[JFFS2] Build fragtree in reverse order
[linux-2.6] / fs / jffs2 / build.c
1 /*
2  * JFFS2 -- Journalling Flash File System, Version 2.
3  *
4  * Copyright (C) 2001-2003 Red Hat, Inc.
5  *
6  * Created by David Woodhouse <dwmw2@infradead.org>
7  *
8  * For licensing information, see the file 'LICENCE' in this directory.
9  *
10  * $Id: build.c,v 1.76 2005/07/30 15:29:27 lunn Exp $
11  *
12  */
13
14 #include <linux/kernel.h>
15 #include <linux/sched.h>
16 #include <linux/slab.h>
17 #include <linux/vmalloc.h>
18 #include <linux/mtd/mtd.h>
19 #include "nodelist.h"
20
21 static void jffs2_build_remove_unlinked_inode(struct jffs2_sb_info *, struct jffs2_inode_cache *, struct jffs2_full_dirent **);
22
23 static inline struct jffs2_inode_cache *
24 first_inode_chain(int *i, struct jffs2_sb_info *c)
25 {
26         for (; *i < INOCACHE_HASHSIZE; (*i)++) {
27                 if (c->inocache_list[*i])
28                         return c->inocache_list[*i];
29         }
30         return NULL;
31 }
32
33 static inline struct jffs2_inode_cache *
34 next_inode(int *i, struct jffs2_inode_cache *ic, struct jffs2_sb_info *c)
35 {
36         /* More in this chain? */
37         if (ic->next)
38                 return ic->next;
39         (*i)++;
40         return first_inode_chain(i, c);
41 }
42
43 #define for_each_inode(i, c, ic)                        \
44         for (i = 0, ic = first_inode_chain(&i, (c));    \
45              ic;                                        \
46              ic = next_inode(&i, ic, (c)))
47
48
49 static inline void jffs2_build_inode_pass1(struct jffs2_sb_info *c, struct jffs2_inode_cache *ic)
50 {
51         struct jffs2_full_dirent *fd;
52
53         D1(printk(KERN_DEBUG "jffs2_build_inode building directory inode #%u\n", ic->ino));
54
55         /* For each child, increase nlink */
56         for(fd = ic->scan_dents; fd; fd = fd->next) {
57                 struct jffs2_inode_cache *child_ic;
58                 if (!fd->ino)
59                         continue;
60
61                 /* XXX: Can get high latency here with huge directories */
62
63                 child_ic = jffs2_get_ino_cache(c, fd->ino);
64                 if (!child_ic) {
65                         printk(KERN_NOTICE "Eep. Child \"%s\" (ino #%u) of dir ino #%u doesn't exist!\n",
66                                   fd->name, fd->ino, ic->ino);
67                         jffs2_mark_node_obsolete(c, fd->raw);
68                         continue;
69                 }
70
71                 if (child_ic->nlink++ && fd->type == DT_DIR) {
72                         printk(KERN_NOTICE "Child dir \"%s\" (ino #%u) of dir ino #%u appears to be a hard link\n", fd->name, fd->ino, ic->ino);
73                         if (fd->ino == 1 && ic->ino == 1) {
74                                 printk(KERN_NOTICE "This is mostly harmless, and probably caused by creating a JFFS2 image\n");
75                                 printk(KERN_NOTICE "using a buggy version of mkfs.jffs2. Use at least v1.17.\n");
76                         }
77                         /* What do we do about it? */
78                 }
79                 D1(printk(KERN_DEBUG "Increased nlink for child \"%s\" (ino #%u)\n", fd->name, fd->ino));
80                 /* Can't free them. We might need them in pass 2 */
81         }
82 }
83
84 /* Scan plan:
85  - Scan physical nodes. Build map of inodes/dirents. Allocate inocaches as we go
86  - Scan directory tree from top down, setting nlink in inocaches
87  - Scan inocaches for inodes with nlink==0
88 */
89 static int jffs2_build_filesystem(struct jffs2_sb_info *c)
90 {
91         int ret;
92         int i;
93         struct jffs2_inode_cache *ic;
94         struct jffs2_full_dirent *fd;
95         struct jffs2_full_dirent *dead_fds = NULL;
96
97         /* First, scan the medium and build all the inode caches with
98            lists of physical nodes */
99
100         c->flags |= JFFS2_SB_FLAG_SCANNING;
101         ret = jffs2_scan_medium(c);
102         c->flags &= ~JFFS2_SB_FLAG_SCANNING;
103         if (ret)
104                 goto exit;
105
106         D1(printk(KERN_DEBUG "Scanned flash completely\n"));
107         jffs2_dbg_dump_block_lists_nolock(c);
108
109         c->flags |= JFFS2_SB_FLAG_BUILDING;
110         /* Now scan the directory tree, increasing nlink according to every dirent found. */
111         for_each_inode(i, c, ic) {
112                 D1(printk(KERN_DEBUG "Pass 1: ino #%u\n", ic->ino));
113
114                 D1(BUG_ON(ic->ino > c->highest_ino));
115
116                 if (ic->scan_dents) {
117                         jffs2_build_inode_pass1(c, ic);
118                         cond_resched();
119                 }
120         }
121
122         D1(printk(KERN_DEBUG "Pass 1 complete\n"));
123
124         /* Next, scan for inodes with nlink == 0 and remove them. If
125            they were directories, then decrement the nlink of their
126            children too, and repeat the scan. As that's going to be
127            a fairly uncommon occurrence, it's not so evil to do it this
128            way. Recursion bad. */
129         D1(printk(KERN_DEBUG "Pass 2 starting\n"));
130
131         for_each_inode(i, c, ic) {
132                 D1(printk(KERN_DEBUG "Pass 2: ino #%u, nlink %d, ic %p, nodes %p\n", ic->ino, ic->nlink, ic, ic->nodes));
133                 if (ic->nlink)
134                         continue;
135                         
136                 jffs2_build_remove_unlinked_inode(c, ic, &dead_fds);
137                 cond_resched();
138         } 
139
140         D1(printk(KERN_DEBUG "Pass 2a starting\n"));
141
142         while (dead_fds) {
143                 fd = dead_fds;
144                 dead_fds = fd->next;
145
146                 ic = jffs2_get_ino_cache(c, fd->ino);
147                 D1(printk(KERN_DEBUG "Removing dead_fd ino #%u (\"%s\"), ic at %p\n", fd->ino, fd->name, ic));
148
149                 if (ic)
150                         jffs2_build_remove_unlinked_inode(c, ic, &dead_fds);
151                 jffs2_free_full_dirent(fd);
152         }
153
154         D1(printk(KERN_DEBUG "Pass 2 complete\n"));
155         
156         /* Finally, we can scan again and free the dirent structs */
157         for_each_inode(i, c, ic) {
158                 D1(printk(KERN_DEBUG "Pass 3: ino #%u, ic %p, nodes %p\n", ic->ino, ic, ic->nodes));
159
160                 while(ic->scan_dents) {
161                         fd = ic->scan_dents;
162                         ic->scan_dents = fd->next;
163                         jffs2_free_full_dirent(fd);
164                 }
165                 ic->scan_dents = NULL;
166                 cond_resched();
167         }
168         c->flags &= ~JFFS2_SB_FLAG_BUILDING;
169         
170         D1(printk(KERN_DEBUG "Pass 3 complete\n"));
171         jffs2_dbg_dump_block_lists_nolock(c);
172
173         /* Rotate the lists by some number to ensure wear levelling */
174         jffs2_rotate_lists(c);
175
176         ret = 0;
177
178 exit:
179         if (ret) {
180                 for_each_inode(i, c, ic) {
181                         while(ic->scan_dents) {
182                                 fd = ic->scan_dents;
183                                 ic->scan_dents = fd->next;
184                                 jffs2_free_full_dirent(fd);
185                         }
186                 }
187         }
188
189         return ret;
190 }
191
192 static void jffs2_build_remove_unlinked_inode(struct jffs2_sb_info *c, struct jffs2_inode_cache *ic, struct jffs2_full_dirent **dead_fds)
193 {
194         struct jffs2_raw_node_ref *raw;
195         struct jffs2_full_dirent *fd;
196
197         D1(printk(KERN_DEBUG "JFFS2: Removing ino #%u with nlink == zero.\n", ic->ino));
198         
199         raw = ic->nodes;
200         while (raw != (void *)ic) {
201                 struct jffs2_raw_node_ref *next = raw->next_in_ino;
202                 D1(printk(KERN_DEBUG "obsoleting node at 0x%08x\n", ref_offset(raw)));
203                 jffs2_mark_node_obsolete(c, raw);
204                 raw = next;
205         }
206
207         if (ic->scan_dents) {
208                 int whinged = 0;
209                 D1(printk(KERN_DEBUG "Inode #%u was a directory which may have children...\n", ic->ino));
210
211                 while(ic->scan_dents) {
212                         struct jffs2_inode_cache *child_ic;
213
214                         fd = ic->scan_dents;
215                         ic->scan_dents = fd->next;
216
217                         if (!fd->ino) {
218                                 /* It's a deletion dirent. Ignore it */
219                                 D1(printk(KERN_DEBUG "Child \"%s\" is a deletion dirent, skipping...\n", fd->name));
220                                 jffs2_free_full_dirent(fd);
221                                 continue;
222                         }
223                         if (!whinged) {
224                                 whinged = 1;
225                                 printk(KERN_NOTICE "Inode #%u was a directory with children - removing those too...\n", ic->ino);
226                         }
227
228                         D1(printk(KERN_DEBUG "Removing child \"%s\", ino #%u\n",
229                                   fd->name, fd->ino));
230                         
231                         child_ic = jffs2_get_ino_cache(c, fd->ino);
232                         if (!child_ic) {
233                                 printk(KERN_NOTICE "Cannot remove child \"%s\", ino #%u, because it doesn't exist\n", fd->name, fd->ino);
234                                 jffs2_free_full_dirent(fd);
235                                 continue;
236                         }
237
238                         /* Reduce nlink of the child. If it's now zero, stick it on the 
239                            dead_fds list to be cleaned up later. Else just free the fd */
240
241                         child_ic->nlink--;
242                         
243                         if (!child_ic->nlink) {
244                                 D1(printk(KERN_DEBUG "Inode #%u (\"%s\") has now got zero nlink. Adding to dead_fds list.\n",
245                                           fd->ino, fd->name));
246                                 fd->next = *dead_fds;
247                                 *dead_fds = fd;
248                         } else {
249                                 D1(printk(KERN_DEBUG "Inode #%u (\"%s\") has now got nlink %d. Ignoring.\n",
250                                           fd->ino, fd->name, child_ic->nlink));
251                                 jffs2_free_full_dirent(fd);
252                         }
253                 }
254         }
255
256         /*
257            We don't delete the inocache from the hash list and free it yet. 
258            The erase code will do that, when all the nodes are completely gone.
259         */
260 }
261
262 static void jffs2_calc_trigger_levels(struct jffs2_sb_info *c)
263 {
264         uint32_t size;
265
266         /* Deletion should almost _always_ be allowed. We're fairly
267            buggered once we stop allowing people to delete stuff
268            because there's not enough free space... */
269         c->resv_blocks_deletion = 2;
270
271         /* Be conservative about how much space we need before we allow writes. 
272            On top of that which is required for deletia, require an extra 2%
273            of the medium to be available, for overhead caused by nodes being
274            split across blocks, etc. */
275
276         size = c->flash_size / 50; /* 2% of flash size */
277         size += c->nr_blocks * 100; /* And 100 bytes per eraseblock */
278         size += c->sector_size - 1; /* ... and round up */
279
280         c->resv_blocks_write = c->resv_blocks_deletion + (size / c->sector_size);
281
282         /* When do we let the GC thread run in the background */
283
284         c->resv_blocks_gctrigger = c->resv_blocks_write + 1;
285
286         /* When do we allow garbage collection to merge nodes to make 
287            long-term progress at the expense of short-term space exhaustion? */
288         c->resv_blocks_gcmerge = c->resv_blocks_deletion + 1;
289
290         /* When do we allow garbage collection to eat from bad blocks rather
291            than actually making progress? */
292         c->resv_blocks_gcbad = 0;//c->resv_blocks_deletion + 2;
293
294         /* If there's less than this amount of dirty space, don't bother
295            trying to GC to make more space. It'll be a fruitless task */
296         c->nospc_dirty_size = c->sector_size + (c->flash_size / 100);
297
298         D1(printk(KERN_DEBUG "JFFS2 trigger levels (size %d KiB, block size %d KiB, %d blocks)\n",
299                   c->flash_size / 1024, c->sector_size / 1024, c->nr_blocks));
300         D1(printk(KERN_DEBUG "Blocks required to allow deletion:    %d (%d KiB)\n",
301                   c->resv_blocks_deletion, c->resv_blocks_deletion*c->sector_size/1024));
302         D1(printk(KERN_DEBUG "Blocks required to allow writes:      %d (%d KiB)\n",
303                   c->resv_blocks_write, c->resv_blocks_write*c->sector_size/1024));
304         D1(printk(KERN_DEBUG "Blocks required to quiesce GC thread: %d (%d KiB)\n",
305                   c->resv_blocks_gctrigger, c->resv_blocks_gctrigger*c->sector_size/1024));
306         D1(printk(KERN_DEBUG "Blocks required to allow GC merges:   %d (%d KiB)\n",
307                   c->resv_blocks_gcmerge, c->resv_blocks_gcmerge*c->sector_size/1024));
308         D1(printk(KERN_DEBUG "Blocks required to GC bad blocks:     %d (%d KiB)\n",
309                   c->resv_blocks_gcbad, c->resv_blocks_gcbad*c->sector_size/1024));
310         D1(printk(KERN_DEBUG "Amount of dirty space required to GC: %d bytes\n",
311                   c->nospc_dirty_size));
312
313
314 int jffs2_do_mount_fs(struct jffs2_sb_info *c)
315 {
316         int i;
317
318         c->free_size = c->flash_size;
319         c->nr_blocks = c->flash_size / c->sector_size;
320 #ifndef __ECOS
321         if (c->mtd->flags & MTD_NO_VIRTBLOCKS)
322                 c->blocks = vmalloc(sizeof(struct jffs2_eraseblock) * c->nr_blocks);
323         else
324 #endif
325                 c->blocks = kmalloc(sizeof(struct jffs2_eraseblock) * c->nr_blocks, GFP_KERNEL);
326         if (!c->blocks)
327                 return -ENOMEM;
328         for (i=0; i<c->nr_blocks; i++) {
329                 INIT_LIST_HEAD(&c->blocks[i].list);
330                 c->blocks[i].offset = i * c->sector_size;
331                 c->blocks[i].free_size = c->sector_size;
332                 c->blocks[i].dirty_size = 0;
333                 c->blocks[i].wasted_size = 0;
334                 c->blocks[i].unchecked_size = 0;
335                 c->blocks[i].used_size = 0;
336                 c->blocks[i].first_node = NULL;
337                 c->blocks[i].last_node = NULL;
338                 c->blocks[i].bad_count = 0;
339         }
340
341         INIT_LIST_HEAD(&c->clean_list);
342         INIT_LIST_HEAD(&c->very_dirty_list);
343         INIT_LIST_HEAD(&c->dirty_list);
344         INIT_LIST_HEAD(&c->erasable_list);
345         INIT_LIST_HEAD(&c->erasing_list);
346         INIT_LIST_HEAD(&c->erase_pending_list);
347         INIT_LIST_HEAD(&c->erasable_pending_wbuf_list);
348         INIT_LIST_HEAD(&c->erase_complete_list);
349         INIT_LIST_HEAD(&c->free_list);
350         INIT_LIST_HEAD(&c->bad_list);
351         INIT_LIST_HEAD(&c->bad_used_list);
352         c->highest_ino = 1;
353
354         if (jffs2_build_filesystem(c)) {
355                 D1(printk(KERN_DEBUG "build_fs failed\n"));
356                 jffs2_free_ino_caches(c);
357                 jffs2_free_raw_node_refs(c);
358 #ifndef __ECOS
359                 if (c->mtd->flags & MTD_NO_VIRTBLOCKS) 
360                     vfree(c->blocks);
361                 else 
362 #endif
363                     kfree(c->blocks);
364                 
365                 return -EIO;
366         }
367
368         jffs2_calc_trigger_levels(c);
369
370         return 0;
371 }