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: gc.c,v 1.155 2005/11/07 11:14:39 gleixner Exp $ 11 * 12 */ 13 14 #include <linux/kernel.h> 15 #include <linux/mtd/mtd.h> 16 #include <linux/slab.h> 17 #include <linux/pagemap.h> 18 #include <linux/crc32.h> 19 #include <linux/compiler.h> 20 #include <linux/stat.h> 21 #include "nodelist.h" 22 #include "compr.h" 23 24 static int jffs2_garbage_collect_pristine(struct jffs2_sb_info *c, 25 struct jffs2_inode_cache *ic, 26 struct jffs2_raw_node_ref *raw); 27 static int jffs2_garbage_collect_metadata(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, 28 struct jffs2_inode_info *f, struct jffs2_full_dnode *fd); 29 static int jffs2_garbage_collect_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, 30 struct jffs2_inode_info *f, struct jffs2_full_dirent *fd); 31 static int jffs2_garbage_collect_deletion_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, 32 struct jffs2_inode_info *f, struct jffs2_full_dirent *fd); 33 static int jffs2_garbage_collect_hole(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, 34 struct jffs2_inode_info *f, struct jffs2_full_dnode *fn, 35 uint32_t start, uint32_t end); 36 static int jffs2_garbage_collect_dnode(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, 37 struct jffs2_inode_info *f, struct jffs2_full_dnode *fn, 38 uint32_t start, uint32_t end); 39 static int jffs2_garbage_collect_live(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, 40 struct jffs2_raw_node_ref *raw, struct jffs2_inode_info *f); 41 42 /* Called with erase_completion_lock held */ 43 static struct jffs2_eraseblock *jffs2_find_gc_block(struct jffs2_sb_info *c) 44 { 45 struct jffs2_eraseblock *ret; 46 struct list_head *nextlist = NULL; 47 int n = jiffies % 128; 48 49 /* Pick an eraseblock to garbage collect next. This is where we'll 50 put the clever wear-levelling algorithms. Eventually. */ 51 /* We possibly want to favour the dirtier blocks more when the 52 number of free blocks is low. */ 53 again: 54 if (!list_empty(&c->bad_used_list) && c->nr_free_blocks > c->resv_blocks_gcbad) { 55 D1(printk(KERN_DEBUG "Picking block from bad_used_list to GC next\n")); 56 nextlist = &c->bad_used_list; 57 } else if (n < 50 && !list_empty(&c->erasable_list)) { 58 /* Note that most of them will have gone directly to be erased. 59 So don't favour the erasable_list _too_ much. */ 60 D1(printk(KERN_DEBUG "Picking block from erasable_list to GC next\n")); 61 nextlist = &c->erasable_list; 62 } else if (n < 110 && !list_empty(&c->very_dirty_list)) { 63 /* Most of the time, pick one off the very_dirty list */ 64 D1(printk(KERN_DEBUG "Picking block from very_dirty_list to GC next\n")); 65 nextlist = &c->very_dirty_list; 66 } else if (n < 126 && !list_empty(&c->dirty_list)) { 67 D1(printk(KERN_DEBUG "Picking block from dirty_list to GC next\n")); 68 nextlist = &c->dirty_list; 69 } else if (!list_empty(&c->clean_list)) { 70 D1(printk(KERN_DEBUG "Picking block from clean_list to GC next\n")); 71 nextlist = &c->clean_list; 72 } else if (!list_empty(&c->dirty_list)) { 73 D1(printk(KERN_DEBUG "Picking block from dirty_list to GC next (clean_list was empty)\n")); 74 75 nextlist = &c->dirty_list; 76 } else if (!list_empty(&c->very_dirty_list)) { 77 D1(printk(KERN_DEBUG "Picking block from very_dirty_list to GC next (clean_list and dirty_list were empty)\n")); 78 nextlist = &c->very_dirty_list; 79 } else if (!list_empty(&c->erasable_list)) { 80 D1(printk(KERN_DEBUG "Picking block from erasable_list to GC next (clean_list and {very_,}dirty_list were empty)\n")); 81 82 nextlist = &c->erasable_list; 83 } else if (!list_empty(&c->erasable_pending_wbuf_list)) { 84 /* There are blocks are wating for the wbuf sync */ 85 D1(printk(KERN_DEBUG "Synching wbuf in order to reuse erasable_pending_wbuf_list blocks\n")); 86 spin_unlock(&c->erase_completion_lock); 87 jffs2_flush_wbuf_pad(c); 88 spin_lock(&c->erase_completion_lock); 89 goto again; 90 } else { 91 /* Eep. All were empty */ 92 D1(printk(KERN_NOTICE "jffs2: No clean, dirty _or_ erasable blocks to GC from! Where are they all?\n")); 93 return NULL; 94 } 95 96 ret = list_entry(nextlist->next, struct jffs2_eraseblock, list); 97 list_del(&ret->list); 98 c->gcblock = ret; 99 ret->gc_node = ret->first_node; 100 if (!ret->gc_node) { 101 printk(KERN_WARNING "Eep. ret->gc_node for block at 0x%08x is NULL\n", ret->offset); 102 BUG(); 103 } 104 105 /* Have we accidentally picked a clean block with wasted space ? */ 106 if (ret->wasted_size) { 107 D1(printk(KERN_DEBUG "Converting wasted_size %08x to dirty_size\n", ret->wasted_size)); 108 ret->dirty_size += ret->wasted_size; 109 c->wasted_size -= ret->wasted_size; 110 c->dirty_size += ret->wasted_size; 111 ret->wasted_size = 0; 112 } 113 114 return ret; 115 } 116 117 /* jffs2_garbage_collect_pass 118 * Make a single attempt to progress GC. Move one node, and possibly 119 * start erasing one eraseblock. 120 */ 121 int jffs2_garbage_collect_pass(struct jffs2_sb_info *c) 122 { 123 struct jffs2_inode_info *f; 124 struct jffs2_inode_cache *ic; 125 struct jffs2_eraseblock *jeb; 126 struct jffs2_raw_node_ref *raw; 127 int ret = 0, inum, nlink; 128 int xattr = 0; 129 130 if (down_interruptible(&c->alloc_sem)) 131 return -EINTR; 132 133 for (;;) { 134 spin_lock(&c->erase_completion_lock); 135 if (!c->unchecked_size) 136 break; 137 138 /* We can't start doing GC yet. We haven't finished checking 139 the node CRCs etc. Do it now. */ 140 141 /* checked_ino is protected by the alloc_sem */ 142 if (c->checked_ino > c->highest_ino && xattr) { 143 printk(KERN_CRIT "Checked all inodes but still 0x%x bytes of unchecked space?\n", 144 c->unchecked_size); 145 jffs2_dbg_dump_block_lists_nolock(c); 146 spin_unlock(&c->erase_completion_lock); 147 BUG(); 148 } 149 150 spin_unlock(&c->erase_completion_lock); 151 152 if (!xattr) 153 xattr = jffs2_verify_xattr(c); 154 155 spin_lock(&c->inocache_lock); 156 157 ic = jffs2_get_ino_cache(c, c->checked_ino++); 158 159 if (!ic) { 160 spin_unlock(&c->inocache_lock); 161 continue; 162 } 163 164 if (!ic->nlink) { 165 D1(printk(KERN_DEBUG "Skipping check of ino #%d with nlink zero\n", 166 ic->ino)); 167 spin_unlock(&c->inocache_lock); 168 continue; 169 } 170 switch(ic->state) { 171 case INO_STATE_CHECKEDABSENT: 172 case INO_STATE_PRESENT: 173 D1(printk(KERN_DEBUG "Skipping ino #%u already checked\n", ic->ino)); 174 spin_unlock(&c->inocache_lock); 175 continue; 176 177 case INO_STATE_GC: 178 case INO_STATE_CHECKING: 179 printk(KERN_WARNING "Inode #%u is in state %d during CRC check phase!\n", ic->ino, ic->state); 180 spin_unlock(&c->inocache_lock); 181 BUG(); 182 183 case INO_STATE_READING: 184 /* We need to wait for it to finish, lest we move on 185 and trigger the BUG() above while we haven't yet 186 finished checking all its nodes */ 187 D1(printk(KERN_DEBUG "Waiting for ino #%u to finish reading\n", ic->ino)); 188 /* We need to come back again for the _same_ inode. We've 189 made no progress in this case, but that should be OK */ 190 c->checked_ino--; 191 192 up(&c->alloc_sem); 193 sleep_on_spinunlock(&c->inocache_wq, &c->inocache_lock); 194 return 0; 195 196 default: 197 BUG(); 198 199 case INO_STATE_UNCHECKED: 200 ; 201 } 202 ic->state = INO_STATE_CHECKING; 203 spin_unlock(&c->inocache_lock); 204 205 D1(printk(KERN_DEBUG "jffs2_garbage_collect_pass() triggering inode scan of ino#%u\n", ic->ino)); 206 207 ret = jffs2_do_crccheck_inode(c, ic); 208 if (ret) 209 printk(KERN_WARNING "Returned error for crccheck of ino #%u. Expect badness...\n", ic->ino); 210 211 jffs2_set_inocache_state(c, ic, INO_STATE_CHECKEDABSENT); 212 up(&c->alloc_sem); 213 return ret; 214 } 215 216 /* First, work out which block we're garbage-collecting */ 217 jeb = c->gcblock; 218 219 if (!jeb) 220 jeb = jffs2_find_gc_block(c); 221 222 if (!jeb) { 223 D1 (printk(KERN_NOTICE "jffs2: Couldn't find erase block to garbage collect!\n")); 224 spin_unlock(&c->erase_completion_lock); 225 up(&c->alloc_sem); 226 return -EIO; 227 } 228 229 D1(printk(KERN_DEBUG "GC from block %08x, used_size %08x, dirty_size %08x, free_size %08x\n", jeb->offset, jeb->used_size, jeb->dirty_size, jeb->free_size)); 230 D1(if (c->nextblock) 231 printk(KERN_DEBUG "Nextblock at %08x, used_size %08x, dirty_size %08x, wasted_size %08x, free_size %08x\n", c->nextblock->offset, c->nextblock->used_size, c->nextblock->dirty_size, c->nextblock->wasted_size, c->nextblock->free_size)); 232 233 if (!jeb->used_size) { 234 up(&c->alloc_sem); 235 goto eraseit; 236 } 237 238 raw = jeb->gc_node; 239 240 while(ref_obsolete(raw)) { 241 D1(printk(KERN_DEBUG "Node at 0x%08x is obsolete... skipping\n", ref_offset(raw))); 242 raw = raw->next_phys; 243 if (unlikely(!raw)) { 244 printk(KERN_WARNING "eep. End of raw list while still supposedly nodes to GC\n"); 245 printk(KERN_WARNING "erase block at 0x%08x. free_size 0x%08x, dirty_size 0x%08x, used_size 0x%08x\n", 246 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size); 247 jeb->gc_node = raw; 248 spin_unlock(&c->erase_completion_lock); 249 up(&c->alloc_sem); 250 BUG(); 251 } 252 } 253 jeb->gc_node = raw; 254 255 D1(printk(KERN_DEBUG "Going to garbage collect node at 0x%08x\n", ref_offset(raw))); 256 257 if (!raw->next_in_ino) { 258 /* Inode-less node. Clean marker, snapshot or something like that */ 259 /* FIXME: If it's something that needs to be copied, including something 260 we don't grok that has JFFS2_NODETYPE_RWCOMPAT_COPY, we should do so */ 261 spin_unlock(&c->erase_completion_lock); 262 jffs2_mark_node_obsolete(c, raw); 263 up(&c->alloc_sem); 264 goto eraseit_lock; 265 } 266 267 ic = jffs2_raw_ref_to_ic(raw); 268 269 #ifdef CONFIG_JFFS2_FS_XATTR 270 /* When 'ic' refers xattr_datum/xattr_ref, this node is GCed as xattr. 271 * We can decide whether this node is inode or xattr by ic->class. */ 272 if (ic->class == RAWNODE_CLASS_XATTR_DATUM 273 || ic->class == RAWNODE_CLASS_XATTR_REF) { 274 BUG_ON(raw->next_in_ino != (void *)ic); 275 spin_unlock(&c->erase_completion_lock); 276 277 if (ic->class == RAWNODE_CLASS_XATTR_DATUM) { 278 ret = jffs2_garbage_collect_xattr_datum(c, (struct jffs2_xattr_datum *)ic); 279 } else { 280 ret = jffs2_garbage_collect_xattr_ref(c, (struct jffs2_xattr_ref *)ic); 281 } 282 goto release_sem; 283 } 284 #endif 285 286 /* We need to hold the inocache. Either the erase_completion_lock or 287 the inocache_lock are sufficient; we trade down since the inocache_lock 288 causes less contention. */ 289 spin_lock(&c->inocache_lock); 290 291 spin_unlock(&c->erase_completion_lock); 292 293 D1(printk(KERN_DEBUG "jffs2_garbage_collect_pass collecting from block @0x%08x. Node @0x%08x(%d), ino #%u\n", jeb->offset, ref_offset(raw), ref_flags(raw), ic->ino)); 294 295 /* Three possibilities: 296 1. Inode is already in-core. We must iget it and do proper 297 updating to its fragtree, etc. 298 2. Inode is not in-core, node is REF_PRISTINE. We lock the 299 inocache to prevent a read_inode(), copy the node intact. 300 3. Inode is not in-core, node is not pristine. We must iget() 301 and take the slow path. 302 */ 303 304 switch(ic->state) { 305 case INO_STATE_CHECKEDABSENT: 306 /* It's been checked, but it's not currently in-core. 307 We can just copy any pristine nodes, but have 308 to prevent anyone else from doing read_inode() while 309 we're at it, so we set the state accordingly */ 310 if (ref_flags(raw) == REF_PRISTINE) 311 ic->state = INO_STATE_GC; 312 else { 313 D1(printk(KERN_DEBUG "Ino #%u is absent but node not REF_PRISTINE. Reading.\n", 314 ic->ino)); 315 } 316 break; 317 318 case INO_STATE_PRESENT: 319 /* It's in-core. GC must iget() it. */ 320 break; 321 322 case INO_STATE_UNCHECKED: 323 case INO_STATE_CHECKING: 324 case INO_STATE_GC: 325 /* Should never happen. We should have finished checking 326 by the time we actually start doing any GC, and since 327 we're holding the alloc_sem, no other garbage collection 328 can happen. 329 */ 330 printk(KERN_CRIT "Inode #%u already in state %d in jffs2_garbage_collect_pass()!\n", 331 ic->ino, ic->state); 332 up(&c->alloc_sem); 333 spin_unlock(&c->inocache_lock); 334 BUG(); 335 336 case INO_STATE_READING: 337 /* Someone's currently trying to read it. We must wait for 338 them to finish and then go through the full iget() route 339 to do the GC. However, sometimes read_inode() needs to get 340 the alloc_sem() (for marking nodes invalid) so we must 341 drop the alloc_sem before sleeping. */ 342 343 up(&c->alloc_sem); 344 D1(printk(KERN_DEBUG "jffs2_garbage_collect_pass() waiting for ino #%u in state %d\n", 345 ic->ino, ic->state)); 346 sleep_on_spinunlock(&c->inocache_wq, &c->inocache_lock); 347 /* And because we dropped the alloc_sem we must start again from the 348 beginning. Ponder chance of livelock here -- we're returning success 349 without actually making any progress. 350 351 Q: What are the chances that the inode is back in INO_STATE_READING 352 again by the time we next enter this function? And that this happens 353 enough times to cause a real delay? 354 355 A: Small enough that I don't care :) 356 */ 357 return 0; 358 } 359 360 /* OK. Now if the inode is in state INO_STATE_GC, we are going to copy the 361 node intact, and we don't have to muck about with the fragtree etc. 362 because we know it's not in-core. If it _was_ in-core, we go through 363 all the iget() crap anyway */ 364 365 if (ic->state == INO_STATE_GC) { 366 spin_unlock(&c->inocache_lock); 367 368 ret = jffs2_garbage_collect_pristine(c, ic, raw); 369 370 spin_lock(&c->inocache_lock); 371 ic->state = INO_STATE_CHECKEDABSENT; 372 wake_up(&c->inocache_wq); 373 374 if (ret != -EBADFD) { 375 spin_unlock(&c->inocache_lock); 376 goto release_sem; 377 } 378 379 /* Fall through if it wanted us to, with inocache_lock held */ 380 } 381 382 /* Prevent the fairly unlikely race where the gcblock is 383 entirely obsoleted by the final close of a file which had 384 the only valid nodes in the block, followed by erasure, 385 followed by freeing of the ic because the erased block(s) 386 held _all_ the nodes of that inode.... never been seen but 387 it's vaguely possible. */ 388 389 inum = ic->ino; 390 nlink = ic->nlink; 391 spin_unlock(&c->inocache_lock); 392 393 f = jffs2_gc_fetch_inode(c, inum, nlink); 394 if (IS_ERR(f)) { 395 ret = PTR_ERR(f); 396 goto release_sem; 397 } 398 if (!f) { 399 ret = 0; 400 goto release_sem; 401 } 402 403 ret = jffs2_garbage_collect_live(c, jeb, raw, f); 404 405 jffs2_gc_release_inode(c, f); 406 407 release_sem: 408 up(&c->alloc_sem); 409 410 eraseit_lock: 411 /* If we've finished this block, start it erasing */ 412 spin_lock(&c->erase_completion_lock); 413 414 eraseit: 415 if (c->gcblock && !c->gcblock->used_size) { 416 D1(printk(KERN_DEBUG "Block at 0x%08x completely obsoleted by GC. Moving to erase_pending_list\n", c->gcblock->offset)); 417 /* We're GC'ing an empty block? */ 418 list_add_tail(&c->gcblock->list, &c->erase_pending_list); 419 c->gcblock = NULL; 420 c->nr_erasing_blocks++; 421 jffs2_erase_pending_trigger(c); 422 } 423 spin_unlock(&c->erase_completion_lock); 424 425 return ret; 426 } 427 428 static int jffs2_garbage_collect_live(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, 429 struct jffs2_raw_node_ref *raw, struct jffs2_inode_info *f) 430 { 431 struct jffs2_node_frag *frag; 432 struct jffs2_full_dnode *fn = NULL; 433 struct jffs2_full_dirent *fd; 434 uint32_t start = 0, end = 0, nrfrags = 0; 435 int ret = 0; 436 437 down(&f->sem); 438 439 /* Now we have the lock for this inode. Check that it's still the one at the head 440 of the list. */ 441 442 spin_lock(&c->erase_completion_lock); 443 444 if (c->gcblock != jeb) { 445 spin_unlock(&c->erase_completion_lock); 446 D1(printk(KERN_DEBUG "GC block is no longer gcblock. Restart\n")); 447 goto upnout; 448 } 449 if (ref_obsolete(raw)) { 450 spin_unlock(&c->erase_completion_lock); 451 D1(printk(KERN_DEBUG "node to be GC'd was obsoleted in the meantime.\n")); 452 /* They'll call again */ 453 goto upnout; 454 } 455 spin_unlock(&c->erase_completion_lock); 456 457 /* OK. Looks safe. And nobody can get us now because we have the semaphore. Move the block */ 458 if (f->metadata && f->metadata->raw == raw) { 459 fn = f->metadata; 460 ret = jffs2_garbage_collect_metadata(c, jeb, f, fn); 461 goto upnout; 462 } 463 464 /* FIXME. Read node and do lookup? */ 465 for (frag = frag_first(&f->fragtree); frag; frag = frag_next(frag)) { 466 if (frag->node && frag->node->raw == raw) { 467 fn = frag->node; 468 end = frag->ofs + frag->size; 469 if (!nrfrags++) 470 start = frag->ofs; 471 if (nrfrags == frag->node->frags) 472 break; /* We've found them all */ 473 } 474 } 475 if (fn) { 476 if (ref_flags(raw) == REF_PRISTINE) { 477 ret = jffs2_garbage_collect_pristine(c, f->inocache, raw); 478 if (!ret) { 479 /* Urgh. Return it sensibly. */ 480 frag->node->raw = f->inocache->nodes; 481 } 482 if (ret != -EBADFD) 483 goto upnout; 484 } 485 /* We found a datanode. Do the GC */ 486 if((start >> PAGE_CACHE_SHIFT) < ((end-1) >> PAGE_CACHE_SHIFT)) { 487 /* It crosses a page boundary. Therefore, it must be a hole. */ 488 ret = jffs2_garbage_collect_hole(c, jeb, f, fn, start, end); 489 } else { 490 /* It could still be a hole. But we GC the page this way anyway */ 491 ret = jffs2_garbage_collect_dnode(c, jeb, f, fn, start, end); 492 } 493 goto upnout; 494 } 495 496 /* Wasn't a dnode. Try dirent */ 497 for (fd = f->dents; fd; fd=fd->next) { 498 if (fd->raw == raw) 499 break; 500 } 501 502 if (fd && fd->ino) { 503 ret = jffs2_garbage_collect_dirent(c, jeb, f, fd); 504 } else if (fd) { 505 ret = jffs2_garbage_collect_deletion_dirent(c, jeb, f, fd); 506 } else { 507 printk(KERN_WARNING "Raw node at 0x%08x wasn't in node lists for ino #%u\n", 508 ref_offset(raw), f->inocache->ino); 509 if (ref_obsolete(raw)) { 510 printk(KERN_WARNING "But it's obsolete so we don't mind too much\n"); 511 } else { 512 jffs2_dbg_dump_node(c, ref_offset(raw)); 513 BUG(); 514 } 515 } 516 upnout: 517 up(&f->sem); 518 519 return ret; 520 } 521 522 static int jffs2_garbage_collect_pristine(struct jffs2_sb_info *c, 523 struct jffs2_inode_cache *ic, 524 struct jffs2_raw_node_ref *raw) 525 { 526 union jffs2_node_union *node; 527 struct jffs2_raw_node_ref *nraw; 528 size_t retlen; 529 int ret; 530 uint32_t phys_ofs, alloclen; 531 uint32_t crc, rawlen; 532 int retried = 0; 533 534 D1(printk(KERN_DEBUG "Going to GC REF_PRISTINE node at 0x%08x\n", ref_offset(raw))); 535 536 rawlen = ref_totlen(c, c->gcblock, raw); 537 538 /* Ask for a small amount of space (or the totlen if smaller) because we 539 don't want to force wastage of the end of a block if splitting would 540 work. */ 541 ret = jffs2_reserve_space_gc(c, min_t(uint32_t, sizeof(struct jffs2_raw_inode) + 542 JFFS2_MIN_DATA_LEN, rawlen), &phys_ofs, &alloclen, rawlen); 543 /* this is not the exact summary size of it, 544 it is only an upper estimation */ 545 546 if (ret) 547 return ret; 548 549 if (alloclen < rawlen) { 550 /* Doesn't fit untouched. We'll go the old route and split it */ 551 return -EBADFD; 552 } 553 554 node = kmalloc(rawlen, GFP_KERNEL); 555 if (!node) 556 return -ENOMEM; 557 558 ret = jffs2_flash_read(c, ref_offset(raw), rawlen, &retlen, (char *)node); 559 if (!ret && retlen != rawlen) 560 ret = -EIO; 561 if (ret) 562 goto out_node; 563 564 crc = crc32(0, node, sizeof(struct jffs2_unknown_node)-4); 565 if (je32_to_cpu(node->u.hdr_crc) != crc) { 566 printk(KERN_WARNING "Header CRC failed on REF_PRISTINE node at 0x%08x: Read 0x%08x, calculated 0x%08x\n", 567 ref_offset(raw), je32_to_cpu(node->u.hdr_crc), crc); 568 goto bail; 569 } 570 571 switch(je16_to_cpu(node->u.nodetype)) { 572 case JFFS2_NODETYPE_INODE: 573 crc = crc32(0, node, sizeof(node->i)-8); 574 if (je32_to_cpu(node->i.node_crc) != crc) { 575 printk(KERN_WARNING "Node CRC failed on REF_PRISTINE data node at 0x%08x: Read 0x%08x, calculated 0x%08x\n", 576 ref_offset(raw), je32_to_cpu(node->i.node_crc), crc); 577 goto bail; 578 } 579 580 if (je32_to_cpu(node->i.dsize)) { 581 crc = crc32(0, node->i.data, je32_to_cpu(node->i.csize)); 582 if (je32_to_cpu(node->i.data_crc) != crc) { 583 printk(KERN_WARNING "Data CRC failed on REF_PRISTINE data node at 0x%08x: Read 0x%08x, calculated 0x%08x\n", 584 ref_offset(raw), je32_to_cpu(node->i.data_crc), crc); 585 goto bail; 586 } 587 } 588 break; 589 590 case JFFS2_NODETYPE_DIRENT: 591 crc = crc32(0, node, sizeof(node->d)-8); 592 if (je32_to_cpu(node->d.node_crc) != crc) { 593 printk(KERN_WARNING "Node CRC failed on REF_PRISTINE dirent node at 0x%08x: Read 0x%08x, calculated 0x%08x\n", 594 ref_offset(raw), je32_to_cpu(node->d.node_crc), crc); 595 goto bail; 596 } 597 598 if (node->d.nsize) { 599 crc = crc32(0, node->d.name, node->d.nsize); 600 if (je32_to_cpu(node->d.name_crc) != crc) { 601 printk(KERN_WARNING "Name CRC failed on REF_PRISTINE dirent ode at 0x%08x: Read 0x%08x, calculated 0x%08x\n", 602 ref_offset(raw), je32_to_cpu(node->d.name_crc), crc); 603 goto bail; 604 } 605 } 606 break; 607 default: 608 printk(KERN_WARNING "Unknown node type for REF_PRISTINE node at 0x%08x: 0x%04x\n", 609 ref_offset(raw), je16_to_cpu(node->u.nodetype)); 610 goto bail; 611 } 612 613 nraw = jffs2_alloc_raw_node_ref(); 614 if (!nraw) { 615 ret = -ENOMEM; 616 goto out_node; 617 } 618 619 /* OK, all the CRCs are good; this node can just be copied as-is. */ 620 retry: 621 nraw->flash_offset = phys_ofs; 622 nraw->__totlen = rawlen; 623 nraw->next_phys = NULL; 624 625 ret = jffs2_flash_write(c, phys_ofs, rawlen, &retlen, (char *)node); 626 627 if (ret || (retlen != rawlen)) { 628 printk(KERN_NOTICE "Write of %d bytes at 0x%08x failed. returned %d, retlen %zd\n", 629 rawlen, phys_ofs, ret, retlen); 630 if (retlen) { 631 /* Doesn't belong to any inode */ 632 nraw->next_in_ino = NULL; 633 634 nraw->flash_offset |= REF_OBSOLETE; 635 jffs2_add_physical_node_ref(c, nraw); 636 jffs2_mark_node_obsolete(c, nraw); 637 } else { 638 printk(KERN_NOTICE "Not marking the space at 0x%08x as dirty because the flash driver returned retlen zero\n", nraw->flash_offset); 639 jffs2_free_raw_node_ref(nraw); 640 } 641 if (!retried && (nraw = jffs2_alloc_raw_node_ref())) { 642 /* Try to reallocate space and retry */ 643 uint32_t dummy; 644 struct jffs2_eraseblock *jeb = &c->blocks[phys_ofs / c->sector_size]; 645 646 retried = 1; 647 648 D1(printk(KERN_DEBUG "Retrying failed write of REF_PRISTINE node.\n")); 649 650 jffs2_dbg_acct_sanity_check(c,jeb); 651 jffs2_dbg_acct_paranoia_check(c, jeb); 652 653 ret = jffs2_reserve_space_gc(c, rawlen, &phys_ofs, &dummy, rawlen); 654 /* this is not the exact summary size of it, 655 it is only an upper estimation */ 656 657 if (!ret) { 658 D1(printk(KERN_DEBUG "Allocated space at 0x%08x to retry failed write.\n", phys_ofs)); 659 660 jffs2_dbg_acct_sanity_check(c,jeb); 661 jffs2_dbg_acct_paranoia_check(c, jeb); 662 663 goto retry; 664 } 665 D1(printk(KERN_DEBUG "Failed to allocate space to retry failed write: %d!\n", ret)); 666 jffs2_free_raw_node_ref(nraw); 667 } 668 669 jffs2_free_raw_node_ref(nraw); 670 if (!ret) 671 ret = -EIO; 672 goto out_node; 673 } 674 nraw->flash_offset |= REF_PRISTINE; 675 jffs2_add_physical_node_ref(c, nraw); 676 677 /* Link into per-inode list. This is safe because of the ic 678 state being INO_STATE_GC. Note that if we're doing this 679 for an inode which is in-core, the 'nraw' pointer is then 680 going to be fetched from ic->nodes by our caller. */ 681 spin_lock(&c->erase_completion_lock); 682 nraw->next_in_ino = ic->nodes; 683 ic->nodes = nraw; 684 spin_unlock(&c->erase_completion_lock); 685 686 jffs2_mark_node_obsolete(c, raw); 687 D1(printk(KERN_DEBUG "WHEEE! GC REF_PRISTINE node at 0x%08x succeeded\n", ref_offset(raw))); 688 689 out_node: 690 kfree(node); 691 return ret; 692 bail: 693 ret = -EBADFD; 694 goto out_node; 695 } 696 697 static int jffs2_garbage_collect_metadata(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, 698 struct jffs2_inode_info *f, struct jffs2_full_dnode *fn) 699 { 700 struct jffs2_full_dnode *new_fn; 701 struct jffs2_raw_inode ri; 702 struct jffs2_node_frag *last_frag; 703 union jffs2_device_node dev; 704 char *mdata = NULL, mdatalen = 0; 705 uint32_t alloclen, phys_ofs, ilen; 706 int ret; 707 708 if (S_ISBLK(JFFS2_F_I_MODE(f)) || 709 S_ISCHR(JFFS2_F_I_MODE(f)) ) { 710 /* For these, we don't actually need to read the old node */ 711 mdatalen = jffs2_encode_dev(&dev, JFFS2_F_I_RDEV(f)); 712 mdata = (char *)&dev; 713 D1(printk(KERN_DEBUG "jffs2_garbage_collect_metadata(): Writing %d bytes of kdev_t\n", mdatalen)); 714 } else if (S_ISLNK(JFFS2_F_I_MODE(f))) { 715 mdatalen = fn->size; 716 mdata = kmalloc(fn->size, GFP_KERNEL); 717 if (!mdata) { 718 printk(KERN_WARNING "kmalloc of mdata failed in jffs2_garbage_collect_metadata()\n"); 719 return -ENOMEM; 720 } 721 ret = jffs2_read_dnode(c, f, fn, mdata, 0, mdatalen); 722 if (ret) { 723 printk(KERN_WARNING "read of old metadata failed in jffs2_garbage_collect_metadata(): %d\n", ret); 724 kfree(mdata); 725 return ret; 726 } 727 D1(printk(KERN_DEBUG "jffs2_garbage_collect_metadata(): Writing %d bites of symlink target\n", mdatalen)); 728 729 } 730 731 ret = jffs2_reserve_space_gc(c, sizeof(ri) + mdatalen, &phys_ofs, &alloclen, 732 JFFS2_SUMMARY_INODE_SIZE); 733 if (ret) { 734 printk(KERN_WARNING "jffs2_reserve_space_gc of %zd bytes for garbage_collect_metadata failed: %d\n", 735 sizeof(ri)+ mdatalen, ret); 736 goto out; 737 } 738 739 last_frag = frag_last(&f->fragtree); 740 if (last_frag) 741 /* Fetch the inode length from the fragtree rather then 742 * from i_size since i_size may have not been updated yet */ 743 ilen = last_frag->ofs + last_frag->size; 744 else 745 ilen = JFFS2_F_I_SIZE(f); 746 747 memset(&ri, 0, sizeof(ri)); 748 ri.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK); 749 ri.nodetype = cpu_to_je16(JFFS2_NODETYPE_INODE); 750 ri.totlen = cpu_to_je32(sizeof(ri) + mdatalen); 751 ri.hdr_crc = cpu_to_je32(crc32(0, &ri, sizeof(struct jffs2_unknown_node)-4)); 752 753 ri.ino = cpu_to_je32(f->inocache->ino); 754 ri.version = cpu_to_je32(++f->highest_version); 755 ri.mode = cpu_to_jemode(JFFS2_F_I_MODE(f)); 756 ri.uid = cpu_to_je16(JFFS2_F_I_UID(f)); 757 ri.gid = cpu_to_je16(JFFS2_F_I_GID(f)); 758 ri.isize = cpu_to_je32(ilen); 759 ri.atime = cpu_to_je32(JFFS2_F_I_ATIME(f)); 760 ri.ctime = cpu_to_je32(JFFS2_F_I_CTIME(f)); 761 ri.mtime = cpu_to_je32(JFFS2_F_I_MTIME(f)); 762 ri.offset = cpu_to_je32(0); 763 ri.csize = cpu_to_je32(mdatalen); 764 ri.dsize = cpu_to_je32(mdatalen); 765 ri.compr = JFFS2_COMPR_NONE; 766 ri.node_crc = cpu_to_je32(crc32(0, &ri, sizeof(ri)-8)); 767 ri.data_crc = cpu_to_je32(crc32(0, mdata, mdatalen)); 768 769 new_fn = jffs2_write_dnode(c, f, &ri, mdata, mdatalen, phys_ofs, ALLOC_GC); 770 771 if (IS_ERR(new_fn)) { 772 printk(KERN_WARNING "Error writing new dnode: %ld\n", PTR_ERR(new_fn)); 773 ret = PTR_ERR(new_fn); 774 goto out; 775 } 776 jffs2_mark_node_obsolete(c, fn->raw); 777 jffs2_free_full_dnode(fn); 778 f->metadata = new_fn; 779 out: 780 if (S_ISLNK(JFFS2_F_I_MODE(f))) 781 kfree(mdata); 782 return ret; 783 } 784 785 static int jffs2_garbage_collect_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, 786 struct jffs2_inode_info *f, struct jffs2_full_dirent *fd) 787 { 788 struct jffs2_full_dirent *new_fd; 789 struct jffs2_raw_dirent rd; 790 uint32_t alloclen, phys_ofs; 791 int ret; 792 793 rd.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK); 794 rd.nodetype = cpu_to_je16(JFFS2_NODETYPE_DIRENT); 795 rd.nsize = strlen(fd->name); 796 rd.totlen = cpu_to_je32(sizeof(rd) + rd.nsize); 797 rd.hdr_crc = cpu_to_je32(crc32(0, &rd, sizeof(struct jffs2_unknown_node)-4)); 798 799 rd.pino = cpu_to_je32(f->inocache->ino); 800 rd.version = cpu_to_je32(++f->highest_version); 801 rd.ino = cpu_to_je32(fd->ino); 802 /* If the times on this inode were set by explicit utime() they can be different, 803 so refrain from splatting them. */ 804 if (JFFS2_F_I_MTIME(f) == JFFS2_F_I_CTIME(f)) 805 rd.mctime = cpu_to_je32(JFFS2_F_I_MTIME(f)); 806 else 807 rd.mctime = cpu_to_je32(0); 808 rd.type = fd->type; 809 rd.node_crc = cpu_to_je32(crc32(0, &rd, sizeof(rd)-8)); 810 rd.name_crc = cpu_to_je32(crc32(0, fd->name, rd.nsize)); 811 812 ret = jffs2_reserve_space_gc(c, sizeof(rd)+rd.nsize, &phys_ofs, &alloclen, 813 JFFS2_SUMMARY_DIRENT_SIZE(rd.nsize)); 814 if (ret) { 815 printk(KERN_WARNING "jffs2_reserve_space_gc of %zd bytes for garbage_collect_dirent failed: %d\n", 816 sizeof(rd)+rd.nsize, ret); 817 return ret; 818 } 819 new_fd = jffs2_write_dirent(c, f, &rd, fd->name, rd.nsize, phys_ofs, ALLOC_GC); 820 821 if (IS_ERR(new_fd)) { 822 printk(KERN_WARNING "jffs2_write_dirent in garbage_collect_dirent failed: %ld\n", PTR_ERR(new_fd)); 823 return PTR_ERR(new_fd); 824 } 825 jffs2_add_fd_to_list(c, new_fd, &f->dents); 826 return 0; 827 } 828 829 static int jffs2_garbage_collect_deletion_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, 830 struct jffs2_inode_info *f, struct jffs2_full_dirent *fd) 831 { 832 struct jffs2_full_dirent **fdp = &f->dents; 833 int found = 0; 834 835 /* On a medium where we can't actually mark nodes obsolete 836 pernamently, such as NAND flash, we need to work out 837 whether this deletion dirent is still needed to actively 838 delete a 'real' dirent with the same name that's still 839 somewhere else on the flash. */ 840 if (!jffs2_can_mark_obsolete(c)) { 841 struct jffs2_raw_dirent *rd; 842 struct jffs2_raw_node_ref *raw; 843 int ret; 844 size_t retlen; 845 int name_len = strlen(fd->name); 846 uint32_t name_crc = crc32(0, fd->name, name_len); 847 uint32_t rawlen = ref_totlen(c, jeb, fd->raw); 848 849 rd = kmalloc(rawlen, GFP_KERNEL); 850 if (!rd) 851 return -ENOMEM; 852 853 /* Prevent the erase code from nicking the obsolete node refs while 854 we're looking at them. I really don't like this extra lock but 855 can't see any alternative. Suggestions on a postcard to... */ 856 down(&c->erase_free_sem); 857 858 for (raw = f->inocache->nodes; raw != (void *)f->inocache; raw = raw->next_in_ino) { 859 860 /* We only care about obsolete ones */ 861 if (!(ref_obsolete(raw))) 862 continue; 863 864 /* Any dirent with the same name is going to have the same length... */ 865 if (ref_totlen(c, NULL, raw) != rawlen) 866 continue; 867 868 /* Doesn't matter if there's one in the same erase block. We're going to 869 delete it too at the same time. */ 870 if (SECTOR_ADDR(raw->flash_offset) == SECTOR_ADDR(fd->raw->flash_offset)) 871 continue; 872 873 D1(printk(KERN_DEBUG "Check potential deletion dirent at %08x\n", ref_offset(raw))); 874 875 /* This is an obsolete node belonging to the same directory, and it's of the right 876 length. We need to take a closer look...*/ 877 ret = jffs2_flash_read(c, ref_offset(raw), rawlen, &retlen, (char *)rd); 878 if (ret) { 879 printk(KERN_WARNING "jffs2_g_c_deletion_dirent(): Read error (%d) reading obsolete node at %08x\n", ret, ref_offset(raw)); 880 /* If we can't read it, we don't need to continue to obsolete it. Continue */ 881 continue; 882 } 883 if (retlen != rawlen) { 884 printk(KERN_WARNING "jffs2_g_c_deletion_dirent(): Short read (%zd not %u) reading header from obsolete node at %08x\n", 885 retlen, rawlen, ref_offset(raw)); 886 continue; 887 } 888 889 if (je16_to_cpu(rd->nodetype) != JFFS2_NODETYPE_DIRENT) 890 continue; 891 892 /* If the name CRC doesn't match, skip */ 893 if (je32_to_cpu(rd->name_crc) != name_crc) 894 continue; 895 896 /* If the name length doesn't match, or it's another deletion dirent, skip */ 897 if (rd->nsize != name_len || !je32_to_cpu(rd->ino)) 898 continue; 899 900 /* OK, check the actual name now */ 901 if (memcmp(rd->name, fd->name, name_len)) 902 continue; 903 904 /* OK. The name really does match. There really is still an older node on 905 the flash which our deletion dirent obsoletes. So we have to write out 906 a new deletion dirent to replace it */ 907 up(&c->erase_free_sem); 908 909 D1(printk(KERN_DEBUG "Deletion dirent at %08x still obsoletes real dirent \"%s\" at %08x for ino #%u\n", 910 ref_offset(fd->raw), fd->name, ref_offset(raw), je32_to_cpu(rd->ino))); 911 kfree(rd); 912 913 return jffs2_garbage_collect_dirent(c, jeb, f, fd); 914 } 915 916 up(&c->erase_free_sem); 917 kfree(rd); 918 } 919 920 /* FIXME: If we're deleting a dirent which contains the current mtime and ctime, 921 we should update the metadata node with those times accordingly */ 922 923 /* No need for it any more. Just mark it obsolete and remove it from the list */ 924 while (*fdp) { 925 if ((*fdp) == fd) { 926 found = 1; 927 *fdp = fd->next; 928 break; 929 } 930 fdp = &(*fdp)->next; 931 } 932 if (!found) { 933 printk(KERN_WARNING "Deletion dirent \"%s\" not found in list for ino #%u\n", fd->name, f->inocache->ino); 934 } 935 jffs2_mark_node_obsolete(c, fd->raw); 936 jffs2_free_full_dirent(fd); 937 return 0; 938 } 939 940 static int jffs2_garbage_collect_hole(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, 941 struct jffs2_inode_info *f, struct jffs2_full_dnode *fn, 942 uint32_t start, uint32_t end) 943 { 944 struct jffs2_raw_inode ri; 945 struct jffs2_node_frag *frag; 946 struct jffs2_full_dnode *new_fn; 947 uint32_t alloclen, phys_ofs, ilen; 948 int ret; 949 950 D1(printk(KERN_DEBUG "Writing replacement hole node for ino #%u from offset 0x%x to 0x%x\n", 951 f->inocache->ino, start, end)); 952 953 memset(&ri, 0, sizeof(ri)); 954 955 if(fn->frags > 1) { 956 size_t readlen; 957 uint32_t crc; 958 /* It's partially obsoleted by a later write. So we have to 959 write it out again with the _same_ version as before */ 960 ret = jffs2_flash_read(c, ref_offset(fn->raw), sizeof(ri), &readlen, (char *)&ri); 961 if (readlen != sizeof(ri) || ret) { 962 printk(KERN_WARNING "Node read failed in jffs2_garbage_collect_hole. Ret %d, retlen %zd. Data will be lost by writing new hole node\n", ret, readlen); 963 goto fill; 964 } 965 if (je16_to_cpu(ri.nodetype) != JFFS2_NODETYPE_INODE) { 966 printk(KERN_WARNING "jffs2_garbage_collect_hole: Node at 0x%08x had node type 0x%04x instead of JFFS2_NODETYPE_INODE(0x%04x)\n", 967 ref_offset(fn->raw), 968 je16_to_cpu(ri.nodetype), JFFS2_NODETYPE_INODE); 969 return -EIO; 970 } 971 if (je32_to_cpu(ri.totlen) != sizeof(ri)) { 972 printk(KERN_WARNING "jffs2_garbage_collect_hole: Node at 0x%08x had totlen 0x%x instead of expected 0x%zx\n", 973 ref_offset(fn->raw), 974 je32_to_cpu(ri.totlen), sizeof(ri)); 975 return -EIO; 976 } 977 crc = crc32(0, &ri, sizeof(ri)-8); 978 if (crc != je32_to_cpu(ri.node_crc)) { 979 printk(KERN_WARNING "jffs2_garbage_collect_hole: Node at 0x%08x had CRC 0x%08x which doesn't match calculated CRC 0x%08x\n", 980 ref_offset(fn->raw), 981 je32_to_cpu(ri.node_crc), crc); 982 /* FIXME: We could possibly deal with this by writing new holes for each frag */ 983 printk(KERN_WARNING "Data in the range 0x%08x to 0x%08x of inode #%u will be lost\n", 984 start, end, f->inocache->ino); 985 goto fill; 986 } 987 if (ri.compr != JFFS2_COMPR_ZERO) { 988 printk(KERN_WARNING "jffs2_garbage_collect_hole: Node 0x%08x wasn't a hole node!\n", ref_offset(fn->raw)); 989 printk(KERN_WARNING "Data in the range 0x%08x to 0x%08x of inode #%u will be lost\n", 990 start, end, f->inocache->ino); 991 goto fill; 992 } 993 } else { 994 fill: 995 ri.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK); 996 ri.nodetype = cpu_to_je16(JFFS2_NODETYPE_INODE); 997 ri.totlen = cpu_to_je32(sizeof(ri)); 998 ri.hdr_crc = cpu_to_je32(crc32(0, &ri, sizeof(struct jffs2_unknown_node)-4)); 999 1000 ri.ino = cpu_to_je32(f->inocache->ino); 1001 ri.version = cpu_to_je32(++f->highest_version); 1002 ri.offset = cpu_to_je32(start); 1003 ri.dsize = cpu_to_je32(end - start); 1004 ri.csize = cpu_to_je32(0); 1005 ri.compr = JFFS2_COMPR_ZERO; 1006 } 1007 1008 frag = frag_last(&f->fragtree); 1009 if (frag) 1010 /* Fetch the inode length from the fragtree rather then 1011 * from i_size since i_size may have not been updated yet */ 1012 ilen = frag->ofs + frag->size; 1013 else 1014 ilen = JFFS2_F_I_SIZE(f); 1015 1016 ri.mode = cpu_to_jemode(JFFS2_F_I_MODE(f)); 1017 ri.uid = cpu_to_je16(JFFS2_F_I_UID(f)); 1018 ri.gid = cpu_to_je16(JFFS2_F_I_GID(f)); 1019 ri.isize = cpu_to_je32(ilen); 1020 ri.atime = cpu_to_je32(JFFS2_F_I_ATIME(f)); 1021 ri.ctime = cpu_to_je32(JFFS2_F_I_CTIME(f)); 1022 ri.mtime = cpu_to_je32(JFFS2_F_I_MTIME(f)); 1023 ri.data_crc = cpu_to_je32(0); 1024 ri.node_crc = cpu_to_je32(crc32(0, &ri, sizeof(ri)-8)); 1025 1026 ret = jffs2_reserve_space_gc(c, sizeof(ri), &phys_ofs, &alloclen, 1027 JFFS2_SUMMARY_INODE_SIZE); 1028 if (ret) { 1029 printk(KERN_WARNING "jffs2_reserve_space_gc of %zd bytes for garbage_collect_hole failed: %d\n", 1030 sizeof(ri), ret); 1031 return ret; 1032 } 1033 new_fn = jffs2_write_dnode(c, f, &ri, NULL, 0, phys_ofs, ALLOC_GC); 1034 1035 if (IS_ERR(new_fn)) { 1036 printk(KERN_WARNING "Error writing new hole node: %ld\n", PTR_ERR(new_fn)); 1037 return PTR_ERR(new_fn); 1038 } 1039 if (je32_to_cpu(ri.version) == f->highest_version) { 1040 jffs2_add_full_dnode_to_inode(c, f, new_fn); 1041 if (f->metadata) { 1042 jffs2_mark_node_obsolete(c, f->metadata->raw); 1043 jffs2_free_full_dnode(f->metadata); 1044 f->metadata = NULL; 1045 } 1046 return 0; 1047 } 1048 1049 /* 1050 * We should only get here in the case where the node we are 1051 * replacing had more than one frag, so we kept the same version 1052 * number as before. (Except in case of error -- see 'goto fill;' 1053 * above.) 1054 */ 1055 D1(if(unlikely(fn->frags <= 1)) { 1056 printk(KERN_WARNING "jffs2_garbage_collect_hole: Replacing fn with %d frag(s) but new ver %d != highest_version %d of ino #%d\n", 1057 fn->frags, je32_to_cpu(ri.version), f->highest_version, 1058 je32_to_cpu(ri.ino)); 1059 }); 1060 1061 /* This is a partially-overlapped hole node. Mark it REF_NORMAL not REF_PRISTINE */ 1062 mark_ref_normal(new_fn->raw); 1063 1064 for (frag = jffs2_lookup_node_frag(&f->fragtree, fn->ofs); 1065 frag; frag = frag_next(frag)) { 1066 if (frag->ofs > fn->size + fn->ofs) 1067 break; 1068 if (frag->node == fn) { 1069 frag->node = new_fn; 1070 new_fn->frags++; 1071 fn->frags--; 1072 } 1073 } 1074 if (fn->frags) { 1075 printk(KERN_WARNING "jffs2_garbage_collect_hole: Old node still has frags!\n"); 1076 BUG(); 1077 } 1078 if (!new_fn->frags) { 1079 printk(KERN_WARNING "jffs2_garbage_collect_hole: New node has no frags!\n"); 1080 BUG(); 1081 } 1082 1083 jffs2_mark_node_obsolete(c, fn->raw); 1084 jffs2_free_full_dnode(fn); 1085 1086 return 0; 1087 } 1088 1089 static int jffs2_garbage_collect_dnode(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, 1090 struct jffs2_inode_info *f, struct jffs2_full_dnode *fn, 1091 uint32_t start, uint32_t end) 1092 { 1093 struct jffs2_full_dnode *new_fn; 1094 struct jffs2_raw_inode ri; 1095 uint32_t alloclen, phys_ofs, offset, orig_end, orig_start; 1096 int ret = 0; 1097 unsigned char *comprbuf = NULL, *writebuf; 1098 unsigned long pg; 1099 unsigned char *pg_ptr; 1100 1101 memset(&ri, 0, sizeof(ri)); 1102 1103 D1(printk(KERN_DEBUG "Writing replacement dnode for ino #%u from offset 0x%x to 0x%x\n", 1104 f->inocache->ino, start, end)); 1105 1106 orig_end = end; 1107 orig_start = start; 1108 1109 if (c->nr_free_blocks + c->nr_erasing_blocks > c->resv_blocks_gcmerge) { 1110 /* Attempt to do some merging. But only expand to cover logically 1111 adjacent frags if the block containing them is already considered 1112 to be dirty. Otherwise we end up with GC just going round in 1113 circles dirtying the nodes it already wrote out, especially 1114 on NAND where we have small eraseblocks and hence a much higher 1115 chance of nodes having to be split to cross boundaries. */ 1116 1117 struct jffs2_node_frag *frag; 1118 uint32_t min, max; 1119 1120 min = start & ~(PAGE_CACHE_SIZE-1); 1121 max = min + PAGE_CACHE_SIZE; 1122 1123 frag = jffs2_lookup_node_frag(&f->fragtree, start); 1124 1125 /* BUG_ON(!frag) but that'll happen anyway... */ 1126 1127 BUG_ON(frag->ofs != start); 1128 1129 /* First grow down... */ 1130 while((frag = frag_prev(frag)) && frag->ofs >= min) { 1131 1132 /* If the previous frag doesn't even reach the beginning, there's 1133 excessive fragmentation. Just merge. */ 1134 if (frag->ofs > min) { 1135 D1(printk(KERN_DEBUG "Expanding down to cover partial frag (0x%x-0x%x)\n", 1136 frag->ofs, frag->ofs+frag->size)); 1137 start = frag->ofs; 1138 continue; 1139 } 1140 /* OK. This frag holds the first byte of the page. */ 1141 if (!frag->node || !frag->node->raw) { 1142 D1(printk(KERN_DEBUG "First frag in page is hole (0x%x-0x%x). Not expanding down.\n", 1143 frag->ofs, frag->ofs+frag->size)); 1144 break; 1145 } else { 1146 1147 /* OK, it's a frag which extends to the beginning of the page. Does it live 1148 in a block which is still considered clean? If so, don't obsolete it. 1149 If not, cover it anyway. */ 1150 1151 struct jffs2_raw_node_ref *raw = frag->node->raw; 1152 struct jffs2_eraseblock *jeb; 1153 1154 jeb = &c->blocks[raw->flash_offset / c->sector_size]; 1155 1156 if (jeb == c->gcblock) { 1157 D1(printk(KERN_DEBUG "Expanding down to cover frag (0x%x-0x%x) in gcblock at %08x\n", 1158 frag->ofs, frag->ofs+frag->size, ref_offset(raw))); 1159 start = frag->ofs; 1160 break; 1161 } 1162 if (!ISDIRTY(jeb->dirty_size + jeb->wasted_size)) { 1163 D1(printk(KERN_DEBUG "Not expanding down to cover frag (0x%x-0x%x) in clean block %08x\n", 1164 frag->ofs, frag->ofs+frag->size, jeb->offset)); 1165 break; 1166 } 1167 1168 D1(printk(KERN_DEBUG "Expanding down to cover frag (0x%x-0x%x) in dirty block %08x\n", 1169 frag->ofs, frag->ofs+frag->size, jeb->offset)); 1170 start = frag->ofs; 1171 break; 1172 } 1173 } 1174 1175 /* ... then up */ 1176 1177 /* Find last frag which is actually part of the node we're to GC. */ 1178 frag = jffs2_lookup_node_frag(&f->fragtree, end-1); 1179 1180 while((frag = frag_next(frag)) && frag->ofs+frag->size <= max) { 1181 1182 /* If the previous frag doesn't even reach the beginning, there's lots 1183 of fragmentation. Just merge. */ 1184 if (frag->ofs+frag->size < max) { 1185 D1(printk(KERN_DEBUG "Expanding up to cover partial frag (0x%x-0x%x)\n", 1186 frag->ofs, frag->ofs+frag->size)); 1187 end = frag->ofs + frag->size; 1188 continue; 1189 } 1190 1191 if (!frag->node || !frag->node->raw) { 1192 D1(printk(KERN_DEBUG "Last frag in page is hole (0x%x-0x%x). Not expanding up.\n", 1193 frag->ofs, frag->ofs+frag->size)); 1194 break; 1195 } else { 1196 1197 /* OK, it's a frag which extends to the beginning of the page. Does it live 1198 in a block which is still considered clean? If so, don't obsolete it. 1199 If not, cover it anyway. */ 1200 1201 struct jffs2_raw_node_ref *raw = frag->node->raw; 1202 struct jffs2_eraseblock *jeb; 1203 1204 jeb = &c->blocks[raw->flash_offset / c->sector_size]; 1205 1206 if (jeb == c->gcblock) { 1207 D1(printk(KERN_DEBUG "Expanding up to cover frag (0x%x-0x%x) in gcblock at %08x\n", 1208 frag->ofs, frag->ofs+frag->size, ref_offset(raw))); 1209 end = frag->ofs + frag->size; 1210 break; 1211 } 1212 if (!ISDIRTY(jeb->dirty_size + jeb->wasted_size)) { 1213 D1(printk(KERN_DEBUG "Not expanding up to cover frag (0x%x-0x%x) in clean block %08x\n", 1214 frag->ofs, frag->ofs+frag->size, jeb->offset)); 1215 break; 1216 } 1217 1218 D1(printk(KERN_DEBUG "Expanding up to cover frag (0x%x-0x%x) in dirty block %08x\n", 1219 frag->ofs, frag->ofs+frag->size, jeb->offset)); 1220 end = frag->ofs + frag->size; 1221 break; 1222 } 1223 } 1224 D1(printk(KERN_DEBUG "Expanded dnode to write from (0x%x-0x%x) to (0x%x-0x%x)\n", 1225 orig_start, orig_end, start, end)); 1226 1227 D1(BUG_ON(end > frag_last(&f->fragtree)->ofs + frag_last(&f->fragtree)->size)); 1228 BUG_ON(end < orig_end); 1229 BUG_ON(start > orig_start); 1230 } 1231 1232 /* First, use readpage() to read the appropriate page into the page cache */ 1233 /* Q: What happens if we actually try to GC the _same_ page for which commit_write() 1234 * triggered garbage collection in the first place? 1235 * A: I _think_ it's OK. read_cache_page shouldn't deadlock, we'll write out the 1236 * page OK. We'll actually write it out again in commit_write, which is a little 1237 * suboptimal, but at least we're correct. 1238 */ 1239 pg_ptr = jffs2_gc_fetch_page(c, f, start, &pg); 1240 1241 if (IS_ERR(pg_ptr)) { 1242 printk(KERN_WARNING "read_cache_page() returned error: %ld\n", PTR_ERR(pg_ptr)); 1243 return PTR_ERR(pg_ptr); 1244 } 1245 1246 offset = start; 1247 while(offset < orig_end) { 1248 uint32_t datalen; 1249 uint32_t cdatalen; 1250 uint16_t comprtype = JFFS2_COMPR_NONE; 1251 1252 ret = jffs2_reserve_space_gc(c, sizeof(ri) + JFFS2_MIN_DATA_LEN, &phys_ofs, 1253 &alloclen, JFFS2_SUMMARY_INODE_SIZE); 1254 1255 if (ret) { 1256 printk(KERN_WARNING "jffs2_reserve_space_gc of %zd bytes for garbage_collect_dnode failed: %d\n", 1257 sizeof(ri)+ JFFS2_MIN_DATA_LEN, ret); 1258 break; 1259 } 1260 cdatalen = min_t(uint32_t, alloclen - sizeof(ri), end - offset); 1261 datalen = end - offset; 1262 1263 writebuf = pg_ptr + (offset & (PAGE_CACHE_SIZE -1)); 1264 1265 comprtype = jffs2_compress(c, f, writebuf, &comprbuf, &datalen, &cdatalen); 1266 1267 ri.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK); 1268 ri.nodetype = cpu_to_je16(JFFS2_NODETYPE_INODE); 1269 ri.totlen = cpu_to_je32(sizeof(ri) + cdatalen); 1270 ri.hdr_crc = cpu_to_je32(crc32(0, &ri, sizeof(struct jffs2_unknown_node)-4)); 1271 1272 ri.ino = cpu_to_je32(f->inocache->ino); 1273 ri.version = cpu_to_je32(++f->highest_version); 1274 ri.mode = cpu_to_jemode(JFFS2_F_I_MODE(f)); 1275 ri.uid = cpu_to_je16(JFFS2_F_I_UID(f)); 1276 ri.gid = cpu_to_je16(JFFS2_F_I_GID(f)); 1277 ri.isize = cpu_to_je32(JFFS2_F_I_SIZE(f)); 1278 ri.atime = cpu_to_je32(JFFS2_F_I_ATIME(f)); 1279 ri.ctime = cpu_to_je32(JFFS2_F_I_CTIME(f)); 1280 ri.mtime = cpu_to_je32(JFFS2_F_I_MTIME(f)); 1281 ri.offset = cpu_to_je32(offset); 1282 ri.csize = cpu_to_je32(cdatalen); 1283 ri.dsize = cpu_to_je32(datalen); 1284 ri.compr = comprtype & 0xff; 1285 ri.usercompr = (comprtype >> 8) & 0xff; 1286 ri.node_crc = cpu_to_je32(crc32(0, &ri, sizeof(ri)-8)); 1287 ri.data_crc = cpu_to_je32(crc32(0, comprbuf, cdatalen)); 1288 1289 new_fn = jffs2_write_dnode(c, f, &ri, comprbuf, cdatalen, phys_ofs, ALLOC_GC); 1290 1291 jffs2_free_comprbuf(comprbuf, writebuf); 1292 1293 if (IS_ERR(new_fn)) { 1294 printk(KERN_WARNING "Error writing new dnode: %ld\n", PTR_ERR(new_fn)); 1295 ret = PTR_ERR(new_fn); 1296 break; 1297 } 1298 ret = jffs2_add_full_dnode_to_inode(c, f, new_fn); 1299 offset += datalen; 1300 if (f->metadata) { 1301 jffs2_mark_node_obsolete(c, f->metadata->raw); 1302 jffs2_free_full_dnode(f->metadata); 1303 f->metadata = NULL; 1304 } 1305 } 1306 1307 jffs2_gc_release_page(c, pg_ptr, &pg); 1308 return ret; 1309 } 1310