build.c 13 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433
  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. #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  13. #include <linux/kernel.h>
  14. #include <linux/sched.h>
  15. #include <linux/slab.h>
  16. #include <linux/vmalloc.h>
  17. #include <linux/mtd/mtd.h>
  18. #include "nodelist.h"
  19. static void jffs2_build_remove_unlinked_inode(struct jffs2_sb_info *,
  20. struct jffs2_inode_cache *, struct jffs2_full_dirent **);
  21. static inline struct jffs2_inode_cache *
  22. first_inode_chain(int *i, struct jffs2_sb_info *c)
  23. {
  24. for (; *i < c->inocache_hashsize; (*i)++) {
  25. if (c->inocache_list[*i])
  26. return c->inocache_list[*i];
  27. }
  28. return NULL;
  29. }
  30. static inline struct jffs2_inode_cache *
  31. next_inode(int *i, struct jffs2_inode_cache *ic, struct jffs2_sb_info *c)
  32. {
  33. /* More in this chain? */
  34. if (ic->next)
  35. return ic->next;
  36. (*i)++;
  37. return first_inode_chain(i, c);
  38. }
  39. #define for_each_inode(i, c, ic) \
  40. for (i = 0, ic = first_inode_chain(&i, (c)); \
  41. ic; \
  42. ic = next_inode(&i, ic, (c)))
  43. static void jffs2_build_inode_pass1(struct jffs2_sb_info *c,
  44. struct jffs2_inode_cache *ic,
  45. int *dir_hardlinks)
  46. {
  47. struct jffs2_full_dirent *fd;
  48. dbg_fsbuild("building directory inode #%u\n", ic->ino);
  49. /* For each child, increase nlink */
  50. for(fd = ic->scan_dents; fd; fd = fd->next) {
  51. struct jffs2_inode_cache *child_ic;
  52. if (!fd->ino)
  53. continue;
  54. /* we can get high latency here with huge directories */
  55. child_ic = jffs2_get_ino_cache(c, fd->ino);
  56. if (!child_ic) {
  57. dbg_fsbuild("child \"%s\" (ino #%u) of dir ino #%u doesn't exist!\n",
  58. fd->name, fd->ino, ic->ino);
  59. jffs2_mark_node_obsolete(c, fd->raw);
  60. /* Clear the ic/raw union so it doesn't cause problems later. */
  61. fd->ic = NULL;
  62. continue;
  63. }
  64. /* From this point, fd->raw is no longer used so we can set fd->ic */
  65. fd->ic = child_ic;
  66. child_ic->pino_nlink++;
  67. /* If we appear (at this stage) to have hard-linked directories,
  68. * set a flag to trigger a scan later */
  69. if (fd->type == DT_DIR) {
  70. child_ic->flags |= INO_FLAGS_IS_DIR;
  71. if (child_ic->pino_nlink > 1)
  72. *dir_hardlinks = 1;
  73. }
  74. dbg_fsbuild("increased nlink for child \"%s\" (ino #%u)\n", fd->name, fd->ino);
  75. /* Can't free scan_dents so far. We might need them in pass 2 */
  76. }
  77. }
  78. /* Scan plan:
  79. - Scan physical nodes. Build map of inodes/dirents. Allocate inocaches as we go
  80. - Scan directory tree from top down, setting nlink in inocaches
  81. - Scan inocaches for inodes with nlink==0
  82. */
  83. static int jffs2_build_filesystem(struct jffs2_sb_info *c)
  84. {
  85. int ret, i, dir_hardlinks = 0;
  86. struct jffs2_inode_cache *ic;
  87. struct jffs2_full_dirent *fd;
  88. struct jffs2_full_dirent *dead_fds = NULL;
  89. dbg_fsbuild("build FS data structures\n");
  90. /* First, scan the medium and build all the inode caches with
  91. lists of physical nodes */
  92. c->flags |= JFFS2_SB_FLAG_SCANNING;
  93. ret = jffs2_scan_medium(c);
  94. c->flags &= ~JFFS2_SB_FLAG_SCANNING;
  95. if (ret)
  96. goto exit;
  97. dbg_fsbuild("scanned flash completely\n");
  98. jffs2_dbg_dump_block_lists_nolock(c);
  99. dbg_fsbuild("pass 1 starting\n");
  100. c->flags |= JFFS2_SB_FLAG_BUILDING;
  101. /* Now scan the directory tree, increasing nlink according to every dirent found. */
  102. for_each_inode(i, c, ic) {
  103. if (ic->scan_dents) {
  104. jffs2_build_inode_pass1(c, ic, &dir_hardlinks);
  105. cond_resched();
  106. }
  107. }
  108. dbg_fsbuild("pass 1 complete\n");
  109. /* Next, scan for inodes with nlink == 0 and remove them. If
  110. they were directories, then decrement the nlink of their
  111. children too, and repeat the scan. As that's going to be
  112. a fairly uncommon occurrence, it's not so evil to do it this
  113. way. Recursion bad. */
  114. dbg_fsbuild("pass 2 starting\n");
  115. for_each_inode(i, c, ic) {
  116. if (ic->pino_nlink)
  117. continue;
  118. jffs2_build_remove_unlinked_inode(c, ic, &dead_fds);
  119. cond_resched();
  120. }
  121. dbg_fsbuild("pass 2a starting\n");
  122. while (dead_fds) {
  123. fd = dead_fds;
  124. dead_fds = fd->next;
  125. ic = jffs2_get_ino_cache(c, fd->ino);
  126. if (ic)
  127. jffs2_build_remove_unlinked_inode(c, ic, &dead_fds);
  128. jffs2_free_full_dirent(fd);
  129. }
  130. dbg_fsbuild("pass 2a complete\n");
  131. if (dir_hardlinks) {
  132. /* If we detected directory hardlinks earlier, *hopefully*
  133. * they are gone now because some of the links were from
  134. * dead directories which still had some old dirents lying
  135. * around and not yet garbage-collected, but which have
  136. * been discarded above. So clear the pino_nlink field
  137. * in each directory, so that the final scan below can
  138. * print appropriate warnings. */
  139. for_each_inode(i, c, ic) {
  140. if (ic->flags & INO_FLAGS_IS_DIR)
  141. ic->pino_nlink = 0;
  142. }
  143. }
  144. dbg_fsbuild("freeing temporary data structures\n");
  145. /* Finally, we can scan again and free the dirent structs */
  146. for_each_inode(i, c, ic) {
  147. while(ic->scan_dents) {
  148. fd = ic->scan_dents;
  149. ic->scan_dents = fd->next;
  150. /* We do use the pino_nlink field to count nlink of
  151. * directories during fs build, so set it to the
  152. * parent ino# now. Now that there's hopefully only
  153. * one. */
  154. if (fd->type == DT_DIR) {
  155. if (!fd->ic) {
  156. /* We'll have complained about it and marked the coresponding
  157. raw node obsolete already. Just skip it. */
  158. continue;
  159. }
  160. /* We *have* to have set this in jffs2_build_inode_pass1() */
  161. BUG_ON(!(fd->ic->flags & INO_FLAGS_IS_DIR));
  162. /* We clear ic->pino_nlink ∀ directories' ic *only* if dir_hardlinks
  163. * is set. Otherwise, we know this should never trigger anyway, so
  164. * we don't do the check. And ic->pino_nlink still contains the nlink
  165. * value (which is 1). */
  166. if (dir_hardlinks && fd->ic->pino_nlink) {
  167. JFFS2_ERROR("child dir \"%s\" (ino #%u) of dir ino #%u is also hard linked from dir ino #%u\n",
  168. fd->name, fd->ino, ic->ino, fd->ic->pino_nlink);
  169. /* Should we unlink it from its previous parent? */
  170. }
  171. /* For directories, ic->pino_nlink holds that parent inode # */
  172. fd->ic->pino_nlink = ic->ino;
  173. }
  174. jffs2_free_full_dirent(fd);
  175. }
  176. ic->scan_dents = NULL;
  177. cond_resched();
  178. }
  179. jffs2_build_xattr_subsystem(c);
  180. c->flags &= ~JFFS2_SB_FLAG_BUILDING;
  181. dbg_fsbuild("FS build complete\n");
  182. /* Rotate the lists by some number to ensure wear levelling */
  183. jffs2_rotate_lists(c);
  184. ret = 0;
  185. exit:
  186. if (ret) {
  187. for_each_inode(i, c, ic) {
  188. while(ic->scan_dents) {
  189. fd = ic->scan_dents;
  190. ic->scan_dents = fd->next;
  191. jffs2_free_full_dirent(fd);
  192. }
  193. }
  194. jffs2_clear_xattr_subsystem(c);
  195. }
  196. return ret;
  197. }
  198. static void jffs2_build_remove_unlinked_inode(struct jffs2_sb_info *c,
  199. struct jffs2_inode_cache *ic,
  200. struct jffs2_full_dirent **dead_fds)
  201. {
  202. struct jffs2_raw_node_ref *raw;
  203. struct jffs2_full_dirent *fd;
  204. dbg_fsbuild("removing ino #%u with nlink == zero.\n", ic->ino);
  205. raw = ic->nodes;
  206. while (raw != (void *)ic) {
  207. struct jffs2_raw_node_ref *next = raw->next_in_ino;
  208. dbg_fsbuild("obsoleting node at 0x%08x\n", ref_offset(raw));
  209. jffs2_mark_node_obsolete(c, raw);
  210. raw = next;
  211. }
  212. if (ic->scan_dents) {
  213. int whinged = 0;
  214. dbg_fsbuild("inode #%u was a directory which may have children...\n", ic->ino);
  215. while(ic->scan_dents) {
  216. struct jffs2_inode_cache *child_ic;
  217. fd = ic->scan_dents;
  218. ic->scan_dents = fd->next;
  219. if (!fd->ino) {
  220. /* It's a deletion dirent. Ignore it */
  221. dbg_fsbuild("child \"%s\" is a deletion dirent, skipping...\n", fd->name);
  222. jffs2_free_full_dirent(fd);
  223. continue;
  224. }
  225. if (!whinged)
  226. whinged = 1;
  227. dbg_fsbuild("removing child \"%s\", ino #%u\n", fd->name, fd->ino);
  228. child_ic = jffs2_get_ino_cache(c, fd->ino);
  229. if (!child_ic) {
  230. dbg_fsbuild("cannot remove child \"%s\", ino #%u, because it doesn't exist\n",
  231. fd->name, fd->ino);
  232. jffs2_free_full_dirent(fd);
  233. continue;
  234. }
  235. /* Reduce nlink of the child. If it's now zero, stick it on the
  236. dead_fds list to be cleaned up later. Else just free the fd */
  237. child_ic->pino_nlink--;
  238. if (!child_ic->pino_nlink) {
  239. dbg_fsbuild("inode #%u (\"%s\") now has no links; adding to dead_fds list.\n",
  240. fd->ino, fd->name);
  241. fd->next = *dead_fds;
  242. *dead_fds = fd;
  243. } else {
  244. dbg_fsbuild("inode #%u (\"%s\") has now got nlink %d. Ignoring.\n",
  245. fd->ino, fd->name, child_ic->pino_nlink);
  246. jffs2_free_full_dirent(fd);
  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. static void jffs2_calc_trigger_levels(struct jffs2_sb_info *c)
  256. {
  257. uint32_t size;
  258. /* Deletion should almost _always_ be allowed. We're fairly
  259. buggered once we stop allowing people to delete stuff
  260. because there's not enough free space... */
  261. c->resv_blocks_deletion = 2;
  262. /* Be conservative about how much space we need before we allow writes.
  263. On top of that which is required for deletia, require an extra 2%
  264. of the medium to be available, for overhead caused by nodes being
  265. split across blocks, etc. */
  266. size = c->flash_size / 50; /* 2% of flash size */
  267. size += c->nr_blocks * 100; /* And 100 bytes per eraseblock */
  268. size += c->sector_size - 1; /* ... and round up */
  269. c->resv_blocks_write = c->resv_blocks_deletion + (size / c->sector_size);
  270. /* When do we let the GC thread run in the background */
  271. c->resv_blocks_gctrigger = c->resv_blocks_write + 1;
  272. /* When do we allow garbage collection to merge nodes to make
  273. long-term progress at the expense of short-term space exhaustion? */
  274. c->resv_blocks_gcmerge = c->resv_blocks_deletion + 1;
  275. /* When do we allow garbage collection to eat from bad blocks rather
  276. than actually making progress? */
  277. c->resv_blocks_gcbad = 0;//c->resv_blocks_deletion + 2;
  278. /* What number of 'very dirty' eraseblocks do we allow before we
  279. trigger the GC thread even if we don't _need_ the space. When we
  280. can't mark nodes obsolete on the medium, the old dirty nodes cause
  281. performance problems because we have to inspect and discard them. */
  282. c->vdirty_blocks_gctrigger = c->resv_blocks_gctrigger;
  283. if (jffs2_can_mark_obsolete(c))
  284. c->vdirty_blocks_gctrigger *= 10;
  285. /* If there's less than this amount of dirty space, don't bother
  286. trying to GC to make more space. It'll be a fruitless task */
  287. c->nospc_dirty_size = c->sector_size + (c->flash_size / 100);
  288. dbg_fsbuild("trigger levels (size %d KiB, block size %d KiB, %d blocks)\n",
  289. c->flash_size / 1024, c->sector_size / 1024, c->nr_blocks);
  290. dbg_fsbuild("Blocks required to allow deletion: %d (%d KiB)\n",
  291. c->resv_blocks_deletion, c->resv_blocks_deletion*c->sector_size/1024);
  292. dbg_fsbuild("Blocks required to allow writes: %d (%d KiB)\n",
  293. c->resv_blocks_write, c->resv_blocks_write*c->sector_size/1024);
  294. dbg_fsbuild("Blocks required to quiesce GC thread: %d (%d KiB)\n",
  295. c->resv_blocks_gctrigger, c->resv_blocks_gctrigger*c->sector_size/1024);
  296. dbg_fsbuild("Blocks required to allow GC merges: %d (%d KiB)\n",
  297. c->resv_blocks_gcmerge, c->resv_blocks_gcmerge*c->sector_size/1024);
  298. dbg_fsbuild("Blocks required to GC bad blocks: %d (%d KiB)\n",
  299. c->resv_blocks_gcbad, c->resv_blocks_gcbad*c->sector_size/1024);
  300. dbg_fsbuild("Amount of dirty space required to GC: %d bytes\n",
  301. c->nospc_dirty_size);
  302. dbg_fsbuild("Very dirty blocks before GC triggered: %d\n",
  303. c->vdirty_blocks_gctrigger);
  304. }
  305. int jffs2_do_mount_fs(struct jffs2_sb_info *c)
  306. {
  307. int ret;
  308. int i;
  309. int size;
  310. c->free_size = c->flash_size;
  311. c->nr_blocks = c->flash_size / c->sector_size;
  312. size = sizeof(struct jffs2_eraseblock) * c->nr_blocks;
  313. #ifndef __ECOS
  314. if (jffs2_blocks_use_vmalloc(c))
  315. c->blocks = vzalloc(size);
  316. else
  317. #endif
  318. c->blocks = kzalloc(size, GFP_KERNEL);
  319. if (!c->blocks)
  320. return -ENOMEM;
  321. for (i=0; i<c->nr_blocks; i++) {
  322. INIT_LIST_HEAD(&c->blocks[i].list);
  323. c->blocks[i].offset = i * c->sector_size;
  324. c->blocks[i].free_size = c->sector_size;
  325. }
  326. INIT_LIST_HEAD(&c->clean_list);
  327. INIT_LIST_HEAD(&c->very_dirty_list);
  328. INIT_LIST_HEAD(&c->dirty_list);
  329. INIT_LIST_HEAD(&c->erasable_list);
  330. INIT_LIST_HEAD(&c->erasing_list);
  331. INIT_LIST_HEAD(&c->erase_checking_list);
  332. INIT_LIST_HEAD(&c->erase_pending_list);
  333. INIT_LIST_HEAD(&c->erasable_pending_wbuf_list);
  334. INIT_LIST_HEAD(&c->erase_complete_list);
  335. INIT_LIST_HEAD(&c->free_list);
  336. INIT_LIST_HEAD(&c->bad_list);
  337. INIT_LIST_HEAD(&c->bad_used_list);
  338. c->highest_ino = 1;
  339. c->summary = NULL;
  340. ret = jffs2_sum_init(c);
  341. if (ret)
  342. goto out_free;
  343. if (jffs2_build_filesystem(c)) {
  344. dbg_fsbuild("build_fs failed\n");
  345. jffs2_free_ino_caches(c);
  346. jffs2_free_raw_node_refs(c);
  347. ret = -EIO;
  348. goto out_free;
  349. }
  350. jffs2_calc_trigger_levels(c);
  351. return 0;
  352. out_free:
  353. #ifndef __ECOS
  354. if (jffs2_blocks_use_vmalloc(c))
  355. vfree(c->blocks);
  356. else
  357. #endif
  358. kfree(c->blocks);
  359. return ret;
  360. }