Merge master.kernel.org:/home/rmk/linux-2.6-arm
[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 1.85 2005/11/07 11:14:38 gleixner 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 *,
22                 struct jffs2_inode_cache *, struct jffs2_full_dirent **);
23
24 static inline struct jffs2_inode_cache *
25 first_inode_chain(int *i, struct jffs2_sb_info *c)
26 {
27         for (; *i < INOCACHE_HASHSIZE; (*i)++) {
28                 if (c->inocache_list[*i])
29                         return c->inocache_list[*i];
30         }
31         return NULL;
32 }
33
34 static inline struct jffs2_inode_cache *
35 next_inode(int *i, struct jffs2_inode_cache *ic, struct jffs2_sb_info *c)
36 {
37         /* More in this chain? */
38         if (ic->next)
39                 return ic->next;
40         (*i)++;
41         return first_inode_chain(i, c);
42 }
43
44 #define for_each_inode(i, c, ic)                        \
45         for (i = 0, ic = first_inode_chain(&i, (c));    \
46              ic;                                        \
47              ic = next_inode(&i, ic, (c)))
48
49
50 static void jffs2_build_inode_pass1(struct jffs2_sb_info *c,
51                                         struct jffs2_inode_cache *ic)
52 {
53         struct jffs2_full_dirent *fd;
54
55         dbg_fsbuild("building directory inode #%u\n", ic->ino);
56
57         /* For each child, increase nlink */
58         for(fd = ic->scan_dents; fd; fd = fd->next) {
59                 struct jffs2_inode_cache *child_ic;
60                 if (!fd->ino)
61                         continue;
62
63                 /* we can get high latency here with huge directories */
64
65                 child_ic = jffs2_get_ino_cache(c, fd->ino);
66                 if (!child_ic) {
67                         dbg_fsbuild("child \"%s\" (ino #%u) of dir ino #%u doesn't exist!\n",
68                                   fd->name, fd->ino, ic->ino);
69                         jffs2_mark_node_obsolete(c, fd->raw);
70                         continue;
71                 }
72
73                 if (child_ic->nlink++ && fd->type == DT_DIR) {
74                         JFFS2_ERROR("child dir \"%s\" (ino #%u) of dir ino #%u appears to be a hard link\n",
75                                 fd->name, fd->ino, ic->ino);
76                         /* TODO: What do we do about it? */
77                 }
78                 dbg_fsbuild("increased nlink for child \"%s\" (ino #%u)\n", fd->name, fd->ino);
79                 /* Can't free scan_dents so far. We might need them in pass 2 */
80         }
81 }
82
83 /* Scan plan:
84  - Scan physical nodes. Build map of inodes/dirents. Allocate inocaches as we go
85  - Scan directory tree from top down, setting nlink in inocaches
86  - Scan inocaches for inodes with nlink==0
87 */
88 static int jffs2_build_filesystem(struct jffs2_sb_info *c)
89 {
90         int ret;
91         int i;
92         struct jffs2_inode_cache *ic;
93         struct jffs2_full_dirent *fd;
94         struct jffs2_full_dirent *dead_fds = NULL;
95
96         dbg_fsbuild("build FS data structures\n");
97
98         /* First, scan the medium and build all the inode caches with
99            lists of physical nodes */
100
101         c->flags |= JFFS2_SB_FLAG_SCANNING;
102         ret = jffs2_scan_medium(c);
103         c->flags &= ~JFFS2_SB_FLAG_SCANNING;
104         if (ret)
105                 goto exit;
106
107         dbg_fsbuild("scanned flash completely\n");
108         jffs2_dbg_dump_block_lists_nolock(c);
109
110         dbg_fsbuild("pass 1 starting\n");
111         c->flags |= JFFS2_SB_FLAG_BUILDING;
112         /* Now scan the directory tree, increasing nlink according to every dirent found. */
113         for_each_inode(i, c, ic) {
114                 if (ic->scan_dents) {
115                         jffs2_build_inode_pass1(c, ic);
116                         cond_resched();
117                 }
118         }
119
120         dbg_fsbuild("pass 1 complete\n");
121
122         /* Next, scan for inodes with nlink == 0 and remove them. If
123            they were directories, then decrement the nlink of their
124            children too, and repeat the scan. As that's going to be
125            a fairly uncommon occurrence, it's not so evil to do it this
126            way. Recursion bad. */
127         dbg_fsbuild("pass 2 starting\n");
128
129         for_each_inode(i, c, ic) {
130                 if (ic->nlink)
131                         continue;
132
133                 jffs2_build_remove_unlinked_inode(c, ic, &dead_fds);
134                 cond_resched();
135         }
136
137         dbg_fsbuild("pass 2a starting\n");
138
139         while (dead_fds) {
140                 fd = dead_fds;
141                 dead_fds = fd->next;
142
143                 ic = jffs2_get_ino_cache(c, fd->ino);
144
145                 if (ic)
146                         jffs2_build_remove_unlinked_inode(c, ic, &dead_fds);
147                 jffs2_free_full_dirent(fd);
148         }
149
150         dbg_fsbuild("pass 2a complete\n");
151         dbg_fsbuild("freeing temporary data structures\n");
152
153         /* Finally, we can scan again and free the dirent structs */
154         for_each_inode(i, c, ic) {
155                 while(ic->scan_dents) {
156                         fd = ic->scan_dents;
157                         ic->scan_dents = fd->next;
158                         jffs2_free_full_dirent(fd);
159                 }
160                 ic->scan_dents = NULL;
161                 cond_resched();
162         }
163         c->flags &= ~JFFS2_SB_FLAG_BUILDING;
164
165         dbg_fsbuild("FS build complete\n");
166
167         /* Rotate the lists by some number to ensure wear levelling */
168         jffs2_rotate_lists(c);
169
170         ret = 0;
171
172 exit:
173         if (ret) {
174                 for_each_inode(i, c, ic) {
175                         while(ic->scan_dents) {
176                                 fd = ic->scan_dents;
177                                 ic->scan_dents = fd->next;
178                                 jffs2_free_full_dirent(fd);
179                         }
180                 }
181         }
182
183         return ret;
184 }
185
186 static void jffs2_build_remove_unlinked_inode(struct jffs2_sb_info *c,
187                                         struct jffs2_inode_cache *ic,
188                                         struct jffs2_full_dirent **dead_fds)
189 {
190         struct jffs2_raw_node_ref *raw;
191         struct jffs2_full_dirent *fd;
192
193         dbg_fsbuild("removing ino #%u with nlink == zero.\n", ic->ino);
194
195         raw = ic->nodes;
196         while (raw != (void *)ic) {
197                 struct jffs2_raw_node_ref *next = raw->next_in_ino;
198                 dbg_fsbuild("obsoleting node at 0x%08x\n", ref_offset(raw));
199                 jffs2_mark_node_obsolete(c, raw);
200                 raw = next;
201         }
202
203         if (ic->scan_dents) {
204                 int whinged = 0;
205                 dbg_fsbuild("inode #%u was a directory which may have children...\n", ic->ino);
206
207                 while(ic->scan_dents) {
208                         struct jffs2_inode_cache *child_ic;
209
210                         fd = ic->scan_dents;
211                         ic->scan_dents = fd->next;
212
213                         if (!fd->ino) {
214                                 /* It's a deletion dirent. Ignore it */
215                                 dbg_fsbuild("child \"%s\" is a deletion dirent, skipping...\n", fd->name);
216                                 jffs2_free_full_dirent(fd);
217                                 continue;
218                         }
219                         if (!whinged)
220                                 whinged = 1;
221
222                         dbg_fsbuild("removing child \"%s\", ino #%u\n", fd->name, fd->ino);
223
224                         child_ic = jffs2_get_ino_cache(c, fd->ino);
225                         if (!child_ic) {
226                                 dbg_fsbuild("cannot remove child \"%s\", ino #%u, because it doesn't exist\n",
227                                                 fd->name, fd->ino);
228                                 jffs2_free_full_dirent(fd);
229                                 continue;
230                         }
231
232                         /* Reduce nlink of the child. If it's now zero, stick it on the
233                            dead_fds list to be cleaned up later. Else just free the fd */
234
235                         child_ic->nlink--;
236
237                         if (!child_ic->nlink) {
238                                 dbg_fsbuild("inode #%u (\"%s\") has now got zero nlink, adding to dead_fds list.\n",
239                                           fd->ino, fd->name);
240                                 fd->next = *dead_fds;
241                                 *dead_fds = fd;
242                         } else {
243                                 dbg_fsbuild("inode #%u (\"%s\") has now got nlink %d. Ignoring.\n",
244                                           fd->ino, fd->name, child_ic->nlink);
245                                 jffs2_free_full_dirent(fd);
246                         }
247                 }
248         }
249
250         /*
251            We don't delete the inocache from the hash list and free it yet.
252            The erase code will do that, when all the nodes are completely gone.
253         */
254 }
255
256 static void jffs2_calc_trigger_levels(struct jffs2_sb_info *c)
257 {
258         uint32_t size;
259
260         /* Deletion should almost _always_ be allowed. We're fairly
261            buggered once we stop allowing people to delete stuff
262            because there's not enough free space... */
263         c->resv_blocks_deletion = 2;
264
265         /* Be conservative about how much space we need before we allow writes.
266            On top of that which is required for deletia, require an extra 2%
267            of the medium to be available, for overhead caused by nodes being
268            split across blocks, etc. */
269
270         size = c->flash_size / 50; /* 2% of flash size */
271         size += c->nr_blocks * 100; /* And 100 bytes per eraseblock */
272         size += c->sector_size - 1; /* ... and round up */
273
274         c->resv_blocks_write = c->resv_blocks_deletion + (size / c->sector_size);
275
276         /* When do we let the GC thread run in the background */
277
278         c->resv_blocks_gctrigger = c->resv_blocks_write + 1;
279
280         /* When do we allow garbage collection to merge nodes to make
281            long-term progress at the expense of short-term space exhaustion? */
282         c->resv_blocks_gcmerge = c->resv_blocks_deletion + 1;
283
284         /* When do we allow garbage collection to eat from bad blocks rather
285            than actually making progress? */
286         c->resv_blocks_gcbad = 0;//c->resv_blocks_deletion + 2;
287
288         /* If there's less than this amount of dirty space, don't bother
289            trying to GC to make more space. It'll be a fruitless task */
290         c->nospc_dirty_size = c->sector_size + (c->flash_size / 100);
291
292         dbg_fsbuild("JFFS2 trigger levels (size %d KiB, block size %d KiB, %d blocks)\n",
293                   c->flash_size / 1024, c->sector_size / 1024, c->nr_blocks);
294         dbg_fsbuild("Blocks required to allow deletion:    %d (%d KiB)\n",
295                   c->resv_blocks_deletion, c->resv_blocks_deletion*c->sector_size/1024);
296         dbg_fsbuild("Blocks required to allow writes:      %d (%d KiB)\n",
297                   c->resv_blocks_write, c->resv_blocks_write*c->sector_size/1024);
298         dbg_fsbuild("Blocks required to quiesce GC thread: %d (%d KiB)\n",
299                   c->resv_blocks_gctrigger, c->resv_blocks_gctrigger*c->sector_size/1024);
300         dbg_fsbuild("Blocks required to allow GC merges:   %d (%d KiB)\n",
301                   c->resv_blocks_gcmerge, c->resv_blocks_gcmerge*c->sector_size/1024);
302         dbg_fsbuild("Blocks required to GC bad blocks:     %d (%d KiB)\n",
303                   c->resv_blocks_gcbad, c->resv_blocks_gcbad*c->sector_size/1024);
304         dbg_fsbuild("Amount of dirty space required to GC: %d bytes\n",
305                   c->nospc_dirty_size);
306 }
307
308 int jffs2_do_mount_fs(struct jffs2_sb_info *c)
309 {
310         int ret;
311         int i;
312         int size;
313
314         c->free_size = c->flash_size;
315         c->nr_blocks = c->flash_size / c->sector_size;
316         size = sizeof(struct jffs2_eraseblock) * c->nr_blocks;
317 #ifndef __ECOS
318         if (jffs2_blocks_use_vmalloc(c))
319                 c->blocks = vmalloc(size);
320         else
321 #endif
322                 c->blocks = kmalloc(size, GFP_KERNEL);
323         if (!c->blocks)
324                 return -ENOMEM;
325
326         memset(c->blocks, 0, size);
327         for (i=0; i<c->nr_blocks; i++) {
328                 INIT_LIST_HEAD(&c->blocks[i].list);
329                 c->blocks[i].offset = i * c->sector_size;
330                 c->blocks[i].free_size = c->sector_size;
331         }
332
333         INIT_LIST_HEAD(&c->clean_list);
334         INIT_LIST_HEAD(&c->very_dirty_list);
335         INIT_LIST_HEAD(&c->dirty_list);
336         INIT_LIST_HEAD(&c->erasable_list);
337         INIT_LIST_HEAD(&c->erasing_list);
338         INIT_LIST_HEAD(&c->erase_pending_list);
339         INIT_LIST_HEAD(&c->erasable_pending_wbuf_list);
340         INIT_LIST_HEAD(&c->erase_complete_list);
341         INIT_LIST_HEAD(&c->free_list);
342         INIT_LIST_HEAD(&c->bad_list);
343         INIT_LIST_HEAD(&c->bad_used_list);
344         c->highest_ino = 1;
345         c->summary = NULL;
346
347         ret = jffs2_sum_init(c);
348         if (ret)
349                 return ret;
350
351         if (jffs2_build_filesystem(c)) {
352                 dbg_fsbuild("build_fs failed\n");
353                 jffs2_free_ino_caches(c);
354                 jffs2_free_raw_node_refs(c);
355 #ifndef __ECOS
356                 if (jffs2_blocks_use_vmalloc(c))
357                         vfree(c->blocks);
358                 else
359 #endif
360                         kfree(c->blocks);
361
362                 return -EIO;
363         }
364
365         jffs2_calc_trigger_levels(c);
366
367         return 0;
368 }