Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * JFFS2 -- Journalling Flash File System, Version 2. | |
3 | * | |
c00c310e | 4 | * Copyright © 2001-2007 Red Hat, Inc. |
1da177e4 LT |
5 | * |
6 | * Created by David Woodhouse <dwmw2@infradead.org> | |
7 | * | |
8 | * For licensing information, see the file 'LICENCE' in this directory. | |
9 | * | |
1da177e4 LT |
10 | */ |
11 | ||
12 | #include <linux/kernel.h> | |
1da177e4 LT |
13 | #include <linux/mtd/mtd.h> |
14 | #include <linux/compiler.h> | |
15 | #include <linux/sched.h> /* For cond_resched() */ | |
16 | #include "nodelist.h" | |
e631ddba | 17 | #include "debug.h" |
1da177e4 LT |
18 | |
19 | /** | |
20 | * jffs2_reserve_space - request physical space to write nodes to flash | |
21 | * @c: superblock info | |
22 | * @minsize: Minimum acceptable size of allocation | |
1da177e4 LT |
23 | * @len: Returned value of allocation length |
24 | * @prio: Allocation type - ALLOC_{NORMAL,DELETION} | |
25 | * | |
26 | * Requests a block of physical space on the flash. Returns zero for success | |
9fe4854c DW |
27 | * and puts 'len' into the appropriate place, or returns -ENOSPC or other |
28 | * error if appropriate. Doesn't return len since that's | |
1da177e4 LT |
29 | * |
30 | * If it returns zero, jffs2_reserve_space() also downs the per-filesystem | |
31 | * allocation semaphore, to prevent more than one allocation from being | |
32 | * active at any time. The semaphore is later released by jffs2_commit_allocation() | |
33 | * | |
34 | * jffs2_reserve_space() may trigger garbage collection in order to make room | |
35 | * for the requested allocation. | |
36 | */ | |
37 | ||
e631ddba | 38 | static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize, |
9fe4854c | 39 | uint32_t *len, uint32_t sumsize); |
1da177e4 | 40 | |
9fe4854c | 41 | int jffs2_reserve_space(struct jffs2_sb_info *c, uint32_t minsize, |
e631ddba | 42 | uint32_t *len, int prio, uint32_t sumsize) |
1da177e4 LT |
43 | { |
44 | int ret = -EAGAIN; | |
45 | int blocksneeded = c->resv_blocks_write; | |
46 | /* align it */ | |
47 | minsize = PAD(minsize); | |
48 | ||
49 | D1(printk(KERN_DEBUG "jffs2_reserve_space(): Requested 0x%x bytes\n", minsize)); | |
ced22070 | 50 | mutex_lock(&c->alloc_sem); |
1da177e4 LT |
51 | |
52 | D1(printk(KERN_DEBUG "jffs2_reserve_space(): alloc sem got\n")); | |
53 | ||
54 | spin_lock(&c->erase_completion_lock); | |
55 | ||
56 | /* this needs a little more thought (true <tglx> :)) */ | |
57 | while(ret == -EAGAIN) { | |
58 | while(c->nr_free_blocks + c->nr_erasing_blocks < blocksneeded) { | |
1da177e4 LT |
59 | uint32_t dirty, avail; |
60 | ||
61 | /* calculate real dirty size | |
62 | * dirty_size contains blocks on erase_pending_list | |
63 | * those blocks are counted in c->nr_erasing_blocks. | |
64 | * If one block is actually erased, it is not longer counted as dirty_space | |
65 | * but it is counted in c->nr_erasing_blocks, so we add it and subtract it | |
66 | * with c->nr_erasing_blocks * c->sector_size again. | |
67 | * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks | |
68 | * This helps us to force gc and pick eventually a clean block to spread the load. | |
69 | * We add unchecked_size here, as we hopefully will find some space to use. | |
70 | * This will affect the sum only once, as gc first finishes checking | |
71 | * of nodes. | |
72 | */ | |
73 | dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size + c->unchecked_size; | |
74 | if (dirty < c->nospc_dirty_size) { | |
75 | if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) { | |
4132ace8 | 76 | D1(printk(KERN_NOTICE "jffs2_reserve_space(): Low on dirty space to GC, but it's a deletion. Allowing...\n")); |
1da177e4 LT |
77 | break; |
78 | } | |
79 | D1(printk(KERN_DEBUG "dirty size 0x%08x + unchecked_size 0x%08x < nospc_dirty_size 0x%08x, returning -ENOSPC\n", | |
80 | dirty, c->unchecked_size, c->sector_size)); | |
81 | ||
82 | spin_unlock(&c->erase_completion_lock); | |
ced22070 | 83 | mutex_unlock(&c->alloc_sem); |
1da177e4 LT |
84 | return -ENOSPC; |
85 | } | |
182ec4ee | 86 | |
1da177e4 LT |
87 | /* Calc possibly available space. Possibly available means that we |
88 | * don't know, if unchecked size contains obsoleted nodes, which could give us some | |
89 | * more usable space. This will affect the sum only once, as gc first finishes checking | |
90 | * of nodes. | |
182ec4ee | 91 | + Return -ENOSPC, if the maximum possibly available space is less or equal than |
1da177e4 LT |
92 | * blocksneeded * sector_size. |
93 | * This blocks endless gc looping on a filesystem, which is nearly full, even if | |
94 | * the check above passes. | |
95 | */ | |
96 | avail = c->free_size + c->dirty_size + c->erasing_size + c->unchecked_size; | |
97 | if ( (avail / c->sector_size) <= blocksneeded) { | |
98 | if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) { | |
4132ace8 | 99 | D1(printk(KERN_NOTICE "jffs2_reserve_space(): Low on possibly available space, but it's a deletion. Allowing...\n")); |
1da177e4 LT |
100 | break; |
101 | } | |
102 | ||
103 | D1(printk(KERN_DEBUG "max. available size 0x%08x < blocksneeded * sector_size 0x%08x, returning -ENOSPC\n", | |
104 | avail, blocksneeded * c->sector_size)); | |
105 | spin_unlock(&c->erase_completion_lock); | |
ced22070 | 106 | mutex_unlock(&c->alloc_sem); |
1da177e4 LT |
107 | return -ENOSPC; |
108 | } | |
109 | ||
ced22070 | 110 | mutex_unlock(&c->alloc_sem); |
1da177e4 LT |
111 | |
112 | 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", | |
113 | 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, | |
114 | c->free_size + c->dirty_size + c->wasted_size + c->used_size + c->erasing_size + c->bad_size, c->flash_size)); | |
115 | spin_unlock(&c->erase_completion_lock); | |
182ec4ee | 116 | |
1da177e4 | 117 | ret = jffs2_garbage_collect_pass(c); |
422b1202 DW |
118 | |
119 | if (ret == -EAGAIN) | |
120 | jffs2_erase_pending_blocks(c, 1); | |
121 | else if (ret) | |
1da177e4 LT |
122 | return ret; |
123 | ||
124 | cond_resched(); | |
125 | ||
126 | if (signal_pending(current)) | |
127 | return -EINTR; | |
128 | ||
ced22070 | 129 | mutex_lock(&c->alloc_sem); |
1da177e4 LT |
130 | spin_lock(&c->erase_completion_lock); |
131 | } | |
132 | ||
9fe4854c | 133 | ret = jffs2_do_reserve_space(c, minsize, len, sumsize); |
1da177e4 LT |
134 | if (ret) { |
135 | D1(printk(KERN_DEBUG "jffs2_reserve_space: ret is %d\n", ret)); | |
136 | } | |
137 | } | |
138 | spin_unlock(&c->erase_completion_lock); | |
2f785402 | 139 | if (!ret) |
046b8b98 | 140 | ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1); |
1da177e4 | 141 | if (ret) |
ced22070 | 142 | mutex_unlock(&c->alloc_sem); |
1da177e4 LT |
143 | return ret; |
144 | } | |
145 | ||
9fe4854c DW |
146 | int jffs2_reserve_space_gc(struct jffs2_sb_info *c, uint32_t minsize, |
147 | uint32_t *len, uint32_t sumsize) | |
1da177e4 LT |
148 | { |
149 | int ret = -EAGAIN; | |
150 | minsize = PAD(minsize); | |
151 | ||
152 | D1(printk(KERN_DEBUG "jffs2_reserve_space_gc(): Requested 0x%x bytes\n", minsize)); | |
153 | ||
154 | spin_lock(&c->erase_completion_lock); | |
155 | while(ret == -EAGAIN) { | |
9fe4854c | 156 | ret = jffs2_do_reserve_space(c, minsize, len, sumsize); |
1da177e4 | 157 | if (ret) { |
ef53cb02 | 158 | D1(printk(KERN_DEBUG "jffs2_reserve_space_gc: looping, ret is %d\n", 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); |
2f785402 | 164 | |
1da177e4 LT |
165 | return ret; |
166 | } | |
167 | ||
e631ddba FH |
168 | |
169 | /* Classify nextblock (clean, dirty of verydirty) and force to select an other one */ | |
170 | ||
171 | static void jffs2_close_nextblock(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb) | |
1da177e4 | 172 | { |
e631ddba | 173 | |
99c2594f AH |
174 | if (c->nextblock == NULL) { |
175 | D1(printk(KERN_DEBUG "jffs2_close_nextblock: Erase block at 0x%08x has already been placed in a list\n", | |
176 | jeb->offset)); | |
177 | return; | |
178 | } | |
e631ddba FH |
179 | /* Check, if we have a dirty block now, or if it was dirty already */ |
180 | if (ISDIRTY (jeb->wasted_size + jeb->dirty_size)) { | |
181 | c->dirty_size += jeb->wasted_size; | |
182 | c->wasted_size -= jeb->wasted_size; | |
183 | jeb->dirty_size += jeb->wasted_size; | |
184 | jeb->wasted_size = 0; | |
185 | if (VERYDIRTY(c, jeb->dirty_size)) { | |
186 | 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", | |
187 | jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size)); | |
188 | list_add_tail(&jeb->list, &c->very_dirty_list); | |
189 | } else { | |
190 | D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to dirty_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n", | |
191 | jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size)); | |
192 | list_add_tail(&jeb->list, &c->dirty_list); | |
193 | } | |
182ec4ee | 194 | } else { |
e631ddba FH |
195 | D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n", |
196 | jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size)); | |
197 | list_add_tail(&jeb->list, &c->clean_list); | |
198 | } | |
199 | c->nextblock = NULL; | |
200 | ||
201 | } | |
202 | ||
203 | /* Select a new jeb for nextblock */ | |
204 | ||
205 | static int jffs2_find_nextblock(struct jffs2_sb_info *c) | |
206 | { | |
207 | struct list_head *next; | |
182ec4ee | 208 | |
e631ddba FH |
209 | /* Take the next block off the 'free' list */ |
210 | ||
211 | if (list_empty(&c->free_list)) { | |
212 | ||
213 | if (!c->nr_erasing_blocks && | |
214 | !list_empty(&c->erasable_list)) { | |
215 | struct jffs2_eraseblock *ejeb; | |
216 | ||
217 | ejeb = list_entry(c->erasable_list.next, struct jffs2_eraseblock, list); | |
f116629d | 218 | list_move_tail(&ejeb->list, &c->erase_pending_list); |
e631ddba FH |
219 | c->nr_erasing_blocks++; |
220 | jffs2_erase_pending_trigger(c); | |
221 | D1(printk(KERN_DEBUG "jffs2_find_nextblock: Triggering erase of erasable block at 0x%08x\n", | |
222 | ejeb->offset)); | |
223 | } | |
224 | ||
225 | if (!c->nr_erasing_blocks && | |
226 | !list_empty(&c->erasable_pending_wbuf_list)) { | |
227 | D1(printk(KERN_DEBUG "jffs2_find_nextblock: Flushing write buffer\n")); | |
228 | /* c->nextblock is NULL, no update to c->nextblock allowed */ | |
1da177e4 | 229 | spin_unlock(&c->erase_completion_lock); |
1da177e4 LT |
230 | jffs2_flush_wbuf_pad(c); |
231 | spin_lock(&c->erase_completion_lock); | |
e631ddba FH |
232 | /* Have another go. It'll be on the erasable_list now */ |
233 | return -EAGAIN; | |
1da177e4 | 234 | } |
e631ddba FH |
235 | |
236 | if (!c->nr_erasing_blocks) { | |
237 | /* Ouch. We're in GC, or we wouldn't have got here. | |
238 | And there's no space left. At all. */ | |
182ec4ee TG |
239 | 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", |
240 | c->nr_erasing_blocks, c->nr_free_blocks, list_empty(&c->erasable_list)?"yes":"no", | |
e631ddba FH |
241 | list_empty(&c->erasing_list)?"yes":"no", list_empty(&c->erase_pending_list)?"yes":"no"); |
242 | return -ENOSPC; | |
1da177e4 | 243 | } |
e631ddba FH |
244 | |
245 | spin_unlock(&c->erase_completion_lock); | |
246 | /* Don't wait for it; just erase one right now */ | |
247 | jffs2_erase_pending_blocks(c, 1); | |
248 | spin_lock(&c->erase_completion_lock); | |
249 | ||
250 | /* An erase may have failed, decreasing the | |
251 | amount of free space available. So we must | |
252 | restart from the beginning */ | |
253 | return -EAGAIN; | |
1da177e4 | 254 | } |
e631ddba FH |
255 | |
256 | next = c->free_list.next; | |
257 | list_del(next); | |
258 | c->nextblock = list_entry(next, struct jffs2_eraseblock, list); | |
259 | c->nr_free_blocks--; | |
182ec4ee | 260 | |
e631ddba FH |
261 | jffs2_sum_reset_collected(c->summary); /* reset collected summary */ |
262 | ||
f04de505 | 263 | #ifdef CONFIG_JFFS2_FS_WRITEBUFFER |
5bf17237 AB |
264 | /* adjust write buffer offset, else we get a non contiguous write bug */ |
265 | if (!(c->wbuf_ofs % c->sector_size) && !c->wbuf_len) | |
266 | c->wbuf_ofs = 0xffffffff; | |
f04de505 | 267 | #endif |
5bf17237 | 268 | |
e631ddba FH |
269 | D1(printk(KERN_DEBUG "jffs2_find_nextblock(): new nextblock = 0x%08x\n", c->nextblock->offset)); |
270 | ||
271 | return 0; | |
272 | } | |
273 | ||
274 | /* Called with alloc sem _and_ erase_completion_lock */ | |
9fe4854c DW |
275 | static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize, |
276 | uint32_t *len, uint32_t sumsize) | |
e631ddba FH |
277 | { |
278 | struct jffs2_eraseblock *jeb = c->nextblock; | |
9fe4854c | 279 | uint32_t reserved_size; /* for summary information at the end of the jeb */ |
e631ddba FH |
280 | int ret; |
281 | ||
282 | restart: | |
283 | reserved_size = 0; | |
284 | ||
285 | if (jffs2_sum_active() && (sumsize != JFFS2_SUMMARY_NOSUM_SIZE)) { | |
286 | /* NOSUM_SIZE means not to generate summary */ | |
287 | ||
288 | if (jeb) { | |
289 | reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE); | |
733802d9 | 290 | dbg_summary("minsize=%d , jeb->free=%d ," |
e631ddba FH |
291 | "summary->size=%d , sumsize=%d\n", |
292 | minsize, jeb->free_size, | |
293 | c->summary->sum_size, sumsize); | |
294 | } | |
295 | ||
296 | /* Is there enough space for writing out the current node, or we have to | |
297 | write out summary information now, close this jeb and select new nextblock? */ | |
298 | if (jeb && (PAD(minsize) + PAD(c->summary->sum_size + sumsize + | |
299 | JFFS2_SUMMARY_FRAME_SIZE) > jeb->free_size)) { | |
300 | ||
301 | /* Has summary been disabled for this jeb? */ | |
302 | if (jffs2_sum_is_disabled(c->summary)) { | |
303 | sumsize = JFFS2_SUMMARY_NOSUM_SIZE; | |
304 | goto restart; | |
1da177e4 LT |
305 | } |
306 | ||
e631ddba | 307 | /* Writing out the collected summary information */ |
733802d9 | 308 | dbg_summary("generating summary for 0x%08x.\n", jeb->offset); |
e631ddba FH |
309 | ret = jffs2_sum_write_sumnode(c); |
310 | ||
311 | if (ret) | |
312 | return ret; | |
313 | ||
314 | if (jffs2_sum_is_disabled(c->summary)) { | |
315 | /* jffs2_write_sumnode() couldn't write out the summary information | |
316 | diabling summary for this jeb and free the collected information | |
317 | */ | |
318 | sumsize = JFFS2_SUMMARY_NOSUM_SIZE; | |
319 | goto restart; | |
320 | } | |
321 | ||
322 | jffs2_close_nextblock(c, jeb); | |
323 | jeb = NULL; | |
34c0e906 FH |
324 | /* keep always valid value in reserved_size */ |
325 | reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE); | |
e631ddba FH |
326 | } |
327 | } else { | |
328 | if (jeb && minsize > jeb->free_size) { | |
fc6612f6 DW |
329 | uint32_t waste; |
330 | ||
e631ddba FH |
331 | /* Skip the end of this block and file it as having some dirty space */ |
332 | /* If there's a pending write to it, flush now */ | |
333 | ||
334 | if (jffs2_wbuf_dirty(c)) { | |
1da177e4 | 335 | spin_unlock(&c->erase_completion_lock); |
e631ddba | 336 | D1(printk(KERN_DEBUG "jffs2_do_reserve_space: Flushing write buffer\n")); |
1da177e4 LT |
337 | jffs2_flush_wbuf_pad(c); |
338 | spin_lock(&c->erase_completion_lock); | |
e631ddba FH |
339 | jeb = c->nextblock; |
340 | goto restart; | |
1da177e4 LT |
341 | } |
342 | ||
fc6612f6 DW |
343 | spin_unlock(&c->erase_completion_lock); |
344 | ||
345 | ret = jffs2_prealloc_raw_node_refs(c, jeb, 1); | |
346 | if (ret) | |
347 | return ret; | |
348 | /* Just lock it again and continue. Nothing much can change because | |
349 | we hold c->alloc_sem anyway. In fact, it's not entirely clear why | |
350 | we hold c->erase_completion_lock in the majority of this function... | |
351 | but that's a question for another (more caffeine-rich) day. */ | |
352 | spin_lock(&c->erase_completion_lock); | |
353 | ||
354 | waste = jeb->free_size; | |
355 | jffs2_link_node_ref(c, jeb, | |
356 | (jeb->offset + c->sector_size - waste) | REF_OBSOLETE, | |
357 | waste, NULL); | |
358 | /* FIXME: that made it count as dirty. Convert to wasted */ | |
359 | jeb->dirty_size -= waste; | |
360 | c->dirty_size -= waste; | |
361 | jeb->wasted_size += waste; | |
362 | c->wasted_size += waste; | |
1da177e4 | 363 | |
e631ddba FH |
364 | jffs2_close_nextblock(c, jeb); |
365 | jeb = NULL; | |
1da177e4 | 366 | } |
e631ddba FH |
367 | } |
368 | ||
369 | if (!jeb) { | |
370 | ||
371 | ret = jffs2_find_nextblock(c); | |
372 | if (ret) | |
373 | return ret; | |
1da177e4 | 374 | |
e631ddba | 375 | jeb = c->nextblock; |
1da177e4 LT |
376 | |
377 | if (jeb->free_size != c->sector_size - c->cleanmarker_size) { | |
378 | printk(KERN_WARNING "Eep. Block 0x%08x taken from free_list had free_size of 0x%08x!!\n", jeb->offset, jeb->free_size); | |
379 | goto restart; | |
380 | } | |
381 | } | |
382 | /* OK, jeb (==c->nextblock) is now pointing at a block which definitely has | |
383 | enough space */ | |
e631ddba | 384 | *len = jeb->free_size - reserved_size; |
1da177e4 LT |
385 | |
386 | if (c->cleanmarker_size && jeb->used_size == c->cleanmarker_size && | |
387 | !jeb->first_node->next_in_ino) { | |
182ec4ee | 388 | /* Only node in it beforehand was a CLEANMARKER node (we think). |
1da177e4 | 389 | So mark it obsolete now that there's going to be another node |
182ec4ee | 390 | in the block. This will reduce used_size to zero but We've |
1da177e4 LT |
391 | already set c->nextblock so that jffs2_mark_node_obsolete() |
392 | won't try to refile it to the dirty_list. | |
393 | */ | |
394 | spin_unlock(&c->erase_completion_lock); | |
395 | jffs2_mark_node_obsolete(c, jeb->first_node); | |
396 | spin_lock(&c->erase_completion_lock); | |
397 | } | |
398 | ||
9fe4854c DW |
399 | D1(printk(KERN_DEBUG "jffs2_do_reserve_space(): Giving 0x%x bytes at 0x%x\n", |
400 | *len, jeb->offset + (c->sector_size - jeb->free_size))); | |
1da177e4 LT |
401 | return 0; |
402 | } | |
403 | ||
404 | /** | |
405 | * jffs2_add_physical_node_ref - add a physical node reference to the list | |
406 | * @c: superblock info | |
407 | * @new: new node reference to add | |
408 | * @len: length of this physical node | |
1da177e4 | 409 | * |
182ec4ee | 410 | * Should only be used to report nodes for which space has been allocated |
1da177e4 LT |
411 | * by jffs2_reserve_space. |
412 | * | |
413 | * Must be called with the alloc_sem held. | |
414 | */ | |
182ec4ee | 415 | |
2f785402 DW |
416 | struct jffs2_raw_node_ref *jffs2_add_physical_node_ref(struct jffs2_sb_info *c, |
417 | uint32_t ofs, uint32_t len, | |
418 | struct jffs2_inode_cache *ic) | |
1da177e4 LT |
419 | { |
420 | struct jffs2_eraseblock *jeb; | |
2f785402 | 421 | struct jffs2_raw_node_ref *new; |
1da177e4 | 422 | |
2f785402 | 423 | jeb = &c->blocks[ofs / c->sector_size]; |
1da177e4 | 424 | |
2f785402 DW |
425 | D1(printk(KERN_DEBUG "jffs2_add_physical_node_ref(): Node at 0x%x(%d), size 0x%x\n", |
426 | ofs & ~3, ofs & 3, len)); | |
1da177e4 | 427 | #if 1 |
2f785402 DW |
428 | /* Allow non-obsolete nodes only to be added at the end of c->nextblock, |
429 | if c->nextblock is set. Note that wbuf.c will file obsolete nodes | |
430 | even after refiling c->nextblock */ | |
431 | if ((c->nextblock || ((ofs & 3) != REF_OBSOLETE)) | |
432 | && (jeb != c->nextblock || (ofs & ~3) != jeb->offset + (c->sector_size - jeb->free_size))) { | |
66bfaeaa DW |
433 | printk(KERN_WARNING "argh. node added in wrong place at 0x%08x(%d)\n", ofs & ~3, ofs & 3); |
434 | if (c->nextblock) | |
435 | printk(KERN_WARNING "nextblock 0x%08x", c->nextblock->offset); | |
436 | else | |
437 | printk(KERN_WARNING "No nextblock"); | |
438 | printk(", expected at %08x\n", jeb->offset + (c->sector_size - jeb->free_size)); | |
2f785402 | 439 | return ERR_PTR(-EINVAL); |
1da177e4 LT |
440 | } |
441 | #endif | |
442 | spin_lock(&c->erase_completion_lock); | |
443 | ||
2f785402 | 444 | new = jffs2_link_node_ref(c, jeb, ofs, len, ic); |
1da177e4 | 445 | |
9b88f473 | 446 | if (!jeb->free_size && !jeb->dirty_size && !ISDIRTY(jeb->wasted_size)) { |
1da177e4 LT |
447 | /* If it lives on the dirty_list, jffs2_reserve_space will put it there */ |
448 | D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n", | |
449 | jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size)); | |
450 | if (jffs2_wbuf_dirty(c)) { | |
451 | /* Flush the last write in the block if it's outstanding */ | |
452 | spin_unlock(&c->erase_completion_lock); | |
453 | jffs2_flush_wbuf_pad(c); | |
454 | spin_lock(&c->erase_completion_lock); | |
455 | } | |
456 | ||
457 | list_add_tail(&jeb->list, &c->clean_list); | |
458 | c->nextblock = NULL; | |
459 | } | |
e0c8e42f AB |
460 | jffs2_dbg_acct_sanity_check_nolock(c,jeb); |
461 | jffs2_dbg_acct_paranoia_check_nolock(c, jeb); | |
1da177e4 LT |
462 | |
463 | spin_unlock(&c->erase_completion_lock); | |
464 | ||
2f785402 | 465 | return new; |
1da177e4 LT |
466 | } |
467 | ||
468 | ||
469 | void jffs2_complete_reservation(struct jffs2_sb_info *c) | |
470 | { | |
471 | D1(printk(KERN_DEBUG "jffs2_complete_reservation()\n")); | |
472 | jffs2_garbage_collect_trigger(c); | |
ced22070 | 473 | mutex_unlock(&c->alloc_sem); |
1da177e4 LT |
474 | } |
475 | ||
476 | static inline int on_list(struct list_head *obj, struct list_head *head) | |
477 | { | |
478 | struct list_head *this; | |
479 | ||
480 | list_for_each(this, head) { | |
481 | if (this == obj) { | |
482 | D1(printk("%p is on list at %p\n", obj, head)); | |
483 | return 1; | |
484 | ||
485 | } | |
486 | } | |
487 | return 0; | |
488 | } | |
489 | ||
490 | void jffs2_mark_node_obsolete(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref) | |
491 | { | |
492 | struct jffs2_eraseblock *jeb; | |
493 | int blocknr; | |
494 | struct jffs2_unknown_node n; | |
495 | int ret, addedsize; | |
496 | size_t retlen; | |
1417fc44 | 497 | uint32_t freed_len; |
1da177e4 | 498 | |
9bfeb691 | 499 | if(unlikely(!ref)) { |
1da177e4 LT |
500 | printk(KERN_NOTICE "EEEEEK. jffs2_mark_node_obsolete called with NULL node\n"); |
501 | return; | |
502 | } | |
503 | if (ref_obsolete(ref)) { | |
504 | D1(printk(KERN_DEBUG "jffs2_mark_node_obsolete called with already obsolete node at 0x%08x\n", ref_offset(ref))); | |
505 | return; | |
506 | } | |
507 | blocknr = ref->flash_offset / c->sector_size; | |
508 | if (blocknr >= c->nr_blocks) { | |
509 | printk(KERN_NOTICE "raw node at 0x%08x is off the end of device!\n", ref->flash_offset); | |
510 | BUG(); | |
511 | } | |
512 | jeb = &c->blocks[blocknr]; | |
513 | ||
514 | if (jffs2_can_mark_obsolete(c) && !jffs2_is_readonly(c) && | |
31fbdf7a | 515 | !(c->flags & (JFFS2_SB_FLAG_SCANNING | JFFS2_SB_FLAG_BUILDING))) { |
182ec4ee TG |
516 | /* Hm. This may confuse static lock analysis. If any of the above |
517 | three conditions is false, we're going to return from this | |
1da177e4 LT |
518 | function without actually obliterating any nodes or freeing |
519 | any jffs2_raw_node_refs. So we don't need to stop erases from | |
520 | happening, or protect against people holding an obsolete | |
521 | jffs2_raw_node_ref without the erase_completion_lock. */ | |
ced22070 | 522 | mutex_lock(&c->erase_free_sem); |
1da177e4 LT |
523 | } |
524 | ||
525 | spin_lock(&c->erase_completion_lock); | |
526 | ||
1417fc44 DW |
527 | freed_len = ref_totlen(c, jeb, ref); |
528 | ||
1da177e4 | 529 | if (ref_flags(ref) == REF_UNCHECKED) { |
1417fc44 | 530 | D1(if (unlikely(jeb->unchecked_size < freed_len)) { |
1da177e4 | 531 | 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 | 532 | freed_len, blocknr, ref->flash_offset, jeb->used_size); |
1da177e4 LT |
533 | BUG(); |
534 | }) | |
1417fc44 DW |
535 | D1(printk(KERN_DEBUG "Obsoleting previously unchecked node at 0x%08x of len %x: ", ref_offset(ref), freed_len)); |
536 | jeb->unchecked_size -= freed_len; | |
537 | c->unchecked_size -= freed_len; | |
1da177e4 | 538 | } else { |
1417fc44 | 539 | D1(if (unlikely(jeb->used_size < freed_len)) { |
1da177e4 | 540 | 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 | 541 | freed_len, blocknr, ref->flash_offset, jeb->used_size); |
1da177e4 LT |
542 | BUG(); |
543 | }) | |
1417fc44 DW |
544 | D1(printk(KERN_DEBUG "Obsoleting node at 0x%08x of len %#x: ", ref_offset(ref), freed_len)); |
545 | jeb->used_size -= freed_len; | |
546 | c->used_size -= freed_len; | |
1da177e4 LT |
547 | } |
548 | ||
549 | // Take care, that wasted size is taken into concern | |
1417fc44 | 550 | if ((jeb->dirty_size || ISDIRTY(jeb->wasted_size + freed_len)) && jeb != c->nextblock) { |
c7c16c8e | 551 | D1(printk("Dirtying\n")); |
1417fc44 DW |
552 | addedsize = freed_len; |
553 | jeb->dirty_size += freed_len; | |
554 | c->dirty_size += freed_len; | |
1da177e4 LT |
555 | |
556 | /* Convert wasted space to dirty, if not a bad block */ | |
557 | if (jeb->wasted_size) { | |
558 | if (on_list(&jeb->list, &c->bad_used_list)) { | |
559 | D1(printk(KERN_DEBUG "Leaving block at %08x on the bad_used_list\n", | |
560 | jeb->offset)); | |
561 | addedsize = 0; /* To fool the refiling code later */ | |
562 | } else { | |
563 | D1(printk(KERN_DEBUG "Converting %d bytes of wasted space to dirty in block at %08x\n", | |
564 | jeb->wasted_size, jeb->offset)); | |
565 | addedsize += jeb->wasted_size; | |
566 | jeb->dirty_size += jeb->wasted_size; | |
567 | c->dirty_size += jeb->wasted_size; | |
568 | c->wasted_size -= jeb->wasted_size; | |
569 | jeb->wasted_size = 0; | |
570 | } | |
571 | } | |
572 | } else { | |
c7c16c8e | 573 | D1(printk("Wasting\n")); |
1da177e4 | 574 | addedsize = 0; |
1417fc44 DW |
575 | jeb->wasted_size += freed_len; |
576 | c->wasted_size += freed_len; | |
1da177e4 LT |
577 | } |
578 | ref->flash_offset = ref_offset(ref) | REF_OBSOLETE; | |
182ec4ee | 579 | |
e0c8e42f AB |
580 | jffs2_dbg_acct_sanity_check_nolock(c, jeb); |
581 | jffs2_dbg_acct_paranoia_check_nolock(c, jeb); | |
1da177e4 | 582 | |
31fbdf7a AB |
583 | if (c->flags & JFFS2_SB_FLAG_SCANNING) { |
584 | /* Flash scanning is in progress. Don't muck about with the block | |
1da177e4 | 585 | lists because they're not ready yet, and don't actually |
182ec4ee | 586 | obliterate nodes that look obsolete. If they weren't |
1da177e4 LT |
587 | marked obsolete on the flash at the time they _became_ |
588 | obsolete, there was probably a reason for that. */ | |
589 | spin_unlock(&c->erase_completion_lock); | |
590 | /* We didn't lock the erase_free_sem */ | |
591 | return; | |
592 | } | |
593 | ||
594 | if (jeb == c->nextblock) { | |
595 | D2(printk(KERN_DEBUG "Not moving nextblock 0x%08x to dirty/erase_pending list\n", jeb->offset)); | |
596 | } else if (!jeb->used_size && !jeb->unchecked_size) { | |
597 | if (jeb == c->gcblock) { | |
598 | D1(printk(KERN_DEBUG "gcblock at 0x%08x completely dirtied. Clearing gcblock...\n", jeb->offset)); | |
599 | c->gcblock = NULL; | |
600 | } else { | |
601 | D1(printk(KERN_DEBUG "Eraseblock at 0x%08x completely dirtied. Removing from (dirty?) list...\n", jeb->offset)); | |
602 | list_del(&jeb->list); | |
603 | } | |
604 | if (jffs2_wbuf_dirty(c)) { | |
605 | D1(printk(KERN_DEBUG "...and adding to erasable_pending_wbuf_list\n")); | |
606 | list_add_tail(&jeb->list, &c->erasable_pending_wbuf_list); | |
607 | } else { | |
608 | if (jiffies & 127) { | |
609 | /* Most of the time, we just erase it immediately. Otherwise we | |
610 | spend ages scanning it on mount, etc. */ | |
611 | D1(printk(KERN_DEBUG "...and adding to erase_pending_list\n")); | |
612 | list_add_tail(&jeb->list, &c->erase_pending_list); | |
613 | c->nr_erasing_blocks++; | |
614 | jffs2_erase_pending_trigger(c); | |
615 | } else { | |
616 | /* Sometimes, however, we leave it elsewhere so it doesn't get | |
617 | immediately reused, and we spread the load a bit. */ | |
618 | D1(printk(KERN_DEBUG "...and adding to erasable_list\n")); | |
619 | list_add_tail(&jeb->list, &c->erasable_list); | |
182ec4ee | 620 | } |
1da177e4 LT |
621 | } |
622 | D1(printk(KERN_DEBUG "Done OK\n")); | |
623 | } else if (jeb == c->gcblock) { | |
624 | D2(printk(KERN_DEBUG "Not moving gcblock 0x%08x to dirty_list\n", jeb->offset)); | |
625 | } else if (ISDIRTY(jeb->dirty_size) && !ISDIRTY(jeb->dirty_size - addedsize)) { | |
626 | D1(printk(KERN_DEBUG "Eraseblock at 0x%08x is freshly dirtied. Removing from clean list...\n", jeb->offset)); | |
627 | list_del(&jeb->list); | |
628 | D1(printk(KERN_DEBUG "...and adding to dirty_list\n")); | |
629 | list_add_tail(&jeb->list, &c->dirty_list); | |
630 | } else if (VERYDIRTY(c, jeb->dirty_size) && | |
631 | !VERYDIRTY(c, jeb->dirty_size - addedsize)) { | |
632 | D1(printk(KERN_DEBUG "Eraseblock at 0x%08x is now very dirty. Removing from dirty list...\n", jeb->offset)); | |
633 | list_del(&jeb->list); | |
634 | D1(printk(KERN_DEBUG "...and adding to very_dirty_list\n")); | |
635 | list_add_tail(&jeb->list, &c->very_dirty_list); | |
636 | } else { | |
637 | D1(printk(KERN_DEBUG "Eraseblock at 0x%08x not moved anywhere. (free 0x%08x, dirty 0x%08x, used 0x%08x)\n", | |
182ec4ee TG |
638 | jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size)); |
639 | } | |
1da177e4 LT |
640 | |
641 | spin_unlock(&c->erase_completion_lock); | |
642 | ||
31fbdf7a AB |
643 | if (!jffs2_can_mark_obsolete(c) || jffs2_is_readonly(c) || |
644 | (c->flags & JFFS2_SB_FLAG_BUILDING)) { | |
1da177e4 LT |
645 | /* We didn't lock the erase_free_sem */ |
646 | return; | |
647 | } | |
648 | ||
649 | /* The erase_free_sem is locked, and has been since before we marked the node obsolete | |
650 | and potentially put its eraseblock onto the erase_pending_list. Thus, we know that | |
651 | the block hasn't _already_ been erased, and that 'ref' itself hasn't been freed yet | |
c38c1b61 | 652 | by jffs2_free_jeb_node_refs() in erase.c. Which is nice. */ |
1da177e4 LT |
653 | |
654 | D1(printk(KERN_DEBUG "obliterating obsoleted node at 0x%08x\n", ref_offset(ref))); | |
655 | ret = jffs2_flash_read(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n); | |
656 | if (ret) { | |
657 | printk(KERN_WARNING "Read error reading from obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret); | |
658 | goto out_erase_sem; | |
659 | } | |
660 | if (retlen != sizeof(n)) { | |
661 | printk(KERN_WARNING "Short read from obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen); | |
662 | goto out_erase_sem; | |
663 | } | |
1417fc44 DW |
664 | if (PAD(je32_to_cpu(n.totlen)) != PAD(freed_len)) { |
665 | 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 |
666 | goto out_erase_sem; |
667 | } | |
668 | if (!(je16_to_cpu(n.nodetype) & JFFS2_NODE_ACCURATE)) { | |
669 | D1(printk(KERN_DEBUG "Node at 0x%08x was already marked obsolete (nodetype 0x%04x)\n", ref_offset(ref), je16_to_cpu(n.nodetype))); | |
670 | goto out_erase_sem; | |
671 | } | |
672 | /* XXX FIXME: This is ugly now */ | |
673 | n.nodetype = cpu_to_je16(je16_to_cpu(n.nodetype) & ~JFFS2_NODE_ACCURATE); | |
674 | ret = jffs2_flash_write(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n); | |
675 | if (ret) { | |
676 | printk(KERN_WARNING "Write error in obliterating obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret); | |
677 | goto out_erase_sem; | |
678 | } | |
679 | if (retlen != sizeof(n)) { | |
680 | printk(KERN_WARNING "Short write in obliterating obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen); | |
681 | goto out_erase_sem; | |
682 | } | |
683 | ||
684 | /* Nodes which have been marked obsolete no longer need to be | |
685 | associated with any inode. Remove them from the per-inode list. | |
182ec4ee TG |
686 | |
687 | Note we can't do this for NAND at the moment because we need | |
1da177e4 LT |
688 | obsolete dirent nodes to stay on the lists, because of the |
689 | horridness in jffs2_garbage_collect_deletion_dirent(). Also | |
182ec4ee | 690 | because we delete the inocache, and on NAND we need that to |
1da177e4 LT |
691 | stay around until all the nodes are actually erased, in order |
692 | to stop us from giving the same inode number to another newly | |
693 | created inode. */ | |
694 | if (ref->next_in_ino) { | |
695 | struct jffs2_inode_cache *ic; | |
696 | struct jffs2_raw_node_ref **p; | |
697 | ||
698 | spin_lock(&c->erase_completion_lock); | |
699 | ||
700 | ic = jffs2_raw_ref_to_ic(ref); | |
701 | for (p = &ic->nodes; (*p) != ref; p = &((*p)->next_in_ino)) | |
702 | ; | |
703 | ||
704 | *p = ref->next_in_ino; | |
705 | ref->next_in_ino = NULL; | |
706 | ||
c9f700f8 KK |
707 | switch (ic->class) { |
708 | #ifdef CONFIG_JFFS2_FS_XATTR | |
709 | case RAWNODE_CLASS_XATTR_DATUM: | |
710 | jffs2_release_xattr_datum(c, (struct jffs2_xattr_datum *)ic); | |
711 | break; | |
712 | case RAWNODE_CLASS_XATTR_REF: | |
713 | jffs2_release_xattr_ref(c, (struct jffs2_xattr_ref *)ic); | |
714 | break; | |
715 | #endif | |
716 | default: | |
27c72b04 | 717 | if (ic->nodes == (void *)ic && ic->pino_nlink == 0) |
c9f700f8 KK |
718 | jffs2_del_ino_cache(c, ic); |
719 | break; | |
720 | } | |
1da177e4 LT |
721 | spin_unlock(&c->erase_completion_lock); |
722 | } | |
723 | ||
1da177e4 | 724 | out_erase_sem: |
ced22070 | 725 | mutex_unlock(&c->erase_free_sem); |
1da177e4 LT |
726 | } |
727 | ||
1da177e4 LT |
728 | int jffs2_thread_should_wake(struct jffs2_sb_info *c) |
729 | { | |
730 | int ret = 0; | |
731 | uint32_t dirty; | |
8fb870df DW |
732 | int nr_very_dirty = 0; |
733 | struct jffs2_eraseblock *jeb; | |
1da177e4 LT |
734 | |
735 | if (c->unchecked_size) { | |
736 | D1(printk(KERN_DEBUG "jffs2_thread_should_wake(): unchecked_size %d, checked_ino #%d\n", | |
737 | c->unchecked_size, c->checked_ino)); | |
738 | return 1; | |
739 | } | |
740 | ||
741 | /* dirty_size contains blocks on erase_pending_list | |
742 | * those blocks are counted in c->nr_erasing_blocks. | |
743 | * If one block is actually erased, it is not longer counted as dirty_space | |
744 | * but it is counted in c->nr_erasing_blocks, so we add it and subtract it | |
745 | * with c->nr_erasing_blocks * c->sector_size again. | |
746 | * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks | |
747 | * This helps us to force gc and pick eventually a clean block to spread the load. | |
748 | */ | |
749 | dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size; | |
750 | ||
182ec4ee TG |
751 | if (c->nr_free_blocks + c->nr_erasing_blocks < c->resv_blocks_gctrigger && |
752 | (dirty > c->nospc_dirty_size)) | |
1da177e4 LT |
753 | ret = 1; |
754 | ||
8fb870df DW |
755 | list_for_each_entry(jeb, &c->very_dirty_list, list) { |
756 | nr_very_dirty++; | |
757 | if (nr_very_dirty == c->vdirty_blocks_gctrigger) { | |
758 | ret = 1; | |
a8c68f32 DW |
759 | /* In debug mode, actually go through and count them all */ |
760 | D1(continue); | |
761 | break; | |
8fb870df DW |
762 | } |
763 | } | |
764 | ||
765 | 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", | |
766 | c->nr_free_blocks, c->nr_erasing_blocks, c->dirty_size, nr_very_dirty, ret?"yes":"no")); | |
1da177e4 LT |
767 | |
768 | return ret; | |
769 | } |