fs/jffs2/build.c

   1 /*
   2  * JFFS2 -- Journalling Flash File System, Version 2.
   3  *
   4  * Copyright © 2001-2007 Red Hat, Inc.
   5  * Copyright © 2004-2010 David Woodhouse <dwmw2@infradead.org>
   6  *
   7  * Created by David Woodhouse <dwmw2@infradead.org>
   8  *
   9  * For licensing information, see the file 'LICENCE' in this directory.
  10  *
  11  */
  12
  13 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  14
  15 #include <linux/kernel.h>
  16 #include <linux/sched.h>
  17 #include <linux/slab.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/mtd/mtd.h>
  20 #include <linux/mm.h> /* kvfree() */
  21 #include "nodelist.h"
  22
  23 static void jffs2_build_remove_unlinked_inode(struct jffs2_sb_info *,
  24                 struct jffs2_inode_cache *, struct jffs2_full_dirent **);
  25
  26 static inline struct jffs2_inode_cache *
  27 first_inode_chain(int *i, struct jffs2_sb_info *c)
  28 {
  29         for (; *i < c->inocache_hashsize; (*i)++) {
  30                 if (c->inocache_list[*i])
  31                         return c->inocache_list[*i];
  32         }
  33         return NULL;
  34 }
  35
  36 static inline struct jffs2_inode_cache *
  37 next_inode(int *i, struct jffs2_inode_cache *ic, struct jffs2_sb_info *c)
  38 {
  39         /* More in this chain? */
  40         if (ic->next)
  41                 return ic->next;
  42         (*i)++;
  43         return first_inode_chain(i, c);
  44 }
  45
  46 #define for_each_inode(i, c, ic)                        \
  47         for (i = 0, ic = first_inode_chain(&i, (c));    \
  48              ic;                                        \
  49              ic = next_inode(&i, ic, (c)))
  50
  51
  52 static void jffs2_build_inode_pass1(struct jffs2_sb_info *c,
  53                                     struct jffs2_inode_cache *ic,
  54                                     int *dir_hardlinks)
  55 {
  56         struct jffs2_full_dirent *fd;
  57
  58         dbg_fsbuild("building directory inode #%u\n", ic->ino);
  59
  60         /* For each child, increase nlink */
  61         for(fd = ic->scan_dents; fd; fd = fd->next) {
  62                 struct jffs2_inode_cache *child_ic;
  63                 if (!fd->ino)
  64                         continue;
  65
  66                 /* we can get high latency here with huge directories */
  67
  68                 child_ic = jffs2_get_ino_cache(c, fd->ino);
  69                 if (!child_ic) {
  70                         dbg_fsbuild("child \"%s\" (ino #%u) of dir ino #%u doesn't exist!\n",
  71                                   fd->name, fd->ino, ic->ino);
  72                         jffs2_mark_node_obsolete(c, fd->raw);
  73                         /* Clear the ic/raw union so it doesn't cause problems later. */
  74                         fd->ic = NULL;
  75                         continue;
  76                 }
  77
  78                 /* From this point, fd->raw is no longer used so we can set fd->ic */
  79                 fd->ic = child_ic;
  80                 child_ic->pino_nlink++;
  81                 /* If we appear (at this stage) to have hard-linked directories,
  82                  * set a flag to trigger a scan later */
  83                 if (fd->type == DT_DIR) {
  84                         child_ic->flags |= INO_FLAGS_IS_DIR;
  85                         if (child_ic->pino_nlink > 1)
  86                                 *dir_hardlinks = 1;
  87                 }
  88
  89                 dbg_fsbuild("increased nlink for child \"%s\" (ino #%u)\n", fd->name, fd->ino);
  90                 /* Can't free scan_dents so far. We might need them in pass 2 */
  91         }
  92 }
  93
  94 /* Scan plan:
  95  - Scan physical nodes. Build map of inodes/dirents. Allocate inocaches as we go
  96  - Scan directory tree from top down, setting nlink in inocaches
  97  - Scan inocaches for inodes with nlink==0
  98 */
  99 static int jffs2_build_filesystem(struct jffs2_sb_info *c)
 100 {
 101         int ret, i, dir_hardlinks = 0;
 102         struct jffs2_inode_cache *ic;
 103         struct jffs2_full_dirent *fd;
 104         struct jffs2_full_dirent *dead_fds = NULL;
 105
 106         dbg_fsbuild("build FS data structures\n");
 107
 108         /* First, scan the medium and build all the inode caches with
 109            lists of physical nodes */
 110
 111         c->flags |= JFFS2_SB_FLAG_SCANNING;
 112         ret = jffs2_scan_medium(c);
 113         c->flags &= ~JFFS2_SB_FLAG_SCANNING;
 114         if (ret)
 115                 goto exit;
 116
 117         dbg_fsbuild("scanned flash completely\n");
 118         jffs2_dbg_dump_block_lists_nolock(c);
 119
 120         dbg_fsbuild("pass 1 starting\n");
 121         c->flags |= JFFS2_SB_FLAG_BUILDING;
 122         /* Now scan the directory tree, increasing nlink according to every dirent found. */
 123         for_each_inode(i, c, ic) {
 124                 if (ic->scan_dents) {
 125                         jffs2_build_inode_pass1(c, ic, &dir_hardlinks);
 126                         cond_resched();
 127                 }
 128         }
 129
 130         dbg_fsbuild("pass 1 complete\n");
 131
 132         /* Next, scan for inodes with nlink == 0 and remove them. If
 133            they were directories, then decrement the nlink of their
 134            children too, and repeat the scan. As that's going to be
 135            a fairly uncommon occurrence, it's not so evil to do it this
 136            way. Recursion bad. */
 137         dbg_fsbuild("pass 2 starting\n");
 138
 139         for_each_inode(i, c, ic) {
 140                 if (ic->pino_nlink)
 141                         continue;
 142
 143                 jffs2_build_remove_unlinked_inode(c, ic, &dead_fds);
 144                 cond_resched();
 145         }
 146
 147         dbg_fsbuild("pass 2a starting\n");
 148
 149         while (dead_fds) {
 150                 fd = dead_fds;
 151                 dead_fds = fd->next;
 152
 153                 ic = jffs2_get_ino_cache(c, fd->ino);
 154
 155                 if (ic)
 156                         jffs2_build_remove_unlinked_inode(c, ic, &dead_fds);
 157                 jffs2_free_full_dirent(fd);
 158         }
 159
 160         dbg_fsbuild("pass 2a complete\n");
 161
 162         if (dir_hardlinks) {
 163                 /* If we detected directory hardlinks earlier, *hopefully*
 164                  * they are gone now because some of the links were from
 165                  * dead directories which still had some old dirents lying
 166                  * around and not yet garbage-collected, but which have
 167                  * been discarded above. So clear the pino_nlink field
 168                  * in each directory, so that the final scan below can
 169                  * print appropriate warnings. */
 170                 for_each_inode(i, c, ic) {
 171                         if (ic->flags & INO_FLAGS_IS_DIR)
 172                                 ic->pino_nlink = 0;
 173                 }
 174         }
 175         dbg_fsbuild("freeing temporary data structures\n");
 176
 177         /* Finally, we can scan again and free the dirent structs */
 178         for_each_inode(i, c, ic) {
 179                 while(ic->scan_dents) {
 180                         fd = ic->scan_dents;
 181                         ic->scan_dents = fd->next;
 182                         /* We do use the pino_nlink field to count nlink of
 183                          * directories during fs build, so set it to the
 184                          * parent ino# now. Now that there's hopefully only
 185                          * one. */
 186                         if (fd->type == DT_DIR) {
 187                                 if (!fd->ic) {
 188                                         /* We'll have complained about it and marked the coresponding
 189                                            raw node obsolete already. Just skip it. */
 190                                         continue;
 191                                 }
 192
 193                                 /* We *have* to have set this in jffs2_build_inode_pass1() */
 194                                 BUG_ON(!(fd->ic->flags & INO_FLAGS_IS_DIR));
 195
 196                                 /* We clear ic->pino_nlink ∀ directories' ic *only* if dir_hardlinks
 197                                  * is set. Otherwise, we know this should never trigger anyway, so
 198                                  * we don't do the check. And ic->pino_nlink still contains the nlink
 199                                  * value (which is 1). */
 200                                 if (dir_hardlinks && fd->ic->pino_nlink) {
 201                                         JFFS2_ERROR("child dir \"%s\" (ino #%u) of dir ino #%u is also hard linked from dir ino #%u\n",
 202                                                     fd->name, fd->ino, ic->ino, fd->ic->pino_nlink);
 203                                         /* Should we unlink it from its previous parent? */
 204                                 }
 205
 206                                 /* For directories, ic->pino_nlink holds that parent inode # */
 207                                 fd->ic->pino_nlink = ic->ino;
 208                         }
 209                         jffs2_free_full_dirent(fd);
 210                 }
 211                 ic->scan_dents = NULL;
 212                 cond_resched();
 213         }
 214         jffs2_build_xattr_subsystem(c);
 215         c->flags &= ~JFFS2_SB_FLAG_BUILDING;
 216
 217         dbg_fsbuild("FS build complete\n");
 218
 219         /* Rotate the lists by some number to ensure wear levelling */
 220         jffs2_rotate_lists(c);
 221
 222         ret = 0;
 223
 224 exit:
 225         if (ret) {
 226                 for_each_inode(i, c, ic) {
 227                         while(ic->scan_dents) {
 228                                 fd = ic->scan_dents;
 229                                 ic->scan_dents = fd->next;
 230                                 jffs2_free_full_dirent(fd);
 231                         }
 232                 }
 233                 jffs2_clear_xattr_subsystem(c);
 234         }
 235
 236         return ret;
 237 }
 238
 239 static void jffs2_build_remove_unlinked_inode(struct jffs2_sb_info *c,
 240                                         struct jffs2_inode_cache *ic,
 241                                         struct jffs2_full_dirent **dead_fds)
 242 {
 243         struct jffs2_raw_node_ref *raw;
 244         struct jffs2_full_dirent *fd;
 245
 246         dbg_fsbuild("removing ino #%u with nlink == zero.\n", ic->ino);
 247
 248         raw = ic->nodes;
 249         while (raw != (void *)ic) {
 250                 struct jffs2_raw_node_ref *next = raw->next_in_ino;
 251                 dbg_fsbuild("obsoleting node at 0x%08x\n", ref_offset(raw));
 252                 jffs2_mark_node_obsolete(c, raw);
 253                 raw = next;
 254         }
 255
 256         if (ic->scan_dents) {
 257                 int whinged = 0;
 258                 dbg_fsbuild("inode #%u was a directory which may have children...\n", ic->ino);
 259
 260                 while(ic->scan_dents) {
 261                         struct jffs2_inode_cache *child_ic;
 262
 263                         fd = ic->scan_dents;
 264                         ic->scan_dents = fd->next;
 265
 266                         if (!fd->ino) {
 267                                 /* It's a deletion dirent. Ignore it */
 268                                 dbg_fsbuild("child \"%s\" is a deletion dirent, skipping...\n", fd->name);
 269                                 jffs2_free_full_dirent(fd);
 270                                 continue;
 271                         }
 272                         if (!whinged)
 273                                 whinged = 1;
 274
 275                         dbg_fsbuild("removing child \"%s\", ino #%u\n", fd->name, fd->ino);
 276
 277                         child_ic = jffs2_get_ino_cache(c, fd->ino);
 278                         if (!child_ic) {
 279                                 dbg_fsbuild("cannot remove child \"%s\", ino #%u, because it doesn't exist\n",
 280                                                 fd->name, fd->ino);
 281                                 jffs2_free_full_dirent(fd);
 282                                 continue;
 283                         }
 284
 285                         /* Reduce nlink of the child. If it's now zero, stick it on the
 286                            dead_fds list to be cleaned up later. Else just free the fd */
 287                         child_ic->pino_nlink--;
 288
 289                         if (!child_ic->pino_nlink) {
 290                                 dbg_fsbuild("inode #%u (\"%s\") now has no links; adding to dead_fds list.\n",
 291                                           fd->ino, fd->name);
 292                                 fd->next = *dead_fds;
 293                                 *dead_fds = fd;
 294                         } else {
 295                                 dbg_fsbuild("inode #%u (\"%s\") has now got nlink %d. Ignoring.\n",
 296                                           fd->ino, fd->name, child_ic->pino_nlink);
 297                                 jffs2_free_full_dirent(fd);
 298                         }
 299                 }
 300         }
 301
 302         /*
 303            We don't delete the inocache from the hash list and free it yet.
 304            The erase code will do that, when all the nodes are completely gone.
 305         */
 306 }
 307
 308 static void jffs2_calc_trigger_levels(struct jffs2_sb_info *c)
 309 {
 310         uint32_t size;
 311
 312         /* Deletion should almost _always_ be allowed. We're fairly
 313            buggered once we stop allowing people to delete stuff
 314            because there's not enough free space... */
 315         c->resv_blocks_deletion = 2;
 316
 317         /* Be conservative about how much space we need before we allow writes.
 318            On top of that which is required for deletia, require an extra 2%
 319            of the medium to be available, for overhead caused by nodes being
 320            split across blocks, etc. */
 321
 322         size = c->flash_size / 50; /* 2% of flash size */
 323         size += c->nr_blocks * 100; /* And 100 bytes per eraseblock */
 324         size += c->sector_size - 1; /* ... and round up */
 325
 326         c->resv_blocks_write = c->resv_blocks_deletion + (size / c->sector_size);
 327
 328         /* When do we let the GC thread run in the background */
 329
 330         c->resv_blocks_gctrigger = c->resv_blocks_write + 1;
 331
 332         /* When do we allow garbage collection to merge nodes to make
 333            long-term progress at the expense of short-term space exhaustion? */
 334         c->resv_blocks_gcmerge = c->resv_blocks_deletion + 1;
 335
 336         /* When do we allow garbage collection to eat from bad blocks rather
 337            than actually making progress? */
 338         c->resv_blocks_gcbad = 0;//c->resv_blocks_deletion + 2;
 339
 340         /* What number of 'very dirty' eraseblocks do we allow before we
 341            trigger the GC thread even if we don't _need_ the space. When we
 342            can't mark nodes obsolete on the medium, the old dirty nodes cause
 343            performance problems because we have to inspect and discard them. */
 344         c->vdirty_blocks_gctrigger = c->resv_blocks_gctrigger;
 345         if (jffs2_can_mark_obsolete(c))
 346                 c->vdirty_blocks_gctrigger *= 10;
 347
 348         /* If there's less than this amount of dirty space, don't bother
 349            trying to GC to make more space. It'll be a fruitless task */
 350         c->nospc_dirty_size = c->sector_size + (c->flash_size / 100);
 351
 352         dbg_fsbuild("trigger levels (size %d KiB, block size %d KiB, %d blocks)\n",
 353                     c->flash_size / 1024, c->sector_size / 1024, c->nr_blocks);
 354         dbg_fsbuild("Blocks required to allow deletion:    %d (%d KiB)\n",
 355                   c->resv_blocks_deletion, c->resv_blocks_deletion*c->sector_size/1024);
 356         dbg_fsbuild("Blocks required to allow writes:      %d (%d KiB)\n",
 357                   c->resv_blocks_write, c->resv_blocks_write*c->sector_size/1024);
 358         dbg_fsbuild("Blocks required to quiesce GC thread: %d (%d KiB)\n",
 359                   c->resv_blocks_gctrigger, c->resv_blocks_gctrigger*c->sector_size/1024);
 360         dbg_fsbuild("Blocks required to allow GC merges:   %d (%d KiB)\n",
 361                   c->resv_blocks_gcmerge, c->resv_blocks_gcmerge*c->sector_size/1024);
 362         dbg_fsbuild("Blocks required to GC bad blocks:     %d (%d KiB)\n",
 363                   c->resv_blocks_gcbad, c->resv_blocks_gcbad*c->sector_size/1024);
 364         dbg_fsbuild("Amount of dirty space required to GC: %d bytes\n",
 365                   c->nospc_dirty_size);
 366         dbg_fsbuild("Very dirty blocks before GC triggered: %d\n",
 367                   c->vdirty_blocks_gctrigger);
 368 }
 369
 370 int jffs2_do_mount_fs(struct jffs2_sb_info *c)
 371 {
 372         int ret;
 373         int i;
 374         int size;
 375
 376         c->free_size = c->flash_size;
 377         c->nr_blocks = c->flash_size / c->sector_size;
 378         size = sizeof(struct jffs2_eraseblock) * c->nr_blocks;
 379 #ifndef __ECOS
 380         if (jffs2_blocks_use_vmalloc(c))
 381                 c->blocks = vzalloc(size);
 382         else
 383 #endif
 384                 c->blocks = kzalloc(size, GFP_KERNEL);
 385         if (!c->blocks)
 386                 return -ENOMEM;
 387
 388         for (i=0; i<c->nr_blocks; i++) {
 389                 INIT_LIST_HEAD(&c->blocks[i].list);
 390                 c->blocks[i].offset = i * c->sector_size;
 391                 c->blocks[i].free_size = c->sector_size;
 392         }
 393
 394         INIT_LIST_HEAD(&c->clean_list);
 395         INIT_LIST_HEAD(&c->very_dirty_list);
 396         INIT_LIST_HEAD(&c->dirty_list);
 397         INIT_LIST_HEAD(&c->erasable_list);
 398         INIT_LIST_HEAD(&c->erasing_list);
 399         INIT_LIST_HEAD(&c->erase_checking_list);
 400         INIT_LIST_HEAD(&c->erase_pending_list);
 401         INIT_LIST_HEAD(&c->erasable_pending_wbuf_list);
 402         INIT_LIST_HEAD(&c->erase_complete_list);
 403         INIT_LIST_HEAD(&c->free_list);
 404         INIT_LIST_HEAD(&c->bad_list);
 405         INIT_LIST_HEAD(&c->bad_used_list);
 406         c->highest_ino = 1;
 407         c->summary = NULL;
 408
 409         ret = jffs2_sum_init(c);
 410         if (ret)
 411                 goto out_free;
 412
 413         if (jffs2_build_filesystem(c)) {
 414                 dbg_fsbuild("build_fs failed\n");
 415                 jffs2_free_ino_caches(c);
 416                 jffs2_free_raw_node_refs(c);
 417                 ret = -EIO;
 418                 goto out_free;
 419         }
 420
 421         jffs2_calc_trigger_levels(c);
 422
 423         return 0;
 424
 425  out_free:
 426         kvfree(c->blocks);
 427
 428         return ret;
 429 }