref_offset(ref), freed_len)); |
514 | jeb->used_size -= freed_len; | |
515 | c->used_size -= freed_len; | |
1da177e4 LT |
516 | } |
517 | ||
518 | // Take care, that wasted size is taken into concern | |
1417fc44 | 519 | if ((jeb->dirty_size || ISDIRTY(jeb->wasted_size + freed_len)) && jeb != c->nextblock) { |
c7c16c8e | 520 | D1(printk("Dirtying\n")); |
1417fc44 DW |
521 | addedsize = freed_len; |
522 | jeb->dirty_size += freed_len; | |
523 | c->dirty_size += freed_len; | |
1da177e4 LT |
524 | |
525 | /* Convert wasted space to dirty, if not a bad block */ | |
526 | if (jeb->wasted_size) { | |
527 | if (on_list(&jeb->list, &c->bad_used_list)) { | |
528 | D1(printk(KERN_DEBUG "Leaving block at %08x on the bad_used_list\n", | |
529 | jeb->offset)); | |
530 | addedsize = 0; /* To fool the refiling code later */ | |
531 | } else { | |
532 | D1(printk(KERN_DEBUG "Converting %d bytes of wasted space to dirty in block at %08x\n", | |
533 | jeb->wasted_size, jeb->offset)); | |
534 | addedsize += jeb->wasted_size; | |
535 | jeb->dirty_size += jeb->wasted_size; | |
536 | c->dirty_size += jeb->wasted_size; | |
537 | c->wasted_size -= jeb->wasted_size; | |
538 | jeb->wasted_size = 0; | |
539 | } | |
540 | } | |
541 | } else { | |
c7c16c8e | 542 | D1(printk("Wasting\n")); |
1da177e4 | 543 | addedsize = 0; |
1417fc44 DW |
544 | jeb->wasted_size += freed_len; |
545 | c->wasted_size += freed_len; | |
1da177e4 LT |
546 | } |
547 | ref->flash_offset = ref_offset(ref) | REF_OBSOLETE; | |
182ec4ee | 548 | |
e0c8e42f AB |
549 | jffs2_dbg_acct_sanity_check_nolock(c, jeb); |
550 | jffs2_dbg_acct_paranoia_check_nolock(c, jeb); | |
1da177e4 | 551 | |
31fbdf7a AB |
552 | if (c->flags & JFFS2_SB_FLAG_SCANNING) { |
553 | /* Flash scanning is in progress. Don't muck about with the block | |
1da177e4 | 554 | lists because they're not ready yet, and don't actually |
182ec4ee | 555 | obliterate nodes that look obsolete. If they weren't |
1da177e4 LT |
556 | marked obsolete on the flash at the time they _became_ |
557 | obsolete, there was probably a reason for that. */ | |
558 | spin_unlock(&c->erase_completion_lock); | |
559 | /* We didn't lock the erase_free_sem */ | |
560 | return; | |
561 | } | |
562 | ||
563 | if (jeb == c->nextblock) { | |
564 | D2(printk(KERN_DEBUG "Not moving nextblock 0x%08x to dirty/erase_pending list\n", jeb->offset)); | |
565 | } else if (!jeb->used_size && !jeb->unchecked_size) { | |
566 | if (jeb == c->gcblock) { | |
567 | D1(printk(KERN_DEBUG "gcblock at 0x%08x completely dirtied. Clearing gcblock...\n", jeb->offset)); | |
568 | c->gcblock = NULL; | |
569 | } else { | |
570 | D1(printk(KERN_DEBUG "Eraseblock at 0x%08x completely dirtied. Removing from (dirty?) list...\n", jeb->offset)); | |
571 | list_del(&jeb->list); | |
572 | } | |
573 | if (jffs2_wbuf_dirty(c)) { | |
574 | D1(printk(KERN_DEBUG "...and adding to erasable_pending_wbuf_list\n")); | |
575 | list_add_tail(&jeb->list, &c->erasable_pending_wbuf_list); | |
576 | } else { | |
577 | if (jiffies & 127) { | |
578 | /* Most of the time, we just erase it immediately. Otherwise we | |
579 | spend ages scanning it on mount, etc. */ | |
580 | D1(printk(KERN_DEBUG "...and adding to erase_pending_list\n")); | |
581 | list_add_tail(&jeb->list, &c->erase_pending_list); | |
582 | c->nr_erasing_blocks++; | |
583 | jffs2_erase_pending_trigger(c); | |
584 | } else { | |
585 | /* Sometimes, however, we leave it elsewhere so it doesn't get | |
586 | immediately reused, and we spread the load a bit. */ | |
587 | D1(printk(KERN_DEBUG "...and adding to erasable_list\n")); | |
588 | list_add_tail(&jeb->list, &c->erasable_list); | |
182ec4ee | 589 | } |
1da177e4 LT |
590 | } |
591 | D1(printk(KERN_DEBUG "Done OK\n")); | |
592 | } else if (jeb == c->gcblock) { | |
593 | D2(printk(KERN_DEBUG "Not moving gcblock 0x%08x to dirty_list\n", jeb->offset)); | |
594 | } else if (ISDIRTY(jeb->dirty_size) && !ISDIRTY(jeb->dirty_size - addedsize)) { | |
595 | D1(printk(KERN_DEBUG "Eraseblock at 0x%08x is freshly dirtied. Removing from clean list...\n", jeb->offset)); | |
596 | list_del(&jeb->list); | |
597 | D1(printk(KERN_DEBUG "...and adding to dirty_list\n")); | |
598 | list_add_tail(&jeb->list, &c->dirty_list); | |
599 | } else if (VERYDIRTY(c, jeb->dirty_size) && | |
600 | !VERYDIRTY(c, jeb->dirty_size - addedsize)) { | |
601 | D1(printk(KERN_DEBUG "Eraseblock at 0x%08x is now very dirty. Removing from dirty list...\n", jeb->offset)); | |
602 | list_del(&jeb->list); | |
603 | D1(printk(KERN_DEBUG "...and adding to very_dirty_list\n")); | |
604 | list_add_tail(&jeb->list, &c->very_dirty_list); | |
605 | } else { | |
606 | D1(printk(KERN_DEBUG "Eraseblock at 0x%08x not moved anywhere. (free 0x%08x, dirty 0x%08x, used 0x%08x)\n", | |
182ec4ee TG |
607 | jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size)); |
608 | } | |
1da177e4 LT |
609 | |
610 | spin_unlock(&c->erase_completion_lock); | |
611 | ||
31fbdf7a AB |
612 | if (!jffs2_can_mark_obsolete(c) || jffs2_is_readonly(c) || |
613 | (c->flags & JFFS2_SB_FLAG_BUILDING)) { | |
1da177e4 LT |
614 | /* We didn't lock the erase_free_sem */ |
615 | return; | |
616 | } | |
617 | ||
618 | /* The erase_free_sem is locked, and has been since before we marked the node obsolete | |
619 | and potentially put its eraseblock onto the erase_pending_list. Thus, we know that | |
620 | the block hasn't _already_ been erased, and that 'ref' itself hasn't been freed yet | |
c38c1b61 | 621 | by jffs2_free_jeb_node_refs() in erase.c. Which is nice. */ |
1da177e4 LT |
622 | |
623 | D1(printk(KERN_DEBUG "obliterating obsoleted node at 0x%08x\n", ref_offset(ref))); | |
624 | ret = jffs2_flash_read(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n); | |
625 | if (ret) { | |
626 | printk(KERN_WARNING "Read error reading from obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret); | |
627 | goto out_erase_sem; | |
628 | } | |
629 | if (retlen != sizeof(n)) { | |
630 | printk(KERN_WARNING "Short read from obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen); | |
631 | goto out_erase_sem; | |
632 | } | |
1417fc44 DW |
633 | if (PAD(je32_to_cpu(n.totlen)) != PAD(freed_len)) { |
634 | 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 |
635 | goto out_erase_sem; |
636 | } | |
637 | if (!(je16_to_cpu(n.nodetype) & JFFS2_NODE_ACCURATE)) { | |
638 | D1(printk(KERN_DEBUG "Node at 0x%08x was already marked obsolete (nodetype 0x%04x)\n", ref_offset(ref), je16_to_cpu(n.nodetype))); | |
639 | goto out_erase_sem; | |
640 | } | |
641 | /* XXX FIXME: This is ugly now */ | |
642 | n.nodetype = cpu_to_je16(je16_to_cpu(n.nodetype) & ~JFFS2_NODE_ACCURATE); | |
643 | ret = jffs2_flash_write(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n); | |
644 | if (ret) { | |
645 | printk(KERN_WARNING "Write error in obliterating obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret); | |
646 | goto out_erase_sem; | |
647 | } | |
648 | if (retlen != sizeof(n)) { | |
649 | printk(KERN_WARNING "Short write in obliterating obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen); | |
650 | goto out_erase_sem; | |
651 | } | |
652 | ||
653 | /* Nodes which have been marked obsolete no longer need to be | |
654 | associated with any inode. Remove them from the per-inode list. | |
182ec4ee TG |
655 | |
656 | Note we can't do this for NAND at the moment because we need | |
1da177e4 LT |
657 | obsolete dirent nodes to stay on the lists, because of the |
658 | horridness in jffs2_garbage_collect_deletion_dirent(). Also | |
182ec4ee | 659 | because we delete the inocache, and on NAND we need that to |
1da177e4 LT |
660 | stay around until all the nodes are actually erased, in order |
661 | to stop us from giving the same inode number to another newly | |
662 | created inode. */ | |
663 | if (ref->next_in_ino) { | |
664 | struct jffs2_inode_cache *ic; | |
665 | struct jffs2_raw_node_ref **p; | |
666 | ||
667 | spin_lock(&c->erase_completion_lock); | |
668 | ||
669 | ic = jffs2_raw_ref_to_ic(ref); | |
0eac940b DW |
670 | /* It seems we should never call jffs2_mark_node_obsolete() for |
671 | XATTR nodes.... yet. Make sure we notice if/when we change | |
672 | that :) */ | |
673 | BUG_ON(ic->class != RAWNODE_CLASS_INODE_CACHE); | |
1da177e4 LT |
674 | for (p = &ic->nodes; (*p) != ref; p = &((*p)->next_in_ino)) |
675 | ; | |
676 | ||
677 | *p = ref->next_in_ino; | |
678 | ref->next_in_ino = NULL; | |
679 | ||
437316d9 | 680 | if (ic->nodes == (void *)ic && ic->nlink == 0) |
1da177e4 | 681 | jffs2_del_ino_cache(c, ic); |
1da177e4 LT |
682 | |
683 | spin_unlock(&c->erase_completion_lock); | |
684 | } | |
685 | ||
686 | ||
687 | /* Merge with the next node in the physical list, if there is one | |
688 | and if it's also obsolete and if it doesn't belong to any inode */ | |
99988f7b | 689 | next_ref = ref_next(ref); |
182ec4ee | 690 | |
99988f7b | 691 | if (next_ref && ref_obsolete(next_ref) && !next_ref->next_in_ino) { |
1da177e4 LT |
692 | spin_lock(&c->erase_completion_lock); |
693 | ||
ca89a517 | 694 | #ifdef TEST_TOTLEN |
99988f7b | 695 | ref->__totlen += next_ref->__totlen; |
ca89a517 | 696 | #endif |
99988f7b DW |
697 | ref->next_phys = ref_next(next_ref); |
698 | if (jeb->last_node == next_ref) jeb->last_node = ref; | |
699 | if (jeb->gc_node == next_ref) { | |
1da177e4 LT |
700 | /* gc will be happy continuing gc on this node */ |
701 | jeb->gc_node=ref; | |
702 | } | |
703 | spin_unlock(&c->erase_completion_lock); | |
704 | ||
99988f7b | 705 | __jffs2_free_raw_node_ref(next_ref); |
1da177e4 | 706 | } |
182ec4ee | 707 | |
1da177e4 LT |
708 | /* Also merge with the previous node in the list, if there is one |
709 | and that one is obsolete */ | |
710 | if (ref != jeb->first_node ) { | |
711 | struct jffs2_raw_node_ref *p = jeb->first_node; | |
712 | ||
713 | spin_lock(&c->erase_completion_lock); | |
714 | ||
99988f7b DW |
715 | while ((next_ref = ref_next(ref)) != ref) |
716 | p = next_ref; | |
182ec4ee | 717 | |
1da177e4 | 718 | if (ref_obsolete(p) && !ref->next_in_ino) { |
ca89a517 | 719 | #ifdef TEST_TOTLEN |
1da177e4 | 720 | p->__totlen += ref->__totlen; |
ca89a517 | 721 | #endif |
1da177e4 LT |
722 | if (jeb->last_node == ref) { |
723 | jeb->last_node = p; | |
724 | } | |
725 | if (jeb->gc_node == ref) { | |
726 | /* gc will be happy continuing gc on this node */ | |
727 | jeb->gc_node=p; | |
728 | } | |
99988f7b | 729 | p->next_phys = ref_next(ref); |
2f785402 | 730 | __jffs2_free_raw_node_ref(ref); |
1da177e4 LT |
731 | } |
732 | spin_unlock(&c->erase_completion_lock); | |
733 | } | |
734 | out_erase_sem: | |
735 | up(&c->erase_free_sem); | |
736 | } | |
737 | ||
1da177e4 LT |
738 | int jffs2_thread_should_wake(struct jffs2_sb_info *c) |
739 | { | |
740 | int ret = 0; | |
741 | uint32_t dirty; | |
742 | ||
743 | if (c->unchecked_size) { | |
744 | D1(printk(KERN_DEBUG "jffs2_thread_should_wake(): unchecked_size %d, checked_ino #%d\n", | |
745 | c->unchecked_size, c->checked_ino)); | |
746 | return 1; | |
747 | } | |
748 | ||
749 | /* dirty_size contains blocks on erase_pending_list | |
750 | * those blocks are counted in c->nr_erasing_blocks. | |
751 | * If one block is actually erased, it is not longer counted as dirty_space | |
752 | * but it is counted in c->nr_erasing_blocks, so we add it and subtract it | |
753 | * with c->nr_erasing_blocks * c->sector_size again. | |
754 | * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks | |
755 | * This helps us to force gc and pick eventually a clean block to spread the load. | |
756 | */ | |
757 | dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size; | |
758 | ||
182ec4ee TG |
759 | if (c->nr_free_blocks + c->nr_erasing_blocks < c->resv_blocks_gctrigger && |
760 | (dirty > c->nospc_dirty_size)) | |
1da177e4 LT |
761 | ret = 1; |
762 | ||
182ec4ee | 763 | D1(printk(KERN_DEBUG "jffs2_thread_should_wake(): nr_free_blocks %d, nr_erasing_blocks %d, dirty_size 0x%x: %s\n", |
1da177e4 LT |
764 | c->nr_free_blocks, c->nr_erasing_blocks, c->dirty_size, ret?"yes":"no")); |
765 | ||
766 | return ret; | |
767 | } |