blob: 5a2dec2b064c945aba23acac93dcee2820019d68 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * JFFS2 -- Journalling Flash File System, Version 2.
3 *
David Woodhousec00c3102007-04-25 14:16:47 +01004 * Copyright © 2001-2007 Red Hat, Inc.
David Woodhouse6088c052010-08-08 14:15:22 +01005 * Copyright © 2004-2010 David Woodhouse <dwmw2@infradead.org>
Linus Torvalds1da177e2005-04-16 15:20:36 -07006 *
7 * Created by David Woodhouse <dwmw2@infradead.org>
8 *
9 * For licensing information, see the file 'LICENCE' in this directory.
10 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 */
12
Joe Perches5a528952012-02-15 15:56:45 -080013#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/kernel.h>
16#include <linux/mtd/mtd.h>
17#include <linux/slab.h>
18#include <linux/pagemap.h>
19#include <linux/crc32.h>
20#include <linux/compiler.h>
21#include <linux/stat.h>
22#include "nodelist.h"
23#include "compr.h"
24
Thomas Gleixner182ec4e2005-11-07 11:16:07 +000025static int jffs2_garbage_collect_pristine(struct jffs2_sb_info *c,
Linus Torvalds1da177e2005-04-16 15:20:36 -070026 struct jffs2_inode_cache *ic,
27 struct jffs2_raw_node_ref *raw);
Thomas Gleixner182ec4e2005-11-07 11:16:07 +000028static int jffs2_garbage_collect_metadata(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
Linus Torvalds1da177e2005-04-16 15:20:36 -070029 struct jffs2_inode_info *f, struct jffs2_full_dnode *fd);
Thomas Gleixner182ec4e2005-11-07 11:16:07 +000030static int jffs2_garbage_collect_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
Linus Torvalds1da177e2005-04-16 15:20:36 -070031 struct jffs2_inode_info *f, struct jffs2_full_dirent *fd);
Thomas Gleixner182ec4e2005-11-07 11:16:07 +000032static int jffs2_garbage_collect_deletion_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
Linus Torvalds1da177e2005-04-16 15:20:36 -070033 struct jffs2_inode_info *f, struct jffs2_full_dirent *fd);
34static int jffs2_garbage_collect_hole(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
35 struct jffs2_inode_info *f, struct jffs2_full_dnode *fn,
36 uint32_t start, uint32_t end);
37static int jffs2_garbage_collect_dnode(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
38 struct jffs2_inode_info *f, struct jffs2_full_dnode *fn,
39 uint32_t start, uint32_t end);
40static int jffs2_garbage_collect_live(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
41 struct jffs2_raw_node_ref *raw, struct jffs2_inode_info *f);
42
43/* Called with erase_completion_lock held */
44static struct jffs2_eraseblock *jffs2_find_gc_block(struct jffs2_sb_info *c)
45{
46 struct jffs2_eraseblock *ret;
47 struct list_head *nextlist = NULL;
48 int n = jiffies % 128;
49
50 /* Pick an eraseblock to garbage collect next. This is where we'll
51 put the clever wear-levelling algorithms. Eventually. */
52 /* We possibly want to favour the dirtier blocks more when the
53 number of free blocks is low. */
Artem B. Bityuckiya42163d2005-03-20 17:45:29 +000054again:
Linus Torvalds1da177e2005-04-16 15:20:36 -070055 if (!list_empty(&c->bad_used_list) && c->nr_free_blocks > c->resv_blocks_gcbad) {
Joe Perches9c261b32012-02-15 15:56:43 -080056 jffs2_dbg(1, "Picking block from bad_used_list to GC next\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -070057 nextlist = &c->bad_used_list;
58 } else if (n < 50 && !list_empty(&c->erasable_list)) {
Thomas Gleixner182ec4e2005-11-07 11:16:07 +000059 /* Note that most of them will have gone directly to be erased.
Linus Torvalds1da177e2005-04-16 15:20:36 -070060 So don't favour the erasable_list _too_ much. */
Joe Perches9c261b32012-02-15 15:56:43 -080061 jffs2_dbg(1, "Picking block from erasable_list to GC next\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 nextlist = &c->erasable_list;
63 } else if (n < 110 && !list_empty(&c->very_dirty_list)) {
64 /* Most of the time, pick one off the very_dirty list */
Joe Perches9c261b32012-02-15 15:56:43 -080065 jffs2_dbg(1, "Picking block from very_dirty_list to GC next\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 nextlist = &c->very_dirty_list;
67 } else if (n < 126 && !list_empty(&c->dirty_list)) {
Joe Perches9c261b32012-02-15 15:56:43 -080068 jffs2_dbg(1, "Picking block from dirty_list to GC next\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 nextlist = &c->dirty_list;
70 } else if (!list_empty(&c->clean_list)) {
Joe Perches9c261b32012-02-15 15:56:43 -080071 jffs2_dbg(1, "Picking block from clean_list to GC next\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 nextlist = &c->clean_list;
73 } else if (!list_empty(&c->dirty_list)) {
Joe Perches9c261b32012-02-15 15:56:43 -080074 jffs2_dbg(1, "Picking block from dirty_list to GC next (clean_list was empty)\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -070075
76 nextlist = &c->dirty_list;
77 } else if (!list_empty(&c->very_dirty_list)) {
Joe Perches9c261b32012-02-15 15:56:43 -080078 jffs2_dbg(1, "Picking block from very_dirty_list to GC next (clean_list and dirty_list were empty)\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 nextlist = &c->very_dirty_list;
80 } else if (!list_empty(&c->erasable_list)) {
Joe Perches9c261b32012-02-15 15:56:43 -080081 jffs2_dbg(1, "Picking block from erasable_list to GC next (clean_list and {very_,}dirty_list were empty)\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -070082
83 nextlist = &c->erasable_list;
Artem B. Bityuckiya42163d2005-03-20 17:45:29 +000084 } else if (!list_empty(&c->erasable_pending_wbuf_list)) {
85 /* There are blocks are wating for the wbuf sync */
Joe Perches9c261b32012-02-15 15:56:43 -080086 jffs2_dbg(1, "Synching wbuf in order to reuse erasable_pending_wbuf_list blocks\n");
Artem B. Bityuckiy3cceb9f2005-03-20 21:43:26 +000087 spin_unlock(&c->erase_completion_lock);
Artem B. Bityuckiya42163d2005-03-20 17:45:29 +000088 jffs2_flush_wbuf_pad(c);
Artem B. Bityuckiy3cceb9f2005-03-20 21:43:26 +000089 spin_lock(&c->erase_completion_lock);
Artem B. Bityuckiya42163d2005-03-20 17:45:29 +000090 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 } else {
92 /* Eep. All were empty */
Joe Perches5a528952012-02-15 15:56:45 -080093 jffs2_dbg(1, "No clean, dirty _or_ erasable blocks to GC from! Where are they all?\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 return NULL;
95 }
96
97 ret = list_entry(nextlist->next, struct jffs2_eraseblock, list);
98 list_del(&ret->list);
99 c->gcblock = ret;
100 ret->gc_node = ret->first_node;
101 if (!ret->gc_node) {
Joe Perchesda320f02012-02-15 15:56:44 -0800102 pr_warn("Eep. ret->gc_node for block at 0x%08x is NULL\n",
103 ret->offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 BUG();
105 }
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000106
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 /* Have we accidentally picked a clean block with wasted space ? */
108 if (ret->wasted_size) {
Joe Perches9c261b32012-02-15 15:56:43 -0800109 jffs2_dbg(1, "Converting wasted_size %08x to dirty_size\n",
110 ret->wasted_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 ret->dirty_size += ret->wasted_size;
112 c->wasted_size -= ret->wasted_size;
113 c->dirty_size += ret->wasted_size;
114 ret->wasted_size = 0;
115 }
116
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 return ret;
118}
119
120/* jffs2_garbage_collect_pass
121 * Make a single attempt to progress GC. Move one node, and possibly
122 * start erasing one eraseblock.
123 */
124int jffs2_garbage_collect_pass(struct jffs2_sb_info *c)
125{
126 struct jffs2_inode_info *f;
127 struct jffs2_inode_cache *ic;
128 struct jffs2_eraseblock *jeb;
129 struct jffs2_raw_node_ref *raw;
David Woodhouse2665ea82007-10-13 11:31:23 +0100130 uint32_t gcblock_dirty;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 int ret = 0, inum, nlink;
KaiGai Koheiaa98d7c2006-05-13 15:09:47 +0900132 int xattr = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133
David Woodhouseced22072008-04-22 15:13:40 +0100134 if (mutex_lock_interruptible(&c->alloc_sem))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135 return -EINTR;
136
137 for (;;) {
138 spin_lock(&c->erase_completion_lock);
139 if (!c->unchecked_size)
140 break;
141
142 /* We can't start doing GC yet. We haven't finished checking
143 the node CRCs etc. Do it now. */
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000144
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 /* checked_ino is protected by the alloc_sem */
KaiGai Koheiaa98d7c2006-05-13 15:09:47 +0900146 if (c->checked_ino > c->highest_ino && xattr) {
Joe Perchesda320f02012-02-15 15:56:44 -0800147 pr_crit("Checked all inodes but still 0x%x bytes of unchecked space?\n",
148 c->unchecked_size);
Artem B. Bityutskiye0c8e422005-07-24 16:14:17 +0100149 jffs2_dbg_dump_block_lists_nolock(c);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 spin_unlock(&c->erase_completion_lock);
David Woodhouseced22072008-04-22 15:13:40 +0100151 mutex_unlock(&c->alloc_sem);
David Woodhouse44b998e2007-04-23 12:11:46 +0100152 return -ENOSPC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 }
154
155 spin_unlock(&c->erase_completion_lock);
156
KaiGai Koheiaa98d7c2006-05-13 15:09:47 +0900157 if (!xattr)
158 xattr = jffs2_verify_xattr(c);
159
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 spin_lock(&c->inocache_lock);
161
162 ic = jffs2_get_ino_cache(c, c->checked_ino++);
163
164 if (!ic) {
165 spin_unlock(&c->inocache_lock);
166 continue;
167 }
168
David Woodhouse27c72b02008-05-01 18:47:17 +0100169 if (!ic->pino_nlink) {
Joe Perches9c261b32012-02-15 15:56:43 -0800170 jffs2_dbg(1, "Skipping check of ino #%d with nlink/pino zero\n",
171 ic->ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 spin_unlock(&c->inocache_lock);
KaiGai Kohei355ed4e2006-06-24 09:15:36 +0900173 jffs2_xattr_delete_inode(c, ic);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 continue;
175 }
176 switch(ic->state) {
177 case INO_STATE_CHECKEDABSENT:
178 case INO_STATE_PRESENT:
Joe Perches9c261b32012-02-15 15:56:43 -0800179 jffs2_dbg(1, "Skipping ino #%u already checked\n",
180 ic->ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 spin_unlock(&c->inocache_lock);
182 continue;
183
184 case INO_STATE_GC:
185 case INO_STATE_CHECKING:
Joe Perchesda320f02012-02-15 15:56:44 -0800186 pr_warn("Inode #%u is in state %d during CRC check phase!\n",
187 ic->ino, ic->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 spin_unlock(&c->inocache_lock);
189 BUG();
190
191 case INO_STATE_READING:
192 /* We need to wait for it to finish, lest we move on
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000193 and trigger the BUG() above while we haven't yet
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 finished checking all its nodes */
Joe Perches9c261b32012-02-15 15:56:43 -0800195 jffs2_dbg(1, "Waiting for ino #%u to finish reading\n",
196 ic->ino);
David Woodhoused96fb992006-04-17 00:19:48 +0100197 /* We need to come back again for the _same_ inode. We've
198 made no progress in this case, but that should be OK */
199 c->checked_ino--;
200
David Woodhouseced22072008-04-22 15:13:40 +0100201 mutex_unlock(&c->alloc_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 sleep_on_spinunlock(&c->inocache_wq, &c->inocache_lock);
203 return 0;
204
205 default:
206 BUG();
207
208 case INO_STATE_UNCHECKED:
209 ;
210 }
211 ic->state = INO_STATE_CHECKING;
212 spin_unlock(&c->inocache_lock);
213
Joe Perches9c261b32012-02-15 15:56:43 -0800214 jffs2_dbg(1, "%s(): triggering inode scan of ino#%u\n",
215 __func__, ic->ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216
217 ret = jffs2_do_crccheck_inode(c, ic);
218 if (ret)
Joe Perchesda320f02012-02-15 15:56:44 -0800219 pr_warn("Returned error for crccheck of ino #%u. Expect badness...\n",
220 ic->ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221
222 jffs2_set_inocache_state(c, ic, INO_STATE_CHECKEDABSENT);
David Woodhouseced22072008-04-22 15:13:40 +0100223 mutex_unlock(&c->alloc_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 return ret;
225 }
226
David Woodhouse0717bf82010-05-19 16:37:13 +0100227 /* If there are any blocks which need erasing, erase them now */
228 if (!list_empty(&c->erase_complete_list) ||
229 !list_empty(&c->erase_pending_list)) {
230 spin_unlock(&c->erase_completion_lock);
Joakim Tjernlund81cfc9f2010-10-07 18:01:44 +0200231 mutex_unlock(&c->alloc_sem);
Joe Perches9c261b32012-02-15 15:56:43 -0800232 jffs2_dbg(1, "%s(): erasing pending blocks\n", __func__);
Joakim Tjernlund81cfc9f2010-10-07 18:01:44 +0200233 if (jffs2_erase_pending_blocks(c, 1))
David Woodhouse0717bf82010-05-19 16:37:13 +0100234 return 0;
Joakim Tjernlund81cfc9f2010-10-07 18:01:44 +0200235
Joe Perches9c261b32012-02-15 15:56:43 -0800236 jffs2_dbg(1, "No progress from erasing block; doing GC anyway\n");
Joakim Tjernlund81cfc9f2010-10-07 18:01:44 +0200237 mutex_lock(&c->alloc_sem);
Josh Cartwright226bb7d2012-03-29 19:34:53 -0400238 spin_lock(&c->erase_completion_lock);
David Woodhouse0717bf82010-05-19 16:37:13 +0100239 }
240
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 /* First, work out which block we're garbage-collecting */
242 jeb = c->gcblock;
243
244 if (!jeb)
245 jeb = jffs2_find_gc_block(c);
246
247 if (!jeb) {
David Woodhouse422b1202008-04-23 15:40:52 +0100248 /* Couldn't find a free block. But maybe we can just erase one and make 'progress'? */
David Woodhouse0717bf82010-05-19 16:37:13 +0100249 if (c->nr_erasing_blocks) {
David Woodhouse422b1202008-04-23 15:40:52 +0100250 spin_unlock(&c->erase_completion_lock);
251 mutex_unlock(&c->alloc_sem);
252 return -EAGAIN;
253 }
Joe Perches5a528952012-02-15 15:56:45 -0800254 jffs2_dbg(1, "Couldn't find erase block to garbage collect!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 spin_unlock(&c->erase_completion_lock);
David Woodhouseced22072008-04-22 15:13:40 +0100256 mutex_unlock(&c->alloc_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 return -EIO;
258 }
259
Joe Perches9c261b32012-02-15 15:56:43 -0800260 jffs2_dbg(1, "GC from block %08x, used_size %08x, dirty_size %08x, free_size %08x\n",
261 jeb->offset, jeb->used_size, jeb->dirty_size, jeb->free_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 D1(if (c->nextblock)
263 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));
264
265 if (!jeb->used_size) {
David Woodhouseced22072008-04-22 15:13:40 +0100266 mutex_unlock(&c->alloc_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 goto eraseit;
268 }
269
270 raw = jeb->gc_node;
David Woodhouse2665ea82007-10-13 11:31:23 +0100271 gcblock_dirty = jeb->dirty_size;
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000272
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 while(ref_obsolete(raw)) {
Joe Perches9c261b32012-02-15 15:56:43 -0800274 jffs2_dbg(1, "Node at 0x%08x is obsolete... skipping\n",
275 ref_offset(raw));
David Woodhouse99988f72006-05-24 09:04:17 +0100276 raw = ref_next(raw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 if (unlikely(!raw)) {
Joe Perchesda320f02012-02-15 15:56:44 -0800278 pr_warn("eep. End of raw list while still supposedly nodes to GC\n");
279 pr_warn("erase block at 0x%08x. free_size 0x%08x, dirty_size 0x%08x, used_size 0x%08x\n",
280 jeb->offset, jeb->free_size,
281 jeb->dirty_size, jeb->used_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 jeb->gc_node = raw;
283 spin_unlock(&c->erase_completion_lock);
David Woodhouseced22072008-04-22 15:13:40 +0100284 mutex_unlock(&c->alloc_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 BUG();
286 }
287 }
288 jeb->gc_node = raw;
289
Joe Perches9c261b32012-02-15 15:56:43 -0800290 jffs2_dbg(1, "Going to garbage collect node at 0x%08x\n",
291 ref_offset(raw));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292
293 if (!raw->next_in_ino) {
294 /* Inode-less node. Clean marker, snapshot or something like that */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 spin_unlock(&c->erase_completion_lock);
David Woodhouse61715862006-05-21 00:02:06 +0100296 if (ref_flags(raw) == REF_PRISTINE) {
297 /* It's an unknown node with JFFS2_FEATURE_RWCOMPAT_COPY */
298 jffs2_garbage_collect_pristine(c, NULL, raw);
299 } else {
300 /* Just mark it obsolete */
301 jffs2_mark_node_obsolete(c, raw);
302 }
David Woodhouseced22072008-04-22 15:13:40 +0100303 mutex_unlock(&c->alloc_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304 goto eraseit_lock;
305 }
306
307 ic = jffs2_raw_ref_to_ic(raw);
308
KaiGai Kohei084702e2006-05-13 15:16:13 +0900309#ifdef CONFIG_JFFS2_FS_XATTR
KaiGai Koheiaa98d7c2006-05-13 15:09:47 +0900310 /* When 'ic' refers xattr_datum/xattr_ref, this node is GCed as xattr.
KaiGai Kohei084702e2006-05-13 15:16:13 +0900311 * We can decide whether this node is inode or xattr by ic->class. */
312 if (ic->class == RAWNODE_CLASS_XATTR_DATUM
313 || ic->class == RAWNODE_CLASS_XATTR_REF) {
KaiGai Kohei084702e2006-05-13 15:16:13 +0900314 spin_unlock(&c->erase_completion_lock);
315
316 if (ic->class == RAWNODE_CLASS_XATTR_DATUM) {
KaiGai Koheic9f700f2006-06-11 10:35:15 +0900317 ret = jffs2_garbage_collect_xattr_datum(c, (struct jffs2_xattr_datum *)ic, raw);
KaiGai Kohei084702e2006-05-13 15:16:13 +0900318 } else {
KaiGai Koheic9f700f2006-06-11 10:35:15 +0900319 ret = jffs2_garbage_collect_xattr_ref(c, (struct jffs2_xattr_ref *)ic, raw);
KaiGai Kohei084702e2006-05-13 15:16:13 +0900320 }
David Woodhouse2665ea82007-10-13 11:31:23 +0100321 goto test_gcnode;
KaiGai Kohei084702e2006-05-13 15:16:13 +0900322 }
323#endif
KaiGai Koheiaa98d7c2006-05-13 15:09:47 +0900324
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 /* We need to hold the inocache. Either the erase_completion_lock or
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000326 the inocache_lock are sufficient; we trade down since the inocache_lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 causes less contention. */
328 spin_lock(&c->inocache_lock);
329
330 spin_unlock(&c->erase_completion_lock);
331
Joe Perches9c261b32012-02-15 15:56:43 -0800332 jffs2_dbg(1, "%s(): collecting from block @0x%08x. Node @0x%08x(%d), ino #%u\n",
333 __func__, jeb->offset, ref_offset(raw), ref_flags(raw),
334 ic->ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335
336 /* Three possibilities:
337 1. Inode is already in-core. We must iget it and do proper
338 updating to its fragtree, etc.
339 2. Inode is not in-core, node is REF_PRISTINE. We lock the
340 inocache to prevent a read_inode(), copy the node intact.
341 3. Inode is not in-core, node is not pristine. We must iget()
342 and take the slow path.
343 */
344
345 switch(ic->state) {
346 case INO_STATE_CHECKEDABSENT:
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000347 /* It's been checked, but it's not currently in-core.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 We can just copy any pristine nodes, but have
349 to prevent anyone else from doing read_inode() while
350 we're at it, so we set the state accordingly */
351 if (ref_flags(raw) == REF_PRISTINE)
352 ic->state = INO_STATE_GC;
353 else {
Joe Perches9c261b32012-02-15 15:56:43 -0800354 jffs2_dbg(1, "Ino #%u is absent but node not REF_PRISTINE. Reading.\n",
355 ic->ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 }
357 break;
358
359 case INO_STATE_PRESENT:
360 /* It's in-core. GC must iget() it. */
361 break;
362
363 case INO_STATE_UNCHECKED:
364 case INO_STATE_CHECKING:
365 case INO_STATE_GC:
366 /* Should never happen. We should have finished checking
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000367 by the time we actually start doing any GC, and since
368 we're holding the alloc_sem, no other garbage collection
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 can happen.
370 */
Joe Perchesda320f02012-02-15 15:56:44 -0800371 pr_crit("Inode #%u already in state %d in jffs2_garbage_collect_pass()!\n",
372 ic->ino, ic->state);
David Woodhouseced22072008-04-22 15:13:40 +0100373 mutex_unlock(&c->alloc_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 spin_unlock(&c->inocache_lock);
375 BUG();
376
377 case INO_STATE_READING:
378 /* Someone's currently trying to read it. We must wait for
379 them to finish and then go through the full iget() route
380 to do the GC. However, sometimes read_inode() needs to get
381 the alloc_sem() (for marking nodes invalid) so we must
382 drop the alloc_sem before sleeping. */
383
David Woodhouseced22072008-04-22 15:13:40 +0100384 mutex_unlock(&c->alloc_sem);
Joe Perches9c261b32012-02-15 15:56:43 -0800385 jffs2_dbg(1, "%s(): waiting for ino #%u in state %d\n",
386 __func__, ic->ino, ic->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 sleep_on_spinunlock(&c->inocache_wq, &c->inocache_lock);
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000388 /* And because we dropped the alloc_sem we must start again from the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 beginning. Ponder chance of livelock here -- we're returning success
390 without actually making any progress.
391
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000392 Q: What are the chances that the inode is back in INO_STATE_READING
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 again by the time we next enter this function? And that this happens
394 enough times to cause a real delay?
395
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000396 A: Small enough that I don't care :)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 */
398 return 0;
399 }
400
401 /* OK. Now if the inode is in state INO_STATE_GC, we are going to copy the
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000402 node intact, and we don't have to muck about with the fragtree etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 because we know it's not in-core. If it _was_ in-core, we go through
404 all the iget() crap anyway */
405
406 if (ic->state == INO_STATE_GC) {
407 spin_unlock(&c->inocache_lock);
408
409 ret = jffs2_garbage_collect_pristine(c, ic, raw);
410
411 spin_lock(&c->inocache_lock);
412 ic->state = INO_STATE_CHECKEDABSENT;
413 wake_up(&c->inocache_wq);
414
415 if (ret != -EBADFD) {
416 spin_unlock(&c->inocache_lock);
David Woodhouse2665ea82007-10-13 11:31:23 +0100417 goto test_gcnode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 }
419
420 /* Fall through if it wanted us to, with inocache_lock held */
421 }
422
423 /* Prevent the fairly unlikely race where the gcblock is
424 entirely obsoleted by the final close of a file which had
425 the only valid nodes in the block, followed by erasure,
426 followed by freeing of the ic because the erased block(s)
427 held _all_ the nodes of that inode.... never been seen but
428 it's vaguely possible. */
429
430 inum = ic->ino;
David Woodhouse27c72b02008-05-01 18:47:17 +0100431 nlink = ic->pino_nlink;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 spin_unlock(&c->inocache_lock);
433
David Woodhouse1b690b42008-05-01 16:59:24 +0100434 f = jffs2_gc_fetch_inode(c, inum, !nlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 if (IS_ERR(f)) {
436 ret = PTR_ERR(f);
437 goto release_sem;
438 }
439 if (!f) {
440 ret = 0;
441 goto release_sem;
442 }
443
444 ret = jffs2_garbage_collect_live(c, jeb, raw, f);
445
446 jffs2_gc_release_inode(c, f);
447
David Woodhouse2665ea82007-10-13 11:31:23 +0100448 test_gcnode:
449 if (jeb->dirty_size == gcblock_dirty && !ref_obsolete(jeb->gc_node)) {
450 /* Eep. This really should never happen. GC is broken */
Joe Perchesda320f02012-02-15 15:56:44 -0800451 pr_err("Error garbage collecting node at %08x!\n",
452 ref_offset(jeb->gc_node));
David Woodhouse2665ea82007-10-13 11:31:23 +0100453 ret = -ENOSPC;
David Woodhouse4fc8a602007-10-13 14:29:39 +0100454 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 release_sem:
David Woodhouseced22072008-04-22 15:13:40 +0100456 mutex_unlock(&c->alloc_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457
458 eraseit_lock:
459 /* If we've finished this block, start it erasing */
460 spin_lock(&c->erase_completion_lock);
461
462 eraseit:
463 if (c->gcblock && !c->gcblock->used_size) {
Joe Perches9c261b32012-02-15 15:56:43 -0800464 jffs2_dbg(1, "Block at 0x%08x completely obsoleted by GC. Moving to erase_pending_list\n",
465 c->gcblock->offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 /* We're GC'ing an empty block? */
467 list_add_tail(&c->gcblock->list, &c->erase_pending_list);
468 c->gcblock = NULL;
469 c->nr_erasing_blocks++;
David Woodhouseae3b6ba2010-05-19 17:05:14 +0100470 jffs2_garbage_collect_trigger(c);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 }
472 spin_unlock(&c->erase_completion_lock);
473
474 return ret;
475}
476
477static int jffs2_garbage_collect_live(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
478 struct jffs2_raw_node_ref *raw, struct jffs2_inode_info *f)
479{
480 struct jffs2_node_frag *frag;
481 struct jffs2_full_dnode *fn = NULL;
482 struct jffs2_full_dirent *fd;
483 uint32_t start = 0, end = 0, nrfrags = 0;
484 int ret = 0;
485
David Woodhouseced22072008-04-22 15:13:40 +0100486 mutex_lock(&f->sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487
488 /* Now we have the lock for this inode. Check that it's still the one at the head
489 of the list. */
490
491 spin_lock(&c->erase_completion_lock);
492
493 if (c->gcblock != jeb) {
494 spin_unlock(&c->erase_completion_lock);
Joe Perches9c261b32012-02-15 15:56:43 -0800495 jffs2_dbg(1, "GC block is no longer gcblock. Restart\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 goto upnout;
497 }
498 if (ref_obsolete(raw)) {
499 spin_unlock(&c->erase_completion_lock);
Joe Perches9c261b32012-02-15 15:56:43 -0800500 jffs2_dbg(1, "node to be GC'd was obsoleted in the meantime.\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 /* They'll call again */
502 goto upnout;
503 }
504 spin_unlock(&c->erase_completion_lock);
505
506 /* OK. Looks safe. And nobody can get us now because we have the semaphore. Move the block */
507 if (f->metadata && f->metadata->raw == raw) {
508 fn = f->metadata;
509 ret = jffs2_garbage_collect_metadata(c, jeb, f, fn);
510 goto upnout;
511 }
512
513 /* FIXME. Read node and do lookup? */
514 for (frag = frag_first(&f->fragtree); frag; frag = frag_next(frag)) {
515 if (frag->node && frag->node->raw == raw) {
516 fn = frag->node;
517 end = frag->ofs + frag->size;
518 if (!nrfrags++)
519 start = frag->ofs;
520 if (nrfrags == frag->node->frags)
521 break; /* We've found them all */
522 }
523 }
524 if (fn) {
525 if (ref_flags(raw) == REF_PRISTINE) {
526 ret = jffs2_garbage_collect_pristine(c, f->inocache, raw);
527 if (!ret) {
528 /* Urgh. Return it sensibly. */
529 frag->node->raw = f->inocache->nodes;
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000530 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 if (ret != -EBADFD)
532 goto upnout;
533 }
534 /* We found a datanode. Do the GC */
535 if((start >> PAGE_CACHE_SHIFT) < ((end-1) >> PAGE_CACHE_SHIFT)) {
536 /* It crosses a page boundary. Therefore, it must be a hole. */
537 ret = jffs2_garbage_collect_hole(c, jeb, f, fn, start, end);
538 } else {
539 /* It could still be a hole. But we GC the page this way anyway */
540 ret = jffs2_garbage_collect_dnode(c, jeb, f, fn, start, end);
541 }
542 goto upnout;
543 }
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000544
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 /* Wasn't a dnode. Try dirent */
546 for (fd = f->dents; fd; fd=fd->next) {
547 if (fd->raw == raw)
548 break;
549 }
550
551 if (fd && fd->ino) {
552 ret = jffs2_garbage_collect_dirent(c, jeb, f, fd);
553 } else if (fd) {
554 ret = jffs2_garbage_collect_deletion_dirent(c, jeb, f, fd);
555 } else {
Joe Perchesda320f02012-02-15 15:56:44 -0800556 pr_warn("Raw node at 0x%08x wasn't in node lists for ino #%u\n",
557 ref_offset(raw), f->inocache->ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 if (ref_obsolete(raw)) {
Joe Perchesda320f02012-02-15 15:56:44 -0800559 pr_warn("But it's obsolete so we don't mind too much\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 } else {
Artem B. Bityutskiye0c8e422005-07-24 16:14:17 +0100561 jffs2_dbg_dump_node(c, ref_offset(raw));
562 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 }
564 }
565 upnout:
David Woodhouseced22072008-04-22 15:13:40 +0100566 mutex_unlock(&f->sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567
568 return ret;
569}
570
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000571static int jffs2_garbage_collect_pristine(struct jffs2_sb_info *c,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 struct jffs2_inode_cache *ic,
573 struct jffs2_raw_node_ref *raw)
574{
575 union jffs2_node_union *node;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 size_t retlen;
577 int ret;
578 uint32_t phys_ofs, alloclen;
579 uint32_t crc, rawlen;
580 int retried = 0;
581
Joe Perches9c261b32012-02-15 15:56:43 -0800582 jffs2_dbg(1, "Going to GC REF_PRISTINE node at 0x%08x\n",
583 ref_offset(raw));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584
David Woodhouse61715862006-05-21 00:02:06 +0100585 alloclen = rawlen = ref_totlen(c, c->gcblock, raw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586
587 /* Ask for a small amount of space (or the totlen if smaller) because we
588 don't want to force wastage of the end of a block if splitting would
589 work. */
David Woodhouse61715862006-05-21 00:02:06 +0100590 if (ic && alloclen > sizeof(struct jffs2_raw_inode) + JFFS2_MIN_DATA_LEN)
591 alloclen = sizeof(struct jffs2_raw_inode) + JFFS2_MIN_DATA_LEN;
592
David Woodhouse9fe48542006-05-23 00:38:06 +0100593 ret = jffs2_reserve_space_gc(c, alloclen, &alloclen, rawlen);
David Woodhouse61715862006-05-21 00:02:06 +0100594 /* 'rawlen' is not the exact summary size; it is only an upper estimation */
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100595
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 if (ret)
597 return ret;
598
599 if (alloclen < rawlen) {
600 /* Doesn't fit untouched. We'll go the old route and split it */
601 return -EBADFD;
602 }
603
604 node = kmalloc(rawlen, GFP_KERNEL);
605 if (!node)
David Woodhouseef53cb02007-07-10 10:01:22 +0100606 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607
608 ret = jffs2_flash_read(c, ref_offset(raw), rawlen, &retlen, (char *)node);
609 if (!ret && retlen != rawlen)
610 ret = -EIO;
611 if (ret)
612 goto out_node;
613
614 crc = crc32(0, node, sizeof(struct jffs2_unknown_node)-4);
615 if (je32_to_cpu(node->u.hdr_crc) != crc) {
Joe Perchesda320f02012-02-15 15:56:44 -0800616 pr_warn("Header CRC failed on REF_PRISTINE node at 0x%08x: Read 0x%08x, calculated 0x%08x\n",
617 ref_offset(raw), je32_to_cpu(node->u.hdr_crc), crc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 goto bail;
619 }
620
621 switch(je16_to_cpu(node->u.nodetype)) {
622 case JFFS2_NODETYPE_INODE:
623 crc = crc32(0, node, sizeof(node->i)-8);
624 if (je32_to_cpu(node->i.node_crc) != crc) {
Joe Perchesda320f02012-02-15 15:56:44 -0800625 pr_warn("Node CRC failed on REF_PRISTINE data node at 0x%08x: Read 0x%08x, calculated 0x%08x\n",
626 ref_offset(raw), je32_to_cpu(node->i.node_crc),
627 crc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 goto bail;
629 }
630
631 if (je32_to_cpu(node->i.dsize)) {
632 crc = crc32(0, node->i.data, je32_to_cpu(node->i.csize));
633 if (je32_to_cpu(node->i.data_crc) != crc) {
Joe Perchesda320f02012-02-15 15:56:44 -0800634 pr_warn("Data CRC failed on REF_PRISTINE data node at 0x%08x: Read 0x%08x, calculated 0x%08x\n",
635 ref_offset(raw),
636 je32_to_cpu(node->i.data_crc), crc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 goto bail;
638 }
639 }
640 break;
641
642 case JFFS2_NODETYPE_DIRENT:
643 crc = crc32(0, node, sizeof(node->d)-8);
644 if (je32_to_cpu(node->d.node_crc) != crc) {
Joe Perchesda320f02012-02-15 15:56:44 -0800645 pr_warn("Node CRC failed on REF_PRISTINE dirent node at 0x%08x: Read 0x%08x, calculated 0x%08x\n",
646 ref_offset(raw),
647 je32_to_cpu(node->d.node_crc), crc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 goto bail;
649 }
650
David Woodhouseb534e702007-10-13 11:35:58 +0100651 if (strnlen(node->d.name, node->d.nsize) != node->d.nsize) {
Joe Perchesda320f02012-02-15 15:56:44 -0800652 pr_warn("Name in dirent node at 0x%08x contains zeroes\n",
653 ref_offset(raw));
David Woodhouseb534e702007-10-13 11:35:58 +0100654 goto bail;
655 }
656
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 if (node->d.nsize) {
658 crc = crc32(0, node->d.name, node->d.nsize);
659 if (je32_to_cpu(node->d.name_crc) != crc) {
Joe Perchesda320f02012-02-15 15:56:44 -0800660 pr_warn("Name CRC failed on REF_PRISTINE dirent node at 0x%08x: Read 0x%08x, calculated 0x%08x\n",
661 ref_offset(raw),
662 je32_to_cpu(node->d.name_crc), crc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 goto bail;
664 }
665 }
666 break;
667 default:
David Woodhouse61715862006-05-21 00:02:06 +0100668 /* If it's inode-less, we don't _know_ what it is. Just copy it intact */
669 if (ic) {
Joe Perchesda320f02012-02-15 15:56:44 -0800670 pr_warn("Unknown node type for REF_PRISTINE node at 0x%08x: 0x%04x\n",
671 ref_offset(raw), je16_to_cpu(node->u.nodetype));
David Woodhouse61715862006-05-21 00:02:06 +0100672 goto bail;
673 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 }
675
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 /* OK, all the CRCs are good; this node can just be copied as-is. */
677 retry:
David Woodhouse2f785402006-05-24 02:04:45 +0100678 phys_ofs = write_ofs(c);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679
680 ret = jffs2_flash_write(c, phys_ofs, rawlen, &retlen, (char *)node);
681
682 if (ret || (retlen != rawlen)) {
Joe Perchesda320f02012-02-15 15:56:44 -0800683 pr_notice("Write of %d bytes at 0x%08x failed. returned %d, retlen %zd\n",
684 rawlen, phys_ofs, ret, retlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 if (retlen) {
David Woodhouse2f785402006-05-24 02:04:45 +0100686 jffs2_add_physical_node_ref(c, phys_ofs | REF_OBSOLETE, rawlen, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 } else {
Joe Perchesda320f02012-02-15 15:56:44 -0800688 pr_notice("Not marking the space at 0x%08x as dirty because the flash driver returned retlen zero\n",
689 phys_ofs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 }
David Woodhouse2f785402006-05-24 02:04:45 +0100691 if (!retried) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 /* Try to reallocate space and retry */
693 uint32_t dummy;
694 struct jffs2_eraseblock *jeb = &c->blocks[phys_ofs / c->sector_size];
695
696 retried = 1;
697
Joe Perches9c261b32012-02-15 15:56:43 -0800698 jffs2_dbg(1, "Retrying failed write of REF_PRISTINE node.\n");
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000699
Artem B. Bityutskiy730554d2005-07-17 07:56:26 +0100700 jffs2_dbg_acct_sanity_check(c,jeb);
701 jffs2_dbg_acct_paranoia_check(c, jeb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702
David Woodhouse9fe48542006-05-23 00:38:06 +0100703 ret = jffs2_reserve_space_gc(c, rawlen, &dummy, rawlen);
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100704 /* this is not the exact summary size of it,
705 it is only an upper estimation */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706
707 if (!ret) {
Joe Perches9c261b32012-02-15 15:56:43 -0800708 jffs2_dbg(1, "Allocated space at 0x%08x to retry failed write.\n",
709 phys_ofs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710
Artem B. Bityutskiy730554d2005-07-17 07:56:26 +0100711 jffs2_dbg_acct_sanity_check(c,jeb);
712 jffs2_dbg_acct_paranoia_check(c, jeb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713
714 goto retry;
715 }
Joe Perches9c261b32012-02-15 15:56:43 -0800716 jffs2_dbg(1, "Failed to allocate space to retry failed write: %d!\n",
717 ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 }
719
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 if (!ret)
721 ret = -EIO;
722 goto out_node;
723 }
David Woodhouse2f785402006-05-24 02:04:45 +0100724 jffs2_add_physical_node_ref(c, phys_ofs | REF_PRISTINE, rawlen, ic);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 jffs2_mark_node_obsolete(c, raw);
Joe Perches9c261b32012-02-15 15:56:43 -0800727 jffs2_dbg(1, "WHEEE! GC REF_PRISTINE node at 0x%08x succeeded\n",
728 ref_offset(raw));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729
730 out_node:
731 kfree(node);
732 return ret;
733 bail:
734 ret = -EBADFD;
735 goto out_node;
736}
737
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000738static int jffs2_garbage_collect_metadata(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 struct jffs2_inode_info *f, struct jffs2_full_dnode *fn)
740{
741 struct jffs2_full_dnode *new_fn;
742 struct jffs2_raw_inode ri;
Artem B. Bityuckiy8557fd52005-04-09 11:47:03 +0100743 struct jffs2_node_frag *last_frag;
David Woodhouseaef9ab42006-05-19 00:28:49 +0100744 union jffs2_device_node dev;
David Woodhouse2e16cfc2009-12-16 03:27:20 +0000745 char *mdata = NULL;
746 int mdatalen = 0;
David Woodhouse9fe48542006-05-23 00:38:06 +0100747 uint32_t alloclen, ilen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 int ret;
749
750 if (S_ISBLK(JFFS2_F_I_MODE(f)) ||
751 S_ISCHR(JFFS2_F_I_MODE(f)) ) {
752 /* For these, we don't actually need to read the old node */
David Woodhouseaef9ab42006-05-19 00:28:49 +0100753 mdatalen = jffs2_encode_dev(&dev, JFFS2_F_I_RDEV(f));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 mdata = (char *)&dev;
Joe Perches9c261b32012-02-15 15:56:43 -0800755 jffs2_dbg(1, "%s(): Writing %d bytes of kdev_t\n",
756 __func__, mdatalen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 } else if (S_ISLNK(JFFS2_F_I_MODE(f))) {
758 mdatalen = fn->size;
759 mdata = kmalloc(fn->size, GFP_KERNEL);
760 if (!mdata) {
Joe Perchesda320f02012-02-15 15:56:44 -0800761 pr_warn("kmalloc of mdata failed in jffs2_garbage_collect_metadata()\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 return -ENOMEM;
763 }
764 ret = jffs2_read_dnode(c, f, fn, mdata, 0, mdatalen);
765 if (ret) {
Joe Perchesda320f02012-02-15 15:56:44 -0800766 pr_warn("read of old metadata failed in jffs2_garbage_collect_metadata(): %d\n",
767 ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 kfree(mdata);
769 return ret;
770 }
Joe Perches9c261b32012-02-15 15:56:43 -0800771 jffs2_dbg(1, "%s(): Writing %d bites of symlink target\n",
772 __func__, mdatalen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773
774 }
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000775
David Woodhouse9fe48542006-05-23 00:38:06 +0100776 ret = jffs2_reserve_space_gc(c, sizeof(ri) + mdatalen, &alloclen,
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100777 JFFS2_SUMMARY_INODE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 if (ret) {
Joe Perchesda320f02012-02-15 15:56:44 -0800779 pr_warn("jffs2_reserve_space_gc of %zd bytes for garbage_collect_metadata failed: %d\n",
780 sizeof(ri) + mdatalen, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 goto out;
782 }
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000783
Artem B. Bityuckiy8557fd52005-04-09 11:47:03 +0100784 last_frag = frag_last(&f->fragtree);
785 if (last_frag)
786 /* Fetch the inode length from the fragtree rather then
787 * from i_size since i_size may have not been updated yet */
788 ilen = last_frag->ofs + last_frag->size;
789 else
790 ilen = JFFS2_F_I_SIZE(f);
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000791
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 memset(&ri, 0, sizeof(ri));
793 ri.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK);
794 ri.nodetype = cpu_to_je16(JFFS2_NODETYPE_INODE);
795 ri.totlen = cpu_to_je32(sizeof(ri) + mdatalen);
796 ri.hdr_crc = cpu_to_je32(crc32(0, &ri, sizeof(struct jffs2_unknown_node)-4));
797
798 ri.ino = cpu_to_je32(f->inocache->ino);
799 ri.version = cpu_to_je32(++f->highest_version);
800 ri.mode = cpu_to_jemode(JFFS2_F_I_MODE(f));
801 ri.uid = cpu_to_je16(JFFS2_F_I_UID(f));
802 ri.gid = cpu_to_je16(JFFS2_F_I_GID(f));
Artem B. Bityuckiy8557fd52005-04-09 11:47:03 +0100803 ri.isize = cpu_to_je32(ilen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 ri.atime = cpu_to_je32(JFFS2_F_I_ATIME(f));
805 ri.ctime = cpu_to_je32(JFFS2_F_I_CTIME(f));
806 ri.mtime = cpu_to_je32(JFFS2_F_I_MTIME(f));
807 ri.offset = cpu_to_je32(0);
808 ri.csize = cpu_to_je32(mdatalen);
809 ri.dsize = cpu_to_je32(mdatalen);
810 ri.compr = JFFS2_COMPR_NONE;
811 ri.node_crc = cpu_to_je32(crc32(0, &ri, sizeof(ri)-8));
812 ri.data_crc = cpu_to_je32(crc32(0, mdata, mdatalen));
813
David Woodhouse9fe48542006-05-23 00:38:06 +0100814 new_fn = jffs2_write_dnode(c, f, &ri, mdata, mdatalen, ALLOC_GC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815
816 if (IS_ERR(new_fn)) {
Joe Perchesda320f02012-02-15 15:56:44 -0800817 pr_warn("Error writing new dnode: %ld\n", PTR_ERR(new_fn));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 ret = PTR_ERR(new_fn);
819 goto out;
820 }
821 jffs2_mark_node_obsolete(c, fn->raw);
822 jffs2_free_full_dnode(fn);
823 f->metadata = new_fn;
824 out:
825 if (S_ISLNK(JFFS2_F_I_MODE(f)))
826 kfree(mdata);
827 return ret;
828}
829
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000830static int jffs2_garbage_collect_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 struct jffs2_inode_info *f, struct jffs2_full_dirent *fd)
832{
833 struct jffs2_full_dirent *new_fd;
834 struct jffs2_raw_dirent rd;
David Woodhouse9fe48542006-05-23 00:38:06 +0100835 uint32_t alloclen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 int ret;
837
838 rd.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK);
839 rd.nodetype = cpu_to_je16(JFFS2_NODETYPE_DIRENT);
840 rd.nsize = strlen(fd->name);
841 rd.totlen = cpu_to_je32(sizeof(rd) + rd.nsize);
842 rd.hdr_crc = cpu_to_je32(crc32(0, &rd, sizeof(struct jffs2_unknown_node)-4));
843
844 rd.pino = cpu_to_je32(f->inocache->ino);
845 rd.version = cpu_to_je32(++f->highest_version);
846 rd.ino = cpu_to_je32(fd->ino);
Artem B. Bityutskiy3a69e0c2005-08-17 14:46:26 +0100847 /* If the times on this inode were set by explicit utime() they can be different,
848 so refrain from splatting them. */
849 if (JFFS2_F_I_MTIME(f) == JFFS2_F_I_CTIME(f))
850 rd.mctime = cpu_to_je32(JFFS2_F_I_MTIME(f));
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000851 else
Artem B. Bityutskiy3a69e0c2005-08-17 14:46:26 +0100852 rd.mctime = cpu_to_je32(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 rd.type = fd->type;
854 rd.node_crc = cpu_to_je32(crc32(0, &rd, sizeof(rd)-8));
855 rd.name_crc = cpu_to_je32(crc32(0, fd->name, rd.nsize));
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000856
David Woodhouse9fe48542006-05-23 00:38:06 +0100857 ret = jffs2_reserve_space_gc(c, sizeof(rd)+rd.nsize, &alloclen,
Ferenc Havasie631ddb2005-09-07 09:35:26 +0100858 JFFS2_SUMMARY_DIRENT_SIZE(rd.nsize));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 if (ret) {
Joe Perchesda320f02012-02-15 15:56:44 -0800860 pr_warn("jffs2_reserve_space_gc of %zd bytes for garbage_collect_dirent failed: %d\n",
861 sizeof(rd)+rd.nsize, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 return ret;
863 }
David Woodhouse9fe48542006-05-23 00:38:06 +0100864 new_fd = jffs2_write_dirent(c, f, &rd, fd->name, rd.nsize, ALLOC_GC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865
866 if (IS_ERR(new_fd)) {
Joe Perchesda320f02012-02-15 15:56:44 -0800867 pr_warn("jffs2_write_dirent in garbage_collect_dirent failed: %ld\n",
868 PTR_ERR(new_fd));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 return PTR_ERR(new_fd);
870 }
871 jffs2_add_fd_to_list(c, new_fd, &f->dents);
872 return 0;
873}
874
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000875static int jffs2_garbage_collect_deletion_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 struct jffs2_inode_info *f, struct jffs2_full_dirent *fd)
877{
878 struct jffs2_full_dirent **fdp = &f->dents;
879 int found = 0;
880
881 /* On a medium where we can't actually mark nodes obsolete
882 pernamently, such as NAND flash, we need to work out
883 whether this deletion dirent is still needed to actively
884 delete a 'real' dirent with the same name that's still
885 somewhere else on the flash. */
886 if (!jffs2_can_mark_obsolete(c)) {
887 struct jffs2_raw_dirent *rd;
888 struct jffs2_raw_node_ref *raw;
889 int ret;
890 size_t retlen;
891 int name_len = strlen(fd->name);
892 uint32_t name_crc = crc32(0, fd->name, name_len);
893 uint32_t rawlen = ref_totlen(c, jeb, fd->raw);
894
895 rd = kmalloc(rawlen, GFP_KERNEL);
896 if (!rd)
897 return -ENOMEM;
898
899 /* Prevent the erase code from nicking the obsolete node refs while
900 we're looking at them. I really don't like this extra lock but
901 can't see any alternative. Suggestions on a postcard to... */
David Woodhouseced22072008-04-22 15:13:40 +0100902 mutex_lock(&c->erase_free_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903
904 for (raw = f->inocache->nodes; raw != (void *)f->inocache; raw = raw->next_in_ino) {
905
Artem Bityutskiyaba54da2006-12-19 15:45:23 +0200906 cond_resched();
907
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 /* We only care about obsolete ones */
909 if (!(ref_obsolete(raw)))
910 continue;
911
912 /* Any dirent with the same name is going to have the same length... */
913 if (ref_totlen(c, NULL, raw) != rawlen)
914 continue;
915
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000916 /* Doesn't matter if there's one in the same erase block. We're going to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 delete it too at the same time. */
Andrew Victor3be36672005-02-09 09:09:05 +0000918 if (SECTOR_ADDR(raw->flash_offset) == SECTOR_ADDR(fd->raw->flash_offset))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 continue;
920
Joe Perches9c261b32012-02-15 15:56:43 -0800921 jffs2_dbg(1, "Check potential deletion dirent at %08x\n",
922 ref_offset(raw));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923
924 /* This is an obsolete node belonging to the same directory, and it's of the right
925 length. We need to take a closer look...*/
926 ret = jffs2_flash_read(c, ref_offset(raw), rawlen, &retlen, (char *)rd);
927 if (ret) {
Joe Perchesda320f02012-02-15 15:56:44 -0800928 pr_warn("%s(): Read error (%d) reading obsolete node at %08x\n",
929 __func__, ret, ref_offset(raw));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 /* If we can't read it, we don't need to continue to obsolete it. Continue */
931 continue;
932 }
933 if (retlen != rawlen) {
Joe Perchesda320f02012-02-15 15:56:44 -0800934 pr_warn("%s(): Short read (%zd not %u) reading header from obsolete node at %08x\n",
935 __func__, retlen, rawlen,
936 ref_offset(raw));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 continue;
938 }
939
940 if (je16_to_cpu(rd->nodetype) != JFFS2_NODETYPE_DIRENT)
941 continue;
942
943 /* If the name CRC doesn't match, skip */
944 if (je32_to_cpu(rd->name_crc) != name_crc)
945 continue;
946
947 /* If the name length doesn't match, or it's another deletion dirent, skip */
948 if (rd->nsize != name_len || !je32_to_cpu(rd->ino))
949 continue;
950
951 /* OK, check the actual name now */
952 if (memcmp(rd->name, fd->name, name_len))
953 continue;
954
955 /* OK. The name really does match. There really is still an older node on
956 the flash which our deletion dirent obsoletes. So we have to write out
957 a new deletion dirent to replace it */
David Woodhouseced22072008-04-22 15:13:40 +0100958 mutex_unlock(&c->erase_free_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959
Joe Perches9c261b32012-02-15 15:56:43 -0800960 jffs2_dbg(1, "Deletion dirent at %08x still obsoletes real dirent \"%s\" at %08x for ino #%u\n",
961 ref_offset(fd->raw), fd->name,
962 ref_offset(raw), je32_to_cpu(rd->ino));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 kfree(rd);
964
965 return jffs2_garbage_collect_dirent(c, jeb, f, fd);
966 }
967
David Woodhouseced22072008-04-22 15:13:40 +0100968 mutex_unlock(&c->erase_free_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 kfree(rd);
970 }
971
Thomas Gleixner182ec4e2005-11-07 11:16:07 +0000972 /* FIXME: If we're deleting a dirent which contains the current mtime and ctime,
Artem B. Bityutskiy3a69e0c2005-08-17 14:46:26 +0100973 we should update the metadata node with those times accordingly */
974
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 /* No need for it any more. Just mark it obsolete and remove it from the list */
976 while (*fdp) {
977 if ((*fdp) == fd) {
978 found = 1;
979 *fdp = fd->next;
980 break;
981 }
982 fdp = &(*fdp)->next;
983 }
984 if (!found) {
Joe Perchesda320f02012-02-15 15:56:44 -0800985 pr_warn("Deletion dirent \"%s\" not found in list for ino #%u\n",
986 fd->name, f->inocache->ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 }
988 jffs2_mark_node_obsolete(c, fd->raw);
989 jffs2_free_full_dirent(fd);
990 return 0;
991}
992
993static int jffs2_garbage_collect_hole(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
994 struct jffs2_inode_info *f, struct jffs2_full_dnode *fn,
995 uint32_t start, uint32_t end)
996{
997 struct jffs2_raw_inode ri;
998 struct jffs2_node_frag *frag;
999 struct jffs2_full_dnode *new_fn;
David Woodhouse9fe48542006-05-23 00:38:06 +01001000 uint32_t alloclen, ilen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 int ret;
1002
Joe Perches9c261b32012-02-15 15:56:43 -08001003 jffs2_dbg(1, "Writing replacement hole node for ino #%u from offset 0x%x to 0x%x\n",
1004 f->inocache->ino, start, end);
Thomas Gleixner182ec4e2005-11-07 11:16:07 +00001005
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 memset(&ri, 0, sizeof(ri));
1007
1008 if(fn->frags > 1) {
1009 size_t readlen;
1010 uint32_t crc;
Thomas Gleixner182ec4e2005-11-07 11:16:07 +00001011 /* It's partially obsoleted by a later write. So we have to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 write it out again with the _same_ version as before */
1013 ret = jffs2_flash_read(c, ref_offset(fn->raw), sizeof(ri), &readlen, (char *)&ri);
1014 if (readlen != sizeof(ri) || ret) {
Joe Perchesda320f02012-02-15 15:56:44 -08001015 pr_warn("Node read failed in jffs2_garbage_collect_hole. Ret %d, retlen %zd. Data will be lost by writing new hole node\n",
1016 ret, readlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 goto fill;
1018 }
1019 if (je16_to_cpu(ri.nodetype) != JFFS2_NODETYPE_INODE) {
Joe Perchesda320f02012-02-15 15:56:44 -08001020 pr_warn("%s(): Node at 0x%08x had node type 0x%04x instead of JFFS2_NODETYPE_INODE(0x%04x)\n",
1021 __func__, ref_offset(fn->raw),
1022 je16_to_cpu(ri.nodetype), JFFS2_NODETYPE_INODE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 return -EIO;
1024 }
1025 if (je32_to_cpu(ri.totlen) != sizeof(ri)) {
Joe Perchesda320f02012-02-15 15:56:44 -08001026 pr_warn("%s(): Node at 0x%08x had totlen 0x%x instead of expected 0x%zx\n",
1027 __func__, ref_offset(fn->raw),
1028 je32_to_cpu(ri.totlen), sizeof(ri));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 return -EIO;
1030 }
1031 crc = crc32(0, &ri, sizeof(ri)-8);
1032 if (crc != je32_to_cpu(ri.node_crc)) {
Joe Perchesda320f02012-02-15 15:56:44 -08001033 pr_warn("%s: Node at 0x%08x had CRC 0x%08x which doesn't match calculated CRC 0x%08x\n",
1034 __func__, ref_offset(fn->raw),
1035 je32_to_cpu(ri.node_crc), crc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 /* FIXME: We could possibly deal with this by writing new holes for each frag */
Joe Perchesda320f02012-02-15 15:56:44 -08001037 pr_warn("Data in the range 0x%08x to 0x%08x of inode #%u will be lost\n",
1038 start, end, f->inocache->ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 goto fill;
1040 }
1041 if (ri.compr != JFFS2_COMPR_ZERO) {
Joe Perchesda320f02012-02-15 15:56:44 -08001042 pr_warn("%s(): Node 0x%08x wasn't a hole node!\n",
1043 __func__, ref_offset(fn->raw));
1044 pr_warn("Data in the range 0x%08x to 0x%08x of inode #%u will be lost\n",
1045 start, end, f->inocache->ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 goto fill;
1047 }
1048 } else {
1049 fill:
1050 ri.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK);
1051 ri.nodetype = cpu_to_je16(JFFS2_NODETYPE_INODE);
1052 ri.totlen = cpu_to_je32(sizeof(ri));
1053 ri.hdr_crc = cpu_to_je32(crc32(0, &ri, sizeof(struct jffs2_unknown_node)-4));
1054
1055 ri.ino = cpu_to_je32(f->inocache->ino);
1056 ri.version = cpu_to_je32(++f->highest_version);
1057 ri.offset = cpu_to_je32(start);
1058 ri.dsize = cpu_to_je32(end - start);
1059 ri.csize = cpu_to_je32(0);
1060 ri.compr = JFFS2_COMPR_ZERO;
1061 }
Thomas Gleixner182ec4e2005-11-07 11:16:07 +00001062
Artem B. Bityuckiy8557fd52005-04-09 11:47:03 +01001063 frag = frag_last(&f->fragtree);
1064 if (frag)
1065 /* Fetch the inode length from the fragtree rather then
1066 * from i_size since i_size may have not been updated yet */
1067 ilen = frag->ofs + frag->size;
1068 else
1069 ilen = JFFS2_F_I_SIZE(f);
1070
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 ri.mode = cpu_to_jemode(JFFS2_F_I_MODE(f));
1072 ri.uid = cpu_to_je16(JFFS2_F_I_UID(f));
1073 ri.gid = cpu_to_je16(JFFS2_F_I_GID(f));
Artem B. Bityuckiy8557fd52005-04-09 11:47:03 +01001074 ri.isize = cpu_to_je32(ilen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 ri.atime = cpu_to_je32(JFFS2_F_I_ATIME(f));
1076 ri.ctime = cpu_to_je32(JFFS2_F_I_CTIME(f));
1077 ri.mtime = cpu_to_je32(JFFS2_F_I_MTIME(f));
1078 ri.data_crc = cpu_to_je32(0);
1079 ri.node_crc = cpu_to_je32(crc32(0, &ri, sizeof(ri)-8));
1080
David Woodhouse9fe48542006-05-23 00:38:06 +01001081 ret = jffs2_reserve_space_gc(c, sizeof(ri), &alloclen,
1082 JFFS2_SUMMARY_INODE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 if (ret) {
Joe Perchesda320f02012-02-15 15:56:44 -08001084 pr_warn("jffs2_reserve_space_gc of %zd bytes for garbage_collect_hole failed: %d\n",
1085 sizeof(ri), ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 return ret;
1087 }
David Woodhouse9fe48542006-05-23 00:38:06 +01001088 new_fn = jffs2_write_dnode(c, f, &ri, NULL, 0, ALLOC_GC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089
1090 if (IS_ERR(new_fn)) {
Joe Perchesda320f02012-02-15 15:56:44 -08001091 pr_warn("Error writing new hole node: %ld\n", PTR_ERR(new_fn));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 return PTR_ERR(new_fn);
1093 }
1094 if (je32_to_cpu(ri.version) == f->highest_version) {
1095 jffs2_add_full_dnode_to_inode(c, f, new_fn);
1096 if (f->metadata) {
1097 jffs2_mark_node_obsolete(c, f->metadata->raw);
1098 jffs2_free_full_dnode(f->metadata);
1099 f->metadata = NULL;
1100 }
1101 return 0;
1102 }
1103
Thomas Gleixner182ec4e2005-11-07 11:16:07 +00001104 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 * We should only get here in the case where the node we are
1106 * replacing had more than one frag, so we kept the same version
Thomas Gleixner182ec4e2005-11-07 11:16:07 +00001107 * number as before. (Except in case of error -- see 'goto fill;'
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 * above.)
1109 */
1110 D1(if(unlikely(fn->frags <= 1)) {
Joe Perchesda320f02012-02-15 15:56:44 -08001111 pr_warn("%s(): Replacing fn with %d frag(s) but new ver %d != highest_version %d of ino #%d\n",
1112 __func__, fn->frags, je32_to_cpu(ri.version),
1113 f->highest_version, je32_to_cpu(ri.ino));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 });
1115
1116 /* This is a partially-overlapped hole node. Mark it REF_NORMAL not REF_PRISTINE */
1117 mark_ref_normal(new_fn->raw);
1118
Thomas Gleixner182ec4e2005-11-07 11:16:07 +00001119 for (frag = jffs2_lookup_node_frag(&f->fragtree, fn->ofs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 frag; frag = frag_next(frag)) {
1121 if (frag->ofs > fn->size + fn->ofs)
1122 break;
1123 if (frag->node == fn) {
1124 frag->node = new_fn;
1125 new_fn->frags++;
1126 fn->frags--;
1127 }
1128 }
1129 if (fn->frags) {
Joe Perchesda320f02012-02-15 15:56:44 -08001130 pr_warn("%s(): Old node still has frags!\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 BUG();
1132 }
1133 if (!new_fn->frags) {
Joe Perchesda320f02012-02-15 15:56:44 -08001134 pr_warn("%s(): New node has no frags!\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 BUG();
1136 }
Thomas Gleixner182ec4e2005-11-07 11:16:07 +00001137
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 jffs2_mark_node_obsolete(c, fn->raw);
1139 jffs2_free_full_dnode(fn);
Thomas Gleixner182ec4e2005-11-07 11:16:07 +00001140
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 return 0;
1142}
1143
David Woodhouse25dc30b2008-04-22 12:12:25 +01001144static int jffs2_garbage_collect_dnode(struct jffs2_sb_info *c, struct jffs2_eraseblock *orig_jeb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 struct jffs2_inode_info *f, struct jffs2_full_dnode *fn,
1146 uint32_t start, uint32_t end)
1147{
1148 struct jffs2_full_dnode *new_fn;
1149 struct jffs2_raw_inode ri;
David Woodhouse9fe48542006-05-23 00:38:06 +01001150 uint32_t alloclen, offset, orig_end, orig_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 int ret = 0;
1152 unsigned char *comprbuf = NULL, *writebuf;
1153 unsigned long pg;
1154 unsigned char *pg_ptr;
Thomas Gleixner182ec4e2005-11-07 11:16:07 +00001155
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 memset(&ri, 0, sizeof(ri));
1157
Joe Perches9c261b32012-02-15 15:56:43 -08001158 jffs2_dbg(1, "Writing replacement dnode for ino #%u from offset 0x%x to 0x%x\n",
1159 f->inocache->ino, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160
1161 orig_end = end;
1162 orig_start = start;
1163
1164 if (c->nr_free_blocks + c->nr_erasing_blocks > c->resv_blocks_gcmerge) {
1165 /* Attempt to do some merging. But only expand to cover logically
1166 adjacent frags if the block containing them is already considered
Thomas Gleixner182ec4e2005-11-07 11:16:07 +00001167 to be dirty. Otherwise we end up with GC just going round in
1168 circles dirtying the nodes it already wrote out, especially
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 on NAND where we have small eraseblocks and hence a much higher
1170 chance of nodes having to be split to cross boundaries. */
1171
1172 struct jffs2_node_frag *frag;
1173 uint32_t min, max;
1174
1175 min = start & ~(PAGE_CACHE_SIZE-1);
1176 max = min + PAGE_CACHE_SIZE;
1177
1178 frag = jffs2_lookup_node_frag(&f->fragtree, start);
1179
1180 /* BUG_ON(!frag) but that'll happen anyway... */
1181
1182 BUG_ON(frag->ofs != start);
1183
1184 /* First grow down... */
1185 while((frag = frag_prev(frag)) && frag->ofs >= min) {
1186
1187 /* If the previous frag doesn't even reach the beginning, there's
1188 excessive fragmentation. Just merge. */
1189 if (frag->ofs > min) {
Joe Perches9c261b32012-02-15 15:56:43 -08001190 jffs2_dbg(1, "Expanding down to cover partial frag (0x%x-0x%x)\n",
1191 frag->ofs, frag->ofs+frag->size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 start = frag->ofs;
1193 continue;
1194 }
1195 /* OK. This frag holds the first byte of the page. */
1196 if (!frag->node || !frag->node->raw) {
Joe Perches9c261b32012-02-15 15:56:43 -08001197 jffs2_dbg(1, "First frag in page is hole (0x%x-0x%x). Not expanding down.\n",
1198 frag->ofs, frag->ofs+frag->size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 break;
1200 } else {
1201
Thomas Gleixner182ec4e2005-11-07 11:16:07 +00001202 /* OK, it's a frag which extends to the beginning of the page. Does it live
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 in a block which is still considered clean? If so, don't obsolete it.
1204 If not, cover it anyway. */
1205
1206 struct jffs2_raw_node_ref *raw = frag->node->raw;
1207 struct jffs2_eraseblock *jeb;
1208
1209 jeb = &c->blocks[raw->flash_offset / c->sector_size];
1210
1211 if (jeb == c->gcblock) {
Joe Perches9c261b32012-02-15 15:56:43 -08001212 jffs2_dbg(1, "Expanding down to cover frag (0x%x-0x%x) in gcblock at %08x\n",
1213 frag->ofs,
1214 frag->ofs + frag->size,
1215 ref_offset(raw));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 start = frag->ofs;
1217 break;
1218 }
1219 if (!ISDIRTY(jeb->dirty_size + jeb->wasted_size)) {
Joe Perches9c261b32012-02-15 15:56:43 -08001220 jffs2_dbg(1, "Not expanding down to cover frag (0x%x-0x%x) in clean block %08x\n",
1221 frag->ofs,
1222 frag->ofs + frag->size,
1223 jeb->offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 break;
1225 }
1226
Joe Perches9c261b32012-02-15 15:56:43 -08001227 jffs2_dbg(1, "Expanding down to cover frag (0x%x-0x%x) in dirty block %08x\n",
1228 frag->ofs,
1229 frag->ofs + frag->size,
1230 jeb->offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 start = frag->ofs;
1232 break;
1233 }
1234 }
1235
1236 /* ... then up */
1237
1238 /* Find last frag which is actually part of the node we're to GC. */
1239 frag = jffs2_lookup_node_frag(&f->fragtree, end-1);
1240
1241 while((frag = frag_next(frag)) && frag->ofs+frag->size <= max) {
1242
1243 /* If the previous frag doesn't even reach the beginning, there's lots
1244 of fragmentation. Just merge. */
1245 if (frag->ofs+frag->size < max) {
Joe Perches9c261b32012-02-15 15:56:43 -08001246 jffs2_dbg(1, "Expanding up to cover partial frag (0x%x-0x%x)\n",
1247 frag->ofs, frag->ofs+frag->size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 end = frag->ofs + frag->size;
1249 continue;
1250 }
1251
1252 if (!frag->node || !frag->node->raw) {
Joe Perches9c261b32012-02-15 15:56:43 -08001253 jffs2_dbg(1, "Last frag in page is hole (0x%x-0x%x). Not expanding up.\n",
1254 frag->ofs, frag->ofs+frag->size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 break;
1256 } else {
1257
Thomas Gleixner182ec4e2005-11-07 11:16:07 +00001258 /* OK, it's a frag which extends to the beginning of the page. Does it live
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 in a block which is still considered clean? If so, don't obsolete it.
1260 If not, cover it anyway. */
1261
1262 struct jffs2_raw_node_ref *raw = frag->node->raw;
1263 struct jffs2_eraseblock *jeb;
1264
1265 jeb = &c->blocks[raw->flash_offset / c->sector_size];
1266
1267 if (jeb == c->gcblock) {
Joe Perches9c261b32012-02-15 15:56:43 -08001268 jffs2_dbg(1, "Expanding up to cover frag (0x%x-0x%x) in gcblock at %08x\n",
1269 frag->ofs,
1270 frag->ofs + frag->size,
1271 ref_offset(raw));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 end = frag->ofs + frag->size;
1273 break;
1274 }
1275 if (!ISDIRTY(jeb->dirty_size + jeb->wasted_size)) {
Joe Perches9c261b32012-02-15 15:56:43 -08001276 jffs2_dbg(1, "Not expanding up to cover frag (0x%x-0x%x) in clean block %08x\n",
1277 frag->ofs,
1278 frag->ofs + frag->size,
1279 jeb->offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 break;
1281 }
1282
Joe Perches9c261b32012-02-15 15:56:43 -08001283 jffs2_dbg(1, "Expanding up to cover frag (0x%x-0x%x) in dirty block %08x\n",
1284 frag->ofs,
1285 frag->ofs + frag->size,
1286 jeb->offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 end = frag->ofs + frag->size;
1288 break;
1289 }
1290 }
Joe Perches9c261b32012-02-15 15:56:43 -08001291 jffs2_dbg(1, "Expanded dnode to write from (0x%x-0x%x) to (0x%x-0x%x)\n",
1292 orig_start, orig_end, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293
Artem B. Bityuckiy8557fd52005-04-09 11:47:03 +01001294 D1(BUG_ON(end > frag_last(&f->fragtree)->ofs + frag_last(&f->fragtree)->size));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 BUG_ON(end < orig_end);
1296 BUG_ON(start > orig_start);
1297 }
Thomas Gleixner182ec4e2005-11-07 11:16:07 +00001298
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 /* First, use readpage() to read the appropriate page into the page cache */
1300 /* Q: What happens if we actually try to GC the _same_ page for which commit_write()
1301 * triggered garbage collection in the first place?
1302 * A: I _think_ it's OK. read_cache_page shouldn't deadlock, we'll write out the
1303 * page OK. We'll actually write it out again in commit_write, which is a little
1304 * suboptimal, but at least we're correct.
1305 */
1306 pg_ptr = jffs2_gc_fetch_page(c, f, start, &pg);
1307
1308 if (IS_ERR(pg_ptr)) {
Joe Perchesda320f02012-02-15 15:56:44 -08001309 pr_warn("read_cache_page() returned error: %ld\n",
1310 PTR_ERR(pg_ptr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 return PTR_ERR(pg_ptr);
1312 }
1313
1314 offset = start;
1315 while(offset < orig_end) {
1316 uint32_t datalen;
1317 uint32_t cdatalen;
1318 uint16_t comprtype = JFFS2_COMPR_NONE;
1319
David Woodhouse9fe48542006-05-23 00:38:06 +01001320 ret = jffs2_reserve_space_gc(c, sizeof(ri) + JFFS2_MIN_DATA_LEN,
Ferenc Havasie631ddb2005-09-07 09:35:26 +01001321 &alloclen, JFFS2_SUMMARY_INODE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322
1323 if (ret) {
Joe Perchesda320f02012-02-15 15:56:44 -08001324 pr_warn("jffs2_reserve_space_gc of %zd bytes for garbage_collect_dnode failed: %d\n",
1325 sizeof(ri) + JFFS2_MIN_DATA_LEN, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 break;
1327 }
1328 cdatalen = min_t(uint32_t, alloclen - sizeof(ri), end - offset);
1329 datalen = end - offset;
1330
1331 writebuf = pg_ptr + (offset & (PAGE_CACHE_SIZE -1));
1332
1333 comprtype = jffs2_compress(c, f, writebuf, &comprbuf, &datalen, &cdatalen);
1334
1335 ri.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK);
1336 ri.nodetype = cpu_to_je16(JFFS2_NODETYPE_INODE);
1337 ri.totlen = cpu_to_je32(sizeof(ri) + cdatalen);
1338 ri.hdr_crc = cpu_to_je32(crc32(0, &ri, sizeof(struct jffs2_unknown_node)-4));
1339
1340 ri.ino = cpu_to_je32(f->inocache->ino);
1341 ri.version = cpu_to_je32(++f->highest_version);
1342 ri.mode = cpu_to_jemode(JFFS2_F_I_MODE(f));
1343 ri.uid = cpu_to_je16(JFFS2_F_I_UID(f));
1344 ri.gid = cpu_to_je16(JFFS2_F_I_GID(f));
1345 ri.isize = cpu_to_je32(JFFS2_F_I_SIZE(f));
1346 ri.atime = cpu_to_je32(JFFS2_F_I_ATIME(f));
1347 ri.ctime = cpu_to_je32(JFFS2_F_I_CTIME(f));
1348 ri.mtime = cpu_to_je32(JFFS2_F_I_MTIME(f));
1349 ri.offset = cpu_to_je32(offset);
1350 ri.csize = cpu_to_je32(cdatalen);
1351 ri.dsize = cpu_to_je32(datalen);
1352 ri.compr = comprtype & 0xff;
1353 ri.usercompr = (comprtype >> 8) & 0xff;
1354 ri.node_crc = cpu_to_je32(crc32(0, &ri, sizeof(ri)-8));
1355 ri.data_crc = cpu_to_je32(crc32(0, comprbuf, cdatalen));
Thomas Gleixner182ec4e2005-11-07 11:16:07 +00001356
David Woodhouse9fe48542006-05-23 00:38:06 +01001357 new_fn = jffs2_write_dnode(c, f, &ri, comprbuf, cdatalen, ALLOC_GC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358
1359 jffs2_free_comprbuf(comprbuf, writebuf);
1360
1361 if (IS_ERR(new_fn)) {
Joe Perchesda320f02012-02-15 15:56:44 -08001362 pr_warn("Error writing new dnode: %ld\n",
1363 PTR_ERR(new_fn));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 ret = PTR_ERR(new_fn);
1365 break;
1366 }
1367 ret = jffs2_add_full_dnode_to_inode(c, f, new_fn);
1368 offset += datalen;
1369 if (f->metadata) {
1370 jffs2_mark_node_obsolete(c, f->metadata->raw);
1371 jffs2_free_full_dnode(f->metadata);
1372 f->metadata = NULL;
1373 }
1374 }
1375
1376 jffs2_gc_release_page(c, pg_ptr, &pg);
1377 return ret;
1378}