fs/jffs2/build.c

   1 /*
   2  * JFFS2 -- Journalling Flash File System, Version 2.
   3  *
   4  * Copyright © 2001-2007 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  */
  11
  12 #include <linux/kernel.h>
  13 #include <linux/sched.h>
  14 #include <linux/slab.h>
  15 #include <linux/vmalloc.h>
  16 #include <linux/mtd/mtd.h>
  17 #include "nodelist.h"
  18
  19 static void jffs2_build_remove_unlinked_inode(struct jffs2_sb_info *,
  20                 struct jffs2_inode_cache *, struct jffs2_full_dirent **);
  21
  22 static inline struct jffs2_inode_cache *
  23 first_inode_chain(int *i, struct jffs2_sb_info *c)
  24 {
  25         for (; *i < INOCACHE_HASHSIZE; (*i)++) {
  26                 if (c->inocache_list[*i])
  27                         return c->inocache_list[*i];
  28         }
  29         return NULL;
  30 }
  31
  32 static inline struct jffs2_inode_cache *
  33 next_inode(int *i, struct jffs2_inode_cache *ic, struct jffs2_sb_info *c)
  34 {
  35         /* More in this chain? */
  36         if (ic->next)
  37                 return ic->next;
  38         (*i)++;
  39         return first_inode_chain(i, c);
  40 }
  41
  42 #define for_each_inode(i, c, ic)                        \
  43         for (i = 0, ic = first_inode_chain(&i, (c));    \
  44              ic;                                        \
  45              ic = next_inode(&i, ic, (c)))
  46
  47
  48 static void jffs2_build_inode_pass1(struct jffs2_sb_info *c,
  49                                         struct jffs2_inode_cache *ic)
  50 {
  51         struct jffs2_full_dirent *fd;
  52
  53         dbg_fsbuild("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                 /* we can get high latency here with huge directories */
  62
  63                 child_ic = jffs2_get_ino_cache(c, fd->ino);
  64                 if (!child_ic) {
  65                         dbg_fsbuild("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                         JFFS2_ERROR("child dir \"%s\" (ino #%u) of dir ino #%u appears to be a hard link\n",
  73                                 fd->name, fd->ino, ic->ino);
  74                         /* TODO: What do we do about it? */
  75                 }
  76                 dbg_fsbuild("increased nlink for child \"%s\" (ino #%u)\n", fd->name, fd->ino);
  77                 /* Can't free scan_dents so far. We might need them in pass 2 */
  78         }
  79 }
  80
  81 /* Scan plan:
  82  - Scan physical nodes. Build map of inodes/dirents. Allocate inocaches as we go
  83  - Scan directory tree from top down, setting nlink in inocaches
  84  - Scan inocaches for inodes with nlink==0
  85 */
  86 static int jffs2_build_filesystem(struct jffs2_sb_info *c)
  87 {
  88         int ret;
  89         int i;
  90         struct jffs2_inode_cache *ic;
  91         struct jffs2_full_dirent *fd;
  92         struct jffs2_full_dirent *dead_fds = NULL;
  93
  94         dbg_fsbuild("build FS data structures\n");
  95
  96         /* First, scan the medium and build all the inode caches with
  97            lists of physical nodes */
  98
  99         c->flags |= JFFS2_SB_FLAG_SCANNING;
 100         ret = jffs2_scan_medium(c);
 101         c->flags &= ~JFFS2_SB_FLAG_SCANNING;
 102         if (ret)
 103                 goto exit;
 104
 105         dbg_fsbuild("scanned flash completely\n");
 106         jffs2_dbg_dump_block_lists_nolock(c);
 107
 108         dbg_fsbuild("pass 1 starting\n");
 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                 if (ic->scan_dents) {
 113                         jffs2_build_inode_pass1(c, ic);
 114                         cond_resched();
 115                 }
 116         }
 117
 118         dbg_fsbuild("pass 1 complete\n");
 119
 120         /* Next, scan for inodes with nlink == 0 and remove them. If
 121            they were directories, then decrement the nlink of their
 122            children too, and repeat the scan. As that's going to be
 123            a fairly uncommon occurrence, it's not so evil to do it this
 124            way. Recursion bad. */
 125         dbg_fsbuild("pass 2 starting\n");
 126
 127         for_each_inode(i, c, ic) {
 128                 if (ic->nlink)
 129                         continue;
 130
 131                 jffs2_build_remove_unlinked_inode(c, ic, &dead_fds);
 132                 cond_resched();
 133         }
 134
 135         dbg_fsbuild("pass 2a starting\n");
 136
 137         while (dead_fds) {
 138                 fd = dead_fds;
 139                 dead_fds = fd->next;
 140
 141                 ic = jffs2_get_ino_cache(c, fd->ino);
 142
 143                 if (ic)
 144                         jffs2_build_remove_unlinked_inode(c, ic, &dead_fds);
 145                 jffs2_free_full_dirent(fd);
 146         }
 147
 148         dbg_fsbuild("pass 2a complete\n");
 149         dbg_fsbuild("freeing temporary data structures\n");
 150
 151         /* Finally, we can scan again and free the dirent structs */
 152         for_each_inode(i, c, ic) {
 153                 while(ic->scan_dents) {
 154                         fd = ic->scan_dents;
 155                         ic->scan_dents = fd->next;
 156                         jffs2_free_full_dirent(fd);
 157                 }
 158                 ic->scan_dents = NULL;
 159                 cond_resched();
 160         }
 161         jffs2_build_xattr_subsystem(c);
 162         c->flags &= ~JFFS2_SB_FLAG_BUILDING;
 163
 164         dbg_fsbuild("FS build complete\n");
 165
 166         /* Rotate the lists by some number to ensure wear levelling */
 167         jffs2_rotate_lists(c);
 168
 169         ret = 0;
 170
 171 exit:
 172         if (ret) {
 173                 for_each_inode(i, c, ic) {
 174                         while(ic->scan_dents) {
 175                                 fd = ic->scan_dents;
 176                                 ic->scan_dents = fd->next;
 177                                 jffs2_free_full_dirent(fd);
 178                         }
 179                 }
 180                 jffs2_clear_xattr_subsystem(c);
 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         /* What number of 'very dirty' eraseblocks do we allow before we
 289            trigger the GC thread even if we don't _need_ the space. When we
 290            can't mark nodes obsolete on the medium, the old dirty nodes cause
 291            performance problems because we have to inspect and discard them. */
 292         c->vdirty_blocks_gctrigger = c->resv_blocks_gctrigger;
 293         if (jffs2_can_mark_obsolete(c))
 294                 c->vdirty_blocks_gctrigger *= 10;
 295
 296         /* If there's less than this amount of dirty space, don't bother
 297            trying to GC to make more space. It'll be a fruitless task */
 298         c->nospc_dirty_size = c->sector_size + (c->flash_size / 100);
 299
 300         dbg_fsbuild("JFFS2 trigger levels (size %d KiB, block size %d KiB, %d blocks)\n",
 301                   c->flash_size / 1024, c->sector_size / 1024, c->nr_blocks);
 302         dbg_fsbuild("Blocks required to allow deletion:    %d (%d KiB)\n",
 303                   c->resv_blocks_deletion, c->resv_blocks_deletion*c->sector_size/1024);
 304         dbg_fsbuild("Blocks required to allow writes:      %d (%d KiB)\n",
 305                   c->resv_blocks_write, c->resv_blocks_write*c->sector_size/1024);
 306         dbg_fsbuild("Blocks required to quiesce GC thread: %d (%d KiB)\n",
 307                   c->resv_blocks_gctrigger, c->resv_blocks_gctrigger*c->sector_size/1024);
 308         dbg_fsbuild("Blocks required to allow GC merges:   %d (%d KiB)\n",
 309                   c->resv_blocks_gcmerge, c->resv_blocks_gcmerge*c->sector_size/1024);
 310         dbg_fsbuild("Blocks required to GC bad blocks:     %d (%d KiB)\n",
 311                   c->resv_blocks_gcbad, c->resv_blocks_gcbad*c->sector_size/1024);
 312         dbg_fsbuild("Amount of dirty space required to GC: %d bytes\n",
 313                   c->nospc_dirty_size);
 314         dbg_fsbuild("Very dirty blocks before GC triggered: %d\n",
 315                   c->vdirty_blocks_gctrigger);
 316 }
 317
 318 int jffs2_do_mount_fs(struct jffs2_sb_info *c)
 319 {
 320         int ret;
 321         int i;
 322         int size;
 323
 324         c->free_size = c->flash_size;
 325         c->nr_blocks = c->flash_size / c->sector_size;
 326         size = sizeof(struct jffs2_eraseblock) * c->nr_blocks;
 327 #ifndef __ECOS
 328         if (jffs2_blocks_use_vmalloc(c))
 329                 c->blocks = vmalloc(size);
 330         else
 331 #endif
 332                 c->blocks = kmalloc(size, GFP_KERNEL);
 333         if (!c->blocks)
 334                 return -ENOMEM;
 335
 336         memset(c->blocks, 0, size);
 337         for (i=0; i<c->nr_blocks; i++) {
 338                 INIT_LIST_HEAD(&c->blocks[i].list);
 339                 c->blocks[i].offset = i * c->sector_size;
 340                 c->blocks[i].free_size = c->sector_size;
 341         }
 342
 343         INIT_LIST_HEAD(&c->clean_list);
 344         INIT_LIST_HEAD(&c->very_dirty_list);
 345         INIT_LIST_HEAD(&c->dirty_list);
 346         INIT_LIST_HEAD(&c->erasable_list);
 347         INIT_LIST_HEAD(&c->erasing_list);
 348         INIT_LIST_HEAD(&c->erase_pending_list);
 349         INIT_LIST_HEAD(&c->erasable_pending_wbuf_list);
 350         INIT_LIST_HEAD(&c->erase_complete_list);
 351         INIT_LIST_HEAD(&c->free_list);
 352         INIT_LIST_HEAD(&c->bad_list);
 353         INIT_LIST_HEAD(&c->bad_used_list);
 354         c->highest_ino = 1;
 355         c->summary = NULL;
 356
 357         ret = jffs2_sum_init(c);
 358         if (ret)
 359                 goto out_free;
 360
 361         if (jffs2_build_filesystem(c)) {
 362                 dbg_fsbuild("build_fs failed\n");
 363                 jffs2_free_ino_caches(c);
 364                 jffs2_free_raw_node_refs(c);
 365                 ret = -EIO;
 366                 goto out_free;
 367         }
 368
 369         jffs2_calc_trigger_levels(c);
 370
 371         return 0;
 372
 373  out_free:
 374 #ifndef __ECOS
 375         if (jffs2_blocks_use_vmalloc(c))
 376                 vfree(c->blocks);
 377         else
 378 #endif
 379                 kfree(c->blocks);
 380
 381         return ret;
 382 }