build.c (12896B)
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 23static void jffs2_build_remove_unlinked_inode(struct jffs2_sb_info *, 24 struct jffs2_inode_cache *, struct jffs2_full_dirent **); 25 26static inline struct jffs2_inode_cache * 27first_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 36static inline struct jffs2_inode_cache * 37next_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 52static 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*/ 99static 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 224exit: 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 239static 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 308static 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 370int 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_sum_exit; 419 } 420 421 jffs2_calc_trigger_levels(c); 422 423 return 0; 424 425 out_sum_exit: 426 jffs2_sum_exit(c); 427 out_free: 428 kvfree(c->blocks); 429 430 return ret; 431}