MOXA linux-2.6.x / linux-2.6.9-uc0 from sdlinux-moxaart.tgz
[linux-2.6.9-moxart.git] / fs / jffs2.org / gc.c
blobb180451d079a123b1f85b2cd501fafbd332eccde
1 /*
2 * JFFS2 -- Journalling Flash File System, Version 2.
4 * Copyright (C) 2001-2003 Red Hat, Inc.
6 * Created by David Woodhouse <dwmw2@redhat.com>
8 * For licensing information, see the file 'LICENCE' in this directory.
10 * $Id: gc.c,v 1.136 2004/05/27 19:06:09 gleixner Exp $
14 #include <linux/kernel.h>
15 #include <linux/mtd/mtd.h>
16 #include <linux/slab.h>
17 #include <linux/pagemap.h>
18 #include <linux/crc32.h>
19 #include <linux/compiler.h>
20 #include <linux/stat.h>
21 #include "nodelist.h"
22 #include "compr.h"
24 static int jffs2_garbage_collect_pristine(struct jffs2_sb_info *c,
25 struct jffs2_inode_cache *ic,
26 struct jffs2_raw_node_ref *raw);
27 static int jffs2_garbage_collect_metadata(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
28 struct jffs2_inode_info *f, struct jffs2_full_dnode *fd);
29 static int jffs2_garbage_collect_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
30 struct jffs2_inode_info *f, struct jffs2_full_dirent *fd);
31 static int jffs2_garbage_collect_deletion_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
32 struct jffs2_inode_info *f, struct jffs2_full_dirent *fd);
33 static int jffs2_garbage_collect_hole(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
34 struct jffs2_inode_info *f, struct jffs2_full_dnode *fn,
35 uint32_t start, uint32_t end);
36 static int jffs2_garbage_collect_dnode(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
37 struct jffs2_inode_info *f, struct jffs2_full_dnode *fn,
38 uint32_t start, uint32_t end);
39 static int jffs2_garbage_collect_live(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
40 struct jffs2_raw_node_ref *raw, struct jffs2_inode_info *f);
42 /* Called with erase_completion_lock held */
43 static struct jffs2_eraseblock *jffs2_find_gc_block(struct jffs2_sb_info *c)
45 struct jffs2_eraseblock *ret;
46 struct list_head *nextlist = NULL;
47 int n = jiffies % 128;
49 /* Pick an eraseblock to garbage collect next. This is where we'll
50 put the clever wear-levelling algorithms. Eventually. */
51 /* We possibly want to favour the dirtier blocks more when the
52 number of free blocks is low. */
53 if (!list_empty(&c->bad_used_list) && c->nr_free_blocks > c->resv_blocks_gcbad) {
54 D1(printk(KERN_DEBUG "Picking block from bad_used_list to GC next\n"));
55 nextlist = &c->bad_used_list;
56 } else if (n < 50 && !list_empty(&c->erasable_list)) {
57 /* Note that most of them will have gone directly to be erased.
58 So don't favour the erasable_list _too_ much. */
59 D1(printk(KERN_DEBUG "Picking block from erasable_list to GC next\n"));
60 nextlist = &c->erasable_list;
61 } else if (n < 110 && !list_empty(&c->very_dirty_list)) {
62 /* Most of the time, pick one off the very_dirty list */
63 D1(printk(KERN_DEBUG "Picking block from very_dirty_list to GC next\n"));
64 nextlist = &c->very_dirty_list;
65 } else if (n < 126 && !list_empty(&c->dirty_list)) {
66 D1(printk(KERN_DEBUG "Picking block from dirty_list to GC next\n"));
67 nextlist = &c->dirty_list;
68 } else if (!list_empty(&c->clean_list)) {
69 D1(printk(KERN_DEBUG "Picking block from clean_list to GC next\n"));
70 nextlist = &c->clean_list;
71 } else if (!list_empty(&c->dirty_list)) {
72 D1(printk(KERN_DEBUG "Picking block from dirty_list to GC next (clean_list was empty)\n"));
74 nextlist = &c->dirty_list;
75 } else if (!list_empty(&c->very_dirty_list)) {
76 D1(printk(KERN_DEBUG "Picking block from very_dirty_list to GC next (clean_list and dirty_list were empty)\n"));
77 nextlist = &c->very_dirty_list;
78 } else if (!list_empty(&c->erasable_list)) {
79 D1(printk(KERN_DEBUG "Picking block from erasable_list to GC next (clean_list and {very_,}dirty_list were empty)\n"));
81 nextlist = &c->erasable_list;
82 } else {
83 /* Eep. All were empty */
84 D1(printk(KERN_NOTICE "jffs2: No clean, dirty _or_ erasable blocks to GC from! Where are they all?\n"));
85 return NULL;
88 ret = list_entry(nextlist->next, struct jffs2_eraseblock, list);
89 list_del(&ret->list);
90 c->gcblock = ret;
91 ret->gc_node = ret->first_node;
92 if (!ret->gc_node) {
93 printk(KERN_WARNING "Eep. ret->gc_node for block at 0x%08x is NULL\n", ret->offset);
94 BUG();
97 /* Have we accidentally picked a clean block with wasted space ? */
98 if (ret->wasted_size) {
99 D1(printk(KERN_DEBUG "Converting wasted_size %08x to dirty_size\n", ret->wasted_size));
100 ret->dirty_size += ret->wasted_size;
101 c->wasted_size -= ret->wasted_size;
102 c->dirty_size += ret->wasted_size;
103 ret->wasted_size = 0;
106 D1(jffs2_dump_block_lists(c));
107 return ret;
110 /* jffs2_garbage_collect_pass
111 * Make a single attempt to progress GC. Move one node, and possibly
112 * start erasing one eraseblock.
114 int jffs2_garbage_collect_pass(struct jffs2_sb_info *c)
116 struct jffs2_inode_info *f;
117 struct jffs2_inode_cache *ic;
118 struct jffs2_eraseblock *jeb;
119 struct jffs2_raw_node_ref *raw;
120 int ret = 0, inum, nlink;
122 if (down_interruptible(&c->alloc_sem))
123 return -EINTR;
125 for (;;) {
126 spin_lock(&c->erase_completion_lock);
127 if (!c->unchecked_size)
128 break;
130 /* We can't start doing GC yet. We haven't finished checking
131 the node CRCs etc. Do it now. */
133 /* checked_ino is protected by the alloc_sem */
134 if (c->checked_ino > c->highest_ino) {
135 printk(KERN_CRIT "Checked all inodes but still 0x%x bytes of unchecked space?\n",
136 c->unchecked_size);
137 D1(jffs2_dump_block_lists(c));
138 spin_unlock(&c->erase_completion_lock);
139 BUG();
142 spin_unlock(&c->erase_completion_lock);
144 spin_lock(&c->inocache_lock);
146 ic = jffs2_get_ino_cache(c, c->checked_ino++);
148 if (!ic) {
149 spin_unlock(&c->inocache_lock);
150 continue;
153 if (!ic->nlink) {
154 D1(printk(KERN_DEBUG "Skipping check of ino #%d with nlink zero\n",
155 ic->ino));
156 spin_unlock(&c->inocache_lock);
157 continue;
159 switch(ic->state) {
160 case INO_STATE_CHECKEDABSENT:
161 case INO_STATE_PRESENT:
162 D1(printk(KERN_DEBUG "Skipping ino #%u already checked\n", ic->ino));
163 spin_unlock(&c->inocache_lock);
164 continue;
166 case INO_STATE_GC:
167 case INO_STATE_CHECKING:
168 printk(KERN_WARNING "Inode #%u is in state %d during CRC check phase!\n", ic->ino, ic->state);
169 spin_unlock(&c->inocache_lock);
170 BUG();
172 case INO_STATE_READING:
173 /* We need to wait for it to finish, lest we move on
174 and trigger the BUG() above while we haven't yet
175 finished checking all its nodes */
176 D1(printk(KERN_DEBUG "Waiting for ino #%u to finish reading\n", ic->ino));
177 up(&c->alloc_sem);
178 sleep_on_spinunlock(&c->inocache_wq, &c->inocache_lock);
179 return 0;
181 default:
182 BUG();
184 case INO_STATE_UNCHECKED:
187 ic->state = INO_STATE_CHECKING;
188 spin_unlock(&c->inocache_lock);
190 D1(printk(KERN_DEBUG "jffs2_garbage_collect_pass() triggering inode scan of ino#%u\n", ic->ino));
192 ret = jffs2_do_crccheck_inode(c, ic);
193 if (ret)
194 printk(KERN_WARNING "Returned error for crccheck of ino #%u. Expect badness...\n", ic->ino);
196 jffs2_set_inocache_state(c, ic, INO_STATE_CHECKEDABSENT);
197 up(&c->alloc_sem);
198 return ret;
201 /* First, work out which block we're garbage-collecting */
202 jeb = c->gcblock;
204 if (!jeb)
205 jeb = jffs2_find_gc_block(c);
207 if (!jeb) {
208 D1 (printk(KERN_NOTICE "jffs2: Couldn't find erase block to garbage collect!\n"));
209 spin_unlock(&c->erase_completion_lock);
210 up(&c->alloc_sem);
211 return -EIO;
214 D1(printk(KERN_DEBUG "GC from block %08x, used_size %08x, dirty_size %08x, free_size %08x\n", jeb->offset, jeb->used_size, jeb->dirty_size, jeb->free_size));
215 D1(if (c->nextblock)
216 printk(KERN_DEBUG "Nextblock at %08x, used_size %08x, dirty_size %08x, wasted_size %08x, free_size %08x\n", c->nextblock->offset, c->nextblock->used_size, c->nextblock->dirty_size, c->nextblock->wasted_size, c->nextblock->free_size));
218 if (!jeb->used_size) {
219 up(&c->alloc_sem);
220 goto eraseit;
223 raw = jeb->gc_node;
225 while(ref_obsolete(raw)) {
226 D1(printk(KERN_DEBUG "Node at 0x%08x is obsolete... skipping\n", ref_offset(raw)));
227 raw = raw->next_phys;
228 if (unlikely(!raw)) {
229 printk(KERN_WARNING "eep. End of raw list while still supposedly nodes to GC\n");
230 printk(KERN_WARNING "erase block at 0x%08x. free_size 0x%08x, dirty_size 0x%08x, used_size 0x%08x\n",
231 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size);
232 jeb->gc_node = raw;
233 spin_unlock(&c->erase_completion_lock);
234 up(&c->alloc_sem);
235 BUG();
238 jeb->gc_node = raw;
240 D1(printk(KERN_DEBUG "Going to garbage collect node at 0x%08x\n", ref_offset(raw)));
242 if (!raw->next_in_ino) {
243 /* Inode-less node. Clean marker, snapshot or something like that */
244 /* FIXME: If it's something that needs to be copied, including something
245 we don't grok that has JFFS2_NODETYPE_RWCOMPAT_COPY, we should do so */
246 spin_unlock(&c->erase_completion_lock);
247 jffs2_mark_node_obsolete(c, raw);
248 up(&c->alloc_sem);
249 goto eraseit_lock;
252 ic = jffs2_raw_ref_to_ic(raw);
254 /* We need to hold the inocache. Either the erase_completion_lock or
255 the inocache_lock are sufficient; we trade down since the inocache_lock
256 causes less contention. */
257 spin_lock(&c->inocache_lock);
259 spin_unlock(&c->erase_completion_lock);
261 D1(printk(KERN_DEBUG "jffs2_garbage_collect_pass collecting from block @0x%08x. Node @0x%08x(%d), ino #%u\n", jeb->offset, ref_offset(raw), ref_flags(raw), ic->ino));
263 /* Three possibilities:
264 1. Inode is already in-core. We must iget it and do proper
265 updating to its fragtree, etc.
266 2. Inode is not in-core, node is REF_PRISTINE. We lock the
267 inocache to prevent a read_inode(), copy the node intact.
268 3. Inode is not in-core, node is not pristine. We must iget()
269 and take the slow path.
272 switch(ic->state) {
273 case INO_STATE_CHECKEDABSENT:
274 /* It's been checked, but it's not currently in-core.
275 We can just copy any pristine nodes, but have
276 to prevent anyone else from doing read_inode() while
277 we're at it, so we set the state accordingly */
278 if (ref_flags(raw) == REF_PRISTINE)
279 ic->state = INO_STATE_GC;
280 else {
281 D1(printk(KERN_DEBUG "Ino #%u is absent but node not REF_PRISTINE. Reading.\n",
282 ic->ino));
284 break;
286 case INO_STATE_PRESENT:
287 /* It's in-core. GC must iget() it. */
288 break;
290 case INO_STATE_UNCHECKED:
291 case INO_STATE_CHECKING:
292 case INO_STATE_GC:
293 /* Should never happen. We should have finished checking
294 by the time we actually start doing any GC, and since
295 we're holding the alloc_sem, no other garbage collection
296 can happen.
298 printk(KERN_CRIT "Inode #%u already in state %d in jffs2_garbage_collect_pass()!\n",
299 ic->ino, ic->state);
300 up(&c->alloc_sem);
301 spin_unlock(&c->inocache_lock);
302 BUG();
304 case INO_STATE_READING:
305 /* Someone's currently trying to read it. We must wait for
306 them to finish and then go through the full iget() route
307 to do the GC. However, sometimes read_inode() needs to get
308 the alloc_sem() (for marking nodes invalid) so we must
309 drop the alloc_sem before sleeping. */
311 up(&c->alloc_sem);
312 D1(printk(KERN_DEBUG "jffs2_garbage_collect_pass() waiting for ino #%u in state %d\n",
313 ic->ino, ic->state));
314 sleep_on_spinunlock(&c->inocache_wq, &c->inocache_lock);
315 /* And because we dropped the alloc_sem we must start again from the
316 beginning. Ponder chance of livelock here -- we're returning success
317 without actually making any progress.
319 Q: What are the chances that the inode is back in INO_STATE_READING
320 again by the time we next enter this function? And that this happens
321 enough times to cause a real delay?
323 A: Small enough that I don't care :)
325 return 0;
328 /* OK. Now if the inode is in state INO_STATE_GC, we are going to copy the
329 node intact, and we don't have to muck about with the fragtree etc.
330 because we know it's not in-core. If it _was_ in-core, we go through
331 all the iget() crap anyway */
333 if (ic->state == INO_STATE_GC) {
334 spin_unlock(&c->inocache_lock);
336 ret = jffs2_garbage_collect_pristine(c, ic, raw);
338 spin_lock(&c->inocache_lock);
339 ic->state = INO_STATE_CHECKEDABSENT;
340 wake_up(&c->inocache_wq);
342 if (ret != -EBADFD) {
343 spin_unlock(&c->inocache_lock);
344 goto release_sem;
347 /* Fall through if it wanted us to, with inocache_lock held */
350 /* Prevent the fairly unlikely race where the gcblock is
351 entirely obsoleted by the final close of a file which had
352 the only valid nodes in the block, followed by erasure,
353 followed by freeing of the ic because the erased block(s)
354 held _all_ the nodes of that inode.... never been seen but
355 it's vaguely possible. */
357 inum = ic->ino;
358 nlink = ic->nlink;
359 spin_unlock(&c->inocache_lock);
361 f = jffs2_gc_fetch_inode(c, inum, nlink);
362 if (IS_ERR(f))
363 return PTR_ERR(f);
364 if (!f)
365 return 0;
367 ret = jffs2_garbage_collect_live(c, jeb, raw, f);
369 jffs2_gc_release_inode(c, f);
371 release_sem:
372 up(&c->alloc_sem);
374 eraseit_lock:
375 /* If we've finished this block, start it erasing */
376 spin_lock(&c->erase_completion_lock);
378 eraseit:
379 if (c->gcblock && !c->gcblock->used_size) {
380 D1(printk(KERN_DEBUG "Block at 0x%08x completely obsoleted by GC. Moving to erase_pending_list\n", c->gcblock->offset));
381 /* We're GC'ing an empty block? */
382 list_add_tail(&c->gcblock->list, &c->erase_pending_list);
383 c->gcblock = NULL;
384 c->nr_erasing_blocks++;
385 jffs2_erase_pending_trigger(c);
387 spin_unlock(&c->erase_completion_lock);
389 return ret;
392 static int jffs2_garbage_collect_live(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
393 struct jffs2_raw_node_ref *raw, struct jffs2_inode_info *f)
395 struct jffs2_node_frag *frag;
396 struct jffs2_full_dnode *fn = NULL;
397 struct jffs2_full_dirent *fd;
398 uint32_t start = 0, end = 0, nrfrags = 0;
399 int ret = 0;
401 down(&f->sem);
403 /* Now we have the lock for this inode. Check that it's still the one at the head
404 of the list. */
406 spin_lock(&c->erase_completion_lock);
408 if (c->gcblock != jeb) {
409 spin_unlock(&c->erase_completion_lock);
410 D1(printk(KERN_DEBUG "GC block is no longer gcblock. Restart\n"));
411 goto upnout;
413 if (ref_obsolete(raw)) {
414 spin_unlock(&c->erase_completion_lock);
415 D1(printk(KERN_DEBUG "node to be GC'd was obsoleted in the meantime.\n"));
416 /* They'll call again */
417 goto upnout;
419 spin_unlock(&c->erase_completion_lock);
421 /* OK. Looks safe. And nobody can get us now because we have the semaphore. Move the block */
422 if (f->metadata && f->metadata->raw == raw) {
423 fn = f->metadata;
424 ret = jffs2_garbage_collect_metadata(c, jeb, f, fn);
425 goto upnout;
428 /* FIXME. Read node and do lookup? */
429 for (frag = frag_first(&f->fragtree); frag; frag = frag_next(frag)) {
430 if (frag->node && frag->node->raw == raw) {
431 fn = frag->node;
432 end = frag->ofs + frag->size;
433 if (!nrfrags++)
434 start = frag->ofs;
435 if (nrfrags == frag->node->frags)
436 break; /* We've found them all */
439 if (fn) {
440 if (ref_flags(raw) == REF_PRISTINE) {
441 ret = jffs2_garbage_collect_pristine(c, f->inocache, raw);
442 if (!ret) {
443 /* Urgh. Return it sensibly. */
444 frag->node->raw = f->inocache->nodes;
446 if (ret != -EBADFD)
447 goto upnout;
449 /* We found a datanode. Do the GC */
450 if((start >> PAGE_CACHE_SHIFT) < ((end-1) >> PAGE_CACHE_SHIFT)) {
451 /* It crosses a page boundary. Therefore, it must be a hole. */
452 ret = jffs2_garbage_collect_hole(c, jeb, f, fn, start, end);
453 } else {
454 /* It could still be a hole. But we GC the page this way anyway */
455 ret = jffs2_garbage_collect_dnode(c, jeb, f, fn, start, end);
457 goto upnout;
460 /* Wasn't a dnode. Try dirent */
461 for (fd = f->dents; fd; fd=fd->next) {
462 if (fd->raw == raw)
463 break;
466 if (fd && fd->ino) {
467 ret = jffs2_garbage_collect_dirent(c, jeb, f, fd);
468 } else if (fd) {
469 ret = jffs2_garbage_collect_deletion_dirent(c, jeb, f, fd);
470 } else {
471 printk(KERN_WARNING "Raw node at 0x%08x wasn't in node lists for ino #%u\n",
472 ref_offset(raw), f->inocache->ino);
473 if (ref_obsolete(raw)) {
474 printk(KERN_WARNING "But it's obsolete so we don't mind too much\n");
475 } else {
476 ret = -EIO;
479 upnout:
480 up(&f->sem);
482 return ret;
485 static int jffs2_garbage_collect_pristine(struct jffs2_sb_info *c,
486 struct jffs2_inode_cache *ic,
487 struct jffs2_raw_node_ref *raw)
489 union jffs2_node_union *node;
490 struct jffs2_raw_node_ref *nraw;
491 size_t retlen;
492 int ret;
493 uint32_t phys_ofs, alloclen;
494 uint32_t crc, rawlen;
495 int retried = 0;
497 D1(printk(KERN_DEBUG "Going to GC REF_PRISTINE node at 0x%08x\n", ref_offset(raw)));
499 rawlen = ref_totlen(c, c->gcblock, raw);
501 /* Ask for a small amount of space (or the totlen if smaller) because we
502 don't want to force wastage of the end of a block if splitting would
503 work. */
504 ret = jffs2_reserve_space_gc(c, min_t(uint32_t, sizeof(struct jffs2_raw_inode) + JFFS2_MIN_DATA_LEN,
505 rawlen), &phys_ofs, &alloclen);
506 if (ret)
507 return ret;
509 if (alloclen < rawlen) {
510 /* Doesn't fit untouched. We'll go the old route and split it */
511 return -EBADFD;
514 node = kmalloc(rawlen, GFP_KERNEL);
515 if (!node)
516 return -ENOMEM;
518 ret = jffs2_flash_read(c, ref_offset(raw), rawlen, &retlen, (char *)node);
519 if (!ret && retlen != rawlen)
520 ret = -EIO;
521 if (ret)
522 goto out_node;
524 crc = crc32(0, node, sizeof(struct jffs2_unknown_node)-4);
525 if (je32_to_cpu(node->u.hdr_crc) != crc) {
526 #if 0 // mask by Victor Yu. 04-07-2007
527 printk(KERN_WARNING "Header CRC failed on REF_PRISTINE node at 0x%08x: Read 0x%08x, calculated 0x%08x\n",
528 ref_offset(raw), je32_to_cpu(node->u.hdr_crc), crc);
529 #endif
530 goto bail;
533 switch(je16_to_cpu(node->u.nodetype)) {
534 case JFFS2_NODETYPE_INODE:
535 crc = crc32(0, node, sizeof(node->i)-8);
536 if (je32_to_cpu(node->i.node_crc) != crc) {
537 #if 0 // mask by Victor Yu. 04-07-2007
538 printk(KERN_WARNING "Node CRC failed on REF_PRISTINE data node at 0x%08x: Read 0x%08x, calculated 0x%08x\n",
539 ref_offset(raw), je32_to_cpu(node->i.node_crc), crc);
540 #endif
541 goto bail;
544 if (je32_to_cpu(node->i.dsize)) {
545 crc = crc32(0, node->i.data, je32_to_cpu(node->i.csize));
546 if (je32_to_cpu(node->i.data_crc) != crc) {
547 #if 0 // mask by Victor Yu. 04-07-2007
548 printk(KERN_WARNING "Data CRC failed on REF_PRISTINE data node at 0x%08x: Read 0x%08x, calculated 0x%08x\n",
549 ref_offset(raw), je32_to_cpu(node->i.data_crc), crc);
550 #endif
551 goto bail;
554 break;
556 case JFFS2_NODETYPE_DIRENT:
557 crc = crc32(0, node, sizeof(node->d)-8);
558 if (je32_to_cpu(node->d.node_crc) != crc) {
559 #if 0 // mask by Victor Yu. 04-07-2007
560 printk(KERN_WARNING "Node CRC failed on REF_PRISTINE dirent node at 0x%08x: Read 0x%08x, calculated 0x%08x\n",
561 ref_offset(raw), je32_to_cpu(node->d.node_crc), crc);
562 #endif
563 goto bail;
566 if (node->d.nsize) {
567 crc = crc32(0, node->d.name, node->d.nsize);
568 if (je32_to_cpu(node->d.name_crc) != crc) {
569 #if 0 // mask by Victor Yu. 04-07-2007
570 printk(KERN_WARNING "Name CRC failed on REF_PRISTINE dirent ode at 0x%08x: Read 0x%08x, calculated 0x%08x\n",
571 ref_offset(raw), je32_to_cpu(node->d.name_crc), crc);
572 #endif
573 goto bail;
576 break;
577 default:
578 printk(KERN_WARNING "Unknown node type for REF_PRISTINE node at 0x%08x: 0x%04x\n",
579 ref_offset(raw), je16_to_cpu(node->u.nodetype));
580 goto bail;
583 nraw = jffs2_alloc_raw_node_ref();
584 if (!nraw) {
585 ret = -ENOMEM;
586 goto out_node;
589 /* OK, all the CRCs are good; this node can just be copied as-is. */
590 retry:
591 nraw->flash_offset = phys_ofs;
592 nraw->__totlen = rawlen;
593 nraw->next_phys = NULL;
595 ret = jffs2_flash_write(c, phys_ofs, rawlen, &retlen, (char *)node);
597 if (ret || (retlen != rawlen)) {
598 printk(KERN_NOTICE "Write of %d bytes at 0x%08x failed. returned %d, retlen %zd\n",
599 rawlen, phys_ofs, ret, retlen);
600 if (retlen) {
601 /* Doesn't belong to any inode */
602 nraw->next_in_ino = NULL;
604 nraw->flash_offset |= REF_OBSOLETE;
605 jffs2_add_physical_node_ref(c, nraw);
606 jffs2_mark_node_obsolete(c, nraw);
607 } else {
608 printk(KERN_NOTICE "Not marking the space at 0x%08x as dirty because the flash driver returned retlen zero\n", nraw->flash_offset);
609 jffs2_free_raw_node_ref(nraw);
611 if (!retried && (nraw == jffs2_alloc_raw_node_ref())) {
612 /* Try to reallocate space and retry */
613 uint32_t dummy;
614 struct jffs2_eraseblock *jeb = &c->blocks[phys_ofs / c->sector_size];
616 retried = 1;
618 D1(printk(KERN_DEBUG "Retrying failed write of REF_PRISTINE node.\n"));
620 ACCT_SANITY_CHECK(c,jeb);
621 D1(ACCT_PARANOIA_CHECK(jeb));
623 ret = jffs2_reserve_space_gc(c, rawlen, &phys_ofs, &dummy);
625 if (!ret) {
626 D1(printk(KERN_DEBUG "Allocated space at 0x%08x to retry failed write.\n", phys_ofs));
628 ACCT_SANITY_CHECK(c,jeb);
629 D1(ACCT_PARANOIA_CHECK(jeb));
631 goto retry;
633 D1(printk(KERN_DEBUG "Failed to allocate space to retry failed write: %d!\n", ret));
634 jffs2_free_raw_node_ref(nraw);
637 if (!ret)
638 ret = -EIO;
639 goto out_node;
641 nraw->flash_offset |= REF_PRISTINE;
642 jffs2_add_physical_node_ref(c, nraw);
644 /* Link into per-inode list. This is safe because of the ic
645 state being INO_STATE_GC. Note that if we're doing this
646 for an inode which is in-code, the 'nraw' pointer is then
647 going to be fetched from ic->nodes by our caller. */
648 nraw->next_in_ino = ic->nodes;
649 ic->nodes = nraw;
651 jffs2_mark_node_obsolete(c, raw);
652 D1(printk(KERN_DEBUG "WHEEE! GC REF_PRISTINE node at 0x%08x succeeded\n", ref_offset(raw)));
654 out_node:
655 kfree(node);
656 return ret;
657 bail:
658 ret = -EBADFD;
659 goto out_node;
662 static int jffs2_garbage_collect_metadata(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
663 struct jffs2_inode_info *f, struct jffs2_full_dnode *fn)
665 struct jffs2_full_dnode *new_fn;
666 struct jffs2_raw_inode ri;
667 jint16_t dev;
668 char *mdata = NULL, mdatalen = 0;
669 uint32_t alloclen, phys_ofs;
670 int ret;
672 if (S_ISBLK(JFFS2_F_I_MODE(f)) ||
673 S_ISCHR(JFFS2_F_I_MODE(f)) ) {
674 /* For these, we don't actually need to read the old node */
675 /* FIXME: for minor or major > 255. */
676 dev = cpu_to_je16(((JFFS2_F_I_RDEV_MAJ(f) << 8) |
677 JFFS2_F_I_RDEV_MIN(f)));
678 mdata = (char *)&dev;
679 mdatalen = sizeof(dev);
680 D1(printk(KERN_DEBUG "jffs2_garbage_collect_metadata(): Writing %d bytes of kdev_t\n", mdatalen));
681 } else if (S_ISLNK(JFFS2_F_I_MODE(f))) {
682 mdatalen = fn->size;
683 mdata = kmalloc(fn->size, GFP_KERNEL);
684 if (!mdata) {
685 printk(KERN_WARNING "kmalloc of mdata failed in jffs2_garbage_collect_metadata()\n");
686 return -ENOMEM;
688 ret = jffs2_read_dnode(c, f, fn, mdata, 0, mdatalen);
689 if (ret) {
690 printk(KERN_WARNING "read of old metadata failed in jffs2_garbage_collect_metadata(): %d\n", ret);
691 kfree(mdata);
692 return ret;
694 D1(printk(KERN_DEBUG "jffs2_garbage_collect_metadata(): Writing %d bites of symlink target\n", mdatalen));
698 ret = jffs2_reserve_space_gc(c, sizeof(ri) + mdatalen, &phys_ofs, &alloclen);
699 if (ret) {
700 printk(KERN_WARNING "jffs2_reserve_space_gc of %zd bytes for garbage_collect_metadata failed: %d\n",
701 sizeof(ri)+ mdatalen, ret);
702 goto out;
705 memset(&ri, 0, sizeof(ri));
706 ri.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK);
707 ri.nodetype = cpu_to_je16(JFFS2_NODETYPE_INODE);
708 ri.totlen = cpu_to_je32(sizeof(ri) + mdatalen);
709 ri.hdr_crc = cpu_to_je32(crc32(0, &ri, sizeof(struct jffs2_unknown_node)-4));
711 ri.ino = cpu_to_je32(f->inocache->ino);
712 ri.version = cpu_to_je32(++f->highest_version);
713 ri.mode = cpu_to_jemode(JFFS2_F_I_MODE(f));
714 ri.uid = cpu_to_je16(JFFS2_F_I_UID(f));
715 ri.gid = cpu_to_je16(JFFS2_F_I_GID(f));
716 ri.isize = cpu_to_je32(JFFS2_F_I_SIZE(f));
717 ri.atime = cpu_to_je32(JFFS2_F_I_ATIME(f));
718 ri.ctime = cpu_to_je32(JFFS2_F_I_CTIME(f));
719 ri.mtime = cpu_to_je32(JFFS2_F_I_MTIME(f));
720 ri.offset = cpu_to_je32(0);
721 ri.csize = cpu_to_je32(mdatalen);
722 ri.dsize = cpu_to_je32(mdatalen);
723 ri.compr = JFFS2_COMPR_NONE;
724 ri.node_crc = cpu_to_je32(crc32(0, &ri, sizeof(ri)-8));
725 ri.data_crc = cpu_to_je32(crc32(0, mdata, mdatalen));
727 new_fn = jffs2_write_dnode(c, f, &ri, mdata, mdatalen, phys_ofs, ALLOC_GC);
729 if (IS_ERR(new_fn)) {
730 printk(KERN_WARNING "Error writing new dnode: %ld\n", PTR_ERR(new_fn));
731 ret = PTR_ERR(new_fn);
732 goto out;
734 jffs2_mark_node_obsolete(c, fn->raw);
735 jffs2_free_full_dnode(fn);
736 f->metadata = new_fn;
737 out:
738 if (S_ISLNK(JFFS2_F_I_MODE(f)))
739 kfree(mdata);
740 return ret;
743 static int jffs2_garbage_collect_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
744 struct jffs2_inode_info *f, struct jffs2_full_dirent *fd)
746 struct jffs2_full_dirent *new_fd;
747 struct jffs2_raw_dirent rd;
748 uint32_t alloclen, phys_ofs;
749 int ret;
751 rd.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK);
752 rd.nodetype = cpu_to_je16(JFFS2_NODETYPE_DIRENT);
753 rd.nsize = strlen(fd->name);
754 rd.totlen = cpu_to_je32(sizeof(rd) + rd.nsize);
755 rd.hdr_crc = cpu_to_je32(crc32(0, &rd, sizeof(struct jffs2_unknown_node)-4));
757 rd.pino = cpu_to_je32(f->inocache->ino);
758 rd.version = cpu_to_je32(++f->highest_version);
759 rd.ino = cpu_to_je32(fd->ino);
760 rd.mctime = cpu_to_je32(max(JFFS2_F_I_MTIME(f), JFFS2_F_I_CTIME(f)));
761 rd.type = fd->type;
762 rd.node_crc = cpu_to_je32(crc32(0, &rd, sizeof(rd)-8));
763 rd.name_crc = cpu_to_je32(crc32(0, fd->name, rd.nsize));
765 ret = jffs2_reserve_space_gc(c, sizeof(rd)+rd.nsize, &phys_ofs, &alloclen);
766 if (ret) {
767 printk(KERN_WARNING "jffs2_reserve_space_gc of %zd bytes for garbage_collect_dirent failed: %d\n",
768 sizeof(rd)+rd.nsize, ret);
769 return ret;
771 new_fd = jffs2_write_dirent(c, f, &rd, fd->name, rd.nsize, phys_ofs, ALLOC_GC);
773 if (IS_ERR(new_fd)) {
774 printk(KERN_WARNING "jffs2_write_dirent in garbage_collect_dirent failed: %ld\n", PTR_ERR(new_fd));
775 return PTR_ERR(new_fd);
777 jffs2_add_fd_to_list(c, new_fd, &f->dents);
778 return 0;
781 static int jffs2_garbage_collect_deletion_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
782 struct jffs2_inode_info *f, struct jffs2_full_dirent *fd)
784 struct jffs2_full_dirent **fdp = &f->dents;
785 int found = 0;
787 /* On a medium where we can't actually mark nodes obsolete
788 pernamently, such as NAND flash, we need to work out
789 whether this deletion dirent is still needed to actively
790 delete a 'real' dirent with the same name that's still
791 somewhere else on the flash. */
792 if (!jffs2_can_mark_obsolete(c)) {
793 struct jffs2_raw_dirent *rd;
794 struct jffs2_raw_node_ref *raw;
795 int ret;
796 size_t retlen;
797 int name_len = strlen(fd->name);
798 uint32_t name_crc = crc32(0, fd->name, name_len);
799 uint32_t rawlen = ref_totlen(c, jeb, fd->raw);
801 rd = kmalloc(rawlen, GFP_KERNEL);
802 if (!rd)
803 return -ENOMEM;
805 /* Prevent the erase code from nicking the obsolete node refs while
806 we're looking at them. I really don't like this extra lock but
807 can't see any alternative. Suggestions on a postcard to... */
808 down(&c->erase_free_sem);
810 for (raw = f->inocache->nodes; raw != (void *)f->inocache; raw = raw->next_in_ino) {
812 /* We only care about obsolete ones */
813 if (!(ref_obsolete(raw)))
814 continue;
816 /* Any dirent with the same name is going to have the same length... */
817 if (ref_totlen(c, NULL, raw) != rawlen)
818 continue;
820 /* Doesn't matter if there's one in the same erase block. We're going to
821 delete it too at the same time. */
822 if ((raw->flash_offset & ~(c->sector_size-1)) ==
823 (fd->raw->flash_offset & ~(c->sector_size-1)))
824 continue;
826 D1(printk(KERN_DEBUG "Check potential deletion dirent at %08x\n", ref_offset(raw)));
828 /* This is an obsolete node belonging to the same directory, and it's of the right
829 length. We need to take a closer look...*/
830 ret = jffs2_flash_read(c, ref_offset(raw), rawlen, &retlen, (char *)rd);
831 if (ret) {
832 printk(KERN_WARNING "jffs2_g_c_deletion_dirent(): Read error (%d) reading obsolete node at %08x\n", ret, ref_offset(raw));
833 /* If we can't read it, we don't need to continue to obsolete it. Continue */
834 continue;
836 if (retlen != rawlen) {
837 printk(KERN_WARNING "jffs2_g_c_deletion_dirent(): Short read (%zd not %u) reading header from obsolete node at %08x\n",
838 retlen, rawlen, ref_offset(raw));
839 continue;
842 if (je16_to_cpu(rd->nodetype) != JFFS2_NODETYPE_DIRENT)
843 continue;
845 /* If the name CRC doesn't match, skip */
846 if (je32_to_cpu(rd->name_crc) != name_crc)
847 continue;
849 /* If the name length doesn't match, or it's another deletion dirent, skip */
850 if (rd->nsize != name_len || !je32_to_cpu(rd->ino))
851 continue;
853 /* OK, check the actual name now */
854 if (memcmp(rd->name, fd->name, name_len))
855 continue;
857 /* OK. The name really does match. There really is still an older node on
858 the flash which our deletion dirent obsoletes. So we have to write out
859 a new deletion dirent to replace it */
860 up(&c->erase_free_sem);
862 D1(printk(KERN_DEBUG "Deletion dirent at %08x still obsoletes real dirent \"%s\" at %08x for ino #%u\n",
863 ref_offset(fd->raw), fd->name, ref_offset(raw), je32_to_cpu(rd->ino)));
864 kfree(rd);
866 return jffs2_garbage_collect_dirent(c, jeb, f, fd);
869 up(&c->erase_free_sem);
870 kfree(rd);
873 /* No need for it any more. Just mark it obsolete and remove it from the list */
874 while (*fdp) {
875 if ((*fdp) == fd) {
876 found = 1;
877 *fdp = fd->next;
878 break;
880 fdp = &(*fdp)->next;
882 if (!found) {
883 printk(KERN_WARNING "Deletion dirent \"%s\" not found in list for ino #%u\n", fd->name, f->inocache->ino);
885 jffs2_mark_node_obsolete(c, fd->raw);
886 jffs2_free_full_dirent(fd);
887 return 0;
890 static int jffs2_garbage_collect_hole(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
891 struct jffs2_inode_info *f, struct jffs2_full_dnode *fn,
892 uint32_t start, uint32_t end)
894 struct jffs2_raw_inode ri;
895 struct jffs2_node_frag *frag;
896 struct jffs2_full_dnode *new_fn;
897 uint32_t alloclen, phys_ofs;
898 int ret;
900 D1(printk(KERN_DEBUG "Writing replacement hole node for ino #%u from offset 0x%x to 0x%x\n",
901 f->inocache->ino, start, end));
903 memset(&ri, 0, sizeof(ri));
905 if(fn->frags > 1) {
906 size_t readlen;
907 uint32_t crc;
908 /* It's partially obsoleted by a later write. So we have to
909 write it out again with the _same_ version as before */
910 ret = jffs2_flash_read(c, ref_offset(fn->raw), sizeof(ri), &readlen, (char *)&ri);
911 if (readlen != sizeof(ri) || ret) {
912 printk(KERN_WARNING "Node read failed in jffs2_garbage_collect_hole. Ret %d, retlen %zd. Data will be lost by writing new hole node\n", ret, readlen);
913 goto fill;
915 if (je16_to_cpu(ri.nodetype) != JFFS2_NODETYPE_INODE) {
916 printk(KERN_WARNING "jffs2_garbage_collect_hole: Node at 0x%08x had node type 0x%04x instead of JFFS2_NODETYPE_INODE(0x%04x)\n",
917 ref_offset(fn->raw),
918 je16_to_cpu(ri.nodetype), JFFS2_NODETYPE_INODE);
919 return -EIO;
921 if (je32_to_cpu(ri.totlen) != sizeof(ri)) {
922 printk(KERN_WARNING "jffs2_garbage_collect_hole: Node at 0x%08x had totlen 0x%x instead of expected 0x%zx\n",
923 ref_offset(fn->raw),
924 je32_to_cpu(ri.totlen), sizeof(ri));
925 return -EIO;
927 crc = crc32(0, &ri, sizeof(ri)-8);
928 if (crc != je32_to_cpu(ri.node_crc)) {
929 #if 0 // mask by Victor Yu. 04-07-2007
930 printk(KERN_WARNING "jffs2_garbage_collect_hole: Node at 0x%08x had CRC 0x%08x which doesn't match calculated CRC 0x%08x\n",
931 ref_offset(fn->raw),
932 je32_to_cpu(ri.node_crc), crc);
933 /* FIXME: We could possibly deal with this by writing new holes for each frag */
934 printk(KERN_WARNING "Data in the range 0x%08x to 0x%08x of inode #%u will be lost\n",
935 start, end, f->inocache->ino);
936 #endif
937 goto fill;
939 if (ri.compr != JFFS2_COMPR_ZERO) {
940 printk(KERN_WARNING "jffs2_garbage_collect_hole: Node 0x%08x wasn't a hole node!\n", ref_offset(fn->raw));
941 printk(KERN_WARNING "Data in the range 0x%08x to 0x%08x of inode #%u will be lost\n",
942 start, end, f->inocache->ino);
943 goto fill;
945 } else {
946 fill:
947 ri.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK);
948 ri.nodetype = cpu_to_je16(JFFS2_NODETYPE_INODE);
949 ri.totlen = cpu_to_je32(sizeof(ri));
950 ri.hdr_crc = cpu_to_je32(crc32(0, &ri, sizeof(struct jffs2_unknown_node)-4));
952 ri.ino = cpu_to_je32(f->inocache->ino);
953 ri.version = cpu_to_je32(++f->highest_version);
954 ri.offset = cpu_to_je32(start);
955 ri.dsize = cpu_to_je32(end - start);
956 ri.csize = cpu_to_je32(0);
957 ri.compr = JFFS2_COMPR_ZERO;
959 ri.mode = cpu_to_jemode(JFFS2_F_I_MODE(f));
960 ri.uid = cpu_to_je16(JFFS2_F_I_UID(f));
961 ri.gid = cpu_to_je16(JFFS2_F_I_GID(f));
962 ri.isize = cpu_to_je32(JFFS2_F_I_SIZE(f));
963 ri.atime = cpu_to_je32(JFFS2_F_I_ATIME(f));
964 ri.ctime = cpu_to_je32(JFFS2_F_I_CTIME(f));
965 ri.mtime = cpu_to_je32(JFFS2_F_I_MTIME(f));
966 ri.data_crc = cpu_to_je32(0);
967 ri.node_crc = cpu_to_je32(crc32(0, &ri, sizeof(ri)-8));
969 ret = jffs2_reserve_space_gc(c, sizeof(ri), &phys_ofs, &alloclen);
970 if (ret) {
971 printk(KERN_WARNING "jffs2_reserve_space_gc of %zd bytes for garbage_collect_hole failed: %d\n",
972 sizeof(ri), ret);
973 return ret;
975 new_fn = jffs2_write_dnode(c, f, &ri, NULL, 0, phys_ofs, ALLOC_GC);
977 if (IS_ERR(new_fn)) {
978 printk(KERN_WARNING "Error writing new hole node: %ld\n", PTR_ERR(new_fn));
979 return PTR_ERR(new_fn);
981 if (je32_to_cpu(ri.version) == f->highest_version) {
982 jffs2_add_full_dnode_to_inode(c, f, new_fn);
983 if (f->metadata) {
984 jffs2_mark_node_obsolete(c, f->metadata->raw);
985 jffs2_free_full_dnode(f->metadata);
986 f->metadata = NULL;
988 return 0;
992 * We should only get here in the case where the node we are
993 * replacing had more than one frag, so we kept the same version
994 * number as before. (Except in case of error -- see 'goto fill;'
995 * above.)
997 D1(if(unlikely(fn->frags <= 1)) {
998 printk(KERN_WARNING "jffs2_garbage_collect_hole: Replacing fn with %d frag(s) but new ver %d != highest_version %d of ino #%d\n",
999 fn->frags, je32_to_cpu(ri.version), f->highest_version,
1000 je32_to_cpu(ri.ino));
1003 /* This is a partially-overlapped hole node. Mark it REF_NORMAL not REF_PRISTINE */
1004 mark_ref_normal(new_fn->raw);
1006 for (frag = jffs2_lookup_node_frag(&f->fragtree, fn->ofs);
1007 frag; frag = frag_next(frag)) {
1008 if (frag->ofs > fn->size + fn->ofs)
1009 break;
1010 if (frag->node == fn) {
1011 frag->node = new_fn;
1012 new_fn->frags++;
1013 fn->frags--;
1016 if (fn->frags) {
1017 printk(KERN_WARNING "jffs2_garbage_collect_hole: Old node still has frags!\n");
1018 BUG();
1020 if (!new_fn->frags) {
1021 printk(KERN_WARNING "jffs2_garbage_collect_hole: New node has no frags!\n");
1022 BUG();
1025 jffs2_mark_node_obsolete(c, fn->raw);
1026 jffs2_free_full_dnode(fn);
1028 return 0;
1031 static int jffs2_garbage_collect_dnode(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
1032 struct jffs2_inode_info *f, struct jffs2_full_dnode *fn,
1033 uint32_t start, uint32_t end)
1035 struct jffs2_full_dnode *new_fn;
1036 struct jffs2_raw_inode ri;
1037 uint32_t alloclen, phys_ofs, offset, orig_end, orig_start;
1038 int ret = 0;
1039 unsigned char *comprbuf = NULL, *writebuf;
1040 unsigned long pg;
1041 unsigned char *pg_ptr;
1043 memset(&ri, 0, sizeof(ri));
1045 D1(printk(KERN_DEBUG "Writing replacement dnode for ino #%u from offset 0x%x to 0x%x\n",
1046 f->inocache->ino, start, end));
1048 orig_end = end;
1049 orig_start = start;
1051 if (c->nr_free_blocks + c->nr_erasing_blocks > c->resv_blocks_gcmerge) {
1052 /* Attempt to do some merging. But only expand to cover logically
1053 adjacent frags if the block containing them is already considered
1054 to be dirty. Otherwise we end up with GC just going round in
1055 circles dirtying the nodes it already wrote out, especially
1056 on NAND where we have small eraseblocks and hence a much higher
1057 chance of nodes having to be split to cross boundaries. */
1059 struct jffs2_node_frag *frag;
1060 uint32_t min, max;
1062 min = start & ~(PAGE_CACHE_SIZE-1);
1063 max = min + PAGE_CACHE_SIZE;
1065 frag = jffs2_lookup_node_frag(&f->fragtree, start);
1067 /* BUG_ON(!frag) but that'll happen anyway... */
1069 BUG_ON(frag->ofs != start);
1071 /* First grow down... */
1072 while((frag = frag_prev(frag)) && frag->ofs >= min) {
1074 /* If the previous frag doesn't even reach the beginning, there's
1075 excessive fragmentation. Just merge. */
1076 if (frag->ofs > min) {
1077 D1(printk(KERN_DEBUG "Expanding down to cover partial frag (0x%x-0x%x)\n",
1078 frag->ofs, frag->ofs+frag->size));
1079 start = frag->ofs;
1080 continue;
1082 /* OK. This frag holds the first byte of the page. */
1083 if (!frag->node || !frag->node->raw) {
1084 D1(printk(KERN_DEBUG "First frag in page is hole (0x%x-0x%x). Not expanding down.\n",
1085 frag->ofs, frag->ofs+frag->size));
1086 break;
1087 } else {
1089 /* OK, it's a frag which extends to the beginning of the page. Does it live
1090 in a block which is still considered clean? If so, don't obsolete it.
1091 If not, cover it anyway. */
1093 struct jffs2_raw_node_ref *raw = frag->node->raw;
1094 struct jffs2_eraseblock *jeb;
1096 jeb = &c->blocks[raw->flash_offset / c->sector_size];
1098 if (jeb == c->gcblock) {
1099 D1(printk(KERN_DEBUG "Expanding down to cover frag (0x%x-0x%x) in gcblock at %08x\n",
1100 frag->ofs, frag->ofs+frag->size, ref_offset(raw)));
1101 start = frag->ofs;
1102 break;
1104 if (!ISDIRTY(jeb->dirty_size + jeb->wasted_size)) {
1105 D1(printk(KERN_DEBUG "Not expanding down to cover frag (0x%x-0x%x) in clean block %08x\n",
1106 frag->ofs, frag->ofs+frag->size, jeb->offset));
1107 break;
1110 D1(printk(KERN_DEBUG "Expanding down to cover frag (0x%x-0x%x) in dirty block %08x\n",
1111 frag->ofs, frag->ofs+frag->size, jeb->offset));
1112 start = frag->ofs;
1113 break;
1117 /* ... then up */
1119 /* Find last frag which is actually part of the node we're to GC. */
1120 frag = jffs2_lookup_node_frag(&f->fragtree, end-1);
1122 while((frag = frag_next(frag)) && frag->ofs+frag->size <= max) {
1124 /* If the previous frag doesn't even reach the beginning, there's lots
1125 of fragmentation. Just merge. */
1126 if (frag->ofs+frag->size < max) {
1127 D1(printk(KERN_DEBUG "Expanding up to cover partial frag (0x%x-0x%x)\n",
1128 frag->ofs, frag->ofs+frag->size));
1129 end = frag->ofs + frag->size;
1130 continue;
1133 if (!frag->node || !frag->node->raw) {
1134 D1(printk(KERN_DEBUG "Last frag in page is hole (0x%x-0x%x). Not expanding up.\n",
1135 frag->ofs, frag->ofs+frag->size));
1136 break;
1137 } else {
1139 /* OK, it's a frag which extends to the beginning of the page. Does it live
1140 in a block which is still considered clean? If so, don't obsolete it.
1141 If not, cover it anyway. */
1143 struct jffs2_raw_node_ref *raw = frag->node->raw;
1144 struct jffs2_eraseblock *jeb;
1146 jeb = &c->blocks[raw->flash_offset / c->sector_size];
1148 if (jeb == c->gcblock) {
1149 D1(printk(KERN_DEBUG "Expanding up to cover frag (0x%x-0x%x) in gcblock at %08x\n",
1150 frag->ofs, frag->ofs+frag->size, ref_offset(raw)));
1151 end = frag->ofs + frag->size;
1152 break;
1154 if (!ISDIRTY(jeb->dirty_size + jeb->wasted_size)) {
1155 D1(printk(KERN_DEBUG "Not expanding up to cover frag (0x%x-0x%x) in clean block %08x\n",
1156 frag->ofs, frag->ofs+frag->size, jeb->offset));
1157 break;
1160 D1(printk(KERN_DEBUG "Expanding up to cover frag (0x%x-0x%x) in dirty block %08x\n",
1161 frag->ofs, frag->ofs+frag->size, jeb->offset));
1162 end = frag->ofs + frag->size;
1163 break;
1166 D1(printk(KERN_DEBUG "Expanded dnode to write from (0x%x-0x%x) to (0x%x-0x%x)\n",
1167 orig_start, orig_end, start, end));
1169 BUG_ON(end > JFFS2_F_I_SIZE(f));
1170 BUG_ON(end < orig_end);
1171 BUG_ON(start > orig_start);
1174 /* First, use readpage() to read the appropriate page into the page cache */
1175 /* Q: What happens if we actually try to GC the _same_ page for which commit_write()
1176 * triggered garbage collection in the first place?
1177 * A: I _think_ it's OK. read_cache_page shouldn't deadlock, we'll write out the
1178 * page OK. We'll actually write it out again in commit_write, which is a little
1179 * suboptimal, but at least we're correct.
1181 pg_ptr = jffs2_gc_fetch_page(c, f, start, &pg);
1183 if (IS_ERR(pg_ptr)) {
1184 printk(KERN_WARNING "read_cache_page() returned error: %ld\n", PTR_ERR(pg_ptr));
1185 return PTR_ERR(pg_ptr);
1188 offset = start;
1189 while(offset < orig_end) {
1190 uint32_t datalen;
1191 uint32_t cdatalen;
1192 uint16_t comprtype = JFFS2_COMPR_NONE;
1194 ret = jffs2_reserve_space_gc(c, sizeof(ri) + JFFS2_MIN_DATA_LEN, &phys_ofs, &alloclen);
1196 if (ret) {
1197 printk(KERN_WARNING "jffs2_reserve_space_gc of %zd bytes for garbage_collect_dnode failed: %d\n",
1198 sizeof(ri)+ JFFS2_MIN_DATA_LEN, ret);
1199 break;
1201 cdatalen = min_t(uint32_t, alloclen - sizeof(ri), end - offset);
1202 datalen = end - offset;
1204 writebuf = pg_ptr + (offset & (PAGE_CACHE_SIZE -1));
1206 comprtype = jffs2_compress(c, f, writebuf, &comprbuf, &datalen, &cdatalen);
1208 ri.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK);
1209 ri.nodetype = cpu_to_je16(JFFS2_NODETYPE_INODE);
1210 ri.totlen = cpu_to_je32(sizeof(ri) + cdatalen);
1211 ri.hdr_crc = cpu_to_je32(crc32(0, &ri, sizeof(struct jffs2_unknown_node)-4));
1213 ri.ino = cpu_to_je32(f->inocache->ino);
1214 ri.version = cpu_to_je32(++f->highest_version);
1215 ri.mode = cpu_to_jemode(JFFS2_F_I_MODE(f));
1216 ri.uid = cpu_to_je16(JFFS2_F_I_UID(f));
1217 ri.gid = cpu_to_je16(JFFS2_F_I_GID(f));
1218 ri.isize = cpu_to_je32(JFFS2_F_I_SIZE(f));
1219 ri.atime = cpu_to_je32(JFFS2_F_I_ATIME(f));
1220 ri.ctime = cpu_to_je32(JFFS2_F_I_CTIME(f));
1221 ri.mtime = cpu_to_je32(JFFS2_F_I_MTIME(f));
1222 ri.offset = cpu_to_je32(offset);
1223 ri.csize = cpu_to_je32(cdatalen);
1224 ri.dsize = cpu_to_je32(datalen);
1225 ri.compr = comprtype & 0xff;
1226 ri.usercompr = (comprtype >> 8) & 0xff;
1227 ri.node_crc = cpu_to_je32(crc32(0, &ri, sizeof(ri)-8));
1228 ri.data_crc = cpu_to_je32(crc32(0, comprbuf, cdatalen));
1230 new_fn = jffs2_write_dnode(c, f, &ri, comprbuf, cdatalen, phys_ofs, ALLOC_GC);
1232 jffs2_free_comprbuf(comprbuf, writebuf);
1234 if (IS_ERR(new_fn)) {
1235 printk(KERN_WARNING "Error writing new dnode: %ld\n", PTR_ERR(new_fn));
1236 ret = PTR_ERR(new_fn);
1237 break;
1239 ret = jffs2_add_full_dnode_to_inode(c, f, new_fn);
1240 offset += datalen;
1241 if (f->metadata) {
1242 jffs2_mark_node_obsolete(c, f->metadata->raw);
1243 jffs2_free_full_dnode(f->metadata);
1244 f->metadata = NULL;
1248 jffs2_gc_release_page(c, pg_ptr, &pg);
1249 return ret;